J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

Ava i l ab l e on l i ne a t www.sc i enced i r ec t . com

www.e l sev i e r . com/ loca te / j p ro t

Quantitative in-depth analysis of the dynamic secretome ofactivated Jurkat T-cells

Elena Bonzon-Kulichenkoa, Sara Martínez-Martínezb, Marco Trevisan-Herraza,Pedro Navarroc, Juan Miguel Redondob, Jesús Vázqueza, b,⁎aLaboratory of Cardiovascular Proteomics, Centro de Biología Molecular “Severo Ochoa” (CSIC-UAM), E-28049, Madrid, SpainbCentro Nacional de Investigaciones Cardiovasculares, E-28029, Madrid, SpaincETH Zurich, Institute of Molecular Systems Biology, HPT C75, 8093 Zurich, Switzerland

A R T I C L E I N F O

⁎ Corresponding author at: Laboratory of CardiovaE-mail address: jvazquez@cbm.uam.es (J.

1874-3919/$ – see front matter © 2011 Elseviedoi:10.1016/j.jprot.2011.08.022

A B S T R A C T

Article history:Received 2 March 2011Accepted 24 August 2011Available online 1 September 2011

Proteins secreted by cells are of the highest biomedical relevance since they play asignificant role in the progression of numerous diseases. However, characterization of theproteins specifically secreted in response to precise stimuli is challenging, since theseproteins are contaminated by cellular byproducts. Here we present a method tocharacterize a dynamic secretome and demonstrate its utility by performing the deepestquantitative analysis to date of proteins secreted by lymphoid Jurkat T-cells uponactivation. Cell-free supernatant proteins were analyzed by using an optimized protocolfor differential 18O/16O-labeling and LC–MS/MS, followed by statistical analysis using arandom-effects model. More than 4000 unique peptides belonging to 1288 proteins wereidentified and a large proportion could be quantified. To determine the proteins whosesecretion was up-regulated upon T-cell activation, protein variance of the null hypothesiswas estimated after protein classification in terms of secretion and ontology usingbioinformatic tools. 62 proteins showed a statistically significant change in abundanceupon cell activation and most of them (49 proteins) were up-regulated. These proteinswere functionally involved mainly in inflammatory response, signal transduction, cellgrowth and differentiation and cell redox homeostasis. Our approach provides apromising technology for the high-throughput quantitative study of dynamic secretomes.

© 2011 Elsevier B.V. All rights reserved.

Keywords:Jurkat T-cellsSecretionSecretomeMass spectrometryQuantitative proteomics18O labeling

1. Introduction

Mediators secreted from cells into the extracellular environ-ment play a significant role in intercellular communicationin multicellular organisms. In humans, about 800 proteinshave been identified in blood plasma and some of them areused as biomarkers of disease [1]. Particularly relevant exam-ples are the proteins secreted by activated lymphocytes,which play a significant role in the progression of several in-flammatory diseases. T-cell secretion could be inferred by

scular Proteomics, Centro deVázquez).

r B.V. All rights reserved.

the blood plasma proteomic profile of patients with chronicor acute lymphocytic leukemia [2–4]. However, blood plasmais very complex in nature, with a wide dynamic protein con-centration range, containing not only T-cell-derived proteins,but also proteins coming from other tissues. Hence, to identifyspecifically the T-cell-derived secretome, the characterizationof the proteins secreted by T-cells in culture is needed. Withthis regard, human Jurkat T leukemic cells are one of themost important model systems for activated T cell study [5].Several authors have investigated the Jurkat T-cell proteome

Biología Molecular “Severo Ochoa” (CSIC-UAM), E-28049, Madrid, Spain.

562 J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

using mostly a two-dimensional gel electrophoresis ap-proach [6–8]. So far, the deepest analysis of the global JurkatT-cell proteome was performed using a combination of sub-cellular fractionation, multiple protein enrichment methodsand LIT-MS/MS, identifying 4928 proteins, which are mainlyintracellular [9]. Recently, the basal secretome of these cellswas characterized using one-dimensional SDS-PAGE and 70nano-LC–MS/MS runs performed on an LTQ-Orbitrap massspectrometer, identifying 2227 proteins [10]. However, noquantitative study has been performed to date on the dynam-ic changes taking place in the Jurkat T-cell secretome uponactivation.

Characterization of the secretome of cells in culture in-volves a number of technical difficulties. The first one isthe presence of contaminant proteins coming from theserum and from damaged cells (intracellular proteins), whichincrease the dynamic range of the sample and can potentiallyhinder the detection of actively secreted proteins. Additional-ly, the high number of proteins identified in a high-throughputexperiment makes necessary the use of bioinformatic tools todeterminewhether these proteins are potentially secretable ornot. This introduces some uncertainty on the results, since ex-perimental evidence demonstrating secretion exists for only30% of all proteins of the human proteome [11]. Moreover, se-cretion mechanisms of many proteins remain elusive [12]; be-sides the classical signal peptide-dependent pathway, at leastfour non-classical protein export routes have been described,which are taken by inflammatory cytokines [13], growth fac-tors [14,15], extracellular matrix components [16], transcrip-tion factors [17], chromatin-binding proteins [18] and parasitesurface proteins [19]. Thus, most bioinformatics tools mainlyrely on the sequence-based prediction of the physico-chemicalproperties thatwouldallowproteins to function inanextracellularenvironment [20]. With this regard, SecretomeP and TMHMM arethe most used and considered to be among the best-performingprograms [20].

In an attempt to develop novel tools for the study of the dy-namic secretome, in this work we present a method that in-cludes optimized sample preparation and bioinformaticanalysis of the data, and demonstrate its practical utility byperforming the deepest quantitative analysis to date of pro-teins secreted by Jurkat T-cells upon activation. A recently-developed quantitative proteomics method [21], which usesSDS solubilization, an improved 18O-labeling protocol, IEFpeptide fractionation and LIT-MS/MS analysis was used forhigh-throughput identification of proteins in the secretomeand at the same time specific determination of the subset ofproteins which showed alterations in their concentration inresponse to T-cell stimulation. The assignment of secretomeproteins specifically involved in activation was performed byusing a validated, null-hypothesis statistical model thattakes into account four independent error sources (fitting,scan, peptide and protein) [22]. This was done by a robust cal-culation of the experimental variance at the protein levelusing the intracellular protein population to determine thenull hypothesis distribution. This population was selected byusing several bioinformatic tools, such as the GO, SecretomePand TMHMM, that allowed classification of all identified pro-teins in terms of secretion. Our approach is potentially appli-cable to other kinds of samples and opens new opportunities

for the high-throughput, systematic study of the dynamicsecretome.

