at SciVerse ScienceDirect

Pancreatology 12 (2012) 124e129

Contents lists available

Pancreatology

journal homepage: www.elsevier .com/locate/pan

Original article

Discovery of diagnostic biomarkers for pancreatic cancer in immunodepletedserum by SELDI-TOF MS

Aiqun Xue, Robert C. Gandy, Liping Chung, Robert C. Baxter, Ross C. Smith*

Department of Surgery, The University of Sydney, Suite 5 Level 5 North Shore Private Hospital, St Leonards, NSW 2065 Australia

Keywords:Pancreatic neoplasmsBiomarkersSerum proteomeImmunodepletionDiagnosis

* Corresponding author. Tel.: þ61 2 94373511; fax:E-mail address: ross.smith@sydney.edu.au (R.C. Sm

1424-3903/$ e see front matter Copyright � 2012, IAdoi:10.1016/j.pan.2012.02.009

a b s t r a c t

Motivation: Reports of serum pancreatic cancer (PC) biomarkers using SELDI-TOF MS have been incon-sistent because different chip surfaces and interference with high-abundant proteins. This studyexamines the influence of these factors on the detection of discriminating diagnostic biomarkers.Methods: Serum from fourteen from patients with PC, disease controls (DC, n ¼ 14) and healthyvolunteers (HV, n ¼ 14) were evaluated by SELDI using H50, IMAC, Q10 and CM10 chips. A furtherevaluation was undertaken after depletion of seven high-abundant proteins using spin cartridges.Results: More protein peaks were detected in whole serum than in depleted serum for IMAC, H50 andQ10 chips: 60 vs 39, 56 vs 48 and 69 vs 65, respectively, while the CM10 found less peaks in serum (27 vs47 peaks). However, there were more differentially expressed peaks in the depleted serum samples forPC vs DC and PC vs HV samples using the H50, Q10 and CM10 ProteinChip arrays, whereas for IMACarrays, more discriminating peaks were seen in non-depleted serum. The highly significant peaksobserved on Q10, CM10 and H50 are consistent with the previous finding of ApoA-I (m/z 27,910e28000)and ApoA-II (m/z 8758 and 17,240). In addition, a number of new discriminating protein peaks werefound on different ProteinChip arrays, notably peaks at m/z 4280 and 7763 on IMAC arrays.Conclusion: This study confirms the diagnostic value of ApoA-I&II and identifies further potential diag-nostic biomarkers for pancreatic cancer when multiple chip surfaces are used with depletion of the mosthighly-abundant proteins.Copyright � 2012, IAP and EPC. Published by Elsevier India, a division of Reed Elsevier India Pvt. Ltd. Allrights reserved.

1. Introduction

The serum of cancer patients contains many proteins whichinteract with the cancer raising the hope of discovering newbiomarkers to aid earlier diagnosis of pancreatic cancer (Rosty C,2002, 1868; Koopmann J, 2004, 860; Yu Y, 2005, 79, Honda K, 2005,10,613) [1e4]. Ciphergen Biosystems (Fremont, CA, USA) hasdeveloped ProteinChip technology that utilizes Surface-enhancedlaser desorption and ionization time of flight mass spectrometry(SELDI-TOP-MS) to facilitate protein profiling of complex biologicalspecimens (Conrade TP, 2004, 163) [5]. It allows the analysis ofserum sample sets to discover proteins whose concentrations areinfluenced by disease states. Recent studies have demonstratedthat the detection of pancreatic cancer will probably benefit froma screen of many individual proteins to improve the positive andnegative predictive value of CA19.9. Five studies have identified

þ61 2 94373522.ith).

P and EPC. Published by Elsevier I

a number of markers (Ehmann, 2007, 205; Takano, 2008; Guo J,2009, 2292; Fiedler GM, 2009, 3812; Xue A, 2010, 391) [6e10]which appear to discriminate serum from pancreatic cancer casesand disease control groups and from healthy volunteers. Thesestudies demonstrated different protein markers but they have useddifferent chip surfaces and therefore different methods of selection.Another important variable is the method of preparation of theserum sample. Some studies have examined undepleted serumwhile other studies filter off the high-abundant proteins such asalbumin, IgG, antitrypsin, IgA, transferrin, haptoglobin and fibrin-ogen because theymay conceal important low-abundance proteins.

