DEVELOPMENT AND CHARACTERIZATION OF AN OPTICAL FIBER BASED INSTRUMENT FOR ULTRAVIOLET RESONANCE RAMAN SPECTROSCOPY OF BIOMOLECULES by LLOYD SHANE GREEK B.Sc. (Engineering) Queen's University at Kingston, 1992 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES Department of Electrical and Computer Engineering and The Biotechnology Laboratory We accept this thesis as conforming to th^required standard THE UNIVERSITY OF BRITISH COLUMBIA April 1998 © Lloyd Shane Greek, In p resent ing this thesis in partial fu l f i lment of the requ i rement s fo r an advanced deg ree at the Univers ity of British C o l u m b i a , I agree that t he Library shall make it freely available for re ference and study. I further agree that pe rmi s s i on fo r extens ive c o p y i n g of this thesis for scholar ly pu rpo se s may be g ranted by the head of my depa r tment or by his o r her representat ives. It is u n d e r s t o o d that c o p y i n g o r pub l i ca t i on of this thesis for f inancial gain shall no t b e a l l o w e d w i t hou t my wr i t ten permi s s ion . Depa r tment of EAecNrsccA C ow^l I V V < L K - t n ^ ^ e f ^ The Univers i ty of British C o l u m b i a ^N^VC-C-VV^VCS^ L_*3o vm DE-6 (2/88) A B S T R A C T Proteins are involved in virtually all natural biological processes, as well as many industrial and clinical applications. The function and activity of a protein are determined by its primary, secondary, tertiary and quaternary structures which dictate how it interacts with other biomolecules and its environment. Determination of protein structure is critical in elucidating mechanisms of protein action and in understanding protein behaviour, and protein-mediated processes and functions. A number of spectroscopic techniques are common for protein structure determination, the choice of which is often determined tradeoffs between (a) level of stucture required, (b) performance, (c) adaptability to in situ and/or in vivo use, and (d) cost. There exists a need for an inexpensive instrumental technique with moderate sensitivity to secondary and tertiary structure and an ability to operate remotely (in vivo and in situ). This document describes the design, development and application of the first fiber-optic linked instrument for ultraviolet resonance Raman spectroscopy (TO-UVRRS) which meets these criteria. A distinctly systems engineering approach to this problem was adopted. Starting with a definition of the problem and design criteria (Chapter 1), design of the FO-UVRRS system proceeded by considering the relationship between the major hardware components (Chapter 2). In optimizing the system, two major problems were encountered which resulted in detailed investigations: (a) stable transmission of high intensity ultraviolet light through new silica optical fibers without catastrophic bulk/surface damage or colour-centre induced solarization (Chapter 3), and (b) characterization and optimization of fiber-optic probe design for use in situ with highly absorbing samples (Chapter 4). Specialized signal-to-noise ratio enhancement techniques were investigated as a further means of improving the system (Chapter 5). The efficacy of the probes was demonstrated through applications to systems of biological import (Chapter 6), including specific and non-specific protein binding. These demonstrations comprise the first reported fiber-optic linked biophysical spectroscopic investigations at deep UV wavelengths and represent a significant contribution to biomolecular spectroscopy. Collectively, the research described here has resulted in novel designs, mathematical models, and optical materials and biophysical data which are immediately useful for UVRRS instrument design as well as other future applications. iv TABLE OF CONTENTS ABSTRACT ii TABLE OF CONTENTS iv LIST OF TABLES ix LIST OF FIGURES x ACKNOWLEDGEMENT xxi FOREWARD xxii Chapter 1. Introduction and Motivation 1 1.1 Overview of Motivation and Approach 1 1.2.1 Proteins - A Brief Introduction 2 1.2.2 Levels of Structure 4 1.2.3 Protein Adsorption 10 1.2.3.1 Introduction to Protein Adsorption 10 1.2.3.2 Thermodynamics of Protein Adsorption 13 1.3 Comparison of Spectroscopic Methods to Determine Protein Structure 18 1.3.1 General Comments on Spectroscopic Methods 18 1.3.2 X-Ray Diffraction 22 1.3.3 Nuclear Magnetic Resonance Spectroscopy 23 1.3.4 Ultraviolet-Visible Absorption Spectrophotometry 24 1.3.5 Electronic Circular Dichroism Spectroscopy 25 1.3.6 Infrared Absorption Spectroscopy 28 1.3.7 Vibrational Circular Dichroism Spectroscopy 31 1.3.8 Fluorescence Spectroscopy 32 1.3.9 Non Resonance Raman Spectroscopy 35 1.3.10 Ultraviolet Resonance Raman Spectroscopy 36 1.3.11 Summary and Outlook 39 1.4 Overview of UVRRS for Protein Studies 40 1.4.1 General Comments on UVRRS 40 1.4.2 Molecular Vibrations 41 1.4.3 Regular Raman Spectroscopy 43 1.4.4 Visible and Ultraviolet Resonance Raman Spectroscopy: Theory 48 1.4.5 Aromatic Amino Acid Resonance Raman Enhancement 54 1.4.6 Peptide Backbone Resonance Raman Enhancement 59 1.4.7 Other Resonance Enhancable Protein Chromophores 61 1.4.8 The Need for Remote UVRRS Probes 62 Chapter 2. General System Design 63 2.1 Overview of System Design 63 2.1.1 Instrument Performance Goals and Design Requirements 63 2.1.2 Hardware Overview 64 2.2 Light Source 66 2.2.1 General Comments on RS and UVRRS Light Sources 66 2.2.2 Light Source For Visible and Near-UV studies 68 2.2.3 Light Source for DUV Studies 69 2.3 Fiber Light Delivery and Collection 74 2.3.2 Fiber-Optic Probe design Considerations 77 2.4 Wavelength Separation 79 2.4.1 General Comments on Raman Spectrometers 79 2.4.2 The Czerny-Turner Grating Based Spectrometer 81 2.4.3 Rayleigh Light Rejection 82 2.5 Optical detection 84 2.5.1 General Comments on Optical Detectors for UVRRS 84 2.5.2 The Intensified Photo-Diode Array and Associated Hardware 85 2.6 Data Collection and Processing 87 2.6.1 Equipment 87 2.6.2 Wavenumber Calibration 87 2.7 System Throughput 89 Chapter 3. Characterization of Silica Fibers for High Intensity UV Transmission... 91 3.1 Overview of Fiber Characterization, Attenuation and Damage Mechanisms 91 3.1.1 Motivation for Fiber Characterization 91 3.1.2 Introduction to Fiber-Optic DUV Light Transmission 92 3.1.3 Low Intensity Loss Mechanisms 94 3.1.4 Catastrophic Surface and Bulk Damage 97 3.1.5 High Intensity UV Loss and Damage Mechanisms 99 3.2 DUV Characterization of Improved-Ultraviolet Fibers for UVRRS 105 3.2.1 Description of New Fibers 105 3.2.2 Catastrophic Damage 109 3.2.3 Characterization with Respect to Energy and Wavelength 112 3.2.4 Characterization with Respect to Length and Area 117 3.2.5 Calculation of Optical Transmission Parameters 119 Chapter 4. Design and Optimization of Fiber-Optic UVRRS Probes 124 4.1 Overview and Background of Probe Design and Optimization 124 4.1.1 Summary of Probe Investigations 124 4.1.2 Background and History of Fiber-Optic RS Probes 124 4.2 Flush Probe Designs 126 4.2.1 Overview of Flush Probe Investigations 126 4.2.2 Rationale, Theory and Simulation of Flush Probe Operation 127 4.2.3 Experimental Results and Discussion 136 4.2.3.1 Materials and Methods for Flush Probe Investigations 136 4.2.3.2 Modeling 139 4.2.3.3 Discussion 142 4.2.3.4 Flush Probe Investigation Conclusions 153 4.2.3.5 Probe Tip Geometry Modifications 154 4.3 Improved Probe Design 157 4.3.1 Rationale for Improved Probe Design 157 4.3.2 Materials and Methods for Angled Probe Construction and Testing 158 4.3.2.1 Probe Fabrication 158 4.3.2.2 UVRRS Data 162 4.3.3 Improved Probe Design and Characterization 162 4.3.3.1 Probe Design Considerations 162 4.3.3.2 UVRRS Data with Improved Probe 163 4.3.4 Optimal Excitation Wavelength 168 4.3.5 Probe Lifetime Considerations 174 4.3.6 Conclusions from Improved Probe Design Work 175 4.4 Theoretical and Geometrical Analytical Limits of UVRRS 177 4.4.1 Overview of Theoretical and Geometrical Limit Investigations 177 4.4.1.1 Summary of Analytical Limit Investigations 177 4.4.1.2 Motivation for Analytical Limit Investigations 177 4.4.2 Theoretical Limits 179 4.4.2.1 General Comments on the Theoretical Limits of UVRRS 179 4.4.2.2 Absorbing Analytes in Non-Absorbing Media 180 4.4.2.3 Absorbing Analytes in Absorbing Media 190 4.4.3 Geometrical Limits 192 4.4.3.1 General Comments on Analytical FO-UVRRS 192 4.4.3.2 Development of FO-UVRRS Simulation 193 4.4.3.3 Determination of Geometrical Analytical Limits to FO-UVRRS 198 4.4.3.4 Working Curves and Analytical Limits 199 4.4.3.5 Mapping the Probe Volume 201 Chapter 5. Signal Processing for UVRRS 203 5.1 Overview of Signal Processing 203 5.2 Signal Characterization 204 5.2.1 Description of Raman Spectra 204 5.2.2 Investigation of the Sampling Process 207 5.2.3 PSD Estimation Methods 208 5.2.3.1 Overview of PSD Estimation Methods 208 5.2.3.2 Non-Parametric PSD Estimation - Welch's method 209 5.2.3.3 Parametric PSD Estimation Using an Autoregressive Model 215 5.3 FFT Based Methods for SNR Enhancement 216 5.3.1 FIR Techniques Used 216 5.3.1.1 Introduction to FLR Techniques 216 5.3.1.2 Determination of Centre of Transition Band 217 5.3.1.3 Performance of FLR Filters 219 vii 5.3.2 IIR Techniques 220 5.4 Regularization Based Methods 221 5.4.1 Theory of Regularization Based Methods 221 5.4.1.1 The Signal Recovery Problem 221 5.4.1.2 Background and Introduction to Regularization Methods 222 5.4.1.3 TPMEM 226 5.4.2 Algorithm and Computational Details 227 5.4.3 Experimental 230 5.4.3.1 Artificial Data Simulations 230 5.4.3.2 Experimental Raman Spectra 231 5.4.3.3 Performance Evaluation 232 5.4.4 Results and Discussion 234 5.4.4.1 Low SNR Signal Enhancement 234 5.4.4.2 Deconvolutions 236 Chapter 6. Biochemical Investigations 241 6.1 Introduction to FO-UVRRS Biochemical Investigations 241 6.1.1 Summary of Biochemical FO-UVRRS Investigations 241 6.1.2 Motivation for FO-UVRRS Biochemical Studies 242 6.1.3 Material and Methods for Biochemical FO-UVRRS 245 6.2 Probe Validation and Performance Characterization 247 6.2.1 RNase TI Denaturation 247 6.2.2 Resonance Enhancement of Phosphotyrosine 250 6.2.3 Analytes Bound to Insoluble Substrates 253 6.2.3.1 Introduction to Insoluble Substrate Binding Studies 253 6.2.3.2 CBD Binding To Cellulose 255 6.2.3.3 Lysozyme Binding to Silica 256 6.2.4 FO-UVRRS Data of DNA 258 6.3 CBD Studies 259 6.3.1 Introduction and Motivation for CBD Studies 259 6.3.1.1 Motivation and Summary of CBD Studies 259 6.3.1.2 Cellulose Binding Domains 260 6.3.1.3 CBDcex 261 6.3.1.4 C B D N ! 263 6.3.2 CBDcex Studies 266 6.3.2.1 Materials and Methods for CBDcex Studies 266 6.3.2.2 CBDcex Results and Discussion 267 6.3.2.3 Conclusions from CBDcex Glycosylation Studies 272 6.3.3 CBDNI Studies 273 6.3.3.1 Materials and Methods for CBDNI Studies 273 6.3.3.2 CBDNI Results and Discussion 274 6 1 3.3 Conclusions from Fiber-Optic UVRRS CBDNI Binding Studies 282 viii Chapter 7. Conclusions and Future Directions 284 7.1 Overview of Conclusions and Future Directions 284 7.2 General System Design 285 7.2.1 Conclusions from System Design 285 7.2.2 Future Directions in System Design 287 7.2.2.1 Ultraviolet Lasers 287 7.2.2.2 Improved Spectrometers 288 7.2.2.3 Improved Rayleigh Light Rejection 289 7.2.2.4 Improved Detectors 290 7.2.2.5 Imaging 291 7.2.2.6 Ultimate Short Term Improvement from Component Changes 291 7.3 Transmission of UV Light Through Optical Fibers 293 7.3.1 Conclusions 293 7.3.2 Future Directions 293 7.4 Fiber-Optic Probes 294 7.4.1 Conclusions from Probe Design and Modelling 294 7.4.2 Future Directions in Probe Design 295 7.5 Signal Processing 296 7.5.1 Conclusions on Signal Processing 296 7.5.2 Future Directions in Signal Processing 296 7.6 Biochemical Applications 297 7.6.1 Conclusions from Biochemical Applications 297 7.6.2 Future Directions in Biochemical Studies 298 7.6.2.1 Improved Signal Understanding and Calibration 298 7.6.2.2 Binding and Adsorption Studies 299 7.6.2.3 Bioanalytical Applications 300 7.6.2.4 Clinical, Environmental and Industrial Applications 301 7.7 Final Concluding Remarks 302 Chapter 8. References 303 Appendix 1. Flush Probe Simulations 318 Appendix 2. Angled/Mirrored Probe Simulation 329 Appendix 3. Signal Processing Programs 341 IX LIST OF TABLES Table L l 16 Primary subprocesses involved in protein adsorption. Table 1.2 21 Comparison of spectroscopic methods for investigating protein structure Table 3.1 121 The estimated saturated induced attenuation coefficient at wavelengths from 215 nm to 240 nm. Table 4.1 143 Geometry and performance of probes used in these studies. Table 4.2 172 Approximate fits and parameters for components of SC(A,). Table 4.3 190 Theoretical optimal limits of detection and limits of quantification for several wavelengths, analytes, and Raman peaks using 100 seconds of 1 mW excitation. Sample volume penetration was set to 600 u.m. Table 4.4 201 Geometrical optimal limits of detection and limits of quantification for several wavelengths, analytes, and Raman peaks obtained using the computer program in appendix 2 and considering 100 seconds of 1 mW excitation. Probe geometry was taken to be a single 600 urn excitation fiber and a single 600 urn angled, mirrored collection fiber. Table 5.1 205 DNA data sets Table 5.2 205 MO-KN data sets Table 5.3 235 Figures of merit for the performance of TPMEM, fourth order Savitzky-Golay filtering (SG-4) and zeroth order Savitzky-Golay filtering (SG-0). Table 7.1 292 Best case estimated system performance improvement factors for planned component additions. LIST OF FIGURES Figure 1.1 3 The twenty common amino acids and their three letter abbreviations. Hydrogen atoms and charges are omitted for clarity. Figure 1.2 5 (A) The geometry of the peptide backbone showing a beta sheet structure with alpha carbons as bullets and hydrogen bonds as dotted lines. (B) The amino acids serine and tyrosine joining in a condensation reaction to form a di-peptide. Figure 1.3 5 "Ribbon" diagram of Horse Heart Myoglobin, a 153 amino acid residue oxygen-storage protein composed almost entirely of a-helices, as denoted by the helical ribbons shown here. Figure 1.4 6 Ribbon diagram of a-chymotrypson, a 247 amino acid residue protein composed of 45% P-sheet (mostly on the left in this figure), 40% random coil (mostly on the right), and 15% a-helix (difficult to see in this figure.) Figure 1.5 43 Schematic representation of Rayleigh (elastic) and Raman (inelastic) scattering. Stokes and anti-Stokes light are scattered at lower and higher energies, respectively, than the incident light. Figure 1.6 44 A schematic representation of the basic setup for Raman spectroscopy. Figure 1.7 47 Low frequency Raman spectrum of neat ethanol taken with a fiber-optic probe and 227 nm, 250 u,W, 15 second excitation. Figure 1.8 48 Raman spectra of methanol, ethanol, and isopropanol taken with a fiber-optic probe and 227 nm, 250 u,W, 15 second excitation. Figure 1.9 49 Energy level diagrams for infrared absorption, Raleigh, Raman, resonance Raman, and fluorescence events. E Q and Ei indicate ground and excited electronic states, v indicates the vibrational energy level of the ground electronic state, and v' indicates the vibrational state of the excited electronic state. Figure 1.10 53 The UV-VIS absorption spectrum of a fictitious molecule (large trace) and the Raman spectra (insets) that may be obtained for visible, non resonance conditions (A); visible or near-UV resonance conditions (B) or deep-UV resonance conditions (C). Figure 2.1 65 Overview of the laser systems used in these studies for the production of (A) 266 nm pulsed UV light, (B) 205-250 nm tunable pulsed UV light, and (C) discrete visible and near-UV laser lines from 363.8 nm to 514 nm. Figure 2.2 70 Detailed schematic of the DUV light system, m: mirror, G: grating, L: lens, osc: laser oscillator, amp: laser amplifier, PR: quarter wave plate polarization rotator, P: silica prism, KDP: potassium dihydrogen phosphate crystal, BBO: P-barium borate crystal. Figure 2.3 71 Pulse width measurement of 355 nm Nd:YAG pulse. Bottom axis divisions are 5 ns. Figure 2.4 74 Pulse shape measurement of 230 nm (doubled 460 nm) pulse. Bottom axis divisions are 5 ns. Figure 2.5 75 Expanded pictorial view of the three concentric layers of a typical optical fiber. Figure 2.6 79 Temporal pulse profiles at 225 nm and ca. 30 uJ/pulse for fiber input (left) and output (right). Note different time scale divisions (2 ns/division on the left and 5 ns/division on the right). Figure 2.7 81 The Czerny-Turner spectrometer and associated hardware. SM: single monochromator, S: bilateral slit assembly, FOC: Fiber-optic coupler, Mj: folding mirror, Mf-: parabolic focusing mirror, G: holographic grating, Mc: collimating mirror, D: detector, DSM: double subtractive monochromator. The dotted lines indicate the general optical path. Figure 2.8 84 The absorption spectrum of acenaphthene dissolved in HPLC-grade methanol. Figure 2.9 86 A schematic diagram of the information flow in the FO-UVRRS system. Solid lines represent electrical signals and dotted lines represent optical signals. Arrowheads indicate the direction of signal flow. XI1 Figure 2.10 88 Left: Top trace shows ethanol calibration spectrum (with peaks used for calibration labeled) using 30 s, 226 nm excitation and a fiber-optic probe. The bottom trace shows the derivative spectrum and arrows indicate the zero crossings used for calibration. Right: Plot of peak wavenumber vs. measured diode position for the spectrum on the left as well as one taken 1 hour later, after completing a set of experiments. In all cases the peak positions are repeatable to within 1 diode. The straight line shows the fit indicating a dispersion of 1.52 cm'Vdiode and a correlation coefficient of R=-0.99993. Figure 2.11 90 A block diagram of system throughput. The nominal figure in brackets indicates the amount of usable optical energy (as a fraction of Nd: YAG third harmonic energy) remaining after each component. The symbol r|P indicates the probe efficiency (collected Raman scattered energy/excitation energy, vide infra, section 4.3). Figure 3.1 96 Low intensity basic attenuation of fused silica in the deep ultraviolet spectral region, adapted from data in Fabian et al. [1993]. Figure 3.2 102 Schematic representations of UV-induced and UV-absorbing colour centers in fused silica. Figure 3.3 108 Total throughput versus time for 56.5 cm segments of 400 um diameter SUV fiber (dotted), 400 u,m diameter IUV fiber (solid), and 600 pm diameter RJV fiber (dashed). Excitation is ca. 50 uJ, 20 Hz, ca. 3 ns, 225 nm pulses. Figure 3.4 110 Micrograph of damaged 400 pm (diam.) optical fiber proximal end showing linear catastrophic damage failure mode (near left side). Figure 3.5 113 Initial (filled symbols) and steady state (open symbols) Eout versus E„, for 400 (squares) and 600 urn (triangles) diameter 56.5 IUV fiber segments. Excitation is 20 Hz, ca. 3 ns, 225 nm pulses. Figure 3.6 114 Initial (squares) and steady state (circles) throughput for a 56.5 cm length of IUV fiber as a function of wavelength for 20 Hz, ca. 3 ns, 50 +/- 4 pj pulses. Figure 3.7 115 Fractional initial throughput vs. wavelength for experimental conditions identical to those of Figure 3.6. xiii Figure 3.8 116 Total induced attenuation vs. wavelength for experimental conditions identical to those of Figure 3.6. Two fiber segments were run at each wavelength and induced attenuation was calculated on the basis of both initial and recovered throughput, resulting in a total of four data points for each wavelength. Figure 3.9 117 Initial energy out of 400 um (diam.) fiber vs. fiber segment length using 225 nm, 20 Hz., ca. 50 uJ into fiber, ca. 3 ns pulses. Figure 3.10 118 Steady-state energy out of 400 um (diam.) IUV optical fiber segment as a function of segment length. Energy in was 225 nm, ~50 uJ/pulse, ca. 3 ns, 20 Hz. pulses. Error bars represent one standard deviation of the mean for 700 pulses. Figure 3.11 119 Total induced attenuation vs. fiber segment length with the experimental conditions the same as Figures 3.9 and 3.10. Inset shows the linear fit to these data. Figure 3.12 120 Induced attenuation per unit length from experiments shown in Figures 3.9 and 3.10. Inset shows results of linear fit to these data. Figure 4.1 130 A diagrammatic representation of the excitation and collection processes in fiber-optic resonance RS. Figure 4.2. 134 Demonstration of different path lengths and angles possible when scattering from a particular volume element to the collection surface and the need for integration over all area elements on the collection surface. See text for explanation of labeled light rays. Figure 4.3 137 EX(0) for a 200 um (diam.) fiber as a function of angle from fiber axis. Figure 4.4 137 COL(a) vs. a for a 600 um (diam.) fiber. Squares are experimental points and dotted line is a Gaussian fit. Figure 4.5 A typical spectrum of a mixture of MO and NO3" showing the peak positions and subtracted backgrounds. 141 xiv Figure 4.6 144 Experimental and modeled working curves for methyl orange. • - probe E, A - probe C, o - probe A. Solid symbols are experimental data, open symbols are modeled data. Figure 4.7 146 Plot of NO3" signal strength vs. MO concentration showing reduction in probe sensitivity with increasing medium absorbance. • - probe E, A - probe C, o- probe A. Solid symbols are experimental data, open symbols are modeled data. Figure 4.8 151 266 nm (10 Hz, 10 ns pulses) UV resonance Raman spectrum of 50 pg/ml Salmon DNA obtained using probe C. The exposure time was 8.8 minutes, the average power at the sample was approximately 1 mW, and the slit width was 500 pm. The letter over each peak denotes the base which gives rise to the peaks (C - cytosine, A - adenine, G -guanine, T - thymine; see ref. 21). The peak near 1050 cm - 1 is due to nitrate, which was added as an internal standard and calibrant. Figure 4.9 153 Pulsed (10 Hz, 10 ns) UV ( 225 nm ) resonance Raman spectrum of the amino acid tryptophan using probe H. The exposure time was 10 minutes, the average power at the sample was approx. 200 pW and the slit width was 300 pm. Figure 4.10 153 A partial working curve for DNA using 266 nm excitation and probe C. The 1484 cm'l peak was used as the signal. Figure 4.11 155 • - Working curve for methyl-orange ( 1400 cm"l peak height vs. [MO]) and a 200 pm diameter, 45° angled excitation - 200 pm diameter collection, fiber probe. • - height of 0.153 M N0 3 " peak vs [MO]. Inset: probe tip diagram. Figure 4.12 159 TOP: A schematic diagram showing the sequential steps in the fabrication of a novel FO probe for UVRRS. A - IUV fiber, B - SUV fiber, C - photoresist, D - aluminum film, E -silicone rubber collar. 1 - apply photoresist to polished fibers, 2 - selectively polish away photoresist and fiber, 3 - apply aluminum film, 4 - remove photoresist, 5 - attach fibers. BOTTOM: micrograph of aluminum mirrored, angled collection fiber (Left), micrograph of completed probe tip (Right). Figure 4.13 161 Four diagrammatic views of the probe tip geometry. Legend is uo follows: (1): excitation fiber, (2): collection fiber, (3) and (4): silicone rubber collars, (5): distal end of excitation fiber, (6): point of maxium distal extent of collection fiber, (7): polished lens, (8): aluminum mirror, (9): effective 'window' for 'viewing', (10) and (11): centre lines of collection and excitation fibers, respectively, (12), (13) and (14): general optical path, (15): field of illumination, (16): approximate field of view, (17), (18), (19), (20) and (21): optical rays representing the normal operation of the probe. Figure 4.14 165 (a) DUV RS data of neat ethanol using (A) flush probe (400 urn diameter excitation fiber/600 urn diameter collection fiber), (B) angled and mirrored probe (400 um diameter excitation fiber/600 urn diameter collection fiber), and (C) angled, mirrored, lensed and faceted probe (600 urn diameter excitation fiber, 600 um diameter collection fiber). Average power into excitation fibers was 1 mW; using 20 Hz, ca. 3 ns, 225 nm pulses; integration time = 30 s. (b) UVRRS data of 200 ug/ml hen egg-white lysozyme. Probe designations are the same as in Fig. 4.14 (a). Average power into excitation fibers was 1.2 mW; using 20 Hz, ca. 3 ns, 230 nm pulses; integration time = 90 s. Sloping backgrounds have been subtracted and spectra have been vertically translated for clarity. Figure 4.15 166 FO-UVRRS data of aqueous cellulose binding domain protein (CBD, 200 ug/ml, ^=227 nm, 18 minutes integration) , aqueous tryptophan (TRP, 100 uM, "K^Hl nm, 5 minutes integration), aqueous poly-l-lysine (PLL, 150 ug/ml, XexC=208 nm, 10 minutes integration) and testosterone dissolved in ethanol (TSTN, 0.5 mM, X«xc=245 nm, 4.5 minutes integration). Asterixes indicate solvent peaks. All data was obtained using 20 Hz, ca. 3 ns pulses. Sloping backgrounds and ca. 1640 cm"1 water peak have been subtracted, and spectra have been vertically translated for clarity. Spectra are not to scale. Figure 4.16 167 Circles: working curve for tryptophan FO-UVRRS using 1010 cm"1 peak height. Squares: ca. 1640 cm"1 water peak height as a function of tryptophan concentration. Excitation was at 227 nm and the probe was of the type shown in Fig. 4.12 with excitation and collection fiber diameters 400 urn and 600 um, respectively. Figure 4.17 174 Sc (normalized to 1) and component terms (normalized to 0.75 for clarity) for the ca. 1010 cm"1 tryptophan residue signal in 50 ug/ml hen egg-white lysozyme using a 400 um (excitation fiber diameter), 600 urn (collection fiber diameter) A/M probe. X indicates experimental point. See text for detail. Figure 4.18 186 Calculated fraction of total Raman scattering that occurs in the probe volume for lysozyme excited at 230 nm (dotted and dotted-dashed lines: 600 urn and 100 urn probe volume depths, respectively) and 206 nm (solid and dashed lines: 600 um and 100 um probe volume depths, respectively) xvi Figure 4.19. 188 Optimal (signal shot-noise limited) signal-to-noise ratio (SNRopt) as a function of concentration assuming 100 s of 1 mW excitation for (A) tryptophan 762 cm'1 W18 line excited at 218 nm, (B) tryptophan 1342 cm"1 W7 line excited at 218 nm, (C) phenylalanine 1207 cm"1 v 7 a line excited at 192 nm, (D) phenylalanine 1606 cm"1 v 8 a line excited at 209 nm, (E) tyrosine 1263 cm"1 v7, line excited at 192 nm, (F) tyrosine 1617 cm"1 vg a line excited at 223 nm, and (G) lysozyme amide III mode excited at 204 nm. Figure 4.20 191 Theoretical optimal detection limits versus medium extinction coefficient (a™) for TRP W18 signal at 218 nm (square symbols) and TYR v g a at 223 nm (circles). All points using 100 seconds of 1 mW excitation over a 600 pm probe depth. Figure 4.21 193 Schematic side view of 90° collection angled/mirrored probe geometry. O indicates the origin of the right handed coordinate system. Vectors labeled as k denote scattered, transmitted, or reflected light rays; while vectors labeled as u denote surface normals or fiber axis vectors. R« and R« indicate the radii of the excitation and collection fibers, respectively; while dV and dA denote differential volume (scattering) and area (collection) elements, respectively Figure 4.22 194 Top view of angled/mirrored improved probe geometry. Symbols as in Figure 4.21. Dotted line represents line of constant x in the plane of the mirror that passes through the point of intersection of k< and the mirror (x=z plane). Figure 4.23. 200 Simulated working curves for lysozyme excited at 227 nm using an ATM probe with a 600 pm (diam.) collection fiber and a 600 pm (circles) or 100 pm (squares) diameter excitation fibers. The values were calculated on a per uJ and per (mbarn/sr) basis. Cumulative (top) and differential (bottom) collected intensity as a function of z for 10 pg/ml (dotted line)and 918 pg/ml (solid line) lysozyme. All intensities are on a per pJ and per (mbarn/sr) basis. Top: Raw (uncalibrated) data from one of the de04h spectra showing DNA signal peaks and the unintensified region. Bottom: Spectrum calibrated for wavenumber. Figure 4.24. 202 Figure 5.1. 206 Figure 5.2 206 Top: Raw (uncalibrated) data from one of the fe08h spectra showing DNA signal peaks and the unintensified region. Bottom: Spectrum calibrated for wavenumber. XVII Figure 5.3 211 PSD estimates from fe08h using different FFT lengths. Figure 5.4 211 95% confidence intervals for one fe08h spectrum using different FFT lengths. Figure 5.5 213 Standard deviation for fe08h spectra using different FFT lengths. Figure 5.6 213 PSD estimates for fe08a using different FFT lengths. Figure 5.7 214 PSD estimates for MO-KN RRS data. Figure 5.8 215 PSD estimates for 266 nm DNA UVRRS data. Figure 5.9 217 SNR vs. transition band centre for spectra in data set fe08h with a filter order of 140. Figure 5.10 218 SNR vs. transition band centre for spectra in data set fe08h with a filter order of 60. Figure 5.11 219 Top: Frequency response of filter with a transition band center and width of 0.1 and 0.02, respectively, and an order of 70. Bottom: magnitude of FFT of raw and filtered data. Figure 5.12 220 Raw and Filtered DNA UVRRS data de04f. The pass band was centred at a normalized frequency of 0.1 and width of 0.02. The filter order was 70. Figure 5.13 229 A flow chart representation of the numerical algorithm TPMEM, implementing a two-point entropy regularization method. Figure 5.14 235 The results of a typical SNR enhancement trial on simulated data. (A) The underlying, simulated Raman spectrum (SNR= °°). (B) The noise corrupted spectrum (SNR=1.22). The spectra recovered using (C) TPMEM (SNR=6.73), (D) zerorth order SG (SNR=3.81), and (E) fourth order SG (SNR=3.42). Figure 5.15 236 An example of the use of TPMEM for SNR enhancement on experimental Raman data. (A) The resonance Raman spectrum of 20 uM aqueous methyl orange excited at 472.7 nm xviii with an integration time of 1 second. (B) The spectrum of A processed using TPMEM. (C) The resonance Raman spectrum of 200 uM aqueous chromate ion excited at 363.8 nm with an integration time of 4 seconds. (D) The spectrum of C processed using TPMEM. Figure 5.16 237 The results of a typical deconvolution trial on simulated data. (A) The underlying, simulated Raman spectrum. (B) The convolved and noise corrupted spectrum (noise standard deviation = 3% of maximum signal height). The spectra recovered using (C) TPMEM, and (D) MEM. Figure 5.17 238 The averaged values of (A) RMSERR, and (B) 1-r (where r is Pearson's correlation coefficient), for the deconvolution trials at noise levels of 1%, 2%, 3%, 4%, and 5%. The error bars represent one standard error of the mean (N=10) in both cases. Figure 5.18 239 An example of the use of TPMEM for deconvolution on experimental Raman data. (A) The Raman spectrum of saturated KNO3 (aq.) excited with 50 mW of 514.5 nm power and integrated for 1 second. (B) The spectrum of A deconvolved using TPMEM. Figure 6.1 246 Expanded view of fiber-optic UVRRS probe tip. A - photosensitization-resistant excitation fiber, B - standard UV-grade collection fiber, C - reflective surface, D - sample volume. Arrows indicate the general optical path. Figure 6.2 248 Fiber-optic UVRRS data of native (top trace, 24° C) and denatured (bottom trace, 87° C) RNase TI (200 ug/mL) excited at 230 nm. The peak near 933 cm"1 is the perchlorate internal standard. Figure 6.3 249 Fiber-optic UVRRS data of RNase TI (50 u.g/mL, 20 minutes integration) excited at 208 nm. Average power at the sample was approximately 50 u.W. The ca. 1640 cm"1 water bend peak and sloping background have been subtracted. Figure 6.4 252 Fiber-optic UVRRS data of L-tyrosine (top trace) and O-phospho-L-tyrosine (bottom trace) excited at 227.5 nm. Smoothing has been applied for clarity. The peak near 933 cm" 1 is the perchlorate internal standard. Figure 6.5 253 (A) Absorption spectra of L-tyrosine (solid line) and O-phospho-L-tyrosine (dotted line). RREPs showing cross sections of the (B) ca. 1176 cm'1, (C) ca. 1210 cm"1, and (D) overlapped 1600/1610 cm'1 bands of tyrosine (solid line, square symbols) and O-phospho-L-tyrosine (dotted line, circular symbols). xix Figure 6.6 256 Fiber-optic UVRRS data obtained using 230.5 nm excitation of free aqueous CBDc«c (top trace) and CBDcex bound to avicel. Water and silica backgrounds have been subtracted, and the spectra have been scaled and vertically translated for clarity. Figure 6.7 257 Fiber-optic UVRRS data obtained using 230.5 nm excitation of free aqueous lysozyme and lysozyme adsorbed to silica. Water and silica backgrounds have been subtracted, and the spectra have been scaled and vertically translated for clarity. Figure 6.8 258 266 nm Fiber-optic UVRRS data of 50 pg/mL salmon-sperm DNA obtained using flush-geometry probe incorporating a 300 pm diameter excitation fiber and a 600 pm diameter collection fiber. Integration time was 9 minutes. Figure 6.9 262 Space fill solution structure of CBDcex showing the relevant tryptophan and asparagine residues as wireframes [structure from Xu et al., 1995]. Figure 6.10 264 Ribbon diagram solution structure of CBDNI showing the relevant tyrosine and tryptophan residues as black wireframes [structure from Johnson et al, 1996A]. Figure 6.11 261 Fiber-optic UVRRS data at 227 nm of CBDccxG- (dotted) and CBDcexG+ (solid). A five point low pass FFT filter was applied to remove high frequency noise. Data was scaled to equal sulfate peak intensities. Figure 6.12 269 Fiber-optic 227 nm UVRRS data of CBDcexG- (dotted) and CBDcexG+ (solid) from Figure 6.11 expanded to show detail between 830 and 930 cm"1. Figure 6.13 270 Fiber-optic 227 nm UVRRS data of CBDcexG- (dotted) and CBDceXG+ (solid) from Figure 6.11 expanded to show detail between 1300 and 1400 cm"1. Figure 6.14 271 Fiber-optic 227 nm UVRRS data of CBDc«G- (dotted) and CBDceXG+ (solid) from Figure 6.11 expanded to show detail between 1500 and 1650 cm"1. Figure 6.15 273 Fiber-optic UVRRS data of free and cellopentaose-bound CBDNI and mutants: (A) free CBDNJWT, (B) free CBD N iY19A, (C) free CBD N iY85A, (D) CBDNIWT and cellopentaose, (E) CBDNIY19A and cellopentaose, and (F) CBDNIY85A and XX cellopentaose. Some Raman peaks have been labeled. The sloping background and water signal have been subtracted and a five point low-pass FFT filter has been applied. Figure 6.16 274 Fiber-optic UVRRS data of distilled water (bottom), 40 mM aqueous MgS0 4 (top) and 2.5 mM cellopentaose and 40 mM MgS04 (middle). The water spectrum has been smoothed with a five point low-pass FFT filter and the spectra have been offset for clarity. All spectra taken at 230 nm with 90 second integration. Figure 6.17 276 Fiber-optic UVRRS C B D N i spectra from Figure 6.15 in expanded detail showing vga/vgb region. C P indicates the presence of 2.5 mM cellopentaose. Figure 6.18 278 Fiber-optic UVRRS CBDNI spectra from Figure 6.15 in expanded detail showing v7, and v 9 i region. C P indicates the presence of 2.5 mM cellopentaose. Figure 6.19 279 Fiber-optic UVRRS CBDNI spectra from Figure 6.15 in expanded detail showing W7 Fermi doublet region. C P indicates the presence of 2.5 mM cellopentaose. Figure 6.20 281 Fiber-optic UVRRS CBDNI spectra from Figure 6.15 in expanded detail showing W17 region. C P indicates the presence of 2.5 mM cellopentaose Figure 6.21 282 Fiber-optic UVRRS CBDNI spectra from Figure 6.15 in expanded detail showing W3 region. C P indicates the presence of 2.5 mM cellopentaose. xxi ACKNOWLEDGEMENTS First and foremost, I am indebted to my Ph.D. supervisor Dr. Robin F. B. Turner and my long-time colleague Dr. Georg Schulze, without either of whom this work could not have been realized. Further, Dr. Charles A. Haynes (Chemical Engieering), Dr. Michael W. Blades (Chemistry) and Dr. Alan Bree (Chemistry), all of The University of British Columbia (UBC), provided invaluable advice, equipment, and knowledge enabling the completion of this project. I am grateful to Chris Barbosa for his assistance in the lab, especially in the Spring of 1997.1 wish to thank Gary Nelson of Polymicro Technologies Inc. (Phoenix, AZ) and Dr. Karl-F. Klein of Fachhochschule Geissen Friedberg (Friedberg, FRG) for their assistance and support regarding the supply and characterization of UV-improved optical fibers. I thank Al Boraston and Jeff Kormos of the Department of Microbiology at UBC for advice, assistance and protein used in the CBD studies of Chapter 6. Dr. C. Nick Pace of Texas A&M University generously provided the RNase TI used in Chapter 6.1 also wish to thank Dr. Alina Kulpa in the Department of Electrical Engineering at UBC for her assistance with aluminum deposition. This research was supported by grants from the Natural Sciences and Engineering Research Council of Canada (NSERC) and the British Columbia Health Research Foundation. I have personally been supported by scholarships and grants from NSERC, UBC, The R. Howard Webster Foundation, The Dorothy Gilbert Foundation, and The Advanced Systems Institute. I would also like to thank my family in Nova Scotia for their never-ending support and my friends here in Vancouver (especially at UBC and in The Biotechnology Laboratory) who have made the past five years so enjoyable. I acknowledge the temporal and spiritual contributions of the schooners Nancy and Bluenose to this work, and my well-being in general. xxii FO REWARD Much of the work in this thesis has already appeared in published, refereed journal articles and conference proceedings. A small amount of Chapter 2 appeared in "materials and methods" or "experimental" sections of several papers [Greek et al, 1995; Greek et al., 1996B; Greek et al., 1997A; Greek et al, 1997]. Some of Chapter 3 is derived from a recent Applied Optics publication [Greek et al, 1997B]. Much of section 4.2 was presented in an earlier Applied Optics publication [Greek et al, 1996B] and the Proceedings of the 22nd Canadian Medical and Biological Engineering Society Conference [Greek et al, 1996A]. Some of the work in section 4.3 also came from the second Applied Optics publication [Greek et al, 1997B] and section 4.4 is currently being reworked for submission to Applied Spectroscopy. The angled/mirrored probe geometry is also the subject of a recent patent application. The second half of Chapter 5 reports work that was also presented in an Applied Spectroscopy publication [Greek et al, 1995]. Substantial portions of section 6.2 were presented at the BiOS '97 section of Photonics West '97 and reported in SPIEProceedings [Greek et al, 1997A]. The author was the principle author on all of the aforementioned publications and had primary responsibility for both their conceptual and practical aspects. The author's Ph.D. supervisor (R. F. B. Turner) was the corresponding author on each of these. Little, if any work from co-authored publications in which the present author was not the first author [Schulze et al, 1993; Schulze et al, 1994; Schulze etal, 1997A; Schulze et al, 1997B; Schulze et al, 1997C] appears directly in this document. An exception to this is a paper co-authored with Klein et al. [Klein et al, 1997], some of the work from which I was directly responsible for and appears in Chapter 3. 1 Chapter 1. Introduction and Motivation 1.1 Overview of Motivation and Approach This thesis describes the development of the first fiber-optic linked ultraviolet resonance Raman spectroscopy system and its use for biomolecular structure studies. This chapter establishes the motivation to accomplish this goal by discussing various aspects of protein structure and their importance to protein function and behaviour (section 1.2). In this regard, the phenomenon of protein binding in general and protein adsorption specifically are addressed as potential end user applications for this technology. The various existing spectroscopic methods currently used to investigate proteins are briefly presented (section 1.3) in order to establish the motivation to develop ultraviolet resonance Raman spectroscopy (UVRRS) for this application (and specifically to develop fiber-optic UVRRS (FO-UVRRS)). Finally, the theory of UVRRS in general and protein UVRRS signals in particular are presented (section 1.4). This chapter is by no means a comprehensive introduction, and the reader is referred to the references cited herein for more detail. Due to the interdisciplinary nature of the research, it is recognized that readers may come from a diverse range of disciplines, including electrical engineering, biochemistry, chemistry, chemical engineering and biotechnology. This chapter attempts to provide background necessary for understanding generally why and how the research presented in this thesis was performed. Basic instrumental and hardware aspects of the research are left for Chapter 2, which deals more with materials and methods. 2 1.2.1 Proteins - A Brief Introduction This section introduces the basic concepts and terminology of protein chemistry for the reader who is not versed in biochemistry. Proteins and polypeptides are biological hetero-polymers consisting essentially of chains of amino acid monomer units linked together by peptide bonds. It is difficult to overstate the importance of proteins as components of biological systems. One of their most important functions is as enzymes, which are biological catalysts that mediate and/or regulate virtually all of the biochemical processes necessary to sustain life. Recently, enzymes have been harnessed to serve as catalysts in a number of new and important industrial processes [Bailey and Olis, 1986]. Some proteins, such as collagen and elastin, have a macro-structural function; others, such as hormones, receptors and antibodies, enable cells to interact with each other and their environment. Such interactions are essential, for example, in regulating growth, metabolism, and immune system function. Blood clotting proteins are responsible for initiating and controlling haemostasis. Proteins are involved in virtually every single one of our life processes. Proteins carry oxygen from our lungs to our muscles (e.g. hemoglobin) and store it there (e.g. myoglobin); they give us motility through the contraction of our muscles (e.g. actin and myosin); they convert food into stored energy (e.g. cytochromes and the enzymes involved in glycolysis); they support us (e.g. collagen and elastin) and keep us warm (e.g. keratin in hair); they stop us from bleeding (e.g. fibrinogen and the other blood-clotting proteins) and protect us from invaders (e.g. the antibodies); and the list could (and does) continue into innumerable volumes. Amino acid monomers have the general formula RCH(NH3+)CC>2", where R represents one of twenty common amino acid side chains (Figure 1.1). When incorporated via a condensation reaction into a protein molecule, the individual monomers are referred to as amino acid residues. For the most part, protein molecular weights lie between one thousand and one million Daltons (1 Dalton = 1 g/mol). Physico-chemically significant properties of the amino acid residues include proton donor/acceptor characteristics Threonyl [Thr] Tryptophanyl LlEP] Tyrosyl Valyl JVaJL Figure 1.1. The twenty common amino acids and their three letter abbreviations. Hydrogen atoms and charges are omitted for clarity. 4 (acidity and alkalinity); relative affinities for water molecules (hydrophobicity or hydrophilicity); ability to covalently bond to side chains of other, non adjacent, amino acid residues via a disulfide linkage (in the case of cysteine residues); ability to interact electrostatically with adjacent or non-adjacent residues; and morphology. The monomeric units link together through a condensation process (Figure 1.2) whereby the carbon of the CO2" (carboxyl) end of one amino acid bonds with the nitrogen of the NH3 + (amino) end of another amino acid, resulting in a C-N (peptide) bond and the release of a molecule of water. 1.2.2 Levels of Structure Figures 1.3 and 1.4 (as well as figures 6.9 and 6.10) show different representations of protein structure. The structural properties of a protein may be broadly divided into the following categories, from most local in scope to most global in scope: primary structure, secondary structure and tertiary structure [Creighton, 1984, pp. 159 - 198]a. a An additional level of structural organization that should be mentioned for completeness is that of quaternary structure. This level will not be discussed here since the role of quaternary structure in protein adsorption processes is incidental and the analytical methodology proposed here is not suitable to analyzing quaternary structural changes. 5 (A) NH > o=c NH > o - c NH < c - o NH > N H < C - O NH o - c > NH < < NH C - O C - O i i (B) H O H O CH, I *H3N - C - CO, + *H3N I H H O CH, I c - CO, I H H O . CH, O I I *H,N - C - C -I H N-I H CH 2 C - CO," + H,0 I H Figure 1.2. (A) The geometry of the peptide backbone showing a beta sheet structure with alpha carbons as bullets and hydrogen bonds as dotted lines. (B) The amino acids serine and tyrosine joining in a condensation reaction to form a di-peptide. Figure 1.3. "Ribbon" diagram of Horse Heart Myoglobin, a 153 amino acid residue oxygen-storage protein composed almost entirely of a-helices, as denoted by the helical ribbons shown here [structure from Evans and Brayer, 1988], 6 The primary structure of a protein refers to the specific sequence of amino acid residues in the polypeptide chain which is specified by the DNA segment (gene) that encodes the information for assembling the protein. The primary sequence, in turn, contains all of the information necessary to describe the subsequent folding pattern of the native protein which dictates its secondary and tertiary structure. Knowledge of the primary structure confers the following information: (a) the total hydrophobicity, which affects adsorption, solubility and interaction with other proteins; (b) the number of titratable (charged) groups, which dictates the electrostatic characteristics of the molecule; and (c) the number of cysteine amino acid residues, which in turn specifies the number of intra-molecular disulfide bonds that may be formed by the protein. Figure 1.4. Ribbon diagram of a-chymotrypson, a 247 amino acid residue protein composed of 45% P-sheet (mostly on the left in this figure), 40% random coii (mostly on the right), and 15% a-helix (difficult to see in this figure) [structure from Tsukada and Blow, 1985]. 7 Secondary structure may be broadly subdivided into two types: ordered and disordered. Ordered secondary structure refers to the presence of locally periodic structure such as a-helices and p-sheets. a-Helices arise due to hydrogen bonding between local residues. Hydrogen bonding between the N-H group of the i t n residue and the C=0 group of the i+4tn results in a helical structure with 3.6 amino acids per turn. P-pleated sheets (see also Figure 1.2) are formed through hydrogen bonding between non-local residues. If the protein chain folds back on itself, then oppositely oriented amino acid residues may hydrogen bond to each other to form the anti-parallel P-pleated sheet structure. The residues that bond to each other in an anti-parallel P-pleated sheet are typically 10 to 20 monomeric units apart on the peptide chain, and oppositely oriented. When two distinct areas of the peptide chain are adjacent, then like-oriented residues may hydrogen bond to form the parallel P-pleated sheet structure. The residues that bond to each other in this structure are typically much further apart than in the anti-parallel structure. Figures 1.3 and 1.4 show the a-helix, P-pleated sheet, and random coil structures in the proteins horse heart myoglobin and a-chymotrypsin. Disordered (random coil) secondary structure implies loops and turns in the main chain, as well as random hydrogen bonding between residues. The primary driving force for secondary structure formation is thought to be the expulsion of water away from hydrophobic amino acid residues [Tanford, 1980, chapter 13]. Expulsion of water from a protein core requires protein atom packing densities to be large enough that no space is available for water molecules; in general, intra-atomic packing densities in folded proteins are similar to polymer glasses and crystals. The extremely compact nature of these ordered secondary 8 structures results in very few rotational degrees of freedom, and hence there is a large entropic penalty associated with their formation. The structures will still form spontaneously, however, provided the increased entropy of liberated solvent molecules and the net negative enthalpy change due to favourable hydrogen bonding outweigh the loss in entropy of the polypeptide chain. Secondary structure is relatively local, even though it may be formed by non-local forces. In general, regions of ordered secondary structure, are flanked by random structures which allow a-helices and P-sheets to repeatedly fold back upon themselves to form a globular conformation, which may include multiple polypeptide chains linked through disulfide bonds. The manner with which this folding occurs is denoted as the tertiary structure of the protein. The tertiary structure defines the solvent accessible surface area of a globular protein. The tertiary structure also defines the number and location of surface titratable groups, which affect the acidity and alkalinity of the resulting protein, as well as the solubility and surface potential. Finally, the tertiary structure determines the surface hydrophobicity:hydrophilicity ratio, which has a large impact on solution and adsorption behaviour. Tertiary structure and changes to it are more difficult to quantify than primary and secondary structure. Changes to tertiary structure are usually examined by considering changes in the environments of specific residues or changes in the relative positions of amino acid residues that are separated along the peptide chain, but may be in close proximity in terms of three-dimensional structure. The combined contributions of secondary and tertiary structure of a protein yield what is referred to as its conformation. 9 Folded proteins, also known as native state proteins, are usually marginally stable at physiological conditions. Through evolution, Nature has apparently optimized proteins for function and biological activity in their native environments, not for stability with respect to changes in those environments. Hence, they may be denatured (unfolded) by relatively modest changes in environmental conditions such as temperature, pH, and ionic strength, as well as by chemical denaturants such as guanidinium hydrochloride. Interaction with surfaces (e.g. adsorption) can also induce denaturation. The biological activity of a protein is intimately connected to its conformation. Changes in the hydrogen bonding of even a single residue can degrade protein function (activity, binding affinity, etc.), and complete unfolding usually reduces protein function to zero. The stability of a protein is defined by thermodynamics as the negative Gibbs energy change (relative to the denatured state) upon folding which for a typical globular protein, is usually only on the order of tens of kilojoules per mole. This is equivalent to the enthalpy of formation of from one to five hydrogen bonds. Considering that there may be hundreds or thousands of hydrogen bonds in a typical protein, this is an extremely small stabilizing energy. It has also been shown that protein unfolding is a co-operative process, which is to say that distinct regions of a large globular protein domain do not unfold sequentially, but rather, after unfolding begins in one region, it progresses rapidly to complete unfolding of the entire structure. This "all or none" active vs. denatured property is characteristic of most globular proteins [Creighton, 1984 p 321; Privalov, 1979]. 10 1.2.3 Protein Adsorption 1.2.3.1 Introduction to Protein Adsorption The initial motivation behind embarking upon the research described in this thesis was the absence of an effective instrumental technique for investigating changes in protein structure upon adsorption at the solid/liquid interface. Such changes had been implicated through direct and indirect means as significant contributors to, and determinants of, the protein adsorption phenomenon [e.g. Norde and Favier, 1992; Jakobsen and Wasacz, 1987; Haynes and Norde, 1994]. It was reasoned that fiber-optic ultraviolet resonance Raman spectroscopy could provide the requisite experimental versatility, robustness, and sensitivity to protein structure to make significant contributions to the field in terms of quantifying adsorption induced structural changes and thereby elucidating adsorption mechanisms. During the development of the fiber-optic linked instrument, the range of applications under consideration has been broadened both through curiosity and necessity, as will be seen in Chapter 6, making the general impact and utility of this instrumentation more far reaching. However, in terms of the author's long-term research interests, the use of the instrument to investigate specific and non-specific protein binding to insoluble substrates remains the primary goal of its development. This section (section 1.2.3) introduces, describes, and discusses this most important aspect of the motivation. Protein adsorption refers to the binding of protein molecules to a surface, with concomitant changes in structure and activity. The phenomenon is extremely important in many natural, clinical and industrial applications. At present, however, protein adsorption is poorly understood, and there are no reliable methods for probing the critical changes in 11 secondary and tertiary structure in proteins as they adsorb to solid substrates. In natural biological processes, adsorbed interfacially active enzymes, such as pancreatic lipase, for example, which controls fat digestion, are often found to act at interfaces. Membrane bound proteins which are intercalated into the lipid bi-layer of a cell are involved in such important processes as photosynthesis, active transport across cell membranes and catalysis. Proteins adsorbed to cell membranes also act as receptors for molecules required for cellular metabolism. Other physiological processes involving protein adsorption include the formation of dental plaque, where an initial layer of proteins is adsorbed and denatures on the surface; this provides an anchor for subsequent cellular adhesion, leading to plaque formation and eventually to such conditions as gingivitis. A similar, but more complex mechanism is responsible for blood coagulation. An improved understanding of protein adsorption would aid in the understanding of many of these biological and biochemical processes. Clinically, an understanding of protein adsorption is driven by the need for improved blood-contacting polymers. Presently, there is no truly biocompatible synthetic polymer and, as a result, all current artificial implant materials have a finite lifetime before protein fouling compromises their performance. The deleterious effects of the adsorption of blood proteins such as serum albumin, thrombin and fibrinogen on artificial implants has received much attention [Morrissey and Stomberg, 1974; Vroman and Leonard, 1991; Vroman, 1987]. The situation is particularly bleak in the area of vascular prostheses, where no synthetic materials have shown promise. Vascular grafts and catheters often fail prematurely due to thrombotic occlusion. If the thrombus detaches from the implant, 12 tissue damage, aneurysm, pulmonary edema, congestive heart failure or stroke may result. With most synthetic heart valves and whole artificial hearts, this process is delayed or avoided through the regular and costly use of natural and synthetic anticoagulants. Similarly, the maximum operational lifetime of implantable biosensors is dictated by the fouling of the artificial membrane which, if functioning properly, allows for selective mass transport/exchange between the sensor and the physiological environment. Several drug delivery systems rely on desorption from polymeric micro-carriers. In summary, with a tool to provide better understanding of the adsorption phenomenon, it may eventually be possible to design truly biocompatible polymers, long lasting implantable biosensors, and improved drug delivery systems. This would result in a significant economic benefit, an increase in life span, and an improved quality of life for patients depending on such implantable devices. Protein adsorption is also responsible for many deleterious and beneficial effects encountered in industrial processes. The phenomenon is used advantageously in the bioprocess industry for immobilized enzyme bioreactors where proteins are adsorbed to surfaces within a bioreactor to catalyze desirable reactions. Some protein purification schemes rely on the adsorption of the desired protein from a dilute solution to a surface, after which it is recovered. However, protein function is strongly dependent on its structure, and often the changes that occur due to adsorption can drastically affect activity. Undesirable industrial effects of protein adsorption include fouling of heat exchangers, ultrafiltration membranes, ship hulls and food processing units. A better 13 understanding of the mechanisms behind adsorption would clearly benefit future design of these industrial processes and result in economic benefits. 