identification of specific reachable molecular targets in human breast cancer using a versatile ex...

RESEARCH ARTICLE

Identification of specific reachable molecular targets in

human breast cancer using a versatile ex vivo

proteomic method

Vincent Castronovo1*, Philippe Kischel1*, Francois Guillonneau2, Laurence de Leval3,Thierry Deféchereux4, Edwin De Pauw2, Dario Neri5 and David Waltregny1

1 Metastasis Research Laboratory, Center for Experimental Cancer Research, University of Liège, Liège, Belgium2 Laboratory of Mass Spectrometry, Department of Chemistry, University of Liège, Liège, Belgium3 Department of Pathology, University Hospital of Liège, Liège, Belgium4 Department of Abdominal Surgery and Transplantation, University Hospital of Liège, Liège, Belgium5 Institute of Pharmaceutical Sciences, Swiss Federal Institute of Technology Zürich, Zürich, Switzerland

Targeting of tumoral tissues is one of the most promising approaches to improve both the effi-cacy and safety of anticancer treatments. The identification of valid targets, including proteinsspecifically and abundantly expressed in cancer lesions, is of utmost importance. Despite state-of-the-art technologies, the discovery of cancer-associated target proteins still faces the limitation,in human tissues, of antigen accessibility to suitable high-affinity ligands such as human mAbbound to bioactive molecules. Terminal perfusion of tumor-bearing mice or ex vivo perfusion ofhuman cancer-bearing organs with a reactive biotin ester solution has successfully led to theidentification of novel accessible biomarkers. This methodology is however restricted to per-fusable organs, and excludes most of the tissues of interest to targeted therapies, e.g. primarybreast cancer and metastases. Herein, we report on the development of a new chemical prote-omic method that bypasses the perfusion step and thus offers the potential to identify accessiblemolecular targets in virtually all types of animal and human tissues. We have validated our newprocedure by identifying biomarkers selectively expressed in human breast carcinoma. Overall,this powerful technology may lay the ground not only for custom-made therapies in cancer, butalso for the development of therapies that need to be selectively delivered in a specific tissue.

Received: November 9, 2006Revised: January 8, 2007

Accepted: January 16, 2007

Keywords:

Biomarker discovery / Biotinylation / Breast cancer / Mass spectrometry / Tumortargeting

1188 Proteomics 2007, 7, 1188–1196

1 Introduction

The identification of biomarkers unique to specific patho-logic processes would be most valuable for the accuratedetection (e.g. by imaging studies) and selective therapy of

diseases [1, 2]. For instance, patients suffering from cancerwould certainly benefit from such developments [3]. Indeed,chemotherapeutic agents, usually designed to take advantageof tumor cell characteristics such as high proliferation rates,unfortunately also target normal cycling cells includinghematopoietic progenitors. The targeted delivery of thesedrugs and other bioactive molecules (e.g. radioisotopes, cyto-kines) to the tumor microenvironment (e.g. proteins specifi-cally expressed in the stromal or vascular compartment ofthe tumor) by means of binding molecules such as recombi-nant human antibodies, would represent a considerable

Correspondence: Dr. Philippe Kischel, Metastasis Research Labo-ratory, Pathology Building, Level-1, Bat. B23, CHU Sart Tilman, B-4000 Liège, BelgiumE-mail: [email protected]: 132-43-66-29-75

Abbreviations: ECM, extracellular matrix; MudPIT, multidimen-sional protein identification technology; TMA, tissue microarray * Both authors contributed equally to this work.

DOI 10.1002/pmic.200600888

© 2007 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.proteomics-journal.com

Proteomics 2007, 7, 1188–1196 Technology 1189

therapeutic improvement [1, 2, 4, 5]. This selective strategywould increase the amount of drugs reaching the tumor withlittle or no toxicity to the host’s healthy tissues [1, 2, 5]. Therecent development of high-throughput proteomic technolo-gies such as MS have facilitated the rapid and accurate iden-tification of small, but complex biological sample mixtures,making target identification easier than ever. For example,gel-free shotgun MS/MS has been recently used to compareglobal protein expression profiles in human mammary epi-thelial normal and cancer cell lines [6]. Unfortunately, a sig-nificant pitfall associated with such an approach is that itprovides no clue as to whether proteins of interest are acces-sible to suitable high-affinity ligands, such as systemicallydelivered mAb, in human tissues. Indeed, specific, yet poorlyaccessible proteins expressed in pathologic tissues areexpected to be of little value for the development of antibody-based anticancer therapies. Strategies that would unveil dis-ease biomarkers not only specifically expressed in pathologictissues but also accessible from the extracellular fluid wouldhelp overcome this limitation.

