Available online at www.sciencedirect.com

Clinical Biochemistry 43 (2010) 168–177

SELDI-TOF biomarker signatures for cystic fibrosis, asthma and chronicobstructive pulmonary disease

Patrícia Gomes-Alves a, Margaret Imrie b, Robert D. Gray b, Paulo Nogueira c, Sergio Ciordia d,Paula Pacheco e, Pilar Azevedo f, Carlos Lopes f, António Bugalho de Almeida f, Micaela Guardiano g,

David J. Porteous b, Juan P. Albar d, A. Christopher Boyd b,1, Deborah Penque a,⁎,1

a Laboratório de Proteómica, Departamento de Genética, INSA-IP, Av. Padre Cruz, 1649-016 Lisboa, Portugalb Medical Sciences (Medical Genetics), University of Edinburgh, Molecular Medicine Centre, Western General Hospital, Edinburgh, UK

c Departamento de Epidemiologia, INSA-IP, Lisboa, Portugald Laboratory of Proteomics, CNB-CSIC, Universidad Autónoma de Madrid, Madrid, Spaine Unidade de Biologia Molecular, Departamento de Genética, INSA-IP, Lisboa, Portugal

f Clínica Universitária de Pneumologia, Hospital Santa Maria, Lisboa, Portugalg Hospital São João, Universidade do Porto, Porto, Portugal

Received 5 March 2009; received in revised form 9 October 2009; accepted 11 October 2009Available online 20 October 2009

Abstract

Objectives: The aim of this work was to establish protein profiles in serum and nasal epithelial cells of cystic fibrosis individuals incomparison with controls, asthma and chronic obstructive pulmonary disease patients for specific biomarker signatures identification.

Design and methods: Protein extracts were analyzed by Surface Enhanced Laser Desorption/Ionization Time-Of-Flight Mass-Spectrometry(SELDI-TOF-MS).

Results: The mass spectra revealed a set of peaks with differential expression in serum and nasal cells among the different groups studied,resulting into peak signatures representative/specific of each pathology. Logistic regressions were applied to those peaks; sensitivity, specificity,Youden's indexes and area under the curve (AUC) of the respective receiver operating characteristic (ROC) curves were compared.

Discussion: Multivariate analysis demonstrated that combination of peaks has a better predictive value than the individual ones. These proteinsignatures may serve as diagnostic/prognostic markers for the studied diseases with common clinical features, or as follow-up assessment markersof therapeutic interventions.© 2009 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.

Keywords: CF; Asthma; COPD; Proteomics; SELDI-TOF-MS; Biomarkers signatures

Introduction

Cystic fibrosis (CF) is the most common lethal geneticdisorder in the Caucasian population [1]. Mutations in the cysticfibrosis transmembrane conductance regulator (CFTR) genedisrupt a cAMP-regulated chloride channel formed by theCFTR protein. CFTR is present at the apical plasma membraneof epithelial cells lining organs such as the sweat glands,airways and pancreas [2]. Lung disease in CF is characterized

⁎ Corresponding author. Fax: +351217526410.E-mail address: deborah.penque@insa.min-saude.pt (D. Penque).

1 Collaborated equally as supervisors.

0009-9120/$ - see front matter © 2009 The Canadian Society of Clinical Chemistsdoi:10.1016/j.clinbiochem.2009.10.006

by thick mucus, airways inflammation, chronic bacterialinfection and frequent exacerbations ultimately leading todeath from respiratory failure [3]. Chronic obstructive pulmo-nary disease (COPD) and asthma are common respiratorydiseases associated with chronic inflammation and structural“remodeling” inappropriate to the maintenance of normal lungfunction [4], possibly due to complex interactions betweenenvironmental and genetic factors. Furthermore, the CFTR genehas been described as a putative genetic risk factor for thesediseases [5]. Therefore, COPD and asthma are interestingcomparators in CF proteomics biomarker discovery studies.

Protein expression profiling using SELDI-TOF-MS has beenused to study novel biomarkers in several diseases namely in CF

. Published by Elsevier Inc. All rights reserved.

169P. Gomes-Alves et al. / Clinical Biochemistry 43 (2010) 168–177

and other pulmonary diseases [6–8]. SELDI-TOF-MS is an arraybased technology combining MS with affinity chromatographyto isolate a subset of proteins from complex biological mixturesaccording to their physicochemical properties [6]. The generatedspectra allow the comparison of differentially expressed proteinsamong different samples [9], providing characteristic diagnosticpatterns that distinguish pathological states [10–13]. Thismethodology provides semi-quantitative MS evaluation ofionized proteins peaks rather than the qualitative evaluationthat is traditionally used in MS methods [14].

