Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

In situ analysis of lateral triterpenoid distribution in plant cuticular waxes using Raman microspectroscopy… Yu, Marcia Mei Lin 2008

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2008_spring_yu_marcia_mei_lin.pdf [ 11.73MB ]
Metadata
JSON: 24-1.0061758.json
JSON-LD: 24-1.0061758-ld.json
RDF/XML (Pretty): 24-1.0061758-rdf.xml
RDF/JSON: 24-1.0061758-rdf.json
Turtle: 24-1.0061758-turtle.txt
N-Triples: 24-1.0061758-rdf-ntriples.txt
Original Record: 24-1.0061758-source.json
Full Text
24-1.0061758-fulltext.txt
Citation
24-1.0061758.ris

Full Text

IN SITU ANALYSIS OF LATERAL TRITERPENOID DISTRIBUTION IN PLANT CUTICULAR WAXES USING RAMAN MICROSPECTROSCOPY AND NON-LINEAR OPTICAL IMAGING by MARCIA MEI LIN YU B.Sc. (Combined Honours), The University of British Columbia, 2002 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Chemistry) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) April 2008 © Marcia Mei Lin Yu, 2008 ABSTRACT The above-ground organs of plants are covered by a protective cuticle, an extracellular membrane performing important physiological and ecological functions, that consists of cuticular wax and the fatty acid-derived polymer cutin. Until the past decade, the cuticular wax was thought to be a homogenous mixture. Hence, previous interpretations relating the chemical composition and biological functions of the cuticular wax were based on the total wax composition, an average taken over the entire area and depth of the cuticle. However, recent selective sampling experiments showed a heterogeneity of the chemical composition between different wax layers. The finding of this heterogeneity imposes the need for a more accurate description of the cuticle in order to understand how the chemical composition determines the biological function. This thesis is aimed at mapping the lateral patterns of cuticular waxes on Prunus laurocerasus leaf surfaces with microscopic resolution to provide spatially resolved chemical information in the interest of describing the cuticle more accurately. Firstly, this thesis examines the post-acquisitional data processing and analysis techniques followed by the investigation of the potential of Raman microspectroscopy for the simultaneous detection of structurally similar triterpenoids in plant cuticles. Relative composition analysis was performed on artificial triterpenoid mixtures and the resulting calculated triterpenoid ratios were consistent with the expected values. Qualitative and quantitative analysis of the linear near infrared (NIR) Raman, coherent anti-Stokes Raman scattering (CARS), and third harmonic generation (THG) spectra of isolated adaxial and abaxial P. laurocerasus cuticles demonstrated the in situ detectability of the triterpenoids using this approach. Raman maps of the adaxial cuticle showed that the triterpenoids accumulate to relatively high concentrations over the periclinal regions of the pavement cells, while the very long chain aliphatic wax constituents are distributed fairly evenly across the entire adaxial cuticle. In the analysis of the abaxial cuticles, the triterpenoids were found to accumulate in greater amounts over the guard cells relative to the pavement cells. The very long chain aliphatic compounds accumulated in the cuticle above the anticlinal cell walls of the pavement cells, and were found at low concentration above the periclinals and the guard cells. ii The main research contributions include evaluating various data processing techniques as candidates for automated implementation and applying different imaging techniques to obtain chemical information about the lateral concentration gradient of the triterpenoid components in the cuticles, with high spatial resolution. This thesis also provides the first direct (i.e. in situ) evidence for lateral spatial heterogeneity of triterpenoids in the cuticle of a model species, P. laurocerasus. This work is expected to impel further structure-function investigation of the cuticular membranes of plants. iii TABLE OF CONTENTS Abstract^ ii Table of Contents^ iv List of Tables vii List of Figures viii List of Abbreviations^ xi Acknowledgements xiii Dedication^ xiv Co-Authorship Statement^ xv Chapter 1. Introduction 1 1.1 Plant Anatomy^ 1 1.2 Leaf Cuticles 5 1.3 Techniques for Cuticular Analysis^ 8 1.3.1 Techniques Providing Spatial Information Only^ 9 1.3.2 Techniques Providing Structural Information Only 10 1.3.3 Techniques Providing Chemical Information Only 11 1.3.4 Techniques Providing Spatial, Structural, and Chemical Information ^ 13 1.4 Raman Spectroscopy^ 16 1.4.1 Classical Theory of Raman Scattering^ 17 1.4.2 Quantum Mechanical Description of Raman Scattering^ 19 1.4.3 Raman Microspectroscopy^ 21 1.5 Non-Linear Spectroscopy^ 25 1.5.1 Coherent Anti-Stokes Raman Scattering (CARS) Spectroscopy^26 1.5.2 Third Harmonic Generation (THG) Spectroscopy^ 28 1.6 Thesis Objectives and Overview^ 30 1.7 References^ 32 Chapter 2. Instrumentation 41 2.1 Raman Microspectroscopy^ 41 2.1.1 Lasers^ 41 2.1.2 Renishaw Raman Microscope^ 42 2.1.3 Confocal Modality^ 44 2.1.4 Software^ 46 2.2 CARS and THG Microspectroscopy^ 47 2.2.1 Laser Set-up 47 2.2.2 Software^ 49 2.3 References 50 Chapter 3. Post-Acquisitional Spectral Processing^ 51 3.1 Introduction^ 51 3.2 Theory 53 3.3 Materials and Methods^ 55 3.3.1 Synthetic Spectra 55 3.3.2 Computation Equipment^ 56 3.3.3 Figures of Merit^ 58 iv 3.4 Selected Baseline Correction Techniques^ 59 3.4.1 Methods Requiring No Explicit Knowledge of p, b, or n: Noise Median Method^ 60 3.4.1.1 Background^ 60 3.4.1.2 Theory 60 3.4.1.3 Implementation 61 3.4.1.4 Results and Discussion^ 61 3.4.1.5 Modes of Failure 63 3.4.1.6 Suitability of Method for Automation^ 63 3.4.2 Methods Requiring Explicit Estimates of b: Signal Removal Method^63 3.4.2.1 Background^ 63 3.4.2.2 Theory 64 3.4.2.3 Implementation 65 3.4.2.4 Results and Discussion^ 65 3.4.2.5 Modes of Failure 67 3.4.2.6 Suitability of Method for Automation^ 67 3.4.3 Methods Requiring Explicit Use of b and n: Manual Method^67 3.4.3.1 Background^ 67 3.4.3.2 Theory 68 3.4.3.3 Implementation 68 3.4.3.4 Results and Discussion^ 69 3.4.3.5 Modes of Failure 71 3.4.3.6 Suitability of Method for Automation^ 71 3.4.4 Methods Requiring Explicit Use of p, b, and n: Maximum Entropy Method^ 71 3.4.4.1 Background^ 71 3.4.4.2 Theory 72 3.4.4.3 Implementation 73 3.4.4.4 Results and Discussion^ 73 3.4.4.5 Modes of Failure 75 3.4.4.6 Suitability of Method for Automation^ 75 3.4.5 Methods Requiring Frequency Information: Wavelet Transform Method^ 75 3.4.5.1 Background^ 75 3.4.5.2 Theory 76 3.4.5.3 Implementation 78 3.4.5.4 Results and Discussion^ 78 3.4.5.5 Modes of Failure 80 3.4.5.6 Suitability of Method for Automation^ 80 3.5 Conclusion^ 82 3.6 References 83 Chapter 4. Raman Microspectroscopic Analysis of Plant Triterpenoids^89 4.1 Introduction^ 89 4.2 Materials and Methods^ 90 4.2.1 Reference Samples 90 4.2.2 Artificial Triterpenoid Mixture Preparation^ 91 4.2.3 Raman Microspectroscopy^ 91 4.2.4 Raman Spectra Processing and Data Analysis^ 91 4.3 Results and Discussion^ 93 4.3.1 Laser Selection 93 4.3.2 Reference Samples 96 4.3.3 Artificial Triterpenoid Mixtures^ 101 4.4 Conclusion^ 105 4.5 References 106 Chapter 5. In situ Analysis of Triterpenoid Compositional Patterns within Prunus laurocerasus Leaf Cuticles^ 109 5.1 Introduction^ 109 5.2 Materials and Methods 111 5.2.1 Reference Compounds^ 111 5.2.2 P. laurocerasus Cuticular Membranes^ 112 5.2.3 Epicuticular Wax Preparation 112 5.2.4 Linear Raman Microspectroscopy 113 5.2.5 Non-linear Spectroscopy^ 113 5.2.6 GC-MS Analysis 114 5.2.7 Scanning Electron Microscopy 114 5.2.8 Histological Cross-Sections^ 115 5.2.9 Ultraviolet-Visible Absorption Spectroscopy^ 115 5.3 Results and Discussion^ 116 5.3.1 Adaxial Leaf Surfaces^ 116 5.3.2 Abaxial Leaf Surfaces 132 5.3.3 Ultraviolet-Visible Absorption Analysis of Adaxial and Abaxial Leaf Surfaces^ 135 5.4 Conclusion^ 137 5.5 References 138 Chapter 6. Conclusion and Future Directions^ 141 6.1 Concluding Remarks^ 141 6.2 Future Directions 143 6.3 Final Remark 145 6.4 References^ 146 vi LIST OF TABLES Table 1.1^ 15 A comparison between the strengths and limitations of the analytical techniques used for cuticular analysis. Table 1.2^ 22 A comparison of the advantages and disadvantages of the three types of imaging modes for Raman microscopy. Table 3.1^ 59 Classification of baseline removal methods based on the type of required information. Table 4.1^ 99 Raman shifts and peak assignment of vibrational modes of triacontanoic acid, a-amyrin and oleanolic acid. Table 4.2^ 104 Actual and calculated experimental relative weight ratios of a-amyrin to oleanolic acid, and oleanolic acid to ursolic acid in the various triterpenoid mixtures. Table 4.3^ 104 Actual and calculated experimental relative weight ratios of nonacosane to triterpenoids, and ursolic acid to oleanolic acid in the tertiary mixture. vii LIST OF FIGURES Figure 1.1^ 3 An illustration depicting the three main vegetative organs in a plant: the root, the stem, and the leaf Figure 1.2 ^ 4 An illustration of a dicotyledon leaf cross-section. Figure 1.3^ 7 An illustration of a cuticle cross-section. Figure 1.4^ 7 Examples of compounds found in the cuticular soluble waxes. Figure 1.5^ 20 Energy level diagram illustrating the quantum mechanical approach to the Raman effect. Figure 1.6^ 22 The different types of Raman mapping experiments. Figure 1.7^ 24 Chemical images obtained by point mapping experiments. Figure 1.8^ 3^ 27 Energy level diagram depicting the principle of CARS spectroscopy. Figure 1.9^ 29 Energy level diagram depicting the principle of THG spectroscopy. Figure 2.1^ 44 Schematic diagram of the Renishaw Ramanscope System 1000. Figure 2.2^ 45 Schematic diagram of a typical confocal microscope set-up. Figure 2.3^ 46 Confocality of the Renishaw Ramanscope System 1000. Figure 2.4^ 48 Schematic diagram of the CARS and THG instrumental set-up. Figure 3.1^ 57 Standardized data and example spectra. viii Figure 3.2^ 62 Baseline removal using noise median method. Figure 3.3^ 66 Baseline removal using signal removal (moving average) method. Figure 3.4^ 70 Baseline removal using manual method. Figure 3.5^ 74 Baseline removal using maximum entropy regularization. Figure 3.6^ 79 Baseline removal using wavelet transforms. Figure 3.7^ 81 Global comparisons for baseline estimation methods based on all spectra. Figure 4.1^ 95 A comparison of different laser excitation wavelengths for nonacosane, oleanolic acid, and ursolic acid. Figure 4.2^ 97 The molecular structure of the four triterpenoids used in this mixture study. Figure 4.3^ 98 Raman spectra of the reference triterpenoids and very long chain aliphatic compounds in the 350-2000 cm -1 spectral region. Figure 4.4^ 103 Raman spectra of artificial triterpenoid mixtures with 1:2, 1:1, and 2:1 relative weight concentrations of a-amyrin:oleanolic acid in the 350-900 cm -1 spectral region. Figure 5.1^ 117 Gas chromatogram of the soluble waxes from isolated adaxial cuticles of P. laurocerasus leaves. Figure 5.2^ 119 A series of NIR Raman spectra acquired at the same position as a function of acquisition order. Figure 5.3^ 121 In situ NIR Raman spectra of isolated cuticular membrane from adaxial surfaces of P. laurocerasus leaves with and without wax, and their difference spectrum. Figure 5.4^ 123 Chemical images of isolated adaxial P. laurocerasus cuticular membrane using NIR Raman microspectroscopy. ix Figure 5.5^ 124 SEM micrographs of isolated adaxial P. laurocerasus cuticular membrane under 250x magnification. Figure 5.6^ 125 Bright field light microscopic image of an isolated adaxial P. laurocerasus cuticular membrane cross-section. Figure 5.7^ 128 Chemical images of the distribution of the different components in isolated adaxial P. laurocerasus cuticular membrane using NIR Raman microspectroscopy. Figure 5.8^ 131 Isolated adaxial P. laurocerasus cuticular analysis using non-linear spectroscopy. Figure 5.9^ 133 SEM micrographs of isolated abaxial P. laurocerasus cuticular membrane under 250x magnification. Figure 5.10^ 134 Isolated abaxial P. laurocerasus cuticular analysis using NIR Raman microspectroscopy. Figure 5.11^ 136 UV-Vis absorption spectrum of isolated P. laurocerasus cuticles from the adaxial and abaxial sides of the leaf. LIST OF ABBREVIATIONS 2D^two-dimensional 3D three-dimensional AFM^atomic force microscopy ATR attenuated total reflectance BCM^baseline correction method BSTFA bis-N,N-(trimethylsilyl)trifluoroacetamide CARS^coherent anti-Stokes Raman scattering CCD charge-coupled device CM^cuticular membrane DMSO dimethyl sulfoxide DNA^deoxyribonucleic acid DSC differential scanning calorimetry EI^electron impact FID flame ionization detector FOM^figures of merit FT Fourier transform FTM^Fourier transform method FWHM full-width half-maximum GC^gas chromatography HeNe helium-neon HNF^holographic notch filter HR-CP/MAS^high resolution-cross polarization/magic angle spinning HR-MAS high resolution-magic angle spinning IR •^infrared LC liquid chromatography MEM^maximum entropy method MM manual method MS^mass spectrometry xi MX^cutin matrix NA numerical aperture NIR^near infrared NMM noise median method NMR^nuclear magnetic resonance OCT optimal cutting temperature OPA^optical parametric amplifier SBR signal-to-baseline ratio SEM^scanning electron microscopy SIMS secondary ion mass spectrometry SNR^signal-to-noise ratio SRM signal removal method ssNMR^solid-state nuclear magnetic resonance TEM transmission electron microscopy THG^third harmonic generation TIR total internal reflectance Ti: Sapph^titanium: sapphire TLC thin-layer chromatography TMSi^trimethylsilyl ToF-SIMS^time-of-flight secondary ion mass spectrometry UV-Vis ultraviolet-visible w/w^weight per weight WTM wavelet transform method XPS^X-ray photoelectron spectroscopy xii ACKNOWLEDGEMENTS First and foremost, I would like to acknowledge my supervisors Dr. Michael Blades and Dr. Robin Turner, as well as Dr. Reinhard Jetter for inspiring me to pursue this project. My graduate school experience has been a memorable one and I am truly grateful for their guidance and their devotion of time and energy towards my education. I would like to thank Dr. Stanislav Konorov, Dr. Georg Schulze, and Dr. Andrew Jirasek for sharing their knowledge and expertise, all the while being patient with my questions. To the Blades and Turner group members, talking about chemistry was fun, but you have to admit, the non-science "artsy" conversations were even better! A special thank you to Dr. Dan Bizzotto, Mr. Ben Clifford, Dr. Bruce Todd, Dr. Guillaume Bussiere, and Mr. Christopher Buschhaus for allowing me to use their instruments and chemical solutions. I am grateful to Mr. Derrick Horne from the UBC Biological Imaging Laboratory for acquiring scanning electron microscopy images of my samples, as well as to the staff of Wax-It Histological Services Inc., for obtaining cross-sectional slices of my cuticle sample. Lastly, I would like to extend a big thank you to my parents and my family for their endless love and support. Words cannot express my sincere gratitude for their continuous encouragement, patience, and understanding. This thesis is dedicated to my parents and my grandparents xiv CO-AUTHORSHIP STATEMENT A version of the following chapters has been previously published. Chapter 3: Jirasek A, Schulze G, Yu MML, Blades MW, and Turner RFB. Accuracy and Precision of Manual Baseline Determination. Applied Spectroscopy, 2004, 58(12), 1488-1499. Schulze G, Jirasek A, Yu MML, Lim A, Turner RFB, and Blades MW. Investigation of Selected Baseline Removal Techniques as Candidates for Automated Implementation. Applied Spectroscopy, 2005, 59(5), 545-574. For the above 2 papers, my contributions mainly involved performing and assisting with the research, which included gathering volunteers to perform the manual baseline correction of various spectra and classifying the subjects' level of experience in baseline determination. Chapter 4: Yu MML, Schulze HG, Jetter R, Blades MW, and Turner RFB. Raman Microspectroscopic Analysis of Triterpenoids Found in Plant Cuticles. Applied Spectroscopy, 2007, 61(1), 32-37. For the above paper, I performed the research, carried out the data processing and analysis, and prepared the manuscript. Chapter 5: Yu MML, Konorov SO, Schulze HG, Blades MW, Turner RFB, and Jetter R. In situ analysis by microspectroscopy reveals triterpenoid compositional patterns within leaf cuticles of Prunus laurocerasus. Planta, 2008, 227(4), 823-834. For the above paper, I performed the linear Raman microspectroscopy portion of the research, carried out all the data processing and analysis, and prepared the manuscript. xv CHAPTER 1 INTRODUCTION "Nature will bear the closest inspection. She invites us to lay our eye level with her smallest leaf, and take an insect view of its plain." — Henry David Thoreau Plants are a vital part of the biosphere mainly due to their ability to convert water and carbon dioxide into oxygen and carbohydrates via photosynthesis. Understanding the structural details of the various parts of plants is important because this knowledge affects how certain structure-function relationships are viewed and interpreted. Often, microscopic analysis is required to explain the biological, chemical, and mechanical behaviours of plants and plant materials. This work will describe the application of Raman microspectroscopy and non-linear optical imaging techniques for the study of the chemical distribution on the surfaces of leaves. More specifically, an emphasis will be placed on the determination of the lateral distribution and localization of triterpenoids on the different regions of the leaf cuticle. This chapter will introduce the general anatomy and composition of plant cuticles, and it will also introduce the reader to the fundamentals of the techniques involved, mainly Raman spectroscopy and non- linear spectroscopy. 1.1 Plant Anatomy There are three main vegetative (non-reproductive) organs in all plants — the root, the stem, and the leaf (Figure 1.1) [1] — and while each organ will be briefly explained, the latter will be described in detail, as this research chiefly pertains to the chemical composition of leaf surfaces. The majority of vascular plants produce roots in soil, with the exception of those that develop roots under water or above ground. The key function of the root is to anchor the plant and to absorb water from the surrounding soil. Other functions include salt intake [2, 3] and assimilate storage [4, 5]. The organ attached to the root is the stem, which functions as a supporting structure and is the location where the xylems and phloems are differentiated [6]. 1 The function of the xylems and phloems is to conduct water and nutrients from the root to the rest of the plant, and to conduct assimilates from source to sink organs, respectively [7]; collectively, the xylems and phloems are known as vascular tissues or vascular bundles. Plants can essentially be divided into two taxonomic groups depending on the number of cotyledons, which are leaves attached to the seedling stem. Plants with one cotyledon are called monocotyledons, which include the grass and the orchid and lily families; those with two cotyledons are called dicotyledons, and these include roses, cherry blossoms, and most fruit trees. The leaves of both monocotyledon and dicotyledon plants share similar internal tissues, including the mesophyll cells and the vascular bundles. Functionally, the mesophyll cells contain chlorophyll and their arrangement in the leaf maximizes photosynthesis and facilitates gas movement within the cell [8]. While the monocotyledons only contain a uniform mesophyll distribution, the dicotyledons have two types of mesophyll cells, the closely packed upright cylindrically-shaped palisade mesophyll cells and the loosely packed smaller spongy mesophyll cells, located on the adaxial (top) and the abaxial (bottom) halves of the leaf, respectively (Figure 1.2). These cells and vascular bundles are confined to the insides of the leaf by the surrounding upper and lower epidermis. 2 leaf Figure 1.1 An illustration depicting the three main vegetative organs in a plant: the root, the stem, and the leaf. Reproduced from http://www.kew.org/ksheets/pdfs/b3plant.pdf (accessed on October 2007) with the permission of the Trustees of the Royal Botanic Gardens, Kew. 3 Bundle sheath cell Xylem PhloemVein. Periclinal cell wall ^Anticlinal cell wall 7— Cuticle Upper epidermis Palisade mesophyll cell Lower epidermis Spongy mesophyll cells Guard cells Stoma Cuticle Figure 1.2 An illustration of a dicotyledon leaf cross-section. The functions of the different cells are described in the text. Reprinted from Purves et al. [9] with the permission of Sinauer Associates, Inc. There are two main types of cells, the pavement cells and the guard cells, comprising the continuous, mono-cellular layer known as the epidermis. An additional cell type known as trichomes may be present on leaf surfaces, but it will not be discussed here as trichomes are not present in the model plant species used in this work. As with most plant cells, the epidermal cells have an outer rigid cell wall which functions to provide strength and support for the plants. The periclinal cell walls of the epidermal cells are those parallel to the surface of the leaf, while the anticlinal cell walls are perpendicular to the surface, or more specifically, where the cell contacts its neighboring cell (Figure 1.2). The pavement cells comprise most of the epidermis and, although the main function of this external layer of cells is to cover the internal leaf 4 components, it also acts as the site of cuticular wax biosynthesis. The guard cells, on the other hand, are located randomly along the surface of the leaf and they function to form, and to open and close the stoma (pl. stomata) by turgor pressure. The stomata are openings in the leaf that allow exchange of CO2 and 02 gases in and out of the leaf. In addition to allowing gas exchange, guard cells also regulate the release of water vapour. These processes are controlled by water pressure; when the guard cells are fully turgid, the stomata are open and the stomata close when the turgidity is decreased. 1.2 Leaf Cuticles The epidermal cells are protected from the environment by a thin, continuous, extra- cellular membrane called the cuticle, that appears early on during leaf development [10-16]. The cuticle is made up of waxes soluble in organic solvents and an insoluble polymeric cutin matrix (Figure 1.3). The soluble waxes can be found as a layer above (epicuticular wax) [17] or embedded within (intracuticular wax) the cutin [18]. Furthermore, some plants also contain additional wax crystals protruding from the epicuticular wax layer [19]. These cuticular waxes are synthesized in the epidermal pavement cell and are eventually transported to the plasma membrane and exported from the cell [20]. The transport process of the lipophilic wax molecules across the cell wall and into the cuticle has not yet been determined, but is hypothesized to be facilitated by specific lipid transfer proteins [21, 22] or non-specific lipid transfer proteins [23-25], the latter of which have been found in the epidermal cell wall and in the cuticle [26, 27]. The soluble waxes are typically mixtures made up of very long chain aliphatic compounds (Figure 1.4), including n-alkanes, primary n-alcohols, n-aldehydes, and fatty acids. The chain lengths of these compounds typically vary from twenty to forty carbons [28]. Some plant cuticular waxes also contain secondary alcohols, alkyl esters, and/or triterpenoid constituents. The cuticular triterpenoids are generally pentacyclic with a hydroxyl functional group in the 313 position [28]. Variations in the structure of the triterpenoids occur at, for example, the cyclic groups which are made up of either five-carbon or six-carbon rings (Figure 1.4). Other variations include different functional groups, as well as the presence of double 5 bonds. The concentration of the triterpenoids relative to the very long chain aliphatic compounds vary: some species have low triterpenoid concentrations, for example Prunus laurocerasus (Laurel cherry) leaf cuticles contain 5 lig/cm2 of triterpenoids and 38 ixg/cm2 of very long chain aliphatic compounds [29]; while others have high triterpenoid concentrations, for example Ligustrum vulgare (European privet) leaf cuticles contain 18 i.tg/cm 2 of triterpenoids and 6 jug/cm2 of very long chain aliphatic compounds [30]. However, it has been observed that the triterpenoids are exclusively located in the intracuticular wax, while the very long chain aliphatic compounds can be found in both the epi- and intracuticular wax layers [29-32]. While the cuticular waxes are mixtures of the homologous series mentioned above, the cutin matrix is an extensive biopolymer comprised of w-hydroxylated fatty acid monomers, predominantly dihydroxy C16 and trihydroxy C18 acids, cross-linked by ester bonds [33, 34]. It is the presence of the mid-chain hydroxyl groups on the cutin monomers that allows for branching to occur [33-37]. Moreover, a non-saponifiable and non-extractable biopolymer known as cutan may also be present in the insoluble matrix [38]. Bearing in mind that there are numerous plant species, it is not surprising that the cuticular compositions, both in the soluble waxes and the insoluble biopolymers, are species dependent. For example, some plant cuticles only have cutin or only cutan, while other species contain both components [39]. Cuticular dynamics and heterogeneity will be discussed in detail in Chapter 5. 6 Epicuticular wax crystals Epicuticular wax ..-----^film^------a. Intracuticular wax + Cutin Cell wall OH OH 0 OH OH OH Figure 1.3 An illustration of a cuticle cross-section. The cuticle is made up of epi- and intracuticular waxes and the cutin matrix. Some species may also have epicuticular wax crystals of various shapes and sizes. The figure on the left is reprinted from Jetter et al. [29] by permission of Blackwell Publishing, and the figure on the right is reprinted from Jeffree [40] by permission of Edward Arnold (Publishers) Ltd. Figure 1.4 Examples of compounds found in the cuticular soluble waxes. Very long chain aliphatic compounds: (a) nonacosane, (b) hentriacontane, (c), octacosanol, (d) triacontanol, (e) octacosanoic acid, and (f) triacontanoic acid. Triterpenoids: (g) lupeol, (h) lupenone, and (i) erythrodiol. 7 As the cuticle is the interface between the environment and the underlying tissues, it provides protection in many ways. For example, the cuticle prevents transpirational water loss [41] and also acts as a water repellent [42] and may repel other small particles like aerosols and dust [43-46]. Other cuticular functions include protecting the plants from fungal invasion [47], pathogens [48-51], and herbivores [52], as well as from ultraviolet radiation which can cause damage to the photosynthetic cells and their DNA [53-56]. It has also been suggested that the cutin may be important in thermoregulation in leaves and fruits [57-59]. Prior to the last decade, quantitative analyses for determining the concentrations of the different wax components were done on the total cuticular wax following the organic solvent extraction of wax from both the epi- and intracuticular layers. Hence, attempts to understand the relationship between chemical composition and biological functions were made based on the assumption that the cuticular waxes were homogenous throughout the cuticle. However, it was recently reported that the cuticles in fact exhibit heterogeneity between the chemical composition of the epi- and intracuticular wax layers [29-31, 60]. This heterogeneity of the cuticle emphasizes the need to explore in some detail the spatial distribution of chemical compounds in this extracellular membrane in order to establish a more accurate relationship between the chemical composition and the biological functions of the cuticular waxes. The following section will introduce the different techniques that have been used to study plant cuticles and their contribution to our current knowledge of the cuticle and its chemical components. 1.3 Techniques for Cuticular Analysis In the past century, there has been a steady increase in the number of plant cuticular research studies. Initial experiments consisted of light microscopy and gravimetric observations, while recent instrumental developments paved the way for atomic and molecular analyses. The selected analytical methods reported here have contributed significantly to our current knowledge of the structure of the plant cuticle. Table 1.1, located at the end of this section, lists the main strengths and weaknesses of these techniques. 8 1.3.1 Techniques Providing Spatial Information Only The first few studies of plant cuticles were done by fixing a transverse slice of a leaf and observing their anatomical features under a light microscope [61-64]. However, due to technical difficulties, only plants with thick cuticles were investigated [65-68]. With improvements in fixation techniques and staining solutions, thin cuticles from developing to mature leaves could be studied [69]. These histological and histochemical investigations showed the surface structure of the cuticle having a smooth or rough layer of wax on the environmental side of the cuticle [11, 69-71]. Furthermore, the cuticle cross-sections showed a variation in the cuticle thickness, where it was the thinnest over the periclinal wall of the epidermal cells, and thickest over the regions where the anticlinal cell walls meet [69, 70]. The localization of cuticular waxes was observed by staining the fixed samples using Sudan IV, as well as by looking at their birefringence when viewed under polarized light. The soluble waxes were found to have negative birefringence [11, 69-71]; after immersing the isolated cuticle in chloroform, no birefringence was observed from the isotropic cutin matrix [69, 72]. This negative birefringence revealed that the cuticular wax molecules have a specific orientation and it has been hypothesized that their long axes are oriented normal to the cuticular membrane [65]. These birefringence observations were used to explain the degree of chemical permeability in Pyrus communis (Bartlett pear) leaves [69]; however, the actual significance of the specific orientation of these molecules in terms of structure and function is not yet known. As mentioned above, light microscopy proved to be beneficial for visualizing the surfaces and the transversal cross-section of cuticles. However, due to its low-resolution images (albeit their relative ease of use), high-resolution microscopy like atomic force microscopy (AFM) and scanning electron microscopy (SEM) have currently become the instruments of choice. Another high resolution technique known as transmission electron microscopy (TEM) has also been applied to study cuticular membranes, but it initially provided little information concerning the wax distribution in the cuticle [11] and sometimes gave contradictory results to those obtained by other techniques [73]. Recently, TEM has been used to determine the thickness of Arabidopsis thaliana cuticles [74], as well as to study the development of the cuticular ultrastructure during the different developmental stages of various plants [75]. Others have shown this technique to be complementary to SEM [76, 77], the latter being the most commonly used technique for 9 determining the surface morphology of leaves and isolated cuticles. The various shapes of epicuticular crystals have been observed on electron micrographs [78-80], which allowed for the classification of the different kinds of crystals to be made [19]. For cuticles lacking the epicuticular crystals, the environmental side of the cuticle appeared smooth [70], while SEM micrographs of the physiological (i.e. internal) side of the isolated cuticles exhibited imprints of the underlying cells [81, 82]. These morphological features have also been observed using AFM. The images produced by AFM scans offer additional information on the topography of the cuticles as they also provide the dimensions of the abovementioned contours giving three- dimensional (3D) information [83-85]. Moreover, Perkins et al. reported that the material properties of the leaf surface like hydrophobicity could be identified by AFM with the use of a hydrophobic probe [86] thus providing partial chemical information; however, the majority of AFM experiments have been performed for surface morphology studies. In situ studies on the regeneration of the epicuticular waxes by AFM have also been performed [87]. Recently, fluorescence spectroscopy has been used as a simple and time-saving tool to efficiently measure the number and dimensions of stomata found on plant surfaces, as a greater fluorescence emission was observed from the guard cells in comparison to the neighbouring cells [88]. 