Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Challenges in the spectrochemical analysis of complex materials Chen, Zhiwen 2015

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

Item Metadata

Download

Media
24-ubc_2015_may_chen_zhiwen.pdf [ 7.73MB ]
Metadata
JSON: 24-1.0167681.json
JSON-LD: 24-1.0167681-ld.json
RDF/XML (Pretty): 24-1.0167681-rdf.xml
RDF/JSON: 24-1.0167681-rdf.json
Turtle: 24-1.0167681-turtle.txt
N-Triples: 24-1.0167681-rdf-ntriples.txt
Original Record: 24-1.0167681-source.json
Full Text
24-1.0167681-fulltext.txt
Citation
24-1.0167681.ris

Full Text

Challenges in the SpectrochemicalAnalysis of Complex MaterialsbyZhiwen ChenB.Sc., Nankai Univeristy, 2008A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Chemistry)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)February 2015© Zhiwen Chen 2015AbstractRaman and infrared spectroscopy offer characteristic molecular vibrational information that en-ables a rapid quantitative and qualitative analysis of many types of samples. Easy to operate andrequiring little sample preparation, these techniques offer great potential for the classificationand quantification of complex materials. The research described in this thesis has sought toapply vibrational spectroscopy and multivariate data analysis to tackle a variety of challenginganalytical problems.In vitro fertilization has relied purely on embryo morphological appearance to select viableembryos. We explore the potential of Raman spectroscopy to profile embryo metabolism. Byanalyzing blank culture media, patient samples and bacteria spent media, we establish thatRaman spectroscopy does not offer sufficient sensitivity to differentiate used culture media fromcontrol. Even using liquid core Teflon-AF 2400 fibre to enhance the Raman signal of aqueoussolution, analytical information still lies beneath the sensitivity limit.Turning to a classification problem relative to variance on a larger scale, we investigate oliveoil as a complex biomaterial. The adulteration of extra virgin olive oil with cheaper vegetableoils presents a serious food integrity problem. We demonstrate Raman spectroscopy can detectcorn, canola, grape seed and walnut oil in extra virgin olive oils from various countries andcultivars, but only at levels greater than 20%. This contrasts with conclusions of many limitedstudies, suggesting Raman spectroscopy reliably detects a 5% adulterant. Our analysis showsthat such high sensitivity relies on olive oils limited to a specific geographic region or cultivar.Bleached kraft pulps represent important economic resources. By referring to wet chem-istry, we apply infrared spectroscopy to study alkaline treated bleached eucalyptus kraft pulp.Infrared spectroscopy shows how alkaline treatment modifies hardwood pulp structure. It alsoclassifies bleached hardwood pulps based on species. Despite the natural biological variancepresented by this material, we establish that spectroscopic analysis can accurately quantify thecontents of hemicelluloses in a large variety of bleached kraft pulps (softwood, hardwood andiiAbstracttheir mixture) in industry.iiiPrefaceI joined in Dr. Ed Grant’s laboratory when Raman project had just started. I partially con-structed and set up the Raman system, wrote a custom LabVIEW program to control the Ramanspectrometer, and also did all the trouble shooting.Chapter 3 is based on the work conducted in Children′s and Women′s Health Centre ofBritish Columbia and Dr. Grant’s laboratory at the University of British Columbia. Dr. AnthonyP. Cheung and Dr. Grant designed the experiment. Hospital physicians performed in vitro fer-tilization routine procedures in Children′s and Women′s Health Centre of British Columbia.Alaleh A. Roodsari and I collected the patient culture media during a period of six months. Ideveloped the Raman methodology for the analysis of culture media, measured the Raman spec-tra for all the samples, and analyzed the data. University of British Columbia Research Ethicsapproved the use of human embryo spent culture media from in vitro fertilization treatment(Certificate Number: H09-00420).In Chapter 4 and 5, Dr. Grant helped me identify the projects. I designed the study anddeveloped the Raman methodologies for the analysis of aqueous solutions and vegetable oilsrespectively. Later, I collected all the Raman spectra and analyzed the data by myself.Chapter 6 and 7 are based on the work conducted in FPInnovations and the University ofBritish Columbia. I identified the project and Dr. Thomas Hu helped me design the study. I con-ducted all the wet chemistry experiments for pulp samples in FPInnovations, took the infraredspectroscopic measurement at Shared Instrument Facility at the Department of Chemistry andanalyzed all the spectra.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Outline of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Vibrational spectroscopy and multivariate analysis . . . . . . . . . . . . . . . . . . . 62.1 Overview of non-invasive vibrational spectroscopy . . . . . . . . . . . . . . . . . 62.2 Raman spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.3 Fourier transform Infrared spectroscopy . . . . . . . . . . . . . . . . . . . . . . . . 142.4 Preprocessing methods and multivariate analysis . . . . . . . . . . . . . . . . . . . 162.4.1 Preprocessing methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.4.2 Multivariate analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Metabolomic analysis of human embryo culture media by Raman spectroscopy . . . . 253.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253.2 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29vTable of Contents3.2.1 Reagents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.2.2 Samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.2.3 Instrumentation and Raman measurements . . . . . . . . . . . . . . . . . . 303.2.4 Spectral preprocessing and multivariate analysis . . . . . . . . . . . . . . . 313.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.3.1 Metabolites in embryo culture media . . . . . . . . . . . . . . . . . . . . . 313.3.2 Analysis of glucose aqueous solution . . . . . . . . . . . . . . . . . . . . . . 323.3.3 Analysis of IVF G-1 medium spiked with glucose and sodium lactate so-lutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.3.4 Analysis of IVF patient samples . . . . . . . . . . . . . . . . . . . . . . . . 363.3.5 Analysis of the culture media spent by the bacteria Bacillus . . . . . . . . 393.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434 Raman analysis of aqueous solutions with Teflon-AF 2400 fibre. . . . . . . . . . . . . 444.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444.2 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464.2.1 Reagents and samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464.2.2 Instrumentation and Raman measurements . . . . . . . . . . . . . . . . . . 464.2.3 Spectral preprocessing and multivariate analysis . . . . . . . . . . . . . . . 484.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 494.3.1 Raman analysis of deionized water . . . . . . . . . . . . . . . . . . . . . . . 494.3.2 Raman analysis of aqueous solutions in 240 µm i.d. Teflon-AF 2400 waveg-uide fibre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.3.3 Raman analysis of aqueous solution in 110 µm i.d. Teflon-AF s2400 waveg-uide fibre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595 Detection of the adulteration of vegetable oil in extra virgin olive oil by Raman spec-troscopy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60viTable of Contents5.2 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 625.2.1 Instrumentation and Raman measurements . . . . . . . . . . . . . . . . . . 625.2.2 Oil samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 635.2.3 Spectral preprocessing and multivariate analysis . . . . . . . . . . . . . . . 645.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645.3.1 Raman analysis of pure vegetable oils . . . . . . . . . . . . . . . . . . . . . 645.3.2 PCA analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.3.3 PCA-LDA analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 766 Infrared analysis of alkaline treated bleached eucalyptus kraft pulps. . . . . . . . . . 786.1 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 806.1.1 Samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 806.1.2 Methods and Instrumentations . . . . . . . . . . . . . . . . . . . . . . . . . 816.2 Results and discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 826.2.1 Peak assignment in ATR-FTIR spectra . . . . . . . . . . . . . . . . . . . . . 826.2.2 Wet chemistry analysis of NaOH treated eucalyptus pulps . . . . . . . . . 866.2.3 ATR-FTIR analysis of bleached eucalyptus kraft pulps treated by NaOHsolution of less than 4 percent . . . . . . . . . . . . . . . . . . . . . . . . . 886.2.4 ATR-FTIR analysis of bleached eucalyptus kraft pulps treated by NaOHsolution of greater than 4 percent . . . . . . . . . . . . . . . . . . . . . . . 906.2.5 Univariate and multivariate analysis of FTIR-ATR spectra and xylan con-tent in bleached kraft pulp . . . . . . . . . . . . . . . . . . . . . . . . . . . 936.2.6 Alkaline treatment of bleached eucalyptus pulps as mercerization . . . . . 956.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 957 Quantification of hemicelluloses in bleached kraft pulps by infrared spectroscopy. . . 977.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 977.2 Experimental . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 997.2.1 Samples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 997.2.2 Methods and instrumentations . . . . . . . . . . . . . . . . . . . . . . . . . 1007.2.3 Data processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100viiTable of Contents7.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027.3.1 Characteristic bands in ATR-FTIR spectra . . . . . . . . . . . . . . . . . . . 1027.3.2 PCA analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057.3.3 PLS analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1087.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1107.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1128 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116viiiList of Tables3.1 The summary of non-invasive techniques used to study embryo spent medium asa predictor of human embryo development and viability. . . . . . . . . . . . . . . 283.2 Analytical figure of merits for the PLS model of 0-10 mM glucose solutions usingNAS analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344.1 Summarized PLS calculation from validation data based on the Raman spectraacquired with 110 µm i.d. Teflon fibre and 1 cm cuvette . . . . . . . . . . . . . . . 554.2 The summary of Teflon enhanced Raman spectroscopy. . . . . . . . . . . . . . . . . 565.1 The summary of pure vegetable oils. . . . . . . . . . . . . . . . . . . . . . . . . . . 635.2 The summary of pure extra virgin olive oils. . . . . . . . . . . . . . . . . . . . . . . 645.3 Assignment of observed bands in Raman spectra of vegetable oils. . . . . . . . . . 655.4 Linear classification between authentic and adulterated extra virgin olive oils byPLS-LDA method. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 736.1 The summary of the studied pulp samples. . . . . . . . . . . . . . . . . . . . . . . . 816.2 Assignment of observed ATR-FTIR bands to bleached hardwood kraft pulp (HW),bleached softwood kraft pulp (SW), cellulose and xylan. . . . . . . . . . . . . . . . 866.3 The summary of calculation results based on PLS and univariate analysis with thefeature at 964 cm−1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 947.1 The summary of the studied samples. . . . . . . . . . . . . . . . . . . . . . . . . . . 1007.2 Assignment of major FTIR-ATR bands for bleached hardwood kraft pulp (HW),bleached softwood kraft pulp (SW) and their mixture. . . . . . . . . . . . . . . . . 1047.3 The summary of PLS calculation results based on 100 models. . . . . . . . . . . . 108ixList of Figures2.1 IR absorption, Raman scattering and Fluorescence. . . . . . . . . . . . . . . . . . . 62.2 Scheme diagrams for dispersive Raman and FT-Raman systems. . . . . . . . . . . 92.3 Conventional and confocal Raman spectroscopy along z axis. . . . . . . . . . . . . 102.4 Scheme diagram for CARS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.5 Various FTIR sampling techniques. (A)transmission FTIR, (B) ATR, (C) PAS and(D) DRIFT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.6 Raman spectra of cooking oils. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.7 First-derivative preprocessing method. . . . . . . . . . . . . . . . . . . . . . . . . . 172.8 Second-derivative preprocessing method. . . . . . . . . . . . . . . . . . . . . . . . 182.9 Iterative polynomial fitting preprocessing method. . . . . . . . . . . . . . . . . . . 182.10 Symlet6 wavelet vs sine filter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.11 Discrete wavelet transform preprocessing method. . . . . . . . . . . . . . . . . . . 203.1 The development of single human embryo from Day 0 to Day 6 during IVF. Thesize of human embryo cell stays unchanged during Day 1 to Day 6. http://medicine.yale.edu/obgyn/yfc/ourservices/invitro/development.aspx Septem-ber 2014. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263.2 (A) Normalized raw spectra of (−) water and (−) glucose aqueous solutions, (B)Normalized DWT spectra of (−) water and (−) glucose aqueous solutions, (C) Plotof reference glucose concentrations versus PLS predicted concentrations ((o) cal-ibration and (o) validation samples), (D) Plot of reference glucose concentrationsversus Net Analyte Signal (NAS) of glucose ((o) calibration and (o) validation sam-ples). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32xList of Figures3.3 (A) Normalized raw and DWT spectra of (−) water, (−) blank G-1 medium, andG-1 media spiked with glucose aqueous solution; (B) Normalized raw and DWTspectra of (−) water, (−) blank G-1 medium, and G-1 media spiked with sodiumlactate aqueous solution, (C) Plot of NAS versus the reference concentrations ofthe added glucose ((o) calibration and (o) validation data), (D) Plot of NAS versusthe reference concentrations of the added sodium lactate ((o) calibration and (o)validation data). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353.4 (A) Normalized DWT spectra of IVF (−) G-1 and (−) G-2 control media; (B) PCAscores plot of (o) G-1 and (o) G-2 control media based on PC1 and PC2. . . . . . . 373.5 (A) Normalized DWT spectra of (−) G-1 control and (−) samples obtained on Day3, (B) Normalized DWT spectra of (−) G-2 control and (−) samples obtained onDay 6, (C) PCA scores plot of (o) G-1 control and (∗) samples obtained on Day 3,(D) PCA scores plot of (o) G-2 control and (♦) samples obtained on Day 6. . . . . . 383.6 (A) Normalized raw and DWT spectra of (−) blank LB media and (−) bacillus spentLB media collected at different time points during culture, (B) Bacillus growthcurve based on OD600. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.7 (A) PCA scores plot of bacillus spent LB media based on PC1 and PC2. (o) LBmedia collected during its lag phase, (M) LB media collected from its log phase,(♦) LB media collected from bacillus stationary phase, (B) OD600 predicted by PLSversus the reference values ((o) calibration and (o) validation data). . . . . . . . . . 403.8 (A) Raman spectra of the culture media obtained on Day 3 from 36 embryos (16patients). (B) Mean-centered Raman spectra of the embryos, which result in adelivery in comparison with those who did not implant. These two graphs arepublished by Seli et al[193]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 424.1 Schematic drawing of two sample cells used in this study. (A) Teflon waveguidefibre mounted in aluminum block with finger tight HPLC fitting, (B) 1cm cuvettemade of aluminum. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474.2 The setup of Teflon Raman spectroscopy. (A) Microscope camera image of 110 µmi.d. Teflon tubing, (B) 240 µm i.d. Teflon tubing filled with 20% ethanol solutionunder exposure to 100 mW 785 nm laser . . . . . . . . . . . . . . . . . . . . . . . . 48xiList of Figures4.3 Raw Raman spectra of water collected with conventional Raman spectroscopy via(−) the single lens objective, (−) 10x objective and (−) 50x objective . . . . . . . . . 494.4 Raman spectra of aqueous solutions acquired with a 240 µm i.d. Teflon fibre and1 cm cuvette by 100 mW laser. (A) 20% ethanol for 1 s exposure, (B) water for10 s exposure (water spectrum collected by cuvette is multiplied by a factor of 3for comparison), (C) 100 mM sodium lactate solution for 10 s exposure (100 mMlactate spectrum collected by cuvette is multiplied by a factor of 3 for compari-son), (D) 0.5-200 mM sodium lactate solutions collected with Teflon fibre for 10 sexposure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524.5 Raman spectra of 20% ethanol solution collected with (−) 1 cm cuvette, (−) 110µm i.d. Teflon fibre and (−) 240 µm i.d. Teflon fibre by 100 mW 785 nm laser for1 s exposure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534.6 Normalized DWT Raman spectra of 1-10 mM sodium lactate solutions obtainedwith (−) a 28 cm long 110 µm i.d. Teflon fibre and (−) 1 cm cuvette by 100 mW785 nm laser for 300 s exposure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544.7 Normalized DWT Raman spectra of IVF G-1 media acquired with (−) 110 µm i.d.Teflon fibre and 10X objective in comparison with that (−) acquired with a cuvetteand single lens . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555.1 Normalized DWT transformed Raman spectra of pure vegetable oils after DWTand normalization to 1441 cm−1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 655.2 PCA scores plot of pure vegetable oils based on (A) PC1 vs. PC2 and (B) PC1vs. PC3. (1 q)canola oil, (2 I)corn oil, (3 )sunflower oil (high oleic acid), (4F)extra virgin olive oil, (5 )peanut oil, (6 )grape seed oil, (7 )safflower oil and(8F)walnut oil. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.3 PCA loading plot of PC1, PC2 and PC3 compared with Raman spectra of purevegetable oils. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 675.4 Three-dimensional PCA scores plot of various extra virgin olive oils. ( )ITA-Italy,(q)SPA-Spain, ( )USA- USA, (F)PER-Peru, (N)TUN-Tunisia, ()GRE-Greece, (H)POR-Portugal and ()AUS-Australia. . . . . . . . . . . . . . . . . . . . . . . . . . . 68xiiList of Figures5.5 Two-dimensional PCA scores plot of (5 I) authentic extra virgin olive oils andsamples adulterated with (1 q)-5%, (2 )-10%, (3 )-20%, (4F)-50% (A) canola,(B) corn, (C) grape seed, and (D) walnut oil. 5% canola and corn adulteratedsamples are not plotted, owing to their serious overlapping with authentic samples. 706.1 General structural formulae for cellulose[274]. . . . . . . . . . . . . . . . . . . . . 796.2 General structural formulae for galactoglucomannans in softwood[274]. . . . . . . 796.3 General structural formulae for arabinoglucuronoxylan in softwood[274]. . . . . . 796.4 General structural formulae for glucuronoxylan in hardwood[274]. . . . . . . . . . 796.5 Raw ATR-FTIR spectra of cellulose, birch xylan, bleached eucalyptus kraft pulp(HW) and bleached softwood kraft pulp (SW). . . . . . . . . . . . . . . . . . . . . . 836.6 Normalized second derivative ATR-FTIR spectra of cellulose, birch xylan, bleachedeucalyptus kraft pulp (HW) and bleached softwood kraft pulp (SW). . . . . . . . . 836.7 NaOH concentration plotted against (A) the dissolved carbohydrates (mainly xy-lan) in extraction filtrate measured by titration method and (B) the remainingxylan in pulp quantified by HPLC. . . . . . . . . . . . . . . . . . . . . . . . . . . . 876.8 PCA analysis of normalized second derivative ATR-FTIR spectra from all alkalinetreated samples. ( ) control bleached eucalyptus kraft pulp; ( ) samples treatedby less than 4 percent NaOH; ( ) samples treated by higher than 4 percent NaOH. 886.9 Normalized second derivative ATR-FTIR spectra of NaOH treated bleached euca-lyptus pulp in spectra region of (A) 870-1200 cm−1 and (B)1200-1500 cm−1. Eu-calyptus pulp treated with (-) 0%, (-) 0.5%, (-) 1.0%, (-) 1.5%, (-) 2.0%, (-) 2.5%,(-), 3.0% and (-) 3.5% NaOH. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 896.10 Normalized second derivative ATR-FTIR spectra of NaOH treated bleached eu-calyptus pulp in spectra region of (A) 870-1200 cm−1 and (B)1200-1500 cm−1.Eucalyptus pulp treated with (−) 0.0%, (−) 1.0%, (−) 2.0%, (−) 3.0%, ( ) 4.0%,( ) 4.5%, ( ) 5.0%, ( ) 5.5% and ( ) 6.0% NaOH. . . . . . . . . . . . . . . . . 916.11 Linear relationship between the absorbance intensity at 964 cm−1 and relativexylan content in pulp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94xiiiList of Figures7.1 Normalized ATR-FTIR spectra of (–) extracted birch xylan, (–) bleached softwoodpulps, (–) bleached hardwood pulps, and (–) mixtures of bleached softwood andhardwood pulps obtained from different kraft mills. . . . . . . . . . . . . . . . . . 1027.2 Normalized DWT FTIR spectra of (–) extracted birch xylan, (–) bleached softwoodpulps, (–) bleached hardwood pulps and (–) mixtures of bleached softwood andhardwood pulps obtained from different kraft mills. . . . . . . . . . . . . . . . . . 1037.3 Normalized DWT transformed ATR-FTIR spectra of bleached eucalyptus kraftpulp treated with (-) 0%, (-) 0.5%, (-) 1.0%, (-) 1.5%, (-) 2.0%, (-) 2.5% and (-)3.0% NaOH. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1057.4 PCA scores plot of normalized DWT transformed ATR-FTIR spectra for com-mercial bleached kraft pulps, laboratory generated pulp mixtures and alkalinetreated pulps. (o), (o) and (o) obtained from kraft mill, (M) NBSK blended withaspen pulp, (M) NBSK blended with maple pulp, (M) NBSK blended with eucalyp-tus pulp, (∗) bleached eucalyptus kraft pulp treated with 0-4% NaOH, (∗) NBSKtreated with 0-6% NaOH. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1067.5 The loading plot of (–) PC1 and (–) PC2 versus normalized DWT transformedATR-FTIR spectra. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1077.6 Normalized raw, DWT and SG-D1 ATR-FTIR spectra of bleached (–) softwood and(–) hardwood kraft pulp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1097.7 Linear relationship between predicted xylan or mannan content by DWT-SNV-PLS and measured concentrations from HPLC carbohydrate analysis. (A) Xylancalibration, (B) xylan validation, (C) mannan calibration, (D) mannan validation. . 1107.8 Selected wavelengths for xylan and mannan from 5 independent runs of GA-PLScalculation. (−) selected wavelengths for mannan and (−) selected wavelengthsfor xylan. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111xivAcknowledgementsI would like to express my sincere gratitude to my supervisor Prof. Ed Grant, who gave memagnificent guidance and support throughout my PhD study. Without your help, I would notbe able to finish this thesis.I would like to thank Dr. Georg Schulze for his great advice and help when I was strugglingwith my data.Special thanks to Dr. Thomas Hu and Dr. Ho Fan Jang, who allowed me to do experiment atFPInnovations. I would also like to thank Dr. Dongbo Yan, Dr. Hao Qi, Micelle Zhao, ShannonHuntley and Youni Kim, who gave me lots of help with the instruments and pulp experiments.I really appreciate my group mates from Prof. Ed Grant’s lab. Dr. Chris Rennick taught mehow to set-up the Raman spectrometer and write LabVIEW program. Dr. Nicolas Saquet, Dr.Jonathan Morrison, Najmeh Tavassoli, Markus Schulz-Weiling, Jachin Hung, Hossein Sadeghi,Alison Bain, Ashton Christy, Mahyad Aghigh and Luke Melo gave me so much encouragementand support throughout the last six years.Last of all, I would like to express my deepest appreciation to my parents and my husband.Without your encouragement, I could not have come so far.xvDedicationThis thesis is dedicated to my parents, my husband and my family. Thanks for your selflesssupport and unconditional love!xviChapter 1IntroductionThe accurate classification and quantification of naturally occurring complex materials presentsome of the most important challenges in the contemporary field of analytical chemistry. Forinstance, confronted with a colon tissue specimen, a practitioner must decide with confidencewhether it is to be classified as normal or carcinoma[1]. In the food industry, vegetable cookingoils market on the basis of their content. Customers expect products that are well classified andregulated. However, the high value of extra virgin olive oil triggers the widespread clandestineadulteration of this product with cheaper vegetable oils[2]. Often, the analysis of a naturalmaterial requires more than a yes/no classification or multi-class identification, and insteadneeds quantification, with the objective of determining precise levels of specific analytes in thesamples. For example, the sugar content often serves as an important criterion in assessing thequality of a food product[3]. For juice[3, 4], honey[5], wine[6, 7] and other products derivedfrom a natural source, food manufacturers are required to certify quality and authenticity byproviding a detailed composition of sugars.In most cases, the traditional solutions of these and other challenging classification andquantification problems have relied upon wet chemistry often followed by chromatographicanalysis. Solid samples require complex preparation to convert them into liquid form beforeanalysis. In an industrial production environment, it is usually far too expensive and time con-suming to characterize every product. As a consequence, problems such as adulteration[8, 9] of-ten go undetected, which can even cause serious consequences for health and well being[10, 11].Sometimes, classification or quantification requires morphological analysis. During treat-ment by in vitro fertilization, practitioners must select and transfer the most viable embryos onthe basis of embryo morphological appearance[12, 13]. This simple decision relies on skilledpractitioners’ visual recognition of embryo features, such as cell numbers, fragmentation, sym-metry etc[12]. The results are thus affected by subjectivity and chance. Even with correct per-ception, the morphological appearance of an embryo does not necessarily reflect embryo chro-1Chapter 1. Introductionmosomal abnormalities[14]. This and other medical analysis, which rely on classification at thecellular level, require a reliable analytical tool to provide fast and accurate diagnosis.Infrared and Raman spectroscopy serve as important analytical techniques in many differ-ent fields. The positions and intensities of spectroscopic features offer characteristic molecularvibrational information, that allow for a quick quantitative and qualitative analysis of sam-ples without complicated sample preparation. They can offer a means to analyze the chemicalcomposition of a solid without physical modification. For this reason, these two techniqueshave proven particularly useful for characterizing complex materials[15–17]. With the advan-tages of little sample preparation, non-destructive analysis, easy operation and fast data collec-tion, infrared and Raman methods have been widely applied in the analysis of food[6, 18, 19],pharmaceutical[20–22], forestry products[23, 24], etc. Infrared spectroscopy, however, is sen-sitive to water and thus has limited utility for aqueous samples or moist solids. Comparedwith infrared spectroscopy, Raman spectroscopy excels in the analysis of water-rich biomedicalsamples, such as biofluids[25, 26], cells[27, 28] and tissues[29, 30].The appealing characteristics of infrared and Raman spectroscopy have motivated manystudies seeking to develop them as alternative tools to solve a variety of natural material classi-fication and quantification problems. In particular, Raman spectroscopy has been proven to pro-vide a 99.9% diagnostic accuracy in the classification of normal, hyperplastic polyps and adeno-carcinomas colon tissue specimens[1]. Differentiation between normal and diseased breast[31],skin[32], or lung[33] tissues have also been demonstrated. Fourier transform infrared spec-troscopy has been proven to quantify the carboxyl content[34], lignin[35], ink concentration[36]in high yield pulp samples. In the food industry, both infrared and Raman spectroscopy havebeen used to distinguish authentic honey or juice samples from those adulterated by the ad-dition of sugar or industry syrup[37, 38]. Many studies have presented cases with these twotechniques, which offer a fast method to achieve an acceptable classification or quantificationresults. In this thesis, I test the extension of Raman and infrared spectroscopy beyond a simpledemonstration stage in an effort to determine whether direct spectroscopic analysis, supportedby sophisticated chemometric data processing, can provide a reliable means to assess real-worldsamples.21.1. Outline of this thesis1.1 Outline of this thesisChapter 2 presents a brief overview of Raman and infrared spectroscopy together with an in-troduction to several multivariate analysis methods commonly used for spectroscopic prepro-cessing and further classification or quantification. Chapters 3 through 7 detail five differentexperimental challenges testing the utility of spectroscopic analysis as a means of classificationand quantification. Chapter 8 summarizes the broad conclusion of this work.In Chapter 3, we apply Raman spectroscopy to classify embryo culture media to predicthuman embryo reproductive viability. Nowadays, many couples experience the problem of in-fertility. This problem is often treated by in vitro fertilization (IVF), as it has the highest successrate among the remedies offered to infertile patients[39, 40]. During each cycle of treatment,practitioners culture human embryos in culture medium for a few days, and then select themost viable embryos on the basis of morphological appearance for transfer to the uterus of themother. Currently IVF succeeds only 30-40% of the cases[40]. In order to prevent failure, physi-cians select and transfer multiple viable embryos simultaneously. However, this practice causesa high incidence of twins and triplets, which results in significant health problems for infants,including an increased rate of preterm delivery or lifelong disabilities[41]. Thus, to increasethe success rate of IVF and reduce the rate of multiple pregnancies, the practice of IVF requiresa reliable method for viable embryo selection. Current clinical methods rely on morphologyscores, which do not necessarily signal intrinsic chromosomal abnormalities[14, 42]. To pre-dict their viability, researchers have used various analytical techniques for the analysis of usedgrowth media to non-invasively study the metabolism of embryos. This chapter offers a criti-cal assessment of Raman spectroscopy as a viable clinical tool for the fast evaluation of embryometabolism and reproductive prognosis.Chapter 4 explores the use of a Teflon liquid core waveguide fibre to enhance the Ramansignal of liquid samples. The inherently weak nature of Raman scattering limits the applicationof spontaneous Raman spectroscopy in dilute biofluid samples. Fibre enhancement offers a wayto amplify the Raman signal for dilute analytes. For this purpose, we use a hollow core capillarymade of Teflon-AF 2400 polymer, which has a refractive index lower than that of water[43, 44].When an excitation laser is focused into such a liquid-filled Teflon fibre with an appropriateacceptance angle, total internal reflection occurs on the liquid-Teflon interface and directs light31.1. Outline of this thesisto propagate along the liquid core fibre. With this increased light path in the liquid core, theTeflon fibre forms a waveguide that significantly enhances Raman signal[45]. In this chapter, wecollect enhanced Raman signal from a liquid-filled Teflon fibre and aim to develop an enhancedRaman system for dilute biofluid samples.Chapter 5 studies the potential of Raman spectroscopy as a fast sensor to identify the adulter-ation of extra virgin olive oil. Olive oil contains many healthy monounsaturated fatty acids andantioxidants, which offer protection from various cancer and cardiovascular diseases[46, 47].Owing to these beneficial nutritional components as well as its pleasant flavor, olive oil has be-come one of the most popular cooking oils, despite its high cost. The high sale price has alsocaused the wholesale adulteration of olive oil with cheaper vegetable oils[2]. Although standardmethods, such as wet chemistry and chromatography, can identify the adulterated samples,these approaches are too expensive and inefficient for routine quality tests of olive oil sampleson the market. Spontaneous Raman spectroscopy suffers from low sensitivity in the analysis ofaqueous solutions, but it yields intense Raman signal for oil samples. In Chapter 5, we demon-strate the effectiveness of Raman spectroscopy for the analysis of a large variety of extra virginolive oils and vegetable oils. Here, we focus on the differentiation of authentic extra virgin oliveoils from those adulterated with cheaper vegetable oil.As vibrational spectroscopy has a far greater effectiveness for the analysis of solid samplescompared with dilute liquid solutions, the next two chapters focus on the application of in-frared spectroscopy as a means to perform classification and quantification for bleached kraftpulp samples. Treatment by alkaline solution chemically modifies pulp. In Chapter 6, we treatbleached eucalyptus kraft pulp with sodium hydroxide solutions of various concentrations. Byapplying infrared spectroscopy and wet chemistry methods, we characterize alkaline treatedbleached eucalyptus kraft pulps and study the effect of sodium hydroxide on the chemical struc-ture of pulp samples. We establish that second derivative spectra preprocessing and multivari-ate analysis serves well to predict the xylan content in these alkaline treated eucalyptus pulpsand other commercial bleached hardwood kraft pulps.Building on this work, Chapter 7 uses infrared spectroscopy to rapidly determine the hemi-celluloses in a wide range of bleached kraft pulp samples. Owing to their high strength and su-perior reinforcement properties, bleached kraft pulps constitute the main materials for a num-ber of widely used paper products. Hemicellulose is the second most abundant component in41.1. Outline of this thesisa pulp, and has a significant effect on its chemical and physical properties. The hemicellulosecontent of a bleached kraft pulp is measured by acid hydrolysis of ground micrometer-sizedsolid pulp samples followed by HPLC analysis of the extract[48]. Here, we explore the use ofattenuated total internal reflected infrared spectroscopy (ATR-FTIR) to quantify the hemicellu-lose content in various bleached kraft pulps, including softwood pulps, hardwood pulps, theirmixtures as well as chemically modified bleached kraft pulps. This wide variety of samples al-lows us to test the wide feasibility of ATR-FTIR as a fast sensor tool for the quantitative analysisof hemicelluloses in various bleached kraft pulp samples.We used the second derivative in Chapter 6 and discrete wavelet transform in Chapter 7 topreprocess the infrared spectra. Both these two preprocessing methods confirm that the spectralvariation displayed at 964cm−1specifically correlates with xylan in the pulp.5Chapter 2Vibrational spectroscopy andmultivariate analysis2.1 Overview of non-invasive vibrational spectroscopyRaman and infrared spectroscopy are well-established vibrational spectroscopic methods. Bothtechniques involve the interaction of light with molecular vibrational modes. Figure 2.1 illus-trates the mechanism of infrared absorption and Raman scattering in terms of the vibrationalenergy levels.S0S101320123IRAbsorptionRayleighScatteringStoke RamanScatteringAnti-Stoke RamanScatteringFluorescenceνvν0 ν0-νvν0 ν0 ν0 ν0+νvFluorescenceVirtual stateFigure 2.1: IR absorption, Raman scattering and Fluorescence.When the energy of the incident photon ν0 corresponds to the transition energy between anyvibrational state and the ground state (ex. νv), then the molecule can absorb the incident radia-62.1. Overview of non-invasive vibrational spectroscopytion. If the incident wavelength falls in the region from 2.5 to 25 µm, mid-infrared absorptionoccurs and the characteristic infrared spectrum appears in the form of absorbance intensity ver-sus wavelength[49]. Unlike infrared absorption, the Raman effect is based on light scattering.By interacting with the incident radiation at frequency ν0, the molecule and photon field forma virtual energy state which relaxes by emitting photons. According to the frequencies of theemitted photons, this radiation can be classified into two categories. One is elastic Rayleigh scat-tering, which has unchanged frequency ν0 and dominates the emitted photon signal. The otheris inelastic Raman scattering, which consists of Stokes scattering with a lower frequency thanthe incident light field, ν0-νv and anti-Stokes scattering, which appears at a higher frequency,ν0+νv[50]. In fact, only 0.0001% of the emitted photons undergo inelastic Raman scattering,leading to the primary issue of Raman spectroscopy - very weak Raman signal[20]. Thus thistechnique has had limited application until the invention of high power lasers and sensitivedetectors[51]. During the Raman process, laser also triggers fluorescence as shown in Figure2.1, which causes another drawback of Raman spectroscopy. In some cases, intense fluorescencecan completely swamp the Raman signal, resulting in the failure of a Raman measurement. Incontrast, infrared spectroscopy does not have this problem.As the vibrational energy levels of each molecule correspond to a specific pattern of stretch-ing and bending motions, the infrared and Raman spectra offer a characteristic fingerprint[52].Although both techniques provide chemical information based on molecular vibrational mo-tions, they sample different molecular responses, and so complement each other. Infrared ab-sorption occurs only when the electronic dipole moment of the molecules changes during amolecular vibration, while the Raman process requires a net change of molecular polarizability[20].For example, a homonuclear diatomic molecule or a molecule with a centre inversion has a zerodipole moment change with vibration but a nonzero polarizability. This allows the Raman pro-cess, but forbids infrared activity[53]. For this reason, Raman spectroscopy has proven a bettertool to