2. Materials and methods

2.1. Jurkat T-cell culture and activation

Jurkat T-cells were grown in RPMI (GIBCO, Invitrogen) con-taining 10% fetal calf serum (FCS, Sigma) supplementedwith L-glutamine plus antibiotics (100 units/ml penicillin and100 μg/ml streptomycin) until the number of 300×106 cells wasreached. Cells were washed three times with serum-free RPMIand left to culture at 1×106 cells/ml in RPMI containing 0.5%FCS. After 12 h, the reduced-serum containingmedia was elim-inated by three washes with PBS and replaced by RPMI withoutserum supplemented with 20 ng/ml phorbol 12-myristate 13-acetate (PMA) (Sigma) plus 1 μM calcium ionophore A23187 (Io)(Sigma) at 1×106 cells/ml. Control cells were cultured inserum-free media containing vehicle (DMSO). All cultureswere incubated at 37 °C and 5% CO2. After 8 h conditionedmedia from two 150 ml flasks were combined, centrifuged at200×g for 5 min to remove cell debris and then at 100,000 g for1 h to remove intracellular vesicles. A supernatant aliquot wasretained for control of successful activation. The numbers of vi-able and dead cells were determined using the trypan blue dyeexclusion assay. The percentage of cell viability was expressedas the ratio of total viable cells to the sum of total viable anddead cells.

2.2. Cell activation analysis

Cell activation by PMA+Io was confirmed by a Western blotagainst NFATc1 and by an ELISA detecting IL-2 secretion. Tothat end, the cell pellet was washed with ice-cold phosphate-buffered saline (PBS), resuspended and incubated for 10 min inice cold lysis buffer (10mM Hepes pH 7.4, 1.5 mM MgCl2,10 mM KCl, 0.2% N-octylglucoside and complete protease inhib-itor cocktail (Roche)). The suspensionwas homogenized in a Pot-ter Elvehjem homogenizer and centrifuged at 400×g for 15min.All steps were performed at 4 °C. After centrifugation, the super-natants were collected as cytosolic protein fraction, protein con-centration was determined by Bradford (Bio-Rad) using BSA asstandard and Laemmli buffer was added. Total cell extractswereboiled for 10min, loadedonto6%or15%SDS-polyacrylamidegels, and proteinswere separated by electrophoresis under reduc-ing conditions. The protein gels were transferred to nitrocellulosemembranes and then incubated in blocking solution (10%, w/vskimmed milk in PBS) for 1 h at room temperature. After severalwashes with PBS-T (0.1% Tween-20 in PBS), the membranes wereincubated for 2 h with the mouse monoclonal anti-NFATc1antibody 7A6 (0.1%, v/v) (Alexis) in PBS-T with 1% bovine serumalbumin or with mouse monoclonal anti-α-tubulin antibody(0.0025%, v/v) (Sigma) as loading control. Membranes were thenwashed three times for 5min each with PBS-T, and incubatedwith peroxidase-labeled goat anti-mouse IgG (Pierce) for 1 h atroom temperature. After three washes with PBS-T and one washwith H2O, membrane-bound antibody was detected with the ECL(enhanced chemiluminescence) detection reagent (AmershamBiosciences).

Fig. 1 – Scheme of the technology proposed in this work forthe systematic analysis of secretomes. For further details,see the text.

563J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

The levels of IL-2 in the Jurkat T-cell conditioned mediawere determined by an ELISA kit (Diaclone) following themanufacturer's instructions.

2.3. Sample pretreatment

Control and treated Jurkat T-cell conditioned media weredried-down by lyophilization, resuspended in 2.5 ml 25 mMammonium bicarbonate buffer pH 8.8, desalted on PD-10 col-umns (GE Healthcare) equilibrated with the same buffer andlyophilized again. Samples were taken up in 200 μl water andprotein concentration was assayed by Bradford (Bio-Rad)using BSA as standard.

2.4. Protein quantification by whole proteome in-geldigestion, peptide 18O labeling and IEF fractionation

These stepswereperformedaccording toour protocol describedin [21]. Briefly, 500 μg of protein from control and treated sam-ples was electrophoresed on the same SDS-PAGE gel until thefront of the run reached the resolving gel, so that thewhole pro-teome became concentrated in the stacking gel. The CoomassieBlue stained proteome band from each lane was cut into cubes(0.5×0.5 cm). Gel slices from the same lane were pooled in atube and digested overnight at 37 °C [23] with 60 ng/μl trypsin(Promega, Madison, WI, USA) at 5:1 protein:trypsin (w/w) ratioin 50mM ammonium bicarbonate buffer pH 8.8 containing10% (v/v) acetonitrile and 0.01% (w/v) 5-cyclohexyl-1-pentyl-ß-D-maltoside (CYMAL-5) [24]. The resulting tryptic peptidesfrom each lane were extracted by 1 h incubation in 12mM am-monium bicarbonate buffer pH 8.8, followed by 1 h incubationin acetonitrile/12 mM ammonium bicarbonate buffer pH 8.8(2:1) (v:v). 25% stock trifluoracetic acid was added to a final con-centration of 1% and samples were desalted on C18 OASIS HLBExtraction cartridges (Waters, Milford; Massachusetts, USA)and dried-down in a Speed Vac.

Dried peptides from each gel slice were differentially labeledwith either H2

16O or H218O (95%, Isotec, Miamisburg, OH) as previ-

ously described [22] with some modifications. Briefly, peptidesfrom the control sample were labeled with 16O, and peptidesfrom the treated sample were labeled with 18O. The labelingwas stopped by removing trypsin beads using a physical filter(Wizard minicolumns, Promega) and by adding to the remain-ing mixture the irreversible trypsin inhibitor TLCK at a finalconcentration of 1 mM from 50mg/ml stock solution inmetha-nol. After 1 h incubation at 37 °C, the two labeled samples weremixed, diluted until 2% acetonitrile and pH adjusted to 3 with1 M ammonium formiate pH 3, desalted on C18 Oasis HLB Ex-traction cartridges using as elution solution 50% acetonitrilein 5 mM ammonium formiate pH 3, and dried down.