In this study, a new, quick, and efficient method wasengineered to identify human breast cancer-associated pro-teins accessible from the extracellular space. Our strategywas inspired by a recently described methodology in whichaccessible proteins within cancer-bearing organs are labeledwith reactive ester derivatives of biotin either by in vivo ter-minal perfusion of rodents [4] or by ex vivo perfusion of hu-man organs [7]. This method allows the covalent linking ofbiotin injected into the arterial vasculature onto primaryamines of proteins that are readily accessible from thebloodstream (e.g. extracellular matrix (ECM) proteins). Al-though powerful, this perfusion technique is restricted toexperimental animals or surgically resected organs vascular-ized by a catheterizable artery (e.g. kidney). Our new methodwas developed for the ex vivo biotinylation of human tissuesoriginating from small biopsies and nonperfusable organs(e.g. cancer lesions present in mastectomy or prostatectomyspecimens).

2 Materials and methods

2.1 Tissue harvesting

Cancerous and nontumoral human breast tissue sampleswere obtained from mastectomy specimens, immediatelysliced and soaked into freshly prepared EZ-link Sulfo NHS-LC biotin (1 mg/mL, Pierce) in PBS (pH 7.4). The timeframe from the excision in the operating theatre to thebiotinylation step required typically less than 10 min. Eachbiotinylation reaction was stopped by a 5 min incubation in50 mM Tris (pH 7.4). Tissue samples were then snap-frozenin liquid nitrogen, except for a tiny portion of each samplethat was directly immersed in formalin and then processedfor further histological and histochemical investigations.Additional tissue samples not included in the biotinylation

procedure were routinely processed for histopathologicaldiagnosis. Controls included adjacent tissue slices incubatedin PBS. The Ethics Committee of the University Hospital ofLiège reviewed and approved the specific protocol used inthis study, and written informed consent was obtained fromall the patients.

2.2 Histochemistry and immunohistochemistry

In order to assess the diffusion depth of reactive biotin esterderivatives in the tissues, formalin-fixed, paraffin-embeddedbreast tissue sections were incubated with avidin-peroxidaseconjugates with the use of the Vectastain ABC kit (VectorLaboratories, Burlingame, CA, USA), according to the man-ufacturer’s instructions. Immunohistochemical experimentswere performed as previously described [8]. For theimmunohistochemical detection of versican, antigen retrie-val was performed by incubating the slides with chon-droitinase (Sigma-Aldrich, Bornem, Belgium). Anti-versican(clone 12C5, Developmental Studies Hybridoma Bank at theUniversity of Iowa, Iowa City, IA, USA) antibody was appliedonto the sections at a dilution of 1:200. Control experimentsincluded omission of the primary antibody in the procedure.The list of normal human tissues and organs included in thetissue microarrays (TMAs) is provided in SupplementaryTable 3 online.

2.3 Sample processing

Pulverization of frozen biotinylated biopsies was performedusing a Mikro-Dismembrator U (Braun Biotech, Melsungen,Germany) and generated tissue powder. Approximately100 mg of tissue powder, both for normal and tumoral breasttissues were resuspended first in a PBS buffer containing aprotease inhibitor cocktail (Complete, Roche Diagnostics,Mannheim, Germany). Homogenates were sonicated(263000) with a 2 mm microprobe and soluble proteins weresubjected to a preclearing step consisting in HSA andimmunoglobulins (IgGs) depletion (Qproteome HSA andIgG Removal Kit, Qiagen). This step was included to limitthe number of HSA and IgG peptides detected by MS, butdid not hinder detection of low abundant proteins (data notshown). Insoluble pellet was resuspended in 2% SDS inPBS, and lysates were sonicated (36300 0). HSA- and IgG-depleted soluble protein fractions and detergent-solubilizedproteins were pooled and boiled for 5 min. Protein con-centration was determined using the BCA protein assayreagent kit (Pierce). Streptavidin–sepharose slurry (Amers-ham Biosciences, 150 mL/mg of total proteins) was equili-brated by three washes in buffer A (1% NP40 and 0.1% SDSin PBS), and protein binding was allowed for 2 h at roomtemperature in a rotating mixer. The resin was then washedtwice with buffer A, twice with buffer B (0.1% NP40, 1 MNaCl in PBS), twice with buffer C (0.1 M sodium carbonatein PBS, pH 11), and once with ammonium hydrogeno-carbonate (50 mM, pH 7.8). Binding of the biotinylated


1190 V. Castronovo et al. Proteomics 2007, 7, 1188–1196

proteins onto the resin and washing efficiencies werechecked by SDS-PAGE, and further either by CBB stainingor by blotting for subsequent detection of biotin by strept-avidin-HRP (Supplementary Fig. 1a). On-resin digestion wascarried out overnight at 377C with agitation, using modifiedporcine trypsin (Promega) in 100 mL final volume of ammo-nium hydrogenocarbonate (pH 7.8). The supernatants werecollected, protein concentration was determined, and onceevaporated, the peptides were resuspended in 0.1% formicacid. Samples were analyzed by nanocapillary LC-electro-spray MS/MS (nLC-ESI MS/MS).

