Macrophages in T cell/histiocyte rich large B cell lymphoma strongly express metal-binding proteins and show a bi-activated phenotype

Authors


Correspondence to: Dr. Sylvia Hartmann, Senckenberg Institute of Pathology, University of Frankfurt, Theodor-Stern-Kai 7, D-60590 Frankfurt, Germany, Tel.: +49–69-6301-4284, Fax: +49–69-6301-5241, E-mail: s.hartmann@em.uni-frankfurt.de

Abstract

Abundant macrophage infiltration in tumors often correlates with a poor prognosis. T cell/histiocyte rich large B cell lymphoma (THRLBCL) is a distinct aggressive B cell lymphoma entity showing a high macrophage content. To further elucidate the role of tumor-associated macrophages in THRLBCL, we performed gene expression profiling of microdissected histiocyte subsets of THRLBCL, nodular lymphocyte predominant Hodgkin lymphoma (NLPHL), Piringer lymphadenitis, sarcoidosis, nonspecific lymphadenitis and monocytes from peripheral blood. In a supervised principal component analysis, histiocytes from THRLBCL were most closely related to epithelioid cells from NLPHL, with both types of cells expressing genes related to proinflammatory and regulatory macrophage activity. Moreover, histiocytes from THRLBCL strongly expressed metal-binding proteins like MT2A, by which histiocytes of THRLBCL can be distinguished from the other histiocyte subsets investigated. Interestingly, the validation at the protein level showed a strong expression of TXN, CXCL9, MT2A and SOD2 not only in macrophages of THRLBCL but also in the tumor cells of NLPHL and classical Hodgkin lymphoma (cHL). Overall, the present findings indicate that macrophages in the microenvironment of THRLBCL have acquired a distinct gene expression pattern that is characterized by a mixed M1/M2 phenotype and a strong expression of several metal binding proteins. The microenvironments in NLPHL and THRLBCL appear to have a similar influence on the macrophage phenotype. The high expression of metal binding proteins in histiocytes of THRLBCL may be diagnostically useful, but a potential pathophysiological role remains to be identified.

A high content of tumor-associated macrophages (TAM) has been shown to be associated with a poor clinical outcome in many malignancies including lymphoma.[1-3] Several gene expression profiling studies on diffuse large B cell lymphoma (DLBCL) reported a stromal or host response signature to be an adverse prognostic factor reflecting the importance of the tumor microenvironment.[4-6] T cell/histiocyte-rich large B cell lymphoma (THRLBCL) is a variant of DLBCL with a striking background of non-neoplastic reactive immune cells composed mostly of T cells and histiocytes/macrophages.[7, 8] The histological appearance is characterized by scattered, single large neoplastic B cells of germinal center (GC) origin[9] which make up only a minority of the infiltrate. Small B cells are virtually absent. THRLBCL usually presents with an advanced stage including liver, spleen and bone marrow infiltration.[10] There are conflicting data on the clinical behavior of THRLBCL with some studies observing a more aggressive course than conventional DLBCL,[10, 11] whereas others found a comparable outcome in both entities.[12, 13]

Although THRLBCL generally is considered to be an aggressive B cell lymphoma, there is a considerable overlap with nodular lymphocyte predominant Hodgkin lymphoma (NLPHL).[14-16] Both diseases show a predominance of young adult males and there is even some overlap in the histologic picture: In both lymphomas the tumor cells represent only a minority in the infiltrate, they show an almost identical immunophenotype with a preserved B cell phenotype and ongoing somatic hypermutation.[9, 17, 18] However, the clinical behavior of NLPHL is usually indolent. In a previous study only few differentially expressed genes between the microdissected tumor cells of both entities were found.[19] However, the microenvironment in NLPHL is considerably different from THRLBCL with only few histiocytes differentiated into epithelioid cells (EC). By pure morphology, histiocytes can be classified in EC and tissue macrophages. EC are usually found in granulomatous diseases like infectious granulomas or sarcoidosis. These lesions are characterized by strong inducible nitric oxide synthase activity and Th1-type cytokine secretion.[20] Applying molecular techniques, histiocytes can be divided into pro-inflammatory/classically activated M1 or anti-inflammatory/alternatively activated M2 subtypes.[21] Considering this dichotomy, TAM have been frequently associated with an M2- or M2-like polarization.[22] Thus far, there is only little evidence whether M1 and M2 polarization is reflected in a different morphologic appearance of histiocytes. Most recent work suggests that there is significant phenotypic heterogeneity of TAM exhibiting various tumor-supporting roles such as angiogenesis, tumor cell invasion and immunosuppression.[23]

In the present study, we therefore aimed to characterize the specific gene expression pattern of histiocytes in THRLBCL in comparison to histiocyte subsets from NLPHL and reactive conditions by gene expression profiling. We also aimed to obtain insight whether these particular histiocytes support tumorigenesis or represent a host reaction.