These variances in technique and patients could lead toapparent non-reproducible results between studies [11]. Differentprotein groups would be expected to adhere to the different chipsurfaces: The ProteinChip H50 array binds proteins throughreverse-phase or hydrophobic interaction chromatography; theCM10 ProteinChip arrays incorporate a carboxylate chemistry(negatively charge) and thus acts as a weak cationic exchanger,while the ProteinChip Q10 array incorporates quaternized ammo-nium groups (positively charged) and thus acts as a strong anion

ndia, a division of Reed Elsevier India Pvt. Ltd. All rights reserved.

A. Xue et al. / Pancreatology 12 (2012) 124e129 125

exchanger; The IMAC30-Cu ProteinChip arrays is coated witha nitrilotriacetic acid functional group to entrap transition metalsfor subsequent metal affinity binding to proteins (Ciphergen Bio-systems). Among the several identified serum markers, ApoA-1,ApoA-II, and Transthyretin were derived from CM10 and IMAC30-Cu (Ehmann, 2007, 205) [6]; ApoC-1 was from CM10 (Takano,2008, 2810) [9]; C14orf16 was from Q10 (Guo, 2009, 2292) [8].Recently, ApoA-II and ApoC-I, have been confirmed as usefuldiscriminators using a single ProteinChip array-H50 (Xue A, 2010,391) [10]. The study by Xue et al.[10] used two separate discoverysets of serum, a total of 240 samples, with a further validation ona further 79 samples using ELISA. Consideration of the above papersit is likely that differences in results of serum biomarker discoveryfor pancreatic cancer appear to be partly due to the different chipsurfaces being used. This paper examines the value of undertakingcomprehensive proteomic profiling using of a range of protein chipswith and without immune-depleted the high-abundance proteins.

2. Material and methods

2.1. Serum and samples preparation

For the comprehensive proteomic analysis of human pancreaticductal adenocarcinoma (PDAC) serum proteome, serum sampleswere collected from PDAC, diseases control (DC) and healthyvolunteer (HV) groups. The DC group included non-neoplasticcontrols (such as mucinous cystadenoma) and pancreatitis. Thecrude serum sample was diluted 1:1 with denaturing buffer 8 MUrea/1% CHAPs, then centrifuged at 12,000 rpm for 5 min. Thesupernatant was diluted to 1:20 with chip binding buffer. All serumsamples were pre-treated by a protein immunodepletion system.Serum samples and the depleted serum samples collected from theimmunodepletion system were analysed using four different Pro-teinChip arrays of SLEDI-TO MS. For this purpose, we collected 14samples per group to allow a statistically significant analysis.

2.2. Seven high-abundant proteins immunodepletion

The multiple affinity removal system (MARS) spin cartridge(Agilent Technologies, Wilmington, DE, USA) uses antibodies toremove albumin, transferrin, haptoglobin, antitrypsin, FibrinogenIgG, and IgA from serum. Briefly, 12 mL of human serumwas diluted20-fold with the manufacturer’s equilibration buffer, filteredthrough a 0.22mmmicrocentrifuge filtration tube, and loaded ontothe antibody resin in a microcentrifuge spin column. The samplewas passed into the column at slow speed (1.5 min, 100 � g, RT) Inboth cases, the column was washed twice with the equilibrationbuffer and centrifuged (2.5 min, 100 � g, RT) to collect the totalunbound plus wash fractions. The bound fraction was collectedwith the manufacturer’s elution buffer and re-equilibrate, and boththe flow-through fraction containing unbound proteins and thebound fractions were concentrated with a 5 K MWCO spinconcentrator (Agilent Technologies, Wilmington, DE, USA) forimmediate analysis or aliquot and stored at �80 �C until use.

2.3. ProteinChip handing protocols

Each sample was analysed on four different array surfaces:hydrophobic (H50), cation-exchange (CM10), anion-exchange(Q10) and metal binding (IMAC30-Cu) arrays. For the four chiptypes tested, chip protocols were following the standard protocolsprovided by the chip manufacture Ciphergen Biosystems. Thedifferent arrays were pre-treated with 50% Ch (ACN), 0.5% tri-fluoroacetic acid (TFA) for H50; 100 mM sodium acetate pH 4.0 forCM10; 100 mM TriseHCl pH 8.0 for Q10, and 100 mM CuSO4,

100 mM sodium acetate (pH4.0), 100 mM phosphate buffer pH 7.0/500 mM NaCl for IMAC30-Cu. Samples were randomly applied toa ProteinChip array surface and the arrays were incubated for 2 hwith gentle shaking. Arrays were then washed three times with10 mL of binding buffer, followed by a final washwith water. Bindingbuffers used for the different arrays were 10% CAN/0.1% TFA forH50; 100 mM sodium acetate pH 4.0 for CM10; 100 mM TriseHClpH 8.0 for Q10; and 100 mM phosphate buffer pH 7.0/500 mMNaCl for IMAC30-Cu. Arrays allowed to air dry. One microliter of50% sinapinic acid was applied twice to the spots. Finally, arrayswere analysed on a PBSIIc ProteinChip Reader (Ciphergen Bio-systems, Inc.).