In the present study, due to the difficulties and risk of samplingtissue from the lower airway, human nasal epithelial cells wereobtained by nasal brushing (NB) and their proteins evaluated bySELDI-TOF-MS.Nasal respiratory epithelium is representative ofthe airway system. Epithelial cells represent approximately 80%–95% of nasal brushings cellular content and CFTR expression canbe detected in both CF and healthy individuals [15,16]. Nasalepithelium also displays characteristic CF ion transport abnor-malities [17,18]. We utilized serum in this investigation as itcontains high protein concentrations which may contain proteinspecies indicative of diseased states, resulting from blood constantperfusion and percolation throughout the body organs [19].Moreover, several other authors have used serum as the electedbody fluid for searching protein biomarkers through SELDI-TOFmethodology, apparently without any caveat of using serum andnot plasma. As suggested by different manuscripts in proteomicsarea, the crucial point in serum/plasma analysis is the use of astandardized sampling, handling and storing procedures [20]. Toensure uniformity in our study, all serum and NB samples wereequally collected, processed and subjected to the same steps priorSELDI-TOF analysis.

The objective of this work was to establish SELDI-TOF-MSto unravel biomarkers patterns of CF in serum and nasalepithelial cells, in a comparative study with COPD, asthma andhealthy subjects. This is the first time to our knowledge thatserum and/or nasal mucosa cells from three different groups ofpatients (CF, asthma and COPD) sharing common chronic lungdisease features were processed and analyzed in parallel toidentify significant specific SELDI derived signatures for thesediseases. As pointed by other authors [21], it is crucial tocompare diseases with common clinical elements in order toenhance biomarkers specificity for the disease studied and notonly to distinguish sick from healthy states.

Design and methods

Subjects and sample collection

This project was approved by ethical committees of INSA,Hospital Santa Maria (HSM) and Hospital São João (HSJ). Afterinformed consent, peripheral blood and/or epithelial cells ofnasal mucosa were collected from patients and healthy subjects.

To ensure uniformity in our study, all serum samples wereequally collected (the same tubes were used and a clinicspecialist collected the blood), processed and subjected to thesame steps prior SELDI-TOF analysis. The same type ofconcerns was held for NB samples.

The majority of the samples were collected in HSM; only afew (CF) patients were from HSJ.

Peripheral blood was collected to both EDTA-tubes, forCFTR genotyping and serum/gel-tubes. A total of 125 sera werecollected: 26 CF, 29 asthma, 16 COPD and 54 healthy controls(all samples were processed in 4 h maximum after collection,the same protocol was used and they were all stored at the sameconditions). All patients were clinically stable and recruitedfrom specialist respiratory clinics in HSM and/or HSJ.

Nasal epithelial cells were isolated by NB technique aspreviously described [15,22]. In brief, cells were collected bybrushing the inferior turbinate and adjacent lateral nasal wallwith an interdental brush (Paro-Isola, Switzerland). Cells werepelleted (3000 rpm in a Picofuge, Stratagene, USA) in ice coldPBS, snap frozen and stored at −80 °C. A total of 49 NBsamples were collected: 14 CF, 20 asthma and 15 healthycontrols (samples were all collected by the same technician andsamples were immediately processed in loco).

SELDI-TOF-MS

Forty microliters of serum was denatured with 60 μL ofdenaturing buffer (9 M Urea, 2% CHAPS, in 50 mM Tris–HCl,Merck). Denatured sera were subjected to SELDI-TOF analysiswith CM10 (cationic exchange surface) (Ciphergen, Freemont,USA) arrays at pH 4. In brief, 10 μL of denaturated serum wasapplied to pre-activated CM10 proteinchip arrays. All chips weretreated with sinapinic acid (Ciphergen, Freemont, USA) matrix(2×0.8 μL/spot) and allowed to air dry. Samples were analyzedon a Protein Biology System-IIc SELDI-TOFmass spectrometer(Ciphergen, Freemont, USA) with a laser intensity of 215 withdeflector set at 5000 Da and a focus mass of 26,500 Da.