Design and Methods

Macrophages of eight THRLBCL, tingible body macrophages of GCs and macrophages from the marginal sinus of six cases of nonspecific lymphadenitis as well as EC of nine NLPHL, five Piringer lymphadenitis (toxoplasmosis infection) and five cases of sarcoidosis were microdissected from frozen tissue sections. Samples of Piringer lymphadenitis and sarcoidosis were included since EC under these conditions morphologically resemble most closely EC of NLPHL, whereas TAM in THRLBCL are morphologically more similar to sinus and starry sky macrophages.

RNA integrity of the samples was assessed before the microdissection procedure was started. For this purpose, the total RNA of all samples was isolated using the RNeasy Mini Kit (QIAGEN, Hilden, Germany) to check the RNA quality with a 2100 Bioanalyzer Nano Assay (Agilent Technologies, Böblingen, Germany). Only samples with an RNA integrity number (RIN) ≥ 7 and an excellent morphology in the frozen tissue were included in the study. About 5-µm thick sections of the frozen tissue were mounted on membrane-covered slides (PALM, Bernried, Germany), which had previously been irradiated under an UV lamp for 30 min. The slides were fixed for 10 sec in 100% ice cold ethanol, then incubated with ice cold hematoxylin containing 200 U/ml RNase inhibitor (Roche, Grenzach, Germany) for 4 min, washed in ice cold RNase-free molecular biology grade water for 2 min, incubated in 2% eosin for 15 sec, washed again, fixed for 10 sec in 100% ice cold ethanol and dried at room temperature for 10 min. For the microdissection procedure a Microdissection Axiovert 200M microscope (PALM) was used. Two hundred twenty histiocytes were chosen from six to twenty serial sections, depending on the number of histiocytes per section, which could be identified with confidence. Macrophages/EC were identified by their typical abundant cytoplasm and oval or elongated nuclei by an expert pathologist. Sinus macrophages were selected from reactive lymph nodes with sinus histiocytosis. Starry sky macrophages were microdissected from reactive lymph nodes with massive follicular hyperplasia. Between five and fifteen bona fide starry sky macrophages were isolated from a single GC, depending on the size of the GC and the number of starry sky macrophages present. Cells were catapulted in PALM adhesive caps and lysed with 2 µl NUGEN Direct Lysis Buffer (NUGEN, Bemmel, The Netherlands). After lysis, RNA was amplified with the WT-Ovation-One-direct-Kit (NUGEN) according to the manufacturer's protocol. The next quality control was done after transcription into cDNA, when the amount and size distribution of cDNA fragments obtained were analyzed on a 2100 Bioanalyzer Nano Assay. After successful amplification, the samples were labeled with the Encore Biotin Module (NUGEN) and hybridized onto Human Gene 1.0 ST Arrays (Affymetrix, Santa Clara, CA). Five samples of classical CD14+CD16 and four samples of the alternatively activated CD14+CD16+ monocytes, which were assumed to be precursors of M2-polarized macrophages,[24, 25] were fluorescence activated cell sorting-purified from peripheral blood of healthy donors with a purity of ≥ 95.6%. Cells were lysed in NUGEN Direct Lysis Buffer and processed as mentioned above. Detailed information on sample purification of the monocytes and statistical analysis is found in the Supporting Information Methods section. Gene expression data are available on GEO (GSE42495).

Immunohistochemistry for confirmation purposes was performed as described previously.[26] Immunohistochemical stainings for MT2A, CD163, TXN, CXCL9, SOD2, S100A8 and FTH1 were performed on paraffin sections of an independent series of lymph nodes from patients with THRLBCL (N = 11), NLPHL (N = 10), sarcoidosis, Piringer and nonspecific lymphadenitis (N = 8, each) as well as Epstein–Barr virus-positive mixed cellularity Hodgkin lymphoma with a high content of macrophages (N = 10). Antibodies and staining procedures are summarized in Supporting Information Table 1.

The study was approved by the local ethics committee of Frankfurt University Hospital.

