Purpose: The tumor microenvironment is formed by many distinct and interacting cell populations, and its composition may predict patients' prognosis and response to therapies. Colorectal cancer is a heterogeneous disease in which immune classifications and four consensus molecular subgroups (CMS) have been described. Our aim was to integrate the composition of the tumor microenvironment with the consensus molecular classification of colorectal cancer.

Experimental Design: We retrospectively analyzed the composition and the functional orientation of the immune, fibroblastic, and angiogenic microenvironment of 1,388 colorectal cancer tumors from three independent cohorts using transcriptomics. We validated our findings using immunohistochemistry.

Results: We report that colorectal cancer molecular subgroups and microenvironmental signatures are highly correlated. Out of the four molecular subgroups, two highly express immune-specific genes. The good-prognosis microsatellite instable–enriched subgroup (CMS1) is characterized by overexpression of genes specific to cytotoxic lymphocytes. In contrast, the poor-prognosis mesenchymal subgroup (CMS4) expresses markers of lymphocytes and of cells of monocytic origin. The mesenchymal subgroup also displays an angiogenic, inflammatory, and immunosuppressive signature, a coordinated pattern that we also found in breast (n = 254), ovarian (n = 97), lung (n = 80), and kidney (n = 143) cancers. Pathologic examination revealed that the mesenchymal subtype is characterized by a high density of fibroblasts that likely produce the chemokines and cytokines that favor tumor-associated inflammation and support angiogenesis, resulting in a poor prognosis. In contrast, the canonical (CMS2) and metabolic (CMS3) subtypes with intermediate prognosis exhibit low immune and inflammatory signatures.

Conclusions: The distinct immune orientations of the colorectal cancer molecular subtypes pave the way for tailored immunotherapies. Clin Cancer Res; 22(16); 4057–66. ©2016 AACR.

Translational Relevance

Targeted therapies have highly improved the treatment of colorectal cancer, such as cetuximab in patients with KRAS wild-type tumors. Recently, immunotherapy using an anticheckpoint antibody, anti–PD-1, has shown strikingly positive effects in patients with microsatellite instable (MSI) tumors. To guide future targeted immunotherapies, it is therefore essential to integrate molecular and immune classifications of colorectal cancer. We analyze herein the immune, inflammatory, angiogenic, and fibroblastic landscape of molecularly defined colorectal cancer subtypes and identify, in addition to the MSI-like subgroup with a Th1/cytotoxic orientation, tumors with high lymphocyte and stromal infiltration, suggesting that the corresponding patients could be treated by a combination of antiangiogenic, anti-inflammatory, and anticheckpoint agents. In addition, we show that canonical and metabolic colorectal cancer subtypes are immune down, suggesting the use of adoptive T-cell therapies. Our data therefore provide immune and molecular subtype integration with prognostic impact and pave the way for personalized colorectal cancer immunotherapy.

Cancers are generally classified according to their localization, histology, and the genomic alterations of the malignant cells, such as chromosomal rearrangements or DNA mutations. Prognosis has been based essentially on tumor extension categorized by the TNM staging method which incorporates local tumor spread (T) and distant lymph node (N) and organ (M) metastases (1). Cancer therapies are proposed based on these classifications including conventional chemotherapies in advanced cancers and personalized therapies targeting products of mutated genes or rearranged genes. Mutational analyses also unveiled unexpected findings, such as the resistance of patients with colorectal cancer exhibiting KRAS mutation to treatment with cetuximab, an anti-EGF receptor antibody (2). These classifications have been complemented by high-throughput transcriptome analyses that identified dominant oncogenic pathways and established prognostic subtypes, as in diffuse large B-cell lymphoma (3), breast cancer (4, 5) or clear-cell renal cell carcinoma (6).

In the last decade, immune classification of cancers has shed new light in patients' care providing prognostic (7) and predictive factors for chemotherapies (8) and immunotherapies, such as immune checkpoint inhibitors (9). Colorectal cancer has been a paradigmatic tumor for immune classifications. Our laboratory has demonstrated that patients whose tumors are highly infiltrated by memory T cells, particularly cytotoxic CD8+ T lymphocytes, had a longer progression-free survival (PFS) and overall survival (OS; refs. 10–14). We have hypothesized that tumor-associated antigens could locally induce antitumor adaptive immune responses and have characterized tertiary lymphoid structures (TLS), adjacent to the tumor nests, that could be sites where antitumor immunity is generated (15). Indeed, we found that high T- and B-cell infiltration and a high expression of genes coding for lymphocyte-attracting chemokines, i.e., CX3CL1, CXCL9, and CXCL10 for T cells (13) and CXCL13 for B cells (14), as well as genes involved in a Th1 orientation (IFNG and TBX21) and cytotoxicity (GZMB and GNLY; ref. 10), are associated with favorable prognosis (10, 12, 14). MSI tumors, with their high mutational load and high leukocyte infiltration, fall perfectly in this category. It has recently been reported that metastatic colorectal cancer tumors with this phenotype responded to treatments with PD-1 immune checkpoint–blocking antibodies that increase the local immune reaction, potentially against tumor-associated antigens (16, 17).

A very recent publication proposed a transcriptomic classification of colorectal cancer into four consensus molecular subtypes (CMS; ref. 18). CMS1, called MSI-like, contains most microsatellite instable (MSI) tumors with mutations in genes encoding DNA mismatch-repair proteins, resulting in high mutational burden. The MSI-like subtype is also enriched for tumors with a CpG-island methylator phenotype (CIMP) and mutations in the BRAF oncogene. CMS2, called canonical, is a subtype with high chromosomal instability (CIN) as well as activation of the Wnt and MYC pathways. CMS3, called metabolic, is enriched in tumors with KRAS mutations and shows a disruption of metabolic pathways. Finally, CMS4, called mesenchymal, has a mesenchymal phenotype and frequent CIMP phenotype. This classification stratifies colorectal cancer into intrinsic subtypes with different prognosis (18). It has been independently well established that the composition of the microenvironment in which the malignant cells grow and expand is essential for predicting patient's prognosis (10, 7) and can be a target for cancer therapies (19).

