A Diagnostic Biopsy-Adapted Immunoscore Predicts Response to Neoadjuvant Treatment and Selects Patients with Rectal Cancer Eligible for a Watch-and-Wait Strategy

Colorectal cancer is the third most common cancer in the world with an increasing incidence especially in younger adults (1). In locally advanced rectal cancer (LARC), neoadjuvant chemoradiotherapy (nCRT) followed by radical surgery is recommended by international guidelines (2, 3). Tumor recurrence and patient's survival are strongly influenced by the quality of response to neoadjuvant treatment (nT; refs. 4–6). Recent advances in the management of patients with LARC have shown that it could be conceivable to avoid amputation of the rectum (preservative strategy; e.g., Watch-and-Wait) in patients with clinical and imaging features compatible with a complete response to nT (7, 8). These patients experience acceptable outcomes; however, about 25% of them develop early tumor regrowth (9). There are currently no molecular markers to predict responses to nCRT and guide treatment decisions (3), such as optimization or modification of nT in nonresponding patients, and better selection of patients eligible to preservative strategy.

Ionizing radiation has the capacity to prime/reinforce an adaptive T-cell–mediated immune response, which acts in the mechanisms of local tumor regression and of distant tumor inhibition and rejection (i.e., the abscopal effect; refs. 10–12). This suggests that the quality and intensity of the natural immune reaction at tumor site before nT could influence the magnitude of response to nT and provide a predictive marker of response. Natural immune reaction at tumor site has been further associated with a favorable prognosis in various cancers (13), including colorectal cancer, treated by surgery alone (14, 15). Recent advances in digital pathology and image analysis have allowed a translation of immune assessment into a clinically relevant application (16). Using these technologies, the first standardized immune-based assay for colon cancer called “Immunoscore” (IS; i.e., the combination of CD3⁺ and CD8⁺ T-cell densities in the tumor and its invasive margin) has been developed. Its robustness and prognostic performance in stage I–III colon cancer have been consolidated through an international validation study (17). Thus, IS provides a reliable evaluation of the natural immune reaction at tumor site.

Preliminary studies in rectal cancer have suggested that the natural immune reaction of tumors could be evaluated on biopsies (18–20), the only sample material available before treatment. A derivation of the IS performed in initial biopsies (IS_B) before nT has the benefit of evaluating the quality of the initial immune response in the tumor and its potential influence on both the degree of response to nT and clinical outcome.

The aim of this study was to determine whether the IS_B correlates with response to nT, the in situ immune status after nT, and clinical outcome. We finally tested whether IS could help to better select the subgroup of patients eligible to a preservative strategy, with acceptable outcome.

Two retrospective consecutive cohorts of patients with LARC (n₁ = 131, n₂ = 118) with available biopsies, treated by nT and radical surgery by total mesorectal excision (TME), were analyzed. Cohort 1 was a monocentric cohort, and cohort 2 was multicentric (Supplementary Table S1 and Supplementary Fig. S1). Inclusion period ranged from 1999 to 2016. nT and surgery criteria were defined by each institution. Overall, 64.2% of patients were male, and the median age at diagnosis was 65 years [interquartile range (IQR), 53.3–74.1]. Patients were treated by nT [short (3.7%) or long (96.3%) course of radiation; 5-fluorouracil–based chemotherapy (82%); 18% did not receive chemotherapy]. According to the clinical TNM (cTNM) classification (UICC TNM 8th edition), rectal tumors were classified as cTNM I (1.2%), II (27.3%), and III (71.5%) based on baseline staging information provided by pelvic magnetic resonance and chest/abdominal computed tomography imaging. An additional cohort of patients (n = 73) with a complete/nearly complete response to nT (ycTNM 0–1), followed by a Watch-and-Wait strategy, was analyzed (Supplementary Table S2). A flow chart illustrating the studied population in each part of the investigations is provided in Supplementary Fig. S2. The median duration of follow-up for disease-free survival (DFS) of the cohort 1+2 was 45.4 months (IQR, 25.7–65.6). Duration of follow-up of each cohort for DFS, time to recurrence (TTR), and overall survival (OS) with the number of events is provided in Supplementary Table S3. The study was performed in accordance with the Declaration of Helsinki. Approval of all the centers' Institutional Review Board committee was obtained for the study. Signed-informed consents from patients were obtained in each center.