2.4. SELDI peak detection

The data were averaged over 200 laser shots for each spot. Massdetection accuracy of PBSIIc was calibrated externally by using theAll-in-1 protein II molecular mass standards (Ciphergen Bio-systems). To minimize experimental variation, samples were ana-lysed concurrently, and no sample was subjected to more than twofreezeethaw cycles. In all experiments, SELDI-TOF MS profiles wereobtained in triplicate for each sample laser intensity was optimizedfor each protein array chip type. Samples were prepared and readon the same day in batches of 42 samples. All samples were readover a period of 8 days, to reduce mass and laser intensity drift.Peak detection was performed using the ProteinChip� DataManager Software (Bio-Rad, Hercules, CA, USA). The SELDI-TOF-MSwas calibrated to known standards, bovine insulin (5734.51 þ1H),equine cytochrome c (12,361.96 þ1H), equine apmyoglobin(16,952.27 þ1H) and rabbit muscle aldolase (39,212.28 þ1H)(SigmaeAldrich, St. Louis, MO, USA) prior to reading. Mass spectrawere generated in the mass to charge range (m/z) 2000e30,000.Proteins and peptide below 1000 Da were deflected. Laser inten-sities were optimized for each chip type using bovine insulin priorto trial commencement. Automatic peak detection was performedwith the following settings: (i) signal-to-noise ratio at 5 for the firstpass and 2 for the second pass, (ii) minimal peak threshold at 5% ofall spectra, (iii) cluster mass window at 0.5% of mass. The resultingfile, containing absolute intensities and m/z ratios, was exportedinto SPSS software (Version 16.0, Chigaco, IL, USA) for subsequentanalysis.

2.5. Reproducibility

Reproducibility was estimated using a control sample (BovineInsulin, m/z 5725e5735, Sigma) which was randomly spotted oneach chip to measure the variability of on-chip spotting and dataacquisition. The percentage coefficient variation (% CV) for massbetweenm/z 5725e5735was less than 0.2% for all runs. ThemedianCV for the intensity of the insulin peak was 8.4 (IQR 6.4e19.5) whennormalisation was undertaken with the whole protein spectrumand 6.8 (IQR 3.7e9.5) when the normalisation was undertakenfrom 5000 to 6000 kDa spectrum. The coefficients of variation (CV)of the reliable peaks were calculated by averaging peak intensityvalues derived from different runs. The average CVs was less than20% for intra- and inter-assay variability.

2.6. Data and statistical analysis

Mass to charge spectra were processed to reduce noise, spot tospot and baseline correction was performed. Spectra were nor-malised between m/z 2000 and 30,000. External controls werenormalised between m/z 2000 and 30,000 and m/z 5000e6000.Common protein peaks were identified between PC, DC and HVusing Biowizard Utility software (Bio-Rad). Coefficients of variation

Fig. 1. Efficiency of removal of 7 high-abundant proteins from human serum by usinga multiple affinity removal spin cartridge (Agilent). Equal amounts of protein fromeach fraction were separated by 4%e20% SDS-PAGE under non-reducing conditionsand visualized by SYPRO Staining. Lane 1, 5 and 9 e molecular weight standards(Benchmen, unstained, Invitrogen); Lane 2 and 6ecrude serum samples 2.18 and 8.16,respectively; Lane 3 and 7 e flow-through fraction for serum samples 2.18 and 8.16respectively; Lane 4 and 8 e bound fraction for serum samples 2.18 and 8.16,respectively.

A. Xue et al. / Pancreatology 12 (2012) 124e129126

were calculated for external controls on each chip type and for eachrun to assess technical variation. Standard deviations of m/z forcommon peaks were recorded. Common peak information wasexported statistical software package (SPSS version 17.0, Chicago).Differences between the intensities of the peaks was tested usinga Kruskal Wallis One-way ANOVA and when there was a significantdifference of P< 0.005 Univariate analysis was performed using theManneWhitney U test for mean peak intensity heights wasundertaken to compare different groups. Statistical significance andthe calculation of these parameters were computed with theStatistical Package for the Social Sciences program (SPSS Software,version 17.0, Chicago, Ill).