Results

Relatedness of different histiocyte subsets

Gene expression profiling was performed for five to nine samples each of sorted peripheral blood monocytes (PBM) and microdissected histiocytes of THRLBCL, NLPHL, sarcoidosis, Piringer and nonspecific lymphadenitis. Unsupervised hierarchical clustering did not separate the 48 profiles into distinct branches, indicating a lack of consistent differences between the histiocyte subsets (not shown). Therefore, the relationship of the different subsets was examined using a principal components analysis of the mean expression vectors of each entity (Fig. 1a). Here, macrophages of GC and sinuses were closely related and showed a large distance to EC of Piringer lymphadenitis and sarcoidosis. Interestingly, TAM of THRLBCL and EC from NLPHL were closely related and located intermediate between all other samples, more closely related to GC and sinus macrophages.

Figure 1.

(a) Relatedness of different histiocyte entities by principal component analysis based on the 496 differentially expressed probesets showing a standard deviation ≥ 1. Red: THRLBCL macrophages, turquoise: EC from NLPHL, yellow: GC tingible body macrophages, gray: sinus macrophages, pink: Piringer EC, orange: Sarcoidosis EC, blue: CD14+ fluorescence activated cell sorting-purified monocytes, green: CD14+CD16+ fluorescence activated cell sorting-purified monocytes. (b) Expression of M1-related genes based on Martinez et al.[30] separates THRLBCL macrophages (red) from M0-polarized PBM (blue and green). (c) Expression of M2-related genes based on Martinez et al.[30] separates THRLBCL macrophages (red) from PBM. It also separates CD14+CD16+ monocytes (green), the precursors of M2-polarized macrophages, from CD14+ monocytes.

TAM in THRLBCL express metal-binding and proinflammatory genes

In THRLBCL macrophages, the highest and most specifically upregulated gene was MT2A compared to all other subsets profiled. MT2A was 21-fold upregulated compared to tingible body macrophages from GC and 9.6-fold compared to PBM (Supporting Information Tables 2 and 3). The only differentially expressed genes between THRLBCL macrophages and NLPHL EC were members of the metallothionein (MT) family (MT2A 9.5-fold, MT1H 4-fold and MT1G 2.4-fold, Supporting Information Table 4). Overexpression of MTs was most specific in THRLBCL macrophages, whereas EC of Piringer lymphadenitis showed a weak expression of MT2A mRNA. Additionally, THRLBCL macrophages overexpressed mRNAs of other metal-binding proteins like the calcium-binding protein S100A8 (4.7-fold), ferritin heavy chain (FTH1, 3.7-fold) and genes involved in redox-reactions (TXN 4.9-fold, SOD2 2.8-fold) compared to GC macrophages. Other proinflammatory genes like the interferon-gamma induced chemokine ligand CXCL9 (5.7-fold upregulated compared to PBM), STAT1 (2.9-fold upregulated compared to GC macrophages) as well as several HLA-associated transcripts (HLA-DQA1 2.4-fold, BAT3 2.3-fold, HLA-DOA 2.2-fold, compared to PBM) were highly expressed in THRLBCL macrophages.

Gene expression in different histiocyte subsets

CD14+CD16 monocytes showed compared to CD14+CD16+ monocytes a strong overexpression of proinflammatory S100A8 and S100A12 genes as reported previously (9- and 10.3-fold).[24, 27] All microdissected samples were compared to the merged group of PBM and showed except sarcoidosis EC a strong upregulation (12- to 15.4-fold) of UBD, which was shown to mediate tumor necrosis factor-alpha induced nuclear factor ‘kappa-light-chain-enhancer’ of activated B-cells (NF-κB) activation.[28] CD209, a C-type lectin pathogen recognition and adhesion receptor, controlling migration and trafficking of macrophages and dendritic cells,[29] was specifically expressed by sinus macrophages. EC of Piringer lymphadenitis, NLPHL and TAM of THRLBCL shared several upregulated genes like MUC12, CXCL9 and SLAMF7. CTSB, GM2A and RFC1 were upregulated in EC of sarcoidosis and Piringer lymphadenitis only.

