Tumor-immune microenvironment (TIME) analyses in LUAD
RNA-Seq data of LUAD, SKCM, and KIRC were analyzed for the transcriptome-based immunoprofiling of the TIME across tumor types [19]. Relative tumor-infiltrating levels of immune cells were estimated in silico using previously published single sample gene set enrichment analysis (ssGSEA)-based approaches. Along with immune signature scores, expression values of three immunotherapy targets (PD-1, PD-L1, CTLA-4), as well as ESTIMATE algorithm-based total immune/stromal scores [21] were also estimated for LUAD. The TIME profile was then subject to hierarchical clustering for immune-based taxonomy (Fig. 1A-B). A heatmap result showed samples were largely segregated into three subgroups according to immune infiltration patterns. A “hot” subgroup (n=136, 26.3%), termed for highly inflamed tumors with elevated levels of immune infiltrations [3, 25], was enriched in CD56dim NK cells, cytotoxic cells, CD8+ T cells, B cells, Th1, as well as regulatory T cells (Treg). An “intermediate” subgroup (n=163, 31.5%) manifested a heterogeneous presence of immune cells at varying levels with intact infiltrations of macrophages, dendritic cells (DC), neutrophils, central and effector memory T cells (Tcm, Tem respectively). An immunologically “cold” subgroup (n=218, 42.2%) [26] was dominantly immune-excluded, exhibited the lowest antigen presentation machinery (APM) score and nominal infiltration of immune cells except Th17, Th2 and CD56bright NK cells (Fig. 1A). To broadly assess potential inhibitory mechanisms in this cohort, we evaluated immune checkpoint gene expression based on the aforementioned TIME subtypes (Fig. 1C-F). PD-1 and CTLA-4 were remarkably upregulated in the hot cluster, closely heeding distribution of tumor-infiltrating CD8+ T cells, Treg, cytotoxic cells and CD56dim NK cells, whereas their levels stagnated in the intermediate and cold subgroups. The PD-L1 level was highest in the hot subgroup and showed a steady decline in the intermediate and cold subgroups. To interrogate the relationship between expression levels of immune checkpoints in the TIME context, we conducted correlation analyses with PD-1, CTLA-4, and PD-L1 in the LUAD cohort. The expression levels of CTLA-4 and PD-1 were strongly related (Spearman’s r=0.75, p<2.2e-16) (Fig. 1G) while correlation of CTLA-4 and PD-L1 was moderate (Spearman’s r=0.52, p<2.2e-16) (Supplementary Fig. 1A). The normalized correlation level[23] between CTLA-4 and PD-1 was also stronger (z score =6.194) than that of CTLA-4 and PD-L1 (z = 4.014). Taken together, our results may suggest the CIT could be more beneficial in LUAD patients belonging to the hot subgroup with copious immune cell infiltrations and strong concurrent expression among the immune checkpoints.
Tumor-immune Microenvironment (Time) Analyses In Kirc
When we assessed a KIRC cohort employing the same computational methodologies, patients were similarly segregated into three TIME subgroups (Fig. 2A-B). Various Teff and B cells were enriched predominantly in the hot subgroup (n=114, 22.5%), and as infiltrations of the majority of T cells decreased, certain immune cell types (Th17, neutrophils, and DCs) remained prevalent in the intermediate group (n=267, 52.8%). The cold subgroup (n=125, 24.7%) showed similar ESTIMATE immune scores to the intermediate (p>0.05, Fig. 2B), but the cold subgroup presented poorer infiltrations of CD8+ T cells, memory T cells and higher infiltration of Treg, NK cells or B cells compared with the intermediate subgroup. In the KIRC cohort, PD-1 and CTLA-4 were highly upregulated in the hot subgroup and downregulated the most in the intermediate group (Fig. 2C-F). In contrast, PD-L1 expression was dispersed throughout the cohort and showed modest upregulation in the hot subgroup (Fig. 2E). When the relationship between PD-1 vs. CTLA-4 (Fig. 2G) was analyzed, expression of these two genes displayed a strong correlation (Spearman’s r=0.75, p<2.2e-16; normalized z score=5.168), whereas PD-L1 and CTLA-4 did not (Spearman’s r=0.25, p=1.5e-08; z score=1.710) (Supplementary Fig. 1B). Similar to LUAD, our results imply patients belonging to the immune-hot subgroups with concordant overexpression of PD-1 and CTLA-4 may more likely benefit from the CIT than those of the intermediate or cold subgroup.