3. Results

3.1. Serum samples and patient characteristics

Fourteen patients with a proven diagnosis of pancreaticadenocarcinoma (PC) had serum sampled for this analysis. Twofurther control groups of 14 patients; the disease controls (DC),who had benign pancreatic conditions and healthy controls (HC),who had no history of pancreatic disease or any other cancer.Patient characteristics indicate that the median age and ratio ofmales to females was similar Table 1.

3.2. The efficiency of seven high-abundant proteinsimmunodepletion

Specificity of the spin cartridge for high-abundant proteinsremoved from human serum was tested by 1-demitinal SDS-PAGEwith SYPRO Protein Gel Staining. Fig. 1 shows that the majorhigh-abundant proteins, such as albumin, IgG, IgA, transferrin,antitrypsin and haptoglobin (MW: 67, 146, 160, 80, 47, and171.9 kDa, respectively), were efficiently removed by a MultipleAffinity Removal Spin Cartridge (Agilent). The remaining proteinsare seen to bemaintained on thewhole and the background ismoreclear particularly in the area between 9 and 30 kDa. However thecleaner appearance possibly allows for better enrichment andvisibility of low-abundant proteins of interest.

3.3. Comparison of the proteomic profiles of serum and depletedserum by 4 types of ProteinChip arrays

A comparison of the protein profile between serum samples andthe depleted serum samples collected from the seven high-abundant proteins immunodepletion system for a single patient

Table 1Patient characteristics.

Number Mean age Range

Pancreatic adenocarcinoma 14 69 48e83Male 9 66 48e83Female 5 73 58e86Healthy Volunteers 14 65 52e79Male 7 57 52e65Female 7 74 69e79Disease controls 14 59 18e85Male 9 56 18e85Female 5 63 32e83Choledocolithiasis 3 70 60e85Chronic pancreatitis 3 42 18e55Neuroendocrine tumour 3 58 49e66Benign common bile duct stricture 2 55 36e74Metastatic carcinoid tumour of pancreas 1 76 N/AMucinous cystic neoplasm 1 83 N/AIslet cell carcinoma 1 32 N/A

is shown in Fig. 2. The samples were managed with the samedilution in a final concentration to demonstrate differencesbetween the proteomic profiles of serum (S) and depleted serum(DS) across four chip types. The number of peaks (common to eachchip type) varies across the four chip types. Peaks common to allfour chip types are expressed to variable intensity in each chip andbetween serum and depleted serum.

Many of the moderate and low intensity protein peaks thatwere detected on the depleted sample were also detectable on theserum sample (Fig. 2). However, m/z 7900 for CM10 and IMAC, m/z14,00e16,000 for CM10 and Q10; m/z 14,000 for H50 and m/z28,000 for CM10, Q10, H50 and IMAC arrays were detected inserum but were reduced in amount or not measurable in depletionserum. Alternatively, m/z 4100 on Q10, m/z 14,000 on IMAC and m/z between 8000 and 10,000 on H50 arrays were below thedetection limit in serum but were increased in the post-depletionsamples.

3.4. SELDI peak quantification between serum and depleted serum

The overall number of peaks detected to be greater than fivetimes the baseline, by SELDI analysis, across three chip arrays(IMAC, H50 and Q10) was higher in serum 60 to 56, 69 thandepleted serum samples, 39 and 48e65, respectively. However,with the CM10 chip array there were 47 peaks identified when thedepleted serum samples were analyzed compared to 27 peaks inserum samples. A total 212 different peaks were identified in serumcompared to 199 in depleted serum. Fifty-five peaks were commonto multiple chip types.

3.5. Discriminatory peaks between PC to DC and HC byNonparametric Testing

Comparing the serum results for the three sample groups(Table 2) therewere significant differences between the intensity of19 proteins at P < 0.005 (Kruskal Wallis test). Fourteen of theseproteins had higher intensity in the PC serum and five had a lowerintensity than in the HV samples. Only two proteins had a higher

Fig. 2. Comparison of the proteomic profiles serum (S) and depleted serum (DF) of a single sample (2.016) using CM10, Q10, IMAC and H50 arrays, respectively. The spectra range isfrom m/z 4000 to 30,000.

A. Xue et al. / Pancreatology 12 (2012) 124e129 127

intensity in PC compared with DC samples. It is interesting thatthese proteins were in the m/z range of 3268e10,257 kDa.