In the era of targeted therapies, particularly immunotherapies that are dependent on the composition of the tumor microenvironment, we undertook to integrate molecular and immune classifications of colorectal cancer, by addressing the question of the immune, inflammatory, angiogenic, and fibroblastic composition of colorectal cancer molecular subtypes. We thus quantified these components in a discovery cohort of 458 CMS-classified colorectal cancer tumors (CIT cohort). It was validated in two independent validation cohorts of 404 (CIT validation cohort) and 526 (PETACC3 cohort) tumors. For this purpose, we applied the MCP-counter algorithm, a computational method able to infer the abundance of nine immune and two other stromal cell populations from a transcriptomic sample. Using this method, we quantified immune and stromal infiltration of the four CMS subtypes of colorectal cancer and found a significant correlation with CMS subtypes, validated the predicted infiltration profiles using immunohistochemistry, and discussed immunotherapeutic approaches that could benefit each subtype.

Public transcriptomic datasets

The complete lists of selected gene expression profiles (GEP), related type, and experimental conditions are given in Supplementary Tables S1, S2, and S3.

Colorectal tumors samples and subtypes annotations

The GEP from 1,750 colorectal tumor samples were collected. The GSE39582 dataset (fresh-frozen samples; Affymetrix HG-U133Plus2.0; n = 566) was used as a discovery cohort (herein termed CIT discovery). Samples from series GSE13067 (n = 74), GSE13294 (n = 155), GSE17536 (n = 177), and GSE33113 (n = 90) were aggregated as a validation meta-series (herein termed CIT validation; fresh-frozen samples; Affymetrix HG-U133Plus2.0; n = 496). Samples from the PETACC3 (ArrayExpress:E-MTAB-990) series (n = 688, formalin-fixed, paraffin-embedded samples, custom Affymetrix microarrays) were used to validate the nondependency of the results on microarray technology and sample processing. The CMS subtype annotation of all tumors analyzed was provided by the Colorectal Cancer Subtyping Consortium (CRCSC). CMS-unclassified samples reduced the numbers of samples analyzed to 458 for the CIT discovery cohort (81% classified), 404 for the CIT validation cohort (81% classified), and 526 for the PETACC3 cohort (76% classified). The total number of colorectal cancer tumors analyzed was therefore 1,388.

Multi-cancers dataset

The GEP of breast (n = 254), colorectal (n = 173), kidney (n = 144), ovarian (n = 97), lung (n = 80), and endometrial (n = 69) were retrieved from expO dataset (GEO:GSE2109).

Microenvironment-purified cells

We screened the GEO database (20) for GEP of purified samples of human immune cells, fibroblasts, and endothelial cells hybridized on Affymetrix HG-U133Plus2.0 microarrays. We collected 1,194 GEP from 80 series, including 1,114 immune, 36 endothelial, and 50 fibroblast samples.

Colorectal tumor cell lines

The Affymetrix HG-U133Plus2.0 GEP from the 55 colorectal tumor cell lines from the Cancer Cell Lines Encyclopedia (21) series were selected as tumor controls.

Immunohistochemistry

Serial 5-μm formalin-fixed paraffin-embedded tissue sections from colorectal cancer were stained using autostainerPlus Link 48 (Dako). Antigen retrieval and deparaffinization were carried out on a PT-Link (Dako) using the EnVision FLEX Target Retrieval Solutions (Dako). The antibodies used are listed in Supplementary Table S4. Peroxidase activity was detected using diaminobenzidine substrate (Dako). Slides stained with anti-CD8 and anti-CD68 were digitalized with a NanoZoomer scanner (Hamamatsu) and digitally quantified with Calopix software (Tribvn). The degree of smooth muscle actin (SMA) expression in the tumor stroma was quantified according to the following grading system: (1) scarce fibroblasts; (2) continuous layer of fibroblast between tumor nests with overall thickness inferior to three cells; (3) continuous layer of fibroblast between tumor nests with overall thickness superior to three cells and fibroblast area <50% of tumor area; and (4) continuous layer of fibroblast between tumor nests with overall thickness superior to three cells and fibroblast area >50% of tumor area.

Microarrays analysis

GEP normalization.

The GEP from microenvironment-purified cells and pan-cancers cell lines were normalized independently for each series, using the frozen Robust Multiarray Average (RMA) method (21a) on each independent series (“fRMA” R package). The RMA-normalized GEP from the CIT colorectal cancer discovery series were downloaded directly from GEO. The GEP from PETACC3 colorectal cancer series were normalized in batch using the RMA method (‘affy’ R package). Each GEP series from the CIT colorectal cancer validation meta-series was normalized independently using frozen RMA method; then the corresponding matrices were combined into one matrix, further normalized with Combat method (22), using series' identifiers as batch variables and no covariates. The GEP from RNA mixture models were normalized using the RMA method. When mapping probesets to HUGO Gene Symbols, the mean across probesets was chosen to represent the gene's expression level.

Use of the MCP-counter algorithm.

MCP-counter is an algorithm that aims to estimate samples' infiltration by various immune and other stromal cell populations using transcriptomic data. Its output can be used to compare cellularly heterogeneous samples for their relative infiltration by eleven cell populations. MCP-counter relies on the identification of so-called “Transcriptomic markers,” which are transcripts specifically expressed by a given cell population and not by the others. These transcriptomic markers have been identified on a discovery series of 1,939 curated gene expression profiles representing cell populations present in the tumor microenvironment. The specificity of their expression for the corresponding cell population was validated on two independent series of 1,596 and 3,208 samples. MCP-counter scores summarize the expression of the transcriptomic markers specific for a given population, and have been validated for their ability to correlate with the fraction of mRNA originating from the corresponding cell population and to cell infiltration estimated by immunohistochemistry (Becht et al.; submitted for publication).

We applied the R (version 3.1.3) package “MCPcounter” version 0.1.0 on each normalized tumor GEP, using probesets as identifiers for the CIT, CIT validation and multi-cancer cohorts, and gene symbols for the PETACC3 cohort. The MCP-counter design and workflow are illustrated in Supplementary Fig. S1. MCPcounter version 0.1.0 is available at http://cit.ligue-cancer.net/?page_id=1243&lang=en.