NB cell pellets were lysed in 100 μL of PBS (25 mlDulbecco's-Gibco PBS with a protease inhibitor cocktail tablet-Roche, Switzerland). Cells were kept on ice for 10 min and thenvortexed for 5 min and centrifuged 10 min at 13,000 rpm, 4°C(Heraeus BiofugeFresco, Germany). Supernatants were collect-ed and quantified using the Micro BCA Protein Assay kit(Pierce, USA). All samples were normalized for an optimizedconcentration of 250 μg/mL. Two chromatographic chipsurfaces and two specific binding conditions were chosen forSELDI analysis of this type of cells: a weak cation exchange atpH 4 (CM10) and a strong anion exchange at pH 10 (Q10).Twenty microliters of sample was added to CM10 and Q10chips in a bioprocessor (Ciphergen, Freemont, USA). Follow-ing 2 h of incubation, several washes were done with therespective buffer (pH 4 or pH 10) in order to remove salts andunbound proteins. Chips were read with a laser intensity of 205with a deflector set at 3000 Da and a focus mass of 11,000 Da.

Data analysis

Mass spectral profiles from each set of experiments wereinitially analyzed on Ciphergen ProteinChip® Software(v3.2.1.1216, Biomarker Edition, Ciphergen, Freemont, USA)to perform clustering and data analysis. Baseline correction wasperformed to enhance the contrast of peaks to baseline using

Table 1Subjects demographic and smoking behavior.

Group Nasal epithelial cells Serum

n Age (years) (SD) Sex Smokers n Age (years) (SD) Sex Smokers

Control 15 29.3 (7.7) 9 F/6 M 3 54 41.6 (14.5) 33 F/21 M 15CF 14 27.2 (7.9) 11 F/3 M 1 26 23.5 (9.8) 14 F/12 M 0Asthma 20 27.8 (8.3) 15 F/5 M 0 29 34.4 (12.1) 21 F/8 M 2COPD – – – – 16 63.2 (5.8) 4 F/12 M 16

Table 2Number of peaks detected by SELDI-TOF and respective differential clustersrevealed by clustering analysis.

No. of peaks NBa Serumb

Q10 pH 10 CM10 pH 4 CM10 pH 4

Total 232 114 137pb0.05 45 46 119pb0.01 23 34 108pb0.001 12 23 95N20,000 Da 1 8 35a Controls vs. asthma vs. CF (n=15 vs. n=20 vs. n=14).b Controls vs. asthma vs. COPD vs. CF (n=54 vs. n=29 vs. n=16 vs. n=26).

170 P. Gomes-Alves et al. / Clinical Biochemistry 43 (2010) 168–177

fitting width of 4 times the expected width; noise wasautomatically measured from 4 to 50 kDa and spectra correctedaccordingly. Data were then normalized for total ion current. Datawith normalization coefficients higher than 2 were identified (anormalization coefficient of 1 reflected individual AUC the sameas the average, b1 suggests a greater AUC than average, and N1 asmaller AUC than average) and examined in order to evaluatespot to spot variability in chip surface binding. Respectiveindividual spectra were excluded whenever they representedpoor quality (e.g., spectrum with no peaks, high noise and forthat not comparable with the other spectra). Automatic peakdetection and clustering were performed and peaks with asignal to noise ratio N3 were selected to perform analysis withinand between groups. Peaks of a similar inferred molecularweight (within 0.3%) were clustered allowing the comparisonacross groups.

Statistical analysis

Statistical analysis between groups was performed using theKruskal-Wallis test. Values of pb0.05 were assumed assignificant. Dunn's Multiple Comparison Test was performedon all significant peaks to investigate differences betweengroups using GraphPad Prism (v.4.02 GraphPad software).Multivariate analysis (logistic regression, Hosmer-Lemeshowgoodness-of-fit) was performed for all significant peaks onSPSS (v.13.0 SPSS for Windows).

Logistic regression was employed to adjust data to amathematical model that simplified its interpretation; a categor-ical variable was predicted from a set of predictor variables.Predicted dependent variable is a function of the probability that aparticular subject will be in one of the categories (disease groups).Models with OR (odd-ratio) p valuesb0.2 were selected aspotential candidates [23]. Sensitivity, specificity and the AUC(area under the curve) of the ROC curves were also measured. Avalidation set and a training set were randomly created from eachgroup of samples to complement the evaluation of the “diagnosticaccuracy” of each individual peak or peak signature. For NBsamples, 6 individuals per group were used for the validation set;for serum samples, as the different groups have very distinctnumbers of individuals, 12 controls, 8 asthma, 8 CF and 6 COPDwere used to compose the respective validation sets. Theremaining samples of each group (both in NB and serum)constituted the particular training sets. True positives, truenegatives, false positives (FP), false negatives (FN), sensitivity,specificity and Youden's index of the validation sets werecalculated. Youden's index (sensitivity+specificity−1) estimatesthe overall test accuracy (cut-off established=0.38).

