Variability in Raman Spectra of Single Human …agbrolo/cell_2010.pdf2010 Society for Applied...

Variability in Raman Spectra of Single Human Tumor CellsCultured in Vitro Correlation with Cell Cycle andCulture Confluency

QUINN MATTHEWS ANDREW JIRASEK JULIAN LUM XIAOBO DUANand ALEXANDRE G BROLODept of Physics amp Astronomy University of Victoria Victoria BC Canada (QM AJ) Deeley Research Centre BC Cancer Agency Victoria

BC Canada (JL XD) and Dept of Chemistry University of Victoria Victoria BC Canada (AGB)

In this work we investigate the capability of Raman microscopy (RM) to

detect inherent sources of biochemically based spectral variability

between single cells of a human tumor cell line (DU145) cultured in vitro

Principal component analysis (PCA) is used to identify differences in

single-cell Raman spectra These spectral differences correlate with (1)

cell cycle progression and (2) changing confluency of a cell culture during

the first 3 to 4 days after sub-culturing Cell cycle regulatory drugs are

used to synchronize the cell cycle progression of cell cultures and flow

cytometry is used to determine the cell cycle distribution of cell cultures at

the time of Raman analysis Spectral variability arising from cell cycle

progression is (1) expressed as varying intensities of protein and nucleic

acid features relative to lipid features (2) well correlated with known

biochemical changes in cells as they progress through the cell cycle and

(3) shown to be the most significant source of inherent spectral variability

between cells Furthermore the specific biomolecules responsible for the

observed spectral variability due to both cell cycle progression and

changes in cell culture confluency can be identified in the first and second

components of principal component analysis (PCA) Our characterization

of the inherent sources of variability in Raman spectra of single human

cells will be useful for understanding subtle spectral differences in RM

studies of single cells

Index Headings Raman microscopy Spectral variability Human cells

Cell biochemistry Cell cycle Confluency Principal component analysis

PCA

INTRODUCTION

In the last decade Raman microscopy (RM) has emerged asa prominent tool for the biochemical analysis of human cellsand tissues A recent study found that Raman spectroscopyprovides biochemical information at comparable levels ofaccuracy and sensitivity as established techniques such asNMR spectroscopy and flow cytometry1 RM is an attractivemodality both for its ability to attain biochemical informationfrom proteins nucleic acids lipids and carbohydrates in asingle acquisition and for its noninvasiveness and nondestruc-tiveness when used with an appropriate choice of laserwavelength and power2ndash4 Furthermore the use of high-powerfocusing objectives confocal optics and sub-micrometerresolution stepping stages enables the analysis of single cellsin vitro whereas fiber-optic-based Raman systems may in thefuture allow the technology to be applied to clinical patients invivo

There has been considerable interest in applying RM to livecells and tissues for cancer detection and diagnosis Due to thecomplexity of a single biological Raman spectrum RM is

commonly used in conjunction with multivariate statisticalmethods such as principal component analysis (PCA) lineardiscriminant analysis (LDA) or cluster analysis RM with theuse of multivariate methods has been successfully applied todiscriminate between healthy and cancerous skin5ndash7 bladder8

and gastric910 tissues Applying RM and multivariate methodsto the analysis of single cells has shown the ability todistinguish between healthy and tumorigenic rat fibroblastcells11 human bone cells12 and human epithelial cells from avariety of organs13 Similar techniques have been employed todiscriminate between different types of human tumor cell linesfrom a mixed sample set14ndash16

The single-cell studies mentioned above have focused on thedifferences in Raman spectra between different cell types (iehealthy vs tumor prostate vs bladder) In addition severalstudies have used RM for analyzing biochemical differencesarising within a population of a single type of cell One pair ofstudies detected spectral changes from cell death via apopto-sis17 and spectral differences between live and dead cells18 andanother pair of studies detected spectral changes due to celldeath via necrosis19 and spectral differences between expo-nentially growing (proliferative) and plateau phase (non-proliferative) cells20 The former pair of studies compared theaveraged Raman spectra of many single cells from eachsample and the latter pair of studies compared spatiallyaveraged spectra obtained from many cells at once from apellet of cells One recent study applied PCA and LDA toRaman spectra of single live cells to discriminate between cellssynchronized in the G0G1 S and G2M phases of the cellcycle and demonstrated accurate discrimination of G0G1 cellsfrom S and G2M cells21

Within a given population of cells one can expectbiochemical differences between individual cells due to anumber of reasons such as cells existing at different points intheir cell cycle (in the case of an asynchronously growing cellculture) or cells growing in cultures of different confluence (inthe case of multiple cell cultures) In all the studies mentionedabove natural biochemical differences exist between individ-ual cells in a sample In all these previous studies (with theexception of the recent cell cycle discrimination study21)spectral variability arising from inherent biochemical differ-ences was either averaged during spectral acquisition or post-processing or was not a relevant or necessary consideration forthe purpose of the study However the spectral differencesbetween cell populations may be very subtle and it isimportant to thoroughly investigate the existing sources ofspectral variability within a given population To ourknowledge there have been no systematic studies on the

Received 22 December 2009 accepted 12 May 2010 Author to whom correspondence should be sent E-mail jirasekuvicca

Volume 64 Number 8 2010 APPLIED SPECTROSCOPY 8710003-7028106408-0871$2000

2010 Society for Applied Spectroscopy

inherent sources of spectral variability that may arise due tobiochemical differences between single cells

In this work we have undertaken an investigation of thecapability of RM to detect inherent sources of spectralvariability within a human tumor cell line (DU145) culturedin vitro PCA is used to observe differences in Raman spectrathat correlate with cells existing in different phases of the cellcycle as well as differences correlating with the confluency ofthe cell culture at the time of Raman analysis Furthermore thebiochemical changes detected by RM that correlate with cellcycle progression are consistent with known biochemicalchanges within cells We also show that the variability fromcell cycle and culture confluency comprises almost all of theinherent variability in a multi-culture data set and is primarilyexplained by the first two PCA components

The results of this work are presented in two studies In thefirst study the inherent variability between cultures of varyingconfluence is investigated by acquiring Raman spectra of cellscollected from asynchronously growing cell cultures harvest-ed 1ndash8 days after sub-culturing In the second study thevariability due to cell cycle in sub-confluent cultures isdirectly examined by acquiring spectra from cells collectedfrom cultures synchronized at specific points in the cell cycleFlow cytometry is used to monitor the cell cycle distributionand viability of all cultures at the time of Raman analysis Weuse adherent cells that have been resuspended and centrifugedinto a pellet from which individual cells are selected using ahigh-power focusing objective and 785 nm laser excitationThis technique provides a very high quality Raman spectrumof a single cell while eliminating any spectral variability(caused by inconsistent cell structure and varying local celldensity) arising when cells are grown and analyzed directly ina monolayer22 Both low-wavenumber (LWN) 600ndash1800cm1 and high-wavenumber (HWN) 2700ndash3100 cm1spectral windows are used in order to determine whetherinformation can be attained equivalently from either windowas some authors have found to be the case for certainapplications23

METHODS

Cell Preparation Standard Cell Culture Human prostatetumor cells (cell line DU145 (ATCC Manassas VA)) werecultured in a sterile environment and grown in T-75 flasks with15 mL of Dulbeccos Modified Eagle Medium (HyCloneLogan UT) supplemented with 10 fetal bovine serum (FBS)(HyClone) Cultures were kept in an incubator at 5 CO2 and37 8C to promote growth Cell stocks were sub-cultured every3 to 4 days by rinsing the cells in phosphate buffered saline(PBS) adding trypsin (HyClone) to detach the cells from theflask and transferring 10ndash20 of the harvested cells to a newflask containing fresh media Before re-incubation the new cellsuspensions were pipetted several times to ensure an evenlydistributed monolayer of cells throughout the flask

Asynchronous Cell Cultures A single T-75 flask wasgrown to 90 confluency and sub-cultured equally into eightidentical T-75 flasks The first flask was harvested for RManalysis 24 hours after sub-culturing and each remaining flaskwas harvested every 24 hours thereafter Each time a flask washarvested the confluency cell cycle distribution and viabilityof the culture was measured as described below

Synchronized Cell Cultures Using established proto-cols2425 2 mM of thymidine (Sigma-Aldrich Oakville ON

Canada) was used to inhibit DNA replication and arrest cells atthe end of G1 phase before the onset of S phase and 100 ngmL of nocodazole (Sigma-Aldrich) was used to preventformation of the mitotic spindle and arrest cells at the end ofG2 phase prior to M phase In the second study presented inthis work cell cultures were treated with thymidine andnocodazole to obtain cultures synchronized at four pointsduring the cell cycle (1) at the G1S boundary (with double-thymidine24 treatment) (2) during S phase (with double-thymidine treatment and re-incubation with drug-free media for3 hours) (3) at the G2M boundary (with thymidine treatmentfor 24 hours re-incubation with drug-free media for 3 hoursnocodazole treatment for 12 hours and mitotic shake-off) and(4) during early G1 phase (with mitotic shake-off from a G2Marrested culture and re-incubation of detached cells with freshmedia for 5 hours)

Cell Cycle Viability and Confluency Analysis The cellcycle distribution and viability of each culture was measuredusing flow cytometry26 following established protocols2728