1.3.2 Techniques Providing Structural Information Only Cuticular analyses have been carried out using nuclear magnetic resonance (NMR) spectroscopy, which can provide structural information on the types of links involved in the cutin matrix. Typically, solid-state NMR (ssNMR) spectra show very broad peaks resulting from anisotropic interactions in the solid sample, as opposed to sharp signals observed in solution NMR resulting from an average of anisotropic interactions. However, by using NMR techniques like high resolution — magic angle spinning (HR-MAS) NMR or high resolution — cross polarization/magic angle spinning (HR-CP/MAS) NMR, major carbon functionalities have been identified in intact biopolymers [89]. Several studies have been done using HR-MAS on Solanum lycopersicum (Tomato) cutin matrices and all data confirm that the S. lycopersicum cutin is a polyester, with traces of primary and secondary alcohols and free fatty acids, as well as 10 olefinic and aromatic carbons [35, 81, 84, 90]. Furthermore, the rigidity and mobility of individual carbon segments in the cuticle, in this case both the cutin and the cuticular wax, were determined using NMR giving further insight to the surface elasticity of plants [35, 81, 91, 92]. 1.3.3 Techniques Providing Chemical Information Only Although the first two techniques listed below do not provide information about the individual chemical components found in the cuticle, they are nevertheless introduced in this section as they provide information about the overall chemical composition of cuticles. Initially, the amount of cuticular wax present in the plant cuticle was determined gravimetrically by extracting the soluble waxes with chloroform [11, 69]. Norris and Bukovac measured the wax content of intact P. communis leaves as well as that of isolated cuticles and reported that the quantities of the surface waxes were comparable, thus implying no significant loss of material during the isolation process [69]. Furthermore, Schmidt and Schônherr quantified the amount of cutin comprising the cuticle of Clivia miniata (Bush lily) leaves and found that the thickness and mass of both the cutin and soluble waxes increased with leaf development [11]. Another method that can provide chemical information about the cuticular wax is differential scanning calorimetry (DSC) [57]. This technique can be used to study the thermal properties of the cuticle by revealing information on the phase behaviour of the cuticular waxes as a function of temperature [57]. DSC thermograms of cuticular waxes in plants and fruits show broad, occasionally overlapping, peaks indicating the melting of several components [91, 93, 94]. There exist a few cases where sharp peaks are observed in the thermograms, indicating the presence of crystalline solids [91, 94]. The chemical composition of the cutin and the soluble waxes can be determined using mass spectrometry (MS). Electron impact (EI) ionization is a type of hard ionization technique that ionizes the molecule for MS analysis with extensive fragmentation of the "parent" molecular ion. This proved to be important for the determination of monomeric constituents of cutin [95]. 11 This was done by hydrolyzing the cutin with porcine pancreatic lipase, an enzyme specific for the hydrolysis of esters of primary alcohols [96], resulting in a mixture of monomeric and oligomeric components [97, 98]. Using this method, the principal monomeric fatty acid constituents of citrus fruit cutins were identified, in addition to providing corroborative evidence with the NMR data about their cross-linking structures [95, 99]. When coupled to an efficient separation technique, the application of MS is further enhanced. Liquid chromatography (LC) MS has been used to characterize S. lycopersicum cutin [100]. When MS is coupled with gas chromatography (GC), it is possible to obtain additional information when working with mixtures, and it thus proved to be the most frequently used method for identifying the various components present in the soluble cuticular waxes [29, 101- 103]. Once the wax compounds have been identified, their concentrations can be determined using GC with a flame ionization detector (FID) [29, 101]. These methods have shown that the exact chemical compositions and concentrations of the waxes vary significantly for various plant species, organs and developmental stages [29, 101, 104-107]. Another chromatographic technique used to separate, isolate and identify the components of the cuticular soluble waxes is thin-layer chromatography (TLC) [108, 109]. Most recently, Gawronska-Grzywacz and Krzaczek used this technique in conjunction with GC-MS to identify all the triterpenoid components in Hieracium pilosella (Mouse-ear hawkweed) [110]. A quantitative, in situ surface analysis technique is X-ray photoelectron spectroscopy (XPS), a tool that measures the elemental composition from the uppermost ca. 10 nm of a surface as a result of the focused X-ray beam. The main elements detected on leaf surfaces by XPS in decreasing atomic concentrations were C, 0, and N, with most of the carbons being involved in C-C or C-H bonds [86, 111]. This is consistent with the fact that the major components in the leaf cuticles are fatty acid derivatives. The elemental concentrations can be determined by calculating the peak areas in the XPS spectra as they are proportional to the mole fractions of the element. Using this method, Barr et al. found that the surface of a leaf is constantly changing even after the leaf has been removed from the plant [112]. 12 Techniques similar in principle to XPS are secondary ion mass spectrometry (SIMS) and time-of-flight secondary ion mass spectrometry (ToF-SIMS), which has just recently been employed for the elemental analysis of P. laurocerasus cuticles [86]. It should be noted that although XPS and SIMS have mainly been used for obtaining chemical information about the cuticle, spatially resolved information could conceivably be obtained by raster scanning the sample. For example, dynamic SIMS has recently been applied to the study of plants (though not cuticles, specifically) and the capability of obtaining spatial information using SIMS was demonstrated by imaging the compartmentalization and quantification of inorganic metals and ions in the different plant tissues and cells [113, 114]. 1.3.4 Techniques Providing Spatial, Structural, and Chemical Information The last two analytical techniques to be presented here, infrared (IR) and Raman spectroscopy, are based on the molecular vibrations. Although the origin of their spectra arise from two mechanistically different processes, both techniques can be used to "characterize" the different functional chemical groups in situ of the various components found in plant and fruit cuticles [115, 116]. In a destructive mode of IR spectroscopy, where the sample preparation involved mixing the cuticle into KBr pellets, the degree of cross-linking in the cutin matrix of developing leaves and fruits was determined by measuring the ratio of the peak belonging to the methylene stretching vibrations to that of the carbonyl C=0 stretching vibration of the ester bond [84]. Fourier transform (FT) IR spectroscopy has been used to 'measure the crystallinity and phase transitions of reconstituted wax, isolated cuticular membranes, and wax on intact leaves, with comparable results for all three types of samples [117]. Spectral variations in adaxial and abaxial surfaces of leaves due to age and amount of sun exposure have been studied using attenuated total reflectance (ATR) spectroscopy measured with FT-IR spectrometers [118]. Although Raman spectroscopy has been used to study plants in general, it has not widely been used for the specific examination of plant and fruit cuticles. Nevertheless, there have been Raman studies of tree bark [119-123] and the secondary cell wall of the black spruce tree [124, 125] revealing spatially resolved information, but they will not be discussed further here as these woody tissues are not the focus of this research. One of the few Raman measurements of plant 13 cuticles reported involves the in vivo studies of barley leaf epicuticular waxes by total internal reflection (TIR) Raman spectroscopy, which provided information about the composition of the waxes [126]. Despite the lack of Raman spectroscopic reports of plant cuticles, the current study will demonstrate its powerful capabilities for the characterization of plant cuticular waxes. 14 Table 1.1 A comparison between the strengths and limitations of the analytical techniques used for cuticular analysis. Technique/Concept Strengths Limitations Atomic force microscopy (AFM) provides information on sample morphology high spatial resolution in situ analysis possible no chemical information Attenuated total reflectance (ATR) spectroscopy provides chemical information of samples in natural state good reproducibility data inaccuracy if the sample does not have good contact with the ATR crystal most ATR crystals have pH limitations Birefringence provides information of chemicalorientation in sample qualitative data only no chemical specificity Differential scanning calorimetry (DSC) provides thermal property and phase transition information destructive Dynamic secondary ion mass spectrometry (SIMS) provides chemical composition information destructive Electron impact mass spectrometry (EI-MS) provides structural information destructive Gas chromatography (GC) provides quantitative information of chemical composition destructive Infrared (IR) spectroscopy provides structural information in situ analysis possible interfering water band Light microscopy ease of use can see visible differences in sample with and without staining qualitative data only Nuclear magnetic resonance (NMR) spectroscopy provides structural information destructive Raman spectroscopy provides structural information in situ analysis possible various wavelengths possible laser source may cause heating and/or photodecomposition of sample fluorescence background 15 (Table 1.1 Continued) Technique/Concept Strengths Limitations Scanning electron microscopy (SEM) provides information on sample morphology very high spatial resolution in situ analysis possible no chemical information Thin-layer chromatography (TLC) achieves separation and isolation of different components large samples required and substance loss during separation not easily checkable destructive Transmission electron microscopy (TEM) provides information on sample morphology high spatial resolution in situ analysis possible no chemical information X-ray photoelectron spectroscopy (XPS) provides elemental composition information can get concentration information in situ analysis possible does not provide molecular information low spatial resolution 1.4 Raman Spectroscopy In 1928, Sir Chandrasekhara Venkata Raman, along with Sir Kariamanikkam Srinivasa Krishnan, focused a filtered beam of sunlight onto a large volume of neat liquid, and with his eye as the detector, discovered the phenomenon of inelastic scattering of light, a phenomenon that is called Raman scattering in honour of its discoverer [127]. Raman observed that when a sample is illuminated with a beam of monochromatic light, not only did the resulting scattered light contain the same frequency as that of the incident light (Rayleigh scattering), but that it also contained light, albeit at a low percentage, of different frequencies (Raman scattering). The shift in frequency is the result of the incoming light interacting with the electric dipole of the 16 molecules. Although the Raman spectra mainly originate from the intrinsic vibrational and rotational transitions, only the vibrational transitions will be considered in this report due to the nature of the current sample (i.e. solid) and instrumentation. 1.4.1 Classical Theory of Raman Scattering In the classical view of Raman scattering, a time-dependent electric field in the incident light, most commonly a laser beam, induces a dipole moment in the oscillating molecule, which in turn radiates electromagnetic radiation. The electric field of the incident radiation, E(t), can be described by the equation, E(t) = E0 cos 27ry ot^ (1.1) where E0 is the amplitude and vo is the frequency of the incident radiation. The induced dipole moment of molecule, ,u(t), can be described by ,u(t) = aE(t) = a E cos Dry ot^ (1.2) where a is the linear polarizability of the molecule, which is a function of the internuclear separation, r. It should be noted that ,u(t) is actually a non-linear function of E(t), and so a more accurate expression of the induced dipole, ,u(t), is 1 ,u(t) = aE(t) + — 1 )6 E(t)- 9 6 + — y E(t) 3 + • • • (1.3) where, as before, E(t) is the time-dependent electric field of the incident radiation, a is the polarizability of the molecule, and /3 and y are the first and second hyper-polarizabilities, respectively. However, since a » 13 » y , the second and third terms are negligible for conventional Raman spectroscopy. 17 The internuclear displacement of an oscillating diatomic molecule, q(t) can be written as q(t) = *go cos2ry k t^ (1.4) Here, 'go is the maximum displacement and vk is the intrinsic frequency of the oscillating molecule. The polarizability tensor can be expanded as a Taylor series in q(t), giving rise to a(q) = ac, + ( IDO( q + aq ) 0 (1.5) where ao is the polarizability at the equilibrium position of the nuclei, (aa/aq )0 is the rate of change of a with respect to the internuclear displacement evaluated at the equilibrium position, and q = r — r 0, the displacement of the nuclei from its equilibrium position, r0, as mentioned above. For vibrations with small atomic displacements, the higher terms are neglected, and thus substituting Equations (1.4) and (1.5) into (1.2) gives rise to as\ ,u(t)= a0E0 cos2rvot + — q0E0 cos2irvo t cos2rcvk t , )0 which can be written as y(t)= a0E0 cos 27Tvot +- 2 1 as  a^ q0E0 [cos{2z(vo + vk )t} + cos {2z(vo — vk )t}1 )0 by applying the trigonometric identity of AB Acos cos0 = — [cos(6 + 0) + cos(0— Ø)]• (1.6) (1.7) (1.8) 18 There are three ways in which the vibrating molecule can scatter light, as shown by Equation (1.7). The first term shows that some of the scattered light has the same frequency as the incoming radiation. This is known as Rayleigh scattering and is dominant in all scattering processes. The origin of the Raman scatter, which occurs for approximately 1 in 10 7 photons, comes from the second and third terms, which involve a change in frequency of the emitted light as a result of an energy transfer between the molecule and the radiation field. The second term is known as anti-Stokes Raman scattering and the scattered light is observed at (vo + vk), while the third term is known as Stokes Raman scattering, occurring at (v o — vk). 1.4.2 Quantum Mechanical Description of Raman Scattering Without going into detailed mathematical derivations, the quantum mechanical description of Raman scattering can be depicted schematically by Figure 1.5. By irradiating a sample with monochromatic light whose energy is less than the first electronic excited state, the molecule is excited to a non-quantized state known as a virtual state. The lifetime of Raman scattering process is extremely short (10 42 -10-13 s), and the molecule re-emits the radiation at vo (Rayleigh), vo + vk (anti-Stokes Raman), and vo — vk (Stokes Raman), where vk represents some characteristic frequency of the molecule corresponding to the energy of the first vibrational state of the ground electronic state. Since the ensemble distribution of vibrational states is characteristic of a given molecular configuration, different compounds can often be identified using Raman spectroscopy. Hence Raman spectra can be interpreted by the characteristic group frequencies which arise from a particular group of atoms having peaks at or near the same frequency; these group frequencies can be used for obtaining information about the presence of certain functional groups in the molecular structure. It will be shown in Chapter 4 that some structurally similar compounds have comparable and undistinguishable Raman spectra, while others have unique spectra that allow the compounds to be effectively distinguished. The latter is especially true for compounds that contain coupled vibrational modes, which can give rise to peaks at frequencies characteristic of the specific structural elements composing the molecule and provide distinct "fingerprints" that can be used for detection and quantitative analysis. Since spontaneous Raman spectroscopy is a linear process, the intensity of the Raman signal is proportional to the power of the incident radiation and the molecular concentration [128]. 19 Under normal conditions, the intensity of the anti-Stokes Raman scattered light is much weaker due to the Maxwell-Boltzmann distribution law which states that the population of the molecules in the ground vibrational state is much larger than that in any excited state. Consequently, the Raman spectra acquired in this report will display peaks from Stokes Raman scattering, basically giving the same information as anti-Stokes Raman scattering. v v,' V ( ^' AE V3 V, V I V() IR absorption Raman Figure 1.5 Energy level diagram illustrating the quantum mechanical approach to the Raman effect. The green arrow represents the incident radiation, while the downward-pointing red and blue arrows represent the Stokes and anti-Stokes Raman scattering, respectively. v n and y r,' represent the vibrational levels of the ground and first excited electronic states, respectively. As shown in Figure 1.5, the difference in energy between the excitation frequency and the Raman frequency corresponds to an IR absorption. The difference between the two techniques, aside from IR being an absorption process and Raman a scattering process, lies in the 20 selection rules determining IR and Raman activity. More explicitly, a normal mode is IR active if there is a change in dipole moment upon the interaction of the incoming radiation with the vibrating molecule, while a change in the polarizability of the molecule is required for the normal mode to be Raman active. In many cases, the vibrational transitions seen in Raman spectroscopy can be observed by IR spectroscopy, thus resulting in similar spectra. However, these two techniques often provide complementary information since some vibrations are Raman active only, while others are IR active only. Additionally, an important practical difference between IR spectroscopy and Raman spectroscopy is that water is a strong interferent in IR absorption measurements, but is a relatively weak Raman scatterer. Hence Raman spectroscopy is often favoured for the analyses of biological samples. 1.4.3 Raman Microspectroscopy The two main differences between Raman microspectroscopy and other conventional Raman techniques are based on the microscope's capabilities of focusing the excitation laser onto the sample using a high numerical aperture (NA) objective and the way in which the scattered light is collected. In addition, confocal microscopy further improves the spatial resolution of the system by using pinhole apertures for both the excitation and detection to isolate a small observation volume in the sample [129]. As a result, the collected signal comes selectively from a thin layer close to the focal plane while all other signals from out-of-focus regions are attenuated by this spatial filtering effect. A full mathematical treatment of the confocal effect can be found elsewhere [130, 131]. A detailed instrumental description is given in the Confocal Modality section in Chapter 2. An important advantage of Raman microspectroscopy is its capability to acquire spectra at a minute and defined area in an unstained sample. This area can be viewed as an individual "pixel", which ultimately can be arrayed in order to render a Raman "image" of the sample [132]. Raman imaging is essentially a form of chemical imaging as it is based on the (spatially resolved) intensities of a specific Raman frequency of the sample. There are three ways in which a Raman image can be obtained: point mapping, line mapping, and global imaging (Figure 1.6) [133]. Table 1.2 lists the advantages and disadvantages of each method. 21 a)  laser b) z heterogeneity in sample c)^ d) x defocused laser Figure 1.6 The different types of Raman mapping experiments. (a) A single spot Raman experiment. (b) Point mapping — the dots represent locations where the spectra are acquired. (c) Line mapping — the line represents the laser focus where the spectra are acquired. (d) Global imaging — an image is acquired where the sample is illuminated by the laser. Reproduced from McCreery [134] with permission of John Wiley & Sons, Inc. Table 1.2 A comparison of the advantages and disadvantages of the three types of imaging modes for Raman microscopy. Imaging mode Advantage Disadvantage Point mapping Full spectrum obtained at each point High spectral resolution Long integration time Line mapping Short integration time Poor spatial resolution Global imaging Laser power evenly distributed Short integration time Only one wavenumber selection per experiment Low spectral resolution 22 In point mapping experiments, a Raman spectrum is collected at each resolved position of a sample, essentially giving information in four independent variables: x position, y position, Raman shift, and intensity. A 2D or 3D Raman image is reconstructed by plotting the intensity of a particular frequency as a function of the position, where the intensity of this Raman frequency is assigned a colour that correlates with the magnitude of the scattering intensity (Figure 1.7). Experimentally, the sample is rastered under a focused laser beam and a sequential collection of spectra is acquired from each spatial position such that each point defines a "pixel" of the composite chemical image. Since a full Raman spectrum is acquired at each spatial position of the sample, any additional subsequent analysis can be done on the individual Raman spectra. Despite the higher spatial resolution and its capabilities of chemical imaging, Raman microspectroscopy has several limitations. For example, due to the small cross-section of Raman scattering, a high average power is required to obtain a high signal-to-noise (SNR) Raman spectrum at each point. Since the laser beam is focused onto a small volume of the sample, this high average power increases the risk of photodamage to the sample. A disadvantage of performing point mapping experiments for chemical imaging is the long integration time since a full Raman spectrum is acquired at each point in a specified area. Depending on the size of the image area and the size of individual pixels, it can thus take several hours to days to acquire a high SNR Raman image. An alternative to point mapping is line mapping, where a laser beam is line-focused onto the sample with a Powell lens [135] .. In line mapping experiments, the sample is rastered in one direction with the laser line parallel to the slit of the spectrometer. With the use of a two- dimensional (2D) charge-coupled detector (CCD), both spatial and spectral information are obtained, where the former is recorded parallel to the slit and the latter is displayed perpendicular to the slit. Another alternative to point mapping is global imaging, where the sample is illuminated with a laser beam covering the entire field of view. The Raman image of a specific wavenumber is recorded on a 2D CCD by using a bandpass filter to reject all wavelengths except the one corresponding to the wavenumber of interest; a different filter is required for each wavelength/wavenumber of interest. Although these latter two techniques reduce the total integration time greatly, point mapping was used throughout this work since true confocality can only be achieved with the use of the pinhole apertures. 23 y/ pm / are' x / pm Figure 1.7 Chemical images obtained by point mapping experiments. Full Raman spectra are acquired and certain peaks are selected to reconstruct a chemical image of that particular species. Reprinted with permission of Renishaw plc. 24 1.5 Non-Linear Spectroscopy Non-linear spectroscopy arises from the interaction of a molecule with the electric field of two or more incident radiation fields, with the exception of second and third harmonic generation spectroscopy which only require one beam. Generally speaking, the electric polarization of a sample can be expanded as a power series and expressed as [136] P = z (1) + x (2) E1 E2 + x (3)E1 E2E3 + • •^ (1.9) where P is the polarization vector, E is the sum total of all applied electric field intensities, and x( `) is the ith-order dielectric susceptibility of the sample, where zo) >> x (2) >> x(3) If the intensity of the incident electric field is low, the higher-order terms are negligible and the polarization, and thus the system, has a linear response to the applied electric field. The term "non-linear" arises from the non-linear response of the system as the higher-order terms manifest upon irradiation with intense lasers [137, 138]. Although x (2) >> Z (3) , the x(2) term vanishes in centrosymmetric and isotropic media [137, 139] and hence, most non-linear effects are the result of the y3) term, which is always present since it does not depend on the anisotropy of the sample. Experimentally, non-linear spectroscopy can be achieved by multi-wave mixing processes that involve the interaction of n laser fields with wave vectors k1 , k2, k3, k, and frequencies v1 , v2, v3, ..., vn with the sample. This results in a coherent signal with wave vector lc, and frequency vs depicted by the momentum conservation law: ks = +k1 + k2 + k3 • • • + kn^(1.10) and the energy conservation law: = ±V1 ±^+ Vn .^ (1.11) 25 The two non-linear spectroscopic techniques described below are four-wave mixing processes relating to this third-order non-linearity. For both methods, the resulting light propagates in the forward direction [140-142] and the intensity of the collected signal is proportional to the third power of the excitation beam, as implied by Equation (1.9). 1.5.1 Coherent Anti-Stokes Raman Scattering (CARS) Spectroscopy Coherent anti-Stokes Raman scattering spectroscopy, also known as CARS spectroscopy, is a non-linear Raman spectroscopic technique that has the potential of providing greater sensitivity and better spatial resolution than linear Raman imaging. CARS is a type of four-wave mixing technique that involves the use of three high peak intensity pulsed laser beams (a pump, a Stokes, and a probe beam centered at frequencies co p, w s, w pr, respectively *) interacting coherently. By tuning the frequency difference of the pump and Stokes beams to that of a Raman-active mode of a sample component or components, a strong CARS signal is produced (Figure 1.8). Furthermore, not only does CARS display high chemical specificity and thus can provide a chemical image of unstained samples, it also reduces any fluorescence interference since the signal is blue-shifted from the excitation frequency. An elaborate mathematical explanation and theoretical work of CARS can be found elsewhere [143-149]. An advantage of CARS is that high spatial resolution is achieved, since the CARS signal is generated mainly from the focal point where the intensities of the three beams are highest. Thus CARS has a 3D optical sectioning capability with approximately 1-gm spatial resolution in the x, y, and z directions. In CARS microscopy, the CARS signal is optimal where all three beams are exactly phase matched [150] which, as a result of its collinear geometry and tight focusing conditions, occurs at the objective lens focus [151, 152]. The two most common detection modes of CARS microscopy are the epi-detection (E-CARS) and forward-detection (F- CARS) modes. In E-CARS, the backward-scattered signals are collected with the same objective used to focus the beam onto the sample. This detection mode is ideal for imaging objects that are smaller than the excitation wavelength as it reduces the non-resonant background * The laser and CARS frequencies are denoted by co to avoid ambiguity with the vibrational state v n . 26 WCARS t AE V 3 V2 V I V0 signal from the solvent [145, 153, 154]. For objects whose sizes are comparable to or larger than the excitation wavelength, the F-CARS detection mode is preferred since a stronger signal is produced due to the constructive interference of the CARS signal in the forward direction [145, 151, 152]. The limitations of CARS include the presence of the non-resonant polarization which is frequency-independent and thus results in an non-specific background signal, thereby reducing the detection sensitivity of weak Raman bands [143]. Additionally, since the CARS signal is quadratically dependent on the molecular concentration [155, 156], the detection of molecules at low concentrations may prove to be problematic. Furthermore, the chemical selectivity of CARS may be reduced depending on the bandwidth of the excitation lasers used. However, despite the complexity of the instrumental set-up, high cost, and the disadvantages listed above, CARS remains the most popular coherent Raman technique. For example, CARS microscopy has been used to study lipid membranes [157-163] and other biological samples [164-166]; however, it has not been widely applied to the study of the chemistry and structure of plant cuticles. Figure 1.8 Energy level diagram depicting the principle of CARS spectroscopy. (Op, Ws, wpr, and WARS refer to the pump, Stokes, probe, and CARS frequency, respectively. (ocARs = (op — (os + (ow. vn represents the vibrational levels of the ground electronic state. 27 1.5.2 Third Harmonic Generation (THG) Spectroscopy Third harmonic generation (THG) spectroscopy involves the absorption of three identical photons from the excitation laser of frequency v0 and the emission of one photon of frequency 3v, (Figure 1.9) [141]. To understand the origin of the THG signal, the equation for the induced polarization of a sample is revisited. For THG, the polarization, P3 , , induced by an electric field, Ev , can be described as P3v oc Z (3)Ev3 ^ (1.12) where, as before, X(3) is the third-order non-linear susceptibility. Consequently, the THG power, /3v has a third-order dependence on the incident power, Iv , and is expressed as /3v a[X (3) 1 2 1 3V • (1.13) In the case where a Gaussian beam is tightly focused, the THG power is given by [141, 142] where /3v oc [Jr Jv 0.