Previous studies attempted to investigate the meaningful biomarkers in prediction of prognosis in ccRCC including lncRNA (21), microRNA (22, 23), circRNA or high-frequency mutants (24). However, the tumor associated immune signature has been less reported. In our study, we identified a total of 14 hub immune signature associated with survival and constructed an integrative TAIG model from the multivariate Cox regression results. We systematically assessed the prognostic value of TAIG that was demonstrated to be an independent prognostic factor versus other risk clinical features via Cox regression models. Not only the patients with high TAIG showed poor survival outcomes, but the TAIG also correlated positively with AJCC-TNM stages, pathological stages or tumor grades. Besides, we further calculated the TMB for each patient and illustrated the mutation profiles in ccRCC. Wilcoxon rank-sum test revealed the superior relativity between TAIG and TMB, and high TMB levels predicted poor survival outcomes and progression. Given the high correlation of TAIG with TMB and several immune-related crosstalk enriched significantly in differential immune signature, we attempted to figure out whether theses signature correlated with immune infiltrates in ccRCC tumor microenvironment. Firstly, we estimated the abundance of immune cells in each ccRCC sample based on CIBERSORT algorithm, and we conducted the differential analysis using Wilcoxon rank-sum test. Then, we additionally assessed the prognostic value of differentially distributed immune cells in two TAIG groups. Interestingly, we observed that most of these differential immune infiltrating cells possessed significantly prognostic value in ccRCC, among which higher infiltrating density of memory activated CD4+ T cell, T follicular helper cells, T regulatory cell or M0 macrophage were hazard factors in high-risk TAIG group, yet higher levels of resting mast cells with resting dendritic cells in low-TAIG group may contribute to tumor suppressors in ccRCC.
The tumor associated immune signature included a total of 14 genes possessing prognostic ability. We found most of genes belonged to cytokines or their corresponding receptors, and the GO enriched items revealed the associated pathways consisted of regulation of cell-cell adhesion, activation of T cell or proliferation. Previous researches uncovered the vital roles of cytokines or chemokines, such as IL-4, IL-18 or CXCL family (25–28) in promotion of tumor inflammatory response associated with prognosis. Among these genes, HLA-G was reported as an immune checkpoint molecular and functioning an inhibitor especially for cytotoxic activity of infiltrating NK cells through ILT2 (29), in accordance with our results. We further established a quantitative model named TAIG as an immune risk score for assessing the hazard levels of each patient. The distributions of all identified 14 immune signatures in two TAIG groups were in agreement with the subsequent Kaplan-Meier analysis, where the hazard immune signature showed higher expression profiles in high-TAIG group. But protective immune genes tend to reveal low expression levels. Though we figured out the clinical significance of TAIG and the tight correlations with TNM stages or tumor grades, whether the combination of TAIG with other risk clinical features could further optimize the predictive model warranted large samples to validate and clinical feasibility should be evaluated again. It is worth mentioning that we discussed the role of TMB in prognosis of ccRCC and analyzed the associations of TMB with TAIG. It has been well recognized that high TMB may yield many neoantigens to stimulate immune response thus correlating with better effect of immunotherapy. Given that previous studies across 33 cancer types have already implicated higher-TMB patients could gain a more favorable prognosis if treated with immunotherapy, otherwise would reveal a worse prognosis compared to lower-TMB patients (30), we hypothesized ccRCC patients with high TAIG and TMB levels might be considered the preferable options of immunotherapies.
Apart from the characterization of immune signature in ccRCC, we also investigated the tumor-infiltrating immune cells that accounted for the indispensable components in immune microenvironment. CIBERSORT, a newly computational approach developed by scholars in Stanford University (31), implemented a deconvolution algorithm to characterize diverse cell types based on gene signature matrix. Avoiding the defects of large material resources or time needed in flow cytometry and immunohistochemistry, we could quantify the immune cell fractions in each patient, especially appropriate for dealing with large samples. In agreement with research by Giraldo NA et al. revealing that tumor-infiltrating and peripheral blood T-cell immunophenotypes predict early relapse in localized ccRCC (32), our study also distinguished the risk T cell subsets, including CD4+ T cell, T follicular helper cells, as well as T regulatory (Treg) cell. Since the tumor associated macrophage (TAM) was recognized as a promoter in tumor progression and reported currently as powerful predictors for outcomes with Tyrosine kinase inhibitors (TKI) therapy in ccRCC, we also found that the M0 macrophage subset correlated positively with OS prognosis which was less reported (33). However, resting mast cells and dendritic cells showed the protective factors in ccRCC, and the dendritic cells was reported as an immune enhancer utilized as baseline in immunotherapy for solid tumors (34, 35). What is more, we also used another method to infer the fractions of immune cells based on the characteristic immune cells marker gene (36). From another aspect, we also demonstrated the underlying relationships between infiltrating immune cells with our identified signature and we illustrated the specific associations between one gene with single cell subset. These prognostic tumor-infiltrating immune cells all correlated with our immune signature, and we proposed the hypothesis that these immune signatures impact the differential infiltrating density of immune cells, thus influencing the prognosis in ccRCC.
Accordingly, we validated our risk signature in another data set from the ICGC, which is publicly available database providing the international community with comprehensive genomic data for various cancer types. The predictive value of TAIG still maintained superior with AUC = 0.72. Though the P value of log-rank test in Kaplan-Meier analysis was 0.083, we considered the marginal difference resulted from the smaller sample size with only 91 patients and the median cutoff was defined improperly that need to be further optimized. Taken together, it is our first attempt to uncover the risk immune signature in ccRCC based on large samples with high-throughput data. Additionally, we discussed the TMB and TAIG-related infiltrating immune cells. Characterization of immune landscape from tumor-associated immune gene signature to relative prognostic immune cell profiles in microenvironment favor our comprehensive understanding of prognosis, even immunotherapy strategies in ccRCC.
However, there existed several problems in our work as the following. Firstly, the correlation between TAIG with TMB or immune infiltrates was calculated based on statistics, the actual regulatory mechanisms among them are warranted for further demonstration. Secondly, the fractions or prognostic value of TAIG-related immune cells might be validated by flow cytometry. Lastly, the clinical significance of TAIG needs to be determined by our own cohort which we are preparing to conduct in next work.