Macrophage polarization

To further evaluate whether the microdissected histiocytic subsets show a predominant polarization, we analyzed the expression of typical M1- and M2-related genes based on the study by Martinez et al.[30] Whereas EC from sarcoidosis showed an M1-skewed phenotype, in tingible body macrophages from GC neither an expression of M1- nor M2-related genes was found. All other subsets showed an expression of both M1 and M2 genes (Figs. 1b and 1c). In TAM of THRLBCL, a median of 22 M1-related genes and 19 M2-related genes were expressed per case, whereas the PBM expressed a median of 14 M1- and 11 M2-related genes per sample. The expression of the respective M1- and M2-related genes was quite heterogeneous in the specific samples (Figs. 1b and 1c, Supporting Information Tables 5 and 6). The top M1 genes CXCL9, CCL5, CCR7, IL6, IL23A, IRF7, IGFBP4, IL15RA, CCL19, AK3, STAT1, PSME2 and SLC2A6 were expressed in all subsets of THRLBCL macrophages. The following M1-related genes were significantly upregulated compared to the PBM (t-test, p < 0.05): CXCL9, CCL5, CCR7, IL23A, IRF7, IGFBP4, IL15RA, CCL19, AK3, STAT1, IL2RA, SLC7A5, GADD45G, PSMA2, PFKFB3, APOL1, SPHK1, IL15, PFKP, NOS2 and IL12B. The number of M2 genes expressed in the macrophages in all THRLBCL cases was less: SEPP1, MS4A4, MMP1, HNMT, PPARG and SLC38A6. Similarly, there were less significantly upregulated M2-related genes in TAM of THRLBCL compared to PBM (t-test, p < 0.05): SEPP1, MS4A4, CERK, SLC4A7, ALOX15, IL10, ARG1, TGFBR2, CTSC, HS3ST1, VEGFA, CXCR4 and FLG2. We also performed gene set enrichment analyses of the supervised comparisons of microdissected samples with CD14+CD16+ PBM. However, due to the strong amplification and the heterogeneity of the microdissected samples no significant enrichment scores neither for M1 nor M2 were obtained (data not shown).

Validation of protein expression in additional primary cases

To further validate the gene expression data and to evaluate whether the upregulated mRNA levels correspond to elevated protein levels, we investigated additional cases of all lesions for protein expression of seven genes (MT2A, CXCL9, TXN, CD163, S100A8, FTH1 and SOD2, Supporting Information Fig. 1, Table 1). These genes were selected as they were most strongly upregulated in the comparison of THRLBCL macrophages compared to GC macrophages. Since particularly the mixed cellularity subtype of classical Hodgkin lymphoma (cHL) frequently presents a microenvironment with abundant EC and macrophages and can represent a differential diagnosis to THRLBCL, we decided to enlarge the series of immunohistochemical stainings and additionally include also histiocyte rich cHL cases. TXN and FTH1 proved as particularly good markers for EC in all lesions but were also expressed in TAM of THRLBCL. This could also be expected from the supervised comparisons of these histiocyte subsets with GC macrophages (FTH1 2.2- to 8.5-fold upregulated and TXN 2.5- to 4.8-fold upregulated). The GC macrophages showed no reaction with these antibodies. CXCL9 was strongly expressed in the TAM of THRLBCL and cHL but showed only an inconsistent expression in EC of some cases of Piringer lymphadenitis (4/8). Expression of CD163 was shared by THRLBCL macrophages, cHL macrophages and in sinus macrophages, but CD163 was negative in EC. S100A8 was more specifically expressed in the TAM of THRLBCL (11/11) and cHL (10/10), but EC and sinus macrophages were negative. A strong expression of MT2A combined with an expression of CD163 was a unique finding in THRLBCL macrophages (11/11, Fig. 2). TAM of cHL expressed MT2A in only 5/10 cases and if positive, the number of MT2A-positive macrophages in the cHL cases was low compared with the total macrophage content (Fig. 2). Thus, the immunohistochemical findings nicely validate the differential expression of the genes identified in the mRNA expression study at the protein level.

Table 1. Immunostainings performed for confirmation purposes on an independent series of lymph node biopsies
 NonspecificGranulomatous epithelioid cellsTHRLBCLNLPHLcHL mixed cellularity
 Nonspecific lymphadenitis (Germinal centre macrophages)Nonspecific lymphadenitis (Sinus macrophages)Sarcoidosis (Epithelioid cells)Piringer (Epithelioid cells)THRLBCL (Macrophages)THRLBCL (Tumor cells)NLPHL (Macrophages)NLPHL (Epithelioid cells)NLPHL (Tumor cells)cHL (Macrophages)cHL (Epithelioid cells)cHL (Tumor cells)
  1. The number of positively staining cases is indicated/the total number of cases stained.