study the vibration of quasi-symmetric covalent bond such as C=C, S-S, C-S stretchesand certain aromatic distortions, whereas infrared spectroscopy produces stronger signals inrelation to dipolar and partially ionic bonds, such as O-H, N-H and C=O[54].72.2. Raman spectroscopy2.2 Raman spectroscopyIndian Physicist Sir C. V. Raman and K. S. Krishnan first observed the inelastic scattering ofsunlight from liquid in 1928, and this phenomenon was later named after Raman[51]. Thistechnique has been developed dramatically, since Raman’s first experiment. In particular, theinvention of the photomultiplier tubes and He-Ne lasers in the 1960s drove substantial newefforts in the study of Raman spectroscopy[51]. In 1985, the first 1064 nm Fourier transformRaman spectroscopy (FT-Raman) systems was demonstrated[55]. After Charge Coupled Device(CCD) detectors became available in the 1980s, Renishaw and Dilor introduced the first com-mercial bench-top Raman systems in 1992[55]. Advances in detectors, laser optics, confocalline-scanning etc. contributed the renaissance of Raman spectroscopy through the 1990s.Spontaneous Raman spectroscopyPrimarily two types of Raman spectrometers exist today, classified as dispersive Raman and FT-Raman systems. Dispersive Raman spectrometers link four principal components: a excitationsource, optics, a monochromator and a CCD (Figure 2.2A). A monochromatic continuous-wavelaser with output ranging from ultraviolet to near infrared is commonly used to illuminatethe sample. Argon (488.0 and 514.5 nm), He-Ne (633 nm) and diodes (785 or 830 nm) havebeen frequently used as the laser sources[56]. Among them, the argon laser gives the strongestRaman signal, owing to the fact scattering intensity varies as the inverse fourth power of thewavelength (λ-4)[57]. However, blue and ultraviolet light also produce more fluorescence, whichcan completely suppress and obscure the Raman signal. Choosing an excitation laser in the nearinfrared region significantly decreases the fluorescence[58]. However, a Nd:YAG laser (1064 nm)as used in FT-Raman is inappropriate, as the regular CCD detector is blind beyond the 1100 nmregion, where most Raman scattering from the 1064 nm laser would appear[59]. Thus, thecombination of an excitation laser at 785 nm and a CCD has proven the most popular choice fordispersive Raman spectroscopy. Following the laser source, optics such as notch filter, lens andfibre, filter the laser source and collect the Raman scattering. Then, a monochromator dispersesthe Raman signal by wavelength, for collection by a multichannel CCD as a Raman spectrum.82.2. Raman spectroscopySpectragraphLaserSampleCCDLaserInGaAs InterferometerSpectraFTSpectraSample(A) (B)Figure 2.2: Scheme diagrams for dispersive Raman and FT-Raman systems.A typical FT-Raman system consists of a 1064 nm Nd:YAG laser, optics, an interferome-ter and a single channel photomultiplier tube (PMT) (Figure 2.2B)[60]. With 1064 nm laser,FT-Raman can analyze samples emitting relatively strong fluorescence. For instance, it allowsfor the nondestructive detection of in-situ carotenoids in Japanese green tea, mandarin orange,spinach leaf and hen’s egg yolk[61]. Agarwal et al. have used FT-Raman spectroscopy to studythe chemical structure of lignin and quantify the content of lignin in various wood and pulpsamples[62–65]. In contrast, Raman scattering at all other laser frequencies fail due to the in-terference of intense fluorescence emitted from samples.Unlike a dispersive Raman system, FT-Raman uses an interferometer and a single channeldetector such as Indium gallium arsenide (InGaAs) to record the interferogram as the movingmirror translates at a slow rate[60]. Afterwards, the application of Fourier transform convertsthe interferogram to the corresponding FT-Raman spectrum. FT-Raman spectroscopy has a ma-jor disadvantage in that it requires a much longer acquisition time (in several minutes) thandispersive Raman spectroscopy, owing to the slow translation velocity of the moving mirror andthe low-quantum-efficiency detector. Thus, such instrument uses a high power Nd:YAG laser(1W) in order to acquire a decent FT-Raman spectrum with high signal-to-noise ratio. But thiscan also cause problems, such as photodecomposition or overheating of samples.Dispersive Raman spectroscopy has become more widely used than FT-Raman spectroscopy.Cost reductions in Diode lasers at 785nm, monochromators and CCD detectors have spurred92.2. Raman spectroscopythe application of dispersive Raman spectroscopy. With the appropriate optical componentsand a microscope system, a conventional dispersive Raman methodology can extend to forma confocal Raman microscope. Particularly, a spatial filter ( a small pinhole aperture ), placedin front of the laser source, can aid in the formation of a T EM00 Gaussian beam, so that theincident laser can form a diffraction-limited focal spot on the sample[66]. Using a microscopeto acquire the Raman image can yield a spatial resolution of 1 µm, allowing for a single pointillumination[60, 67]. Exposure to the excitation laser stimulates Raman scattering at the focuson the sample. A second pinhole, inserted in front of the spectrograph, eliminates all the out-of-focus Raman light (Figure 2.3)[67, 68]. Generally, a modern Raman microscope can obtaina depth resolution of 1-2 µm, depending on the the selection of objective, pinhole size, laserwavelength, sample as well as the instrumental setup[69].Conventionaldispersive RamanSpectroscopy Confocal Raman SpectroscopyConfocal pinholeSampleSampleSpectrometerapertureSpectrometerapertureZ axisObjective ObjectiveFigure 2.3: Conventional and confocal Raman spectroscopy along z axis.In another configuration of confocal Raman microscopy, a 50 µm core diameter optical fibreserves as a pinhole while delivering the laser into the Raman system[70, 71]. Then the back-scattered Raman signal is collected by another 100 µm core diameter optical fibre, which also102.2. Raman spectroscopyserves as a second pinhole[70, 71]. Confocal Raman microscope collects the scattered Ramansignal at the sharp focal point, whereas the conventional dispersive Raman spectroscopy col-lects Raman signal within a large depth-of-field. Thus, conventional Raman spectroscopy ismore suitable to analyze liquid solutions or any sample with bulk volume, as the intensity ofRaman signal is proportional to the illuminated volume[57]. On the other hand, confocal Ra-man microscopy has the notable advantage in the analysis of micro-volume samples or sampleswith various depth profiling. It has successfully provided chemical analysis or two dimensionalchemical mapping for samples such as tissue[30, 72, 73], cells[27, 68, 74, 75], mineral[76–78],material[79–81], drugs[82, 83] etc. In contrast, conventional Raman spectroscopy is quite lim-ited in these applications. Both confocal and conventional Raman spectroscopy are very usefultechniques for the analysis of solids or biological samples, and easily provide the chemical in-formation at the spot where the laser is focused. However, the major drawback lies their lowsensitivity. Sometimes it takes hours to acquire a high-quality Raman spectrum or make a two-dimensional map.Microscope used in this work was not confocal, and this presented us with some limitationswhen analyzing system that requires a narrow depth of field, such as a thin biological materialon a slide.Enhanced Raman spectroscopyIn addition to spontaneous Raman scattering, enhanced Raman spectroscopy significantly in-creases Raman signal and has undergone substantial studies. According to the enhancementmechanism, enhanced Raman spectroscopy can be primarily classified into three different cat-egories: resonance Raman spectroscopy (RRS), surface enhanced Raman spectroscopy (SERS)and coherent anti-Stokes Raman spectroscopy (CARS).By tuning the laser near an electronic transition of a specific analyte or chromophoric group,RRS can provide an enhancement factor up to 106 relative to spontaneous Raman spectroscopy[51,84]. For instance, the electronic transition of amide bonds in protein and peptides lies around200 nm[85]. With the laser set in this deep ultraviolet region, RRS greatly enhances the charac-teristic Raman bands of amide I(C=O), II and III, which highly relate with protein secondarystructure[86]. Its high selectivity simplifies the spectrum and avoids fluorescence emissionlonger than 260 nm, so RRS is a sensitive technique in the analysis of the secondary struc-112.2. Raman spectroscopyture of proteins and peptides[85]. As the laser shifts to the visible range, samples such as thecarotenoids, pigment, dyes or other colored molecules also become RRS active[87]. Despite sev-eral advantages, the practical application of RRS is still quite limited, as it requires tuning anexcitation laser to a specific wavelength near the electric transition of a molecule. This compli-cates the rejection of Rayleigh scattering. A fixed notch filter will not work, so RRS requires aspecial spectrometer for general application. Furthermore, strong fluorescence emission fromother components in a sample matrix can completely obscure the Raman signal of interest. An-alytes in resonance with the radiation can also undergo photodecomposition[87].SERS is another type of enhanced Raman spectroscopy. It uses nanoparticle colloids or othermetallic substrates to amplify the Raman response[88–90]. SERS produces signal enhancementof as much as 1011-1012 for molecules adsorbed on the surface of nanoparticle[91, 92]. Thisdramatic enhancement is attributed to two possible mechanisms, chemical enhancement andelectromagnetic enhancement[91, 93]. Under exposure to the laser, the excitation of electronson the nanoparticle stimulates a surface plasmon resonance, enhancing the electromagneticfield in the range of 10-100 nm from nanoparticle surface[93, 94]. Dynamic charge trans-fer between the nanoparticle and adsorbed molecules contributes to the so-called the chemi-cal enhancement[93, 95]. Heterocyclic nitrogen-containing compounds or thiol molecules havehigher affinity to gold and silver nanoparticles and produce intense SERS signal[96, 97]. As a re-sult, SERS has been utilized to study thiolated DNA[98], neurotransmitter[99, 100], and aminoacids[101]. In another application, SERS has also been developed as a biosensor for the detectionof various cancers or particular diseases[102–105]. Generally, SERS-active substrates, attachedwith nitrogen- or sulfur-containing molecules as Raman reporters, are functionalized with probemolecules (i.e. antibodies)[106]. If this biosensor specifically links to the target molecules, thenan intense SERS signal of reporters will be observed from the specimen, signifying the existenceof cancer cells or other disease[106]. In addition, SERS using an excitation source in ultravioletregion, also known as surface enhanced resonance Raman spectroscopy (SERRS), can provideeven larger enhancement due to the combination of SERS and RRS effect[107]. SERS and SERRSare sensitive techniques, but poor reproducibility and complicated SERS substrates limit theirreal-world measurement. Moreover, SERS is not active, if the targeted analytes cannot bind withSERS substrates.Contrasting with Raman spectroscopy using single continuous-wave laser, non-linear CARS122.2. Raman spectroscopyspectroscopy uses three strong pulse lasers. CARS, a multi-photon process, illuminates the sam-ple with a pump beam νp, a Stokes beam νs and a probe beam νpr simultaneously to produce acoherent Raman response νp-νs+νpr (Figure 2.4)[108]. In many cases, pump beam also serves asνp νs νpr νp-νs+νprvvCARSVirtual statesFigure 2.4: Scheme diagram for CARS.a probe beam (νp=νpr) with a fixed frequency, while Stokes beam is tuned to vary the frequencydifference between the pump beam and Stokes beam[109]. When the frequency difference νp-νsmatches the frequency of molecular vibration mode νv, then coherent anti-Stokes Raman scat-tering νp+νs occurs and its intensity can be amplified by as much as five order of magnitudeover spontaneous Raman spectroscopy[109]. CARS microscopy is much more sensitive and effi-cient than conventional Raman microscopy for real-time chemical mapping. Contrary to opticaland fluorescence microscopy, it does not require staining samples with various chromophores orspecific fluorescent probes, which may inherently affect the metabolism of living cells[109, 110].Thus far, CARS has been successfully applied in the chemical imaging of various tissues[111],living cells[112], lipid membranes[108], material[113] and other complexes.In addition to RRS, SERS, SERRS and CARS, other strategies have also been applied to over-come the inherently weak spontaneous Raman scattering responses. Stimulated Raman spec-troscopy (SRS), another nonlinear Raman spectroscopy, yields a resolution of 10 cm−1 spectrumin 50 femtosecond, allowing for the analysis of chemical or biochemical reaction dynamics[114].Tip enhanced Raman spectroscopy (TERS) features a nanometers gold- or silver-coated tip to en-hance Raman signal for the molecules near the tip apex[115]. Despite the appeal of enhancedRaman spectroscopy, the complexity, the poor reproducibility as well as the expense of instru-mentations or SERS-active substrates have limited their wide application. If a sample itself pro-132.3. Fourier transform Infrared spectroscopyduce sufficient Raman signal, then spontaneous Raman spectroscopy is highly recommended.2.3 Fourier transform Infrared spectroscopyFourier transform infrared spectroscopy (FTIR) is also a useful spectroscopic technique. A regu-lar FTIR spectrometer consists of an infrared radiation source, a Michelson interferometer and asingle channel detector. Globar silicon is a common radiation source, emitting the mid-infraredradiation ranging from 2.5 to 25 µm[49, 116]. The incident beam enters into the interferome-ter, where the beam splitter splits the laser into two halves and delivers them to a fixed and amoving mirror respectively[23]. The reflected lasers recombine and focus on the sample, whilethe detector measures the interferogram of the transmitted radiation[23, 49]. FTIR is similarto FT-Raman, as both use an interferometer and single-channel detector. However, FT-Ramanuses a monochromatic continuous-wave laser at 1064 nm, whereas FTIR uses a broadband mid-infrared radiation source. Moreover, the majority of Raman spectrometers collect Raman scat-tering in a back-scatter mode, but various sampling techniques are available for FTIR (Figure2.5).SSSSGas(A) (B) (C)Microphone(D)Figure 2.5: Various FTIR sampling techniques. (A)transmission FTIR, (B) ATR, (C) PAS and (D) DRIFT.Transmission FTIR measures the absorbance intensity of radiation transmitted through thesample, so it is more suitable for the transparent samples (Figure 2.5A). Opaque samples have tobe ground to micrometer size and evenly mixed with potassium bromide as a pellet for measure-ment, so transmission FTIR evaluates the overall sample. Attenuated total reflectance (ATR),diffuse reflectance Infrared Fourier transform (DRIFT) and photoacoustic (PAS) tend to analyze142.3. Fourier transform Infrared spectroscopythe surface, but their penetration depths and mechanisms differ.ATR measures the absorbance intensity of a internally reflected beam (Figure2.5B). As theinfrared light contacts the sample through the higher-refractive-index ATR crystal (i.e. Zinc Se-lenide or Germanium), total internal reflection occurs on the interface with evanescent beampenetrating few micrometers into the sample surface. For instance, Forsskahl et. al. haveshowed a penetration depth of 1-2.2 µm for paper samples using ATR[117]. Thus, ATR ex-cels in studying interfaces[118] or coating materials[119, 120]. ATR spectrometer can maintaina constant penetration depth for the analyzed sample by applying a constant pressure to thesample hold to the ATR crystal.The penetration depth of PAS is deeper than ATR. For the pulp samples, PAS can reach ap-proximately 3-10 µm in depth[117]. In PAS (Figure 2.5C), infrared beam strikes the samplethrough IR-transparent gas (i.e. helium)[49]. The absorption of the incident beam producesthe heat within the sample, which diffuses to the surrounding gas[49, 121]. The sudden ther-mal expansion of the gas generates a photoacoustic signal, which is recorded by a sensitivemicrophone[49, 121]. Unlike transmission FTIR or DRIFT, PAS allows nondestructive chemi-cal analysis of opaque or strong scattering samples and requires minimum sample preparation.Many works have reported the application of PAS for wood, pulp, soil and coal analysis[34, 122–125].DRIFT detects diffusely reflected infrared light in all angles (Figure 2.5D), so it offers muchlarger penetration depth than ATR and PAS. In wood samples, it ranges from 37 to 138 µmbased on the wood density[117]. The measured absorbance intensity is inversely proportionalto the light scattering, so this technique requires a reduction of the sample particle size to 2 µmor less in order to minimize light scattering[49, 60]. The ground sample is often blended withpotassium bromide for measurement.Due to its high sensitivity to water, FTIR is very limited when it comes to the analysis ofaqueous solutions analysis. The broad and intense water bands centered at 3450 and 1640cm−1 dominate the infrared spectra[60]. Even if this does not interfere detecting the signal ofinterest, it still increases the acquisition time to acquire a satisfactory FTIR spectrum for theweakly absorbing analytes. In order to be able to quantify the sugar contents in juice samples orstandard solutions, Duarte et. al. have co-added 512 scans of single-reflectance ATR spectra[3].Alternatively, a cylindrical ATR liquid cell can induce multiple reflectances of infrared beam and152.4. Preprocessing methods and multivariate analysisenhance infrared signal[126]. Another possible solution is to dry the liquid samples before themeasurement. Budinova et. al. have dried blood serum or whole blood on a polyethylene (PE)card and determined its glucose and cholesterol concentrations using 64 scans of transmissionFTIR spectra[127]. However, this method relies on the fortuitous fact the detected glucose andcholesterol features do not overlap with background polyethylene signal[127].FITR has also been developed for microanalysis. A conventional infrared microscope allowsthe measurement in transmission, ATR or reflectance mode, whereas Raman microscope collectsthe back-scattered signal. As the incident radiation ranges from 2.5-25 µm, the diffraction limitof infrared microscope is limited to more than 10 µm, depending on set up[128]. It also failsto provide confocal depth profiling or confocal imaging as Raman microscope. Both the atmo-spheric CO2 and the moisture in the sample can disturb the final infrared spectra[54, 129]. Thus,studies of biological samples more often use Raman microscopy. However, infrared microscopehas a pronounced advantage for samples that emit strong fluorescence. A FTIR microscope cannondestructively characterize or chemically image pigments, dyes and binding media in culturalstatues and paintings[116, 130]. It is also a preferred technique for hydrocarbon fluid inclusionsin minerals[131, 132]. In addition to these features, infrared microscope can provide rich infor-mation for molecules with strong dipole moments, well complementing Raman spectroscopy.Thus, the two techniques together have been used to study tissues, minerals, etc[133–135].2.4 Preprocessing methods and multivariate analysis2.4.1 Preprocessing methodsBoth infrared and Raman spectra always contain some undesirable baseline and noise. In par-ticular, vibrational bands in Raman spectra often superimpose on a sloping background causedby fluorescence. Infrared spectral bands often suffer from band overlapping. Spectral pretreat-ment can resolve overlapping bands and amplify small features, which can simplify the spectralanalysis. Here, we use several Raman spectra from cooking oils (Figure 2.6) as an example toreview several often-used automated preprocessing methods.162.4. Preprocessing methods and multivariate analysis400 600 800 1000 1200 1400 1600 1800 20000123456 x 105Raman Shift (cm−1)Raman Intensity / arb. units  Canola oilExtra virgin olive oilPeanut oilFigure 2.6: Raman spectra of cooking oils.Simple and robust, first-derivative methods can easily remove the baseline, but it shifts thepeak positions and decreases the spectral signal-to-noise ratio (Figure 2.7).400 600 800 1000 1200 1400 1600 1800 2000−1−0.500.511.5Raman Shift (cm−1)Raman Intensity / arb. units  1st Derivative Canola oilExtra virgin olive oilPeanut oilFigure 2.7: First-derivative preprocessing method.Second-derivative method also removes the sloping baseline and enhances the spectral res-olution, but the spectral signal-to-noise ratio is a lot worse than the first-derivative and originalspectra (Figure 2.8). Therefore, derivative methods are usually combined with Savitzky-Golay172.4. Preprocessing methods and multivariate analysisaveraging, which smooths the spectrum by using least-squares polynomial fitting over a pre-scribed window of the spectrum[136, 137]. Even so, they still cause difficulties in the analysisof the spectra with low signal-to-noise ratio.400 600 800 1000 1200 1400 1600 1800 2000−2−1.5−1−0.500.511.522.5Raman Shift (cm−1)Raman Intensity / arb. units  2nd derivative Canola oilExtra virgin olive oilPeanut oilFigure 2.8: Second-derivative preprocessing method.In addition to smoothing, polynomial functions also fits the sloping baseline well. The de-rived fourth- or fifth-order polynomials resemble the fluorescent baseline in Raman spectra[43].Various automated baseline subtraction methods based on iterative polynomial fitting have been400 600 800 1000 1200 1400 1600 1800 2000−0.200.20.40.60.811.2Raman Shift (cm−1)Raman Intensity / arb. units  Iterative polynomial fitting (4th order) Canola oilExtra virgin olive oilPeanut oilFigure 2.9: Iterative polynomial fitting preprocessing method.182.4. Preprocessing methods and multivariate analysisreported[138–140]. The often-used algorithm proposed by Lieber repeatedly removes Ramanbands from an underlying polynomial fitted baseline until the fitted baseline converges[138].Usually, the whole process only takes 25-200 iterations[138]. As shown in Figure 2.9, the fourthorder of polynomial function effectively extracts the Raman features from the sloping baselinein the region between 800 and 1800 cm−1, but it clearly shows some limitations from 350 to 800cm−1.Fourier transform (FT)[141, 142] and wavelet transform (WT)[147–149] offer alternativestrategies. Spectra are decomposed on the basis of the waves of different frequencies, as base-line, signal and noise represent low, middle and high frequency components respectively. Theinfinite sine and cosine functions serve as FT basis functions, and FT can remove noise andbaseline by setting high-frequency and low-frequency cut-offs. However, this method is sen-sitive to the noise in the spectrum, requiring a change of frequency cut-off from spectrum tospectrum[141, 142]. This makes it difficult to apply FT to a large number of spectra automati-cally. Another drawback of FT is the loss of time-domain information in the frequency domain.The method of Windowed Fourier transform (WFT) addresses this problem by setting a givensize of window for all basis functions[150]. However, the requirement of a constant windowsize still limits this method[150]. Since its first introduction in chemistry, WT has been inten-sively used for infrared and Raman spectra[147–149]. In WT, wavelets serve as basis functionsto represent original spectrum. Unlike sine and cosine functions, a wavelet is a finite waveformoscillation and many famous wavelet families such as Symlet, Daubechies, Gaussian, Mexicanhat etc. exist (Figure 2.10).0 2 4 6 8 10−1−0.500.511.5  (A)0 2 4 6 8 10−1−0.500.511.5  (B) sin(x)symlet6Figure 2.10: Symlet6 wavelet vs sine filter.192.4. Preprocessing methods and multivariate analysisWavelets such as Daubechies or Symlet are preferred for vibrational spectra, because theyresemble the shape of vibrational bands[149, 151]. Once the mother wavelet is selected, it isscaled and translated, creating a set of wavelets basis. Then the raw spectrum is mathematicallydescribed as a sum over of all scaled and translated wavelets. It is commonly agreed that WT issuperior to FT[149, 150, 152]. As the shifted and scaled wavelets provide far more features thansinusoids, fewer wavelets are needed to represent the raw spectrum[142]. The window width ofwavelets changes as the scale varies, so the resolution of wavelet transformed spectrum is notlimited as FT or WFT[150]. WT is more suitable for performing automatic baseline correctionthan FT[142, 153]. Due to its wide application, WT related toolboxes such as WaveLab850400 600 800 1000 1200 1400 1600 1800 2000−1−0.500.511.52Raman Shift (cm−1)Raman Intensity / arb. units  Discrete wavelet transform Canola oilExtra virgin olive oilPeanut oilFigure 2.11: Discrete wavelet transform preprocessing method.and Wavelets are available for Matlab, easily allowing uses that are well adapted to vibrationalspectral analysis[154–156]. Sometimes WT change the shape of spectra (Figure 2.11), but thisdoes not affect the Raman features of interest. More importantly, WT does not shift the peakpositions like first-derivative method.2.4.2 Multivariate analysisIn a next step, preprocessed spectra are subject to multivariate analysis for further classificationor regression. Each spectrum x presents as a vector with n dimensional variables (wavenumberor Raman shift). We arrange it as a row vector, so m number of measurements constitute a rect-202.4. Preprocessing methods and multivariate analysisangular matrix X withm rows and n columns. The number of variables far surpasses the numberof measurements, but many spectral variables are redundant and uninformative. Multivariateanalysis allows us to observe the overall variance distribution among the samples and searchfor the variables most significantly related to that variance. Methods such as principal compo-nent analysis (PCA), linear discriminate analysis (LDA), partial least squares (PLS), and supportvector machines (SVM) represent useful chemometrics tools for spectral analysis, and they arebriefly introduced below. If it is not particularly specified, a bold uppercase letter represents amatrix, and a bold lowercase letter represents a vector.PCA is one of the simplest methods to reduce dimensionality and visualize high dimensionaldata. It linearly transforms the data matrix X into a new coordinate system, where new variables,known as principal components (PC), lie orthogonal to each other. These new PCs order indescending variance, so the projection of data along the pc1 direction has the largest variance,followed by pc2, pc3 and so on. Equation 2.1 mathematically describes the decomposition ofPCA,X = TPT + E = t1pcT1 + t2pcT2 + ...+ tjpcTj + E (2.1)where X is a m × n data matrix with m number of observations and n number of variables, Tis a m × j scores matrix, P is a j × n loading matrix, E is a m × n residue matrix, tj is a m × 1score vector and pcj is a 1 × n PC vector[157]. In order to find these new PCs, the covariancematrix of the original dataset (XTX) is calculated first and then decomposed with singular valuedecomposition (SVD) to solve the eigenvectors and eigenvalues as Equation 2.2,XTX = WΣ2WT (2.2)where each column of W is an eigenvector, whose eigenvalue lies on the corresponding diagonalelement of Σ[158]. The eigenvectors, that have relatively large eigenvalues, serve as PCs andconstitute the loading matrix P. The projection of the original data into the loading (PC) spacegives the scores matrix T as Equation 2.3,T = XP (2.3)The first few PCs usually account for 80-90% of the variability, which are good enough to de-212.4. Preprocessing methods and multivariate analysisscribe the original high-dimension data. Thus, plotting the scores against the first two or threePCs can reveal the variance of a dataset and allow us to directly visualize the original highdimensional data. Vibration spectroscopy combined with PCA has been employed to classifyvarious samples such as bacteria[159], cell[68], edible oil[160], gasoline[161], tissue[162] and soon.LDA, also known as Fisher’s linear discriminant, is a linear classifier and aims to seek aprojection vector ω, along which the projected data are maximally differentiated[163]. LDApredicts the class information as Equation 2.4,y = xω (2.4)where y is the class label (i.e. 1 or -1), ω is the projection vector, and x is a spectrum arranged asa row vector[163]. Fisher proposed the solution to ω by maximizing the ratio of the differencebetween the projected means to the sum of the projected variance as Equation 2.5,maxωJ(ω) =∣∣∣µ˜1 − µ˜2∣∣∣2S˜12 + S˜22 (2.5)where µ˜1 is the projected mean of class 1, µ˜2 is the projected mean of class 2, S˜1 is the projectedvariance of class 1, and S˜2 is the projected variance of class 2[163]. LDA analysis requires allthe class information, as it is a supervised method, compared with unsupervised PCA. In somecases, the classification results predicted by LDA are better than PCA. Nevertheless, the perfor-mance of LDA gets worse as the number of variables increases. Direct application of LDA to thespectral analysis fails due to the high dimensionality of spectra data, so dimension reductionshould be performed before LDA (i.e. PCA-LDA)[164–166]. LDA assumes samples from eachclass are normally distributed and their projected means are significantly different[167]. For thedata, which do not satisfy this condition, LDA cannot classify samples as well as SVM or othernonlinear classifiers such as quadratic discriminant analysis.SVM is another powerful supervised learning technique. It can perform linear or nonlinearregression and classification for problems, which are limited in the original dimension space. Inclassification, it transforms the spectral data x to the higher dimension space φ(x) via a kernelfunction φ[168]. SVM searches for an optimal hyperplane (ωT φ(x)+b=0) to maximally separate222.4. Preprocessing methods and multivariate analysistwo categories of samples. Those who satisfy ωT φ(x)+b≥1 are assigned to class 1, whereas thosethat satisfy ωT φ(x)+b≤-1 are assigned to the other class -1[168, 169]. The larger the boundarygap between these two decision domains, the smaller error the classification can achieve. Infact, two classes may overlap in some degree. Thus, not only does SVM need to maximize theboundary gap, but also simultaneously minimize the total distance of the misclassified samplesfrom their corresponding decision boundary[169]. In order to solve ω and b for such optimalhyperplane, we refer to Equation 2.6,minω,ε,bJ(ω, ε) =∣∣∣ω2∣∣∣2+ cn∑i=1εi (2.6)where |ω2|2 is a reciprocal of the distance between two decision boundaries, εi is the distance ofi-th misclassified sample to its decision boundary, and c is a positive constant adjusting the tradeoff between a wide boundary gap and small misclassification error[168, 169]. SVM was devel-oped for binary classification originally, but it has also been adapted to multiple classificationproblems associated with Raman spectra or infrared spectra[1, 170–172]. In a similar fashion,SVM solves linear or nonlinear regression problem. SVM regression (SVR) or Least Squares SVM(LS-SVM) has been developed for quantitative analysis of Raman or infrared spectra[173–176].PLS is a often-used linear quantification tool owing to its simplicity and good performance[175].In PLS, a new set of variables, also referred as latent vectors, decomposes the spectral dataset Xand the response data Y simultaneously as Equation 2.7 and 2.8,X = TPT + E (2.7)Y = TWQT + F = UQT + F (2.8)where T and U are the score matrices, P and Q are the loading matrices, E and F are the residuematrices, and W is the regression weight matrix[177]. Here, latent vectors are the columns of T,which account for the maximal covariance between X and Y[157, 177]. Furthermore, Y can beestimated on the basis of a linear regression model between the spectra and the response valuesas Equation 2.9,Yˆ = XB (2.9)232.4. Preprocessing methods and multivariate analysisB = (PT)−1WQT (2.10)where Yˆ is the predicted response vector and B is the regression coefficient vector, which canbe calculated from W, P and Q[157, 178]. The nonlinear iterative partial least squares (NI-PALS) algorithm has been developed to solve the regression coefficient to construct a linear PLSregression model[157, 178]. Many papers in the literature have reported the application of var-ious spectroscopic techniques combined with PLS to quantify the chemical composition andproperties of the analyzed samples[123, 179–181]. In comparison with other regression toolssuch as SVM and neural network, PLS is much simpler and does not need to optimize extraparameters, which significantly relate to the efficiency of the regression model. However, SVMexcels in the analysis of NIR data affected by nonlinear temperature variation, due to its abilityto handle nonlinear regression[175].Matlab offers very powerful software for use in matrix calculations and interface with otherprograms written in C++ and Fortran. For chemists, it is quite difficult and time-consumingto program detailed algorithms for classification or regression. Toolboxes built in Matlab suchas the Statistics toolbox and Bioinformatics toolbox, provide PLS and PCA related functions,so we can easily call these functions to perform multivariate analysis of the spectral data. Inaddition, some useful toolboxes are available online, and they are free for academic use. Forinstance, the Statistical Pattern Recognition toolbox, published by Vojteˇch Franc and Va´clavHlava´cˇ, is a well developed classification toolbox and includes various classifiers such as LDA,PCA, SVM, Quadratic discriminant analysis and Bayesian classification[182, 183]. The classi-fiers listed in this toolbox have been used to classify vibrational spectra and address differenttypes of classification problems[1, 184–186]. For regression, many papers in the literature havereported various algorithms to improve PLS models by removing uninformative features. PLS-Genetic algorithm (PLS-GA) toolbox, introduced by Leardi and Lupia´n˜ez, is able to build a PLSmodel with the most relevant variables, improving our knowledge of spectra further[187, 188].Likewise, iToolbox utilizes moving window PLS or interval PLS to construct a feature selectedregression model[189, 190].24Chapter 3Metabolomic analysis of human embryoculture media by Raman spectroscopy3.1 IntroductionAccording to the United States National Health Statistics Reports, during the period from 2006to 2010, 12% of women aged 15-44 experienced difficulties getting pregnant or carrying a full-term pregnancy[191]. Another report based on the 2009-2010 Canadian Community HealthSurvey, put the proportion of infertile Canadian women aged 18-44 in a range from 11.5% to15.7%[192]. One possible solution to this infertility problem is assisted reproductive technology(ART). Among the three types of ART, in vitro fertilization (IVF), gamete intrafallopian transferand zygote intrafallopian transfer, IVF results in the highest success rate for pregnancy, andaccounts for more than 99% of ART treatment[39, 40, 193, 194].A typical IVF cycle proceeds as follows[194–196]. On Day 0, oocytes are retrieved (Figure3.1A) and inseminated with sperm (natural IVF, Figure 3.1B) or intracytoplasmic sperm injec-tion (ICSI, Figure 3.1C). On Day 1, an egg with two pronuclei (Figure 3.1D) is placed in a newdrop of culture medium for further incubation. On Day 2, the embryo remains in a cleavagestate and usually grows to 2-4 cells by the end of day (Figure 3.1E). On Day 3, the embryo de-velops to 4-8 cells (Figure 3.1F). Physicians then select a multiple set of the most viable embryosand transfer them to the female uterus. The remaining embryos are placed in another type ofmedium and continue to grow. On Day 4, the viable embryo can reach the morula stage, whichconsists of approximately 12-15 cells in a compact cell cluster (Figure 3.1G). On day 5, the vi-able embryo forms a blastocyst (Figure 3.1H). A viable blastocyst embryo can also be transferredowing to its high success rate of implantation[13]. Other viable embryos are cultured to Day 6(Figure 3.1I) and cryopreserved for a future IVF cycle in case the current cycle fails.253.1. Introduction100-200 μmFigure 3.1: The development of single human embryo from Day 0 to Day 6 during IVF. The size ofhuman embryo cell stays unchanged during Day 1 to Day 6. http://medicine.yale.edu/obgyn/yfc/ourservices/invitro/development.aspx September 2014.Regardless of which day embryos are transferred, multiple embryos with high viability arealways selected on the basis of their morphologic appearance, then transferred simultaneously[197].All knowledge of embryo viability is based on cleavage rate and morphologic development[12,198, 199]. The selection rules may vary between clinics, but generally speaking, practitionersfollow the standards developed by the Society for Assisted Reproductive Technologies (SART)[12,200].However, this morphology-dependent selection rule leads both to relatively low success ratesand high multiple pregnancy rates. The current practice of multiple embryo transfer causes seri-ous health problems associated with the gestation of multiple infants, such as preterm delivery,low birth weight, lifelong disabilities etc[41]. The mortality rates of twins and triplets are fiveand thirteen times, respectively, those of a single infant in the first year of life.