1.2.3.2 Thermodynamics of Protein Adsorption As with any chemical reaction, a protein will adopt a new structure, dissolve, adsorb, desorb, etc., if the change in Gibbs free energy in going to the new state is negative; that is, if ^Gold_>new = AHM_>new - TASold_>new < 0 (1.1) then the change will occur spontaneously, provided kinetic barriers are not prohibitively high. The magnitude of A G o i j ^ n ^ is indicative of how strongly the new state is favoured over the old state. It should be noted that this expression applies to the entire system, including protein, solvent, and surface (if any). The Gibbs energy change is made up of contributions from changes in enthalpy and entropy, and depending on the process and environment, one or the other may dominate. Enthalpy (H) changes arise from changes in the internal energy (U), pressure, and/or volume of the system (H=U+PV), however for the systems in which we are interested, the PV changes are negligible. Negative contributions to enthalpy usually result from physical, or chemical bond formation. Entropy (S) is a measure of the number of degrees of freedom of the system. Bonds restricting the rotational or translational freedom of the protein or solvent will result in negative contributions to entropy. It is generally accepted that there are four dominant forces driving protein adsorption [Haynes and Norde, 1993]: (a) the hydrophobic effect, (b) electrostatic 14 attraction, (c) Lewis acid-base interactions (e.g. hydrogen bonding), and (d) entropic effects due to conformational changes upon adsorption. In protein adsorption, the hydrophobic effect refers to the entropy increase associated with dehydration of solvent-exposed hydrophobic protein and sorbent surfaces which come in intimate contact during binding. Electrostatic attraction between groups of unlike charge on a protein and surface may drive adsorption through a resulting energy decrease. However this interaction is not dominant, as demonstrated by the fact that proteins will adsorb (often to a significant extent) on surfaces of like charge. Lewis interactions are clearly important when polar surface groups are present. As yet, it is unclear to what extent entropic forces drive the adsorption process. It is reasonable to expect that these effects contribute significantly to overall adsorption behaviour, since an enormous amount of entropy can be recovered through an increase in peptide and side-chain rotational degrees of freedom if the protein unfolds on the surface. Direct evidence of such surface induced unfolding, although limited due to the inadequacy of presently used instruments, has been reported in the literature [MacRitchie,1986; Vroman, 1967; Norde, 1976]. Circular dichroism transmission spectroscopy has provided evidence of protein unfolding at surfaces [Norde and Favier, 1992; Kondo et al, 1991; McMillan and Walton, 1974]. Several groups have used fluorescence to show that significant structural changes occur upon adsorption [Burghardt and Axelrod, 1983; Beissinger and Leonard, 1980]. Calorimetric studies have been used to quantify these adsorption induced conformational changes [Haynes et al, 1993;], and have shown conclusively the cooperative nature of protein unfolding in solution [Privalov, 1973; Privalov, 1979]. However, as discussed below, each of these 15 techniques is hindered (or rendered useless) by the chemical natures of most sorbent surfaces. Thus, their application to the general study of the mechanism of protein adsorption is severely limited. There are several more subtle considerations affecting overall adsorption behaviour. For example, the size and shape of the protein in question, as determined by its tertiary structure, dictate the diffusional properties of the protein in solution. Smaller, more compact proteins diffuse to the surface faster, and hence are the first to adsorb in a multi-component mixture. However, larger proteins are able to make more contacts with the surface, and hence it is energetically more favourable for large proteins to displace adsorbed small proteins. This leads to the Vroman effect, which describes the process by which a layer of low molecular weight proteins is initially adsorbed to a surface, followed by their displacement by larger proteins [e.g. Vroman, 1987; Vroman and Leonard, 1991]. The basic question that must be answered in order to develop a rational approach to the design of biocompatible materials is... what is it about proteins and surfaces that make them attract? Table 1.1 summarizes the contributions to AG, AH, and TAS of the various interactions that are thought to affect adsorption behaviour. Entropy, for instance, is intimately linked to secondary and tertiary conformational changes in the protein, and hence any scheme to elucidate the influence of entropy on adsorption behaviour requires a tool to probe such changes. Changes in local water structure can also lead to large changes in entropy. 16 SUBPROCESS CONTRIBUTION TO IMPORTANT AG , H « PARAMETERS Hydrophobic Effect AH < or > 0 Hydrophobicity of the AS>0 sorbent and protein surface AG < or > 0 Overlap of sorbent and AH < or > 0 Distribution of charge and protein surface electric AS < or > 0 dielectric constants before fields AG < or > 0 and after adsorption Ion transfer between AH<0 Valency and size of surface and sorbent AS<0 transferred ions A G > 0 Changes in protein AH>0 Structural stability of the structure AS>0 protein molecule A G < 0 Table 1.1: Primary subprocesses involved in protein adsorption. Severe limitations are inherent in past approaches to investigating conformational changes due to adsorption. Circular dichroism spectroscopy can elucidate static structural changes between native and steady state adsorbed structures. However, because this method examines changes in polarization upon transmission of light through a heterogeneous medium, it is plagued by low signal strengths and, hence, requires signal averaging over long periods. This reduces temporal resolution to such an extent that dynamic studies are not possible. More importantly, because the technique requires matching solvent and sorbent refractive indices, it has to date only been applied to a single surface, silica, which is not biologically or clinically significant. Infrared (IR) absorption spectroscopy has been used with limited success to determine the fraction of adsorbed protein carbonyl groups in contact with the sorbent [Morrissey ar.d Stromberg, 1974]. However, IR studies require transparent surfaces and are hampered by the strong infrared 17 absorption due to water molecules. Efforts using electron paramagnetic resonance (EPR) spectroscopy [Pireaux, 1992] and nuclear magnetic resonance (NMR) spectroscopy [Benko et ai, 1975] have met with failure due to their inability to deal with complex, heterogeneous media. Calorimetric studies using differential scanning micro-calorimeters, which measure heat capacity as the system temperature is ramped, have been very successful in determining changes in thermodynamic properties during the protein adsorption process, however they are unable to directly observe structural changes, per se. There is a need for a novel instrument which can elucidate adsorption induced structural changes, provide dynamic information, operate in heterogeneous aqueous solution, examine arbitrary surfaces, and function using realistic concentrations of proteins. Additionally, in order to correlate this structural information with thermodynamic properties, the instrument should be able to operate remotely (e.g. in a calorimeter). In principle, Remote ultraviolet resonance Raman spectroscopy satisfies all of these criteria. A more thorough discussion and literature review of Raman spectroscopy and its use for protein studies will be presented in the following section. 18 1.3 Comparison of Spectroscopic Methods to Determine Protein Structure 1.3.1 General Comments on Spectroscopic Methods In order to establish the appropriate applications niche for the FO-UVRRS instrument described in this thesis, it is first necessary to examine the efficacy and roles of existing techniques to determine protein structure. In this section the most common instrumental methods to determine protein structure are briefly described. They include X-ray diffraction (XD), multi-dimensional, multi-nuclear magnetic resonance spectroscopy (NMR), ultraviolet-visible absorption spectrophotometry (UV-VIS), electronic circular dichroism spectroscopy (ECD), infrared absorption spectroscopy (IR), vibrational circular dichroism spectroscopy (VCD), fluorescence spectroscopy (FS), regular (non-resonance) Raman spectroscopy (RS), and ultraviolet resonance Raman spectroscopy (UVRRS). All of these techniques are spectroscopic in nature and can, to some extent at least, deal with aqueous (D 20 in the case of NMR) solutions. The exception to this is X-ray crystallography which is included because of its stature as a "Gold Standard" in calculating atomic positions in a protein molecule in the solid state. Furthermore, with the exception of NMR, all the methods described below make use of optical (LR, visible, or UV) spectroscopy. In this sense, it is relatively straightforward to compare the theoretical bases and relative capabilities of the techniques. It is first necessary to establish a fair set of criteria for comparison. I propose the following set: (1) level of structure examined (e.g. primary, secondary, etc.); (2) limit of detection (concentration of protein required for a quantitative determination of structure); (3) sensitivity (change in signal caused by a change in the aspect of protein structure 19 under consideration); (4) selectivity (ability to examine different aspects of structure independently); (5) time response (this is especially important in studies to determine kinetics, and in general to maximize the sample throughput of the instrument); (6) range of applications (ability to operate in situ and or in vivo in various environments - this includes, in particular, isolated (remote), turbid and/or highly absorbing samples); (7) instrument size (this can be an important practical criterion for many applications, especially where portability is an issue); and (8) cost (the importance of this is self evident). It will become evident in choosing between the techniques that there is a tradeoff between these criteria. Furthermore, in many cases it is possible to trade off between these criteria for a given instrument, e.g. it is often possible to exchange time response for sensitivity in many of these techniques by increasing the data acquisition time to collect more signal. Additionally, some of the optical techniques are absorption based (UV-VIS, ECD, IR, VCD), while others are emission or scattering based (RS, UVRRS, FS) and this distinction in itself may impact upon the decision to choose a particular technique. There is no "best" spectroscopic technique for all investigations of aqueous protein structure. Any comparison between the techniques must explicitly involve a consideration of the particular system under investigation with particular attention paid to the criteria listed above, especially items 1 to 4. For example, UV-VIS is a sensitive, cheap, rapid technique to determine the concentration of a protein based upon its tryptophan residue content; however if one is more interested in the environment and hydrogen bonding of the tryptophan and other aromatic residues, then UVRRS is a better choice, and if detailed information concerning the environment and conformation of a specific residue with any 20 arbitrary position and environment within the protein, then NMR would be an appropriate choice. Recently, Havel etal. [Havel, 1996B] reviewed the common spectroscopic techniques used for protein structural determination. The reference by Andrade and HJady [Andrade and Hlady, 1986] provides a good review of some optical spectroscopic methods used to investigate protein adsorption. Table 1.2 provides a starting point in comparing the relative capabilities and limitations of the techniques It is evident from Table 1.2 that the relatively information-rich techniques (XD, NMR, and to some extent VCD, LR and RS) require relatively large amounts or concentrations of proteins and/or make use of difficult or expensive experimental arrangements. The more robust, inexpensive techniques (UV-VIS, ECD and FS) offer lower detection limits, but are lacking in information content. UVRRS has established a niche as a technique with moderately low limits of detection and moderately high sensitivity and selectivity to secondary structure and aromatic amino acid residue tertiary structure [e.g. Austin et al, 1993A]. The primary limitations of the technique inhibiting its more widespread use are practical issues relating to expense, complexity, and experimental inflexibility of the hardware components of a conventional UVRRS system. The fiber-optic probes and their numerical analyses described in this thesis address these limitations and enable, for the first time, routine in situ studies. 21 Technique Structure Sensitivity Advantages Disadvantages XD 2 ° & 3 ° atomic resolution, detailed structures which include side-chain orientations and interactions very slow, very expensive, limited to proteins that crystallize, aqueous systems not possible, in situ studies impossible NMR 2 ° & 3 ° atomic resolution, structural dynamics information slow, expensive, limited to small proteins, requires significant amount of protein, sample must be pure UV-VIS local 3° inexpensive, fast, easy to operate, almost universally available and understood, moderately low limits of detection poor resolution, relatively little information, poor sensitivity, limited mainly to aromatic residues, little or no 2° structure information ECD local 2° & 3° relatively simple operation poor resolution, poor selectivity, in situ studies difficult or impossible IR 2° widely available, simple operation, sensitive to a variety of structural elements water interference, high concentrations, overlapping peaks limit selectivity VCD 2° sensitive probe of secondary structure high concentrations, water interference, equipment is rare, in situ studies difficult FS local 3° low concentrations, structural dynamics, in situ studies relatively easy only useful for fluorescent chromophores, no 2° structure information RS 2° & local 3° in situ studies relatively easy, sensitive to a variety of structural elements overlapped peaks limit specificity, high concentrations required, fluorescence background UVRRS 2° & local 3° excellent specificity provided by resonance, little background, moderately low detection limits, specific structural information expensive equipment, relatively difficult to operate, not commercially available as a stand alone instrument, until now no in situ capability Table 1.2. Comparison of spectroscopic methods for investigating protein structure. XD: X-ray Diffraction, NMR: nuclear magnetic resonance, UV-VIS: UV-visible absorption, ECD: electronic circular dichroism, IR: infrared, VCD: vibrational circular dichroism, FS: fluorescence, RS: Raman, UVRRS: UV resonance Raman, 2°: secondary, 3°: tertiary. 22 A comprehensive description and comparison of the spectroscopic techniques listed in Table 1.2 would be an ambitious project which lies beyond the scope of this document. However a brief description of the techniques is warranted for the purpose of reference and comparison in providing a scientific context in which to view the FO-UVRRS system that is the subject of this thesis. To that end, sections 1.3.2 to 1.3.10 provide brief descriptions of each of the techniques and establish the motivation for using UVRRS and developing fiber-optic UVRRS probes. As an introduction to spectroscopy for protein structure determination, the reviews by Havel et al. [1996B], Campbell and Dwek [1984], and the citations therein may be referred to for greater detail. An excellent introduction to the basic theory of molecular spectroscopy (suitable for engineers) is provided by Ingle and Crouch [1988, Chapter 10], and the text by Skoog and Leary [1992] provides an introduction to a variety of spectroscopic techniques. The reader versed in protein spectroscopy for biophysical investigations may skip or skim sections 1.3.2 to 1.3.8 without a serious loss of continuity. 1.3.2 X-Ray Diffraction X-ray diffraction (XD) or x-ray crystallography is a very high resolution technique that makes use of the diffraction of short wavelength light by a crystalline protein to determine protein structure with atomic accuracy [see, e.g., Chapter 12 in Campbell and Dwek, 1984]. Nearly monochromatic X-rays are directed at a protein crystal and a diffraction pattern of regions of destructive and constructive interference is observed. The position, magnitude, and phase of the diffraction pattern bright spots can be used to 23 reconstruct the atomic coordinates of the protein with a resolution approaching the wavelength of the x-rays used (typically ca. 0.1 nm = 1 Angstrom). This technique, in addition to being expensive and time consuming, is limited to solid samples of proteins that crystallize well. XD is not used as a routine method for determining changes in protein structure, but remains the method of choice when high resolution determinations of protein structure (including side chain positions) are required. XD can also measure protein and ligand structure in a bound complex. 1.3.3 Nuclear Magnetic Resonance Spectroscopy Nuclear magnetic resonance spectroscopy [see, e.g. Gronenborn and Clore, 1996; Bruice, 1995] (NMR) is the method of choice for obtaining extremely high resolution structural information of proteins in aqueous (D2O) solution provided samples are pure, the protein is small (< ca. 25 kDa), high concentrations can be obtained, and high sample throughput is not required. Most modern NMR instrumentation setups and methodologies involve placing a sample in a strong magnetic field, perturbing the resulting proton magnetization with a burst of radio-frequency (RF) radiation, and recording the free induction decay immediately thereafter. Without delving into further detail, this may be extended to multiple "dimensions" by allowing for various combinations and permutations of preparation, evolution, mixing, and detection periods, thereby creating 2D, 3D, or 4 D proton (and 15N) NMR. By observing proton-proton magnetic interactions that are due to nuclear Overhauser effects, a set of magnetic environment specific inter-proton distances may be derived. Advanced computer algorithms may then be applied to these data to 24 obtain atomic coordinates at resolutions better than 5 Angstroms. While multidimensional proton NMR provides solution protein structures with resolutions comparable to XD, its main drawbacks lie in its expense, demand for computational power, and high required concentrations. NMR experiments must build up global structural information from what is inherently very local information and consequently for many applications that do not require such high resolution, the use of NMR is "experimental overkill". In this sense, proton NMR, like XD, is not really comparable to the optical spectroscopic techniques that either examine particular isolated molecular chromophores (e.g. aromatic amino acids) or more global arrangements of amide chromophores (e.g. a-helices, 0-sheets, random coil). Additionally, many in situ and in vivo problems are simply not amenable to solution by proton NMR due to the large number of hydrogen atoms associated with the water solvent. NMR is generally not suitable for protein adsorption studies for several reasons. Apart from the drawbacks mentioned above, difficulties are encountered owing to the use of complex, heterogeneous media, including heterogeneities in, and interfering signals from the sorbent surfaces. 1.3.4 Ultraviolet-Visible Absorption Spectrophotometry Ultraviolet-visible absorption spectrophotometry (UV-VIS) is a mature technique that makes use of the fact that different chromophores of a protein absorb light to excite electronic transitions at different wavelengths and with different strengths [see, e.g., Skoog and Leary, 1992; Ingle and Crouch, 1988; Campbell and Dwek, 1984]. A UV-VIS 25 spectrum consists of a plot of absorbance (log T i n/T o u t) versus wavelength. Inexpensive UV-VIS instruments have been ubiquitous as biochemical laboratory instruments for many years and are commonly applied to the measurement of concentrations of proteins using the absorption peaks of the aromatic amino acids near 280 nm. The primary drawback of UV-VIS as a method to investigate protein structure is the fact that its broad, unresolved peaks result in it being relatively information-poor and insensitive to changes in structure. This stems from the fact that the strength and energy of the broad electronic transitions probed by UV-VIS are left relatively unperturbed upon changes to overall protein structure. An additional failing of the technique occurs because, like all absorption based techniques, it is rather cumbersome to perform in situ. While changes in aspects of protein structure may induce only subtle changes in the absorption spectrum, these can sometimes be revealed through derivative spectroscopy [Havel, 1996B] which has, in particular, proved to be a good technique to measure tyrosine exposure to water. The lack of structural information in UV-VIS spectra and the problems associated with light scattering generally preclude the use of this technique for investigating adsorbed proteins. It should be noted, however, that UV-VIS is often used in protein adsorption studies as an analytical method to determine protein concentration in solution and subsequently, through a mass balance, adsorbed protein amounts. 1.3.5 Electronic Circular Dichroism Spectroscopy While UV-VIS is relatively information-poor, further information concerning protein structure may be garnered from UV absorption by examining the difference in 26 absorption between left and right circularly polarized light. This is the basis for electronic circular dichroism spectroscopy (ECD) of proteins [Mulkerrin, 1996; Ito et al, 1993; Cherney, 1979] which is performed using an instrument called a spectropolarimeter. An ECD spectrum is sometimes presented as a plot of Ae (EL-£R) VS. wavelength, where e refers to the molar absorptivity of either left or right circularly polarized light, as denoted by the appropriate subscript. More often, however, it is presented as a plot of the mean residue weight ellipticity ( [0]MRW ), which is directly proportional to Ae, versus wavelength. Since CD spectra are inherently difference spectra, they usually possess both positive and negative bands. Rosenfeld [Rosenfeld, 1928] developed the modern quantum mechanical theory of optical activity and Kirkwood and Moffitt furthered this in applying it to chiral activity of polymers [Kirkwood, 1937; Moffitt, 1956; Moffitt etal, 1957]. Significant contributions to the understanding of protein CD spectra have been made by Manning and Woody [Manning, 1989; Woody, 1968; Manning and Woody, 1989]. Proteins possess several optically active chromophores in the amide linkages, aromatic amino acids, and disulfide bonds. ECD is arguably the most commonly used technique for routine determinations of secondary structural content of proteins, and is therefore worthy of further discussion in order to establish why FO-UVRRS may be an appropriate tool to supplant it in some applications. The amide chromophore has optically active n —>7C* and T t - M t * electronic transitions, near 220 nm and 190 nm, respectively. The theory behind ECD is relatively complex and beyond the scope of this document; however the result is that the common secondary structure motifs (a-helices, parallel P-sheets, anit-paraiiel P-sheets and random 27 coil structures) have distinct ECD resulting from differences in the dihedral angles \\f and . It is therefore possible to assign secondary structure content to a protein on the basis of its ECD spectrum by comparing it to known spectra of the secondary structural components. Owing to interferences in protein amide ECD spectra from aromatic amino acid chromophores and perturbations from different electronic environments, however, a more accurate approach is to first establish a basis set of spectra using ECD spectra from proteins whose secondary structural content has been determined from XD (or NMR). Advanced fitting algorithms may then be used to determine the structural content of the protein or peptide under investigation. There are many problems and caveats with the use of ECD for routine determinations of protein secondary structure [Manning, 1989]. Fundamental problems arise due to interferences from changes in optical activity of aromatic amino acids and disulfide bonds, as well as from changes in the electronic environments of the secondary structure motifs under consideration. The latter effect may perturb the electron cloud and change the ECD spectrum without a corresponding change in secondary structure. The broad peaks characterizing the various structural elements overlap, thereby further reducing selectivity. Practical problems arise due to the nature of the instrumentation and the fact that ECD is an absorption, rather than emission or scattering based technique. For example, studies of adsorbed proteins are made difficult due to absorption, reflection, and possible optical activity of the substrate. Most of the adsorption studies that have been done using ECD [e.g. Norde and Favier, 1992] have made use of dilute dispersions of ultrafint, solvent-refractive-inJex matched particles. Because of this, the surfaces used 28 have generally been limited to silica, although good results have recently been obtained for adsorption to a special type of Teflon [Maste et al., 1996]. Furthermore, true in situ ECD studies are difficult or impossible, and, to the author's knowledge, a remote (fiber-optic) ECD instrument has never been demonstrated. 1.3.6 Infrared Absorption Spectroscopy Infrared absorption spectroscopy (IR) [see, e.g. Skoog and Leary, 1992; Ingle and Crouch, 1988; Campbell and Dwek, 1984] is another technique commonly used for determinations of aspects of protein structure, including assignments of secondary structural content [Surewicz and Mantsch, 1996]. LR provides, in many ways, similar or complementary information to RS or UVRRS, and therefore warrants special attention in this overview. LR involves transmitting broadband infrared light through a sample and thereby determining the wavelengths at which the sample absorbs. Early LR instruments were dispersive, scanning devices incorporating diffraction gratings; however these have largely been supplanted by Fourier transform LR (FT-IR) instruments that demonstrate superior optical throughput and signal-to-noise ratio (SNR). While absorption bands in the visible and UV regions of the spectrum arise due to electronic transitions, mid-IR absorption bands arise due to changes in the vibrational states of the molecule. Molecular vibrations are IR absorption active if they result in a change in the dipole moment of the molecule. LR spectra generally consist of a series of peaks which occur, for most practical purposes, in the wavenumber range from 500 to 2000 cm"1 (5 to 20 pm wavelength). The 29 particular wavelength of a molecular vibration may be determined through a process known as normal mode analysis. The most common use of ER in protein investigations is as an assay of secondary structure. Elliot and Ambrose [1950] were the first to demonstrate a correlation between secondary structure and infrared spectra of proteins (specifically the amide bands). It is now known that there are nine amide vibrational bands in TR spectra, however due to interference from non-amide signals, insufficient strength, or inadequate sensitivity to changes in secondary structure, usually the amide I band (primarily C=0 stretching coupled with N-H bend and C-N stretch) occurring between 1600 and 1700 cm"1 is primarily of interest. Each secondary structure motif produces a particular amide I signature. However^ since most globular proteins consist of regions with different secondary structures, a complex amide I region results, consisting of overlapping amide signals from cc-helices, P-sheets, and random coil structures, as well as contributions from interfering species. Analysis usually proceeds using one of two methods: (a) factor analysis, making use of a calibration set of spectra of proteins of known structure, or (b) resolution enhancement and deconvolution techniques that improve resolution and reduce overlap to the extent that a meaningful analysis may be performed. Because molecular vibrations are more conformationally dependent than electronic transitions, and because IR spectra consist of a larger number of narrower peaks than UV-VIS or ECD, IR spectroscopy of proteins has greater specificity than UV-VIS and exhibits greater sensitivity to changes in secondary structure. However, these advantages are only gained at the cost of poorer limits of detection, thus necessitating the use of higher protein 30 concentrations, a consequence which is rarely desirable, and in many cases impossible. Furthermore, there exist several other drawbacks of IR spectroscopy for protein structure determination. Foremost among these are the facts that (a) water absorbs strongly in the mid-LR, and (b) as an absorption technique, in situ studies are difficult. These problems can be addressed to some extent, although not overcome, through the use of an attenuated total internal reflectance (ATR) attachment which makes use of absorption of the infrared evanescent wave that decays exponentially into the sample solution outside of a dielectric waveguide. LR has been used to investigate protein adsorption to surfaces with some degree of success. In an early study, Morrissey and Stromberg investigated the conformations of adsorbed blood proteins using the LR amide I band [Morrissey and Stromberg, 1974]. Subsequently, ATR-FTLR has been used to investigate adsorbed protein conformation in several studies [e.g. Castillo et al, 1985; Pitt et al, 1987; Jakobsen and Wasacz, 1987; Chittur et al, 1987; Steadman et al, 1992; Barbucci and Magnani, 1994; Ball and Jones, 1995; Buijs et al, 1996] which generally utilized the amide I and/or amide ILT. LR bands. In most of these studies, LR spectra analysis provided only qualitative information concerning secondary structure, i.e. detecting increases or decreases in a-helix content, but not actually determining the fraction of the protein molecule in a helical conformation. IR is rarely used for investigating tertiary structure in adsorption studies. The main drawbacks of LR for blood protein adsorption studies were enumerated by Chittur et al. [Chittur et al, 1987] as follows: (1) spurious signals introduced in subtracting the water background; (2) general complexity of the FTLR spectra (lack of specificity); (3) poor baseline stability; 31 (4) general similarity of protein IR spectra (lack of sensitivity); (5) general interference from microenvironmental changes, solutes, and surfaces; and (6) interference from unidentified plasma components. 1.3.7 Vibrational Circular Dichroism Spectroscopy Vibrational circular dichroism spectroscopy (VCD) [Keiderling, 1996] is a relatively new technique for investigating protein structure through an examination of the optical activity of infrared absorption bands [Nafie, 1984]. This is done by producing an IR difference spectrum of left and right circularly polarized light transmitted through a sample. Although VCD instruments are not commercially available at present, they may be constructed easily through modification of either FT-ER or dispersive ER. systems [Keiderling, 1981]. Dispersive VCD systems provide faster data acquisition; however FT-ER VCD systems result in higher resolution. Further, accessories to convert FT-ER instruments to VCD instruments are commercially available from some vendors. The relationship between VCD and IR is analogous to that between ECD and UV-VIS. Alternatively, VCD can be considered a fusion of ECD and ER. The strength of VCD as a method for determining secondary structure derives from the fact that, like ER, the ground electronic state vibrational transitions that are probed arise due to a variety of specific amide vibrations involving all of the relevant atoms, and therefore exhibit sensitivity to changes in hydrogen bonding and strain that are indicative of changes in secondary structure. Unlike ER, however, the optical activity aspect of VCD is derived from interferences in the polarization and intensity character of the transmitted light which 32 arise directly from the 3D arrangement of amide chromophores and thereby provide a more global measure of secondary structure. Like ECD, VCD spectra generally have both positive and negative bands in the optical activity of each vibrational transition examined. Particular secondary structure motifs exhibit characteristic bandshapes and shifts. Although theoretical models exist to correlate VCD spectra and molecular structure, at present these are too complex for all but the smallest molecules, and empirical models and calibrations are generally relied upon. At this time, VCD has many drawbacks precluding its more widespread use. First, integration times can be as long as two to three hours using a FT-IR VCD instrument, and protein concentrations in the range of 5-20% are usually required. VCD suffers additionally from the same failings as IR, namely solvent interference from water (necessitating the use of D 2 0 and limiting studies to the amide I band), and a lack of reported remote probes for in situ studies. Thus, while from a biophysical point of view VCD provides nearly the ideal form of optical/molecular interaction information for a certain class of investigations, at present it suffers from severe practical difficulties and has not, to the author's knowledge, been used for protein adsorption studies. 1.3.8 Fluorescence Spectroscopy Fluorescence spectroscopy (FS) involves the excitation of a molecule (or chromophore) to an excited electronic (and possibly vibrational) state, followed by vibrational relaxation to the ground vibrational state of the excited electronic state, electronic relaxation (accompanied by photon emission) to a possibly excited vibrational 33 state of the ground electronic state, and finally possible relaxation to the ground vibrational state. FS requires a light source (usually a lamp or laser), collection optics, a wavelength selective device (usually a grating monochromator, prism, or filter), and a detector (often a PMT or CCD array). In addition to the excellent detection limits and time response deriving from the high quantum efficiencies for fluorescence of many chromophores, this arrangement is also very amenable to incorporation in a remote probe using optical fibers and operates well in aqueous environments. There exist two common methods of FS for investigations of protein structure. First, the intrinsic fluorescence of tryptophan excited around 280-300 nm and detected near 330-350 nm exhibits excellent sensitivity and selectivity to the microenvironment of tryptophan residues. Shifts in the wavelength of maximum fluorescence intensity resulting from changes in TRP microenvironment can provide additional information. For example, fluorescence intensity shifts are often used as markers of protein folding/unfolding kinetics, since tryptophan residues are usually buried in folded globular proteins. Upon unfolding and exposure to water, the intrinsic fluorescence of the indole ring system is quenched, reducing the fluorescence intensity by around 80%. The second common method of employing FS for protein investigations include a family of techniques called energy transfer (ET) methods [e.g. Haas, 1996]. These make use of energy transfer between a donor and acceptor label attached to a protein to determine average distances between acceptor and donor, distance distributions, and rates of change of distances. ET is of considerable use in the determination of protein folding and binding kinetics. 34 FS exhibits excellent limits of detection, excellent time resolution and throughput, is relatively inexpensive, performs well in aqueous solution, and is amenable to remote use with fiber-optic probes. However, in contrast with many of the other techniques, fundamental limitations of FS preclude its more widespread use for protein structural investigations. Structurally sensitive intrinsic protein FS is limited to a very small number of chromophores and ET methods require special site-specific covalently attached labels. These methods provide only very limited information concerning a particular aspect of tertiary structure, and FS is incapable of providing secondary structure information. Thus, although robust and signal-rich, FS is information-poor. It is for this reason that it finds many of its uses in determination of kinetics rather than true structural investigations. FS has been used, for example, for investigations of protein adsorption to surfaces in studies investigating quantification of adsorbed protein amount [Horsley et al, 1987], protein adsorption kinetics [Anderson et al, 1987], adsorbed protein thermal stability [Steadman et al, 1992], adsorbed fluorescence lifetime and anisotropy [Maste et al, 1996], to name only a few. In almost all cases, FS data is considered as "all or none" and used to detect gross changes in tertiary structure (relating to tryptophan or label environments or positions) that occur upon adsorption or unfolding. Quantitative information may be present in the form of fluorescence intensities or fluorescence maximum wavelength; however this can usually only be used to quantify the amount of protein in each state, rather than the extent to which the protein conformation has changed. 35 1.3.9 Non Resonance Raman Spectroscopy The Raman effect refers to the inelastic scattering of incident photons from a molecule with an associated change in the vibrational energy state of the molecule. It was first discovered by, and subsequently named after the Indian physicist C. V. Raman in 1928 [Raman and Krishnan, 1928], an accomplishment for which he was awarded the Nobel Prize in 1930. Raman spectroscopy is performed by illuminating a sample with monochromatic (laser) light, collecting the light scattered from the sample, separating and detecting the frequency components. Normal (non-resonance) Raman spectroscopy [see, e.g., Tensmeyer and Kauffman, 1996] (RS) occurs when photon excitation energy is not near that of an electronic transition of the chromophore associated with the vibration under consideration. RS is most often performed using light in the visible or near infrared (NLR) regions of the spectrum. The theory and instrumentation for both RS and resonance RS is further discussed in section 1.4 below. Raman spectra are usually plotted as intensity (in counts or counts/s) versus downward frequency shift from the excitation line, in wavenumbers. Like LR, RS probes the vibrational modes of a molecule and, in fact, many of the same vibrational modes are detected in both types of spectra. However, unlike LR, a vibration is Raman active only if it changes the polarizability of the molecule. Thus, RS is in many ways complementary to LR. RS shares many of the advantages of LR, including a sensitivity to protein secondary structure via the amide vibrations, an ability to investigate a variety of other protein vibrational modes, and the existence of relatively inexpensive, commercially available equipment. For our purposes, it is significant that the structurally sensitive amide 36 III vibration is more pronounced in Raman than in ER spectra. Unlike ER, water peaks do not interfere to any great extent and it is also quite compatible with the use of fiber-optic probes for in situ analysis [e.g. McCreery et al., 1983; Myrick and Angel, 1990; Greek et ai, 1996B]. These advantages over ER arise primarily because RS is a scattering, rather than an absorption technique. The recent development of the Raman analogue to VCD, known as Raman optical activity [Hecht et al., 1991] (ROA), is notable in that it may be able to provide similar information to VCD, but without some of the drawbacks (specifically water interference and lack of remote probes). Unfortunately, RS also shares a few of the same drawbacks as ER, including relatively poor limits of detection and a large number of overlapping peaks that limit selectivity. Protein concentrations in the percentage (w/w) range are often required. Additionally, visible excitation RS is often accompanied by a large fluorescence background that degrades SNR through background shot noise. This can be overcome using Fourier transform RS (FT-RS) and excitation in the NER (typically 1064 nm); however the 1/X.4 dependence of the scattering cross section reduces signal intensity. 1.3.10 Ultraviolet Resonance Raman Spectroscopy The cross-section for Raman scattering from a particular vibrational mode can be increased substantially if the excitation photon energy is comparable to that of an electronic transition of the chromophore associated with the vibration. This is the basis for resonance Raman spectroscopy (RRS), which, for many biological chromophores, involves excitation in the deep ultraviolet (DUV), hence ultraviolet resonance Raman 37 spectroscopy (UVRRS) [e.g. Thamann, 1996; Asher, 1993A; Asher, 1993B; Austin etal., 1993A]. While UVRRS instrumentation is superficially similar to that for RS, significant differences exist due to the greater difficulties in operating in the DUV spectral region. Theory, instrumentation, and use of UVRRS are discussed in much greater detail in section 1.4 and Chapter 2. Resonance enhancements of Raman signals in the DUV spectral region may be obtained for the amide backbone; tryptophan, tyrosine, and phenylalanine residues; and to a lesser extent, histidine and proline residues. All of these signals are sensitive, to a greater or lesser extent, to secondary (in the case of the amide vibrations) or tertiary (in the case of the amino acid residues) structure. Specifically, the positions and relative peak heights of the amide bands may be correlated quantitatively with secondary structure (percentage a-helix, 13-sheet, random coil, etc.) while the resonance enhanced amino acid bands can in many cases be correlated with such measures of tertiary structure as environment hydrophobicity, hydrogen bonding to specific atoms, and bond angles (see sections 1.4.3 to 1.4.5, below). A certain degree of selectivity is conferred upon any vibrational technique (e.g. LR, RS) by virtue of the relatively large number of moderately resolved peaks; however, since the electronic transitions that give rise to the resonance effects for the various chromophores have different energies, a much more powerful method of selectivity is available in UVRRS experiments by judiciously choosing the excitation wavelength to match the electronic transition energy of the chromophore under investigation. In some cases, the resonance effect may permit limits of detection approaching or exceeding those of FS. With present technology, in most cases resonance enhancable chromophores may be investigated in the micromolar range. An additional advantage is gained through excitation at wavelengths shorter than approximately 260 nm. In this region, the fluorescence background that is so problematic in visible RS is reduced in intensity and shifted away from the Raman signature region (500 - 2000 cm"1), thereby improving SNR. Finally, the use of UVRRS provides another dimension of information, known as the resonance Raman enhancement profile (RREP) which consists of a plot of Raman cross section versus excitation wavelength. In many cases the RREP is also highly sensitive to structure. ROA measurements may also be possible using UVRRS. While there exists a significant literature base concerning the use of RS and UVRRS for investigations of protein structure, to my knowledge neither RS nor UVRRS have been used to investigate adsorbed proteins. RS and visible resonance RS have enjoyed a modest degree of success in studies investigating adsorption of small aromatic molecules to silica [Simpson and Harris, 1990; Schick and Sun, 1994; Matzner et al, 1994] and carbon [Kagan and McCreery, 1995]. Surface enhanced Raman spectroscopy (SERS) has been used to examine proteins adsorbed to surfaces [e.g. Vidugiris et al, 1988; Hobara etal, 1994]. Although SERS has some impressive performance characteristics, it is very limited in that it requires very specific surfaces (e.g. gold or silver) which have little biological significance and will not be discussed further. 39 1.3.11 Summary and Outlook The methods discussed in sections 1.3.2 to 1.3.9, above may be generally classified as having (a) good information content but poor limits of detection (ER, RS, VCD, ECD), (b) good to excellent limits of detection but poor information content (FS, UV-VIS), or (c) information providing atomic resolution, but with great expense and experimental complexity (XD, NMR). Further, only FS, RS, and to some extent LR are truly amenable to in situ studies. The quality, sensitivity, and selectivity of the information provided using UVRRS make it unique amoung the various spectroscopic methods and establish its potential utility for use in a large class of investigations. However, significant practical difficulties must be overcome to permit its more widespread use. Foremost among these is laser technology for UVRRS (section 2.2) which is often expensive, of low duty cycle, and only partially tunable. Advances have recently and are presently being made which promise to overcome this. The disadvantage that is primarily addressed and overcome in this thesis is the lack of remote probes for routine in situ studies. This has been due to UV-induced damage to optical fibers (addressed in Chapter 3) and the lack of appropriate probe designs (addressed in Chapter 4). 40 1.4 Overview of UVRRS for Protein Studies 1.4.1 General Comments on UVRRS As mentioned, resonance Raman spectroscopy (RRS) makes use of an interaction with excited electronic states to greatly increase the vibrational-inelastic (Raman) scattering from a chromophore. RRS began to evolve from RS in the 1960's and 1970's when the quantum mechanical approach used to describe the change in scattering cross section with wavelength was fully developed [Albrecht and Huteley, 1971; Hutely and Jacobs, 1969; Albrecht, 1960; Tang and Albrecht, 1970; Tang and Albrecht, 1968; Peticolas et al, 1970]. Early studies tended to concentrate on chromophores with visible or near UV resonances (e.g. methyl-orange, potassium chromate, chlorophylls, metalloproteins, etc.) due to the (relative) ease of obtaining laser light to investigate these resonances. However, most resonance enhanced protein and DNA chromophores can only be excited with narrow-band DUV excitation and therefore had to wait for the development of a tunable DTP/ laser system [e.g. Asher et al, 1983]. A complete discussion of the equipment used to effect UVRRS will be left to Chapter 2. In sections 1.4.2 to 1.4.6 the theory of RS and UVRRS is presented briefly and its application to protein chromophores is discussed. Several excellent reviews of UVRRS have appeared in the literature [e.g. Austin e/a/.,1993A; Austin etal, 1993B; Asher, 1993A; Asher, 1993B; Thamann, 1996], and several books [e.g. Long, 1977; Ferraro and Nakamoto, 1994] have done an exemplary job at describing and discussing RS and RRS. These may be consulted for further details, as only the relevant results are presented here. Notable among the reviews is that by Asher [1993A, 1993B] for its concise, 41 comprehensible review of the field, and that by Austin et al. [1993A] for its comprehensive and complete discussion of protein UVRRS. 1.4.2 Molecular Vibrations RS and UVRRS (as well as LR) provide measures of the frequencies of molecular vibrations and the extent and manner in which these vibrations interact with the traveling periodic electric field of light waves (or photons, depending on whether one adopts a classical or quantum mechanical viewpoint). Vibrations of any sort occur when there exists a restoring force acting in the opposite direction to the displacement from equilibrium of a particle or system. Sinusoidal (simple harmonic) motion results when the restoring force is directly proportional to the displacement from equilibrium. This condition also implies that the system potential well may be described by a parabola whose vertex occurs at the equilibrium point. For example, the two atoms in a diatomic molecule experience a force that, for small displacements, is roughly proportional to the displacement from equilibrium with a force constant K, in units of restoring force per unit length of displacement. Solution to the appropriate differential equation shows that any such diatomic molecules (with masses mi and m2) undergo a fundamental vibration with a frequency given by equation 1.2. where p is the reduced mass of the system given by p=mini2/(mi+m2). Clearly, a "stiffer" bond (higher K) results in higher vibrational frequencies; usually this is also associated with a stronger bond. However since K is really a measure of the curvature of the (1.2) 42 potential well near equilibrium, this need not always be the case. It is also clear that the greater the difference in masses between the atoms in a heterodiatomic molecule, the higher the resulting vibrational frequency. For an actual diatomic molecule, however, a quantum mechanical approach and the (stationary state) Schrddinger equation for this system (equation 1.3) must be solved d2y/ Z7^nfr. 1 „ 2s dq2 h2 where VJ/ is the wavefunction, q is the net interatomic displacement from equilibrium, h is Plank's constant, and E is the energy. This may be solved to give £ „ = A v [ y + i ) (1.4) where u is any positive integer and v is given by the classical formula in equation 1.2. This result indicates that the energy associated with a given vibration may change only by discrete units of hv. These results are for a harmonic oscillator with a parabolic potential well. In fact, the actual potential well of a diatomic molecule is more accurately described by a Morse potential which approaches infinity as the interatomic distance approaches zero, and approaches a finite value as the interatomic distance approaches infinity (the dissociation energy). This results in a series of possible energy values for a given vibration that are spaced by less and less energy as energy increases, and approach a continuum at the dissociation energy. The discrete values of vibrational energy changes are the origin of the individual lines associated with Raman and infrared spectra. At first glance, the vibration of a large, polyatomic molecule appears rather complex and without any overall simple periodicities. However, a mathematical analytical •(E--Kq<)y, = 0 (1.3) 43 technique called normal mode analysis may be used to deconvolve such complex motion into normal modes of vibration. In a particular normal mode, each atom undergoes simple harmonic motion at the same frequency (although possibly with zero amplitude). A polyatomic molecule with N atoms has 3N degrees of freedom. Since 6 of these are taken up by pure translational or rotational motion of the entire molecule, there exists 3N-6 normal modes of vibration (3N-5 for linear molecules). For molecules with particular symmetries, some of these may be degenerate (i.e. result in the same vibrational frequency). 1.4.3 Regular Raman Spectroscopy The Raman effect refers to the inelastic scattering of light from a molecule with a concomitant change in the vibrational energy of the molecule (Figure 1.5). Generally, L I G H T Figure 1.5. Schematic representation of Rayleigh (elastic) and Raman (inelastic) scattering. Stokes and anti-Stokes light are scattered at lower and higher energies, respectively, than the incident light. 44 Raman scattering from a molecule may result in increased (anti-Stokes) or decreased (Stokes) energy of the scattered light, and is usually accompanied by a large amount of elastically scattered Rayleigh light. Raman spectroscopy involves the scattering of narrow-band incident light from a molecule and detecting the transient changes in the vibrational energy states of the molecule, through the associated changes in the energy of the scattered photons (Figure 1.6). Classically, this may be thought of as the interaction or coupling of the periodically varying electric field of the incident light with the molecular vibration, mediated by the induced dipole of the molecule through its polarizability. The electric field of the incident light may be described by equation 1.5, E^^E^co^Kv.t) (1.5) By definition, the polarization or dipole moment of the dipole induced by an electric field is proportional to the polarizability, a, of the molecule and is therefore given by equation 1.6. P = aE = aE^ cos(2;rvy) (1.6) WAVELENGTH COMPONENTS SPATIALLY DISPERSED MONOCHROMATIC MANY WAVELENGTH H LIGHT COMPONENTS MULTI-CHANNEL DETECTOR Figure 1.6. A schematic representation of the basic setup for Raman spectroscopy. 45 (It should be noted that polarizability is, in general, described by a tensor.) If the molecule is undergoing vibration, the polarizability may depend upon the internuclear distance, q. For small displacements (i.e. near the equilibrium point) such a dependence may be approximated as a Taylor series truncated to two terms, viz. (da) a(q) = ao+ — q (1.7) \dqJ0 where the subscript 0 here denotes quantities at the equilibrium position. This may be substituted back into equation 1.6 along with the explicit time dependent sinusoidal molecular vibration with frequency vm (given by q=qocos(27tvmt)) to give, after simplification, equation 1.8: ( da\ P{t) = cc0E^ cos(2;rvy}+ tfo^™ cos(2;rvy) co&xvj) Kcq) o (1.8) The trigonometric product in the second term may be expanded, resulting in equation 1.9: W^ = »o^ maxcos(2^ v)+d -A 9o^ »x(cos(2^ (v0 + vm» + cos(2^ (v0 - vm)t)) 2\oqJ0 (1.9) Such a periodically changing polarization (induced dipole moment) results in radiation at each of the frequencies present in the sinusoidal terms of equation 1.9. Several useful results may be obtained from this classical analysis: (1) some of the incident light is elastically scattered at a frequency v0, (2) Raman scattered light occurs shifted both up (anti-Stokes shifted) and down (Stokes shifted) by the frequency (vm) of the molecular 46 vibration, and (3) the intensity of the Raman scattered light is directly proportional to the derivative of the polarizability with respect to vibrational displacement. For a harmonic oscillator, quantum mechanics also dictates that the change in vibrational energy state must be described by Ao=±l and therefore energy shifts between incident and scattered light of ±hv describe Stokes (-) and anti-Stokes (+) shifted scattering. While these shifts may be either vibrational excitations (+) or relaxations (-), they are usually observed as excitations since at room temperature the vast majority of vibrational modes exist in the ground state, as described by the Boltzmann distribution, viz. = (1.10) Fg where F e and F g are the fractions of the population in excited and ground states, respectively; kT is the product of Boltzmann's constant and absolute temperature; and AE is the difference in energy between ground and excited states. The selection rules used to determine whether a particular normal mode is Raman active are based upon the use of symmetry to determine whether the polarizability changes during the vibration. For small molecules, this may be accomplished by inspection through consideration of the probable changes in the polarizability tensor (via the polarizability ellipse). For larger molecules, however, an approach using symmetry point-groups derived from group theory must be adopted. This will not be discussed here. Since Raman spectra are probes of vibrational energies, it would be natural to plot them as scattered intensity, I, versus photon energy. Alternatively, considering the wave 47 nature of light, they could be plotted as I versus the frequency (f) or the wavelength (k) of the scattered light, since E=hf=hc/X.. In order to make the abcissa directly proportional to vibrational energy, and to keep the magnitude of the ordinate values to a manageable level, Raman spectra are conventionally plotted as I versus wavenumber shift, Av. The wavenumber shift is obtained as Av= l/Xiua - 1 Matter and is usually plotted in units of cm'1. Figures 1. 