For cell cycle analysis over 200 000 cells were extracted froma culture and fixed in 70 ethanol to permeabilize the cellmembrane RNase A (Qiagen Inc Mississauga ON Canada)was added to a concentration of 1 mgmL in order to degradecellular RNA Propidium iodide (PI) (Sigma) was subsequentlyadded to a final concentration of 50 lgmL After 30 minutesthe suspension was centrifuged re-suspended in a buffer (PBSthorn 1 FBS) and kept on ice until analysis For flow cytometrycollection the relative DNA content of 100 000 cells wasmeasured using a BD FACSCalibur Flow Cytometer (BDBiosciences Mississauga ON Canada) Cell counts wererecorded by the flow cytometer during sample acquisitionRelative fractions of cells in each phase of the cell cycle (G1 Sor G2) were determined by performing a nonlinear least-squares fit to the measured data using functions representativeof the expected distributions of DNA content for each cellcycle phase (Matlab The Mathworks Natick MA)

For culture viability assessment over 100 000 cells wereextracted from the culture and split into two equal parts Onepart was stained with PI at a concentration of 5 lgmL and theother part was left untreated to serve as a control 20 000 cellsfrom each sample underwent flow cytometry analysis within 15minutes of staining and the fractions of live (no PI signal) anddead (positive PI signal) cells in the PI stained sample weredetermined

Cell culture confluency is defined as the percentage of thesurface area of the culture flask covered by cells Confluencyestimates of each of the asynchronous cell cultures wereobtained prior to harvesting by acquiring low magnificationimages of five different regions of the cell culture Each imagewas imported into Matlab and the fraction of the image coveredby cells was calculated Because the confluency is neverconsistent throughout the entire surface area of the flask theaveraged confluency from the five regions was used as anestimate of the overall confluency of the culture

Raman Microscopy To prepare cells for Raman analysiscultures were harvested by rinsing with PBS to remove deadcells and debris adding trypsin to detach the remaining livecells and centrifuging to discard the trypsin supernatant Cellswere re-suspended in growth media and centrifuged into apellet in a 200 lL plastic vial Vials were kept on ice until RManalysis (1ndash6 hours) upon which the chosen pellet wastransferred to a low-fluorescence quartz disc (Technical Glass

872 Volume 64 Number 8 2010

Products Painesville OH) in order to minimize spectralcontributions from the sample substrate All Raman spectrawere acquired within 2 hours of transferring the pellet to thequartz disc We have observed no spectral variations thatcorrelate with the time of sample removal from the ice bathsuggesting that any effect of removing cells from the ice bath toan exposed environment has negligible impact on our RManalysis within this two-hour time interval

Raman analysis was performed on an InVia RamanMicroscope (Renishaw Inc Hoffman Estates IL) with a1003 dry objective (NAfrac14 09) (Leica Microsystems WetzlarGermany) and a 1200 linesmm diffraction grating A 785 nmcontinuous wave diode laser (Renishaw) was used for sampleexcitation providing a laser power density at the sample of05 mWlm3 The size of the sampling volume wasmeasured to be 2 3 5 3 10 lm these dimensions allow asingle acquisition to represent the spectrum of a single cellRaman spectra were acquired from 20 individual cells fromeach sample chosen from the top layer of the cell pellet (Fig1) Spectra were collected at 30-second acquisitions per cellthe LWN window (600ndash1800 cm1) and the HWN window(2700ndash3100 cm1) spectra were acquired in succession foreach cell The wavenumber range for each spectral windowwas covered in a single acquisition using the RenishawrsquosSynchroScan operation mode

Spectral Processing Prior to PCA analysis each spectrumwas processed to remove cosmic rays increase the signal-to-noise ratio via spectral smoothing subtract a baseline arisingfrom the quartz substrate and biological fluorescence andnormalize to the amount of biological material within thesampling volume All data processing was performed withMatlab

Spectral smoothing was performed with an in-house versionof the previously described two-point maximum entropymethod2930 which has been particularly successful whenapplied to Raman spectra3132 In this work a very modestamount of smoothing was applied in order to maintain fidelityof the sharpest Raman peaks in the spectra

The large number of spectra collected in this studynecessitated the use of automated baseline removal methods

An effective and robust baseline removal method is critical forthis work in order to remove sources of variability arising fromvarying levels of fluorescence or quartz substrate contamina-tion (addressed further in the Discussion section) For the LWNspectral window we used a modified signal removal method33

This method was chosen due to the mixture of sharp and broadfeatures throughout the spectral window and the need for ahighly conformal baseline around the regions of quartzcontamination (800 cm1 and 1050 cm1) and a broaderbaseline from 1100 to 1800 cm1 where many overlappingpeaks give rise to broad Raman features For the HWN spectralwindow a three-point linear interpolated baseline was found tobe sufficient for baseline removal

After baseline removal the principal remaining source ofvariability between spectra is the overall intensity of the Ramanfeatures arising from the variable amount of biologicalmaterial within the sampling volume In the present work thisvariability arises from slightly different physical shapes andorientations of each cell in the cell pellet To remove thissource of variability each spectrum was normalized to the totalarea under the baseline-corrected spectrum Other authors haveaddressed this issue of intensity variability by normalizing tothe area under the CH deformation peak at 1450 cm1 thoughtto be proportional to the total amount of biological materialwithin the sampling volume1718 In our work we have foundthat the CH deformation peak can vary independently of otherRaman peaks and therefore may not be suitable as anormalization peak in all cases For example in this studywe report that one of the most significantly varying Ramanpeaks between cells arises from CH2 deformation in lipids at1438 cm1 which affects the area of the CH deformation peakas well due to its close spectral proximity Furthermore wehave found that the method of normalizing to the total areaunder the baseline-corrected spectrum is suitable for both theLWN and HWN spectral windows

Principal component analysis was performed using standardalgorithms (Matlab) PCA calculations were performed sepa-rately on the LWN and HWN window data sets to facilitate anindependent comparison and corroboration of results obtainedfrom each window In this work spectral variability arisingfrom the quartz substrate was easily identifiable in a singlePCA component from the LWN window the quartz compo-nent was therefore removed and the PCA calculation wasrepeated on the filtered set of spectra It is important to note thatthis action does not affect the other LWN window componentsbut only redistributes the variance explained by the excludedcomponent among the remaining components

RESULTS

Single DU145 Cell Spectrum In the LWN spectralwindow the Raman spectrum of a single DU145 cell (Fig2a) contains multiple contributions from proteins lipids andnucleic acids Spectral features of proteins arise from aromaticamino acids (phenylalanine tryptophan and tyrosine) amidegroups of secondary protein structures (a-helices b-sheets andrandom coils) and the stretching or deformation of carbonatoms bonded with nitrogen hydrogen or other carbon atomsNucleic acid features include contributions from individualRNA and DNA bases (adenine thymine guanine cytosineand uracil) as well as from the sugar-phosphate backbone ofDNA A number of different lipid features are also detectablethroughout the spectral window In the HWN window the

FIG 1 Optical image of a portion of a cell pellet acquired with a 1003objective The 785 nm focused laser spot is shown relative to the selected cellwhich is highlighted by the white circle

APPLIED SPECTROSCOPY 873

spectrum (Fig 2b) is a superposition of broad featuresdominated by the stretching of various lipid and protein CH2

and CH3 groups There are also weak contributions fromfrac14CHstretching in lipids and from aromatic groups in both nucleicand amino acids A detailed listing of the molecularassignments111834ndash38 for all spectral features observed in thiswork is provided in Table I

Study 1 Asynchronous Cell Cultures Cell CycleConfluency and Viability The cell cycle distributions andculture confluencies for the eight samples in this study (Fig 3)are typical for asynchronous cells growing to confluency inculture From 24 to 72 hours after sub-culturing the distributionamong the three phases is fairly constant at 50 G1 20G2 and 30 S Between 72 and 96 hours we see an increasein the G1 phase and a decrease of both the G2 and S phasesAfter 96 hours the G2 content remains relatively constantwhereas the S content decreases and the G1 content increasesuntil about 168 hours One element that is not measurable withthis method of cell cycle analysis is the fraction of cells inlsquolsquoG0rsquorsquo phase a state of cellular quiescence Cellular quiescenceis only achievable during G1 phase usually soon after celldivision therefore G0 cells are indistinguishable from G1 cellsby the flow cytometry methods used in this study

The viability of the harvested cells was determined with flowcytometry prior to Raman analysis Dead cells will usuallydetach from the growth substrate and subsequently be rinsedoff and discarded during the harvesting procedure However asmall percentage of dead cells will always remain in aharvested culture For this study viability tests proved thatall of the first seven samples (24ndash168 hours after sub-culturing)had a viability of 98 (ie less than 2 dead cells) and theeighth and final sample (192 hours after sub-culturing) had aviability of 95

First Principal Component For the 160 cell spectracollected in this study the first PCA components (Fig 4)represent the most significant source of spectral variability ineach data set (526 of the total variance for the LWN

window 886 for the HWN window) By comparison withthe known Raman shifts (Fig 2 Table I) the features in thePCA components for both the LWN and HWN window areidentifiable as arising from variability in the Raman intensity ofpeaks in the original data set therefore one can assign amolecular origin to the features in the components The PCAcomponents consist of both positive and negative features anyspectrum that is assigned a higher (ie more positive) PCAscore for a given component will have a proportionately higheramount of the positive features and a lower amount of thenegative features from that component It should be noted thatthe positive or negative nature of the features is purely arbitraryand only holds meaning with respect to the sign of thecorresponding PCA scores Any component can be reflectedabout zero with a corresponding change of sign for all scoresfor that component without altering the results of the PCAtransformation