[41]. IVF cy-cles using fresh donor eggs during 2010, transferred on average of 2 to 3 embryos, resultingin 36.9% clinical pregnancies per cycles, of which 57.1% produced singleton live births and24.8% resulted in multiple-infant live births[201]. Such low pregnancy success rates coupledwith a high percentage of multiple-infant live births constitutes the main drawbacks of the cur-rent embryos selection criteria based on morphological appearance. For this reason, clinics and263.1. Introductionphysicians need a more reliable method to perform single embryo transfer (SET) of a selectedmost-viable embryo.Some clinics experience consistent failures of three cycles even upon transfer of embryoswith good morphologic scores to a normal uterus[14, 42]. The major cause for these failures ischromosomal abnormality, which cannot be monitored via embryo morphology using a microsc-ope[14]. To address this issue, non-invasive techniques have been recently proposed to assessthe degree of embryo viability by measuring the metabolomic changes to the embryo spentculture medium[202–204]. The embryo culture medium provides the embryo with an envi-ronment similar to that of the human female reproductive tract. The key components of cul-ture media are salts, buffer, amino acids, energy substrates (i.e. glucose, pyruvate and lactate),human serum albumin, etc[205, 206]. The metabolism of a growing embryo induces changesin the metabolites of the culture medium. These metabolites could serve as small-moleculebiomarkers to track the physiology and functional phenotype of embryo at the cellular level.At the precompaction stage, the embryo is in the mode of pyruvate-based metabolism, and thepredominant energy substrates are pyruvate and lactate[203, 206–208]. It also needs to con-sume a low level of glucose and some specific amino acids such as aspartate[206, 209]. Oncethe embryonic genome is activated, the embryo switches to a glucose-based metabolism. Ithas been observed that the consumption of glucose significantly increases when the embryogrows from morula to blastocyst[203, 206]. After the embryo is transferred, metabolites inthe remaining culture media are analyzed. Table 3.1 summarizes previous studies performedby many researchers in the effort to develop predictors of embryo development and viability.Among often-used techniques, microfluorescence and HPLC both require complicated samplepreparation[207–212]. Spectroscopic methods offer greater promise for practical clinical ap-plication. Compared with NMR spectroscopy, NIR and Raman spectroscopy use a simpler in-strumental setups and easier measurement procedures. However, literature results based onNIR studies are contradictory[213–216], and the effectiveness of NIR is still questionable, ow-ing in particular to its sensitivity to water. In the present work, we explore the use of Ramanspectroscopy to study embryo spent culture media collected during Day 3 and Day 6 to predictembryo viability for IVF treatment.273.1.IntroductionTable 3.1: The summary of non-invasive techniques used to study embryo spent medium as a predictor of human embryo development andviability.Method Transfer Metabolites Outcome Ref.Microfluorescence Day 1-6 Glucose, pyruvate Embryo which develops into blastocyst has higher pyruvate [207]uptake during Day 2-5 and higher glucose uptake on Day 6.Microfluorescence Day 2-6 Glucose, pyruvate, Normal developed embryos have higher pyruvate and [208]lactate glucose uptake and lactate production than arrested ones.HPLC(fluorescence) ET on Day 2 18 amino acids The turnover of asparagine, glycine and leucine highly [209]correlates with pregnancy and live birth.HPLC(fluorescence) Day 2-3, Day 5 18 amino acids The turnover of 18 amino acids are different between [210]normal developed embryos and arrested ones.HPLC-MS ET on Day 3 Amino acids Embryos which result in pregnancy or nonpregnancy can be [211]classified by Coomans plot of amino acids chromatographs.HPLC-MS and NMR ET on Day 3 Amino acids by HPLC Both methods differentiate embryos which result in pregnancy [212]Sample matrix by NMR or not. NMR signals from phospholipid, cholesterol andtriglycerides correlate with pregnancy rate.Raman ET on Day 3,5 Sample matrix Viability indices predicted by Raman spectra correlate with [194]0 or 100% delivery rate of embryos.Raman and NIR ET on Day 3 Sample matrix Viability indices predicted by Raman or NIR spectra correlate [193]with 0 or 100% delivery rate of embryos.NIR SET on Day 2,3 Sample matrix Viability indices predicted by NIR correlate with reproducible [213]potential but independent of morphology grades of embryos.NIR ET on Day 5 Sample matrix Viability indices predicted by NIR correlate with 0 or 100% [214]implantation rate of embryos.NIR SET on Day 2,5 Sample matrix Pregnancy rate is not improved when NIR is added to [215]morphology scores.NIR ET on Day 2,3,5 Sample matrix Pregnancy rate is not improved when a commercial NIR is [216]added to morphology scores.NMR ET on Day 3 Glucose,pyruvate Embryos resulting in implantation have higher glutamate [217]lactate, amino acids concentration in media.Viability indices predicted byNMR correlate with 0 or 100% delivery of embryos.ET stands for embryo transfer.283.2. Experimental3.2 Experimental3.2.1 ReagentsGlucose, sodium lactate and Luria Broth (LB) were purchased from Sigma-Aldrich (Canada). G-1TM PLUS (VitroLife, Englewood, USA) was supplied by Children′s and Women′s Health Centreof British Columbia. Physicians there culture human embryos in VitroLife G-1 medium fromDay 0 to Day 3 and in Vitrolife G-2 from Day 4 to Day 6. Bacillus subtilis was kindly providedby Dr. Elena Polishchuk (UBC Biological Services Laboratory).3.2.2 SamplesThis work focuses on three different sample sets. The first sample set contains aqueous solutionsprepared in our laboratory. Glucose and sodium lactate standard solutions were prepared asfollows. 20 aqueous solutions of glucose, ranging from 0.5 mM to 10 mM in 0.5 mM increments,were prepared by diluting the glucose standard solution. To consider the effect of sample matrix,G-1 was spiked with the glucose standard solution. The added glucose in G-1 ranged from0.13 to 22.2 mM. Likewise, a set of G-1 solutions was spiked with the sodium lactate standardsolution, yielding the added lactate concentrations in G-1 ranging from 0.28 to 15.56 mM.The second sample set consists of IVF specimens collected from Children′s and Women′sHealth Centre of British Columbia during a period of six months. Hospital physicians and labstaffs performed the IVF routine procedures. Eggs were retrieved and fertilized on Day 0. Onthe next day, they placed each two-pronuclei embryo in a 50 µL droplet of G-1 medium in a petridish overlaid with oil (Day 1). An embryo-free 50 µL droplet of G-1 medium was also placed inthe same petri dish. It underwent the same incubation condition and served as an experimentcontrol. On Day 3, viable embryos were selected based on their morphological scores, andtransferred. Sometimes, owing to a lack of viable embryos, transfer was postponed. Instead,physicians placed each embryo in a 50 µL droplet of G-2 medium and extended its culture fromDay 3 to Day 6. On Day 6, top-grade blastocyst embryos were selected and transferred.Once embryos were transferred on Day 3 or Day 6, the petri dishes containing the usedand control media were immediately placed in a dry ice box and transported to our lab for Ra-man spectroscopic analysis. Patient clinical pregnancy results and embryo morphology scores,recorded by hospital physicians, served as classification standards to evaluate embryo spent293.2. Experimentalculture media as a predictor of embryo viability.A third sample set constitutes bacillus used LB media sampled at different time points dur-ing culture. 498 mL sterile LB was placed in a 1 L Erlenmeyer flask for bacillus culture. Stockbacillus subtilis was revived in 15 mL LB medium. We then sampled 2 mL bacillus-containingLB and centrifuged it for 5 min. After the supernatant was removed, the bacillus pellet waswashed with 0.9% NaCl solution. This NaCl solution was filtered, and 2 mL LB medium wasused to re-suspend the pellet. This seed solution was inoculated to 498 mL LB medium forculture. The Erlenmeyer flask was kept in a shaker conditioned at 37◦C and 200 rpm. Duringculture, 1 mL LB was sampled at different time points. Optical density (OD600), measured byabsorption at a wavelength of 600 nm, estimated the number of viable bacillus in the LB media.After the sample was centrifuged at 13000 rpm for 10 min, the supernatant was collected forRaman spectroscopic analysis under the same experimental condition as the IVF patient sam-ples.3.2.3 Instrumentation and Raman measurementsRaman spectra were recorded using the 785±0.3 nm output of a continuous-wave laser (Inno-vative Photonic Solutions) for excitation. A home-built Raman fibre-optic probe integrates withan Olympus BX50 microscope to collect Raman scattering from a micro-volume sample. Thisprobe connects to the 785 nm excitation laser source and spectrometer (Princeton InstrumentActon SP2300) via multimode fibre optics. For each Raman measurement, 18 µL sample wasplaced in a micro-cuvette made of aluminum. The laser is focused into the sample solution viaa 5x long-working-distance single lens objective (f=25 mm, D=13 mm).This lens collects the backscattered Raman signal and passes it through a dichroic beamsplitter and a spatial filter. The probehead then refocuses this light into a six-around-one multi-mode fibre bundle for transmission to the spectrograph. At the bundle exit, the seven fibres arealigned vertically and connected to the spectrometer. The spectrometer is equipped with a 600-groove/mm diffraction grating. The entrance slit width is set to 200 µm. A thermoelectrically-cooled CCD detector (Princeton Instrument Acton PIXIS 100) with a chip size of 1341 x 100pixels detects the final signal in the Raman shift range of 300-2000 cm−1. The entire Ramansystem is controlled by a custom LabVIEW program. Each spectrum was acquired by co-adding20 scans of 15 s integration time. The laser power at the sample measured about 100 mW.303.3. Results3.2.4 Spectral preprocessing and multivariate analysisDiscrete wavelet transform (DWT) was applied to the original Raman spectra using Wavelab850toolbox in Matlab 2009a[154]. Using a Symlet5 filter, the raw spectrum was decomposed into aseries of wavelet coefficients consisting of one approximation (CA6) plus six detail (CD1, CD2...CD6) coefficients. Signal reconstructed from CA6 represents the low frequency background,while that reconstructed from CD1 and CD2 contains the high frequency noise. The final DWTspectrum was calculated by subtracting the noise and background signal from the raw spectrum.All preprocessed spectra were normalized to a constant intensity of one for the water feature at1640 cm−1. A standard normal variate (SNV) correction was performed on each spectrum bydividing the mean-centered DWT spectrum by the spectral standard deviation:xij,SNV = (xij − x¯i)/√∑nj=1(xij − x¯i)2n− 1(3.1)where xij stands for the absorbance intensity at the j-th wavenumber position in the i-th spec-trum, x¯i is the mean value of the i-th spectrum and n is the total number of points in thespectrum[218].3.3 Results3.3.1 Metabolites in embryo culture mediaAlthough the precise chemical compositions of G-1 and G-2 are proprietary, their contents mustprovide similar nutrients as the female oviduct and uterus. In particular, G-1 must resem-ble oviduct fluid, which contains approximately 10.5 mM lactate, 0.5 mM glucose, 0.32 mMpyruvate together with several amino acids for the development of embryos in precompactionstage[206]. G-2 resembles uterine fluid and must offer an environment containing approxi-mately 3.15 mM glucose, 0.10 mM pyruvate, 5.2 mM lactate together with all essential aminoacids[206].According to Table 3.1, embryos consume glucose, lactate and amino acids from the culturemedium as they develop. In principle, Raman spectroscopy can offer a overall profile of the cul-ture medium with the aim of evaluating the total change of metabolites in the sample matrix.However, Raman spectroscopy is not as sensitive as chromatography or fluorescence, so the ob-313.3. Resultsservation of a difference in Raman spectra requires a large variation in the abundant componentssuch as glucose and lactate. For this reason, we prepare glucose and sodium lactate standardsolutions. Studying these aqueous solutions serves to benchmark our Raman spectrometer, andhelps us optimize the experimental parameters (i.e. exposure time and laser power) for patientspecimens.3.3.2 Analysis of glucose aqueous solutionThe patient specimens have a volume of 25-30 µL. Accordingly, we fabricated a small aluminumcuvette to hold a 18 µL sample for each Raman measurement. The intensity of the Raman signalis proportional to the incident laser power, exposure time and illuminated sample volume[57].Due to weak Raman signal, we tried to increase the exposure time to improve its sensitivity andLimit of Detection (LOD) with the given sample volume. Here, we choose 100 mW laser power.The maximum exposure time we set is 300 s, as longer exposure than 300 s caused sampleevaporation.With this configuration, we acquired Raman spectra from 0-10 mM glucose solutions. Itis very difficult to observe any glucose features in the raw spectra (Figure 3.2A). Instead, the0.5 1 1.5 2 2.5 30246810Net Analyte Signal (NAS) of GlucoseRef. Glucose Conc. (mM) D  0 2 4 6 8 100246810Reference Glucose Conc. (mM)PLS Predicted Conc. (mM) C  400 600 800 1000 1200 1400 1600 1800 2000−101234Raman Shift (cm−1)Raman Intensity / arb. units.1640A400 600 800 1000 1200 1400 1600 1800 2000−2−1012Raman Shift (cm−1)Raman Intensity / arb. units.1072B1128CalibrationValidationCalibrationValidationFigure 3.2: (A) Normalized raw spectra of (−) water and (−) glucose aqueous solutions, (B) NormalizedDWT spectra of (−) water and (−) glucose aqueous solutions, (C) Plot of reference glucose concentrationsversus PLS predicted concentrations ((o) calibration and (o) validation samples), (D) Plot of referenceglucose concentrations versus Net Analyte Signal (NAS) of glucose ((o) calibration and (o) validationsamples).323.3. Resultswater bending band at 1640 cm−1 dominates the spectra[219]. After DWT and normalizationpretreatment, the most distinguishing features appear at 1072 cm−1 (C-O stretching) and 1128cm−1 (C-O-H bending) (Figure 3.2B)[220]. Such apparently weak Raman features verify the lowsensitivity of this technique, suggesting that its LOD probably remains in the scale of millimolar.We divided the complete dataset into calibration and validation dataset, then applied PLS to thecalibration dataset to construct a regression model as Equation 3.2,ycal = Xcal ∗b (3.2)where ycal is a m× 1 concentration vector with m number of calibration samples, Xcal is a m× nmatrix of spectra with m spectra and n wavelengths, and b is a n×1 regression vector calculatedfrom PLS analysis. Using this method, we predict the glucose concentrations of the unknownsamples (validation dataset). In Figure 3.2C, the linear relationship between the reference glu-cose concentration and PLS predicted values demonstrates Raman spectroscopy can quantifyglucose aqueous solution in the range of 0-10 mM under current experimental conditions.Figure of merits (i.e. LOD, sensitivity, etc.) serve to evaluate the calibration model. However,such diagnostics were originally developed for univariate models, in which the analyte concen-tration is linearly related to an univariate signal recorded in a single channel of detection[221].The direct estimation of a LOD from a multivariate model was limited until the introductionof Net Analyte Signal (NAS) by Farber and Kowalski[221]. For each spectrum, an NAS vector,which contains only spectral information of the target analyte, is calculated[221, 222]. As it alsolies orthogonal to the background and all the other interfering components, the Euclidean normof such vector, NAS, serves as a scalar response correlating with the amount of the analyte ineach sample[221, 222]. Linear least square fitting solves the regression coefficient and allowsone to construct an univariate calibration model as Equation 3.3,ycal = nascal ∗ bnas (3.3)where ycal is the m-by-1 concentration vector with m number of calibration samples, nascal isthe m-by-1 vector consisting of m number of NAS scalar responses, and bnas is the regressioncoefficient[222]. Researchers have used this NAS method to assess the LOD, limit of quantifica-tion (LOQ) and sensitivity of a PLS model constructed by NIR or FTIR[222, 223]. Figure 3.2D333.3. Resultsplots the NAS of glucose versus concentration, where a linear relationship starts around 2 mM.Table 3.2 summarizes the calculated Figure of merits from NAS of glucose solutions. The rootTable 3.2: Analytical figure of merits for the PLS model of 0-10 mM glucosesolutions using NAS analysis.Figure ofmeritNAS(glucose) Formula[222, 223]Accuracy RMSEE 0.343√∑ni=1(yi,cal−yˆi,cal )2nRMSEP 0.492√∑ni=1(yi,val−yˆi,val )2nLinearity Slope 5.186(Calibration) Intercept -2.249R2 0.991Sensitivity 0.193 1/bnasLOD 0.781 3.3 ∗ δa/SensitivityLOQ 2.368 10 ∗ δa/Sensitivitya: δ is the noise calculated from the standard deviation of the NAS of 3 blank(water) spectra.mean square error of estimation (RMSEE) 0.343 suggests a small deviation between the refer-ence values and NAS predicted values in the calibration set, and the root mean square error ofprediction (RMSEP) 0.492 indicates a good agreement in the validation set. The slope, interceptand correlation coefficient are estimated using the reference and the NAS predicted glucose con-centrations from the calibration set. The low level of sensitivity (0.193) as well as the high levelof LOD (0.781 mM) and LOQ (2.368 mM) demonstrate that Raman spectroscopy is an insen-sitive technique for the analysis of dilute aqueous solutions. This explains why the data dotslower than 2.4 mM largely deviate from the linear region, as shown in Figure 3.2.3.3.3 Analysis of IVF G-1 medium spiked with glucose and sodium lactatesolutionsTo consider the effect of sample matrix, we compare the Raman spectra of water with G-1medium spiked with water, glucose and sodium lactate solutions (Figure 3.3A and 3.3B). Thissample matrix does not give too much fluorescence background, or other difficulties in metabo-lite analysis by Raman spectroscopy. Comparing with pure water, we observe several dominant343.3. Results400 600 800 1000 1200 1400 1600 1800 2000−6−4−2024Raman Shift (cm−1)Raman Intensity / arb. units.10721128518 1464A400 600 800 1000 1200 1400 1600 1800 2000−6−4−2024Raman Shift (cm−1)Raman Intensity / arb. units.860540B1 2 3012345678NAS of Added LactateRef. Conc. of Added Lactate (mM)LOD=0.197  RMSEE=0.473RMSEP=0.532D  0.5 1.5 2.501234567NAS of Added GlucoseRef. Conc. of Added Glucose (mM)LOD=0.498  RMSEE=0.494RMSEP=0.521C  Figure 3.3: (A) Normalized raw and DWT spectra of (−) water, (−) blank G-1 medium, and G-1 mediaspiked with glucose aqueous solution; (B) Normalized raw and DWT spectra of (−) water, (−) blank G-1 medium, and G-1 media spiked with sodium lactate aqueous solution, (C) Plot of NAS versus thereference concentrations of the added glucose ((o) calibration and (o) validation data), (D) Plot of NASversus the reference concentrations of the added sodium lactate ((o) calibration and (o) validation data).features at 540, 860, 986, 1012, 1128 and 1464 cm−1 in the Raman spectrum of G-1 medium.The band at 1012 cm−1 is attributed to the abundant bicarbonate ion[224]. The features at 518,1128 and 1464 cm−1 are assigned to glucose (Figure 3.3A)[219, 220], as the added glucose in G-1 medium (0.13 to 22.2 mM) increases the intensities at these positions. Signal at 540 (νOCO)and 860 (νC-COO-) cm−1 (Figure 3.3B) correlates with sodium lactate[225, 226]. This can alsobe confirmed by the increased signal at 540 and 860 cm−1 with an increased concentration ofthe added sodium lactate (0.28 to 15.56 mM).To estimate the LOD and prediction uncertainty of the added glucose and lactate in G-1medium by Raman spectroscopy, we divide each dataset into calibration and validation dataset, and then we apply NAS. The linear relationship between the reference concentrations andthe NAS of the added glucose (Figure 3.3C) or sodium lactate (Figure 3.3D) suggests Ramanspectroscopy can detect their contents with an average of prediction error around 0.5 mM. In353.3. ResultsG-1 medium, the minimum concentrations of the added glucose and lactate, that Raman spec-troscopy can detect, are about 0.498 and 0.197 mM respectively.G-1 medium consists of approximately 10.5 mM lactate, 0.5 mM glucose, 0.32 mM pyruvateand a few amino acids. We can conclude that Raman spectroscopy will fail to measure the slightchanges in such low levels of glucose, owing to its 0.498 mM LOD and 0.521 mM RMSEP inG-1 medium. On the other hand, it is more likely, that Raman spectroscopy can detect a changein lactate concentration, as the G-1 medium contains 10.5 mM lactate and Raman spectroscopyoffers a lower LOD (0.197 mM) for sodium lactate in this sample matrix. As long as the singlehuman embryo produces enough sodium lactate to make a relative change by more than 0.2 mM.Raman spectroscopy will differentiate single embryo spent medium from the blank. Likewise,if an embryo, which results in pregnancy, produces 0.2 mM lactate greater than that resulting innonpregnancy, Raman spectroscopy ought to predict embryo viability by analyzing its culturemedia.LOD and prediction error depend on the molecular Raman activity and instrument settingsuch as laser power, laser focus size, acquisition time, etc. Stronger Raman signal, higher laserpower or longer acquisition time used during the measurement will enable the detection oflower concentrations by Raman spectroscopy. However, the available embryo spent culture me-dia are very limited in volume. Only 18 µL could be used for Raman measurement. Althoughthe aluminum cuvette assists heat transfer, 100 mW laser power and 5 minutes exposure rep-resent maximum for this sample volume. Qu et al. have measured glucose concentration inpre-ultrafiltered blood sample by illuminating a 100 mW 785 nm laser for 5 minutes[26]. TheirRMSEE for glucose was 0.38 mM[26]. Ren and Arnold have studied a six-component mixedsolution in aqueous phosphate buffer with 40 s integration time under exposure of 183 mW 785nm laser and reported 0.32 mM and 0.46 mM RMSEE for glucose and lactate respectively[220].Compared with their conclusion, our NAS results shown in Figure 3.3 are very reasonable, cer-tifying current instrument parameters and methodology for real patient samples.3.3.4 Analysis of IVF patient samplesDuring embryo culture, an embryo-free culture medium was placed along with the embryo-containing media in the same petri dish. These embryo-free initial G-1 and secondary G-2 mediaserve as good experiment controls, and Figure 3.4A displays their normalized DWT spectra.363.3. ResultsThe major difference arises from the different absorbance intensities of the lactate feature at400 600 800 1000 1200 1400 1600 1800 2000−4−3−2−10123Raman Shift (cm−1)Raman Intensity / arb. units−2 −1 0 1 2−0.8−0.6−0.4−0.200.20.40.60.811.2PC 1PC 2  G−1 controlG−2 controlFigure 3.4: (A) Normalized DWT spectra of IVF (−) G-1 and (−) G-2 control media; (B) PCA scores plotof (o) G-1 and (o) G-2 control media based on PC1 and PC2.860 cm−1, owing to the two different concentrations of lactate in G-1 and G-2 media. Instead ofcomparing every single band, multivariate variate analysis such as PCA and PLS can evaluate theoverall difference. In Figure 3.4B, PCA clearly differentiates the G-1 controls from G-2 controls.Now the next question is whether Raman spectroscopy can discriminate the control samplesfrom those spent by a single human embryo. Setting aside the final goal, to predict embryoviability by Raman spectroscopic analysis of a embryo spent culture media, it is an importantand necessary preliminary step to compare the embryo-free and embryo-containing media, astheir results can help us study the difference in the culture media spent by embryos resultingin pregnancy and non-pregnancy. A failure to get correct classification differentiating betweencontrol media and patient samples, will call into question the future conclusion that Ramanspectroscopy can predict embryo viability.According to Figure 3.5C and 3.5D, the sample distributions in PCA scores plot show thatRaman spectroscopy does not have the capacity to differentiate controls from the patient sam-ples obtained either on Day 3 or Day 6. In fact, all the principal component scores for thesesamples randomly mix together. I have carefully checked the data against higher principalcomponents. Unfortunately, no choice improves the classification. Moreover, we observe no dif-ferences between the Raman spectra of control media and patient samples obtained on Day 3373.3. Results−3−2.5−2−1.5−1−0.5 0 0.5 1 1.5 2−1.5−1−0.500.511.5PC 1PC 2D−2 −1.5 −1 −0.5 0 0.5 1 1.5 2−1−0.500.51PC 1PC 2C400 600 800 1000 1200 1400 1600 1800 2000−3−2−10123Raman Shift (cm−1)Raman Intensity / arb. units.A400 600 800 1000 1200 1400 1600 1800 2000−3−2−10123Raman Shift (cm−1)Raman Intensity / arb. units.BFigure 3.5: (A) Normalized DWT spectra of (−) G-1 control and (−) samples obtained on Day 3, (B)Normalized DWT spectra of (−) G-2 control and (−) samples obtained on Day 6, (C) PCA scores plot of(o) G-1 control and (∗) samples obtained on Day 3, (D) PCA scores plot of (o) G-2 control and (♦) samplesobtained on Day 6.(Figure 3.5A) or Day 6 (Figure 3.5B). Supervised PLS analysis also fails to perform the correctclassification.Not surprisingly, PLS analysis cannot correlate the Raman spectra of patient samples withtheir associating morphology scores or clinical pregnancy outcomes. This is completely con-tradictory to the published articles, which assert that Raman spectra of embryo spent mediacorrelate with the clinical pregnancy outcomes[193, 194].In another article, Gott et al. have studied the metabolism of a single human embryo fromDay 1 to Day 6 using ultramicrofluorescence, and reported a production of 43.6 pmol/embryo/hlactate on Day 2.5 and 95.4 pmol/embryo/h lactate on Day 5.5 from a normal developed embryo[208].Let us assume that each embryo spends 60 hours in one 50 µL drop of G-1 or G-2 medium be-fore transfer. Then the total production of the lactate from this embryo increases lactate by0.05 mM in G-1 medium on Day 3 and 0.1 mM in G-2 medium on Day 6. This is far below theLOD of lactate by conventional Raman spectroscopy, which explains why both PCA and PLS383.3. Resultsanalysis fail to differentiate the control media from the patient samples. Here, we only con-sider the production of lactate from a normal developed embryo in comparison with the blankmedium. The abnormal developed embryo also produces lactate, but it does not yield as muchas normal one[208]. This implies that singular difference in culture media lactate produced by anormal and abnormal developing embryos cannot be detected by Raman spectroscopy. Admit-tedly, Raman spectroscopy is an analytical tool that evaluates the overall sample matrix insteadof a single component. Our estimation focuses only on the lactate to predict us an inconclusiveanalysis. However, among metabolites, lactate shows a strong Raman signal and represents thepredominate energy substrate for the development of embryo from Day 1 to Day 3[206]. Evenif an embryo normally develops to blastocyst stage, the newly expanded blastocyst only have atotal of 58.3±8.1 cells on Day 5 and 84.4 ±5.7 cells on Day 6[227]. Changes in metabolite con-centrations produced by such a few number of cells appeared to fall below the limit of Ramandetection.3.3.5 Analysis of the culture media spent by the bacteria BacillusI attribute the failure of Raman spectroscopy in the analysis of the single human embryo usedculture medium to its low sensitivity. A further experiment based on bacillus culture supportsthis conclusion. The bacillus have been grown in LB medium, and this used culture medium hasbeen collected at different time points for Raman analysis (Figure 3.6B). As bacillus reproducesexponentially, Raman spectroscopy monitors the change of metabolite concentrations in theseused media. The intensities of the bands at 654, 942 and 1426 cm−1 obviously increase with theincrease of bacillus cells (Figure 3.6A).393.3. Results400 600 800 1000 1200 1400 1600 1800 2000−6−5−4−3−2−10123Raman Shift (cm−1)Raman Intensity / arb. unitsA0 100 200 300 400 500 600 70000.20.40.60.811.21.41.61.82Time (min)OD600  lag phaselog phasestationery phaseBFigure 3.6: (A) Normalized raw and DWT spectra of (−) blank LB media and (−) bacillus spent LB mediacollected at different time points during culture, (B) Bacillus growth curve based on OD600.Here, I also plot the samples based on PC1 and PC2 (Figure 3.7A). Samples clearly distribute−6 −4 −2 0 2 4−1.5−1−0.500.511.52PC 1PC 2  A0 0.5 1 1.5 200.20.40.60.811.21.41.6Reference OD600Predicted OD600  PLS        RMSEE=0.128RMSEP=0.087R=0.982    laglogstationaryFigure 3.7: (A) PCA scores plot of bacillus spent LB media based on PC1 and PC2. (o) LB media collectedduring its lag phase, (M) LB media collected from its log phase, (♦) LB media collected from bacillus sta-tionary phase, (B) OD600 predicted by PLS versus the reference values ((o) calibration and (o) validationdata).along PC1 in the scores plot. Blue circles represent the media collected from the lag phase,followed by the red triangles collected from the log phase. On the far side of PC1, the blackdiamonds correspond to the spent media collected from bacillus in the stationary phase. ThePCA scores plot thus reflects the development of bacillus like the corresponding bacillus growthcurve. In this example, Raman spectroscopy successfully detects the change of metabolites in403.4. Discussionbacterial spent LB media. However, here, millions of bacillus cells caused the changes observedby Raman spectroscopy.As these bacillus cells do not have established morphologic scores or pregnancy outcomeslike human embryos, I cannot use PLS to predict the viability of bacillus cells. Instead, PLSregression analysis can roughly estimate the density of bacteria during their log phase. Figure3.7B plots the predicted OD600 versus the reference values. The RMSEE of the PLS calibrationset is 0.128, while the RMSEP and R of PLS validation set are 0.087 and 0.982 respectively.After the conversion from OD600 to cell density, the prediction error of Raman spectroscopyis 1.28x107 cell/mL in PLS calibration set and 0.87x107 cell/mL in PLS validation set. Theseresults demonstrate how impossible it is for Raman spectroscopy to analyze the culture mediumspent by a single human embryo, which only contains few cells on Day 3 and Day 6. Admittedly,human embryos and bacteria are completely different and their metabolism also differ, but theanalysis of bacillus spent LB provides a dramatic qualitative example to support our previousconclusion.3.4 DiscussionMany studies have applied spontaneous Raman spectroscopy for the analysis of biofluid sam-ples. It has been reported Raman spectroscopy can measure concentrations of total protein,triglyceride, albumin, glucose, cholesterol and urea in human serum samples[26, 228, 229]. Ra-man spectroscopy can also quantify the glucose in human urine samples[230]. As an in-lineprobe, it offers a means to monitor the concentrations in real time of glutamine, glutamate,glucose, lactate and ammonium in bioreactors of mammalian cells or Escherichia coli[231, 232].However, all these studies limit analysis to glucose, lactate, urea or other abundant components,which are not only Raman active and also result in relatively strong Raman signal. Without spe-cial sample treatments in advance, it is impossible to use Raman spectroscopy to detect traceamounts of analytes or slight changes in culture media spent by only a few cells.It is also worth emphasizing that all embryo-containing media are covered by a layer of min-eral oil to prevent evaporation and contamination during embryo culture. After the embryo isremoved, only 25-30 µL of medium remains under the oil. By carefully inserting a pipet tipinto the remaining medium through the thin oil layer, sometimes I still obtain medium contam-413.4. Discussioninated by oil. Our back-scattered Raman system focuses the laser just below the sample surface,and collects Raman signal in 180◦ geometry. The oil floating on the sample surface causes un-desired signal, which seriously contaminates the Raman spectrum. As a consequence, the oilsignal perturbs PCA or PLS analysis, which seeks to correlate Raman spectra with associatedembryo viability.In contrast with these results, Scott, Seil et al. conclude that PLS analysis correlates Ramanspectra with pregnancy outcomes of transferred embryos [193, 194]. They used the Scandina-vian G-1 media (VitroLife AB, Goteborg, Sweden) for Day 1 to Day 3. We used the same G-1medium from VitroLife, but their Raman spectra completely differ from ours (Figure 3.8). TheFigure 3.8: (A) Raman spectra of the culture media obtained on Day 3 from 36 embryos (16 patients). (B)Mean-centered Raman spectra of the embryos, which result in a delivery in comparison with those whodid not implant. These two graphs are published by Seli et al[193].Raman bands in the fingerprint region are normally quite narrow and have a Lorentzian shape,but no such peaks appear in Figure 3.8A. They did not discuss how to obtain clean samples fromoil. Moreover, they did not refer to any embryo-free media along with the embryo-containingmedia in the same petri dish as control samples. Comparison between control samples andembryo spent media need to be considered before directly reaching the conclusion that Ramanspectroscopy can predict the embryo viability by analyzing spent culture media.423.5. Conclusion3.5 ConclusionRaman spectroscopic analysis combined with multivariate analysis such as PCA and PLS failsto correlate with embryo pregnancy outcomes. In particular, PCA analysis of Raman spectracannot differentiate the embryo-free control samples from those media spent by a single humanembryo. PLS analysis of Raman spectra also fails to correlate with embryo morphology scoresor pregnancy outcomes. In my opinion, Raman spectroscopy is not sensitive enough to observethe slight metabolite changes occurring in a 50 µL droplet of culture medium, caused by thedevelopment of a single human embryo from Day 1 to Day 3 or from Day 3 to Day 6. In contrast,Raman spectroscopy detects the changes of the chemical compositions in those LB media spentby the exponential growth of bacillus. The corresponding PCA and PLS analysis demonstratethat Raman detectable changes arise from the metabolism of millions bacteria cells as opposedto the few cells active in a developing human embryo. What is more, the available embryospent culture media are very limited in volume and covered by oil. It remains a challengeto sample clean culture media without the contamination of oil. Therefore, for future study,a more sensitive analytical methodology is needed to study the single human embryo spentculture media to correlate with their associated morphology scores or pregnancy outcomes.43Chapter 4Raman analysis of aqueous solutionswith Teflon-AF 2400 fibre4.1 IntroductionRaman spectroscopy offers a quick means to perform chemical analysis simply by focusing thelaser on a sample. This method is particularly well suited to the gross characterization of bioflu-ids. For instance, Raman spectroscopy has been shown to non-invasively determine the con-centrations of glucose, cholesterol, triglycerides and albumin in serum or blood samples[26,228, 233, 234]. Abu-Absi et. al. have demonstrated Raman spectroscopy can quantify lactate,ammonium, glutamate in mammalian cell culture bioreactors as an in-line probe[231]. It alsoprovides characteristic signatures for body fluids and has recognized its potential for forensicbody fluid identification[25, 235].Despite its advantages, we note that most studies focus on the most abundant analytes ina biofluid. The weak Raman scattering process hinders its application for the determinationof low concentration components. As discussed in the previous chapter, Raman spectroscopyhas proven insufficiently sensitive to detect slight changes of metabolites in the spent culturemedium of a single human embryo compared with embryo-free controls. The present chap-ter addresses this deficiency by exploring the feasibility of an enhanced Raman technique forbiofluid analysis with improved sensitivity for spontaneous Raman scattering, in the hope thatit can differentiate single human embryo spent culture media from control samples.Several enhanced Raman spectroscopy methods are available. Surface enhanced Ramanspectroscopy (SERS) produces a 105 fold enhancement of melamine signal and allows the detec-tion of concentration as low as 0.5 µg/mL of melamine in liquid milk[236]. It also successfullymeasures the physiological concentration of creatinine in human serum or urine, while sponta-444.1. Introductionneous Raman spectroscopy fails[88, 237]. Although SERS’s large enhancement is appealing, itspoor reproducibility and the complexity of SERS-active substrates make it difficult for routinebiofluid analysis. Resonance Raman spectroscopy (RRS) also enhances Raman scattering, but itis more often used to study specific proteins or enzymes[85, 238]. For the purpose of obtainingan overall profile of small molecules such as lactate, glucose and amino acids present in biofluid,RRS cannot help.The analysis of liquid solutions contained in a wave-guiding Teflon fibre also benefits fromsignificantly increased Raman intensities[70, 239–241]. Teflon fibre consists of a hollow coretubing made of an amorphous fluoropolymer Teflon-AF 2400 by DuPont company[43]. Theadvantage of this Teflon material over others lies in a refractive index of 1.29 in the visible andnear-infrared region. This refractive index is lower than that of water (n=1.33) as well as mostother liquids[44]. This allows total internal reflection at the liquid-Teflon interface, as long asthe laser strikes the interface from the higher-refractive-index liquid medium at an appropriateangle. When aqueous solution fills the Teflon tubing, a laser can propagate along the liquid corefibre, which forms waveguide[45]. As a result, the increased light path in the aqueous solutionin Teflon fibre enhances the intensity of the backscattered Raman signal. In Particular, Pelletierand Altkorn report an 80 times enhancement for lysozyme aqueous solution in Teflon fibre,which enables them to detect as low as 54 µM lysozyme[242]. The detection limit of isopropanolin water is reduced by a factor of more than 1000 using Teflon fibre[241]. In contrast with SERSand RRS, Teflon enhanced Raman spectroscopy needs only a Teflon waveguide fibre. It does notneed any SERS-active substrate or a tunable laser system as RRS. Therefore, for practical use, itpromises a much easier and more stable utility than the other two methods.The present study configures a Teflon enhanced Raman system by coupling a backscatteredRaman probe to a Teflon liquid core fibre. During the IVF cycle, human embryo primarilydepends on lactate for the first three days, and the variation in the concentration of lactateconstitutes the predominant change in the Day 3 medium. For other biofluids, lactate is also acommonly studied metabolite. Thus, sodium lactate is selected as a target analyte and a seriesof sodium lactate aqueous solutions are prepared. We then test the Teflon enhanced Ramanspectrum using these lactate aqueous solutions together with a blank embryo culture medium,and comprehensively evaluate the potential of this method for the analysis of single humanembryo spent culture media obtained from IVF treatment.454.2. Experimental4.2 Experimental4.2.1 Reagents and samplesTwo samples of Teflon-AF 2400 tubing were obtained from Biogeneral (San Diego, USA). One is50 cm in length with a 240 µm inner diameter (i.d.) and a 890 µm outer diameter (o.d.). Theother is 28 cm in length with a 110 µm i.d. and a 890 µm o.d. Sodium lactate was purchasedfrom Sigma-Aldrich (Canada). IVF G-1TM PLUS media (VitroLife, Englewood, CO) was suppliedby Children′s and Women′s Health Centre of British Columbia.20% ethanol solution was prepared by diluting ethanol with deionized water. Two sets ofsodium lactate aqueous solutions were prepared. The first set consisted of the following con-centrations of sodium lactate: 0.5, 1, 2, 4, 8, 12, 24, 50, 100 and 200 mM. Three spectra wereacquired for each sample to test the reproducibility of Teflon waveguide enhanced Raman spec-troscopy. The large differences in lactate concentrations also allowed us to detect the charac-teristic bands of lactate and optimize the integrated Teflon Raman system. The second set ofsolutions consisted of twenty-one samples, ranging from 0 to 10 mM in increments of 0.5 mM.We measured these solutions using both Teflon Raman spectroscopy and conventional Ramanspectroscopy. For the purpose of comparing two methods, one spectrum was acquired for eachsample.4.2.2 Instrumentation and Raman measurementsThe Raman microscope system consists of a continuous-wave laser (Innovative Photonic Solu-tions) at 785±0.3 nm, a Raman fibre-optic probehead, an Olympus BX-51 microscope, a spec-trometer (Princeton Instrument Acton SP2300) and a thermo-electrically cooled CCD detector(Princeton Instrument Acton PIXIS 100). The Raman fibre-optic probehead integrates with themicroscope to collect Raman scattering from a micro-volume sample. A multimode optical fi-bre delivers the excitation laser to Raman probehead, where a lens collimates the laser, and abandpass filter eliminates the fibre Raman background. Then a dichroic beam splitter deliv-ers the laser into the Olympus microscope. A 5x single lens objective (f=25 mm, D=13 mm,NA=0.25) focuses the laser into a liquid core Teflon fibre (Figure 4.1A) or a 1 cm cuvette (Figure4.1B). The backscattered Raman signal, collected via the same lens, passes through the dichroicbeam splitter and spatial filter, then refocuses into a six-around-one fibre bundle. At the bundle464.2. Experimentalexit, the seven fibres are aligned vertically and inserted into the spectrometer with a 200 µm en-trance slit. This spectrometer is also equipped with a 600-groove/mm diffraction grating, whichdisperses the Raman signal onto a 1341 x 100 pixel CCD.The Teflon waveguide is mounted in an aluminum block (Figure 4.1A) using a standard 1/16HPLC finger tight fitting (not shown). The size of this aluminum block is similar to a microscopeglass slide, so that it can sit tightly on the microscope stage. Due to the hydrophobicity ofTeflon, the liquid surface at the proximal end of Teflon tubing is either concave or convex. Thisadds difficulty in coupling the laser into fibre and acquiring reproducible Raman spectra. Toovercome this problem, we cut a circle groove, 2 mm in diameter and 1 mm in depth, aroundthe proximal end of Teflon fibre. As the aqueous solution is slowly injected into the Teflon fibre,the excess solution fills the groove. As a result, the proximal end of Teflon tubing is immersedin the solution and a relatively flat solution surface is achieved.A BFigure 4.1: Schematic drawing of two sample cells used in this study. (A) Teflon waveguide fibre mountedin aluminum block with finger tight HPLC fitting, (B) 1cm cuvette made of aluminum.A custom LabVIEW program controls this Raman system and its microscope stage. Thesmallest step size of this microscope stage (Prior Scientific, USA) is 1 µm in X, Y and Z directions,allowing fine adjustment of fibre position. The Raman microscope is also equipped with a videocamera, which provides a clear image of the cross section of Teflon tubing. Figure 4.2A displaysthe camera image of 110 µm i.d. Teflon tubing set on the microscope stage, and Figure 4.2Bexhibits the setup of liquid core Teflon Raman spectroscopy. In contrast with Teflon waveguideenhanced Raman spectroscopy, conventional Raman spectroscopic analysis simply focuses thelaser into an aluminum cuvette with 1 cm in depth (Figure 4.1B). This study calculates theenhancement factor of Teflon waveguide enhanced Raman spectroscopy relative to conventional474.2. ExperimentalRaman spectroscopy (cuvette) by comparing Raman spectra collected with the same sample,laser power and acquisition time.