Protein identification

Two peaks from each sample type (NB and serum) presentinga significant p value (pb0.05) in clustering and assessed onrelative abundance and quality in the original spectra wereselected for protein identification. Pooled samples of each groupwere utilized. Protein purification and identification wereperformed as described by Miguet et al. [24]. After eachpurification step, fractions were evaluated on CM10 or goldchips (to evaluate the protein captured without any selectivity)(Ciphergen, Freemont, USA). Fractions containing the highestratio of proteins of interest were enriched byReverse Phase Beads(PLRP-S 300 Å Beads, ScharLab), dried and resuspended inSample Buffer, loaded on a 16% tricine-gel [25] and stainedovernight using colloidal coomassie blue (Safe-stain, Invitrogen)or silver staining.Gel bands of appropriatemolecular weight wereexcised. Half of the band was destained and protein eluted fromthe gel (passive elution, described in [24]); the other half wasdigested with trypsin (Promega, Madison, WI, USA) for MS/MSanalysis on a MALDI-TOF/TOF (4800Plus MALDI-TOF/TOF™ Analyzer, Applied Biosystems). Protein identificationwas accepted when pb0.05. One missed cleavage per peptidewas allowed.A peptide tolerance of 100 ppm and a fragmentmasstolerance of 0.8Da were considered and Cys carbamidomethyla-tion and Met oxidation were set as fixed and variable amino acidmodifications, respectively. A taxonomic restriction to Homosapiens sequences was included (NCBInr: total—5401388sequences, Homo sapiens—187188 sequences).

Results

Subjects and demographics

Patient demographics and smoking habits are given inTable 1. All CF patients have two identified CFTR mutations.

Fig. 1. SELDI-TOF-MS spectra at three different magnifications. Representation of denaturated serum spectra obtained on a CM10 pH 4 chip, individual spectrum per group.

171P.Gom

es-Alves

etal.

/Clinical

Biochem

istry43

(2010)168–177

one

172 P. Gomes-Alves et al. / Clinical Biochemistry 43 (2010) 168–177

Two patients with asthma and one with COPD present one CFmutation. All controls are apparently healthy individuals withno CF mutations (Supplementary Table 1).

SELDI-TOF

Profiling analysis revealed a large number of peaks thatdifferentiate between disease and healthy individuals both inserum samples and NB (Table 2). A total of 45 and 46 peakswere accessed with a p Kruskal–Wallisb0.05, respectively, inQ10 pH 10 and CM10 pH 4 NB analysis (control vs. CF vs.asthma). In serum, 119 peaks were significantly differentiallyexpressed in the CM10 pH 4 analysis, where controls, CF,asthma and COPD were compared.

Representative spectra from serum study are shown in Fig. 1(see NB spectra on Supplementary Fig 1). In serum, a total of 84clusters were considered relevant for additional statisticalanalysis based on Kruskal–Wallis analysis (pb0.05 and m/zbetween 4000 and 20,000).

Dunn's comparison test demonstrated 5 peaks with pb0.05differentiating only CF from controls; the most significantdifferentiating peak was at m/z=18,637, pb0.001 (Fig. 2).Other peaks (14 peaks with pb0.05) differentiated CF fromcontrols and from one of the other pulmonary pathologies, suchas peak m/z=15,887 (CF vs. controls pb0.001 and CF vs.asthma pb0.01). Others seem to differentiate better one of theother pathologies from controls but not CF; peak m/z=17,410differentiated asthma from controls with a pb0.01 and peak m/z=15,148 differentiated COPD from controls with a pb0.05(Table 3 and Supplementary Table 2).

In NB cells, Q10 pH 10 arrays presented 44 eligible peaks andCM10 pH 4 arrays presented 39 peaks, yielding a total of 83clustered protein peaks with a significant p value and m/zbetween 4000 and 20,000. As in serum, different “types” ofpeaks are observed: peaks that differentiate CF from controls andasthma (e.g., peak m/z=13,873 with a pb0.001 and pb0.05,respectively), peaks that differentiate asthma from both CF andcontrols (e.g., peak m/z=10,313 with a pb0.001 for bothcomparisons), others differentiate only controls from CF butnot asthma from either of them (e.g., peak m/z=9618 witha pb0.05), among other possible combinations (Table 3

Fig. 2. Relative intensities of protein cluster 18,637 Da in serum from control

and Supplementary Table 2). Fig. 3 illustrates the referredexamples.