Supervised tests of differential expression.

ANOVA tests were used to assess the dependency of genes or MCP-counter scores to the molecular subgroups classification. Student t tests were used to investigate differential expression of genes between subgroups or cell line phenotypes. To test for differential level of the MCP-counter scores in a given molecular subgroup, Student t tests against the cohort's median MCP-counter score were used. To test for differential level of MCP-counter scores between molecular subgroups, pairwise one-tailed t tests with Bonferroni correction were used (Supplementary Table S5).

List of immune-related genes.

We curated a list of genes known to encode proteins with immunomodulatory functions (Supplementary Table S6). Representatives of the chemokines, chemokine receptors, interleukins, interleukins receptors, TNF and TNF receptors, growth factors, interferons and interferon receptors, inhibitory receptors and their ligands, TLR, and class I MHC gene families were included.

1-colorectal cancer molecular subgroups show distinct expression patterns of immune and stromal signatures

We used the MCP-counter algorithm to obtain estimates of tumor infiltration by immune and other stromal cell populations. A brief presentation of MCP-counter design and workflow is provided in Supplementary Fig. S1. In the CIT and CIT validation cohorts, this algorithm revealed that the molecular subgroups showed consistent and distinct patterns for the immune, endothelial, and fibroblastic cell populations' abundance estimates (Fig 1A). Tumors of the MSI-like and mesenchymal subtypes had a high expression of lymphoid (Fig 1A and B) as well as myeloid cell–specific genes (Fig 1A and C), thus exhibiting a strong immune and inflammatory contexture, whereas tumors of the canonical and metabolic subtypes had low expression of the lymphocytic and myeloid signatures. Tumors of the MSI-like and mesenchymal subtypes differed in that MSI-like samples exhibited a higher cytotoxic-cells abundance estimate, reflecting high infiltration by activated CD8+ and natural killer (NK) cells. Granulocyte-specific transcripts were poorly discriminative (Fig 1A and C). In addition, mesenchymal samples exhibited a high expression of the fibroblastic and endothelial cell abundance estimates, compatible with highly vascularized and inflammatory tumors that have a high density of cancer-associated fibroblasts (CAF) in their microenvironments (Fig 1A and D).

Figure 1.

Immune and stromal signatures of the four molecular subgroups of colorectal cancer. A, heatmap showing the level of the of the nine immune and two other stromal MCP-counter abundance estimates among two transcriptomic cohorts of colorectal cancer patients that were classified in four molecularly defined colorectal cancer subgroups. Distributions of the (B) lymphocytic, (C) myeloid, and (D) stromal abundance estimates across subgroups in the two cohorts. *, P < 0.05; **, P < 0.001; and ***, P < 0.0001 compared with the cohort's median using a Student t test.

Figure 1.

Immune and stromal signatures of the four molecular subgroups of colorectal cancer. A, heatmap showing the level of the of the nine immune and two other stromal MCP-counter abundance estimates among two transcriptomic cohorts of colorectal cancer patients that were classified in four molecularly defined colorectal cancer subgroups. Distributions of the (B) lymphocytic, (C) myeloid, and (D) stromal abundance estimates across subgroups in the two cohorts. *, P < 0.05; **, P < 0.001; and ***, P < 0.0001 compared with the cohort's median using a Student t test.

Close modal

The immune infiltrations in the four subtypes predicted by MCP-counter were confirmed using immunohistochemical analyses in a subset of 38 randomly selected tumors from the CIT discovery cohort. CD8+ T cells and CD68+ macrophages were quantified within the tumor center. These analyses showed a significant correlation between the density of CD8+ cells in the tumor and the cytotoxic abundance estimate from transcriptomic analyses (P = 2.10−5; r = 0.67) and between CD68+ macrophages and the monocytic-lineage abundance estimates (P = 1.10−5; r = 0.68). We confirmed that the MSI-like and the mesenchymal-like subgroups had higher densities of CD8 T cells and CD68 macrophages than the canonical and metabolic subtypes, validating the transcriptomic predictions (Fig 2A and B). In addition, we performed SMA immunohistochemical labeling. The mesenchymal subtype had the highest SMA grading, supporting the fact that the transcriptomic fibroblastic signature was reflecting a high presence of CAF (Fig 2C and D).

Figure 2.

Immunohistochemical characterization of the four colorectal cancer subgroups. A, distributions of the densities of tumor-infiltrating CD8+ T cells in the four subgroups. B, distributions of the densities of tumor-infiltrating CD68+ macrophages in the four subgroups. P values were assessed using the Kruskal–Wallis test. C, representative tumor areas of each SMA-staining grades. SMA-positive areas are labeled in brown. D, distributions of each SMA grades in the four subgroups. P values were assessed using the Fisher exact test.

Figure 2.

Immunohistochemical characterization of the four colorectal cancer subgroups. A, distributions of the densities of tumor-infiltrating CD8+ T cells in the four subgroups. B, distributions of the densities of tumor-infiltrating CD68+ macrophages in the four subgroups. P values were assessed using the Kruskal–Wallis test. C, representative tumor areas of each SMA-staining grades. SMA-positive areas are labeled in brown. D, distributions of each SMA grades in the four subgroups. P values were assessed using the Fisher exact test.