2.4 MS

Peptide separation by RP LC was performed on an UltimateLC system (LC Packings) completed by a Famos autosamplerand a Swichos II Microcolumn switching device for sampleclean-up, fractionation, and preconcentration. Sample (5 mgin 20 mL at 0.25 mg/mL 0.1% formic acid) was first trapped ona SCX micro precolumn (500 mm id, 15 mm length, packedwith MCA50 bioX-SCX 5 mm; LC Packings) at a flow rate of30 nL/min followed by a micro precolumn cartridge(300 mm id, 5 mm length, packed with 5 mm C18 Pep-Map100; LC Packings). After 5 min, the precolumn wasconnected with the separating nanoflow column (75 mm id,15 cm length, packed with 3 mm C18 PepMAp100; LC Pack-ings) equilibrated in mobile phase A (0.1% formic acid in2:98 of ACN/degassed milliQ water). A linear elution gra-dient was applied with mobile phase B (0.1% formic acid in80:20 of ACN/degassed milliQ water) from 10 to 40% span-ning on 95 min. The outlet of the LC system was directlyconnected to the nanoelectrospray source of an Esquire HCTIT mass spectrometer (Bruker Daltonics, Germany), con-trolled by Esquire Control v5.2 and Hystar v3.0 (from theBruker Compass software bundle). Mass data acquisitionwas performed in the mass range of 50–2000 m/z using thestandard-enhanced mode (8100 m/z per second). For eachmass scan, a data-dependent scheme picked the 3 mostintense doubly or triply charged ions to be selectively isolatedand fragmented in the trap. The resulting fragments wereanalyzed using the Ultra Scan mode (m/z range of 50–3000at 26 000 m/z per second). SCX-trapped peptides were step-wise eluted with five salt concentrations (10, 20, 40, 80, and200 mM), each followed by the same gradient of mobilephase B.

2.5 Data processing and mgf file generation

Raw spectra were formatted in DataAnalysis software (BrukerDaltonics, v3.4 build 150). Portion of the chromatogram con-taining signal (i.e. with base peak chromatogram signal above50 000 arbitrary units) was processed to extract and deconvo-lute MS/MS spectra, without smoothing or background sub-traction. An S/N of 3 was applied to filtrate-irrelevant data inthe MS/MS spectra and generate the mass list. Chargedeconvolution was performed on both MS and MS/MS spec-

tra. A retention time of 1.5 min was allowed for compoundelution to minimize detection redundancy of parents ofidentical masses and charge states. Both deconvoluted andundeconvoluted data were incorporated in the mgf file.

2.6 Database searching

Protein identification was performed using different data-bases and different softwares featuring different built-inalgorithms. Proteins were first identified using the mini-mally redundant Swiss-Prot human protein database [9](release 49.5, SIB; Switzerland, 13 799 human entries),through the MS/MS ion search algorithm of the MASCOTsearch engine (MASCOT and MASCOT Daemon v2.1.0) [10]running on a local 4-processor computer cluster. The masstolerances of precursor and fragmented ions were set at 0.6and 0.3, respectively; allowed modifications were partial oxi-dization of methionines. One miscut was also allowed.Stringent filtration was deliberately used: the multi-dimensional protein identification technology (MudPIT)scoring was employed (this scoring is more aggressive thanthe usual standard scoring, and allow to remove protein thathave high protein scores essentially because they have a largenumber of low-scoring peptide matches), and ions score cut-off was set to 15. By setting the threshold to this value, all ofthe very low scoring (i.e. random peptide matches) were cutout, and homologous proteins were more likely to collapseinto a single hit. Furthermore, the absolute probability (P)was set to 0.01 (i.e. less than 1% probability of a randommatch). Despite this filtering, protein hits were manuallyinspected, particularly for interesting proteins identified byonly one peptide. These precautions were taken to ascertainthe accuracy of protein identified and reported in Table 1.The full protein list generated with the above parameterscontained 670 proteins (Supplementary Table 4). Additionalsearches were performed against the NCBI nonredundantdatabase (NCBInr, release 20 060 131, 134 668 humanentries) through the Phenyx web interface (PWI 2.1, Gene-Bio, Geneva, Switzerland). The mass tolerance of precursorions was set at 0.6; allowed modifications were partial oxidi-zation of methionines and partial lysin modifications withLC biotin. One miscut was also allowed. Proteins of interestlisted in Table 1 were identified by both Phenyx and MAS-COT search engines. The false–positive rate was estimated,for each sample, by dividing the number of peptides found inthe randomized Swiss-Prot database by the number of iden-tified peptides from the normal Swiss-Prot database, accord-ing to the following formula: fp = n random/n normal,where fp is the estimated false–positive rate, n random is thenumber of peptides identified (queries after filtering) fromthe random Swiss-Prot database, and n normal is the num-ber of peptides identified (queries after filtering) from thenormal Swiss-Prot database. Considering the ten normalbreast and the ten matched breast tumors, fp is equal to1.72 6 0.26% (Mean 6 SEM).