Clean peptide pools were taken up in the focusing bufferconsisting of 5% glycerol and 2% IPG buffer pH 3–5.6, 3–10 or4–7 (GE Healthcare) and separated into 24 fractions by IEFusing the Agilent 3100 OFFgel fractionator, a 24 cm Immobi-line DryStrip pH 3–10 (GE Healthcare) and the standard meth-od for peptides recommended by the manufacturer. Therecovered fractions were acidified with 20 μl 1 M ammoniumformiate pH 3, cleaned-up using OMIX C18 tips (Varian), elutedwith 50% acetonitrile in 5 mM ammonium formiate pH 3 anddried-down prior to RP-HPLC–LIT analysis.

2.5. LC–MS/MS analysis and peptide identification

All samples were analyzed by LC–MS/MS using a Surveyor LCsystem coupled to a linear ion trap mass spectrometer modelLTQ (Thermo-Finnigan, San Jose, CA, USA) as previously de-scribed [25,26]. The LTQ was operated in a data-dependentZoomScan- and MS/MS-switching mode [27]. Zoom targetparameters, number of microscans, normalized collisionenergy, and dynamic exclusion parameters were as previouslydescribed [25].

Fig. 3 – Classification of identified proteins in Jurkat T-cellconditionedmedium in terms of secretion pathways. The pro-portion of proteins termed as extracellular by the GO; as secretedby the signal peptide-dependent pathway and by a non-classicalpathway by the SecretomeP software; and of proteins found intheextracellular spaceaccording toexperimental evidence in theliterature, is shown. Proteinsnot found in the set of 4928proteinspreviously identified in the Jurkat global lysate obtained from [9]were also considered as secretable.

Fig. 2 – Fraction-wise distribution of identified peptides (A) and peptide fractionation profile (B) by OFFgel IEF along the pHrange 3–11. The number of unique peptides (Np) and proteins (Nq) identified with an FDR≤5% is shown.

564 J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

Protein identification was carried out as previously de-scribed [25] using SEQUEST algorithm (Bioworks 3.2 package,Thermo Finnigan). TheMS/MS raw files were searched againstthe Human Swissprot database (Uniprot release 14.0, 19,929sequence entries for human) supplementedwith porcine tryp-sin. The same collection of MS/MS spectra was searchedagainst a pseudo-inverted database constructed from thesame database. SEQUEST searches were performed allowingoptional (Methionine oxidation, Lysine and Argininemodifica-tion of +4 Da) and fixedmodifications (Cysteine carboxamido-methylation), 2 missed cleavages, 2 amu or 1.2 amu masstolerance for precursor or fragment ions, respectively. Deter-mination of False Discovery Rate (FDR) as a measure of statis-tical significance of peptide identification was performed byusing the Probability Ratio method [28] and a refined FDR cal-culation method, that employs separate inverted databasesand a competition strategy, as we have recently described [29].

2.6. Computational analysis of potentiallysecreted proteins

From all the identified proteins at an FDR<5% a list of poten-tially secreted proteins was constructed that met at leastone of the following criteria: (1) proteins annotated as extra-cellular (GO:0005576, 0031012, 0005615 and 0070062), belong-ing to the cell surface (GO:0009986) or located on the externalside of plasma membrane (GO:0009897) according to theGene Onthology (GO); (2) proteins reported as extracellular inthe literature; (3) proteins containing a signal peptide accord-ing to the SecretomeP 2.0 and thereby identified as potentiallysecreted via the classical pathway; and (4) proteins with aSecretomeP 2.0 score exceeding 0.5, which were classified assecreted by a non-classical pathway [30]. Also, (5) from thelist of proteins fulfilling criteria (3) or (4) were subtracted theones containing more than 2 or at least one transmembranehelix, respectively, predicted by the TMHMM server [31]. Addi-tionally, (6) we considered as potentially secreted thoseremaining proteins that did not meet any of the previous se-cretion criteria, but were not present among the list of 4928proteins reported in the Jurkat global cell lysate study [9]. To

prevent overlap among secretion criteria, proteins were listedunder only one category.

2.7. Peptide quantification and statistical analysis

Peptide quantification from ZoomScan spectra and labelingefficiency calculation were performed as described [32]; [33]using QuiXoT, a program written in C# in our laboratory.Data statistical analysis was done on the basis of a recentlydeveloped and validated random-effects model that includesfour different sources of variance: at the spectrum-fitting,scan, peptide and protein levels [22]. Only ZoomScan spectracorresponding to peptide matches with a false discovery ratelower than 5% were used for quantification.

565J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

2.8. Functional protein analysis

Protein log2-ratios standardized according to their estimatedvariances (zq values) were classified in terms of the GeneOnthology Biological Process. The significance of enrichmentfor different categories was assessed by Student's t-test.

Fig. 5 – Cumulative frequency plot of the standardizedvariable at the protein level for the 709 quantified proteinsin the Jurkat T-cell conditioned medium (A) and afterseparating all quantified proteins into three functionalcategories (B). These categories were: intracellular(black line), extracellular (blue line) and functionally relatedto apoptosis (green line). In (A) and (B) the red line representsthe normal distribution N(0,1), and the black line thedistribution with either σq

2=0.24 or σq2=0.11, respectively.

Proteins detected by only 1 peptide per protein with aFDR>1% were excluded from the analysis. (C) Analysis ofprotein abundance changes in the population of 709quantified proteins from Jurkat T-cells upon PMA+Ioactivation. A variance at the protein level (σq

2) of 0.11 wasused to determine the statistically significant proteinabundance changes at FDRq<5% (indicated by black points).