Patients were compared according to the degree of tumoral response to nT, using different tumor regression grade scoring systems: (i) the Dworak classification (21) defined as complete (Dworak 4), near complete (Dworak 3), moderate (Dworak 2), minimal (Dworak 1), and no regression (Dworak 0); (ii) the neoadjuvant rectal (NAR) score classification (5), calculated using the equation [5pN-3(cT-pT) + 12]^2/9.61, and classified as low (<8), intermediate (8–16), and high (>16); (iii) the ypTNM stage, i.e., the postsurgical pathologic T and N evaluation; and (iv) downstaging of the tumor (4), defined as complete (ypT0N0), intermediate (ypT1-2N0), or weak/absent (ypT3-4 or N+). For patients who underwent surgery, the events were local, systemic recurrences and death from the date of surgery for DFS, recurrences for TTR, and death from any cause for OS. All patients who were managed with the Watch-and-Wait strategy were considered to have clinical complete response (ycTNM0) and were offered a strict surveillance protocol.

Initial biopsies of all patients performed for diagnosis purpose were retrieved from all centers. Two formalin-fixed paraffin-embedded (FFPE) tumor tissue sections of 4 μm were processed for IHC with antibodies against CD3⁺ (2GV6, 0.4 μg/mL; Ventana) and CD8⁺ (C8/144B, 3 μg/mL; Dako) according to the previously described protocol (17) revealed with the Ultraview Universal DAB IHC Detection Kit (Ventana) and counterstained with Mayer's hematoxylin.

Digital images of stained tissue sections were obtained with a 20× magnification and a resolution of 0.45 μm/pixel (Nanozoomer HT). Delimitation of the tumoral component excluding normal tissue and low-/high-grade dysplasia-associated lesions was performed by an experienced pathologist (C. Lagorce). The mean densities of CD3⁺ and CD8⁺ T cells in the tumor region were determined with a dedicated IS module of the Developer XD image analysis software (Definiens; Fig. 1A). The mean and distribution of the staining intensities were monitored providing an internal staining quality control. A final quality check was performed to remove nonspecific staining detected by the software. Determination of IS_B was directly derived from the methodology used to determine the Immunoscore® (HalioDx, France) in the international validation cohort of IS in colon cancers which have shown a strong interobserver reproducibility (17). CD3⁺ and CD8⁺ T-cell densities in the tumoral region of each patient were compared with that obtained for the whole cohort of patient and converted accordingly into percentile. Then, the mean of the two percentiles (CD3 and CD8) was translated into one of the three IS_B categories (Fig. 1B): IS_B low (0%–25%), IS_B intermediate (>25%–70%), and IS_B high (>70%–100%). The IS_B determination was performed blinded to the study endpoint.

Total RNA from 20-μm FFPE tumor tissue sections from all patients for which both biopsies and the corresponding surgical specimen post-nT was available (cohorts 1 and 2; n = 62) and from patients with colorectal cancer not treated with nT (n = 13) was isolated using the RecoverAll Total Nucleic Acid Isolation Kit (Ambion Thermo Fisher). Distribution of tumor extension T and N stages among patients with or without nT did not display any statistical difference. The quality and quantity of the isolated RNA were measured using Agilent RNA 6000 Nano kit (Agilent Technologies) and NanoDrop 2000 (Thermo Fisher Scientific), and 100–400 ng RNA of each sample was processed using an in-house panel of 44 immune-related genes (Nanostring Technologies; Supplementary Table S4). Reporter-capture probe pairs were hybridized, and the probe/target complexes were immobilized and counted on the nCounter analyzer. Background subtraction was applied to raw data, and normalization based on the geometric mean of positive control and internal housekeeping genes (GUSB, SP2) was performed using the nSolver Analysis software, version 2.5.