7 and 1.8 show the molecular specificity of Raman, as well as typical lineshapes, linewidths, and Raman shift magnitudes. 6000 883.3 cm ,-1 5000 h Raylei^i Line 1051.6 cm-1 (23255 nm) T — 4000 h 1095.2 cm"1 (23279 nm) 0 0(tfcrr) ZECtfcm1 433,Var) 600(tfcrr) airtfcnl 1C00(1/crr) VCL\Vcn) 227 nm 228 nm 229 nm 230 nm 231 nm 232 nm 233 nm Raman Shift Figure 1.7. Low frequency Raman spectrum of neat ethanol taken with a fiber-optic probe and 227 nm, 250 pW, 15 second excitation. 48 8000 r |T 6000 ; 8 Intensity Intensity • 1 e ™ 2000 -& 0 •J \ f \ j \ yf^^*^ \ Isopropanol 800 1000 1200 1400 1600 Raman Shift (cm"1) Figure 1.8. Raman spectra of methanol, ethanol, and isopropanol taken with a fiber-optic probe and 227 nm, 250 u,W, 15 second excitation (scaled and vertically translated). 1.4.4 Visible and Ultraviolet Resonance Raman Spectroscopy: Theory The resonance Raman effect describes the large increase in Raman scattering cross section often observed when the excitation photon energy approaches the energy of an electronic transition of the chromophore associated with the particular vibration. Figure 1.9 shows the energy level diagrams of FS, IR, UV-VIS, RS, and RRS for comparison. 49 V'=3 v'=2 v ' = l E,.v'=0 v=3 v=1 1 Eo,v=0 T INFRARED S T O K E S ANT1-STOKES RAYLEIGH RAMAN RESONANCE RAMAN FLUORESCENCE Figure 1.9. Energy level diagrams for infrared absorption, Raleigh, Raman, resonance Raman, and fluorescence events. E 0 and E i indicate ground and excited electronic states, v indicates the vibrational energy level of the ground electronic state, and v' indicates the vibrational state of the excited electronic state. The theory behind resonance enhancement has been presented in the references cited above, and the reader is referred to sources by Spiro and Czernuszewicz [1995], Wang and Van Wart [1993], Ferraro and Nakamoto [1994], and Albrecht and Huteley [1971] for quite readable presentations. The labels of the initial and final vibrational states of the ground electronic state are designated m and «, respectively; the excitation frequency and intensity to be Vo and Io, 50 respectively; the electronic transition associated with the resonance is denoted e. The intensity (at a fixed concentration), I™, of any Raman band associated with the m->n vibrational transition and resulting in light scattered at a frequency of Vo-v™ is given by [Clark etal., 1974]: where (ctpo)™ denotes the component of the polarizability tensor describing the vibration associated with the m->n vibrational transition with column and row indices of p and a, and a™ is the scattering cross-section. The analysis proceeds by considering the expression for (ape)™,, viz. [Clark etal., 191'4] where Ven, and vm are the frequencies that would be associated with the transitions e-»m and e—>n, respectively; Te is the bandwidth of the state which keeps the resonance finite even at Ven, = vo; and Mab denotes the electronic transition moment from state a to state b given by, for example, equation 1.13: where the *F terms represent the total wavefunctions of the particular states and p<, represents the a* component of the electric dipole moment. It is clear from the form of equation 1.11 and the denominator on the right hand side of equation 1.12 that when the excitation frequency, Vem, is not near that of an electronic transition, Vo, then the scattered intensity (or Raman cross section) increases pa (1.11) (1.12) (1.13) 51 approximately as v 0 4 (the vibrational frequency generally being very small compared to vo) with a sloping background produced by the changing denominator of equation 1.12. On the other hand, as Vem approaches vo, the resonance condition is met and the scattering cross section increases very much faster than v 0 4 The damping term /T e prevents the cross section from increasing without bound. Thus far, this analysis does not indicate how different vibrational modes are enhanced differently (selectivity). This requires more explicit consideration of the quantum mechanical wavefunctions and proceeds [see, e.g. Clark etal., 1974; Homborg and Preetz, 1976; Ferraro and Nakamoto, 1994] by expressing equation 1.12 as the sum of two components, each of which results in a particular mechanism for resonance enhancement: KL^+B o-14> The A-term is given by equation 1.15. A-=M^r W W ( U J ) where, as above, n and m refer to the vibrational states of the ground electronic state, u denotes a particular vibrational state of the excited electronic state, T u is the bandwidth of this state, and M« is the pure electronic transition moment from the ground state to the excited electronic state. The A-term described by equation 1.15 has two Franck-Condon overlap integrals in the numerator, and hence A-term resonance enhancement is sometimes called Franck-Condon or FC resonance enhancement. This is the most commonly encountered mechanism for resonance enhancement. Since vibrational wave-functions are orthogonal, 52 both and become zero unless the vibration's equilibrium position is changed upon electronic excitation. This occurs only for totally symmetric vibrations, and thus only these are resonance enhanced via the so-called A term. Another result (which is simply stated here) is that the vibrations that distort the molecule in the same direction as electronic excitation result in the greatest FC overlap and hence the greatest A-term resonance enhancement. It is now clear how the enhancement and selectivity of RRS result. By exciting a sample with photons having energies hvo, near an electronic transition of a particular chromophore, Raman scattering from ground state vibrational modes of that chromophore whose wavefunctions produce significant Franck-Condon overlap with the excited state wavefunction will be greatly enhanced. This enhancement can be of a factor on the order of 104 more than non-resonance excitation with visible light, and factors of 108 more than excitation in the infrared at 1064 nm, a wavelength commonly used with Fourier-transform Raman instruments. Furthermore, if tunable radiation is available, this confers a significant degree of selectivity upon the technique since, by judiciously selecting excitation wavelengths to be in resonance with certain chromophores, only particular vibrational modes of a large molecule need be excited. This selectivity significantly improves the technique's resolution and ease of interpretation via the elimination of non-information-bearing overlapping peaks and resulting simplification of the spectrum. For example, the heme group of a metalloprotein such as hemoglobin may be investigated using excitation near 400 nm for the heme group, 225 nm for tryptophan and tyrosine, 240 nm for tyrosinate, and 200 nm for the secondary structural information contained in the protein 53 amide backbone modes (vide infra). Figure 1.10 shows schematically the different spectra that might occur using excitation at various points relative to peaks in the UV-VIS absorption spectrum of a fictitious molecule. Excitation in the deep ultraviolet spectral region is often required to obtain resonance enhancement in many biological and organic molecules. Ultraviolet RRS (UVRRS) is usually distinguished as a separate technique due to the uniqueness of the instrumentation required and some of the additional considerations, caveats and advantages. Moreover, excitation in the ultraviolet is accompanied by a significant reduction or elimination in the fluorescence background that degrades SNR in visible Wavelength Figure 1.10. The UV-VIS absorption spectrum of a fictitious molecule (large trace) and the Raman spectra (insets) that may be obtained for visible, non resonance conditions (A); visible or near-UV resonance conditions (B) or deep-UV resonance conditions (C). 54 Raman studies. This arises due to (a) the shifting of the fluorescence further away from the excitation line than the Raman spectrum, and (b) the higher probability of non-radiative processes occurring in the relaxation of UV-excited electronic states. Special considerations also exist relating to sample heating, saturation, and photodecomposition. The B term in equation 1.14 involves two excited electronic states, labeled e and s, and is given by equations 1.16 and 1.17 [see, e.g. Clark et ai, 1974; Nakamoto and Ferraro, 1994; Spiro and Czernuszewicz, 1995]. state. H is the electronic Hamiltonian and Q is a vibration normal coordinate. The B term will not be dealt with further here except to note that it provides a mechanism for resonance enhancement of vibrational modes that are non-totally symmetric. While the majority of biological UVRRS investigations make use of A term enhancement, B term enhancement is important for several molecules, most notably heme proteins. 1.4.5 Aromatic Amino Acid Resonance Raman Enhancement The aromatic amino acids are phenylalanine (PHE, F), tryptophan (TRP, W), tyrosine (TYR, Y), and histidine (HIS, H). The side chains of each of these molecules possess electronic transitions that allow for resonance enhancement of vibrational modes (1.16) (1.17) where a, and v, are the transition dipole moment and frequency of the s excited electronic 55 via the A and B terms mentioned above. Histidine is not resonance enhanced in the same manner or to the same extent as PHE, TRP, or TYR and is investigated considerably less frequently in the literature; therefore it will not be discussed further here. UVRRS data from the aromatic amino acids (along with peak assignments, estimations of scattering cross sections, and a discussion of enhancement mechanisms) was first reported in papers by Spiro [Rava and Spiro, 1985; Rava and Spiro, 1984; Fodor et al., 1989], Asher [Johnson et al., 1984; Asher et al., 1986], and co-workers. Saturation effects were investigated and accurate measures of scattering cross sections were subsequently obtained by Su etal. [1990]. Several groups have correlated changes in aromatic amino acid UVRRS data with environmental and conformational changes, and these are the basis of the use of the technique to examine changes in protein tertiary structure. Austin et al. [1993A] provide a very good review of this material and the references cited therein may be consulted for further detail. The side chain of phenylalanine consists of a benzene ring attached through a methylene group to the cc-carbon on the main chain, and thus its electronic and vibrational transitions (and hence its resonance Raman spectra) are derived from and similar to those of benzene. Resonance enhancement of phenylalanine arises due to the B b^, L a and Lb electronic transitions at 188 nm, 207 nm, and 255 nm, respectively. Resonance with the B,,b transition produces considerable enhancement of Raman scattering, while the L a transition produces considerably less enhancement and enhancement from the L b transition is negligible. Although changes in peak intensities have been reported to result from changes in environment polarity, phenylalanine UVRRS peaks do not exhibit noticeable 56 shifts relating to environmental change. Phenylalanine signals are not used to any extent in the work in this thesis; however one should be aware of the existence of enhancements occurring especially at wavelengths shorter than 220 nm for peaks at 1000, 1082, 1207, 1586, and 1606 cm"1. The side chain of tyrosine consists of a phenol group (OH substituted benzene ring) which is attached in the para (p) position through a methylene group to the cc-carbon on the main chain. The electronic absorption transitions occur at 193 nm (Ba>b), 223 nm (L,), and 275 nm (Lb). Like phenylalanine, the Lb transition does not provide any significant degree of enhancement, however the tyrosine L, transition provides significantly more enhancement of certain modes than does phenylalanine and is considerably more separated from the B^ b transition. The vibrational modes of the aromatic amino acids have little dependence on the main chain, and thus UVRRS data for tyrosine resemble those of p-cresol [Takeuchi, 1988; Harada, 1986]. The OH group of the phenol is mildly acidic (pKa=10.07), and its deprotonation results in red shifts of the electronic absorption maxima to 198 nm (B,,b), 240 nm (L.) and 293 nm (Lb) and changes in the relative resonance enhancement of the various peaks [Fodor et al., 1989]. Deprotonation also results in shifts in the positions of the Raman peaks which are most pronounced for the v g a (shifts from 1617 to 1601 cm"1 upon deprotonation) and v8b (shifts from 1601 to 1558 cm'1 upon deprotonation) peaks [Su etal, 1990; Fodor etal, 1989; Asher et al., 1986]. Additionally, both the V g , and v8b peaks have been reported to be linearly correlated with the hydrogen bond strength of the phenolic hydrogen [Rodgers et al., 1992]. 57 The side chain of tryptophan consists of an indole ring connected to the a-carbon of the main chain through a methylene group attached at the indole carbon 3. The indole ring consists of a five-membered pyrrole ring fused to a benzene ring at two carbons adjacent to the pyrrole nitrogen. This fusion causes a large change to both the electronic (UV-VIS) and vibrational spectrum of tryptophan compared with PHE and TYR. The pyrrole nitrogen has very little basic character (pK, of conjugate acid = -3.8) due to the fact that its non-bonded pair of electrons are required to establish aromaticity (and hence stability) of the ring. The normal modes of vibration and peak assignments have been determined [Takeuchi and Harada, 1986], and unlike PHE and TYR, the convention is to denote them with the Latin letter W, rather than v. The three prominent UV absorption peaks of tryptophan occur at 280 nm (L, overlapped with Lb), 220 nm (Bb) and 195 nm (B,). Of these, the excitation near the 220 nm transition produces the greatest degree of resonance enhancement. This results in the 760 cm" (W18), 880 cm"1 (W17), 1010 cm (W16), ca. 1350 cm"1 (W7 Fermi doublet), and 1550 cm"1 (W3) vibrational being the most prominent UVRRS spectral features. UVRRS data of tryptophan residues in proteins contain tertiary structural information from several sources, and the most useful of these will be listed here. First, the Raman shift of the W17 mode has been established as a marker of hydrogen bonding involving the nitrogen on the indole ring [Miura et al, 1989; Miura et al, 1988], shifting to lower wavenumbers upon accepting a hydrogen bond. Perhaps the most useful of the tryptophan signals for tertiary structure characterization is the partially overlapping pair of bands found near 1340 cm"1 and 1360 cm"1, the ratio of which is a sensitive indicator of the 58 polarity (hydrophilic or hydrophobic) of the tryptophan environment [Miura et al, 1988; Harada and Takeuchi, 1986]. A stronger 1360 cm"1 component is indicative of a more hydrophobic environment, with an approximately 2:1 ratio (1360:1340 cm"1) being typical for buried tryptophan and near 1:1 being more typical for exposed residues or free aqueous tryptophan. The Raman shift of the strongly enhanced 1550 cm"1 (W3) band is known to be sensitive to the torsional angle between the indole ring and the bond between the a-carbon of the main chain and the methylene carbon of the side chain. Before leaving the topic of aromatic amino acid UVRRS, three caveats should be mentioned. First, in dealing with proteins with more than one residue of the particular type being investigated, only an overall average of the signals from the various residues is obtained. Unlike NMR, there is yet no easy way to investigate, for example, one particular tryptophan residue apart from the other tryptophan residues in the protein. In cases where many of the residues are in close proximity and/or serve closely related biochemical functions, such as in the CBDcs protein that will be discussed in Chapter 6, this may be acceptable, if not desirable. Otherwise, this problem may be addressed, to some extent at least, by some judicious combination of the following: (a) careful consideration of three-dimensional protein structure and likely structural changes and interactions, (b) difference spectroscopy, (c) excitation at multiple wavelengths to take advantage of changes in RREPs upon changes in structure/environment, and (d) mutational studies. The second major caveat is the existence of saturation phenomena [Su et al, 1990], which may arise due to the inadequacies of present excitation and sample exposure hardware and result in artificially low signals for certain Raman bands. The pulsed, Q-switched, Nd: YAG 59 pumped, doubled dye laser used for most of the work in this thesis is particularly problematic in this respect due to its low duty cycle ( < 10"8) which results in high peak powers. The final caveat is the formation of phototransient and photodegradation species, which is also associated with high peak powers and low laser duty cycles. This problem is most noticeable in tyrosine spectra and can lead to spurious signals and an exacerbation of the saturation effects [see, e.g., Austin et al, 1993; Johnson et al., 1986]. 1.4.6 Peptide Backbone Resonance Raman Enhancement As discussed in section 1.2, free aqueous amino acids (except proline) exist in solution as zwitterions, with a basic amino end and an acidic carboxyl end linked by the a-carbon, to which the particular side group is attached. These form peptides through a condensation reaction that results in peptide bonds consisting of amide groups linking adjacent amino acids and separating the a-carbons. A %->%* excitation of the amide chromophore results in a broad absorption band with a maximum near 185 nm, and is associated with most of the resonance enhancement of the amide vibrational modes [see, e.g. Austin et al, 1993A; Dudik et al, 1985]. The amide vibrational modes have been determined using normal mode analysis of N-methylacetamide (NMA) [Sugawara et al, 1984; Krimm and Bandekar, 1986; Krimm, 1987] (except for amide S, which has been discussed in detail by Wang et al. [Wang et al, 1991]) and in UVRRS data consist primarily of the vibrations labeled amide I (mostly C=0 stretch, found near 1650 cm'1), amide II (mostly N-H deformation coupled with C-N stretch, found near 1560 cm"1), amide S (mostly C a -H deformation, found near 1390 cm"1), and amide III (C-N stretch 60 coupled with N-H deformation and a smaller component from C-C stretch, found near 1300 cm"1). While IR has been used for some time to investigate protein secondary structure through the amide vibrations (see section 1.3.6, above) the correlation of changes in amide Raman spectra with protein secondary structure was pioneered by Williams et al. [1983, 1981] using the amide I Raman band and visible (non-resonance) excitation. Not long after investigations of the resonance enhancement of the amide chromophore Raman signals began [Dudik et al., 1985; Mayne et ai, 1985], the groups of both Asher and Spiro pioneered the use of the technique for investigating protein secondary structure [Copeland and Spiro, 1986; Copeland and Spiro, 1987; Song etal., 1988; Song and Asher, 1989]. Several reviews of protein UVRRS have discussed the use of the amide band for secondary structure quantification in detail [e.g. Carey, 1982; Austin et al., 1993A; Austin etal, 1993B; Thamann, 1996]. Characteristic changes in the positions (Raman shifts) of the amide I and III bands can be correlated with overall secondary structural content (% a-helix, % P-sheet, % unordered), while the intensities of the amide II and amide S bands demonstrate a strong (negative) linear correlation with a-helix content. 61 1.4.7 Other Resonance Enhanced Protein Chromophores While the phenylalanine, tyrosine, tryptophan, and the amide backbone are generally the most common chromophores investigated using UVRRS, there exist several others worthy of mention. First, histidine shows resonance enhancement (albeit weaker than PHE, TYR and TRP) with RREP maxima near 204 and 218 nm [Caswell and Spiro, 1986; Hudson and Mayne, 1986]. Complexes of histidine give rise to markedly different spectra and certain bands may exhibit structural sensitivity. Being /mino rather than amino acids, the unique structure of proline residues in proteins gives rise to both a red shift in the RREP as compared with amide RREPs, and a change in vibrational modes [see, e.g., Austin et al, 1993A]. This results in a strong, structurally sensitive imide II band near 1500 cm"1 that can effectively be resonance enhanced using wavelengths shorter than 220 nm. The disufide-link chromophore exhibits weak resonance enhancement in the ultraviolet; however it is more commonly investigated using non-resonance visible excitation and its importance in UVRRS is more due to its role in laser-induced protein damage via photolysis of disulfide-links. Finally, while resonance Raman spectroscopy of the heme groups of metalloproteins is of considerable importance, this occurs in the visible or near UV, and as a consequence experimental arrangements are quite different, and will not be discussed here. The reader is referred to one of the review papers on the subject for further information [e.g. Spiro and Czernuszewicz, 1995; Wang and Van Wart, 1993; Asher, 1993A; Asher, 1993B]. 62 1.4.8 The Need for Remote UVRRS Probes The past ten to fifteen years has seen the development of the theory, calibration, and methodology of protein UVRRS to the extent that it has recently been used in a number of studies to extract useful information. For example, UVRRS has been used to investigate cytochrome c conformation under various conditions [Jordan et al, 1995; Copeland and Spiro, 1985], viral capsid assembly and folding [Tuma etal., 1996], human tumor necrosis factor receptor conformation [Tuma etal., 1995], and amino acid signals from E. coli [Britton et al, 1988], to name but a few. However, to become a more commonly applied laboratory technique in industry, academia, and clinical settings, UVRRS sample introduction techniques must be facilitated to support ease of use and higher sample throughput. The fiber-optic probes described in this thesis address this shortcoming. A more recent and exciting development has been the use of RS [Kramer et al., 1996; Frank etal., 1996; Schaeberie etal., 1996; Clary etal, 1997; Shim etal, 1997] and UVRRS [Manoharan et al, 1994; Manoharan et al, 1997; Mahadevan-Jansen et al, 1997] as tools for disease (specifically cancer) diagnosis. The sensitivity and specificity of UVRRS for the various chromophores in proteins (see above) and polynucleotides (DNA and RNA) [Mukerji et al, 1996; Wheeler et al, 1996; Mukerji et al, 1995; Takeuchi and Sasamori, 1995; Zhao etal, 1995; Fodor and Spiro, 1986; Fodor etal, 1985] make it a promising technique for these applications. The usefulness of an in situ remote fiber-optic probe in such applications is self evident. 63 Chapter 2. General System Design 2.1 Overview of System Design 2.1.1 Instrument Performance Goals and Design Requirements The intent of the research was to design, develop, analyze and characterize a system incorporating optical fibers to perform remote ultraviolet resonance Raman spectroscopy of protein molecules. If possible, the system was to be useful for investigating proteins bound to insoluble substrates as well as soluble substrates. The discussion presented in sections 1.3 and 1.4 above, along with certain practical considerations, led to the following goals for instrument performance: (PI) The instrument must be able to obtain UVRRS signals from tryptophan, tyrosine, and the protein amide backbone. (P2) Signal acquisition must be remote in the sense that the sample may be arbitrarily positioned with respect to the instrument, may undergo some motion during acquisition, and may be opaque except for a small access port. (P3) The instrument must be capable of obtaining signals in a reasonable period of time (10 minutes). (P4) Peak optical powers (or irradiances) at the sample must not be unduly large (Pma < = 1 MW/cm2). (P5) Signals must be relatively stable and experiments must be repeatable (i.e. 1/f noise must be kept small). Furthermore, the specific hardware design requirements imposed by the performance goals above and the available technology may be described as follows: 64 (Dl) A tunable (ca. 200-300 nm), narrow-band (<5 cm'1) ultraviolet light source with an average power of around 1 mW and peak powers low enough (duty cycle high enough) to satisfy point P4, above. (D2) A sampling method incorporating optical fibers and appropriate fiber-optic probe designs to (a) efficiently deliver the excitation light to a highly absorbing sample, (b) efficiently collect Raman scattered light from a highly absorbing sample, and (c) efficiently deliver the collected light to the spectrometer. (D3) A spectrometer (possibly multi-stage) with the following characteristics (a) efficient coupling from the collection optical-fiber, (b) adequate rejection of spurious Rayleigh (elastically scattered) light, (c) adequate spectrometer throughput to obtain usable signal levels at the detector, and (d) adequate dispersion and resolution to differentiate between reasonably closely spaced (ca. 30 cm"1) Raman peaks. (D4) A multi-channel detector with good sensitivity and adequately low noise in the ultraviolet region of the spectrum. 2.1.2 Hardware Overview Several intermediate systems were used in the design and development of the system and fiber-optic probes prior to their final implementation. It should also be noted that practical and financial constraints in many cases precluded the use of the "ideal" equipment available at the time. Furthermore, significant advances continue to be made in the fields of ultraviolet laser development, optical detector technology, and spectrometer design which will certainly result in the appearance of new, improved UVRRS systems 65 (which could be based upon the designs described herein) in the future. In any event, the primary element of novelty in the system design consisted of the fiber-optic probes, the first of their kind for use with UVRRS. Figure 2.1 shows an overview of the various systems used. Each system consisted of a light source (a cw Argon-ion laser, a frequency quadrupled Nd: YAG laser, or a frequency-doubled Nd: YAG-pumped dye laser), laser-to-fiber coupling optics, appropriate fiber-optic probes, an optical-fiber adapter containing fiber-to-spectrometer coupling optics, a single or triple monochomator (incorporating either 1200 G/mm or 3600 G/mm holographic diffraction gratings), and intensified photodiode array detector, and a personal computer. Q Nd:YAG Laser 266 nm B BBO Xtal Dye Laser 450 nm h-Q Nd:YAG Laser 355 nm Argon Ion Laser 472.7 nm Coupling Optics x EB— Fiber-Optic Probe Fiber-Optic Adapter Intensified Diode-Array Diode-Array Controller Analyte Solution - j * ^ 0.67m Single-Monochromator Data Acquisition & Control Figure 2.1. Overview of the laser systems used in these studies for the production of (A) 266 nm pulsed UV light, (B) 205-250 nm tunable pulsed UV light, and (C) discrete visible and near-UV laser lines from 363.8 nm to 514 nm. 66 2.2 Light Source 2.2.1 General Comments on RS and UVRRS Light Sources Prior to the development of the laser in 1960 the standard light source for RS was the 435.8 nm line of the mercury lamp (although interestingly, the very first Raman experiments used focused, filtered sunlight for excitation and the human eye for the detector). Lamps are poor choices for Raman excitation due to their broad spectral peaks and spatial incoherence. These undesirable characteristics impeded the growth of RS as a laboratory instrumental technique compared with the more rapid maturation and widespread use of LR, for which adequate light sources have existed for some time. The pulsed ruby laser was the first to demonstrate lasing action; however other lasing media with superior characteristics were subsequently developed, and lasers soon supplanted lamps in RS applications thereby allowing the technique to realize a more rapid technology growth curve. Laser is an acronym standing for Tight amplification by the stimulated emission of radiation. The central component of any laser is the gain medium which must consist of a material with at least three, and usually four (or more) energy levels. Examples of common gain media include helium/neon (HeNe) gas mixtures, neodymium doped glass or yttrium-aluminum-garnet (Nd: Glass or Nd:YAG), semiconductor junctions (diode lasers), organic dyes (dye lasers), mixtures of noble gasses and halogens (excimers), titanium doped sapphire (TiS lasers), cadmium in helium gas (HeCd), and noble gas plasmas (argon-ion and krypton-ion). The gain medium is excited (or pumped) using electrical or optical means to populate the higher energy levels which may then be stimulated by the 67 periodic electric field of passing photons to emit light at the same frequency, phase, and direction, thereby providing amplification. Placing the gain medium in an optical resonator (cavity) consisting of one highly reflective mirror and one partially (usually >90%) reflecting mirror results in the standard laser configuration. The optical resonator confers the following advantages: increase in total optical power of most lasers, decrease in bandwidth of the laser line, and improved temporal and spatial coherence. The properties of lasers that make them ideal sources for Raman excitation are as follows: (a) monochromatic emission (narrow band), (b) spatial coherence (facilitates transmission and focusing), (c) temporal coherence (allows for efficient generation of higher harmonics), and (d) high spectral brightness. Nonlinear crystals such as beta-Barium Borate (BBO), potassium dihydrohen phosphate (KDP), and lithium niobate (LiNbOs) may be used to produce higher harmonics (frequency doubling, tripling, etc.) For UVRRS, key considerations are the wavelength and duty cycle of the laser system. Production of tunable, high duty cycle and high average power narrow band UV sources is difficult, and at present there exists no ideal source for UVRRS excitation. A complete description of lasers and their operation is beyond the scope of this document, and the reader is referred to one of the many excellent references [e.g. Yariv, 1992; Milonni and Eberly, 1988; Hecht, 1992]. Discussions of the various lasers used for RS instruments can be found elsewhere [Ferraro and Nakamoto, 1994; Gerrard and Bowley, 1989]. 68 2.2.2 Light Source For Visible and Near-UV studies Some of the preliminary studies reported in Chapter 4 made use of continuous wave (cw) blue or near-UV light. Being considerably easier to work with, use of a cw source operating at a wavelength that resulted in resonant excitation of a model compound facilitated the initial development of the system and fiber-optic probes. Furthermore, it permitted early studies and modelling of inner-filtering effects experienced by the probes, which proved invaluable in modifying designs and choosing experimental parameters in later deep-UV studies. A Spectra Physics Stabilite 2017 argon-ion laser (see Figure 2.1) was used to produce visible and near-UV cw light. A prism assembly between the plasma tube and the rear high reflector could be used to choose a particular argon ion laser transition. The higher-power laser lines at 488 and 514 nm were initially used to optimize the system with non-resonant analytes (usually KNO3) and for SNR enhancement studies (Chapter 5). The deep blue laser line at 472.7 nm was used to produce resonance with methyl-orange, and the near UV line at 363.3 nm was used to produce resonance Raman signals from chromate ion. The pulsed 266 nm UV output (ca. 10 ns pulses at 10 or 20 Hz) of a quadrupled Lumonics HY-400 Nd: YAG laser was used for initial studies of pulsed UV transmission through optical fibers, and to obtain fiber-optic UVRRS data of DNA. Although a small amount of reversible fiber photosensitization was observed using this system, it had longer pulse lengths and a longer duty cycle compared with the tunable DUV system discussed in section 2.2.3, below, which therefore facilitated the use of standard-UV optical fibers. 69 This effectively made 266 nm operation a different regime, and it is therefore included in the "visible and near-UV" systems section. Nevertheless, pulsed fiber-optic UVRRS at 266 nm was an important step between visible RRS and true DUV UVRRS. 2.2.3 Light Source for DUV Studies This section describes the tunable, pulsed, narrow band light source that was assembled and used in the final stages of the research to perform fiber-optic UVRRS. The large size, low duty cycle, high peak powers, and generally awkward and unstable operation of the doubled-pulsed dye laser system make it far from an ideal light source for UVRRS, especially a fiber fiber-optic delivery system. Although a considerably superior light source was originally part of the design of the system, funding considerations precluded its incorporation in the system that was eventually realized. Chapter 7 (conclusions and future work) provides a short discussion of recommendations and future possibilities for light sources used with fiber-optic UVRRS systems. The data presented in this thesis should therefore be considered as a proof of concept or lower-limit for the quality of data obtainable using fiber-optic UVRRS. Figure 2.2 shows the general arrangement of the components used. The basic operation of the tunable, pulsed DUV system involved using the 355 nm third harmonic of the Lumonics JTY-400 Q-switched Nd: YAG to pump a Quanta-Ray PDL-1 pulsed dye laser. For our purposes, the PDL-1 could operate between 410 and 500 nm using Exalite 417 and 428, and Coumarin 420, 440, 460 and 480 dyes. The use of the coumarin dyes where possible was preferred for safety reasons as the p-dioxane solvent required for use 70 with the exalite dyes was highly toxic, mutagenic, and explosive. The output of the PDL-1 was doubled using a CSK Optronics (Los Angeles, California, U.S.A.) Super Doubler and a P-barium-borate crystal to produce pulsed light in the range from 205 to 250 nm. The laser was coupled into the excitation fiber using either a short focal length microscope objective (for 472.7 nm excitation) or a car. 10 cm focal length lens (for 266 nm and deep UV excitation). The Nd: YAG was of the oscillator/amplifier configuration. A single flashlamp pumped both the oscillator and amplifier Nd: YAG rods and Q switching was accomplished using a Pockels cell. The repetition rate (up to 20 Hz), Q-switching, and flash lamp discharge voltage were controlled with a remote unit. The Q-switched 1064 nm fundamental was rated at a maximum of 425 mJ per pulse. The fundamental was internally extra-cavity doubled using a KDP crystal housed in a thermostated oven to produce the 532 nm second harmonic. This was subsequently either doubled again to produce the 266 nm fourth harmonic for DNA UVRRS, or mixed with the fundamental to produce the 355 > £ 7 KDP-1 KDP-2 1^064 nm \ T^t^is •^flnshlnmp j l f f l >• fTij ^ amp. * Nd:YAG Laser ^ TV omp 2 Pulsed Dye Laser Freq. Doubler Figure 2.2. Detailed schematic of the DUV light system, m: mirror, G: grating, L: lens, osc: laser oscillator, amp: laser amplifier, PR: quarter wave plate polarization rotator, P: silica prism, KDP: potassium dihydrogen phosphate crystal, BBO: P-barium borate crystal. 71 nm third harmonic used to pump the PDL-1. The energies of the (Q-switched) 355 and 266 nm lines were rated at 95 and 40 mJ/pulse, respectively. These operations occurred in a second thermostated oven and the choice of doubling or mixing depended on the arrangement of the crystals therein. An internal prism assembly could be arranged to select the desired beam to exit the laser chassis and direct the undesired beam to a beam dump. The 355 nm Q-switched pulse width was measured at an oscillator voltage of 655 V using a Thorlabs (Newton, NJ) Det-Si 2 high speed silicon photodetector with a rise time of less than 1 ns and a Tektronix (Beaverton, OR) model TDS5684A 1 GHz, 5 GS/s digitizing oscilloscope and was found to be roughly Gaussian with a FWHM of 11 ns (Figure 2.3). Tek EHjjTS single Seq S.OOCS/s I T ] . . . . . . . . 1 1 1 1 1 1 • • • • i i '• • • • T " ' i i i i i —i—r—i—i— ...JJ 1—{— ™j—|— ....]——i—j—i—1.~ i~ i . . . . A . . . A N H""*rH—1~ —j—•[—1—i™ i i t H " " 1 i t i . . . . _ (_ . . yr->, . - V MfH 2.66 V M5.06nsCh2/ 1.48 V Figure 2.3. Pulse width measurement of 355 nm Nd:YAG pulse. Bottom axis divisions are 5 ns. 72 The optics in the PDL-1 required a vertically polarized pump beam, and therefore the horizontally polarized 355 nm Nd:YAG beam was rotated by 90° using a CVI Laser (Albequerque, NM) 355 nm quarter wave plate set with the optical axis at 45° to the horizontal. The PDL-1 consisted of an oscillator and preamplifier dye cells supplied by and continuously recirculated through a common oscillator dye reservoir, and an amplifier cell supplied by and continuously recirculated through the amplifier dye reservoir. The dye concentrations used were those recommended by the dye laser manufacturer (Quanta Ray) or, when these were not available (as in the case of the Exalite dyes), those recommended by the dye manufacturer. After initial alignment, the optimization procedure recommended by the manufacturer was used to adjust the positions of the dye laser optical components to produce a maximum optical power output. Dye concentration optimization was performed at one point, but was found to provide a negligible improvement in efficiency over the recommended concentrations and was therefore discontinued. Each dye provided a tuning range of between about 10 and 20 nm over which they had quoted efficiencies of between 5 and 20% (=Edye/EpUmp). It was difficult to establish a measured efficiency, owing to the inadequate damage threshold of the energy meter head that was available to measure Nd:YAG energies under Q-switched (i.e. ns pulse) conditions. However, the measurements that were done indicated an efficiency of 8% using a coumarin 460 dye (which had been in use for ca. 10 hours) operating at 460 nm and a Nd:YAG oscillator voltage of 650V. This is consistent with the 5 to 20% value mentioned above and the nominal 10% value for coumarin 460 given by Brackmann 73 [1994]. The dye laser beam was vertically polarized and consisted of a spot that was roughly circular to elliptical with a diameter of around 0.5 cm. The CSK (Culver City, CA) Super Doubler consisted of a Galilean beam-reducing telescope assembly, a BBO crystal and housing, a recollimating lens, and a prism to separate fundamental (dye laser, visible) and second harmonic (UV) beams. Manual fine and rough adjustments were available to control the phase matching angle of the crystal and thereby optimize doubling efficiency. All components were housed in a common chassis which was kept dry to protect the hygroscopic BBO crystals using plastic boats containing silica gel desiccant, which was changed periodically. The efficiency of the Super Doubler was measured to be 9.1% with 1 mJ, 460 nm dye laser pulses. Determining the width of the DUV pulses is important for several reasons, including (a) controlling saturation, (b) measuring non-linear parameters of the optical fibers used to deliver excitation energy (see Chapter 3), (c) predicting the amount of energy delivered to the sample, and (d) choosing an appropriate gate width for the photo-diode array intensifier (vide infra). A procedure identical to the one used to measure the Nd:YAG pulse width (above) was used to measure the DUV pulse width at 230 and 225 nm and it was found to be approximately 3 to 4 ns (Figure 2.4). Notably, the pulse shape was decidedly non-Gaussian and in many cases appeared to be bimodal, which resulted in difficulties in interpreting the results of Chapter 3. 74 E— . . T . -1 -»-c J _ i . JL i . . . .._ T : x •—t—i—i—i— —— i—S—i—I—|—§-—i-"^ —— --'-++++-- l - M - l - — J S J j ™t— - -s—-»s—*—i—1—i—i— 1 1 T t : T MM 2.06 V M S.oOns f i.28 V Figure 2.4. Pulse shape measurement of 230 nm (doubled 460 nm) pulse. Bottom axis divisions are 5 ns. 2.3 Fiber Light Delivery and Collection The primary objective of this thesis was the development of fiber-optic probes for UVRRS, with the ultimate goal of obtaining UVRRS data in situ. Optical fibers were conceived in the mid 1960's for optical communications applications and were first successfully realized at Corning Glass in 1970. Optical fibers usually consist of a cylindrical core material, a concentric (annular) cladding layer, and a concentric jacketing layer (see Figure 2.5). The purpose of the jacketing layer is solely to improve the mechanical properties of the fiber, and sometimes a "buffer" layer is added between cladding and jacketing layers to increase mechanical strength and chemical resistance even further. The operation of optical fibers as light guides depends upon core index of 75 refraction, ru, being higher than the cladding index of refraction, rid. Although a variety of core and cladding materials may be used in the manufacture of optical fibers, including plastics, liquids, silver halide glass, chalcogenide glass, silica glass, and fused silica, the choice of optical fiber composition depends upon the wavelength and application. For DUV work, few choices exist and we will consider only fibers composed of high purity fused silica cores, and high purity, fluorine-doped fused silica claddings. Additionally, the Figure 2.5. Expanded/exploded pictorial view of the three concentric layers of a typical optical fiber. manufacturer-controlled hydroxyl content of all of the fibers used in these studies was high (generally 500 - 1000 ppm). The cladding fluorine doping serves to depress the index of refraction relative to the core, while the high OH content is known to improve UV transmission. As will be seen in Chapter 3, long term, high intensity UV light transmission through optical fibers presents special problems which were only overcome using a novel fiber composition. Core Cladding Jacket 76 Considering geometrical (ray) optics, when nco>nC|, light can be propagated down an optical fiber through multiple, lossless reflections at the core-cladding interface by virtue of total internal reflection, provided the condition of equation 2 . 1 is met. 6, >0= sin" ( 2 . 1 ) As well, it is clear that rays coupled into an optical fiber from an external medium will only result in losslessly guided rays within the fiber if the angle between the fiber axis and the incident ray is less than Gacc, the acceptance angle, as given by equation 2 . 2 , viz. n%ineaa:=^n2co-nll^NA. ( 2 . 2 ) where NA, the numerical aperture, is a measure of the light gathering capability of the fiber; n is the index of refraction of the medium; and 8 i and 0 C are the angle of incidence and critical angle, respectively. Strictly speaking, equation 2 . 2 is only valid for meridional rays. Skew rays have larger acceptance angles, as given by equation 2 . 3 , where y is the angle made at the point of incidence on the core-cladding interface by the fiber radius and the projection of the ray onto a plane normal to the fiber axis and passing through the point of incidence. f i a n ^ = — ( 2 . 3 ) cos/ The acceptance angle defines the boundary of the conical region from which light entering a fiber may be accepted, and which light exiting a fiber illuminates. When used in fiber-optic probes, this will also define the outer boundaries of the effective sample volume. 77 While the geometrical ray optical approximation is sufficient for qualitative and semi-quantitative explanations of optical-fiber light transmission characteristics, it fails to explain more detailed aspects of fiber transmission including the intensity distribution within the fiber core and cladding. For this, Maxwell's equations must be used to obtain the wave equations for E and H, which are solved for the appropriate media properties and boundary conditions [see, e.g., Agbo etal, 1993; Yariv, 1992; Donhowe, 1994]. These will not be discussed in detail here, except to note that the solution to the wave equations result in a number of possible "guided modes", each of which produces a specific time-averaged intensity profile over the cross section of the fiber. The reader is referred to one of the cited references for more detail. The transmission loss mechanisms in optical fibers and their light collection/delivery properties are of fundamental importance in this research, however discussion of these topics is left to Chapter 3. 2.3.2 Fiber-Optic Probe design Considerations Any fiber-optic probe for optical spectroscopy consists of at least one excitation fiber and one collection fiber, although in single fiber bifurcated probes the two may be fused into a single fiber at the sample end. If multiple fibers are used, they are usually aligned in close proximity at the sample end. Often, the excitation and collection fiber end faces lie adjacent in the same plane, a geometry that will be referred to as "flush". Probes of this type were used for initial investigations (vide infra, section 4.1) to establish a "baseline" to use in comparison with later, high-performance designs (vide infra, sections 4.2-4.4). The beam is usually coupled into the excitation fiber using a lens with a 78 numerical aperture less than or equal to the fiber numerical aperture. To avoid end face damage and improve coupling efficiency, the fiber is often positioned past the point of the beam waist to ensure a diverging beam at the air/fiber interface and thereby circumvent or mitigate any self focusing effects which might reduce efficiency or cause internal damage. In this work, a Newport (Irvine, CA) model M-10X (NA=0.25, f=14.8 mm) microscope objective was used to couple cw visible (argon ion) light into the fiber, for which transmission efficiencies of up to 85% were routinely achieved. Similar efficiencies were achieved for the 363.8 nm cw UV argon line using a Newport model U-27X (NA=0.13, f=5.77 mm) microscope objective. Alignment of the fiber axis and beam directions was established by eye and optimized for power output with micrometer adjustments. For pulsed light, the short focal length of the microscope objective resulted in very high peak irradiances that caused laser-induced plasma breakdown of the air at the beam waist due to the small beam diameter, high energies and short pulse lengths. To alleviate this, a ca. 10 cm focal length lens was used for coupling in pulsed experiments. The relationship between spot size at the waist and focal length for a low-divergence Gaussian beam is given by equation 2.4, viz. w0=^- (2.4) 7TW0 where wo' is the spot size at the waist, w0 is the spot size at the lens, and f is the lens focal length. The excitation light was delivered through the excitation fiber to the sample where it irradiated a conical effective sample volume defined by the fiber numerical aperture and coupling conditions. Figure 2.6 shows the 225 nm pulse width into and out of a 1 m length 79 of fiber, indicating a negligible change in width. The collection fiber may collect some of the light scattered from the excitation volume, which may then be guided and delivered to the spectrometer. Detailed analytical, numerical, and empirical analyses of the excitation/collection process are presented in Chapter 4. 2.4 Wavelength Separation 2.4.1 General Comments on Raman Spectrometers Raman scattered light in these experiments may be treated as isotropic and the collected light in a RS experiment contains Stokes and anti-Stokes Raman components, a large Rayleigh component at the laser wavelength, possibly underlying fluorescence, and sometimes other interfering signals (e.g. phosphorescence, room light contamination, fiber fluorescence, fiber Raman scattering). The optical spectrometers used in RS serve to decode or separate the various components of the Raman scattered light while rejecting the Rayleigh light. Important considerations in discussing spectrometers include Tek EUjJQj Single Seq S.OOGS/s i [ T-TeKBHtB 5.00GS/S -t-Figure 2.6. Temporal pulse profiles at 225 nm and ca. 30 pJ/pulse for fiber input (left) and output (right). Note different time scale divisions (2 ns/division on the left and 5 ns/division on the right). 80 dispersion, resolution, throughput (efficiency), free spectral range, linearity, size, and price. Generally, the wavelengths of the desired (Raman shifted) light lie very much closer to the much larger excitation (Rayleigh) signal and require significantly higher resolution than in FS. This is especially true for UVRRS, since Raman shifts are constant in wavenumber (or frequency or energy) rather than wavelength, while the dispersion of grating spectrometers is roughly constant in wavelength. This implies that UVRRS spectrometers must have very high dispersion and resolving power relative to the small optical spectrometers often used for FS. Raman spectrometers are usually based upon reflective diffraction gratings, interferometers (Fourier transform (FT) RS), or, more recently, acousto optical tunable filters (AOTFs). Only the grating based spectrometers used in these studies (and generally common for UVRRS investigations) will be discussed here, however Chapter 7 (future work) will provide an outlook for the use of FT-UVRRS, AOTF-UVRRS, as well as prism based spectrometers. Section 2.4.2 discusses the Czerney-Turner spectrometer and fiber-coupling optics that are central to the instrumentation, while section 2.4.3 discusses further means of rejecting Rayleigh light. The references by Ferraro and Nakamoto [1994] and Gerrard and Bowley [1989] provide decent introductions to Raman spectrometers. Note that while the term spectrograph is more correct for dispersive multi-channel Raman instrumentation, the terms monochromator and spectrometer are more common and will be used here 81 2.4.2 The Czerny-Turner Grating Based Spectrometer Figure 2.7 shows the arrangement of components in and around the wavelength dispersive spectrometer. The central piece of hardware was a McPherson model 207 spectrometer (Acton, MA). The model 207 has a focal length of 0.67 meters and was of Czerny-Turner design. The gratings used were holographic "Snap-In" gratings with a ruled area of 116x136 mm (resulting in an effective aperture of f/4.7) and were either 1200 G/mm (for visible studies) or 3600 G/mm (for UV studies). The wavelength range of the system was quoted as extending to wavelengths as short as 185 nm The 1200 G/mm grating had a large scratch on it which may have significantly deteriorated the instrument's stray light rejection for visible studies. Two entrance slit assemblies were available and arranged at 90°. A double-subtractive monochromator (vide infra) was mounted on one slit assembly, while the other was available for direct attachment of the fiber-optic adapter. : FOC Figure 2.7. The Czerny-Turner spectrometer and associated hardware. ISM: single monochromator, S: bilateral slit assembly, FOC: Fiber-optic coupler, Mj: folding mirror, Mf: parabolic focusing mirror, G: holographic grating, Mc: collimating mirror, D: detector, DSM: double subtractive monochromator. The dotted lines indicate the general optical path. 82 The active entrance slit could be selected using a folding mirror. The collection fibers were coupled (f-matched) to the spectrometer using a Fiber Optic/Lens/Filter Adapter Assembly (model 132, also from McPherson). This consisted (from fiber to spectrometer) of a fiber holder, a filter holder chamber, an adjustable two-lens system, and a mounting bracket. The adapter was customized in-house to accept unconnectorized fibers mounted in chucks. 2.4.3 Rayleigh Light Rejection One of the main sources of noise in these studies was Rayleigh light which was spuriously directed to the detector at the focal plane due to imperfections in the spectrometer. While the 3600 G/mm grating exhibited superior performance over the 1200 G/mm grating in this respect, this spurious or "stray" light was especially troublesome in the experiments using suspended, insoluble particles of silica or cellulose (see Chapter 6). Rayleigh line rejection strategies for RS that have been reported in the literature include double-subtractive monochromators [e.g. Huang and Yu, 1988], holographic filters [e.g. Everall, 1992], chevron notch filters [Puppels et al, 1990], atomic vapour filters [Pelletier, 1992], crystalline colloidal arrays [Flaugh et al., 1984; Tse etal., 1995; Weissman et al, 1996], and so-called solution filters [e.g. Chou et al, 1991; Kleimeyer et al, 1996]. The lack of availability of simple, inexpensive, sharp cutoff, high efficiency Rayleigh line rejection filters continues to impede the field of UVRRS; however progress is being made in this area (see Chapter 7). 83 For the majority of DUV experiments, the single spectrometer was used without an additional Rayleigh rejection filter. Therefore, many spectra were accompanied by a significant amount of Rayleigh background, and this often proved to be the limiting noise. This proved adequate for the development and characterization of the optical fiber instrument; however adequate SNR could not be obtained in protein adsorption and binding experiments without the addition of an external Rayleigh rejection filter. A McPherson (Acton, MA) model 275DS 0.2 meter double-subtractive monochromator was obtained, and used after being sent back to the manufacturer for the installation of focal plane reversing optics. However its throughput was so low (-7%) that it did not result in SNR improvement except in a few experiments that had both very large signal and very high Rayleigh contamination. A solution filter was finally chosen to reject Rayleigh light iri protein adsorption and binding experiments. Solution filters consist of a solution of molecules (often organic) having sharp cut-offs on the red side of an absorption peak near the excitation wavelength of interest. Acenaphthene dissolved in spectroscopic grade methanol and placed in a 1 cm cylindrical cell in the filter holder chamber of the fiber-optic adapter proved to perform well for rejecting Rayleigh light in the 225 to 235 nm range without unduly attenuating the signal. The absorption spectrum of acenaphthene is shown in Figure 2.8, and the concentration could be adjusted to control the Rayleigh line extinction. 84 2.5 Optical detection 2.5.1 General Comments on Optical Detectors for UVRRS For many years, the standard detector for RS studies was the photomultiplier tube (RMT), which continues to boast nearly unmatched sensitivity, noise characteristics, and time response at a relatively low cost. The PMT has a significant drawback, however, in that it is a single-channel instrument, and from the 1980's to the early 1990's PMTs were largely supplanted in RS applications by multi-channel detectors such as the photo-diode array (PDA) and charge transfer devices (CTDs) which include both the charge coupled device (CCD) and charge injection device (CLD). The multi-channel detectors may also incorporate a multi-channel plate (MCP) image intensifier to improve gain and gating 80000 60000 \ acenaphthene § 40000 i V GO 20000 n i . i . i . i . i . i . i — I . I 224 226 228 230 232 234 236 238 240 242 wavelength (nm) Figure 2.8. The absorption spectrum of acenaphthene dissolved in HPLC-grade methanol. 85 times at the expense of resolution, noise, and cost. Several reviews of multi-channel detectors have appeared in the literature [Talmi, 1987; Sweedler etal., 1988; Epperson et al, 1988; Grossman, 1989; Gerrard and Bowley, 1989; Hanley etal, 1996; Lerner, 1996]. Talmi [1987] identified the main advantages of multi-channel detectors over PMTs as temporal, SNR, accuracy, disposition, gating, alignment, kinetics, continuous monitoring, and data processing. 