Tumor Infiltrating Lymphocytes Signature as a New Pan-Cancer Predictive Biomarker of Anti PD-1/PD-L1 Efficacy

Abstract

Tumor immune infiltrates are associated with tumor prognosis in many cancer types. However, their capacity to predict the efficacy of checkpoint inhibitors is poorly documented. We generate three signatures that evaluate in different ways these infiltrates: lymphoid- and myeloid-alone signatures, and a combined signature of both named the TIL (tumor-infiltrating lymphocyte) transcriptomic signature. We evaluate these signatures in The Cancer Genome Atlas Program (TCGA) Pan-Cancer cohort and four cohorts comprising patients with melanoma, lung, and head and neck cancer treated with anti-PD-1 or anti-CTLA-4 therapies. We observe using TCGA Pan-Cancer cohort that this TIL or lymphoid-alone signature accurately estimates prognosis in most cancer types and outperforms histological TIL evaluation or myeloid signature alone. Both TIL and lymphoid signatures are correlated with response rate to immunotherapy. Combining lymphoid signature or TIL with tumor mutational burden generates a score that is highly efficient in predicting response to immunotherapy. In different series of patients treated with checkpoint inhibitors for non-small cell lung cancer, head and neck cancer, and melanoma, we observed that TIL or lymphoid signature were associated with outcome. These data demonstrate that a simple TIL or lymphoid signature could be used as a Pan-Cancer prognostic and predictive biomarker to estimate patient survival under checkpoint inhibitors.

Keywords: biomarkers; biostatistics; immunotherapy; tumor-infiltrating lymphocytes.

Figure 1

Evaluation of the tumor-infiltrating lymphocyte (TIL), lymphoid, and myeloid scores in the Pan-Cancer TCGA cohort. (A–C): Boxplots representing the distribution of the TIL (A), lymphoid (B), and myeloid (C) scores normalized between 0 and 1 according to tumor type for TCGA cohorts; for each score, tumor types are ordered by corresponding median values. (D): Spearman correlation matrix estimated for leukocyte, lymphoid, myeloid, TIL, and H&E TIL scores. Correlations were computed across all cancer types. H&E: hematoxylin and eosin.

Figure 2

Relation between TIL, lymphoid, and myeloid scores respectively and MCP-counter abundances. (A): Spearman correlation matrix estimated for lymphoid, myeloid, and TIL scores as well as absolute abundances of eight immune and two stromal cell populations estimated by MCP-counter. (B–D): Heatmaps showing, for each cancer type, the regression coefficients estimated when fitting a LASSO model to describe the relationship between TIL score (B), lymphoid score (C), and myeloid score (D) and 10 cell population abundances. Bar plots in rows represent the number of variables selected in the LASSO model.

Figure 3

Evaluation of the prognostic role of the TIL and lymphoid scores in the Pan-Cancer TCGA cohort. (A,B): Log hazard ratios estimated by univariate Cox models for (A) overall survival (OS) and (B) progression-free interval (PFI) respectively associated with the TIL, lymphoid, myeloid, immunophenoscore (IPS) and IMmuno-PREdictive Score (IMPRES) scores, for each TCGA cancer type. Blue cells represent good prognosis and red cells represent poor prognosis. Black cells represent a p value for a log rank test less than 0.05. (C,D): For each significant TCGA tumor type from panel A and B respectively, bar plots are showing the area under the curve (AUC) estimated for overall survival (C) or progression-free interval (D) univariate survival models involving the five scores. For each tumor, the five scores are ordered decreasingly; the first score is the one that best predicts the outcome.

Figure 4

Evaluation of the predictive role of the TIL and lymphoid scores in the Pan-Cancer TCGA cohort. (A–C): Correlation of median TIL (A), myeloid (B), and lymphoid (C) scores with objective response rate (ORR) to anti-PD-1/PD-L1 therapy across 21 TCGA cancer types. (D): Spearman correlation matrix estimated for lymphoid score, TMB, CD274 gene expression, expanded immune gene (EIG) signature, and objective response rate (ORR) to anti-PD-1/PD-L1 therapy. Correlations were computed across 21 TCGA cancer types.

Figure 5

Prognostic role of TIL and lymphoid scores in comparison to other scores in a control cohort. (A–F): Kaplan–Meier survival analysis based on the TIL (A,C) and lymphoid (B,D–F) scores for progression-free survival; patients from the pooled Montreal and Dijon cohort (A,B), Van Allen (C,D), GSE136961 (E), and GSE93157 (F) were stratified according to the cutoff obtained from maximally selected rank statistics. (G–J): Bar plots showing area under the curve (AUC) estimated for progression-free interval in univariate survival models involving TIL, lymphoid, IMPRES, IPS, and Tumor Immune Dysfunction and Exclusion (TIDE) scores (when score was available), CD8A genes and EIG signature for pooled Montreal and Dijon (G), Van Allen (H), GSE136961 (I), and GSE93157 (J) cohorts.