3. Results

3.1. Preparation and characterization of proteins in theconditioned medium of Jurkat T-cells

The major technical difficulty in characterizing the secretomeof cells in culture is obtaining a sample containing mainlyproteins secreted via specific cellular mechanisms and notas a consequence of nonspecific events such as cell lysis.Also, the presence of standard serum concentration (10%) dur-ing conditioned media generation dramatically reduces theability to detect low-abundance proteins. Moreover, serum isknown for its activating properties on Jurkat T-cells, therebyhindering the performance of activation assays with otherstimuli. However, complete deprivation of serum for the rest-ing period plus the several hours needed for gene expressionto produce significant changes in protein secretion upon cellactivation, brings about cell breakdown and a concomitant in-crease in contaminants of intracellular origin that may beconfounded with the real secretome. In order to reduce thedeleterious effects of a complete absence of serum, cellswere cultured in a medium containing 0.5% serum, whichwas low enough to leave them in the resting state. Underthese experimental conditions, Jurkat T-cells maintainedtheir ability to activate upon PMA+Io treatment. This wasconfirmed by the typical induction of NFATc1 expression [34]detected by Western blot analysis (Supplementary Fig. S1),and IL-2 secretion [35] (data not shown). PMA+Io is knownas one of the most potent stimuli for lymphocyte activationand cellular protein release. PMA stimulates both exocyticpathways and shedding of membrane anchored proteins[36–40]. Although a long-term PMA+Io treatment also elicitsJurkat T-cell apoptosis [41], after the treatment cell viabilityremained similar between control and treated samples (79%versus 68%).

Fig. 4 – Control of quantification artifacts. (A) Distribution of labeexperiment. (B) Effect of partial digestions and (C) Met oxidationcorresponds to the statistical model and is as used in previous p

A scheme of the entire protocol followed to determinechanges in secretome composition produced by cell stimula-tion is presented in Fig. 1. Basically, after elimination of cell

ling efficiencies as function of log2-ratios in the globalon peptide quantification. The meaning of the symbolsublications [21,22].

Table 1 – Proteins from conditioned medium changing their expression upon PMA+Io activation of Jurkat T-cells. (FDRq<0.05).

566JO

UR

NA

LO

FPR

OT

EO

MIC

S75

(2011)

561–571

2.12.52.93.54.03.12.72.42.62.72.22.22.32.12.22.32.63.1

3.203.843.864.41

− 5.46− 4.27− 3.76− 3.70− 3.70− 3.68− 3.25− 3.15− 3.00

2.942.962.993.914.17

2.20E-023.08E-033.21E-036.02E-045.55E-068.30E-044.25E-034.89E-034.97E-034.67E-031.95E-022.57E-023.37E-023.87E-023.75E-023.37E-022.81E-031.25E-03

Protein biosynthesisImmune response

downdowndowndown

upupupupupupupupup

downdowndowndowndown

P04080|CYTB Cystatin-B 2.32.12.32.12.12.32.12.32.02.32.32.32.5

− 3.30− 3.11− 3.10− 3.09− 3.05− 3.04− 3.03− 2.99− 2.87

3.003.063.073.13

1.64E-022.86E-022.89E-022.84E-023.02E-023.18E-023.31E-023.39E-024.82E-023.39E-023.11E-023.01E-022.60E-02

Unknown Regulation of apoptosis

Protein foldingUnknownUnknownUnknown

Actin cytoskeleton organizationUnknown

Negative regulation of cell proliferationmRNA processing

Negative regulation of cell proliferationmRNA processing

Signal transduction

upupupupupupupupup

downdowndowndown

YesYesYesYesYesYesYesYesYesYesYesYesYes

YesYesYesYesNoNoNoNoNoNoNoNoNoNoNoNoNoNo

P12814|ACTN1 Alpha-actinin-1Q9Y536|PAL4B Peptidylprolyl cis-trans isomerase A-like 4B P24666|PPAC Low molecular weight phosphotyrosine protein phosphataseO60361|NDK8 Putative nucleoside diphosphate kinaseQ14011|CIRBP Cold-inducible RNA-binding proteinP07737|PROF1 Profilin-1O75368|SH3L1 SH3 domain-binding glutamic acid-rich-like proteinP98179|RBM3 Putative RNA-binding protein 3Q9NR30|DDX21 Nucleolar RNA helicase 2P25705|ATPA ATP synthase subunit alpha, mitochondrial precursorQ9UKF6|CPSF3 Cleavage and polyadenylation specificity factor subunit 3 - Q9Y570|PPME1 Protein phosphatase methylesterase 1

P60842|IF4A1 Eukaryotic initiation factor 4A-IQ9NZ08|ERAP1 Endoplasmic reticulum aminopeptidase 1Q92896|GSLG1 Golgi apparatus protein 1 precursorQ99523|SORT Sortilin precursorQ8NC51|PAIRB Plasminogen activator inhibitor 1 RNA-binding proteinP30046|DOPD D-dopachrome decarboxylaseP05387|RLA2 60S acidic ribosomal protein P2Q01469|FABP5 Fatty acid-binding protein, epidermalP99999|CYC Cytochrome cP62750|RL23A 60S ribosomal protein L23aO75369|FLNB Filamin-BP14317|HCLS1 Hematopoietic lineage cell-specific proteinP54727|RD23B UV excision repair protein RAD23 homolog BQ12905|ILF2 Interleukin enhancer-binding factor 2Q92922|SMRC1 SWI/SNF-related matrix-associated actin-dependent regulator Q12882|DPYD Dihydropyrimidine dehydrogenase [NADP+] precursorO75367|H2AY Core histone macro-H2A.1P11388|TOP2A DNA topoisomerase 2-alpha

Cell adhesionVesicle organization

Regulation of mRNA stabilityMelanin biosynthetic process

Translational elongationLipid metabolic process

ApoptosisTranslational elongation

Signal transductionPositive regulation of cell proliferation

UnknownmRNA processing

Signal transductionProtein biosynthesis

Chromatin modificationDNA repair

Proteins are sorted by zq, whose magnitude is shaded according to the color scales at the top. Secretion prediction was performed using the bioinformatic tools described in Materials and methods. (*) proteins not identifiedpreviously in the Jurkat T-cell secretome [41].