Multivariate analysis of SELDI-TOF data

We applied logistic regressions to the clusters and comparedsensitivity, specificity and the AUC of the respective ROCcurves (Supplementary Table 3). Youden's index and thesignificance of the logistic regression were considered aspossible “checkpoints” to discriminate between good and badpotential biomarkers among peaks studied. In both NB andserum samples, we could find several peaks that satisfy the cut-offs established (Supplementary Table 3).

We observed that in general the combinations of two or morepeaks do improve the discrimination between groups. Of themost significant biomarkers/signatures, some are displayed inTable 4.

Comparisons of the training and validation sets revealed thatin the majority of the cases the validation set confirmed theperformance of individual peaks or peak signatures todiscriminate between two groups presenting equivalent valuesof sensitivity, specificity and Youden's indexes. Some cases,however, show some discordance between the two sets (Table 4and Supplementary Table 3).

Protein purification and identification

After an initial discovery phase to identify potential novelbiomarkers, the same instrument was used for the fractionationand purification of some proteins followed by MS analysis.

Some peaks were chosen for protein identification, namelypeaks with m/z 10,860 and 15,887 in serum (CM10 array), and8455 and 10,830 in NB cells (CM10 array). The purification/isolation protocol used was the same for both types of samples,only differing in the buffers used, accordingly with the surfacesemployed on SELDI-TOF profiling. Of the bands of possibleinterest (confirmed by passive elution and SELDI-TOF-MSanalysis, m/z≈15,887), only one yielded a positive ID(identification by MS/MS), identified as a hemoglobinsubunit-beta, with accession number gi|61679604 (NCBInr),sequence coverage 24%, mascot score 89 (pb0.05), and two

s, asthma, COPD and CF patients. Data obtained on a CM10 pH 4 assay.

Table 3Number of significant peaks (pb0.05) resultant from Dunn's test.

Serum NB

Comparisons No. of peaks ⁎ Comparisons No. of peaks ⁎

CM10 pH 4 CM10 pH 4 Q10 pH 10

CF vs. control (exclusive) 5 Control vs. asthma (exclusive) 5 17CF vs. control and asthma or COPD 14 Control vs. CF (exclusive) 2 6CF vs. control, asthma and COPD 42 Asthma vs. CF (exclusive) 4 5Asthma or COPD vs. control 6 Asthma vs. CF and control 15 13Asthma and COPD vs. CF 2 CF vs. asthma and control 8 0Asthma vs. COPD and/or CF 3Control vs. COPD and CF 2Other combinations 9

Dunn's test was applied to serum and NB peaks with a Kruskal-Wallis pb0.05 (see also Table 2 and Supplementary Table 1).⁎ pb0.05.

173P. Gomes-Alves et al. / Clinical Biochemistry 43 (2010) 168–177

ions were sequenced: LLVVYPWTQR (score 20) andEFTPPVQAAYQK (score 34).

Discussion

CF remains the most common lethal autosomal recessivegenetic disease without an effective treatment. By using aproteomics-based approach, we searched proteins that may beused to monitor the progression of disease and lead to moreeffective intervention strategies. We also looked for biomarkersof asthma and COPD, as these pathologies share great part oftheir symptomatic features with CF lung pathology.

Applying SELDI-TOF-MS technology, we were able togenerate both NB and serum proteomic profiles of differentpulmonary pathologies. These protein signatures revealedseveral possible biomarkers that differentiate, with statisticalsignificance, patients from controls; furthermore, it also allowed

Fig. 3. Relative intensity of different protein clusters in NB cells from controls, asthmDa) and on a Q10 pH 10 assay (9618 Da).

distinguishing CF from asthma and COPD patients. This high-throughput technology allows the screening of a large panel ofbiomarker candidates in biological samples, contrasting withanalyses where predefined proteins are investigated. As aconsequence of the differential protein binding to thechromatographic surfaces, low abundance proteins can also beaccessed and their relative expression compared betweensamples. The small amount of sample required for SELDI-TOF analysis is also a great advantage of this technology.

Several studies have described before the application of thistechnology to the analysis of different biological samples,namely urine [26], serum [26–28], plasma [29], sputum [8] andBALF [7], among others, for diagnosis in numerous and diversepathologies. Our study is the first to describe serum profiling inCF, as well as its comparison with asthma and COPD. The useof serum instead of plasma does not present any criticaldisadvantage in our study as we are not doing a peptidomic

a and CF patients. Data obtained on a CM10 pH 4 assay (13,873 Da and 10,313

Table 4Examples of multivariate analysis of SELDI-TOF generated protein clusters in serum and NB.

