Tumor immunogenic signature (TIGS)
Unsupervised hierarchical clustering of all genes sequenced in the discovery cohort revealed three clusters of coexpressing genes. Refining these results using k-means (k = 3) clustering generated three stable clusters of genes and three clusters of patients (non-inflamed, borderline, and inflamed) [Fig. 1a]. Pathway analysis of these gene clusters distinguished them as cancer testis antigen genes, genes associated with the inflammation response, and other immune and neoplasm genes [Supplementary Tables S2, S3]. The 161 genes associated with the inflammation response were termed the TIGS, as the expression of these genes closely followed the degree of inflammation presented by each of the three patient clusters [Fig. 1b]. The distributions of the immunogenic scores of the all samples in each of sample cluster were used to establish boundaries between the strong, moderate, and weak immunogenic score groups.
To assess agreement of the algorithmic TIGS with observed immune cell infiltration, we analyzed the distribution of immunogenic score within three major types of CD8 infiltration patterns estimated by IHC (infiltrating/strongly infiltrating, non-infiltrating, and excluded) [Fig. 1c-e]. As expected, the median immunogenic score of infiltrating/strongly infiltrating samples (n = 493) was 54.85, whereas the median immunogenic score of noninfiltrating samples (n = 403) was significantly lower (median = 34.84; p = 2.22E-16). Interestingly, excluded phenotype (n = 26) of immune infiltration had a median immunogenic score similar to the strongly/moderately infiltrating phenotype (median = 50.83; p = 0.31), but significantly higher than the noninfiltrating pattern (p = 0.00032) [Fig. 1f].
TIGS and clinical outcomes
To assess clinical utility, TIGS was used to classify a previously published retrospective cohort of 242 samples with ICI outcomes (melanoma, NSCLC, and RCC) into strongly, moderately, and weakly immunogenic groups [11, 12, 14] [Fig. 2a]. Strongly immunogenic tumors showed higher objective response rate (ORR) compared to weakly immunogenic tumors (37% vs 23%; p = 0.06) to checkpoint inhibition in the retrospective cohort. Tumor type-specific analysis showed similar results in melanoma (53% vs. 33%; p = 0.27), NSCLC (36% vs. 14%; p = 0.05), and RCC (25% vs 16%; p = 0.8) [Fig. 2b] [Supplementary Table S4].
Next, we investigated the impact of immunogenic score on overall survival in the retrospective cohort. Even though there was no significant difference in overall survival of strongly inflamed compared to weakly inflamed tumors (p = 0.19), we observed a clear separation of median survival between the two groups (25.6 months vs. 13.8 months) [Fig. 2c]. Multivariate analysis using Cox proportional hazard model revealed that weakly inflamed TIGS category had a significantly high hazard ratio (HR = 1.83 [1.09–3.06]; p = 0.022) compared to strongly inflamed category [Supplementary Fig. S3]. We further investigated the source of this survival difference by performing tumor type-specific survival analysis, which showed that most of the survival difference can be attributed to NSCLC cases (p = 0.0012; 15.4 months vs. 7.63 months) [Fig. 2d-f, Supplementary Fig. S4-S6; Supplementary Table S5]. This NSCLC survival effect was supported by the multivariate Cox proportional hazard analysis where melanoma (HR = 0.39 [0.24–0.66]; p < 0.001) and RCC (HR = 0.44 [0.24–0.81]; p = 0.008) had significantly less effect on overall survival difference compared to NSCLC [Supplementary Fig. S3]. Age, gender and TMB status had no significant association to overall survival (p > 0.05) [Supplementary Fig. S3]. Interestingly, multivariate analysis of PD-L1 IHC status showed that negative cases showed trend towards worse survival (HR = 1.51[0.93–2.45]; p = 0.095) [Supplementary Fig. S3].
TIGS and traditional biomarkers
To further investigate the utility of TIGS, we studied the predictive capacity of TIGS in conjunction with traditional biomarkers for response to ICI therapy such as PD-L1 expression and TMB high. The combination of TIGS and PD-L1 shows an additive effect on objective response rate to ICI therapy in the retrospective cohort [Fig. 3a]. A similar effect was observed for TMB [Fig. 3b]. In general, PD-L1+, strongly immunogenic patients had the highest clinical response rate for all three cancer types (excluding single-sample groups), and PD-L1-, weakly immunogenic patients had the lowest response rate (or in the case of melanoma, the second-lowest). Interestingly, PD-L1 and TMB in combination did not show a similar effect [Supplementary Fig. S7]. In melanoma, TMB high, strongly inflamed patients had an ORR of 72.73%, while TMB low, strongly inflamed patients had a response rate of 16.67%.
Finally, combining TIGS with PD-L1 and TMB status for NSCLC, melanoma and RCC, the prediction of objective response becomes even more robust [Supplementary Fig. S7]. A significantly higher [p = 0.0001] objective response rate of 69.23% was observed for PD-L1 positive, TMB high, strongly inflamed tumors, compared to an objective response rate of only 10.53% for PD-L1 negative, non-TMB high, weakly inflamed tumors.
TIGS and cell proliferation
In order to gain more comprehensive insight into the TME and its effect on immunotherapy response, an understanding of both immune and neoplastic influences is required. To achieve this, we combined TIGS with a previously published emerging biomarker of cell proliferation [12, 14]. Combining TIGS subgroups with cell proliferation classes of highly, moderately, and poorly proliferative tumors significantly improves objective response separation, where highly proliferative, inflamed tumors [55%] have significantly higher objective response to ICI therapy than poorly proliferative, non-inflamed tumors [14.28%; p = 0.0006] [Fig. 4a]. Tumor type-specific analysis were not performed due to small sample sizes within each group.
Further evidence demonstrated significant survival differences between different combinations of TIGS and cell proliferation [p = 0.012] [Fig. 4b]. Importantly, we noted that strongly inflamed and highly [median = not achieved; p = 0.025] or moderately [median = 16.2 months; p = 0.025] proliferative tumors had significantly better survival compared to weakly inflamed, highly proliferative tumors [median = 7.03 months] [Tables S6, S7]. Even though this difference was not statistically significant for individual tumor types, median OS was not reached for strongly inflamed and highly proliferative tumors for NSCLC and melanoma [Supplementary Fig. S9, S10, S11]. Multivariate Cox proportional hazard analysis showed no additional effect of age, gender, PD-L1 IHC status, and TMB status on the overall survival [Supplementary Fig. S8]. This data suggests and we hypothesize that both T cell proliferation and tumor cell proliferation contribute to the signal in highly inflamed and highly proliferative tumors, whereas only tumor cell proliferation appears to contribute to the measurement of highly proliferative, weakly inflamed tumors [Fig. 5]. Therefore, combining biomarkers of both neoplastic and immune influences as described could facilitate a more comprehensive understanding of the tumor immune microenvironment and likelihood of response to ICIs.