567JO

UR

NA

LO

FPR

OT

EO

MIC

S75

(2011)

561–571

568 J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

debris and intracellular vesicles by centrifugation, the sampleswere desalted and subjected to “in gel” digestion under well-controlled conditions using a robust protocol described previ-ously [21]. The peptideswere then subjected to 16O/18O-labeling,mixed together, IEF-separated and analyzed bymass spectrom-etry.More than 4000 uniquepeptides belonging to 1288proteinswere identified in this study (Supplementary Table S1); fromthese, 546 proteins (Supplementary Table S2) have not beenidentified in a previous Jurkat T-cell secretome study [42]. Thismay be explained by the different cell culture setup, togetherwith the high resolving power of the OFFgel IEF separation, inwhich more than 65% of the identified peptides were found inonly one fraction and more than 80% in one or two fractions(Fig. 2A). A representative distribution of identified peptidesalong the pH range is presented in Fig. 2B, showing a pI distribu-tion in good agreement with OFFgel analysis reported previous-ly in other proteomes [21].

We then analyzed whether the proteins identified in theconditioned medium could be there as a consequence of aphysiological secretion process or due to a mere cell breakdown.We considered as potentially secreted the proteins fulfilling anyof the following criteria (see Material andmethods): 1) the proteinis annotated as extracellular in theGeneOnthology (GO) database;2) there is bibliographical evidence in the literature that the candi-date is a secreted protein; 3) the protein contains a signal peptidefor the classical secretion pathway according to the SecretomePalgorithm; or 4) the protein has no signal peptide predicted butgives a scorehigher than0.5 by theSecretomePalgorithm, indicat-ing it is a good candidate for non-classical secretion. In addition, 5)theTMHMMalgorithmpredicted integralmembraneproteins thatcould not be categorized in the classical or non-classical secretionpathways. For someof the proteins secreted via the classical path-way, the TMHMM algorithm predicted a transmembrane helixlocated more than 40 amino acids away from the N-terminus ora second transmembrane helix that could be acting as a stop-transfer sequence, suggesting they were anchored to the plasmamembrane. However, since a signal peptide was predicted andpossibility exists for these proteins to be released via sheddingfrom the plasma membrane, and PMA is known for promotingthis secretion route, we also considered these proteins as poten-tially secreted. Additionally, if the TMHMM algorithm predictedmore than 2 transmembrane helixes for the signal-peptide con-taining proteins, those proteinswere considered as integralmem-brane proteins and were eliminated from the list of potentiallysecreted candidates. On the other hand, for the proteins secretedby the non-classical route, the TMHMM prediction of at least onetransmembrane helix eliminated the candidate from the list ofpotentially secreted proteins.

A further criterion was also used on the basis of a previousreport [9], which we have used to generate a database contain-ing predominantly intracellular proteins. In that study 4928proteins were identified in Jurkat T-cell global lysate by anexhaustive, high-throughput HPLC–MS/MS peptide identifica-tion, using a repetitive injection approach, subcellular frac-tionation and multiple enrichment methods, accumulating atotal of 1707 HPLC runs [9]. In our study, 198 proteins (15% ofthe total) were not previously identified in the global lysate.Since in our work we used the same mass spectrometer (LTQlinear ion trap) and analyzed a considerably lower number offractions (24 HPLC runs) than in the global lysate study, we

considered it unlikely that we were able to identify proteinsbelonging to the global lysate that were not identified in theprevious work. Therefore, it was reasonable to considerthese new 198 proteins identified in the conditioned mediaas secretion candidates and thus were included in the list ofpotentially secreted proteins. The final classification of pro-teins identified in the conditioned medium according totheir secretion potential is depicted in Fig. 3. As shown, al-most 60% (762 proteins) of the identified proteins could beconsidered as potentially secreted.

3.2. Analysis of quantitative data

Besides the peptide identification information gathered fromthe collection of MS/MS spectra, our experimental designallowed to compare protein abundance before and after JurkatT-cell stimulation, providing a critical information about theeffect of the stimulus on protein secretion. Quantitative infor-mation by 18O/16O-labeling was obtained at the individualpeptide level from the ZoomScan spectra performed beforethe MS/MS spectra, as described previously [32], and proteindigestion and labeling efficiency were carefully controlled[33]. As expected, an excellent labeling of all peptides wasobtained (Fig. 4A), ensuring that labeling did not introduceartifacts in peptide quantification.

Analysis of quantitative datawas performed using a statisti-cal model described previously [21,22] that has also been thor-oughly tested on a set of different proteomes [21]. A veryuseful advantage of this statistical model is that it considersseparately the different sources or error, allowing the user tocontrol the quality of the whole process going from samplepreparation to MS analysis. The variances at the scan and pep-tide levels (σS

2=0.008 (95% C.I., 0.010–0.013) and σP2=0.028 (95%

C.I., 0.014–0.029)) were similar to the ones obtained using otherproteomes [21,22], suggesting that MS analysis and protein di-gestion were performed within the expected standards. Otherpossible quantification artifacts at the peptide level, such as un-balanced partial digestion or methionine oxidation were alsocontrolled, as described previously [21,22]. As shown in Fig. 4Band C, the variability introduced by partial digestions and me-thionine oxidation on individual peptide quantification wasnegligible. After controlling for all potential artifacts, wewere fi-nally able to quantify more than 2390 unique peptides comingfrom 709 proteins (Supplementary Table S3), from which 385(54%) were classified as potentially secreted.

On the other hand, the calculated protein variance was twoorders of magnitude higher (σQ

2 =0.24) than that of other pro-teomes analyzed previously in our laboratory [21]. Moreover,as shown in Fig. 5A, the cumulative normality plot of the stan-dardized variable at the protein level significantly deviatedfrom the theoretical curve, suggesting the existence of morethan one population of proteins in terms of variance. The dis-persion at the protein level is related to the variance of themethod used to prepare protein extracts and to the biologicalvariance. When protein extraction is performed in a way thatdoes not introduce biases for relative comparison of proteinlevels in the samples, we have systematically observed thatthe variance at the protein level is close to zero [21,22]. How-ever, in the kind of experiments performed in this work, thepreparation of the secretome sample requires further steps

569J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

of manipulation and a large proportion of secretome proteinsare expected to increase upon PMA+Io treatment, making itdifficult to estimate the true protein variance associated tothe null hypothesis (i.e., that none of the proteins undergoesexpression changes). In order to deepen our insight into thesources of dispersion at the protein level in our experiment,we set out to analyze protein variance in relation to the natureof the proteins identified in the conditioned medium.