Close modal

Having analyzed patterns related to microenvironment cell populations, we focused on features related to immune cells function and migration. We thus analyzed the expression of genes encoding molecules involved in T-cell chemotaxis, activation and inhibition, inflammation and complement components, angiogenesis as well as MHC1 molecules (Fig 3; Supplementary Table S6). The four consensus molecular subgroups again showed strikingly reproducible data across the two independent cohorts. The MSI-like subtype exhibited a high expression of genes coding for T-cell–attracting chemokines [CXCL9 (13), CXCL10 (13), and CXCL16] or involved in the formation of tumor-adjacent tertiary lymphoïd structures (CXCL13; refs. 23, 24), as well as the Th1 cytokines IFNG and IL15, all of which have been shown to correlate with good prognosis in colorectal cancer (10, 12, 13, 14). In contrast, the mesenchymal subtype exhibited a high expression of the myeloid chemokine CCL2, complement components (C1QA, C1QB, C1QC, C1R, C1S, C3, C3AR1, C5AR1, C7, CFD, CFH, and CFI), angiogenic factors (VEGFB, VEGFC, and PDGFC), and immunosuppressive molecules [TGFB1, TGFB3, LGALS1 (25), and CXCL12]. CD274 and PDCD1LG2, the genes encoding the PD-1 ligands, were highly expressed in MSI-like tumors but also in some tumors of the mesenchymal group. Strikingly, MHC1 genes, whose products present peptides to CD8+ T cells, were poorly expressed in the poorly infiltrated canonical subtype.

Figure 3.

Expression of functionally relevant immune genes among the four subgroups in the two cohorts. The heatmaps on the left represent the level of expression of the genes. Rows were centered and scaled. Red denotes a higher expression, and blue a lower expression. The heatmaps on the right represent the P value of a Student t test against the cohort median, for each gene.

Figure 3.

Expression of functionally relevant immune genes among the four subgroups in the two cohorts. The heatmaps on the left represent the level of expression of the genes. Rows were centered and scaled. Red denotes a higher expression, and blue a lower expression. The heatmaps on the right represent the P value of a Student t test against the cohort median, for each gene.

Close modal

We were able to reproduce these results on an independent cohort, called “PETACC3,” of 526 colorectal cancer samples, whose RNA was extracted from paraffin-embedded tissues and hybridized on another microarray platform, indicating strong reproducibility (Fig. 4).

Figure 4.

Results are reproducible on the independent PETACC3 colorectal cancer cohort (526 CMS-classified samples). A, heatmap showing the level of the abundance estimates of the immune and stromal signatures in the PETACC3 colorectal cancer transcriptomic cohort that was classified according to the four molecularly defined colorectal cancer subgroups. Distributions of the (B) lymphocytic, (C) myeloid, and (D) stromal abundance estimates across subgroups in the two cohorts. *, P < 0.05; **, P < 0.001; and ***, P < 0.0001 compared with the cohort's median using a Student t test. E, the heatmap on the left represents the level of expression of the genes. Rows were centered and scaled. Red denotes a higher expression, and blue a lower expression. The heatmap on the right represents the P value of a Student t test against the cohort median, for each gene.

Figure 4.

Results are reproducible on the independent PETACC3 colorectal cancer cohort (526 CMS-classified samples). A, heatmap showing the level of the abundance estimates of the immune and stromal signatures in the PETACC3 colorectal cancer transcriptomic cohort that was classified according to the four molecularly defined colorectal cancer subgroups. Distributions of the (B) lymphocytic, (C) myeloid, and (D) stromal abundance estimates across subgroups in the two cohorts. *, P < 0.05; **, P < 0.001; and ***, P < 0.0001 compared with the cohort's median using a Student t test. E, the heatmap on the left represents the level of expression of the genes. Rows were centered and scaled. Red denotes a higher expression, and blue a lower expression. The heatmap on the right represents the P value of a Student t test against the cohort median, for each gene.

Close modal

2-mesenchymal cells induce an inflammatory and angiogenic tumor microenvironment

The poor-prognostic CMS4 colorectal cancer subgroup is characterized by a high fibroblastic MCP-counter score, an estimate of cellular abundance, as well as a high expression of the myeloid and endothelial cells scores. We found that the fibroblastic score highly correlated with the endothelial (P < 10−15 on the three cohorts, Pearson r = 0.84, 0.84, 0.82 for CIT, CIT validation, and PETACC3, respectively) and myeloid ones (P < 10−15 on the three cohorts, Pearson r = 0.6, 0.46, 0.46 for CIT, CIT validation, and PETACC3, respectively; Fig 5A). In contrast, there was no correlation between the fibroblastic and cytotoxic cells abundance estimates (Fig. 5A). Correlations between the fibroblastic score and both the endothelial and myeloid cell scores were also observed in breast, lung, and ovary cancers, and confirmed in colorectal cancer (Fig 5B), suggesting that the immune contexture found in mesenchymal colorectal cancer tumors also exists in these cancers. In kidney cancer, the correlation between the fibroblast and the myeloid scores was weaker, and it was absent in endometrium cancer (Fig 5B).

Figure 5.

The fibroblast abundance estimate correlates with endothelial and myeloid cells abundance estimates in colorectal cancer and other cancers. Scatterplots representing the relationships between the endothelial cells, myeloid cells, and cytotoxic cells MCP-counter scores (cellular abundance estimates) compared with the fibroblast MCP-counter score (A) in the two colorectal cancer cohorts (B) across six cancers, including colorectal cancer, in the expO dataset.

Figure 5.

The fibroblast abundance estimate correlates with endothelial and myeloid cells abundance estimates in colorectal cancer and other cancers. Scatterplots representing the relationships between the endothelial cells, myeloid cells, and cytotoxic cells MCP-counter scores (cellular abundance estimates) compared with the fibroblast MCP-counter score (A) in the two colorectal cancer cohorts (B) across six cancers, including colorectal cancer, in the expO dataset.