Table 1. Selection of proteins identified in the breast tumors (n = 10) and associated adjacent normal breast tissues (n = 10)

Swiss-Prot no.

Protein name Sequencecoverage(lowest–highest) (%)

MudPIT score(lowest–highest)

Location Nontumoralbreast

Tumoralbreast

P13611 Versican core protein precursor 0.2–1.5 40–221 E 0 10P08727 Keratin, type I cytoskeletal 19 10–55 212–1270 C 1 10P02751 Fibronectin precursor 0.7–12 56–3688 E 3 10Q15063 Periostin precursor 1–26 60–3383 E 5 10P12111 Collagen type VI alpha 3 7–26 73–573 E 10 10P51884 Lumican precursor 13–33 313–2009 E 10 10P05783 Keratin, type I cytoskeletal 18 4–25 49–543 C, N 0 8O95994 Anterior gradient protein 2 homolog precursor 6–42 39–775 ER 1 7P24821 Tenascin precursor 1–11 41–1471 E, Mb 1 5P02533 Keratin, type I cytoskeletal 14 3–17 75–417 C 5 1

The proteins were identified from the Swiss-Prot database (release 49.5) with MASCOT (v2.1). The numbers indicate in how many breastsamples each protein was detected. Location of the proteins was determined using both the Human Protein Reference Database(www.hprd.org) and Bioinformatic Harvester (harvester.embl.de). C: cytoplasmic, E: extracellular, ER: endoplasmic reticulum, Mb: plasmamembrane, and N: nuclear. See Supplementary Table 2 for detailed information concerning the peptides corresponding to each of theseproteins.

2.7 TMAs

Medium-density TMAs comprising 0.6 mm cores of variousnormal human tissues were assembled as previously descri-bed [11, 12]. Formalin-fixed, paraffin-embedded blocks ofnormal human tissues were retrieved from the Departmentof Pathology of the University Hospital of Liège. Most of thenormal tissues were from surgical specimens, except forneuronal tissues, which were obtained from autopsies.Initial sections were stained with haematoxylin and eosin toverify histology. Two TMAs were generated with the use of amanual tissue arrayer (Beecham Instruments). The list ofnormal tissue and organ specimens used for the generationof the TMAs is provided in Supplementary Table 3. Duplicatecores were included in the TMA for each specific tissue ororgan analyzed. A total of 120 cores were examined forimmunohistochemical expression of versican. Immuno-staining intensity was scored as absent, weak, moderate, orstrong by two observers (L. d. L. and D. W.).

3 Results

3.1 Principle of the approach and tissue biotinylation

Nonperfusable tissues were first incubated in a reactive bio-tin solution. Biotinylated proteins were purified on strept-avidin resin, and then directly digested on-resin, followed byshotgun MS analysis of the peptides (Fig. 1, SupplementaryFig. 1a and b).

We first investigated the effect of incubation time on themagnitude of tissue biotinylation, as assessed by histo-chemistry [4]. Diffusion extent is obviously dependent not

only on the soaking time, but also on the thickness and com-position of the tissue of interest. Increased soaking times weretested on breast tissue specimens with variable thicknesses. Ahistochemical analysis of tissue sections with streptavidin-peroxidase conjugates showed that labeling extent, indicativeof biotin penetration in the tissues, increased over time(Fig. 2). Since an immersion time of 20 min was found ap-propriate for 2 mm-thick human breast tissue slices, thisprocedure was used in all further experiments. Reproducibil-ity of the biotinylation step, evaluated by examining labeledprotein patterns of different samples from the same tumorspecimen, was found consistent (Supplementary Fig. 1c).