In a first attempt to make a conservative estimation of theprotein variance in the null hypothesis we calculated the pro-tein variance in the subpopulation of quantified proteins thatcould not be classified as secreted (NQ=384, SupplementaryTable S4), obtaining a lower protein dispersion (σQ

2 =0.17).However, the cumulative normality plot of the standardizedvariable at the protein level did not appreciably improve, indi-cating that this population was still heterogeneous in terms ofprotein variance (data not shown). To obtain a homogeneouspopulation, the set of potentially non-secreted proteins wasfurther classified according to their biological function by theGene Onthology. As shown in Fig. 5B, proteins involved in ap-optosis were clearly displaced from the rest of the proteinsand this category was significantly displaced from the nulldistribution (p<0.05) (Supplementary Table S5). The remain-ing population of intracellular proteins, from which the onesidentified by only 1 peptide per protein with FDR>1% werealso excluded (NQ=176, Supplementary Table S5), followed apure normal distribution N(0,1) (Fig. 5B) and was finally usedto make a conservative estimate of protein variance for thenull hypothesis (σQ

2 =0.11).

3.3. Identification of secreted proteins specifically regulatedby T-cell activation

With this new value of protein variance we set out to discrimi-nate the secreted proteins specifically regulated by PMA+Io inJurkat T-cell conditioned medium. In this analysis we includedthe entire set of quantified proteins, so that the protein classifi-cation performed previously did not introduce any bias on thefinal result. As shown in Fig. 5C and Table 1, 62 proteins showeda statistically significant change in their abundance at anFDRq<5%. From these, 11 proteins (26%) were not identified inthe previous report of basal Jurkat secretome [9], and in goodagreement with the nature of this subproteome, more than79% (49 proteins) of the proteins changing their expressionwere up-regulated (Fig. 5C, Table 1).

Most of the proteins secreted by Jurkat T-cells upon PMA+Ioactivation are involved in inflammatory response, signal trans-duction, cell growth and differentiation and cell redox homeo-stasis. Among these proteins there is a group of cytokine-related proteins of the greatest biological relevance, which in-cluded the macrophage migration inhibitory factor (MIF), theS100A9 protein and the insulin-like growth factor-binding pro-tein 2. MIF is a pro-inflammatory cytokine playing an essentialrole in the pathophysiology of host immune and inflammatoryresponse [43]. It is secreted by macrophages, anterior pituitarycells, and endothelial cells activating both macrophages andT-lymphocytes [44–46]. MIF is involved in the induction of in-flammatory cytokines, nitric oxide and superoxide anion, andin the regulation of macrophage and lymphocyte proliferation.

This cytokine plays an important role in the pathogenesis ofseveral diseases such as cancer, septic shock and allergiesamong others [47]. This is the first report of MIF detection byMS in the secretome of Jurkat T-cells.

The S100A9 protein belongs to the low molecular masscalcium-binding S100 proteins. Although S100A9 is localizedpredominantly in the cytoplasm, it is secreted in humanmonocytes by an energy-consuming pathway which is de-pendent on an intact microtubule network and involves pro-tein kinase C [48]. Moreover, S100A9 serum levels areincreased in patients suffering from a number of inflamma-tory disorders including cystic fibrosis, rheumatoid arthritisand chronic bronchitis [49]. Additionally, this protein wasdemonstrated to be a potent chemotactic factor for murinemyeloid cells [50] and to enhance migration of monocytesacross endothelial cells [51]. IGFBP2 is up-regulated in differ-ent types of tumors, including leukemia, while is found atvery low levels in normal tissues [52]. This protein has beenshown to bind integrins [53] and to activate Akt signaling[52], an important event for cell proliferation. AlthoughS100A9 and IGFBP2 were previously identified in the basalsecretome of Jurkat T-cells [10], our work is the first reportthat gives evidence that the secretion of these proteins is en-hanced in response to Jurkat T-cell activation.

In summary, in this work we present a method for thequantitative analysis of secretomes and present the deepestinsight to date into the effect of activation on T-cell secre-tome. This was accomplished thanks to our robust protocolfor quantitative high-throughput analysis of proteomes by18O labeling [21], together with a novel computational algo-rithm for data analysis [33] and a statistical model supportedby a validated null hypothesis [22], which we have used inconjunction with bioinformatics tools for the ontological anal-ysis of the data. The proteomic technology used in this studypaves a way towards systematic analysis of dynamic secre-tomes in other biological models.

Supplementary materials related to this article can befound online at doi:10.1016/j.jprot.2011.08.022.

Acknowledgment

This work was supported by grants BIO2006-10085, SAF 2006-8348, SAF 2009-03736 and BIO2009-07990 from Spain's Minis-terio de Educación y Ciencia, CAM BIO/0194/2006 from Comuni-dad de Madrid, the Fondo de Investigaciones Sanitarias(Ministerio de Sanidad y Consumo, Instituto Salud Carlos III,RECAVA RD06/0014) and by an institutional grant by FundaciónRamón Areces to CBMSO.

R E F E R E N C E S

[1] Schenk S, Schoenhals GJ, de Souza G, Mann M. A highconfidence, manually validated human blood plasma proteinreference set. BMC Med Genomics 2008;1:41.

[2] Cochran DA, Evans CA, Blinco D, Burthem J, Stevenson FK,Gaskell SJ, et al. Proteomic analysis of chronic lymphocyticleukemia subtypes with mutated or unmutated Ig V(H) genes.Mol Cell Proteomics 2003;2:1331–41.

570 J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

[3] Cui JW, Wang J, He K, Jin BF, Wang HX, Li W, et al. Proteomicanalysis of human acute leukemia cells: insight into theirclassification. Clin Cancer Res 2004;10:6887–96.

[4] Cui JW, Wang J, He K, Jin BF, Wang HX, Li W, et al.Two-dimensional electrophoresis protein profiling as ananalytical tool for human acute leukemia classification.Electrophoresis 2005;26:268–79.