Statistical analyses and data visualizations were performed using the R software version 3.5.1 with the add-on survival, survminer, ggpubr, ggplot2, rms, and coin packages. The associations between IS_B and clinical characteristics were assessed through the χ² or Fisher tests of independence. Association level between CD3⁺ and CD8⁺ cell densities was measured by Pearson's correlation coefficient r and related P value. Survival univariate analyses were performed using the log-rank test and the Cox proportional hazards model. Survival curves were estimated by the Kaplan–Meier method. The log-rank test for trend from the survminer package was performed to detect ordered differences in survival curves. Multivariate survival analyses were performed with Cox proportional hazards model to test the simultaneous influence of all covariates. The proportional hazards assumption for each covariate was tested using the cox.zph function. The relative importance of each parameter to survival risk was assessed by the χ² from Harrell's rms R package. The association between IS_B and nT ordinal response level was assessed using a unilateral linear-by-linear association test. The associations between nT response levels, CD3⁺ T-cell densities, CD8⁺ T-cell densities, and gene intensities were assessed by Kendall's correlation test, T test, and Mann–Whitney U test. The Wilcoxon test adjusted to control FDR by using the Benjamini and Hochberg procedure was used to test treatment response level in transcriptional analysis. The ycTNM staging and IS_B were included in the proportional odds ordinal logistic regression model to predict good histopathologic response to nT. P values < 0.05 were considered statistically significant. Principal component analysis (PCA) was performed with PCA and fviz_pca_ind functions from packages FactoMineR and factoextra. Linear weighted kappa was used to measure the agreement between resected tumors and biopsy samples in IS calculation.

CD3⁺ lymphocytes and cytotoxic CD8⁺ cells were assessed on initial tumor biopsies performed for diagnosis purpose of LARC (n = 322) treated by nT. The immunostaining intensity was monitored to ensure a valid detection and counting of stained cells with the image analysis software (Fig. 1A). Seven patients were excluded after biomarker quality control (2.8%), and 4 patients were excluded after clinical data quality control (1.2%). The median densities of CD3⁺ and CD8⁺ T cells in the tumor were 1,363 cells/mm² and 274 cells/mm², respectively (Supplementary Fig. S3A). The CD3⁺/CD8⁺ T-cell ratio was highly variable among patients, with a coefficient of determination (r²) between both markers of 0.58 (Supplementary Fig. S3B). IS_B was derived from the CD3⁺ and CD8⁺ T-cell densities (Fig. 1B). CD3 and CD8 densities in the tumor were converted into percentiles referring to the densities observed in all patients. IS_B mean percentile of CD3 and CD8 was calculated for each biopsy (IS_B mean score). No statistical difference for the mean score was observed between the two cohorts [cohort 1: 51.9 ± 29.1, cohort 2: 48.8 ± 23.4 (mean ± SD); P = 0.36]. After converting the mean score into IS_B scoring system, overall 22.7%, 52.5%, and 24.8% of patients had IS_B low, intermediate, and high, respectively. Of note, IS_B intermediate category was more represented in cohort 2 (61.9%), as compared with cohort 1 (43.5%). Representative tumor biopsies from patients with IS_B low and IS_B high are shown in Fig. 1C.

Distribution analysis of IS_B did not display any association with age, sex, or tumor location (Supplementary Table S1). The magnitude and reproducibility of the IS_B prognostic performance were tested in two independent cohorts. In cohort 1 (n₁ = 131), a significant difference in DFS between patients stratified by IS_B was observed [P test for trend (P_tft) = 0.012; HR_{(High vs. Low)} = 0.21; 95% confidence interval (CI), 0.06–0.78 and illustrated by Kaplan–Meier curves; Fig. 1D]. Patients with IS_B high were at low risk of relapse, with the 5-year DFS of 91.1% (95% CI, 82.0–100) versus 65.8% (95% CI, 49.8–86.9) in patients with IS_B low. These results were confirmed in second independent cohort [n₂ = 118; P_tft = 0.021; HR_{(High vs. Low)} = 0.25; 95% CI, 0.07–0.86; Fig. 1E]. Identical results were obtained when removing the 3 patients with UICC-TNM stage I tumors (Supplementary Fig. S4). In pooled analysis (n = 249), a significant difference between patient's groups stratified by IS_B was evidenced by univariate analysis (Supplementary Table S5) illustrated by Kaplan–Meier curves for TTR (P < 0.001), DFS (P < 0.005), and OS (P = 0.04; Supplementary Fig. S5).