Close modal

The coordination of these cell population abundance estimates across human tumors led us to hypothesize that fibroblasts promote angiogenesis and inflammatory cells recruitment in the mesenchymal colorectal cancer tumors' microenvironment. Because tumor samples correspond to a mixture of tumor cells and microenvironment cells, transcriptomic samples of pure cell populations were used to investigate the cellular origin of the inflammatory and angiogenic signatures of the mesenchymal molecular subgroup (Supplementary Tables S1 and S3). We first identified the genes upregulated in the mesenchymal subtype compared with each of the other subtypes (Student t tests against each of the other three subtypes, all P < 0.05; Supplementary Table S6). We then investigated the expression of these genes by immune, stromal, and malignant cells (Fig. 6). B, T, and NK lymphocytes, as well as colorectal cancer cell lines, each overexpressed only a small subset of these genes. Fibroblasts had the highest expression for the proangiogenic factors VEGFB, VEGFC, and PDGFC, the immunosuppressive factors LGALS1, CXCL12, PTGS1, and TGFB3, and the complement components C1S, C1R, CFH, C7, and CFHR2, and can thus promote angiogenesis and immunosuppression. Endothelial cells had the highest expression of the myeloid chemoattractant CCL2, the angiogenic factor PDGFB, and immunosuppressive molecules TGFB1 and TGFB2. Finally, monocytic cells expressed complement components (C1QA, C1QC, C3, C3AR1, and C5AR1) and chemokines attracting macrophages (CCL19 and CCL23). These results suggest that these three cell populations could foster inflammation, angiogenesis, and immunosuppression in mesenchymal colorectal cancer tumors.

Figure 6.

Inflammatory, angiogenic, and suppressive molecules overexpressed in mesenchymal tumors are highly expressed by fibroblastic, endothelial, and monocytic cells. Expression of the genes specifically upregulated in mesenchymal tumors and related to inflammation, angiogenesis, immunosuppression, and immune cell functional orientations, in homogeneous samples of immune, stromal, or colorectal cancer cell lines (Supplementary Table S1). Black frames indicate that the corresponding cell population has the highest expression of the gene.

Figure 6.

Inflammatory, angiogenic, and suppressive molecules overexpressed in mesenchymal tumors are highly expressed by fibroblastic, endothelial, and monocytic cells. Expression of the genes specifically upregulated in mesenchymal tumors and related to inflammation, angiogenesis, immunosuppression, and immune cell functional orientations, in homogeneous samples of immune, stromal, or colorectal cancer cell lines (Supplementary Table S1). Black frames indicate that the corresponding cell population has the highest expression of the gene.

Close modal

In the last decade, the interplay between tumors and the immune system has emerged as a critical aspect of tumor biology and is strongly associated with the host ability to control tumor growth and to respond to therapies. Incorporating precise immune-related information in descriptive cancer-classification studies or in prospective clinical trials is therefore critical. In the present work, we apply the MCP-counter method to study the heterogeneity of the immune, inflammatory, angiogenic, and fibroblastic tumor microenvironment of the consensus molecular classification of colorectal cancer.

The expression of the immune cell abundance estimates, enriched by the analysis of a large array of functionally relevant genes, in three colorectal cancer cohorts stratified using a recently published molecular classification revealed a strong association between the tumor cell phenotype and both the composition and the functional orientation of its immune microenvironment. Notably, we demonstrate that mesenchymal tumors are associated with a proinflammatory, proangiogenic, and immunosuppressive microenvironment.

In the three cohorts, two subgroups were characterized by high expression of immune signatures: the expected MSI-rich CMS1 group and the unexpected mesenchymal CMS4 group. Strikingly, although the MSI-like group correlated with favorable patient's prognostic in terms of RFS (18), the mesenchymal subgroup of patients had the worst prognosis of the four subgroups (18). We describe for the first time a group of colorectal cancer tumors with high lymphoid gene expression associated with poor prognosis for the patients. This subgroup is characterized by an extensive tumor infiltration by CAF (Fig. 1A and Fig. D and Fig. 2C and D), correlating with high angiogenesis and myeloid cells infiltration (Fig. 1A and D, and Fig. 5A). Recent studies reported an extensive infiltration by stromal fibroblasts in mesenchymal colorectal cancer tumors (26, 27). These studies additionally showed that the presence of stromal fibroblasts and TGFß signaling enhanced tumors' metastatic capacities and abilities to form xenografts in mice. Our results suggest that in addition to this prometastatic effect, CAF promote inflammation and angiogenesis in mesenchymal colorectal cancer tumors. We hypothesize that this strong inflammatory component hampers the positive value of the Th1/CD8+ T cells in these tumors, by repressing the antitumor activity of cytotoxic T cells while fueling tumor growth, angiogenesis, and stroma remodeling. These findings are reminiscent of our recent observations in clear-cell renal cell carcinoma, a paradigmatic tumor for high angiogenesis (28), in which we identified a poor-prognosis subgroup with high CD8+ T cells infiltration with high inflammatory and immunosuppressive contexture (6, 29).