MT2A0/80/83/82/811/110/113/100/107/105/100/107/10
CD1631/88/82/80/811/110/115/100/100/1010/100/100/10
TXN0/81/88/88/810/112/115/1010/100/102/107/108/10
CXCL90/80/80/84/811/111/114/100/101/107/100/104/10
SOD21/82/85/88/811/112/115/107/105/106/1010/102/10
S100A80/83/82/81/810/110/115/101/100/1010/100/100/10
FTH12/82/87/87/810/110/112/106/100/102/108/100/10
Figure 2.

Expression of MT2A (a, c, e) and CD163 (b, d, f) in cases of THRLBCL, NLPHL and cHL. Macrophages in THRLBCL (a, b) show a strong coexpression of MT2A and CD163. Even if numerous macrophages were visible in NLPHL (c, d, diffuse variant) and cHL (e, f) in the CD163-immunostaining, the number and intensity of MT2A expressing macrophages was much lower.

Interestingly, MT2A was also expressed in the tumor cells of NLPHL, the LP cells, in 7/10 cases, but never in the tumor cells of THRLBCL. In cHL MT2A was expressed in the Hodgkin- and Reed-Sternberg cells (7/10 cases), occasionally in a small subpopulation of TAM and in a subset of lymphocytes (Fig. 3). Additionally, we observed expression of CXCL9 and TXN in the tumor cells of THRLBCL in 2 and 1 of 11 cases, respectively (Table 1, Fig. 3) as well as CXCL9 expression in the LP cells of one NLPHL case. CXCL9 and TXN were more frequently expressed in the Hodgkin- and Reed-Sternberg cells of cHL (8 and 4 of 10 cases, respectively, Table 1).

Figure 3.

Expression of proinflammatory factors found upregulated in THRLBCL macrophages in the tumors cells of NLPHL (a, c, e, g) and cHL (b, d, f, h), 400×. MT2A (a, b), CXCL9 (c, d), TXN (e, f), SOD2 (g, h).

Discussion

In the present study, we performed gene expression profiling of different histiocyte subsets in neoplastic and reactive lesions. We aimed to further characterize TAM in THRLBCL and found a large overlap of upregulated genes in TAM of THRLBCL with EC from NLPHL as well as from Piringer lymphadenitis and sarcoidosis. This is surprising since TAM have been associated with an M2 polarization[22] whereas granulomatous diseases usually display a Th1 phenotype and show proinflammatory properties.[20] In line with these observations we found an expression of proinflammatory factors like FTH1 and TXN in all kinds of EC and TAM of THRLBCL. Ferritin heavy chain upregulation correlates with increased iron retention, described as a bacteriostatic mechanism and usually found in M1-polarized macrophages.[31] Additional genes with proinflammatory properties like CXCL9, STAT1 and HLA-associated transcripts—usually expressed in M1-polarized activated macrophages[32]—were expressed in TAM of THRLBCL. By applying immunohistochemistry, we validated S100A8 expression specifically in TAM of THRLBCL. Additionally, we observed an expression of the proinflammatory S100A8 protein, which can generate reactive oxygen species, in TAM of cHL.[33] However, we did not only observe an upregulation of M1-related genes in THRLBCL macrophages, but also genes typical for M2 polarization like the scavenger receptor CD163 as well as CCL18 (2.7- and 2.1-fold upregulated compared to PBM).[32] Surprisingly, CD163 has recently also been identified on a subset of M1 polarized macrophages in psoriasis.[34] Expression of both M1- and M2-related genes by Martinez et al.[30] was observed in the histiocytes of THRLBCL, NLPHL, Piringer lymphadenitis as well as sinus macrophages. Therefore, we observed a bi-activated phenotype with expression of both pro- and anti-inflammatory factors in TAM of THRLBCL, but also in EC of NLPHL and Piringer lymphadenitis. This biactivated phenotype has also been described for TAM of squamous cell carcinoma.[35] However, this is an unexpected finding as TAM of many tumors were characterized as M2-like,[21, 36] such as the tumor tolerogenic microenvironment as it was observed in THRLBCL in the study by Van Loo et al.[37] This study focused on the tumor tolerogenic effect of TAM in THRLBCL mediated by VSIG4, which we confirm to be expressed in the macrophages in 6 of 8 THRLBCL in the present study. However, in the same study by van Loo et al.[37] also several genes could be identified encoding for proteins that are upregulated in macrophages and dendritic cells upon treatment with interferon-γ, confirming at least a partial M1 polarization. Since EC in Piringer lymphadenitis (toxoplasmosis infection) and NLPHL also expressed M2-related genes, they differed from EC in sarcoidosis not only on morphologic grounds (scattered distribution and not forming dense granulomas) but also in their gene expression program. In line with our findings, a mixed Th1/Th2 response was observed in mice after vaccination with Toxoplasma gondi DNA or protein.[38] Granuloma forming EC may therefore be indicative of an M1 phenotype, as observed in the present study in sarcoidosis and in a previous study in tuberculosis.[20] A potential role as morphologic favorable prognostic factor should therefore be evaluated in further studies.