The negative features in the first PCA component for theLWN window (Fig 4a) are dominated by lipid contributionsfrom cholesterol CH2 twisting CH2 and CH deformation andCndashC Cfrac14C and Cfrac14O stretching The CndashC features at 10651080 and 1127 cm1 the CH feature at 1460 cm1 and theCfrac14C feature at 1656 cm1 have the same Raman shifts as theprotein contributions from CndashN stretching CH deformationand a-helix amide groups respectively As such it isimpossible to determine whether or not this variability issolely lipid or protein in nature or some combination of bothhowever as the rest of the negative features in the componentare uniquely lipid in origin it follows that the negative natureof these features arises in part from lipids as well The negativefeature at 844 cm1 was not identifiable The positive featuresin the LWN component are almost exclusively nucleic acid andprotein in origin with the exception of a weak positivecontribution from choline at 719 cm1 Nucleic acid featuresarise from DNA and RNA bases and from the DNA backboneProtein features arise from aromatic amino acids (phenylala-nine tryptophan and tyrosine) and from b-sheet amide groups

FIG 2 Raman spectra of a single DU145 cell for the (a) LWN and (b) HWN spectral windows the Raman shift and molecular origin of identifiable features areprovided111834ndash38 Abbreviations (p) protein (l) lipid (d) DNARNA (A) adenine (T) thymine (G) guanine (C) cytosine (U) uracil (Phe) phenylalanine (Tyr)tyrosine (Trp) tryptophan (bk) backbone (def) deformation (tw) twist (sym) symmetric (asym) asymmetric and (str) stretch


Interestingly it is known that the aromatic amino acids aremost likely to be found in a b-sheet conformation and lesslikely to be found in an a-helix or coiled structure39 As suchan increase in signal from the aromatic amino acids should becorrelated with an increase in signal from b-sheet amidegroups which we see to be the case here For the HWNwindow (Fig 4b) the positive features can be assigned to thesymmetric and asymmetric stretching of CH3 groups in bothproteins and lipids The negative features however arise fromthe symmetric and asymmetric stretching of CH2 groups inlipids alone To summarize the negative features in both the

LWN and the HWN window are primarily due to lipidswhereas the positive features in both the LWN and the HWNwindow are primarily due to nucleic acids and proteins (inparticular from amino acids and b-sheet amides for the LWNwindow)

The PCA scores (Fig 5) determine how much of thevariability explained by the first components (Fig 4) isexpressed in each of the 160 cell spectra in each data setNote that positive scores are correlated with increased nucleicacid and protein content and negative scores are correlatedwith increased lipid content The scores for both windows

TABLE I Molecular assignments for spectra of DU145 cells Superscript numbers indicate references used for particular assignmenta

Raman shift (cm1)

Molecular assignment

DNARNA Proteins Lipids

622 CndashC tw Phe3435

644 CndashC tw Tyr3435

669 G T3435

700 Cholesterol35

719 Choline343538

728 A34ndash36

759 Trp ring br3435

784 U C T ring br34ndash36

OndashPndashO str bk3436

811 OndashPndashO str RNA34 OndashPndashO str38

828 OndashPndashO asym str3436 Tyr ring br34ndash37

853 Tyr ring br34ndash37

877 Acyl C2ndashC1 str3438

936 CndashC sym str bk a-helix1134ndash37

975 Head CndashC str3438

1003 Phe sym ring br1134ndash37

1032 CndashH Phe113436

1065 CndashN str3436 Chain CndashC str1134353738


1094 PO2 bk B-from1134ndash36 CndashN str1136 Chain CndashC str343738

1100 PO2 bk A-from3537


1158 CndashC CndashN str34

1175 CndashH Tyr Phe113436

1208 CndashC6H6 str Phe Trp Tyr1134ndash37

1230 Amide III rand coil18

1246 Amide III b-sheet1837

1255 Amide III b-sheet a-helix111835ndash37

1267 Amide III a-helix1837 frac14CH def1838

1300 CH2 tw1134353738

1320 C1134 CH def1834

1340 A G1134ndash36 CH def1834

1374 T113536

1421 A G1136 CH2 bk35

1438 CH2 def11353738

1450 CH def111834ndash37 CH def1834

1460 CH def1834 CH def1834

1486 A G3536

1577 A G1134ndash36

1607 Cfrac14C Phe Tyr34

1618 Cfrac14C Tyr Trp34

1656 Amide I a-helix1837 Cfrac14C str1118353738

1660 Amide I1134ndash36

1669 Amide I rand coil37

1685 Amide I b-sheet37

1743 Cfrac14O str34353738

2853 CH2 sym str3435ndash38

2888 CH2 asym str3435ndash38

2935 CH3 sym str1137 CH3 sym str113538

2960ndash2980 CH3 asym str1137 CH3 asym str11353738

3008 frac14CH str3738

3060 Aromatics37 Aromatics37

a Abbreviations (A) adenine (U) uracil (C) cytosine (T) thymine (G) guanine (Trp) tryptophan (Tyr) tyrosine (Phe) phenylalanine (br) breathing (bk) backbone(def) deformation (tw) twist (sym) symmetric (asym) asymmetric and (str) stretch


show the same overall trend between 24 and 48 hours aftersub-culturing there is an increase in the average nucleic acidand protein content relative to the average lipid contentfollowed by a steady decrease in the average nucleic acid and

protein content relative to the average lipid content from 48 to192 hours after sub-culturing Furthermore the relativepositions of the individual cell scores are consistent betweenthe LWN and HWN windows For example cells 30 and 39

FIG 3 Flow cytometry analysis of cell cycle distributions for the asynchronous cell cultures Time indicates the incubation time of the culture after sub-culturingCulture confluency (Conf) and cell cycle phase fractions were calculated as described in the Materials and Methods section

FIG 4 First PCA components from the asynchronous cell cultures study (a) LWN window (526 of total variance) (b) HWN window (886 of total variance)The Raman shift and molecular origin of identifiable features are provided111834ndash38 Abbreviations (p) protein (l) lipid (d) DNARNA (A) adenine (T) thymine(G) guanine (C) cytosine (U) uracil (Phe) phenylalanine (Tyr) tyrosine (Trp) tryptophan (bk) backbone (def) deformation (tw) twist (sym) symmetric (asym)asymmetric and (str) stretch


(Fig 5) have respectively the highest and lowest scores from

the 48-hour LWN window sample set and the same two cells

have respectively the highest and lowest scores from the

corresponding HWN window sample set It is worth empha-

sizing that the intra-sample variability in the PCA scores for a

given PCA component arises from the same source of spectral

variability as the inter-sample variability and simply reflects the

intrinsic biochemical heterogeneity of each cell culture

To show that the variability predicted by the PCA analysis is

directly observable in the original data the Raman and

difference spectra for two cells (cells 30 and 148) having a

large separation in their PCA scores (Fig 5) are shown in Fig

6 along with the PCA components for comparison All of the

major features in the components are directly observable in the

corresponding difference spectrum for each spectral window

without any rescaling of the difference spectra

Second Principal Component The second PCA component

for the LWN window (Fig 7a) explains 101 of the total

variance and the corresponding component for the HWN

window (Fig 7b) explains only 17 of the total variance

Assigning a molecular origin to the features in the second

components is more difficult than for the first components

FIG 5 PCA scores for the first components from the asynchronous cell cultures study for the (a) LWN and (b) HWN window Scores for all 160 cells are groupedby time of harvest after sub-culturing The average score and standard deviation are shown for each sample for visualization of the trends in the data The Ramanspectra of cell 30 and cell 148 are shown in Fig 6

FIG 6 Raman and difference spectra for two cells (30 and 148) having a large difference in PCA score (Fig 5) for the first PCA component The first PCAcomponents have been offset and rescaled for comparison with the unscaled difference spectra Wavenumbers are provided for any known features in thecomponents (Fig 4) that are also observable in the difference spectra


especially for the HWN window where the percent variance

explained is very low and there is a small number of known

molecules contributing to the HWN spectra (Fig 2b Table I)

The only feature in the HWN window that corresponds with a

known wavenumber is the symmetric stretching of CH3 groups

at 2935 cm1 (Fig 7b) although the accuracy of this

assignment is uncertain However for the LWN window

(Fig 7a) almost all of the major features can be assigned with

confidence The positive features arise from amino acids

amide groups in b-sheet and random coil conformation and a

combined contribution from the nucleic acid bases A and G

and CH deformation in proteins The origin of the positive

feature at 1120 cm1 is unknown The negative features

include a strong contribution from choline as well as

contributions from OndashPndashO stretching in lipids and RNA the

nucleic acid bases A and G and a combined contribution from

lipid frac14CH deformation and a-helix amide groups The sharp

negative feature at 1660 cm1 arises from amide groups as

well but whether it arises from a certain protein conformation

or from amide groups in general is unknown It is also unclear

as to why contributions from the nucleic acids A and G appear

in both the positive and negative features of the component

Despite the uncertainty of the molecular origins of the

features in the second PCA components (especially for the

FIG 7 Second PCA components from the asynchronous cell cultures study (a) LWN window (101 of total variance) (b) HWN window (17 of totalvariance) The Raman shift and molecular origin of identifiable features are provided111834ndash38 Abbreviations (p) protein (l) lipid (d) DNARNA (A) adenine (G)guanine (Phe) phenylalanine (Tyr) tyrosine (def) deformation (sym) symmetric and (str) stretch

FIG 8 PCA scores for the second components from the asynchronous cell cultures study for the (a) LWN and (b) HWN window Scores for all 160 cells aregrouped by time of harvest after sub-culturing The average score and standard deviation is shown for each sample for visualization of the trends in the data TheRaman spectra of cells indicated by arrows are shown in Fig 9