$Figure 4.2: The setup of Teflon Raman spectroscopy. (A) Microscope camera image of 110 µm i.d. Teflontubing, (B) 240 µm i.d. Teflon tubing filled with 20% ethanol solution under exposure to 100 mW 785nm laser4.2.3 Spectral preprocessing and multivariate analysisTo calculate the enhancement factor of Teflon Raman spectrum, an automated method withiterative fifth-order polynomial fitting was applied to raw spectra to subtract the baseline[138].We then calculated the peak height ratio of Teflon waveguide enhanced Raman features andthose observed by conventional Raman spectroscopy as an enhancement factor.Before partial least square (PLS) regression analysis, we applied discrete wavelet transform(DWT) to original spectra using Wavelab850 toolbox in Matlab 2009a[154, 155]. With the Sym-let5 filter, we subtracted the noise and background from Raman spectra, followed by normaliz-ing a constant intensity of one for the water feature at 1640 cm−1.The normalized DWT spectra were separated to calibration and validation dataset, then sub-jected to PLS regression analysis to predict sodium lactate concentrations. The number of PLSfactor was set as 5. The root mean square error of estimation (RMSEE) combined with leave-one-out cross validation was used to evaluate the prediction uncertainty of the calibration model,while the root mean square error of prediction (RMSEP) was calculated to assess the validationdataset.484.3. Results4.3 Results4.3.1 Raman analysis of deionized waterWater serves as a good standard to assess the performance of Raman system, as it has a sim-ple Raman spectrum. Figure 4.3 plots water spectra collected with conventional Raman spec-troscopy via the single lens objective, 10x objective and 50x objective. The single lens clearly400 600 800 1000 1200 1400 1600 1800 20000.60.811.21.41.61.82 x 104Raman Shift (cm−1)Raman Intensity / arb. units.  1640lens10Xobj50XobjFigure 4.3: Raw Raman spectra of water collected with conventional Raman spectroscopy via (−) thesingle lens objective, (−) 10x objective and (−) 50x objective .yields the highest intensity counts for the characteristic band of water at 1640 cm−1. 10x objec-tive decreases the relative intensity of 1640 cm−1 decreases to half of that given by the singlelens. Another broad signal, appearing from 1300 to 1500 cm−1 in the same spectrum, arisesfrom the glass material in this 10x objective. Unlike the single lens and 10x objective, the per-formance of 50x objective is the worst. Not only is the intensity of the overall spectrum verylow, also it is quite difficult to observe the water band at 1640 cm−1. Instead, the interferencesignal from the 50x objective itself dominates the spectrum. Due to its poor performance, weexclude this 50x objective from this study.The enhancement of Teflon Raman spectroscopy depends strongly on the coupling efficiencybetween the laser and the Teflon fibre as well as the propagation of the beam along the liquidcore of Teflon fibre (through total internal reflection). To achieve optimal coupling, the incidentbeam should satisfy the following two conditions: First, the laser focal point must be smaller494.3. Resultsthan the internal diameter of the Teflon fibre[43]. Second, the half angle of the focused lasershould be less than the acceptance angle (θmax) of the filled Teflon fibre, so that Teflon fibrecan waveguide the incident laser[43]. This acceptance angle is determined by the numericalaperture (N.A.) of the filled Teflon fibre (Equation 4.1).N.A. = n ∗ sinθmax =√n2core −n2clad (4.1)Here, the slight difference in the refractive index between water (1.33) and Teflon-AF 2400 (1.29)provides the water filled Teflon fibre a maximum numerical aperture of 0.32 and an acceptanceangle of 18◦. The N.A. of the lens or objective determines the half angle of the incident laser, sug-gesting only optics with an N.A. smaller than 0.32 should be chosen for use with this Teflon fibre.Unfortunately, the often-used 20x and 50x objective do not satisfy this requirement. Among allthe available lenses and objectives, the single lens is the first option. For a smaller focal point,the only option is the 10x objective despite its lower collection efficiency. For present purpose,we adopt the single lens objective as the default collection lens.4.3.2 Raman analysis of aqueous solutions in 240 µm i.d. Teflon-AF 2400waveguide fibreWe couple the Teflon waveguide fibre with our Raman system and perform coarse adjustmentby focusing a very weak laser into the hollow tubing with the assistance of microscope cameraimage. Then we inject 20% ethanol solution into Teflon fibre for further fine adjustment, usingits intense Raman signal within a short 1 s acquisition time. This allows an immediate feedbackafter each slight adjustment and helps locate the optimal position, at which the filled Teflonfibre yields the maximum Raman signal. The refractive index of 20% ethanol aqueous solu-tion is similar to that of water, so switching from 20% ethanol solution to any aqueous samplesdoes not require realignment of the fibre position. Figure 4.4A compares the Raman spectraof 20% ethanol solution, acquired with a Teflon waveguide fibre and a cuvette. The conven-tional Raman spectrum features a relatively flat baseline. In contrast, Teflon enhanced Ramanspectrum significantly shifts upwards owing to the fluorescence light emitted from Teflon-AF2400[70]. After baseline subtraction by fifth-order-polynomial curve fitting, the enhancementfactors based on the peak height ratio at 880, 1049, 1090 and 1283 cm−1 stay at 7.0, 6.9, 6.7, 6.7,504.3. Resultsand 6.5 respectively. A slight decrease of enhancement occurs as the Raman shift increases, asalso reported by Qi and Berger[243].Once the optimal position is achieved, water is injected. Figure 4.4B plots the Raman spectraof water, collected with Teflon fibre and cuvette. The enhancement factor at 1640 cm−1 remainsaround 4. Surprisingly, the enhancement is much lower, and the enhanced spectrum does notseem as significant as in 20% ethanol. A lot of efforts was made to optimize the fibre position,but none improved the enhancement. We note Li et al. have also observed a dramatic discrep-ancy between the enhancement of 0.1 M aqueous NaHCO3 (∼20) and benzene (∼120) using thesame Teflon fibre[244]. Compared with the conventional Raman spectrum of water in Figure4.4B, three extra bands at 713, 755 and 831 cm−1 appear in the enhanced water spectrum. Af-ter comparing them with Teflon spectrum, we could identify them as Teflon Raman signal. Infact, these three bands also appear in the enhanced Raman spectra of 20% ethanol, but they areoverwhelmed by the strong signal from ethanol.After water, Figure 4.4C presents the Raman spectra of 100 mM sodium lactate. The charac-teristic bands of lactate appear at 540, 831, 1045, 1085, 1420 and 1455 cm−1[226]. The enhance-ment factors at 540, 860 and 1640 cm−1 are 5.7, 5.2 and 5.4. The Teflon peak at 831 cm−1 standsvery close to 860 cm−1, so this interference undoubtedly suppresses the lactate feature at 860cm−1 and lowers the corresponding enhancement factor. Likewise, if any other Raman signal ofinterest appear in the region from 700 to 870 cm−1, the Teflon signal will disturb the analysis.In contrast, the spectrum collected with cuvette does not show this problem.The difficulty of uniformly coupling the incident laser into the Teflon fibre decreases thereproducibility of the enhanced spectra compared with conventional spectra. To overcome thisproblem, normalization is necessary. Here, DWT transformed spectra are normalized to a con-stant intensity of one at water feature 1640 cm−1. For the first set of sodium lactate solutionsranging from 0.5 to 200 mM, three spectra were collected for each sample. The normalizedtriple spectra from each sample overlie well (Figure 4.4D), indicating the effective correction ofthis pretreatment method. We also find it is very difficult to observe the lactate signal using 10s exposure of a 100 mW laser, when the concentration decreases to lower than 25 mM. Longerexposure time and smaller i.d. Teflon tubing may improve detection of lower concentrations oflactate.514.3.Results400 600 800 1000 1200 1400 1600 1800 200000.511.522.533.544.5 x 104Raman Shift (cm−1)Raman Intensity / arb. units.  88043210491090128314601640400 600 800 1000 1200 1400 1600 1800 20001234567891011 x 104Raman Shift (cm−1)Raman Intensity / arb. units.  713755831 1247−13171364400 600 800 1000 1200 1400 1600 1800 200024681012 x 104Raman Shift (cm−1)Raman Intensity / arb. units.  5407137558318601045108514201455400 600 800 1000 1200 1400 1600 1800 2000−2−10123456Raman Shift (cm−1)Raman Intensity / arb. units.  540 71375583186010451085 1420145593020%EtOH20%EtOH−TFWater−TFWaterTeflon100mM Lactate−TF100mM LactateFigure 4.4: Raman spectra of aqueous solutions acquired with a 240 µm i.d. Teflon fibre and 1 cm cuvette by 100 mW laser. (A) 20% ethanol for1 s exposure, (B) water for 10 s exposure (water spectrum collected by cuvette is multiplied by a factor of 3 for comparison), (C) 100 mM sodiumlactate solution for 10 s exposure (100 mM lactate spectrum collected by cuvette is multiplied by a factor of 3 for comparison), (D) 0.5-200 mMsodium lactate solutions collected with Teflon fibre for 10 s exposure.524.3. Results4.3.3 Raman analysis of aqueous solution in 110 µm i.d. Teflon-AF s2400waveguide fibreWe use the 20% ethanol solution to assess the performance of Teflon fibres with different internaldiameters. We observe a stronger Raman signal from ethanol and a higher baseline, as theinternal diameter of Teflon fibre decreases from 240 µm to 110 µm (Figure 4.5).400 600 800 1000 1200 1400 1600 1800 200012345678 x 104Raman Shift (cm−1)Raman Intensity / arb. units.  4328801049109012831460164020%EtOH20%EtOH−TF240um20%EtOH−TF110umFigure 4.5: Raman spectra of 20% ethanol solution collected with (−) 1 cm cuvette, (−) 110 µm i.d. Teflonfibre and (−) 240 µm i.d. Teflon fibre by 100 mW 785 nm laser for 1 s exposure.In order to estimate the enhancement of this technique, we also collect the Raman spectrumfrom this 20% ethanol solution with a 1-cm cuvette. On the basis of this, 110 µm i.d. Teflonfibre yields approximately 14 times enhancement for 20% ethanol solution, whereas 240 µmi.d. Teflon fibre yields about 7 times enhancement. Altkorn et al. have also discussed theeffect of internal diameter on the enhancement[70]. The Teflon fibre with a smaller internaldiameter usually has a higher attenuation loss, which arises from the higher reflection loss owingto the inhomogeneous surface of the inner fibre wall[70]. Despite the significant increase ofattenuation loss with a decrease of internal diameter, 50 µm i.d. Teflon fibre has about 4 timesgreater efficiency to couple the signal back to Raman system than 150 µm i.d. fibre, ultimatelyleading to a larger enhancement of Teflon fibre at smaller core diameter[70]. As a result, wechoose 110 µm i.d. Teflon fibre together with 300 s acquisition time (co-adding 20 spectra of 15s exposure) to detect lower concentrations of lactate.Unfortunately, long exposure of 100 mW causes local over-heating and air bubbles at the534.3. Resultsproximal end of the Teflon tubing. This hinders the acquisition of high-quality Raman spectra.A 1 mm thick quartz window, placed on top of the groove, relieves the over-heating and main-tains the stability of this technique. It also flattens the solution held in the groove, so the lasercan still focus into the Teflon fibre through this thin quartz window. It is necessary to carefullyremove the quartz window and clean the Teflon fibre after finishing the measurement of eachsolution. It would be better to have a complete interface, allowing the discharge of waste solu-tion, cleaning and refilling with a new sample into the Teflon fibre without touching the quartzwindow. Such a setup was tested, but trapped air bubbles at the proximal end of the Teflon tub-ing scattered the incident laser and prevented the coupling of the laser into the Teflon tubing.For present purpose, all the data were acquired by simply placing a quartz window on top ofholder.Figure 4.6 compares the performance of 110 µm i.d. Teflon fibre and 1 cm cuvette basedon the Raman spectra of 0-10 mM sodium lactate aqueous solutions. The Teflon signal at 831cm−1 interferes with the lactate feature at 860 cm−1. However, the enhanced spectra presenthigher signal-to-noise ratio than conventional spectra. PLS analysis of both datasets is com-pared. 10 samples are selected as calibration set and the remaining 11 samples serve as a vali-400 600 800 1000 1200 1400 1600 1800 2000−2−1.5−1−0.500.511.5Raman Shift (cm−1)Raman Intensity / arb. units.86054014201045 1455860Figure 4.6: Normalized DWT Raman spectra of 1-10 mM sodium lactate solutions obtained with (−) a 28cm long 110 µm i.d. Teflon fibre and (−) 1 cm cuvette by 100 mW 785 nm laser for 300 s exposure.dation set. We calculate RMSEP from validation dataset to evaluate the prediction uncertaintyby these two methods (Table 4.1). 110 µm i.d. Teflon only slightly improves RMSEP from 0.564to 0.401 mM, compared with conventional method. However, such small difference in RMSEP544.3. ResultsTable 4.1: Summarized PLS calculation from validation data based on the Raman spectra acquired with110 µm i.d. Teflon fibre and 1 cm cuvetteSample No. [Lactate] Predicted [Lactate] Predicted [Lactate]by 110 µm i.d. Teflon by 1 cm cuvette1 0.51 0.08 1.302 5.59 4.91 5.243 6.60 6.00 6.354 7.62 8.09 7.315 8.13 7.82 7.456 8.64 8.70 7.907 9.65 9.87 9.038 1.02 0.99 1.379 2.03 1.63 2.5510 3.05 3.00 3.8311 3.56 4.03 4.02RMSEP 0.401 0.564,indicates the prediction uncertainties of these two methods have the same scale. Thus, we con-clude that the Teflon waveguide enhanced Raman spectroscopy is still insensitive to analyze thesingle embryo spent culture media obtained from IVF treatment.From Figure 4.4 to Figure 4.6, Teflon signal appears in all Raman spectra, suggesting thecurrent system is not well optimized. Thus, we tried to use the 10x objective (N.A. 0.25) toimprove the coupling and eliminate the Teflon signal. Even though 10x objective offers a smallerfocal point, Teflon features still exist and suppress the lactate signal (Figure 4.7).400 600 800 1000 1200 1400 1600 1800 2000−1.5−1−0.500.511.5Raman Shift (cm−1)Raman Intensity / arb. units.  860Blank G−1−TF−10X objBlank G−1−lensFigure 4.7: Normalized DWT Raman spectra of IVF G-1 media acquired with (−) 110 µm i.d. Teflon fibreand 10X objective in comparison with that (−) acquired with a cuvette and single lens .554.4. DiscussionDecreasing the laser focal point does not help solve the problem. Both the single lens and10x objective satisfy the acceptance angle of aqueous solution filled Teflon fibre. A lot of effortwas made to optimize the coupling between the laser and Teflon fibre, but none could removethe Teflon signal in Raman spectra. In this circumstance, this problem probably arises fromdiffuse scattered laser on the inhomogeneous surface of the inner fibre wall owing to imperfectmanufacture. The scattered laser light also produces Teflon Raman signal, which are collectedback by our Raman microscope system. In an article published by Qi and Berger, we note bandsfrom Teflon also appear in their spectra of aqueous samples[43].As discussed in the previous session, the collection efficiency of 10x objective is about half ofthe single lens. The little enhancement contributed by Teflon fibre together with the low collec-tion efficiency of 10x objective does not show any advantage of this Teflon technique in Figure4.7. Even worse, Teflon signal in the region from 731 to 831 cm−1 and glass signal between 1200and 1400 cm−1 dominate the enhanced spectrum of G-1 medium.4.4 DiscussionFrom our experimental data, Teflon waveguide fibre yields very low Raman enhancement. Incontrast, the enhancement factors reported by other references vary from 4.5 to 500 (Table 4.2).The wide range of the enhancement factors suggests the complexity of this technique. We shouldTable 4.2: The summary of Teflon enhanced Raman spectroscopy.ID Length λex Sample Enhancement Ref.(µm) (cm) (nm) relative to240 50 785 20% ethanol 7; 1cm cuvette exp. result240 50 785 100 mM sodium lactate 6; 1cm cuvette exp. result110 28 785 20% ethanol 14; 1cm cuvette exp. result600 17 830 20% ethanol 8; 4cm cuvette [243]600 17 830 0 - 150 mM Creatinine 4.5; 4cm cuvette [43]500 100 785 100 mM Na2CO3 20; 1cm cuvette [244]500 100 785 Benzene 120; 1cm cuvette [244]150 393 785 Methanol 37-9; 1.5cm vial [70]150 393 532 Methanol 56; 1.5cm vial [70]50 121.5 532 water 500; 1cm cuvette [242]50 100 785 10% 2-propanol 80; glass vial [241]50 92 532 60 mM phenylalanine 90, vial [70]50 41 785 60 mM phenylalanine 58; vial [70]564.4. Discussionnot compare these enhancement numbers directly, because many experimental factors affect theenhancement.First, the enhancement relies upon the wavelength of excitation laser. Altkorn et al. havemeasured the Raman spectra of methanol using a 393 cm long, 150 µm i.d. Teflon fibre and a1.5 cm vial under exposure of 532 nm and 785 nm laser[70]. 532 nm excitation yields a higherenhancement, owing to its smaller attenuation loss in methanol-filled Teflon fibre than the 785nm laser[70].Second, the liquid used to fill the Teflon fibre also plays an important role in the enhance-ment. Benzene-filled Teflon fibre yields around 120 times enhancement, while the same Teflonfibre filled with 0.1 M Na2CO3 solution only gives 20 times enhancement[244]. We also ob-served this phenomenon when we compared the enhancement factor between 20% ethanol and100 mM sodium lactate aqueous solution. To quantitatively evaluate the Raman intensity en-hanced by Teflon Raman spectroscopy in backscatter mode, Altkron et al. have proposed anequation (Equation 4.2),PR =PLK2α(1− e−2αx) (4.2)where PR is the Raman intensity, PL is the laser power, α is the loss coefficient of Teflon fibre,K is the constant related with the Raman cross section of the analyte, and x is the Teflon fibrelength[70]. As the fibre length is much greater than 1/α, PR approaches its peak value at PR,maxas Equation 4.3.PR,max =PLK2α(4.3)PR,max significantly decreases with an increase of the loss coefficient α. As water has more than20 times greater absorption coefficient than benzene[244], the laser decays much faster in aque-ous solutions. The larger loss coefficient of aqueous solution compared to that of benzene yieldsa smaller PR,max in aqueous solution and a smaller enhancement as well. Likewise, this modelalso explains the higher enhancement of 532 nm excitation than 785 nm laser.Third, the inner diameter of Teflon tubing also affects the enhancement of Teflon fibre. Wehave compared the enhancement factor of 20% ethanol solution in 110 and 240 µm i.d. Teflontubing. The smaller the inner diameter of Teflon fibre, the higher the enhancement factor.574.4. DiscussionWe also note the enhancement reported by Qi and Berger is close to what we found in mystudy[43, 243]. They have explained such low enhancement factor using another quantitativemodel including the efficiency of total internal reflection (R≤1)[243]. Instead of simply assum-ing the loss coefficient as a constant α in Altkorn’s model, Qi and Berger have pointed out the losscoefficient is a function of propagation angle[243]. Only a fractional beam total internally re-flects, while the rest diffusely scatters or experiences other non-absorptive mechanisms[243]. Bymeasuring the transmitted power at the end of Teflon fibres of various lengths, they obtain a R of0.966 in their setup. Based on this number, the predicted enhancement factor from their modelmatches well with an experimental enhancement factor of 9 for 20% ethanol in the 830 nm Ra-man spectrum[243]. In another study, Altkorn and Pelletier have also mentioned the significanteffect of imperfect inner fibre wall on the enhancement of Teflon Raman spectroscopy[242]. Awater-filled 1 m long 50 µm i.d Teflon fibre yields 120 times enhancement and transmits 20% of532 nm laser, while another 1.21 m long 50 µm i.d Teflon fibre yields 500 times enhancementand transmits 43% of laser[242]. Thus, the scattering loss on the inner fibre wall must be one ofthe primary causes for the low enhancement of our waveguide.In our Raman system, the Olympus microscope integrates with a Raman probe using a mul-timode fibre to deliver the laser and a six-around-one multimode fibre bundle to collect thebackscattered Raman signal. This results in a deep depth-of-field, so that the incident lasermay not completely enter into Teflon fibre at the focal point. The uncoupled laser undoubtedlyinteracts with Teflon and produces Teflon signal. Meanwhile, the six-around-one multimodefibre bundle collects as much backscattered Raman signal as possible, including the signal fromaqueous solution and Teflon fibre. In this circumstance, the instrument setup itself accounts forthe low enhancement of this technique as well as those unavoidable Teflon features.Based on the discussion above, liquid core Teflon fibre does enhance Raman signal. Toachieve a good enhancement, however, it places demanding requirements on the instrumen-tal setup, Teflon fibre, and efficient coupling between laser and fibre. Any slight deviation leadsto a low enhancement. According to Qi and Berger, the enhancement factor of creatinine aque-ous solution is only 4.5[43]. We also observe a similar low enhancement for lactate aqueoussolution. Furthermore, it is very difficult to align the laser, locate the optimal position and in-ject the sample without any displacement of fibre. Although Teflon fibre only consumes severalmicroliters of samples, it requires the expense of far more solution to avoid air bubbles in the584.5. Conclusionhydrophobic Teflon fibre. A 25 µL embryo used medium is not enough for this method. In con-trast, Raman spectroscopy using a cuvette is much easier and faster. The RMSEP and RMSEEof sodium lactate by PLS analysis suggest these two methods lie in the same scale. Thus, thistechnique is not recommended to analyze IVF samples. A more sensitive analytical method isneeded to study this type of sample.4.5 ConclusionRaman spectroscopy coupled with Teflon waveguide fibre was studied. Relative to 1 cm cuvette,a 50 cm long 240 µm i.d. Teflon fibre yields 6 times enhancement for 100 mM sodium lactatesolution and 7 times enhancement for 20% ethanol. Likewise, a 28 cm long 110 µm i.d. Teflonfibre offers a 14 times enhancement for 20% ethanol.In contrast with conventional Raman spectroscopy, Teflon Raman signal and low enhance-ment actually hinder the practical use of this technique in the analysis of IVF specimens. Factorssuch as the wavelength of excitation laser, the inner diameter of Teflon fibre, the liquid used tofill the fibre core, the setup of Raman system and the overall reflection efficiency (R) highly re-late with the manufacture quality of Teflon fibre, determine the enhancement of this technique.Among them, the quality of Teflon fibre and the instrument setup probably have the greatesteffect on the enhancement. Any imperfect inner fibre wall or improper selection of instrumentalcomponents lead to losses associated with diffuse scattering or inappropriate coupling betweenthe laser and Teflon fibre and dramatically decrease the enhancement.Generally, it is very difficult to couple the laser into Teflon fibre and obtain the reproduciblespectra, as any slight displacement or small air bubbles appearing in Teflon fibre completelychange the spectrum. The reproducibility of this technique is much worse than cuvette-basedconventional Raman spectroscopy. PLS analysis was applied to Raman spectra of 0-10 mMsodium lactate aqueous solutions, collected with 110 µm i.d. Teflon fibre and 1 cm cuvetterespectively. Teflon enhanced Raman spectroscopy does not exhibit any obvious sensitivity im-provement. Therefore, Teflon Raman spectroscopy is not suitable to analyze IVF culture mediaspent by single human embryo. Instead, more sensitive analytical methods are needed.59Chapter 5Detection of the adulteration ofvegetable oil in extra virgin olive oil byRaman spectroscopy5.1 IntroductionOlive oil, processed from the fruit of olive trees, mainly consists of monounsaturated fatty acids.Studies suggest that these acids represent a healthier dietary fat than saturated fatty acids, asthey can help reduce the low-density (LDL)-cholesterol and total cholesterol levels in the blood-stream, resulting in a lower risk of cardiovascular disease[245, 246]. They are also superior topolyunsaturated fatty acids, as they are less likely to get oxidized[247]. Furthermore, the pheno-lic compounds present in olive oil play important roles as antioxidants, offering demonstratedprotection from (colon, breast, skin) cancer and cardiovascular disease[46, 47].High quality olive oils contain a particularly large amounts of antioxidants, such as squaleneand various phenolic compounds[248]. These beneficial nutritional elements combined witholive oil’s pleasant flavor, have increased the worldwide popularity of olive oil, even though itcosts considerably more than other cooking oils. This high sale price has created an incentive toadulterate olive oil with cheaper vegetable cooking oils. Adulteration presents a serious problemfor olive oil industry. For instance, the UC Davis Olive Oil Center reported in 2011 that 73% ofimported extra virgin olive oil in a representative sample of the California market sample failedto pass the standard purity test of the International Olive Council (IOC)[2]. Therefore, thedetection and identification of vegetable oil adulterants in olive oil have significant importancefor the olive oil industry.In order to properly regulate olive oil production and monitor its authenticity, the IOC has605.1. Introductionestablished standard tests based on wet chemistry and chromatographic methods (i.e. HPLC,GC) for the identification of vegetable oils and the quality of olive oil. However, the expenseof complicated sample preparation and time-consuming separation make it impossible to effi-ciently test anything more than a small fraction of the large amount of olive oil on the market.On the other hand, in recent years, numerous reports have claimed that spectroscopic methodswith simpler operation and sample preparation have the potential to replace classical character-ization methods. 31P and 1H Nuclear Magnetic Resonance (NMR) combined with multivariateanalysis distinguish a variety of genuine Greek virgin olive oils from adulterated oils containinga 5% component of cheaper vegetable oils[249, 250]. Synchronous fluorescence spectroscopyreportedly quantifies cheaper vegetable oils in virgin olive oil diluted 1% in hexane with a de-tection of limit of 3-4% for sunflower, corn, soybean, rapeseed or olive-pomace, but 13.8% forwalnut oil[251, 252]. Compared with NMR and fluorescence spectroscopy, vibrational spec-troscopy methods, such as Fourier Transform Infrared spectroscopy (FTIR), Near Infrared spec-troscopy (NIR) and Raman spectroscopy, have gained increasing attention, owing to their simplesetup and success in the classification of different vegetable oils. There are now many studies us-ing either FTIR[253–257], NIR[256, 258–260] or Raman[256, 261–264] to discriminate betweengenuine and adulterated olive oils.However, contrary to frequent reports in the spectroscopic literature, the detection of oliveoil adulteration is not a simple task. To begin with, there exist hundreds of different varietiesof olive trees. Most olive oils come from the Mediterranean region. Australia, Peru, and theUSA also produce a substantial amount of olive oils. As a result, the chemical components (i.e.saturated, monounsaturated and polyunsaturated fatty acids) of olive oil vary from sample tosample owing to such factors as olive tree variety (cultivar), climate, country of origin and geo-graphical region, ripeness, production system, etc. The ideal analytical method must recognizethis large variety of olive oils as being authentic and distinguish them from adulterated oils.However, spectroscopic methods have succeeded in classifying extra virgin olive oils based onthe country of origin[265]. Tapp and coauthors, for example, have used FTIR to identify extravirgin olive oils from Portugal, Italy, Greece, and Spain[265]. Even extra virgin olive oils har-vested from a single country vary by geographic regions, because of climate, soil conditions,olive cultivars, etc. Downey et al. have applied NIR analysis to differentiate extra virgin oliveoils on the basis of harvest location in such area of Greece as Crete, Peloponnese and other615.2. Experimentallocations[266]. By performing FTIR analysis, Gurdeniz et al. have classified extra virgin oliveoils of three Turkish cultivars (Erkence, Ayvalik and Nizip) and their mixtures (Erkence-Nizipand Ayvalik-Nizip)[267]. Casale et al. have used NIR and FTIR to classify extra virgin olive oilsfrom cultivars, Casaliva, Leccino and Frantoio, grown in three different regions of Italy[268].Further, FTIR has also been used to evaluate the freshness of virgin olive oil from the Garda re-gion as sampled for different storage times and conditions [269]. From these reports, it appearsthat vibrational spectroscopy can successfully differentiate a wide variety of olive oil samples.In this circumstance, whether vibrational spectroscopy can still simultaneously distinguish awide variety of authentic olive oil samples from the adulterated ones are unknown.Constrained sampling remains as a major issue with all of previous adulteration studies.Of particular note is the fact that many of the samples investigated were purchased from localgrocery stores, evidently without detailed information about the origin or cultivar. Furthermore,for cases in which such cultivar details are provided, only a limited number of samples werestudied. This lack of breadth and specificity very likely leads to small-scope, over-optimisticmodels for the detection of extra virgin olive oil adulteration in general. The need for newanalytical methodologies for the identification of different vegetable oils and the detection ofadulteration is driven by consumer demand for a wide variety oils of different quality and fromdifferent geographic regions.The present work seeks to use Raman spectroscopy as a means to study the effect of au-thentic variety on classification when distinguishing between genuine extra virgin olive oils andadulterated oils. In order to comprehensively assess various extra virgin olive oils for adulter-ation with edible vegetable oil, we purposely choose extra virgin olive oils grown and producedin different countries, and also consider a variety of adulterants, including canola, corn, grapeseed and walnut oil for binary adulteration systems.5.2 Experimental5.2.1 Instrumentation and Raman measurementsRaman spectra were recorded using 785±0.3 nm output of a continuous-wave laser (InnovativePhotonic Solutions) for excitation. A home-built Raman fibre-optic probe integrates this laserwith an Olympus BX50 microscope to collect Raman scattering from oil samples. This probe di-625.2. Experimentalrects the backscattered signal to a spectrometer (Princeton Instrument Acton SP2300) via mul-timode fibre optics. A 5x lens objective focuses the laser into a 2 mL sample of oil, contained ina standard 2 mL HPLC vial (Canadian Life Science). The same lens collects the backscatteredRaman signal and passes it through a dichroic beam splitter and a spatial filter. The probeheadthen refocuses this light into a six-around-one multimode fibre bundle for transmission to thespectrograph. At the bundle exit, the seven fibres are aligned vertically and connected to thespectrometer. The spectrometer is equipped with a 600-groove/mm diffraction grating. Theentrance slit width is set to 200 µm. A thermoelectrically-cooled CCD detector (Princeton In-strument Acton PIXIS 100) with a chip size of 1341 x 100 pixels detects the signal in the Ramanshift range of 300-2000 cm−1. A custom LabVIEW program controls the entire Raman system.The integration time for each oil sample was set to 0.5 s with 20 co-added accumulations. Threespectra were acquired for each individual sample.5.2.2 Oil samplesFifteen monovarietal extra virgin olive oils with defined variety were purchased from Vancou-ver Olive Oil Company. Oils supplied by this company were tested by Australia Modern OlivesLaboratories (a member of the IOC). An additional four extra virgin olive oil samples withoutdefined variety and the other vegetable oils were purchased from local markets. Table 5.1 sum-marizes the information on the pure vegetable oils and Table 5.2 summarizes the informationon all of these pure extra virgin olive oils. Adulterated samples were prepared by randomlychoosing one extra virgin olive oil and mixing one particular vegetable oil (canola, corn, grapeseed or walnut oil) using weight percentages of 5%, 10%, 20% and 50%.Table 5.1: The summary of pure vegetable oils.Type of Oil NumberCanola 2Corn 2Sunflower(High Oleic acid) 1Peanut 1Grape seed 2Safflower 2Walnut 1Extra virgin olive oil 19635.3. ResultsTable 5.2: The summary of pure extra virgin olive oils.Type of Oil Origin Variety Production year QuantityEVOO1 Spain not declared not clear 1EVOO2-4 Italy not declared not clear 3EVOO5 Australia Picual 2012 1EVOO6 Portugal Corbrancosa 2012 1EVOO7 Spain Manzanilla 2012 1EVOO8 USA (California) Arbosana 2012 1EVOO9 Greece Koroneiki 2012 1EVOO10 Tunisia Chetoui 2012 1EVOO11 Spain Martena 2012 1EVOO12 Tunisia Chemlali 2012 1EVOO13 Italy Nocellara 2012 1EVOO14 Italy Cerasuola 2012 1EVOO15 USA (California) Arbequina 2012 1EVOO16 Italy Coratina 2012 1EVOO17 Peru Barnea 2013 1EVOO18 Peru Picual 2013 1EVOO19 Spain Picual 2013 15.2.3 Spectral preprocessing and multivariate analysisRaw Raman spectra were