Alternatively, the depleted serum samples demonstratedsignificant differences between peak intensities for 16 proteins

Table 2Protein/peptide peaks demonstrating differences in intensity between patientgroups in serum samples.

Results for serum

m/z Chip Pancreatic cancer Disease controls Healthy volunteers

Median IQR Median IQR Median IQR

3268 IMAC 7 8.38 8.5 11.5 22*** 14.633958 IMAC 28 14.63 45* 45 68.25**** 634280 IMAC 16 12.25 24 25.5 56.5**** 65.385757 Q10 5.5 3.5 4.5 6.13 2.5*** 1.255836 Q10 3 3.38 2.5 3.47 1*** 15836 H50 2.2 2.75 1.5 8.6 1*** 0.57763 CM10 52.25 48.3 60.75 74 9.25*** 22.87763 IMAC 275 121 327 111 162*** 81.17924 CM10 7 3.88 11 6.8 4.75** 3.757927 IMAC 33 14.9 38.5 20 20*** 12.57966 CM10 7.25 5 8.75 7 3*** 2.257970 IMAC 39.75 16.6 44.5 14 21.5*** 9.638137 IMAC 52.75 19.25 63.5 27 30.5*** 14.758139 CM10 9.75 7.38 12.25 11.63 4.5*** 3.758758 Q10 19 16.5 24.25 19 38.9**** 7.388760 H50 17 27 30 22.1 61.7*** 33.99285 CM10 9.75 10.75 8.25 10.75 4.25**** 2.889346 Q10 36.25 7.38 36.5 13.75 50*** 20.7510,257 CM10 2.75 1.88 1.5* 1.25 1*** 0.75

IQR ¼ Interquartile range.* Indicates P < 0.05 for difference with PC values (ManneWhitney test)y indicatesdifference between PC and HV samples * for P< 0.05 ** for P< 0.01, *** for P< 0.005and **** for P < 0.001.

from the m/z range of 4620e29,075 kDa (Table 3). Seven of thesehad a higher intensity in the PC compared to the HV samples. Eightproteins had different intensities in the PC samples compared to theDC and three of these had higher intensities in the PC group.

Table 3Protein/peptide peaks demonstrating differences in intensity between patientgroups in depleted serum samples.

Results for depleted serum

m/z Chip type Pancreatic cancer Disease controls Healthy volunteers

Median IQR Median IQR Median IQR

4620 Q10 31 34.5 22.25* 12.7 16*** 7.254634 IMAC 28.5 12.5 22 8.5 15.7*** 6.45802 CM10 7.5 14.8 6 93 2.8** 9.58458 Q10 1.5 2.5 3** 4 3.5*** 2.28565 Q10 7 12.75 21.25** 22.8 19*** 118759 H50 10.5 7 14.5**** 20.5 23.5**** 249283 IMAC 145 60 127 74 66*** 9010,252 IMAC 14.5 12 6* 6 2.7**** 8.410,256 CM10 8.5 6.5 7.7 6.2 2.7*** 1.710,360 IMAC 6 3.5 3.5* 2 2*** 312,826 IMAC 4 1 3 2 4.5* 2.517,232 Q10 21 19.5 33.25 17.63 47*** 1117,241 CM10 2 5 6 5.2 7.5** 3.717,436 Q10 13 5.25 13 6.75 17.5** 4.527,910 Q10 13 8.75 25.9** 18 24.5*** 12.2529,075 CM10 1.5 2 2** 0.36 2** 1

IQR ¼ Interquartile range.* Indicates P < 0.05 for difference with PC values (ManneWhitney test)y indicatesdifference between PC and HV samples * for P< 0.05 ** for P< 0.01, *** for P< 0.005and **** for P < 0.001.

A. Xue et al. / Pancreatology 12 (2012) 124e129128

4. Discussion

This serum study confirms the presence of diagnosticbiomarkers in serum through the use of SELDI-TOF MS to screena number of samples. However to acquire a broad screen anddiscover the maximal number of potential biomarkers for pancre-atic cancer requires the use of four different chip surfaces to takeadvantage of the different adherence properties and both serumand depleted serum have been used. The MARS spin column thatwas used removed the seven most abundant proteins allowing fora greater amount of the low abundant proteins to be added to thechips without the interference of the single from the large proteins.This increased the number of proteins which demonstrateddifferent concentrations between PCa and DC samples from two inserum to eight in depleted serum. It was also interesting that in theserum samples only one of the larger proteins within m/z rangefrom 10,000 to 30,000 kDa was able to differentiate PCa sampleswhile nine peaks were differential in the depleted serum samples.