174P.Gom

es-Alves

etal.

/Clinical

Biochem

istry43

(2010)168–177

Comparisons between all groups are presented (FN, false negatives; FP, false positives; AUC, area under the curve).a Cut-offs are sets where sensitivity and specificity are given equal importance.

175P.Gom

es-Alves

etal.

/Clinical

Biochem

istry43

(2010)168–177

176 P. Gomes-Alves et al. / Clinical Biochemistry 43 (2010) 168–177

analysis of human blood specimens [30]. We believe that theprofile of peaks we analyzed is mainly composed of proteins,not peptides, and as so, the potentially clotting resultantpeptides if retained would not mask the range of peak signaturesin which we are focused (4–20 kDa, see below).

We also used nasal epithelial cells obtained by nasalbrushing a non-invasive method that allows the easy samplingof cells from the superficial respiratory mucosa which isrepresentative of the airway system, to evaluate proteinsignatures in CF, asthma and control subjects. AlthoughBALF [7] and sputum [8] profiling in CF has been alreadydescribed, the present study and the samples investigated canconstitute a valid complement to the understanding of lungpathology in these chronic pulmonary diseases.

Our data revealed large numbers of potential biomarkerseven applying very restrictive conditions in data analysis.Previous studies performed by our group [7] suggested thatpeaks below 4 kDa should be excluded as a means ofguarantee that spectra would not include any matrix/back-ground noise peaks. As also described, peaks above 20 kDaare not well discriminated by SELDI-TOF-MS. An optimummass range of 4–20 kDa can represent, on the other hand, thelost of some relevant proteins, both in low and high molecularweight range. Then again, the number of differentiating peaksidentified can also be influenced by the fact that some peaksrepresent double charged proteins (appear twice in the m/zspectra), some proteins can be detected in more than onesurface (in the case of NB samples) and fragments of the sameprotein can correspond to different peaks of the spectrum. Alltogether, it can explain the high numbers of peaks evaluated.To ensure uniformity in our study, all samples were equallyprepared and generated data subjected to the same steps priorto analysis.

Although our data revealed a large number of possiblebiomarkers in these pulmonary pathologies, its interpretation isnot simple and must be very cautious.

We only identified one protein in serum, hemoglobinsubunit-beta (m/z=15,887). Due to their low MW and lowconcentration, proteins were visualized in all SELDI-TOFscreenings but not in 1D gels, confirming that SELDI-TOF is amore sensitive technology. Alternatives to tricine-gels are beingevaluated, namely AOT-gels that seem to better resolve proteinsbelow 10 kDa.

Hemoglobin subunit-beta (beta-Hb) has already beendescribed as being differentially expressed in serum of ovariancancer patients [27]. In our study, beta-Hb is differentiallyexpressed in CF patients in comparison to controls and asthmaindividuals suggesting erythrocytes lysis and beta-Hb releasedue to sample handling, but we previously discarded all samplesthat presented hemolysis to reduce this impairment. Studiesdemonstrated that oxidative stress in erythrocytes and theirmembranes may result in hemolysis [31]. CF patients have beendescribed to have oxidative stress imbalance, and our results cantranslate an increased susceptibility of CF erythrocytes tooxidative injury [31,32].

SELDI-TOF-MS allows detection of multiple biomarkerpatterns to characterize complex diseases rather than only

individual markers. Some reported studies propose that proteinprofiles can be used as potential diagnostic tests [27,33], or alsoas a signature for early assessment of therapeutic efficacywithout any protein identification [26]. Moreover, when used asprotein signatures, these biomarkers improve sensitivity andspecificity in the diagnosis of different illness states [26,27,34].The statistical analysis performed in the present studydemonstrates that the combination of biomarkers can in facthave a better predictive value than the individual markers.However, some of the selected biomarkers were found to be ofless value, i.e., they do not contribute to an enhancement in thedifferentiating power of peak signatures. Comparison betweengroups resulted in different combinations of peaks, revealingwhich are more representative of each pathology. We not onlysearched for patterns of biomarkers that differentiate CF fromcontrol healthy individuals, but we also sought for proteinsignatures that discriminate CF lung disease from asthma andCOPD.