preprocessed as follows. For each sample, three collected Raman spec-tra were averaged into one. The baseline of each individual Raman spectrum was removed byapplying discrete wavelet transform (DWT)[154, 270]. Background subtracted Raman spectrawere normalized to a constant intensity of one at 1441 cm−1 as shown in Figure 5.1.Multivariate analysis such as principal component analysis (PCA) was performed with Mat-lab (R2009a). DWT was performed with the Wavelab 850 toolbox[154, 155], and linear discrim-inate analysis (LDA) was performed with STPRtool toolbox[182, 183].5.3 Results5.3.1 Raman analysis of pure vegetable oilsFigure 5.1 shows the normalized Raman spectra of pure vegetable oils in the fingerprint regionof 300 to 2000 cm−1. Major Raman bands occur at 835, 866, 968, 1078, 1120, 1263, 1303, 1441,1658 and 1748 cm−1. Based on previous works, the corresponding band assignments are listedin Table 5.3[256, 261–263, 271]. Among them, the most distinguishable features appear around835, 866, 968, 1120, 1263 and 1658 cm−1 (Figure 5.1). Corn, grape seed, and walnut oil do not645.3. Results400 600 800 1000 1200 1400 1600 1800 2000−1.5−1−0.500.511.522.5Raman Shift (cm−1)Raman Intensity / arb. units WalnutGrape seedCornCanolaSafflowerSunflower PeanutEVOO1441165817481303126310789681120866835722Figure 5.1: Normalized DWT transformed Raman spectra of pure vegetable oils after DWT and normal-ization to 1441 cm−1.Table 5.3: Assignment of observed bands in Raman spectra of vegetable oils.Raman shift (cm−1) Functional Group Vibrational Mode Reference1748 RC=OOR C=O stretch [261–263, 271]1658 cis-RHC=CHR C=C stretch [261–263, 271]1441 -CH2 -C-H bend (scissor) [261–263, 271]1303 -CH2 -C-H bend (twist) [261–263, 271]1263 cis-RHC=CHR =C-H bend (scissor) [261, 263, 271]1078 -(CH2)n- C-C stretch [261–263, 271]866 -(CH2)n- C-C stretch [263]have a peak at 1120 cm−1, while canola, sunflower (high oleic acid), peanut, safflower and extravirgin olive oil have a feature at this position.The Raman intensities of the peaks at 1263 and 1658 cm−1 obviously increase in going fromextra virgin olive oil to walnut oil (Figure 5.1), suggesting an increased fraction of cis-RHC=CHRbonds. The double bond on both monounsaturated and polyunsaturated fatty acids contributesto this cis-RHC=CHR Raman signal, so oils, especially like walnut, grape seed, corn, canolaetc, containing a greater proportion of polyunsaturated fatty acid, display stronger scatteringat Raman shifts of 1263 and 1658 cm−1. Likewise, the addition of vegetable oil with higherpolyunsaturated fatty acids as adulterant into extra virgin olive oil undoubtedly increases thesetwo bands as well. This conforms with observations of Zhang et al., who have found an increase655.3. Resultsin the band intensities at 1263 and 1658 cm−1 with increasing weight percentage of soybeanoil in olive oil[262]. Thus, the signal correlating with cis-RHC=CHR in Raman spectroscopicanalysis must play an important role in differentiating various vegetable oils and adulteratedextra virgin olive oils.5.3.2 PCA analysisWe apply PCA to the normalized Raman spectra of 30 pure vegetable oils. From Figure 5.2A,the scores plot based on PC1 and PC2 gives a good classification of different pure vegetable oils.Walnut oil appears the furthest away from extra virgin olive oil, followed by grape seed, corn,−4 −3 −2 −1 0 1−0.3−0.2−0.100.10.20.30.4 11223444444444 444444 4456 6778PC1PC2−4 −3 −2 −1 0 1−0.100.10.20.30.40.51 12234444444 4444 44566778PC1PC3Figure 5.2: PCA scores plot of pure vegetable oils based on (A) PC1 vs. PC2 and (B) PC1 vs. PC3. (1q)canola oil, (2 I)corn oil, (3 )sunflower oil (high oleic acid), (4F)extra virgin olive oil, (5 )peanut oil,(6 )grape seed oil, (7 )safflower oil and (8F)walnut oil.canola, sunflower (high oleic acid) and safflower oils. Peanut oil seems quite similar to extravirgin olive oil. We see here that different varieties of extra virgin olive oils form a large cluster,which almost overlaps with peanut oil. The distribution of different vegetable oils along the axisof PC1 and PC2 (Figure 5.2A) agrees well with the evident increase in the relative intensities ofthe peaks at 1263 and 1658 cm−1, suggesting PC1 and PC2 correlate with these features. This isindeed confirmed by the PC1 and PC2 loading plot shown in Figure 5.3. The PC1 loading plotexhibits features with large negative values at 1263 and 1658 cm−1, indicating that samples withhigher absorbance intensities at 1263 and 1658 cm−1 tend to move in the negative direction onthe PC1 axis. On the other hand, features in the region from 835 to 866 cm−1 together with 1658665.3. Resultscm−1 dominate the PC2 loading plot. In particular, samples with higher absorbance intensitiesat both 866 and 1658 cm−1 as well as lower absorbance value at 835 cm−1 shift in the positivedirection on the PC2 axis. As a result, in Figure 5.2A, both walnut and canola oil appear on thetop of PC2 compared with grape seed and corn oil on the opposite direction.200 400 600 800 1000 1200 1400 1600 1800 2000 2200−1−0.500.511.5Raman Shift (cm−1)Raman Intensity / arb. unitsPC1PC2PC3SpectraFigure 5.3: PCA loading plot of PC1, PC2 and PC3 compared with Raman spectra of pure vegetable oils.Although oil samples produce intense Raman signal, accompanying fluorescence can de-crease the spectral signal-to-noise ratio. Among the spectra of pure vegetable oils in Figure 5.1,peanut oil is affected by fluorescence most seriously, significantly reducing its spectral signal-to-noise ratio. PC3 picks up this fluorescence component in the spectral region from 300 to 400cm−1 and from 1800 to 2000 cm−1. In the PCA scores plot based on PC1 and PC3 (Figure 5.2B),the peanut oil and one grape seed oil sample appear on the top of PC3 due to the higher fluo-rescence than other samples. In total, Raman spectroscopy can satisfactorily classify differenttypes of vegetable oils using PC1, PC2 and PC3.Since there are many varieties of extra virgin olive oils, PCA is also performed on the Ramanspectra of multiple samples of extra virgin olive oil. In this three-dimensional PCA scores plotshown in Figure 5.4, extra virgin olive oils class roughly by country.675.3. Results−0.2 −0.10 0.10.2 0.3−0.2−0.100.1−0.15−0.1−0.0500.05  PC3PC1PC2ITA−16−CoratinaPOR−6−CorbrancosaITA−14−CerasuoloSPA−19−PicualITA−4GRE−9−KoroneikiSPA−11−Martena ITA−13−NocelleraUSA−8−Arbosana TUN−10−ChetouiSPA−7−Manzanilla USA−15−ArbequinaPER−18−Picual PER−17−BarneaAUS−5−PicualSPA−1 ITA−3TUN−12−ChemlaliITA−2Figure 5.4: Three-dimensional PCA scores plot of various extra virgin olive oils. ( )ITA-Italy, (q)SPA-Spain, ( )USA- USA, (F)PER-Peru, (N)TUN-Tunisia, ()GRE-Greece, (H) POR-Portugal and ()AUS-Australia.For instance, extra virgin olive oils (cultivar of Nocellara, Coratina, and Cerasuola and oneunknown cultivar) grown in Italy group together. Extra virgin olive oils from Spain (cultivarof Martena, Picual, Manzanilla), Tunisia (cultivar of Chetoui and Chemlali) or USA (cultivarof Arbequina and Arbosana) also form groups based on country of origin. However, there arealso some cases in which extra virgin olive oils from the same country do not group together.Two samples of extra virgin olive oil from Peru (cultivar of Picual and Barnea) are separatedand one unknown cultivar sample from Spain and two unknown cultivar samples from Italy arealso separated from their respective groups. This is reasonable, because extra virgin olive oilsharvested from the same country can still vary due to the effect of geographical region, climate,cultivar and other factors. While the number of extra virgin olive oil samples considered hereis very limited, they nevertheless illustrate those variations among the samples from differentcountries.After PCA analysis of pure vegetable oils and extra virgin olive oils, we further perform PCAanalysis on genuine and adulterated extra virgin olive oils. In the two-dimensional PCA plotsshown in Figure 5.5A and Figure 5.5B, the detection of canola or corn oil in different varieties ofextra virgin olive oils should be possible when there is 20% or more of the adulterant, becausethese points are separated by PC1 and PC2. However, for 10% there is little overlap between thegenuine and adulterated results. Samples consisting of 5% corn and canola oil were also studied685.3. Resultsbut are not shown here, as they are indistinguishable from the genuine samples. However, inFigure 5.5C and Figure 5.5D, the detection of adulterated extra virgin olive oils with grape seedand walnut oil is possible at level around 5%. This can be explained by the fact that, comparedwith canola or corn oil, grape seed and walnut oil show more intense features at 1263 and 1658cm−1 (Figure 5.1) and lie further apart from the extra virgin olive oils in the PC1 direction(Figure 5.2A). Regardless of which adulterant oils is present, clusters of 50% adulterated extravirgin olive oil always group together tighter than those of extra virgin olive oil alone, becausethe Raman features from the high adulterated vegetable oils reduces the effective scatter owingto natural variance in the extra virgin olive oils in the PCA plots.695.3.Results−0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8−0.1−0.0500.050.10.150.20.25222 22222222333 33333333 4 44 44444445555555 555555555555CanolaPC1PC2(A)−0.5 0 0.5 1−0.15−0.1−0.0500.050.122222222223 33333 3333444444 44445555555 555555555555CornPC1PC2(B)−3 −2.5 −2 −1.5 −1 −0.51.351.41.451.51.551.61.651.71.751.811111111111 1122222 222222 22 2233333 33 3333 3344444 44 4444 44 44555555 55555555555PC1PC2Grapeseed(C)−3.5 −3 −2.5 −2 −1.5 −1−1.9−1.85−1.8−1.75−1.7−1.65−1.6−1.55−1.5−1.45111111111111222222 22 22222333333 3333334 44444 44 44444555555 55555555555PC1PC2Walnut(D)Figure 5.5: Two-dimensional PCA scores plot of (5 I) authentic extra virgin olive oils and samples adulterated with (1 q)-5%, (2 )-10%, (3)-20%, (4F)-50% (A) canola, (B) corn, (C) grape seed, and (D) walnut oil. 5% canola and corn adulterated samples are not plotted, owing totheir serious overlapping with authentic samples.705.3. Results5.3.3 PCA-LDA analysisLDA is a supervised linear classification model, very commonly used to classify on the basis ofspectra. We process normalized Raman spectra by PCA to reduce their dimension from 1340spectral elements to 5 principal components, which account for more than 99% of the variancein the original dataset. Then we perform LDA with these 5 dimensional variables as X data andthe corresponding class responses as Y-data. The class of pure extra virgin olive oil is assignedto 1 in the Y-data, while adulterated extra virgin olive oil is 2 in the Y-data.For each dataset, we randomly select one third of the samples as a test set, while using theremaining samples as a training set to build up a LDA model. In particular, from the indi-vidual vegetable oil adulterated dataset (the first 17 rows of Table 5.4), we randomly select 6genuine and 5 adulterated extra virgin olive oil samples for the testing dataset and use the other13 genuine and 9 or 10 adulterated extra virgin olive oil samples for the training dataset. Forthe dataset referring to mixtures of four different vegetable oil adulterants (the last three rowsof Table 5.4), 6 genuine and 16 adulterated extra virgin olive oil samples (4 canola/corn/grapeseed/walnut adulterated samples) are randomly selected in the testing dataset, while the re-maining 13 genuine and 32 corresponding adulterated extra virgin olive oil samples (8 canola/c-orn/grape seed/walnut adulterated samples) are used in the training dataset. This sample par-titioning is repeated 50 times for each dataset, so that the final classification performances areevaluated using all 50 models. In this circumstance, we can more comprehensively evaluateclassification performances between genuine and adulterated samples without being affectedby the varieties of genuine extra virgin olive oil samples. The classification performance criteriaare listed as below,CCR =2∑n=1Number of correctly classified samples in class(n)Total number of samples(5.1)Sensitivity =Number of true positive samplesTotal number of samples(5.2)Specif icity =Number of true negative samplesTotal number of samples(5.3)where CCR is the correctly classified rate, true positive samples are true extra virgin olive oils715.3. Resultsand true negative samples are true adulterated extra virgin olive oils.725.3.ResultsTable 5.4: Linear classification between authentic and adulterated extra virgin olive oils by PLS-LDA method.Group vs. pure EVOO Training TestingCCR(%) Sensitivity (%) Specificity (%) CCR(%) Sensitivity (%) Specificity (%)10% canola vs. EVOO 88.95 87.54 91.25 76.00 72.33 81.5020% canola vs. EVOO 100.00 100.00 100.00 96.40 95.00 98.0050% canola vs. EVOO 100.00 100.00 100.00 100.00 100.00 100.0010% corn vs. EVOO 80.73 82.62 78.00 71.09 70.33 72.0020% corn vs. EVOO 94.82 91.85 99.11 90.73 87.33 94.8050% corn vs. EVOO 100.00 100.00 100.00 99.82 99.67 100.005% grape seed vs. EVOO 100.00 100.00 100.00 98.60 98.33 98.8010% grape seed vs. EVOO 100.00 100.00 100.00 99.82 99.67 100.0020% grape seed vs. EVOO 100.00 100.00 100.00 100.00 100.00 100.0050% grape seed vs. EVOO 100.00 100.00 100.00 100.00 100.00 100.005% walnut vs. EVOO 98.10 99.38 96.00 86.80 92.33 78.5010% walnut vs. EVOO 99.82 99.69 100.00 96.40 96.33 96.5020% walnut vs. EVOO 100.00 100.00 100.00 100.00 100.00 100.0050% walnut vs. EVOO 98.10 99.38 96.00 86.80 92.33 78.5010% any vs. EVOO 88.12 93.54 86.17 81.82 80.67 82.2520% any vs. EVOO 97.38 99.23 96.69 93.91 93.67 94.0050% any vs. EVOO 100.00 100.00 100.00 99.36 99.67 100.00735.3. ResultsBased on the summary results from the testing dataset in Table 5.4, Raman spectroscopy canachieve only around 70% correct classification rate for the detection of 10% of corn or canolain extra virgin olive oil and above 90% correct classification rate for the detection of 20% ofcorn and canola in extra virgin olive oil. For 20% corn and canola adulterated extra virgin oliveoil, around 95% accuracy of specificity is obtained, which means that 95% of the fake samplespointed out by Raman spectroscopy are truly adulterated by corn or canola oil.In contrast, we have more confidence to detect the adulteration of extra virgin olive oil withgrape seed or walnut oil by Raman spectroscopy, because Raman spectra of pure walnut andgrape seed oil differ much more from those of extra virgin olive oils as compared to corn andcanola oil (Figure 5.1 and Figure 5.2). Therefore, it is much easier for Raman spectroscopyto identify extra virgin olive oil adulterated with grape seed or walnut oil. We can achievemore than 95% accuracy of specificity identifying 10% grape seed or walnut oil in extra virginolive oil. The analysis performs slightly worse in detecting 5% walnut oil adulterated extravirgin olive oil than 5% grape seed oil adulterated samples. However, as a general rule, themore distributed the vegetable oils are in PCA scores (Figure 5.2), the more successful Ramanspectroscopy should be at detecting adulterated extra virgin olive oil.In any situation of practical interest, the type of adulterant or adulterants will not be knownbeforehand. Therefore, we also study systems that are more complex than extra virgin oliveoil adulterated with a single vegetable oil. For any given extra virgin olive oil sample, we donot know whether or not it is adulterated or which type of oil it is adulterated with. We canuse Raman spectroscopy as a tool for rapid screening prior to a standard analysis with HPLCor GC. According to Table 5.4, at a 50% adulteration level, the model can identify genuinesamples from adulterated ones with a 99.36% correct classification rate, 99.67% sensitivity and100% specificity. Thus, if this model identifies the tested sample to be adulterated, we are100% sure that the tested sample is not genuine extra virgin olive oil. If the tested samplepasses this discrimination, it can further be verified using the model based on genuine and 20%adulterated extra virgin olive oils, which shows a 93.91% correct classification rate, 93.67%sensitivity, 94.00% specificity. That is to say, if the tested sample is identified to be adulteratedthen the accuracy is 94.00%.In our case, including all regions and all cultivars, the correct classification rate of genuineand adulterated extra virgin olive oils is only possible when the adulterant (either corn or canola745.3. Resultsoil) is at least 20%. For lower adulterant concentrations, i.e. 5% and 10%, classification isnot possible. This limit is much higher than previous claims (i.e. 1% or 5%)[256, 261, 263,264, 271, 272], despite the fact that some of that work considered the adulteration of differentgrades of olive oils. For instance, in references different grades of olive oil samples are used(instead of only adulterating extra virgin olive oil)[261, 264, 272]. Zhang et al. have used sixbrands of extra virgin olive oils and virgin olive oils samples that were purchased in the USAsupermarkets, four of which originated from Italy and two of which came from Spain[272]. Theconclusion that Raman spectroscopy can distinguish genuine olive oil from 5% adulterated oliveoil rests upon a PCA testing model that includes three Italian olive oils, one Spanish olive oil andadulterated samples of these oils, and a PCA training model that includes one Italian olive oil,one Spanish olive oil and adulterated samples of these oils. Obviously, these six olive oil samplesfail to represent the complete variance of genuine olive oils that are available worldwide, oreven in Spain or Italy. Baeten and Meurens have studied the adulteration problem using oneCoratina olive oil and its adulterated samples with 1-10% trilinolein as a calibration model,then further tested the adulteration problem using six virgin olive oils of the Hojiblanca andPicual varieties[264]. However, this successful classification between six genuine virgin oliveoils and adulterated samples with 1%, 5%, and 10% of soybean, corn and olive-pomace oil isover-optimistic, because only 3 varieties of virgin olive oils are used and the authors did notclarify the degree of discrepancy among these samples[264]. In another paper, twenty-threecommercial brands of olive oil samples including different grades, such as extra virgin oliveoil, virgin olive oil, pure olive oil and olive oil, were purchased at supermarkets in the USA,China and Spain[261]. Those samples are either from Spain or Italy and the adulterated oliveoil samples are prepared based on only six olive oil samples instead of randomly choosing anyone from that twenty-three olive oil library. Moreover, according to the report from the UCDavis Olive Oil Center, whether these commercial brands of olive oil sample are authentic arestill questionable[2].For cases where only extra virgin olive oils are considered, Yang and Irudayaraj are ableto predict the concentration of adulterant in the range of 1 to 5% using NIR, FTIR, and Ramanspectroscopy combined with PLS[256]. However, the details of extra virgin olive oil (the numberof samples, varieties, country of origin, etc.) are not provided at all. If they choose only oneolive oil and blend it with the other adulterant vegetable oil, the estimated prediction error will755.4. Conclusioncertainly be very small, but this method cannot accurately predict adulteration in other oliveoils. Lo´pez-Dı´ez et al. have studied five cultivars of Italian extra virgin olive oil from severalgeographic regions and also noted the clear differentiation between the group of extra virginolive oils from the Sardinian region and others from the Peninsula region[271]. However, theirfurther study of admixtures of hazelnut oil and extra virgin olive oil by PLS uses only one extravirgin olive oil and one refined hazelnut oil[271]. In contrast, our result that the detection ofvegetable oil in extra virgin olive oil is not possible for adulterant concentrations as low as 5%accounts for a realistic variety of genuine extra virgin olive oil samples, so our conclusions aremore conservative. Based on previous analysis, the reason for the discrepancy is because weconsider the enormous varieties of genuine extra virgin olive oil samples normally encounteredin the marketplace.It is also worthwhile to mention that results from studies using FTIR spectroscopy are similarto those obtained by Raman spectroscopy[253–257]. Most of the work concluding that FTIRspectroscopy can distinguish genuine olive oil from 5% or less adulterated olive oil either doesnot consider the varieties of olive oils or only include a limited number of similar samples[253–256]. For example, de la Mata et al. chose a large variety of olive oil samples (virgin oliveoils, blends of refined and virgin olives, pomace-olive oils and extra virgin olive oils of variouscultivars) for a study of olive oils in edible oil blends[257]. They find that FTIR-ATR combinedwith PLS-DA (partial least squares-discriminant analysis) only differentiates oil blends withgreater than 50% of olive oil from those with less than 50% of olive oil[257]. The large variabilityin olive oil samples also leads to model failure in quantifying olive oil weight percentage, whenthis component accounts for more than 70% of sample [257].5.4 ConclusionWe have found that the large variety of extra virgin olive oils in the marketplace affects theclassification between genuine extra virgin olive oil and canola, corn, grape seed or walnut oiladulterated ones as determined by Raman spectroscopy. When we consider a large variety ofextra virgin olive oil samples, Raman spectroscopy combined with PCA-LDA cannot reliablydistinguish between genuine and 5% or 10% adulterated extra virgin olive oil. However, classi-fication between genuine and 20% or 50% adulterated extra virgin olive oil can be realized with765.4. Conclusionhigh accuracy.Thus, it is too optimistic to conclude that Raman spectroscopy can be used as a fast tool toreplace HPLC or GC for global classification in the sphere of an enormous variety of genuineand adulterated extra virgin olive oils. However, the simplicity of Raman spectroscopy makesit appealing as a fast online testing tool to screen adulterated extra virgin oil, it might wellbe used to reduce the number of tested sample before more complicated and time-consumingchromatographic tests. Samples that fail using Raman spectroscopy combined with PCA-LDAcan be identified as adulterated extra virgin olive oil and need no further testing. Samples thatpass ought to be evaluated with standard chromatographic methods.A clear library of various extra virgin olive oils is needed. We might improve the correct clas-sification rate between genuine and adulterated extra virgin olive oil, if we pay more attentionto appropriate selection of genuine extra virgin oil samples in the training dataset. Secondly,nonlinear classification models or further preprocessing methods should be explored to improvethe accuracy of classification between genuine and adulterated extra virgin olive oil samples. Fi-nally, peanut, safflower, sunflower (high Oleic acid) oil, olive-pomace or refined olive oil shouldalso be included as adulteration sources to further improve the scope of Raman spectroscopyfor discriminating various genuine and adulterated samples of extra virgin olive oil as they aremore similar to genuine extra virgin olive oils than other vegetable oils. In all, developing afast Raman sensor to completely replace chromatographic methods and accurately identify theadulterated extra virgin olive oil in the market still has a long way to go.77Chapter 6Infrared analysis of alkaline treatedbleached eucalyptus kraft pulpsAbout 40-45 percent by weight of a typical wood-chip feedstock is composed of celluloses[273].Cellulose consists of linear-chain glucose joined by β-linkages(1-4) glycosidic bonds (Figure6.1)[273, 274]. Native cellulose has a degree of polymerization of as much as 10,000, fallingtypically to 1,000 in bleached kraft pulps[274]. The carbohydrate polymers in wood typicallyinclude a heterpolysaccharide fraction as large as 20 percent. Collectively referred to as hemi-celluloses, these substances form amorphous polymer phases with degrees of polymerizationfrom 50 to 300[273, 275].Hemicelluloses appear in various forms. Two of chief importance in softwood are gluco-mannan (Figure 6.2) and arabinoglucuronoxylan (Figure 6.3)[273, 276]. Glucuronoxylan is themajor hemicellulose in hardwood (Figure 6.4)[273, 276]. Unlike mannan, xylan resembles cel-lulose but consists of β-linkages(1-4) between xylose in the backbone. In hardwood xylan, about10 percent of xylose units are substituted at the C-2 position by a 4-O-methyl-α-D-glucuronicacid (MeGlcA) side group, while some C-2 and C-3 positions are acetylated[276]. Softwood xy-lan is similar to hardwood xylan, but it has MeGlcA and arabinose residues instead of MeGlcAand acetyl groups[276]. Kraft pulping liberates wood fibres as a result of substantial delignifi-cation, which also causes some complex polysaccharides degrade. For instance, MeGlcA largelyconverts to unsaturated hexenuronic acid (hexA), which is largely removed by bleaching[277].Acetyl substituent is removed at the early stage of kraft cooking[277]. As a consequence, fullybleached kraft pulps contain only cellulose, hemicellulose and very little residual lignin.Xylan is known to improve inter-fiber bonding and leads to a pulp or paper with highertensile strength[273]. Researchers have studied the effect of xylan on a variety of pulp proper-ties. Generally speaking, the chemical removal of xylan from kraft pulp results in lower tensile78Chapter 6. Infrared analysis of alkaline treated bleached eucalyptus kraft pulps. . .Figure 6.1: General structural formulae for cellulose[274].Figure 6.2: General structural formulae for galactoglucomannans in softwood[274].Figure 6.3: General structural formulae for arabinoglucuronoxylan in softwood[274].Figure 6.4: General structural formulae for glucuronoxylan in hardwood[274].strength, while the addition of xylan significantly increases the strength[273, 278, 279]. There-fore, it is important to gauge the influence of xylan on fibre properties after pulping and bleach-ing. Improved knowledge of the role of xylan will increase understanding of the fibre structure,leading ultimately to fibre engineering and the manufacture of products with special propertieson demand.Fourier Transform Infrared Spectroscopy (FTIR) provides rich molecular information by796.1. Experimentalwhich to distinguish cellulose from mannan and xylan. Therefore, FTIR is often used to char-acterize regular pulp samples and chemically modified pulp samples, particularly followingalkaline treatment applied to extract hemicellulose. High concentration of alkaline solutioncan remove the major hemicelluloses and transform cellulose crystal structure from I to II, alsoknown as mercerization ([277, 280].Considerable research has been conducted to characterize alkaline treated pulps with staticFTIR in order to understand fibre structure or the mechanism of cellulose mercerization[280–287]. Among these reports, analysis has largely relied on the original spectra without prepro-cessing, which serves adequately for well-resolved cellulose signal. However, deeper informa-tion exists in the complicated fingerprint regions of IR spectra, and these are harder to discernwithout multivariate approaches.In the present study, we have created a calibration set of pulps with varying xylan con-tent by treating a commercial bleached eucalyptus kraft pulp with different concentrations ofNaOH. Alkaline base degrades the hemicelluloses to form polysaccharides and individual xyloseand mannose. Applying second derivative pretreatment to Attenuated Total Internal Reflection(ATR) - FTIR spectra and referring to results from wet chemistry methods, we have establishedthat preprocessing can uncover information on how alkaline treatment affects the xylan contentand cellulose crystal structure in the pulps. Combining these samples of varying xylan contentwith other commercial bleached hardwood kraft pulps, we have proven the utility of ATR-FTIRas a fast sensor to quantify xylan in commercial bleached kraft pulps.6.1 Experimental6.1.1 SamplesCommercial bleached hardwood kraft pulps derived from different wood feedstocks (eucalyp-tus, aspen, maple) were obtained from industrial source. We selected one bleached eucalyptuspulp for treatment with NaOH solutions of concentration from 0.5 to 6.0 weight percent inincrement of 0.5 percent, to serve as laboratory samples with systematically varying xylan com-position. Table 6.1 details this sample set. Extracted birch wood xylan and Whatman filterpaper purchased from Sigma served as xylan and cellulose sources for comparison.806.1. ExperimentalTable 6.1: The summary of the studied pulp samples.Sample NaOH % Xylan(HPLC) Titration Information1 0.0 16.61 0.0 Eucalyptus A2 0.5 15.79 1.49 Eucalyptus A3 1.0 14.62 3.3 Eucalyptus A4 1.5 12.94 5.21 Eucalyptus A5 2.0 11.47 7.31 Eucalyptus A6 2.5 9.76 9.13 Eucalyptus A7 3.0 8.51 10.83 Eucalyptus A8 3.5 7.14 12.59 Eucalyptus A9 4.0 6.12 13.45 Eucalyptus A10 4.5 N.A. 14.69 Eucalyptus A11 5.0 N.A. 13.29 Eucalyptus A12 5.5 5.84 14.29 Eucalyptus A13 6.0 6.5 13.36 Eucalyptus A14 0.0 19.61 N.A. Aspen15 0.0 19.61 N.A. Aspen16 0.0 16.94 N.A. Eucalyptus B17 0.0 21.99 N.A. Maple18 0.0 15.73 N.A. Eucalyptus C6.1.2 Methods and InstrumentationsHemicelluloses extraction by sodium hydroxideAir-dried bleached eucalyptus kraft pulp samples weighing 1.5 g were immersed in 100 mLNaOH and allowed to react for 1 hour at room temperature (25◦C). At end of the reaction, thepulp suspension was filtered through a plastic mesh filter. 10 mL filtrate was collected for ananalysis of alkaline soluble carbohydrates by titration. The treated pulps were neutralized with1 N HCl and washed with deionized water. Pulp samples were dewatered and stored at 4◦Cfor use to prepare laboratory handsheets, as well as spectroscopic analysis and HPLC standardcarbohydrate analysis.Chemical analysis of alkaline soluble hemicelluloses by titration method (TAPPI methodT235)[291]10 mL samples of the filtrates obtained as above were mixed with 10 mL 0.5 N potassium dichro-mate and 30 mL concentrated H2SO4. Mixtures were allowed to react for 15 min in order tocomplete the acid hydrolysis and oxidation of the carbohydrates in the filtrate. 50 mL water wasthen added with 4 drops of ferroin indicator after the solution was cooled to room temperature.816.2. Results and discussionThese solutions were titrated with 0.1 N ferrous ammonium sulfate solution to a purple endpoint. A blank titration was carried out using the same volume of the NaOH solution used forthe hemicellulose extraction in order to correct for background consumption of the titrant.HPLC standard method for pulp carbohydrate analysis[48]All pulp samples were ground to a size of 40 mesh. 150 mg samples of ground pulp werehydrolyzed individually by treatment with 3 mL 72% H2SO4 in a 30◦C water bath for one hour.The mixture was diluted with 84 g of deionized water and autoclaved at 121◦C for anotherhour. The subsequent liquid samples were filtered for HPLC carbohydrate analysis. HPLC sugarconcentrations detected electrochemically were calculated back to the relative weight percent ofxylan in the pulp.FTIR-ATR AnalysisWe recorded ATR-FTIR spectra of all pulps as handsheets using PerkinElmer FrontierTM spec-trometer, equipped with a Zinc Selenide (ZnSe) ATR sampling accessory. Spectra were collectedover the range from 4000 to 600 cm−1 with a resolution of 4 cm−1. Each spectrum consisted of1701 data points and acquired as a result of 20 co-added scans. We collected three spectra fromthree different locations on each handsheets.Data processingEach spectrum was preprocessed by taking its second derivative and smoothing by applyinga nine-point Savizky-Golay routine. We normalized these second derivative IR spectra to aconstant intensity of a prominent feature at 1162cm−1. We performed principal componentanalysis (PCA) and partial least squares (PLS) regression using Matlab 2009a.6.2 Results and discussion6.2.1 Peak assignment in ATR-FTIR spectraFTIR spectra of cellulose, birch xylan, fully bleached softwood and hardwood pulps are shownin Figure 6.5. Corresponding normalized second derivative spectra are shown in Figure 6.6.826.2. Results and discussionDominant features appear in the regions from 2800 to 3700 and 600 to 1800 cm−1. Note espe-cially, the range from 3000 to 3700 cm−1, where we see cellulose hydrogen bonds, and 2800-3000 cm−1, where C-H stretching appears[288]. IR spectra between 600 and 1800 cm−1 aremuch more complicated. Generally speaking, the interval between 1300 and 1400 cm−1 relates1000 1500 2000 2500 3000 3500 400000.10.20.30.40.5Wavenumber (cm−1)Absorbance  OH stretchingCH stretchingCH bendingC−O, C−H and ring stretching HWSWCelluloseXylanFigure 6.5: Raw ATR-FTIR spectra of cellulose, birch xylan, bleached eucalyptus kraft pulp (HW) andbleached softwood kraft pulp (SW).800 900 1000 1100 1200 1300 1400 1500 1600 1700−1.5−1−0.500.511.522.5Wavenumber (cm−1)Absorbance  896814 870964984998103210561110 1162123412461280131413361372 142814561482 1640160514649741204138210401416HWSWCelluloseXylanFigure 6.6: Normalized second derivative ATR-FTIR spectra of cellulose, birch xylan, bleached eucalyptuskraft pulp (HW) and bleached softwood kraft pulp (SW).to C-H bending, C-H scissoring and CH2 wagging in cellulose[282, 288, 290]. The region below836.2. Results and discussion1300 cm−1 is rich with C-O-C asymmetric stretching from glycosidic linkages, C-O stretching,C-O-C deformation[282, 292, 293].All the pulp samples in our study were fully bleached and contained nearly zero lignin.Therefore, the fingerprint region from 600 to 1500 cm−1 relates primarily to cellulose and hemi-celluloses. Characteristic bands of lignin, such as 1510, 1595, 1740 and 1770 cm−1 features,could not be found in bleached pulp samples[292].It has been reported that, owing to the MeGlcA, hexA and acetyl substituents in the xylanof the holocellulose samples, 1740 (C=O stretching of carboxylic acid), 1605 (C=O stretchingof carboxylate: salt or ester form) and 1246 cm−1 (C-O-R stretching) correlate strongly with thexylan concentration in holocellulose samples containing only cellulose and xylan[276, 289, 294].As shown in Figure 6.5, only the extracted birch xylan shows strong bands at 1605 and 1246cm−1. The absence of the feature at 1740 cm−1 occurs, because carboxyl groups are transformedto the salt or ester form after kraft pulping and bleaching. However, all of 1246, 1740 and 1605cm−1 features disappear in bleached pulp samples, as kraft pulping and bleaching lead to asubstantial removal of MeGlcA, hexA and acetyl branches[277, 289]. The remaining carboxylcontent arises from the degraded polysaccharides or little residual of lignin[295]. This impliesthat these two features might not persist in accurately reflecting the xylan concentration inbleached kraft pulp.It is well known that the fingerprint region of IR spectra below 1000 cm−1 is very compli-cated and suffers from overlapping band. Compared with Figure 6.5, our second derivativepretreatment in Figure 6.6 obviously helps to relieve the problem of overlapping bands. As seenin Figure6.6, the most distinguishable features among these four samples occur in prominentpeaks at 814, 870, 896, 964 cm−1 and in the region between 1230 and 1280 cm−1. Bands at814 and 870 cm−1 are characteristic of glucomannan in softwood hemicellulose[123, 296], sothese two features do not appear in the IR spectra of cellulose, birch xylan and hardwood pulpsamples.The band at 896 cm−1 is suggestive of the polysaccharides β-glycosidic linkages, and allthese samples have this signal in common[297]. The band at 964 cm−1 represents anotherprominent feature that differentiates cellulose from xylan in bleached softwood and hardwoodpulps. This feature shows high absorbance intensity in hardwood pulp and low intensity insoftwood pulp, compared with nearly zero in pure cellulose, suggesting that this feature relates846.2. Results and discussionclosest to xylan. Although birch xylan does not show a peak exactly at this position, a verybroad band at 974 cm−1 could relate to 964 cm−1 in standard pulps, because the region from968 to 979 cm−1 has been assigned to -CH3 and C-OH scissoring in synthetic xylan structuralunits[299]. The fact that xylan is very sensitive to the surrounding environment[277] and ad-sorbed water[298], could well shift the band to 964 cm−1 in pulp.However, this band is not often discussed in pulp spectroscopy, owing to its location in thecomplex fingerprint region. It is difficult to resolve without second derivative preprocessing.However, Bjarnestad and Dahlman have pointed out that this wavenumber strongly influencesPLS models used to quantify chemical composition in high-yield kraft pulp by photoacousticFTIR[34].Further, bands at 1336 (O-H in-plane bending) and 1428 cm−1(C-H scissoring) are bandscharacteristic of cellulose[282, 287, 290]. Thus, they have been commonly used to correlatewith cellulose crystallinity[288, 290]. In total, based on literature assignment and comparingsecond derivative IR spectra among cellulose, hardwood xylan, bleached softwood and eucalyp-tus pulps, we can assign major IR bands as summarized in Table 6.2.856.2. Results and discussionTable 6.2: Assignment of observed ATR-FTIR bands to bleached hardwood kraft pulp (HW), bleachedsoftwood kraft pulp (SW), cellulose and xylan.Wavenumber Band assignment HW SW Cellulose Xylan Ref.814 Mannan - 814 - - [123, 296]870 Mannan - 870 - - [123, 296]896 β-glycosic linkage, 896 896 894 896 [123, 297]C-O-C asymmetric stretch964-979 C-OH, CH3 bend 964 964 - 968-979 [299]982-984 C-O stretch at C6 984 984 982 - [280]in cellulose998-1000 C-O stretch in cellulose 1000 1000 998 - [280, 287, 292]1032 C-O deformation at C6 1032 1032 1032 - [282, 293]in cellulose1056 C-O stretch at C6 1056 1056 1056 - [280, 282, 287]in cellulose1110 Antisymmetric ring stretch 1110 1110 1110 1118 [280, 282, 293]1162-1164 C-O-C antisymmetric 1162 1162 1162 1164 [282, 292, 296]bridge ring stretch1204 -OH in plane bend in cellulose 1204 1204 1204 1212 [282, 292]1232-1246 C-O stretch 1232 1232 1234 1244 [276, 280, 292, 294]1278-1280 C-H bend in cellulose 1280 1280 1278 - [282, 292]1314 CH2 wag 1314 1314 1314 1316 [282, 288, 290]1336 O-H in-plane bend 1336 1336 1336 - [282, 288, 290]1372 C-H bend in cellulose 1372 1372 1372 - [282, 292]1428 CH2 scissor 1428 1428 1428 1416 [287, 290, 299]1456 O-H in plane bend 1456 1456 1456 - [282]in cellulose1464 C-H asymmetric deformation - - - 1464 [293]in xylan and lignins1605 C=O stretch - - - 1605 [34, 