^(3) i4kbzj= .1 % e a (1 + 2i.02 • (1.14) (1.15) In Equation (1.15), z is the axial position, zik = 3k, — k3 ,, is the phase mismatch, and b=kvw,2 is the confocal parameter of the excitation beam with k v as the wave vector at the fundamental wavelength and wo2 as the beam waist diameter. The integration of Equation (1.15) 28 is over the volume of excitation within the sample and calculations of this integration show that a THG signal is only produced when Ak > 0 . The sign of the Gaussian beam phase is reversed at the focal point; therefore, in a completely uniform sample, no THG signal is observed since the third harmonic signal produced in front and behind the focal point destructively interfere with one other, thereby producing a net of zero THG signal [142]. Figure 1.9 Energy level diagram depicting the principle of THG spectroscopy. The resulting THG signal is three times the frequency (3v0) of the incoming beam (v0). Experimentally, a laser beam is tightly focused in the centre of the sample, where the THG light is produced. As stated above, if the sample is homogeneous, then no THG signal is generated. However, if there is inhomogeneity around the focal point due to changes in refractive index, absorption coefficient, or non-linear optical susceptibility, then a non-zero THG signal is observed [167, 168]. High spatial resolution is achieved in THG spectroscopy providing optical sectioning of unstained samples, since the third harmonic light is generated in very close proximity to the beam focus. The disadvantages of this non-linear spectroscopic technique are the lack of chemical specificity and the complexity of the THG image interpretation. Similar to CARS spectroscopy, even though THG spectroscopy has recently been applied to the imaging of biological tissues [169-173] and among them plant tissues [167, 169, 174, 175], this technique has not yet been applied to the study of plant cuticles specifically. 29 1.6 Thesis Objectives and Overview There has been extensive research on plant cuticle biochemistry and biology, thus providing valuable information on the chemical composition of the cuticle. Even so, only the bulk chemical composition is typically reported. In terms of the chemical distribution, our current knowledge is limited to the two cuticular layers. Furthermore, there has been very limited research on the in situ analysis of cuticular triterpenoids, pentacyclic compounds that are commonly found in almost all plants. Due to the extensive research already performed on and much known chemical information of P. laurocerasus leaves [29, 86, 101], this particular species was chosen as the model species. Since a difference in the chemical composition was found on the epicuticular and intracuticular layers, the obvious next question is whether a gradient exists in the lateral dimension as well. Therefore, the ultimate objective of this work is to determine the lateral distribution of very long chain aliphatic compounds and triterpenoids on leaf cuticles using various non-destructive methods that allow for in situ analysis. More specifically, this thesis will address the following questions: • What instrumental methods are available for analyzing the components of plant cuticular waxes and which is the method of choice for analyzing the lateral distribution of the triterpenoid components? • What post-acquisitional data processing is required/advantageous prior to data analysis? • How does one choose the best and most efficient data processing technique? • What excitation wavelength would be optimal and why? • Can Raman microspectroscopy be used to distinguish between structurally similar triterpenoid compounds? • How do the quantitative results obtained by Raman microspectroscopy compare with those obtained using GC? • How do the chemical images obtained by Raman point mapping experiments, CARS imaging, and THG imaging compare to the morphological images obtained by SEM? 30 Following this introduction, Chapter 2 of this thesis will describe the instrumental set-up of the various instruments used in this work. As with most biological samples, the cuticles exhibit a fluorescence background, and thus various forms of baseline correction methods will be addressed in Chapter 3. Prior to handling P. laurocerasus leaf cuticle samples, the results of several calibration experiments performed to validate the data analysis method are discussed in Chapter 4. This is followed by the triterpenoid distribution analysis of isolated adaxial and abaxial cuticles in Chapter 5. Finally, Chapter 6 concludes the thesis with suggestions on further research, based on the chemical imaging techniques developed/described in this thesis, that may provide additional information about the localization of certain chemical compounds in plant cuticles, and hence further insights into their structure and biological function. 31 1.7 References [1] Eames AJ. Morphology of Vascular Plants: Lower Groups; McGraw-Hill Book Company: New York, 1936. [2] Kramer PJ in Plant Physiology: A Treatise; Steward FC, Ed.; Academic Press: New York, 1959; Vol. 2, pp 607-726. [3] Steward FC and Sutcliffe JF in Plant Physiology: a Treatise; Steward FC, Ed.; Academic Press: New York, 1959; Vol. 2, pp 253-478. [4] Artschwager E. Journal of Agricultural Research 1926, 33, 143-176. [5] Esau K. Anatomy of Seed Plants; John Wiley and Sons: New York, 1960. [6] Esau K. Plant Anatomy; 2nd ed.; John Wiley and Sons, Inc.: New York, 1965. [7] Stern KR. Introductory Plant Biology; 3rd ed.; Wm. C. Brown Publishers: Dubuque, 1985. [8] Nobel PS, Zaragoza LJ, and Smith WK. Plant Physiology 1975, 55, 1067-1070. [9] Purves WK, Orians GH, and Heller HC. Life: The Science of Biology; 4th ed.; Sinauer Associates, Inc.: New York, 1994. [10] WattendorffJ and Holloway PJ. Annals of Botany 1980, 46, 13-28. [11] Schmidt HW and Schtinherr J. Planta 1982, 156, 380-384. [12] Riederer M and Schtinherr J. Planta 1988, /74, 127-138. [13] Tenberge KB. Canadian Journal of Botany 1992, 70, 1467-1487. [14] Suzuki K, Miyake H, Taniguchi T, and Maeda E. Japanese Journal of Crop Science 1993, 62, 116-125. [15] Chamberlin MA, Horner HT, and Palmer RG. International Journal of Plant Sciences 1994, 155, 421-436. [16] Rodkiewicz B, Fyk B, and Szczuka E. Sexual Plant Reproduction 1994, 7, 287-289. [17] Baker EA in The Plant Cuticle; Cutler DF, Alvin KL and Price CE, Eds.; Academic Press: London, 1982; Vol. 10, pp 139-165. [18] Walton TJ. Waxes, cutin and suberin; Academic Press: London, 1990. [19] Barthlott W, Neinhuis C, Cutler D, Ditsch F, Meusel I, Theisen I, and Wilhelmi H. Botanical Journal of the Linnean Society 1998, 126, 237-260. [20] Kunst L, Jetter R, and Samuels AL in Biology of the Plant Cuticle; Riederer M and Muller C, Eds.; Blackwell Publishing Ltd.: Oxford, 2006, pp 182-215. 32 [21] Sterk P, Booij H, Schellekens GA, Vankammen A, and Devries SC. Plant Cell 1991, 3, 907-921. [22] Moreau P, Bessoule JJ, Mongrand S, Testet E, Vincent P, and Cassagne C. Progress in Lipid Research 1998, 37, 371-391. [23] Kader JC. Annual Review of Plant Physiology and Plant Molecular Biology 1996, 47, 627-654. [24] Arondel V, Vergnolle C, Cantrel C, and Kader JC. Plant Science 2000, 157, 1-12. [25] Rogers DP and Bankaitis VA. International Review of Cytology - A Survey of Cell Biology 2000, 197, 35-81. [26] Thoma S, Kaneko Y, and Somerville C. Plant Journal 1993, 3, 427-436. [27] Pyee J, Yu HS, and Kolattukudy PE. Archives of Biochemistry and Biophysics 1994, 311, 460-468. [28] Jetter R, Kunst L, and Samuels AL in Biology of the Plant Cuticle; Riederer M and Muller C, Eds.; Blackwell Publishing Ltd.: Oxford, 2006, pp 145-181. [29] Jetter R, Schaffer S, and Riederer M. Plant Cell and Environment 2000, 23, 619-628. [30] Buschhaus C, Herz H, and Jetter R. New Phytologist 2007, 176, 311-316. [31] Buschhaus C, Herz H, and Jetter R. Annals of Botany 2007, 100, 1557-1564. [32] Jetter R. 2007, Personal communication. [33] Holloway PJ in The Plant Cuticle; Cutler DF, Alvin KL and Price CE, Eds.; Academic Press: London, 1982, pp 45-85. [34] Kolattukudy PE. Canadian Journal of Botany 1984, 62, 2918-2933. [35] Stark RE, Yan B, Ray AK, Chen Z, Fang X, and Garbow JR. Solid State Nuclear Magnetic Resonance 2000, 16, 37-45. [36] Villena JF, Dominguez E, and Heredia A. Journal of Plant Physiology 2000, 156, 419- 422. [37] Kolattukudy PE in Advances in Biochemical Engineering/Biotechnology; Scheper T, Ed.; Springer: Berlin, 2001, pp 4-49. [38] Nip M, Tegelaar EW, Deleeuw JW, Schenck PA, and Holloway PJ. Naturwissenschaften 1986, 73, 579-585. [39] Jeffree CE in Plant Cuticles: an Integrated Functional Approach; Kerstiens G, Ed.; Bios Scientific Publishers Ltd.: Oxford, 1996, pp 33-82. 33 [40] Jeffree CE in Insects and the Plant Surface; Juniper BE and Southwood R, Eds.; Edward Arnold: London, 1986, pp 23-64. [41] SchOnherr J. Planta 1976, 131, 159-164. [42] Holloway PJ. Pesticide Science 1970, 1, 156-163. [43] Barthlott W and Neinhuis C. Planta 1997, 202, 1-8. [44] Wagner P, Furstner R, Barthlott W, and Neinhuis C. Journal of Experimental Botany 2003, 54, 1295-1303. [45] Otten A and Herminghaus S. Langmuir 2004, 20, 2405-2408. [46] Bargel H, Koch K, Cerman Z, and Neinhuis C. Functional Plant Biology 2006, 33, 893- 910. [47] I316e E in Lipid Biotechnology; Kuo TM and Gardner HW, Eds.; Marcel Dekker: New York, 2002, pp 231-248. [48] Carver TLW and Thomas BJ. Plant Pathology 1990, 39, 367-375. [49] Carver TLW, Thomas BJ, Ingerson-Morris SM, and Roderick HW. Plant Pathology 1990, 39, 573-583. [50] Carver TLW and Gurr SJ in Biology of the Plant Cuticle; Riederer M and Muller C, Eds.; Blackwell Publishing Ltd.: Oxford, 2006, pp 368-397. [51] Heredia A. Biochimica Et Biophysica Acta-General Subjects 2003, 1620, 1-7. [52] Eigenbrode SD and Espelie KE. Annual Review of Entomology 1995, 40, 171-194. [53] Day TA and Neale PJ. Annual Review of Ecology and Systematics 2002, 33, 371-396. [54] HollOsy F. Micron 2002, 33, 179-197. [55] Jordan BR. Functional Plant Biology 2002, 29, 909-916. [56] Kakani VG, Reddy KR, Zhao D, and Sailaja K. Agricultural and Forest Meteorology 2003, 120, 191-218. [57] Luque P and Heredia A. Plant Physiology and Biochemistry 1997, 35, 251-256. [58] Casado CG and Heredia A. Biochimica Et Biophysica Acta-Biomembranes 2001, 1511, 291-296. [59] Casado CG and Heredia A. Biomacromolecules 2001, 2, 407-409. [60] Guhling 0, Kinzler C, Dreyer M, Bringmann G, and Jetter R. Journal of Chemical Ecology 2005, 31, 2323-2341. [61] Priestley JH. Botany Review 1943, 9, 593-616. [62] Roelofsen PA. Handbuch der Pflanzenanatomie 1959, 4, 86-90. 34 [63] Roelofsen PA. Handbuch der Pflanzenanatomie 1959, 4, 257-268. [64] Frey-Wyssling A and Miihlethaler K. Ultrastructural Plant Cytology; Elsevier: Amsterdam, 1965. [65] Meyer M. Protoplasma 1938, 29, 552-586. [66] Roelofsen PA. Acta Botanica Neerlandic 1952, 1, 99-114. [67] Anderson DB. Jahrbiicher fur wissenschaftliche Botanik 1928, 69, 501-515. [68] Sitte P and Rennier R. Planta 1963, 60, 19-40. [69] Norris RF and Bukovac MJ. American Journal of Botany 1968, 55, 975-983. [70] Wilson LA, Cain RF, and Beavers DV. American Journal of Botany 1972, 59, 722-728. [71] Merida T, SchOnherr J, and Schmidt HW. Planta 1981, 152, 259-267. [72] Mazliak P in Progress in Phytochemistry; Reinhold L and Lienschitz Y, Eds.; Interscience Publishers: New York, 1968, pp 49-111. [73] Canet D, Rohr R, Charnel A, and Guillain F. New Phytologist 1996, 134,571-577 . [74] Franke R, Briesen I, Wojciechowski T, Faust A, Yephremov A, Nawrath C, and Schreiber L. Phytochemistry 2005, 66, 2643-2658. [75] Heide-JOrgensen HS. Planta 1991, 183, 511-519. [76] Cook ME and Graham LE. International Journal of Plant Sciences 1998, 159, 780-787. [77] Ensikat HJ, Neinhuis C, and Barthlott W. International Journal of Plant Sciences 2000, 161, 143-148. [78] Schieferstein RH and Loomis WE. Plant Physiology 1956, 31, 240-247. [79] Martin JT and Juniper BE. The Cuticles of Plants; Edward Arnold: London, 1970. [80] Schieferstein RH and Loomis WE. American Journal of Botany 1959, 46, 625-635. [81] Round AN, Yan B, Dang S, Estephan R, Stark RE, and Batteas JD. Biophysical Journal 2000, 79, 2761-2767. [82] Viougeas MA, Rohr R, and Charnel A. New Phytologist 1995, 130, 337-348. [83] Yang HS, An HJ, Feng GP, and Li YF. LWT-Food Science and Technology 2005, 38, 571-577. [84] Benitez JJ, Matas AJ, and Heredia A. Journal of Structural Biology 2004, 147, 179-184. [85] Mechaber WL, Marshall DB, Mechaber RA, Jobe. RT, and Chew FS. Proceedings of the National Academy of Sciences of the United States of America 1996, 93, 4600-4603. [86] Perkins MC, Roberts CJ, Briggs D, Davies MC, Friedmann A, Hart CA, and Bell GA. Planta 2005, 221, 123-134. 35 [87] Koch K, Neinhuis C, Ensikat HJ, and Barthlott W. Journal of Experimental Botany 2004, 55,711-718. [88] Karabourniotis G, Tzobanoglou D, Nikolopoulos D, and Liakopoulos G. Annals of Botany 2001, 87, 631-639. [89] Zlotnik-Mazori T and Stark RE. Macromolecules 1988, 21, 2412-2417. [90] Deshmukh AP, Simpson AJ, and Hatcher PG. Phytochemistry 2003, 64, 1163-1170. [91] Schreiber L, Schorn K, and Heimburg T. European Biophysics Journal with Biophysics Letters 1997, 26, 371-380. [92] Reynhardt EC and Riederer M. Journal of Physics D-Applied Physics 1991, 24, 478-486. [93] Aggarwal P. Thermochimica Acta 2001, 367, 9-13. [94] Casado CG and Heredia A. Journal of Experimental Botany 1999, 50, 175-182. [95] Ray AK and Stark RE. Phytochemistry 1998, 48, 1313-1320. [96] Hamosh M. Lingual and Gastric Lipases: Their Role in Fat Digestion; CRC Press: Boca Raton, 1990. [97] Purdy RE and Kolattukudy PE. Biochemistry 1975, 14, 2832-2840. [98] Brown AJ and Kolattukudy PE. Archives of Biochemistry and Biophysics 1978, 190, 17- 26. [99] Ray AK, Chen ZJ, and Stark RE. Phytochemistry 1998, 49, 65-70. [100] Osman SF, Gerard HC, Fett WF, Moreau RA, and Dudley RL. Journal of Agricultural and Food Chemistry 1995, 43, 2134-2137. [101] Jetter R and Schaffer S. Plant Physiology 2001, 126, 1725-1737. [102] Kunst L and Samuels AL. Progress in Lipid Research 2003, 42, 51-80. [103] Christie WW. Lipid Analysis; 3rd ed.; The Oily Press: Bridgewater, 2003. [104] Markstadter C, Federle W, Jetter R, Riederer M, and Holldobler B. Chemoecology 2000, 10,33-40. [105] Spencer GF, Plattner RD, and Miwa T. Journal of the American Oil Chemists Society 1977, 54, 187-189. [106] Tulloch AP. Chemistry of waxes of higher plants; Elsevier: Amsterdam, 1976. [107] Bianchi G in Waxes: Chemistry. Molecular Biology and Functions; Hamilton RJ, Ed.; The Oily Press: Dundee, 1995, pp 175-222. [108] Holloway PJ and Challen SB. Journal of Chromatography 1966, 25, 336-346. 36 [109] Holloway PJ in CRC Handbook of Chromatography. Lipids; Mangold HK, Zweig G and Sherma J, Eds.; CRC Press: Boca Raton, 1984; Vol. 1, pp 347-380. [110] Gawronska-Grzywacz M and Krzaczek T. Journal of Separation Science 2007, 30, 746- 750. [111] Hietala T, Mozes N, Genet MJ, Rosenqvist H, and Laakso S. Colloids and Surfaces B- Biointerfaces 1997, 8, 205-215. [112] Barr TL, Seal S, Hardcastle SE, Maclaurin MA, Chen LM, and Klinowski J. Bulletin of the Polish Academy of Sciences-Chemistry 1997, 45, 17-30. [113] Derue C, Gibouin D, Demarty M, Verdus MC, Lefebvre F, Thellier M, and Ripoll C. Microscopy Research and Technique 2006, 69, 53-63. [114] Dickinson M, Heard PJ, Barker JHA, Lewis AC, Mallard D, and Allen GC. Applied Surface Science 2006, 252, 6793-6796. [115] Luque P, Bruque S, and Heredia A. Archives of Biochemistry and Biophysics 1995, 317, 417-422. [116] Ramirez FJ, Luque P, Heredia A, and Bukovac MJ. Biopolymers 1992, 32, 1425-1429. [117] Merk S, Blume A, and Riederer M. Planta 1998, 204, 44-53. [118] da Luz BR. New Phytologist 2006, 172, 305-318. [119] Yamauchi S, Kikuchi Y, and Kurimoto Y. Materials Science Research International 2000, 6, 227-228. [120] Pinzaru SC, Leopold N, and Kiefer W. Talanta 2002, 57, 625-631. [121] Yamauchi S and Kurimoto Y. Journal of Wood Science 2003, 49, 235-240. [122] Creagh DC and Otieno-Alego V. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 2004, 213, 670-676. [123] Frosch T, Schmitt M, and Popp J. Journal of Physical Chemistry B 2007, 111, 4171- 4177. [124] Agarwal UP and Atalla RH. Planta 1986, 169, 325-332. [125] Agarwal UP. Planta 2006, 224, 1141-1153. [126] Greene PR and Bain CD. Colloids and Surfaces B: Biointerfaces 2005, 45, 174-180. [127] Raman CV and Krishnan KS. Nature 1928, 121, 501-502. [128] Jayasooriya UA and Jenkins RD in An Introduction to Laser Spectroscopy; 2nd ed.; Andrews DL and Demidov AA, Eds.; Kluwer Academic/Plenum Publishers: New York, 2002, pp 77-104. 37 [129] Puppels GJ in Fluorescent and Luminecent Probes for Biological Activity; 2nd ed.; Mason WT, Ed.; Academic Press: New York, 1999, p 337. [130] Turrell G, Delhaye M, and Dhamelincourt P in Raman Microscopy: Developments and Applications; Turrell G and Corset J, Eds.; Academic Press: Cambridge, 1996, pp 27-49. [131] Baldwin KJ, Batchelder DN, and Webster S in Handbook of Raman Spectroscopy: From the Research Laboratory to the Process Line; Lewis IR and Edwards HGM, Eds.; Marcel Dekker, Inc.: New York, 2001, pp 145-190. [132] Turrell G and Corset J. Raman Microscopy: Developments and Applications; Academic Press: Cambridge, 1996. [133] Barbillat J in Raman Microscopy: Developments and Applications; Turrell G and Corset J, Eds.; Academic Press: Cambridge, 1996, pp 175-200. [134] McCreery RL. Raman Spectroscopy for Chemical Analysis; John Wiley & Sons, Inc.: New York, 2000. [135] Bewsher A, Powell I, and Boland W. Applied Optics 1996, 35, 1654-1658. [136] Bloembergen N. Non-Linear Optics; W. A. Benjamin, Inc.: New York, 1965. [137] Boyd RW. Nonlinear Optics; 2nd ed.; Academic Press: San Diego, 2003. [138] Mukamel S. Principles of Nonlinear Optical Spectroscopy; Oxford University Press: New York, 1995. [139] Andrews DL and Meech SR in An Introduction to Laser Spectroscopy; 2nd ed.; Andrews DL and Demidov AA, Eds.; Kluwer Academic/Plenum Publishers: New York, 2002, pp 233-256. [140] Rodriguez LG, Lockett SJ, and Holtom GR. Cytometry Part A 2006, 69A, 779-791. [141] Barad Y, Eisenberg H, Horowitz M, and Silberberg Y. Applied Physics Letters 1997, 70, 922-924. [142] Boyd RW. Nonlinear Optics; Academic Press: New York, 1992. [143] Volkmer A. Journal of Physics D-Applied Physics 2005, 38, R59-R81. [144] Volkmer A, Cheng JX, and Xie XS. Physical Review Letters 2001, 87, 023901. [145] Cheng JX, Volkmer A, and Xie XS. Journal of the Optical Society of America B-Optical Physics 2002, 19, 1363-1375. [146] Milner M, Squier J, de Lange CA, and Brakenhoff GJ. Journal of Microscopy-Oxford 2000, 197, 150-158. 38 [147] Potma EO, de Boeij WP, and Wiersma DA. Journal of the Optical Society of America B- Optical Physics 2000, 17, 1678-1684. [148] Hashimoto M and Araki T. Journal of the Optical Society of America a-Optics Image Science and Vision 2001, 18, 771-776. [149] Cheng JX and Xie XS. Journal of the Optical Society of America B-Optical Physics 2002, 19, 1604-1610. [150] Bjorklund GC. IEEE Journal of Quantum Electronics 1975, QE11, 287-296. [151] Cheng JX and Xie XS. Journal of Physical Chemistry B 2004, 108, 827-840. [152] Cheng JX. Applied Spectroscopy 2007, 61, 197A-208A. [153] Cheng JX, Vollmer A, Book LD, and Xie XS. Journal of Physical Chemistry B 2001, 105, 1277-1280. [154] Potma EO, Jones DJ, Cheng JX, Xie XS, and Ye J. Optics Letters 2002, 27, 1168-1170. [155] Holtom GR, Thrall BD, Chin BY, Wiley HS, and Colson SD. Traffic 2001, 2, 781-788. [156] Levenson MD and Kano SS. Introduction to Nonlinear Laser Spectroscopy; Academic Press: San Diego, 1988. [157] Nan XL, Potma EO, and Xie XS. Biophysical Journal 2006, 91, 728-735. [158] Li L, Wang HF, and Cheng JX. Biophysical Journal 2005, 89, 3480-3490. [159] Nan XL, Cheng JX, and Xie XS. Journal of Lipid Research 2003, 44, 2202-2208. [160] Mtiller M and Schins JM. Journal of Physical Chemistry B 2002, 106, 3715-3723. [161] Mtiller M, Schins JM, Nastase N, Wurpel S, and Brakenhoff FGJ. Biophysical Journal 2002, 82, 175A-175A. [162] Potma EO and Xie XS. ChemPhysChem 2005, 6, 77-79. [163] Wurpel GWH, Rinia HA, and Mailer M. Journal of Microscopy-Oxford 2005, 218, 37- 45. [164] Kano H and Hamaguchi H. Optics Express 2005, 13 , 1322-1327. [165] Kee TW and Cicerone MT. Biophysical Journal 2005, 88, 362A. [166] Wang HF, Fu Y, Zickmund P, Shi RY, and Cheng JX. Biophysical Journal 2005, 89, 581-591. [167] Mailer M, Squier J, Wilson KR, and Brakenhoff GJ. Journal of Microscopy-Oxford 1998, 191, 266-274. [168] Oron D, Yelin D, Tal E, Raz S, Fachima R, and Silberberg Y. Journal of Structural Biology 2004, 147, 3-11. 39 [169] Debarre D, Supatto W, Pena AM, Fabre A, Tordjmann T, Combettes L, Schanne-Klein MC, and Beaurepaire E. Nature Methods 2006, 3, 47-53. [170] Debarre D and Beaurepaire E. Biophysical Journal 2007, 92, 603-612. [171] Tai SP, Lee WJ, Shieh DB, Wu PC, Huang HY, Yu CH, and Sun CK. Optics Express 2006, 14, 6178-6187. [172] Schaller RD, Johnson JC, and Saykally RJ. Analytical Chemistry 2000, 72, 5361-5364. [173] Debarre D, Pena AM, Supatto W, Boulesteix T, Strupler M, Sauviat MP, Martin JL, Schanne-Klein MC, and Beaurepaire E. Medecine Sciences 2006, 22, 845-850. [174] Millard AC, Wiseman PW, Fittinghoff DN, Wilson KR, Squier JA, and Muller M. Applied Optics 1999, 38, 7393-7397. [175] Chu SW, Chen IH, Liu TM, Chen PC, Sun CK, and Lin BL. Optics Letters 2001, 26, 1909-1911. 40 CHAPTER 2 INSTRUMENTATION While integrated instrument systems for linear Raman microspectroscopy have long been commercially available, their configurations vary widely. Instrumentation for non-linear optical spectroscopy still requires complex custom experimental set-ups. This chapter describes the specific instrumental set-ups for spontaneous confocal Raman, coherent anti-Stokes Raman scattering (CARS), and third harmonic generation (THG) microspectroscopy. These same set- ups are utilized for all subsequent chapters unless otherwise specified. 2.1 Raman Microspectroscopy The first Raman microspectrometer was assembled by Tomas Hirschfeld in 1973 [1] after Michel Delhaye and Michel Migeon introduced the concept of Raman microspectroscopy in 1966 [2]. Although two other Raman microspectrometers were subsequently introduced in 1974 [3, 4], it was Hirschfeld' s instrument that defined the basic principles of Raman microspectroscopic instrumentation and was later commercialized [5]. These commercial instruments can be used with various excitation lasers and allow for point-by-point analysis, as well as global imaging, of the sample. Confocality, explained here, can also be achieved with the commercial Raman microspectrometers combined with the use of charge-coupled device (CCD) detectors. 2.1.1 Lasers The spontaneous Raman experiments were performed using three different excitation wavelengths. Visible excitation was achieved using an Argon 514 nm laser (Laser Physics UK Ltd., Cheshire, UK) and a Renishaw RL633 helium-neon (HeNe) 633 nm laser (Renishaw, Gloucester, UK). Two types of NIR lasers with a wavelength of 785 nm were used during this research. Initially, a 700 to 1000 nm tunable titanium:sapphire (Ti:Sapph, TiAl 2O3) MIRA 41 Model 900D laser (Coherent Laser Group, CA) was used as the NIR excitation laser. The pumping source for the MIRA laser was a single-frequency green 532 nm output Verdi TM V-10 solid-state diode-pumped frequency-doubled neodymium vanadate (Nd:YVO4) laser (Coherent), which has a maximum output power of 10 W. The green light from the pump laser is converted to the desired 785 nm by tuning the laser output of the MIRA. The stability of the wavelength was measured with a Miniature Fiber Optic spectrometer model SD2000 (Ocean Optics, Inc., Dunedin, FL). The laser beam was directed through a holographic laser bandpass filter model HLBF-785.0 (Kaiser Optical Systems, Inc., MI) with a >90% throughput transmission and a narrow-pass bandwidth of less than 2 nm. This laser system was later replaced by a Renishaw RL785 diode laser (Renishaw). There are several advantages of using a diode laser, among them its ease of use, affordability and stability. Power stability was a key factor in the decision to replace the MIRA laser with the diode laser: the power level instability of the MIRA laser lead to unacceptable variance in peak height measurements made over periods of more than one hour, which did not permit quantitative comparison of sequential measurements to be made at widely separated grid locations in a sample. The output power of both lasers was measured at the beginning of each day using a FieldMaster power meter (Coherent). 2.1.2 Renishaw Raman Microscope The NIR Raman experiments were performed using the Renishaw Ramanscope System 1000 (Renishaw) coupled to an infinity corrected Leica DMLB microscope (Leica Microsystems, Richmond Hill, ON, Canada), while the visible Raman experiments were performed using the Renishaw inVia Raman Microscope System (Renishaw) coupled to a Leica DMLM microscope (Leica Microsystems). Microscope objectives of 5x/0.12 numerical aperture (NA), 20x/0.40 NA, 50x/0.75 NA, and 100x/0.75 NA magnification were available to focus the laser onto the sample and for collecting the Raman scattered light from the sample. The power at the tip of the microscope objectives was measured with a handheld LaserCheck power meter (Coherent) at the beginning of each experiment. For scanning purposes, the microscope is equipped with a motorized, programmable xyz stage (ProScan Series, Prior Scientific Inc., MA), where the smallest step sizes for the lateral plane is 0.1 gm while that for the vertical axis is 0.01 42 gm. The microscope is also coupled to a video camera, which allows visible image capture and real-time sample image display on a computer monitor to facilitate sample positioning. The optical path for the Raman spectrometer is provided in Figure 2.1. The incoming laser is passed through a beam expander to obtain a uniform excitation beam. For the Ti:Sapph MIRA laser, a 10x beam expander was used, however, this was replaced with a 1 x beam expander for the diode laser (as suggested by Renishaw [6]). Both the excitation beam and the Raman scattered radiation are directed to the holographic notch filter (HNF), where the former is reflected off the surface while the latter passes through the filter. Typically, a 50/50 or 70/30 beam splitter is used, but this current method of using HNFs, which optimizes the coupling efficiency, permits more than 90% of the Raman signal from the sample to reach the grating spectrometer [7] all the while rejecting the Rayleigh radiation. The slit width of the spectrometer was set to 10 gm and with the 1200 lines/mm grating, the spectral resolution was measured to be approximately 4 cm 1. The Raman microscope is equipped with a —70 °C thermoelectrically cooled AIMO CCD02-06 CCD array detector (e2v technologies, Essex, UK), with a resolution of 0.97 cm-l/pixel [8]. The Raman system is calibrated at various times throughout the experiment to ensure consistency among the Raman shifts between spectral acquisitions. The calibration of the wavenumber offset was done by acquiring a spectrum of a piece of silicon semiconductor wafer and calibrating the spectrum to its characteristic phonon band at 520 cm -1 [9]. 43 EV to computer .41-- 7S5 nni laser input A Figure 2.1 Schematic diagram of the Renishaw Ramanscope System 1000. The laser beam comes in from the back (A) and is directed to the holographic notch filter (B). The excitation beam (red) is directed to the microscope (C) and the Raman signal (brown) follows the same pathway back. The Raman light is then dispersed by a grating (D) and detected by a CCD detector (E). Reproduced with permission of Renishaw plc. 2.1.3 Confocal Modality The confocal capabilities of the microscope allow spectra from within a well-defined region of the sample to be acquired by accepting light from the region of interest and rejecting light from regions that are not at the focal point, ultimately improving the spatial resolution. Traditionally, this has been achieved with the use of pinhole apertures and a dichroic beam splitter (Figure 2.2) [10, 11]. The size of the detector pinhole aperture determines how much light enters the detector and from which optical path angles. By choosing a small pinhole aperture, only the light originating from the in-focus plane will reach the detector, while all other light originating from out-of-focus planes are blocked by the pinhole. Therefore, acquiring spectra while moving the stage in the axial dimension allows for optical sectioning of the sample, and a subsequent three-dimensional (3D) reconstruction, to be made [12]. 44 dichroic beam splitter pinhole aperture PMT emission filter >-‹ rf --t x-y scanners objective focal plane r Figure 2.2 Schematic diagram of a typical confocal microscope set-up. The light pathway shown in solid green is detected by the detector under this set of confocal parameters. Hence only the signal coming from the focused spot on the focal plane is detected. Reproduced from http://www.gonda.ucla.edu/bri_core/confocal.htm (accessed on December 2007) with permission of van der Wulp, TNO Netherlands. The Renishaw Raman microscope uses a slightly different method from the common pinhole aperture mentioned above, to achieve the same confocal results. The key components involved in the ability to perform confocal experiments are the slit width and the image area of the CCD detector. The slit serves as a spatial filter for the light entering the grating spectrograph, while the CCD image area provides further spatial filtering at a 90 ° angle (Figure 2.3), where collectively, they act as the detector pinhole aperture described above. All Raman experiments were performed with the microscope in the confocal mode. The confocal parameters when acquiring spectra using a 50x/0.75 NA microscope objective were a 45 CCD detector slit slit width of 10 gm and a CCD image area of 3 x 576 pixels. This resulted in a lateral spatial resolution of approximately 5 gm and an axial spatial resolution of approximately 5 gm, the latter determined by Renishaw [7]. Figure 2.3 Confocality of the Renishaw Ramanscope System 1000. Adjustment of the slit opening and CCD detector area rejects light from out-of-focus regions thereby allowing images of different sections of a specimen to be obtained. Modified with permission of Renishaw plc. 2.1.4 Software The Raman signals detected by the CCD camera were collected using the Renishaw WiRETM (Windows® based Raman Environment, version 1.3.30, Renishaw) and Grams/32® 46 (Graphic Relational Array Management System, version 4.14, level II, Galactic Industries Corporation, Salem, NH) software. Baseline corrections were done using HiQ (National Instruments Corporation, Austin, TX) while data analyses were performed using Matlab (Mathworks, Natick, MA). 2.2 CARS and THG Microspectroscopy One of the first non-linear optical microscopes was developed in the 1970s using the principle of second harmonic generation to image crystals [13, 14]. Although Duncan et al. used a CARS microscope with a dye laser system to image D 20 in onion cells in 1982 [15], it wasn't until the introduction of a mode-locked femtosecond dye laser [16], as well as high resolution CARS [17] and THG imaging [18], that prompted non-linear microspectroscopy to flourish. In this report, the terms CARS/THG spectroscopy, CARS/THG imaging, and CARS/THG microscopy will be used interchangeably. 2.2.1 Laser Set-up The experimental set-up up for the non-linear measurements is shown in Figure 2.4. The CARS experiments were done using a forward-detection (F-CARS) set-up since a stronger CARS signal is produced in the forward direction as a result of the constructive interference of the coherent radiation fields. A Ti:Sapph laser (Spectra Physics, Mountain View, CA) equipped with a Mai Tai oscillator and a Spitfire HPR amplifier, produced an output beam at 800 nm comprising 150 fs pulses at a 1 kHz repetition rate with 2 mJ energy per pulse. The Ti:Sapph output beam was split into two branches: one was used as a probe beam in the CARS experiment, the other was used to pump an optical parametric amplifier (OPA) (Light Conversion, Vilnius, Lithuania) with wavelength tunability from 0.4 to 2.8 gm. The second harmonic of the variable OPA idler output was used for the CARS pump beam, while the OPA signal output was used for the CARS Stokes beam. The pump, Stokes, and probe beams were collinearly combined with the use of dielectric mirrors and focused onto the sample through a 40x/0.65 NA microscope objective (Leica Microsystems). The sample was placed on an 47 ^/ .