We investigated whether the prognostic value associated to IS_B was at least partly a consequence of a relationship between IS_B and the quality of the nT response. The quality of response to nT was assessed 6 to 8 weeks after nT by imaging (ycTNM) and microscopic examination of the resected tumor, by the Dworak classification, a tumor regression grading system, ypTNM, downstaging, and the NAR score. In our cohorts (n = 249 patients), high CD3⁺ and CD8⁺ T-cell densities were significantly associated with a good response to nT evaluated by both Dworak classification and ypTNM staging (all P < 0.005; Fig. 2A). The mean of CD3⁺ and CD8⁺ percentiles (IS_B mean score) were correlated with the NAR score, Dworak classification, and ypTNM staging (Fig. 2B). The IS_B level and distribution were positively correlated with tumor response to nT (Fig. 2C). IS_B high patients were not found in the nonresponder Dworak 0 group, and 52.9% of patients with undetectable tumor cells (i.e., the Dworak 4 group) were IS_B high (P = 0.0006). The same correlation was observed with the ypTNM, tumor downstaging, and NAR (Fig. 2C). Good responders to nT were 6 times more frequent in the IS_B high group than in the IS_B low group according to the NAR scoring system (Supplementary Fig. S6). Immune consequences of nT were then investigated on post-nT tumor samples (Dworak 0–4; n = 62) by analyzing 44 immune-related genes (Supplementary Table S4). Gene expression levels were highly variable among patients (Supplementary Fig. S7). Unsupervised hierarchical clustering showed that 30.6% (n = 19) of patients presented with signs of local immune activation after nT (Fig. 3A). The immune activation status after nT was positively correlated with the densities of CD3⁺ and CD8⁺ T cells (i.e., IS_B) before treatment (Fig. 3B; P = 0.006). Nonresponder tumors (Dworak 0–1) presented a similarly low level of immune-related genes expression compared with tumors not treated by nT (Fig. 3A). Patients with a partial/complete response to nT had a significantly higher expression of genes associated with adaptive immunity (CD3D, CD3E, CD3Z, and CD8A), Th1 orientation (TBX21/Tbet and STAT4), activation (CD69), cytotoxicity (GZMA, GZMH, GZMK, and PRF1), immune checkpoints (CTLA-4 and LAG3), and chemokines (CCL2, CCL5, and CX3CL1), as compared with nonresponders to nT (Fig. 3C). This suggests a link between the quality of the natural adaptive cytotoxic immune response (IS_B), the presence of a post-nT immune activation, and the degree of response to nT. Gene expression data analysis through a PCA visualization further reinforced the putative link existing between the response to nT and the immune environment by showing distinct patterns of gene expression depending on the degree of response to nT (Fig. 3D). The combination of the second and third dimensions was the most accurate to discriminate responders/nonresponder patients.