Serum/Plasma is a promising source of biomarkers, and pro-teomic approaches have been developed to identify novel plasmabiomarkers. However, the complexity of protein interactions anddynamic protein concentration range are obstacles for identifyinglow abundant proteins (<10 ng mL�1) which account for only 5e6%of total protein identifications in immunodepleted plasma. The 50most abundant plasma proteins account for 90% of cumulativespectral counts and precursor ion intensities andmay conceal thoseproteins in low concentration [12]. Thus Rudnick et al. consider thatuntargeted proteomic analyses using current LC-MS/MS platforms-even with immunodepletion-cannot be expected to efficientlydiscover low-abundance, disease-specific biomarkers in plasma[12]. Prefractioning methods using immunodepletion and a bead-based library of combinatorial hexapeptide technology [13] indi-cated that the methods were complementary for identifying

Table 4Applications of serum proteomics in diagnosis and Prognosis of pancreatic Cancer.

Studies Type of array Protein peak (m/z) Biomarker (ID) Senof d

Ehmann2007[6]

IMACa

CM10a13,76117,27028,079

TransthyretinApoA-IIApoA-I

89/9

Takano, 2008[9] CM10a 66306420

ApoC-Ib N/D

Guo,2009[8]

Q10a 40164155479128,068

C14orf16d

(m/z 28,068)91.4(PC83.3(PC

Fiedler,2009[7]

MALDI-TOFMS

38845959

PlateletFactor 4e

86.3

Xue, 2010[10] H50a 66208759/17,247

ApoC-IApoA-II

AUC(PC80.6(PC

Honda 2005[1] 8,766, 17,272, 28,080,and 14,779 m/z

AUC(PC97.2

Kooperman2004[2]

WCXIMAC

3,146, 12,861, 3,885,3,967, and 8,929, 3,473,5,903, 16,008, 3667,7,441, 12,861

(PCv

Navaglia2009[17]

IMAC 1211, 1802, 1526, 3359,3519, 7903, 1802,

a : SELDI-TOF MS.b : both m/z 6630 and 6420 were identified as ApoC-I.c : AUC means the area under the ROC curve.d : m/z 28.068 was identified as C14orf16.e : m/z 3884 was identified as Platelet factor 4. N/D: not detected.

proteins [14]. Comparing the %CV of relative intensities for low andhigh m/z measurements revealed six-fold more peaks were foundin albumin-depleted than unfractionated serum at both high andlowm/z. However, sporadic and systematic variation in efficiency ofalbumin depletion by spin columns may contribute significantpreanalytical bias and requires ongoing QC of the albumin deple-tion process by quantification of albumin concentration to assessdepletion efficiency [15].

Different chip surfaces had different results, as a result of thedifferent surface characteristics, but there were some protein peaksthat were identified on multiple chip surfaces. In the depletedserum samples it is apparent that there were two proteins whichwere seen at a slightly differentm/z on different chips, one of whichwas 17,232 on Q10 and 17,241 on the CM10 chips is consistent withApoA-II and the other was m/z 10,252 (IMAC) and 10,256 (CM10)and this protein needs to be identified. Shin et al.([16] also founda peak at 10,256 to be elevated in the serum in breast cancersubjects compared to controls but this peak was not identified. Inthe serum samples the m/z peaks at 5836, 7763, 1724 7966 and8137 were duplicated on different chip surfaces but each of the fourchip types were involved in the pairing. Thus it would not bepossible to reduce the number of chips because of the sameproteins being easily identified on all chips.

The level of raw information from each spectrum varies withchip type and high-abundant proteins removal. Peaks common toboth serum and depleted serum have variable intensities. Someproteins become much more clearly defined when depletedserum is analysed, as can be seen in Fig. 2, while others are moreeasily recognised in the serum samples. It is particularly inter-esting that the discrimination between patient groups wasdiminished for m/z > 10 kDa proteins in serum but proteins inthis range showed significance in the depleted serum samples.Binding of these proteins to the larger proteins such as albumin

sitivity/Specificityiscovery set

Classificationgroups

Sensitivity/Specificityof validation set

CA19.9 þ markers

2 N/D N/D N/D

ND AUCc:0.66 N/D

/100v. HV)/88.5v. DC)

81.2/85.7(PC v. DC)

82.8/92.2(PC v. HV)82.8/88.9(PC v. DC)

85.4/96.2(PC v. DC)

/97.6 ND AUC: 0.833(PC v. HV)AUC:0.829(PC v. DC)