HWN window) the scores for both windows still show thesame general trend from 24 to 120 hours after sub-culturingthere is an overall increase in the average scores and after120 hours the average scores appear to remain relativelyconstant until decreasing slightly between 168 and 192 hoursHowever the relative positions of the individual cell scoresbetween the LWN and HWN windows are not consistentTherefore the similar trends between the two windows maynot be the result of the same biomolecular changes occurringwithin the cells

To determine whether the variability predicted by the secondPCA components is directly observable in the original data (asit was with the first PCA components (Fig 6)) the Raman anddifference spectra for two cells (cells 137 and 19 for theLWN window and cells 114 and 31 for the HWN window)having a large separation in their PCA scores (Fig 8) areshown in Fig 9 along with the PCA components forcomparison For the LWN window all of the major featuresin the component are observable in the LWN differencespectrum However the features in the HWN component arenot observable in the HWN difference spectrum

Other Principal Components The third PCA componentfor the LWN window (not shown) explains 76 of the totalvariance and is dominated by a sharp derivative-like featurecentered at the wavenumber of the sharp phenylalanine ringbreathing peak at 1003 cm1 This feature in the PCAcomponent indicates variability arising from a shift in thecalibration of the Raman system over time The trend displayedby the score plots for this component (also not shown)correlates well with a known drift in the Raman calibrationover the eight-day sample collection period which wasmonitored by measuring peak shifts of the 520 cm1 featureof an instrument-based silicon sample before and after the dailyRaman collection During daily collections it was verified thatthe initial calibration of the system was within 05 cm1 of thecalibration performed on the first day of collection By

inspection of various pairs of spectra with large differencesin their scores for the third PCA component it was found thatthe maximum shift in the position of the phenylalanine peak at1003 cm1 was less than one pixel (1 pixel rsquo 09 cm1 at1003 cm1) for all 160 spectra collected during the eight daysof data collection

For spectra with the most outlying scores for the third PCAcomponent corrections to the shift were attempted using linearinterpolation and were successful in reducing by a few percentthe total amount of variance explained by the third PCAcomponent However due to the sharpness of many peaks inthe LWN spectrum there will always be slight shifts measuredin the peak positions due to experimental limitations whichwill translate into variability brought out in the PCA analysisIn this study the third PCA component is the last componentfor the LWN window that displayed any measurable trend inthe score plots furthermore each of the remaining 156components explain less than 3 of the total variance andlikely have little to no biological significance The same can besaid for the remaining 157 PCA components for the HWNwindow each of which explains less than 1 of the totalvariance The remaining PCA components will account forresidual variance arising from random sources of spectralvariability such as organelle positioning within cells orinstrument noise

Study 2 Synchronized Cell Cultures Cell CycleSynchronization Cell cultures were synchronized at fourdifferent points in the cell cycle at the G1S boundary at 3hours into S phase at the G2M boundary and at early G1phase For the first culture 83 of the cells weresuccessfully arrested either in late G1 or early S phase (Fig10a lsquolsquoG1Srsquorsquo) Three hours after release from an identical G1Sarrest 19 of the second culture remained in G1 phasewhereas 64 of the culture was measured to be in S phase(Fig 10b lsquolsquoG1S thorn3 hrsrsquorsquo) For the third culture a distinct G1peak was not observed after G2M synchronization Therefore

FIG 9 Raman and difference spectra for two cells (137 and 19 for the LWN window and 114 and 31 for the HWN window) having a large difference in PCAscore (Fig 8) for the second PCA component The second PCA components have been offset and rescaled for comparison with the unscaled difference spectraWavenumbers are provided for any known features in the components (Fig 7) that are also observable in the difference spectra


the combined fraction of cells in G1 or S phase was estimatedto be 26 with at least 74 of the cells successfully arrestedat the G2M boundary (Fig 10c lsquolsquoG2Mrsquorsquo) Five hours afterharvesting and re-incubating cells from an identical G2Marrest 21 of the fourth culture was determined to be left inG2 phase while 75 of the culture was now found in G1phase (Fig 10d lsquolsquoG2Mthorn5 hrsrsquorsquo) Since the fourth culture wasseeded with cells that were primarily at the G2M boundarythe G1 cells in the fourth culture must be less than 4 to 5 hoursinto G1 phase

First Principal Component The first PCA component forthe LWN window (Fig 11a) explains 516 of the totalvariance and is very similar to the corresponding componentfrom the asynchronous cell cultures study (Fig 4a) whichexplained 526 of the total variance As in the previous studythe negative features in the component are dominated by lipidcontributions from cholesterol CH2 twisting CH2 and CH

deformation and CndashC Cfrac14C and Cfrac14O stretching with anadditional negative contribution from choline which previous-ly contributed as a weak positive feature in the asynchronousstudy There is also a new negative feature at 1267 cm1which is a combined contribution from lipidfrac14CH deformationand a-helix amide groups this feature correlates with theexisting negative combined contribution from lipid Cfrac14Cstretching and a-helix amides at 1656 cm1 The previouslyobserved negative features at 844 and 1127 cm1 are notobserved here The positive features in the LWN component asin the previous study are exclusively nucleic acid and proteinin origin with contributions from DNA and RNA bases theDNA backbone aromatic amino acids and b-sheet amidegroups In this study there are additional positive contributionsfrom tyrosine at 853 cm1 thymine at 1374 cm1 and randomcoil amide groups at 1230 cm1 The previously observedpositive feature at 811 cm1 is not observed here The first

FIG 10 Flow cytometry analysis of cell cycle distributions for the synchronized cell cultures Synchronization was performed using thymidine and nocodazole asdescribed in the Materials and Methods section

FIG 11 First PCA components from the synchronized cell cultures study (a) LWN window (516 of total variance) (b) HWN window (866 of total variance)The Raman shift and molecular origin of identifiable features are provided1118ndash34-38 Abbreviations (p) protein (l) lipid (d) DNARNA (A) adenine (T) thymine(G) guanine (C) cytosine (U) uracil (Phe) phenylalanine (Tyr) tyrosine (Trp) tryptophan (bk) backbone (def) deformation (tw) twist (sym) symmetric (asym)asymmetric and (str) stretch


PCA component for the HWN window (Fig 11b) explains866 of the total variance and is nearly identical to thecorresponding component from the asynchronous cell culturesstudy (Fig 4b) which explained 886 of the total varianceAs before the positive features arise from the symmetric andasymmetric stretching of CH3 groups in both proteins andlipids whereas the negative features arise from the symmetricand asymmetric stretching of CH2 groups in lipids alone

The PCA scores for the first components (Fig 12) show thesame trend for both the LWN and the HWN window Betweenthe G1S culture and the S-phase culture there is a slightincrease in the average nucleic acid and protein content relative

to the average lipid content There is no observable difference

in the average scores between the S-phase culture and the G2

M culture However between the G2M culture and the early

G1-phase culture there is a decrease in the average nucleic acid

and protein content relative to the average lipid content As was

the case for the PCA scores for the first components from the

asynchronous study (Fig 5) the relative positions of the

individual cell scores are consistent between the LWN and

HWN windows For example cells 63 and 75 have

respectively the highest and lowest scores from the lsquolsquoG2M

thorn5 hrsrsquorsquo LWN window sample set and the same two cells have

FIG 12 PCA scores for the first components from the synchronized cell cultures study for the (a) LWN and (b) HWN window The average score and standarddeviation are shown for each sample for visualization purposes

FIG 13 Second PCA components from the synchronized cell cultures study (a) LWN window (77 of total variance) (b) HWN window (21 of total variance)The Raman shift and molecular origin of identifiable features are provided111834ndash38 Abbreviations (p) protein (l) lipid (d) DNARNA (A) adenine (T) thymine(G) guanine (C) cytosine (U) uracil (Phe) phenylalanine (Tyr) tyrosine (Trp) tryptophan (bk) backbone (def) deformation (tw) twist (sym) symmetric (asym)asymmetric and (str) stretch


respectively the highest and lowest scores from the corre-sponding HWN window sample set (Fig 12)

Second Principal Component The second component forthe LWN window (Fig 13a) explains 77 of the totalvariance and the corresponding component for the HWNwindow (Fig 13b) explains only 21 of the total varianceNeither the LWN nor HWN window components have anysimilarity to the second components from the asynchronouscell cultures study (Fig 7) For the LWN window all featuresare easily identifiable except for the feature at 1402 cm1 Thenegative features include multiple contributions from thearomatic amino acids with additional contributions fromcholine and OndashPndashO stretching in nucleic acids The positivefeatures are made up of contributions from nucleic acid basesand the DNA backbone a-helix and b-sheet amide groups inproteins and CH2 twisting Cfrac14C stretching and both CH2 andfrac14CH deformation in lipids For the HWN window two broadnegative features are observed which possibly arise from theasymmetric stretching of CH2 groups in lipids and thesymmetric stretching of CH3 groups in proteins and lipids

The PCA scores for the LWN window (Fig 14a) show adistinct increase in the average score for the G2M culture Thisincrease is correlated with increased amounts of nucleic acidbases DNA conformational proteins and CH2 and Cfrac14Cgroups in lipids and decreased amounts of aromatic aminoacids choline and OndashPndashO groups in nucleic acids The scoresfor the HWN window do not have any relationship to the LWNwindow scores and do not appear to provide much meaningfulbiochemical information except that the highest scores aremostly observed in the early G1-phase culture