The microenvironment of mesenchymal colorectal cancer tumors, characterized by high fibroblastic, endothelial, and myeloid densities, extends to other cancers than colorectal cancer (Fig. 5A). It is thus tempting to postulate that similar immune, inflammatory, and immunosuppressive microenvironments might also be found in these tumors, indicating that similar therapies aimed at modifying the tumor microenvironment could be applied to cohorts of cancers of different origins and locations exhibiting a mesenchymal phenotype. In particular, antiangiogenic treatments and/or inhibitors of LGALS1-encoded protein (31) should be tested in mesenchymal colorectal cancer and the mesenchymal-like tumors. The mesenchymal subgroup also exhibits an angiogenic and inflammatory signature which is probably the consequence of their high fibroblastic infiltration. Angiogenesis and inflammation are intertwined pathways, which both fuel tumor growth through the production of survival and proliferative signals and by favoring blood supply (31). Yet, because the mesenchymal subtype is highly infiltrated by CD8+ T cells, one could expect it to be associated with favorable outcome (7). However, an extensive number of studies have shown that inflammatory and angiogenic microenvironments were associated with the inhibition of antitumor cytotoxic T-cell immune responses, notably through the inhibition of the maturation of dendritic cells (31). Immature dendritic cells deliver inhibitory secondary signals to T cells upon antigen presentation, inhibiting their activation. MSI-like is the other “immune-high subgroup” of colorectal cancer. This group contains most patients harboring MSI tumors, and is known to be associated with a good prognosis, and to feature a strong CD8+ T-cell infiltration. Strikingly, MSI-like is the group featuring the highest expression of class I MHC genes (Fig. 3; Supplementary Table S6), as well as genes specific for cytotoxic lymphocytes (Fig. 1A; Supplementary Table S5) or attracting memory T cells (CXCL9 or CXCL10), activating T cells (IFNG), supporting proliferation of T and NK cells (IL15) and helping in the formation of TLS (CXCL13), where antitumor adaptive immune responses are likely shaped (ref. 32; Fig. 3; Supplementary Table S6). High expression of these genes has been reported to be associated with good prognosis in colorectal cancer (7, 10, 13, 14). The cellular composition and functional orientation of the MSI-like subgroup thus suggest that it corresponds to tumors with a high Immunoscore (33). CXCL13 and IL15 have been shown to be produced by the tumor cells (12, 14), whereas IFNG is clearly produced by the infiltrating cells. MSI-like is also characterized by a lower expression of the myeloid and endothelial cells signatures (Fig. 1A, C, and D) as well as angiogenesis-inducing genes (Fig. 3; Supplementary Table S6). It is therefore likely that the MSI-like subgroup contains highly immunogenic tumors, in the context of mild inflammation and angiogenesis, which results in the generation of antitumor adaptive immune responses educated in tumor-adjacent TLS (34). Effector memory CD8 T+ cells (34) and B cells (35) would then control the growth and metastasis in this subgroup (36), as exemplified in non-small cell lung cancer (NSCLC) (37). IFNG produced by infiltrating T cells is known to induce a phenomenon called “adaptive resistance” by increasing the expression of the inhibitory checkpoint molecule PD-1 on T cells (38) and of its ligands CD274 (PD-L1; ref. 38) and PDCD1LG2 (PD-L2; ref. 39) on the tumor cells, which may result in inefficient antitumor T-cell reaction (40). It is striking that MSI-like tumors also show the highest expression of PD-L1 and PD-L2 genes, followed by mesenchymal tumors (Fig. 3; Supplementary Table S6). These results prompt to treat colorectal cancer MSI-like patients with agents blocking the PD-1/PD-L1 pathway, such as anti–PD-1 and anti–PD-L1. Recent evidence using in-situ immunohistochemical staining of immune checkpoints molecules supports the use of anticheckpoint immunotherapies in patients with MSI tumors (16), and a recent clinical trial showed that patients with MSI tumors responded to PD-1 blockade (17). Because MSI-like subtype is highly enriched for patients with MSI tumors but also includes a group of microsatellite stable (MSS) patients (41), the use of molecular classifications might help identify responders to PD-1 blockade therapies among MSS patients, and nonresponders among patients with MSI tumors, and we therefore advocate to investigate the molecular subgroups of anti–PD-1 and anti–PD-L1-treated colorectal cancer patients.

Tumors of the canonical and metabolic subgroups were characterized by poor infiltration by immune cells and low class I MHC expression (Fig. 3; Supplementary Table S6), and are thus most likely poorly immunogenic, which may explain the low tumor T-cell infiltration in these subgroups. The use of bispecific antibodies targeting a tumor-associated antigen (42), such as carcinoembryonic antigen (43), could enhance these tumors' immunogenicity.

Because the transcriptomic classification of colorectal cancer is strongly associated with different immune and stromal contextures, the present work paves the way of novel classifications of colorectal cancer tumors, based on the relationships between the phenotype of the cancer cell and the corresponding immune and stromal profile of its microenvironment, potentially identifying the most appropriate treatments, including antiangiogenic drugs and immunotherapies.

Cell lines

The cell lines correspond to transcriptomes of cell lines downloaded from GEO (dataset GSE36133), and were not cultivated in the laboratory.

P. Laurent-Puig is a consultant/advisory board member for Amagen, Astra-Zeneca, Boehringer-Ingelheim, INTEGRAGEN, Merck-Serono, and Sanofi. W.H. Fridman is a consultant/advisory board member for Curetech, Efranat, Pierre Fabre Medicament, Sandoz, Sanofi, and Servier. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Becht, A. de Reyniès, N.A. Giraldo, P. Laurent-Puig, W.H. Fridman

Development of methodology: E. Becht, A. de Reyniès, N.A. Giraldo, L. Lacroix

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Becht, N.A. Giraldo, B. Buttard, J. Selves, C. Sautès-Fridman, P. Laurent-Puig

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Becht, A. de Reyniès

Writing, review, and/or revision of the manuscript: E. Becht, A. de Reyniès, N.A. Giraldo, C. Pilati, J. Selves, C. Sautès-Fridman, W.H. Fridman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Lacroix, W.H. Fridman

Study supervision: A. de Reyniès, W.H. Fridman

The authors acknowledge members of the Centre d'Imagerie Cellulaire et de Cytométrie “CICC” plateform of the Cordeliers Research Center and the “Plateforme Biopuces et Séquençage” of the IGBMC for their respective technical expertise. The efforts of Gene Expression Omnibus, arrayExpress, the expression project for Oncology and the International Genomics Consortium, and all the teams that shared their GEP results are greatly acknowledged. They also thank Ivo Natario for his help with data collection and Lubka Roumenina, Gabriela Bindea, Jerome Galon, Bernhard Mlecnik, Estelle Devevre, Audrey Lupo, and Marie-Caroline Dieu-Nosjean for their fruitful discussions.