[5] Sundrud MS, Torres VJ, Unutmaz D, Cover TL. Inhibition ofprimary human T cell proliferation by Helicobacter pylorivacuolating toxin (VacA) is independent of VacA effectson IL-2 secretion. Proc Natl Acad Sci USA 2004;101:7727–32.

[6] Grolleau A, Bowman J, Pradet-Balade B, Puravs E, Hanash S,Garcia-Sanz JA, et al. Global and specific translational controlby rapamycin in T cells uncovered by microarrays andproteomics. J Biol Chem 2002;277:22175–84.

[7] Tu X, Huang A, Bae D, Slaughter N, Whitelegge J, Crother T,et al. Proteome analysis of lipid rafts in Jurkat cellscharacterizes a raft subset that is involved in NF-kappaBactivation. J Proteome Res 2004;3:445–54.

[8] Woo AJ, Dods JS, Susanto E, Ulgiati D, Abraham LJ.A proteomics approach for the identification of DNA bindingactivities observed in the electrophoretic mobility shift assay.Mol Cell Proteomics 2002;1:472–8.

[9] Wu L, Hwang SI, Rezaul K, Lu LJ, Mayya V, Gerstein M, et al.Global survey of human T leukemic cells by integratingproteomics and transcriptomics profiling. Mol CellProteomics 2007;6:1343–53.

[10] Wu CC, Hsu CW, Chen CD, Yu CJ, Chang KP, Tai DI, et al.Candidate serological biomarkers for cancer identified fromthe secretomes of 23 cancer cell lines and the human proteinatlas. Mol Cell Proteomics 2010;9:1100–17.

[11] Nair R, Rost B. Mimicking cellular sorting improves predictionof subcellular localization. J Mol Biol 2005;348:85–100.

[12] Nickel W. The mystery of nonclassical protein secretion. Acurrent view on cargo proteins and potential export routes.Eur J Biochem 2003;270:2109–19.

[13] Rubartelli A, Cozzolino F, Talio M, Sitia R. A novel secretorypathway for interleukin-1 beta, a protein lacking a signalsequence. EMBO J 1990;9:1503–10.

[14] Jackson A, Friedman S, Zhan X, Engleka KA, Forough R,Maciag T. Heat shock induces the release of fibroblast growthfactor 1 from NIH 3T3 cells. Proc Natl Acad Sci USA 1992;89:10691–5.

[15] Mignatti P, Morimoto T, Rifkin DB. Basic fibroblast growthfactor, a protein devoid of secretory signal sequence, isreleased by cells via a pathway independent of theendoplasmic reticulum-Golgi complex. J Cell Physiol1992;151:81–93.

[16] Hughes RC. Secretion of the galectin family of mammaliancarbohydrate-binding proteins. Biochim Biophys Acta1999;1473:172–85.

[17] Joliot A, Maizel A, Rosenberg D, Trembleau A, Dupas S,Volovitch M, et al. Identification of a signal sequencenecessary for the unconventional secretion of Engrailedhomeoprotein. Curr Biol 1998;8:856–63.

[18] Gardella S, Andrei C, Ferrera D, Lotti LV, Torrisi MR, BianchiME, et al. The nuclear protein HMGB1 is secreted bymonocytes via a non-classical, vesicle-mediated secretorypathway. EMBO Rep 2002;3:995–1001.

[19] Denny PW, Gokool S, Russell DG, Field MC, Smith DF.Acylation-dependent protein export in Leishmania. J BiolChem 2000;275:11017–25.

[20] Emanuelsson O, Brunak S, von Heijne G, Nielsen H. Locatingproteins in the cell using TargetP, SignalP and related tools.Nat Protoc 2007;2:953–71.

[21] Bonzon-Kulichenko E, Perez-Hernandez D, Nunez E,Martinez-Acedo P, Navarro P, Trevisan-Herraz M, et al.A robust method for quantitative high-throughput analysis

of proteomes by 18O labeling. Mol Cell Proteomics 2011;10M110 003335.

[22] Jorge I, Navarro P, Martinez-Acedo P, Nunez E, Serrano H,Alfranca A, et al. Statistical model to analyze quantitativeproteomics data obtained by 18O/16O labeling and linear iontrap mass spectrometry: application to the study of vascularendothelial growth factor-induced angiogenesis inendothelial cells. Mol Cell Proteomics 2009;8:1130–49.

[23] Shevchenko A, Tomas H, Havlis J, Olsen JV, Mann M. In-geldigestion for mass spectrometric characterization of proteinsand proteomes. Nat Protoc 2006;1:2856–60.

[24] Katayama H, Tabata T, Ishihama Y, Sato T, Oda Y, Nagasu T.Efficient in-gel digestion procedure using5-cyclohexyl-1-pentyl-beta-D-maltoside as an additive forgel-based membrane proteomics. Rapid Commun MassSpectrom 2004;18:2388–94.

[25] Lopez-Ferrer D, Martinez-Bartolome S, Villar M, Campillos M,Martin-Maroto F, Vazquez J. Statistical model for large-scalepeptide identification in databases from tandem massspectra using SEQUEST. Anal Chem 2004;76:6853–60.

[26] Ortega-Perez I, Cano E, Were F, Villar M, Vazquez J, RedondoJM. c-Jun N-terminal kinase (JNK) positively regulates NFATc2transactivation through phosphorylation within theN-terminal regulatory domain. J Biol Chem 2005;280:20867–78.

[27] Jorge I, Casas EM, Villar M, Ortega-Perez I, Lopez-Ferrer D,Martinez-Ruiz A, et al. High-sensitivity analysis of specificpeptides in complex samples by selected MS/MS ionmonitoringand linear ion trap mass spectrometry: application to biologicalstudies. J Mass Spectrom 2007;42:1391–403.

[28] Martinez-Bartolome S, Navarro P, Martin-Maroto F,Lopez-Ferrer D, Ramos-Fernandez A, Villar M, et al. Propertiesof average score distributions of SEQUEST: the probabilityratio method. Mol Cell Proteomics 2008;7:1135–45.