2.5.2 The Intensified Photo-Diode Array and Associated Hardware The detector chosen was a proximity-focused MCP-equipped intensified PDA (IPDA, Princeton Instruments/EG&G model 1455B-700-G, Princeton, NJ). This was a linear array of 1024 individual 25 u,m diameter (centre-to-centre), 2.5 mm high elements; however only 700 elements were covered by the image intensifier and therefore "active" for spectroscopy. The MCP could be gated as fast as 5 ns or operated in cw mode and was equipped with a knob to control the voltage and therefore the gain of the MCP from approximately 0.01 to 1.5 counts per photoelectron. The dynamic range was 14 bits. The IPDA was thermoelectrically cooled to -25°C to reduce dark noise and continually flushed with dry nitrogen to prevent damage from moisture condensation or freezing. The detector MCP was gated using an EG&G FG-100 gate pulse generator capable of producing -200 Volt pulses with rise and fall times from 2.5 to 9 ns, pulse widths (FWHM) from 3.5 to 2500 ns, and a delay (relative to trigger) from 20 to 1700 ns. The detector and gate pulse generator were controlled (through an Intel 486 based PC) 86 using an EG&G model ST 120 controller which allowed for up to 1000 s of integration time and had a 33 ms array readout time. Figure 2.9 shows schematically the relationships and information flow between the detector, controller, gate pulse generator, and laser. Briefly, the Nd: YAG Pockels cell that controlled the Q-switching of the pump laser also provided a TTL trigger which was sent via the controller to the gate pulse generator. The gate pulse generator waited for the user-selected delay time which was always adjusted to optimize the signal and was usually around 60 ns, but depended upon the Nd: YAG oscillator voltage and the geometrical separation of the optical and electrical components. The gate pulse generator then sent a -ADJUSTABLE TIME DELAY GATE PULSE GENERATOR CONTROLLER COMPUTER Nd:YAG LASER DYE LASER AND FREQUENCY DOUBLER FIBER OPTIC PROBE AND EXPERIMENT/SAMPLE INTENSIFIER DETECTOR Figure 2.9. A schematic diagram of the information flow in the FO-UVRRS system. Solid lines represent electrical signals and dotted lines represent optical signals. Arrowheads indicate the direction of signal flow. 87 200 Volt pulse with a width chosen by the user to be as narrow as possible without incurring any signal loss. Pulse-to-pulse jitter in the timing proved to be surprisingly minimal. After the user-selected integration time, the controller read off the array and sent it to the PC. 2.6 Data Collection and Processing 2.6.1 Equipment Data was collected and stored as binary files using an Intel 486 based PC running the Princeton Instruments (Princeton, NJ) CSMA data acquisition software. A background spectrum was collected for each series of experiments and subtracted from each experimental spectrum to remove the systematic component of the read-out noise. The data was calibrated, processed, and plotted off-line using Matlab version 4.2 (The Math Works, Natick, MA), Microcal Origin (Northhampton, MA), and/or Borland C++ version 4.0 (Scotts Valley, CA). 2.6.2 Wavenumber Calibration Dispersive RS systems are usually calibrated by using (a) plasma lines from the cavity of the gas laser being used (argon-ion systems only), (b) lines from an external gas lamp (e.g. a neon gas lamp), or (c) the lines from a well-characterized analyte (or calibrant). Graves presents a short, readable review of calibration procedures [Graves, 1989]. Indene and tetrachloromethane are common calibrants; however neat ethanol was 88 chosen for these studies due to its relatively low toxicity, lack of UV absorption or resonances, and miscibility with water (which facilitated probe cleaning after calibration). The calibration proceeded by obtaining a ca. 30 s exposure spectrum of neat ethanol under conditions identical to those of the subsequent experiment. This spectrum was then smoothed with a five-point moving polynomial (Savitzky-Golay) filter and differentiated. Peaks in the original spectra produced zero crossings in the differentiated spectra and provided accurate measures of peak position. The peak positions (in diode numbers) were then plotted against the known Raman shifts of the peaks, and a calibration curve was obtained using a linear fit. This was then applied to the experimental spectra. Figure 2.10 shows the calibration process diagramatically. The system was calibrated before every series of spectra. In all cases, the calibration proved to be very linear and repeatable (provided that instrumental parameters Figure 2.10. Left: Top trace shows ethanol calibration spectrum (with peaks used for calibration labeled) using 30 s, 226 nm excitation and a fiber-optic probe. The bottom trace shows the derivative spectrum and arrows indicate the zero crossings used for calibration. Right: Plot of peak wavenumber vs. measured diode position for the spectrum on the left as well as one taken 1 hour later, after completing a set of experiments. In all cases the peak positions are repeatable to within 1 diode. The straight line shows the fit indicating a dispersion of 1.52 cm'Vdiode and a correlation coefficient of R=-0.99993. 89 were not changed in the interim). It was especially important to recalibrate after any changes in fiber-to-spectrometer coupling or alignment, which is likely to shift the calibration. 2.7 System Throughput It is instructive to examine the transmission of optical power through the system and the sources and possible mechanisms of loss. Studies were undertaken to measure the efficiencies of all parts of the system. In particular, chapters 3 and 4 focus in detail on the dependencies of the efficiencies of the excitation fiber and probe configuration, respectively, and provide guidelines for using this information to obtain optimal system performance. Figure 2.11 shows schematically the overall system throughput with approximations for numerical efficiencies. All of these efficiencies are moderately to highly wavelength dependent. The dye laser and doubling optics are each approximately 10% efficient. The efficiency of the excitation fiber is approximately 80% for cw, visible excitation, and considerably less (see Chapter 3) for pulsed DUV excitation due to nonlinear effects and photosensitization. The probe tip geometry efficiency (collected Raman photons/excitation photons) depends upon geometrical and physical-chemical considerations, and is discussed thoroughly in Chapter 4. The collection fiber efficiency is approximately 80%, with loss mechanisms due to Fresnel reflections and fiber Rayleigh scattering. The fiber to spectrometer coupling was measured to be 59% efficient at 488 nm because of losses from Fresnel reflections in the fiber-optic adapter and inevitably 90 imperfect f-matching. The single monochromator was found to be 25% efficient at 488 nm, primarily due to energy losses to unused diffraction orders. Nd:YAG DYE LASER (1.00) (0.1) DOUBLING OPTICS (0.01) EXCITATION FIBER (0.003) PROBE TIP GEOMETRY AND SAMPLE CONDITIONS (0.003.1,) DETECTOR (O.C0004TJ,) MONOCHROMATOR (0.00031%) SPECTROMETER COUPLER (0.00 12TV) | COLLECTION | FIBER (0-002!%) Figure 2.11. A block diagram of system throughput. The nominal figure in brackets indicates the amount of usable optical energy (as a fraction of Nd: YAG third harmonic energy) remaining after each component. The symbol TIP indicates the probe efficiency (collected Raman scattered energy/excitation energy, vide infra, section 4.3). 91 Chapter 3. Characterization of Silica Fibers for High Intensity UV Transmission 3.1 Overview of Fiber Characterization, Attenuation and Damage Mechanisms 3.1.1 Motivation for Fiber Characterization Chapters 1 and 2 demonstrated that the ability of ultraviolet resonance Raman spectroscopy (UVRRS) to determine structural, environmental and analytical information concerning biomolecules at low concentration in aqueous solution make it a powerful bioanalytical and biophysical technique. Unfortunately, its utility to date has been limited by experimental requirements which preclude in situ or in vivo studies in most cases. We have developed the first high performance fiber-optic probes suitable for long-term use in DUV-UVRRS (see Chapter 4). The probes incorporate recently developed improved ultraviolet (IUV) fibers that do not exhibit the rapid solarization and throughput decay that had previously hampered the use of optical fibers for delivering pulsed or high-intensity cw, deep-ultraviolet light. In this chapter these IUV fibers are characterized with respect to their efficacy at transmitting pulsed, DUV laser light. Potential users of RS often require that spectroscopy be performed in situ. In many cases, this precludes the use of a regular excitation/collection geometry comprising a sample cell, focusing and collection optics. This is especially true if (a) the analyte or matrix is strongly absorbing, resulting in severe inner-filter effects; (b) by necessity the desired sample volume is contained in a non-transparent enclosure such as a bioreactor, process stream, organism, measuring instrument, etc.; or (c) the sample is in a radioactive, high temperature, toxic, or otherwise experimentally harsh environment. A very successful method of overcoming many of these experimental obstacles for visible, non-resonance 92 RS has been fiber-optic excitation and collection; however this approach has not previously been successfully applied to ultraviolet resonance RS in the deep-UV (DUV) spectral region. Fiber-optic probe designs reported to date are generally unsuitable for UVRRS because of the twin problems of inner filtering and UV-induced damage to the excitation fiber. In fact, in addition to the notable lack of compact, inexpensive light sources and spectrometers for UVRRS, one of the main barriers to commercialization of UVRRS technology has been the lack of a convenient sample exposure method. In this chapter, the transmission of high intensity UV light through optical fibers will be addressed, and the following chapter will address the problem of inner filtering. 3.1.2 Introduction to Fiber-Optic DUV Light Transmission Low-intensity UV transmission by UV-grade (high OFT content) fused silica fibers with fluorine-doped cladding is limited by the Rayleigh scattering of the pure silica core. This increases as 1/X4, and dictates that such fibers can efficiently ( < 1.0 dB/m attenuation, discounting Fresnel losses) transmit very low intensity light at wavelengths as short as 200 nm. Bending losses and micro-cracks may also sometimes make important contributions to the overall low-intensity loss through a fiber. However, the use of optical fibers for the transmission of high intensity (e.g. pulsed or high power cw laser) UV light has long been plagued by a severe solarization (or photosensitization) associated with the formation of E' and NBOH colour centers in the UV-irradiated silica [Karlitschek et al., 1966B; Weeks, 1994; Fabian et al, 1991; Fabian etal, 1993; Karlitschek etal, 1995]. The result of this is a rapid, irreversible throughput decay, primarily due to broad colour 93 center absorption peaks centered around 215 and 260 nm for E' and NBOH centers, respectively (vide infra, section 3.1.5). The problem is compounded by effects such as two-photon absorption, stimulated Raman scattering, and stimulated Brillouin scattering, resulting in a rapid increase in the non-linear absorption coefficient with decreasing wavelength in the UV spectral region. As well, when high intensity pulses are used, catastrophic thermal or mechanical damage to the fiber may also be encountered. Recently, the development and characterization of improved ultraviolet (IUV) optical fibers for pulsed UV applications has been reported [Karlitschek et al, 1996A; Fabian et al, 1993; Klein etal, 1996; Karlitschek et al, 1996B; Klein etal, 1995]. Samples of these fibers were obtained from Polymicro Technologies Inc. (Phoenix, AZ) and characterized their performance in the wavelength range most important for UVRRS of biomolecules (i.e. from 205 to 250 nm). Moreover, these fibers were incorporated in the design of the first long-term-stable, high-performance UVRRS fiber-optic probes reported to date. In a previous paper [Greek et al, 1996A], the author first discussed and modeled the use of RS probes in highly absorbing samples, such as those associated with RRS and UVRRS, and the problems associated with inner-filtering (sample self-absorbance) effects when using standard (flush) fiber-optic RS probes. Although high quality UVRRS data of DNA at 266 nm were obtained, the ultimate SNR and probe lifetime were severely limited when used at wavelengths shorter than 250 nm because of limitations in excitation efficiency stemming from the fiber composition, and limitations in collection efficiency stemming from the probe geometry. In this thesis I describe a novel geometry, composition and fabrication method for a superior UVRRS probe that 94 overcomes many of these limitations and permits routine, stable, high performance, in situ fiber-optic UVRRS with sensitivities comparable to those obtained using standard UVRRS sample introduction techniques. 3.1.3 Low Intensity Loss Mechanisms Mechanisms of low intensity loss in optical fibers may be divided into two broad categories: losses at the interfaces (i.e. fiber end-faces) and internal losses. Losses at the interfaces are primarily made up of Fresnel losses and mode-mismatch losses. Fresnel losses occur due to reflections that take place at the boundaries between media of different indices of refraction (different dielectric constants). There is therefore a Fresnel loss at both the entrance and exit of the optical fiber. The ratio of the fields of reflected to incident light is known as the Fresnel reflection coefficient, r, and is different for s and p polarization; however for normal incidence rw is given by equation 3.1, viz. 90 n2+nx where n2 and ni indicate that the light is incident from a medium with refractive index ni to a medium with refractive index n2. The (normally incident) reflected and transmitted intensity (or power) ratios are given by equations 3.2 and 3.3, respectively. General expressions for these quantities for the case of non-normal angles of incidence are given in Chapter 4 in the context of fiber-optic probe simulation. The index of refraction k90° 90° (3.2) (3.3) 95 for fused silica for blue or green light (450-550 nm) is approximately 1.46; however it rises rapidly in DUV wavelengths to 1.52 at 226 nm and approximately 1.55 at 200 nm. This results in larger Fresnel losses in fused silica optical fibers used in the DUV spectral region with the total losses from Fresnel reflections at 226 nm calculated as 4.9 % when the entrance face is in air and the exit face is in an aqueous medium (index of refraction ca. 1.33), as would generally be encountered with a fiber-optic probe. This figure is made up of a 4.26 % loss at the entrance compounded with a 0.6% loss at the exit face. The relative closeness of silica and water indices of refraction result in relatively small losses at the exit face. In a study by Klein et al. [1996] using 308 nm excimer pulses, the Fresnel losses in 200 um core-diameter fibers were estimated experimentally to be 8% in air. Mode mismatch losses occur because the spatial profile of the laser beam intensity distribution does not exactly match the spatial profiles of the modes supported by the fiber, and thus not all of the laser optical energy can be coupled into guided modes of the fiber. These losses obviously depend on laser beam spatial profile and coupling conditions, and are therefore quite variable. However, the 308 nm results of Klein et al. [1996] for 200 u,m core-diameter fibers showed ca. 5% loss from mode mismatch, and this may be taken as a rough guide. The primary low intensity internal loss mechanism for most optical fibers is the intrinsic (basic) attenuation of the material. For all-silica fibers in the DUV region, this comes primarily from Rayleigh scattering out of the fiber, the cross section for which increases as 1/A.4. Another contributor to the basic attenuation at high photon energies (>6 eV, >200 nm) in the DUV region is the tail of the intrinsic absorption profile of the silica 96 material that arises due to its ca. 10 eV bandgap. Other losses occur due to OH absorption in the near and mid infrared; however these may be neglected for DUV studies. Figure 3.1 shows the low intensity basic attenuation of pure silica fibers. Other low-intensity linear losses occur due to defects in the silica structure that give rise to absorbing "colour centers". These will be discussed further in section 3.1.5 in terms of UV-induced colour centers; however it should be noted that a small concentration of colour centers may also exist owing to the fabrication process and/or fiber stoichiometry. Bending or micro-bending losses occur when the fiber is coiled or bent thereby decreasing the angle of incidence of a ray at the core-cladding interface on the outside of the bend and causing losses when this fails to exceed the critical angle. 1.4 1.2 1.0 - \ E | 0 . 8 \ s IS 0.6 5 minutes) exposure. There were no significant differences in the behaviour of SUV fiber segments from the different manufacturers. The IUV fibers do not exhibit a monotonic decay; rather, after an initial drop during the first 50-200 pulses, the throughput stabilizes, and occasionally recovers somewhat. The fractional throughput and initial decay are dependent upon the absolute energy/pulse incident upon the fiber, wavelength, fiber diameter, fiber segment length, and beam parameters (spot size and divergence). Figure 3.3 shows typical energy throughput versus time plots for 400 pm diameter, 56.5 cm long, IUV and SUV samples, and a 600 pm diameter, 56.5 cm IUV sample. This length is typical for the excitation fibers of the fiber-optic UVRRS probes that we have used for biomolecular studies (see Chapters 4 and 6). 108 30 25 20 exposure recovery (no exposure) -I - - I E 50 100 Time (s) Figure 3.3. Total throughput versus time for 56.5 cm segments of400 pm diameter SUV fiber (dotted), 400 pm diameter IUV fiber (solid), and 600 pm diameter IUV fiber (dashed). Excitation is ca. 50 pJ, 20 Hz, ca. 3 ns, 225 nm pulses. The initial (time=0) attenuation is due to Fresnel losses, mode mismatching, basic linear attenuation, and nonlinear attenuation. The additional attenuation over and above the initial attenuation is referred to as the induced attenuation, and arises solely from the formation of color centers. It is clear that, since a finite asymptotic limit of energy throughput exists, after a certain number of pulses the fiber reaches an equilibrium state with the rate of formation of color centers equal to their rate of passivation. In an experiment designed to investigate the recovery time of the IUV fibers, it was found that complete recovery occurred in a timescale of around 10 seconds for the experimental conditions used here. By comparison, Karlitschek et al. [Figure 2 in Karlitschek et al., 1995] reported time scales on the order of 10 minutes for iecovery of 109 SUV fibers exposed to pulsed (3 ns, 10 Hz) 266 nm excitation and Klein [K. -F. Klein, personal communication] reported recovery times on the order of hours for deuterium-lamp exposed IUV fibers. The deuterium lamp studies were obviously done at lower fluences and higher sensitivities than the experiments reported here, and therefore the length-specific induced attenuation and subsequent recovery were considerably smaller. It is likely, then, that the same sort of recovery behaviour also occurs in addition to the 10 second time scale recovery in these experiments; however it is masked by its smaller effect. These differences in the formation and passivation kinetics of identical color centers imply the existence of a distinctly heterogeneous population of silicate bond environments with different susceptibilities to color centre formation. This conclusion is important in the formulation of any future models to describe this process and to identify and delimit the operating regimes for these fibers. Moreover, it is clear that a detailed examination of recovery in these fibers, including fitting multiple recovery time constants, may provide important new information regarding the underlying processes that govern fiber behaviour in the DUV spectral region. 3.2.2 Catastrophic Damage Catastrophic damage was not observed using cw argon-ion radiation. Pulsed UV radiation from either source (quadrupled Nd: YAG or doubled dye laser) was found to result in catastrophic damage when pulse energies above a certain threshold were used. Relatively large pulse-to-pulse energy variations (jitter) exacerbated this situation and necessitated the use of average pulse energies significantly lower than the estimated 110 threshold for catastrophic damage. The fiber and focusing optics were adjusted to make the beam incident upon the fiber entrance face after passing through the beam waist and diverging, thereby avoiding or mitigating problems associated with an internal focal point or self focusing. Attempts were made to adjust the spot size at the entrance face to approximately 2/3 of the fiber cross-sectional area, as suggested by Beck et al. [1993]; however the DUV (<250 nm) beam spot generally had an aspect ratio of approximately 1.5 and was therefore usually arranged to completely span the fiber entrance face in the long (vertical) axis, and underfill it in the shorter (horizontal) axis. Susceptibility to catastrophic damage increased if the beam and fiber axes were skew, and so every effort was made to ensure that they were parallel. The estimated error in aligning them parallel was 1°. Three distinct modes of catastrophic damage were observed. All of these failure modes occurred near or at the entrance face, the reason for which will become clear in Figure 3.4. Micrograph of damaged 400 um (diam.) optical fiber near laser-entrance end showing linear catastrophic damage failure mode (near left side). I l l section 3.2.4. At very high energies and/or small spot sizes (compared to the entrance face area) the entrance face was pitted or fractured resulting in a nearly 100% reduction in throughput. This mode of failure could be avoided by following the guidelines listed above. The most common mode of catastrophic damage encountered was the linear damage occurring on the outside of the fiber, as first identified and investigated by Allison and coworkers [see section 3.1.4; Allison et al, 1985; Allison et al, 1987]. This consisted of a long (several hundred micron), narrow (ca. 50 micron) textured crack or crevice starting within about one centimeter from the entrance face (Figure 3.4). The appearance of such a damage feature did not reduce the throughput as severely as the entrance face damage mentioned above, but did reduce the performance significantly (typically ca. 20 -50%). This damage mechanism appeared to be highly correlated with the use of polishing rather than cleaving of the excitation fiber entrance face, and therefore after its identification, all excitation fiber entrance faces were cleaved. The final mode of catastrophic failure identified consisted of light scattering or emitting points or cracks that developed in the first few centimeters of the fiber. These had a similar effect on throughput as the linear damage mechanism mentioned above; however it is less clear why they originate. While other catastrophic damage mechanisms tend to occur almost immediately after exposure to optical radiation of suitably high intensity, these cracks or points would often not appear until after hours of nearly continuous use at a constant (or even decreasing) pulse energy. This evidence would seem to indicate that they arise from (a) low probability, randomly occurring, abnormally high energy laser pulses; (b) a low-112 probability random defect formation process in the fiber; or (c) a defect formation process with a non-linear cumulative exposure dependence. Fused silica optical fiber catastrophic damage threshold decreases from approximately 3.6 mJ at 308 nm to 0.6 mJ at 193 nm using 200 pm (diam.) fibers and 20 ns pulses [Klein et al., 1996]. However, the fused silica non-linear attenuation coefficient, 3, increases from 3.6xl0"7 at 308 nm to 2xl0"3 cm/MW at 193 nm and, as will be shown in section 3.2.3, under the experimental conditions used for these studies, the induced attenuation increases from near zero dB/m at wavelengths longer than 250 nm to ca. 7 dB/m at 215 nm (see Figure 3.8). These results imply that catastrophic damage becomes relatively less important as excitation is extended deeper into the UV. Put differently, with high-fluence DUV excitation, doubling the excitation energy may not result in a statistically significantly increased overall throughput, but may increase the probability of catastrophic damage by an order of magnitude. 3.2.3 Characterization with Respect to Energy and Wavelength The IUV samples exhibit improved UV transmission capabilities when compared with the SUV fibers. In particular, for a given (sub-damage threshold) input pulse energy (E;„), they display a constant average output pulse energy (Eout). At low values of E u , , E o ^ varies approximately linearly with input pulse energy. However, at higher TLm levels, transmission efficiency (ne = E o u t /Ein) drops off due to the non-linear effects described by the P-term in eq. 1. This ultimately limits the total throughput of a probe excitation fiber. For example, using 20 Hz, 225 nm, ca. 3 ns pulses, we have found that the maximum 113 sustainable throughput of a 56.5 cm length of 400 urn diameter fiber is around 240 u.W. Figure 3.5 shows typical Eout vs. E i „ curves for 56.5 cm lengths of 400 and 600 um core-diameter IUV fibers. The implications of these results for spectroscopy are twofold: (1) for a given fiber diameter, a limiting practical E ; , , exists above which no significant increase in E ^ is observed, but there is an increased probability of catastrophic surface damage; and (2) where not prohibited by a high sample extinction coefficient [see Greek et al., 1996A], the use of larger diameter excitation fibers is advised as the maximum Eout scales approximately linearly with cross-sectional area, as shown in Figure 3.5. Energy In (uJ) Figure 3.5. Initial (filled symbols) and steady state (open symbols) E„ut versus E„, for 400 (squares) and 600 um (triangles) diameter 56.5 IUV fiber segments. Excitation is 20 Hz, ca. 3 ns, 225 nm pulses. Figures 3.6, 3.7 and 3.8 show the results of a series of wavelength variations for a 56.5 cm length of IUV fiber using 20 Hz, ca. 3 ns pulses with energies of 50 +/- 4 uJ. This energy level was chosen for a 400 urn diameter fiber since increasing it further does not 114 significantly increase the transmitted energy, but does significantly increase the risk of catastrophic damage. Induced attenuation from the Actuv term of eq. 1 can be inferred from the difference between the initial and plateau throughput values and is nearly absent at 240 nm, but increases sharply as wavelength decreases and then stabilizes in the region from 205 - 220 nm, as would be expected from the position of the E' center absorption maximum. The initial throughput, which is dependent primarily on P in eq. 1, drops monotonically with wavelength. It is clear that, in the DUV spectral region using 50 uJ input pulse energies, significant contributions to the overall attenuation are made by both Aou and P terms. Caution is required in using these data to obtain numerical values of Aa„v and P, however, as throughput is very sensitive to temporal and spatial beam profile. These calculations and further discussion of the optical properties of IUV fibers will be reported in future work. 30 25 3 *> S 15 " J 10 5 0 200 210 220 230 240 Wavelength (nm) Figure 3.6. Initial (squares) and steady state (circles) throughput for a 56.5 cm length of IUV fiber as a function of wavelength for 20 Hz, ca. 3 ns, 50 +/- 4 pj pulses. 115 Ultimately, the appropriate figure of merit for a fiber used as a UVRRS probe excitation fiber is the average power sustainable over long periods, as data acquisition times for UVRRS can range up to an hour. In all cases, a stable energy throughput plateau was observed for IUV fibers, the value of which decreased sharply from 240 to 220 nm, and stabilized thereafter to the lower wavelength limit of 205 nm. The results show that, under these experimental conditions, a minimum of 100 u,W average power is available for spectroscopy, increasing up to 600 uW at 240 nm. The implications for fiber-optic UVRRS of these results are that UVRRS of analytes with resonances near 240 nm (e.g. tyrosinate and certain nucleotides) can be effected with much greater average power than for those with resonances around 220 nm (e.g. TYR, TRP and certain nucleotides). The 0.7 f 0.-1 03 *•*—» ' c — 0.3-O H °'2~ UL o.H 0.0- —i ' 1 • 1 • 1— 215 220 225 230 I 1 1— 235 240 X.(nm) Figure 3.7 Fractional initial throughput vs. wavelength for experimental conditions identical to those of Figure 3.6. 116 further decrease in wavelength to 205 nm needed to enhance structurally sensitive UVRRS signals of the protein backbone does nor carry with it any significantly further penalty compared with 220 nm excitation. 51 • 4-• • • m 2, 3-1 • • .4-2 'to 0.0--15 ~1 -10 ~1 10 15 Collection Angle (degrees) Figure 4.4. COL(a) vs. a for a 600 um (diam.) fiber. Squares are experimental points and dotted line is a Gaussian fit. 138 carefully determined distance, on a 4096 element photo-diode array (Reticon 1020). An appropriate correction was then applied to take into account the fact that the probe would be used in water, where n=1.33 rather than 1.00. COL(a) was determined using the methods of Zhu and Yappert [1992A] in which the test-fiber was scanned through the excitation profile of a fiber for which EX(0) had been previously determined. The plot of intensity out of the test fiber-versus angular position of the test-fiber was then normalized with respect to EX(0) of the excitation fiber to produce COL(a). Figures 4.3 and 4.4 show examples of measured EX(0) and COL(a) profiles. Analytes used. The analytes used in these studies were methyl-orange (MO), KNO3, salmon sperm DNA, and monomeric tryptophan. MO is a strong absorber around 470 nm and exhibits an associated resonance enhancement of a Raman band near 1400 cm"1 due to this electronic absorption peak [Bernstein, 1973]. KN0 3 does not absorb or exhibit a resonance at 472.7 nm; however N O 3 has a strong, sharp Raman band near 1050 cm"1. The purine and pyrimidine bases of DNA exhibit strong resonance Raman enhancement when excited at 266 nm [Fodor et al., 1985; Fodor et al., 1986]. Tryptophan exhibits a resonance when excited near 225 nm [Johnson et al, 1984; Asher et al, 1986]. Instrumentation. The RS system used in these studies was as described in Chapter 2. Briefly, excitation was effected in one of three ways, depending upon the wavelength used. For MO and KN0 3 , the blue 472.7 nm line of a Spectra Physics Stabilite 2017 Ar+ laser was used. For DNA, the pulsed 266 nm UV output (12 ns pulses at 10 Hz) of a quadrupled Lumonics HY-400 Nd:YAG laser was used. For tryptophan, the 355 nm third harmonic of the HY-400 Nd:YAG was used to pump a Quanta-Ray PDL-1 pulsed dye 139 laser operating at 450 nm using a Coumarin-460 dye, the output of which was doubled using a CSK Optronics (Los Angeles, California, U.S.A.) Super Doubler and a P-barium-borate crystal to produce pulsed 225 nm light. The laser was coupled into the excitation fiber using either a short focal length microscope objective (for 472.7 nm excitation) or a 10 cm focal length lens (for 225 nm and 266 nm excitation). For the visible experiments, a 100 ml graduated cylinder was used to hold the sample, while in the UV experiments, a 1 ml glass tube was used. In all cases, scattered light from the collection fiber was coupled into a 0.67 m McPherson 207 single monochromator using a McPherson fiber-optic adapter. Holographic gratings of 1200 G/mm and 3600 G/mm were used for visible and UV studies, respectively. Spectra were detected using a Princeton Instruments IRY-700B intensified photo-diode array, which was used in continuous and gated modes for visible and pulsed-UV excitation, respectively. 4.2.3.2 Modeling Equation 4.7 was discretized as follows COL(a)co 1 R l i i i i i i i i i i i i i I 200 400 600 800 1000 1200 1400 1600 Raman Shift (cm"1) Figure 4.9. Pulsed (10 Hz, 10 ns) UV ( 225 nm ) resonance Raman spectrum of the amino acid tryptophan using probe H. The exposure time was 10 minutes, the average power at the sample was approx. 200 pW and the slit width was 300 pm. 4.2.3.4 Flush Probe Investigation Conclusions The use of fiber-optic probes for resonance Raman spectroscopy using cw visible i ' i > i i •••"r i • ' i 1 i i ary units) - -1 >» - -Intensi " . I . I i . i i . i i -20 0 20 40 60 80 100 120 140 DNA concentration (pg/ml) 160 Figure 4.10. A partial working curve for DNA using 266 nm excitation and probe C. The 1484 cm-1 peak was used as the signal. 154 and pulsed UV excitation has been investigated. A numerical model that can reliably calculate working curves for fiber-optic probes operating in an absorbing matrix and with absorbing analytes has been formulated and validated. Several practical considerations for the design and use of fiber-optic probes for RRS and UVRRS have been discussed and applying these considerations, the first remote UVRR spectra excited and collected using fiber-optic probes have been presented. Further work remains to be done on optimizing the fiber geometry and throughput in order to improve the SNR of the spectra and to optimize the probe geometry for a given application, however the results of this work provided guidelines for achieving this goal which were used in the work presented in the following section (4.2) to obtain substantial improvements in probe performance. 4.2.3.5 Probe Tip Geometry Modifications 155 The two sample-end fiber-face modifications that we initially evaluated were: 1) polishing fibers at angles and 2) polishing lenses. By polishing 45° angles on the tips of the fibers and applying a thin film of aluminum to the bevelled surface to effect a 45° reflector, the dead space can be eliminated and 90° collection may be effected. Figure 4.11 shows a working curve for methyl orange and a novel double 200 nm fiber probe with a 45°-polished excitation fiber. One problem with this arrangement, however, is that excitation light may directly impinge upon the collection fiber resulting in a degraded SNR due to stray laser-frequency light in the spectrometer. This may be overcome to some 5000 4000 c O 3000 D ) X 2000 cu CD 1000 0 1E-7 1E-6 0.00001 0.0001 Methyl Orange Concentration (M) Figure 4.11. • - Working curve for methyl-orange ( 1400 cm-1 peak height vs. [MO] ) and a 200 um diameter, 45° angled excitation - 200 um diameter collection, fiber probe. • - height of 0.153 M N0 3 - peak vs [MO]. Inset: probe tip diagram. 156 extent by polishing a lens on the excitation fiber tip in order to collimate the excitation light. It should be noted, however, that due to the fact that the index of refraction of water is closer to that of silica than is air, the focal distance of the lens is significantly increased. In these visible RRS investigations, it was found that angled-tip-probes are extremely useful when dealing with very absorbing solutions. These preliminary investigations provided the motivation and impetus to develop the probes presented in section 4.3 (vide infra). 157 4.3 Improved Probe Design 4.3.1 Rationale for Improved Probe Design Chapter 1 discussed the ability of UVRRS to determine structural, environmental and analytical information concerning low-concentration aqueous biomolecules, which makes it a powerful bioanalytical and biophysical technique. Unfortunately, its utility has been limited by experimental requirements which preclude in situ or in vivo studies in most cases. The work from section 4.2 provided the necessary understanding of the operation of fiber-optic RS probes in highly absorbing samples, and investigations presented in Chapter 3 established the guidelines for transmitting high intensity DUV light through optical fibers. Armed with this knowledge, the first high-performance fiber-optic probes suitable for long-term use in pulsed UVRRS applications in the DUV were developed, and are presented in this section. The probes incorporate recently developed improved ultraviolet (TUV) fibers, discussed and characterized in Chapter 3, that do not exhibit the rapid solarization and throughput decay that had previously hampered the use of optical fibers for delivering pulsed, deep-ultraviolet light. A novel 90° mirrored collection geometry is used to overcome the inner filtering effects which plague flush probe geometries. Prototype probes are used to obtain pulsed UVRRS data of aromatic amino acids, proteins, and hormones, at low concentrations, with 205-240 nm pulsed excitation. Efficient probe geometries and fabrication methods are presented. The performance of the probes for examining resonance-enhanced Raman signals from absorbing chromophores is investigated and the optimal excitation wavelength is shown to be significantly red-shifted from the resonance Raman enhancement profile (RREP) 158 maximum. Generally applicable procedures for determining optimal experimental conditions are introduced. The previous section (4.2), first discussed and modelled the use of RS probes in highly absorbing samples, such as those associated with RRS and UVRRS, and the problems associated with inner-filtering (sample self-absorbance) effects when using standard (flush) fiber-optic RS probes. Although high quality UVRRS data of DNA at 266 nm were obtained, the ultimate SNR and probe lifetime were severely limited when used at wavelengths shorter than 250 nm because of limitations in excitation efficiency stemming from the fiber composition, and limitations in collection efficiency stemming from the probe geometry. The novel geometry, composition and fabrication methods described in this section result in a superior UVRRS fiber-optic probe that overcomes all of these limitations and permits routine, high-performance, in situ fiber-optic UVRRS with sensitivities comparable to those obtained using standard UVRRS sample introduction techniques. 4.3.2 Materials and Methods for Angled Probe Construction and Testing 4.3.2.1 Probe Fabrication. The new probes were fabricated using an rUV fiber for excitation and a SUV fiber for collection (see Figure 4.12). Several centimeters of jacketing material was removed from either end of both fibers, and they were cleaved and/or polished until no scratches were observable under 40x magnification. The excitation fiber was used either as prepared or with a polished micro-lens (radius of curvature ca. 1 fiber diameter). The preparation of the collection fiber involved first coating the tip with positive photoresist (S1813) and baking it at 95° for 20 minutes. The fiber tip and photoresist were then selectively polished away to leave an exposed 45° face. Polishing was accomplished using successively finer grits of emery paper (400, 600, 1200, 4000 grit), followed by successively finer grits of lapping film (5, 3, 1 and 0.3 pm). The fiber tips were rinsed with distilled water between Figure 4.12. TOP: A schematic diagram showing the sequential steps in the fabrication of a novel FO probe for UVRRS. A - IUV fiber, B - SUV fiber, C -photoresist, D - aluminum film, E - silicone rubber collar. 1 - apply photoresist to polished fibers, 2 - selectively polish away photoresist and fiber, 3 - apply aluminum film, 4 - remove photoresist, 5 - attach fibers. BOTTOM: micrograph of aluminum mirrored, angled collection fiber (Left),micrograph of completed probe tip (Right), polishing stages. They were then cleaned by immersion in a 10% (v/v) solution of ammonium hydroxide for 20 s, and blown dry with dry nitrogen. Immediately thereafter, an aluminum reflecting surface (ca. 200 - 300 nm thickness) was deposited on the exposed 160 surface using a CHA Industries Autotech II high-vacuum evaporator operating at 2 x 10"6 Torr. The mirror thickness was monitored using an Inficon 321 Film Thickness Monitor. After deposition, the remaining photoresist and unwanted aluminum on top of the photoresist were removed with warm acetone, leaving an aluminum mirror only on the polished, 45° surface of the distal (sample) end of the collection fiber. The collection and excitation fibers were aligned, fastened with thin (0.3 mm i.d.) silicone rubber collars (Baxter medical grade silicone tubing, McGraw Pk., IL), and attached using a silicone rubber adhesive (Dow Corning Silastic medical adhesive). During attachment, proper alignment of the collection.fiber with respect to the excitation fiber was determined using a 40x dissection microscope. The tips were rinsed with distilled, deionized water before and after every use. The tips were periodically cleaned with a mild detergent to help prevent long-term fouling. Both ends all fibers were periodically inspected for damage. Figure 4.13 shows several diagrammatic views of the resulting probe tip. Briefly, the excitation fiber illuminates a roughly cylindrical or conical excitation zone (volume), and the collection fiber 'views' part of this volume at approximately 90° through a 'window' consisting of the curved outside wall of the collection fiber that is adjacent to the excitation zone. Some of the collected light is redirected by the mirror to be coupled into guided modes of the collection fiber, and thereby delivered to the spectrometer. A detailed description of probe operation is presented in section 4.4.2. Figure 4.13. Four diagrammatic views of the probe tip geometry. Legend is as follows: (1): excitation fiber, (2): collection fiber, (3) and (4): silicone rubber collars, (5): distal end of excitation fiber, (6): point of maxium distal extent of collection fiber, (7): polished lens, (8): aluminum mirror, (9): effective 'window' for 'viewing', (10) and (11): centre lines of collection and excitation fibers, respectively, (12), (13) and (14): general optical path, (15): field of illumination, (16): approximate field of view, (17), (18), (19), (20) and (21): optical rays representing the probe operation. 162 4.3.2.2 UVRRS Data. The pulsed UV light source, monochromator and detector system have been described previously (Chapter 2; figs. 2.1, 2.2, and 2.6; and section 4.2.3.1). The only modification to the system was the use of Coumarin 440, 460, and 480; Exalite 428; and Stilbene 420 laser dyes in order to cover the wavelength range from 205.5 nm to 240 nm. 4.3.3 Improved Probe Design and Characterization 4.3.3.1 Probe Design Considerations. The design philosophy used to overcome inner filtering effects and improve overall probe efficiency was to attempt to maximize overlap between the theoretical excitation and collection volumes, while minimizing the average distance between the excited volume and the collection optic, and keeping the fiber axes parallel to achieve a small fiber tip size. The most common design of fiber-optic visible RS probes at present are flush probes, in which the fiber end faces lie nearly adjacent and in the same geometric plane; however flush probes do a poor job of addressing the aforementioned design issues in that the excitation and collection cones do not overlap to any great extent near the probe tip (see, e.g., Figures 4.1 and 4.2). Significant overlap is not achieved for these probes until the distance from a given volume element to the excitation or collection fiber face is such that the analyte absorbance seriously begins to attenuate both the excitation and scattered light [Greek et al., 1996A; Greek et al., 1996B]. Previously, two approaches have been investigated to improve upon the flush design - aligning the excitation and collection fiber axes at some angle [Myrick etal., 1990; McLachlan etal., 1986] and, more recently, 163 beveling the probe tip to effect a refraction-induced angular difference between the axes of the fiber and its excitation (or collection) cone [Cooney et al., 1996A; Cooney et al., 1996B]. In addition to practical difficulties in alignment, the former approach leads either to small inter-axial angles between cones (resulting in both limited overlap and long average path lengths of scattered and collected light rays), or, if large angles are used, a impractically large and awkward probe tip. The latter approach does provide for a compact probe tip and, unlike the design presented here, works very well with solid samples; however the degree of overlap is governed by the bevel angle and this is limited due to losses of both excitation and scattered light stemming from Fresnel reflections, which increase with increasing bevel angle. Further, since the index of refraction of water is much closer to silica than is that of air, the cone overlap is not as good in aqueous solution as in air. The design introduced here and described above makes use of a 90° collection geometry. This ensures a very high degree of overlap between the excitation and collection cones, especially in the critical area near the probe tip itself. Because the excitation light is within the volume of collection and in close proximity to the collection optic immediately after exiting the excitation fiber, a larger fraction of the scattered light can be collected and inner filtering effects in absorbing samples are mitigated. Although the total overlap between excitation and collection cones is greater for a flush-probe design when considered over all space, most of this overlap region occurs quite far from the probe tip, and not in the critical volume near the probe tip. The design was further improved by polishing a lens on the excitation-fiber sample-end-tip and polishing three 164 facets on the reflective collection surface. It was reasoned that these modifications concentrated excitation and collection along the axis of the excitation fiber, thereby increasing the effective numerical aperture of the probe. Additionally, the lens helped prevent excitation light from being directly incident upon the collection optic, thereby reducing noise due to stray light in the spectrometer. An additional advantage of the 90° collection geometry is that it should minimize collected Rayleigh scattered light, which demontrates a l+cos2(p dependence, where cp is the angle between the incident and Rayleigh scattered light. 4.3.3.2 UVRRS Data with Improved Probe We used the 1454 cm'1 vibration (in-plane HCH deformation) of neat ethanol as a benchmark non-absorbing analyte and calibrant, and the ca. 1010 cm'1 W16 line of tryptophan in 200 pg/ml lysozyme as a benchmark absorbing analyte. Figure 4.14 shows spectra obtained using probes of several flush and angled designs. The new design demonstrates superior performance. 165 I i i i i I i i i i i i i i i i i i i 600 800 1000 1200 1400 1600 600 800 10001200140016001800 Raman Shift (cm1) Figure 4.14. (a) DUV RS data of neat ethanol using (A) flush probe (400 um diameter excitation fiber/600 um diameter collection fiber), (B) angled and mirrored probe (400 um diameter excitation fiber/600 um diameter collection fiber), and (C) angled, mirrored, lensed and faceted probe (600 urn diameter excitation fiber, 600 um diameter collection fiber). Average power into excitation fibers was 1 mW; using 20 Hz, ca. 3 ns, 225 nm pulses; integration time = 30 s. (b) UVRRS data of 200 ug/ml hen egg-white lysozyme. Probe designations are the same as in Fig. 4.14 (a). Average power into excitation fibers was 1.2 mW; using 20 Hz, ca. 3 ns, 230 nm pulses; integration time = 90 s. Sloping backgrounds have been subtracted and spectra have been vertically translated for clarity. In the final analysis, the efficacy of the probe must be judged by the quality of data that may be obtained using it, and by the ease with which it is obtained. We have found that the angled/mirrored (A/M) probes are exceptionally reliable and easy to use. The probe may be arbitrarily positioned; relative motion between the sample, laser, and spectrometer is not a problem; and sample volumes as small as 10 uL have been investigated. Figure 4.15 shows example spectra of cellulose binding domain protein (CBD, vide infra, Chapter 6), tryptophan (TRP), testosterone (TSTN), and poly-l-lysine 166 Raman Shift (cm"1) Figure 4.15. FO-UVRRS data of aqueous cellulose binding domain protein (CBD, 200 pg/ml, AexC=227 nm, 18 minutes integration) , aqueous tryptophan (TRP, 100 pM, Xexc=r227 nm, 5 minutes integration), aqueous poly-l-lysine (PLL, 150 pg/ml, X^^IOS nm, 10 minutes integration) and testosterone dissolved in ethanol (TSTN, 0.5 mM, A«CC =245 nm, 4.5 minutes integration). Asterixes indicate solvent peaks. All data was obtained using 20 Hz, ca. 3 ns pulses. Sloping backgrounds and ca. 1640 cm"1 water peak have been subtracted, and spectra have been vertically translated for clarity. Spectra are not to scale. (PLL) obtained at wavelengths of 245, 227 or 208 nm. Clearly visible in the 227 nm CBD spectrum are the ca. 1555 (W3), 1350 (W7), 878 (W17) and 760 cm"1 (W18) TRP modes, which are known to be sensitive to tertiary structure [see, e.g., Austin etal., 1993A] The 208 nm PLL spectrum clearly shows the amide (A) I, II, III, and S bands that exhibit sensitivity to secondary structure [Copeland and Spiro, 1987]. In addition, high-quality UVRRS data of the catecholamine neurotransmitters serotonin, dopamine, epinephrine and norepinephrine have been obtained [data not shown, refer to Schulze et al., 1997B]. 167 J i • • i 10 100 1000 Tryptophan Concentration (uM) Figure 4.16. Circles: working curve for tryptophan FO-UVRRS using 1010 cm"1 peak height. Squares: ca. 1640 cm"1 water peak height as a function of tryptophan concentration. Excitation was at 227 nm and the probe was of the type shown in Fig. 4.12 with excitation and collection fiber diameters 400 um and 600 um, respectively. Figure 4.16 shows plots of the intensity (peak height) of the W16 line from aqueous TRP (resonance enhanced using 227 nm excitation) and of the ca. 1640 cm"1 water bending mode as a function of TRP concentration. These data were obtained using an A/M design with excitation and collection fiber diameters of 400 um and 600 urn, respectively. The increasing concentration of the absorbing analyte causes a monotonic decrease in the signal from water, which is at a constant 55.5 M concentration and could essentially be viewed as an internal standard. With this probe and experimental conditions, the TRP signal exhibits a maximum at an optimal concentration of approximately 80 uM. This value depends upon probe geometry, the presence of other absorbing species, and excitation wavelength, as will be discussed in section 4.3.4, below. 168 4.3.4 Optimal Excitation Wavelength. A problem often confronted in these studies is to determine the optimal excitation wavelength for fiber-optic UVRRS. Early on, it was found that simply choosing an excitation wavelength corresponding to the maximum of the resonance-enhanced scattering cross-section did not result in the optimum signal. The total amount of Raman signal detected by a particular fiber-optic UVRRS system depends upon both the wavelength and analyte concentration, ca, and can be expressed as the product of the eight terms in equation 4.9, viz. S(X,c) = /IU(XKO where L is the laser power in pW; t is the integration time in seconds; r\e, r\s and r\m are the efficiencies (unitless) of excitation fiber, collection fiber and monochromator, respectively; o is the Raman scattering cross section of the vibrational mode of interest; and rjd is the sensitivity (counts/photon) of the detector. We define the normalized probe efficiency, r)p, as the number of Raman scattered photons collected by the probe from a 1 pM solution of a perfectly non-absorbing analyte with a Raman cross section of 1 mbarn/sr (if c and a are given in pM and mbarn/sr, respectively), using 1 pJ of excitation. The wavelength dependence of rjp stems from the wavelength dependence of the molar absorptivities of the various species in the sample, any increase in which exacerbates the inner-filtering problems and reduces the total amount of collected light. This definition gives the quantity Ltriecorip the interpretation of "the number of photons collected by the collection fiber". 169 Since, r|m, r\d, and L are only weakly dependent on wavelength, for practical purposes these terms, time, and the concentration can be combined in a constant coefficient, Kc, to yield the wavelength dependent expression for the collected Raman signal from a given probe and analyte at a fixed concentration, viz. Se(X) = ^ 7.(^W^W*7. W (41°) where the additional subscript, c, denotes a fixed concentration. If approximations to just the shapes of the four wavelength-dependent terms are available, it is a simple matter to determine the optimal wavelength through numerical or graphical methods. A rough approximation for r|e with a typical laser energy of 50 uJ/pulse is easily obtained by fitting the data from Figure 3.6 with a sigmoidal function. The plot of a as a function of wavelength is known as the resonance Raman enhancement profile (RREP). Although the RREP is usually described using the terms introduced by Albrecht and coworkers [e.g. Albrecht and Hutley, 1971], a Gaussian fit to literature values provides sufficient accuracy for these purposes. The length-specific low intensity linear attenuation is proportional to l/X4 and therefore r|s is approximated well by fitting the literature values to such a function and using this to calculate the wavelength-dependent fractional transmission of the known length of collection fiber. Approximating r\p is not as straightforward, but can be accomplished by considering how the signal from a fixed concentration of a non-absorbing analyte decreases as the medium absorbance is increased; such plots were modelled in the previous section for flush probe geometries. If water is now considered to be the analyte, and it is recognized that TRP concentration is directly proportional to medium 170 absorbance, then the trace for the 1640 cm'1 water peak in Figure 4.16 is an example of such a plot. If it is scaled by taking into account the transmission of the collection fiber, the spectrometer throughput and the quantum efficiency of the detector, then this trace represents the number of photons Raman scattered by water that are actually collected at a given TRP concentration. Since this function depends only on the sample absorbance, the abscissa may be replaced by absorbance (in length"1 units) by multiplying TRP concentration by its molar absorptivity at the given excitation wavelength, giving a plot that is specific to the probe only, and independent of the particular absorbing species involved. Information for the specific analyte/medium combination may then be introduced though e(k) for the specific analyte of interest. Finally, since the analyte (water in this case) is non-absorbing, this trace can be scaled linearly (i.e. maintaining its shape) to give the curve for any non-absorbing analyte with arbitrary a and c. Specifically, it can be scaled to a=l mbarn, c=l pM, and an incident energy of 1 pJ, in which case the trace becomes the number of Raman scattered photons collected from a 1 pM solution of a perfectly non-absorbing analyte having a Raman cross section of 1 mbarn in an absorbing medium described by e(X) using 1 pJ of excitation energy. This is exactly the definition for jjp given above. Therefore, if the signal from a non-absorbing analyte of concentration c„a and scattering cross section a™ in the presence of an absorbing species with concentration ca and molar absorptivity ea is measured to be S„a(ca), then the signal from the same non-absorbing analyte as a function of medium absorbance (extinction coefficient), a™ (in length"1 units), is given by equation 4.11, viz. Sna{<*m) = Sna(caza) (4.11) 171 This measured result may then be used to find the normalized probe efficiency using equation 4.12, viz.. Although this definition states that the analyte is non-absorbing, the inner-filtering effects of the analyte are taken into account by considering its absorbance to be that of the medium. Therefore, if the abscissa is transformed to wavelength using an approximation for e(k) (widely available from literature and reference data), then the shape of the plot is the same as that of np. The 1640 cm"1 peak-height trace in Fig. 4.16 fits well to an exponential decay as a function of absorbance, and absorbance (or extinction coefficient, in m"1 units) is easily obtained by multiplying the (fixed) TRP concentration by an approximation to the literature or measured values of e at the various wavelengths. It should be noted here that any absorbing analyte of known e(k) could have been used in place of TRP, as r\p(X,c) is characteristic of the probe only, and independent of the analyte. Since equation 4.12 shows how to use plots such as those in Figure 4.16 to obtain the normalized probe efficiency as a function of medium extinction coefficient, a„„ and am can be found from equation 4.13, viz. where the summation is taken over all of the species present in the sample, then t|p(A,,c) for a particular sample may be found using equation 4.14, viz. (4.12) a, = Z<*s,(A.) (4.13) m (4.14) na 172 Eq. ! Term Approximation Fitted Parameters Data Source (4.15) T]e=(Ae-Be)/(l+exp((X-Xe)/we))+Be A^O.0825, Be=0.595, j Fig. 3.6 Xe=231.4 nm, we=3.24 nm (4.16) I 8 e=(0.798*Ae/we) exp (-2*(X-Kf/^s2) + Be AZ = 5.08xi66 nmMTW 1 1 Data B 8=1.73xl0 4M- 1cm- 1, | not ^ 218.6 nm, we = 17.7 nm j shown (4.17) | % T|p=ApeVc Ap= 3.34x10^ hvmbarn"1 | Fig. 4.16 pJVT1 pJ"1, kp= 2.90 cm, c = 50 pg/ml = 3.4 pM (4.18) | 0"wi6 CJwie = (0.798*A • ' ' • • • Tip . - * . *' v . • ~ \, \ x \ s c M 0.0 i I . I . I i . i 220 225 230 235 Wavelength (nm) 240 245 Figure 4.17. Sc (normalized to 1) and component terms (normalized to 0.75 for clarity) for the ca. 1010 cm"1 tryptophan residue signal in 50 ug/ml hen egg-white lysozyme using a 400 um (excitation fiber diameter), 600 urn (collection fiber diameter) A/M probe. X indicates experimental point. See text for detail. 