The most specifically expressed genes in TAM of THRLBCL were members of the MT family. MTs are cysteine-rich metal-binding proteins which bind and release copper and hereby modulate redox cycling.[39] There are conflicting data on MT functions showing prooxidant functions[40] as well as antioxidant and antiapoptotic effects.[41] MTs are expressed in the tumor cells of various cancers.[42-44] Interestingly, we observed expression of MT2A protein as well as CXCL9, TXN and SOD2, which are strongly expressed in histiocytes of THRLBCL, also in the tumor cells of cHL. Expression of CXCL9 and TXN in the tumor cells of cHL and DLBCL has previously been described,[45-47] and TXN expression in the tumor cells of DLBCL was associated with decreased patient survival.[47] Unfortunately, the number of cHL cases investigated in the present study was too small to draw any conclusions from the heterogeneous expression patterns observed in terms of clinicopathological correlations. This should be evaluated in further studies. Since we only rarely observed an expression of these proteins in the tumor cells of THRLBCL, CXCL9 and TXN expression by the tumor cells does not appear to be a major pathophysiological factor in THRLBCL. Whereas the Hodgkin- and Reed-Sternberg cells in cHL show a great plasticity and adopt functions of other cell lineages,[48-51] tumor cells in THRLBCL generally appear to have a preserved B cell phenotype,[19, 52, 53] and do not show this plasticity. They might therefore benefit from the high amount of macrophages in the microenvironment, in order to shape the tumor microenvironment for their benefit. However, this hypothesis should be corroborated in further studies.

In summary, THRLBCL macrophages as well as EC from NLPHL and Piringer lymphadenitis share many common features like expression of both genes related to M1 and M2 functions. However, TAM in THRLBCL show a unique expression of mRNAs of metal-binding proteins, like the members of the MT family. The specific combination of a strong expression of MT2A and CD163 in TAM of THRLBCL may therefore help in the differential diagnosis to other histiocyte rich lymphoma infiltrates. Macrophages in THRLBCL adopt proinflammatory functions which can otherwise be observed in tumor cells of NLPHL and cHL. Paradoxically, secretion of proinflammatory factors by macrophages in THRLBCL therefore seems to be an important factor for tumor survival. Whether it is also responsible for the extensive tumor dissemination, usually observed in THRLBCL,[7] should be evaluated in further studies.

Acknowledgements

The authors thank Sabine Albrecht, Gaby Haley, Christiane Kehm, Ralf Lieberz, Yvonne Michel, Christiane Wenk and Emilie Bittoun for excellent technical assistance and Dr. Xavier Sagaert for kind help in the selection of cases. This project was funded by the Deutsche Forschungsgemeinschaft (grant HA 6145/1-1). Ralf Küppers is supported by the Deutsche Forschungsgemeinschaft (KU1315/7-1). Christian Steidl is supported by the Cancer Research Society (Steven E. Drabin fellowship). Randy D. Gascoyne is supported by a Canadian Institutes for Health Research grant (178536) and the Terry Fox Foundation (Program Project grant 019001). Thomas Tousseyn is supported by the K.U.Leuven (grant GOA/11/010) and the FWO-Vlaanderen (G081411N). The authors report no potential conflict of interest.

Authors Contribution

S.H.: Experimental design, microdissection, analysis and interpretation of array data, immunohistochemistry, drafting of the manuscript; T.T., A.H., M.P.: pathologic evaluation, supplied essential material; C.D.: development of evaluation algorithms, statistical analysis of data; P.F.: microdissection, analysis of data; H.H.: fluorescence activated cell sorting procedure, interpretation of data; R.K., C.S.: interpretation of data, drafting of the manuscript; R.G., F.F., C.P.W.: pathologic evaluation, supplied essential material, drafting of the manuscript; M.L.H.: experimental design, drafting of the manuscript.

Ancillary