We investigated whether the IS_B could provide valuable prognostic information when combined with clinical and pathological criteria available: (i) before nT [i.e., initial imaging, cTNM (UICC TNM 8th edition)], (ii) after nT (i.e., imaging post-nT, ycTNM), and (iii) after surgery (pathologic examination, ypTNM). In Cox multivariate analysis, IS_B was a stronger predictive marker of DFS than other clinicopathologic parameters including cTNM (IS_B high vs. IS_B low: HR, 0.2; P < 0.001) and ycTNM (IS_B high vs. IS_B low: HR, 0.25; P = 0.039). IS_B further remained a significant independent parameter associated with DFS when combined to ypTNM (Table 1). The important contribution of IS_B for the prediction of the relapse and/or death compared with other clinical parameters is illustrated in Fig. 4A. It is known that the accuracy of the complete response post-nT defined by imaging is imperfect. Only 25% to 50% of clinical complete responders have no residual tumor (i.e., complete histologic response; refs. 22–24). IS_B combined to imaging post-nT (ycTNM) increased the accuracy of prediction of histologic good responders (ypTNM 0–I) as compared with ycTNM alone (Fig. 4B). Three out of 32 patients with good response to nT (ycTNM = 0–I, n = 32) experienced distant relapses, and no local relapse was observed. Importantly, no relapse was observed in IS_B high patients (Fig. 4C). Thus, IS_B could help to select patients who could achieve a very favorable outcome and be eligible to a Watch-and-Wait strategy.

In a series of patient treated by the Watch-and-Wait strategy (n = 73), we retrieved the initial diagnostic biopsies to evaluate the IS_B and the associated clinical outcome. Overall, 23%, 51%, and 26% were classified as IS_B high, IS_B intermediate, and IS_B low, respectively. Time to relapse was significantly different among patients stratified for IS_B [P_{(High vs. Low)} = 0.025; Fig. 5A]. No evidence of relapse was noticed during the follow-up period in IS_B high patients. Under the Cox proportional hazards regression model, the 5-year probability of survival without recurrence ranged from 46% to 89% according to IS_B mean score (Fig. 5B). In Cox multivariable analysis, IS_B was related to the patient's TTR, independent of age, tumor location, and the cTNM classification (UICC TNM 8th edition; P = 0.04; Supplementary Table S6). The relative contribution of IS_B to the prediction of occurrence of disease relapse is illustrated in Fig. 5C.

This work highlights the links between: (i) the quality of natural intratumor immunity evaluated by the IS_B, (ii) the intensity of in situ immune reaction post-nT, (iii) the extent of the tumor regression post-nT, and (iv) the clinical impact in terms of prevention of tumor recurrence and survival. From a clinical point of view, IS_B provides a reliable estimate of both the quality of response after nT and of the risk of recurrence and death in patients with LARC. IS_B combined with imaging could further identify patients with a complete clinical response who can benefit from a close surveillance strategy post-nT, thus avoiding a disabling and useless rectal amputation surgery.

IS_B can be performed on routine diagnosis biopsies without any additional medical procedure. The rigorous and standardized quantification of immune cell infiltrates was achieved as for the IS colon study (17).

In the current study, IS_B was positively and significantly correlated with tumor response to nT. This observation is consistent with our previous preliminary result (18) and with studies using an optical semiquantitative evaluation of immune cell infiltrates (19, 20, 25). In the IS_B low group (22.7% of the cohort), only 5% of patients experienced complete response (low NAR score), suggesting that an optimization or modification of nT such as adjunctive therapies (26), immunotherapy (27), or drug repositioning may provide greater benefits for these patients in order to achieve a better response. We evidenced an association between signs of in situ cytotoxic adaptive immune response and inflammatory interferon type I–associated molecules production post-nT and the response to treatment. Type I IFNs play a key role in antitumoral immunity by promoting the maturation and presentation capacity of dendritic cells and their migration to lymph nodes (28). This immune state was influenced by the quality and the intensity of the natural immune response preexisting before nT. IS_B high could not only favor nT-dependent tumor cell death, but also promote the presence of resident immune components that could be essential to avoid local recurrences in organ preservative strategies such as Watch-and-Wait. Of note, a few IS_B high patients did not achieve a good response, highlighting that treatment resistance is also guided by independent tumor-intrinsic factors (29) or the presence of a suppressive microenvironment (30). Neoadjuvant treatment with development of clinical complete response post-nT has raised the possibility of organ-preserving strategies, as radical resection of the rectum results in functional outcomes, immediate morbidity, and even mortality rates (31). However, imaging after nCRT (ycTNM) has low accuracy in predicting pathologic complete response due to over- or under-staging (32). Importantly, no relapse was observed in good responders with IS_B high patients. In addition, IS_B increased the accuracy prediction for very good responders (ypTNM 0–I) evaluated by imaging and identified a subgroup of patients treated with an organ-preserving strategy (Watch-and-Wait) with a very favorable outcome. No biomarker is currently available to help selection of good responders eligible to the Watch-and-Wait strategy (9). These results may have significant implication in selecting potential candidates for organ preservation including patients with IS_B high and complete clinical response to nT, but also those with a delayed complete clinical response (i.e., “nearly complete responders”) that are presently classified as incomplete responders (33).