123, 292]in carboxylate group1640 O-H of adsorbed water in pulp 1640 1640 1640 1640 [282, 292]6.2.2 Wet chemistry analysis of NaOH treated eucalyptus pulpsWe treated bleached eucalyptus pulp with NaOH solutions of different concentrations from 0 to6 percent. Afterwards, we used oxidation and titration to determine alkaline soluble carbohy-drates (mainly xylan) in the final NaOH filtrate. We performed further pulp carbohydrate analy-sis using standard HPLC method to quantify the relative xylan content in the treated pulps. Ta-ble 6.1 lists the results of these measurements. The amount of dissolved carbohydrates (mainlyxylan) in the filtrate and the relative xylan content left in pulp are plotted against the NaOHconcentration in Figure 6.7A and Figure 6.7B respectively.866.2. Results and discussion0 1 2 3 4 5 602468101214NaOH %Dissolved carbohydrates %  0 1 2 3 4 5 64681012141618NaOH %Xylan %  Figure 6.7: NaOH concentration plotted against (A) the dissolved carbohydrates (mainly xylan) in extrac-tion filtrate measured by titration method and (B) the remaining xylan in pulp quantified by HPLC.An increase in NaOH concentration from 0 to 4 percent increases the dissolved carbohy-drates (mainly xylan) in the filtrate from 0 to 13.45 percent, while the relative xylan content inthese treated pulps decreases from 16.61 to 6.12 percent. Moreover, both Figure 6.7A and Fig-ure 6.7B show that the NaOH concentration (0-4 percent) correlates linearly with the amount ofdissolved carbohydrates (mainly xylan) in the filtrate or remaining xylan content in pulp. Thiscorrelation suggests that the presence of NaOH in this concentration range favors the release ofxylan from bleached eucalyptus kraft pulps. Sun et. al. have reported similar results, wherepartially delignified polar wood (hardwood) samples were extracted with 1.5, 3.0, 5.0, 7.5, 8.5%NaOH, and also showed a decrease of xylose content in treated samples with an increase ofNaOH concentration[300].With an increased NaOH concentration from 4 to 6 percent, the filtrate retains dissolvedcarbohydrates at a level around 14 percent. A constant xylan fraction of 6 percent remains inthe alkaline treated pulps. Unlike the observations at lower concentrations, increasing NaOHconcentration beyond 4 percent to as much as 6 percent does not increase the amount of ex-tracted xylan, as shown clearly in Figure 6.7. Instead, the amount of dissolved xylan in thefiltrate and relative xylan content left in the pulp tends to remain roughly same. We concludethat the residual 6 percent xylan is not easily accessible to hydrolysis by NaOH. Based on HPLCanalysis, NaOH solutions in the range from 4 to 6 percent extract approximately 65 percent of876.2. Results and discussionthe xylan in bleached eucalyptus pulp fibre, leaving the balance in the fibres. This result impliesthat there exist two types of xylan in bleached eucalyptus fibre, one that is very easily accessibleto alkaline hydrolysis, and another that is not. This inaccessible xylan probably binds tightlywith cellulose fibrils[277].6.2.3 ATR-FTIR analysis of bleached eucalyptus kraft pulps treated by NaOHsolution of less than 4 percentTo obtain a first measure of the IR spectrochemical variance of these treated pulps, we per-formed PCA on the entire FTIR spectral data set derived from the thirteen NaOH treated sam-ples. Figure 6.8 shows a PCA scores plot based on PC1 versus PC2 for all the samples. The−1 −0.5 0 0.5 1 1.5−0.8−0.6−0.4−0.200.20.40.6PC1PC2Figure 6.8: PCA analysis of normalized second derivative ATR-FTIR spectra from all alkaline treatedsamples. ( ) control bleached eucalyptus kraft pulp; ( ) samples treated by less than 4 percent NaOH; ( )samples treated by higher than 4 percent NaOH.distribution of the samples shows that PC1 correlates closely with NaOH concentration. Thetreated fibres show two stages with a transition point occurring at 4 percent NaOH. Treatmentwith NaOH solutions of concentration lower than 4 percent causes the PC2 score to graduallyincrease, while above 4 percent the PC2 score decreases with increasing NaOH concentration.As discussed below, these results accord with data obtained by wet chemistry methods.Figure 6.9 shows FTIR spectra in the region from 870 to 1120 cm−1 and 1200 to 1500 cm−1 forbleached eucalyptus kraft pulps treated with solutions containing from 0 to 4 percent NaOH.886.2. Results and discussionIn Figure 6.9A, the major changes occur at 896, 964, 1010, 1056 and 1084 cm−1. Features at900 950 1000 1050 1100−1−0.500.51Wavenumber (cm−1)Absorbance  89694896498410321056108411101010974(A)1200 1250 1300 1350 1400 1450 1500−0.2−0.100.10.20.30.4  123412461262128013141336 1372 14281454148212041214 134813581464Wavenumber (cm−1)Absorbance(B)Figure 6.9: Normalized second derivative ATR-FTIR spectra of NaOH treated bleached eucalyptus pulpin spectra region of (A) 870-1200 cm−1 and (B)1200-1500 cm−1. Eucalyptus pulp treated with (-) 0%, (-)0.5%, (-) 1.0%, (-) 1.5%, (-) 2.0%, (-) 2.5%, (-), 3.0% and (-) 3.5% NaOH.1234, 1246, 1314, 1372 and 1428 cm−1 also show obvious variations in Figure 6.9B. It has beenestablished that the ratio of IR absorbance intensity of features at 1428 and 896 cm−1 can beused to estimate the crystallinity index of cellulose, which defines the percentage of crystallinecellulose in the total cellulose including both crystalline and amorphous regions[288, 301].As the concentration NaOH treatment increases from 0 to 4 percent, the absorbance of the896.2. Results and discussionfeature at 896 cm−1 gradually decreases, while that at 1428 cm−1 subtly increases. This sug-gests that the crystallinity index of cellulose increases in NaOH treated pulps. Wan et. al. haveobserved similar results following treatment of unbleached eucalyptus kraft pulps with an 8percent of NaOH solution using extraction time from 0 to 9 hours[301]. Their analysis of cellu-lose crystallinity index based on this FTIR method has been confirmed by X-ray diffraction[301].Another important feature in Figure 6.9A is the band 964 cm−1. The decrease in intensity of thisband correlates well with an increase in the concentration of NaOH concentration and a decreaseof xylan content remaining in pulp.Bands at 1314 and 1372 cm−1 have been demonstrated to be very sensitive to cellulosecrystallinity[276], so the growth of these features in Figure 6.9B further supports the increase ofcellulose crystallinity in these NaOH treated samples. Therefore, conventional chemical analy-sis and FTIR spectroscopy taken together confirm that treatment by NaOH solutions of concen-tration lower than 4 percent, not only extracts accessible xylan but also increases the cellulosecrystallinity of the pulp.6.2.4 ATR-FTIR analysis of bleached eucalyptus kraft pulps treated by NaOHsolution of greater than 4 percentThe FTIR spectra in Figure 6.10 focus on bleached eucalyptus kraft pulp samples treated byNaOH solution in the 4 to 6 percent concentration range, highlighting the regions from 870 to1120 cm−1 and 1200 to 1500 cm−1. Figure 6.10A shows clearly that features at 896, 1000, 1010,1110, and 1280 cm−1 undergo major wavenumber shifts. With increasing NaOH concentrationfrom 4 to 6 percent, the band at 896 cm−1 starts to increase in intensity and simultaneouslyshifts to lower wavenumber. Recall that this 896 cm−1 feature simply decreases in intensity asNaOH concentration increases in the lower range from 0 to 4 percent.In similar fashion, increasing NaOH concentration in the 4 to 6 percent range increases theabsorbance intensity of the feature at 1000 cm−1 and shifts its peak position to lower wavenum-ber 996 cm−1. This again sharply contrasts with the fact that nearly no changes occur at thisband position in pulp treated by NaOH solution with concentration less than 4 percent. Theband at 1110 cm−1, assigned to in plane ring stretching[280, 282], gradually shifts to higherwavenumber, while the band at 1162 cm−1, assigned to the C-O-C asymmetric stretching vibra-tion of cellulose[282, 292, 296], shifts to lower wavenumber (not shown).906.2. Results and discussion900 950 1000 1050 1100−1−0.500.51Wavenumber (cm−1)Absorbance  8969489629841032105610841110101097410189961006(A)10001200 1250 1300 1350 1400 1450 1500−0.2−0.100.10.20.30.4Wavenumber (cm−1)Absorbance  123412461262128013141336 1372 14281454148212041214134813581464(B)Figure 6.10: Normalized second derivative ATR-FTIR spectra of NaOH treated bleached eucalyptus pulpin spectra region of (A) 870-1200 cm−1 and (B)1200-1500 cm−1. Eucalyptus pulp treated with (−) 0.0%,(−) 1.0%, (−) 2.0%, (−) 3.0%, ( ) 4.0%, ( ) 4.5%, ( ) 5.0%, ( ) 5.5% and ( ) 6.0% NaOH.Clearly the spectroscopic variations with higher NaOH concentrations differ from effectsobserved at lower concentrations of NaOH solutions. These comparisons of FTIR spectra ofsamples treated using low-alkaline 0 to 4 percent NaOH solutions with those obtained for sam-ples treated with higher alkalinity 4 to 6 percent solutions establish that the associated chemicaltransformations are completely different.This is plainly reflected in the fact that treated samples distribute in two PCA classes. More-916.2. Results and discussionover, major band shifts comparable to those discussed above have also been reported followingtreatment of wood fibres by 0, 2.0, 9.1, 10.7, 15.3 percent NaOH solutions[280]. The authorshave observed that treatment with concentration of NaOH higher than 10.7 percent causes bandposition shifts at 896, 1162 and 1425 cm−1, indicating that the cellulose structure started totransform from crystalline form I to II[280].In Figure 6.10B, we can see that the position of the band 1280 cm−1, assigned to C-H bend-ing in cellulose I, also shifts to a lower wavenumber, 1278 cm−1. This indicates the formationof crystalline cellulose II[282]. Together with these major band shifts, we also observe somechanges in absorbance intensity. For instance, in Figure 6.10A the band at 1056 cm−1, assignedto C-O stretching at the C6 position in cellulose, gradually decreases with NaOH treatment,suggesting a decrease in the amount of cellulose I[280].In Figure 6.10B, the fact that the absorbance intensity of cellulose -CH2 bending at band1428 cm−1 starts to decrease also indicates a decrease of cellulose I[282]. Referring to the crys-tallinity index of cellulose, we see an increase of the band 896 cm−1 in Figure 6.10A and a de-crease of the band at 1428 cm−1 in Figure 6B, suggesting a decrease in the cellulose crystallinity.Thus, with the transformation from cellulose I to II, the cellulose crystallinity decreases. Yuehas reported similar results using X-ray diffraction to study cotton treated by 15 and 20 per-cent NaOH solutions[287]. Their X-ray diffraction confirms that the change of cotton cellulosecrystalline structure from I to II decreases cellulose crystallinity index[287].Lastly, although the relative xylan content does not change with treatment by 4 to 6 percentNaOH solutions in this work, we still observed a subtle increase at 962 cm−1. As discussedabove, the xylan polymer structure is very sensitive to its environment[277], so the change incellulose crystalline structure is very likely to explain this variation in the xylan spectrum at962 cm−1 owing to the change its environment.Our titration and HPLC analysis in Figure 6.7A and Figure 6.7B taken together with ourFTIR results suggest that treatment of pulp with solution having NaOH concentration above 4percent favors chemical modification of cellulose crystalline structure as opposed to dissolvingmore xylan. Using X-ray diffraction, Liu and Hu have studied the effects of NaOH concentra-tion ranging from 0 to 24 percent on the cellulose crystallinity index of bamboo fibres with a97 percent cellulose content[302]. They conclude that treatment with base simply caused reac-tions in the amorphous regions and cellulose fibril surfaces for NaOH concentrations lower than926.2. Results and discussionthat required to promote the transformation of cellulose crystalline structure I to II[302]. How-ever, at higher base concentrations, hydroxide ions penetrate the cellulose internal lattice andchemically disturb cellulose crystalline structure[302]. This implies that the accessible xylan inbleached eucalyptus kraft pulps, removed by NaOH concentration less than 4 percent, existson the surface of cellulose microfibril or appear in a free form, while the inaccessible xylan isstrongly bound on the inner surface of cellulose fibril, where it cannot be reached with NaOHtreatment.6.2.5 Univariate and multivariate analysis of FTIR-ATR spectra and xylan contentin bleached kraft pulpAs discussed in the previous section, NaOH solutions with concentration from 0 to 4 percentcan effectively modify the xylan content in bleached eucalyptus kraft pulp, while NaOH solu-tions in 4.5 to 6 percent concentration range remove the accessible xylan and transform cellulosecrystalline structure from I to II, leading to large peak shifts in the IR spectra. Therefore, well-controlled efforts to spectroscopically quantify the relative xylan content in bleached hardwoodkraft pulp must focus on the samples treated by NaOH concentrations less than 4 percent. Ap-plying this constraint, we combine this subset of NaOH treated bleached eucalyptus kraft pulpswith other commercial bleached hardwood kraft pulps (Table 6.1) for univariate and multivari-ate analysis. For univariate analysis, we plotted absorbance intensity at several band positionsagainst relative xylan composition. We found the best correlation using the band at 964 cm−1.We thus select five samples from this dataset for calibration and use the remaining samplesfor validation (Table 6.3). Then, applying univariate analysis of 964 cm−1 and PLS multivariateanalysis, we use the validation dataset to predict the xylan content from FTIR spectra.Corresponding calculation results are summarized in Table 6.3. As shown in Figure 6.11, aplot of xylan concentration versus the band intensity at 964 cm−1 yields an excellent regressionfor the validation data. This good univariate correlation with xylan content applies not onlyto alkaline treated eucalyptus samples, but also extends as well to other commercial bleachedhardwood kraft pulps (maple, aspen, eucalyptus), produced in different mills. It is clear to seethat these two methods predict xylan content very close to those measured by HPLC.936.2. Results and discussionTable 6.3: The summary of calculation results based on PLS and univariate analysis with the feature at964 cm−1.Calibration Information HPLC Band 964 cm−1 PLS1 Eucalyptus A-0% NaOH 16.61 16.32 16.582 Eucalyptus A-1.5% NaOH 12.94 12.56 12.933 Eucalyptus A-3.0% NaOH 8.51 8.77 8.554 Eucalyptus A-4.0 % NaOH 6.12 6.24 6.105 Maple 21.99 22.28 22.01R2 0.998 1.000RMSEE 0.281 0.029Validation Information HPLC Band 964 cm−1 PLS1 Eucalyptus A-0.5% NaOH 15.79 15.59 16.302 Eucalyptus A-1.0% NaOH 14.62 14.14 14.153 Eucalyptus A-2.0% NaOH 11.47 11.44 11.594 Eucalyptus A-2.5% NaOH 9.76 9.57 10.095 Eucalyptus A-3.5% NaOH 7.14 7.32 7.486 Aspen A 19.61 20.03 20.137 Aspen A 19.61 19.48 19.868 Eucalyptus B 16.94 15.66 15.839 Eucalyptus C 15.73 15.48 15.28R2 0.988 0.983RMSEP 0.499 0.5280 0.05 0.1 0.15 0.20510152025Absorbance Intensity at 964 cm−1Xylan content in pulp %   y = 69.05*x + 7.098R2=0.988MapleAspenEucalyptusNaOH treated EucalyptusFigure 6.11: Linear relationship between the absorbance intensity at 964 cm−1 and relative xylan contentin pulp.946.3. ConclusionThe regression between HPLC results and the univariate intensities at 964 cm−1 yields astrong correlation (R=0.988) for the validation dataset. A similar result (R=0.983) is obtainedfrom the regression between HPLC results and PLS predicted xylan contents. Although neitherunivariate ATR-FTIR spectroscopic analysis nor PLS multivariate analysis can be as accurateas HPLC, it can provide a rapid estimate of xylan content in bleached pulp. The root meansquare error of prediction (RMSEP) are very similar, 0.499 percent and 0.528 percent respec-tively. Thus, FTIR offers great promise as a means for rapid classification of bleached hardwoodkraft pulp, such as eucalyptus, aspen, maple for the amount of xylan.6.2.6 Alkaline treatment of bleached eucalyptus pulps as mercerizationAlkaline treatment (mercerization) is widely used to modify natural wood fibre. In a separatespectrochemical study, Gwon et al. treated natural wood fibres with 0, 2.0, 9.1, 10.7 and 15.3percent solutions of NaOH, and observed major FTIR band shifts owing to the conversion ofthe cellulose I to II at 10.7 percent[280]. In the present study, we note such change initiated bytreatment with NaOH solution as dilute as 4 percent. We can attribute the difference betweenthese results to the difference in the samples. Gwon’s samples were natural wood fibres contain-ing 28.1% lignin, 3.4% extractives, 27.5% hemicelluloses and 41.0% celluloses[280], while oursamples consist of bleached eucalyptus kraft pulp, which had already gone through kraft pulp-ing and bleaching. In our case, polysaccharide polymers in the bleached eucalyptus kraft pulpshad already experienced partial degradation and were more vulnerable to NaOH in aqueoussolution. Further tests can be carried out to study the fibre structure of this bleached eucalyptuskraft pulp.6.3 ConclusionThis study confirms that ATR-FTIR spectrochemical analysis provides rich molecular structureinformation about cellulose and hemicellulose in bleached kraft pulps and alkaline treatedpulps. Second derivative preprocessing enhances spectral resolution in the IR fingerprint re-gion. The changes in band absorbance intensities and shifts in band positions elucidate thetransformation of cellulose crystallinity index and the modification of molecular structure in-duced by alkaline treatment. Moreover, second derivative ATR-FTIR spectra combined with956.3. Conclusionunivariate analysis of the band at 964 cm−1 or PLS multivariate analysis serve well to predictxylan content in bleached hardwood kraft pulp with about 0.5 percent accuracy. This establishesthe feasibility of ATR-FTIR as a fast tool for the quantification of xylan content in bleached kraftpulps. The strong correlation of the band at 964 cm−1 in the second derivative ATR-FTIR spec-trum accounts well for xylan.96Chapter 7Quantification of hemicelluloses inbleached kraft pulps by infraredspectroscopy7.1 IntroductionThe paper industry relies upon bleached kraft pulp as the main raw material for the manufac-ture of printing paper, tissue and packaging owing to its high strength and superior reinforce-ment properties. Softwood fibres average 3-4 mm in length and 40 µm in width, compared withthe smaller 1 mm long and 20 µm wide dimension of hardwood fibres[303]. Thus, softwoodpulp has higher tendency to join together and form a web, and shows inherently stronger abilitythan bleached hardwood kraft pulp[303]. On the other hand, hardwood pulp yields productswith more uniform and smoother surfaces[303, 304]. As a result, paper products dictate variousblends of softwood and hardwood fibres.Regardless of the fibre type, cellulose and hemicellulose polymer dominate the compositionof materials aligned along the fibre axis. The proportion of hemicellulose significantly affectspulp chemical and physical properties[278, 301]. In practice, advanced manufacturing requiresa means to rapidly estimate the amount of hemicelluloses in pulp fibre in process. The standardmethod used to accurately determine the amount of hemicelluloses hydrolyzes pulp with con-centrated sulfuric acid to form individual sugars[48]. The subsequent solution is autoclaved andassayed by HPLC carbohydrate analysis[48]. HPLC using an electrochemical detector provides avery sensitive method for sugar analysis, but this method is too expensive and time-consumingfor use in real time process control. Acid hydrolysis of solid pulp, followed by high temperatureautoclaving, requires the extensive handling of 72 percent sulfuric acid and as much as four977.1. Introductionhours’ sample preparation time before HPLC analysis. Tappi Test Method T235 provides an al-ternative method to roughly estimate the amount of hemicelluloses[291]. Here, a 10% solutionof NaOH extracts the hemicelluloses. Afterwards, the filtrate is hydrolyzed with concentratedsulfuric acid and titrated with ferrous ammonium sulfate to estimate the total hemicellulose.However, this method gives no information on the exact amount of xylan and mannan in thesample.In contrast with such traditional methods, spectroscopic instrumentation promises an im-portant new means for analysis and real time process control in the paper industry. Many stud-ies have shown that near infrared (NIR) and Raman spectroscopy can determine lignin contentand kappa number in pulp samples[305, 306]. Wallbacks et al. have demonstrated the feasibil-ity of NIR as a means of quantifying glucose and xylose in birch kraft pulp[307]. However, NIRspectroscopy yields only broad absorption bands with few chemical features, and this limitsits application as a general method for measuring hemicellulose in pulp samples represent-ing a broad range of species. In this aspect, Fourier transform infrared spectroscopy (FTIR)is more appealing, as it offers substantial information relevant to molecular structure. It alsopermits a variety of different sampling techniques such as transmission, diffuse reflectance in-frared Fourier transform spectroscopy (DRIFT), attenuated total reflectance (ATR) and photo-acoustic spectroscopy (PAS). Extensive work has explored the application of FTIR to the analysisof pulp[23, 34, 35, 123, 181, 293, 307–315]. DRIFT and PAS have emerged as preferred tech-niques, owing to their ability to sample bulk volumes. DRIFT requires the grinding of pulpsamples into powders and blending with KBr to form pellets for measurement. PAS does not in-volve complicated sample preparation. Bjarnestad and Dahlman have used PAS associated withPLS to quantify the chemical compositions of bleached softwood and hardwood pulps, obtainedfrom different manufacturing processes[123].ATR offers another nondestructive sampling method particularly suited to opaque pulp sam-ples with greater ease of operation than PAS and DRIFT. Hemicelluloses are primarily located onthe cell wall of the pulp fibres, so the detection of xylan and mannan does not need a deep pen-etration depth of measurement. For pulp samples, ATR can penetrate about 2 µ in depth[117],which benefits in the collection of infrared signal from xylan and mannan.It has yet to be applied as a fast sensor for quantitatively measuring the hemicellulose com-position in bleached kraft pulp. Most IR studies have been limited to a degree by focusing on the987.2. Experimentalraw or first derivative spectra. In the present study, we apply discrete wavelet transform (DWT)and standard normal variate (SNV) methodologies to preprocess and explore the effect of pre-treatment methods on regression models. By including bleached softwood kraft pulp, bleachedhardwood kraft pulp and their blends as well as chemically modified bleached kraft pulp, wehave endeavored to demonstrate the wide feasibility of ATR-FTR as a means of classifying andquantifying the hemicellulose composition of in-process pulp stream.7.2 Experimental7.2.1 SamplesWe obtained 19 bleached kraft pulps including bleached hardwood kraft pulp, bleached soft-wood kraft pulp and blends of bleached softwood and hardwood kraft pulp from different in-dustrial mills. Softwood pulps, constituting mostly northern bleached softwood kraft (NBSK)pulp, were collected from Scandinavia as well as British Columbia, Alberta and eastern Canada.Hardwood pulps were derived from different wood feedstocks including three different sourcesof eucalyptus, aspen and maple. An additional five pulps, consisting of blends of softwoodand hardwood pulps, were produced by kraft cooking and bleaching mixtures of softwood andhardwood chips together. We further increased the number and variety of samples by blendingbleached softwood and hardwood kraft pulp in our laboratory. For this, we randomly selectedone NBSK and blended it with bleached maple kraft pulp from 0 to 100% in increments of 5%by weight. Then, we blended this NBSK with aspen or eucalyptus pulp in a similar fashion. Inorder to test the wide applicability of ATR-FTIR spectroscopy, we also included chemically mod-ified softwood and hardwood pulps, which were treated by NaOH solutions of concentration in0 to 6 percent range. Table 7.1 summaries all of the samples. For spectroscopic measurement,we prepared standard laboratory handsheets with weight of 20g/m2 from each of these wet pulpsamples.997.2. ExperimentalTable 7.1: The summary of the studied samples.Sample Type Num. Origin Information1 Softwood 12 Kraft mill NBSK from Canada2 Hardwood 5 Kraft mill Aspen(1), eucalyptus(3),maple(1)3 Mixture 5 Kraft mill Unknown4 Mixture A 22 Lab NBSK mixed with aspen5 Mixture B 22 Lab NBSK mixed with eucalyptus6 Mixture C 22 Lab NBSK mixed with maple7 Chemical pulp 8 Lab NaOH(0-6%) treated eucalyptus8 Chemical pulp 33 Lab NaOH(0-6%) treated NBSK7.2.2 Methods and instrumentationsHPLC standard method for pulp carbohydrate analysis[48]All pulp samples were ground to 40 mesh. 15 mg ground pulp was hydrolyzed with 3 mL 72%sulfuric acid in a 30◦C water bath for one hour. The mixture was further diluted with 84 gdeionized water and then autoclaved at 121◦C for another hour. The subsequent liquid samplewas filtered for HPLC carbohydrate analysis. The sugars detected by HPLC were used determinethe relative weight percent of xylan and mannan in the pulp.ATR-FTIR AnalysisWe recorded FTIR spectra of all pulp samples as handsheets using PerkinElmer FrontierTM spec-trometer. The ATR sampling accessory was equipped with a single reflection Zinc Selenide(ZnSe) crystal. Spectra were collected over the range from 4000 to 600 cm−1 with a resolutionof 4 cm−1. Each spectrum consisted of 20 co-added scans over 1701 data points. We collectedthree spectra from three different locations on each handsheet.7.2.3 Data processingWe applied discrete wavelet transform (DWT) to IR spectra with the Symlet8 filter using Wave-lab850 toolbox under Matlab 2009a[154, 155]. Each IR spectrum was decomposed into a seriesof wavelets described by the fourth approximation wavelet coefficient (CA4) plus the four detailwavelet coefficients (CD4, CD3, CD2, CD1). CA4 refers to the low frequency baseline, while CD2and CD1 contain high frequency noise. We calculated the final DWT spectrum by subtractingthese noise and baseline signal contributions, described by CD2, CD1 and CA4, from the raw IR1007.2. Experimentalspectrum. These spectra, pretreated by DWT, were normalized by applying the SNV method asrepresented in Equation 7.1[218],xij,SNV = (xij − x¯i)/√∑nj=1(xij − x¯i)2n− 1(7.1)where xij represents the absorbance intensity at the j-th wavenumber position in the i-th spec-trum, x¯i is the mean value of the i-th spectrum and n is the total number of points in the IRspectrum. Three normalized DWT spectra were averaged to one spectrum to represent eachsample.Multivariate analysisWe recorded IR spectra from 4000 to 600 cm−1, but subjected only the region from 1900 to600 cm−1 to multivariate analysis. Knowing that xylan and mannan features are predominantlylocated in this fingerprint region, confining calculation over this interval avoids uninformativefeatures. Multivariate analysis, including principal component analysis (PCA) and partial leastsquares (PLS) were performed using Matlab 2009a. IR spectra, arranged as row vectors, consti-tuted the data matrix of predictors X. The data matrix Y contained the response consisting ofthe corresponding relative xylan or mannan content, obtained from HPLC analysis. Includingsamples with the maximum and minimum concentrations of xylan or mannan, we randomlyselected two thirds of samples for calibration. The rest served as validation data to objectivelyevaluate the calibration model. We repeated this data partition 100 times to produce 100 dif-ferent calibration and validation datasets. Final results account for the average performance ofthese 100 datasets. The root mean square error and correlation coefficient between measuredand PLS predicted values served as principal criteria. In particular, we used leave-one-out crossvalidation combined with the root mean square error of estimation (RMSEE) to assess the per-formance of the calibration model, and the root mean square error of prediction (RMSEP) to testthe adequacy of validation.1017.3. Results7.3 Results7.3.1 Characteristic bands in ATR-FTIR spectraFigure 7.1 shows normalized IR spectra of bleached softwood kraft pulps, bleached hardwoodkraft pulps and their blends obtained from different industry mills in the region from 600 to1800 cm−1. The samples in our study consist of fully bleached kraft pulp. They contain nearlyWavenumber (cm−1)Absorbance814896986−10581106116212061335 142816401605137212801246600 800 1000 1200 1400 1600 1800−10123456Figure 7.1: Normalized ATR-FTIR spectra of (–) extracted birch xylan, (–) bleached softwood pulps, (–)bleached hardwood pulps, and (–) mixtures of bleached softwood and hardwood pulps obtained fromdifferent kraft mills.zero lignin content, so the characteristic bands of lignin (i.e. 1510, 1595, 1740 and 1770 cm−1)do not appear in Figure 7.1[292, 293]. In order to focus on the bands associated with hemi-celluloses, this study excludes the region from 3000 to 3700 cm−1, which pertains to cellulosehydrogen bonds, and the C-H stretching from 2800 to 3000 cm−1. The most distinguishing fea-ture among various pulp samples appears at 814 cm−1. This is characteristic band of mannan,which points to the samples containing softwood pulp[123]. The band at 896 cm−1 correlates tothe asymmetric stretching of the C-O-C β-glycoside linkage in cellulose and hemicellulose[292].Bands at 1335 cm−1 (OH in-plane bending), 1372 (C-H bending) and 1428 (CH2-scissoring) arecharacteristic of cellulose[288, 292]. Features at 1372 and 1428 cm−1 strongly relate to cellulosecrystallinity[287, 288].Although the IR fingerprint region provides substantial information relevant to molecular1027.3. Resultsstructure, these bands form very complicated pattern and suffer from significant band overlap.The application of DWT pretreatment improves to resolve many overlapping bands. Figure 7.2displays the normalized DWT spectra, corresponding to the raw spectra in Figure 7.1. Majorbands are labeled with their positions, and detailed assignments refer to Table 7.2.600 800 1000 1200 1400 1600 1800−6−4−20246810Wavenumber (cm−1)Absorbance 814852898964986100210281058110611621206 13351234 142812788641640124697216051372Figure 7.2: Normalized DWT FTIR spectra of (–) extracted birch xylan, (–) bleached softwood pulps, (–)bleached hardwood pulps and (–) mixtures of bleached softwood and hardwood pulps obtained fromdifferent kraft mills.Interestingly, we note that different pulp samples present only little difference between 900and 1000 cm−1 in Figure 7.1, but their differences become amplified after DWT pretreatment. Inaddition to the mannan-related signal, the band at 964 cm−1 in Figure 7.2 clearly differentiatessoftwood pulp, hardwood pulp and their mixtures. At this position, the hardwood pulp ex-hibits the largest relative absorption intensity followed by mixtures of hardwood and softwoodpulp and then softwood pulp, suggesting that the band at 964 cm−1 relates to the xylan con-tent. A broad and strong band around 972 cm−1 appears in extracted birch xylan, which seemsrelated to the pulp band at 964 cm−1. Kacˇura´kova´ et al. have studied the synthetic xylooligosac-charide model compounds and assigned the region from 968 to 979 cm−1 to -CH3 and C-OHscissoring in xylan[299]. They have also found that xylan is very sensitive to the surroundingenvironment[299]. Thus, it might be reasonable that the band at 972 cm−1 in the extracted birchxylan shifts to 964 cm−1 in pulp samples.1605 and 1246 cm−1 in birch xylan are another two distinguishable features from those in1037.3. ResultsTable 7.2: Assignment of major FTIR-ATR bands for bleached hardwood kraft pulp (HW), bleached soft-wood kraft pulp (SW) and their mixture.Bands assignment HW SW Mixture Ref.Mannan - 814 - [123, 296]Mannan - 864-876 864-876 [123, 296]β-glycosic linkage, C-O-C asymmetric stretch 898 898 898 [293]C-OH, CH3 bend 964 958 960 [299]C-O stretch at C6 in cellulose 986 986 986 [280]C-O stretch in cellulose 1002 1002 1002 [280, 287, 292]C-O deformation at C6 in cellulose 1028 1028 1028 [293]C-O stretch at C6 1058 1058 1058 [280, 287]Antisymmetric ring stretch 1106 1106 1106 [280, 282, 293]C-O-C antisymmetric bridge ring stretch 1162 1162 1162 [282, 292]-OH in plane bend in cellulose 1206 1206 1206 [282, 292]C-O stretch - 1234 1234 [292]CH bend in cellulose 1278 1278 1278 [282, 292]CH2 wag in cellulose 1314 1314 1314 [282, 288, 290]O-H in-plane bend in cellulose 1335 1335 1335 [282, 288, 290]C-H bend 1372 1372 1372 [282, 292]CH2-scissor in cellulose 1426 1426 1426 [287, 290, 299]O-H of adsorbed water in pulp 1640 1640 1640 [282, 292]pulp samples. The appearance of these two features are mainly attributed to carboxylate ions inthe side branch of xylan (4-O-methyl-D-glucuronic acid (MeGlcA) or unsaturated hexenuronicacid (hexA))[123, 290]. In contrast, the absence of these two bands in bleached kraft pulp sam-ples confirms the substantial cleavage of MeGlcA and hexA, owing to the kraft pulping andbleaching process[277, 289]. Thus, we cannot rely on these two features to accurately estimatexylan content.In a separate experiment, we treated bleached eucalyptus kraft pulp with NaOH solutionsof different concentrations from 0 to 3.5 percent in increments of 0.5 percent. The amountof extracted xylan increased with increasing NaOH concentration. We acquired IR spectra ofthese alkaline treated pulp samples in order to compare the spectroscopic consequence of thisextraction. Figure 7.3 shows that the prominent differences appear around 964 cm−1. As theremaining xylan content in the pulp decreases, the band at 964 cm−1 gradually decreases. Thisaccords with the discussion above that this region of IR spectra correlates well with the xylancontent in pulp samples, suggesting its feasibility for quantitatively measuring the xylan contentin pulp fibre, despite the fact that it is not assigned as xylan characteristic band.1047.3. Results900 950 1000 1050 1100 1150 1200 1250 1300−4−3−2−10123456Wavenumber (cm−1)Absorbance  89896498610281234120612781106Figure 7.3: Normalized DWT transformed ATR-FTIR spectra of bleached eucalyptus kraft pulp treatedwith (-) 0%, (-) 0.5%, (-) 1.0%, (-) 1.5%, (-) 2.0%, (-) 2.5% and (-) 3.0% NaOH.Infrared spectra depend sensitively on the details of molecular structure information. Incomplex mixture of large molecules, it is too complex to associate each feature with individualchemical component, especially for the bands that appear in the IR fingerprint region. Insteadof assigning small features to various components, we base our classification on multivariateanalysis and rapidly analyze the overall variance of the dataset.7.3.2 PCA analysisFigure 7.4 presents the PCA scores plot for all the studied samples. Here we plot principalcomponent 1 (PC1) against PC2. These two components account for 83.89% and 9.20% of thetotal variance. Circles represent raw bleached kraft pulps obtained from different industrialmills. A group of red circles refers to softwood pulps, mostly NBSK, collected across Canada. Afew black circles, appearing right beside the red circles, are blends of softwood and hardwoodpulps. On the other side of Figure 7.4 display points for hardwood kraft pulps as green circles.Judging from the fact that the black circles fall closer to red circles than green circles, we caneasily recognize that these mixture samples are dominated by softwood pulp. As mentionedabove, the bleached hardwood pulps are derived from three wood species: eucalyptus, mapleand aspen. The three completely different groups of green circles in this PCA scores plot confirmthis, indicating the ability of ATR-FTIR to classify the pulp samples on the basis of their species.1057.3. Results−8 −6 −4 −2 0 2 4 6 8 10−4−202468PC1PC2Alkaline modified SW  Alkaline modified EucalyptusSWEucalyptusMapleAspenSW and HW mixtureFigure 7.4: PCA scores plot of normalized DWT transformed ATR-FTIR spectra for commercial bleachedkraft pulps, laboratory generated pulp mixtures and alkaline treated pulps. (o), (o) and (o) obtained fromkraft mill, (M) NBSK blended with aspen pulp, (M) NBSK blended with maple pulp, (M) NBSK blendedwith eucalyptus pulp, (∗) bleached eucalyptus kraft pulp treated with 0-4% NaOH, (∗) NBSK treated with0-6% NaOH.In addition to the raw pulps, Figure 7.4 places three types of laboratory blended samples inthis scores plot as triangles. We blended NBSK with eucalyptus, maple and aspen respectively,with weight percentage ranging from 0 to 100% in increments of 5%. These mixtures tend tolinearly distribute along PC1 between softwood and hardwood pulp. With increased hardwoodpulp content, the blends differentiate in PC space according to the species of hardwood pulp.Lastly, the black and red asterisks denote alkaline treated NBSK and eucalyptus pulp samples.They tend to move in the negative direction on the PC1 axis, as the decreased content of xylanremains in the pulp samples owing to the increased NaOH treatment concentration. Noting thedistribution of hardwood pulp in the region of large positive PC1, we can conclude that PC1correlates with the xylan content of pulp.Figure 7.5 plots the PC1 and PC2 loadings, together with the normalized DWT IR spectraas a function of wavenumber. This allows us to compare PC loadings to IR spectral intensities,enabling a search for the most significant wavenumber positions in terms of the difference we seein the PCA scores plot. The PC1 loading, representing 83.89% of the total variability, correlates1067.3. Resultswith features that have a high positive value at 964 cm−1 and low negative values at 814, 870and 940 cm−1. This implies that samples with a higher absorbance value at band 964 cm−1 andlower absorbance values at bands 814 and 870 cm−1 tend to have larger PC1 score, causing themto distribute in the positive direction on the PC1 axis.Wavenumber (cm−1)PC loading  814 870 96494012301106994600 800 1000 1200 1400 1600 1800−0.5−0.4−0.3−0.2−0.100.10.2Figure 7.5: The loading plot of (–) PC1 and (–) PC2 versus normalized DWT transformed ATR-FTIRspectra.As discussed above, we assign the band at 964 cm−1 to C-O stretching, correlating stronglywith the xylan content remaining in the pulp, while 814 and 870 cm−1 are characteristic bandsof mannan. This analysis suggests