175 the degradation of a 400 urn diameter fiber used an average of 2 hours per day is a 25% loss over 100 days; however further work is underway to more accurately characterize this effect and its dependency on exposure. Other factors affecting probe lifetime are protein fouling, especially serious when conducting in vivo experiments, and mirror damage. Damage to the aluminum surface can occur over the long term (weeks or months) through a slow desorption process, or rapidly (minutes or hours) if exposed to certain harsh chemical environments such as extremes of pH. These effects can be mitigated by applying a silicone or epoxy cap to the mirrored surface. Overall, useful probe lifetimes of greater than four months have been achieved, and we continue working to improve this figure. 4.3.6 Conclusions from Improved Probe Design Work It has been demonstrated that fiber-optic probes can be used to obtain UVRRS data at wavelengths shorter than 250 nm. Excitation light can be delivered efficiently using IUV fibers that have recently become available. High energy throughputs (> 20 pj/pulse), throughput efficiencies (>10 %), and long probe lifetimes (>100 hours) have been achieved by incorporating these fibers into UVRRS probes. Factors that reduce the energy throughput efficiency include increasing fiber length, decreasing fiber diameter, increasing input energy, and the use of wavelengths close to 215 nm. Collection efficiency has been increased and inner-filtering effects encountered in absorbing biological media have been addressed by using a novel technique to fabricate probes using angled mirrors to effect collection at 90°, thereby mitigating the deleterious effects of sample and medium 176 absorbance, while at the same time increasing the effective numerical aperture of the collection fiber and therefore further increasing the collection efficiency. The efficacy of these probes has been validated by using them to obtain remotely-collected UVRRS data of proteins, hormones, and free aromatic amino acids, the quality of which is comparable to spectra from the literature collected using instrumentation that incorporated traditional sample introduction techniques. A working curve of tryptophan indicates that the operating range of the probe is at least from 10 pM to more than 1 mM for that biologically significant analyte. The optimal excitation wavelength can be calculated for any analyte of interest by making approximations for the wavelength-dependent efficiencies and RREP, and maximizing their product. The optimal wavelength is found to be significantly red-shifted from the RREP maximum. This analysis is generally applicable to fiber-optic scattering or emission spectroscopy in the DUV spectral region and/or with absorbing analytes. 177 4.4 Theoretical and Geometrical Analytical Limits of UVRRS 4.4.1 Overview of Theoretical and Geometrical Limit Investigations 4.4.1.1 Summary of Analytical Limit Investigations UVRRS is a sensitive, specific, and versatile technique for bioanalytical and biophysical investigations. Section 4.3 reported the first system capable of fiber-optic UVRRS. In this section the theory to determine the theoretical maximum number of Raman scattered photons, maximum SNR, and minimum detection limit for resonance enhanced Raman spectroscopy is presented. Although these considerations are discussed in the context of FO-UVRRS, they are also more generally valid. Further, a simulation of the angled probes previously introduced is presented and used to characterize their performance in terms of collected signal intensity and spatial distribution of collected Raman photons. The theoretical limits of detection are calculated to be on the order of 10" 1 1 M or less for aromatic amino acids. However, it is shown that in practice it will be difficult to achieve limits of detection less than 10"9 M. 4.4.1.2 Motivation for Analytical Limit Investigations The author's interests are in developing FO-UVRRS instrumentation in order to make the technique more amenable to applications in biotechnology. To this end, the author recently described and demonstrated the first effective FO-UVRRS probes for deep-UV application [Greek et al, 1997A; Greek et al, 1997B; Greek et al, 1996B; Schulze et al, 1997B]. As discussed in section 4.3, these probes incorporate a 90° collection geometry and achieved micromolar detection limits for the aromatic amino acids 178 (resonance enhanced with excitation near 230 nm) using integration times on the order of minutes, with hundreds of micro-Watts of optical power delivered to the sample. In the course of considering design improvements, the problem arose of determining the ultimate limits of these improvements. In this sense, there are two ways of defining the limits. Theoretical limits are imposed by the physical limits of the technique which are in turn defined by the sample properties of scattering cross section and absorbance (and possibly saturation and sample damage); while practical limits are imposed by current technology, in this case the collection geometry of the FO-UVRRS probes is considered. In earlier work [section 4.2; Greek et al., 1996A; Greek et al., 1996B] the excitation and collection process of standard (flush) fiber probes for use in visible resonance Raman spectroscopy was simulated. In this section, expressions are derived for the theoretical analytical limits of UVRRS, and the operation and practical limits of FO-UVRRS are investigated using computer simulations of the improved probe design geometry. While this analysis is formulated to consider the analytical limits of UVRRS, these are directly linked to the technique's limits as a tool for biophysical investigations. For example, we are interested in using FO-UVRRS to investigate changes in protein structure upon adsorption to suspended particles. Ultimately, the lowest surface coverage that we can investigate is defined by the analytical limits discussed here. It is imperative to understand these limits and use them to determine the feasibility of a proposed UVRRS project if limits of detection or quantification are in question. 179 4.4.2 Theoretical Limits 4.4.2.1 General Comments on the Theoretical Limits of UVRRS. The theoretical limit of detection comes from collecting every possible Raman scattered photon from the sample volume, delivering them with 100% efficiency to a perfectly efficient spectrometer incorporating a detector with 100% quantum efficiency and zero dark current and read noise. This must all be done with 100% rejection of Rayleigh scattering, underlying fluorescence, and spectrally overlapping Raman scattering from solvents and other interfering species. While none of these idealities can ever be achieved in practice, recent and anticipated improvements in detector, spectrometer, and filter technology make many of the assumptions more realistic. For example, quantum efficiencies of high end state-of-the-art CCD arrays can be as high as 80% in the DUV region with read noise values as low as 3 e" rms and dark currents of less than 1 eThour (for L N 2 cooled detectors). Newly available filters for the DUV can have pass-band efficiencies of well over 50% with optical densities of more than 3 in the stop-band and transition-band widths of only a few nm. The selectivity and lack of fluorescence inherent in UVRRS can reduce interfering fluorescence and solvent Raman signals to negligible levels. Therefore, the challenge in the future will be to collect as many of the Raman scattered photons as possible (to approach the idealized 100% collection assumption) by using an optimal geometry. Theoretical aspects of this challenge are addressed in this section. 180 4.4.2.2 Absorbing analytes in non-absorbing media The total number of photons (N„) Raman scattered into 47t steradians from a Raman-active vibrational mode with a differential Raman scattering cross section of a (in units of area/scatterersteradian), using a laser pulse of transverse cross sectional area A with N; photons illuminating a volume containing N m scattering centres, is given by equation 4.20, below, N, = 4K^-Nma (4.20) A It is more common and convenient to use the concentration, Cm, and the sample volume, V, rather than N m , and these may be introduced by making the substitution Nm=CmV, viz., Ns = 4n^-cja (4.21) A This assumes that the intensity of the beam does not change significantly in the volume under consideration, i.e. that the analyte and medium are non-absorbing, or that the path length of the laser through the sample volume is short enough such that the absorption of incident photons is negligible. For any real analyte (with a finite molar absorptivity) it is necessary to calculate the total number of Raman scattered photons by integrating the differential number of scattered photons from a differential volume element through which Ni/A is approximately constant. If the laser pulse is travelling in the z direction and has an energy of E and a wavelength of A,, then the following substitutions (equations 4.22-4.25) may be made. dNs = 4n^-cmadV (4.22) A 181 V = Az (4.23) dV = Adz (4.24) Nt=—=— (4.25) hv he This results in equation 4.26, which expresses dN8 in terms of the commonly measured experimental parameters of energy, wavelength, and concentration. EX dNs = An cjjdz (4.26) he Since the analyte is absorbing, E (and hence N/A) is a function of z. If the sample has a specific absorptivity of 8 and is assumed to exist in solution in the semi-infinite half-space defined by z>0, then the energy at any z is given by equation 4.27, which, when substituted into equation 4.26, results in equations 4.28 and 4.29. £(z) = ^ 10"*"r (4.27) = 4 ^ V ] 0 - V i ; ( 4 2 g ) he = 4xE0Acmo-e_Hl0)eCm2dz ( 4 2 9 ) he Where Eo is the energy at incidence (z=0). The total number of Raman scattered photons produced by the pulse from its initial incidence to any value of z is given by the integral in equation 4.30 which may be evaluated to give N,-7-,,=o in equation 4.31, where the subscripted notation a=0 denotes the fact that the absorbance of the medium is zero; that 182 is, the only absorbing centres in the sample are the analyte molecules themselves. An evaluation of N, for samples with a finite medium absorbance will follow. Ns z a=0 = ]dNs = W ^ f g ( - ln f lO^r) ^ ^ 0 h c 0 r W ^ Y i _ e ( - I n 0 ° ) M l (4 31) 'jcelnfiom J v ' ' v/jceln(lO)^ We are interested in collecting the maximum possible number of Raman scattered photons from the sample. This number cannot exceed the total number of Raman scattered photons produced, which is equal to equation 4.30 evaluated in the limit that z approaches infinity, and is given as equation 4.32: N ^ Q t e ( 4.3 2 ) Equation 4.32 provides somewhat of a "gold standard" for analytical UVRRS (fiber-optic or otherwise) in that this is the maximum signal that can ever be generated or collectedfrom all space. It is futile to strive to collect any more photons than this number. As intuitively expected, the total number of Raman scattered photons is seen to be directly dependent upon the scattering cross section and inversely dependent upon the specific absorptivity. That is, as the Raman scattering cross section increases with respect to the absorptivity, incident photons have a higher probability of interacting through a Raman scattering process rather than absorption. It is interesting to note that, ceteris paribus, N„ is also directly dependent on the wavelength of excitation. This is a consequence of the fact that longer wavelength photons have lower energies, and so for a given energy, longer 183 wavelength pulses will have a greater number of photons. In practice, this dependence on wavelength is not observed due to the much greater wavelength dependence of a, which increases approximately as 1/V under non-resonance conditions, and as |1/((1A<.2 -l/V)+ir)l near resonance conditions [see, e.g., equation 1.15; Austin etal., 1993; Ferraro and Nakamoto, 1994; Albrecht and Hutley, 1971] (where hc/Xr is the energy of the electronic transition associated with the resonance under consideration). What is perhaps more surprising is the fact that equation 4.32 shows that Ng,*,,,^ is completely independent of the concentration of scatterers. While this seems counter-intuitive at first, it is easily rationalized by considering Nsaoa=Q, the number of scattered photons per incident photon. This is a more useful basis of comparison of different analytes when the excitation conditions are identical. This may be obtained by dividing eq. 4.32 by the photon energy, and results in equation 4.33. An o~ . ..a , . ,^,00,0 = 7 - 7 - ^ - = 5.46- (4.33) ln(10) s s This shows that Nsaoa=0, the excitation photon-specific total Raman scattering, is simply dependent upon the ratio of the differential Raman scattering cross section of the Raman-active vibrational mode in question (in m2 molecule"1 sr'1), to the absorptivity of the molecule (in l/(molecules m"2). For UVRRS, the scattering cross-section and molar absorptivity are highly correlated, since the physical origin of both the cross-section resonance enhancement and the absorption of UV light are associated with specific electronic transitions. Specifically, both the molar absorptivity and the A-term (Franck-Condon overlap enhancement) have a direct dependence on the dipole transition moment 184 of the electronic transition. Thus, as a increases due to resonance enhancement, so too will e usually increase. The ratio of a to 8 will therefore not increase at the same rate as either o increases or e decreases. For example, the Raman scattering cross-section of the ca. 1010 cm'1 W16 vibration of tryptophan increases from 25 mbarn molecule"1 sr"1 at 240 nm to 1908 mbarn molecule"1 sr"1 at 223 nm due to a resonance enhancement associated with the Bb electronic transition at 218 nm. The molar absorptivity also increases from ca. 2000 M"1 cm"1 at 240 nm to ca. 37 500 M"1 cm'1 at 223 nm due primarily to the same electronic transition. These are increases in o and 8 by factors of 76 and 19, respectively, in going from 240 nm to 223 nm. However the ratio o7e, and therefore the total Raman scattering, changes by only a factor of 4.1 [all cross sections from Su et al. and Austin et al., 1993]. Put differently, although the scattering cross section increases by nearly two orders of magnitude, the total number of Raman scattered photons increases by significantly less than an order of magnitude upon tuning the excitation to the RREP optimum (at ca. 220 nm) from 240 nm. While equation 4.33 may seem to indicate that it may be possible to have more Raman scattered photons than incident photons (for analytes with very low absorptivities and high Raman scattering probabilities) intuitively this is impossible and is, in fact, not the case. In the limit of an analyte that does not "absorb" at all in the usual sense (i.e. via photon-induced electronic transitions), but only scatters, e will be small, but finite, and determined by the scattering itself. Thus, in the limit that the probability for all mechanisms of photon/analyte interaction except Raman (or Rayleigh) scattering approach zero, Nsaoa=0 approaches unity. Although the total amount of Raman scattering produced by 185 an absorbing analyte in a non-absorbing medium throughout the entire semi-infinite space under consideration is independent of the concentration, the amount of Raman scattering from any finite section of the volume through which the laser pulse travels is quite dependent on concentration. The form of equation 4.31 shows, as would be intuitively expected, that higher concentration result in a larger fraction of the Raman scattering occurring proximal to the point of incidence of the laser pulse and the medium. The above analysis may be used to calculate the minimum (or optimum) theoretical limit of detection (LODop t ) , which corresponds to the lowest concentration at which the SNR is equal to 3. When considering equation 4.32, however, we find that the LODopt calculation is meaningless, since, for a semi-infinite volume, the number of Raman scattered photons is independent of the concentration of analytes (for zero medium absorbance). In practice, however, it is usual (and possible) only to collect scattered photons from a small, well defined volume having an axial extent ranging from several microns (in Raman microscopy, for example) to several millimeters. For the case of the improved fiber-optic UVRRS probes presented in section 4.3 and discussed later in this section, the effective penetration length (the length over which photons can possibly be collected by the collection optic) of this volume in the z-direction is very close to the diameter of the optical fiber used as the collection optic (vide infra). The diameter of the multimode optical fibers that we normally use as collection fibers is ca. 600 um, and the smallest fibers that can practically be used are ca. 100 um in diameter. Thus these distances will be used as reference sample volume penetrations, although the approach is generally applicable to any penetration. The amount of signal collected from such a 186 1.0 § 0.8 5 H 0.6 Q_ cr H 0.4 ractiori o L L 0.0 ( /->:-.-:--.'-' r 7 i . 1 • i . i ) 100 200 300 400 500 Concentration (ug/ml) Figure 4.18. Calculated fraction of total Raman scattering that occurs in the probe volume for lysozyme excited at 230 nm (dotted and dotted-dashed lines: 600 um and 100 um probe volume depths, respectively) and 206 nm (solid and dashed lines: 600 um and 100 um probe volume depths, respectively) volume is given by equation 4.31. It is intuitively obvious and clear from the form of this equation that the higher the sample extinction coefficient, the larger the fraction of the total Raman scattered light that occurs in the sample volume under consideration. This fraction is equal to N v / and is given in equation 4.34. N lysj,a=0 _ j _ g-111(10)^ (4 24) Y V j , » , a = 0 Figure 4.18 shows this fraction as a function of concentration for the sample penetrations under consideration for lysozyme resonance enhanced at 206 nm (e.g. for amide band enhancement) and 230 nm (e.g. for TRP or TYR band enhancement). It is clear from this plot that, for analytes with the molar absorptivities and concentrations commonly used 187 with UVRRS, a significant fraction of the Raman scattering occurs within an excitation-penetration depth that is practical as a sample volume (<1 mm). It is therefore conceivable that the probes described in section 4.3 could be used to collect a relatively large fraction of the total scattered photons from absorbing samples. This analysis should be distinguished from previous analyses [e.g. Cooney etal., 1996A; Cooney etal., 1996B; Zhu and Yappert, 1992A; Zhu and Yappert, 1996B] which show the fraction of collected power in a non-absorbing sample as a function of z for various fiber probe designs, rather than the fraction of total scattered photons in an absorbing sample that we show here. For RS, the theoretically optimum SNR (SNRopt) is obtained when all Raman scattered photons are collected and the detector noise (dark and readout) and background optical noise are both zero. This results in the SNR being limited by the signal shot noise, which is proportional to the square root of the signal, and thus the optimum SNR is simply the square root of the number of Raman scattered photons. Since the values for the parameters a and 8 are easily measured or obtained from the literature for most analytes and wavelengths of interest, it is a simple matter to calculate SNRopt. Figure 4.19 shows SNRopt as a function of concentration for a 600 pm sample volume penetration and several resonance enhanced analyte/wavelength combinations. These calculations were done assuming 1 mW of excitation and 100 seconds of integration. These values were chosen as it is the author's belief that they represent reasonable upper limts for integration time and excitation power for which the underlying idealized assumptions may be achievable. 188 On the other hand, LODopt can be obtained from the curves such as those in Figure 4.19 by observing the concentration at which they intersect the line defined by SNRopt=3. The optimal limit of quantification (LOQopt), defined by the concentration at which SNRopt=10, may be calculated similarly. Table 4.3 shows LODopt and LOQopt for a range of analytes, wavelengths and sample volume penetrations depths. This analysis is easily extended to any analyte and wavelength, and may include additional known instrumental efficiency and noise terms by using SNR=riN s,Z ( i/(N8A a+G 2 a+o 2b+...) 0" 5, where a 2 a, C M , , etc. are the standard deviations of the additional instrumental noise terms (assumed to be Gaussian) and rj is the efficiency of the system (given by the average number of scattered o l i i i i i i i i i I 0 1000 2000 3000 4000 5000 Concentration (flvl) Figure 4.19. Optimal (signal shot-noise limited) signal-to-noise ratio (SNRopt) as a function of concentration assuming 100 s of 1 mW excitation for (A) tryptophan 762 cm"1 W18 line excited at 218 nm, (B) tryptophan 1342 cm"1 W7 line excited at 218 nm, (C) phenylalanine 1207 cm"1 v 7 a line excited at 192 nm, (D) phenylalanine 1606 cm"1 Vg a line excited at 209 nm, (E) tyrosine 1263 cm'1 V7» line excited at 192 nm, (F) tyrosine 1617 cm"1 Vg a line excited at 223 nm, and (G) lysozyme amide III mode excited at 204 nm. 189 photons resulting in a detector count divided by the total number of scattered photons). For example, if the dark current is 1 count/second, then the resulting additional dark current noise is od = (100 s x 1 count/s)°5 = 10, and S N R o p t for 1 nM tryptophan excited at 218 nm (assuming 600 um sample volume depth) changes from 13.2 to 10.5. If the quantum efficiency is further reduced to 50%, then S N R o p t is further reduced to 5.3. The results presented in Table 4.3 show that in principle, picomolar or lower detection limits may be achievable with milliWatts of average power at the sample, sample volumes on the order of (600 um)3 = 200 nL, and integration times of ca. 1 minute. The relatively low limits for lysozyme amide bands reflect the large number of amide chromophores within a single lysozyme molecule. These values (all < 10"11 M) are more than 10s times lower than the current detection limit for aromatic amino acids using similar experimental conditions. This is indicative of the fact that many of the idealized assumptions are still far from valid; in particular severe Rayleigh noise, dark current noise, read noise, low spectrometer throughput and low detector quantum efficiency with the present spectrometer/detector combination conspire to raise the detection limits by many orders of magnitude. Chapter 7 discusses changes in the experimental arrangement which are presently being implemented or planned that should allow the actual limits to come much closer to the theoretical optimum. 190 Analyte Peak Wavelength LOD O D t (fM) LOQ O D t (fM) TRP W l S ^ c m " 1 ) 218 nm 60 580 TRP W7(1342 cm-1) 218 nm 400 3950 PHE v 7 a (1207 cm-1) 192 nm 50 555 PHE v 8 a (1606 cm-1) 209 nm 250 2800 TYR v 7 a (1263 cm-1) 192 nm 25 265 TYR v 8 a (1617 cm'1) 223 nm 400 4050 Lysozyme Amide III 204 nm 6 70 Table 4.3. Theoretical optimal limits of detection anc limits of quantification for several wavelengths, analytes, and Raman peaks using 100 seconds of 1 mW excitation. Sample volume penetration was set to 600 pm. 4.4.2.3 Absorbing Analytes in Absorbing Media In an absorbing medium, the propagating laser pulse is attenuated by both the absorbance of the analytes as well as other absorbing molecules which are not under consideration for detection. If the "medium" is considered to be all of the sample with the exception of the molecule possessing the analytic chromophore and it is described by an extinction coefficient, a (in length"1 units), then equation 4.27 may be generalized to equation 4.35 below. E(z) = E0l (4.35) The analysis presented above for the case of non-absorbing media may then be repeated to obtain expressions incorporating medium absorbance for the quantities described above. These appear in equations 4.36 to 4.38, below. f . >\ = 47£0Acmcr ^ he 1 N, (eem+a)\r{\0l 47cEQA cma ^ - m ^ - ^ j ( 4 3 6 ) s,<*>,a /*cin(l0) ecm+a (4.37) 191 Medium Extinction Coefficient (17cm) Figure 4.20. Theoretical optimal detection limits versus medium extinction coefficient (an,) for TRP W18 signal at 218 nm (square symbols) and TYR v 8 a at 223 nm (circles). All points using 100 seconds of 1 mW excitation over a 600 urn probe depth. 4n ln(l0) £Cm+a (4.38) These demonstrate that as the extinction coefficient of the medium increases, the total number of Raman scattered photons decreases only slowly when e c „ , » a, and quite quickly when a >« ecv N w is reduced to half of N ^ N ) when a^ec,. Figure 4.20 shows the behaviour of SNRopt as a function of medium extinction coefficient for two analyte/wavelength combinations. These calculations are particularly important if the analyte of interest is in a relatively opaque medium (e.g. blood). 192 4.4.3 Geometrical Limits 4.4.3.1 General Comments on Analytical FO-UVRRS The analysis presented above provides targets to aim for in the design of FO-UVRRS excitation and collection optics. If high-performance spectrometers and detectors are available, how close real systems can come to these limits is determined by the excitation and collection geometry. The angled FO-UVRRS probes we have designed in section 4.3 not only provide for a robust and versatile instrument, but through the physical proximity of the excitation and collection optics, may also facilitate more closely approaching these theoretical limits. In the following analysis, we investigate the excitation and collection process for angled FO-UVRRS probes and demonstrate a generally applicable method to determine the practical analytical limits of FO-UVRRS. Further, we discuss the results in terms of probe design modifications that may improve probe performance. 193 SIDE VIEW Figure 4.21. Schematic side view of 90° collection angled/mirrored probe geometry. O indicates the origin of the right handed coordinate system. Vectors labeled as k denote scattered, transmitted, or reflected light rays; while vectors labeled as u denote surface normals or fiber axis vectors. Re and Rc indicate the radii of the excitation and collection fibers, respectively; while d V and dA denote differential volume (scattering) and area (collection) elements, respectively 4.4.3.2 Development of FO-UVRRS Simulation The angled FO-UVRRS probes that we have developed have been discussed previously. Briefly, they consist of an excitation leg incorporating an improved ultraviolet optical fiber and one or more standard UV-grade fused silica collection fibers that incorporate reflective surfaces at 45° angles. Figures 4.21 and 4.22 show diagrams of the 194 TOP M E W Figure 4.22. Top view of angled/mirrored improved probe geometry. Symbols as in Figure 4.21. Dotted line represents line of constant x in the plane of the mirror that passes through the point of intersection of kt and the mirror (x=z plane). probe tip and a representation of the excitation and scattering process introducing the variables that will be used in the simulation below. In the discussion that follows the elevated A symbol (circumflex) indicates a unit vector, added subscripts x, y, or z indicate a particular vector component, and • denotes scalar product of two vectors. We will assume that the light exiting the excitation fiber is collimated, propagating in the z direction, and has a uniform intensity distribution across the excitation fiber face. The collimated beam assumption is a close approximation since (1) the numerical aperture of the fiber in air is small (ca. 0.22) and the light exits into a medium of relatively close refractive index (water, n=1.33; cw. silica, n «1.5), (2) lenses are usually polished on the tips of the excitation fibers in order to further collimate the exiting light, and (3) the 195 assumption is relevant only for a small distance (maximum 600 um) in the z direction. The assumption of uniform intensity distribution, however, is worthy of more discussion. In the past, we and other groups have used angular intensity distribution profiles (e.g. section 4.2) in the simulation of fiber-optic spectroscopic probes. Many of these profiles were obtained under the assumption that the near-field intensity distribution resembles the far-field intensity distribution, which is more easily measured. In fact, there is evidence that this is not the case. Several investigations [e.g. Setchell, 1992; Hillrichs etal., 1996] have indicated that when coupling light into fibers under normal conditions, the near-field intensity is relatively uniform with respect to radial position. Thus, using this approximation not only facilitates the calculation that follows, but may also be more accurate. The number of photons Raman scattered isotropically into 4K steradians due to a laser pulse with a fluence of N/A (photons/m2) from a differential volume element containing scatterers with a concentration C, and a differential Raman scattering cross section of a is given by equation 4.22 or 4.26, where E(z) is obtained by considering equation 4.27 or 4.35. The analysis proceeds by examining the scattering from an excited volume element with coordinates Q e in the sample volume to an area element on the collection optic surface at coordinates Q c by considering a scattered ray in the direction k»=Qc-Qe. The number of photons transmitted through an area element of the collection optic surface is given by equation 4.39, dNc = dNX^O-1'^^ U™*L (4.39) 196 where 0; is the angle between the direction of the scattered ray (kg) and the unit vector in the direction normal to the surface element, u c ; U is the magnitude of k,; e c and ac are the analyte absorptivity and sample absorbance, respectively, at the scattering wavelength; and T a v g is the Fresnel transrnittance averaged assuming equal p and s polarizations, viz. T = 1 - 0.5 ^nSi cos#, -nw cosd,^ + (nw cosc9, -nSj cos0. ynSi cosOt +nw cosOJ \nw cos0{ +nSi cosOj (4.40) Where nsi and nw are the indices of refraction of silica and water, respectively, and 0i and 0t are the incident and transmitted angles (measured from the interface normal), respectively. This analysis assumes that the angular efficiency of collection is determined only by (a) the solid angle subtended at Q e by the area element at Q c , and (b) the Fresnel coefficients. The final step is to determine whether or not the transmitted ray gets collected. The direction of the transmitted ray, k<, may be obtained by considering the vector form of Snell's law, which results in equation 4.41. in. cos0t -n, cos0,) A n " kt = ^ ,J-un +^k, (4.41) The trajectory of the transmitted photons is described by the parametric equation 4.42: Q{r) = a + rkt (4.42) where Q t is the position and x is a dimensionless parameter. The mirrored surface is defined by the plane x=z and the boundaries of the collection fiber, given by the equation 197 y2+(x-Rc)2=Rc2. The value for x at the point of intersection with the plane x=z may be obtained as follows: T = / c ~ * c (4.43) The x and y points of intersection of the transmitted ray and the x=z plane are obtained by the substitution of equation 4.43 into equation 4.42 and are given by equations 4.44 and 4.45. xr=xc + tkt^ (4.44) yr = ye + *tj, (4.45) This intersection point is on the mirrored surface if and only if the following inequality is satisfied: yl+{xr-Rc)2 cos(0c) (4.48) 198 Strictly speaking, this is only true for meridional rays, with skew rays being collected under somewhat less strict conditions (see equation 2.3). However, higher order skew rays are subject to greater attenuation and are not efficiently coupled into the spectrometer; thus this approximation is a good one. The calculation of the total number of collected photons proceeds by integrating over all volume elements (at points Qe) in the sample volume and, for each volume element, integrating over all surface elements (at points Qc) on the collection surface. For each ray from Q e to Q c, the amount of light transmitted into the fiber can be calculated as in equations 4.27, 4.35, 4.39, and 4.40; while the inequalities in equations 4.46 and 4.48 can be evaluated as TRUE or FALSE to determine if this light is collected. This process may be expressed by the following integral: Nc= JJJ \\BdNc (4.49) excitation collection volume surface where dNc is from equation 4.39 and the coefficient B is one if inequalities 4.46 and 4.48 evaluate as TRUE, and zero otherwise. It is also necessary to check that a particular ray under consideration is incident upon the area element under consideration from the outside of the collection fiber, as opposed to first passing through one side of the fiber and subsequently being incident from the inside. 4.4.3.3 Determination of Geometrical Analytical Limits to FO-UVRRS Equation 4.49 was discretized and, along with equations 4.39 to 4.48, became the basis of a computer program designed to simulate the improved probes (see Appendix 2). 199 The program was written, compiled and run using Borland C++ version 4.0 on a 60 MHz Intel Pentium based PC. Execution times varied, but were generally on the order of one hour for one concentration and a particular probe geometry. For a given resonance enhanced Raman vibration, equation 4.49 can be used to determine the optimum (maximum) number of photons collected by an angled probe (of the type described above), if the fiber radii, excitation energy, analyte concentration and scattering cross-section, medium absorbance at excitation and collection wavelengths, and analyte absorptivity at excitation and collection wavelength are specified. Since this equation is linear in excitation energy and scattering cross-section, all of the calculations may be done, for example, using 1 pJ as the excitation energy and 1 mbarn total scattering cross-section, and then scaling the result to the energy (power x time) and cross section under consideration. If this is done for a 1 pM concentration of a non-absorbing analyte for a series of medium extinction coefficients, ac, then the resulting plot of N c versus is the same as r\p, the normalized probe efficiency presented in section 4.3. 4.4.3.4 Working Curves and Analytical Limits Numerical evaluation of equation 4.49 for a fixed probe geometry and a range of analyte concentrations at a given wavelength results in a working curve for the specified probe/analyte/wavelength combination. Such curves are shown in Figure 4.23 for tryptophan residues in lysozyme excited using 227 nm excitation. The usefulness of working curves for determining operating ranges and optimal concentrations was discussed in sections 4.2 and 4.3. 200 Additionally, the approach followed above in calculating SNRopt and LODopt may be repeated to obtain these optimum analytical quantities for a fiber-optic probe geometry. Since the working curves are very linear at low concentrations, several points on the working curve may be obtained in this area and fit to a linear approximation. The square root of this linear fit is the geometrical SNRopt, and the concentration at which the square 1E21 1E22 1E23 1E24 Lysozyme Concentration (molecules/m3) Figure 4.23. Simulated working curves for lysozyme excited at 227 nm using an A/M probe with a 600 pm (diam.) collection fiber and a 600 pm (circles) or 100 pm (squares) diameter excitation fibers. The values were calculated on a per pJ and per (mbarn/sr) basis. root equals 3 and 10 are the geometrical LODopt and LOQopt, respectively. This analysis was done and Table 4.4 shows the analytical limits imposed by the probe geometry for the same analyte/wavelength combinations evaluated in Table 4.3. It is clear that practical geometrical analytical limits of detection are ca. 3 orders higher than the ultimate theoretical limits; however they are still at ca. nanomolar levels or lower. 201 Analyte Peak Wavelength L O D o D t (nM) L O Q o o t (nM) TRP W18 (762 cm-1) 218 nm 0.017 0.18 TRP W7 (1342 an 1) 218 nm 0.112 1.25 PHE v 7. (1207 cm'1) 192 nm 0.014 0.153 PHE v 8 a (1606 cm'1) 209 nm 0.075 0.84 TYR v 7 a (1263 cm'1) 192 nm 0.007 0.073 TYR v8a(1617cm'1) 223 nm 0.12 1.3 Lysozyme Amide III 204 nm 0.002 0.020 Table 4.4. Geometrical optimal limits of detection and limits of quantification for several wavelengths, analytes, and Raman peaks obtained using the computer program in appendix 2 and considering 100 seconds of 1 mW excitation. Probe geometry was taken to be a single 600 um excitation fiber and a single 600 um angled, mirrored collection fiber. 4.4.3.5 Mapping the Probe Volume To aid in conceptualization and evaluation of new designs, it is instructive to consider the origin of the collected light vis-a-vis the cylindrical probe volume. The code written to evaluate equation 4.49 can easily be modified to produce plots of total collected Raman scattering or Raman scattering from a differential z-slice as a function of z. It may also provide maps of the total collected energy from volume elements in the plane y=0, or in any plane of constant z. Figure 4.24 shows cumulative and differential collected intensity as a function of z for two different concentrations. The initial increase in differential collected intensity (despite the Beer-Lambert attenuation of the excitation light) is due to the fact that volume elements closer to z=Rc see a larger accessible area of the mirror. The subsequent decrease in differential collected signal is a result of both the Beer-Lambert attenuation and the reduced accessible mirror area for z » R c . 202 l 1 r 0.0000 0.0002 0.0004 0.0006 Ftenet ration From Beit at ion Fiber (m) 0.0010 -> N IS 0.0008H 15 .8 05 .18 0.00064 £ 0.0004 0.0002Hi "8 1 s 0.0000 l 1 0.0300 0.0006 W 0.000025 0.000020 -i 0.000015 H 0.000010 H 0.000005 0.000000 0.0000 0.0002 0.0004 0.0006 0.0008 Penetration From Dotation Fiber (m) Figure 4.24. Cumulative (top) and differential (bottom) collected intensity as a function of z for 10 pg/ml (dotted line)and 918 pg/ml (solid line) lysozyme. All intensities are on a per pJ and per (mbarn/sr) basis. 203 Chapter 5. Signal Processing for UVRRS 5.1 Overview of Signal Processing There are several important applications that require some kind of on-line signal processing or, more generally, mathematical manipulation of Raman spectral data [Gilbert, 1990]. Two of the more significant and frequently performed manipulations are SNR enhancement and deconvolution of the instrumental lineshape. SNR enhancement in Raman spectroscopy is particularly important due to the inherently weak nature of the Raman effect and the frequent existence of a large fluorescence or stray light background which degrades the SNR due to shot noise. An effective SNR enhancement method would have the effect of improving the detection limit for a given integration time or, alternatively, reducing the integration time for a given detection limit. Several digital filtering techniques have received attention for enhancement of extremely low SNR spectra [Bialkowski, 1988A; Bialkowski, 1988B]. Matched filtering and adaptive peak detection have been considered for such applications [Dyer and Hardin, 1985]. However these and other digital filtering methods usually require either a detailed a priori knowledge of the expected signal, as in the case of matched filtering, or the rather arbitrary assignment of filter parameters, as in the case of adaptive peak detection. Deconvolution to remove the effects of instrumental distortion on the underlying signal facilitates the interpretation of spectra and can also reveal spectral features which have been distorted by the finite resolution of the spectrometer (e.g. closely spaced adjacent peaks can appear as a single broad peak). An effective deconvolution method would allow the use of a larger slit width, while still obtaining spectra with resolutions 204 similar to those obtained without deconvolution using a smaller slit width. Thus, one may obtain superior spectrometer throughput without sacrificing resolution. Fourier deconvolution (FD), Fourier self deconvolution (FSD) [Gilbert, 1990; Mantsch and Moffatt, 1993] and the maximum entropy method (MEM) [Davies et al, 1991; Graves, 1989] have been considered for resolution enhancement. Few papers providing a quantitative comparison of methods for SNR enhancement and deconvolution have appeared in the literature. In particular, very little attention has been given to the use of so-called regularization methods for Raman signal recovery. Additionally, the significant body of knowledge concerning signal processing amassed over the past half century in the field of electrical engineering has largely been ignored by chemists. This chapter investigates the application of standard DSP methods to Raman spectroscopy and also presents a new regularization method for deconvolution and SNR enhancement, validates it, compares it to existing methods using simulated data, and demonstrates its use in an application involving experimental data. 5.2 Signal Characterization 5.2.1 Description of Raman Spectra Raman spectra are usually plotted as intensity, I (in detector counts) as a function of frequency shift, Av, away from the excitation line. Av could be measured in units of Hz, however at optical frequencies this results in unwieldy large numbers (~ 1015 Hz). Instead, it is observed that Av = (viaser-vi) - (c/Xiascr-c/Xi) =c(lA.ia8er-l/A,i), and so (1/Xiaser-1/Xi) is proportional to the frequency shift, and has units of cm"l, also called wavenumbers. Raman spectra are then plotted as I vs. wavenumber, in counts vs. cm"1. Most Raman spectra consist of a series of intensity peaks in the range of 500 to 2000 cm"1. The Raman 205 spectra that were used in this section were those of DNA, obtained using 266 nm excitation, and a mixture of Potassium Nitrate ( K N O 3 ) and Methyl-Orange dye (denoted by MO-KN), obtained using 472.7 nm excitation. Examples of these spectra are shown in Figures 5.1 and 5.2. The data sets used in the investigations are denoted by five characters. The DNA data sets each consisted of two spectra collected from a mixture of 0.67 mole/litre K N O 3 and a given DNA concentration, the MO-KN data sets each consisted of eight spectra collected from a 0.15 Mole/litre K N O 3 concentration, and a given Methyl-Orange concentration. The data sets used are summarized in the following two tables: Table 5.1. DNA data sets DATA SET NAME DNA CONCENTRATION (mg/ml) DE04B 0 DE04D 0.05 DE04F 0.10 DE04H 0.15 Table 5.2. MO-KN data sets DATA SET NAME METHYL-ORANGE CONCENTRATION (Molar) FE08A 0 FE08B 10-7 FE08G 2X10-5 FE08H 5xl0-5 FE08K 3xl0-4 206 m c 2 -i c -1 ' 1 ' 1 • 1 • 1— 0 200 400 600 800 Diode Number 1000 120 — i 1 1 1 1 1 • 1 • 1 • 1 ' 1 1 900 1000 1100 1200 1300 1400 1500 1600 Raman Shift (cm-1) Figure 5.1. Top: Raw (uncalibrated) data from one of the de04h spectra showing DNA signal peaks and the unintensified region. Bottom: Spectrum calibrated for wavenumber. in c _ Q) C J 1 ' 1 ' 1 200 400 600 Diode Number 800 1000 1200 600 —I— 800 T 1000 1600 1800 2000 i j 1 1 r -1200 1400 Raman Shift (cm-1) Figure 5.2. Top: Raw (uncalibrated) data from one of the fe08h spectra showing DNA signal peaks and the unintensified region. Bottom: Spectrum calibrated for wavenumber. 207 5.2.2 Investigation of the Sampling Process There are several stages that ultimately lead to the signal being sampled by the detector. As mentioned in chapters 1 and 2, the scattered light, containing many optical frequency components, passes through a slit, is dispersed by the diffraction grating, is amplified by the image-intensifier, and is finally detected by the photo-diode array; where the optical signal from each diode is converted to an electrical signal, digitized, and stored in memory. Each of these steps imposes its own response function on the signal, changing it in some way. The detailed analysis of the exact response of the entire, cascaded system is a detailed problem in Fourier-optics which is well beyond the scope of this section. However, there are four important things to note in the sampling process: (1) the sequence being sampled is a sequence of intensity vs. optical frequency, and is called a spectrum. This will cause some ambiguities in nomenclature, as the Fourier-domain representation of the Raman spectrum is the "spectrum of the Raman spectrum". Also, the abscissa of the Raman spectrum is in units of frequency, which will cause some more ambiguity, as frequency is also used for the abscissa of Fourier domain representations. Please note the context in which these terms are used in this section. (2) the units of the abscissa for the data (Raman spectrum) are cm~l (or wavenumbers) and so the units for the abscissa in the Fourier domain are cm (or 1 /wavenumbers). (3) The finite spectrometer slit width; the distortion introduced by the dispersive element, reflecting/focusing optics, and image intensifier; and the finite width of the individual diodes all contribute to an Instrumental Point Spread Function (IPSF), which is convolved with the underlying signal and results in the detected Raman signal being spread or broadened compared with the actual, 2 0 8 underlying signal. (4) The spatial separation of the centres of the individual diodes of the detector array is equivalent to an optical frequency separation (in cm"l) which is determined by the inter-diode distance, the wavelength of excitation, and the dispersion of the diffraction gratings. Thus, the sampled sequence is a sequence of intensity values sampled at a 'frequency' of l/(optical frequency separation). In the case of the spectrograph used in these investigations, the optical frequency separation is slightly over 1 cm - 1 , and so the equivalent sampling 'frequency' is approximately 0.98 cm (depending upon the exact excitation wavelength). These units and normalized units will be used in this section. If the sampling frequency is 0.98 cm; then by the sampling theorem, the signal must be bandlimited to less than 0.49 cm in order to avoid aliasing. Raman signals are generally very broad, without sharp features, and the instruments used generally have high dispersion, and thus no significant aliasing is expected. 5.2.3 PSD Estimation Methods 5.2.3.1 Overview of PSD Estimation Methods The goal of power spectral density (PSD) estimation methods is to determine the frequency content of a signal. This is necessary in the case of Raman signals in order to (a) characterize the nature of the signal in order to detect differences between signals, (b) design effective frequency-domain filters to remove high frequency noise, and (c) determine the minimum sampling interval (i.e. - to determine the maximum inter-photo-diode distance for a given dispersion, or to determine the minimum dispersion for a given inter-diode distance). Because the signals under consideration here are non-deterministic, statistical methods must be used. PSD estimation methods may be broadly divided into two classes, non-parametric methods and parametric methods. Non-parametric methods make no assumptions concerning the underlying processes that created the signal, while parametric methods do. 5.2.3.2 Non-Parametric PSD Estimation - Welch's method The non-parametric method of PSD estimation that was investigated was Welch's method [Proakis and Manolakis, 1992; Krauss et. al., 1994], available in MATLAB using the MATLAB PSD function. This is one of several so-called "periodogram-based" PSD methods (the other two main periodogram-based methods being the Bartlett PSD and Blackman-Tukey PSD, neither of which were investigated). The detailed justification for these methods, and considerations in their use is found elsewhere [Proakis and Manolakis, 1992]. These periodogram-based methods involve estimating the PSD using the magnitude-squared of the FFT. This results in a PSD estimate at n/2 equally spaced points, from an original n point vector. An important consideration in PSD estimation techniques is how the expectation value of the estimator compares to the actual PSD of the true process. It turns out that for PSD techniques based on the periodogram method, the expectation value of the estimate is equal to the true value convolved with the window function's frequency. This results in some leakage from one frequency into another. There is a well-known leakage/resolution trade-off in deciding on what window type to use. Most applications of Welch's method use a Bartlett or Harming window. 210 The other important consideration is the variance of the estimate. Periodogram PSD-estimation techniques do not become more accurate as longer stretches of data are used in the squared-FFTs. In fact, the variance stays at the square of its expectation value, which implies that the standard deviation is 100% of the expectation value in each frequency bin. A way to overcome this is to divide the input vector into K equal-length segments and produce a periodogram estimate separately for each. These may then be added together to produce a final estimate with a variance in each channel reduced by a factor of K (standard deviation reduced by K 1 / 2). Unfortunately, because the FFT size is also reduced by a factor of K, the resolution of the PSD estimate is reduced. The variance may be further reduced by using overlapping data segments. This essentially increases the number of averaged segments without decreasing the length of each segment. Unfortunately, the reduction in variance is no longer linear with overlap, due to the fact that adjacent segments are no longer independent. The further decrease in variance with overlap is partially due to the fact that it overcomes the problem that when using non-rectangular windows, much of the data near the window periphery is 'thrown away'. Proakis and Manolakis [Proakis and Manolakis, 1992] introduce a figure-of-merit, Q, in order to evaluate non-parametric PSD methods: Q=expectation value/variance. With a triangle window and 50% overlap in Welch's method, Q=1.39NAf. The MATLAB psd function computes a PSD estimate using Welch's method. One may specify the length of FFT, the number of overlapping samples, the type of detrending used, and the type of window used. In all cases I used the Hanning window. The default detrend option 'linear' was also used in order to remove the best straight line from the 211 psd est. nfft= 64 psd est. nfft= 128 -20 -40 m 33. s 0 'c D) CO E -20 0 0.2 0.4 psd est. nfft= 256 Q 8 -40 or o z 0 -20 -40 i 0 -20 -40 3 0 psd e; 2 0 »t. nfft= 51; 4 > 0.2 0.4 Freq. (cm) 0.2 0.4 Freq. (cm) Figure 5.3. PSD estimates from fe08h using different FFT lengths. prewindowed segments. Furthermore, the PSD function can be used to return the frequencies for each discrete PSD estimate, and a vector of confidence intervals for a given level of confidence. Several PSD estimation investigations were conducted using both the DNA (DNA) and the Methyl-Orange-KN03 (MO-KN) data. 95 percent conf. intnfft= 64 95 percent conf. intnfft= 128 0 0.5 _ 95 percent conf. intnfft= 256 ^ 60 o o I 40 2 CL 10 20 co 0 0.5 80 60 40 20 0 0.5 95 percent conf. intnfft= 512 0.5 Freq. (cm) Freq. (cm) Figure 5.4. 95% confidence intervals for one fe08h spectrum using different FFT lengths. 212 In the first investigation, the MATLAB program psdal (see Appendix 3 for all MATLAB programs) and eight similar MO-KN data sets (fe08h) were used in order to investigate the effects of changing the FFT length, nfft, on the PSD estimate. Raman spectra with 512 points were used in all cases, and the number of overlap samples was equal to nfft/2. Values of nfft of 64, 128, 256 and 512 were used. The results, averaged over the eight PSD estimates, are shown in Figures 5.3, 5.4 and 5.5. Figure 5.3 shows that the variance of the PSD goes down with decreasing nfft; however, so too does the spectral resolution. Note that each of the PSD estimates in Figure 5.3 is normalized to a maximum value of 0 dB. Figure 5.4 shows the estimated 95% confidence intervals (not normalized) for one of the 8 Raman spectra, as produced by the PSD routine. Figure 5.5 shows the o standard deviation at each point (normalized to max(s)=0dB) for the eight PSD estimates using each of the four nfft values. This investigation was also performed on two other eight-spectra data-sets, fe08a and fe08k; the results for fe08a appear in Figure 5.6, and were similar to those for fe08k (not shown). These data indicated that with increasing FFT length in the PSD routine, as expected, the standard deviation and the resolution increased. As a compromise between variance and resolution, a value nfft=128 was chosen for subsequent studies. 0 -10 -20 -30 s (n=8). nffl= 64 3 0.5 s (n=8). nfft= 256 Freq. (cm) 0 -20 •40 -60 s(n=8).nfft=128 0 0.5 s(n=8).nfft=512 0 . 0.5 Freq. (cm) Figure 5.5. Standard deviation for fe08h spectra using different FFT lengths. -20 -40 m "O a. o psd est. nfft= 64 0 0.2 0.4 psd est. nffl= 256 0.2 0.4 Freq. (cm) -20 -40 i 0 -20 -40 psd est. nflt= 128 0 0.2 0.4 psd est. nfft= 512 0.2 0.4 Freq. (cm) Figure 5.6. PSD estimates for fe08a using different FFT lengths. 214 In the next investigation (see file psd_a2 in Appendix 3) attempts were made to optimize the number of overlapping samples. The FFT length was kept at the previously determined 128 samples. Little difference was found, and so the number of overlapping samples was left at 64 (=nfft/2) as suggested in the literature. psd est file: fe08a psd est file: fe08b 0 -20 -40 CQ 4) „ 3 o 'c D) TO E -20 0 0.2 0.4 psd est file: fe08g -40 J - W l AS -20 -40 0 -20 -40 \ 0 0.2 0.4 psd est file: fe08k 0.2 0.4 Freq. (cm) 0.2 0.4 Freq. (cm) Figure 5.7. PSD estimates for MO-KN RRS data. Next, using 128 sample FFTs and 64 sample overlap, PSD estimates were produced for 4 different data-sets (each of different analyte concentration), of each DNA and MO-KN (see psd_a4 in Appendix 3). The results appear in Figures 5.7 and 5.8 for MO-KN and DNA, respectively. The results indicate that in each case, most of the signal information is confined to the spectral region between Fourier-frequencies of 0 and 0.1 cm. Also, it is apparent that for the MO-KN (fe08) spectra, there is a broadband (white-ish) noise background at around -40 dB. For the DNA (de04) spectra, the noise background is around -25 dB. Furthermore, the MO-KN seem to show a small (ca.. -30 215 dB below max, or +10 dB above background) peak at a frequency of around 0.25 cm, corresponding to approximately 1/4 of the sampling frequency. psd est. file: de04b 0 -20 -40 m or o 0 0.2 0.4 psd est. file: de04f 02 0.4 Freq. (cm) psd est. file: de04d 0 -20 -40 0 -20 -*0 \ 0 0.2 0.4 psd est. file: de04h 0 02 0.4 Freq. (cm) Figure 5.8. PSD estimates for 266 nm DNA UVRRS data. 5.2.3.3 Parametric PSD Estimation Using an Autoregressive Model This method of PSD estimation essentially involves modeling the Raman spectrum as the impulse-response of an all-poles infinite impulse response (IIR) filter of order n. The frequency response of the filter is then equal to an estimate of the PSD of the signal. MATLAB can compute the filter coefficients using the linear predictive coding Ipc routine. This routine accepts as input the signal and the autoregressive (or filter) order n, and gives, as output, the all-poles IIR-filter denominator coefficients. These may then be used to calculate the frequency response using the MATLAB function freqz. This is an interesting advantage, as the PSD estimate may now be computed at any frequency, not just nfft/2 discrete frequencies, as was the case with Welch's method, above. The 216 derivation and explanation of the underlying operations used in this technique are beyond the scope of this project and the reader is referred to the appropriate references [Burg, 1967; Proakis and Manolakis, 1992; Kauppinen and Saario, 1993; Press et. al, 1992; Krauss et. al, 1994]. This type of PSD estimation technique is sometimes called maximum entropy PSD estimation; however it should not be confused with the maximum entropy method discussed later. A program was written (see psdb in Appendix) to effect this method of PSD estimation. One Raman spectrum from the fe08 MO-KN data set was used in this investigation, and the auto-regressive order n was varied as n=l, 4, 16, 64, 256 and 1024. The overall morphology of the PSD estimate is similar to that produced by Welch's method, except smoother. Welch's method proved to be more useful for PSD estimation of Raman signals. The main advantage of the Ipc technique is known to be its ability to distinguish rather easily between signal components with very closely spaced frequencies; however this was not a requirement in these investigations. 