3.2 MS analysis

Expression profiles of biotinylated, accessible proteins in sevenductal and three lobular human breast carcinomas (Supple-mentary Table 1 online) and their matched nontumoral coun-terparts were then determined using the MudPIT technique,based on 2-D separation of tryptic peptide digest using nLC-ESIMS/MS. Reproducibility of the method was assessed by com-paring the protein lists generated by multiple runs of the samesample, and by comparing protein lists generated from threedifferent samples from the same tumor (SupplementaryFig. 1c). To assess the specificity of the streptavidin column,nonbiotinylated breast cancer samples were processed and an-alyzed by MS. The number of identified proteins was much lessin nonbiotinylated breast samples than in biotinylated sam-ples. Proteins found in nonbiotinylated samples (about tenproteins, data not shown) were most likely “sticky” proteinsthat unspecifically bound to the streptavidin beads, and con-sisted mainly of serum proteins (e.g. serum albumin, hemo-globin chains, IgGs) and cytokeratins.



Figure 1. Schematic description of the method used in this studyfor the identification of accessible proteins in cancerous andnontumoral human breast tissues. Ex vivo biotinylation of mat-ched healthy and diseased human tissue samples was carried outby incubating the tissues in a solution containing reactive esterderivatives of biotin. Biotinylated proteins from each tissue sam-ple were extracted and HSA and IgG were removed. Biotinylatedproteins were then captured on streptavidin resin, and submittedto on-resin proteolytic digestion after thorough washes. Biotin-ylated accessible proteins were then sequenced by shotgun MSusing nLC-ESI MS/MS. Identified accessible proteins differen-tially expressed in the tissue samples were further validated byimmunohistochemical analysis.

Table 1 shows a selection of ten accessible proteins iden-tified from biotinylated samples with high confidence (seeSupplementary Table 2 for scores, mass of precursors andcharge states of peptides corresponding to each protein, andSupplementary Table 4 for the full protein list). Several pro-teins of this list appeared to be cancer- or breast cancer-asso-ciated proteins (e.g. cytokeratins, anterior gradient proteinhomolog 2, periostin, and versican, among others). Interest-ingly, the membrane antigen ErbB2 was also identified byMS in the single tumor (out of the ten tumors tested) thatwas considered as ErbB2-positive by routine immunohisto-chemical assessment (Supplementary Table 1 online). Bioti-nylated proteins included extracellular and plasma mem-brane proteins, but also included a non-negligible fraction ofintracellular proteins. The reasons for this observation havealready been proposed elsewhere [4, 13], and include intra-cellular protein leakage when tissues are sliced, biotin pene-tration, and strong interactions between cytoplasmic andmembrane proteins. Nevertheless, our comparative analysisof nontumoral versus cancerous breast tissues identified sev-eral proteins known to be preferentially localized in theextracellular compartment. Stromal proteins that are selec-tively expressed in cancer tissues are indeed prime candi-dates for tumor targeting strategies because (i) they areexpected to be more accessible than intracellular proteins,(ii) they are often present in high quantities, and (iii) tumorcells frequently induce changes in the stromal compartment.This “reactive stroma” [14] may create a permissive and sup-portive environment contributing to cancer progression [15].Therefore, identification of stromal proteins specificallyinvolved in cancer is of considerable interest [5].

3.3 Identification of potential targets

Searching for accessible ECM proteins, we identified versi-can as being systematically and specifically (10 out of the 10sample pairs tested) detected in the breast cancer samples,but not in the matched normal counterparts (Table 1). Otherproteins were found to be differentially expressed, includingthe anterior gradient protein 2 homolog (AGR2), cyto-keratins, and periostin.

Versican, a large chondroitin-sulfate proteoglycan,secreted by stromal cells in the ECM, is a recognized celladhesion and motility modulator that may facilitate tumorcell invasion and metastasis [16, 17]. Anti-versican immuno-reactivity is increased in the peritumoral stromal matrices ofbreast [17, 18], prostate [19], and colon [20] cancers. In addi-tion, increased accumulation of versican in the stroma sur-rounding breast cancer cells is associated with a higher riskof relapse in node-negative, primary breast cancer [17, 18].Versican was identified from several peptides (up to five, seeSupplementary Table 2), and MS spectra of versican peptideswere thoroughly examined (Supplementary Fig. 1b). How-ever, due to the limitations inherent to the MS method used(discussed below), validation of the potential targets appearsmandatory.