AUC: 1.00m/z3884 þ CA19.9

: 0.90v. HV)/84.5v. DC)

93% PC84% HV71% DC

AUC: 0.86(PC v. HV)AUC 0.68(PC v. DC)

AUC:0.96(PC v. HV)0.90(PC v. DC)

0.978v HV)/94.4

97.2%/94.4% (PC v HV)90.9%/91.1%

100% identified

DC) 65%/98% (PCvDC)62%/89%

PC 96.4%CP 56.8%

Diabetes 100%PC 76.7%CP 96.8%

A. Xue et al. / Pancreatology 12 (2012) 124e129 129

and ApoA-II increases the variability of measurement andpresumably reduces the precision of measurement. Despite thismany peaks are differentially expressed between the threegroups.

A number of studies using SELDI-TOF MS have identifiedbiomarkers which discriminate the serum from controls from thatof pancreatic cancer patients. However, to gain a reasonable diag-nostic reliability multiple marker panels which include CA 19-9[Ehmann et al., 2007; Guo et al., 2009; Honda et al., 2005; Koop-mann et al.,2004; Navaglia et al., 2009] [1,2,6,8,16,17] are required.Different chip surfaces have been used by these studies where[Koopmann et al., 2004] used IMACCu2þ and weak cationicexchange Guo et al.[Guo et al., 2009] used strong anion-exchangeprotein chips Yu et al.[Yu et al., 2005] Ns Honda et al usedIMACCu2þ but there is a significant variation in the protein panelsdiscovered (Table 4). In the study by Ehmann et al where IMAC-Cuand CM10 chips were used a three protein panel was identified andthe proteins were further identified as ApoA-II, ApoA-I and trans-thyretin[Ehmann et al., 2007]. A further study from our group (Xueet al.) using the H50 chip identified ApoA-II and ApoC-I as impor-tant discriminating proteins in serum. The proteins included inthese studies have some consistencies about them but there area number of differences which make interpretation difficult. It is ofinterest that these studies have indicated that there are consistentchanges in serum and that the characterisation of these changesmay have important implications for improving the diagnosis ofpancreatic cancer and improving our understanding of the processof pancreatic cancer.

Apolipoproteins have been shown to be important diagnosticbiomarkers for PCa in this and four other publications [1,5,9,10]indicating that they are involved with an important biologicalprocess in pancreatic cancer. Pancreatic cancer grows rapidly but isnot sufficiently for 18Fluorodeoxyglucose positron emissiontomography to reliably diagnose primary cancers [18] becauseglucose utilization isn’t greater in the tumour than the adjacenttissue. Alternatively there is evidence that pancreatic cancer is lipidavid [19] and that there are lipoprotein receptors in cancer cellswhich increase the uptake of lipid [20]. Lipoproteins may also besubstrates for metalloproteinases in the neoplastic stroma. Thismay be an important finding because it opens the possibility oftargeting pancreatic cancer in the future.

Only 55 protein peaks were present on more than one chipwhen the results of the four chips were examined, although 212were present in serum and 199 in depleted serum. It is interestingthat in the depleted serum samples the range of m/z for discrimi-nating proteins ranged from 3000 kDa to 30,000 kDa while inserum the proteins which were greater than 10,000 kDa were notdiscriminatory. The larger protein size is most likely to be influ-enced by the removal of the seven most abundant proteins. Thus,comparison of DC and PC results from depleted serum found 8discriminating proteins while there were only 2 in serum samples.Thus although there were similar numbers of peaks to study thedepleted serum was more likely to demonstrate differencesbetween groups. Although the study was too small to undertakemultivariate analysis and to define statistically robust biomarkerpanels, it demonstrates the importance of serum preparation if themaximal benefit is to be obtained.

The highly significant peaks observed in Q10, CM10 and H50are consistent with the finding of Apolipoprotein A-I (m/z

27,910e28000) and A-II (m/z 8758 and 17,232) in a larger study atour institution, and by other groups (Table 4). In this study,a number of new protein peaks were found such as m/z 4280 and7763 with IMAC-Cu arrays in crude serum samples, particularlyprotein peak at m/z 9285 with IMAC-Cu arrays and m/z 2304 and9412 with H50 arrays in depleted serum samples. It demonstratesthat using multiple chip arrays and immune-depleted serum mayresult in revealing more novel detection markers for pancreaticcancer.

References

[1] Honda K, Hayashida Y, Umaki T, Okusaka T, Kosuge T, Kikuchi S, et al. Possibledetection of pancreatic cancer by plasma protein profiling. Cancer Res 2005;65(22):10613e22.