Other Principal Components The third PCA componentfor the LWN window (not shown) explains 51 of the totalvariance Some features in this component are similar to thosein the second PCA component from the asynchronous cellcultures study (Fig 7a) including a strong negative contribu-tion from choline at 719 cm1 and a positive contribution fromphenylalanine at 1003 cm1 However the PCA scores for thiscomponent (also not shown) do not show any significant trend

or discrimination between samples The fourth and fifthcomponents show features representative of slight x-axiscalibration shifts but as all the spectra in this study werecollected in a single day the system calibration was veryconsistent for all samples as such each component explainsonly 3 of the total variance Each of the remainingcomponents for the LWN window explain less than 2 of thetotal variance and likely have little to no biological significanceand account for any residual variance arising from randomsources of variability The same can be said for all theremaining components for the HWN window each of whichexplains less than 1 of the total variance

DISCUSSION

Study 1 Asynchronous Cell Cultures The results of this8-day study show that when Raman spectra are acquired fromsingle DU145 cells taken from multiple cell cultures overmultiple days with different times between sub-culturing andRaman acquisition for each culture there are primarily twoindependent sources of inherent variability observed in theRaman spectra These two sources of variability are represent-ed in this study by the first and second PCA components (Figs4 and 7)

First Principal Component For the entire 8-day data set inthis study the first PCA component explains 526 of the totalvariance for the LWN window data set When searching for abiological origin for this component an important consider-ation is that no matter which subset of the total 8-day data set isinput into PCA this same component is always observed as theprimary source of variability and typically explains 35 to 60of the total variance For example if the data for only the firstfour days is input into PCA the variance explained is 373however if only the data for the last four days is used thevariance explained is 513 No matter how many days worthof data are input into PCA or which days are chosen theprimary features of the component do not change namely thepositive features arise from the same nucleic acid and proteinmolecules and the negative features arise from the same lipid

FIG 14 PCA scores for the second components from the synchronized cell cultures study for the (a) LWN and (b) HWN window The average score and standarddeviation are shown for each sample for visualization purposes


molecules as those assigned to the first component in this study(Fig 4a) These properties of the LWN component are also truefor the HWN component (Fig 4b) except that the percentvariance explained is typically 75 to 90 of the total varianceIt is also important to note that over the course of previousexperiments (not shown) we have collected Raman spectrafrom thousands of single cells No matter which subset ofpreviously collected data is input into PCA the first PCAcomponents presented in this study (Fig 4) are reproduced asthe primary source of variability

The most likely biological origin for the source of variabilityexpressed by the first PCA components is the biochemicalvariability due to cell cycle (examined further below in thediscussion of study 2) In this study the PCA scores for thefirst PCA component for both the LWN and the HWNwindow show the trend of a steady decrease in the averagecellular nucleic acid and protein content relative to the averagelipid content from 48 to 192 hours after sub-culturing (Fig 5)Furthermore there is a definite correlation between the steadilyincreasing fraction of cells in G1 phase as measured by flowcytometry (Fig 3) and the steady decrease in the nucleic acidand protein content of individual cells relative to the lipidcontent as measured by RM and calculated by PCA (Fig 5)Interestingly the flow cytometry results show that the fractionof cells in G1 phase begins to stop steadily increasing around120 to 144 hours after sub-culturing whereas the RM and PCAanalyses show that the relative nucleic acid and protein contentcontinues to decrease steadily from 120 to 192 hours (Fig 5)This discrepancy is likely in part due to a continual increase inthe fraction of G1 cells that have entered into a non-proliferating quiescent lsquolsquoG0rsquorsquo phase from 120 to 192 hourswhich is not detectable by the methods used in this study It isknown that quiescent cells have a much lower RNA contentthan actively cycling G1 cells as well as a decreased amount ofcertain proteins required for cell cycle progression4041 Thediscrepancy is also likely in part due to the flow cytometryobservation that the fraction of cells in S phase continues todecrease from 120 to 192 hours and reaches a minimum of8 at 192 hours after sub-culturing However an increasedfraction of quiescent cells and a decreased fraction of S-phasecells are both indicators of a less proliferative cell culturewhich is an expected trend as cells are left for longer periods oftime after sub-culturing

It is important to note that the results of this study are inagreement with two previous Raman studies2042 both ofwhich compared the average bulk Raman spectra of exponen-tially growing cells (G1 fraction 50) to plateau-phase cells(G1 fraction 80) One of these studies20 found that theproteinlipid RNAlipid and DNAlipid ratios were allstatistically higher for exponentially growing cells as deter-mined by fitting biochemical component spectra to themeasured LWN and HWN window averaged Raman spectraThis same study also identified the spectral regions (and thecorresponding molecules assigned to those regions) thatyielded significant averaged spectral differences betweensamples With similar methods the other study42 demonstratedthat increased fractions of both protein and nucleic acid contentin exponentially proliferating cells were correlated withdecreased fractions of lipid and glycogen content as comparedto plateau-phase cells The results presented here on RM ofsingle cells corroborate and extend these previous Ramanresults for bulk samples Our study identifies which individual

molecular sub-groups are most responsible for the observedchanges in Raman spectra such as the strong contribution fromCH2 deformation in lipids in the first PCA component for theLWN window (Fig 4a) Our PCA analysis also demonstratesthat changes in the relative lipid content in a cell aremathematically anti-correlated with changes in both the proteinand nucleic acid content in a cell this result is consistent bothwith previous Raman results42 and with the prior knowledgethat the RNA-to-protein ratio is relatively constant within a cellthroughout the cell cycle40 Our results extend previous Ramanstudies by showing that the changes in biochemical composi-tion due to cell cycle can be directly observed in single cellspectra (ie Fig 6) and that the changes can be readilyobserved as a continuous process as a cell culture moves froman exponentially growing culture (24 to 96 hours after sub-culturing) to a confluent non-exponential culture (120 to 192hours after sub-culturing) Finally as discussed below ourstudy shows that there is another significant source ofvariability (arising from cell culture confluency) that isdetectable when performing RM on single cultured cells inaddition to the variability in the nucleic acid and proteincontent relative to the lipid content

Second Principal Component For the entire 8-day data setin this study the second PCA component explains 101 ofthe total variance for the LWN window data set Howeverunlike the first PCA component the amount of varianceexplained by this component is highly dependent on whichsubsets of the total data set are input into PCA For examplethe variance explained is maximized at 167 when only thedata for the first five days is input into PCA However if thedata for the first two days are excluded the variance explaineddrops from 101 to 48 and if the first three days areexcluded the variance explained drops further to 24 Whenthe first four days or more are excluded the varianceexplained becomes less than 2 and the component is nolonger recognizable These properties of the LWN componentare also true for the HWN component (Fig 7b) in the HWNcase the percent variance explained is maximized at 33 whenonly the data for the first five days is input into PCA yet thecomponent is not observed when the first four or more days areexcluded as was the case for the LWN component Thedependency of the second PCA component on the choice ofsample subset is consistent with the corresponding PCA scores(Fig 8) which steadily increase up to five days after sub-culturing and remain fairly constant from five to eight daysafter sub-culturing

A definitive biological origin for the second PCA componentis unclear especially for the HWN window where themolecular origin of the features is unknown (Fig 7b)However there is a strong correlation between the trend ofthe PCA scores (Fig 8) and the measured confluency of the cellcultures (Fig 3) which is in turn related to the amount of timethe culture was left to incubate after sub-culturing In thisstudy the cell cultures steadily increase their confluency untilapproximately five days after sub-culturing after which there isvery little room left to grow and the confluency remainsrelatively constant at 90 The confluency trend matches thetrend of the PCA scores which steadily increase up to five daysafter sub-culturing and remain fairly constant afterwardsFurthermore as discussed above if only the data from days5 to 8 is input into PCA (ie only the data collected once theculture had reached 90 confluency) then the second PCA


component is not observed at all This implies that thecomponent is directly caused by biochemical changes related toa sub-confluent culture growing during the first three to fourdays after sub-culturing The primary biomolecules responsiblefor this source of variability in the Raman spectra can beidentified in the second PCA component for the LWN window(Fig 7a) To the authorrsquos knowledge this study is the firstspectroscopic measurement of inherent biochemical variabilityin a cell culture that is correlated with the changing confluencyof a cell culture during the first three to four days after sub-culturing which is shown here to be independent of theexisting variability arising from cell cycle progression Wehave observed this source of variability in several previousexperiments with DU145 cells (not shown) in which cellcultures were harvested for Raman analysis one to two daysafter sub-culturing It should be noted that although this sourceof variability is shown here to be significant for DU145 cells itmay not be a characteristic of all in vitro cultured cell lines

Study 2 Synchronized Cell Cultures In theory theeffects of the cell cycle regulatory drugs thymidine andnocodazole are reversible such that when the drug is removedand replaced by fresh media the cells in the culture willprogress synchronously through the rest of their mitotic cycleIn practice whole culture synchronization is unfeasible and cellcultures become desynchronized very quickly4344 Further-more a certain fraction of the culture will not be immediatelyreleased (or released at all) from the drug-induced arrestHowever large fractions of cells (ie 75) can indeed besynchronized for short periods of time (typically less than 24hours) and the cell cycle distribution can indeed besignificantly altered from that of an untreated asynchronouspopulation Despite the known shortcomings of using drugs forcell cycle control drug treatment is still the easiest and simplesttechnique for significantly altering the cell cycle distributionand achieving a high yield of mostly synchronized cells It isimportant to note that the level of synchronization reported inthis study (Fig 10) is comparable to the level of synchroni-zation reported by both the recent study using RM for cellcycle discrimination where synchronization was performed byserum starvation and cell cycle regulatory drugs21 and anotherprevious study that investigated the infrared spectroscopicdifferences between cells in different stages of the cell cyclewhere synchronization was performed by centrifugal elutria-tion45