This work was supported by the “Institut National de la Santé et de la Recherche Médicale,” the University Paris-Descartes, the University Pierre et Marie Curie, the Institut National du Cancer (2011-1-PLBIO-06-INSERM 6-1), CARPEM (CAncer Research for PErsonalized Medicine), Labex Immuno-Oncology (LAXE62_9UMS872 FRIDMAN), the Fondation ARC pour la recherche sur le cancer, the Cancéropôle Ile-de-France, Institut National du Cancer (2011-1-PLBIO-06-INSERM 6-1,PLBIO09-088-IDF-KROEMER), the Universidad de los Andes School of Medicine (N.A. Giraldo), and Colciencias (N.A. Giraldo). E. Becht is supported by B3MI doctorate fellowship, and N.A. Giraldo by PPATH doctorate fellowship.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Sobin
L
,
Gospodarowicz
M
,
Wittekind
C
.
Tnm classification of malignant tumors, 7th ed.
Oxford
:
Wiley-Blackwell
; 
2009
.
2.
Lièvre
A
,
Bachet
J
,
Le Corre
D
,
Boige
V
,
Landi
B
,
Emile
J
, et al
Kras mutation status is predictive of response to cetuximab therapy in colorectal cancer
.
Cancer Res
2006
;
66
:
3992
5
.
3.
Alizadeh
AA
,
Eisen
MB
,
Davis
RE
,
Ma
C
,
Lossos
IS
,
Rosenwald
A
, et al
Distinct types of diffuse large b-cell lymphoma identified by gene expression profiling
.
Nature
2000
;
403
:
503
11
.
4.
Perou
CM
,
Sørlie
T
,
Eisen
MB
,
van de Rijn
M
,
Jeffrey
SS
,
Rees
CA
, et al
Molecular portraits of human breast tumours
.
Nature
2000
;
406
:
747
52
.
5.
Sørlie
T
,
Perou
CM
,
Tibshirani
R
,
Aas
T
,
Geisler
S
,
Johnsen
H
, et al
Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications
.
Proc Natl Acad Sci USA
2001
;
98
:
10869
74
.
6.
Beuselinck
B
,
Job
S
,
Becht
E
,
Karadimou
A
,
Verkarre
V
,
Couchy
G
, et al
Molecular subtypes of clear cell renal cell carcinoma are predictive of sunitinib response in the metastatic setting
.
Clin Cancer Res
2015
;
21
:
1329
39
.
7.
Fridman
WH
,
Pagès
F
,
Sautès-Fridman
C
,
Galon
J
. 
The immune contexture in human tumours: Impact on clinical outcome
.
Nat Rev Cancer
2012
;
12
:
298
306
.
8.
Andre
F
,
Pusztai
L
. 
Molecular classification of breast cancer: Implications for selection of adjuvant chemotherapy
.
Nat Clin Pract Oncol
2006
;
3
:
621
32
.
9.
Taube
JM
,
Klein
A
,
Brahmer
JR
,
Xu
H
,
Pan
X
,
Kim
JH
, et al
Association of pd-1, pd-1 ligands, and other features of the tumor immune microenvironment with response to anti-pd-1 therapy
.
Clin Cancer Res.
2014
;
20
:
5064
74
.
10.
Galon
J
,
Costes
A
,
Sanchez-Cabo
F
,
Kirilovsky
A
,
Mlecnik
B
,
Lagorce-Pagès
C
, et al
Type, density, and location of immune cells within human colorectal tumors predict clinical outcome
.
Science
2006
;
313
:
1960
4
.
11.
Pagès
F
,
Berger
A
,
Camus
M
,
Sanchez-Cabo
F
,
Costes
A
,
Molidor
R
, et al
Effector memory T cells, early metastasis, and survival in colorectal cancer
.
N Engl J Med
2005
;
353
:
2654
66
.
12.
Mlecnik
B
,
Bindea
G
,
Angell
HK
,
Sasso
MS
,
Obenauf
AC
,
Fredriksen
T
, et al
Functional network pipeline reveals genetic determinants associated with in situ lymphocyte proliferation and survival of cancer patients
.
Sci Transl Med
2014
;
6
:
228ra37
.
13.
Mlecnik
B
,
Tosolini
M
,
Charoentong
P
,
Kirilovsky
A
,
Bindea
G
,
Berger
A
, et al
Biomolecular network reconstruction identifies T-cell homing factors associated with survival in colorectal cancer
.
Gastroenterology
2010
;
138
:
1429
40
.
14.
Bindea
G
,
Mlecnik
B
,
Tosolini
M
,
Kirilovsky
A
,
Waldner
M
,
Obenauf
AC
, et al
Spatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer
.
Immunity
2013
;
39
:
782
95
.
15.
Dieu-Nosjean
M
,
Goc
J
,
Giraldo
NA
,
Sautès-Fridman
C
,
Fridman
WH
. 
Tertiary lymphoid structures in cancer and beyond
.
Trends Immunol
2014
;
35
:
571
80
.
16.
Llosa
NJ
,
Cruise
M
,
Tam
A
,
Wicks
EC
,
Hechenbleikner
EM
,
Taube
JM
, et al
The vigorous immune microenvironment of microsatellite instable colon cancer is balanced by multiple counter-inhibitory checkpoints
.
Cancer Discov
2015
;
5
:
43
51
.
17.
Le
DT
,
Uram
JN
,
Wang
H
,
Bartlett
BR
,
Kemberling
H
,
Eyring
AD
, et al
Pd-1 blockade in tumors with mismatch-repair deficiency
.
N Engl J Med
2015
;
372
:
2509
20
.
18.
Guinney
J
,
Dienstmann
R
,
Wang
X
,
de Reyniès
A
,
Schlicker
A
,
Soneson
C
, et al
The consensus molecular subtypes of colorectal cancer
.
Nat Med
2015
;
21
:
1350
1356
.
19.
Chen
DS
,
Mellman
I
. 
Oncology meets immunology: The cancer-immunity cycle
.
Immunity
2013
;
39
:
1
10
.
20.
Edgar
R
,
Domrachev
M
,
Lash
AE
. 
Gene expression omnibus: NCBI gene expression and hybridization array data repository
.
Nucleic Acids Res
2002
;
30
:
207
10
.
21.
Barretina
J
,
Caponigro
G
,
Stransky
N
,
Venkatesan
K
,
Margolin
AA
,
Kim
S
, et al
The cancer cell line encyclopedia enables predictive modelling of anticancer drug sensitivity
