T-Cell Infiltration and Clonality May Identify Distinct Survival Groups in Colorectal Cancer: Development and Validation of a Prognostic Model Based on The Cancer Genome Atlas (TCGA) and Clinical Proteomic Tumor Analysis Consortium (CPTAC)

Abstract

Predicting the survival outcomes of patients with colorectal cancer (CRC) remains challenging. We investigated the prognostic significance of the transcriptome and tumour-infiltrating lymphocyte T-cell receptor (TIL/Tc-TCR) repertoire and analysed TIL/Tc-TCR sequences of The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumor Analysis Consortium (CPTAC) CRC cohorts. Using a multivariate Cox regression, we tested whether TIL/Tc-TCR repertoire, patient and tumour characteristics (stage, sidedness, total non-synonymous mutations, microsatellite instability (MSI) and transcriptional signatures) correlated with patient overall survival (OS) and designed a prognostic nomogram. A multivariate analysis (C-index = 0.75) showed that only patient age, disease stage, TIL/Tc degree of infiltration and clonality were independent prognostic factors for OS. The cut-offs for patients’ allocation to TIL/Tc abundance subgroups were determined using a strategy of maximally selected rank statistics with the OptimalCutpoints R package. These were “high”, “low” and “very high” (90 th percentile) TIL/Tc infiltration-stratified OS (median not reached, 67 and 44.3 months; p < 0.001); the results were validated in the CPTAC cohort. TIL/Tc clonality was prognostic (median OS in “high” vs. “low” clonality not reached and 67.3 months; p = 0.041) and independent of TIL/Tc infiltration. Whilst tumour sidedness was not prognostic, the “very highly” infiltrated tumours were prevalent among right-sided CRCs (p = 0.039) and showed distinct immunological features, with lower Th1 signature (p = 0.004), higher PD-L1 expression (p < 0.001) and likely enrichment in highly suppressory IL1R1+ Tregs (FoxP3 and IL1R1 overexpression, p < 0.001). TIL/Tc abundance and clonality are independent prognosticators in CRC and, combined with clinical variables, refine risk stratification. We identified a subset of CRCs with “very high” TIL/Tc infiltration, poor prognosis and distinct genetic and immunologic features, which may benefit from alternative therapeutic approaches. These results need validation in prospective patient cohorts.

Keywords: T-cell antigen receptor; colorectal neoplasms; nomograms; prognostic factors; prognostic model.

Figure 1

Prognostic factors for overall survival in patients from The Cancer Genome Atlas colorectal cancer cohort. (a) Forest plot showing hazard ratio (HR) for mortality with 95% C.I. for significant covariates retained in the multivariable Cox regression prognostic model calculated using the fast–backwards method and the Akaike information criterion as a stopping rule and summarised in Supplementary Materials Table S1 (cases = 337, Cox model for OS global p < 0.001). Age, HR 0.03, p < 0.0001; stage (localised), HR 0.34, p < 0.0001; TIL/Tc clonality (high), HR 0.86, p = 0.002; TIL/Tc infiltration (low), HR 0.55, p = 0.011; TIL/Tc infiltration (very high), HR 1.43, p < 0.001. (b) Derived nomogram for survival probability integrating clinical variables and TIL/Tc features. (c) Survival curves for patients with high (pink) or low (blue) TIL/Tc clonality (log-rank p = 0.041). (d) Survival curves for patients with “low” (blue), “high” (green) or “very high” (orange) TIL/Tc abundance (log-rank p < 0.001).

Figure 2

TIL/Tc infiltration according to CRC sidedness. (a) Estimated number of T cell genomes in the sample, indicative of TIL/Tc abundance, according to the cancer site. The dotted horizontal line represents the threshold for very high TIL/Tc abundance (90th percentile). (b) The same cases are grouped according to anatomical location and TIL/Tc abundance (“low or high” = below the threshold indicated in panel (a); “very high” = above the same threshold. Fisher’s exact test p = 0.039). Horizontal bars in (a) represent median values.

Figure 3

Relationship between TIL/Tc abundance and mutational landscape. (a) Total non-synonymous mutation count in cancers with low (median = 159), high (median = 119) and very high (median = 108) TIL/Tc abundance (Kruskal–Wallis analysis of variance test p = 0.5170). (b) Distribution of the second dominant SNV in cases with low/high and very high TIL/Tc abundance; relative to total SNV, the prevalence of C > A was 6% vs. 12%, the prevalence of C > G was 39 vs. 30%; the prevalence of T > A was 26 vs. 21%; the prevalence of T > C was 14 vs. 21% and the prevalence of T > C was 16 vs. 15% (Chi-square test p = 0.4146). (c) MANTIS–MSI scores in cancers with low (median = 0.35), high (median = 0.35) and very high (median = 0.26) TIL/Tc abundance (Kruskal–Wallis analysis of bariance test p = 0.0893). Horizontal bars in (a,c) represent median values. ns, not significant.

Figure 4

Oncoplot of the 20 most frequent mutations in “not highly infiltrated” vs. “very highly infiltrated” CRCs. (a) Top mutated genes in the “not highly infiltrated” (“low” plus “highly infiltrated) CRCs. (b) Top mutated genes in the “very highly” infiltrated CRCs. The upper chart of the diagrams represents the number of hits for every single patient. On the left are the genes’ names; the grid represents the gene status for every patient (the colour code for the alteration subtypes is in the legend on the bottom); on the right, the percentual frequency and the absolute number of each gene alteration.

Figure 5

Unfavourable tumour microenvironment in the “very highly” infiltrated CRCs in the TCGA cohort. (a) Th1 infiltration score as measured by xCell [33] in the “low/highly infiltrated” vs. “very highly infiltrated” tumours (median 0.14 vs. 0.11, respectively, p = 0.004). (b) PD-L1 expression in the “low/highly infiltrated” vs. “very highly infiltrated” tumours (median z-score 47.3 vs. 145.7, respectively, p < 0.001). (c) FOXP3 expression in the “low/highly infiltrated” vs. “very highly infiltrated” tumours (median 41.32 vs. 115.70, respectively, p < 0.001). (d) IL1R1 expression in the “low/highly infiltrated” vs. “very highly infiltrated” tumours (median 510.3 vs. 1026, respectively, p < 0.001). ** p <0.01; **** p < 0.0001.