^ 1 1 \ I pump \., ^ 1\^\ Ti-Sapphire laser probe To computerBoxcar averager Stokes OPA automated, motorized xyz stage and a chemical image, with 150 cm -1 spectral resolution, was obtained by scanning the sample at a fixed wavelength. A similar experimental set-up was used for the THG experiments except that only one beam with a wavelength of 1248 nm was used for sample excitation. The CARS and THG signals were collimated with a 20x/0.50 NA microscope objective (Leica Microsystems) and transmitted through a spectrometer (McPherson, Chelmsford, MA) to filter out the pump radiation and all spurious signals generated by other non-linear processes. The signals were detected with a photomultiplier tube (Hamamatsu, Bridgewater, NJ) and fed into a boxcar averager (Stanford Research Systems, Sunnyvale, CA) prior to collection. With this set-up, a spatial resolution of 2 gm in the x, y, and z dimensions was achieved for both the CARS and THG experiments. filter Delay^Delay line I^line 2 Figure 2.4 Schematic diagram of the CARS and THG instrumental set-up. The probe, Stokes, and pump laser beams for the F-CARS experiments are indicated by the solid, dashed, and dotted lines, respectively, while the resulting CARS signal is indicated by the dot-dashed line. Similarly for THG experiments, the excitation beam and the THG signal are indicated by the solid and dot-dashed lines, respectively. 48 2.2.2 Software The CARS and THG signals were collected using a custom Labview program (National Instruments Corporation). No post-acquisitional data processing was performed aside from normalizing the data to the highest signal value. Chemical images of the normalized raw data were then generated in Surfer (Golden Software, Golden, CO). 49 2.3 References [1] Hirschfeld T. Journal of the Optical Society of America 1973, 63, 476-477. [2] Delhaye M and Migeon M. Comptes Rendus Hebdomadaires des Seances de l'Academie des Sciences Serie B 1966, 262, 1513-1516. [3] Delhaye M and Dhamelincourt P. IVth International Conference of Raman Spectroscopy, 1974, Brunswick, ME, USA. [4] Rosasco GJ, Etz ES, and Cassatt WA. IVth International Conference of Raman Spectroscopy, 1974, Brunswick, ME, USA. [5] Turrell G and Corset J. Raman Microscopy: Developments and Applications; Academic Press: Cambridge, 1996. [6] Prusnick T. 2003, Renishaw technician. Personal communication. [7] Renishaw. 2000, Spectroscopy Products Division, UK. [8] Renishaw. 2004, Spectroscopy Products Division, UK. [9] Kovalchenko A, Gogotsi Y, Domnich V, and Erdemir A. Tribology Transactions 2002, 45, 372-380. [10] Itoh J. Journal of Advanced Science 2001, 13, 7-10. [11] Schrof W, Klingler J, Heckmann W, and Horn D. Colloid and Polymer Science 1998, 276, 577-588. [12] Puppels GJ, Colier W, Olminkhof JHF, Otto C, Demul FFM, and Greve J. Journal of Raman Spectroscopy 1991, 22, 217-225. [13] Gannaway JN and Sheppard CJR. Optical and Quantum Electronics 1978, 10, 435-439. [14] Hellwarth R and Christensen P. Optics Communications 1974, 12, 318-322. [15] Duncan MD, Reintjes J, and Manuccia TJ. Optics Letters 1982, 7, 350-352. [16] Denk W, Strickler JH, and Webb WW. Science 1990, 248, 73-76. [17] Zumbusch A, Holtom GR, and Xie XS. Physical Review Letters 1999, 82, 4142-4145. [18] Barad Y, Eisenberg H, Horowitz M, and Silberberg Y. Applied Physics Letters 1997, 70, 922-924. 50 CHAPTER 3 POST-ACQUISITIONAL SPECTRAL PROCESSING * Vibrational spectra often require baseline removal before further data analysis can be performed. This is especially true for biological samples where a fluorescence background is typically observed [1]. This chapter presents a comparative study of the methods most commonly used with Raman spectra and each of the methods presented here will be assessed to determine which baseline removal technique is most suitable for the triterpenoid, and eventually the cuticle, spectra that will ultimately be used for quantitative analysis. While various baseline removal techniques were investigated, only the methods demonstrating the highest baseline removal qualities for each category listed below are described in this chapter due to space consideration. 3.1 Introduction Acquired spectra often contain undesirable elements such as noise and background features in addition to the desired signal itself. Hence, it may be desirable, or even necessary, to separate the signal from both background and noise. The specific reasons for baseline removal are manifold and application dependent. These may range from relatively simple requirements such as presentation, to more exacting requirements such as quantification or preparation for further numerical processing, and the stringency required of any baseline correction method (BCM) may vary accordingly. Moreover, the accuracy obtainable, modes of failure, and computation time required vary between, and sometimes within, different BCMs and an understanding of these trade-offs facilitate the selection of a BCM suitable to the problem at hand * A version of this chapter has been published. Jirasek A, Schulze G, Yu MML, Blades MW, and Turner RFB. Accuracy and Precision of Manual Baseline Determination. Applied Spectroscopy, 2004, 58(12), 1488-1499. Schulze G, Jirasek A, Yu MML, Lim A., Turner RFB, and Blades MW. Investigation of Selected Baseline Removal Techniques as Candidates for Automated Implementation. Applied Spectroscopy, 2005, 59(5), 545-574. 51 Ruckstuhl et al. observed that the available literature on BCMs is widely dispersed amongst many fields of research and commented that useful review articles are relatively scarce [2]. This highlights the widespread interest in, and utility of, BCMs and underscores the difficulties in undertaking a systematic study of the available methods. Not only do they have to be retrieved from the literature of different specialty areas, but they may also have to be reformulated to make valid comparisons that reveal their relative merits. Early reports of baseline correction that appear in the literature were predominantly based on hardware implementations or modifications [3-6]. In 1969, some of the first computer-based methods designed for the automated removal of baselines were published [7, 8]. Activity in the development and use of BCMs accelerated through the 1970s and 1980s, partly facilitated by the increasing availability of computing equipment [9-49]. The first comparisons, reviews, and evaluations of BCMs also appear during this period [50-54]. Since then, a few additional review and/or evaluation articles have appeared [55, 56], as well as a great number of new methodological developments. Some of these methods were specifically designed for automation, while others clearly offer such potential. It is likely that increases in small computer performance, improved instrumentation yielding vast amounts of data to be processed, and the increasing costs of manual methods of baseline removal will provide continued driving forces for automation. Measured spectra can be considered to consist of the signals of interest plus a background, both convolved with an instrumental blurring function, and with noise added to the result. Intuitively, the background is taken to comprise low frequency spectral components, while noise is considered to have high frequency components. The signals of interest usually have frequency components between those of the background and noise. This somewhat simplified view is adopted to recast a number of different BCMs to facilitate comparisons between them. It is acknowledged that, in reality, the situation is more complex since different sources of noise, with widely different frequency components, are typically present [57]. On occasion, one may even have an interest in studying the noise itself or in the background itself However, these additional dimensions would unnecessarily complicate the comparative study presented here and are neglected. 52 Implicitly, the real analytical problem is essentially a signal recovery problem, and baseline correction is an intermediate step towards this goal. Nevertheless, a variety of approaches have been designed specifically for the baseline removal step. These methods differ, in general, by the extent or type of explicit knowledge or assumptions about the spectral components such as background, noise, blurring function, etc., required, and/or in the order with which they are dealt. Clearly, to include every baseline removal method developed to date would be intractable (although an extensive bibliographic reference list is offered as an aid to interested readers). In this chapter, techniques that are designed to be applied to spectral data a posteriori (i.e. after spectra have been acquired using whatever instrumentation is appropriate) are focused upon. Additional techniques rely, at least in part, on the instrumentation itself [3-6, 58-65] and they will not be discussed further. Of the non-instrumental methods, BCMs that represent techniques that (a) require no a priori information, or (b) do require some a priori information (e.g. regarding the noise, baseline, blurring function, signal, etc.) have been chosen. These represent a cross-section of methods both conceptually and in terms of the underlying assumptions they make about and treatment of the data. In this chapter, the problem of baseline correction in general theoretical terms is introduced, followed by the materials and methods used in the evaluation of the BCMs. Following this introduction, each BCM to be evaluated is discussed separately in some detail. Each BCM discussion "module" includes some of its history, theory of operation, implementation, evaluation on a wide range of synthetic spectra, advantages and disadvantages, modes of failure encountered, and suitability for automation. 3.2 Theory A given vector y = {y l , y2, ....y,} of i observed frequency domain spectral intensities can be modeled as the sum of an ideal spectrum s and a background b, convolved with a blurring function p, with noise n added to the result giving y=((s+b)*p)+n^ (3.1) 53 where * denotes convolution. The noise amplitude versus frequency distribution is often taken to be Gaussian [2], but is perhaps more correctly described as Poissonian [66]. Either way, the two distributions become highly similar under some conditions and one could consider their means as contributing to the baseline b while their variances constitute the noise n of Equation (3.1). The issue of noise and its characterization is also addressed elsewhere [67]. The problem for the data analyst is now to recover s from Equation (3.1), hence to determine s = ((y — n)* p - 1 )— b^ (3.2) with p -1 being the inverse of the blurring function. It is clear that a solution to Equation (3.2) requires knowledge of p -1 , b, and n. Such knowledge is often incomplete or totally lacking, and the approaches taken to the recovery of s reflect then, in general, various ways in dealing with these deficits. Thus, the discussion and evaluation of BCMs are based mainly on a classification of their treatment, whether implicit or explicit, of the incomplete information. The main purpose of this classification is to impose some order on the collection of methods to facilitate discussion. It has as added benefit an indication of the relative extent of information required for baseline correction, but does not reflect on the computational complexity of the various methods. The missing or incomplete information can often be determined either explicitly such as the characterization of the blurring function p, implicitly by way of assumptions and/or estimates, or be ignored altogether. For example, noise is often assumed to be independent with a zero-mean normal distribution [2, 68]. The standard deviation, a n , of the noise distribution is then estimated from a signal-free subset of the observed data y. Likewise, the background b is assumed to consist of broad, slowly varying, correlated features [2, 62, 65-69] and an estimate of its intensity can be made from background-containing but signal-free subsets ofy. 54 3.3 Materials and Methods 3.3.1 Synthetic Spectra Standardized spectra consisting of 1001 channels were created to simulate, primarily, spectra obtained from the vibrational spectroscopies (Raman, IR, etc). However, results presented here may be applied to other data types with relatively narrow peaks (e.g. NMR) as well, but potentially not to those with broad peaks (e.g. UV). A cumulative normal function (sigmoid) was used as the baseline. The slope of the linear part of the baseline was set to one of four values: 0.0, 0.1, 0.3, or 1.0. In the latter case, the baseline amplitude (baseline maximum - baseline minimum) reached a maximum value of 1.0. Seven Lorentzian peaks were added to every baseline with peak centers at 340, 455, 532, 584, 618, 641, 656 channels, respectively. Hence, one part of the spectrum was congested but the other part not. Lorentzians had a full width at half maximum (FWHM) of 5.7 channels, except for the first peak (at position 340) that had a width of 10 channels. They were then convolved with a 5-channel standard deviation normal distribution to simulate the instrumental blurring function. All Lorentzian signals had the same height within a spectrum, but the heights were varied between spectra through 0.0, 0.05, 0.1, 1.0, and 10.0, respectively, hence effectively varying the signal-to-baseline ratio (SBR), i.e. signal strength (signal maximum - signal minimum) relative to baseline amplitude (baseline maximum - baseline minimum). Therefore, for every baseline with a given slope, 5 spectra with Lorentzians of different height were generated. Noise was added to each of the spectra above to give signal-to-noise ratios (SNR) of 2, 3, 5, 10, and 100. Since there were 4 baseline slopes, each with 5 intensities of peaks added, and each of these with 5 different levels of noise added, a total of 100 synthetic spectra were generated. (Both SBR and SNR, when used in the context of a spectrum, are based on the most intense signal in the spectrum; when no signal was present, no noise was added.) The different synthetic baselines used are shown in Figure 3.1(a) and a sample convoluted synthetic signal, 55 before superpositioning on these baselines, in Figure 3.1(b). Example synthetic spectra with various values of SNR, SBR and slope are show in Figure 3.1(c, d). 3.3.2 Computation Equipment The automated BCMs investigated here were implemented in HiQ script (National Instruments Corporation) running under Windows 98 (Microsoft, Redmond, WA) on a 1 GHz AMD Athlon platform (AMD, Sunnyvale, CA). Manual baseline removals were done using the Grams/32® (Galactic Industries Corporation) software. Points corresponding to baseline intensity were chosen at the edges of a given spectrum as well as between signal peaks. These points were chosen and interpolated at the discretion of the volunteers. Baselines were linearly or quadratically interpolated (depending on the volunteer) between the chosen points, hence producing a baseline spectrum. 56 30 20 -10 2.2 1.0  — Mid-point slope = 0 — Mid-point slope = 0.1 — Mid-point slope = 0.3 Mid-point slope = 1.0 uncongested --V. congested 0.8 1.8 1 4 1.0 J 400 0.0 600^8000 400 Channel Index 800 200 Channel Index (a)^ (b) 0 ^ 200 ^ 400^600 ^ 800 ^ 1000 ^ 200 ^ 400^600 ^ 800 ^ 1000 Channel Index Channel Index (c) (d) Figure 3.1 Standardized data and example spectra. (a) Baselines of midpoint slope = 0 (black), 0.1 (dark gray), 0.3 (gray), and 1.0 (light gray) used in the generation of synthetic spectra. Example synthetic spectra consisting of seven Lorentzian peaks (b) free of baseline and noise; (c) with SNR = 3, SBR = 10, and slope = 0.1; and (d) with SNR = 10, SBR = 0.1, and slope = 1.0. Spectrum (b) has been divided into uncongested and congested regions. 57 3.3.3 Figures of Merit Due to the number of methods tested and the number of spectra subjected to each method, reporting multiple figures of merit (FOM) for each trial, as well as the representation of the results of baseline estimation, became problematic. Hence, bulk FOM were chosen which best represent the trends in the baseline estimation capabilities of each method. To this end, the correlation coefficient was calculated between estimated and given baselines for both the congested and uncongested regions of each spectrum, as well as for the total spectrum. Furthermore the chi squared (X2) value between the baseline estimate and the actual baseline was calculated for the entire spectrum. This gives an indication of offsets between estimated and actual baselines. Furthermore, for each method representative examples of baseline estimates as well as baseline-removed spectra are presented. This allows for a qualitative visual inspection of the types of distortions imposed by the given method on the baseline-removed spectra. Both the correlation coefficient and X2 were chosen as FOM for the following reason. Although the x2 statistic is often used as a FOM, it represents a bulk value and hence the nature of the error cannot be inferred. A result with a large x2 value may nevertheless have a very high correlation coefficient indicating that the estimated and actual baselines are separated by a constant offset. In contrast, a smaller x2 value could be associated with a small correlation coefficient indicative of pronounced distortions in the estimated baseline despite average values similar to that of the actual baseline. For every method discussed, a figure containing bar graphs for these FOM (correlation coefficients and x2 values) is given. Each bar graph reflects differences between theoretical and estimated baselines on individual spectra with spectral attributes as shown on the abscissa. The bar graph panels, from top to bottom, show the correlation coefficients for uncongested, congested, and total ("complete") spectral regions, and the x2 values for the total spectral region, respectively. The abscissa shows the attributes for each of the 100 individual spectra with SNR nested within SBR nested within slope. For example, the rightmost bar in the top panel of Figure 3.2 shows the correlation coefficient for the uncongested spectral region (see Figure 3.1) for a spectrum with SNR = 100, SBR = 10, and slope = 1.0. 58 To gain insight into the overall ability of each method to estimate the baseline, regardless of SBR or SNR, a mean correlation coefficient and X2 over all SNR and SBR (i.e. 100 spectra), as well as the standard deviation of this mean, was calculated for each method over the entire spectrum. Finally, the mean time and the standard deviation of this mean, required by each method to perform the baseline estimation for a given spectrum, were calculated. 3.4 Selected Baseline Correction Techniques Only one baseline correction method belonging to each of the categories in Table 3.1 was selected due to space consideration; for the evaluation of all the methods listed in Table 3.1, the reader is directed to the feature article by Schulze et al. [70]. As mentioned before, these categories reflect the amount and type of prior, assumed, or additional knowledge available. Since it is not practical to include all methods, some have been omitted. Omission does not reflect on the value of these methods and their particular approaches or features. Methods not described here include those based on principal component analysis or singular value decomposition [71-74] and Bayesian methods [66, 75-79]. Also not included were methods primarily aimed at mixture resolution and calibration [11, 22, 27, 30, 36, 37, 40, 44, 48, 49, 68, 78-84]. The latter methods often require information about the signal s and involve the resolution of background and signal components through modeling. Table 3.1 Classification of baseline removal methods based on the type of required information. Class Method Methods requiring no explicit knowledge of p, b, or n Noise median method (NMM) First derivative method (FDM) Methods requiring estimates of b Artificial neural networks (ANN) Threshold-based classification (TBC) Signal removal methods (SRM) Composite baseline method (CBM) Spectral shift methods (SSM) Methods requiring estimates of b and n Manual methods (MM) Methods requiring use of p, b, and n Maximum entropy method (MEM) Methods requiring information about - frequency Fourier transform method (FTM) Wavelet transform method (WTM) 59 3.4.1 Methods Requiring No Explicit Knowledge of p, b, or n: Noise Median Method 3.4.1.1 Background Introduced by Friedrichs in 1995 to remove baseline distortion in NMR spectra [85], this method estimates the baseline as the median value in a moving window. If the median is based on extrema in the window, and if the window is suitably large, the median is not unduly affected by signal peaks. This method could be said to have antecedents in minimum search and interpolation methods based on the identification of local and global minima in different successive segments of a spectrum and in methods based on the means in different successive segments of a spectrum [7, 16, 17, 19, 21, 24, 31, 45, 47]. Lewis and Chatwin [86] also used segment minima to estimate the position of the baseline. In later work, a median filter has been used by Keselbrener et al. [87]. 3.4.1.2 Theory The data are modeled here implicitly in the form y = (s * p) + (b* p)+ n^ (3.3) with the objective to estimate b *p ,--- b. Since the baseline itself is not modeled (e.g. with some specific function), but locally estimated, it is considered a 'model-free' method [85]. In this method, the baseline is estimated as the median value of all those points within a window of given size. When the points are rank ordered, the median value is the middle value of the set, implying that there are as many points greater than the median as there are smaller than the median. It is generally unimportant how much the values are greater (or smaller), as long as an equal number are greater and smaller than the median, respectively. For a suitably large window, the presence of a peak simply means that some of the numbers above the median are much greater (in value or intensity) than they would be in the presence of only noise. Under these conditions, there would be neither more nor fewer points above the median. The median 60 remains unaffected by the presence of a signal if the signal is relatively narrow. For wider signals, the effects of the signal can be attenuated by considering the median of all the extrema, rather than the median of all the points, in the window. Hence, in the presence of a signal, the baseline is defined by the requirement of equal representation of local extrema above and below the baseline. If the window is too narrow relative to the signal, this interpretation no longer holds, leading to errors in baseline determination, and a wider window has to be used (see also `Modes of Failure' below). Within a window of suitable size, which defines a local neighborhood, the baseline is approximated by the median value of the local extrema. A given element of y is an extremum if both of its neighbors are either strictly larger or strictly smaller than itself. The estimated baseline obtained by moving the window along the spectrum can then be smoothed by convolving it with a Gaussian function to remove sharp discontinuities where these exist. 3.4.1.3 Implementation A fixed window size of 400 points (approximately 3 times that of the congested spectral region peak) was used. A standard deviation of 5 was used for the Gaussian smoothing function applied to the estimated baseline to reduce sharp discontinuities in case any were present [85]. 3.4.1.4 Results and Discussion The quantitative results are shown in Figure 3.2(a) and comparative examples of baseline correction on two spectra (SNR 10, SBR 0.1, slope 0.1; SNR 100, SBR 10, slope 1.0) are shown in Figure 3.2(b). The results, consistent with the literature (see below), indicated that the method tends to have difficulties with congested spectral regions. Further difficulties are observed in high SBR spectra (i.e. spectra with pronounced signal). Nevertheless, it produced relatively good results and was also comparatively fast as shown in Figure 3.7. 61 1^1^1^1^1^1^1^1^1^1^1^11111 IIII^1^1^1^1^1^1^1^I^1 1^1^1^11 1 1111 ^1111^1^1 1 1^1^1^1 ^1111 1 1111^1^1^1^1^1 1 1 1 1 1 1^1 1^1^1111 ^1^1^1 1 I I I I' I I I 1 —^-I^I I titilmIllmitittliii111111mItlilltillimi milmtItitilittiltm milIttlItmilm itill I co N co tn O^0 8^8 co ul 0 '-e' 0 co^co ul N 0^0 .- .- '^'0 • co . c1,1 r, u, 0^0 .-C^8'a t4c m 1 u, 0 .- cLI 1^l,c m 1 . 0 °^'o o cn Ln 0 tn tn co . 0 8 co ■ A 0^0 8^'."8 • - . co^'''co 111^. 0 0 8 cn^en^cn LO^LII^N 0^0 0 8^8^8 cn^co t. 0^L 0 .- °0 0 0 8 an.^0^7^q.^0^- .^2^7^q^•.^0.- .^2^-^R^qo^d^,-^,- q^1St^-^q^q a^d^.- 0.0 0.1 0.3 1.0 SNR SBR Slope .:11c.) 1 00 0.50 s...._. ,-. a) 'c...3, 0.00 a) 0 zn c' 1.00 0.50 C...) '81) 0.00 0 g0 c.-) -0.50 '&1 a.) 1.00 0." C...) il fl ,•=' 0.50 0.00 104 10 2 1o° 10 -2 10-4 CMPIL 1040%, 0.10 0.00 040 , - 10^I^I^I ^1^1^1^1^1^1^i^I^I^I^I^11^siii 3_R/ 2— 200^400^600^800^1000 Channel Index Figure 3.2 Baseline removal using noise median method. (a) Correlation coefficient between theoretical and estimated baselines for the uncongested, congested, and total spectral region of the given spectra. Also shown is the x2 between theoretical and estimated baselines for the total spectral region. (b) (i) Baseline removed (gray trace) and initial input spectrum (less noise and true baseline, black trace) for SNR = 10, SBR = 0.1, and baseline slope = 0.1; (ii) estimated baseline (gray trace) and input baseline (black trace), SNR, SBR, and baseline slope as for (i); (iii) and (iv) as for (i) and (ii) with SNR = 100, SBR = 10, baseline slope = 1.0. 62 This method requires that a suitable window size be determined. A window size of at least twice the FWHM is recommended for uncongested spectral regions and a larger window for congested and low SNR regions. 3.4.1.5 Modes of Failure This method depends on the presence of noise and will fail if very little or no noise is present. Furthermore, if the window is too small relative to a peak or a congested region and is moved along until it reaches this peak or congested region, most of the extrema in the window would derive from the peak or congested peaks and the baseline estimate will become biased. If too wide a window is used, changes in the baseline cannot be tracked accurately. Ideally, one should use as narrow a window as is consistent with adequate processing of broad peaks and/or congested spectral regions. Friedrichs describes a method of implementing an elastic window that is automatically adjusted to incorporate a suitable number of extrema [85]. Also, although such problems were not encountered, it is possible that the choice and width of the smoothing function could fail to remove sharp discontinuities in the estimated baselines. 3.4.1.6 Suitability of Method for Automation Given that few parameters have to be determined and that generally good results can be obtained for a range of SNR, SBR, and slope, this method is suitable for automation purposes. Incorporating the suggestions for an elastic window [85] may further enhance its suitability for automation. 3.4.2 Methods Requiring Explicit Estimates of b: Signal Removal Method 3.4.2.1 Background Signal removal methods (SRM) require the estimation of the baseline with some procedure such as polynomial fitting to the entire spectrum. Points more than a given distance 63 above this baseline estimate are stripped from the spectrum by replacing them with the values of the baseline estimate. After stripping, a new baseline estimate is generated and the procedure is often iterated until new baseline estimates change little or not at all between iterations. Because peaks are iteratively stripped from the spectra, they do not have to be explicitly located in advance. One of the first reported SRMs was that by Ralston and Wilcox in 1969 using a moving window smoothing procedure and 10 iterations [8]. Variants using polynomials, splines, moving averages or other smoothing filters, and wavelets followed [42, 88-90]. Other approaches, to a lesser or greater extent related to signal removal or peak stripping, have also been advanced [12, 20, 26, 28, 36, 40, 55, 66, 80, 91]. 3.4.2.2 Theory SRMs were further developed and refined in-house with the aim to effect automated baseline correction. No smoothing step is applied and the data are implicitly modeled as in Equation (3.3). A crude estimate of the baseline, b , is obtained from y using either a low-order polynomial fit, a spline fit, or a large window filter (e.g. moving average or Savitzky-Golay zeroth order filter), depending on the variant employed. A high threshold is then established based on the standard deviation a of y (e.g. 56), and the value of points above the threshold value from the estimated baseline, if any, are reduced by some pre-specified amount (e.g. 25% of its value) to yield a residual spectrum r. The value of the threshold is then reduced by some small amount (e.g. 1%) and the process iterated by obtaining a new b from r and reducing those values of r exceeding the corresponding value of b plus current threshold until a stopping criterion is reached. The stopping criterion can be based on either the SNR of the residual spectrum r (here defined as maximum of r — b /standard deviation of r — b) or on a predetermined cut-off value for the threshold. Where spline or moving averages are used to determine b , the splines are fitted through segments with initial lengths about 5-10% of the spectrum or with the window size set at about 5-10% of the spectrum, respectively. On successive iterations, the segment or the window is given a random component that can be biased 64 to gradually reduce or increase its size. The effect of this approach is to remove the convoluted signals from the original data y, leaving (b *p) + n from which b *p is obtained with the last polynomial fit, spline fit, or moving average operation on r. If, instead of the final estimate b based on the final r, the final r itself is subtracted from the initial spectrum y, noise is also removed from the recovered spectrum. Hence, initially r = (s *p) + (b *p) + n and finally r (b *p)+ n, thus b^b *p. 3.4.2.3 Implementation An initial window size of 51 channels was used, which was subsequently increased by 5% per iteration. In every channel, 25% of that portion of the peak extending above the current threshold was removed. The threshold was initially set to 5 times the standard deviation of the original spectrum. The threshold was adjusted to 90% of its previous value after each iteration. A stopping criteria of either SNR = 3 or a threshold level less than 0.05 was used, whichever was attained first. 3.4.2.4 Results and Discussion The quantitative results are shown in Figure 3.3(a) and comparative examples of baseline correction on two spectra (SNR 10, SBR 0.1, slope 0.1; SNR 100, SBR 10, slope 1.0) are shown in Figure 3.3(b) for the moving average variant. The results showed fairly good consistency under most conditions, but with some difficulties in noisy spectra and congested spectral regions. Computation times were comparatively moderately short as shown in Figure 3.7. 65 1 11 Ilk II 1Iii0i I 11111)31111111111111 IV 1 . 1111 111111 R N _,^,,C] . 0 0 . 0 0 R LO 0 0 V) til 0 0 V> O c,-,^R^,,^R 111^.^LO^lf) 0^0 0 0 8^8^8^8 v., In 0 E C,7 . 0 8 el^el .^In 0 0 8^0 In . 0 8 R), 0 (,) VI^V) In^.^ln 0 0 0 -0 8 -00^0^0 C) V/ 0 80 t+) . -8 oo o^7^q^•6^co^.- ino^7^q^•6^o^,-^,_ ein^.-^o^q 6^ci^,-^,_ o^8^,-^o^0 c;^6^ci^.