5.3 FFT Based Methods for SNR Enhancement 5.3.1 FIR Techniques Used. 5.3.1.1 Introduction to FIR Techniques The finite impulse response (FIR) filters that were designed and tested in this investigation were obtained using the Remez exchange algorithm. The theory behind this method will not be explained here, and the reader is referred to the appropriate references [Krauss, 1994; Proakis and Manolakis, 1992; Jackson, 1989]. Suffice to say that it is 217 115 110 105 o 100 (0 £ 90 -S «° 80 70 60 0. Figure 5.10. SNRv order of 60. n= 60 D4 0.06 0.08 0.1 0.12 0.14 0.16 0.18 Centre of trans, band (normalized to Fs=1) s. transition band centre for spectra in data set fe08h with a filter 219 5.3.1.3 Performance of FIR Filters A large number of FIR filters of various orders and transition bands were designed and used; space prevents the display of all of the results. The conclusion is that a PSD -10 -20 -30 -40 -50 -10 -20 -30 -40 -50 200 400 600 800 1000 I I \mi J fll.l, ,,L ... . I i .J 200 400 600 Frequency Bin Index 800 1000 Figure 5.11. Top: Frequency response of filter with a transition band center and width of 0.1 and 0.02, respectively, and an order of 70. Bottom: magnitude of FFT of raw and filtered data. estimate may be used to design an FIR filter using the Remez exchange algorithm that is very effective at removing high frequency noise from Raman spectra. Figure 5.11 shows the frequency response of the filter (top) and the input and output (bottom). Figure 5.12 shows the input and output of one of these filters. The output is shifted by an amount equal to (filter order)/2. It is seen that the SNR is quite significantly improved and that high frequency noise is effectively filtered out. An unexpected result that was found in one case was that there could be an optimum in the filter length - a monotonic increase in SNR with filter length had been expected. 220 5.3.2 IIR Techniques An elliptical IIR filter was also developed using MATLAB (see iira in Appendix 3); however there was no improvement over the FIR case. Two problems were found with the IIR design. First, ignoring the initial conditions caused serious start-up distortion in the output, and second, the non-linearity of the phase response caused a distortion. Both of these problems were overcome by using the MATLAB filtfilt routine, which filters in the 1400 I 1200 1000 800 -5 600 400 200 0 0 200 400 600 800 1000 1200 diode number Figure 5.12. Raw and Filtered DNA UVRRS data de04f. The pass band was centred at a normalized frequency of 0.1 and width of 0.02. The filter order was 70. forward direction, reverses the sequence and runs it back through the filter. This results in zero phase distortion and matched initial conditions [Krauss, 1994, pp. 2-102]. Some numerical problems were encountered in designing higher order filters using ellip. t mm * J 221 5.4 Regularization Based Methods 5.4.1 Theory of Regularization Based Methods 5.4.1,1 The Signal Recovery Problem The measured spectra under consideration are modeled as underlying signals (usually consisting of a series of peaks) convolved with IPSF and augmented with random noise. The spectra in section 5.4 are considered to be vectors of finite length with each component representing the measurement from a single channel of an array detector or a single sampling from a one channel detector and scanning monochromator arrangement. Denoting the underlying signal vector as x, the measured signal as m, the IPSF as b and the noise as n, this may be written as m = x*b + n (5.1) where * represents convolution. The vector x is the spectrum that would be obtained by a spectrometer with perfect resolution in the absence of any random noise. The IPSF is a consequence of the limited resolution of any spectrometer due to finite slit widths and broadening imposed by the detector. The IPSF typically takes on a symmetrical, peak shape that depends upon the spectrometer and detector characteristics. Its effect is to smooth and increase the spectral width of sharp features. The noise may come from several sources including shot noise from the signal or background, thermal noise in the optical detector, digital truncation noise due to the finite word length of the computer, and electrical noise in the detector circuitry, to name a few. The noise arising in different channels is considered to be zero mean and uncorrelated, i.e. the expectation value for the autocorrelation of n is zero for all autocorrelation lags except zero. In practice, the noise 222 is often described by a Gaussian probability distribution. Often, if the effects of the DPSF can be considered negligible (or if they are not of interest), then the model described by eq. 5.1 simplifies to m = x + n (5.2) In either case, the signal recovery problem is to obtain as good an estimation i , of x, as possible from the instrumentally distorted and noise corrupted measured spectrum. 5.4.1.2 Background and Introduction to Regularization Methods Signal enhancement in very low-SNR spectra (i.e. spectra with an SNR near the detection limit, defined as an SNR of 3) has been attempted using a variety of methods, none of which are entirely satisfactory due to the fact that they require either a priori knowledge of the expected signal morphology or the use of parameters for which optimal values are not known. For example, in examining the use of matched filtering (MF) and adaptive peak detection (APD) for SNR enhancement of severely degraded Raman spectra, Dyer and Hardin [1985] point out that the performance of MF depends heavily on an a priori knowledge of peak morphology, while the performance of APD depends heavily on the correct choice of four filter parameters. Due to their prevalence in spectroscopic applications and their somewhat relaxed requirements for a priori signal knowledge and filter parameters, zero and fourth order Savitzky-Golay (SG) polynomial smoothing filters will be used in this paper as reference methods for comparison with the TPMEM method presented here. The theory behind 223 these filters and considerations for their use have been presented in detail elsewhere [Bialkowski, 1988A; Savitzky and Golay, 1964; Bromba and Ziegler, 1981]. The most obvious method to deconvolve the effects of the IPSF is Fourier deconvolution (FD). The convolution represented in eq. 5.1 can be calculated in Fourier space as the component by component product of the Fourier transforms of the underlying signal and the IPSF. If, in the absence of noise, the Fourier transform of the measured signal is divided by the Fourier transform of the IPSF, the result is the Fourier transform of the underlying signal, which may then be inverse-Fourier -transformed to recover the . desired underlying vector. FD is distinct from FSD which attempts to deconvolve the entire line shape, i.e. both natural and instrumental effects, in order to measure peak position and shifts. In practice, the presence of even a small amount of random noise in the measured data when using FD results in an unacceptable amount of noise in the recovered vector. This problem may be overcome to some extent for low levels of noise through the use of prefiltering and/or by appropriately weighting the components of the Fourier domain representation of x in order to attenuate high frequency noise components. However, this imposes a distortion on the resulting lineshapes and also requires a priori knowledge of the expected signal in order to correctly choose the filter design. The second method which has received attention recently is the Maximum Entropy Method. It should be noted that the MEM considered here is distinct from Burg's maximum entropy method [Burg, 1975; Kauppinen and Saario, 1993], which is a power spectral density estimation method. MEM belongs to a particular class of inverse problem solutions, referred to as regvlarization methods [pp. 804-808, Press etal, 1992]. These 224 proceed by solving for the desired estimate of the underlying signal vector x, by minimizing a multidimensional function that is the sum of some smoothing function A(i), and some goodness-of-fit function B(±), subject to some constraint on B(±). In practice, this is done by minimizing eq. 5.3 for the vector t, and then adjusting Lagrange multiplier X, so that the constraint on B is met. F(x) = A(x) + XB(x) (5.3) MEM, and the TPMEM method to be presented here, use the x2 statistic as the goodness-of-fit function, with the constraint that x2=NP, where NP is the number of points in the spectrum, viz. NP = NP (5.4) where a 2 is the noise variance in channel j, and ±, m, and b are as defined previously. This constraint arises due to the statistical expectation that x2 lies in the range of NP +/- (NP)1/2 95% of the time. MEM uses the negative entropy (-S) as the smoothing function to be minimized (hence maximum entropy), viz. A A NP\ A(x) = -S(x)=Z J=I| A ( \ Id A vy J (5.5) In practice, for smoothing purposes, MEM often uses a further term, C, representing a constraint equating the totals of the recovered and measured signals, viz. 225 A W A NP (5-6) C(x) = Xx,- = /=i »=i and a further Lagrange multiplier p, to constrain C. Thus-MEM proceeds by solving for x, X and p such that the following equation is minimized subject to the constraints in equations 5.4 and 5.6 above, viz. Much of the attention in the MEM literature has focused on the theoretical justification for using eq. 5.5 [Jaynes, 1957; Frieden, 1972; Shannon, 1948], which essentially involves considering the underlying spectrum to be a probability distribution, and on investigations into algorithms to solve for t. These will not be discussed further here, and the reader is referred to the references cited above. It should be noted that, despite its information theoretic justification, MEM is merely one of several effective nonlinear regularization methods. Press et al. [p823 Press et al., 1992], point out that "Its peculiarities become strengths when applied to the reconstruction of incomplete Fourier data of images that are expected to be dominated by very bright point sources, but which also contain interesting, low intensity, extended sources. For images of some other character, there is no reason to suppose that MEM methods will generally dominate other regularization schemes, either ones already known or yet to be invented". Here a new regularization method (TPMEM) is introduced, which replaces the entropy of eq. 5.5 by the sum of two-point entropy measures across the entire spectrum. By using the models represented by eq. 5.2 or eq. 5.1, this method may be used for SNR enhancement of very low SNR spectra or for deconvolving noisy spectra to remove the MEM = -S + Xx2 + \xC (5.7) 226 instrumental distortion and noise, respectively. TPMEM requires only estimates of the random noise statistics and, for deconvolution, the IPSF, which are easily obtained in practice. Furthermore, TPMEM does not require any a priori knowledge of the expected signal shape, or any other parameters. This represents a considerable advantage over other methods of SNR enhancement and deconvolution. 5.4.1.3 TPMEM MEM considers the entire spectrum to be a probability distribution and proceeds by attempting to maximize the 'probability' of the recovered spectrum by maximizing its entropy subject to some constraints, as mentioned above. In contrast, the TPMEM method presented here considers each two adjacent points to be a two-point probability distribution and maximizes the sum of the entropy measures of these two-point probability distributions subject to the goodness-of-fit constraint of eq. 5.4. If the values of the underlying spectrum in detector channels i and i+1 are Xi and x;+i, respectively, then the probabilities p and q associated with channels i and i+1, respectively, may be written as * (5-8A) *• A A X/+X/+1 A g = X , + 1 (5-8B) " A A X/+X/+1 The two-point entropy, first defined by Shannon, [Shannon, 1948] is -(p log p +q log q) and represents a measure of choice or information in the two-point probability *•> /•> ^ distribution. Denoting the two-point entropy of points i and i+1 as S2j, their sum, denoted by SS2, may be written as NP-l NP-l SS2= £ £ 2 , = Yj-iplogp + qlogq) (5.9) i=l ;=1 TPMEM proceeds by using the single function, SS2, which represents the sum of two-point entropies, as the smoothing function A in eq. 5.3. Thus, TPMEM solves for the spectral estimation vector ±, and Lagrange multiplier X, which minimize the equation Fm=-SS2 + Xf (5.10) again subject to the constraint of equation 5.4. One important difference between TPMEM and MEM that is immediately evident is that the entropy of eq. 5.5 is invariant under a complete reordering of the data points, while SS2 of eq. 5.9 is highly dependent on the ordering of the data points. In practice, this means that regularization methods using entropy usually make use of a constraint on a third function (eq. 5.6) and a second Lagrange multiplier (p in eq. 5.7) in order to impose a smoothness on the recovered data. The TPMEM method uses only two functions and a single Lagrange multiplier (eq. 5.10). This confers an advantage of computational simplicity and speed upon TPMEM and also obviates the need for the somewhat arbitrarily imposed constraint of eq. 5.6. 5.4.2 Algorithm and Computational Details The algorithm used to implement TPMEM is shown schematically in Figure 5.13. When used for SNR enhancement only (i.e. not for deconvolution) the IPSF b, was taken 228 to be a Kronecker delta function. All computer programs were written and compiled as 16 bit DOS applications for the Intel 80486 processor using Borland C++ v. 4.0 on a 60 MHz Pentium microcomputer. Starting with a flat estimate (i.e. xt = (EmjVNP; i=0,NP) for the recovered values f, and upper and lower estimates for the Lagrange multiplier X, a conjugate gradient method [pp. 420-425 Press etal, 1992; Polak, 1971] modified to introduce a step in the direction -VF7M was used to minimize eq. 5.10. Proper convergence was determined by observing a negligible (<10"8) fractional change in FTM upon successive iterations of the conjugate gradient method, as well as a small value for the magnitude of the gradient of FTM- After convergence of the optimization algorithm, the %2 fit between t*b and m was checked. A, was then adjusted using a bisection method until the condition of eq. 5.4 was met. In practice, we required that %2=NP to within a tolerance of+/- 5. Few convergence problems were encountered for small (100 point) spectra; however the larger (600 point) spectra occasionally failed to converge properly. 229 Initialize estimated spectrum to uniform values and read In measured values Set X = X w minimize F1M= -SS2+XxJ, and set the resulting x =X2«* Set minimize Fw= -SS2+XX* , and set the resulting x -x\>w Set X^O^+X^J/2, and minimize Fw= -SS2+Xx' Write results to file Figure 5.13. A flow chart representation of the numerical algorithm TPMEM, implementing a two-point entropy regularization method. MEM deconvolution was used for comparison and was implemented by minimizing eq. 5.4 using a conjugate gradient method modified to periodically use straight steepest descent steps. A bisection root finding method was used to determine the correct value of X to within the tolerance mentioned above for the TPMEM solution. This tolerance was somewhat tighter than the usual +/- (NP)Vl tolerance commonly used in MEM solutions in order that we could more objectively compare the TPMEM and MEM solutions. The Lagrange multiplier p was adjusted until the constraint on eq. 5.6 was met to within a fractional tolerance of 6%, a value chosen based on previous experience. Few convergence problems were encountered. 230 The SG solutions were obtained by calculating the SG coefficients and convolving them with the measured data. Zero and fourth order SG solutions were obtained using a filter width optimized to produce the smallest average RMS error (see eq. 5.12) between the uncorrupted simulated spectra and the SG filtered spectra. Thus, the SG filtered spectra are, theoretically, the best results obtainable using such SG filters. This, in a sense, represents the use of a priori signal knowledge; however in practice, filter widths between 1 and 2 times the full-width-at-half-maximum (FWHM) of the underlying feature produce similar results. 5.4.3 Experimental 5.4.3.1 Artificial Data Simulations. In order to evaluate the performance of TPMEM in an objective manner, simulated measured spectra were created and used as input to the programs. In all cases these were based on 100 point prototype spectra with Lorentzian peaks defined by (5.11) r where c is the peak amplitude (in arbitrary units), i is the channel number, io is the peak center channel number and T is the half-width-at-half-maximum (HWHM) of the Lorentzian. For the SNR enhancement trials, 28 simulated spectra were produced, each with one randomly positioned Lorentzian peak with a random HWHM between 2 and 12 channels, and were corrupted with random additive Gaussian distributed noise such that the ratio of noise standard deviation to maximum peak height was 35%. 231 For the deconvolution trials, each simulated spectrum consisted of 3 Lorentzians with random heights between 2 and 7 channels, random widths between 2 and 12 channels, and random center positions between channels 10 and 90. Each of these spectra was convolved with an IPSF consisting of a normalized Gaussian peak with a FWHM of 8.29 which was truncated to 35 points. The justification for using this as the IPSF in these trials follows in the next section. Five groups of ten spectra were used in order to obtain results at 1%, 2%, 3%, 4% and 5% noise levels, where the standard deviation of the random Gaussian noise is expressed as a percentage of the maximum signal height. 5.4.3.2 Experimental Raman Spectra The experimental arrangement was described in chapters 2 and 4. A 1200 G/mm holographic grating was used with 514.5 nm and 472.7 nm excitation, while a 3600 G/mm grating was used with 363.8 nm excitation. An entrance slit width of 250 um was used in all cases. A Spectra Physics Stabilite 2017 argon-ion laser was used to excite the sample at 514.5 nm, 472.7 nm or 363.8 nm. A fused silica fiber optic probe consisting of a 100 um core diameter excitation fiber and a 300 um core diameter collection fiber aligned and glued parallel and flush at one end was used to excite the sample and collect the spectra. The laser beam was focused into the excitation fiber using a microscope objective and the collection fiber was delivered to the monochromator entrance slit via a fiber optic adapter. The spectra collected were 1024 points long; however they were truncated to 600 point spectra to remove the unintensified regions at either end of the spectra. Backgrounds 232 were subtracted from the measured spectra by fitting and subtracting a third order polynomial to 6 to 10 judiciously chosen points on the background. The IPSF of the monochromator/detector combination was estimated by de-tuning the laser cavity and observing the shape of the 506.2 nm plasma emission line from the argon-ion laser using a 250 um slit width. It was observed to be Gaussian to a good approximation and the least squares fit Gaussian to the measured shape resulted in a F W H M of 8.29 channels and a %2G fit between the 100 point plasma line spectrum and the estimated Gaussian of 42.9, which indicates that there is no statistically significant difference between the measured IPSF and the Gaussian approximation. Note that %2G = ZCrPSFmes-IPSFestf/IPSFes, * y? of equation 5.4, where J P S F m e s and I P S F ^ are the measured IPSF and Gaussian estimate, respectively. The fact that the measured IPSF does not have the expected lineshape associated with a spectrometer incorporating an unintensified linear photo diode array [Burton and Blades, 1987] is probably due to distortions introduced by the photo-cathode and the intensifier of the detector. 5.4.3.3 Performance Evaluation The performance of the methods for both SNR enhancement and deconvolution on the simulated data was evaluated on the basis of the average RMS error between the known true spectrum and the recovered spectrum, viz. RMSERR _ i NP A . Z(*-xY 7 = 1 NP (5.12) 233 as well as by the average Pearson correlation coefficient, r, between the true and recovered data, viz. NP A A I (*,.-*)(*,-*) (5.13) 1=1 r = NP _ I NP A A ^ Z ( ^ - ^ ) 2 ^ Z ( x , - x ) 2 Where the horizontal bar denotes the average of all components. RMSERR is a standard measure of fit; Pearson's r was also used as it is a better measure of correspondence between the trends of the true and recovered spectra. Since RMSERR=0 while r=l for a perfect fit, 1-r was used as the figure of merit, in order to make both figures of merit return smaller values for better fits. For the SNR enhancement trials, the methods were additionally evaluated on the basis of Q, the ratio of the SNRs before and after processing, i.e. SNRPMCESSED unprocessed In this section the SNR is defined somewhat conservatively as 1 2> , . - b - g d i ) SNR = '-'o-o.5H>fw ( 5 1 5 ) Sbgd where y represents either the measured spectrum m, or the recovered spectrum t; FTWFfM refers to the HWFTM of the spectral peak of interest; io is the center channel of the peak; bgd is the background vector, and Sbgd is the standard deviation of the background. This SNR represents the ratio of the average signal peak height (measured in a window of 234 width HWHM+1, centered at the peak maximum) to the standard deviation of the noise (calculated as the square root of the variance of the measured spectrum in a 26 point region that did not contain any spectral features). Williams [Williams, 1991] has discussed some of the problems in using the SNR as a figure of merit; however, as we used a large number of points to calculate both the noise variance and the signal mean, the SNR is a useful figure of merit in this work. (Williams suggested that for large samples, SNRs differing by a factor of 1.5 can be considered statistically significant at the 95% confidence level). 5.4.4 Results and Discussion 5.4.4.1 Low SNR Signal Enhancement. The results of a typical SNR enhancement trial using simulated data are presented in Figure 5.14, which shows the underlying spectrum, the noise-corrupted spectrum, and the recovered spectrum using TPMEM, zeroth order SG filtering, and fourth order SG filtering, as described above. Table 5.3 shows the resulting RMSERR, 1-r value, and SNR enhancement ratio, Q, for each method. TPMEM shows substantially improved performance over zeroth and fourth order SG filtering. Figure 5.15 demonstrates the use of TPMEM on 600 point experimental spectra of aqueous methyl orange and aqueous potassium chromate (see reference by Bernstein [1973] for peak locations and assignments). The SNR of two (overlapping) methyl orange peaks between 1300 cm"1 and 1500 cm"1, as well as the 853 cm"1 peak of chromate ion have been significantly improved by TPMEM. Additionally, the methyl orange peaks near 235 1600 cm"1 and 1200 cm"1 appear to be resolved. TPMEM is clearly a very effective signal processing method for SNR enhancement of severely noise degraded spectra. Additionally, TPMEM requires no a priori information concerning the expected underlying spectra. A significant drawback of TPMEM (and other regularization methods) is that it is very much slower than the polynomial smoothing methods or FIR filtering methods, and is not guaranteed to converge. FIGURE OF MERIT METHOD TPMEM SG4 SG-0 RMSERR 0.426+/-0.35 0.648+/-0.056 0.897+/-0.059 1-r 0.097+/-0.013 0.171 +/-0.016 0.341 +/-0.019 Q 6.53 +/-0.77 3.35 +/-0.26 1.80+/-0.17 Table 5.3. Figures of merit for the performance of TPMEM, fourth order Savitzky-Golay filtering (SG-4) and zeroth order Savitzky-Golay filtering (SG-0). The errors represent +/- one standard error of the mean (N=28). For RMSERR and 1-r, lower numbers imply superior performance; while higher Q values imply improved performance. 236 800 1000 1200 1400 1600 1800 Raman Shift (cm1) Figure 5.15. An example of the use of TPMEM for SNR enhancement on experimental Raman data. (A) The resonance Raman spectrum of 20 pM aqueous methyl orange excited at 472.7 nm with an integration time of 1 second. (B) The spectrum of A processed using TPMEM. (C) The resonance Raman spectrum of 200 pM aqueous chromate ion excited at 363.8 nm with an integration time of 4 seconds. (D) The spectrum of C processed using TPMEM. 5.4.4.2 Deconvolutions The results of a typical deconvolution trial using simulated data are shown in Figure 5.16, which shows the underlying spectrum, the convolved and noise-corrupted signal, and the spectra recovered using both MEM and TPMEM. Figure 5.17 shows the 237 18 16 i . | i | i | i 14 units) / V //M \ • intensity (arbitrary i - A f A\ -intensity (arbitrary i "c 1 \ / \ 2 / \ / \ -n * D ... -^J " . I . I . I . I . u 20 40 60 80 100 Channel Figure 5.16. The results of a typical deconvolution trial on simulated data. (A) The underlying, simulated Raman spectrum. (B) The convolved and noise corrupted spectrum (noise standard deviation = 3% of maximum signal height). The spectra recovered using (C) TPMEM, and (D) MEM. averaged values of RMSERR and 1-r for MEM and TPMEM at various noise levels. Each bar represents the average of 10 deconvolutions, and the error bars represent the standard error of the mean. TPMEM is clearly an effective method of deconvolving instrumentally distorted spectra and appears to show some improvement over MEM. It is difficult to compare the methods on the basis of convergence time, as this depends on the choice of several parameters including: the starting guess for x, the initial upper and lower guesses for the Lagrange multipliers X and p, the convergence tolerances Noise a Figure 5.17. The averaged values of (A) RMSERR, and (B) 1-r (where r is Pearson's correlation coefficient), for the deconvolution trials at noise levels of 1%, 2%, 3%, 4%, and 5%. The error bars represent one standard error of the mean (N=10) in both cases, on the routines to minimize equations 5.7 and 5.10, the convergence tolerances on the routines to solve for X and p, the minimization algorithm used, the compiler used, and the 239 computer hardware used, among others. These dependencies notwithstanding, it may be said that for similar choices for all of these parameters, TPMEM is generally faster than MEM, probably due to the existence of the extra term in the equation to be minimized for MEM. For example, for the 3% noise deconvolution trials, the average convergence time for TPMEM was 104 seconds, while for MEM it was 164 seconds, an increase of 62%. While this represents a significant improvement, TPMEM is still too slow to use in a real time application without considerable effort to optimize the algorithm and its parameters and/or running on a more efficient hardware platform. A demonstration of the resolution improvement possible with 600 point 700 800 900 1000 1100 1200 1300 Raman Shift (cm"1) Figure 5.18. An example of the use of TPMEM for deconvolution on experimental Raman data. (A) The Raman spectrum of saturated KNO3 (aq.) excited with 50 mW of 514.5 nm power and integrated for 1 second. (B) The spectrum of A deconvolved using TPMEM. experimental data using TPMEM is shown in Figure 5.18. It can be seen that the resolution of the NO3" peak near 1050 cm"1 has been considerably improved by TPMEM. The FWHM of this peak changed from 13.4 cm"1 to 10.1 cm"1 after processing. As well, a 240 considerable reduction in noise and increase in peak height is observed. It should be noted that further resolution enhancement could be obtained by using an estimated IPSF that was wider than the measured IPSF (i.e. over deconvolving), indicating that this method could also be used in a manner analogous to FSD. 241 Chapter 6. Biochemical Investigations 6.1 Introduction to FO-UVRRS Biochemical Investigations 6.1.1 Summary of Biochemical FO-UVRRS Investigations UVRRS has the potential to become a sensitive, specific and versatile bioanalytical and biophysical technique for routine investigations of proteins, DNA, and their monomeric components, as well as a variety of smaller, physiologically important aromatic molecules (e.g. the catecholamine neurotransmitters). The transition of UVRRS from a complex, specialized spectroscopic method to a common laboratory assay depends upon several developments, including a robust sample introduction method permitting routine, in situ analysis in standard laboratory environments. To this end, we recently developed the first fiber-optic probes suitable for deep-UV pulsed laser UVRRS [Chapter 4; Greek et al., 1996A; Greek etal., 1996B; Schulze et al., 1997A; Schulze, 1996]. In this chapter, this work is extended by demonstrating the applicability of such probes to studies of biochemical relevance. In particular, in section 6.2 the usefulness of the probes is validated by investigations of the resonance enhancement profiles of phosphotyrosine compared with L-tyrosine, thermal denaturation of RNase T.l, and specific and non-specific protein binding to substrates and surfaces. The probes are used to obtain information of significance in biochemistry, biophysical chemistry, and biotechnology in section 6.3, where the binding of Cellulose Binding Domain (CBD) proteins to their substrates is investigated in more detail, and new conclusions regarding these interactions are drawn. The advantages, disadvantages and problems of the probes are discussed with reference to sample conditions and probe design considerations. 242 6.1.2 Motivation for FO-UVRRS Biochemical Studies Determination of protein and DNA structure is a problem that arises in a wide variety of problems and applications in the fundamental and applied biomedical sciences. Several spectroscopic methods can be applied to this problem including x-ray diffraction, nuclear magnetic resonance, derivative UV/VIS absorption spectroscopy, infrared absorption, electronic circular dichroism, vibrational circular dichroism, and several forms of Raman spectroscopy [Chapter 1; Havel, 1996A]. Raman spectroscopy directly provides the frequencies of Raman-active molecular vibrational modes through the measured shifts in energy of scattered photons as compared with incident photon energy. The scattering cross-section for Raman events can be increased dramatically if excitation is effected using photons with energies in resonance with electronic transitions of the chromophore involved in the vibration. For many biologically relevant chromophores, these transitions are in the ultraviolet region of the spectrum, hence ultraviolet resonance Raman spectroscopy (UVRRS). For example, with 220 nm excitation, the ca. 1604 cm"1 Vs, in-plane ring-stretch vibration of phenylalanine is enhanced by a factor of approximately 106 compared with 1064 nm excitation [calculated from data in Asher et al., 1986]. As discussed in Chapter 1, UVRRS of aromatic amino acids can act as both an analytical technique and a sensitive probe of physico-chemical environment, especially hydrogen bonding and hydrophobic interactions [Miura et al., 1988; Austin et al, 1993A]. UVRRS of the protein amide backbone vibrational modes provides quantitative secondary structural information [Copeland and Spiro, 1987; Song and Asher, 1989]. Selectivity is 243 conferred upon the technique from both the uniqueness of the vibrational frequencies (and hence the Raman shifts) of the molecular vibrational modes as well as the different patterns of resonance enhancement. Unfortunately, while the technologies for long wavelength (> 300 nm) fiber-optic Raman spectroscopy [Plaza et al, 1986; Myrick et al., 1990; Cooney et al., 1996A; Cooney etal., 1996B] and non-fiber-optic UVRRS [e.g. Thamann, 1996; Austin etal., 1993A; Asher, 1993A; Asher, 1993B] have made great strides in the past 15 years, the combination of fiber-optic probes and UVRRS has only recently been accomplished [Chapter 4; Greek et al, 1997B; Greek et al, 1996A; Greek et al, 1996B; Schulze, 1996]. This has been made possible through the availability of new optical-fibers [Chapter 3; Greek et al, 1997B Klein et al, 1997; Klein et al, 1996; Karlitschek et al, 1996B; Karlitschek et al, 1997] with superior abilities for transporting high intensity UV energy, as well as a more complete understanding of the operation of fiber-optic probes in highly absorbing media. The optical properties of silica fibers and biological samples still present limitations in the use and quality of data obtainable using these probes; however progress is being made in overcoming the remaining obstacles. Chapter 6 presents our experiences with the use of these probes in our efforts to further develop their efficacy to the point where they can be incorporated in a versatile instrument for rapid, routine studies of protein and DNA under a variety of experimental conditions. Many of the advantages of fiber-optic UVRRS are the same as those for other remote, fiber-optic-linked spectroscopic techniques, including that (1) no sample alignment is necessary, (2) the sample can be contained in an opaque vessel, (3) relative 2 4 4 motion between sample, laser, and spectrometer does not present problems, (4) samples can be in environments of extreme temperature, radiation, or biohazard, (5) by using the sample container of another instrument (e.g. a UV-VIS spectrophotometer cuvette) UVRRS can easily be performed on the same sample in parallel with other analytical methods, and (6) small sample volumes can easily be used. In addition, certain advantages of using fiber-optic probes are specific to UVRRS. For example many of the sample introduction techniques commonly used with non-fiber-optic UVRRS, including the use of 135° backscattering, rotating sample cells, and flow systems, are not easily applicable to standard laboratory conditions. Despite these many advantages, there still exist certain disadvantages to using fiber-optic probes for in situ UVRRS. While the 90° collection geometry described in Chapter 4 represents a significant improvement over flush fiber-optic probe designs [Chapter 4; Greek et al., 1996A; Greek et al., 1996B] and provides a reduction in the total optical path length through the sample compared with typical non-fiber-optic visible-Raman schemes, it usually suffers more from sample self-absorption than do non-fiber-optic backscattering techniques. In practice, this limits the maximum sample extinction coefficient and reduces the dynamic range of the instrument [see Chapter 4]. For example, high quality protein spectra can be obtained at 230 nm in under 10 minutes using the present instrument with protein concentrations only in the range of ca. 10 p.g/mL to 1 mg/mL. While a single-bifurcated-fiber probe effecting backscattering could, in principle, overcome this problem, in preliminary tests a prototype of this design suffered from severe interference and SNR degradation arising from silica Raman scattering and Fresnel 245 reflection at the probe/sample interface. Further work is needed in the area of probe design to improve SNR and dynamic range. The second serious disadvantage of fiber-optic UVRRS stems from the poor throughput of even the photosensitization-resistant fibers at wavelengths shorter than approximately 220 nm [see, e.g., Figure 3.6] when using high intensity pulses. With the present low-duty-cycle system, this severely limits the power available at the sample at these wavelengths; however this problem should be largely mitigated (if not eliminated) by new cw and quasi-cw lasers. 6.1.3 Material and Methods for Biochemical FO-UVRRS The UVRRS system has been described in chapters 2 and 4; briefly, the doubled, Nd: YAG-pumped pulsed dye laser system provided 20 Hz, 3 ns pulses tunable from 207 nm to 250 nm. The monochromator/detector system consisted of a 0.67 m single monochromator incorporating a 3600 G/mm grating and an IPDA detector. Standard UV-grade and solarization-resistant-UV-grade optical fibers used in the fabrication of Raman probes were obtained from Polymicro Technologies Inc. (Phoenix, AZ). Absorption spectra were obtained using a Milton Roy Spectronic 601 spectrophotometer. The fiber-optic probes used in these studies (Figure 6.1) consisted of a single, solarization-resistant fiber, cleaved and polished (sometimes with a lens), aligned with, and attached to, a second collection fiber incorporating a 45° polished face and a reflective surface. Flush probe designs, such as those commonly used for fiber-optic visible Raman spectroscopy, were found to be unsuitable for UVRRS due to the longer average path lengths of excitation and scattering involved, and hence the increased problem of inner 246 filtering (sample self-absorbance). Designs incorporating standard UV-grade fused-silica excitation fibers are plagued by solarization problems which severely limit the average power delivered to the sample. Standard UV-grade fibers are used as collection fibers since the peak power in the collection fiber is extremely low. The probe used to collect the data in this paper incorporated 600 um (diameter) fibers for both collection and excitation unless otherwise specified. Figure 6.1. Expanded view of fiber-optic UVRRS probe tip. A - photosensitization-resistant excitation fiber, B - standard UV-grade collection fiber, C - reflective surface, D - sample volume. Arrows indicate the general optical path. 247 6.2 Probe Validation and Performance Characterization 6.2.1 RNase T l Denaturation RNase T l is a small (104 amino acid) single domain protein concerning which there exists a large amount of structural and thermodynamic information [Hu et al., 1992; Schmid et al., 1996; Garrett, et al., 1996] hence, it is often used in studies of protein folding, stability, and denaturation. In many studies the protein is unfolded thermally (denaturation temperature at pH 7 ~ 50 °C) or chemically and monitored using the intrinsic fluorescence of tryptophan excited near 280 nm and measured near 320 nm. Since RNase T l contains only one tryptophan residue (the fluorescence of which is reduced by approximately 80% upon unfolding) it is convenient to detect unfolding in this manner. However, a technique such as fiber-optic UVRRS has the potential to provide a large amount of additional structural and environmental information concerning not only the single tryptophan residue, but also tyrosine residues and the overall secondary structure (via the amide modes) in situ, and possibly in parallel with other techniques. RNase T l was dissolved at a concentration of 200 pg/mL in 1.5 mL of 40 mM NaClCu in 2 mL flat bottom microcentrifuge tubes. FO-UVRRS data excited at 230 nm and integrated for 30 minutes were obtained at both 24 °C and 87 °C using approximately 70 pJ/pulse into the input end of the excitation fiber (corresponding to approximately 500 pW at the sample). Figure 6.2 shows the results scaled so that the area of the 933 cm'1 C104" internal standard peak is the same in each (sloping background at ca. 1640 cm"1 water bend mode subtracted). It is clear that the cross sections of the tryptophan bands near 1006, 1350, and 1555 cm"1 are reduced considerably upon denaturation, while 248 tyrosine modes at ca. 1175 and 1615 cm"1 show little change. Furthermore, band analysis of the environmentally sensitive ca. 1350 cm"1 doublet indicates that the area ratio of the 1360 to 1340 cm"1 components is reduced from 2.2 to 1.2 upon denaturation. This is consistent with a significant reduction in the hydrophobicity of the tryptophan environment [Miura et al., 1988], as would be expected upon denaturation. Figure 6.2. Fiber-optic UVRRS data of native (top trace, 24° C) and denatured (bottom trace, 87° C) RNase TI (200 ug/mL) excited at 230 nm. The peak near 933 cm"1 is the perchlorate internal standard. Although these results show that fiber-optic UVRRS could prove to be a useful tool for in situ studies of protein denaturation, and stability, the real utility of fiber-optic UVRRS in these studies would come from quantification of protein secondary structure using the resonance enhanced amide vibrations obtained with excitation wavelengths 249 shorter than about 210 nm. Figure 6.3 shows fiber-optic UVRRS data of RNase TI obtained using 208 nm excitation. While this may demonstrate that fiber-optic probes can be used to detect the structurally sensitive amide bands in situ, it also shows that instrumental effects presently preclude their use for rapid, routine analysis of protein secondary structure. However, with the advent of new cw and quasi-cw deep UV lasers that will allow both higher throughput and smaller probes (hence less severe inner-filtering effects), as well as with improvements in probe design, it is anticipated that acquisition times and SNRs comparable to those obtainable at 230 nm will soon be the norm at much shorter wavelengths. co S "E cio; 800 1000 1200 1400 Raman Shift (cm"1) 1600 1800 Figure 6.3. Fiber-optic UVRRS data of RNase TI (50 iig/mL, 20 minutes integration) excited at 208 nm. Average power at the sample was approximately 50 uW. The ca. 1640 cm"1 water bend peak and sloping background have been subtracted. 250 6.2.2 Resonance Enhancement of Phosphotyrosine The phosphorylation of tyrosine residues by protein tyrosine kinases and phosphotyrosine phosphatases is known to be an extremely important event in signal transduction and regulation of many intracellular processes, including growth, differentiation, metabolism, cell cytoskeletal function, T-cell action, bone maintenance, learning and memory [Kefalas et al, 1995; Tonks and Neel, 1996; Zenner etal, 1995]. Relatively little is known about the mechanisms and pathways for many of these complex processes. The development of a rapid, sensitive, and selective method to detect tyrosine phosphorylation could potentially provide a new avenue to help elucidate mechanisms that underlie biochemical pathways that involve protein phosphorylation. It was with this motivation that we began an investigation to determine if a UVRRS-based instrument could fill such a role. UVRRS data of aqueous L-tyrosine (Y) and O-phospho-L-tyrosine (OPLY) in the wavenumber range of approximately 800 to 1700 cm-1 were obtained at wavelengths from 208 nm to 237.5 nm using the fiber-optic probe described above. Samples ranging in concentration from 50 pM to 500 pM (depending on wavelength) and volumes of 1.5 mL were used at room temperature with continuous agitation provided by a 7 x 2 mm magnetic stir bar. The samples were unbuffered and the pH was generally between 5.0 and 6.5. The main features present in both the Y and OPLY spectra were the v 9 a (ca. 1176 cm' l), v 7 a (ca. 1210 cm"1), and v8a/gb (overlapped at ca. 1610 cm"1) bands. Figure 6.4 shows typical spectra at 227.5 nm. Raman scattering cross-sections were determined from peak areas using ratios with the 933 cm"1 perchlorate signal, which acted as an internal standard 251 of known cross-section and concentration. A first-approximation analysis indicated that the difference in sample absorbance between the 933 cm"1 internal standard and the measured Raman bands produced at most a 1.5% error in the measured cross-sections; therefore no absorption correction was applied. Figure 6.5 shows absorption spectra and RREPs of Y and OPLY. The RREP of Y obtained in these studies matches the literature data [Johnson et al, 1984; Asher et al, 1986; Su et al, 1990] both quantitatively and qualitatively from 237.5 nm to approximately 225 nm. At shorter wavelengths the measured cross-section continues to decrease as opposed to the continued increase reported in the literature for decreasing wavelength. A purely instrumental explanation for this difference was initially ruled out because (a) the internal standard signal did not show any unusual behavior, and (b) the measured OPLY cross-section continued to increase down to 215 nm. While the pH in these studies differed from the literature data, this is unlikely to be the major source of the discrepancy since tyrosine has no titratable hydrogens with pKa values in the range of 5-7. Saturation or photodegradation are also unlikely candidates owing to the fact that the laser irradiance at the sample was not excessive (< 4 MW/cm2 at 215 nm). Since the source of this discrepancy remains to be determined, we will consider only the data from 225 nm to 237.5 nm. 252 It is clear from the RREPs that the resonance enhancement profiles of Y and OPLY are significantly different. For example, at 233.75 nm, the cross-section of the ca. 1176 cm-1 band is four times higher in Y than in OPLY. Further, there exist differences in the wavenumber shifts of certain Raman bands in OPLY compared with Y. For example, the Vo a band was found to be centered at 1177.4 cm"1 in Y and 1174.0 cm-1 in OPLY. The 30000 800 1000 1200 1400 Raman Shift (cm-1) 1600 Figure 6.4. Fiber-optic UVRRS data of L-tyrosine (top trace) and O-phospho-L-tyrosine (bottom trace) excited at 227.5 nm. Smoothing has been applied for clarity. The peak near 933 cm'1 is the perchlorate internal standard. v 7 i band was shifted up 2.4 cm"1 from 1209.0 in Y to 1211.4 in OPLY; while the peak of the overlapped Vga/sb bands was shifted down 3.6 cm' 1 from 1615.6 cm'1 in Y to 1612.0 cm"1 in OPLY. All of these differences were significant at the a=0.05 level of significance using a minimum of eleven degrees of freedom in a two-tailed Student t-test. However, although they are statistically significant, they are relatively small shifts and would 253 therefore be difficult to use to differentiate between Y and OPLY in practice. Using the differences in RREPs would probably be a better approach to this problem. 8000 H 3001 240 200 210 220 230 wavelength (nm) 240 200 210 220 230 wavelength (nm) 240 Figure 6.5. (A) Absorption spectra of L-tyrosine (solid line) and O-phospho-L-tyrosine (dotted line). RREPs showing cross sections of the (B) ca. 1176 cm"1, (C) ca. 1210 cm"1, and (D) overlapped 1600/1610 cm"1 bands of tyrosine (solid line, square symbols) and O-phospho-L-tyrosine (dotted line, circular symbols). 6.2.3 Analytes Bound to Insoluble Substrates 6.2.3.1 Introduction to Insoluble Substrate Binding Studies The binding of a protein to its substrate is a ubiquitous event in biochemical systems. Specific binding involves the recognition by a specific region of a protein for a certain substrate, while non-specific binding (or adsorption) refers to a more general process in which the protein adheres to a surface with no exactly specified or required 254 orientation or points of contact. Investigations of binding in which the protein, substrate, and resulting protein-substrate complex are all soluble are investigated in section 6.4. As discussed in Chapter 1, one of the author's main research interests is the determination of changes in protein structure upon binding of soluble proteins to insoluble substrates using UVRRS. Fiber-optic probes greatly facilitate such studies; however they also incur at least four experimental problems: (1) diffuse and specular reflection of excitation light into the collection fiber from the substrate surface can severely degrade the SNR due to excess stray light, (2) silica Raman scattering from the excitation fiber can be reflected into the collection fiber, (3) it can be difficult to get enough total substrate surface area in the sample volume, and (4) Raman signals from the substrate can interfere with the protein signals. Problem (1), above, was addressed to some extent by incorporating a 1 cm path length cell containing 160 pM acenapthene dissolved in methanol at the entrance to the spectrometer. The molar absorptivity of acenapthene changes very rapidly in the range of approximately 220 to 240 nm and by adjusting the concentration, it can be used to remove unwanted Rayleigh light without significantly attenuating the desired signal. Problem (2) proved to be more troublesome and could only be addressed by obtaining spectra of a blank containing only the insoluble substrate. This could later be subtracted from the raw signal spectrum, or at least used to determine which peaks came from the protein. Problem (3) was successfully overcome by using suspensions of small (ca. 1-10 pm) particles of the substrate. For the substrates under consideration in this study (silica and cellulose), interfering signals as described in problem (4) above were not encountered. 255 6.2.3.2 CBD Binding To Cellulose The model system used in these studies for specific binding was the cellulose binding domain of an exoglucanase of Cellulomonas fimi (CBDcex) and its substrate, cellulose. CBDcex is known to bind irreversibly and with a strong affinity for cellulose [Ong etal., 1991; Ong etal, 1993; see also section 6.3]. Cellulose particles (avicel, 3.3 mg/mL, diameter ca. 10 um) were suspended in 1.5 mL of aqueous CBDcex (1 mg/mL) and allowed to equilibrate with continuous agitation for at least 12 hours, after which the cellulose was spun down at 14 000 rpm for 4 minutes. The supernatant was removed and the avicel was resuspended in 1.5 mL of 50 mM KC1. This was repeated three times, after which no protein could be detected in the supernatant using 280 nm absorption spectrophotometry. UVRRS was effected at 230.5 nm using the probe described in section 6.1.3, above. The equivalent concentration of bound CBD in the suspension was estimated at 50 ug/mL and there was effectively no unbound CBD. The substrate was kept continually suspended using a small magnetic stir bar. Figure 6.6 shows the spectra of free CBDcex and CBD C e x bound to avicel. Good quality data could not be obtained below 800 cm"1 because of Rayleigh and silica interference. Clearly visible in the bound spectrum are tryptophan signals at ca. 1010, 1350, and 1555 cm"1, as well as a tyrosine signal near 1615 cm"1. 256 1555cm"1 600 800 1000 1200 1400 1600 1800 2000 2200 Raman Shit (cm"1) Figure 6.6. Fiber-optic UVRRS data obtained using 230.5 nm excitation of free aqueous CBDcex (top trace) and CBDcex bound to avicel. Water and silica backgrounds have been subtracted, and the spectra have been scaled and vertically translated for clarity. 6.2.3.3 Lysozyme Binding to Silica Non-specific binding was investigated using lysozyme adsorbed to silica particles. Several studies have established that lysozyme binds irreversibly to silica [Norde and Favier, 1992; Norde and Anusiem, 1992] and it is often used as a model system to investigate protein adsorption. In the present study, 500 pg/mL lysozyme was allowed to come to equilibrium in a 40 mg/mL silica suspension (80% of silica particles had a diameter between 1 and 5 pm) for at least 12 hours. After adsorption, the same rinsing, resuspension, and UVRRS procedures were followed as for the CBD/avicel system. 257 Figure 6.7 shows the spectra of free and silica-adsorbed lysozyme. Clearly visible in the bound spectrum are the ca. 1010, 1350, and 1555 cm"1 tryptophan bands. Both the CBD/avicel experiment and the lysozyme/silica experiment indicate that it is possible to detect UVRRS signals from proteins bound to insoluble substrates using fiber-optic probes. At present SNRs are not sufficient for quantification of protein structure, however this preliminary data demonstrates that this goal is, in principle, achievable. 1555 cnT1 1009 cm"1 : Figure 6.7. Fiber-optic UVRRS data obtained using 230.5 nm excitation of free aqueous lysozyme and lysozyme adsorbed to silica. Water and silica backgrounds have been subtracted, and the spectra have been scaled and vertically translated for clarity. 258 6.2.4 FO-UVRRS Data of DNA The RREPs of the DNA nucleotides exhibit several maxima in the deep UV [Fodor et ai, 1985; Fodor and Spiro, 1986]. High quality fiber-optic UVRRS data of salmon-sperm DNA has been obtained at 266 nm (Figure 6.8). In an attempt to enhance the structurally sensitive DNA Raman bands using 216 nm excitation, however, the SNR was severely limited as described above. Further work is needed to improve the SNR to the extent that reliable structural information can be obtained. T 1 1 1 1 1 1 1 r I 1 1 1 1 1 1 i i i i i i i I 900 1000 1100 1200 1300 1400 1500 1600 Raman Shift (cm"1) Figure 6.8. 266 nm Fiber-optic UVRRS data of 50 ug/mL salmon-sperm DNA obtained using flush-geometry probe incorporating a 300 nm diameter excitation fiber and a 600 um diameter collection fiber. Integration time was 9 minutes. 259 6.3 CBD Studies 6.3.1 Introduction and Motivation for CBD Studies 6.3.1.1 Motivation and Summary of CBD Studies Cellulose [reviewed, e.g., in Lehninger et al, 1993] is an insoluble structural polysaccharide consisting of a homopolymer of D-glucose units connected through pi—>-4 linkages and, of any macromolecule, constitutes the largest fraction of the world's biomass [Rose, 1991]. Although most animals cannot digest cellulose, certain species of bacteria and fungi secrete cellulytic enzymes that breakdown cellulose to its monomeric glucose units which may then be metabolized. Animals that can use cellulose as a food source (e.g. termites, cattle, sheep, goats, camels and giraffes) are generally involved in a symbiotic relationship with such organisms living in their digestive tracts. Cellulases often consist of separate and independently active catalytic and binding domains connected by polypeptide linker rich in proline and threonine (typically ca. 10-20 residues long). The cellulose binding domains (CBDs) bind to the cellulose and act to increase the local concentration of catalytic domains, thereby increasing the reaction rate over the catalytic domain alone [Gilkes et al., 1988]. As well, they may be involved in disrupting the local cellulose structure, thereby facilitating attack by the catalytic domain [Din et al, 1993]. There are two main reasons to study the binding of a CBD to its substrate. It is possible to create fusion proteins of a CBD with another enzyme (or cytokine, growth factor, transport or storage protein) which may then be immobilized on a cellulose substrate and used in an industrial process. Also, as will be discussed in sections 6.2.1.2 to 6.2.1.4, below, CBD-substrate binding is an interesting model system or case study for 260 investigating the specific binding of a protein to a polymeric substrate. The possible importance of aromatic amino acids (especially tyrosine and tryptophan) in their substrate binding make CBDs good candidates for investigation as model systems for UVRRS. In section 6.3, significant preliminary UVRRS results are presented for two CBD proteins. It is shown that fiber-optic UVRRS can be used to provide biophysical information concerning CBDs with integration times as short as 12 minutes. The SNR is still too low to make definitive conclusions based solely on UVRRS data; however when interpreted in the context of available NMR, calorimetric, binding and fluorescence data, they provide important new information. Further, SNR improvements discussed in chapters 4 and 7 should enable fiber-optic UVRRS to tackle more challenging problems with greater rigor in the very near future. 6.3.1.2 Cellulose Binding Domains There are at least 160 known CBDs, all with between 36 and 200 amino acid residues, and these have been divided into 13 families (denoted by Roman numerals I through XIII) based on amino acid sequence similarity [Tomme et al, 1985; Gilkes et al., 1991]. Family I CBDs are produced by fungi, while most others are bacterial. Although most CBDs derive from cellulase enzymes, many also come from xylanases and, to a lesser extent, from hydrolases and other enzymes. CBDs may be classified as coming from endoglucanases or exoglucanases and occurring internally or at the C or N termini of the cellulase enzyme. Only two solution structures and two crystal structures of CBDs have been solved [Xu et al, 1995; Johnson et al, 1996; Kraulis et al, 1989], and there is 261 generally an incomplete biochemical characterization and understanding of the binding process of these biologically and potentially industrially important proteins. 6.3.1.3 CBDcex CBDcex is the 110 residue binding domain from the P-l-4-exoglucanase enzyme Cex produced by Cellulomonas fimi (C. fimi) which bonds to many forms of cellulose substrates with varying degrees of affinity [Ong et al, 1993]. CBDcex is a family JJ CBD that, like many CBDs from this family, has a flat hydrophobic binding face rather than a binding cleft. This face is one side of a P barrel formed from nine of the ten antiparallel P-strands that form the molecule. Binding of CBDcex to cellulose is known to be entropically driven [Creagh et al, 1996] and the exposed TRP residues (W17, W54 and W72) on the binding face have been implicated in the binding process with preliminary data from chemical modification studies [Bray et al, 1996], NMR data [Xu et al, 1995] and mutational studies [Boraston, 1997]. Figure 6.9 shows the structure of CBDcex and points out the aforementioned TRP residues and an important asparagine residue (N24, vide infra). The tryptophans are thought to bind via hydrophobic interactions; however it is also possible that hydrogen bonds could be involved. Furthermore, involvement of these TRP residues has not yet been directly and conclusively established and more information is needed concerning the free and bound tertiary structure of CBDcex vis a vis these TRP residues. FO-UVRRS may be an ideal tool for these investigations since CBDcex has only two other TRP residues, which are buried and therefore the signals from these should not change significantly upon binding. 