This study has some limitations. The immune densities associated with predefined cut points (i.e., 25th and 70th percentiles) are closely linked to the clinical characteristics of the studied cohort. The densities used as cut point are relevant to patients with LARC treated by nT before surgery. In addition, assessment of IS_B was performed on initial biopsies; this implies analysis of only a small fraction of the tumor (10%–15% of the surface of cut from a tumor block available after TME) and no analysis of the invasive margin absent on biopsies. In order to evaluate the correspondence of IS_B and IS in resected tumor, we analyzed 33 colon cancer biopsies and their associated resected tumor, and we found a partial correlation between these two specimens (Supplementary Table S7; kappa = 0.45, P = 0.0004). All discrepancies were only observed between two consecutive categories of IS. Despite this limited surface analysis and the absence of invasive margin, the prognostic value of the IS_B was retained suggesting the accuracy of the immune evaluation on initial diagnostic biopsy when the surgical piece is unavailable or is impossible to analyze due to architectural changes secondary to the nT. In addition, performing IS on postoperative specimen would not allow an assessment of its predictive value of the response to nT. Furthermore, due to the deep histologic modifications after nT (no clear delimitation of the tumor and its invasive margin), an IS on post-nT specimen is not feasible. The study was performed on patients who came from different countries and received standard-of-care treatment in real-life clinical practice. Despite the size of the specimen and the multiple types of patient care, the strong and constant prognosis value associated with IS_B highlights the robustness of the test and its generalizability. Prognostic parameters such as mismatch repair, KRAS, and BRAF status not available in our study were not included in multivariate analysis with IS scoring system. However, MSI+ cases are rare in rectal cancer (<5%; ref. 34), and we recently evidenced that IS was an independent prognostic parameter for survival when associated with MSI, KRAS, and BRAF status in colonic cancers (35). Most of the rectal cancers included in this study were adenocarcinomas. A subanalysis by histologic subtypes could not be performed due to the large multicentric character of the cohorts studied, with heterogeneous level of histopathologic description and the obvious small effect of mucinous carcinomas, signet ring cell carcinoma, or tumor budding to address their relative prognostic impact with enough power. This study emphasizes the importance of the initial diagnostic biopsies, often done in private practices, and not easily available in some cases. Patients with rectal cancer would benefit from a close partnership between private pathology practices, clinics, and teaching hospitals in order to initially assess their immune status (IS_B). This material could become in the near future essential and be part of the personal medical file of patients with rectal cancer as it is the sole material available before any nT. IS_B may facilitate a personalized multimodal treatment of rectal cancer particularly in patients with IS_B high tumors at baseline and with signs of tumor regression by imaging. These patients should benefit the most from the conservative strategy and in turn preserve their quality of life.

In conclusion, our results indicate that IS_B could be used: (i) to predict tumor response after nT, (ii) to restage local disease after nT, and (iii) to predict clinical outcome. This method may facilitate a personalized multimodal treatment of rectal cancer particularly in patients with IS_B high tumors at baseline and with signs of tumor regression by imaging. These patients should benefit the most from the conservative strategy and in turn preserve their quality of life. IS_B is yet to be validated on larger Watch-and-Wait cohorts both retrospectively and prospectively. Such validations are planned in international collaboration studies using the International Watch-and-Wait Database and in the OPERA ongoing clinical trial (NCT02505750).