The combination of potential biomarkers to a specific patternassociated with CF in comparison with asthma, COPD andhealthy condition provided a set of results that should define CFspecific proteome (in serum and NB cells) with high precision.Based on the preferential combinations proposed, furtherstudies on a larger number of samples, namely NB cells, willconsolidate the statistical significance of our results. COPD NBanalysis is also of utmost importance to complete the study. Asmentioned before, comparisons of the training and validationset results yielded a great concordance between them. However,the relative reduced number of samples constituting the sets,namely the validation set, can influence the statistical power ofsome results. We tried to choose an equilibrated experimentdesign and consequent data analysis, given the limitation insamples availability, which would give us the perception of thediagnostic accuracy of the peaks or patterns of peaks chosen in arandom group of individuals.

Although not critical for establishment of valid signaturepatterns, biomarker identification is important for increasing ourbiological knowledge about CF lung disease and also in termsof following measurements using other technologies. Therefore,it is our intention to progress with identification experimentsand define a robust and reliable protein pattern that specificallydifferentiates CF from healthy individuals and other chronicpulmonary diseases. Nowadays, the difficulties in CF diagnosisand prognosis have been concerned with the atypical forms ofthe disease and with the incapacity of predicting the severity oflung disease progression. Although lung disease is almost fullypenetrant late in advanced stages of CF, its expressivity is themost heterogeneous of all CF related phenotypes and severityand progression are highly variable and only partially correlatedwith the CFTR genotype. Recurrent infections and inflamma-tion responses trigger the destruction of the airways and acomplex spectrum of other factors may contribute to progres-sion and severity of lung disease. Once such factors areidentified, allowing the monitoring of CF progression, theultimate goal of SELDI proteomics is direct clinical application.Clinicians would subsequently be able to delineate risk factorsfor susceptible patients, predict how well a patient will respond

177P. Gomes-Alves et al. / Clinical Biochemistry 43 (2010) 168–177

to a particular treatment and monitor improvements madeduring the course of treatment. As so, biomarker patternsand subsequent identification should provide the basis fordevelopment of assays that allow prediction of lung diseaseprogression and optimization of therapeutic intervention in CF,asthma and COPD.

Acknowledgments

Work supported by FCT/FEDER research grants POCTI/MGI/40878/2001 and POCI/SAU-MMO/56163/2004, FCT/Poly-Annual Funding Program and FEDER/Saúde-XXI-Pro-gram (Portugal). PGA is a recipient of FCT-PhD fellowshipSFRH/BD/17744/2004.

Appendix A. Supplementary data

Supplementary data associated with this article can befound, in the online version, at doi:10.1016/j.clinbiochem.2009.10.006.

References

[1] Collins FS. Cystic fibrosis: molecular biology and therapeutic implica-tions. Science 1992;256(5058):774–9.

[2] Rosenstein BJ, Zeitlin PL. Cystic fibrosis Lancet 1998;351(9098):277–82.[3] Welsh MJ, Smith AE. Cystic fibrosis Scientific American 1995;273(6):

52–9.[4] Jeffery PK. Remodeling in asthma and chronic obstructive lung disease. Am

J Respir Crit Care Med 2001;164(10 Pt 2):S28–38.[5] Tiddens H, Silverman M, Bush A. The role of inflammation in airway

disease: remodeling. Am J Respir Crit CareMed 2000;162(2 Pt 2):S7–S10.[6] Reddy G, Dalmasso EA. SELDI ProteinChip(R) Array Technology:

protein-based predictive medicine and drug discovery applications. JBiomed Biotechnol 2003;2003(4):237–41.

[7] Macgregor G, Gray RD, Hilliard TN, et al. Biomarkers for cystic fibrosislung disease: Application of SELDI-TOF mass spectrometry to BAL fluid.J Cyst Fibros 2008.

[8] Gray RD, Macgregor G, Noble D, et al. Sputum proteomics ininflammatory and suppurative respiratory diseases. Am J Respir CritCare Med 2008.

[9] Petricoin EF, Liotta LA. SELDI-TOF-based serum proteomic patterndiagnostics for early detection of cancer. Curr Opin Biotechnol 2004;15(1):24–30.

[10] Lebrecht A, Boehm D, Schmidt M, Koelbl H, Grus FH. Surface-enhancedlaser desorption/ionisation time-of-flight mass spectrometry to detectbreast cancer markers in tears and serum. Cancer genomics and proteomics2009;6(2):75–83.

[11] Seibert V, Wiesner A, Buschmann T, Meuer J. Surface-enhanced laserdesorption ionization time-of-flight mass spectrometry (SELDI TOF-MS)and ProteinChip technology in proteomics research. Pathol Res Pract2004;200(2):83–94.

[12] Engwegen JY, Mehra N, Haanen JB, et al. Validation of SELDI-TOF MSserum protein profiles for renal cell carcinoma in new populations. LabInvestig J Tech Methods Pathol 2007;87(2):161–72.