- 0.0 0.1 0.3 1.0 CU C.) 1 00 0.50 0.00 1.00 CLCL)o '772 0.50 t5.1) 0.00 0 0 ct L., -0.50 1.00 0.50 E 0.00 4104 102 10° 10 2 1 0-4 SNR SBR Slope 0.12 0.08 0.04 0.00 -0.04 1.0 20 cC 1E^10 mit^1 1 1 1 1^Ii eti f Î 3 iv 2 200^400^600^800^1000 Channel Index Figure 3.3 Baseline removal using signal removal (moving average) method. (a) Correlation coefficient between theoretical and estimated baselines for the uncongested, congested, and total spectral region of the given spectra. Also shown is the x2 between theoretical and estimated baselines for the total spectral region. (b) (i) Baseline removed (gray trace) and initial input spectrum (less noise and true baseline, black trace) for SNR = 10, SBR = 0.1, and baseline slope = 0.1; (ii) estimated baseline (gray trace) and input baseline (black trace), SNR, SBR, and baseline slope as for (i); (iii) and (iv) as for (i) and (ii) with SNR = 100, SBR = 10, baseline slope = 1.0. oo 66 3.4.2.5 Modes of Failure This general approach requires the selection of initial window sizes, rates of change of these window sizes, and amounts of peaks in excess of current thresholds to remove from the spectra. Choosing too small an initial window size will result in more of the peaks being considered part of the baseline, while too large an initial window size will result in prominent parts of the baseline being considered as belonging to a peak and hence removed. A choice of initial window size of about three times that of the average FWHM of spectral peaks proved satisfactory. If a larger window is used, it is advisable to reduce the window size on subsequent iterations. Reducing peaks by larger amounts will speed up computation time but may reduce the accuracy with which the baseline is estimated. For low SNR spectra, outliers due to noise are removed from the residual spectrum r along with peaks, hence the baseline is generally underestimated but retains its shape. This problem can be addressed by adjusting the final baseline-corrected spectrum by a constant amount to eliminate the offset. 3.4.2.6 Suitability of Method for Automation This general approach requires relatively few parameters to be determined, is modestly fast, consistently good under most conditions, and therefore suitable for automation. Residual baseline artifacts could potentially be removed in a second pass of the baseline-corrected spectrum. 3.4.3 Methods Requiring Explicit Use of b and n: Manual Method 3.4.3.1 Background Manual methods (MM) consist of selecting manually or "by eye" a number of points from a spectrum deemed to lie on the baseline. The baseline is then interpolated between these points using linear, polynomial, or spline interpolations. The performance of the automated baseline removal methods is often assessed by comparison with either theoretical, baseline-free 67 spectra or with manual baseline removal methods (i.e. baseline chosen and removed by investigator). Hence, the manual baseline removal is often regarded as a "gold standard" against which the automated techniques are compared. Indeed, many automated methods such as NMM, threshold-based classification, composite baseline, and other methods related to them, such as minimum search and interpolation methods, could be seen as attempts to simulate these MMs. Several MM implementations, evaluations, and comparisons with new methods can be found in the literature [32, 34, 41, 56, 67, 89, 92-94]. 3.4.3.2 Theory In the global variant of this approach, a number of more-or-less evenly spaced points judged to be on the true baseline are identified. Line segments, a low-order polynomial, or splines are then fitted through these points to estimate the baseline. In the local variant, points judged to be on the baseline at the start and end of peaks are identified and a linear interpolation is made to estimate the baseline below spectral peaks. In both variants, the data are modeled implicitly as in Equation (3.3) and the selection of points judged to be on the true baseline involves some explicit accounting of baseline and noise characteristics even though these are not mathematically formulated. It is the latter factor that may vary from person to person and within the same person based on experience. Hence automated numerical methods have been pursued in an attempt to minimize the need for judgment calls and to improve reproducibility. Therefore, some results based on MMs are included to provide an indication of their comparative precision and accuracy as well as to provide a benchmark for other methods. 3.4.3.3 Implementation The 100 spectra were randomly assigned to 10 groups of 10 spectra each. Volunteers familiar with spectroscopic concepts were recruited to remove the baselines from one or more of these 10 sets of spectra so that, in total, triplicates of each set were available for analysis. The shapes of the individual baselines, number of simulated peaks, and the peak positions were not revealed to the volunteers prior to their spectral subtractions. 68 3.4.3.4 Results and Discussion The results were based on the means of the triplicate baseline estimates. The quantitative results are shown in Figure 3.4(a) and comparative examples of baseline correction on two spectra (SNR 10, SBR 0.1, slope 0.1; SNR 100, SBR 10, slope 1.0) are shown in Figure 3.4(b). Computation times were moderately long as shown in Figure 3.7, making it less than ideal when dealing with a large number of spectra. This method produced results that were, generally speaking, very accurate and precise, confirming its general utility and eminent position as a standard of baseline removal. The baseline estimates by volunteers were adversely affected by low SNR, high SBR, highly sloping baselines (implying regions of strong curvature), and spectral congestion. Additionally, a bias effect was observed due to experience level. This experience bias suggested that both experienced and inexperienced volunteers tend to disagree on the choice of the baseline when it is "buried" in the noise. When the experience bias is removed, the accuracy and precision of the baseline determination in all four factors (SNR, SBR, slope, congestion) are greatly increased. 69 104 102 100 10-2 10-4 15 0 0) U C C C") U 114 Fo' - o a 0 C_J 0 1.00 0.50 0.00 -0.50 1.00 0.50 0.00 -0.50 1.00 0.50 0.00 -0.50 I^I^I^I^I^I^I^I^I^1^t 1^I^1^1^1 1^1 iv All 'ill _.... il ill 1111 I'll I 11^11 ' III I^1111111 illi ( 11 I 1111 1111 1111 1111 _ 11111111161110111 111111111 1111 1401111111^I - 1111^1111^11111_111111111D11111111 ■1111 11111^...^. Ill 11111111 111 111I l^I^I 1 1111^i111111111 1 1111 ̂ III III^i )IIE 1111 II I i I^I 1111 I I^11111)11 III^I I III ill I 1 . 11 11 1^11 1111 11111 -11114111111111011101 1 1111011^1111101001110111111111111101110111 1111010111110111^11111 !!! II!! &I t !Mill ^IIc,i,^1L.O 8 1.o  ̂ -8 .o -8 .o -8 4,.o -E IJ ‘7.n^. O^o^o °^7.'^- '.1 o -8 0 o -E lc 1,.1 u-,^. 0^0 ^O -8^-8^0 i^f 'u1-, o 8 . O E ' 'Ld o -8 el o -8 in  cUl-, o f 'kg",^ N o 8 in o 8 .^2^,-^q^R6^o^.-^,S, c?^8^,-^0^•o •^d 0^2^',..^c!^ed - d^.- R^,-^c?^q• o^.-^,2 0.0 0.1 0.3 1.0 SNR SBR Slope 0 . 1 0 . 0 -0. 1 1 . 1 C 1. 0 • ^1 0 8 4 0 0 ^ 200 ^ 400^600 ^ 800 ^ 1000 Channel Index Figure 3.4 Baseline removal using manual method. (a) Correlation coefficient between theoretical and estimated baselines for the uncongested, congested, and total spectral region of the given spectra. Also shown is the x2 between theoretical and estimated baselines for the total spectral region. (b) (i) Baseline removed (gray trace) and initial input spectrum (less noise and true baseline, black trace) for SNR = 10, SBR = 0.1, and baseline slope = 0.1; (ii) estimated baseline (gray trace) and input baseline (black trace), SNR, SBR, and baseline slope as for (i); (iii) and (iv) as for (i) and (ii) with SNR = 100, SBR = 10, baseline slope = 1.0. 70 3.4.3.5 Modes of Failure Despite the favourable results described above, the accuracy and precision of baseline determination of spectra with varying SNR was affected by varying SBR, slope, congestion, and user experience. Spectra with high SBR resulted in an overestimation and underestimation of the baseline by approximately 30% in the congested and uncongested regions, respectively, while overall, the peaks in the both low and high SBR spectra were generally overestimated. Congested spectral regions also adversely affected both accuracy and precision. In general, subjects overestimated the baseline in regions of highly congested spectral features by about 20%. The over- and underestimation of the baseline in highly sloped and congested spectra are improved with user experience. 3.4.3.6 Suitability of Method for Automation Not applicable. 3.4.4 Methods Requiring Explicit Use of p, b and n: Maximum Entropy Method 3.4.4.1 Background The maximum entropy method (MEM) is a way to reconstruct a spectrum in such a manner that it is as smooth as possible while its convolution with the instrumental blurring function conforms as closely as possible to the original spectrum. The smoothness of the reconstruction, governed by its entropy, and the closeness of its convolution with the blurring function to the original spectrum, governed by the X2 value, are traded off against each other with a regularization parameter X to find the optimum reconstruction. The MEM variant discussed here implements separate reconstructions of background and signal peaks [64]. The background reconstruction is convolved with a blurring function several times wider than the blurring function used for the signal peaks to ensure that its features are broad and slowly-varying. The sum of the background and signal reconstructions is then compared to the original spectrum for 71 fidelity. Possibly the first use of this method for baseline removal was that by Durman and Wood in 1988 [95]. 3.4.4.2 Theory Durman and Wood [95] modeled the data as in Equation (3.1) and Phillips and Hamilton [69] in a modified version of Equation (3.3): y=(s* p)+(b*q)+n,^ (3.4) where q is a broad blurring function with a width of 5 times or more than that of the blurring function p. The signal s and baseline b are then separately reconstructed in such a way that their entropy is maximized (negative entropy minimized) and that, after convolving with the blurring function(s), the added signal and baseline .9 conform to the measured data y within acceptable limits. MEM involves a trade-off between the smoothness constraints imposed by the entropy function, S, S =^(-9 + tb ilnN \^(3.5) i=1^ E where E is an arbitrarily small constant, and the demand that the reconstructed spectrum .9' agree with the observed data within k \2z2 = 1,‘ 2 ^< k a where k is the number of points in the spectrum and a is based on n. Computation then proceeds by minimization of a weighted sum of Equations (3.5) and (3.6) (x 2 + AS) where the weight (Lagrange multiplier or regularization parameter) 0 < A < Go also has to be optimized. It should be noted that several formulations of entropy functions exist [96, 97]. (3.6) =1 72 3.4.4.3 Implementation This method was implemented with a 3-point entropy window analogous to Greek et al. [98], 15-point and 75-point Gaussian blurring functions for the signal and background components of the spectra, respectively, and a A, (background) of 1000 and a A, (signal) of 0.01. The Levenberg-Marquardt method [99] was used for optimization. 3.4.4.4 Results and Discussion The quantitative results are shown in Figure 3.5(a) and comparative examples of baseline correction on two spectra (SNR 5, SBR 10, slope 1.0; SNR 100, SBR 10, slope 1.0) are shown in Figure 3.5(b). The results on those spectra processed indicated that baseline estimation was difficult in all regions, especially in more congested spectral regions. Computation times were very long as shown in Figure 3.7 and precluded effective optimization. This method requires that the instrumental blurring function be determined or known. It is a very slow method due to its computational load and it made optimization of the various parameters impractical. It is therefore likely that better results could be achieved as computation systems improve. An important consideration is that the potential for deconvolution and denoising is offered by this method along with baseline removal [98]. 73 0 0 0 ^ 0 0.0 0.1 0.3 1 .0 SNR SBR Slope 0 0 0 wilooliol VIII ^'u,nr7,1114.1( 1,1 i^"4i 9 .2^„ 2 8 8 8 8 8 8 5 1:41 0 0^0 0 0 8 8 8 8 E 8 8 8 8 0 V.• 1.00 0.50 ?°. 0.00 .:2, -0.50 0 L72. a..) 0 1, 1.00 0.50 ao 0.00 1-. -0.50 1.00 0.50 ao 0.00 U -0.50 104 102 1 100 10 -2 10 -4 8 V 600200 ^ 400 800 ^ 1000 Channel Index Figure 3.5 Baseline removal using maximum entropy regularization. (a) Correlation coefficient between theoretical and estimated baselines for the uncongested, congested, and total spectral region of the given spectra. Also shown is the x2 between theoretical and estimated baselines for the total spectral region. (b) (i) Baseline removed (gray trace) and initial input spectrum (less noise and true baseline, black trace) for SNR = 5, SBR = 10, and baseline slope = 1.0; (ii) estimated baseline (gray trace) and input baseline (black trace), SNR, SBR, and baseline slope as for (i); (iii) and (iv) as for (i) and (ii) with SNR = 100, SBR = 10, baseline slope = 1.0. Note: due to the computational demands of this method, only six spectra were processed. 74 3.4.4.5 Modes of Failure Congested spectra hinder optimization and convergence failure may occur. Since the same general approach is used to estimate both the signal and baseline components of the data, the baselines under peaks are generally overestimated. This difficulty can be countered by using a broader blurring function for the background, but at the cost of losing background estimation accuracy. The selection of the background blurring function size relative to the peak blurring function size (a factor of 5 was used here) will therefore affect the outcome of this method. Selecting too large a stopping tolerance will lead to early termination of the procedure and poor results while too low a stopping tolerance will increase processing times and possibly lead to convergence failure. The choice of the optimization method (e.g. Levenberg-Marquardt, conjugate gradient descent, etc.) and the selection or determination of regularization parameters (Lagrange multipliers) may also affect the accuracy of the results and the time required to achieve convergence. 3.4.4.6 Suitability of Method for Automation Due to its computational load and the difficulty in determining appropriate parameters the MEM is not very attractive, as it currently stands, for automation. However, this assessment may have to be revised in the future as computation equipment becomes more powerful. 3.4.5 Methods Requiring Frequency Information: Wavelet Transform Method 3.4.5.1 Background When representing a given function as a sum of sinusoids such as in the Fourier Transform Method (FTM), quite a large number of sines and cosines of increasing frequency may have to be used to cancel each other out in featureless regions of the spectrum. Instead, a given function could be represented as a sum of wavelets. Wavelets are localized in space and also in frequency [97]. The spatial localization of wavelets is a unique feature not found in their 75 sinusoidal counterparts (sine, cosine). This makes wavelets more suited to represent a large class of signals. For example, spectra can be represented with far fewer wavelets than sinusoids. Wavelets also allow for a representation of spectra at different resolution scales [96] and this is of interest here since it offers the potential to discriminate between high frequency noise, low frequency baseline, and mid-frequency signals at these different levels of resolution. Wavelets have been used to estimate baselines in capillary electrophoresis data by iterative peak stripping and wavelet application [90]. This method is analogous to the SRMs, but employs wavelet transforms instead of moving averages, polynomials, or splines to estimate the baseline. The effects of wavelet shrinking schemes on estimating baselines in synthetic and real data at different resolution scales have also been investigated [100, 101]. 3.4.5.2 Theory Wavelet decomposition proceeds by the recursive application of a matrix of wavelet filter coefficients to a spectrum modeled as in Equation (3.3). The spectrum must have 2 n resolution elements (zero padding of data may be necessary). The wavelet filter coefficients constitute quadrature mirror filters, i.e. the first filter passing smooth data and the second passing rough or detail data, as shown in the equation representing the Daubechies fourth order wavelet transform, Co^C 1^C2^C3 C3 —C2 C 1 —Co Co^C1^C2^C3 C3 —C2 C1 —Co S i d1 S2 d2 (3.7) • • • CO C3 —C2 C2^C3 C1 —Co Co^Cl C3 —C2 5n/2 dn12_ C2^C3 Cl  — Co YO y1 Y2 _Yn_ 76 In Equation (3.7), the wavelet smooth-pass filter coefficients are shown in row 1 and that of the rough-pass filter in row 2 of the wavelet filter coefficient matrix (c matrix). The requirements of these filter coefficients and how to derive them can be found elsewhere [97]. The spectral data are denoted by the vector y and the smooth and detail wavelet coefficients by the s i and di coefficients, respectively. A spectrum vector y processed in this way gives rise to a vector of interleaved wavelet coefficients that alternately specify the smooth and detail (i.e. non-smooth) components of the original spectrum as shown in Equation (3.7). If these wavelet coefficients are sorted into smooth and detail sections, the smooth coefficients would amount to one-half the size of the original spectrum vector. Recursivity is effected by applying the (adjusted) wavelet filter coefficient matrix to the smooth wavelet coefficients only. This results in another vector with interleaved 'smooth' and 'detail' coefficients, with the new number of smooth coefficients now only one quarter of the size of the original spectrum vector. This procedure is repeated until only two smooth (i.e. the smoothest) wavelet coefficients remain. At each step, the detail wavelet coefficients are retained and added to those from previous steps. Using the inverse wavelet filter coefficient matrix, the original spectrum then can be reconstructed from its smooth and detail wavelet coefficients. Note that the coefficients of the wavelet itself are called the wavelet filter coefficients while the elements of the filtered 'spectra' are called wavelet coefficients. As described above, the wavelet filter coefficients are repeatedly applied to the (smooth) wavelet coefficients until only two remain. The last two wavelet coefficients are the mother function coefficients. At some point during the iterated application of the wavelet filter coefficients, the resulting smooth coefficients may closely approximate the baseline. If the process is terminated at this level of resolution, the baseline could then be estimated from the smooth wavelet coefficients. 77 3.4.5.3 Implementation The wavelet transforms using the Daubechies wavelets [102] (i.e. wavelet filter coefficients) of order 4 were implemented and the process was terminated at resolution 2 6 to obtain the baseline estimate. The baselines were then estimated by linear interpolation from the smooth coefficients obtained at this resolution level which were fewer (32) than the number of points in the padded spectra (2048). 3.4.5.4 Results and Discussion The quantitative results are shown in Figure 3.6(a) and comparative examples of baseline correction on two spectra (SNR 10, SBR 0.1, slope 0.1; SNR 100, SBR 10, slope 1.0) are shown in Figure 3.6(b). The results indicated that baseline estimation was less successful with high SNR spectra and in congested regions of spectra. Computation times were comparatively moderately long as shown in Figure 3.7. The drawbacks of this method are the need to establish the appropriate resolution level to obtain the best baseline estimate, the need to pad the data, and the need to obtain extraneous information such as a suitable set of wavelets (i.e. filter coefficients). Padding generally increases the computation time. This method has the advantage that the first few applications of the wavelet filter coefficients will filter out high frequency noise. In addition, thresholding the wavelet coefficients permits both denoising and baseline estimation to be accomplished. 78 1.00 0.50 0.00 -0.50 1.00 0.50 0.00 -0.50 1.00 0.50 0.00 -0.50 104 102 100 10 - 2 10'4 -0.1 • ^ 1 00 0 8 4 0 t^ iv I 11111M111111111111111111111111111110111 -111111111 MO^MOO MN ' IIMINITIMI111111111111111111F iiiilliiiimliilIlmmlli' ' ME HMI-- 711111 -------- HEIR 1111111111111111111111 ,„,,„,,,,„,,„„ „, „„,„„,„„,,„, ,„ 1111111111111111111illiii '0 8 N 0 8 N^I^N N^N^N '0^'0 8^8: N^c,^m^m "0 '0 0 8 8- - 0 8 m 0 -: CO^m^m^V) "-'0^'0^0^0 -8 -8^o^8 - CO : V)^V)^V) "01 0 0 ^8 ^8^8 ----- 0) 0 8 VI U-'0 .^ill^'-^q^•a^0.- ino^'''^0^•a^6^, 'n^,^o^00^• •^oo 0^..- o^0^-^o^0• 0^• •^ci o^o^.- 0.0 0.1 0.3 1.0 SNR SBR Slope 0.1 0.0 0^200^400^600^800^1000 Channel Index Figure 3.6 Baseline removal using wavelet transforms. (a) Correlation coefficient between theoretical and estimated baselines for the uncongested, congested, and total spectral region of the given spectra. Also shown is the x2 between theoretical and estimated baselines for the total spectral region. (b) (i) Baseline removed (gray trace) and initial input spectrum (less noise and true baseline, black trace) for SNR = 10, SBR = 0.1, and baseline slope = 0.1; (ii) estimated baseline (gray trace) and input baseline (black trace), SNR, SBR, and baseline slope as for (i); (iii) and (iv) as for (i) and (ii) with SNR = 100, SBR = 10, baseline slope = 1.0. 79 3.4.5.5 Modes of Failure Inappropriate wavelet and resolution level choices are detrimental to baseline estimation. Some wavelets may be better suitable for some types of spectra; however, only the Daubechies wavelets were implemented here. The selection of resolution level involves a trade-off between the degree to which the curvature of the baseline could be approximated and the degree to which residual effects of peaks, especially wider and/or overlapping peaks, can be countered. Finally, the selection of thresholds may affect the results. 3.4.5.6 Suitability of Method for Automation This method has further potential for automated implementation (i.e. different from that related to the SRMs [90]). Difficulties are the mentioned selection of a suitable wavelet, automation of the stopping criterion (i.e. the proper level of resolution to represent the baseline well), and thresholding for the wavelet coefficients. 80 1 .0 0.8 0.6 0.4 0.2 105 1 0 -1 .•••••■ c./7 1 06 10-2 a) E o tztEe.,^1,•-^a.)•,- ,..).^c)- ..0 CL-,-,I^CA^0 ao - 0 a) a. C Ca a) -cat C Ca Figure 3.7 Global comparisons for baseline estimation methods based on all spectra. (Top) Mean correlation coefficient and variance. (Middle) Mean )(2 and standard deviation. Note log scale. (Bottom) Mean time and standard deviation to perform one baseline removal calculation. Note log scale. For the signal removal methods: m, p, and s refer to moving average, polynomial, and spline fit, respectively. Note also that this figure includes results from methods that were evaluated and are included here for completeness, but are not explicitly described in the text due to space consideration. 81 3.5 Conclusion A number of different baseline correction methods, their specific approaches to the problem of baseline removal, their implementations, results, advantages and disadvantages, and suitability for automation have been presented. The overall measures of accuracy and efficiency (i.e. processing speed) of the various techniques have also been compared. The evaluation of the various BCMs were critical to determine which method was best suited for the baseline correction of the Raman spectra of triterpenoid mixtures and isolated Prunus laurocerasus (Laurel cherry) leaf cuticles presented in Chapters 4 and 5. Aside from the desired accuracy and efficiency of these techniques, the level of computational demand was included in the criteria for the selection of the BCM to be used for the Raman spectra. Taking all this into consideration, the noise median, signal removal method, and manual method were the top three choices (Figures 3.2, 3.3, 3.4, and 3.7). However, due to the slow processing speed, labour- intensiveness, and lack of automation for the manual method compared to the other two methods, it was decided that this baseline correction method would not be practical when dealing with a large data set. Additional factors were considered when choosing between the noise median and the signal removal methods: 1) the ability of a baseline correction method to consistently provide an accurate estimate of the baseline given the large variability in SNR, SBR, and baseline slopes and shapes; and 2) the reproducibility of the BCM to estimate a baseline with high precision when performing baseline correction repeatedly on the same input data. In the end, the signal removal method was ultimately chosen for the spectral data processing used in Chapters 4 and 5. 82 3.6 References [1] Tu AT. Raman Spectroscopy in Biology: Principles and Applications; Wiley- Interscience: New York, 1982. [2] Ruckstuhl AF, Jacobson MP, Field RW, and Dodd JA. Journal of Quantitative Spectroscopy & Radiative Transfer 2001, 68, 179-193. [3] Wilson JD and McInnes CAJ. Journal of Chromatography 1965, 19, 486-494. [4] Baumann F and Tao F. Journal of Gas Chromatography 1967, 5, 621-626. [5] Deighton MO. IEEE Transactions on Nuclear Science 1969, NS16, 68-75. [6] Hager RN and Anderson RC. Journal of the Optical Society of America 1970, 60, 1444- 1449. [7] Fujii I, Inouye T, Muto H, and Onodera K. Analyst 1969, 94, 189-197. [8] Ralston HR and Wilcox GE. Proceedings of the international conference on modern trends in activation analysis II Gaithersburg, MD, 1968, National Bureau of Standards, Washington, D.C. [9] Goldberg B. Journal of Chromatographic Science 1971, 9, 287-292. [10] Robertson A, Prestwich WV, and Kennett TJ. Nuclear Instruments & Methods 1972, 100, 317-324. [11] Vandeginste BGM and Degalan L. Analytical Chemistry 1975, 47, 2124-2132. [12] Kawarasaki Y. Nuclear Instruments & Methods 1976, 133, 335-340. [13] Gans P and Gill JB. Applied Spectroscopy 1977, 31, 451-455. [14] Pearson GA. Journal of Magnetic Resonance 1977, 27, 265-272. [15] Atakan AK, Blass WE, and Jennings DE. Applied Spectroscopy 1980, 34, 369-372. [16] Hippe Z, Bierowska A, and Pietryga T. Analytica Chimica Acta-Computer Techniques and Optimization 1980, 4, 279-290. [17] Woerlee EFG and Mol JC. Journal of Chromatographic Science 1980, 18, 258-266. [18] Dietrich W and Gerhards R. Journal of Magnetic Resonance 1981, 44, 229-237. [19] Kipiniak W. Journal of Chromatographic Science 1981, 19, 332-337. [20] Steenstrup S. Journal of Applied Crystallography 1981, 14, 226-229. [21] Westmeier W. Nuclear Instruments & Methods 1981, 180, 205-210. [22] Johansson GI. X-Ray Spectrometry 1982, 11, 194-200. 83 [23] Burgess DD and Tervo RJ. Nuclear Instruments & Methods in Physics Research 1983, 214, 431-434. [24] Tervo RJ, Kennett TJ, and Prestwich WV. Nuclear Instruments & Methods in Physics Research 1983, 216, 205-218. [25] Daubenfeld JM, Boubel JC, and Delpuech JJ. Journal of Magnetic Resonance 1985, 62, 195-208. [26] Klevit RE. Journal of Magnetic Resonance 1985, 62,551-555. [27] Osten DW and Kowalski BR. Analytical Chemistry 1985, 57, 908-917. [28] Glaser S and Kalbitzer HR. Journal of Magnetic Resonance 1986, 68, 350-354. [29] Otting G, Widmer H, Wagner G, and Wuthrich K. Journal of Magnetic Resonance 1986, 66, 187-193. [30] Spaink HA, Lub TT, Otjes RP, and Smit HC. Analytica Chimica Acta 1986, 183, 141- 154. [31] Thomsen KN, Pedersen JN, and Pind N. Analytica Chimica Acta 1986, 184, 133-142. [32] Barsukov IL and Arseniev AS. Journal of Magnetic Resonance 1987, 73, 148-149. [33] Cameron DG and Moffatt DJ. Applied Spectroscopy 1987, 41, 539-544. [34] Mann CK and Vickers TJ. Applied Spectroscopy 1987, 41, 427-430. [35] Marion D and Bax A. Journal of Magnetic Resonance 1988, 79, 352-356. [36] Ryan CG, Clayton E, Griffin WL, Sie SH, and Cousens DR. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 1988, 34, 396-402. [37] Tougaard S. Surface and Interface Analysis 1988, 11, 453-472. [38] Marion D and Bax A. Journal of Magnetic Resonance 1989, 83, 205-211. [39] Marion D, Ikura M, and Bax A. Journal of Magnetic Resonance 1989, 84, 425-430. [40] Maxwell JA, Campbell JL, and Teesdale WJ. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 1989, 43, 218-230. [41] Zolnai Z, Macura S, and Markley JL. Journal of Magnetic Resonance 1988, 80, 60-70. [42] Grol3wendt B. Nuclear Instruments & Methods 1971, 93, 461-472. [43] Hegedus LL and Petersen EE. Journal of Chromatographic Science 1971, 9,551-553. [44] Shirley DA. Physical Review B 1972, 5, 4709-4714. [45] Kennett TJ, Prestwich WV, and Tervo RJ. Nuclear Instruments & Methods in Physics Research 1981, 190, 313-323. 84 [46] Froning JN, Olson MD, and Froelicher VF. Journal of Electrocardiology 1988, 21, S 149- S157. [47] Caesar F and Klier M. Chromatographia 1974, 7, 526-529. [48] Sherwood PMA in Practical surface analysis: by Auger and X-ray photoelectron spectroscopy; Briggs D and Seah MP, Eds.; John Wiley and Sons: New York, 1983, pp 445-475. [49] Van Espen P, Janssens K, and Nobels J. Chemometrics and Intelligent Laboratory Systems 1986, /, 109-114. [50] O'Haver TC and Green GL. Analytical Chemistry 1976, 48, 312-318. [51] Maddams WF. Applied Spectroscopy 1980, 34, 245-267. [52] Anderson DJ and Walters RR. Journal of Chromatographic Science 1984, 22, 353-359. [53] Burgess DD. Nuclear Instruments & Methods in Physics Research Section a- Accelerators Spectrometers Detectors and Associated Equipment 1984, 221, 593-599. [54] Ling YC, Vickers TJ, and Mann CK. Applied Spectroscopy 1985, 39, 463-470. [55] Vekemans B, Janssens K, Vincze L, Adams F, and Van Espen P. Spectrochimica Acta Part B-Atomic Spectroscopy 1995, 50, 149-169. [56] Vickers TJ, Wambles RE, and Maim CK. Applied Spectroscopy 2001, 55, 389-393. [57] Skoog D. Principles of Instrumental Analysis; Saunders: Philadelphia, 1985. [58] Epstein MS, Rains TC, and Ohaver TC. Applied Spectroscopy 1976, 30, 324-329. [59] Skogerboe RK, Lamothe PJ, Bastiaans GJ, Freeland SJ, and Coleman GN. Applied Spectroscopy 1976, 30, 495-500. [60] Hoult DI, Chen CN, Eden H, and Eden M. Journal of Magnetic Resonance 1983, 51, 110-117. [61] Henrichs PM, Hewitt JM, and Young RH. Journal of Magnetic Resonance 1986, 69, 460- 466. [62] Mosier-Boss PA, Lieberman SH, and Newbery R. Applied Spectroscopy 1995, 49, 630- 638. [63] Bell SEJ, Bourguignon ESO, and Dennis A. Analyst 1998, 123, 1729-1734. [64] Bell SEJ, Bourguignon ESO, Dennis AC, Fields JA, McGarvey JJ, and Seddon KR. Analytical Chemistry 2000, 72, 234-239. [65] Matousek P, Towrie M, and Parker AW. Journal of Raman Spectroscopy 2002, 33, 238- 242. 85 [66] Padayachee J, Prozesky V, von der Linden W, Nkwinika MS, and Dose V. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 1999, 150, 129-135. [67] Jirasek A, Schulze G, Yu MML, Blades MW, and Turner RFB. Applied Spectroscopy 2004, 58, 1488-1499. [68] Brown CD, Vega-Montoto L, and Wentzell PD. Applied Spectroscopy 2000, 54, 1055- 1068. [69] Phillips AJ and Hamilton PA. Analytical Chemistry 1996, 68, 4020-4025. [70] Schulze G, Jirasek A, Yu MML, Lim A, Turner RFB, and Blades MW. Applied Spectroscopy 2005, 59, 545-574. [71] Hasenoehrl EJ, Perkins JH, and Griffiths PR. Analytical Chemistry 1992, 64, 705-710. [72] Holland JK, Kemsley EK, and Wilson RH. Journal of the Science of Food and Agriculture 1997, 75, 391-400. [73] Balcerowska G and Siuda R. Vacuum 1999, 54, 195-199. [74] Iwata T and Koshoubu J. Applied Spectroscopy 1994, 48, 1453-1456. [75] von der Linden W, Dose V, Padayachee J, and Prozesky V. Physical Review E 1999, 59, 6527-6534. [76] von der Linden W, Dose V, and Fischer R. Maxent96: Proceedings of the maximum entropy entropy conference, 1996, Port Elizabeth, South Africa. [77] Fischer R and Dose V in Monographs of official statistics: Bayesian methods; George EI, Ed.; Eurostat: Luxembourg, 2001, pp 145-154. [78] Fischer R and Dose V in Bayesian Interference and Maximum Entropy Methods in Science and Engineering; Fry RL, Ed.; American Institute of Physics: Melville, 2002, pp 143-154. [79] Fischer R, Dose V, Hanson M, and von der Linden W in Bayesian Interference and Maximum Entropy Methods; Smith CR, Rychert JT and Erickkson GJ, Eds.; American Institute of Physics: Melville, 2001, pp 193-212. [80] Ryan CG, Cousens DR, Sie SH, Griffin WL, Suter GF, and Clayton E. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 1990, 47, 55-71. [81] Raso MA, Tortajada J, and Accion F. Computers & Chemistry 1991, 15, 29-35. 86 [82] Tokutaka H, Ishihara N, Nishimori K, Kishida S, and Isomoto K. Surface and Interface Analysis 1992, 18, 697-704. [83] Ryan CG. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 2001, 181, 170-179. [84] Ryan CG, van Achterbergh E, Yeats CJ, Drieberg SL, Mark G, McInnes BM, Win TT, Cripps G, and Suter GF. Nuclear Instruments & Methods in Physics Research Section B- Beam Interactions with Materials and Atoms 2002, 188, 18-27. [85] Friedrichs MS. Journal of Biomolecular Nmr 1995, 5, 147-153. [86] Lewis DM and Chatwin PC. Boundary-Layer Meteorology 1995, 72, 53-85. [87] Keselbrener L, Keselbrener M, and Akselrod S. Medical Engineering & Physics 1997, 19, 481-484. [88] Schulze HG. An In-vivo Resonance Raman Neuroprobe. PhD Thesis, 1996, The University of British Columbia. [89] Lieber CA and Mahadevan-Jansen A. Applied Spectroscopy 2003, 57, 1363-1367. [90] Wang Y and Mo JY. Chemical Journal on Internet 2003, 5, 052015. [91] Kneen MA and Annegarn HJ. Nuclear Instruments & Methods in Physics Research Section B-Beam Interactions with Materials and Atoms 1996, 109, 209-213. [92] Della Lunga G and Basosi R. Journal of Magnetic Resonance Series A 1995, 112, 102- 105. [93] Della Lunga G, Pogni R, and Basosi R. Journal of Magnetic Resonance Series A 1994, 108, 65-70. [94] Koradi R, Billeter M, Engeli M, Guntert P, and Wuthrich K. Journal of Magnetic Resonance 1998, 135 , 288-297. [95] Durman RP and Wood DJ in Analytical Applications of Spectroscopy; Creaser CS and Davies MC, Eds.; The Royal Society of Chemistry: London, 1988, pp 235-237. [96] Starck JL, Pantin E, and Murtagh F. Publications of the Astronomical Society of the Pacific 2002, 114,1051-1069. [97] Press WH, Teukolsky SA, Vetterling WT, and Flannery BP. Numerical Recipes in C; Cambridge University Press: Cambridge, 1994. [98] Greek LS, Schulze HG, Blades MW, Bree AV, Gorzalka BB, and Turner RFB. Applied Spectroscopy 1995, 49, 425-431. 87 [99] Marquardt DW. Journal of the Society of Industrial and Applied Mathematics 1963, 11, 431-441. [100] Chang W and Vidakovic B. Computational Statistics & Data Analysis 2002, 40, 317-328. [101] Amato U and Sapatinas T. Technical Report 243/02, 2002, Instituto per le Applicazioni del Calcolo, Naples, Italy. [102] Daubechies I. Communications on Pure and Applied Mathematics 1988, 41, 909-996. 88 CHAPTER 4 RAMAN MICROSPECTROSCOPIC ANALYSIS OF PLANT TRITERPENOIDS * Prior to working with biological samples, it is essential to evaluate the efficacy of the quantitative data analysis methods proposed for enhancing the accuracy and precision of the measured result. This chapter will demonstrate the suitability of the methods selected from those evaluated in Chapter 3 for the analysis of binary mixtures of varied known concentrations of triterpenoids typically found on plant surfaces. 4.1 Introduction The primary surfaces of higher plants are covered by a thin membrane known as the cuticle, essentially forming the interface between the above-ground environment and the epidermal cells of the plant. Plant cuticles consist of a polymeric cutin matrix and cuticular waxes, which can be found embedded within the cutin matrix (intracuticular wax), and as a layer covering the matrix (epicuticular wax) [1-3]. As mentioned in Chapter 1, the main components comprising the soluble waxes are very long chain aliphatic compounds. Some plants also have cuticular triterpenoids, which are pentacyclic compounds made up of six isoprene units. Amyrin isomers are commonly found on most plant leaves either in their native form or as oxidized derivatives, as seen in Prunus laurocerasus (Laurel cherry), where ursolic acid and oleanolic acid (derivatives of a- and 13-amyrin, respectively) are the two main triterpenoids found in the cuticular wax [4]. Studies of plant triterpenoids are important, as these compounds can be therapeutically useful. In fact, triterpenoids produced by plants have been found to possess anti-inflammatory, * A version of this chapter has been published. Yu MML, Schulze HG, Jetter R, Blades MW, and Turner RFB. Raman Microspectroscopic Analysis of Triterpenoids Found in Plant Cuticles. Applied Spectroscopy, 2007, 61(1), 32-37. 89 anti-ulcer, anti-hyperlipidemic, anti-tumor, and hepatoprotective properties [5-11]. For example, the a- and P-amyrins found in the resin of Protium heptaphyllum (Breu branco verdadeiro) are the major triterpenoid components that display anti-inflammatory and gastroprotective effects [12-15]. Similarly, oleanolic acid and ursolic acid provide hepatoprotection [16-18], in addition to being anti-inflammatory [19, 20], anti-bacterial [21, 22], antifungal [23, 24], insecticidal [25], anti-HIV [26, 27], diuretic [28], and antidiabetogenic [29]. The mechanisms by which these triterpenoids offer these pharmacological effects are discussed elsewhere [8]. Although this thesis will not delve further into the therapeutic effects of triterpenoids, it is still essential to examine how their distribution on plant surfaces may contribute to the different cuticular functions. Hence, Raman microspectroscopy has been utilized to study the chemical distribution of the cuticles. The ability of confocal Raman microspectroscopy to determine the spatial distribution of different chemical components within the epicuticular and intracuticular layers is particularly important in plants as it could potentially provide further insights in determining the function of the different wax regions in the cuticle. Different compounds can be identified using Raman spectroscopy by sharp peaks characteristic of the specific structural elements comprising the molecule, providing distinct "fingerprints" that can be used for detection and quantitative analysis. This technique was previously used to study the surface of the mango fruit, where the aliphatic components of the wax were found to be non-uniformly distributed within the epicuticular wax layers [30]. 4.2 Materials and Methods 4.2.1 Reference Samples The triterpenoid reference samples used were as follows: a-amyrin, 13-amyrin, ursolic acid, and oleanolic acid (Sigma-Aldrich, Oakville, ON, Canada). The very long chain aliphatic reference samples used were nonacosane, hentriacontane, octacosanol, triacontanol, and triacontanoic acid (Sigma-Aldrich) which are the most abundant very long chain aliphatic 90 components found in P. laurocerasus [4] and hence were chosen to represent the major wax compounds. All reference samples were powdered or crystalline solids. 4.2.2 Artificial Triterpenoid Mixture Preparation Two different methods were used to prepare the artificial triterpenoid mixtures. One method consisted of dissolving the triterpenoid compounds in chloroform (Sigma-Aldrich). The solution was then placed on a glass microscope slide or cover slip for recrystallization. The second method was carried out by preparing nujol mulls of the triterpenoid mixtures. For both methods, binary mixtures of different triterpenoids and/or very long chain aliphatic compounds with various relative weight ratios were prepared. The recrystallized mixtures and nujol mulls were placed on microscope slides and three replicate Raman spectra were collected at each of 15 random locations on each sample. 4.2.3 Raman Microspectroscopy Raman spectra were obtained using a confocal Raman microscope (Renishaw) equipped with a 514 nm Argon, 633 nm HeNe, or 785 nm diode laser source and a CCD detector. A 20x/0.40 numerical aperture (NA) or 50x/0.75 NA microscope objective (Leica Microsystems) was used to focus the laser onto a 15 or 5 gm spot on the sample, respectively. The laser power at the focal point was measured to be approximately 120-130 mW for the 20x/0.40 NA objective and 40-50 mW for the 50x/0.75 NA objective. Spectra were collected for 30 s over the range of 350-2000 cni l . 4.2.4 Raman Spectra Processing and Data Analysis By comparing the various baseline correction methods discussed in Chapter 3, it was decided that the spectra of all the samples were to be baseline corrected using the zeroth order Savitzky-Golay filter [31]. Briefly, this moving average technique involves the calculation of 91 the average of a small set of n data of a sequence, x, followed by the averages of successive shifts of the same interval resulting in a new sequence of smoothed data, s, ( 1 \ i+n-1 Si = -^x •^ (4.1) j=i To remove the peaks completely, a large window (i.e. large n) is chosen so that the resulting smoothed vector approximates the baseline of the spectrum. Where no peak is present, the smoothed vector will approximate the baseline fairly closely, but where a peak is present, the smoothed vector will tend to be elevated above the baseline because of the contribution of the peak to the moving average. That part of the peak (or a fraction thereof if one wishes to proceed more cautiously) that extends above the smoothed vector is then subtracted from the original spectrum, and the procedure is repeated. Because some part of the peak has now been subtracted, its influence on the smoothed vector will be reduced and the smoothed vector will be closer to the true baseline. The procedure is iterated until a stopping criterion is reached (i.e. a preset number of iterations reached, or when successive changes in the smoothed vector become very small). The smoothed vector obtained at this point is then subtracted from the original spectrum thus resulting in a baseline-free spectrum. The measured intensities of each spectrum were then normalized to the amplitude of the strongest peak prior to data analysis. The mixture spectra were modeled as linear combinations of the reference spectra [32, 33], and calculated using the equation M = DkAk^ (4.2) k where M is the mixture spectrum, ak is the coefficient of compound k, and Ak is the reference spectrum of compound k. The ratio of the coefficients of the different compounds in the binary mixture was compared to determine the relative concentration ratio of the two compounds. Quantitative results reported are mean values with standard deviations. 92 4.3 Results and Discussion 4.3.1 Laser Selection Most Raman work with biological samples is done in the NIR region as interference due to fluorescence is minimized in this region. When fluorescence is not an issue, it may be advantageous to perform the experiments in the visible region since the intensity of the Raman signal is greater at shorter wavelengths [34-36]. However, even if fluorescence is present to a significant extent in the Raman spectra in these wavelength regions, this problem can still be addressed by using appropriate baseline removal techniques such as those discussed in Chapter 3. Thus, the spectra of nonacosane, oleanolic acid and ursolic acid were acquired using both an Argon laser (514 nm) and a NIR laser (785 nm) to assess which wavelength would be optimal for this research. The instrumental parameters were all kept the same to avoid any ambiguities, and all spectra were normalized to the tallest peak for direct comparison. The region from 400 to 1500 cm -1 , the so-called "fingerprint region", provides the most spectral information required to identify and distinguish the different compounds. Hence, the criteria for choosing the wavelength to perform further Raman analysis were based on the spectral resolution, intensity, and signal-to-noise ratio (SNR) of the peaks in this region. As expected, the fluorescence background of all the spectra obtained using the Argon 514 nm laser was greater than that of the spectra obtained using the NIR 785 nm laser. When the fluorescence background was corrected for using the moving average technique, the spectra of all three compounds obtained using the Argon 514 nm and NIR 785 nm lasers were found to be similar (Figure 4.1). However, it can be seen in Figures 4.1 that despite correcting for the baseline, the lower signal-to-baseline ratio (SBR) of the raw spectra acquired with the Argon laser resulted in noisier baseline-corrected spectra compared to those obtained using the NIR laser. Taking this into consideration and keeping in mind that the NIR range allows for deeper penetration without damaging the sample [37, 38], it was decided that all further Raman work would be done using the NIR 785 nm laser. It should be noted that a third laser, a HeNe 633 nm laser, was also utilized to acquire the Raman spectra of the three compounds (data not shown). However, the Raman spectra of the 93 compounds obtained using the HeNe laser were not as high quality as those obtained using either the Argon 514 nm or the NIR 785 nm laser and hence, this laser was not considered for future Raman work. 94 1.1 (a) 0.9 0.7 - 0.3 - 0.5 - 0.1 - -0.1 0.5 - O 0.3 0.1 - -0.1 (c) 0.9 - 0.7 - 400^600^800^1000^1200^1400^1600^1800^2000 1.1 (b) 0.7 0.9 0.5 - 0.3 - 0.1 - -0.1 400^600^800^1000^1200^1400^1600^1800^2000 400^600^800^1000^1200^1400^1600^1800^2000 Raman shift (cm -1 ) Figure 4.1 A comparison of different laser excitation wavelengths for (a) nonacosane, (b) oleanolic acid, and (c) ursolic acid. The magenta and blue lines refer to the Argon 514 nm and NIR 785 nm lasers, respectively. 95 4.3.2 Reference Samples Since the reference samples were powdered or crystalline solids, several measurements were acquired at different points within the sample to minimize crystalline effects. Comparisons of these spectra displayed minimal variations and so the multiple measurements were averaged to obtain the basis spectra used for the mixture analysis. The triterpenoids used in this study are a-amyrin, 0-amyrin, ursolic acid, and oleanolic acid (Figure 4.2). The normalized Raman spectra of the wax reference samples, which include the very long chain aliphatic compounds as well as the triterpenoids, are shown in Figure 4.3. The spectra of all the triterpenoids are rich with peaks corresponding to CH vibrations from the extensive ring structures of the molecules. Peaks evident in Figure 4.3 include CH2, CH3, and CCH bends (1200-1400 cm I ), ring breathing stretches (700-750 cm' and 1000-1190 cm -1 ), CH3 rocking (900-950 cm 1 ), and CCC and CCO bends (440-480 cm -1 ) (Table 4.1) [39]. Although the only structural difference between a-amyrin and [3-amyrin is the location of one methyl group, their Raman spectra are unique and readily distinguishable. The position of the methyl group clearly affects vibrational modes at several wavenumbers, but mainly those associated with the ring breathing. Furthermore, a notable distinction between the a- and 0- amyrin spectra is evident within the broad feature at 1455 cm -1 . The spectrum of 13-amyrin has an extra peak at 1438 cm-I corresponding to the CH2 bend of C-29, which is not seen in the a-amyrin spectrum as the methyl group is attached at this carbon. An additional acid group on the basic amyrin structure results in a significant increase in intensity of the peak at 562 cm -I , as well as several slight shifts and variations in peak height at other wavenumbers. Similar to their alcohol counterparts, the two acid triterpenoids are distinguishable by their unique peaks, for example those at 745 cm -I and 1716 cm-I for oleanolic acid and ursolic acid, respectively. 96 OH Figure 4.2 The molecular structure of the four triterpenoids used in this mixture study: (a) a-amyrin, (b) 13-amyrin, (c) ursolic acid and (d) oleanolic acid. The spectra of the reference very long chain aliphatic compounds are all very similar to one another despite the differences in chain lengths and functional groups. Although Figure 4.3 only shows the spectral range of 350-2000 cm 1, it should be noted that there is a group of strong peaks in the 2500-3100 cm -1 region corresponding to CH stretches of the long aliphatic chains. The rest of the peaks correspond to CH2 bends (1300-1500 cm -1 ), CC stretches (1000-1200 cm-1 ), and CH2 rocking (890 cm -1 ) [40-42]. Although it is difficult to distinguish which specific very long chain aliphatic compounds are present in a sample and to determine their relative concentrations by Raman spectroscopy, the spectra can be used to discern the presence of a very long chain aliphatic component. 97 20001000 ,^► 500 ,^I^, 1500 coco^N.co o^co  co^co AAPA AL v:\I CTC.) co0 ^1,- OO (a) (b) (c) (d) co CO - (f) (g) (h) Raman shift (cm 1) Figure 4.3 Raman spectra of the reference triterpenoids and very long chain aliphatic compounds in the 350-2000 cm -I spectral region. (a) a-amyrin, (b) 13-amyrin, (c) ursolic acid, (d) oleanolic acid, (e) triacontanoic acid, (f) triacontanol, (g) octacosanol, (h) hentriacontane, and (i) nonacosane. The spectra have been normalized to the tallest peak. 98 Table 4.1 Raman shifts and peak assignment of vibrational modes of triacontanoic acid s , a-amyrin and oleanolic acid. Triacontanoic a-Amyrin Oleanolic Approximate assignment of acid (cm -1 ) (cm-) acid (cm -1 ) vibrational mode" 1658 (m) 1658 (m) v(C=C) 1482 (m, sh) 6(CH2), S(CH3) 1463 (m, sh) 1455 (ms) 1460 (ms) 5(CH2), S(CH3) 1440 (s) 1444 (ms, sh) 5(CH2), 5(CH3) 1434 (m, sh) 5(CH2), 5(CH3) 1417 (m) CH3 deformation 1385 (mw) 1386 (mw) 5(CH2), 5(CH3) 1364 (w, br) 8(CH2), 8(CH3) 1355 (mw) 5(CH2), 5(CH3) 1342 (m) 1338 (mw, br) 5(CH2), 8(CH3) 1316 (m) 1314 (m) 5(CH2), 6(CH3) 1295 (s) 1302 (m) 1307 (mw, sh) 5(CH2), 8(CH3), CH2 deformation 1282 (w) 1274 (mw, sh) 5(CCH) 1256 (mw) 1261 (m) 5(CCH) 1231 (m) 1228 (m, sh) 5(CCH) 1216 (m) 1218 (mw) 5(CCH) 1202 (m) 1203 (mw, sh) 6(CCH) 1171 (mw) 1189 (m) 1190 (m) v(CC) ring breathing v(CC) 1170 (m) 1163 (m, br) v(CC) ring breathing 1130 (s) 1119 (m, sh) 1142 (w) 1136 (mw) v(CC) v(CC) 1110 (w) 1106 (mw) 1096 (w) 1091 (mw) 1080 (w) 1061 (s) 1065 (w) v(CC) 1048 (mw, sh) 1035 (m) v(CO) 1030 (m) 1016 (w, sh) 1007 (mw) 1010 (w) 5(CC) ring breathing 988 (m) 988 (w) 958 (ms) 962 (vw) 959 (m, sh) 950 (m) p(CH3) 938 (mw, sh) 937 (mw) p(CH3) 926 (m) p(CH3) 914 (m) 922 (m) p(CH3) 99 (Table 4.1 Continued) Triacontanoic^a-Amyrin^Oleanolic acid (cm-1 )^(cm-1) acid (cm -1) Approximate assignment of vibrational mode" 904 (w) p(CH2) 890 (w)^ p(CH2) 881 (mw)^886 (m) 855 (mw) 856 (mw) 826 (w, sh)^830 (mw) 819 (w) 816 (mw) 804 (mw)^803 (mw) 794 (m) 775 (mw)^779 (vw) 746 (w, sh) 745 (ms, sh)^v(CC) ring 726 (ms)^729 (s)^v(CC) ring 678 (s) 680 (s) v(CC) ring 657 (m) 608 (ms)^603 (s)^8(CC) ring deformation 592 (m, sh) 564 (m)^564 (ms) 538 (s) 533 (ms) 511 (vw)^508 (vw) 493 (ms, sh)^496 (vw) 488 (ms) 462 (m)^472 (m, br)^8(CCC) 442 (mw) 447 (m) 8(CCO) 412 (mw)^406 (m, br) 394 (vw) 378 (mw) 8(CCC) 351 (mw) s = strong, ms = medium strong, m = medium, mw = medium weak, w = weak, sh = shoulder, br = broad It is acknowledged that triacontanoic acid contains Raman bands associated with the acid functional group, but these bands are obscured by the stronger CC stretches and are thus not included in the table. 11. Triacontanoic acid and oleanolic acid band assignments are based on Faolain et al. [43] and Brody et al. [39], respectively. 100 4.3.3 Artificial Triterpenoid Mixtures The goal here was to prepare a triterpenoid mixture sample with spatially uniform concentrations of each component, which should give similar quantitative results regardless of the spot at which the spectrum is acquired. When doing quantitative analysis to determine the concentrations of the different components in a sample, a key component is the internal standard. Originally, it was hoped that n-tetracosane be used as the internal standard as this compound had been used for this purpose in GC analysis by several investigators [44-46]. However, since an internal standard would not be added to the in situ analysis of leaf cuticles, it was decided that no internal standard would be added to the mixtures as well. The suitability of the decomposition analysis of the wax mixture without the use of an internal standard is investigated below. Since the sample preparation of almost all previous quantitative analyses of cuticular waxes include the dissolution of the waxes in chloroform, it was decided to follow suit and prepare the artificial wax mixtures by dissolving the triterpenoid compounds of various relative weight ratios in chloroform. This technique, however, had several limitations. The first problem was the presence of the glass background signal produced as a result of the recrystallized mixture being too thin. The glass background can be removed by subtracting the spectrum of the glass microscope slide from the mixture spectrum. However, since the broad silica band, centered at around 1375 cmd , lies in the region where the triterpenoid peaks are found, at low concentrations, these desired triterpenoid peaks can be masked by the relatively strong glass signal. The most problematic feature of this technique, however, was the inconsistency in the relative ratios obtained by decomposing the spectra of the recrystallized mixtures when compared with those of other spectra acquired at different locations in the same sample. This may have been caused by a spatial non-uniformity of the mixture concentration between the different components during the recrystallization process. For this reason, a different "solvent" was used to prepare the wax mixtures. Nujol and KBr pellets are often used in IR analysis of solids because of their optical transparency in this region. For this study nujol, a chemically inert mineral oil, was chosen as the matrix for its similarity in composition to the other components found in wax (i.e. very long chain aliphatic compounds) and to facilitate homogenous distribution of the different reference 101 components. The Raman spectrum of nujol is quite sparse with no interfering peaks in the triterpenoid fingerprint region of the spectrum. Triterpenoid mixtures were prepared by combining different triterpenoids in nujol. Mixtures of nonacosane and ursolic acid, chosen to represent the very long chain aliphatic and triterpenoid compounds, respectively, were also prepared and analyzed. It was observed that the relative intensities of the nonacosane peaks were greater than those of ursolic acid, even at a 1:1 ratio (w/w). However, ursolic acid was still detectable in a 10:1 mixture (w/w) of nonacosane to ursolic acid, which was close to the detection limit of the triterpenoid. Jetter et al. determined the very long chain aliphatic to triterpenoid ratio of intact P. laurocerasus leaves to be 7.5:1 by GC analysis [4], which implies that the triterpenoid peaks should be detectable by Raman spectroscopy in biological samples from this species. Figure 4.4 is a stack plot of binary mixtures of triterpenoids containing 1:2, 1:1, and 2:1 a-amyrin and oleanolic acid. The region from 500-750 cm -1 shows most pronounced variations in peak intensities as the relative concentrations of the two compounds are varied. Thus, when the linear combination calculations were performed on the various mixtures, only the pure triterpenoids were used as the reference spectra in the range of 430-900 cm -1 . Table 4.2 lists the calculated relative concentration of the triterpenoids in the two sets of mixtures, one that contained a-amyrin and oleanolic acid, and the other that contained ursolic acid and oleanolic acid, and shows that the calculated values are in accordance with their actual relative weight ratios. As noted earlier, the addition of the carboxylic acid group to amyrin significantly alters its Raman spectrum, which aided the quantitative analysis. The fact that the results of the quantitative analysis of the oleanolic acid and ursolic acid mixtures were reasonable emphasizes the capability of Raman microspectroscopy to effectively distinguish between two structural isomers with very similar Raman spectra. 102 (a) (b) (c) 400 ^ 600 ^ 800 Raman shift (cm -I ) Figure 4.4 Raman spectra of artificial triterpenoid mixtures with (a) 1:2, (b) 1:1, and (c) 2:1 relative weight concentrations of a-amyrin:oleanolic acid in the 350-900 cm -I spectral region. The shaded areas highlight spectral regions with apparent changes in peak heights as the relative concentrations are varied. 103 Table 4.2 Actual and calculated experimental relative weight ratios of (a) a-amyrin to oleanolic acid, and (b) oleanolic acid to ursolic acid in the various triterpenoid mixtures. Sample mixture Actual RelativeWeight Ratio Experimental Ratio % Difference 0.54 0.50 +/- 0.10 7.1 (a) a-Amyrin : Oleanolic acid 1.03 1.02 +/- 0.09 0.7 1.95 1.94 +/- 0.13 0.7 0.46 0.45 +/-0.06 1.7 (b) Oleanolic acid : Ursolic acid 1.00 0.97 +/- 0.09 3.0 2.05 1.83 +/- 0.25 10.7 In addition to working with binary mixtures composed only of triterpenoids, a mixture containing nonacosane, ursolic acid, and oleanolic acid with a weight ratio of 28:3:1, respectively, was prepared to provide a better representation of the wax found in the cuticles of P. laurocerasus leaves [4]. For this mixture, nonacosane was used to represent all the very long chain aliphatic compounds since it was observed (as noted above) that the spectra of these compounds were nearly identical. The resulting experimental ratios for this mixture were once again in accordance (within acceptable error margins) with their actual relative weight ratios (Table 4.3). Table 4.3 Actual and calculated experimental relative weight ratios of (a) nonacosane to triterpenoids, and (b) ursolic acid to oleanolic acid in the tertiary mixture. Compounds Actual RelativeWeight Ratio Experimental Ratio % Difference (a) Nonacosane : Triterpenoids (b) Ursolic acid : Oleanolic acid 6.8  3.1 8.0 +/- 0.2 3.8 +/- 1.3 17.6 22.6 104 4.4 Conclusion Fluorescence was present in the visible and NIR Raman spectra of the reference triterpenoid compounds, but this problem was solved by performing the baseline correction method selected from Chapter 3. Sample illumination in the NIR region produced baseline- corrected spectra with superior SNR compared with visible excitation. Additionally, chloroform, a common solvent for the quantitative analysis of soluble plant waxes with GC, was found to be unsatisfactory for artificial triterpenoid mixture analysis by Raman spectroscopy due to the partial segregation of the different components during recrystallization resulting in a spatial non- uniform concentration distribution. More importantly, however, this chapter demonstrates the capability of utilizing Raman microspectroscopy for the relative quantification of structurally similar triterpenoids in a mixture. Triterpenoids in an aliphatic environment are detectable at concentrations well below the relative ratio of the very long chain aliphatic to triterpenoid compounds experimentally observed in leaf wax of P. laurocerasus. These results portend the utility for using this approach to analyze the spatial distribution of cuticle components in situ. Thus, Raman microspectroscopy has been shown to be a promising method to complement GC-MS analyses which have allowed the identification and bulk quantification of the different components in wax layers and crystals of plants, but which provided no spatial information about the distribution of the components. 105 4.5 References [1] Baker EA in The Plant Cuticle; Cutler DF, Alvin KL and Price CE, Eds.; Academic Press: London, 1982; Vol. 10, pp 139-165. [2] Holloway PJ in The Plant Cuticle; Cutler DF, Alvin KL and Price CE, Eds.; Academic Press: London, 1982; Vol. 10, pp 1-32. [3] Walton TJ. Waxes, cutin and suberin; Academic Press: London, 1990. [4] Jetter R, Schaffer S, and Riederer M. Plant Cell and Environment 2000, 23, 619-628. [5] Hu ZH. Acta Pharmacologica Sinica 1988, 23, 553-560. [6] Liu J, Liu YP, Parkinson A, and Klaassen CD. Journal of Pharmacology and Experimental Therapeutics 1995, 275, 768-774. [7] James LP, Mayeux PR, and Hinson JA. Drug Metabolism and Disposition 2003, 31, 1499-1506. [8] Liu J. Journal of Ethnopharmacology 1995, 49, 57-68. [9] Hsu HY and Tsai HJ. Cancer Detection and Prevention 2000, 24, 89. [10] Liu H. Journal of Ethnopharmacology 2005, 100, 92-94. [11] Li J, Guo WJ, and Yang QY. World Journal of Gastroenterology 2002, 8, 493-495. [12] Oliveira FA, Lima RCP, Cordeiro WM, Vieira GM, Chaves MH, Almeida FRC, Silva RM, Santos FA, and Rao VSN. Pharmacology Biochemistry and Behavior 2004, 78, 719- 725. [13] Oliveira FA, Vieira-Junior GM, Chaves MH, Almeida FRC, Florencio MG, Lima RCP, Silva RM, Santos FA, and Rao VSN. Pharmacological Research 2004, 49, 105-111. [14] Oliveira FA, Vieira-Junior GM, Chaves MH, Almeida FRC, Santos KA, Martins FS, Silva RM, Santos FA, and Rao VSN. Planta Medica 2004, 70, 780-782. [15] Najid A, Simon A, Cook J, Chablerabinovitch H, Delage C, Chulia AJ, and Rigaud M. FEBS Letters 1992, 299, 213-217. [16] Liu J, Liu YP, Mao Q, and Klaassen CD. Fundamental and Applied Toxicology 1994, 22, 34-40. [17] Saraswat BS, Visen PKS, Dayla R, Agarwal DP, and Patnaik GK. Indian Journal of Pharmacology 1996, 28, 232-239. [18] Kim KA, Lee JS, Park HJ, Kim JW, Kim CJ, Shim IS, Kim NJ, Han SM, and Lim S. Life Sciences 2004, 74, 2769-2779. 106 [19] Baricevic D, Sosa S, Della Loggia R, Tubaro A, Simonovska B, Krasna A, and Zupancic A. Journal of Ethnopharmacology 2001, 75, 125-132. [20] Banno N, Akihisa T, Tokuda H, Yasukawa K, Higashihara H, Ukiya M, Watanabe K, Kimura Y, Hasegawa J, and Nishino H. Bioscience Biotechnology and Biochemistry 2004, 68, 85-90. [21] Kowalewski Z, Kortus M, Ediza W, and Koniar H. Archivum Immunologiae et Therapiae Experimentalis 1976, 24, 115-119. [22] Sattar A, Bankova V, Kujumgiev A, Galabov A, Ignatova A, Todorova C, and Popov S. Pharmazie 1995, 50, 62-65. [23] Jeong TS, Hwang EI, Lee HB, Lee ES, Kim YK, Min BS, Bae KH, Bok SH, and Kim SU. Planta Medica 1999, 65, 261-263. [24] Tang HQ, Hu J, Yang L, and Tan RX. Planta Medica 2000, 66, 391-393. [25] Marquina S, Maldonado N, Garduno-Ramirez ML, Aranda E, Villarreal ML, Navarro V, Bye R, Delgado G, and Alvarez L. Phytochemistry 2001, 56, 93-97. [26] Kashiwada Y, Nagao T, Hashimoto A, Ikeshiro Y, Okabe H, Cosentino LM, and Lee KH. Journal of Natural Products 2000, 63, 1619-1622. [27] Ma C, Nakamura N, Hattori M, Kakuda H, Qiao J, and Yu H. Journal of Natural Products 2000, 63, 238-242. [28] Alvares ME, Maria AO, and Saad JR. Phytotherapy Research 2002, 16, 71-73. [29] Yoshikawa M and Matsuda H. Biofactors 2000, 13, 231-237. [30] Prinsloo LC, du Plooy W, and van der Merwe C. Journal of Raman Spectroscopy 2004, 35, 561-567. [31] Savitzky A and Golay MJE. Analytical Chemistry 1964, 38, 1627-1639. [32] Brennan JF, Romer TJ, Lees RS, Tercyak AM, Kramer JR, and Feld MS. Circulation 1997, 96, 99-105. [33] Osten DW and Kowalski BR in Computerized Quantitative Infrared Analysis, ASTM STP 934; McClure GL, Ed.; American Society for Testing and Materials: Philadelphia, 1984, pp 6-35. [34] McCreery RL. Raman Spectroscopy for Chemical Analysis; John Wiley & Sons, Inc.: New York, 2000. [35] Ferraro JR, Nakamoto K, and Brown CW. Introductory Raman spectroscopy; 2nd ed.; Academic Press: San Diego, 2003. 107 [36] Agarwal UP. Planta 2006, 224, 1141-1153. [37] Buschman HP, Marple ET, Wach ML, Bennett B, Schut TCB, Bruining HA, Bruschke AV, van der Laarse A, and Puppels GJ. Analytical Chemistry 2000, 72, 3771-3775. [38] Schaar JA, Mastik F, Regar E, den Uil CA, Gijsen FJ, Wentzel JJ, Serruys PW, and van der Stehen AFW. Current Pharmaceutical Design 2007, 13, 995-1001. [39] Brody RH, Edwards HGM, and Pollard AM. Biopolymers 2002, 67, 129-141. [40] Kalyanasundaram K and Thomas JK. The Journal of Physical Chemistry 1976, 80, 1462- 1473. [41] Edwards HGM and Falk MJP. Spectrochimica Acta Part A-Molecular and Biomolecular Spectroscopy 1997, 53, 2685-2694. [42] Amorim da Costa AM, Marques MPM, and Batista de Carvalho LAE. Vibrational Spectroscopy 2004, 35, 165-171. [43] Faolain EO, Hunter MB, Byrne JM, Kelehan P, Lambkin HA, Byrne HJ, and Lyng FM. Journal of Histochemistry and Cytochemistry 2005, 53, 121-129. [44] NO R, Markstadter C, Hauke V, and Riederer M. Phytochemical Analysis 1998, 9, 112- 118. [45] Gniwotta F, Vogg G, Gattmann V, Carver TLW, Riederer M, and Jetter R. Plant Physiology 2005, 139, 519-530. [46] Guhling 0, Kinzler C, Dreyer M, Bringmann G, and Jetter R. Journal of Chemical Ecology 2005, 31, 2323-2341. 