that the different chemical compositions of the pulp samples,mostly referring to xylan and mannan, determine the distribution of samples along with PC1,and account for the 83.89% variability. This could also well explain the distribution of alkalinetreated NBSK pulps, that appear on the leftmost of PC1, and hardwood pulps, that displayon the rightmost of PC1 with the order maple (21.99% xylan), aspen (19.61% xylan) and feweucalyptus pulps (around 15-16% xylan) from right to left. Unlike PC1, PC2 has a high positivevalue at 994 cm−1 and negative values at 814, 870 and 1106 cm−1. The changes occurring around994 and 1106 cm−1 in the IR spectra arise mostly as a result of NaOH treatment, which alsodissolves mannan and leads to the decreased absorbance intensities at 814 and 870 cm−1 .1077.3. Results7.3.3 PLS analysisWe have developed PLS models to quantitatively predict xylan and mannan contents in bleachedkraft pulps. Table 7.3 summaries the statistical data calculated from one hundred PLS modelsand also describes the effect of different pretreatment methods on the quality of our PLS models.Table 7.3: The summary of PLS calculation results based on 100 models.Calibration ValidationRange Mean Method R AverageRMSEEσ RMSEE R AverageRMSEPσ RMSEPXylan 1.49-21.33 10.81 - 0.990 0.755 0.038 0.990 0.738 0.092SNV 0.996 0.507 0.022 0.995 0.503 0.060SG-D1-SNV 0.996 0.463 0.021 0.996 0.453 0.045DWT-SNV 0.996 0.457 0.022 0.996 0.442 0.050Mannan 0-7.35 4.56 - 0.978 0.458 0.033 0.977 0.451 0.064SNV 0.980 0.427 0.022 0.979 0.430 0.052SG-D1-SNV 0.994 0.247 0.017 0.993 0.256 0.022DWT-SNV 0.995 0.226 0.014 0.994 0.233 0.021We refer to RMSEE and RMSEP to evaluate the deviation between the measured concentra-tions and predicted values. The correlation coefficient (R) serves as another important gaugereflecting the degree of linear relationship. A good PLS model always has as small RMSEE andRMSEP as possible, along with an R value close to 1. In order to obtain high quality of infraredsignal, ATR-FTIR requires flat sample surface so as to contact well with the ATR crystal. In thepresent study, pulp samples were measured as handsheets. Further, we normalized each spec-trum using SNV to remove the effect of different fibre morphology. Table 7.3 shows that SNVsignificantly improves the quality of PLS model. In contrast with PLS analysis of the originalspectra, SNV pretreatment reduces one third of xylan estimation error by decreasing RMSEEfrom 0.755 to 0.507 and RMSEP from 0.738 to 0.503.Owing to the dominant cellulose signal in IR spectra, we reason that proper pretreatmentcombined with normalization method could further enhance the weak hemicellulose signal andachieve a better PLS model. Thus, before performing SNV, we try to treat IR spectra by takingthe first derivative followed by a 9 points Savitzky-Golay smoothing (SG-D1) or applying DWT.Prediction of mannan shows the effectiveness of these two pretreatment methods, reducing theestimation error in RMSEE and RMSEP by a factor of two compared with SNV. SG-D1-SNVreduces mannan RMSEE from 0.427 to 0.247 and RMSEP from 0.430 to 0.256. DWT-SNV suc-1087.3. Resultscessfully improves mannan RMSEE from 0.427 to 0.226 and RMSEP from 0.430 to 0.233. Moresignificant, R increases from 0.980 to 0.995 for the mannan calibration model and from 0.979 to0.994 for the corresponding validation model.From Table 7.3, we note that DWT is slightly better than the first derivative pretreatment.As we know, the first derivative method decreases spectral signal-to-noise ratio. It may thuscause problems for spectral analysis, especially when the original IR spectrum does not have agood signal-to-noise ratio. As a result, it is quite common to smooth the first derivative spectrawith a moving average window. The wider the window, with more data dots included, the betterthe smoothed spectra obtained. However, one drawback of this is the loss of several data pointsat the beginning and the end of spectrum due to the moving average. In addition, the broadwindow may over-smooth subtle signal in the IR spectra. For that reason, we chose only 9 datapoints for SG-D1. Figure 7.6 illustrates raw, DWT and SG-D1 IR spectra acquired from bleachedsoftwood and hardwood kraft pulps. From this figure, one obviously sees that both DWT andWavenumber (cm−1)AbsorbanceSW raw IRHW raw IRSW DWT IRHW DWT IRSW SG−D1 IRHW SG−D1 IR600 800 1000 1200 1400 1600 1800−2−1.5−1−0.500.511.522.5Figure 7.6: Normalized raw, DWT and SG-D1 ATR-FTIR spectra of bleached (–) softwood and (–) hard-wood kraft pulp.SG-D1 enhance spectral resolution and amplify small features. Reconstructed signals usingDWT are consistent with those appearing in the raw IR spectra. For instance, the vertical dottedlines at 814, 898, 986 and 1106 cm−1 in Figure 7.6 confirm that DWT transformed spectra showthe correct peak position and range. While SG-D1 shifts the band position. We thus concludethat DWT provides a much more straightforward and accurate means to qualitatively analyze1097.4. DiscussionIR bands than the first derivative spectra.Figure 7.7 graphically displays the measured hemicellulose contents versus the results pre-dicted by the DWT-SNV-PLS method. Here, a, b and R are slope, intercept and correlation coeffi-cient obtained from a PLS regression. Combined with the error estimation summarized in Table7.3, Figure 7.7 strongly supports our conclusion that FTIR-ATR associated with DWT-SNV-PLSis able to rapidly and accurately determine the xylan and mannan compositions in bleachedkraft pulps (softwood pulps, hardwood pulps and mixtures) and alkaline treated bleached kraftpulps.0 5 10 15 2005101520Measured  (%)Predicted (%)Xylan calibration  a=0.996b=0.056R=0.996(A)0 5 10 15 2005101520Measured (%)Predicted (%)Xylan validation  a=0.981b=0.172R=0.996(B)0 2 4 6 802468Measured (%)Predicted (%)Mannan calibrationa=0.999b=0.003R=0.995(C)0 2 4 6 802468Measured  (%)Predicted  (%)Mannan validationa=0.965b=0.198R=0.994(D)Figure 7.7: Linear relationship between predicted xylan or mannan content by DWT-SNV-PLS and mea-sured concentrations from HPLC carbohydrate analysis. (A) Xylan calibration, (B) xylan validation, (C)mannan calibration, (D) mannan validation.7.4 DiscussionSubstantial literature reports that the feature at 814 cm−1 in IR spectra corresponds to the char-acteristic band of mannan. We infer that the band at 964 cm−1 strongly relates to the xylan1107.4. Discussioncontent in pulp samples by qualitatively comparing various bleached kraft pulps and alkalinetreated pulps at the beginning of this study. For further confirmation of this, we apply a geneticalgorithm (GA) in concert with PLS regression to quantitatively search for the xylan-relatedsignal. Here, we chose Riccardo’s PLS-GA toolbox to perform feature selection[187, 188], asGA is a very useful optimization tool. Although there are a large number of feature selectionmethods integrated with PLS (ex. moving window PLS[316], interval PLS[190], uninformativevariable elimination PLS[317] etc.), Riccardo’s PLS-GA algorithm emphasizes the autocorrela-tion between the adjacent wavelengths[188]. It can locate the most correlated spectral regions,while the others often pick up dispersed variables[187]. In this way, we just rely on the numer-ical calculation without considering band assignment. Usually, five independent runs of GAmust be performed as the GA-selected variables are not 100% reproducible.Wavenumber (cm−1)Absorbance600 800 1000 1200 1400 1600 1800−12−10−8−6−4−20246810Figure 7.8: Selected wavelengths for xylan and mannan from 5 independent runs of GA-PLS calculation.(−) selected wavelengths for mannan and (−) selected wavelengths for xylan.Figure 7.8 shows the wavelengths selected for xylan and mannan by five GA-PLS runs ofDWT spectra. GA selects three dominant regions (940-980, 1020-1032 and 1222-1240 cm−1) forxylan, while the region from 800 to 840 cm−1 corresponds to mannan in all of five GA runs. TheGA-selected mannan wavelengths are consistent with the mannan characteristic band at 814cm−1, confirming the correct performance of GA. The most xylan relevant wavelengths arisefrom C-O stretching, in accord with the discussion above. Although literature largely ignoresthe C-O stretching regions around the bands at 964 cm−1 in connection with xylan, we note that1117.5. ConclusionBjarnestad and Dahlman have found this feature significant in their PLS regression model[123].Thus, the results from Riccardo’s GA-PLS support our conclusion that the C-O stretching regionaround 964 cm−1 significantly predict the xylan content in bleached kraft pulp samples.In order to explore the feasibility of ATR-FTIR spectroscopy as a means for rapidly deter-mining the hemicellulose content in bleached kraft pulp samples, we included a wide varietyof samples. The use of blended pulps is very common in paper industry, but no one has testedthe performance of FTIR spectroscopy to quantify the hemicellulose content in this type of sam-ple. NaOH treated pulp samples, representing chemically modified pulps, can test the detectionlimit of ATR-FTIR spectroscopy. The constructed PLS models above show that ATR-FTIR spec-troscopy successfully measures the content of xylan in bleached kraft pulps (softwood pulps,hardwood pulps and their blends), NBSK treated with 0-6 percent NaOH solutions and bleachedeucalyptus kraft pulp treated with 0-4 percent NaOH solutions. However, it fails for bleachedeucalyptus kraft pulp treated with 4.5-6 percent NaOH solutions, yielding a large deviation be-tween measured and predicted concentrations. Thus, our final PLS model for the prediction ofxylan excluded these samples. Here, 4.5-6 percent NaOH aqueous solutions modified the cellu-lose crystalline structure from I to II in the bleached eucalyptus pulp we used, while 6 percentNaOH solution was still not concentrated enough to initialize such transformation in NBSK. Thechanges of cellulose crystalline structure from I to II have been demonstrated by the shifts ofIR bands at 897, 983 and 1162 cm−1[280, 287], which degraded the PLS model and leaded to alarge deviation between the predicted and measured xylan concentration. We conclude, there-fore, ATR-FTIR fails to quantify the xylan content for the pulp samples with modified cellulosecrystalline structure. Otherwise, ATR-FTIR has a great potential to rapidly and accurately deter-mine the xylan and mannan contents for normally bleached kraft pulps (softwood, hardwoodand their mixtures) in industry.7.5 ConclusionWe used ATR-FTIR combined with DWT-SNV-PLS to determine the hemicellulose content inbleached kraft pulps and NaOH treated kraft pulps. Pretreatment by DWT followed by SNVcan effectively enhance spectral resolution and improve the PLS regression model based on IRspectra. Both the qualitative comparison of the IR spectra from various pulp samples and the1127.5. Conclusionquantitative calculation using GA-PLS demonstrate that the features around 964 cm−1 stronglycorrelate with the xylan content in pulp samples. PCA analysis of these IR spectra could differ-entiate various pulp samples and classify bleached hardwood pulps on the basis of species. Inthe end, ATR-FTIR associated with DWT-SNV-PLS accurately predicted the xylan and mannancompositions for the bleached kraft pulp samples. However, this technique encounters limita-tion when samples show a different cellulose crystalline structure from natural cellulose.113Chapter 8ConclusionThis thesis demonstrates success in several new applications of vibrational spectroscopy withmultivariate data analysis. The characteristic molecular vibrational information obtained offersunique fingerprint spectrum for individual sample. The analysis of these signatures has enabledthe rapid classification of complex materials. For example, we show that infrared spectroscopycan discriminate a wide set of bleached kraft pulps, including softwood pulps, hardwood pulps,their mixtures, as well as alkaline modified bleached kraft pulps. We also demonstrate its po-tential to classify bleached hardwood kraft pulps on the basis of their species (maple, aspen,eucalyptus).In related work, we show that Raman spectroscopy rapidly classifies different vegetable oils,and also differentiates a wide variety of genuine extra virgin olive oils from those adulteratedwith more than 20% corn, canola, grape seed or walnut oil. This detection limit appears to beless sensitive than other claims, suggesting that Raman spectroscopy can detect a 5% adulter-ant. Our work shows that this difference arises from the large variance among the extra virginolive oils available from a wide variety of countries and cultivars. Work claiming higher sen-sitivity has selected olive oils from a specific geographic region or cultivar. Extending Ramanspectroscopy to the classification of variance on a much larger scale represents a substantial ad-vance, while showing that the problem of uncorrelated biological variance must be addressedfurther in bioanalytical chemistry.The analysis of the absorbance intensities and the positions of peaks in spectra enable us tocharacterize complex materials quantitatively or qualitatively. Infrared spectroscopy providesrich information about the molecular structure of cellulose and hemicelluloses in bleached kraftpulps or alkaline treated bleached eucalyptus kraft pulps. It sensitively detects an increase ofcellulose crystallinity in alkaline treated bleached eucalyptus kraft pulps as base concentrationis increased, and signal a decrease in crystallinity when cellulose structure transforms from I toII. For the hemicellulose in bleached eucalyptus pulp, our measurements recognize xylan in two114Chapter 8. Conclusionforms: (1) xylan easily accessible to NaOH, which appears either in a free form or on the surfaceof cellulose microfibril; and (2) xylan inaccessible to base solution, which is strongly bound onthe inner surface of cellulose fibril.Our work, in Chapters 6 and 7, applies this methodology to rapidly and accurately quan-tify the hemicellulose content in a wide variety of bleached kraft pulps, including softwoodpulps, hardwood pulps, mixtures of softwood and hardwood pulps as well as alkaline modifiedbleached kraft pulps. In comparison with the standard chromatographic measurement, ATR-FTIR is much easier to operate and more cost-effective and efficient. However, we also note thatinfrared spectroscopy fails to accurately predict the xylan content in bleached eucalyptus kraftpulps, whose cellulose structure converts to cellulose II. The transformation of cellulose struc-ture from I to II significantly shifts bands, leading to a completely different pattern of spectrumfrom that of natural cellulosic material having cellulose structure I.Finally, in a test of the limits of Raman classification and quantification, we show that, withoptimal laser power and exposure time, Raman spectroscopy can detect no more than a 0.498mM of added glucose and a 0.197 mM of added lactate in IVF G-1 culture medium. As a result, itdoes not have sufficient sensitivity to differentiate the single human embryo spent culture mediafrom control. It is also impossible to use Raman spectroscopy to quantify the total uptake orproduction of the metabolites from a single human embryo. Even spending efforts in couplingRaman spectroscopy with Teflon-AF 2400 to enhance the Raman signal of aqueous solution, westill could not achieve the sensitivity limit required by IVF samples.115Bibliography[1] Widjaja, E., Zheng, W., Huang, Z., Classification of colonic tissues using near-infrared Ra-man spectroscopy and support vector machines. International journal of oncology 2008,32, 3, 653-662.[2] Frankel, E. N., Mailer, R. J., Wang, S. C., Shoemaker, S. F., Guinard, J., Flynn, J. X., Sturzen-berger, N. D., Evaluation of Extra Virgin Olive Oils Sold in California. 2011.[3] Duarte, I. F., Barros, A., Delgadillo, I., Almeida, C., Gil, A. M., Application of FTIR spec-troscopy for the quantification of sugars in mango juice as a function of ripening. Journalof agricultural and food chemistry 2002, 50, 3104-3111.[4] Cadet, F., Offmann, B., Direct spectroscopic sucrose determination of raw sugar canejuices. Journal of agricultural and food chemistry 1997, 45, 166-171.[5] Tewari, J., Irudayaraj, J., Quantification of Saccharides in Multiple Floral Honeys UsingFourier Transform Infrared Microattenuated Total Reflectance Spectroscopy. Journal ofagricultural and food chemistry 2004, 52, 3237-3243.[6] Bauer, R., Nieuwoudt, H., Bauer, F. F., Kossman, J., Koch, K. R., Esbensen Kim H., FTIRspectroscopy for grape and wine analysis. Analytical chemistry 2008, 80, 1371-1379.[7] Moreira, J. L., Santos, L., Spectroscopic interferences in Fourier transform infrared wineanalysis. Analytica Chimica Acta 2004, 513, 263-268.[8] Cordella, C., Moussa, I., Martel, A. C., Sbirrazzuoli, N., Lizzani-Cuvelier, L., Recent de-velopments in food characterization and adulteration detection: technique-oriented per-spectives. Journal of agricultural and food chemistry 2002, 50, 1751-1764.[9] Rodriguez-Saona, L. E., Allendorf, M. E., Use of FTIR for rapid authentication and de-tection of adulteration of food. Annual review of food science and technology 2011, 2,467-483.[10] Guan, N., Fan, Q., Ding, J., Zhao, Y., Lu, J., Ai, Y., Xu, G., Zhu, S., Yao, C., Jiang, L., Miao,J., Zhang, H., Zhao, D., Liu, X., Yao, Y., Melamine-contaminated powdered formula andurolithiasis in young children. New England Journal of Medicine 2009, 360, 11, 1067-1074.[11] Zhang, Q., Liu, C., Sun, Z., Hu, X., Shen, Q., Wu, J., Authentication of edible vegetableoils adulterated with used frying oil by Fourier Transform Infrared Spectroscopy. Foodchemistry 2012, 132, 3, 1607-1613.116Bibliography[12] Racowsky, C., Vernon, M., Mayer, J., Ball, G. D., Behr, B., Pomeroy, K. O., Wininger D.,Gibbons, W., Conaghan J., Stern, J. E., Standardization of grading embryo morphology.Fertility and Sterility 2010, 94, 1152-1153.[13] Rienzi, L., Ubaldi, F., Iacobelli, M., Romano, S., Minasi, M. G., Ferrero, S., Sapienza, F.,Baroni, E., Greco, E., Significance of morphological attributes of the early embryo. Repro-ductive biomedicine online 2005, 10, 5, 669-681.[14] Margalioth, E. J., Ben-Chetrit, A., Gal, M., Eldar-Geva, T., Investigation and treatment ofrepeated implantation failure following IVF-ET. Human Reproduction 2006, 21, 3036-3043.[15] Kuo, M. L., McClelland, J. F., Luo, S., Chien, P.-L., Walker, R. D., Hse, C. Y., Applicationsof infrared photoacoustic spectroscopy for wood samples. Wood and fiber science 1988,20, 132-145.[16] Smith, G. D., Clark, R. J. H., Raman microscopy in archaeological science. Journal of Ar-chaeological Science 2004, 31, 8, 1137-1160.[17] Cnopnras, A., Single crystal Raman spectra of forsterite, fayalite, and monticellite. Amer-ican Mineralogist 1991, 76, 110L1109.[18] Li-Chan, E., The applications of Raman spectroscopy in food science. Trends in Food Sci-ence & Technology 1996, 7, 361-370.[19] Lin, M., He, L., Awika, J., Yang, L., Ledoux, D. R., Li, H., Mustapha, A., Detection ofMelamine in Gluten, Chicken Feed, and Processed Foods Using Surface Enhanced RamanSpectroscopy and HPLC. Journal of Food Science 2008, 73, T129-T134.[20] Vankeirsbilck, T., Vercauteren, A., Baeyens, W., Van der Weken, G., Verpoort, F., Vergote,G., Remon, J. P., Applications of Raman spectroscopy in pharmaceutical analysis. Trendsin Analytical Chemistry 2002, 21, 869-877.[21] Johansson, J., Spare´n, A., Svensson, O., Folestad, S., Claybourn, M., Quantitative trans-mission Raman spectroscopy of pharmaceutical tablets and capsules. Applied spec-troscopy 2007, 61, 11, 1211-1218.[22] Wartewig, S., Neubert, R. H., Pharmaceutical applications of Mid-IR and Raman spec-troscopy. Advanced drug delivery reviews 2005, 57, 8, 1144-1170.[23] Leclerc, D. F., Fourier transform infrared spectroscopy in the pulp and paper industry.Encyclopedia of Analytical Chemistry 2000, 8361-8388.[24] Workman Jr, J. J., Infrared and Raman spectroscopy in paper and pulp analysis. AppliedSpectroscopy Reviews 2001, 36, 2-3, 139-168.[25] Sikirzhytskaya, A., Sikirzhytski, V., Lednev, I. K., Raman spectroscopic signature of vagi-nal fluid and its potential application in forensic body fluid identification. Forensic sci-ence international 2012, 216, 1, 44-48.117Bibliography[26] Qu, J. Y., Wilson, B. C., Suria, D., Concentration measurements of multiple analytes inhuman sera by near-infrared laser Raman spectroscopy. Applied Optics 1999, 38, 5491-5498.[27] Xie, C., Dinno, M. A., Li, Y. Q., Near-infrared Raman spectroscopy of single opticallytrapped biological cells. Optics letters 2002, 27, 4, 249-251.[28] Nijssen, A., Schut, T. C. B., Heule, F., Caspers, P. J., Hayes, D. P., Neumann, M. H., Pup-pels, G. J., Discriminating basal cell carcinoma from its surrounding tissue by Ramanspectroscopy. Journal of Investigative Dermatology 2002, 119, 1, 64-69.[29] Movasaghi, Z., Rehman, S., Rehman, I. U., Raman Spectroscopy of Biological Tissues.Applied Spectroscopy Reviews 2007, 42, 493-541.[30] Schaeberle, M. D., Kalasinsky, V. F., Luke, J. L., Lewis, E. N., Levin, I. W., Treado, P. J.,Raman chemical imaging: histopathology of inclusions in human breast tissue. Analyticalchemistry 1996, 68, 11,1829-1833.[31] Frank, C. J., McCreery, R. L., Redd, D. C., Raman spectroscopy of normal and diseasedhuman breast tissues. Analytical chemistry 1995, 67, 777-783.[32] Fendel, S., Schrader, B., Investigation of skin and skin lesions by NIR-FT-Raman spec-troscopy. Fresenius’ journal of analytical chemistry 1998, 360, 5, 609-613.[33] Huang, Z., McWilliams, A., Lui, H., McLean, D. I., Lam, S., Zeng, H., Near-infrared Ramanspectroscopy for optical diagnosis of lung cancer. International journal of cancer 2003,107, 1047-1052.[34] Bhardwaj, N. K., Dang, V. Q., Nguyen, K. L., Determination of Carboxyl Content inHigh-Yield Kraft Pulps Using Photoacoustic Rapid-Scan Fourier Transform Infrared Spec-troscopy. Analytical chemistry 2006, 78, 6818-6825.[35] Dang, V. Q., Bhardwaj, N. K., Hoang, V., Nguyen, K. L., Determination of lignin con-tent in high-yield kraft pulps using photoacoustic rapid scan Fourier transform infraredspectroscopy. Carbohydrate Polymers 2007, 68, 489-494.[36] Pan, J., Nguyen, K. L., Development of the photoacoustic rapid-scan FT-IR-based methodfor measurement of ink concentration on printed paper. Analytical chemistry 2007, 79, 6,2259-2265.[37] Kelly, J. F. D., Downey, G., Detection of Sugar Adulterants in Apple Juice Using FourierTransform Infrared Spectroscopy and Chemometrics. Journal of agricultural and foodchemistry 2005, 53, 3281-3286.[38] Li, S., Shan, Y., Zhu, X., Zhang, X., Ling, G., Detection of honey adulteration by highfructose corn syrup and maltose syrup using Raman spectroscopy. Journal of Food Com-position and Analysis 2012, 28, 69-74.[39] U.S. Department of Health and Human Services Centers for Disease Control and Preven-tion, 2006 ART report, 1-584.118Bibliography[40] Centers for Disease Control and Prevention. Assisted Repro-ductive Technology (ART) Report 2011. Retrieved from http://nccd.cdc.gov/DRHART /Apps/NationalSummaryReport.aspx September 2014.[41] Schieve, L. A., Peterson, H. B., Meikle, S. F., Jeng, G., Danel, I., Burnett, N. M., Wilcox, L.S., Live-birth rates and multiple-birth risk using in vitro fertilization. JAMA 1999, 282,19, 1832-1838.[42] Das, M.D., Holzer, H. E., Recurrent implantation failure: gamete and embryo factors.Fertility and Sterility 2012, 97, 1021-1027.[43] Qi, D., Berger, A. J., Quantitative concentration measurements of creatinine dissolvedin water and urine using Raman spectroscopy and a liquid core optical fibre. Journal ofbiomedical optics 2005, 10, 031115.[44] Qi, D., Berger, A. J., Correction method for absorption-dependent signal enhancement bya liquid-core optical fiber. Applied Optics 2006, 45, 489-494.[45] Qi, D., Berger, A. J., Chemical concentration measurement in blood serum and urinesamples using liquid-core optical fibre Raman spectroscopy. Applied optics 2007, 46, 10,1726-1734.[46] Owen, R. W., Giacosa, A., Hull, W. E., Haubner, R., Wu¨rtele, G., Spiegelhalder, B., Bartsch,H., Olive-oil consumption and health: the possible role of antioxidants. The lancet oncol-ogy 2000, 1, 107-112.[47] Frankel, E. N., Chemistry of Extra Virgin Olive Oil: Adulteration, Oxidative Stability, andAntioxidants. Journal of Agricultural and Food Chemistry 2010, 58, 5991-6006.[48] Sluiter, A., Hames, B., Ruiz, R., Scarlata, C., Sluiter, J., Templeton, D., Crocker, D., Deter-mination of structural carbohydrates and lignin in biomass. Laboratory Analytical Proce-dure 2008.[49] Hsu, C. P. Sherman, Chapter 15 Infrared Spectroscopy, Handbook of Instrumental Tech-niques for Analytical Chemistry. Prentice Hall 1997.[50] Reyes-Goddard, J. M., Barr, H., Stone, N., Photodiagnosis using Raman and surface en-hanced Raman scattering of bodily fluids. Photodiagnosis and Photodynamic Therapy2005, 2, 223-233.[51] Asher, S. A., UV resonance Raman spectroscopy for analytical, physical, and biophysicalchemistry. Analytical chemistry 1993, 65, 201A-210A.[52] Kudelski, A., Analytical applications of Raman spectroscopy. Talanta 2008, 76, 1-8.[53] Hollas, J. M., Modern Spectroscopy. John Wiley & Sons 2004.[54] Movasaghi, Z., Rehman, S., ur Rehman, D. I., Fourier transform infrared (FTIR) spec-troscopy of biological tissues. Applied Spectroscopy Reviews 2008, 43, 134-179.119Bibliography[55] Adar F., Delhaye M., DaSilva E., Evolution of instrumentation for detection of the Ra-man effect as driven by available technologies and by developing applications, Journal ofchemical education 2007, 84, 50-60.[56] Yang, D., Ying, Y., Applications of Raman spectroscopy in agricultural products and foodanalysis: A review. Applied Spectroscopy Reviews 2011, 46, 539-560.[57] Pelletier, M. J., Quantitative analysis using Raman spectrometry. Applied spectroscopy2003, 57, 20A-42A.[58] Wang, Y., McCreery, R. L., Evaluation of a diode laser/charge coupled device spectrometerfor near-infrared Raman spectroscopy. Analytical Chemistry 1989, 61, 2647-2651.[59] McCreery R. L., Modern Raman spectroscopy: Has the sleeping giant finally awoken?University of Alberta 2009.[60] Douglas A. Skoog, F. James Holler, Stanley R. Crouch, Principles of Instrumental AnalysisSixth Edition. Thomson Brooks/Cole 2007.[61] Ozaki, Y., Cho, R., Ikegaya, K., Muraishi, S., Kawauchi, K., Potential of near-infraredFourier transform Raman spectroscopy in food analysis. Applied spectroscopy 1992, 46,1503-1507.[62] Agarwal, U. P., Ralph, S. A., FT-Raman Spectroscopy of Wood: Identifying contributionsof Lignin and Carbohydrate Polymers in the Spectrum of Black Spruce (Picea mariana).Applied spectroscopy 2012, 51, 1648-1655.[63] Agarwal, U. P., Terashima, N., FT-Raman study of dehydrogenation polymer (DHP)lignins. In 12th ISWPC International Symposium on Wood and Pulping Chemistry, Madi-son, WI, 2013 June, 9-12.[64] Agarwal, U. P., Weinstock, I. A., Atalla, R. H., FT-Raman spectroscopy for direct measure-ment of lignin concentrations in kraft pulps 2003. Tappi Journal, 2, 22-26.[65] Agarwal, U. P., Landucci, L. L., FT-Raman investigation of bleaching of spruce thermome-chanical pulp. Journal of pulp and paper science 2004, 30(10), 269.[66] De Grauw, C. J., Sijtsema, N. M., Otto, C., Greve, J., Axial resolution of confocal Ramanmicroscopes: Gaussian beam theory and practice. Journal of Microscopy, 1997, 188, 273-279.[67] Confocal Raman microscopy general overview, Princeton Instruments 2007.[68] Chan, J. W., Taylor, D. S., Zwerdling, T., Lane, S. M., Ihara, K., Huser, T., Micro-Ramanspectroscopy detects individual neoplastic and normal hematopoietic cells. Biophysicaljournal 2006, 90, 648-656.[69] Everall, N., Depth profiling with confocal Raman microscopy, part I. Spectroscopy -Springfield then Eugene then Duluth- , 19, 22-33.120Bibliography[70] Altkorn, R.,Koev, I.,Pelletier, M. J., Raman performance characteristics of Teflon--AF 2400liquid-core optical-fiber sample cells. Applied spectroscopy 1999, 53, 1169-1176.[71] Weber, W. H., Zanini-Fisher, M., Pelletier, M. J., Using Raman microscopy to detect leaksin micromechanical silicon structures. Applied spectroscopy 1997, 51, 123-129.[72] Shetty, G., Kendall, C., Shepherd, N., Stone, N., Barr, H., Raman spectroscopy: elucidationof biochemical changes in carcinogenesis of oesophagus. British journal of cancer 2006,94, 1460-1464.[73] Kidder, L. H., Kalasinsky, V. F., Luke, J. L., Levin, I. W., Lewis, E. N., Visualization ofsilicone gel in human breast tissue using new infrared imaging spectroscopy. Naturemedicine 1997, 3, 235-237.[74] Notingher, I., Raman spectroscopy cell-based biosensors. Sensors 2007, 7, 1343-1358.[75] Xie, C.,Li, Y. Q., Confocal micro-Raman spectroscopy of single biological cells using op-tical trapping and shifted excitation difference techniques. Journal of Applied Physics2003, 93, 2982-2986.[76] Frost, R. L., Tran, T. H., Kristof, J., The structure of an intercalated ordered kaolinite,aRaman microscopy study. Clay Minerals 1997, 32, 587-596.[77] Martinez-Ramirez, S., Sanchez-Cortes, S., Garcia-Ramos, J. V., Domingo, C., Fortes, C.,Blanco-Varela, M. T., Micro-Raman spectroscopy applied to depth profiles of carbonatesformed in lime mortar. Cement and Concrete Research 2003, 33, 2063-2068.[78] Miura, M., Arai, S., Mizukami, T., Raman spectroscopy of hydrous inclusions in olivineand orthopyroxene in ophiolitic harzburgite: Implications for elementary processes inserpentinization. Journal of Mineralogical and Petrological Sciences 2011, 106, 91-96.[79] Liu, Z., Davis, C., Cai, W., He, L., Chen, X., Dai, H., Circulation and long-term fate offunctionalized, biocompatible single-walled carbon nanotubes in mice probed by Ramanspectroscopy. Proceedings of the National Academy of Sciences 2008, 105, 1410-1415.[80] Colomban, P., Dinh, H. M., Tournie´, A., Jauzein, V., Origin of the variability of the me-chanical properties of silk fibres: 3. Order and macromolecule orientation in Bombyxmori bave, hand-stretched strings and Nephila madagascarensis spider fibres. Journal ofRaman Spectroscopy 2012, 43, 1042-1048.[81] Eichhorn, S. J., Sirichaisit, J.,Young, R. J., Deformation mechanisms in cellulose fibres,paper and wood. Journal of Materials Science 2001, 36, 3129-3135.[82] Breitenbach, J., Schrof, W., Neumann, J., Confocal Raman-spectroscopy: analytical ap-proach to solid dispersions and mapping of drugs. Pharmaceutical research 1999, 16,1109-1113.[83] Mattha¨us, C., Kale, A., Chernenko, T., Torchilin, V., Diem, M., New ways of imaginguptake and intracellular fate of liposomal drug carrier systems inside individual cells,based on Raman microscopy. Molecular pharmaceutics 2008, 5, 287-293.121Bibliography[84] Robert, B., Resonance Raman spectroscopy. Photosynthesis research 2009, 101, 147-155.[85] Arzhantsev, S., Vilker, V., Kauffman, J., Deep-Ultraviolet (UV) Resonance Raman Spec-troscopy as a Tool for Quality Control of Formulated Therapeutic Proteins. Applied spec-troscopy 2012, 66, 1262-1268.[86] Huang, C. Y., Balakrishnan, G., Spiro, T. G., Protein secondary structure from deep-UVresonance Raman spectroscopy. Journal of Raman Spectroscopy 2006, 37, 277-282.[87] Efremov, E. V., Ariese, F., Gooijer, C., Achievements in resonance Raman spectroscopy.Analytica Chimica Acta 2008, 606, 119-134.[88] Wang, H., Malvadkar, N., Koytek, S., Bylander, J., Reeves, W. B., Demirel, M. C., Quan-titative analysis of creatinine in urine by metalized nanostructured parylene. Journal ofbiomedical optics 2010, 15, 027004.[89] Talley, C. E., Jackson, J. B., Oubre, C., Grady, N. K., Hollars, C. W., Lane, S. M., Huser, T.R., Nordlander, P., Halas, N. J., Surface-enhanced Raman scattering from individual Aunanoparticles and nanoparticle dimer substrates. Nano Letters 2005, 5, 1569-1574.[90] Haynes, C. L., Van Duyne, R. P., Nanosphere lithography: A versatile nanofabrication toolfor studies of size-dependent nanoparticle optics. The Journal of Physical Chemistry B2001, 105, 5599-5611.[91] Willets, K. A., Van Duyne, R. P., Localized surface plasmon resonance spectroscopy andsensing. Annual Review of Physical Chemistry 2007, 58, 267-297.[92] Blackie, E. J.,Ru, E. C. L.,Etchegoin, P. G. Single-molecule surface-enhanced Raman spec-troscopy of nonresonant molecules. Journal of the American Chemical Society 2009, 131,14466-14472.[93] Doering, W. E., Nie, S., Single-molecule and single-nanoparticle SERS: examining theroles of surface active sites and chemical enhancement. The Journal of Physical ChemistryB 2002, 106, 311-317.[94] Zhang, J., Li, X., Sun, X., Li, Y., Surface enhanced Raman scattering effects of silver col-loids with different shapes. The Journal of Physical Chemistry B 2005, 109, 12544-12548.[95] Kneipp, K., Wang, Y., Kneipp, H., Perelman, L. T., Itzkan, I., Dasari, R. R., Feld, M. S., Sin-gle molecule detection using surface-enhanced Raman scattering (SERS). Physical reviewletters 1997, 78, 1667-1670.[96] Pristinski, D., Tan, S., Erol, M., Du, H., Sukhishvili, S., In situ SERS study of Rho-damine 6G adsorbed on individually immobilized Ag nanoparticles. Journal of RamanSpectroscopy 2006, 37, 762-770.[97] Moskovits, M., Surface-enhanced Raman spectroscopy: a brief retrospective. Journal ofRaman Spectroscopy 2005, 36, 485-496.122Bibliography[98] Barhoumi, A., Zhang, D., Tam, F., Halas, N. J., Surface-Enhanced Raman Spectroscopy ofDNA. Journal of the American Chemical Society 2008, 130, 5523-5529.[99] Volkan, M., Stokes, D. L., Vo-Dinh, T., Surface-enhanced Raman of dopamine and neu-rotransmitters using sol-gel substrates and polymer-coated fiber-optic probes. Appliedspectroscopy 2000, 54, 1842-1848.[100] Dijkstra, R. J., Scheenen, W. J., Dam, N., Roubos, E. W., Ter Meulen, J. J., Monitoringneurotransmitter release using surface-enhanced Raman spectroscopy. Journal of Neuro-science Methods 2007, 159, 43-50.[101] Suh, J. S., Moskovits, M., Surface-enhanced Raman spectroscopy of amino acids and nu-cleotide bases adsorbed on silver. Journal of the American Chemical Society 1986, 108,4711-4718.[102] Sha, M. Y., Xu, H., Natan, M. J., Cromer, R., Surface-enhanced Raman scattering tags forrapid and homogeneous detection of circulating tumor cells in the presence of humanwhole blood. Journal of the American Chemical Society 2008, 130, 17214-17215.[103] Mohs, A. M., Mancini, M. C., Singhal, S., Provenzale, J. M., Leyland-Jones, B., Wang, M.D., Nie, S., Hand-held spectroscopic device for in vivo and intraoperative tumor detection:contrast enhancement, detection sensitivity, and tissue penetration. Analytical chemistry2010, 82, 9058-9065.[104] Grubisha, D. S., Lipert, R. J., Park, H.-Y., Driskell, J., Porter, M. D., Femtomolar detectionof prostate-specific antigen: an immunoassay based on surface-enhanced Raman scatter-ing and immunogold labels. Analytical chemistry 2003, 75, 5936-5943.[105] Haes, A. J., Chang, L., Klein, W. L., Van Duyne, R. P., Detection of a biomarker forAlzheimer’s disease from synthetic and clinical samples using a nanoscale optical biosen-sor. Journal of the American Chemical Society 2005, 127, 2264-2271.[106] Schlu¨cker, S., SERS microscopy: nanoparticle probes and biomedical applications.ChemPhysChem 2009, 10, 1344-1354.[107] Moore, B. D., Stevenson, L., Watt, A., Flitsch, S., Turner, N. J., Cassidy, C., Graham, D.,Rapid and ultra-sensitive determination of enzyme activities using surface-enhanced res-onance Raman scattering. Nature biotechnology 2004, 22, 1133-1138.[108] M¨uller, M., Schins, J. M., Imaging the thermodynamic sate of lipid membranes with mul-tiplex CARS microscopy. The Journal of Physical Chemistry B 2002, 106, 3715-3723.[109] Volkmer, A., Vibrational imaging and microspectroscopies based on coherent anti-StokesRaman scattering microscopy. Journal of Physics D: Applied Physics 2005, 38, R59-R81.[110] Chen, K. H., Spontaneous Raman and coherent anti-Stokes Raman spectroscopy of in-frared multiphoton-excited molecules. Doctoral dissertation 1989, Harvard University.[111] Balu, M., Liu, G., Chen, Z., Tromberg, B. J., Potma, E. O., Fiber delivered probe for efficientCARS imaging of tissues. Opt Express 2010, 18, 2380-2388.123Bibliography[112] Kano, H., Hamaguchi, H. O., Vibrationally resonant imaging of a single living cellby supercontinum-based multiplex coherent anti-Stokes Raman scattering microspec-troscopy. Optics express 2005, 13, 1322-1327.[113] Dufresne, E., Corwin, E., Greenblatt, N., Ashmore, J., Wang, D., Dinsmore, A., Cheng, J.,Xie, X., Hutchinson, J., Weitz, D., Flow and fracture in drying nanoparticle suspensions.Physical review letters 2003, 91, 224501.[114] Kukura, P., McCamant, D. W., Mathies, R. A., Femtosecond stimulated Raman spec-troscopy. Annual Review of Physical Chemistry 2007, 58, 461-488.[115] Pettinger, B., Ren, B., Picardi, G., Schuster, R., Ertl, G., Nanoscale probing of adsorbedspecies by tip-enhanced Raman spectroscopy.. Physical review letters 2004, 92, 096101.[116] Joseph, E., Application of FTIR microscopy to cultural heritage materials. Doctoral dis-sertation 2009, Universita` di Bologna.[117] Forsska˚hl, I., Kentta¨, E., Kyyro¨nen, P., Sundstro¨m, O., Depth profiling of a photochemi-cally yellowed paper. Part II: FT-IR Techniques. Applied Spectroscopy 1995, 49, 163-170.[118] Hind, A. R., Bhargava, S. K., McKinnon, A., At the solid/liquid interface: FTIR/ATR-thetool of choice. Advances in colloid and interface science 2001, 93, 91-114.[119] Tang, C. Y., Kwon, Y. N., Leckie, J. O., Probing the nano- and micro-scales of reverseosmosis membranes-A comprehensive characterization of physiochemical properties ofuncoated and coated membranes by XPS, TEM, ATR-FTIR, and streaming potential mea-surements. Journal of Membrane Science 2007, 287, 146-156.[120] Go¨bel, R., Krska, R., Kellner, R., Seitz, R. W., Tomellini, S. A., Investigation of differentpolymers as coating materials for IR/ATR spectroscopic trace analysis of chlorinated hy-drocarbons in water. Applied spectroscopy 1994, 48, 678-683.[121] McClelland, J. F., Jones, R. W., Luo, S., Seaverson, L. M., A practical guide to FTIR pho-toacoustic spectroscopy. CRC Press 1993, Boca Raton, Florida, 101-144.[122] Pandey, K. K., Theagarajan, K. S., Analysis of wood surfaces and ground wood by diffusereflectance (DRIFT) and photoacoustic (PAS) Fourier transform infrared spectroscopictechniques. Holz als Roh-und Werkstoff 1997, 55, 383-390.[123] Bjarnestad, S., Dahlman, O., Chemical compositions of hardwood and softwood pulpsemploying photoacoustic Fourier transform infrared spectroscopy in combination withpartial least-squares analysis. Analytical chemistry 2002, 74, 5851-5858.[124] Rockley, M. G., Devlin, J. P., Photoacoustic infrared spectra (IR-PAS) of aged and fresh-cleaved coal surfaces. Applied spectroscopy 1980, 34, 407-408.