B. Mlecnik reports patents WO2014023706A1 and WO2017194556A1 on methods and kits for screening patients with cancer, licensed and with royalties paid from HalioDX. F. Bibeau reports personal fees from Amgen; grants and personal fees from BMS; personal fees from MSD, Merck, and Bayer; and grants and personal fees from AstraZeneca outside the submitted work. D. Tougeron reports personal fees from Sanofi (board), MSD (board); personal fees and other from AMGEN (board, travel for congress), Servier (board, travel for congress); personal fees from Roche (board), BMS (board), AstraZeneca (board), and Bayer (board) outside the submitted work. R.O. Perez reports personal fees from Johnson & Johnson and personal fees from Roche outside the submitted work. J.-P. Gerard reports grants from Centre Antoine Lacassagne outside the submitted work. J. Galon reports personal fees from HalioDx (HalioDx co-founder) and personal fees from INSERM (licence) during the conduct of the study; grants from Imcheck Therapeutics, AstraZeneca, MedImmune, IObiotech, Janssen, Perkin Elmer, Ultivue; personal fees from BMS (consulting), IOBiotech (consulting), Northwest Biotherapeutics (consulting), Gilead (consulting), Amgen (consulting), Catalym (consulting), and Sanofi (consulting) outside the submitted work; in addition, the author has a patent for biomarkers and methods for predicting survival pending, licensed, and with royalties paid from HalioDx & INSERM. F. Pagès reports grants and personal fees from BMS; grants from HalioDx; personal fees from Gilead, Merck, Roche, and Janssen outside the submitted work; in addition, the author has a patent for WO2007045996 on an in vitro method for the prognosis of progression of a cancer and of the outcome, and a patent for WO/2013/186374 on the quantification of immune cells in tumor tissues, both owned by HalioDX and both applications by INSERM. No potential conflicts of interest were disclosed by the other authors.

C. El Sissy: Conceptualization, data curation, formal analysis, investigation, writing-original draft, writing-review and editing. A. Kirilovsky: Conceptualization, data curation, formal analysis, investigation, writing-original draft, writing-review and editing. M. Van den Eynde: Resources, data curation. A.-M. Muşină: Resources, data curation. M.-G. Anitei: Resources, data curation. A. Romero: Resources, data curation, formal analysis, investigation. F. Marliot: Data curation, formal analysis. A. Junca: Resources, data curation. J. Doyen: Resources, data curation. B. Mlecnik: Formal analysis, software. N. Haicheur: Data curation, formal analysis. T. Fredriksen: Data curation, formal analysis, methodology. C. Lagorce: Data curation, supervision. A. Jouret-Mourin: Resources, data curation. D. Leonard: Resources, data curation. F. Bibeau: Resources, data curation. S. Iseas: Resources, data curation. E.L. Roca: Resources, data curation. A.M. Cabanne: Resources, data curation. C.A. Vaccaro: Resources, data curation. J.P. Santino: Resources, data curation. E. Huertas: Resources, data curation. D. Tougeron: Resources, data curation. C. Carvalho: Resources, data curation. N. Figueiredo: Resources, data curation. R.O. Perez: Resources, data curation. A. Habr-Gama: Resources, data curation. V. Scripcariu: Resources, data curation. J.-P. Gerard: Resources, data curation, writing-review and editing. J. Galon: Investigation, supervision, funding acquisition. G. Zeitoun: Conceptualization, supervision, data curation, writing-original draft, writing-review and editing. F. Pagès: Conceptualization, data curation, supervision, funding acquisition, investigation, methodology, writing-original draft, writing-review and editing.

We thank Anatomopathology Departments (Belgium, Romania, CAL Nice, Medipath Nice, CHPG Monaco) for the retrieval of tumor biopsies and collaboration. The authors would like to acknowledge Magdalena Benetkiewicz (Sc.D., her work was funded by Université de Paris) for the editorial assistance.

The work was supported by Fondation ARC; Association pour la Recherche contre le Cancer; National Institute of Health and Medical Research (INSERM); CARPEM; AP-HP; University Paris Descartes; and INCA translationnel.

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.