262 Figure 6.9. Space fill solution structure of CBDcex showing the relevant tryptophan and asparagine residues as wireframes [structure from Xu et al., 1995]. In section 6.2.3, an investigation of the binding of CBDcex to cellulose (avicel particles) using UVRRS at 230 nm was attempted. While tryptophan and tyrosine signals were visible in the spectrum of the bound C B D , SNRs were too low to make quantitative determinations of changes in tertiary structure (with respect to changes in tryptophan environment hydrophobicity and hydrogen bonding). It is expected that improvements wrought from new narrow-band-pass Rayleigh rejection filters, improved probe designs and 224 nm quasi-cw HeAg lasers (see Chapter 7) will, in the very near future, make such quantification possible. Until that time, however, it was decided to investigate another interesting aspect of CBDcex tertiary structure relating to TRP environment and hydrogen bonding. When expressed in yeast, such as Pichia pastoris (as is sometimes desired when CBD fusion proteins are being produced), many proteins can be glycosylated at asparagine residues [Trimble etal., 1991]. CBDcex has 5 such potential glycosylation sites (N24, 263 N29, N75, N87 and N103). It is known that glycosylation at N24 (controlled by mutating out the asparagines at the other four potential glycosylation sites) reduces the binding affinity of CBDcex for cellulose. There are three potential uses for N24 glycosylation: (1) modulation of the binding affinity, (2) applications for elution in chromatography, and (3) as a probe of the CBDcex binding face. Here, FO-UVRRS was used to investigate the interaction of the attached sugar and the binding face in section 6.2.2, below. Since it is known that (a) CBDcex tryptophans are involved in binding (although little is known about their tertiary structure and mechanisms), and (b) glycosylation reduces binding, it interesting to investigate how glycosylation changes tertiary structure (with respect to TRP residues) even in the unbound state. 6.3.1.4 CBDNI CBDNI is the family IV, 152 residue, N-terminal CBD from the endoglucanase CenC of C. fimi. CenC has two CBDs in tandem (NI, N2); however these are known to be independently active. CBDNI consists of ten P-strands forming two anti-parallel f}-sheets with a five stranded binding cleft consisting of a central strip of hydrophobic residues flanked on either side by polar (potentially hydrogen bonding) groups [Johnson et al., 1996A; Johnson et al., 1996B]. Figure 6.10 shows the solution structure of CBDNI. CBDNI is unique in that it does not bind to crystalline cellulose; however it is one of the few CBDs known to bind to soluble oligosaccharides (cellotriaose and longer polymers) to an appreciable extent. In contrast to CBDcex, this binding is known to be enthalpically driven [Tomme et al., 1996] through favourable hydrogen bonding or Van der Waals 264 W137 Figure 6.10. Ribbon diagram solution structure of C B D N i showing the relevant tyrosine and tryptophan residues as black wireframes [structure from Johnson et al, 1996A], interactions. Further, binding free energy change magnitude is known to increase with oligosaccharide length from cellotriaose to cellopentaose, after which it becomes constant, indicating that cellopentaose just fills the entire binding cleft [Tomme et al., 1996]. There are six TYR and two TRP residues in CBDNI and four of these (W16, Y19, Y43 and Y85) are on or near the binding cleft and show significant NMR chemical shifts upon oligosaccharide binding, possibly implicating them in the binding mechanism [Johnson et al., 1996A; Johnson etal, 1996B; Tomme etal, 1996]. Further, mutation of either Y19 or Y85 to alanine (which is still hydrophobic, but, unlike TYR, has no hydrogen bonding capacity) is known to reduce CBD N i affinity for cellopentaose by a factor of approximately 100 [Kormos, 1997]. UVRRS may be able to provide useful data to improve our understanding of CBD N i oligosaccharide binding vis a vis W16, Y19, Y43 and Y85. 265 In the studies reported in section 6.2.3, below, the involvement of these residues in the binding of C B D N i to its substrate was investigated. This was done by considering wild type CBDNI (WT) and two mutants (Y19 or Y85 mutated to alanine, denoted by Y 1 9 A and Y85 A ) binding to cellopentaose. I also look at differences in tertiary structure between the various mutants in the absence of oligosaccharide. 266 6.3.2 CBDcex Studies 6.3.2.1 Materials and Methods for CBDcex Studies Wild-type CBDcex (CBDcexWT), and glycosylated and unglycosylated CBDcex (denoted CBDcexGH- and CBDcexG-, respectively) mutants were obtained. In the mutants, all possible asparagine-containing yeast glycosylation-recognition sequences (ASN-X-(SER or THR), X * PRO) except the N24 glycosylation site were mutated at one amino acid to a non-aromatic residue, thereby ensuring that the protein was glycosylated only at N24 and that no additional aromatics were added. N-glycosylation by Pichia pastoris is known to produce linked oligosaccharides with up to three branches and ca. 10 mannose and glucose residues [Trimble etal., 1991]. The mutants all contained a six-histidine (H6) residue tag which aided in purification. FO-UVRRS data of each of these were obtained at 227 and 230 nm using ca. 400 pW of average power and ca. 30 minute integration times. The concentrations were 20 p M in all cases. All spectra were calibrated using the spectra of ethanol taken before and after each experiment. At 227 nm, an 80 mM Na2S04 internal standard was used providing an internal reference intensity signal at ca. 984 cm"1. 267 6.3.2.2 C B D c e X Results and Discussion Figure 6.11 shows the UVRRS spectra of C B D c e x G + and C B D c e x G - at 227 nm scaled so that the internal standard has the same height in each. The 230 nm data is similar, but of lower SNR, and is not shown. Clearly visible in the spectra are the sulfate internal standard line at ca. 984 cm"1; the tryptophan modes W 1 8 (ca. 760 cm"1), W 1 7 (ca. 880 cm"1), W 1 6 (ca. 1010 cm"1), W 1 0 (ca. 1240 cm"1), W 7 (ca. 1350 cm"1), and W 3 (ca. 1555 cm"1); and tyrosine modes vga/vgb (overlapped at ca. 1610 cm"1). The line at ca. 1480 cm"1 is difficult to assign, but may be from tryptophan ( W 5 ) , or histidine. Additionally, the ca. 850 cm"1 and 1210 cm"1 (v7a) tyrosine lines may be visible. It is difficult to explain the absence of the usually strong tyrosine v 9 a line that is normally found at ca. 1175 cm"1. The W3 s so42 W 1 6 I j v ° - ' )unts) W 1 8 / f * \j j : CO : W 7 Inter .-• W 1 7 i j : v 8 . / v 7 . : : ' i I . I . I . i 1 . 1 600 800 1000 1200 1400 Raman Shift (cm") 1600 1800 2000 Figure 6.11. Fiber-optic UVRRS data at 227 nm of C B D c e x G - (dotted) and CBDcexCri - (solid). A five point low pass FFT filter was applied to remove high frequency noise. Data was scaled to equal sulfate peak intensities. 268 presence of a large number of histidines due to the H6 tag used in purification was problematic because of both the possible HIS signal near 1480 cm"1 and the increased inner filtering effects owing to the increased molar absorptivity stemming from the increased HIS content. Viewing the overall appearance of the spectra, the most immediately evident change upon glycosylation in both the 227 and 230 nm spectra is the reduction in intensity of the tryptophan lines with respect to both the internal standard and the tyrosine lines. This reduction is on the order of about 50% for the 227 nm spectra and 20 % at 230 nm. Reductions in tryptophan signal intensity are correlated with a decrease in tryptophan hydrophobicity and/or the breaking of a hydrogen bond for at least one of the five C B D c e x tryptophan residues [Liu et al, 1989; Rodgers etal, 1992; Austin etal, 1993A]. Considering the magnitude of the change and the locations of the C B D c « TRP residues (two of which are buried), it is likely that the changes are occurring to at least two of W17, W54, and W72. Another curious and presently unexplainable feature of these spectra is the absence of the TYR Fermi doublet at ca. 850 cm"1 in the C B D c e x G + spectra as compared with the C B D o x G - spectra. 269 870 cm '•884 cm"1 J i : i i i i i i . 840 860 880 900 920 Raman Shift (cm') Figure 6.12. Fiber-optic 227 nm UVRRS data of C B D c e x G - (dotted) and C B D c e x G + (solid) from Figure 6.11 expanded to show detail between 830 and 930 cm"1. To delve further into the effects of N24 glycosylation on the tertiary structure of CBDcex , individual areas of the 227 nm spectra were examined in more detail. Figures 6.12, 6.13, and 6.14 show the regions at 830-930 cm"1, 1300-1400 cm"1, and 1400-1650 cm"1, respectively. Apart from the unexplainable absence of the TYR Fermi doublet in Figure 6.12, this figure also indicates a large shift (from 870 cm to 884 cm"1) in the W17 line. The corresponding wavenumbers in the 230 nm spectra were 873 and 879 cm'1, a smaller but still significant shift. Lower W17 wavenumbers indicate stronger hydrogen bonds at the indole nitrogen [see, e.g., Miura et al., 1988; Austin et al., 1993A], and a shift of this magnitude is strong evidence that a moderate to strong hydrogen bond is completely broken upon glycosylation. Since no substrate (cellulose) is present in these experiments, it must be an intramolecular hydrogen bond. 270 • 1348 cm"1 -4—< linoo) \ 1365 cm"1 J—> Intensi - \ i 1345 cm"1 . ™ - i \ 1362 cm i . i . i *. .-' 1300 1320 1340 1360 1380 1400 Raman Shift (cm") Figure 6.13. Fiber-optic 227 nm UVRRS data of C B D c e x G - (dotted) and CBDc=xG+ (solid) from Figure 6.11 expanded to show detail between 1300 and 1400 cm"1. The TRP Fermi doublet region is shown in greater detail in Figure 6.13. The extremely high intensity of the ca. 1345 cm"1 component of the doublet relative to the ca. 1365 cm"1 component in C B D c e x G - is again indicative of strong indole nitrogen hydrogen bonding, and possibly a hydrophilic environment [Austin et al., 1993A; Miura et al., 1988; Jordan et al., 1995]. A very hydrophilic environment (e.g. aqueous tryptophan) alone is known to result in an I1365/I1340 ratio of about 1:1, which is the observed ratio for the C B D c e x G + . It is unlikely, then, that the change in I1365/I1340 upon glycosylation is due to a more hydrophilic environment, but rather this is further evidence that there is a TRP indole nitrogen hydrogen bond present in C B D c e x G - that is not present in C B D c e x G + . 271 The bond-angle sensitive W3 line at ca. 1555 cm"1 and the H-bond sensitive overlapped TYR v 8 a/v 8 b lines at ca. 1610 cm'1 are shown in greater detail in Figure 6.14. Apart from the aforementioned intensity reduction, W3 remains at the same wavenumber for both C B D c e x G - and CBDcexG+, indicating that the TRP residues remain at approximately the same average side chain torsional bond angle, although there is a low frequency shoulder in the C B D c e x G - spectrum that does not appear in the C B D c s x G + spectrum. The v8a/v8b envelope has a maximum at 1610 cm"1 for C B D c e x G - and double maxima at 1606 and 1616 cm"1 for C B D C e x G + . to 13 to c 0) 1555 cm"1 1610 cm" 1616 cm" 1553 cm'1 1500 1520 1620 1640 1540 1560 1580 1600 Raman Shift (cm") Figure 6.14. Fiber-optic 227 nm UVRRS data of C B D c « G - (dotted) and C B D c c x G + (solid) from Figure 6.11 expanded to show detail between 1500 and 1650 cm"1. 6.3.2.3 Conclusions from CBDcex Glycosylation Studies The fiber-optic UVRRS data points to the existence of a TRP hydrogen bond C B D c e x G - that is broken in C B D c e x G + . This is likely due to some combination of hydrophobic interaction between the sugar and TRP residues, and steric hindrance. 273 6.3.3 C B D ™ Studies 6.3.3.1 Materials and Methods for C B D N i Studies Wild-type C B D N I and Y85A and Y19A C B D N i mutants were obtained and diluted (in distilled water) to a concentration of 150 pg/ml. FO-UVRRS data were obtained from ca. 1.5 ml sample volumes with a 40 mM MgS04 internal standard at 230 nm using twelve minutes of integration time and ca. 500 pW of optical power at the sample. All samples were continuously agitated using a 7 x 2 mm magnetic stir bar. The twelve minutes of integration consisted of eight independent 90 second acquisitions; no significant differences were visible for any sample within these eight acquisitions. These experiments _ i i i i i i i i i 800 1000 1200 1400 1600 Raman Shift (cm"1) Figure 6.15. Fiber-optic UVRRS data of free and cellopentaose-bound C B D N I and mutants: ( A ) free C B D N 1 W T , ( B ) free C B D N i Y 1 9 A , ( C ) free C B D N I Y 8 5 A , ( D ) C B D N I W T and cellopentaose, ( E ) C B D N I Y l 9 A and cellopentaose, and (F) C B D N I Y 8 5 A and cellopentaose. Some Raman peaks have been labeled. The sloping background and water signal have been subtracted and a five point low-pass FFT filter has been applied. 274 were repeated with identical, fresh samples containing 2.5 mM cellopentaose. With this concentration of oligosaccharide, ca. 100% of the WT CBDNI would be bound to substrate and ca. 50% of the Y85A and Y19A CBDNI mutants would be bound. The relatively short total integration time of 12 minutes was chosen for three reasons: (1) ensuring that photo-damage to the protein did not occur, (2) as proof-of-concept for a rapid in situ tertiary structure assay, and (3) because only limited amounts of the protein were available, the experiments could not easily be repeated multiple times. Spectra were also taken of water, 40 mM MgS04, and 2.5 mM cellopentaose in 40 mM MgSCv 6.3.3.2 CBDNI Results and Discussion Figure 6.15 shows the 230 nm fiber-optic UVRRS data of CBDNiWT, Y85A and Y19A with and without the addition of 2.5 mM cellopentaose. Figure 6.16 shows the 14000 12000 - \ Intensity (counts) 4000 2000 600 800 1000 1200 1400 1600 1800 Raman Shift (cm"1) Figure 6.16. Fiber-optic UVRRS data of distilled water (bottom), 40 mM aqueous MgS0 4 (top) and 2.5 mM cellopentaose and 40 mM MgSCv (middle). The water spectrum has been smoothed with a five point low-pass FFT filter and the spectra have been offset for clarity. All spectra taken at 230 nm with 90 second integration. 275 spectra of water, 40 mM MgS0 4 and 2.5 mM cellopentaose in 40 mM MgS04. As expected, water shows only the broad peak at around 1640 cm"1 on a sloping stray-light Rayleigh background. The 40 mM MgSCv shows only the additional ca. 984 cm'1 peak used as an internal standard, and the 2.5 mM cellopentaose shows no additional signals, as sugar vibrations are not resonance enhanced. Thus, any changes in the spectra upon the addition of the oligosaccharide derive from changes in the TRP or TYR signals. No silica features were observed in any of the spectra. The C B D N i spectra show, in addition to the internal standard peak (the water peak was subtracted) the TRP lines W18, W17, W16, W10, W7, an W3; and the TYR lines v9 a, v7a, and v8 a/v8 b. The TYR Fermi doublet at 830/850 cm"1 may also be discerned close to, or as a shoulder of, W17. In observing the six spectra of Figure 6.15, certain information may be obtained by considering changes in the relative intensities of the peaks. In comparing the spectra of the proteins in the absence of oligosaccharide, it is clear that while CBDNI WT and CBDNIY85A are very similar vis a vis relative peak heights, CBD N iY19A has significantly reduced intensities (relative to the internal standard peak) for both TRP and TYR vibrations. Since hydrogen bonding of both TRP and TYR residues results in a red shifted RREP and a large increase in 230 nm scattering cross section, this data is strongly indicative of intramolecular hydrogen bonds in the wild-type protein on Y19 and at least one TRP residue that do not exist in the Y19A mutant. It is logical to conclude that such a hydrogen bond would be between Y19 and a nearby TRP residue (either W16 or W137). In comparing the intensities of the spectra for a given protein with and without the addition of cellopentaose it is clear that TRP signals undergo little intensity changes upon 276 CBD s\ N1 / \ CBDN 1WT "CBDN 1Y19A CBDN1Y85A / \ W T - /1 r \ / \ free / 1 \ I ' r e e / / \ \ free to -c o f 1 \ \Y19A / \ Y 8 5 A • / \C P / \ c p j \ C P I 1 I 1 I 1 I 1590 1605 1620 1635 1 ' 1 ' 1 1 1 1 ' 1 1 1 1 1 1590 1 605 1 620 1635 1590 1605 1620 1635 Raman Shift (cm1) I ' I 1 I 1 I 1590 1605 1620 1635 Figure 6.17. Fiber-optic UVRRS CBDNI spectra from Figure 6.15 in expanded detail showing Vga/vgb region. C P indicates the presence of 2.5 mM cellopentaose. the addition of cellopentaose for both C B D N 1 W T and CBD N iY85A; in contrast, TRP signal intensities are increased upon addition of cellopentaose to CBDNIY19A. This may indicate that the aforementioned exposed TRP residue that is available for hydrogen bonding in CBDNIY19A (but hydrogen bonded to Y19 in C B D N i W T and CBDNIY85A) is able to hydrogen bond to the cellopentaose. Further conclusions concerning changes in tertiary structure upon mutation and binding may be obtained by noting peak shifts in the spectra. Figures 6.17 to 6.21 show various signature regions of the spectra in greater detail and positioned to aid in detecting differences between the three free proteins, and between free and bound states of each protein. Figure 6.17 shows the H-bond-sensitive T Y R vga/vgb regions of the spectra. It 277 should be noted that some of the vga peak positions are unrealistically high (~ 1620 to 1622 cm"1); however this may be partially due to the underlying, weaker TRP Wl mode, and in any case peak shifts are more significant for this work. Apart from a shoulder on CBD N iY19A at ca. 1599 cm"1 (which may indicate a lack of H-bonding compared with the wild type and Y85A mutant, consistent with the discussion above), there are no significant differences between the free protein spectra. The CBDNIWT data show a strong downward shift in at least one component of the Vga peak to 1615 from 1622 cm"1, and a downward shift in the ca. 1595 cm"1 v 8 o shoulder, both possibly indicating the formation of a hydrogen bond to the phenol ring. The CBD N iY19A data show similar downward shifts (to 1615 from 1619 cm"1 for v8a), while the CBD N iY85A peak and shoulder positions remain virtually unchanged at 1617 cm"1. The reader should again be reminded ca. 100 % of the C B D N I W T is bound in the presence of cellopentaose, as compared with only ca. 50 % of the mutants. Therefore, any spectral changes upon binding would be relatively more pronounced in the wild-type spectra. Taken together, these data are evidence that Y19 is not involved in hydrogen bonding to the substrate, but Y85 does form a hydrogen bond to the cellopentaose. 278 Figure 6.18 shows the spectral region between 1150 and 1300 cm"1, which includes the TYR modes v 7 a (ca. 1210 cm"1) v 9 a (ca. 1175 cm"1), and the TRP W 1 0 (ca. 1238 cm"1) mode. These particular modes are relatively poorly resolved and of low SNR, so care must be taken in their interpretation. Although the W 1 0 mode appears to undergo intensity and morphology changes upon mutation and binding (esp. in C B D c e x Y 1 9 A ) , it is not clear if these are significant, and better data is needed. Further, W 1 0 ( C - H + C 3 - C c x t stretch) has not previously been investigated for sensitivity to tertiary structure. More work is needed in this area. Similarly, the TYR v 7 a and V o a modes are known to be sensitive to phenolic hydrogen bonding [Takeuchi et al., 1989]; however the relationships manifesting this sensitivity are not yet fully elucidated. The frequency of Vo» is thought to shift down upon hydrogen bond formation and is also related to C - O H dihedral angle [Takeuchi et al, CBD. ->—i—>—r 1150 1200 1250 CBDN1WT CBDN1Y19A CBDN1Y85A Raman Shift (cm") Figure 6.18. Fiber-optic UVRRS C B D N i spectra from Figure 6.15 in expanded detail showing v 7 a and v 9 a region. C P indicates the presence of 2.5 mM cellopentaose. 279 C B D N 1 Y 8 5 A 1300 1325 1350 1375 14001300 1325 1350 1375 14001300 1325 1350 1375 14001300 1325 1350 1375 1400 Raman Shift (cm1) Figure 6.19. Fiber-optic UVRRS C B D N L spectra from Figure 6.15 in expanded detail showing W7 Fermi doublet region. C P indicates the presence of 2.5 mM cellopentaose. 1989]. The CBDNIY85A undergoes an apparent 6 cm"1 downshift upon addition of cellopentaose compared with a 13 cm"1 downshift for CBDNIY19A. This may also indicate increased direct involvement in binding of Y85 compared with Y19 (as determined above using intensity considerations and Vsa and vsb data); however relatively little weight should be assigned to these particular results because (a) SNR is relatively poor, (b) W10, V7* and v 9 a modes are poorly resolved and overlap, (c) the many components from the six different TYR residues complicate interpretation, (d) significant differences in these bands exist even between the three free proteins, (e) little change is seen for CBDNI WT upon cellopentaose addition, and, most importantly, (f) literature data for interpretation of these results is scarce and not highly informative. 280 Figure 6.19 shows the TRP W7 mode Fermi doublet region from 1300 to 1400 cm"1 in greater detail. Two or three components per TRP residue usually constitute W7 and the ratio of the intensities of the ca. 1360 to ca. 1340 cm'1 signals is known to be sensitive to TRP residue environment hydrophobicity. Again, problems arise in interpretation of W7 due to the large number of unresolved, overlapping peaks. At present, it is not possible to draw any conclusions until higher SNR, higher resolution data is available. Figure 6.20 shows the region from 800 to 900 cm"1, including the T Y R Fermi doublet and the H-bond sensitive TRP W17 mode. The TYR signals are weak and the high frequency (ca. 850 cm"1) component of the TYR Fermi doublet is only poorly resolved as a shoulder on the much larger ca. 875 cm'1 W17 signal. The lower frequency component at ca. 830 cm"1 is better resolved and undergoes changes upon the addition of cellopentaose, which may be an additional indication of the involvement of TYR is substrate binding; however it is of relatively poor SNR and the literature [e.g: Austin et al., 1993] casts some uncertainty on its proper interpretation in UVRRS spectra. In general, the W17 line appears to remain unchanged upon the addition of cellopentaose, indicating that TRP residues are not involved in hydrogen bonding to the oligosaccharide substrate. However, comparison of the W17 mode for each of the three free proteins provides further evidence supporting the argument that Y19 is H-bonded to W16 or W137 in CBDNIWT. CBDNIWT and CBDNIY85A both posses a low-frequency W17 component (a peak and shoulder in C B D N I W T and CBD N iY85A, respectively) at ca. 870 cm"1, indicating the existence of a strongly hydrogen bonded TRP residue. The W17 signal 281 from C B D N I Y 1 9 A has no such component and, in fact, has a high frequency component at ca. 885 cm"1-which is absent from both C B D N i W T and C B D N I Y 8 5 A and indicates the existence of a TRP residue not involved in hydrogen bonding. These data indicate that upon mutation of Y 1 9 to alanine, a TRP residue hydrogen bond is broken. Figure 6.21 shows the TRP W3 mode which is sensitive to side chain torsional angle. There appears to be a significant downshift in W3 frequency upon addition of cellopentaose to C B D N I W T which may indicate the involvement of a TRP residue in substrate binding; however since no significant differences are apparent in any of the other comparisons and SNR is low in any case, little weight can be applied to this conclusion. More work and better data is needed. 800 825 850 875 900 800 825 850 875 900 800 825 850 875 900 800 825 850 875 900 Raman Shift (cm 1 ) Figure 6.20. Fiber-optic UVRRS CBDNi spectra from Figure 6.15 in expanded detail showing W17 region. CP indicates the presence of 2.5 mM cellopentaose 282 6.3.3.3 Conclusions from Fiber-Optic UVRRS CBDNI Binding Studies Three main facts are supported by the fiber-optic UVRRS CBDNi binding data: (a) there is a TRP residue hydrogen bond in C B D N I W T that is present in CBDNiY85 A but not CBDNIY19A, (b) there is a tyrosine hydrogen bond in C B D N I W T that is present in CBDNIY85A, but not CBDNIY19A, (c) tyrosine hydrogen bonds are formed upon the addition of cellopentaose to C B D N I W T and CBDNIY19A, but not CBDNI Y85A. Further, binding data tells us that affinity for oligosaccharides is significantly reduced for both mutants with respect to the wild-type, and NMR data indicates the involvement of Y19, Y43, Y85, and W16 and W137 in binding. Occum's razor dictates that we accept the simplest explanation that is consistent with all of these data. In my opinion, this would be that there exists in C B D N I W T a hydrogen bond between Y19 and either W16orW137 C B D M Y 8 5 A Raman Shift (cm'1) Figure 6.21. Fiber-optic UVRRS CBDNI spectra from Figure 6.15 in expanded detail showing W3 region. C P indicates the presence of 2.5 mM cellopentaose 283 (both of which are in relatively close proximity). This H-bond stabilizes the orientations of the two H-bond partners to promote hydrophobic interactions between either or both of the partners with the substrate and/or prevent steric hindrance of either or both with the substrate. Judging from the solution structure, binding with W16 would favour the former and W137 the latter. Y85 interacts with the substrate via hydrogen bonding. It must be emphasized that more and better data are needed to validate these tentative conclusions. Specifically, higher resolution, higher SNR data (potentially with the improved laser, probe design, and interference filter mentioned in Chapter 7, along with longer integration times if more protein is available) are needed to allow for meaningful band fitting, deconvolution and statistical analysis which would solidify these results. However, these studies and their results do demonstrate that fiber-optic UVRRS probes can be used for meaningful, routine, in situ biochemical and biophysical analysis. 284 Chapter 7. Conclusions and Future Directions 7.1 Overview of Conclusions and Future Directions This thesis described the design, construction, characterization and use of the first reported fiber-optic UVRRS system. The objective of the research described herein was to develop an instrument based on UVRRS for in situ studies of protein binding with a special emphasis on protein adsorption, and this has largely been accomplished. The work that resulted was an exercise in systems engineering combined with biophysical and bioanalytical chemistry. At times it both drew from, and contributed to, areas as diverse as optical design, laser-systems, analytical spectroscopy, optical materials characterization, protein biophysical chemistry, and signal processing. In doing so, (a) a tunable DUV light source, spectrometer and detector were assembled as a system; (b) new optical fibers were characterized with respect to pulsed DUV transmission, and insights into their optical materials characteristics were obtained; (c) standard flush fiber-optic probe designs were modelled and characterized for use in highly absorbing samples; (d) a new, superior angled/reflecting, DUV-transmitting probe geometry and composition was introduced, implemented, characterized and modelled for use in UVRRS studies; (f) ultimate theoretical and geometrical analytical limits were determined; (f) UVRRS data of proteins, amino acids, DNA, hormones and neurotransmitters were obtained in situ with these probes; (g) UVRRS data were characterized as signals and novel high-performance SNR enhancement schemes were designed and implemented; (h) proof-of-concept data were obtained validating the probes' usefulness for bioanalytical, protein-unfolding, RREP-measuring, and specific and non-specific binding studies; and (i) mechanisms of cellulose-285 binding-domain function were investigated vis a vis the involvement of TRP and TYR residues. Ultimately, the performance of the fiber-optic UVRRS system was limited not by the optical fibers or probe geometry, but rather by the DUV laser system (low duty cycle), spectrometer (low throughput, inadequate Rayleigh rejection short focal length), and detector (poor quantum efficiency and high dark noise) relative to equipment which is now commercially available. This chapter briefly summarizes the most important insights and conclusions from chapters two through seven. I also point out where improvements may be made to the instrument with current technology that was unavailable for this work, and where future work is likely (or in my opinion, ought) to be directed. It has been said recently by a pundit in the field that Raman spectroscopy is "an awakening giant" [Borman, 1997] and I hope that this work has in some small way contributed to her early rising. 7.2 General System Design 7.2.1 Conclusions from System Design In many ways, the field of UVRRS is technology driven. The primary novel aspect of the technology for the system design presented here were the fiber-optic probes, conclusions from which are discussed in section 7.3, below. The remainder of the system (tunable DUV light source, spectrometer and detector) was in most ways conventional (and in some cases, antiquated). The low duty cycle, short pulse 20 Hz. Nd: YAG pumped doubled tunable dye laser system required the use of high peak powers to obtain usable average powers. Such systems, although common ten years ago for UVRRS, have now 286 been largely supplanted by high-repetition rate excimer pumped dye lasers and other cw or quasi-cw systems (see section 7;2.2.1, below). This research was constrained by financial and equipment availability limitations which dictated the use of now-outdated Nd: YAG and dye lasers, and very "minimalist" frequency doubling equipment, all from different vendors. The present system not only exacerbated the fiber solarization problem, but also necessitated continuous vigilance in detecting and preventing sample photodegradation, system instabilities and safety hazards. In conclusion, although the author was fortunate and grateful for the generosity of the researchers who donated equipment to enable this project, in a certain sense the project's success must be considered to be in spite of the particular DUV light source used. The IPDA detector used in these studies was state-of-the-art when the project began five years ago; however the rapid pace of progress in the field of multichannel spectrographs detectors has since rendered it obsolete as well. Present CCD systems have quantum efficiencies higher than the IPDA by factors of up to five, and dark currents lower by orders of magnitude. The Czerny-Turner design monochromator and 3600 G/mm grating are still relatively standard in UVRRS applications, despite some of their drawbacks (e.g. large size and low throughput). The 0.67 m focal length of the monochromator combined with the blurring produced by the detector's MCP intensifier and the large slit widths used resulted in only fair resolution, as was evident in the data in sections 6.2 and 6.3. Finally, the poor throughput (<10%) of the double subtractive monochromator rendered it almost entirely useless for stray light rejection. 287 7.2.2 Future Directions in System Design 7.2.2.1 Ultraviolet Lasers Incorporation of a higher duty cycle DUV light source in the system is a necessity to obtain the full benefits of the fiber-optic probes, improve the quality of data obtained, reduce the size of the instrument and increase its reliability and versatility. Presently available cw options include intracavity doubled argon- and krypton-ion lasers. These have 100% duty cycles, but only provide certain discrete DUV lines (notably 229, 244 and 257 nm for the Ar+ and 206 nm for the Kr+). A high-duty-cycle tunable quasi-cw option is the mode-locked, frequency-quadrupled titanium sapphire systems with ca. 5 ps pulse lengths and 80 MFfz repetition rates, resulting in a duty cycle of 4x10 ,^ an improvement of approximately 6000 over the present duty cycle. At the very least, a high repetition rate (ca. kHz) excimer or Nd: YAG laser in place of the present 20 Hz pump laser would improve the duty cycle by a factor of ca. 50. Based on the data from Chapter 3, it is likely that, with any of these options, solarization and non-linear effects in the excitation fibers could be eliminated completely, thereby improving overall energy delivery to the sample. Further, much higher average powers would be achievable at much lower peak powers, thereby simultaneously increasing throughout and reducing the probability of photodamage. In itself, a switch to a cw laser could increase overall signal intensity by a factor of ca. 20, simply by allowing the use of 10 mW excitation instead of the ca. 500 pW levels mentioned throughout most of this thesis. Unfortunately, all of these options are expensive and physically large, and would therefore not be appropriate for incorporation in an instrument that will see widespread use. 288 There are, however, several promising DUV laser options on the horizon. Small (ca. HeNe laser sized), low price, quasi-cw metal vapour lasers with fundamental wavelengths from 200 to 300 nm will soon be available (e.g. HgAg laser with 224 nm quasi-cw line). In the longer term, semiconductor diode lasers based on InGaAIN technology may have the capability to offer bandgaps anywhere in the 2-6.2 eV range, resulting in lasers with wavelengths anywhere from ca. 200 to 600 nm, depending upon stoichiometry [Fonce and Bour, 1997]. Like most semiconductor diode laser technology, these could be small, relatively inexpensive, and possibly tunable over a narrow range. The recent development of room temperature blue diode lasers also opens the possibility of an intracavity-doubled external-cavity blue diode laser to provide DUV light. 7.2.2.2 Improved Spectrometers In all likelihood, large, reflection-grating-based dispersive spectrometers will remain the norm for non-imaging UVRRS applications in the near future; however there are several exciting developments and possibilities that will probably supplant this technology eventually. A dispersive spectrometer based upon antireflection-coated fused-silica prisms may offer throughput approaching 100% (compare with ca. 30% for a single grating monochromator with significant losses in other diffraction orders), superior stray light rejection, and lower cost. The dispersion curve for silica indicates that at wavelengths shorter than about 230 nm, this may indeed be a possibility. Acousto-optic tunable filters (AOTFs) operating in the DUV represent a very small, inexpensive alternative to the large grating based designs. Although they must be electronically scanned to obtain an entire 289 spectrum (thereby sacrificing the multiplex signal advantage enjoyed by the combination of grating spectrometers and multichannel detectors), for many applications this may be acceptable. In combination with a semiconductor or metal-vapour DUV laser, PMT, DA/AD board, fiber-optic probe, and notebook-computer, this would result in a small, portable, inexpensive, versatile and robust instrument that could be incorporated in almost any laboratory, industrial or clinical setting. Since the electronic scanning may be done much more rapidly than mechanical scanning, 1/f noise would be expected to be less of a problem with such a system than in a single-channel scanning grating system. Another possibility is a DUV-FT-UVRRS instrument. This would require a true-cw source, but would not surrender the multiplex advantage and would have most of the other advantages of the AOTF system. 7.2.2.3 Improved Rayleigh Light Rejection Contamination from stray Rayleigh light is a considerable problem in UVRRS studies. In many cases, we found that while signal strength continued to increase as concentration was reduced (i.e. we were operating on the falling, high-concentration part of the working curves discussed in Chapter 4), SNR was rapidly degraded due to increased stray Rayleigh light background. In many cases, the lower detection limit of our system is imposed by shot noise from the Rayleigh background. Until very recently, narrow transition-band optical filters with high pass-band efficiencies suitable for Rayleigh light rejection in UVRRS applications were not available for DUV applications. It is now possible, however, to obtain devices having stop-band optical densities on the order of 290 three and pass band efficiencies of 80% with transition bands of only a few nanometers. This should significantly improve SNR and, when added to an AOTF and CCD array, may also permit rapid, inexpensive UVRRS imaging, albeit not confocal. While it would depend upon the particular experimental conditions, I believe that a reasonable estimate for the reduction in noise provided by such filters when used with low-concentration samples could be an order of magnitude. We have obtained such a filter (for 228 nm rejection) and will soon be incorporating it in the system. 7.2.2.4 Improved Detectors DUV multichannel spectroscopic detector technology has advanced rapidly since the mid to late 1980's, and appears to be beginning to plateau. Large (ca. 512 x 1024, or larger) back-thinned, anti-reflection coated, cooled CCD arrays with quantum efficiencies of greater than 30% (at DUV wavelengths), dark currents as low as 1 eThour and read noise of less than 8 counts (rms) are now widely available. When the spectrum is detector noise limited (e.g. in very highly absorbing samples or with a good Rayleigh rejection filter), I believe that a good estimate for the improvement provided by such a system over the present JJPDA is a factor of 3 in detected photons and a factor of 5 in reduced noise. We are in the process of procuring a cooled, back-thinned, antireflection coated CCD and will soon report on its performance. 291 7.2.2.5 Imaging Much of the cutting edge research being done in the many areas of optical spectroscopy at present concerns imaging; certainly, this will extend to UVRRS. Although the large extinction coefficients of most biological samples in the DUV region will make imaging at depths greater than a few hundred microns quite difficult, for many applications this penetration is sufficient. Since UVRRS image data will have four dimensions (two spatial, one Raman shift, and one intensity dimension) and CCD data has only three dimensions (two spatial and one intensity) it will be necessary to either (a) mechanically scan the excitation light in time through one of the two spatial dimensions, and use a standard imaging spectrometer to detect, at each time, the Raman spectrum at each point along a line, or (b) use a non-dispersive scanning spectrometer (e.g. an AOTF or FT instrument) to detect two spatial dimensions and intensity at each time, and scan the wavenumber shift in time. By using a 'coherent' (or ordered) fiber-optic bundle, it may be possible to effect UVRRS imaging in situ (with, e.g., a FT or AOTF based system) which may lead naturally to medical applications. Such systems have been reported for visible RS [Skinner et al., 1996]. As well, confocal and/or multi-photon excitation imaging may be possible. 7.2.2.6 Ultimate Short Term Improvement from Component Changes In the short term, system performance may be improved as discussed above by incorporating into the system (a) a higher duty cycle laser, (b) a high performance Rayleigh rejection filter, (c) a prism spectrometer, (d) multiple (up to 6) collection fibers, 292 and (e) a cooled, back-thinned, anti-reflection coated CCD array. Table 7.1 shows the factors of improvement estimated to be derived from each of these. Hence, an SNR improvement of up to ca. 864x50=43200 is possible with these short term modifications to the system, all of which are either presently in progress (Rayleigh filter, more fibers and CCD detector) or planned for the short term (new UV laser and prism spectrometer). It is not unreasonable, then, to strive for sub-nanomolar detection limits for the aromatic amino acids, which would begin to approach the geometrical detection limits discussed in section 4.4. It should be noted that the figures for signal increase are probably more reliable than those for noise decrease since (a) they derive mostly from easily measured and/or well known device parameters, (b) unlike the noise figures, they do not depend on sample conditions, and (c) eventually, a limit to the decrease in noise is reached when SNR is limited by signal shot noise. Component Signal Increase Multiplier Noise Decrease Divider UV laser 20 1 Rayleigh Rejection Filter 0.8 ca. 10 Prism Spectrometer 3 1 Multiple Collection Fibers 6 1 New CCD Detector 3 5 TOTAL: 864 50 Table 7.1. Best case estimated system performance improvement factors for planned component additions. 293 7.3 Transmission of UV Light Through Optical Fibers 7.3.1 Conclusions The work in Chapter 3 showed that high-intensity, pulsed DUV light could be delivered with fused silica IUV optical fibers at wavelengths as low as 206 nm and efficiencies on the order of tens of percent when using 0.5-1.0 meter lengths of fiber and 20 Hz pulse energies of 50 pJ or greater, resulting in average delivered powers of hundreds of microWatts. However, with the present low-duty-cycle system, the excitation fiber is still the limiting component in determining power delivered to the sample, since the fiber throughput rapidly saturates (plateaus) with increasing pulse energy. The results also indicated that a higher duty cycle laser would significantly increase overall system performance. In general, this work added to the understanding of the theory and practice of UV transmitting optical fibers and provided some predictive ability, albeit incomplete and to a large extent empirical. 7.3.2 Future Directions Future work in the area of UV transmitting optical fiber characterization will undoubtedly continue, driven by remote sensing applications such as ours as well as medical applications (e.g., laser surgery). This area of research is somewhat peripheral to our main focus of instrument development and biochemical applications; therefore most of the future work in this area stemming directly from this will involve characterizing IUV fibers for use with any new laser system that is acquired. 294 7.4 Fiber-Optic Probes 7.4.1 Conclusions from Probe Design and Modelling The results from section 4.2 permit an enhanced understanding of the operation of flush fiber-optic Raman probes in absorbing media. A new angled, reflecting design incorporating a ca. 90° collection geometry and improved ultraviolet-transmitting fibers was introduced in section 4.3 and used in Chapter 6. It was shown to be superior to standard, flush designs. UVRRS data of various small and large biomolecules were obtained in timescales on the order of minutes with hundreds of micro Watts of excitation power. Detection limits on the order.of micromolar were achieved for the aromatic amino acids using ca. 230 nm excitation. Although demonstrated operation of the probe at wavelengths shorter than 220 nm remains somewhat problematic due to reduced excitation fiber throughput and increased inner-filtering. Given the results from chapters 2, 3, and 4, it is likely that the systems modifications discussed in section 7.2, above, and presently being effected will overcome this difficulty. A procedure for determining the optimal experimental conditions was introduced and used to show that the excitation wavelength to achieve maximum signal strength is generally significantly red-shifted from the RREP maximum. In section 4.4, the excitation and collection processes were modelled and optimal detection limits in the picomolar range were shown to be possible; however it was also shown that nanomolar detection limits may be more reasonable goals in the short term. The angled/mirrored probe simulation showed that the prototype probes used in chapters 4 and 6, while effective, still do not realize the full potential of the design. 295 7.4.2 Future Directions in Probe Design In the short term, certain improvements on the existing design are easily realizable. For the most part, the prototype probes used in this thesis made use of relatively large diameter fibers and only a single collection fiber. Using multiple, equivalent, collection fibers will increase the collected signal arithmetically. The 'six-around-one* geometry is common for flush probe designs and could also be incorporated as an angled/mirrored design. However, special care must be taken with such a probe because, unlike a six-around-one flush design (a) the orientation of the collection fibers is critical; in particular, the mirror centerline normal should intersect the excitation fiber axis, and (b) mass transport problems may be encountered in such a geometry since it would form a sample volume enclosed on seven of eight sides. The use of a higher duty cycle laser as discussed above would permit the incorporation of smaller diameter excitation fibers in the design. This would, in turn, permit the use of smaller diameter collection fibers. The combination would improve probe performance in absorbing samples as well as improve resolution. In the longer term, more radical changes to the probe design may be necessary to more closely approach the analytical limits discussed in section 4.4. These may include the use of single fibers, bifurcated fibers, and single-mode fibers as well as incorporating Rayleigh and/or silica-Raman rejection filters as parts of the probes themselves. It may also be desirable to construct ad hoc probes optimized for specific applications (e.g. protein adsorption). We have a continuing interest in the area of probe design and we will soon report on newer, even higher perfromance designs. 296 7.5 Signal Processing 7.5.1 Conclusions on Signal Processing The author would argue that not using an SNR enhancement or noise removal scheme is the equivalent of throwing away data. From its physical basis and from the PSD estimates of Chapter 5, it is expected and known that Raman signals are band-limited to low frequencies. For example, we showed a high frequency cut-off of at a frequency of ca. 0.1 cm; however, this will depend to some extent on instrument resolution. Failing to remove high-frequency random noise when a method is available is the equivalent of introducing extra noise. Traditional FFT based and regularization based methods were investigated and shown to be effective for SNR enhancement of Raman spectroscopy data. This is significant, since analytical, bioanalytical and biophysical chemists (many of the end users of such technology) have traditionally resorted to methods such as boxcar averaging or Savitzky-Golay filtering, which may filter out low-frequency, signal-rich components of the spectrum, and/or provide insufficient attenuation at higher, noise-containing frequencies. 7.5.2 Future Directions in Signal Processing Regularization methods may become more widespread as computing power is increased and minimization schemes with improved speed and convergence are realized. I believe that a new regularization function may also be desirable. Optimization of the regularization functions that have been used to date (e.g. entropy, two-point entropy, sum 297 of squares of the low frequency Fourier components) implies an a priori expectation of a perfectly flat spectrum. While this is a reasonable first approximation, in fact we usually have good reason to expect a series of peaks with a certain range of shapes, widths and amplitudes. If this a priori expectation could be incorporated in a regularization function, this should result in improved performance. On the other hand, since the underlying peaks are known to be Lorentzian-Gaussian convolutions, a Savitzky-Golay-like method in which a measured point was mapped to a filtered point by fitting a window around the measured point to a Voight profile may result in improved performance compared with Savitzky-Golay filtering. 7.6 Biochemical Applications 7.6.1 Conclusions from Biochemical Applications The results of Chapter 6 showed that the fiber-optic probes could be used to obtain the tertiary structure sensitive protein TRP and TYR modes, the secondary structure sensitive protein amide modes, the DNA nucleotide base modes as well as signals from free amino acids, hormones and neurotransmitters. High quality protein spectra could be obtained in only tens of minutes, and could be used to detect protein unfolding. The RREP of phosphotyrosine was measured and compared to L-tyrosine. Proof-of-concept data were obtained for proteins specifically and non-specifically bound to insoluble substrates. The cellulose binding domain proteins CBDcex and CBDNI were investigated in more detail. It was found that glycosylation of CBDcex at N24 resulted in the breaking of a 298 TRP hydrogen bond. The CBD N i investigations provided evidence that at least one TYR residue (probably Y85) is involved in binding to oligosaccharides via hydrogen bonding, and Y19, while not involved in hydrogen bonding to the substrate, seems to be hydrogen bonded to a TRP residue (W16 or W137) in the wild-type protein which may promote binding to the substrate by stabilizing the orientations of either or both residues and thereby promoting hydrophobic interactions with the substrate or preventing steric hindrance of the substrate. However, more data is needed in this area. In general, the utility of the fiber-optic UVRRS system for rapid, in situ, qualitative and quantitative investigations of binding events, despite the inadequacies of the present system, nominally meets the goals set out in Chapter 1 of the thesis. 7.6.2 Future Directions in Biochemical Studies 7.6.2.1 Improved Signal Understanding and Calibration Much of the work that has been done correlating changes in peak positions, bandshapes, and RREP shapes with changes in secondary and tertiary structure was done more than five years ago, prior to the development of improved DUV light sources and low-noise, high-quantum-efficiency CCDs. The author is of the opinion that repeating some of this work could be very beneficial in terms of correlating more subtle changes in spectrum morphology with structural or molecular environmental changes. This is especially true for weak and/or overlapping peaks which, with modern high-SNR instrumentation and signal processing procedures, could certainly yield more information. Furthermore, a concentrated effort to investigate UVRRS signals from resonance 299 enhancable chromophores in small, synthetic peptides and small model proteins for peak positions and RREPs as a function of solvent physico-chemical parameters (polarity, hydrogen bonding capacity, salt concentration) and adjacent/nearby residue identity could yield a wealth of data and significantly increase the utility of UVRRS by providing a solid basis for data interpretation. In the long run, it is the author's belief that using a tunable high-resolution, high-SNR system to obtain UVRRS data at a series of closely-spaced excitation wavelengths, and plotting this data as a two dimensional contour plot as a function of wavenumber and excitation wavelength may provide a great deal more data than the standard ID spectra that are generally considered at present. Specifically, the extra dimension in such plots may facilitate (a) detection and quantification of structural changes, (b) detection of protein-protein, or more generally protein-substrate interaction, and (c) deconvolution of overlapping bands and their assignment to specific residues within the protein. 7.6.2.2 Binding and Adsorption Studies Given that the SNR for the binding studies reported in section 6.2 was high enough to permit detection of the bound CBD and lysozyme on insoluble "cellulose and silica, respectively, it is likely that even a single order of magnitude improvement in SNR would permit quantitative determinations of protein structures when bound to insoluble substrates. Also, given that most of the noise in these experiments came from Rayleigh background, it is possible that the addition of the narrow-band interference filter discussed in section 7.2.2.3 would, in itself, enable such studies. The full noise advantage of an 300 improved CCD camera would not be realized in such studies, since optical, and not detector noise, is limiting. However, the combination of improved detector quantum efficiency and the use of the higher duty cycle laser would improve SNR by up to ca. (3*20)°'5=7.7, or nearly an order of magnitude. The success of the fiber-optic UVRRS CBDNi/oligosaccharide binding study reported in section 6.3.3, even with relatively short integration times (twelve minutes), low average powers (ca. 500 uW) and the inadequacies of the present system demonstrates the utility of the fiber-optic UVRRS instrument and bode well for future applications. With the improvements from the new components discussed in section 7.2.2, the improved probe designs from section 7.4, signal processing discussed in section 7.5 and advances in signal calibration and interpretation suggested in section 7.6.2.1, it is quite likely that a specialized fiber-optic UVRRS instrument could be realized for rapid detection and quantification of protein-protein and protein substrate interactions. This could provide unique and complimentary information to techniques such as calorimetry and surface plasmon resonance, which are presently used in such application, and could be used in areas such as drug screening and fundamental biophysical and biochemical studies. 7.6.2.3 Bioanalytical Applications Single molecule detection is sometimes held up as the "holy grail" for an analytical technique. Given the results of section 4.4, it is unlikely that UVRRS, in itself, can achieve such a goal. In fact, despite its remarkable selectivity, UVRRS is rarely used in bioanalytical applications at all. However, we have demonstrated the ability of fiber-optic 301 UVRRS to detect micromolar quantities of hormones, amino acids, and neurotransmitters from effective sample volumes on the order of 30 nL. The author believes that there exists a large class of bioanalytical problems for which the combination of selectivity, sensitivity and information content in UVRRS would render it an appropriate solution. These would include resonance enhancable biomolecules with relatively low fluorescence quantum efficiencies that may be found in the presence of other interfering, fluorescing analytes. 7.6.2.4 Clinical, Environmental and Industrial Applications There are numerous other fields where a versatile, remote instrument with the combined physical and analytical capabilities of fiber-optic UVRRS could find applications. For example, UVRRS has recently been shown to be effective in the diagnosis of certain cancers [Manoharan et al., 1994; Clary et al., 1997], and the utility of fiber-optic probes for such applications is self evident. Also, given the range of resonance enhancable chromophores, it is likely that a small, inexpensive fiber-optic UVRRS instrument could find considerable application in clinical biochemistry. In the area of environmental monitoring, UVRRS is useful for most aromatic molecules, which constitute a large number of environmental contaminants, and provides vastly greater selectivity than fluorescence spectroscopy. In combination with fiber-optic probes, remote UVRRS could be ideal for certain dangerous or inaccessible environmental monitoring applications. Finally, in any industrial process where resonance enhancable chromophores are involved as products or contaminants, fiber-optic UVRRS could be used for process 302 monitoring. In particular this might include petrochemical, drug production and bioprocess monitoring applications. 7.7 Final Concluding Remarks Two of the most human characteristics are the making of tools and a desire to understand their own surroundings. I have made an optical tool and used it to understand some of our biochemical environment. My sincere hope is that the tool, the understanding gained from using it, and the experience that I gained might continue to be useful or interesting in some way to others. 