.
Nature
2012
;
483
:
603
7
.
21a.
McCall
MN
,
Bolstad
BM
,
Irizarry
RA
. 
Frozen robust multiarray analysis (fRMA)
.
Biostatistics
2010
;
11(2)
:
242
253
.
22.
Johnson
WE
,
Li
C
,
Rabinovic
A
. 
Adjusting batch effects in microarray expression data using empirical Bayes methods
.
Biostatistics
2007
;
8
:
118
27
.
23.
de Chaisemartin
L
,
Goc
J
,
Damotte
D
,
Validire
P
,
Magdeleinat
P
,
Alifano
M
, et al
Characterization of chemokines and adhesion molecules associated with T cell presence in tertiary lymphoid structures in human lung cancer
.
Cancer Res
2011
;
71
:
6391
9
.
24.
Gu-Trantien
C
,
Loi
S
,
Garaud
S
,
Equeter
C
,
Libin
M
,
de Wind
A
, et al
Cd4+ follicular helper T cell infiltration predicts breast cancer survival
.
J Clin Invest
2013
;
123
:
2873
92
.
25.
Soldati
R
,
Berger
E
,
Zenclussen
AC
,
Jorch
G
,
Lode
HN
,
Salatino
M
, et al
Neuroblastoma triggers an immunoevasive program involving galectin-1-dependent modulation of T cell and dendritic cell compartments
.
Int J Cancer
2012
;
131
:
1131
41
.
26.
Isella
C
,
Terrasi
A
,
Bellomo
SE
,
Petti
C
,
Galatola
G
,
Muratore
A
, et al
Stromal contribution to the colorectal cancer transcriptome
.
Nat Genet
2015
;
47
:
312
9
.
27.
Calon
A
,
Lonardo
E
,
Berenguer-Llergo
A
,
Espinet
E
,
Hernando-Momblona
X
,
Iglesias
M
, et al
Stromal gene expression defines poor-prognosis subtypes in colorectal cancer
.
Nat Genet
2015
;
47
:
320
9
.
28.
Cairns
P
. 
Renal cell carcinoma
.
Cancer Biomark
2010
;
9
:
461
73
.
29.
Giraldo
NA
,
Becht
E
,
Pagès
F
,
Skliris
G
,
Verkarre
V
,
Vano
Y
, et al
Orchestration and prognostic significance of immune checkpoints in the microenvironment of primary and metastatic renal cell cancer
.
Clin Cancer Res
2015
;
21
:
3031
40
.
30.
Astorgues-Xerri
L
,
Riveiro
ME
,
Tijeras-Raballand
A
,
Serova
M
,
Neuzillet
C
,
Albert
S
, et al
Unraveling galectin-1 as a novel therapeutic target for cancer
.
Cancer Treat Rev
2014
;
40
:
307
19
.
31.
Mantovani
A
,
Allavena
P
,
Sica
A
,
Balkwill
F
. 
Cancer-related inflammation
.
Nature
2008
;
454
:
436
44
.
32.
Dieu-Nosjean
M
,
Goc
J
,
Giraldo
NA
,
Sautès-Fridman
C
,
Fridman
WH
. 
Tertiary lymphoid structures in cancer and beyond
.
Trends Immunol
2014
;
35
:
571
80
.
33.
Galon
J
,
Mlecnik
B
,
Bindea
G
,
Angell
HK
,
Berger
A
,
Lagorce
C
, et al
Towards the introduction of the ‘immunoscore’ in the classification of malignant tumours
.
J Pathol
2014
;
232
:
199
209
.
34.
Camus
M
,
Tosolini
M
,
Mlecnik
B
,
Pagès
F
,
Kirilovsky
A
,
Berger
A
, et al
Coordination of intratumoral immune reaction and human colorectal cancer recurrence
.
Cancer Res
2009
;
69
:
2685
93
.
35.
Germain
C
,
Gnjatic
S
,
Tamzalit
F
,
Knockaert
S
,
Remark
R
,
Goc
J
, et al
Presence of b cells in tertiary lymphoid structures is associated with a protective immunity in patients with lung cancer
.
Am J Respir Crit Care Med
2014
;
189
:
832
44
.
36.
Becht
E
,
Goc
J
,
Germain
C
,
Giraldo
NA
,
Dieu-Nosjean
M
,
Sautès-Fridman
C
, et al
Shaping of an effective immune microenvironment to and by cancer cells
. Cancer Immunol Immunother
2014
;
63
:
991
7
.
37.
Goc
J
,
Germain
C
,
Vo-Bourgais
TKD
,
Lupo
A
,
Klein
C
,
Knockaert
S
, et al
Dendritic cells in tumor-associated tertiary lymphoid structures signal a th1 cytotoxic immune contexture and license the positive prognostic value of infiltrating cd8+ t cells
.
Cancer Res
2014
;
74
:
705
15
.
38.
Taube
JM
,
Anders
RA
,
Young
GD
,
Xu
H
,
Sharma
R
,
McMiller
TL
, et al
Colocalization of inflammatory response with b7-h1 expression in human melanocytic lesions supports an adaptive resistance mechanism of immune escape
.
Sci Transl Med
2012
;
4
:
127ra37
.
39.
Latchman
Y
,
Wood
CR
,
Chernova
T
,
Chaudhary
D
,
Borde
M
,
Chernova
I
, et al
Pd-l2 is a second ligand for pd-1 and inhibits t cell activation
.
Nat Immunol
2001
;
2
:
261
8
.
40.
Francisco
LM
,
Sage
PT
,
Sharpe
AH
. 
The PD-1 pathway in tolerance and autoimmunity
.
Immunol Rev
2010
;
236
:
219
42
.
41.
Marisa
L
,
de Reyniès
A
,
Duval
A
,
Selves
J
,
Gaub
MP
,
Vescovo
L
, et al
Gene expression classification of colon cancer into molecular subtypes: characterization, validation, and prognostic value
.
PLoS Med
2013
;
10
:
e1001453
.
42.
Bargou
R
,
Leo
E
,
Zugmaier
G
,
Klinger
M
,
Goebeler
M
,
Knop
S
, et al
Tumor regression in cancer patients by very low doses of a t cell-engaging antibody
.
Science
2008
;
321
:
974
7
.
43.
Oberst
MD
,
Fuhrmann
S
,
Mulgrew
K
,
Amann
M
,
Cheng
L
,
Lutterbuese
P
, et al
Cea/cd3 bispecific antibody medi-565/amg 211 activation of t cells and subsequent killing of human tumors is independent of mutations commonly found in colorectal adenocarcinomas
.
MAbs
2014
;
6
:
1571
84
.

Supplementary data