[2] Koopmann J, Zhang Z, White N, Rosenzweig J, Fedarko N, Jagannath S, et al.Serum diagnosis of pancreatic adenocarcinoma using surface-enhanced laserdesorption and ionization mass spectrometry. Clin Cancer Res 2004;10(3):860e8.

[3] Rosty C, Goggins M. Early detection of pancreatic carcinoma. Hematol OncolClin North Am 2002;16(1):37e52.

[4] Yu Y, Chen S, Wang LS, Chen WL, Guo WJ, Yan H, et al. Prediction ofpancreatic cancer by serum biomarkers using surface-enhanced laserdesorption/ionization-based decision tree classification. Oncology 2005;68(1):79e86.

[5] Conrads TP, Hood BL, Issaq HJ, Veenstra TD. Proteomic patterns as a diagnostictool for early-stage cancer: a review of its progress to a clinically relevant tool.Mol Diagn 2004;8(2):77e85.

[6] Ehmann M, Felix K, Hartmann D, Schnolzer M, Nees M, Vorderwulbecke S,et al. Identification of potential markers for the detection of pancreatic cancerthrough comparative serum protein expression profiling. Pancreas 2007;34(2):205e14.

[7] Fiedler GM, Leichtle AB, Kase J, Baumann S, Ceglarek U, Felix K, et al. Serumpeptidome profiling revealed platelet factor 4 as a potential discriminatingPeptide associated with pancreatic cancer. Clin Cancer Res 2009;15(11):3812e9.

[8] Guo J, Wang W, Liao P, Lou W, Ji Y, Zhang C, et al. Identification of serumbiomarkers for pancreatic adenocarcinoma by proteomic analysis. Cancer Sci;2009.

[9] Takano S, Yoshitomi H, Togawa A, Sogawa K, Shida T, Kimura F, et al. Apoli-poprotein C-1 maintains cell survival by preventing from apoptosis inpancreatic cancer cells. Oncogene 2008;27(20):2810e22.

[10] Xue A, Scarlett CJ, Chung L, Butturini G, Scarpa A, Gandy R, et al. Discovery ofserum biomarkers for pancreatic adenocarcinoma using proteomic analysis.Br J Cancer 2010;103(3):391e400.

[11] Kawada N. Cancer serum proteomics in gastroenterology. Gastroenterology2006;130(6):1917e9.

[12] Tu C, Rudnick PA, Martinez MY, Cheek KL, Stein SE, Slebos RJ, et al. Depletionof abundant plasma proteins and limitations of plasma proteomics.J Proteome Res 2010;9(10):4982e91.

[13] Ernoult E, Bourreau A, Gamelin E, Guette C. A proteomic approach for plasmabiomarker discovery with iTRAQ labelling and OFFGEL fractionation. J BiomedBiotechnol 2010;2010:927917.

[14] Ye H, Sun L, Huang X, Zhang P, Zhao X. A proteomic approach for plasmabiomarker discovery with 8-plex iTRAQ labeling and SCX-LC-MS/MS. Mol CellBiochem 2010;343(1e2):91e9.

[15] Seam N, Gonzales DA, Kern SJ, Hortin GL, Hoehn GT, Suffredini AF. Qualitycontrol of serum albumin depletion for proteomic analysis. Clin Chem 2007;53(11):1915e20.

[16] Shin S, Cazares L, Schneider H, Mitchell S, Laronga C, Semmes OJ, et al. Serumbiomarkers to differentiate benign and malignant mammographic lesions.J Am Coll Surg 2007;204(5):1065e71.

[17] Navaglia F, Fogar P, Basso D, Greco E, Padoan A, Tonidandel L, et al. Pancreaticcancer biomarkers discovery by surface-enhanced laser desorption and ioni-zation time-of-flight mass spectrometry. Clin Chem Lab Med 2009;47(6):713e23.

[18] Delbeke D, Martin WH. PET and PET/CT for pancreatic malignancies. SurgOncol Clin N Am 2010;19(2):235e54.

[19] Wang F, Kumagai-Braesch M, Herrington MK, Larsson J, Permert J. Increasedlipid metabolism and cell turnover of MiaPaCa2 cells induced by high-fat dietin an orthotopic system. Metabolism 2009;58(8):1131e6.

[20] Cao WM, Murao K, Imachi H, Yu X, Abe H, Yamauchi A, et al. A mutant high-density lipoprotein receptor inhibits proliferation of human breast cancercells. Cancer Res 2004;64(4):1515e21.