Figure 2. Evaluation of increasing immersion times on the depth of tissue biotinylation. Thin slices of breast cancer tissues were soaked inthe biotinylation solution for various periods of time, as indicated. Extent of tissue biotinylation was assessed using histochemistry, asdescribed in Section 2. Original magnification: 1006.

AGR2, a human homolog of the Xenopus laevis cementgland anterior gradient protein, is a potentially secreted pro-tein. Interestingly, Tammen et al. [21] extracted peptidesfrom tissue slices, and also identified AGR2. Although AGR2was found to be a negative regulator of p53 [22], its functionremains largely unknown. AGR2 is, however, overexpressedin ER positive breast cancer cells and lesions [23, 24], as wellas in pancreatic [25] and prostate [26] cancer.

Periostin, a soluble heparin-binding N-glycosylatedECM-associated protein, is thought to contribute to celladhesion and motility [27]. It has been recently shown thathuman periostin is differentially expressed in both primarycolon carcinomas and metastatic tumors [28]. Interestingly,we found that periostin was differentially expressed in hu-man kidney cancer by searching for accessible targets usingour recently described perfusion method [29].

Finally, cytokeratins 18 and 19 were preferentiallydetected in tumor samples. This result is also in good agree-ment with the literature, since cytokeratins 18 and 19 werefound to be among the most abundant cytokeratins in carci-nomas [30].

3.4 Validation of a potential target, versican

A first validation step was performed with anti-versicanimmunostaining in human breast tissues (representativeexamples are shown in Fig. 3). While versican expressionaround normal breast glands and ducts was generally absent,the protein harbored moderate to strong immunoreactivityin the stromal compartment of the ten breast cancers ana-lyzed. These results, together with those from the above-referenced literature, are consistent with our MS analysisdetecting versican only in neoplastic tissues. In view of thesedata, it is tempting to propose versican as a potential targetprotein for breast cancer treatment. However, an ideal targetprotein should not only be specifically expressed in tumors,but also be absent in normal tissues. Since the distribution ofversican expression in the normal human body was largelyunknown, we undertook to examine in detail the presence ofthis ECM proteoglycan using TMAs comprising a wide vari-ety of normal human tissues and organs. Analysis of versi-can expression by immunoperoxidase staining revealed thatanti-versican immunoreactivity was absent in most of thenormal tissues, with only moderate reactivity detected in the

placenta and some areas of the CNS (Fig. 3 and Supplemen-tary Table 3 online). Since the CNS is not accessible to circu-lating, conjugated antibodies unless blood brain barrier isdisrupted [31], these in situ expression analyses furtheridentified versican as a promising target for antibody-basedanticancer treatments.

4 Discussion

Using a chemical proteomic method, we have identifiedpotentially accessible proteins that are of special interest fortargeted therapies in breast cancer lesions. Although power-ful enough to unveil biomarkers “as is”, our technique wouldmost probably benefit from future implementations of state-of-the-art proteomic techniques. For instance, MS analysiswas used to detect preferably “on–off” proteins, but a farmore challenging situation arises when there are only quan-titative differences among the sample sets, and these differ-ences were not addressed in this study. In addition, becauseMudPIT is now well established in many MS labs, we chosethis technique to validate our method for identifying acces-sible targets in tissues from nonperfusable organs as a “proofof principle”. However, it might be desirable to use MALDIas well, especially in combination with ESI (the two methodsbeing renowned to complement each other), to generate amore exhaustive, near-complete list of proteins. Indeed,shotgun LC-MS analyses of complex mixtures only yieldinformations for a fraction of relevant peptides in a singleanalytical run [32], leading to a phenomenon commonlyreferred to as “analytical incompleteness” [33]. Confidence ofanalytical completeness of our data (Supplementary Fig. 1c)was in good agreement with previously reported studies [34],and it is therefore obvious that incompleteness associatedwith a single analytical run per sample may influence thenumber of proteins identified. One could thus argue that ourmethod is biased toward the most abundant antigens. Thiswas, however, not an issue in our search for a reachable tar-get, since preferred targets for ligand-based tumor targetingapplications are ideally abundant, yet differentially expressedextracellular or membrane proteins. In this regard, ourmethod was powerful enough to unveil extracellular proteinsthat are specifically and differentially expressed in tumors,and among them, versican appeared to be the best accessible



Figure 3. Validation of versicanby in situ expression analyses.(a) Representative examples ofversican expression in cancer-ous and matched nontumoralhuman breast cancer tissues, asassessed by immunohisto-chemistry. Paraffin sections ofhuman breast tissues were sub-jected to immunoperoxidasestaining, as described in Section2. Original magnification: 100x.(b) Representative examples ofversican expression by immuno-peroxidase staining in variousnormal human tissues andorgans, as indicated (completelist provided in SupplementaryTable 3 available online), usingTMAs.

target. All the above-discussed points underline the needfor strong validation of the potential targets identified byMS. We suggest that validation of MS data should includein situ expression analyses of the target candidates not onlyin the tissues of interest, but also in normal tissues andorgans, for example, by using immunohistochemicalmethods.