Tumor-immune Microenvironment (Time) Analyses In Skcm
Three immune-subgroups were also observed for the SKCM cohort (Fig. 3A-B) after unsupervised clustering of the TIME profiles. The first subgroup signifying a hot TIME (n=88, 18.6%) showed the highest infiltration of CD8+ T cells, Th1, CD56dim NK cells, cytotoxic cells, Treg, as well as the majority of antigen presenting cells (APC) into tumors. The intermediate group (n=221, 46.8%) showed lower degrees of infiltration compared with the hot group, while the cold group was almost completely immune-excluded with few exceptions (Th2, eosinophils and Tcm among others) (n=163, 34.5%). We documented PD-1, CTLA-4 and PD-L1 were largely upregulated in the hot subgroup but downregulated the most in an immune-excluded cold subgroup (Fig. 3C-F). While CTLA-4 levels between the intermediate and the cold subgroups were different with statistical significance (p=0.036), the difference was much less dramatic compared with PD-1 (p<2.2e-16), suggesting the possibility that CTLA-4 is not expressed in strong accordance with PD-1 or with the TIME context (Fig. 3C-D). This discordance unseen in previous tumor types was further corroborated when correlation analyses were conducted with the expression levels of immune checkpoints. The correlation between PD-1 vs. CTLA-4 (Spearman’s r=0.58, p<2.2e-16, normalized z score=4.306; Fig. 3G) in the SKCM cohort was relatively weaker than the correlations documented in LUAD (z score=6.194) and KIRC (z score=5.168) (Fig. 1G and Fig. 2G respectively). The discordance between CTLA-4 and PD-1 levels in SKCM may explain previously reported survival outcome which slightly favored CIT over monotherapy across clinically relevant subgroups [11].
Tumor mutational burden (TMB) as an independent measure depicting TIME of LUAD
Recent studies have illustrated that the expression levels of immune-related genes [27, 28], degree of CD3+ tumor-infiltrating lymphocytes and TMB could be employed for predicting the clinical efficacy of immunotherapy in NSCLC [12, 29]. Building upon these findings, we evaluated correlations among the genomic markers of the LUAD cohort in the context of the TIME immune-subgroups. Nonsynonymous TMB values (Fig. 4A-B), ssGSEA-based T cell scores (Fig. 4C-D), or ESTIMATE immune scores (Supplementary Fig. 2B&Fig. 1B) were ordered according to the unsupervised clustering result shown in Fig. 1A. While T cell or immune scores were arranged as anticipated (e.g. highest in the hot subgroup and lowest in the cold subgroup), TMB sank in the intermediate subgroup and the level was comparable between the hot and cold subgroups (p>0.05; Fig. 4B), yielding the weak correlations with immune-infiltration patterns (Supplementary Fig. 2C-D). To more broadly investigate correlations between TMB and the estimated levels of 24 immune cells, we selected marker genes specific for immune cell types, and performed correlative analyses for each gene with TMB. While associations between TMB and expression of most of the marker genes for the immune cells were negligible, the genes specific for CD56dim NK cells, T helper cells, cytotoxic cells or activated DC (aDC) were positively correlated with TMB (Fig. 4E).
Previous reports have unraveled confounding impacts of tumor purity on TMB evaluation and the expression pattern of immune-related genes[30]. We documented that the purity of the LUAD tumors was inversely correlated with immune scores, stromal scores or infiltration of cytolytic NK cells, and that the tumor purity was not significantly correlated with TMB (Supplementary Fig. 3). To account for the tumor purity as confounding factors, it was demonstrated through partial Spearman’s correlation analyses that the levels of most tumor-infiltrating immune cells were negatively associated with TMB except APM (Spearman’s r=0.125, p=0.006), Th2 (r=0.361, p=3.4e-16) and CD56dim NK cells (r=0.216, p=1.77e-06) (Fig. 4F). Taken together, our data reflect the highly variable and independent dynamics between TMB and immune cell-mediated inflammatory responses across LUAD patients (Fig. 4G).
CTLA-4 and PD-1 are strongly associated in LUAD regardless of TMB
To further elucidate whether expression of different immune checkpoints are dependent upon TMB in LUAD, we divided the LUAD cohort into two groups: TMBHigh (TMB>median (=207.5)) and TMBLow (TMB≤median) (Supplementary Fig. 2A). Immune checkpoint expression was upregulated in the TMBHigh group (p<0.05; Fig. 5A-C) and TMB was weakly associated with expression of PD-1 and PD-L1 (Fig. 5D-F). The result was consistent when the expression levels of the immune checkpoints were adjusted for tumor purity: TMB was correlated with PD-L1 (r=0.172, p=0.0002) and PD-1 (r=0.148, p=0.001) but marginally with CTLA-4 (r=0.092, p=0.042) (Fig. 4F). We then investigated whether correlation between each immune checkpoint depends on TMB. Our results exhibited strong associations between CTLA-4 vs. PD-1 (Spearman’s r =0.74 for TMBHigh; r=0.75 for TMBLow; both p<2.2e-16) and moderate relationships between CTLA-4 vs. PD-L1 (Spearman’s r=0.53 for TMBHigh; r=0.5 for TMBLow; both p<2.2e-16) and PD-1 vs. PD-L1 (Spearman’s r=0.56 for TMBHigh; r=0.49 for TMBLow; both p≤1.1e-15), regardless of TMB (Fig. 5G-I). Altogether, our findings highlight the notion that the TIME context is independent of TMB [17] and that measuring TMB alone cannot serve as an established biomarker to predict T cell activation status in the TIME and thus sensitivity of the CIT in LUAD.