[13] Han KQ, Huang G, Gao CF, et al. Identification of lung cancer patients by

serum protein profiling using surface-enhanced laser desorption/ionizationtime-of-flight mass spectrometry. Am J Clin Oncol 2008;31(2):133–9.

[14] Penque D. Proteomic biomarker discovery for the monogenic diseasecystic fibrosis. Expert Rev Proteomics 2007;4(2):199–209.

[15] Penque D, Mendes F, Beck S, et al. Cystic fibrosis F508del patients haveapically localized CFTR in a reduced number of airway cells. Lab InvestigJ Tech Methods Pathol 2000;80(6):857–68.

[16] Carvalho-Oliveira I, Efthymiadou A, Malho R, et al. CFTR localization innative airway cells and cell lines expressing wild-type or F508del-CFTRby a panel of different antibodies. J Histochem Cytochem 2004;52(2):193–203.

[17] Grubb BR, Pickles RJ, Ye H, et al. Inefficient gene transfer by adenovirusvector to cystic fibrosis airway epithelia of mice and humans. Nature1994;371(6500):802–6.

[18] Davies MG, Geddes DM, Alton EW. The effect of varying tonicity onnasal epithelial ion transport in cystic fibrosis. Am J Respir Crit Care Med2005;171(7):760–3.

[19] Adkins JN, Varnum SM, Auberry KJ, et al. Toward a human blood serumproteome: analysis by multidimensional separation coupled with massspectrometry. Mol Cell Proteomics 2002;1(12):947–55.

[20] Hsieh SY, Chen RK, Pan YH, Lee HL. Systematical evaluation of theeffects of sample collection procedures on low-molecular-weight serum/plasma proteome profiling. Proteomics 2006;6(10):3189–98.

[21] Cottingham K. Basic and clinical proteomics researchers: the great divide?J Proteome Res 2008;7(3):840–3.

[22] Beck S, Penque D, Garcia S, et al. Cystic fibrosis patients with the 3272-26A-NG mutation have mild disease, leaky alternative mRNA splicing,and CFTR protein at the cell membrane. Human Mutat 1999;14(2):133–44.

[23] Hosmer DW, Lemeshow D. Applied logistic regression. John Wiley andSons; 1989.

[24] Miguet L, Bogumil R, Decloquement P, et al. Discovery andidentification of potential biomarkers in a prospective study of chroniclymphoid malignancies using SELDI-TOF-MS. J Proteome Res 2006;5(9):2258–69.

[25] Schagger H. Tricine-SDS-PAGE. Nat Protoc 2006;1(1):16–22.[26] Moreno S, Ibraghimov-Beskrovnaya O, Bukanov NO. Serum and urinary

biomarker signatures for rapid preclinical in vivo assessment of CDKinhibition as a therapeutic approach for PKD. Cell cycle (Georgetown, Tex2008;7(12).

[27] Kozak KR, Su F, Whitelegge JP, Faull K, Reddy S, Farias-Eisner R.Characterization of serum biomarkers for detection of early stage ovariancancer. Proteomics 2005;5(17):4589–96.

[28] Yang SY, Xiao XY, Zhang WG, et al. Application of serum SELDIproteomic patterns in diagnosis of lung cancer. BMC cancer 2005;5:83.

[29] Bowler RP, Canham ME, Ellison MC. Surface enhanced laser desorption/ionization (SELDI) time-of-flight mass spectrometry to identify patientswith chronic obstructive pulmonary disease. Copd 2006;3(1):41–50.

[30] Tammen H, Schulte I, Hess R, et al. Peptidomic analysis of humanblood specimens: comparison between plasma specimens and serum bydifferential peptide display. Proteomics 2005;5(13):3414–22.

[31] Benabdeslam H, Abidi H, Garcia I, Bellon G, Gilly R, Revol A. Lipidperoxidation and antioxidant defenses in cystic fibrosis patients. ClinChem Lab Med 1999;37(5):511–6.

[32] Laskowska-Klita T, Chelchowska M. Antioxidant status in erythrocytes ofcystic fibrosis children. Acta Biochim Pol 2001;48(1):283–5.

[33] Petricoin EF, Ardekani AM, Hitt BA, et al. Use of proteomic patterns inserum to identify ovarian cancer. Lancet 2002;359(9306):572–7.

[34] Zinkin NT, Grall F, Bhaskar K, et al. Serum proteomics and biomarkersin hepatocellular carcinoma and chronic liver disease. Clin Cancer Res2008;14(2):470–7.