A main goal of this second study is to directly examinewhether the variability expressed by the first PCA components(addressed above in the discussion for study 1) is indeed dueto biochemical differences between cells at different points inthe cell cycle The first PCA components for this study haveprimarily the same features as the first PCA components for theasynchronous cell cultures study and both the LWN and HWNwindow components explain approximately the same amountof the total variance as the corresponding components in theasynchronous cell cultures study However in this study allfour cultures were harvested on the same day (after each weresynchronized) and the confluency of each culture was 60as such all four cultures should have very few quiescent cellsand we would not expect to measure any variability due to thedifferences in confluency between the cultures

The variability in the nucleic acid and protein contentrelative to the lipid content in single cells between the foursynchronized cell cultures is expressed by the PCA scores for

the first components (Fig 12) The slight shift to a higheraverage relative nucleic acid and protein content between thelsquolsquoG1Srsquorsquo and the lsquolsquoG1Sthorn3 hrsrsquorsquo cultures correlates with the flowcytometry measured shift from 83 of the first cultureexisting at the G1S boundary to 64 of the second cultureprogressing through S phase (Fig 10) This shift is consistentwith expected changes in the biochemical content for S-phasecells which contain increased levels of RNA and protein ascompared to G1 cells40 and an increased amount of DNA dueto the active DNA replication that occurs during S phase Thereis no observable shift in the scores between the lsquolsquoG1Sthorn3 hrsrsquorsquoand the lsquolsquoG2Mrsquorsquo cultures (Fig 12) even though the lsquolsquoG2Mrsquorsquoculture has over 74 of its cells at the G2M boundarycompared to only 16 of the cells in G2 phase for the lsquolsquoG1Sthorn3 hrsrsquorsquo culture This lack of separation in the scores for thefirst components may seem at odds with a known increase inthe overall RNA and protein content of G2M cells ascompared to late S-phase cells40 however the PCA scoresfor the first components only represent changes in nucleic acidand protein content relative to the total lipid content which isalso increasing throughout G2 phase in preparation for mitoticdivision

The most significant change in the scores for the first PCAcomponents occurs as a decrease in the average nucleic acidand protein content between the lsquolsquoG2Mrsquorsquo and the lsquolsquoG2M thorn5hrsrsquorsquo cultures which undergo a transition from a culture with74 of its cells in a G2M phase to a culture with 75 ofits cells existing within the first five hours of G1 phase Theobserved change in the relative nucleic acid and protein contentis consistent with previous biochemical experiments that haveshown that the lowest levels of RNA and protein are foundwithin the first few hours of G1 phase immediately followingcell division4041 Our observations are also consistent with theresults of the recent RM study for cell cycle discrimination21

in which the successful discrimination between S or G2M cellsand G0G1 cells was due to increased nucleic acid and proteincontent relative to lipid content in both S and G2M cells asmeasured in the LWN spectral window This previous studyalso reported poor discrimination between S and G2M cellsbased on nucleic acid and protein content relative to lipidcontent21 which we also observe here for both spectralwindows (Fig 12)

In our study it is interesting to note that the PCA scores forthe LWN window (Fig 12a) for the lsquolsquoG2Mthorn5 hrsrsquorsquo culture arewell split into two subgroups 60 of the cells have PCAscores 1 (low relative amount of nucleic acid and protein)and 40 of the cells have PCA scores 0 (high relativeamount of nucleic acid and protein) This split is matched bythe cell cycle distribution for this culture (Fig 10) which isdistinctly separated into two groups 75 of the cells in earlyG1 phase and 21 of the cells in G2 phase with only 4of the cells in S phase The relative positions of the scores forthis culture are similar for the HWN window but theseparation between the two subgroups is less distinct (Fig12b) In summary these results confirm that the mostsignificant source of Raman spectral variability between cellsin a culture which is expressed in this work by the first PCAcomponents can be confidently attributed to biochemicalchanges arising from the progression of individual cellsthrough their mitotic cycle

The features in the second PCA components for this study(Fig 13) are different from the features in the second PCA


components for the asynchronous cell cultures study (Fig 7)The scores for the LWN and HWN windows do not showsimilar trends which suggests they each arise from differentsources of biochemical variability between cultures The scoresfor the HWN component (Fig 14b) do not show any cleartrend or separation between cultures and the features in theHWN component (Fig 13b) are not clearly attributed to aunique set or class of biomolecules As such it is difficult toassign a biological meaning to the HWN window resultsHowever the scores for the LWN component (Fig 14a)distinctly separate the lsquolsquoG2Mrsquorsquo cells from the other cultureswith an increase in the average PCA score According to thecorresponding PCA component (Fig 13) the increase in scoresfor the G2M cells corresponds primarily with a decrease inaromatic amino acids choline and OndashPndashO groups in RNAandor lipids correlating with an increase in nucleic acid basesDNA a-helix and b-sheet amide groups and CH2 frac14CH andCfrac14C lipid groups The biological reason for these changes isunclear but may be related to changes in the cell biochemistryin preparation for mitotic division Alternatively the variabilitycould arise as a temporary cellular response to the nocodazoletreatment for synchronization of the G2M culture If thesemeasured changes are indeed due to a natural source ofvariability inherent to G2M cells this component was likelynot observed in the asynchronous cell culture study due to thelow fraction of G2 cells in each of the asynchronous culturesand the presence of other larger sources of variability such asthe changing confluency and shifts in the x-axis calibrationFurther work using different synchronization techniques wouldbe required to determine whether this source of variability isindeed an inherent characteristic of G2M cells

Spectral Variability and Principal Component AnalysisMany of the results in this work depend on the accurateassignment of a molecular origin to features in the PCAcomponents However achieving confidence in the validity ofsuch assignments is only possible if all external sources ofspectral variability that are not inherent to the biochemicalcomposition of the cells have been removed prior to PCAimplementation Sources of variability that arise includevariability in the intensity and shape of the fluorescent baseline(originating from a sample substrate or from the cellularmaterial itself) or variability induced in a data set by animproper spectral normalization technique If an externalsource of variability happens to contribute preferentially to acertain sample (or samples) in the data set the PCA algorithmwill faithfully correlate the variability from the external sourcewith any variability that is inherent to the sample in questionwhich is possibly the variability of interest in the experiment

An example of this issue which arose during the course ofthis work is the variability in intensity of spectral contributionsfrom the quartz substrate For the lsquolsquoG2Mthorn5 hrsrsquorsquo culture fromthe synchronized cell culture study the pellet of cells used forRaman acquisition was only a few cell layers thick as opposedto tens of cell layers thick for the other cultures in the study Assuch there was a slightly greater quartz contribution observedin all the spectra collected from the lsquolsquoG2Mthorn5 hrsrsquorsquo cells If avery conformal baseline was not applied for baseline correctionof the lower half of the LWN window (600ndash1200 cm1) itwas found that the variability due to quartz became significantto the degree that the quartz variability became correlated withother sources of variability inherent to the lsquolsquoG2M thorn5 hrsrsquorsquoculture As the conformity of the baseline was reduced the

resultant increase in quartz variability became observable in thefirst PCA component as recognizable quartz features and manybiological features from the first PCA component began toappear in the component that was originally dominated byquartz features alone

These considerations require extreme care when developingand implementing automated spectral processing methodssuch as spectral smoothing or baseline correction algorithmswhen large multi-sample data sets are prepared for PCAanalysis However the PCA components themselves can aid inthe identification of external sources of variability during thedevelopment and implementation of spectral processingtechniques as long as the spectral features of the externalsources are known It should be noted that performing Ramananalysis with the HWN window is significantly simpler sincebaseline removal is easier due to the absence of substratecontributions and fluorescence in this spectral region Howev-er the simplicity advantage comes at the cost of a significantdecrease in the amount of biochemical information availableas compared to the LWN window Furthermore the strongspectral contributions from water in the HWN window maybecome significant if the methods presented here are applied tothe Raman analysis of cells in an aqueous environment

Low-Wavenumber versus High-Wavenumber SpectralWindows Both the LWN and HWN windows were analyzedindependently throughout this work to determine whetherinformation can be obtained equivalently from either windowWe have found that biochemical variability due to cell cycle isclearly observable in either window and the spectraldifferences are directly observable in the original data for bothwindows (Figs 6a and 6b) However the LWN windowprovides information from many more biomolecules includingmultiple contributions from nucleic acids which are notobserved as sources of cell cycle variability in the HWNwindow The variability due to changes in cell cultureconfluency after sub-culturing is more apparent in the LWNwindow due to the strong contributions from the featuresidentified in the second PCA component from the asynchro-nous cell cultures study (Fig 4a) Although the trends of thePCA scores for the second components are similar for both theLWN and HWN windows (Fig 5) the molecular origins of thecorresponding features in the HWN window component areuncertain (Fig 4b) Furthermore the spectral differencesarising from this source of variability are directly observablein the original data only for the LWN window (Fig 9)Therefore in this case the LWN window provides spectro-scopic information that is not available in the HWN windowWe have also shown that the LWN window is sensitive tobiochemical changes unique to the G2M sample from thesynchronized cell cultures study whereas in the HWN windowno spectroscopic differences were observed for the samesample