303 Chapter 8. References Agbo, S. O. Chapter 40.1 in "The Electrical Engineering Handbook" IEEE/CRC Press, Boca Raton (1993). Albrecht, A. C. and Hutley, M. C. J. Chem. Phys., 55, 4438 (1971). Albrecht, A. C. J. Chem. Phys. 49, 1144 (1968). Allison, S. W.; Cates, M. R.; Gillies, G. T. and Noel, B. W. Opt. Eng., 26, 538 (1987). Allison, S. W.; Gillies, G. T.; Magnusen, D. W. and Pagano, T. S. Appl. Opt., 24, 3140 (1985). Anderson, A. B.; Darst, S. A. and Robertson, C. R. Chapter 20 in Proteins at Interfaces, ACS Symposium Series 343 (Brash and Horbett, eds.), 306 (1987). Andrade, J. D. and HJady, V. Advances in Polymer Science, 79, 1 (1986). Angel, S. M. and Myrick, M. LAppl. Opt., 9, 1350 (1990). Asher, S. A. Anal. Chem. 65, 201A (1993A). Asher, S. A. Anal. Chem. 65, 59A(1993B). Asher, S. A.; Johnson, C. R. and Murtaugh, J. Rev. Sci. Instrum. 54, 1657 (1983). Asher, S. A.; Ludwig, M. and Johnson, C. RJ. Am. Chem. Soc. 108, 3186 (1986). Austin, J. C ; Jordan, T. and Spiro, T. G. "Ultraviolet Resonance Raman Studies of Proteins and Related Model Compounds", in Advances in Spectroscopy, vol. 20A, Biomolecular Spectoscopy, R. J. H. Clark and R. E. Hester, eds., 55 (1993A). Austin, J. C ; Rodgers, K. R. and Spiro, T. G. "Protein Structure from Ultraviolet Resonance Raman Spectroscopy," Methods in Enzymology (J. F. Riordan and B. L. Vallee, eds.), 226, 374 (1993B). Bailey, J. E. and Ollis, D. F. "Biochemical Engineering Fundamentals" McGraw-Hill, New York (1986). Ball, A. and Jones, R. A. L. Langmuir 11, 3542 (1995). Barbucci, R. and Magnani, A. Biomaterials 15, 955 (1994). Beck, T.; Deng, N.; Richter, K Laser Focus WorldOctober, 111 (1993). 304 Beissinger, R. L. and Leonard, E. F. Proceedings of the American Society of Artificial Internal Organs, 3, 160 (1980). Bello, J. M.; Narayanan, V. A.; Stokes, D. L. and Vo-Dinh, T. Anal. Chem. 62, 2437 (1990). Benko, B.; Vuk-Pavlovic, S.; Dezelic, G. and Maricic, S. J. Colloid Interface Sci. 52, 444 (1975). Berstein, H. J. Chapter 37 in "Advances in Raman Spectroscopy", (J. P. Mathieu, Ed.) Heyden and Son, London, 305-316 (1973). Bialkowski, S. E. Anal. Chem. 60, 355A (1988A). Bialkowski, S. E.Anal. Chem. 60, 403A(1988B). Boraston, A. Personal communication and Ph. D. dissertation (in progress) (1997). Borman, S. Chemical and Engineering News March 31, 32 (1997). Brackmann, U. "Lambdachrome Laser Dyes", Lambda Physik, Gottingen, FRG (1994). Bray, M. R.; Johnson, P. E.; Gilkes, N. R.; Mcintosh, L. P.; Kilburn, D. G. and Warren, R. A. J. Prot. Sci. 5, 2311 (1996). Britton, K. A.; Dalterio, R. A.; Nelson, W. H.; Britt, D. and Sperry, J. F. Appl. Spectrosc. 42, 782 (1988). Bromba, M. U. A. and Ziegler, H. Appl. Spectrosc. 53, 1583 (1981). Bruice, P. Y. Chapter 14 in "Organic Chemistry", Prentice-Hall, New Jersey, 1995. Buijs, J.; Norde, W. and Lichtenbelt, J. W. T. Langmuir 12, 1605 (1996). Burg, J. P. "Maximum Entropy Spectral Analysis", Ph.D. Dissertation, Stanford University, Stanford, California (1975). Burghardt, G. P.; Axelrod, D. Biochem. 22, 979 (1983). Burton, L. L. and Blades, M. W. Spectrochim. Acta 42B, 513 (1987). Campbell, I. D. and Dwek, R. A. "Biological Spectroscopy", Benjamin/Cummings, Menlo Park (1984). 305 Carey, P. R. "Biochemical Applications of Raman and Resonance Raman Spectroscopies", Academic Press, New York (1982). Castillo, E. J.; Koenig, J. L.; Anderson, J. M. and Lo, J. Biomaterials 6, 338 (1985). Caswell, D. S. and Spiro, T. G. J. Am. Chem. Soc. 108, 6470 (1986). Cherney, E. "The Molecular Basis of Optical Activity" Wiley Interscience, New York, (1979). Chittur, K. K.; Fink, D. J.; Hutson, T. B.; Gendreau, R. M.; Jakobsen, R. J. and Leininger, R. I. Chapter 23 in Proteins at Interfaces, ACS Symposium Series 343 (Brash and Horbett, eds.), 362, (1987). Chou, Pi-Tai, Studer, S. L. and Martinez, M. L. Appl. Spectrosc. 45, 513 (1991). Clark, R. J. H., Ellestad, O. H. and Mirchell, P.D. Appl. Spectrosc, 28, 73 (1974) Clary, C. E.; Dergachev, A. Y.; Kurdryavtsev, A.; Mirov, S. B. and Gillespie, G. Y. Proc. SPIE 2980, paper no. 23 (in press, 1997). Cooney, T. F.; Skinner, H. T. and Angel, S. M. Appl. Spectrosc. 50, 836 (1996A). Cooney, T. F.; Skinner, H. T. and Angel, S. MAppl. Spectrosc. 50, 849 (1996B). Copeland, R. A. and Spiro, T. G. Biochem. 26, 2134 (1987). Copeland, R. A. and Spiro, T. G. Biochem. 24, 4960 (1985). Copeland, R. A. and Spiro, T. G. J. Am. Chem. Soc. 108, 1281 (1986). Coutinho, J. B.; Gilkes, N. R.; Warren, R. A. J.; Kilburn, D. G. and Miller, R. C. Mol. Microbiol. 6, 1243 (1992). Creagh, A. L.; Ong, E.; Jervis, E.; Kilburn, D. G. and Haynes, C. A. Proc. Natl. Acad Sci. USA 93, 12229 (1996). Creighton, T. E. "Proteins", Freeman, New York (1984). Davies, S.; Packer, K. J., Baruya, A. and Grant, A. I. Chapter 4 in Maximum Entropy in Action, B. (Buck and V. A. MacAulay Eds.) Clarendon Press, Oxford (1991) Din, N.; Forsythe, I. J.; Burtnick, L. D.; Gilkes, N. R.; Miller, R. C ; Warren, R. A. J. and Kilburn, D. G. Mol. Microbiol. 11, 747 (1994). 306 Donhowe, M. Chapter 4 in "Photonic Devices and Systems" Hunsperger, R. G. (Ed.) Marcel Dekker, Inc. New York (1994). Dudik, J. M.; Johnson, C. R. and Asher, S. A. J. Phys. Chem. 89, 3805 (1985). Dyer, S. A. and Hardin, D. S. Appl. Spectrosc. 39, 655 (1985). Elliot, A. and Ambrose, E. J. Nature, 165, 921 (1950). Epperson, P. M.; Sweedler, J. V.; Bilhorn, R. B.; Sims, G. R. and Denton, M. B. Anal. Chem. 60, 327A (1988). Evans, S. V. and Brayer, G. D. J. Biol. Chem. 263, 4263 (1988). Fabian, H.; Grzesik, U.; Hillrichs, G. and Neu, W. "Optical fibers with enhanced performance for excimer laser power transmission at 308 nm," Proc. SPIE 1693, 24 (1993). Fabian, H.; Grzesik, U.; Worner, K.-H. and Klein, K. -F. "Optical Fibers for UV applications," Proc. SPIE, 1513, 168 (1991). Faraday, M. Ann. Chim. Phys. 25, 99 (1825). Ferraro, J. R. and Nakamoto, K. "Introductory Raman Spectroscopy" pp. 1-94, Academic Press, San Diego, 1994. Flaugh, P. L.; OT>onnell, S. E. and Asher, S. A. Appl. Spectrosc. 38, 847 (1984). Fodor, P. A.; Rava, R. P.; Hays, T. R. and Spiro, T. G. J. Am. Chem. Soc. 107, 1520 (1985). Fodor, S. P. A. and Spiro, T. G. J. Am. Chem. Soc. 108, 3198 (1986). Fodor, S. P. A.; Copeland, R. A.; Grygon, C. A. and Spiro, T.G. J. Am. Chem. Soc. I l l , 5509 (1989). Fodor, S. P. A.; Rava, R. P.; Copeland, R. A.; and Spiro, T.G. J. of Raman Spectrosc. 17, 471 (1986). Fonce, F. A. and Bour, D. P. Nature. 386, 351 (1997). Frank, C. J.; McCreery, R. L. and Redd, D. C. Anal Chem. 67, 777 (1995). Frieden, B. R. J. Opt. Soc. Am. 62, 511 (1972). 307 Garrett, J. B.; Mullins, L. S. and Raushel, F. M. Protein Sci. 5, 204 (1996). Gerrard, D. L. and Bowley, H. J. Chapter 3 in "Practical Raman Spectroscopy" Gardiner and Graves (Eds.) 55-102, Springer-Verlag, Berlin, 1989. Gilbert, A. S. "Data Manipulation in Spectroscopy," in Computer Methods in UV, Visible and IR Spectroscopy, W. O. George and H. A. Willis, Eds. (Royal Society of Chemistry, London, 1990), Chap. 4, pp. 37-52. Gilkes, N. R.; Henrissat, B.; Kilburn, D. G.; Miller. R. C. and Warren, R. A. J. Microbiological Reviews, 55, 303 (1991). Gilkes, N. R.; Warren, R. A. J.; Miller, R. C. and Kilburn, D. G. J. Biol. Chem. 263, 10401 (1988). Gratzer, W. B. Tech. Protein Enzyme Biochem., B108, 1 (1978). Graves, P. R. "Calibration and Data Processing", Chapter 4 in Practical Raman Spectroscopy, D. J. Gardiner and P. R. Graves, Eds. (Springer-Verlag, Berlin, 1989), Chap. 4, pp. 77-101. Greek, L. S.; Schulze, H. G.; Blades, M. W.; Bree, A. V.; Gorzalka, B. B. and Turner, R. F. B. Appl. Spectrosc, 49, 425 (1995). Greek, L. S.; Schulze, H. G.; Blades, M. W.; Haynes, C. A. and Turner, R.F.B. Proceedings of The 22nd Canadian Medical and Biological Engineering Society Conference (Charlottetown, Prince Edward Island, June 26-29), pp. 78-79, (1996A). Greek, L. S.; Schulze, H. G ; Blades, M. W.; Haynes, C. A.; Turner, R. F. B. and Klein, K. F. Applied Optics (Biomedical Optics division) Accepted and In Press (1997B). Greek, L. S.; Schulze, H. G ; Haynes, C. A.; Blades, M. W. and Turner, R. F. B. Appl. Opt., 35, 4086 (1996B). Greek, L. S.; Schulze, H. G.; Haynes, C. A.; Blades, M. W. and Turner, R. F. B. Proc. SPIE 2982, BiOS '97, The International Biomedical Optics Symposium of Photonics West '97, San Jose, CA, February 8-14, 1997 (1997A). Gronenborn, A. M. and Clore, G. M. Chapter 9 in "Spectroscopic Methods for Determining Protrein Structure in Solution", H. A. Havel (Ed.) pp 190-246, VCH Publishers, New York, 1996 Grossman, W. E. L. Journal of Chemical Education, 66, 697 (1989). 308 Haas, E. Chapter 3 in "Spectroscopic Methods for Determining Protrein Structure in Solution", H. A. Havel (Ed.) pp 28-61, VCH Publishers, New York, 1996 Hanley, Q. S.; Earle, C. W.; Pennebaker, F. M.; madden, S. P. and Denton, M. B. Anal Chem. 68, 661A(1996). Harada, I. and Takeuchi, H. in Advances in Infrared and Raman Spectroscopy, (R. J.H. Clark and R.E. Hester, Eds.) Vol. 13, Wiley, New York, (1986). Havel, H. A. (ed.), Spectroscopic Methods for Determining Protein Structure in Solution, VCH Publishers, New York, (1996A). Havel, H. A. Chapter 4 in "Spectroscopic Methods for Determining Protrein Structure in Solution", H. A. Havel (Ed.) VCH Publishers, New York (1996). Haynes, C. A. and Norde, W. Colloids Surfaces B 2, 517 (1994). Hecht, J. "Understanding Lasers" IEEE Press, New York (1992). Hecht, L.; Che, D. and Nafie, L. A. Appl Spectrosc. 45, 18, (1991). Heiman, D.; Zheng, X. L.; Sprunt, S.; Goldberg, B. B. and Isaacs, E. D. Proc. SPIE, 1055, 96 (1989). Hitzler, H.; Leclerc, N.; Klein, K. -F.; Greulich, K. -O. and Wolfrum, J. Proc. SPIE, 1023, 249 (1988). Hitzler, H.; Pfleiderer, C ; Leclerc, N.; Wolfrum, J.; Greulich, K.O. and Fabian, H. J. Non Cryst. Solids, 149, 107 (1994). Hobara, D.; Niki, K.; Zhou, C ; Chumanov, G. and Cotton, T. M. Colloids and Surfaces A, 93, 241 (1994). Homborg, H. and Preetz, W. Spectrochimica Acta, 32A, 709 (1976). Horsley, D.; Herron, J.; HJady, V. and Andrade, J. D. Chapter 19 in Proteins at Interfaces, ACS Symposium Series 343 (Brash and Horbett, eds.), 291, (1987). Hu, C.-Q.; Sturtevant, J. M.; Thomson, J. A.; Erickson, R. E. and Pace, C. N. Biochem. 31, 4876 (1992). Huang, Y. and Yu, P. Rev. Sci. Instrum. 59, 190 (1988). Hudson, B. S. and Mayne, L. C. Methods inEnzymology 130, 331 (1986). 309 Hutely, M. C. and Jacobs, D. J. Chem. Phys. Letters 3, 711 (1969). Ingle, J. D. and Crouch, S. R. Chapter 12 in "Spectrochemical Analysis" Prentice Hall, Englewood Cliffs (1988). Ito, H., Arakawa, Y., and L'Haya, Y. J. Chem. Phys. 98, 8835 (1993). Jackson, L. B. "Digital Filters and Signal Processing, 2nd Ed." Kluwer Academic Press, Boston (1989). Jakobsen, R. J. and Wasacz, F. M. Chapter 22 in Proteins at Interfaces, ACS Symposium Series 343 (Brash and Horbett, eds.), 339, (1987). Jaynes, E. T. Phys. Rev. 106, 620 (1957). Jiaying, M. and Zhong, L. Appl. Spectrosc. 45, 1302 (1991). Johnson, C. R.; Ludwig, M. and Asher, S. A. J. Am. Chem. Soc. 108, 905 (1986). Johnson, C. R.; Ludwig, M.; O'Donnell, S. and Asher, S. A. J. Am. Chem. Soc. 106, 5008 (1984). Johnson, P. E.; Joshi, M. D.; Tomme, P.; Kilburn, D. G. and Mcintosh, L. P. Biochem. 35, 14381 (1996A). Johnson, P. E.; Tomme, P.; Joshi, M. D. and Mcintosh, L. P. Biochem. 35, 13895 (1996B). Jordan, T.; Eads, J. C. and Spiro, T. G. Prot. Sci. 4, 716 (1995). Kagan, M. R. and McCreery, R. L. Langmuir 11, 4041 (1995). Karlitschek, P.; Hillrichs, G. and Klein, K. -F. Opt. Comm. 116, 219 (1995). Karlitschek, P.; Klein, K. -F.; Hillrichs, G. and Grzesik, U. Proc. SPIE 2677, 127 (1996B). Karlitschek, P.; Klein, K.-F. and Hillrichs, G. Proc. SPIE 2966, (in press, 1996A). Karlitschek, P.; Klein, K.-F. and Hillrichs, G. in Laser Induced Damage in Optical Materials, H. E. Bennett, A. H. Guenther, M. R. Kozlowski, B. E. Newnam and M. J. Soileau, eds., Proc. Soc. Photo-Opt. Instrum. Eng. 2966 (in press, April 1997). Kauppinen, J. K. and Saario, E. K. Appl. Spectrosc. 47, 1123 (1993). 310 Kefalas, P.; Brown, T. R. and Brickell, P. M. Int. J. Biochem. Cell Biol. 27, 551 (1995). Keiderling, T. A. Appl. Spectrosc. Rev. 17, 189 (1981). Keiderling, T. A. Chapter 8 in "Spectroscopic Methods for Determining Protrein Structure in Solution", (H. A. Havel, ed.) VCH Publishers, New York (1996). Kercel, S. W.; Roberts, M. J. and Garrison, A. A. Proc. SPIE 1336, 144 (1990). Kirkwood, J. G., J. Chem. Phys., 5, 479 (1937). Kitriotis, D. and Merkle, L. D. Appl. Opt. 28, 949 (1989). Kleimeyer, J. A.; Fister III, J. C ; Zimmerman, J. and Harris, J. M. Appl. Spectrosc. 50, 1597(1996). Klein, K. -F.; Hillrichs, G ; Karlitschek, P. and Grzesik, U. Proc. LASERmed 95, paper 107b (1995). Klein, K. -F.; Hillrichs, G ; Karlitschek, P.; and Mann, K. Proc. SPIE, 2966 (in press, 1996). Klein, K.-F.; Huttel, S.; Schulze, H. G ; Greek, L. S.; Blades, M. W.; Haynes, C. A. and Turner, R.F.B. Proc. SPIE 2979, (in press, 1997). Kondo, A.; Oku, S. and Higashitani, K. J. Colloid Interface Sci. 143, 214 (1991). Kormos, J. Personal Communications and Unpublished Data (1997). Kramer, J. R.; Dasari, R. R. and Feld, M. S. Nature Medicine 2 1079 (1996). Kraulis, P. J.; Clore, G. M.; Nilges, M.; Jones, T. A.; Petterson, G.; Knowles, J. and Gronenborn, A. M. Biochem. 28, 7241 (1989). Krauss, T. P., Shure, L. and Little, J. N. "Signal Processing Toolbox for Use with Matlab" The Math Works Inc., Mass., USA, 1994. Krimm, S. and Bandekar, J. Adv. Protein. Chem. 38, 181 (1986). Krimm, S. in "Vibrational Spectra and Structure", vol. 16, Elsevier, New York, p. 1 (1987). Lehninger, A.L.; Nelson, D. L. and Cox, M. M. "Principles of Biochemistry" pp. 298 (Worth, New York, 1993). 311 Lerner, E. J. Laser Focus World. 32, 103 (1996). Levy, D. H.; Gleason, K. K.; Rothschild, M. and Sedlacek, J. H. C. J. Appl. Phys. 73, 2809 (1993). Liu, G. Y.; Grygon, C. A. and Spiro, T. G. Biochem. 28, 5046 (1989). Lombardi, D. R.; Mann, C. K. and Vickers, T. J. Appl. Spectrosc, 49, 220 (1995). Long, D. A. "Raman Spectroscopy", McGraw-Hill, New York (1977). Ma, J. and Li, Y. Appl. Spectrosc. 48, 1529 (1994). MacRitchie, F. Advances in Colloid and Interface Science 25, 341 (1986). Mahadevan-Jansen, A.; Ramanujam, N.; Utzinger-Brunner, U.; Yazdi, Y.; Follen-Mitchell, M. and Richards-Kortum, R. R. Proc. SPIE 2980, paper no. 21 (in press, 1997). Manenkov, A. A. and Nechitailo, V. S. Proc. SPIE 1441, 392 (1990). Mann, K.; Gerhardt, H.; Pfeifer, G. and Wolf, R. Proc. SPIE 1624, 436 (1991). Manning, M. C. and Woody, R. W. Biochem. 28, 8609 (1989). Manning, M. C. J. Pharm. Chem. 7, 1103 (1989). Manoharan, R.; Boustany, N. N.; Deinum, G.; Myles, J.; Crowe, J.; Dasari. R. R. and Feld, M. S. Proc. SPIE 2980, paper no. 22 (in press, 1997). Manoharan, R.; Wang, Y.; Boustany, N.; Brennan III, J. F.; Baraga, J. J.; Dasari, R. R.; Van Dam, J.; Singer, S. and Feld, M. S. Proc. SPIE, 2328, 128 (1994). Mantsch, H. H. and Moffatt, D. J. in "Recent Experimental and Computational Advances in Molecular Spectroscopy", pp. 113-124, R. Fausto (ed.), Kluwer Academic Publishers, Ponta Delgada, Azores, 1993. Marteau, P.; Zanier-Szydlowski, N.; Aoufi, A.; Hotier, G. and Cansell, F. Vibrational Spectroscopy 9, 101 (1995). Maste, M. C. L.; Pap, E. H. W.; Hoek, A. V.; Norde, W. and Visser, A. J. W. G. Journal of Colloid and Interface Science 180, 632 (1996). Matzner, R. A.; Bales, R. C. and Pemberton, J. E. Appl. Spectrosc. 48, 1043 (1994). Mayne, L. C ; Ziegler, L. D., and Hudson, B. J. J. Phys. Chem. 89, 3805, (1985). 3!2 McCreery, R. L.; Fleischmann, M. and Hendra, P. Anal. Chem. 55, 146 (1983). McLachlan, R. D.; Jewett, G. L. and Evans, J. C. "Fiber Optic Probe For Sensitive Raman Analysis," US patent 4,573,761, (4 March, 1986). McMillan, C. R. and Walton, A. G. J. Colloid Interface Sci. 48, 345 (1974). Milonni, P. W. and Eberly, J. H. "Lasers", Wiley, New York, (1988). Miura, T.; Takeuchi, H. and Harada, I. Biochem. 27, 88 (1988). Miura, T.; Takeuchi, H. and Harada, I. J. Raman Spectrosc. 20, 667 (1989). Moffitt, W. J. J. Chem. Phys. 25, 467 (1956) Moffitt, W. J.; Fitts, D. D., and Kirkwood, J. G. Proc. Natl. Acad Sci. USA, 43, 723 (1957). Morrissey, B. W. and Stomberg, R. R. J. Colloid Interface Sci. 46, 152 (1974). Mukerji, I.; Schiber, M. C ; Spiro, T. G. and Fresco, J. R. Biochem. 34, 14300 (1995). Mukerji, I.; Shiber, M. C ; Fresco, J. R. and Spiro, T. G. Nucleic Acids Reserach 24, 5013 (1996). Mulkerrin, M. G. Chapter 2 in "Spectroscopic Methods for Determining Protrein Structure in Solution", H. A. Havel (Ed.) pp 5-27, VCH Publishers, New York, 1996 Myrick, M. L. and Angel, S. M. Appl. Spectrosc. 44, 565 (1990). Myrick, M. L.; Angel, S. M. and Desiderio, R. Appl. Opt. 29, 1333 (1990). Nafie, L. A. Adv. Infrared Raman Spectr. 11, 49, (1984). Norde, W. and Anusiem, A.C.I. Colloids Surfaces 66, 73 (1992). Norde, W. and Favier, J. P. Colloids and Surfaces 64, 87 (1992). Ong, E.; Gilkes, N. R.; Miller, R. C ; Warren, R.A.J. and Kilburn, D. G. Enzyme. Microb. Technol. 13, 59 (1991). Ong, E.; Gilkes, N. R.; Miller, R. C ; Warren, R.A.J. and Kilburn, D.G. Biotechnol. Bioeng., 42, 401 (1993). 313 Pelletier, M. J. Appl. Spectrosc. 45, 395 (1992). Peticolas, W. L., Nafie, L., Stein, P., and Fanconi, B. J. Chem. Phys. 52, 1576 (1970). Pitt, W. G.; Spiegelberg, S. H. and Cooper, S. L. Chapter 21 in Proteins at Interfaces, ACS Symposium Series 343 (Brash and Horbett, eds.), 325, (1987). Plaza, P.; Dao, N. Q.; Jouan, M.; Fevrier, H. and Saisse,H. Appl. Opt., 25, 3448 (1986). Polak, E. "Computational Methods in Optimization", Academic Press, New York (1971). Press, W. H.; Teukolsky, A. A.; Vetterling, W. T. and B. P. Flannery, Numerical Recipes in C (Cambridge, Cambridge, 1992). Privalov, P. L. Advances in Protein Chemistry 33, 167 (1979). Privalov, P. L. and Khechinashvili, N. N. J. Mol. Biol. 86, 665 (1974). Proakis, J. G. and Manolakis, D. G. "Digital Signal Processing: Principles, Algorithms and Applications, 2nd ed." Macmillan, New York (1992). Puppels, G. J.; Huzinga, A., Krabbe, H. W.; de Boer, H. A.; Gijsbers, G. and de Mul, F. F. M. Rev. Sci. Instrum. 61, 3709 (1990). Rainer, F. and Deaton, T. F. Appl. Opt. 21, 1722 (1982). Raman, C. V. and Krishnan, K. S. Nature 121, 501 (1928). Rava, R. P. and Spiro, T. G. "Resonance Enhancement in the Ultraviolet Raman Spectra of Aromatic Amino Acids," J. Phys. Chem. 89, 1856 (1985). Rava, R. P.; and Spiro, T. G. J. Am. Chem. Soc. 106, 4062 (1984). Rodgers, K. R.; Su, C ; Subramaniam, S., and Spiro, T.G. J. Am. Chem. Soc. 114, 3697 (1992). Rose, S. "The Chemistry of Life" Penguin, London, (1991). Rosenfeld, L. Z., Phys. 52, 161 (1928). Rothschild, M. and Abad, H. Opt. Lett. 8, 653 (1983). Savitzky A. and Golay, M. J. E. Anal. Chem. 36, 1624 (1964). 314 Schaeberie, M. D.; Kalasinsky, V. F.; Luke, J. L.; Lewis, E. N.; Levin, I. W. and Treado, P. J. Anal. Chem. 68, 1829 (1996). Schick, G. A. and Sun, Z. Thin Solid Films 248, 86 (1994). Schmid, F. X.; Freeh, C ; Scholz, C. and Walter, S. Biol. Chem. 311, 417 (1996). Schulze, H. G. "The Development of a Fiber-Optic Probe for the In Vitro Resonance Raman Spectroscopy of Neurotransmitters," Ph.D. Thesis, The University of British Columbia, (1996). Schulze, H. G.; Greek, L. S.; Blades, M. W.; Bree, A. V.; Gorzalka, B.B. and Turner, R. F. B. Proc. SPIE 2982, (in press 1997A). Schulze, H. G.; Greek, L. S.; Blades, M. W.; Bree, A. V.; Gorzalka, B.B. and Turner, R. F. B. Proc. SPIE 2982, (in press 1997B). Schulze, H. G.; Greek, L. S.; Blades, M. W.; Bree, A. V.; Gorzalka, B.B., Klein, K.-F. and Turner, R. F. B. Proc. SPIE 2982 (in press 1997C). Schulze, H G ; Blades, M.W.; Bree, A.V.; Gorzalka, B.B.; Greek, L.S.; and Turner, R.F.B. Appl. Spectrosc. 48, 50 (1994). Schulze, H.G; Greek, L.S.; Gorzalka, B.B.; Bree, A.V.; Blades, M.W. and Turner, R.F.B. Journal of Neuroscience Methods 56, 155 (1995). Schwab, S. D. and McCreery, R. L. Anal. Chem. 56, 2199 (1984). Schwab, S. D.; McCreery, R. L. and Gamble, F. T. Anal. Chem. 58, 2486 (1986). Setchell, R. E. Proc. SPIE 1848, 15 (1992). Shannon, C. E. Bell System Tech. J. 27, 379 (1948). Shim, M. G. and Wilson, B. C. Proc. SPIE 2980, paper no. 27 (1997). Silin, A. R. and Lace, L. A. J. Non-Cryst. Solids 149, 54 (1992). Simpson, S. F. and Harris, J. M. J. Phys. Chem., 94, 4649 (1990). Skinner, H. T.; Cooney, T. F.; Sharma, S. K. and Angel, S.M. Appl. Spectrosc. 50, 1007 (1996). Skoog, D. A. and Leary, J. J. "Principles of Instrumental Analysis", Saunders, Fort Worth (1992). 315 Skuja, L. J. Non Cryst. Solids, 179, 51 (1994). Soileau, M. J . ; Van Stryland, E. W. and Williams, W. E. Proc. SPIE 1541, 110, (1985). Song, S. and Asher, S. A. J. Am. Chem. Soc. I l l , 4295 (1989). Song, S.; Asher, S. A.; Krimm, S.; Bandekar, J . J . Am. Chem. Soc. 110, 8547, (1988). Spiro, T. G. and Czernuszewicz, R. S. Methods in Enzymology, 246, 416 (1995). Steadman, B. L.; Thompson, K. C ; Middaugh, C. R.; Matsuno, K.; Vrona, S.; Lawson, E. Q. and Lewis, R. V. Biotech. Bioeng. 40, 8 (1992). Strickland, E. H. CRC Crit. Rev. Biochem. 113 (1974). Su, C ; Wang, Y. and Spiro, T. G. J. Raman Spectrosc, 21, 435 (1990). Sugawara, Y.; Hirakawa, A. Y. and Tsuboi, M. J. Mol. Spectrosc 108, 206 (1984). Surewicz, W. K . ; Mantsch, H. H. Chapter 7 in "Spectroscopic Methods for Determining Protrein Structure in Solution", H. A. Havel (Ed.) pp 135-162, VCH Publishers, New York, 1996 Sweedier, J . V.; Bilhorn, R. B.; Epperson, P. M.; Sims, G. R. and Denton, M. B. Anal. Chem. 60, 282A (1988). Takeuchi, H. and Harada, I. Spectrochim. Acta 42A, 1069 (1986). Takeuchi, H. and Sasamori, J . Biopolymers 55, 359 (1995). Takeuchi, H.; Watanabe, N. and Harada, I. Spectrochim. Acta, 4 4 A , 749 (1988). Talmi, Y. Proc. SPIE S22, 132 (1987). Tanford, C. "The Hydrophobic Effect" Wiley, New York (1980). Tang, J . and Albrecht, A. C. Chapter 2 in "Raman Spectroscopy vol. 2" (H. A. Szymanski, ed.) Plenum, New York, 1970. Tang, J . and Albrecht, A. C. J. Chem. Phys. 49, 1144 (1968). Taylor, R. S.; Leopold, K. E.; Brimacombe, R. K. and Mihailov, S. Appl. Opt. 27, 3124 (1988). 316 Tensmeyer, L. G. and Kauffman, E. W. Chapter 5 in "Spectroscopic Methods for Determining Protrein Structure in Solution", (H. A. Havel ed.) VCH Publishers, New York, (1996). Thamann, T. J. "Probing Local Protein Structure with Ultraviolet Resonance Raman Spectroscopy", Chapter 6 in Spectroscopic Methods for Determining Protein Structure in Solution, (H. A. Havel, ed.) 96 (1996). Tomme, P.; Creagh, L. A.; Kilburn, D. G. and Haynes, C. A. Biochem. 35, 13885 (1996). Tomme, P.; Warren, R. A. J.; Miller, R. C ; Kilburn, D. G. and Gilkes, N. R. "Cellulose Binding Domains - Classification and properties" in Enzymatic Degradation of insoluble carbohydrates (J. N. Saddler and M. H. Penner, eds.) ACS, Washington (1985). Tonks, N. K. and Neel, B. G. Cell, 87, 365 (1996). Trimble, R. B.; Atkinson, P. H.; Tschopp, J. F ; Townsend, R. R. and Maley, F. J. Biol Chem. 266, 22807 (1991). Tse, A. S.; Zhijun, W. and Asher, S. A. Macromolecules 28, 6533 (1995). Tuma, R.; Bamford, J. H. K.; Bamford, D. H.; Russell, M. P. and Thomas, G. J. J. Mol Biol, 257, 87 (1996). Tuma, R.; Russell, M.; Rosendahl, M. and Thomas, G. J. Jr. Biochemistry 34, 15150 (1995). Tsukada, H. and Blow, D. M. J. Mol Biol. 184, 703 (1985). Vidugiris, G. -J. A.; Gudavicius, A. V.; Razumas, V. J. and Kulys, J. J. Eur. Biophys. J. 17, 19 (1989). Vroman, L. "Blood" Natural History Press, Washington (1967). Vroman, L. and Leonard, E. F. Biofouling 4, 81 (1991). Vroman, L. Ann. N. Y. Acad Sci. 516, 300 (1987). Wang, Y. and Van Wart, H.E. "Raman and Resonance Raman Spectroscopy," in Methods in Enzymology (J. F. Riordan and B. L. Vallee, eds.), 226, 319 (1993). Wang, Y.; Purrello, R.; Jordan, T. and Spiro, T.G. J. Am. Chem. Soc. 113, 6359 (1991). Weeks, R. A. J. Non Cryst. Solids, 179, 1 (1994). 317 Weeks, R. A.; Magruder III, R. H. and Wang, P. W. J. Non-Cryst. Solids 149, 122 (1992). Weissman, J. M.; Sunkara, H. B.; Tse, A. S. and Asher, S. A. Science 274, 959 (1996). Wheeler, G. V.; Jolles, B.; Miskovsky, P. and Chinsky, L. Journal of Biomolecular Structure and Dynamics 14, 91 (1996). Williams, R. R. Anal. Chem. 63, 1638 (1991). Williams, R.W. and Dunker, A. K. J. Mol. Biol. 152, 783 (1981). Woody, R. J. J. Chem. Phys., 49, 4797 (1968). Xu, G.-Y.; Ong, E.; Gilkes, N. R.; Kilburn, D. G.; Muhandiram, D. R.; Marees, H.-B.; Carver, J. P.; Kay, L. E. and Harvey, T. S. Biochem. 34, 6993 (1995). Yariv, A. "Quantum Electronics", Wiley, New York, (1992). Zenner, G.; zur Hausen, J. D.; Burn, P. and Mustelin, T. BioEssays 17, 967 (1995). Zhao, X.; Lu, D.; Jiang, S.; Mao, C ; Fan, Y.; An, C. and Li, Z. Science in China Series B 38, 555 (1995). Zhu, Z. Y. and Yappert, M. C. Appl. Spectrosc. 46, 912 (1992A). Zhu, Z. Y. and Yappert, M. C. Appl. Spectrosc. 46, 919 (1992B). Appendix 1. Flush Probe Simulation #include #include #include i^nclude #include #include #include #define EO 1.0 /* total energy per pulse */ #define THETA 0.165806279 /* acceptance angle of fibers in radians */ #define CT 3e-2 /* convergence crriterion */ #define PI 3.141592654 #define NRANSI include "nr.h" include "nrutil.h" /* written by: Shane Greek 14 November, 1994 Last Updated: 12 May, 1995 */ /* This program simulates the excitation and collection process of a direct-sense double cylindrical fiber, flush, parallel, fiber optic probe It integrates the exc-coll function (see notes, Shane Greek, Fall 94 to Winter/spring '95) over the area of overlap of exc and coll acceptance cones. It provides, for each des'd concentration, a file showing the differential and cumulative amounts of Raman scattered energy collected from 'z-slices* (i.e. slices perp. to the fiber axes and parallel to the fiber faces) for an absorbing analyte. It also calculates the signal due to a '1' concentration of a non-abs analyte in the presence of an absorbing one (e.g. Raman at 472.7 nm of a mixture of KN03 (non-abs , peak @ 1048 cm-1) and Methyl Orange (abs, peak ca. 1400 cm-1)) It also produces a log file summarizing the results of all of the concentrations in one file. In essence, it computes a quintuple-integral (3 integrals over the volume of overlap and two over the area of collection) - the discretized integration elements of the area are fixed throughout the integration; however the volume elements get larger further away from the fiber faces for efficiency. This program models the expts done By Shane Greek, Georg Schulze and Onita Basu in early 1995 using KN03 and Methyl Orange (see Shane Greek's lab book, Onita Basu's Chem Eng undergrad thesis It accepts as input: 1. Ex and Coll Fiber diameters in microns 2. Fiber separation 3. Analyte absorpivity (1 /Molar*micrometers) at exc and coll wavelength 4. The abs analyte concentrations 5. exc and coll eff profiles (as a function of theta in degs) These are params defining a polynomial (exc profile) and a Gaussian (coll profile) */ /* function evaluates polynomial c of order nc-1 at x : */ double polval(float x, int nc, float c[]) { int ic; double result; result=c[l]; for(ic=2;ic<=nc;ic++) result+=x*result+c[ic]; return result; } int main(void) { /* output files : */ FILE *fout,*flog; /* polynomial (exc) and gaussian (coll) param input files : */ FILE *fex,*fco,*ftst; /* file names: */ char namel[12],log[12],exname[12],coname[12]; char ans; /* coords of dv excited (xe,ye,z) and dA for collection (xc,yc,0) */ double z,xc,yc,xe,ye; double rc,re; /* radii of cones at z */ /* zA2, zA2+(xc-xe)A2, : precomp saves time: */ double z2,z2sc; /* limits of integration: */ double xemin,xemax,yemin,yemax,xcmin,xcmax,ycmin,ycmax; double xint,yint; /* pos of intersection of exc and coll circles */ 320 double DW; /* differential width , um */ double DA; /* differential area = sqr(DW) */ double zO; /* dist of imaginary point source below exc. fiber face */ double le; /* distance light ray travels through solution to dv */ double lc; /* distance from dV scatt. element to dA coll. element */ double K; /* multiplicative coeff = flu*CONC*dv*DA*z */ double K2; /* multiplicative coeff for non abs = flu*dv*DA*z */ double ZdvCos; /* z divided by cos theta - used to check if scattered ray is accepted by coll fiber */ double expc; /*precalc of exp(AbsE*lc)/lcA3 */ double dve;/*volume scattering element-excited*/ /* increments for x,y, and z in excitation cone */ double dxe,dye,dze; double SEP; /* separation of fiber cores */ double RC; /* radius of collection fiber */ double RE; /* radius of excitation fiber */ double ASEP; /* axial separation =S+RC+RE */ float *CONC; /* concentration vector of analyte, mg/ml */ /* vectors w polynomial & gaussian params for exc. & coll profile ests: */ float *polyex,*gaussco; /* look up table vectors of above polynomials : */ float *exvec,*covec; double comax; /* max val of covec, for normalization */ double flu; /* fluence at dv*/ /* E coll. from one z slice, total energy collected */ double eslice,ecoll; /* for absorbing analyte */ /* energy per z slice,total coll.E: non-abs analyte */ double esna,ecna; /* maximum val of esna: */ double esnmax; /* absorptivity coeff at exc and coll wavelength, 1/um */ double AbsE,AbsC; /* Molar absorptivity of analyte at exc and collection wavelengths, l/(um*mg/ml): */ double EpsE,EpsC; /* absorptivity of medium at excitation and collection wavelengths, 1/um: */ double MedAbE,MedAbC; double ExSum; /* integral to normalize excit. eff. profile */ double C; /* exc eff profile normalization coeff */ int icon,ncon; /* looping var, # different concentrations to run */ int anidx; /* integer index 1-100 for angle 0-10 degs for coll*/ int anex; /* integer index 1-100 for angle 0-10 degs for exc*/ int i,j,k; /* polynomial looping variables */ /* time variables to determine elapsed time to convergence */ time_ttl,t2; /* allocate space for vectors: (w NR routine) */ CONC=vector(l,30); polyex=vector( 1,10); gaussco=vector( 1,7); exvec=vector( 1,100); covec=vector( 1,100); /* loop over user input */ do{ printf("\n\nW'); fflush(stdin); puts("Input radius of exc. fiber in microns..."); scanf\"%lf&RE); fflush(stdin); puts("Input radius of coll. fiber in microns..."); scanf("%lf', &RC); puts("Input separation of two fibers in microns..."); scanf("%lf',&SEP); ASEP=RE+RC+SEP; puts("Input spec.abs. of analyte at exc. wavelength (l/(um*mg/ml))"); scanf("%lf,,&EpsE); puts("Input spec.abs. of analyte at coll. wavelength (l/(um*mg/ml))"); scanf("%lf',&EpsC); /* puts("Input abs. of medium at exc. wavelength (1/um)"); scanfT%lf',&MedAbE); puts("Input abs. of medium at coll. wavelength (1/um)"); scanf^yoh^&MedAbC); */ MedAbE=0.0; MedAbC=0.0; puts("Input number of concentrations to run"); scanf("%d",&ncon); for(icon=l ;icon<=ncon;icon++) { printf("Input concentration to # %3d...",icon); scanf("%f\&CONC[icon]); } /* size of area elements on collection face: */ 322 DW=RC/40.0; DA=SQR(DW); printf("RC= %8.2f um RE= %8.2f um \n", RC, RE); printf(,,ASEP= %8.2f um SEP= %8.2f um \n",ASEP,SEP); printf("DW= %8.2fumDA= %8.2f um \n\n",DW,DA); printf("MOLAR ABS in l/[um*mg/ml] of analyte at: \n"); printf("excit. wavelength: %12.6f coll. wavelength: %12.6f\n",EpsE,EpsC); printf("ABS in 1/um of medium at: \n"); printf("excit. w.l.: %12.6f coll. w.l.: %12.6f\n\n",MedAbE,MedAbC); printf("Concentrations: \n"); printf(" -\n"); icon=l; while(icon<=ncon) { printf("Conc[ %2d ]=%12e",icon,CONC[icon]); icon++; if(icon<=ncon) printfC Conc[ %2d ]=%12e \n",icon,CONC[icon]); icon++; } printf("\n\n\n"); fflush(stdin); puts("Input first three letters of file name for out file..."); gets(namel); puts("Input name for log file (including dot extension)..."); gets(log); puts("Input name of file with poly, coeffs for EXC. profile..."); gets(exname); puts("Input name of file with Gaussian params for COLL. profile..."); gets(coname); printf("Are these values correct (y/n)"); ans=getcharO; putchar(ans); } while(ans != 'y'); /* end loop over user input */ strcat(namel,"A01.dat"); /* prepare extension for O/P files */ 323 /* read in exc profile polynomial coefficients: */ if((fex=fopen(exname,"r"))!=NULL) { for(i=l;i<=ll;i++) { fscanf(fex,"%f,,&poryex[i]); } fclose(fex); } else printfTERROR MESSASGE"); printf("\n\n\n"); /* read in coll profile gaussian coefficients: */ if((fco=fopen(coname,"r"))!=NULL) { for(j=lj<=7u++) { fscardTfco, "%f',&gaussco[j]); } fclose(fco); } else printf("ERROR MESSASGE"); /* precalculate 100 vals of covec for lookup vector */ comax=0.0; forcomax) comax=covec[j]; } /* NORMALIZE: */ for(j=lj<=100j++) { 324 covec[j]=covecrj]/comax; } /* precalculate 100 vals of exvec for lookup table */ for(i=l;i<=100;i++) { exvec[i]=polval(-1 *(i-1)*0.10,11 ,polyex); if (exvec[i]<0.0) exvec[i]=0.0; if((i*0.1)>7.5) exvec[i]=0.0; } /* compute integral of exvec*theta*d(theta) for normalization later */ /* theta in radians */ /* d(theta)=0.1 degree = 1.7453e-3 radians */ ExSum=0.0; for(i=l;i<=100;i++) ExSum+=exvec[i]*1.7453e-3*(i-l)*10/57.3; ifX(ftst=fopen("ftst.tst^ ,w"))!=NULL) { for(k=l;k<=100;k++) { tprintfl[ftst,"0/of\nM,exvecrk]); } fclose(ftst); } else printf("ERROR MESSASGE"); printfr." \n STARTING LOOP OVER Z SLICES \n"); hX(flog==fopenOog,"wM))!=NULL) /* open log file */ { /* write various parameters to log file: */ fprintftflog,"RE= %8.2f RC=%8.2f SEP= %8.2f (um)\n",RE,RC,SEP); fprintf(flog,,,DW= %8.2f DA= %8.2f (um)\n",DW,DA); fprintf(flog,"EpsE= %12.6G EpsC= %12.6G (l/[um*mg/ml])\n",EpsE,EpsC); fprintf(flog,"MedAbE= %12.6G MedAbC= %12.6G (l/um)\n",MedAbE,MedAbC); /* begin loop over various concentrations */ for(icon=l ;icon<=ncon;icon-H-) 325 { /* determine next file name - format is xxxAyy.dat where xxx is any three letters as input above and yy is 01, 02 etc incrementing by one for each successive concentration value */ namel[4]=48+icon/10; namel [5]=(icon-(icon/10)* 10)+48; if((fout=fopen(namel,,,w"))!=NULL) /* open icon'th file for O/P */ { /* minus one times absporp coeff (1/um)- the factor Of 2.302 (=ln(10)) takes ito account the fact that absorption coeff is defined on the basis of exponents of 10, whereas in this program I use eAx */ AbsE=-2.30258509*(CONC[icon]*EpsE+MedAbE); /*at excitation lambda */ AbsC=-2.30258509*(CONC[icon]*EpsC+MedAbC); /*at collection lambda */ eslice=le-22; /* initialize collected energy/slice variables */ esna=le-22; esnmax=esna; ecoll=le-22; /* initialize total collected energy variables */ ecna=le-22; /*initialize z value to first intersections of cones: */ z=SEP/(2*tan(THETA))+50.0; z+=0.1; /* distance below fiber face of (imaginary) point source */ zO=RE/(tan(THETA)); tl=time(NULL); /* start time */ /* loop over z slices, converge when the signal from one z-slice of the non-abs anaylte is a fraction CT of its max value */ while((((esna/esnmax)>CT)&&(z<25000))||(z<1000.0)) { ecoll+=eslice; ecna+=esna; if(esna>esnmax) esnmax=esna; printf("C= %6.3e z=%6.0f Ec= %6.3e ES/Emx= %6.3e ", CONC [icon] ,z, ecoll, esna/ esnmax); t2=time(NULL); printf("t= %ld \n",t2-tl); fprintf(fout,"%10.6f %10.6E %10.6E %ld \n",z,ecoll,ecna,t2-tl); eslice=0.0; esna=0.0; z2=SQR(z); re=RE+z*tan(THETA); rc=RC+z*tan(THETA); /* x position of intersection of exc and coll circles */ xint=(SQR(re)-SQR(rc)+SQR(ASEP))/(2*ASEP); xemin=ASEP-rc; xemax=re; xe=xemin; yint=sqrt(SQR(re)-SQR(xint)); dxe=((re)/70.0)*((2.5-2.5*exp(-l*z/8000))+0.8); dye=dxe; dze=dxe; ZdvCos=1.013905099*z; /* z/cos(9.5 degs) */ /* normalization coefficient for excitation eff. profile: (see notes, Shane Greek, Feb/March/April 1995) :*/ OSQR(re)/(2* SQR(z+zO)*ExSum); /* loop over x direction in exitation cone */ while(xe<=xemax) { if(xe #include #include #include #include #include #include #define EO 1.0 /* total energy per pulse in microjoules*/ #define PI 3.141592654 #defineNRANSI include "nr.h" include "nrutil.h" /* written by: Shane Greek updated: 26,27,28 January, 2,3,4 February, 27 June 1997 */ /* This program simulates the excitation and collection process of an angled, mirrored fiber optic probe of the type reported in 1996/97 in various papers. It was created by modifying uvfib.c It calculates for a Raman cross section of 1 mbarn, using an energy into the sample of 1 microjoule. Since integral is linear in these, the final answer can be scaled to any sigma, Ein All units in the final integration are in metres, molecules, and photons. It accepts as input: 1. Ex and Coll Fiber diameters in microns 2. Fiber separation 3. Analyte absorpivity (l/(molec/mA3)m) at exc and coll wavelength 4. The abs analyte concentration in molec/mA3 5. Medium absorbance (mA-l) @ exc and coll wavelengths It integrates over the volume defined by the cylinder of excitation. It assumes that the beam exiting the excitation fiber is roughly collimated. The angular excitation profile is assumed to be flat; however it is possible to add any arbitrary profile. The angular efficiency of collection is defined simple by angular dependence of Fresnel reflection/transmission */ int main(void) { /* output files : */ FILE *fout,*flog,*fprof,*fzprof; /* file names: */ char name 1 [ 12],log[ 12],profile[ 12],zprof[ 12]; char ans; /* coords of dv excited (xe,ye,z) and dA for collection (xc,yc,0) */ double z,zc,xc,yc,thc,xe,ye; double zbias; /* amount that exc fiber is recessed from z=0 plane */ /* limits of integration: */ double LAM; /* wavelength in metres */ double xemin,xemax,yemin,yemax,zcmin,zcmax,thcmn,thcmx; /* coords of trans, ray when it reaches the reflecting (x=z) plane: */ double xr,yr; ./* Fresnel reflection coefficients: */ double rl2s,rl2p; double argu; /* sqrt argument */ double Dzc; /* differential width, um */ double DA; /* differential area = RE*Dthc*Dzc */ double Dthc; /* differential angle */ double lc; /* distance from dV scatt. element to dA coll. element */ double dve;/*volume scattering element-excited*/ /* increments for x,y, and z in excitation cone */ double dxe,dye,dze; double dNs; /* differntial scattered from dV, collected at dA */ /* collected amounts from single volume elements, used for spat, profile: */ double NdV,NdVz,temp; double Fres; /* squared Fresnel reflection coeff */ double sigma; /* differential raman scat, x-sect, in mA2/(molec sr) */ double SEP; /* separation of fiber cores */ double RC; /* radius of collection fiber */ double RE; /* radius of excitation fiber */ float *CONC; /* concentration vector of analyte, molec/mA3 */ double fluO,flu; /* fluence z=0, dv*/ /* E coll. from one z slice, total energy collected */ double eslice,ecoll; /* for absorbing analyte */ double Nscat; /* total scattered photons in sample volume */ /* absorptivity coeff at exc and coll wavelength, 1/m */ double AbsE,AbsC; /* Molar absorptivity of analyte at exc and collection wavelengths, l/(m*molec/mA3): */ double EpsE,EpsC; /* absorptivity of medium at excitation and collection wavelengths, 1/m: */ double MedAbE,MedAbC; /* normalization coefficients: */ double ExSum; /* integral to normalize excit. eff. profile */ double C; /* exc eff profile normalization coeff */ int icon,ncon; /* looping var, # different concentrations to run */ int an_idx; /* integer index 1-100 for angle 0-10 degs for coll*/ int anex; /* integer index 1-100 for angle 0-10 degs for exc*/ int i; /* polynomial looping variables */ float *ks,*kt,*kr; /* scattered, refracted (trans.) & reflect unit vectors */ float *unc,*ur; /* coll surface & reflect surface normal unit vector */ double cos_i,sin_i,cos_t,sin_t,cos_r; /*sines and cosines of angles*/ float ksmag,krmag,ktmag; /* scattered vector magnitude */ float ni,nt; /* refract index of water(incident), silica(transmitted) */ /* time variables to determine elapsed time to convergence */ timet tl,t2; /* allocate space for vectors: (w NR. routine) */ CONC=vector(l,30); ks=vector(l,3); kt=vector(l,3); kr=vector(l,3); ur=vector(l,3); unc=vector(l,3); /* reflective surface normal unit vector: */ ur[l]=1.0/(sqrt(2.0)); ur[3]=-1.0/(sqrt(2.0)); ur[2]=0; /* indices of refraction of water and silica at LAM: */ ni=1.33; nt=1.50; /* cross sect in barns = 1 mbarn = 0.001 * le-28 mA2/(molec*sr) */ sigma=le-31; /* default vals for testing: (comment out if. req.) */ LAM=227e-9; RE=200e-6; RC=330e-6; SEP=50e-6; EpsE=2.66e-21; /* absorptivities in mA2/molec */ EpsC=8.85e-22; MedAbE=0.0; /* med. extinction coefficients in 1/m */ MedAbC=0.0; ncon=2; /* 2 concentrations */ Nscat-0.0; zbias=0e-6; /* amount that exc fiber is recessed from z=0 plane*/ /* loop over user input */ do{ /* user entered input- comment out to use default values: */ printf("\n\n\n"); fflush(stdin); puts("Input radius of exc. fiber in metres..."); scanf("%lf&RE); fflush(stdin); puts("Input radius of coll. fiber in metres..."); scanfO'o/olf, &RC); puts("Input separation of two fibers in metres..."); scanfr%lf',&SEP); puts("Input spec.abs. of analyte at exc. wavelength (l/(m*molec))"); scanfC'yolf'.&EpsE); puts("Input spec.abs. of analyte at coll. wavelength (l/(m*molec))"); scanfr%lf',&EpsC); puts("Input excitation wavelength in metres"); scarif("%lf\&LAM); /* Medium extinction coefficient, comment out if not used: */ /* puts("Input abs. of medium at exc. wavelength (1/m)"); scanfC%lf',&MedAbE); puts("Input abs. of medium at coll. wavelength (1/m)"); scarnT/Toh^&MedAbC); 333 puts("Input number of concentrations to run"); scanf("%d",&ncon); /* initialize vars for profiling: */ NdV=0.0; NdVz=0.0; /* read in analyte concentrations: */ for(icon= 1 ;icon<=ncon;icon++) { printf("Input concentration (molec/mA3) # %3d...",icon); scanf("%e",&CONC[icon]); } printf("RC= % 1.4e m RE= % 1.4e m \n", RC, RE); printf("SEP= %1.4e m \n",SEP); printf("MOLAR ABS in mA2/molec of analyte at: \n"); printf("excit. wavelength: %1.4e coll. wavelength: %1.4e\n",EpsE,EpsC); printf("ABS in 1/m of medium at: \n"); printf("excit. w.l.: %1.4e coll. w.L: %1.4e\n\n",MedAbE,MedAbC); printf("Concentrations: \n"); printf(" -\n"); icon=l; while(icon<=ncon) { printf("Conc[ %2d ]=%12e",icon,CONC[icon]); icon++; if(icon<=ncon) printf(" Conc[ %2d ]=%12e \n",icon,CONC[icon]); icon-H-; } printf("\n\n\n"); fflush(stdin); puts("Input first three letters of file name for out file..."); gets(namel); puts("Input first three letters of file name for profile file..."); gets(profile); puts("Input first three letters of file name for Z profile file..."); gets(zprof); 334 puts("Input name for log file (including dot extension)..."); gets(log); printf("Are these values correct (y/n)"); ans=getcharO; putchar(ans); } while(ans != 'y'); /* end loop over user input */ strcat(namel,"A01.dat"); /* prepare extension for O/P files */ strcat(profile,"B01.dat"); /* prepare extension for profile files */ strcat(zprof,"C01 .dat"); /* prepare extension for z profile files */ printf(" \n STARTING LOOP OVER Z SLICES \n"); /* volume resolution of integral: */ dxe=RE/40.0; dye=dxe; dze=dxe; dve=dxe*dye*dze; /* angular resolution over collection face: */ Dthc=(2*PI)/90.0; if^ (fprof=fopen(profile,"w"))!=NULL) printf("Profile file is opened \n"); if((fzproMopen(zprof,"w"))!=NULL) printf("Z Profile file is opened \n"); if((flog=fopen(log,"w"))!=NULL) /* open log file */ { /* write various parameters to log file: */ fprintf(flog,"RE=%1.4e RC=%1.4e SEP= %1.4e (m)\n",RE,RC,SEP); fprintfitflog, "dze= %1.4e (m)\n",dze); fprintpog,"EpsE= %12.6G EpsC= %12.6G (mA2/molec)\n",EpsE,EpsC); fprintf(flog,"MedAbE= %12.6G MedAbC= %12.6G (l/m)\n",MedAbE,MedAbC); /* begin loop over various concentrations */ for(icon= 1 ;icon<=ncon;icon++) { 335 /* determine next file name - format is xxxAyy.dat where xxx is any three letters as input above and yy is 01, 02 etc incrementing by one for each successive concentration value */ name 1 [4]=48+icon/10; namel[5]=(icon-(icon/10)*10)+48; if((fout=fopen(namel,"w"))!=NULL) /* open icon'th file for O/P */ { /* minus one times absporp coeff (1/m)- the factor Of 2.302 (=ln(10)) takes ito account the fact that absorption coeff is defined on the basis of exponents of 10, whereas in this program I use e*x */ AbsE=-2.30258509*(CONC[icon]*EpsE+MedAbE); /*at excitation lambda */ AbsC=-2.30258509*(CONC[icon]*EpsC+MedAbC); /*at collection lambda */ eslice=0.0; /* initialize collected photons/slice var. */ ecoll=0.0; /* initialize total collected photon var. */ z=290e-6; /* integration starts at z=0 */ /* fluence at z=0 in units of photons/mA2: */ flu0K(E0*le-6*LAM)/((6.626e-34)*(3e8)))/(PI*SQR(RE)); tl=time(NULL); /* start time */ /* loop over z slices, stop when z is exceeds the maximum possible extent of the probe (collection) volume */ /* while inside of collection volume: */ /* while(z<(2*RC+(2*RE*tan(15.0*PI/180.0)))) */ while(z<310e-6) { ecoll+=eslice; printf("C=%l.lez-%l.le Nc=%1.2e delNc=%1.2e Ns=% 1.1 e",CONC[icon],z,ecoll,eslice,Nscat); t2^ime(>fULL); printf(" t=%ld \n",t2-tl); fprintf(fout,"%1.4e %1.4e %1.4e %1.4e %ld \n",z,Nscat,ecoll,eslice,t2-tl); eslice=0.0; /* x limits of integration on excitation fiber go from point on the core/cladding interface nearest the point of contact of exc & coll fibers to opposite end of exc fiber along x axis: */ xemin=2*RC+SEP+dxe; xemax=2*RC+2*RE+SEP; xe=xemin; /* loop over x direction in exitation cone */ while(xe<=xemax) { /* y limits on exc go from x-axis of exc fiber to line delimited by circle that is exc fiber core boundry: (add small amt so never get sqrt(neg))*/ yemin=0; argu=SQR(RE)-SQR(xe-(2*RC+SEP+RE)); if (argu<0.0) argu=le-12; yemax=sqrt(argu); ye=yemin; /* loop over y direction in excitation cone */ while(ye<=yemax) { /* set collec. z limits 25 degs above and below : */ zcmin=z-((xe-(2*RC))*tan(15.0*PI7180.0)); zcmax=z+((xe-(2*RC))*tan(15.0*PI/180.0)); if (zcmax>(2*RC)) zcmax=2*RC; Dzc=(zcmax-zcmin)/50.0; zc=zcmin; /* fluence (Energy/Area) at dv: */ flu=fluO*exp(AbsE*(z+zbias)); /* Total photons scattered from dV: */ dNs=4*PI*flu*CONC[icon]*dve*sigma; Nscat+=dNs; /* loop over z direction of collection fiber face */ while(zc<=zcmax) { thcmn=-l*PI/2.0; thcmx=PI/2.0; if(zc>RC) { thcmn=-1 *acos((zc-RC)/RC); thcmx=-1 *thcmn; } /* size of differential area element */ DA=Dzc*Dthc*RC; thc=thcmn; /* loop over y direction of collection face */ while(thc<=thcmx) { /* outward pointing unit vector normal to coll. surface: */ unc[l]=cos(thc); unc[2]=sin(thc); unc[3]=0.0; /* rectangular coords of differential collection element */ xc=RC+RC*unc[l]; yc=RC*unc[2]; /* calculate scattered unit vector ks and scattering path length: */ ks[l]=xc-xe; ks[2]=yc-ye; ks[3]=zc-z; ksmag=sqrt(SQR(ks[l])+SQR(ks[2])+SQR(ks[3])); lc=ksmag; /""normalize: : */ for(i=l;i<=3;i++) { ks[i]=ks[i]/ksmag; } /*calc. cosine of incident angle=dot prod of ks and unc: (but add a -1 factor due to angle/dir definitions) */ cos_i=0.0; for(i=l;i<=3;i++) { cos_i+=-1.0*ks[i]*unc[i]; } argu=l-SQR(cos_i); if (argu<0.0) argu=le-12; sin_i=sqrt(argu); sin_t=(ni * sin_i)/nt; argu=l-SQR(sin_t); if (argu<0.0) argu=le-12; cos_t=sqrt(argu); /* calculate trans, (refracted) unit vector : (see Hecht and Zajac, p65, 1979)*/ for(i=l;i<=3;i-H-) { kt[i]=(-1 *(nt*cos_t-ni*cos_i))*(unc[i]/nt); kt[i]+=(ni*ks[i])/nt; } ktmag=sqrt(SQR(kt[l])+SQR(kt[2])+SQR(kt[3])); /*normalize: : */ for(i=l;i<=3;i-H-) { kt[i]=kt[i]/ktmag; } /* collect only if transmitted ray hits mirror, mirror is in z=x plane, ray hits mirror if when it passes through z=x plane, it is within the cylinder defiend by the collection fiber: */ yr=yc+((zc-xc)/(kt[ 1 ]-kt[3]))*kt[2]; xr=xc+((zc-xc)/(kt[ 1 ]-kt[3]))*kt[ 1 ]; if (sqrt(SQR(yr)+SQR(xr-RC))>RC) Fres=0.0; /* if ray hits mirror, calc. trans, energy fraction at water/silica interface using Fresnel reflection coeff. as given by Reitz, Mifford, & Christy, "Foundations of Electriomagnetic Theory" 3rd ed, pp385-391. consider incident ray to be 1/2 p and 1/2 s polarized: */ else { /* Fresnel refl. coeff. for s-polarization:*/ rl2s=(ni*cos_i-nt*cos_t)/(ni*cos_i+nt*cos_t); /* Fresnel refl. coeff. for p polarization: */ rl2p=(nt*cos_i-ni*cos_t)/(nt*cos_i+ni*cos_t); Fres=0.5 * ((1 -SQR(r 12s))+(l -SQR(r 12p))); } /* dont collect if ray is incident from the INSIDE of the collection fiber! */ if(cos_i<0.0) Fres=0.0; /* dot product of kt and ur is the same as cosine of angle between transmitted ray and reflective surface normal: */ cos_r=0.0; for(i=l;i<=3;i++) { cos_r+=-1.0*kt[i]*ur[i]; } /* calculate reflected unit vector: */ for(i=l;i<=3;i++) { kr[i]=kt[i]+(2*cos_r*ur[i]); } krmag=sqrt(SQR(kr[ 1 ])+SQR(kr[2])+SQR(kr[3])); /*normalize: : */ for(i=l;i<=3;i++) { kr[i]=kr[i]/krmag; } /* only include if angle between kr and ka (ray & collec fiber axis vector) < 8.0 degs: */ if(((-1.0*kr[3])>0.990268)&&(cos_i>0.0)) { temp=Fres*(dNs*exp(lc*AbsC)*DA*cos_i)/(4*PI*SQR(lc)); ifiye<(dye/2.0)) NdV+=temp; if(fabs(z-(300e-6))<(dze)) NdVz+=temp; eslice+=temp; } thc+=Dthc; } zc+=Dzc; } /* store profile data and increment y (excit.): (comment out profile data when not needed) */ /* */ if(ye<(dye/2.0)) rprintf(fprof,"%1.3e %1.3e %1.3e \n",xe,z,NdV); if(fabs(z-(300e-6))<(dze)) fprintf(fzprof,H%1.3e %1.3e %1.3e \n",xe,ye,NdVz); NdV=0.0; /* reset photons collect, from dV for y profile */ NdVz=0.0; /* reset photons collect, from dV for z profile */ ye+=dye; } xe+=dxe; } z+=dze; } /* end time */ t2=time(NULL); /* write final results for present cone, to log file: */ fprintf(flog,"\n\n FILE : %12s CONC: %1.4e\n",namel,CONC[icon fprint^flog/TIME: %81d \n",t2-tl); fprintffflog," TOTAL COLLECTED : %1.4e\n",ecoll); fprintf(flog," TOTAL SCATTERED : % 1 4e\n",Nscat); fclose(fout); } /* end output file opening IF statement */ else printf("error messageW); } /* end loop over various concentrations */ fclose(flog); } /* end log file opening IF statement */ else printf("error message"); fclose(fzprof); /* close z=300 profile file */ fclose(fprof); /*close y=0 profile file */ return 0; Appendix 3. Signal Processing Programs PSDA1: % file to estimate PSD of Raman signal from a series of spectra format compact; %prevents extra line feeds clear;clf; %clears variables and figure\ name=input('What are first 5 letters of file name? \'s'); %create 2 sub-plotted figure windows: figure(l); subplot(2,2,l); figure(2); subplot(2,2,l); figure(3); subplot(2,2,l); %95 percent confidence limit for PSD conf. interval calc: p=0.95 %sampling frequency for PSD routine Fs=0.98; %calculate 4 different PSDs at nfft=64,128,256,512 for j=1:4; %size of FFT nfft=32*2Aj; %reset data matrix data=[]; %50 percent overlapping points used in PSD routine noverlap=nfft/2; %Process 8 spectra for i=2:9; %determine proper ending of file using integer division fhame=[name,int2str(i-fix(i/10)*10),,.prn1]; %read in data from file pid_a=fopen(fname,'r'); temp=fscanf(pid_a,,%g %g',[2 infj); %discard unintensified portions of detector read-out: temp=(temp(2,289:800))'; fclose(pid_a); %periodogram est. of PSD using MATLAB PSD routiune: Pxx,Pxxc,fxx]=psd(temp,nfft,Fs,[],noverlap,p); %Combine consecutive psd estimates in 1 matrix data=[data Pxx]; end; %plot (normalized) PSD estimate, average of 8 PSD estimates data_mean= 10 * log 10(mean(data')); figure(l); subplot(2,2j); plo^fxXjdatamean-max^atamean),'-'); axis([0,Fs/2,-50,5]); title(sprintf('psd est. nfft= %3d',nfft)); grid; %plot std dev of 8 spectral estimates data_std=10*logl0(std(data')); figure(2); subplot(2,2,j); plot(fxx,data_std-max(max(data_std))); axis auto; title(sprintf('s (n=8). nffi= %3d',nfft)); grid; %plot std dev of 4th spectral estimate Pxxc=10*logl0(std(data')); figure(3); subplot(2,2j); plot(fxx,Pxxc); axis auto; title(sprintfC95 percent conf. int.nfft= %3d',nfft)); grid; figure(l); subplot(2,2,4); xlabel('Freq. (cm)'); subplot(2,2,3); xlabel('Freq. (cm)'); subplot(2,2,3); ylabelCNORMALIZED magnitude (dB)'); figure(2); subplot(2,2,4); xlabel('Freq. (cm)'); subplot(2,2,3); xlabel('Freq. (cm)'); subplot(2,2,3); ylabel('std dev (n=8) (dB)'); figure(3); subplot(2,2,4); xlabel('Freq. (cm)'); subplot(2,2,3); xlabel('Freq. (cm)'); subplot(2,2,3); ylabel('95 percent conf. int. (dB)'); PSD_A2: % file to estimate PSD of Raman signal from a series of spectra format compact; %prevents extra line feeds clear;clf; %clears variables and figure\ name=input(What are first 5 letters of file name? ','s'); %create 2 sub-plotted figure windows: figure(l); subplot(2,2,l); figure(2); subplot(2,2,l); %sampling frequency for PSD routine Fs=2*pi; %confidence level for PSD to calculate confidence intervals p=0.95; %size of FFT nfft=128; forj=l:4; %reset data matrix data=[]; %number of overlapping points used in PSD routine noverlap=32*(j-l); %Process 8 spectra for i=2:9; %determine proper ending of file using integer division fiiame=[name,int2str(i-fix(i/10)*10),'.prn']; %read in data from file pid_a=fopen(fname,V); temp=fscannjpid_a,'%g %g',[2 infj); %discard unintensified portions of detector read-out: temp=(temp(2,289:800))'; fclose(pida); %periodogram est. of PSD using MATLAB PSD routiune: [Pxx,fxx]=psd(temp,nfft,Fs,[],noverlap); %Combine consecutive psd estimates in 1 matrix data=[data Pxx]; end; %plot (normalized) PSD estimate, average of 8 PSD estimates data_mean= 10 * log 10(mean(data')); figure(l); subplot(2,2j); plot(fxx,data_mean-max(data_mean),'-'); axis([0,Fs/2,-50,5]); title(sprintfCpsd est. # overlap= %3d',noverlap)); %plot std dev of 8 spectral estimates data_std=10*logl0(std(data')); figure(2); subplot(2,2j); plot(fxx,data_std-max(max(data_std))); axis auto; title(sprintf('s (n=8). # overlap= %3d',noverlap)); end; figure(l); subplot(2,2,4); xlabel(Treq. (cm)1); subplot(2,2,3); xlabel(Treq. (cm)1); subplot(2,2,3); ylabel('magnitude (dB)'); PSD_A4: % file to estimate PSD of Raman signal from a series of spectra format compact; %prevents extra line feeds clear;clf; %clears variables and figure\ %95 percent confidence limit for PSD conf. interval calc: p=0.95 %sampling frequency for PSD routine Fs=0.98; %size of FFT nfft-128; %reset data matrix data=zeros(l,512); %50 percent overlapping points used in PSD routine noverlap=64; %accept 4 files forj=l:4 %files name: name=input('What are first 5 letters of file name? ','s'); %Process 2 spectra for i=l:2; %determine proper ending of file using integer division fhame=[name,int2str(i-fix(i/l 0)* 10),'.prn']; %read in data from file pid_a=fopen(fhame,Y); temp=fscanf(pid_a,,%g %g',[2 infj); %discard unintensified portions of detector read-out: temp=(temp(2,289:800)); fclose(pid_a); data=data+temp; end; %periodogram est. of PSD using MATLAB PSD routiune: [Pxx,Pxxc,fxx]=psd(data, nfft,Fs, [], noverlap,p); %plot (normalized) PSD estimate, average of 8 PSD estimates Pxx=10*loglO(Pxx); figure(l); subplot(2,2j); plot(fxx,Pxx-max(Pxx),,-,); axis([0,Fs/2,-50,5]); title(sprintfCpsd est. file: %s',name)); grid; end; figure(l); subplot(2,2,4); xlabel('Freq. (cm)'); subplot(2,2,3); xlabel(Treq. (cm)'); subplot(2,2,3); ylabelCNORMALlZED magnitude (dB)'); PSD_B: %f file to estimate PSD of Raman signal from a series of spectra using AR format compact; %prevents extra line feeds clear;clf; %clears variables and figure\ name=inputCWhat are first 5 letters of file name? ','s'); subplot(3,2,l); %Fs=input('What is sampling frequency (Fs)? '); Fs=0.98; forj=I:6; subplot(3,2j); n^^'-l) data=[]; for i=2:8; %fhame command adds rest of ending to file using integer division fhame=[name,int2str(i-fix(i/10)*10),,.prn,]; pid_a=fopen(fharne,,r'); temp=fscanf(pid_a,,0/og %g',[2 infj); temp=(temp(:,200:800))'; fclose(pid_a); a=lpc(temp(:,2),n); [Pxx,fxx]=freqz(l,a,2048); Pxx=20*log 10(abs(Pxx)); Pxx=Pxx-max(Pxx); data=[data Pxx]; end; data=mean(data'); plot(fxx,data); axis([0,0.5,min(data),max(data)]); title(sprintfCMEM psd est, n= %4d \n file=%s',n,name)); xlabel(Treq (cm)'); end; FIRA: %m-file to investigate fir filter for Raman spectra 348 format compact; %prevents extra line feeds clear;clf; %clears variables and figure\ name=input('What are first 5 letters of file name? ','s'); %sampling freq: Fs=0.98; fwidth=0.01; forj=l:25 fcutoff=0.04+0.005*j; data=[]; n=60; %filter order fparm=[0 fcutoff-fwidth fcutoff+fwidth 1]; %pass/stop band edges of LP filter m=[l 10 0]; %mag. resp of LPF wparm=[l 1]; %rel weighting of ripples in pass/stop bands %design filter using remez exchange algorithm: b=remez(n, fparm, m, wparm); Hb=fn(b,1024V; %initialize snr vector peak=[]; for i=2:8; %fname command adds rest of ending to file using integer division fhame=[name,int2str(i-fix(i/l 0)* 10),'.prn']; %read in data from file pid_a=fopen(fhame,,rl); temp=fscanf(pid_a,,%g %g',[2 inf]); temp=(temp(2,200:800))'; fclose(pid_a); %fft data Ftemp=fft(temp,1024); %convolve data with FIR LPF impulse resp Ftemp=Ftemp. *Hb; %inverse fft for filtered result temp=ifft(Ftemp); xpeak=212-n/2; %add channel 182 peak val to vector, subtract background first: peak=[peak (temp(xpeak)-0.5 *(temp(xpeak-82)+temp(xpeak+68)))]; data=[data temp]; end; snr=[snr (mean(peak')/std(peak'))]; xsnr=[xsnr fcutofFJ; end; plot(xsnr,snr,'o'); axis auto; title(sprintf('n= %4d',n)); xlabel('Centre of trans, band (normalized to Fs=l)'); ylabel('Signal to Noise Ratio'); grid; IIR_A: %f file to estimate PSD of Raman signal from a series of spectra format compact; %prevents extra line feeds clear;clf; %clears variables and figure\ name=inputCWhat are first 5 letters of file name? ','s'); %Fs=input('What is sampling frequency (Fs)? '); Fs=2*pi; n=9; data=[]; [b,a]=ellip(n,2,40,0.2); for i=2:8; %fname command adds rest of ending to file using integer division fhame=[name,int2str(i-fix(i/10)*10),'.prn']; pid_a=fopen(fhame,,r'); templ=fscanf(pid_a,,%g %g',[2 inf]); templ=(templ(2,200:800))'; fclose(pid_a); xi=zeros( 100,1 )+mean(temp 1(1:10)); yi=xi; zi=filtic(b,a,yi,xi); temp=filter(b,a,temp 1 ,zi); data=[data temp]; end; data=mean(data'); plot(20*log 10(abs(freqz(b,a, 1024)))); plot(abs(data)); axis([0,1024,0,max(data)]); title(sprintf('n= %4d',n)); grid;