In summary, we searched for specific and accessible bio-markers of human breast cancer, the most frequent form ofcancer and the second leading cause of cancer death inwomen living in industrialized countries. Our approach isalong the line of the work of Celis et al. [35, 36], who analyzedthe proteins found in the interstitial fluid that perfuses eitherthe breast tumor microenvironment or the adipose tissue,



and the work of Tammen et al. [21], who analyzed the pep-tides and low molecular weight proteins extracted from tis-sues. Both methods identified also potentially “accessible”proteins (without chemical modifications) and were highlypromising sources of biomarkers. To the best of our knowl-edge, the method described herein is novel because it allowsto also identify accessible proteins which are unlikely to leakout of the tissue such as extracellular proteins and trans-membrane proteins. Our methodology is theoretically appli-cable not only to cancerous tissues, but also to other patho-logic tissues resulting from inflammatory, degenerative,metabolic, and genetic alterations. We demonstrated that ourmethod is easy to perform and versatile enough to beexploited in the future for complete mapping of primarytumors and associated metastases. The accessibility of theidentified proteins may be particularly suited for targetingstrategies using an intravenous route. This method mayfacilitate the choice of custom-made treatments, targetingonly accessible proteins effectively found in routinely pro-cessed diseased tissues. Indeed, it may be anticipated thatonce the most relevant, accessible biomarkers specific for apatient’s disease are identified, high-affinity ligands such asrecombinant human antibodies and their fragments, can beprepared to assess the precise localization of the pathologiclesions and to selectively destroy them [37]. For instance, thehuman antibody L19, a specific ligand of the extracellular Bdomain of fibronectin, a biomarker of several cancer types[38, 39], is currently being tested in clinical trials, both as animaging tool (conjugated to radioactive iodine [40]) and as atherapeutic agent (fused with human interleukin-2 [41, 42]).We believe that our innovative approach will promote thedevelopment of targeted therapies and may thus represent asignificant step toward a clean and effective fight againstdiseases and particularly cancer.

We thank P. Heneaux and S. Pierard at the Metastasis Re-search Laboratory for their technical assistance. This work wassupported by the EU FP6 framework program STROMA. Grantsponsors of this work include also the National Fund for ScientificResearch (Belgium), the Centre Anti-Cancéreux près l’Universitéde Liège, the Léon Frédéricq foundation, TELEVIE, and theInteruniversity Attraction Pole Program – Belgian Science Policy.The Esquire HCTdevice belongs to the GIGA proteomic platform.Philippe Kischel is a Research Fellow of the Belgian NationalFund for Scientific Research.

5 References

[1] Wu, A. M., Senter, P. D., Nat. Biotechnol. 2005, 23, 1137–1146.

[2] Adams, G. P., Weiner, L. M., Nat. Biotechnol. 2005, 23, 1147–1157.

[3] Hortobagyi, G. N., Semin. Oncol. 2004, 31, 21–27.

[4] Rybak, J. N., Ettorre, A., Kaissling, B., Giavazzi, R. et al., Nat.Methods 2005, 2, 291–298.

[5] Neri, D., Bicknell, R., Nat. Rev. Cancer 2005, 5, 436–446.

[6] Sandhu, C., Connor, M., Kislinger, T., Slingerland, J., Emili,A., J. Proteome Res. 2005, 4, 674–689.

[7] Castronovo, V., Waltregny, D., Kischel, P., Roesli, C. et al.,Mol. Cell. Proteomics 2006, 5, 2083–2091.

[8] Waltregny, D., Bellahcene, A., Van Riet, I., Fisher, L. W. et al.,J. Natl. Cancer Inst. 1998, 90, 1000–1008.

[9] Nesvizhskii, A. I., Aebersold, R., Mol. Cell. Proteomics 2005,4, 1419–1440.

[10] Perkins, D. N., Pappin, D. J., Creasy, D. M., Cottrell, J. S.,Electrophoresis 1999, 20, 3551–3567.

[11] Kononen, J., Bubendorf, L., Kallionimeni, A., Barlund, M. etal., Nat. Med. 1998, 4, 844–847.

[12] Perrone, E. E., Theoharis, C., Mucci, N. R., Hayasaka, S. et al.,J. Natl. Cancer Inst. 2000, 92, 937–939.