Spectral Variability and Cell Size All of our single-cellRM measurements are acquired with a fixed sampling volume(2 3 5 3 10 lm in x-y-z) that is aligned with the center of theselected cell (see Fig 1) Therefore there is the possibility ofobserving spectral differences that correlate simply with sizedifferences in the cell population For example previousauthors16 have noted that a smaller cell will have a highersurface area to volume ratio than a larger cell and may thereforeyield more biochemical signals from cell membrane lipids andproteins relative to cytoplasmic and nuclear biomolecules


However these authors used direct measurements of the size ofeach selected cell obtained in suspension during opticaltweezers RM acquisition to show that cell size had nocorrelation with the ability of RM and PCA to biochemicallydiscriminate between two cell lines of different average size16

In our work monitoring absolute cell size via direct opticalmeasurements of the cells selected would be inaccurate due tothe lack of three-dimensional cell geometry information in thepellet However the relative cell size distribution for a givenculture is well described by the flow cytometry measurement offorward scatter intensity which is acquired from cells insuspension simultaneously with PI fluorescent intensityacquired for cell cycle analysis

To verify that the dominant sources of inherent spectralvariability observed in this work (as described by the first andsecond PCA components in study 1) are not simply due tochanging cell size we have analyzed the forward scatterintensity distributions of all eight cell cultures used in study 1From 24 to 72 hours after sub-culturing there is no detectablechange in the cell size distribution between cultures From 72to 96 hours there is a detectable shift in the measureddistribution towards lower forward scatter intensities which isindicative of a higher proportion of smaller cells in the culture

From 96 to 192 hours after sub-culturing there is no furtherdetectable change in the cell size distribution between culturesHowever our PCA analysis shows that the largest source ofspectral variability observed in this study (first PCA compo-nent) displays a steady trend of continuing spectral differencesoccurring from 48 to 192 hours after sub-culturing (Fig 5)Furthermore the second largest source of spectral variability(second PCA component) displays a trend of continuingspectral differences occurring from 24 to 120 hours after sub-culturing (Fig 8) Neither of these sources of variabilitycorrelate with the observed changes in the relative cell sizedistribution Therefore any spectral variability arising fromdifferences in cell size must be explained by one of the manylower variance PCA components each of which explains lessthan 3 of the total variance for the LWN window and lessthan 1 of the total variance for the HWN windowDifferences in cell size may introduce significant spectralvariability when comparing cell lines with large differences inaverage size but within a single cell line our results show thatcell size is not a significant source of spectral variability

CONCLUSION

We have shown that the inherent variability in Ramanspectra of single human tumor cells cultured in vitro iscorrelated with biochemical changes arising from (1) cell cycleprogression and (2) the confluency of a cell culture during thefirst three to four days after sub-culturing

The variability between single-cell Raman spectra arisingfrom cell cycle progression is expressed as varying intensitiesof protein and nucleic acid features relative to lipid featuresRaman spectra acquired from synchronized cell cultures showa continual increase in the average nucleic acid and proteincontent relative to lipid content as cells progress from early G1phase to the G1S boundary and into S phase The molecularorigins of the Raman features affected by cell cycle progression

have been identified for both the LWN and the HWN spectralwindows by the features of the first PCA components (Figs 4and 11) Our PCA analysis has shown that in the absence ofadditional external sources of variability cell cycle variabilitytypically accounts for 40ndash60 of the total variance if the LWNwindow is used and 75ndash90 if the HWN window is usedBecause there will always be some level of variability inbiochemical composition between cells due to the cell cyclethe characterization of cell cycle variability presented in thiswork may be useful for future Raman studies in order todistinguish the inherent cell cycle variability between cellsfrom other independent sources of variability

The molecular origins of the Raman features that producevariability correlated with the changing confluency of a cellculture have been identified for the LWN spectral window bythe features of the second PCA component from the study ofasynchronous cell cultures (Fig 7a) In our work with DU145cells this source of variability can explain up to 17 of thetotal variance if the LWN window is used The characterizationof this variability as presented in this work may be animportant consideration for future Raman studies involvingcomparisons between cell cultures harvested at different timeintervals after sub-culturing For example if cell cultures areallowed to incubate after sub-culturing for three to four daysbefore Raman analysis the variance explained by this source ofvariability is greatly reduced and may facilitate the observationof other more subtle spectral differences between cell cultures

ACKNOWLEDGMENTS

The authors gratefully acknowledge funding from the National Science andEngineering Research Council the Canadian Foundation for Innovation andthe Western Economic Diversification program We would also like to thankthe staff of the Deeley Research Centre at the BC Cancer Agencyrsquos VancouverIsland Centre for providing the initial DU145 cell stocks and technicalassistance with cell culture and flow cytometry

1 J Mourant J Dominguez S Carpenter K Short T Powers RMichalczyk N Kunapareddy A Guerra and J Freyer J Biomed Opt11 064024 (2006)

2 G Puppels J Olminkhof G Segersnolten C Otto F Demul and JGreve Exp Cell Res 195 361 (1991)

3 I Notingher S Verrier H Romanska A Bishop J Polak and L HenchSpectrosc Int J 16 43 (2002)

4 I Notingher J Jones S Verrier I Bisson P Embanga P Edwards JPolak and L Hench Spectrosc Int J 17 275 (2003)

5 A Nijssen T Schut F Heule P Caspers D Hayes M Neumann and GPuppels J Invest Dermatol 119 64 (2002)

6 J Choi J Choo H Chung D Gweon J Park H Kim S Park and COh Biopolymers 77 264 (2005)

7 C Lieber S Majumder D Billheimer D Ellis and A Mahadevan-Jansen J Biomed Opt 13 024013 (2008)

8 B de Jong T Bakker K Maquelin T van der Kwast C Bangma DKok and G Puppels Anal Chem 78 7761 (2006)

9 S Teh W Zheng K Ho M Teh K Yeoh and Z Huang J Biomed Opt13 034013 (2008)

10 S Teh W Zheng K Ho M Teh and K Yeoh J Raman Spectrosc 40908 (2009)

11 K Omberg J Osborn S Zhang J Freyer J Mourant and J SchoonoverAppl Spectrosc 56 813 (2002)

12 L Notingher G Jell P Notingher I Bisson O Tsigkou J Polak MStevens and L Hench J Mol Struct 744 179 (2005)

13 N Stone C Kendall N Shepherd P Crow and H Barr J RamanSpectrosc 33 564 (2002)

14 C Krishna G Sockalingum G Kegelaer S Rubin V Kartha and MManfait Vib Spectrosc 38 95 (2005)

15 P Crow B Barrass C Kendall M Hart-Prieto M Wright R Persad andN Stone Brit J Cancer 92 2166 (2005)

16 T Harvey E Faria A Henderson E Gazi A Ward N Clarke MBrown R Snook and P Gardner J Biomed Opt 13 064004 (2008)

This shift is consistent with our cell cycle analysis (Fig 3) where between72 and 96 hours we observe a sharp increase in the fraction of G1 phasecells which are typically smaller than S-phase and G2-phase cells


17 S Verrier I Notingher J Polak and L Hench Biopolymers 74 157(2004)

18 I Notingher S Verrier S Haque J Polak and L Hench Biopolymers 72230 (2003)

19 N Kunapareddy J Freyer and J Mourant J Biomed Opt 13 054002(2008)

20 K Short S Carpenter J Freyer and J Mourant Biophys J 88 4274(2005)

21 R Swain G Jell and M Stevens J Cell Biochem 104 1427 (2008)

22 J Mourant R Gibson T Johnson S Carpenter K Short Y Yamada andJ Freyer Phys Med Biol 48 243 (2003)

23 S Koljenovic T B Schut R Wolthuis B de Jong L Santos P CaspersJ Kros and G Puppels J Biomed Opt 10 031116 (2005)

24 M Whitfield L Zheng A Baldwin T Ohta M Hurt and W MarzluffMol Cell Biol 20 4188 (2000)

25 M Whitfield G Sherlock A Saldanha J Murray C Ball K AlexanderJ Matese C Perou M Hurt P Brown and D Botstein Mol Biol Cell13 1977 (2002)

26 J Gray F Dolbeare M Pallavicini W Beisker and F Waldman Int JRadiat Biol 49 237 (1986)

27 A Tomida J Yun and T Tsuruo Int J Cancer 68 391 (1996)

28 Y Zhang N Fujita and T Tsuruo Oncogene 18 1131 (1999)

29 A Jirasek Q Matthews M Hilts G Schulze M Blades and R TurnerPhys Med Biol 51 2599 (2006)

30 Q Matthews A Jirasek N Virji-Babul A Babul and T CheungBiomed Signal Process Control 3 78 (2008)

31 L Greek H Schulze M Blades A Bree B Gorzalka and R TurnerAppl Spectrosc 49 425 (1995)

32 H Schulze R Foist A Jirasek A Ivanov and R Turner ApplSpectrosc 61 157 (2007)

33 G Schulze A Jirasek M Yu A Lim R Turner and M Blades ApplSpectrosc 59 545 (2005)

34 I Notingher and L Hench Expert Rev Med Devic 3 215 (2006)35 C Krafft T Knetschke A Siegner R Funk and R Salzer Vib

Spectrosc 32 75 (2003)36 N Uzunbajakava A Lenferink Y Kraan B Willekens G Vrensen J

Greve and C Otto Biopolymers 72 1 (2003)37 A Synytsya P Alexa J Besserer J D Boer S Froschauer R Gerlach