108 CHAPTER 5 IN SITU ANALYSIS OF TRITERPENOID COMPOSITIONAL PATTERNS WITHIN PRUNUS LAUROCERASUS LEAF CUTICLES * This chapter explores the usage of linear Raman microspectroscopy, coherent anti-Stokes Raman scattering (CARS) spectroscopy, and third harmonic generation (THG) spectroscopy to determine the spatial distribution of triterpenoids in the cuticles of Prunus laurocerasus (Laurel cherry) leaves. The complementing results from the three independent methods are used to interpret the significance of the observed triterpenoid distribution. 5.1 Introduction The epidermal surfaces of higher plants are compound structures exhibiting heterogeneity on both cellular and sub-cellular levels. The surfaces of leaves, like those of most above-ground organs, are formed by characteristic patchwork patterns of pavement cells, guard cells and trichomes [1, 2]. While trichomes are present only in certain species and predominantly at early stages of leaf development, pavement cells are ubiquitous and make up the largest portion of leaf surfaces. The majority of plant species are hypostomatous, with guard cells being interspersed between pavement cells only on the lower, abaxial, surface of the leaf. On the sub-cellular level, two major regions can be distinguished on the surface of each pavement cell, formed by either the periclinal or the anticlinal walls. All epidermal cells are covered by a thin extra-cellular membrane, known as the cuticle, that serves as a protective barrier against the environment. The cuticle consists of an insoluble polymeric matrix (cutin) and soluble cuticular waxes. The latter can be found embedded within the matrix (intracuticular wax) and as a layer deposited onto the outer surface of cutin * A version of this chapter has been published. Yu MML, Konorov SO, Schulze HG, Blades MW, Turner RFB, and Jetter R. In situ analysis by microspectroscopy reveals triterpenoid compositional patterns within leaf cuticles of Prunus laurocerasus. Planta, 2008, 227(4), 823- 834. 109 (epicuticular wax). Previous studies have shown that the cuticular waxes are mainly composed of very long chain aliphatic compounds and other characteristic components, such as triterpenoids, sterols and flavonoids [3]. The waxes, along with the cutin matrix, play important physiological and ecological roles in the minimization of water loss [4], protection against UV radiation [5, 6], as well as resistance against pathogens [7, 8] and herbivores [9]. In order to fully understand these functions of plant cuticles, it is important to study how the waxes are distributed on the plant surfaces. The chemical composition of the total cuticular wax mixture is commonly studied by extracting the compounds from the plant surface and analyzing it with gas chromatography-mass spectrometry (GC-MS) [10, 11]. Fairly recently, new protocols were developed that allowed the separate sampling of epicuticular and intracuticular wax layers for GC-MS analysis [12, 13]. When these adhesive sampling methods were employed on the leaf cuticle of P. laurocerasus, one of the model species for investigations into the physiological roles of waxes, drastic compositional differences between both layers of waxes were revealed. The epicuticular wax was found to consist exclusively of very long chain aliphatic compounds, while the intracuticular wax contained both very long chain aliphatics and the triterpenoids ursolic acid and oleanolic acid. The finding of such transversal gradients raised the question whether further heterogeneity also exists in the lateral distribution of wax compounds along the leaf surface. However, it cannot be answered by GC-MS, since this technique requires bulk samples and thus cannot provide laterally resolved chemical information. Moreover, GC-MS is a destructive and labor-intensive method involving several steps of sample preparation, thus limiting further progress in the chemical understanding of plant surfaces. A number of techniques, alternative or complementary to GC-MS, have recently been used to study plant surfaces as described in Chapter 1. Scanning electron microscopy (SEM) and atomic force microscopy (AFM) were employed to map morphological features on the leaf surfaces with high spatial resolution [14-18], but these methods do not give any chemical information. Fourier-transform infrared (FT-IR) spectroscopy in transmission near-infrared (NIR) [19] or in attenuated total reflectance (ATR) mode [20, 21], total internal reflectance (TIR) 110 spectroscopy [22], X-ray photoelectron spectroscopy (XPS) [14, 23], time-of-flight secondary ion mass spectrometry (ToF-SIMS) [14], nuclear magnetic resonance (NMR) [24], and radiolabelling [25] experiments have also been used to study the surface morphology and chemistry of plants and fruits. Although all these techniques provide useful chemical information about cutin and cuticular wax, none have been employed to study the lateral distribution of wax compounds in leaf cuticles in situ. It was shown in Chapter 4 that oleanolic acid and ursolic acid, two structural isomers differing only in the position of one methyl group, could be effectively distinguished and quantified in artificial wax mixtures using Raman microspectroscopy. The current investigation therefore is aimed at the in situ Raman microspectroscopic mapping of cuticular triterpenoids on a cellular and sub-cellular level. Since spontaneous Raman spectroscopy is a linear process, the intensity of the Raman peak is directly proportional to the amount of substance in the sampled spot [26, 27]. Therefore, plotting the intensity of a certain peak as a function of location gives rise to a map displaying the spatial distribution of a particular substance. For this purpose, the cuticular membranes (CMs) were isolated from the hypostomatous adaxial and stomatous abaxial sides of P. laurocerasus leaves, and investigated using two Raman spectroscopic techniques, near infrared (NIR) Raman microspectroscopy and CARS spectroscopy, as well as THG spectroscopy. The Raman-based techniques provide chemical selectivity and spatially resolved concentration/density information, while THG spectroscopy provides information about spatial gradients in optical properties which reflect subtle compositional gradients that may not be discernible by the vibrational spectroscopic methods. 5.2 Materials and Methods 5.2.1 Reference Compounds The reference samples, selected to represent the major wax components of P. laurocerasus, were ursolic acid, oleanolic acid, and nonacosane (Sigma-Aldrich). All reference compounds were powdered or crystalline solids. Several Raman measurements were acquired at different points within the sample to minimize crystalline effects. Comparison of these spectra 111 displayed minimal variations and so the multiple measurements were averaged to obtain the basis spectra used for the cuticular analysis. 5.2.2 P. laurocerasus Cuticular Membranes Leaves of P. laurocerasus were harvested from trees growing in the Botanical Garden of the University of Wiirzburg, Germany, as well as from local hedges. Disks of 1.5 to 2 cm in diameter were excised from mature leaves. Since the cuticle is attached to the epidermal cells by a pectinaceous layer and cellulose fibrils [28, 29], both the adaxial and abaxial CMs were enzymatically isolated based on Orgell's technique [30], as modified by Yamada et al. [31] and Petracek and Bukovac [32], by immersing the disks into a sodium citrate buffer containing 2% (w/w) pectinase (MP Biomedicals, Solon, OH), 2% (w/w) cellulase (MP Biomedicals), and 1 mM sodium azide (NaN3) to prevent bacterial growth. Adaxial and abaxial CMs were identified by the presence of the stomata in the latter CM when viewed under a microscope. In order to minimize background signal from the glass, the CM samples were mounted onto microscope slides modified by drilling holes into them underlying the field of view, and no cover slips were used. For cuticle thickness determination, several isolated adaxial CMs were weighed on an electronic microbalance (Sartorius AG, Goettingen, Germany) with a precision of ± 1 jug. 5.2.3 Epicuticular Wax Preparation Epicuticular waxes were removed from mature P. laurocerasus leaves by placing the 2-cm leaf disks onto a spatula with approximately 1 mL of distilled water, as described by Riedel et al. [33]. The bottom side of the spatula was then gently placed on top of liquid nitrogen, causing the water to freeze. With the water still frozen, the leaf was dislodged leaving behind the epicuticular wax layer trapped in the ice. The ice was then placed onto a microscope slide and upon melting, the water was removed by capillary action with a paper towel. The remaining wax was allowed to dry completely before Raman measurements were made. Due to the fragility of the epicuticular wax, regular microscope slides, as opposed to the custom open-hole slides, were used. 112 5.2.4 Linear Raman Microspectroscopy NIR Raman spectra were acquired with a confocal Raman microscope (Renishaw) equipped with a 785 nm diode laser source and a charge-coupled device (CCD) detector. A 50x/0.75 numerical aperture (NA) objective (Leica Microsystems) was employed to focus the laser to a 5-pim spot on the sample, which was irradiated with 40-50 mW of laser power. Spectra were collected for 60 s over the range of 350-2000 cm -1 at 4 cm -1 spectral resolution. A raster scan was performed on selected areas of the isolated cuticle with 5-i.tm step sizes in both the x and y directions; scans were done in triplicate. The NIR Raman spectra of all the samples were baseline corrected prior to data analysis using the zeroth order Savitzky-Golay filter [34] as outlined in Chapters 3 and 4, and the measured intensities of each spectrum were normalized to the amplitude of the strongest peak. The relative concentrations of the triterpenoids were calculated by treating the CM spectra as linear combinations of the reference spectra [35] as described in Chapter 4. Chemical images were produced by choosing peaks in the Raman spectrum unique to triterpenoids and other components of the cuticle and plotting its peak height as a function of location. It should be noted that scans of several different locations on the same and on different isolated CMs were performed. Due to the consistency of the results, only one representative set of chemical maps is shown. 5.2.5 Non-linear Spectroscopy The non-linear signals were collected using the instrumental set-up described in Chapter 2. The energies of the pump, Stokes, and probe beams were varied from 1 to 10 nJ, while the wavelengths of the different beams were selected in accordance to the desired Raman shift. The signals were collected for 0.30 s per spatial point. No post-acquisitional data processing was performed aside from normalizing the data to the highest signal value. Chemical images were produced by plotting the normalized signal as a function of position. 113 5.2.6 GC-MS Analysis Cuticular wax was extracted from cuticles of mature P. laurocerasus leaves by immersing the isolated adaxial CM in chloroform for 30 s at room temperature. The solvent was then removed under a steady stream of N2 while being heated at 50°C. Prior to GC analysis, the compounds were treated with bis-N,N-(trimethylsilyl)trifluoroacetamide (BS TFA, Sigma- Aldrich) in pyridine (30 min at 70 °C) to convert all hydroxyl-containing compounds into their corresponding trimethylsilyl (TMSi) derivatives. The composition of the cuticular wax was studied using capillary GC (5890N, Agilent, Avondale, PA; column 30 m HP-1, 0.32 mm i.d., df = 0.1 gm, Agilent) with the He carrier gas inlet pressure programmed for a constant flow of 1.4 ml/min coupled to a mass spectrometer (5973N, Agilent). GC was carried out with a temperature-programmed on-column injection at 50 °C, 2 min isothermal at 50°C, temperature ramp from 50 °C to 200°C at 40°C/min, 2 min isothermal at 200 °C, another temperature ramp from 200°C to 320°C at 3 °C/min, followed by 30 min isothermal at 320°C. Individual wax components were identified by comparing their mass spectra with those of authentic standards and the literature data. Relative concentrations of the different components in the mixtures were studied using capillary GC with a flame ionization detector (GC-FID). The GC temperature program was the same as stated above, with the exception of the H2 carrier gas inlet pressure regulated for a constant flow of 2 ml/min. Five replicates were done for quantitative analysis. Relative concentrations were determined by comparing the integrated peak areas. 5.2.7 Scanning Electron Microscopy SEM images of adaxial and abaxial CMs isolated from mature P. laurocerasus leaves were obtained at the Bioimaging Facility at The University of British Columbia. The samples were mounted onto a SEM sample holder and were coated with 2 nm of gold with the Nanotech SEMPrep II Sputter Coater (Nanotech, Manchester, UK) for better electrical conductivity of the sample. A 30 kV scanning electron microscope (model 54700, Hitachi High-Technologies, Schaumburg, IL) was used to obtain images at various magnification ranging from 250x to 2500x magnification, with an accelerating voltage of 500 V and a working distance of 23.5 mm. The samples were tilted 45° for better image contrast. 114 5.2.8 Histological Cross-Sections Transversal cross-sections of the isolated adaxial CMs from mature P. laurocerasus leaves were obtained at the Wax-It Histology Services at The University of British Columbia. The samples were sliced using two different techniques. The first technique involved embedding the sample in paraffin for sectioning. For this technique, the sample was first dehydrated with several solutions of ethanol of increasing concentration (70%, 95%, and 100%), followed by immersing it in 100% xylene (Sigma-Aldrich) as the clearing agent. Following processing, the sample was embedded in paraffin and sectioned using a microtome (Leica Microsystems) at 5 gm thickness. The second technique involved the cryo-sectioning of the sample. The sample was cryo-embedded by freezing the sample with liquid nitrogen-cooled 2-methyl butane onto a cryostat chuck (Leica Microsystems) using OCT (optimal cutting temperature) as the embedding medium. The samples were cryo-sectioned at Slum slices. The samples were kept at —20 °C until examined. The adaxial cuticle cross-sections were viewed under an inverted Motic AE31 microscope (Motic Inc. Ltd., Xiamen, China) using the 20x/0.40 NA and 40x/0.60 NA CCIS ® (colour corrected infinity optical system) microscope objectives (Motic). A photographic image of the sample was obtained with a Nikon D100 digital single lens-lens reflex (SLR) camera (Nikon Corporation, Tokyo, Japan). A micrometer (Wild Heerbrugg Ltd., Heerbrugg, Switzerland) with divisions as low as 10 gm was used to measure the thickness of the cuticles. 5.2.9 Ultraviolet-Visible Absorption Spectroscopy The absorption spectra of isolated adaxial and abaxial CMs from mature P. laurocerasus leaves were recorded with a HP 8425A diode array spectrophotometer (Hewlett-Packard Company, Houston, TX). The isolated CM samples were fixed on drilled microscope slides to minimize signal from the glass. The absorption spectra were taken in the wavelength range of 250-800 nm. 115 5.3 Results and Discussion 5.3.1 Adaxial Leaf Surfaces The adaxial side of P. laurocerasus leaves is devoid of stomata and trichomes, and is therefore formed only by a single type of epidermal cells. There were no significant differences in the chemical composition of waxes among these pavement cells, thus providing an opportunity for a detailed analysis of the surface patterning on a subcellular level. A series of experiments were carried out to map lateral chemical patterns, combining near infrared (NIR) Raman microspectroscopy, coherent anti-Stokes Raman scattering (CARS), and third harmonic generation (THG) spectroscopy. Initially, conventional chemical analyses had to be performed in order to verify the overall wax composition of the samples used in the present investigation, and to select relevant compounds representing the wax mixtures in reference samples. To this end, CMs were isolated from the adaxial side of P. laurocerasus leaves, and the soluble waxes were extracted with chloroform and analyzed by GC-MS (Figure 5.1). In accordance with literature reports [13], the main constituents of the wax mixture were very long chain aliphatic compounds (58%) and pentacyclic triterpenoids (42%). The two major triterpenoids were identified as ursolic acid and oleanolic acid and were found to be present in a ratio of approximately 3:1, further confirming previously published data. The relative amounts of non-triterpenoids to triterpenoids, however, were found to be approximately 1.4:1, which is lower than the literature data of 7.5:1 [13]. These differences in relative concentration values may have arisen from the different preparation and extraction conditions used in both studies, as well as from the use of different samples (i.e. intact leaves versus isolated adaxial cuticles). 116 600 - 500 d 400 - 300 - 200- 100- _dill 44-11.-14,..AAAJIn.....ALAWL..d 0 -^ I^• 10^15^20 25^30^35^40^45^50 a 4111,,,L^. Time (min) Figure 5.1 Gas chromatogram of the soluble waxes from isolated adaxial cuticles of P. laurocerasus leaves. The labeled peaks are (a) nonacosane at 19 min, (b) hentriacontane at 23 min, (c) oleanolic acid at 30 min, and (d) ursolic acid at 31 min. NIR Raman analysis of the "peeled" epicuticular wax layers from leaves using water as the peeling agent proved to be unsuccessful. Similar to the chloroform-dissolved triterpenoid mixture analysis in Chapter 4, the wax layer was too thin resulting in a large glass background from the microscope slide. Furthermore, although the peaks from the very long chain aliphatic compound were clearly observed in the artificial mixture spectra, no such peaks could be observed in the epicuticular wax spectra. While other methods of epicuticular wax layer removal using glycerol [36], triethylene-glycol [36], aqueous gum arabic [12, 13, 33, 37] or chloroform- moistened paintbrushes [37] have been reported, these latter solvents were not considered here for further epicuticular wax analyses as their spectra would interfere with the desired peaks. Thus, it was decided that NIR Raman microspectroscopy is not an ideal method for studying the components of epicuticular wax of leaves. 117 Prior to any Raman analysis of the isolated CMs, it was necessary to determine the optimal instrumental parameters and to verify that the samples were not damaged. The latter can be assessed by acquiring a spectrum at the beginning of the experiment and at a certain time after the sample has been irradiated with the laser. If the sample were severely damaged after the intense laser radiation, variations in the peak height and positions would be observed. This can also be confirmed by looking at SEM regions before and after laser irradiation, which showed no visible sample destruction even at 250x magnification. As with most biological samples, the Raman spectra of the isolated adaxial cuticles of P. laurocerasus leaves displayed a large fluorescence background, making quantitative measurements difficult if analyses were done on the raw data. Furthermore, the baseline slope was steepest in the spectral region of most interest, hence requiring careful baseline correction of the spectra. Once again, due to the large number of spectra acquired, the zeroth order Savitzky-Golay filter was utilized as the automated baseline correction method. Another observation supporting the absolute need of baseline correction was the decreasing baseline slope as a function of time (Figure 5.2(a)). Initially, it was thought that there perhaps may have been some sample degradation, but when the baseline-corrected spectra were compared, it could clearly be seen that there were minimal variations in the peak heights and no shifts in the peak positions (Figure 5.2(b)). The gradual decrease in the baseline slope was suspected to be due to photobleaching of the sample [22]. 118 40000 ^ 35000 30000 — 25000 *c7 20000 15000 10000 — 5000 — a) 0 ^ 400^600^800 1000 1200 1400 1600 1800 2000 9000 ^ 8000 — 7000 — 6000 5000 — *E7i 4000 — 3000 — —1000 400^600^800 1000 1200 1400 1600 1800 2000 2000 — 1000 — Raman shift (cm -1 ) Figure 5.2 A series of NIR Raman spectra acquired at the same position as a function of acquisition order. (a) With each successive acquisition, the baseline becomes progressively lower. The different colours represent the spectrum as a function of acquisition order. The top blue spectrum is the first acquired spectrum with each consecutive spectrum having a lower baseline, until the last acquired spectrum shown in golden orange. (b) After baseline correction using a zeroth order Savitzky-Golay filter, the overlaid spectra show little variation in peak height and position. b) 119 In a first Raman spectroscopic experiment, full NIR Raman spectra were acquired. The baseline-free NIR Raman spectrum of isolated P. laurocerasus CM is shown in Figure 5.3(a). Based on the peak assignments for ursolic acid, oleanolic acid and nonacosane listed in Chapter 4, the peaks in the lower wavenumber region of the P. laurocerasus CM spectrum could be assigned to triterpenoids. These peaks correspond to the ring CC stretching and breathing modes (700-750 cm-1 and 1000-1190 cm 1 ), CH3 rocking (900-950 cm'), and CCC and CCO bends (440-480 cm'). Although the triterpenoids also contribute to the peaks in the higher wavenumber region, these peaks had previously been assigned mainly to very long chain wax constituents. A few of these peaks could be assigned to the cutin matrix (MX), as they remained unchanged in the MX spectrum acquired after immersing the CM into chloroform to remove all soluble wax (Figure 5.3(b)). The MX spectrum showed very few peaks in the low wavenumber region, further supporting the interpretation that the peaks in the corresponding region of the CM spectrum belonged to extractable compounds. The difference spectrum obtained by subtracting the MX spectrum from the CM spectrum (Figure 5.3(c)) reflected the spectrum of the cuticular wax, and thus enabled the further assignment of a number of peaks in the wavenumber region from 1050-1650 cm' to non-triterpenoid wax constituents. Accordingly, the strong peaks at 1061 cm-1 and 1130 cm' were described in Chapter 4 as characteristic CC stretches for very long chain alkanes and fatty acids. Initially, bright field light microscopy showed a characteristic pattern of the pavement cells in the CMs isolated from the adaxial side of P. laurocerasus leaves (Figure 5.4(a)). The cell outlines appeared as dark lines, suggesting zones of increased cuticle thickness. This was supported by SEM micrographs showing that the physiological outer side of the cuticle was smooth, while the side that had originally been attached to the epidermal cell walls was rough with cuticular ridges following the anticlinal cell wall lines (Figure 5.5). Similar evidence (light microscopy, SEM, TEM) has previously been provided for various other plant species, and it is therefore well-established that cuticular ridges typically fill the valleys in the topography of epidermal cells, most pronouncedly present in the surface regions over the junction of anticlinal walls of adjacent cells. It can therefore be concluded that the P. laurocerasus cuticles have locally varying thickness similar to those of other species, with relatively thin membranes spanning the periclinal areas and much thicker zones over the anticlinals. These biological structures seen in light micrographs can now be used to interpret chemical maps generated by Raman microspectroscopy. 120  cutin + wax (a) cutin (b) io o-O co I T )- Lo cv^ .. •^ co^0co^co co oO^ co T 1 ILc) 0 'D eocv^ .... c 0 , co co P1 mc o I^o: -. 2 RLc)^Lc) 1^,--- Lc)^cr,c.c)^— co ,--iA t\ j I difference spectrum (a) - (b) (c) kk,f1, 500^1000^1500^2000 Raman shift (cm- 1 ) Figure 5.3 In situ NIR Raman spectra of isolated cuticular membrane from adaxial surfaces of P. laurocerasus leaves (a) with and (b) without wax, and (c) their difference spectrum. 121 It was shown in Chapter 4 that the intensity of selected Raman peaks could be used to assess the relative quantities of triterpenoids and very long chain aliphatics in wax mixtures. Based on the findings from the above experiment, the Raman peak at 728 cm', originating from the CC stretch in the ring vibration unique to oleanolic and ursolic acids (see above), was used as a characteristic marker for mapping the cyclic triterpenoids in the wax of P. laurocerasus. A plot of this peak as a function of lateral position on the adaxial P. laurocerasus leaf surface (Figure 5.4(b)) showed that the triterpenoids are present in the cuticle overlying all pavement cells. The triterpenoid signal was relatively strong in the cuticle regions over anticlinal cell walls and weaker in the periclinal regions. In addition, the peak at 1130 cm' is a characteristic signal for the very long chain aliphatic constituents of the P. laurocerasus wax and the chemical image (Figure 5.4(c)) plotted from the original NIR Raman dataset shows that these non-triterpenoid wax constituents also accumulated in the cuticle regions above the anticlinal walls rather than the periclinal areas. It should be noted that this result reflects the absolute amounts of triterpenoids and very long chain aliphatic compounds in the different lateral regions of the cuticle, but not their relative concentrations. The absolute quantities are probably varying largely due to differences in cuticle thickness between (thin) periclinal and (thick) anticlinal areas. It is therefore not clear whether additional differences exist in local triterpenoid and non-triterpenoid fractions between both cuticle areas. 122 a) b) 1 0.8 0.6 0.4 0.2 c) 1 0.8 0.6 0.4 0.2 0 Figure 5.4 Chemical images of isolated adaxial P. laurocerasus cuticular membrane using NIR Raman microspectroscopy. (a) Bright field light microscopic image of a region on the isolated adaxial P. laurocerasus cuticular membrane and the corresponding Raman maps based on the intensity of the bands at (b) 728 cm -1 and (c) 1130 cm -1 , which correspond to peaks unique to triterpenoids and nonacosane, respectively. Bar = 10 gm 123 54700 0.5kV 23 4mm x250 SE(M) 3/23107 (a) (b) Figure 5.5 SEM micrographs of isolated adaxial P. laurocerasus cuticular membrane under 250x magnification. (a) The environmental side of the cuticle is seen as a relatively smooth surface compared to the (b) physiological side of the cuticle. An axial Raman mapping experiment could potentially provide further insight to the different chemical distribution. Unfortunately, the thickness of the isolated CM was found to be too thin for optical sectioning of the sample to be performed. The overall thickness of several isolated CMs was determined using four different methods. The first method was very crude that involved the placement of the CMs between cover slips and determining what the difference in the gauge readings were. The second method involved calculating the thickness from the mass per unit surface [38, 39] by multiplying the weight of several CMs assuming a density of 1000 kg/m3 [40, 41]. The third method involved the deduction of the thickness from the frequency of the refractive index and the interference fringes of the cuticle [42]. The latter method was unsuccessful as no interference fringes could be observed. However, the first and second methods gave crude values of the CM thickness ranging from 6-12 !lin. The fourth method consisted of fixing the isolated adaxial cuticles and obtaining their physical cross-sections to determine the thickness of the different regions of the cuticle (Section 5.2.8). Although paraffin embedding is the more popular technique for histological analysis, the CM samples were found to be too fragile for this technique and thus another common technique, cryo-sectioning, was performed. Overall, this fourth method proved to be most accurate since a high-resolution image of the magnified cuticle cross-section could be obtained (Figure 5.6). However, due to the 124 fragility of the sample, especially when cutting it into 5-[tm thin slices, it is hard to say whether the sample thickness was compromised upon slicing. Nevertheless, the histology revealed that the periclinal cuticle thickness ranged from 8-10 j.tm, while the anticlinal ridges were relatively thicker, varying from 11-14 Figure 5.6 Bright field light microscopic image of an isolated adaxial P. laurocerasus cuticular membrane cross-section. Each division on the ruler at the right of the image corresponds to 10 Using the dataset for triterpenoid and non-triterpenoid distributions, the relative amounts (i.e. percentages) of these cuticle constituents across the cuticle can now be addressed. In particular, the key question here is whether the enhanced triterpenoid signal for anticlinal areas in the NIR Raman maps is: (i) only due to the greater thickness (and hence volume) of the cuticle in these regions, or (ii) also due to a relatively higher percentage of the triterpenoids in these regions? This question can be resolved by performing a decomposition analysis on the CM Raman spectra. Assuming that the CM spectra are linear combinations of the basis components, 125 coefficients can be calculated for each basis component to show how much each particular component contributes to the overall spectrum. The resulting coefficients can be interpreted as percentages of each compound at a given spot of the cuticle; plotting the coefficients of selected compounds as a function of location displays the concentration distribution of that component within the selected region. Furthermore, as seen in Chapter 4, the ratio of the coefficients can be used to compare the relative concentrations of the different compounds. The basis components used in the present decomposition analysis were ursolic acid, oleanolic acid, nonacosane, and an isolated MX cuticular membrane. Nonacosane was used as a single compound representing all the very long chain aliphatic compounds, as it was shown in Chapter 4 that there is very little variation between the Raman spectra of various very long chain alcohols, alkanes, and fatty acids. Using the same region as Figure 5.4, it is an interesting observation that the two- dimensional (2D) map of the ursolic acid (Figure 5.7(a)) and oleanolic acid (Figure 5.7(b)) coefficients show higher triterpenoid percentages in the periclinal regions than above the anticlinals. This finding is in apparent contradiction with the results of the first Raman experiment, where the chemical images of the 728 cm -1 peak showed greater triterpenoid intensity in the anticlinal regions. However, both results taken together show that the triterpenoids had lower local concentrations above the anticlinal cell walls, but accumulated to higher absolute amounts there due to the greater thickness of the cuticle in this region. Conversely, the triterpenoids were enriched at relatively high local concentrations in the periclinal cuticles of the pavement cells on the adaxial cuticle of P. laurocerasus. A gradient existed between the relatively high concentration of triterpenoids in the periclinal region and the low concentration above the anticlinals. The 2D map of the nonacosane coefficient was relatively uniform (Figure 5.7(d)), implying that the very long chain aliphatic compounds have relatively even local percentages across the entire adaxial cuticle. Mapping the ratio between the very long chain aliphatic compounds and triterpenoids further illustrated the lateral gradient described above, with higher relative amounts of very long chain aliphatic compounds in the anticlinal regions and higher triterpenoid concentrations in the periclinal regions (Figure 5.7(e)). The relative amounts of the compounds of interest in the adaxial cuticle of P. laurocerasus can be summarized as ratios between key components. Data analysis of the NIR 126 Raman spectroscopic experiment showed that the average ratio between non-triterpenoid compounds (i.e. very long chain aliphatics) on the one hand and triterpenoids on the other, was 1.2:1. The average ratio between ursolic acid and oleanolic acid within the class of triterpenoids was 2.6:1. These results are slightly lower, but overall very similar to those obtained by GC-FID analysis. This comparison of data acquired by two very different techniques illustrates how valuable Raman spectroscopy is, as it not only agrees with the GC results, but can further give detailed information on the localization of different components. 