[125] Du, C., Linker, R., Shaviv, A., Identification of agricultural Mediterranean soils usingmid-infrared photoacoustic spectroscopy. Geoderma 2008, 143, 85-90.124Bibliography[126] Mendelson, Y., Clermont, A. C., Peura, R. A., Lin, B. C., Blood glucose measurement bymultiple attenuated total reflection and infrared absorption spectroscopy. Biomedical En-gineering, IEEE Transactions on 1990, 37, 458-465.[127] Budinova, G., Salva, J., Volka, K., Application of molecular spectroscopy in the mid-infrared region to the determination of glucose and cholesterol in whole blood and inblood serum. Applied spectroscopy 1997, 51, 631-635.[128] Cazes J., Analytical Instrumentation Handbook Third edition. CRC Press, November 302004.[129] Tague, T., Infrared and Raman Microscopy: Complimentary or redundant techniques?Microscopy and Microanalysis 2007, 13, 1696-1697.[130] Prati, S., Joseph, E., Sciutto, G., Mazzeo, R., New advances in the application of FTIRmicroscopy and spectroscopy for the characterization of artistic materials. Accounts ofchemical research 2010, 43, 792-801.[131] Pironon, J., Sawatzki, J., Letter: NIR FT-Raman microspectroscopy of fluid inclu-sions: Comparisons with VIS Raman and FT-IR microspectroscopies. Geochimica et Cos-mochimica Acta 1991, 55, 3885-3891.[132] Barres, O., Burneau, A., Dubessy, J., Pagel, M., Application of micro-FT-IR spectroscopyto individual hydrocarbon fluid inclusion analysis. Applied spectroscopy 1987, 41, 1000-1008.[133] Bo¨cker, U., Ofstad, R., Wu, Z., Bertram, H. C., Sockalingum, G. D., Manfait, M., Ege-landsdal, B., Kohler, A., Revealing covariance structures in Fourier transform infraredand Raman microspectroscopy spectra: a study on pork muscle fiber tissue subjected todifferent processing parameters. Applied spectroscopy 2007, 61, 1032-1039.[134] Krafft, C., Codrich, D., Pelizzo, G., Sergo, V., Raman and FTIR microscopic imaging ofcolon tissue: a comparative study. Journal of biophotonics 2008, 1, 154-169.[135] Frost, R. L., Martens, W., Williams, P. A., Kloprogge, J. T., Raman and infrared spectro-scopic study of the vivianite-group phosphates vivianite, baricite and bobierrite. Miner-alogical Magazine 2002, 66, 1063-1073.[136] Afseth, N. K., Segtnan, V. H., Wold, J. P., Raman spectra of biological samples: A study ofpreprocessing methods. Applied spectroscopy 2006, 60, 1358-1367.[137] Savitzky, A., Golay, M. J., Smoothing and differentiation of data by simplified least squaresprocedures. Analytical chemistry 1964, 36, 1627-1639.[138] Lieber, C. A., Mahadevan-Jansen, A., Automated method for subtraction of fluorescencefrom biological Raman spectra. Applied spectroscopy 2003, 57, 1363-1367.[139] Zhao, J., Lui, H., McLean, D. I., Zeng, H., Automated autofluorescence background sub-traction algorithm for biomedical Raman spectroscopy. Applied spectroscopy 2007, 61,1225-1232.125Bibliography[140] Beier, B. D.,Berger, A. J., Method for automated background subtraction from Ramanspectra containing known contaminants. Analyst 2009, 134, 1198.[141] Mosier-Boss, P. A., Lieberman, S. H., Newbery, R., Fluorescence rejection in Raman spec-troscopy by shifted-spectra, edge detection, and FFT filtering techniques. Applied spec-troscopy 1995, 49, 630-638.[142] Schulze, G., Jirasek, A., Yu, M. M. L., Lim, A., Turner, R. F. B., Blades, M. W., Investiga-tion of selected baseline removal techniques as candidates for automated implementation.Applied spectroscopy 2005, 59, 545-574.[143] Alsberg, B. K., Woodward, A. M., Kell, D. B., An introduction to wavelet transforms forchemometricians: a time-frequency approach. Chemometrics and Intelligent LaboratorySystems 1997, 37, 215-239.[144] Bos, M.,Vrielink, J. The wavelet transform for preprocessing IR spectra in the identifi-cation of mono-and di-substituted benzenes. Chemometrics and Intelligent LaboratorySystems 1994, 23, 115-122.[145] Hu, Y., Jiang, T., Shen, A., Li, W., Wang, X., Hu, J., A background elimination methodbased on wavelet transform for Raman spectra. Chemometrics and Intelligent LaboratorySystems 2007, 85, 94-101.[146] Ehrentreich, F.,Su¨mmchen, L., Spike removal and denoising of Raman spectra by wavelettransform methods. Analytical chemistry 2001, 73, 4364-4373.[147] Alsberg, B. K., Woodward, A. M., Kell, D. B., An introduction to wavelet transforms forchemometricians: a time-frequency approach. Chemometrics and Intelligent LaboratorySystems 1997, 37, 215-239.[148] Bos, M.,Vrielink, J. The wavelet transform for preprocessing IR spectra in the identifi-cation of mono-and di-substituted benzenes. Chemometrics and Intelligent LaboratorySystems 1994, 23, 115-122.[149] Hu, Y., Jiang, T., Shen, A., Li, W., Wang, X., Hu, J., A background elimination methodbased on wavelet transform for Raman spectra. Chemometrics and Intelligent LaboratorySystems 2007, 85, 94-101.[150] Ehrentreich, F.,Su¨mmchen, L., Spike removal and denoising of Raman spectra by wavelettransform methods. Analytical chemistry 2001, 73, 4364-4373.[151] Liu, Z., Cai, W., Shao, X., A weighted multiscale regression for multivariate calibration ofnear infrared spectra. Analyst 2009, 134, 261-266.[152] Shao, X., Pang, C., Su, Q., A novel method to calculate the approximate derivative pho-toacoustic spectrum using continuous wavelet transform. Fresenius’ journal of analyticalchemistry 2000, 367, 525-529.126Bibliography[153] Shao, L., Griffiths, P. R., Automatic baseline correction by wavelet transform for quantita-tive open-path Fourier Transform Infrared Spectroscopy. Environmental science & tech-nology 2007, 41, 7054-7059.[154] Buckheit, J., Chen, S., Donoho, D., Johnstone, I., Scargle, J. D., Wavelab reference manual.Version 0.700, 1995.[155] Matlab toolbox downloaded from http://statweb.stanford.edu/˜wavelab/Wavelab_850/download.html September 2014.[156] MATLAB and Wavelet Toolbox Release 2009a, The MathWorks, Inc., Natick, Mas-sachusetts, United States.[157] Geladi, P.,Kowalski, B. R. Partial Least-Squares regression: a tutorial. Analytica ChimicaActa 1986, 185, 1-17.[158] Smith, L. I., A tutorial on Principal Components Analysis. Cornell University 2002.[159] Lin, M., Al-Holy, M., Al-Qadiri, H., Kang, D.-H., Cavinato, A. G., Huang, Y., Rasco, B.A., Discrimination of Intact and Injured Listeria monocytogenesby Fourier Transform In-frared Spectroscopy and Principal Component Analysis. Journal of agricultural and foodchemistry 2004, 52, 5769-5772.[160] Woo, Y. A., Kim, H. J., Chung, H., Classification of cultivation area of ginseng radix withNIR and Raman spectroscopy. Analyst 1999, 124, 1223-1226.[161] Tan, K. M., Barman, I., Dingari, N. C., Singh, G. P., Chia, T. F., Tok, W. L., Toward the devel-opment of Raman spectroscopy as a nonperturbative online monitoring tool for gasolineadulteration. Analytical chemistry 2013, 85, 1846-1851.[162] Deinum, G., Rodriguez, D., Romer, T. J., Fitzmaurice, M., Kramer, J. R., Feld, M. S., His-tological classification of Raman spectra of human coronary artery atherosclerosis usingprincipal component analysis. Applied spectroscopy 1999, 53, 938-942.[163] Farag, A. A., Elhabian, S. Y., A Tutorial on Data Reduction. University of Louisville 2008.[164] Crow, P., Molckovsky, A., Stone, N., Uff, J., Wilson, B., WongKeeSong, L. M., Assessment offiberoptic near-infrared Raman spectroscopy for diagnosis of bladder and prostate cancer.Urology 2005, 65, 1126-1130.[165] Molckovsky, A., Song, L. M. W. K., Shim, M. G., Marcon, N. E., Wilson, B. C., Diagnosticpotential of near-infrared Raman spectroscopy in the colon: Differentiating adenomatousfrom hyperplastic polyps. Gastrointestinal Endoscopy 2003, 57, 396-402.[166] Crow, P., Barrass, B., Kendall, C., Hart-Prieto, M., Wright, M., Persad, R.,Stone, N., Theuse of Raman spectroscopy to differentiate between different prostatic adenocarcinomacell lines. British journal of cancer 2005, 92, 2166-2170.[167] Pohar, M., Blas, M., Turk, S., Comparison of logistic regression and linear discriminantanalysis: a simulation study. Metodolski Zvezki 2004, 1, 143-161.127Bibliography[168] Aliferis, C. F., Tsamardinos, I. Support Vector Machines. Vanderbilt University 2004.[169] Luts, J., Ojeda, F., Van de Plas, R., De Moor, B., Van Huffel, S.,Suykens, J. A. K., A tutorialon support vector machine-based methods for classification problems in chemometrics.Analytica Chimica Acta 2010, 665, 129-145.[170] Harz, M.,Ro¨sch, P.,Peschke, K. D.,Ronneberger, O.,Burkhardt, H.,Popp, J., Micro-Ramanspectroscopic identification of bacterial cells of the genus Staphylococcus and dependenceon their cultivation conditions. Analyst 2005, 130, 1543.[171] Caetano, S., U¨stu¨n, B., Hennessy, S., Smeyers-Verbeke, J., Melssen, W., Downey, G., Buy-dens, L., Heyden, Y. V., Geographical classification of olive oils by the application of CARTand SVM to their FT-IR. Journal of Chemometrics 2007, 21, 324-334.[172] Martin, T. C., Moecks, J., Belooussov, A., Cawthraw, S., Dolenko, B., Eiden, M., Frese,von, J., K hler, W., Schmitt, J., Somorjai, R., Udelhoven, T., Verzakov, S.,Petrich, W., Clas-sification of signatures of Bovine Spongiform Encephalopathy in serum using infraredspectroscopy. Analyst 2004, 129, 897-901.[173] Ferra˜, M. F., Godoy, S. C., Gerbase, A. E., Mello, C., Furtado, J. C., Petzhold, C. L., Poppi,R. J., Non-destructive method for determination of hydroxyl value of soybean polyol byLS-SVM using HATR/FT-IR. Analytica Chimica Acta 2007, 595, 114-119.[174] Balabin, R. M., Smirnov, S. V., Melamine detection by mid- and near-infrared (MIR/NIR)spectroscopy: A quick and sensitive method for dairy products analysis including liquidmilk, infant formula, and milk powder. Talanta 2011, 85, 562-568.[175] Thissen, U., Pepers, M., U¨stu¨n, B., Melssen, W. J., Buydens, L. M. C., Comparing supportvector machines to PLS for spectral regression applications. Chemometrics and IntelligentLaboratory Systems 2004, 73, 169-179.[176] Wu, D., He, Y., Feng, S., Sun, D. W., Study on infrared spectroscopy technique for fastmeasurement of protein content in milk powder based on LS-SVM. Journal of Food Engi-neering 2008, 84, 124-131.[177] Abdi, H., Partial least squares regression (PLS-regression). Encyclopedia for researchmethods for the social sciences 2003, 792-795.[178] Wold, S., Sjo¨stro¨m, M., Eriksson, L., PLS-regression: a basic tool of chemometrics. Chemo-metrics and Intelligent Laboratory Systems 2001, 58, 109-130.[179] Henrique, C. M., Teo´filo, R. F., Sabino, L., Ferreira, M. M. C., Cereda, M. P., Classifica-tion of cassava starch films by physicochemical properties and water vapor permeabilityquantification by FTIR and PLS. Journal of Food Science 2007, 72, E184-E189.[180] Dyrby, M.,Engelsen, S. B.,Nørgaard, L.,Bruhn, M.,Lundsberg-Nielsen, L., Chemometricquantitation of the active substance (containing C- N) in a pharmaceutical tablet us-ing near-infrared (NIR) transmittance and NIR FT-Raman spectra. Applied Spectroscopy2002, 56, 579-585.128Bibliography[181] Hoang, V., Bhardwaj, N. K., Nguyen, K. L., A FTIR method for determining the contentof hexeneuronic acid (hexA) and Kappa number of a high-yield kraft pulp. CarbohydratePolymers 2005, 61, 5-9.[182] Franc, V., Hlava´c, V., Statistical pattern recognition toolbox for Matlab. Prague, Czech:Center for Machine Perception, Czech Technical University 2004.[183] Matlab toolbox downloaded from http://cmp.felk.cvut.cz/cmp/software/stprtool/September 2014.[184] Tan, C., Qin, X., Li, M., Comparison of chemometric methods for brand classification ofcigarettes by near-infrared spectroscopy. Vibrational Spectroscopy 2009, 51, 276-282.[185] Xie, C., Mace, J., Dinno, M. A., Li, Y. Q., Tang, W., Newton, R. J., Gemperline, P. J., Identi-fication of single bacterial cells in aqueous solution using confocal laser tweezers Ramanspectroscopy. Analytical chemistry 2005, 77, 4390-4397.[186] Tan, A., Bi, W., Identification of spilled oils by NIR spectroscopy technology based onKPCA and LSSVM. In Proc. of SPIE 2011, 8193, 81933K-1.[187] Leardi, R., Application of genetic algorithm-PLS for feature selection in spectral data sets.Journal of Chemometrics 2000, 14, 643-655.[188] Leardi, R., Lupia´n˜ez Gonza´lez, A., Genetic algorithms applied to feature selection in PLSregression: how and when to use them. Chemometrics and Intelligent Laboratory Systems1998, 41, 195-207.[189] Leardi, R., Nørgaard, L., Sequential application of backward interval partial least squaresand genetic algorithms for the selection of relevant spectral regions. Journal of Chemo-metrics 2005, 18, 486-497.[190] Nørgaard, L., Saudland, A., Wagner, J., Nielsen, J. P., Munck, L., Engelsen, S. B., Intervalpartial least-squares regression (iPLS): A comparative chemometric study with an exam-ple from near-infrared spectroscopy. Applied Spectroscopy 2000, 54, 413-419.[191] National Center for Health Statistics Infertility and Impaired Fecundity in the UnitedStates,1982-2010: Data From the National Survey of Family Growth. 2013, 1-19.[192] Bushnik, T., Cook, J. L., Yuzpe, A. A., Tough, S., Collins, J., Estimating the prevalence ofinfertility in Canada. Human Reproduction 2012, 27, 738-746.[193] Seli, E., Sakkas, D., Scott, R., Kwok, S. C., Rosendahl, S. M., Burns, D. H., Noninvasivemetabolomic profiling of embryo culture media using Raman and near-infrared spec-troscopy correlates with reproductive potential of embryos in women undergoing in vitrofertilization. Fertility and Sterility 2007, 88, 1350-1357.[194] Scott, R., Seli, E., Miller, K., Sakkas, D., Scott, K., Burns, D. H., Noninvasive metabolomicprofiling of human embryo culture media using Raman spectroscopy predicts embryonicreproductive potential: a prospective blinded pilot study. Fertility and Sterility 2008, 90,77-83.129Bibliography[195] Kyla Jones, In vitro Fertilization, BIO108: Human Biology, UNC-Asheville, s2008.[196] Starr, C., McMillan, B., Human Biology Eighth Edition. Brooks Cole Press, December 29,2008.[197] U.S. Department of Health and Human Services Centers for Disease Control and Preven-tion, 2010 ART Clinic Report-Full. 1-534.[198] Gardner, D. K., Sakkas, D., Assessment of Embryo Viability: The Ability to Select a SingleEmbryo for Transfer-a Review. Placenta 2003, 24, S5-S12.[199] Racowsky, C., Stern, J. E., Gibbons, W. E., Behr, B., Pomeroy, K. O., Biggers, J. D., Nationalcollection of embryo morphology data into Society for Assisted Reproductive TechnologyClinic Outcomes Reporting System: associations among day 3 cell number, fragmentationand blastomere asymmetry, and live birth rate. Fertility and Sterility 2011, 95, 1985-1989.[200] Hossain, A., Potential for Empowering and Broadening the Application of the SART Em-bryo Grading System.Journal of Clinical Embryology 2012, 15, 3, 54-66.[201] National Center for Chronic Disease Prevention and Health Promotion 2010 ART Na-tional Summary Report Presentations. 2014, 1-60.[202] Nagy, Z. P., Sakkas, D., Behr, B., Symposium: Innovative techniques in human embryo vi-ability assessmentNon-invasive assessment of embryo viability by metabolomic profilingof culture media (’metabolomics’). Reproductive biomedicine online 2008, 17, 502-507.[203] Botros, L., Sakkas, D., Seli, E., Metabolomics and its application for non-invasive embryoassessment in IVF. Molecular human reproduction 2008, 14, 679-690.[204] Singh, R., Sinclair, K. D., Metabolomics: approaches to assessing oocyte and embryo qual-ity. Theriogenology 2007, 68, S56-S62.[205] Gruber, I., Klein, M., Embryo culture media for human IVF: which possibilities exist?Journal of the Turkish German Gynecological Association 2011, 12, 110-117.[206] Lane, M., Gardner, D. K., Embryo culture medium: which is the best? Best Practice &Research Clinical Obstetrics & Gynaecology 2007, 21, 83-100.[207] Hardy, K., Hooper, M., Handyside, A. H., Rutherford, A. J., Winston, R., Leese, H. J., Non-invasive measurement of glucose and pyruvate uptake by individual human oocytes andpreimplantation embryos. Human Reproduction 1989, 4, 188-191.[208] Gott, A. L., Hardy, K., Winston, R., Leese, H. J., Non-invasive measurement of pyruvateand glucose uptake and lactate production by single human preimplantation embryos.Human Reproduction 1990, 5, 104-108.[209] Brison, D. R., Identification of viable embryos in IVF by non-invasive measurement ofamino acid turnover. Human Reproduction 2004, 19, 2319-2324.130Bibliography[210] Houghton, F. D., Hawkhead, J. A., Humpherson, P. G., Hogg, J. E., Balen, A. H., Ruther-ford, A. J., Leese, H. J., Non-invasive amino acid turnover predicts human embryo devel-opmental capacity. Human Reproduction 2002, 17, 999-1005.[211] Marhuenda-Egea, F. C., Martnez-Sabater, E., Gonslvez-lvarez, R., Lledo´, B., Ten, J., Bern-abeu, R., A crucial step in assisted reproduction technology: human embryo selectionusing metabolomic evaluation. Fertility and Sterility 2010, 94, 772-774.[212] Marhuenda-Egea, F. C., Gonsa´lvez-Alvarez, R., Martı´nez-Sabater, E., Lledo´, B., Ten, J.,Bernabeu, R., Improving human embryos selection in IVF: non-invasive metabolomic andchemometric approach. Metabolomics 2010, 7, 247-256.[213] Seli, E., Vergouw, C. G., Morita, H., Botros, L., Roos, P., Lambalk, C. B., Yamashita N.,Kato O., Sakkas, D., Noninvasive metabolomic profiling as an adjunct to morphology fornoninvasive embryo assessment in women undergoing single embryo transfer. Fertilityand Sterility 2009, 94, 535-542.[214] Ahlstro¨m, A., Wikland, M., Rogberg, L., Barnett, J. S., Tucker, M., Hardarson, T., Cross-validation and predictive value of near-infrared spectroscopy algorithms for day-5 blas-tocyst transfer. Reproductive biomedicine online 2011, 22, 477-484.[215] Hardarson, T., Ahlstro¨m, A.., Rogberg, L., Botros, L., Hillensjo, T., Westlander, G., Sakkas,D., Wikland, M., Non-invasive metabolomic profiling of Day 2 and 5 embryo culturemedium: a prospective randomized trial. Human Reproduction 2011, 27, 89-96.[216] Sfontouris, I., Zorzovilis, I., Lainas, G., Lainas, T., Petsas, G., Sakkas, D., Non-invasivemetabolomic analysis using a commercial NIR instrument for embryo selection. Journalof human reproductive sciences 2013, 6, 133.[217] Seli, E., Botros, L., Sakkas, D., Burns, D. H., Noninvasive metabolomic profiling of embryoculture media using proton nuclear magnetic resonance correlates with reproductive po-tential of embryos in women undergoing in vitro fertilization. Fertility and Sterility 2008,90, 2183-2189.[218] Barnes, R. J., Dhanoa, M. S., Lister, S. J., Standard normal variate transformation andde-trending of near-infrared diffuse reflectance spectra. Applied Spectroscopy 1989, 43,772-777.[219] Mathlouthi, M., Vinh Luu, D., Laser-Raman spectra of d-glucose and sucrose in aqueoussolution. Carbohydrate Research 1980, 81, 203-212.[220] Ren, M., Arnold, M. A., Comparison of multivariate calibration models for glucose, urea,and lactate from near-infrared and Raman spectra. Analytical and Bioanalytical Chem-istry 2007, 387, 879-888.[221] Lorber, A., Faber, K., Kowalski, B. R., Net analyte signal calculation in multivariate cali-bration. Analytical Chemistry 1997, 69, 1620-1626.131Bibliography[222] Braga, J. W. B., Poppi, R. J., Figures of merit for the determination of the polymorphicpurity of carbamazepine by infrared spectroscopy and multivariate calibration. Journalof pharmaceutical sciences 2004, 93, 2124-2134.[223] Valderrama, P., Braga, J. W. B., Poppi, R. J., Variable selection, outlier detection, andfigures of merit estimation in a partial least-squares regression multivariate calibrationmodel. A case study for the determination of quality parameters in the alcohol indus-try by near-infrared spectroscopy. Journal of agricultural and food chemistry 2007, 55,8331-8338.[224] Frantz, J. D., Raman spectra of potassium carbonate and bicarbonate aqueous fluids atelevated temperatures and pressures: comparison with theoretical simulations. Chemicalgeology 1998, 152, 211-225.[225] Shah, N. C., Lyandres, O., Walsh, J. T., Glucksberg, M. R., Van Duyne, R. P., Lactate and se-quential lactate-glucose sensing using surface-enhanced Raman spectroscopy. AnalyticalChemistry 2007, 79, 6927-6932.[226] Cassanas, G., Morssli, M., Fabregue, E., Bardet, L., Vibrational spectra of lactic acid andlactates. Journal of Raman spectroscopy 1991, 22, 409-413.[227] Hardy, K., Handyside, A. H., Winston, R. M., The human blastocyst: cell number, deathand allocation during late preimplantation development in vitro. Development 1989, 107,597-604.[228] Berger, A. J., Koo, T.-W., Itzkan, I., Horowitz, G., Feld, M. S., Multicomponent blood anal-ysis by near-infrared Raman spectroscopy. Applied optics 1999, 38, 2916-2926.[229] Carmona, P., Molina, M., Calero, M., Bermejo-Pareja, F., Martnez-Martn, P., Toledano,A., Discrimination Analysis of Blood Plasma Associated with Alzheimer’s Disease UsingVibrational Spectroscopy. Journal of Alzheimer’s Disease 2013, 34, 911-920[230] Dou, X., Yamaguchi, Y., Yamamoto, H., Doi, S., Ozaki, Y., A highly sensitive, compactRaman system without a spectrometer for quantitative analysis of biological samples. Vi-brational Spectroscopy 1997, 14, 199-205.[231] Abu-Absi, N. R., Kenty, B. M., Cuellar, M. E., Borys, M. C., Sakhamuri, S., Strachan, D. J.,Hausladen, M. C., Li, Z. J., Real time monitoring of multiple parameters in mammaliancell culture bioreactors using an in-line Raman spectroscopy probe. Biotechnology andbioengineering 2010, 108, 1215-1221.[232] Lee, H. L. T., Boccazzi, P., Gorret, N., Ram, R. J., Sinskey, A. J., In situ bioprocess monitor-ing of Escherichia coli bioreactions using Raman spectroscopy. Vibrational Spectroscopy2004, 35, 131-137.[233] Berger, A. J., Wang, Y., Feld, M. S., Rapid, noninvasive concentration measurements ofaqueous biological analytes by near-infrared Raman spectroscopy.Applied optics 1996,35, 1, 209-212.132Bibliography[234] Berger, A. J., Itzkan, I., Feld, M. S., Feasibility of measuring blood glucose concentrationby near-infrared Raman spectroscopy. Spectrochimica Acta Part A 1997, Molecular andBiomolecular Spectroscopy, 53, 2, 287-292.[235] Sikirzhytski, V., Sikirzhytskaya, A., Lednev, I. K., Multidimensional Raman SpectroscopicSignatures as a Tool for Forensic Identification of Body Fluid Traces: A Review. Appliedspectroscopy 2011, 65, 11, 1223-1232.[236] Zhang, X. F., Zou, M. Q., Qi, X. H., Liu, F., Zhu, X. H., Zhao, B. H., Detection of melaminein liquid milk using surface-enhanced Raman scattering spectroscopy. Journal of RamanSpectroscopy 2010, 41, 12, 1655-1660.[237] Stosch, R., Henrion, A., Schiel, D., Gu¨ttler, B., Surface-enhanced Raman scattering basedapproach for quantitative determination of creatinine in human serum. Analytical chem-istry 2005, 77, 22, 7386-7392.[238] Moog, R. S., McGUIRL, M. A., Cote, C. E., Dooley, D. M., Evidence for methoxatin(pyrroloquinolinequinone) as the cofactor in bovine plasma amine oxidase from reso-nance Raman spectroscopy. Proceedings of the National Academy of Sciences 1986, 83,22, 8435-8439.[239] Holtz, M., Dasgupta, P. K., Zhang, G., Small-Volume Raman Spectroscopy with a LiquidCore Waveguide. Analytical Chemistry 1999, 71, 2934-2938.[240] Tian, Y., Zhang, L.; Zuo, J., Li, Z., Gao, S., Lu, G., Raman sensitivity enhancement foraqueous absorbing sample using Teflon-AF 2400 liquid core optical fibre cell. AnalyticaChimica Acta 2007, 581, 154-158.[241] Marquardt, B. J., Vahey, P. G., Synovec, R. E., Burgess, L. W., A Raman waveguide detectorfor liquid chromatography. Analytical chemistry 1999, 71, 4808-4814.[242] Pelletier, M. J., Altkorn, R., Raman sensitivity enhancement for aqueous protein samplesusing a liquid-core optical-fiber cell. Analytical chemistry 2001, 73, 6, 1393-1397.[243] Qi, D., Berger, A. J., Quantitative analysis of Raman signal enhancement from aqueoussamples in liquid core optical fibers. Applied spectroscopy 2004, 58, 10, 1165-1171.[244] Song, L., Liu, S., Zhelyaskov, V., El-Sayed, M. A., Application of liquid waveguide toRaman spectroscopy in aqueous solution. Applied spectroscopy 1998, 52, 1364-1367.[245] Kris-Etherton, P. M., Pearson, T. A., Wan, Y., Hargrove, R. L., Moriarty, K., Fishell, V.,Etherton, T. D., High-monounsaturated fatty acid diets lower both plasma cholesteroland triacylglycerol concentrations. The American journal of clinical nutrition 1999, 70, 6,1009-1015.[246] Mattson, F. H., Grundy, S. M., Comparison of effects of dietary saturated, monounsatu-rated, and polyunsaturated fatty acids on plasma lipids and lipoproteins in man. Journalof lipid research 1985, 26, 2, 194-202.133Bibliography[247] Stark, A. H., Madar, P. Z., Olive oil as a functional food: epidemiology and nutritionalapproaches. Nutrition reviews 2002, 60, 6, 170-176.[248] Owen, R. W., Mier, W., Giacosa, A., Hull, W. E., Spiegelhalder, B., Bartsch, H., Pheno-lic compounds and squalene in olive oils: the concentration and antioxidant potentialof total phenols, simple phenols, secoiridoids, lignansand squalene. Food and ChemicalToxicology 2000, 38, 647-659.[249] Vigli, G., Philippidis, A., Spyros, A., Dais, P., Classification of Edible Oils by Employing31P and 1H NMR Spectroscopy in Combination with Multivariate Statistical Analysis.A Proposal for the Detection of Seed Oil Adulteration in Virgin Olive Oils. Journal ofAgricultural and Food Chemistry 2003, 51, 5715-5722.[250] Fragaki, G., Spyros, A., Siragakis, G., Salivaras, E., Dais, P., Detection of Extra Virgin OliveOil Adulteration with Lampante Olive Oil and Refined Olive Oil Using Nuclear MagneticResonance Spectroscopy and Multivariate Statistical Analysis. Journal of Agricultural andFood Chemistry 2005, 53, 2810-2816.[251] Poulli, K. I., Mousdis, G. A., Georgiou, C. A., Synchronous fluorescence spectroscopy forquantitative determination of virgin olive oil adulteration with sunflower oil. Analyticaland Bioanalytical Chemistry 2006, 386, 1571-1575.[252] Poulli, K. I., Mousdis, G. A., Georgiou, C. A., Rapid synchronous fluorescence method forvirgin olive oil adulteration assessment. Food Chemistry 2007, 105, 369-375.[253] Tay, A., Singh, R. K., Krishnan, S. S., Gore, J. P., Authentication of Olive Oil Adulteratedwith Vegetable Oils Using Fourier Transform Infrared Spectroscopy. LWT - Food Scienceand Technology 2002, 35, 99-103.[254] Gurdeniz, G., Ozen, B., Detection of adulteration of extra-virgin olive oil by chemometricanalysis of mid-infrared spectral data. Food Chemistry 2009, 116, 519-525.[255] Lerma-Garcı´a, M. J., Ramis-Ramos, G., Herrero-Martı´nez, J. M., Simo´-Alfonso, E. F., Au-thentication of extra virgin olive oils by Fourier-transform infrared spectroscopy. FoodChemistry 2010, 118, 78-83.[256] Yang, H., Irudayaraj, J., Comparison of near-infrared, Fourier transform-infrared, andFourier transform-Raman methods for determining olive pomace oil adulteration in extravirgin olive oil. Journal of the American Oil Chemists’ Society 2001, 78, 889-895.[257] de la Mata, P., Dominguez-Vidal, A., Bosque-Sendra, J. M., Ruiz-Medina, A., Cuadros-Rodrı´guez, L., Olive oil assessment in edible oil blends by means of ATR-FTIR and chemo-metrics. Food Control 2012, 23, 449-455.[258] o¨zttu¨rk, B., Yalin, A., o¨zdemir, D., Determination of olive oil adulteration with vegetableoils by near infrared spectroscopy coupled with multivariate calibration. Journal of NearInfrared Spectroscopy 2010, 18, 191.134Bibliography[259] Downey, G., McIntyre, P., Davies, A. N., Detecting and quantifying sunflower oil adulter-ation in extra virgin olive oils from the eastern mediterranean by visible and near-infraredspectroscopy. Journal of Agricultural and Food Chemistry 2002, 50, 5520-5525.[260] Christy, A. A., Kasemsumran, S., Du, Y., Ozaki, Y., The detection and quantification ofadulteration in olive oil by near-infrared spectroscopy and chemometrics. Analytical Sci-ences 2004, 20, 935-940.[261] Zou, M. Q., Zhang, X. F., Qi, X. H., Ma, H. L., Dong, Y., Liu, C. W., Guo, X., Wang, H., RapidAuthentication of Olive Oil Adulteration by Raman Spectrometry. Journal of Agriculturaland Food Chemistry 2009, 57, 6001-6006.[262] Zhang, X. F., Zou, M. Q., Qi, X. H., Liu, F., Zhang, C., Yin, F., Quantitative detection ofadulterated olive oil by Raman spectroscopy and chemometrics. Journal of Raman Spec-troscopy 2011, 42, 1784-1788.[263] El-Abassy, R. M., Donfack, P., Materny, A., Visible Raman spectroscopy for the discrimi-nation of olive oils from different vegetable oils and the detection of adulteration. Journalof Raman Spectroscopy 2009, 40, 1284-1289.[264] Baeten, V., Meurens, M., Morales, M. T., Aparicio, R., Detection of virgin olive oil adulter-ation by Fourier transform Raman spectroscopy. Journal of Agricultural and Food Chem-istry 1996, 44, 2225-2230.[265] Tapp, H. S., Defernez, M., Kemsley, E. K., FTIR Spectroscopy and Multivariate AnalysisCan Distinguish the Geographic Origin of Extra Virgin Olive Oils. Journal of Agriculturaland Food Chemistry 2003, 51, 6110-6115.[266] Downey, G., McIntyre, P., Davies, A. N., Geographic classification of extra virgin oliveoils from the eastern Mediterranean by chemometric analysis of visible and near-infraredspectroscopic data. Applied spectroscopy 2003, 57, 158-163.[267] Gurdeniz, G., Tokatli, F., Ozen, B., Differentiation of mixtures of monovarietal olive oilsby mid-infrared spectroscopy and chemometrics. European Journal of Lipid Science andTechnology 2007, 109, 1194-1202.[268] Casale, M., Sinelli, N., Oliveri, P., Di Egidio, V., Lanteri, S., Chemometrical strategies forfeature selection and data compression applied to NIR and MIR spectra of extra virginolive oils for cultivar identification. Talanta 2010, 80, 1832-1837.[269] Sinelli, N., Cosio, M. S., Gigliotti, C., Casiraghi, E., Preliminary study on application ofmid infrared spectroscopy for the evaluation of the virgin olive oil ”freshness”. AnalyticaChimica Acta 2007, 598, 128-134.[270] Cai, W., Wang, L., Pan, Z., Zuo, J., Xu, C., Shao, X. Application of the wavelet transformmethod in quantitative analysis of Raman spectra. Journal of Raman Spectroscopy 2001,32, 207-209.135Bibliography[271] Lo´pez-Dı´ez, E. C., Bianchi, G., Goodacre, R., Rapid Quantitative Assessment of the Adul-teration of Virgin Olive Oils with Hazelnut Oils Using Raman Spectroscopy and Chemo-metrics. Journal of Agricultural and Food Chemistry 2003, 51, 6145-6150.[272] Zhang, X., Qi, X., Zou, M., Liu, F., Rapid Authentication of Olive Oil by Raman Spec-troscopy Using Principal Component Analysis. Analytical Lett. 2011, 44, 2209-2220.[273] Sjo¨berg, J., Characterization of chemical pulp fiber surfaces with an emphasis on thehemicelluloses. Doctoral dissertation, KTH, 2003.[274] Pine, C. S., Birch, S., Chemical Composition of Wood.[275] Wyman, C. E., Decker, S. R., Himmel, M. E., Brady, J. W., Skopec, C. E., Viikari, L, Hy-drolysis of cellulose and hemicellulose. Polysaccharides 2005, Structural diversity andfunctional versatility, 1, 1023-1062.[276] A˜kerholm, M., Ultrastructural aspects of pulp fibers as studied by dynamic FT-IR spec-troscopy, Doctoral dissertation, KTH, 2003.[277] Teleman, A., Larsson, P. T., Iversen, T., On the accessibility and structure of xylan in birchkraft pulp. Cellulose 2001, 8, 209-215.[278] Scho¨nberg, C., Oksanen, T., Suurna¨kki, A., Kettunen, H., Buchert, J., The importance ofxylan for the strength properties of spruce kraft pulp fibres. Holzforschung 2001, 55,639-644.[279] dos Santos Muguet, M. C., Pedrazzi, C., Colodette, J. L., Xylan deposition onto eucalyptpulp fibers during oxygen delignification. Holzforschung 2011, 65, 605-612.[280] Gwon, J. G., Lee, S. Y., Doh, G. H., Kim, J. H., Characterization of chemically modifiedwood fibers using FTIR spectroscopy for biocomposites. Journal of Applied Polymer Sci-ence 2010, 3212-3219.[281] Oh, S. Y., Yoo, D. I., Shin, Y., Seo, G., FTIR analysis of cellulose treated with sodiumhydroxide and carbon dioxide. Carbohydrate Research 2005, 340, 417-428.[282] Carrillo, F., Colom, X., Sun˜ol, J. J., Saurina, J., Structural FTIR analysis and thermal char-acterisation of lyocell and viscose-type fibres. European Polymer Journal 2004, 40, 2229-2234.[283] Sun, J. X., Sun, X. F., Sun, R. C., Su, Y. Q., Fractional extraction and structural characteri-zation of sugarcane bagasse hemicelluloses. Carbohydrate Polymers 2004, 56, 195-204.[284] Fang, J. M., Sun, R. C., Tomkinson, J., Isolation and characterization of hemicelluloses andcellulose from rye straw by alkaline peroxide extraction. Cellulose 2000, 7, 87-107.[285] Dinand, E., Vignon, M., Chanzy, H., Heux, L., Mercerization of primary wall cellulose andits implication for the conversion of cellulose I- cellulose II. Cellulose 2002, 9, 7-18.[286] Mwaikambo, L. Y., Ansell, M. P., Chemical modification of hemp, sisal, jute, and kapokfibers by alkalization. Journal of Applied Polymer Science 2002, 84, 2222-2234.136Bibliography[287] Yue, Y. A comparative study of cellulose I and II fibers and nanocrystals. Master disserta-tion, Louisiana State University, 2011.[288] A˜kerholm, M., Hinterstoisser, B., Salme´n, L., Characterization of the crystalline structureof cellulose using static and dynamic FT-IR spectroscopy. Carbohydrate Research 2004,339, 569-578.[289] A˜kerholm, M., Salme´n, L., Dynamic FTIR spectroscopy for carbohydrate analysis of woodpulps. Journal of pulp and paper science 2002, 28, 245-249.[290] A˜kerholm, M., Salme´n, L., Interactions between wood polymers studied by dynamic FT-IR spectroscopy. Polymer 2001, 42, 963-969.[291] Test method T235, Alkali solubility of pulp at 25◦C. Tappi Press, Atlanta.[292] Popescu, C.-M., , M.-C., Singurel, G., Vasile, C., Argyropoulos, D. S., Willfor, S., Spectralcharacterization of eucalyptus wood. Applied Spectroscopy 2007, 61, 1168-1177.[293] Chen, H., Ferrari, C., Angiuli, M., Yao, J., Raspi, C., Bramanti, E., Qualitative and quan-titative analysis of wood samples by Fourier transform infrared spectroscopy and multi-variate analysis. Carbohydrate Polymers 2010, 82, 772-778.[294] Dammstro¨m, S., Salme´n, L., Gatenholm, P., On the interactions between cellulose andxylan, a biomimetic simulation of the hardwood cell wall. Bioresources 2008, 4, 3-14.[295] Dang, Z. The Investigation of Carboxyl Groups of Pulp Fibers during Kraft Pulping, Alka-line Peroxide Bleaching, and TEMPO-mediated Oxidation.Doctoral dissertation, GeorgiaInstitute of Technology, 2007.[296] Kacˇura´kova´, M., Capek, P., Sasinkova´, V., Wellner, N., Ebringerova´, A., FT-IR study ofplant cell wall model compounds: pectic polysaccharides and hemicelluloses. Carbohy-drate Polymers 2000, 43, 195-203.[297] Mansfield, S. D., Jong, E. de, Stephens, R. S., Saddler, J. N., Physical characterization ofenzymatically modified kraft pulp fibers. Journal of biotechnology 1997, 57, 205-216.[298] Kacˇura´kova´, M., Wilson, R. H., Developments in mid-infrared FT-IR spectroscopy of se-lected carbohydrates. Carbohydrate Polymers 2001, 44, 291-303.[299] Kacˇura´kova´, M., Belton, P. S., Wilson, R. H., Hirsch, J., Ebringerova´, A., Hydration prop-erties of xylan-type structures: an FTIR study of Xylooligosaccharides. Journal of the Sci-ence of Food and Agriculture 1998, 77, 38-44.[300] Sun, R. C., Fang, J. M., Tomkinson, J., Geng, Z. C., Liu, J. C., Fractional isolation, physico-chemical characterization and homogeneous esterification of hemicelluloses from fast-growing poplar wood. Carbohydrate Polymers 2001, 44(1), 29-39.[301] Wan, J., Wang, Y., Xiao, Q., Effects of hemicellulose removal on cellulose fiber structureand recycling characteristics of eucalyptus pulp. Bioresource Technology 2010, 101, 4577-4583.137Bibliography[302] Liu, Y., Hu, H., X-ray diffraction study of bamboo fibers treated with NaOH. Fibers andPolymers 2008, 9, 735-739.[303] Shackford, D. L., A comparison of pulping and bleaching of kraft softwood and euca-lyptus pulps. In Proceedings of the 36th International Pulp and Paper Congress 2003October.[304] Nanko, H., Button, A., Swann, C. E., The World of Market Pulp. WOMP LLC 2005.[305] Ibrahim, A., Oldham, P. B., Conners, T. E., Schultz, T. P., Rapid characterization of woodpulp lignin by Fourier transform Raman spectroscopy. Microchemical journal 1997, 56,393-402.[306] Sun, Z., Ibrahim, A., Oldham, P. B., Schultz, T. P., Conners, T. E., Rapid lignin measure-ment in hardwood pulp samples by near-infrared Fourier transform Raman spectroscopy.Journal of agricultural and food chemistry 1997, 45, 3088-3091.[307] Wallbacks, L., Edlund, U., Norden, B., Berglund, I., Multivariate characterization of pulpusing solid-state 13C NMR, FTIR, and NIR. Tappi journal 1991, 74.[308] Meder, R., Gallagher, S., Mackie, K. L., Bo¨hler, H., Meglen, R. R., Rapid determination ofthe chemical composition and density of Pinus radiata by PLS modelling of transmissionand diffuse reflectance FTIR spectra. Holzforschung 1999, 53, 261-266.[309] Schultz, T. P., Templeton, M. C., McGinnis, G. D., Rapid determination of lignocellu-lose by diffuse reflectance Fourier transform infrared spectrometry. Analytical chemistry.1985, 57, 2867-2869.[310] Rodd´s Brudenell River Resort, P., Island, E., Qualitative and Quantitative Analysis ofLignocellulosic Biomass using Infrared Spectroscopy. 2009.[311] Coimbra, M. A., Barros, A., Rutledge, D. N., Delgadillo, I., FTIR spectroscopy as a tool forthe analysis of olive pulp cell-wall polysaccharide extracts. Carbohydrate Research 1999,317, 145-154.[312] Ferraz, A., Baeza, J., Rodriguez, J., Freer, J., Estimating the chemical composition ofbiodegraded pine and eucalyptus wood by DRIFT spectroscopy and multivariate anal-ysis. Bioresource Technology 2000, 74, 201-212.[313] Backa, S., Brolin, A., Determination of pulp characteristics by diffuse reflectance FTIR.Tappi journal 1991, 74, 218-226.[314] Faix, O., Bo¨ttcher, J. H., Determination of phenolic hydroxyl group contents in milledwood lignins by FTIR spectroscopy applying partial least-squares (PLS) and principalcomponents regression (PCR). Holzforschung-International Journal of the Biology, Chem-istry, Physics and Technology of Wood 1993, 47, 45-49.[315] Malkavaara, P., Ale´n, R., A spectroscopic method for determining lignin content of soft-wood and hardwood kraft pulps. Chemometrics and Intelligent Laboratory Systems 1998,44, 287-292.138Bibliography[316] Jiang, J. H., Berry, R. J., Siesler, H. W., Ozaki, Y., Wavelength interval selection in mul-ticomponent spectral analysis by moving window partial least-squares regression withapplications to mid-infrared and near-infrared spectroscopic data. Analytical chemistry2002, 74(14), 3555-3565.[317] Centner, V., Massart, D. L., de Noord, O. E., de Jong, S., Vandeginste, B. M., Sterna, C.,Elimination of uninformative variables for multivariate calibration. Analytical chemistry1996, 68(21), 3851-3858.139

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:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0167681/manifest

Comment

Related Items