M Loewe M Moosburger I Obstova P Quicken B Sosna K Volkaand M Wurkner Int J Radiat Biol 80 581 (2004)

38 D Borchman D Tang and M Yappert Biospectroscopy 5 151 (1999)39 M Levitt Biochemistry 17 4277 (1978)40 H Crissman Z Darzynkiewicz R Tobey and J Steinkamp Science

(Washington DC) 228 1321 (1985)41 Z Darzynkiewicz T Sharpless L Staianocioco and M Melamed Proc

Nat Acad Sci Biol 77 6696 (1980)42 J Mourant K Short S Carpenter N Kunapareddy L Coburn T

Powers and J Freyer J Biomed Opt 10 031106 (2005)43 S Cooper Cell Mol Life Sci 60 1099 (2003)44 S Cooper G Iyer M Tarquini and P Bissett Cell Tissue Res 324 237

(2006)45 S Boydston-White T Gopen S Houser J Bargonetti and M Diem

Biospectroscopy 5 219 (1999)


inherent sources of spectral variability that may arise due tobiochemical differences between single cells

In this work we have undertaken an investigation of thecapability of RM to detect inherent sources of spectralvariability within a human tumor cell line (DU145) culturedin vitro PCA is used to observe differences in Raman spectrathat correlate with cells existing in different phases of the cellcycle as well as differences correlating with the confluency ofthe cell culture at the time of Raman analysis Furthermore thebiochemical changes detected by RM that correlate with cellcycle progression are consistent with known biochemicalchanges within cells We also show that the variability fromcell cycle and culture confluency comprises almost all of theinherent variability in a multi-culture data set and is primarilyexplained by the first two PCA components

The results of this work are presented in two studies In thefirst study the inherent variability between cultures of varyingconfluence is investigated by acquiring Raman spectra of cellscollected from asynchronously growing cell cultures harvest-ed 1ndash8 days after sub-culturing In the second study thevariability due to cell cycle in sub-confluent cultures isdirectly examined by acquiring spectra from cells collectedfrom cultures synchronized at specific points in the cell cycleFlow cytometry is used to monitor the cell cycle distributionand viability of all cultures at the time of Raman analysis Weuse adherent cells that have been resuspended and centrifugedinto a pellet from which individual cells are selected using ahigh-power focusing objective and 785 nm laser excitationThis technique provides a very high quality Raman spectrumof a single cell while eliminating any spectral variability(caused by inconsistent cell structure and varying local celldensity) arising when cells are grown and analyzed directly ina monolayer22 Both low-wavenumber (LWN) 600ndash1800cm1 and high-wavenumber (HWN) 2700ndash3100 cm1spectral windows are used in order to determine whetherinformation can be attained equivalently from either windowas some authors have found to be the case for certainapplications23

METHODS

Cell Preparation Standard Cell Culture Human prostatetumor cells (cell line DU145 (ATCC Manassas VA)) werecultured in a sterile environment and grown in T-75 flasks with15 mL of Dulbeccos Modified Eagle Medium (HyCloneLogan UT) supplemented with 10 fetal bovine serum (FBS)(HyClone) Cultures were kept in an incubator at 5 CO2 and37 8C to promote growth Cell stocks were sub-cultured every3 to 4 days by rinsing the cells in phosphate buffered saline(PBS) adding trypsin (HyClone) to detach the cells from theflask and transferring 10ndash20 of the harvested cells to a newflask containing fresh media Before re-incubation the new cellsuspensions were pipetted several times to ensure an evenlydistributed monolayer of cells throughout the flask

Asynchronous Cell Cultures A single T-75 flask wasgrown to 90 confluency and sub-cultured equally into eightidentical T-75 flasks The first flask was harvested for RManalysis 24 hours after sub-culturing and each remaining flaskwas harvested every 24 hours thereafter Each time a flask washarvested the confluency cell cycle distribution and viabilityof the culture was measured as described below

Synchronized Cell Cultures Using established proto-cols2425 2 mM of thymidine (Sigma-Aldrich Oakville ON

Canada) was used to inhibit DNA replication and arrest cells atthe end of G1 phase before the onset of S phase and 100 ngmL of nocodazole (Sigma-Aldrich) was used to preventformation of the mitotic spindle and arrest cells at the end ofG2 phase prior to M phase In the second study presented inthis work cell cultures were treated with thymidine andnocodazole to obtain cultures synchronized at four pointsduring the cell cycle (1) at the G1S boundary (with double-thymidine24 treatment) (2) during S phase (with double-thymidine treatment and re-incubation with drug-free media for3 hours) (3) at the G2M boundary (with thymidine treatmentfor 24 hours re-incubation with drug-free media for 3 hoursnocodazole treatment for 12 hours and mitotic shake-off) and(4) during early G1 phase (with mitotic shake-off from a G2Marrested culture and re-incubation of detached cells with freshmedia for 5 hours)

Cell Cycle Viability and Confluency Analysis The cellcycle distribution and viability of each culture was measuredusing flow cytometry26 following established protocols2728

For cell cycle analysis over 200 000 cells were extracted froma culture and fixed in 70 ethanol to permeabilize the cellmembrane RNase A (Qiagen Inc Mississauga ON Canada)was added to a concentration of 1 mgmL in order to degradecellular RNA Propidium iodide (PI) (Sigma) was subsequentlyadded to a final concentration of 50 lgmL After 30 minutesthe suspension was centrifuged re-suspended in a buffer (PBSthorn 1 FBS) and kept on ice until analysis For flow cytometrycollection the relative DNA content of 100 000 cells wasmeasured using a BD FACSCalibur Flow Cytometer (BDBiosciences Mississauga ON Canada) Cell counts wererecorded by the flow cytometer during sample acquisitionRelative fractions of cells in each phase of the cell cycle (G1 Sor G2) were determined by performing a nonlinear least-squares fit to the measured data using functions representativeof the expected distributions of DNA content for each cellcycle phase (Matlab The Mathworks Natick MA)

For culture viability assessment over 100 000 cells wereextracted from the culture and split into two equal parts Onepart was stained with PI at a concentration of 5 lgmL and theother part was left untreated to serve as a control 20 000 cellsfrom each sample underwent flow cytometry analysis within 15minutes of staining and the fractions of live (no PI signal) anddead (positive PI signal) cells in the PI stained sample weredetermined

Cell culture confluency is defined as the percentage of thesurface area of the culture flask covered by cells Confluencyestimates of each of the asynchronous cell cultures wereobtained prior to harvesting by acquiring low magnificationimages of five different regions of the cell culture Each imagewas imported into Matlab and the fraction of the image coveredby cells was calculated Because the confluency is neverconsistent throughout the entire surface area of the flask theaveraged confluency from the five regions was used as anestimate of the overall confluency of the culture

Raman Microscopy To prepare cells for Raman analysiscultures were harvested by rinsing with PBS to remove deadcells and debris adding trypsin to detach the remaining livecells and centrifuging to discard the trypsin supernatant Cellswere re-suspended in growth media and centrifuged into apellet in a 200 lL plastic vial Vials were kept on ice until RManalysis (1ndash6 hours) upon which the chosen pellet wastransferred to a low-fluorescence quartz disc (Technical Glass











RESULTS

















Raman shift (cm1)





669 G T3435

700 Cholesterol35

719 Choline343538

728 A34ndash36

759 Trp ring br3435























1300 CH2 tw1134353738

1320 C1134 CH def1834


1374 T113536

1421 A G1136 CH2 bk35

1438 CH2 def11353738


1460 CH def1834 CH def1834

1486 A G3536

1577 A G1134ndash36







1743 Cfrac14O str34353738





































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS

























































RESULTS

















Raman shift (cm1)





669 G T3435

700 Cholesterol35

719 Choline343538

728 A34ndash36

759 Trp ring br3435























1300 CH2 tw1134353738

1320 C1134 CH def1834


1374 T113536

1421 A G1136 CH2 bk35

1438 CH2 def11353738


1460 CH def1834 CH def1834

1486 A G3536

1577 A G1134ndash36







1743 Cfrac14O str34353738





































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS





























































Raman shift (cm1)





669 G T3435

700 Cholesterol35

719 Choline343538

728 A34ndash36

759 Trp ring br3435























1300 CH2 tw1134353738

1320 C1134 CH def1834


1374 T113536

1421 A G1136 CH2 bk35

1438 CH2 def11353738


1460 CH def1834 CH def1834

1486 A G3536

1577 A G1134ndash36







1743 Cfrac14O str34353738





































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS




















































Raman shift (cm1)





669 G T3435

700 Cholesterol35

719 Choline343538

728 A34ndash36

759 Trp ring br3435























1300 CH2 tw1134353738

1320 C1134 CH def1834


1374 T113536

1421 A G1136 CH2 bk35

1438 CH2 def11353738


1460 CH def1834 CH def1834

1486 A G3536

1577 A G1134ndash36







1743 Cfrac14O str34353738





































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS












































































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS







































































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS














































































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS



















































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS











































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS





































































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS





















































DISCUSSION


































CONCLUSION





ACKNOWLEDGMENTS













































































CONCLUSION





ACKNOWLEDGMENTS






































































CONCLUSION





ACKNOWLEDGMENTS




























































CONCLUSION





ACKNOWLEDGMENTS




















































CONCLUSION





ACKNOWLEDGMENTS












































































Date post:	10-Jul-2020
Category:	Documents
Upload:	others
View:	0 times
Download:	0 times

Variability in Raman Spectra of Single Human …agbrolo/cell_2010.pdf2010 Society for Applied...

Documents