127 c) 6.5 5.5 4.51 3.5 2.5 1.5 0.5  0.2 0.15 0.1 0.05 0 a) d) 0.28 0.23 0.18 0.13 0.08 b) 0.12 0.1 0.08 0.06 0.04 0.02 0 e) 3 2.5 2 1.5 1 0.5 Figure. 5.7 Chemical images of the distribution of the different components in isolated adaxial P. laurocerasus cuticular membrane using NIR Raman microspectroscopy. The decomposition coefficient value maps are show in (a) for ursolic acid and (b) for oleanolic acid, while the relative coefficient ratio map of ursolic acid to oleanolic acid is shown in (c). Similarly, the decomposition coefficient value map for the very long chain aliphatic compounds is shown in (d), while relative coefficient ratio map of the very long chain aliphatic compounds to triterpenoids (both ursolic acid and oleanolic acid collectively) is shown in (e). The corresponding bright field light microscopic image of a region on the isolated adaxial P. laurocerasus cuticular membrane is shown in (f). Bar = 10 tint 128 In further experiments, CARS and THG spectroscopy were employed to clarify the chemical mapping of adaxial P. laurocerasus cuticles. As described in Chapter 1, CARS imaging is a Raman spectroscopic technique that is based on the signal detection of a selected wavenumber range in the Raman spectrum instead of acquiring a broadband spectrum. The spectral resolution in this case is dictated by the spectral bandwidth of the probe pulse. Due to its coherent nature, CARS is capable of mapping particular chemical components by 2D scanning with especially high sensitivity, since the signal is much stronger than the non-resonant background when the beams are tuned to a specific vibrational mode. The signal from the non- resonant background, which arises from the electronic contributions to x (3) , is further reduced by using NIR excitation pulses since the long excitation wavelengths are below most electronic transitions. Bright field light microscopy was used, as before, to identify individual pavement cells and anticlinal ridges in the CMs isolated from the adaxial side of P. laurocerasus leaves (Figure 5.8(a)). It was shown in Figures 4.3 and 5.3 that the peaks in the fingerprint region arise only from the triterpenoid vibrations with minimal interference from that of the very long chain aliphatic compounds and MX. Hence, a CARS image of composite cuticular triterpenoid distribution was obtained by selecting an appropriate marker frequency (wavenumber) from the previous Raman experiments. Similar to the NIR Raman maps, the Raman peak at 728 cm" ) , originating from the CC stretch in the ring vibration unique to oleanolic and ursolic acids (see above), was again used to selectively detect the cyclic triterpenoid compounds in the wax. The CARS image tuned to this wavenumber showed enhanced intensity in the anticlinal regions of the adaxial cuticle of P. laurocerasus (Figure 5.8(b)). Again, this can be interpreted as a preferred accumulation of triterpenoids in these areas of the cuticle, with lower absolute quantities of these compounds being present in the periclinal areas. It should be noted that the use of femtosecond lasers results in low spectral resolution and does not allow for the spectra of the two individual triterpenoids to be distinguished from one another. Higher spectral resolution, if required, can be achieved by decreasing the spectral bandwidth of the probe pulse (i.e. longer pulse duration) [431 However, a decrease in the intensity of the CARS signal is consequently observed since the CARS signal is linearly 129 proportional to the intensity of the probe beam field, the latter which itself is linearly proportional to the spectral bandwidth [44-46]. In THG images, only inhomogeneous regions produce signal and therefore appear light while homogeneous areas appear relatively dark. Keeping in mind that THG spectroscopy emphasizes compositional and density gradients as mentioned in Chapter 1, the observed inhomogeneities may arise from differences in chemical composition, as well as from differences in refractive index, absorption coefficient, or non-linear susceptibility. The THG images of the adaxial CM of P. laurocerasus (Figure 5.8(c)) showed that the ridges along the anticlinal junctions of adjacent pavement cells were relatively homogeneous, while the periclinal regions of the cuticle appeared more inhomogeneous. This result can be interpreted in combination with the previous finding of lateral gradients along the surface of the pavement cells (see above) and transversal gradients among different layers of the cuticle. 130 a) b) 1 0.8 0.6 I 0.4 0.2 0 c) 1 0.8 0.6 0.4 0.2 0 Figure 5.8 Isolated adaxial P. laurocerasus cuticular analysis using non- linear spectroscopy. (a) Bright field light microscopic image of a region on the isolated adaxial P. laurocerasus cuticular membrane. (b) CARS image tuned to the triterpenoid ring CC stretching vibrational frequency at 728 cm -1 , and (c) THG image of the selected region. Bar = 20 131 It had been shown that the triterpenoids are restricted to the intracuticular layer of the adaxial P. laurocerasus leaf cuticle, and that the overlying epicuticular wax film contains exclusively very long chain wax compounds [12]. Averaging across the entire adaxial surface, both the intra- and epicuticular layers were found to contain approximately equal wax densities of 15 and 13 lig/cm2 [13], respectively, in a total CM density of 333 1.1g/cm2 [47]. The overall adaxial wax composition was thus characterized by a transversal gradient with high concentrations of triterpenoids in the relatively thick inner layer and zero concentration in the thinner outer layer. It can be assumed that the same layered arrangement of waxes exists across the entire adaxial surface, including both periclinal and anticlinal regions of the pavement cells. The current results for the lateral distribution of triterpenoids must therefore be superimposed with the transversal gradients. The relatively thin cuticles spanning the periclinal regions of the pavement cells were found to contain relatively high concentrations of triterpenoids (as compared to anticlinal regions) and can therefore be expected to have especially short and steep gradients between the intracuticular and epicuticular layers. The transversal gradients in triterpenoid concentrations thus cause heterogeneity of the cuticle that is highest above the periclinal cell walls and lowest above the anticlinals. The THG and Raman mapping results largely confirm this interpretation. 5.3.2 Abaxial Leaf Surfaces The key difference between the adaxial and abaxial CMs of P. laurocerasus leaves is the presence of both pavement cells and guard cells in the latter, as observed in the SEM images (Figure 5.9). The bright field microscopic images of isolated abaxial cuticles showed clear outlines of both these cell types (Figure 5.10(a)). The pavement cells resembled those on the adaxial side of the leaf, both in their irregular polygonal geometry and in the distribution of cuticular triterpenoids between anticlinal and periclinal regions above the cells. Similar to the adaxial CM, the very long chain aliphatic compounds were found in higher concentration along the anticlinal regions of the abaxial pavement cells than over the periclinal walls. The stomata were found to be closed in the isolated cuticles. The patterning of triterpenoids and aliphatics on the surface of the guard cells differed strikingly from the neighboring pavement cells. The NIR Raman map of the 728 cm-1 triterpenoid peak showed relatively high intensity for guard cells 132 (a) S4700 1.5kV 11.9mm x250 SE(L) 4/4/07 200umS4700 1.5kV 12.0mm x250 SE(L) 4/4/07 1111 (Figure 5.10(b)), whereas the NIR Raman map of the 1130 cm -1 peak, characteristic for the very long chain aliphatic wax compounds, had lower intensity in the guard cell than in the pavement cells (Figure 5.10(c)). The component coefficients from the decomposition analysis showed similar trends as the peak intensity maps, further confirming that triterpenoids accumulate to higher levels on the surfaces of guard cells than on pavement cells, while very long chain aliphatic compounds are deposited along the CM surface with the opposite gradient. (b) Figure 5.9 SEM micrographs of isolated abaxial P. laurocerasus cuticular membrane under 250x magnification. Similar to the adaxial cuticles, (a) the environmental side of the cuticle is seen as a relatively smooth surface compared to the (b) physiological side of the cuticle. The circular features are impressions of the stomata and their surrounding guard cells. 133 a) b) 1 0.8 0.6 0.4 0.2 0 c) 1 0.8 0.6 1 0.4 0.2 0 -0.2 Figure 5.10 Isolated abaxial P. laurocerasus cuticular analysis using NIR Raman microspectroscopy. (a) Bright field light microscopic image of a region on the isolated abaxial P. laurocerasus cuticular membrane. Raman maps based on the intensity of the bands at (b) 728 cm -1 and (c) 1130 cm -1 , which correspond to peaks unique to triterpenoids and nonacosane, respectively. Bar = 10 iirn. 134 5.3.3 Ultraviolet-Visible Absorption Analysis of Adaxial and Abaxial Leaf Surfaces The NIR Raman, CARS, and THG results shown above reveal the chemical gradients of the different wax components across the isolated adaxial and abaxial CMs of P. laurocerasus leaves. One of the proposed functions of cuticular triterpenoids includes protection against harmful UV rays [5, 6]. Krauss et al. previously showed that the transmittance of radiation of isolated P. laurocerasus cuticles decreases substantially at around 350-400 nm, with the transmittance minimum centered at around 290 nm [48]. Figure 5.11 shows the UV-Vis absorption spectra of the isolated adaxial and abaxial CMs. Both spectra have an absorption maximum at around 280 nm. The adaxial CM has an additional absorption maximum at around 310 nm. These absorption maxima correspond to their ability to screen against UV-B (280-315 nm) and UV-A (315-400 nm) radiation [49]. Thus, it can be seen that both spectra are in accordance to the previously published data. Combining the NIR Raman, CARS, THG, and UV- Vis results, the following interpretation is made. Since the harmful UV rays shine mainly on the adaxial side of the leaves, it is not surprising, therefore, to see the absorption maxima of the adaxial CMs to be located where greatest protection against UV-B and UV-A radiation would be required. Given that the abaxial CMs also contain the same chemical composition, their UV-Vis absorption spectra should also show high absorption in the UV region, as observed in Figure 5.11. In addition, the high local triterpenoid concentration over the periclinal regions in the cuticles as observed in Figure 5.7(b) and 5.8(c), could be interpreted as a means to provide greater protection against the UV rays. 135 1.2 0 250 300 350 400 450 500 550 600 650 700 750 800 Wavelength (nm) Figure 5.11 UV-Vis absorption spectrum of isolated P. laurocerasus cuticles from the adaxial (blue) and abaxial (magenta) sides of the leaf. 136 5.4 Conclusion This chapter illustrated the capability of NIR Raman micro spectroscopy, CARS, and THG for the in situ investigation of the chemical distribution of wax components in plant cuticles. Spontaneous Raman microspectroscopy is a well-known and well-established method, which gives full spectral information about the chemical composition of the investigated sample. The main disadvantage of spontaneous Raman spectroscopy is its low signal, requiring long exposure times and high average power of excitation radiation. Non-linear optical methods, such as CARS and THG, give higher signal at lower average power without a luminescence or fluorescence background, which significantly decreases the measurement time. Since the signal is generated only in the focal area of the incident beams, these non-linear methods also have true three-dimensional (3D) imaging capabilities with high spatial resolution. Using the abovementioned techniques, the absolute amounts of cuticular triterpenoids can be mapped with high spatial resolution, and a linear decomposition analysis gives the corresponding maps of relative concentrations. It was shown that the different cell types of the P. laurocerasus leaf epidermis differ in the absolute amounts of cuticular triterpenoids, and that the triterpenoid amounts and relative concentrations further differ among periclinal and anticlinal regions of epidermal pavement cells. These lateral gradients in wax composition are superimposed with additional transversal gradients that have previously been shown for the same tissue and species. Overall, the cuticles of P. laurocerasus are characterized by complex wax compositional patterns, mostly due to varying triterpenoid concentrations, on both the subcellular and cellular levels. These findings illustrate, albeit only for one plant species so far, that the cuticle is a far more complex structure than previously thought. Raman microspectroscopy can be used to investigate whether the cuticles of other plant species show a similar degree of spatial heterogeneity. Depending on the results, the local distribution of wax constituents may have to be taken into account when considering correlations between the chemical composition of plant cuticles and their biological functions. 137 5.5 References [1] Fischer RA. Journal of Experimental Botany 1973, 24, 387-399. [2] Meidner H and Edwards M. Plant Cell and Environment 1996, 19, 503-503. [3] Wen M, Au J, Gniwotta F, and Jetter R. Phytochemistry 2006, 67, 2494-2502. [4] Riederer M and Schreiber L. Journal of Experimental Botany 2001, 52, 2023-2032. [5] Barnes JD, Percy KE, Paul ND, Jones P, McLaughlin CK, Mullineaux PM, Creissen G, and Wellburn AR. Journal of Experimental Botany 1996, 47, 99-109. [6] Reicosky DA and Hanover JW. Plant Physiology 1978, 62, 101-104. [7] Carver TLW and Thomas BJ. Plant Pathology 1990, 39, 367-375. [8] Carver TLW, Thomas BJ, Ingerson-Morris SM, and Roderick HW. Plant Pathology 1990, 39, 573-583. [9] Eigenbrode SD and Espelie KE. Annual Review of Entomology 1995, 40, 171-194. [10] Kunst L and Samuels AL. Progress in Lipid Research 2003, 42, 51-80. [11] Christie WW. Lipid Analysis; 3rd ed.; The Oily Press: Bridgewater, 2003. [12] Jetter R and Schaffer S. Plant Physiology 2001, 126, 1725-1737. [13] Jetter R, Schaffer S, and Riederer M. Plant Cell and Environment 2000, 23, 619-628. [14] Perkins MC, Roberts CJ, Briggs D, Davies MC, Friedmann A, Hart CA, and Bell GA. Planta 2005, 221, 123-134. [15] Canet D, Rohr R, Charnel A, and Guillain F. New Phytologist 1996, 134, 571-577. [16] Mechaber WL, Marshall DB, Mechaber RA, Jobe RT, and Chew FS. Proceedings of the National Academy of Sciences of the United States of America 1996, 93, 4600-4603. [17] Benitez JJ, Matas AJ, and Heredia A. Journal of Structural Biology 2004, 147, 179-184. [18] Yang HS, An HJ, Feng GP, and Li YF. LWT-Food Science and Technology 2005, 38, 571-577. [19] Veraverbeke EA, Lammertyn J, Nicolai BM, and Irudayaraj J. Journal of Agricultural and Food Chemistry 2005, 53, 1046-1051. [20] Ribeiro da Luz B. New Phytologist 2006, 1 72 , 305-318. [21] Merk S, Blume A, and Riederer M. Planta 1998, 204, 44-53. [22] Greene PR and Bain CD. Colloids and Surfaces B: Biointerfaces 2005, 45, 174-180. [23] Genet MJ, Jacques C, Mozes N, Van Hove C, Lejeune A, and Rouxhet PG. Surface and Interface Analysis 2002, 33, 601-606. 138 [24] Round AN, Yan B, Dang S, Estephan R, Stark RE, and Batteas JD. Biophysical Journal 2000, 79, 2761-2767. [25] Gilly C, Rohr R, and Charnel A. Annals of Botany 1997, 80, 139-145. [26] Hendra PJ, Jones CH, and Warnes GM. Fourier Transform Raman Spectroscopy Instrumentation and Chemical Applications; Ellis Horwood: Chichester, 1991. [27] Boyd RW. Nonlinear Optics; Academic Press: New York, 1992. [28] Juniper BE and Jeffree CE. Plant Surfaces; Edward Arnold: London, 1983. [29] Martin JT and Juniper BE. The Cuticles of Plants; Edward Arnold: London, 1970. [30] Orgell WH. Plant Physiology 1955, 30, 78-80. [31] Yamada Y, Wittwer SH, and Bukovac MJ. Plant Physiology 1964, 39, 28-32. [32] Petracek PD and Bukovac MJ. Plant Physiology 1995, 109, 675-679. [33] Riedel M, Eichner A, and Jetter R. Planta 2003, 218, 87-97. [34] Savitzky A and Golay MJE. Analytical Chemistry 1964, 38, 1627-1639. [35] Brennan JF, Romer TJ, Lees RS, Tercyak AM, Kramer JR, and Feld MS. Circulation 1997, 96, 99-105. [36] Ensikat HJ, Neinhuis C, and Barthlott W. International Journal of Plant Sciences 2000, 161, 143-148. [37] Dragota S and Riederer M. Annals of Botany 2007, 100, 225-231. [38] Viougeas MA, Rohr R, and Charnel A. New Phytologist 1995, 130, 337-348. [39] Franke R, Briesen I, Wojciechowski T, Faust A, Yephremov A, Nawrath C, and Schreiber L. Phytochemistry 2005, 66, 2643-2658. [40] Riederer M and Schtinhen J. Ecotoxicology and Environmental Safety 1984, 8, 236-247. [41] Schreiber L and Schtinherr J. Planta 1990, 182, 186-193. [42] Landais P and Rochdi A. Energy & Fuels 1990, 4, 290-295. [43] Cheng JX and Xie XS. Journal of Physical Chemistry B 2004, 108, 827-840. [44] Boyd RW. Nonlinear Optics; 2nd ed.; Academic Press: San Diego, 2003. [45] Levenson MD and Kano SS. Introduction to Nonlinear Laser Spectroscopy; Academic Press: San Diego, 1988. [46] Mukamel S. Principles of Nonlinear Optical Spectroscopy; Oxford University Press: New York, 1995. [47] Schreiber L and Riederer M. Oecologia 1996, 107, 426-432. 139 [48] Krauss P, Markstadter C, and Riederer M. Plant Cell and Environment 1997, 20, 1079- 1085. [49] Caldwell MM in Encyclopedia of Plant Physiology; Lange OL, Nobel CB, Osmond H and Ziegler H, Eds.; Springer Verlag: New York, 1981; Vol. 12A, pp 169-197. 140 CHAPTER 6 CONCLUSION AND FUTURE DIRECTIONS The goal of this research was to determine whether Raman microspectroscopy is capable of providing useful information about the little-known spatial distribution of chemical compounds found on the surfaces of leaves. This goal has been achieved not only with linear Raman microspectroscopy, but also with the use of two non-linear optical imaging techniques, all giving rise to consistent trends, as described in Chapter 5. The earlier chapters of this thesis described the methods for post-acquisitional data processing, as well as the validation of the experimental and data analysis methods to be performed. The main findings and contributions to science and technology embodied in each chapter, starting with the baseline removal techniques to the determination of triterpenoid distribution, are presented below. 6.1 Concluding Remarks One of the major problems encountered during the data analysis was the large fluorescence baseline present in all of the spectra. Hence, this research also critically evaluated the different types of baseline correction techniques to determine which method is suitable for automation, while still providing acceptable results. It was found that the results from using the different techniques varied due to their dependence on the signal-to-noise ratio (SNR), the signal-to-baseline ratio (SBR), the slope of the baseline, and the amount of spectral congestion. Although the manual correction of baselines is still thought to be the "gold standard" for baseline removal, this proved to be an impractically time-consuming method when dealing with a large data set. Thus, taking the data manipulation and processing times into consideration, the signal removal method was found to be the best automated baseline removal technique and was therefore chosen for all the data processing done in this work. In the process of evaluating the optimal method for the specific needs of this research, a critical review of the relative strengths and weaknesses of all the most common methods were made available [1, 2] to spectroscopists in other fields to aid their selection of appropriate data processing methods. 141 When multiple laser sources are available, it is wise to first determine which wavelength provides the best spectral information of the sample in question. For the analysis of the triterpenoids and very long chain aliphatic compounds, the 785 nm NIR laser source produced the highest quality spectra with relatively better SNR in the fingerprint region of the spectrum, compared to those obtained with the 514 nm Argon laser or the 633 nm HeNe laser. However, even with the NIR laser source, the spectra of the very long chain aliphatic compounds were still found to be indistinguishable from each other, making the identification and quantification of these compounds very difficult or impossible without more sophisticated signal processing and data analysis techniques. The triterpenoids, on the other hand, with their cyclic components and various functional groups, were clearly distinguishable. In fact, their spectra were found to be sufficiently different to allow for the quantification of relative concentrations of the different compounds in mixtures of triterpenoids. The data analysis techniques evaluated in Chapter 3 and developed and validated in Chapter 4 were put to use to provide in situ lateral information on the triterpenoid distribution on isolated cuticles of mature Prunus laurocerasus (Laurel cherry) leaves. This species was chosen as the model species for this research since the chemical composition of the cuticle has been well characterized by various other methods [3-5]. Linear NIR Raman, CARS, and THG microscopic analyses of isolated cuticles have been shown to provide complementary, as well as contributing new, information to GC-MS, XPS, and SEM. However, what distinguishes the techniques used in this research and makes them more powerful is their ability to perform in situ analysis thus providing chemical information with high spatial resolution. The chemical images produced by mapping the intensity of the baseline-corrected spectra showed that the concentrations of the triterpenoids and the very long chain aliphatic compounds were greatest along the anticlinal areas in the isolated cuticles. This was observed by performing point mapping experiments using linear NIR Raman and CARS spectroscopy. This concentration gradient was found to be due to the absolute concentration of these compounds. The decomposition analysis showed that the relative distribution of the triterpenoid and the very long chain aliphatic compounds along the different regions in the cuticle were in fact not the same as their absolute concentrations. This is an important finding since it shows that the intensity of the raw peaks cannot be solely used to determine the distribution of a particular 142 compound when analyzing thin samples. The decomposition maps of the relative distribution of the different compounds using linear NIR Raman spectroscopy shows that the triterpenoids accumulate at the periclinal regions while the very long chain aliphatic compounds are more or less evenly distributed along the adaxial surface. A similar trend was observed in the abaxial surface except that the triterpenoids seem to accumulate to a greater amount over the cuticle of the guard cells, while no very long chain aliphatic compounds were detectable in these areas. Furthermore, THG results showed a higher THG signal over the periclinal regions indicating heterogeneity over these regions. Combining the information obtained in this work with that already known about the transversal distribution of the triterpenoid and very long chain aliphatic compounds, as well as the UV-Vis absorption properties of the cuticles, it can be hypothesized that the presence of triterpenoids in the periclinal areas of the cuticles provide further protection to the underlying cells from harmful agents like UV rays, insects, and/or pathogens. 6.2 Future Directions Previously, oleanolic acid and ursolic acid were distinguished from each other using two- dimensional (2D) correlation spectroscopy on their Raman spectra without the use of a confocal Raman microscope [6]. Instead, Mello et al. dissolved pure oleanolic acid and ursolic acid in successive increments of dimethyl sulfoxide (DMSO) and collected the Raman spectra using a low resolution Raman spectrometer with approximately 15 cm -1 spectral resolution. As they were using a low-resolution spectrometer, the spectra of oleanolic acid and ursolic acid were indistinguishable; several peaks that were identified in Chapter 4 were not observed in their spectra. However, from their 2D asynchronous spectra, which give information about out-of- phase spectral variations, they were able to differentiate the two compounds and determine the different degrees of solvation with DMSO. Their technique only focused on studying the two triterpenoids individually, and no analyses were done using their technique to distinguish the isomers in a mixture. Thus, although the research done by Mello et al. provided important information about the steric hindrance above the rings of the triterpenoids, the decomposition method described in this thesis provides a better means, both qualitatively and quantitatively, for effectively distinguishing oleanolic acid and ursolic acid for in situ cuticular analysis. Nevertheless, it is still quite remarkable that two structurally similar isomers could be 143 distinguished by 2D correlation spectroscopy from their low-resolution spectra. Conceivably, with the comparably high-resolution Raman spectra obtained by Raman microspectroscopy, 2D correlation spectroscopy could potentially provide additional information about triterpenoid conformation and their microenvironment. When applying new techniques or when determining the improvement of a certain analytical method for the analysis of biological specimens, it is helpful to do the analysis on a model species. In this research, P. laurocerasus was used as the model species since the structure and composition of their cuticle has been well studied [3-5, 7]. The chemical imaging and relative quantification described in this research provided further information about the localization of the different cuticular compounds. Thus, in order to determine whether the results obtained here are observed in other species, the next step would be to use these techniques on the analysis of other plant cuticles. Preliminary studies on isolated adaxial cuticles of Ficus elastica (Rubber tree), Garcinia spicata (Garcinia), and Tetrastigma voinierianum (Chestnut pine) leaves using NIR Raman microspectroscopy have been performed. However, the peaks belonging to the triterpenoids in the spectra of the abovementioned cuticles were not as clearly identifiable as those found in the P. laurocerasus cuticle spectra. The reason behind this observation is not known and thus it would be appropriate to investigate the cause of this issue, which may also require a more considerable effort in the signal processing step. Aside from studying isolated cuticles, it is worthwhile to study intact P. laurocerasus leaves and do in vivo analyses. However, a problem that one may encounter when working with whole leaves is the presence of peaks arising from cellular components in the leaf. These peaks may come from chlorophyll, cellulose, proteins, and nucleic acids among other things. In this case, the confocal capabilities of the Raman microscope become crucial. While it is true that optical sectioning can be performed using a confocal Raman microscope, it has been observed that the degree of instrument confocality varies with the type of objectives used [8-11]. For example, dry metallurgical objectives work well when focused on the surface of the sample. However, they are rated as being the worst choice, compared to oil and water immersion objectives, for depth resolution and intensity due to the spherical aberration of the laser intensity distribution when performing an axial scan [10]. In fact, several studies have shown that uncorrected dry metallurgical objectives produce a reduced depth scale [8, 9, 11]. Oil or water 144 immersion objectives may thus be better suited for confocal depth profiling experiments. For cuticular waxes, however, application of the immersion oil directly onto the sample may not be suitable because it may alter the sample and/or the Raman spectrum of the oil may interfere with the Raman spectrum of the sample. For this reason, water immersion objectives may be more desirable since, although not as good as oil immersion objectives, water immersion objectives are comparably better than the metallurgical ones for confocal depth profiling experiments [8, 9]. An alternative approach for performing depth profiling experiments is to analyze a physical cross-section of the sample. It was stated in Chapter 5 that due to the fragility of the sample, a cross-section of the isolated cuticle was only obtained by cryo-section. It may, however, be practical to design and construct a sample mount that would enable the cuticle to be tightly secured without causing sample deformation, thus allowing a clean cross-sectional cut to be made without further sample preparation. Ideally, this sample mount could easily be affixed to the motorized stage of the microscope. In terms of non-linear spectroscopy, a technique employing microstructure fiber is presently being developed in our group for biological analysis using two-photon luminescence microscopy. As this is still in its preliminary stages of development, the utility of this non-linear technique for the analysis of plant cuticles is not yet known. However, our preliminary results show that the isolated cuticles of F. elastica leaves exhibit this two-photon luminescence. Further investigation is required for the proper interpretation of this result. 6.3 Final Remark Raman spectroscopy and non-linear optical imaging techniques were used in this study for the microscopic chemical analysis of leaf cuticles. The findings in this thesis show how valuable interdisciplinary research can be by bringing together experts from different fields and combining ideas and expertise/experience. As an analytical chemistry student, I never imagined I would be studying botany. Now, I look forward to reading about all the new discoveries in plant sciences, especially those made possible by spectroscopic techniques. 145 6.4 References [1] Jirasek A, Schulze G, Yu MML, Blades MW, and Turner RFB. Applied Spectroscopy 2004, 58, 1488-1499. [2] Schulze G, Jirasek A, Yu MML, Lim A, Turner RFB, and Blades MW. Applied Spectroscopy 2005, 59, 545-574. [3] Jetter R, Schaffer S, and Riederer M. Plant Cell and Environment 2000, 23, 619-628. [4] Jetter R and Schaffer S. Plant Physiology 2001, 126, 1725-1737. [5] Perkins MC, Roberts CJ, Briggs D, Davies MC, Friedmann A, Hart CA, and Bell GA. Planta 2005, 221, 123-134. [6] Mello C, Crotti AEM, Vessecchi R, and Cunha WR. Journal of Molecular Structure 2006, 799, 141-145. [7] Krauss P, Markstadter C, and Riederer M. Plant Cell and Environment 1997, 20, 1079- 1085. [8] Everall NJ. Applied Spectroscopy 2000, 54,1515-1520. [9] Everall NJ. Applied Spectroscopy 2000, 54, 773-782. [10] Everall N, Lapham J, Adar F, Whitley A, Lee E, and Mamedov S. Applied Spectroscopy 2007, 61, 251-259. [11] Baldwin KJ and Batchelder DN. Applied Spectroscopy 2001, 55, 517-524. 146

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0061758/manifest

Comment

Related Items