[13] Peirce, M. J., Begum, S., Saklatvala, J., Cope, A. P., Wait, R.,Proteomics 2005, 5, 2417–2421.

[14] Walker, R. A., Breast Cancer Res. 2001, 3, 143–145.

[15] De Wever, O., Mareel, M., J. Pathol. 2003, 200, 429–447.

[16] Evanko, S. P., Angello, J. C., Wight, T. N., Arterioscler.Thromb. Vasc. Biol. 1999, 19, 1004–1013.

[17] Ricciardelli, C., Brooks, J. H., Suwiwat, S., Sakko, A. J. et al.,Clin. Cancer Res. 2002, 8, 1054–1060.

[18] Suwiwat, S., Ricciardelli, C., Tammi, R., Tammi, M. et al.,Clin. Cancer Res. 2004, 10, 2491–2498.

[19] Ricciardelli, C., Mayne, K., Sykes, P. J., Raymond, W. A. et al.,Clin. Cancer Res. 1998, 4, 963–971.

[20] Mukaratirwa, S., van Ederen, A. M., Gruys, E., Nederbragt,H., J. Comp. Pathol. 2004, 131, 259–270.

[21] Tammen, H., Kreipe, H., Hess, R., Kellmann, M. et al., BreastCancer Res. Treat. 2003, 79, 83–93.

[22] Pohler, E., Craig, A. L., Cotton, J., Lawrie, L. et al., Mol. Cell.Proteomics 2004, 3, 534–547.

[23] Fletcher, G. C., Patel, S., Tyson, K., Adam, P. J. et al., Br. J.Cancer 2003, 88, 579–585.

[24] Thompson, D. A., Weigel, R. J., Biochem. Biophys. Res.Commun. 1998, 251, 111–116.

[25] Missiaglia, E., Blaveri, E., Terris, B., Wang, Y. H. et al., Int. J.Cancer 2004, 112, 100–112.

[26] Zhang, J. S., Gong, A., Cheville, J. C., Smith, D. I., Young, C.Y., Genes Chromosomes Cancer 2005, 43, 249–259.

[27] Gillan, L., Matei, D., Fishman, D. A., Gerbin, C. S. et al., Can-cer Res. 2002, 62, 5358–5364.

[28] Bao, S., Ouyang, G., Bai, X., Huang, Z. et al., Cancer Cell.2004, 5, 329–339.

[29] Castronovo, V., Waltregny, D., Kischel, P., Roesli, C. et al.,Mol. Cell. Proteomics 2006, 5, 2083–2091.

[30] Quinlan, R. A., Schiller, D. L., Hatzfeld, M., Achtstatter, T. etal., Ann. N. Y. Acad. Sci. 1985, 455, 282–306.

[31] Baloyannis, S. J., Gonatas, N. K., J. Neuropathol. Exp. Neu-rol. 1979, 38, 519–531.

[32] Durr, E., Yu, J., Krasinska, K. M., Carver, L. A. et al., Nat. Bio-technol. 2004, 22, 985–992.

[33] Wilkins, M. R., Appel, R. D., Van Eyk, J. E., Chung, M. C. et al.,Proteomics 2006, 6, 4–8.

[34] Elias, J. E., Haas, W., Faherty, B. K., Gygi, S. P., Nat. Methods2005, 2, 667–675.



[35] Celis, J. E., Gromov, P., Cabezon, T., Moreira, J. M. et al., Mol.Cell. Proteomics 2004, 3, 327–344.

[36] Celis, J. E., Moreira, J. M., Cabezon, T., Gromov, P. et al., Mol.Cell. Proteomics 2005, 4, 492–522.

[37] Schrama, D., Reisfeld, R. A., Becker, J. C., Nat. Rev. DrugDiscov. 2006, 5, 147–159.

[38] Zardi, L., Carnemolla, B., Siri, A., Petersen, T. E. et al., Embo.J. 1987, 6, 2337–2342.

[39] Pini, A., Viti, F., Santucci, A., Carnemolla, B. et al., J. Biol.Chem. 1998, 273, 21769–21776.

[40] Berndorff, D., Borkowski, S., Sieger, S., Rother, A. et al., Clin.Cancer Res. 2005, 11, 7053s–7063s.

[41] Ebbinghaus, C., Ronca, R., Kaspar, M., Grabulovski, D. et al.,Int. J. Cancer 2005, 116, 304–313.

[42] Menrad, A., Menssen, H. D., Expert Opin. Ther. Targets 2005,9, 491–500.


identification of specific reachable molecular targets in human breast cancer using a versatile ex...

Documents