[29] Navarro P, JS VZ. A refined method to calculate FalseDiscovery Rates for peptide identification using decoydatabases. J Proteome Res 2009.

[30] Bendtsen JD, Jensen LJ, Blom N, Von Heijne G, Brunak S.Feature-based prediction of non-classical and leaderlessprotein secretion. Protein Eng Des Sel 2004;17:349–56.

[31] Krogh A, Larsson B, von Heijne G, Sonnhammer EL. Predictingtransmembrane protein topology with a hidden Markovmodel: application to complete genomes. J Mol Biol 2001;305:567–80.

[32] Lopez-Ferrer D, Ramos-Fernandez A, Martinez-Bartolome S,Garcia-Ruiz P, Vazquez J. Quantitative proteomics using16O/18O labeling and linear ion trap mass spectrometry.Proteomics 2006;6(Suppl 1):S4–S11.

[33] Ramos-Fernandez A, Lopez-Ferrer D, Vazquez J. Improvedmethod for differential expression proteomics usingtrypsin-catalyzed 18O labeling with a correction for labelingefficiency. Mol Cell Proteomics 2007;6:1274–86.

[34] Northrop JP, Ho SN, Chen L, Thomas DJ, Timmerman LA,Nolan GP, et al. NF-AT components define a family oftranscription factors targeted in T-cell activation. Nature1994;369:497–502.

[35] Martinez-Martinez S, Gomez del Arco P, Armesilla AL,Aramburu J, Luo C, Rao A, et al. Blockade of T-cell activationby dithiocarbamates involves novel mechanisms ofinhibition of nuclear factor of activated T cells. Mol Cell Biol1997;17:6437–47.

[36] Fan H, Derynck R. Ectodomain shedding of TGF-alpha andother transmembrane proteins is induced by receptortyrosine kinase activation and MAP kinase signalingcascades. EMBO J 1999;18:6962–72.

[37] Martinez-Lacaci I, Johnson GR, Salomon DS, Dickson RB.Characterization of a novel amphiregulin-related molecule in12-O-tetradecanoylphorbol-13-acetate-treated breast cancercells. J Cell Physiol 1996;169:497–508.

571J O U R N A L O F P R O T E O M I C S 7 5 ( 2 0 1 1 ) 5 6 1 – 5 7 1

[38] Salo T, Turpeenniemi-Hujanen T, Tryggvason K.Tumor-promoting phorbol esters and cell proliferationstimulate secretion of basement membrane (type IV)collagen-degrading metalloproteinase by human fibroblasts.J Biol Chem 1985;260:8526–31.

[39] Rossner S, Mendla K, Schliebs R, Bigl V. Protein kinase Calphaand beta1 isoforms are regulators of alpha-secretoryproteolytic processing of amyloid precursor protein in vivo.Eur J Neurosci 2001;13:1644–8.

[40] Zheng Y, Saftig P, Hartmann D, Blobel C. Evaluation of thecontribution of different ADAMs to tumor necrosis factoralpha (TNFalpha) shedding and of the function of theTNFalpha ectodomain in ensuring selective stimulatedshedding by the TNFalpha convertase (TACE/ADAM17). J BiolChem 2004;279:42898–906.

[41] Rodriguez-Tarduchy G, Sahuquillo AG, Alarcon B, Bragado R.Apoptosis but not other activation events is inhibited by amutation in the transmembrane domain of T cell receptorbeta that impairs CD3zeta association. J Biol Chem 1996;271:30417–25.

[42] Wu CC, Chen HC, Chen SJ, Liu HP, Hsieh YY, Yu CJ, et al.Identification of collapsin response mediator protein-2 as apotential marker of colorectal carcinoma by comparativeanalysis of cancer cell secretomes. Proteomics 2008;8:316–32.

[43] Calandra T, Roger T. Macrophage migration inhibitory factor: aregulator of innate immunity. Nat Rev Immunol 2003;3:791–800.

[44] Bernhagen J, Calandra T, Mitchell RA, Martin SB, Tracey KJ,Voelter W, et al. MIF is a pituitary-derived cytokine thatpotentiates lethal endotoxaemia. Nature 1993;365:756–9.

[45] Bacher M, Metz CN, Calandra T, Mayer K, Chesney J, Lohoff M,et al. An essential regulatory role for macrophage migration

inhibitory factor in T-cell activation. Proc Natl Acad Sci USA1996;93:7849–54.

[46] Nishihira J, Koyama Y, Mizue Y. Identification of macrophagemigration inhibitory factor (MIF) in human vascularendothelial cells and its induction by lipopolysaccharide.Cytokine 1998;10:199–205.

[47] Lue H, Kleemann R, Calandra T, Roger T, Bernhagen J.Macrophagemigration inhibitory factor (MIF): mechanisms ofaction and role in disease. Microbes Infect 2002;4:449–60.

[48] Rammes A, Roth J, Goebeler M, Klempt M, Hartmann M, SorgC. Myeloid-related protein (MRP) 8 and MRP14,calcium-binding proteins of the S100 family, are secretedby activated monocytes via a novel, tubulin-dependentpathway. J Biol Chem 1997;272:9496–502.

[49] Washburn MP, Wolters D, Yates III JR. Large-scale analysisof the yeast proteome by multidimensional proteinidentification technology. Nat Biotechnol 2001;19:242–7.

[50] Cornish CJ, Devery JM, Poronnik P, Lackmann M, Cook DI,Geczy CL. S100 protein CP-10 stimulates myeloid cellchemotaxis without activation. J Cell Physiol 1996;166:427–37.

[51] Eue I, Pietz B, Storck J, Klempt M, Sorg C. Transendothelialmigration of 27E10+ humanmonocytes. Int Immunol 2000;12:1593–604.

[52] Dunlap SM, Celestino J, Wang H, Jiang R, Holland EC, FullerGN, et al. Insulin-like growth factor binding protein 2promotes glioma development and progression. Proc NatlAcad Sci USA 2007;104:11736–41.

[53] Wang GK, Hu L, Fuller GN, Zhang W. An interaction betweeninsulin-like growth factor-binding protein 2 (IGFBP2) andintegrin alpha5 is essential for IGFBP2-induced cell mobility.J Biol Chem 2006;281:14085–91.