mTOR pathway gene mutations predict response to immune checkpoint inhibitors in multiple cancers

We searched for PathCards database (https://​pathcards.​genecards.​org) and identified genes involved in mTOR pathway. We compared the MSK-IMPACT, PGDx elio Tissue Complete, and Foundation One, the three widely used next generation sequencing (NGS) panel in clinical trial of TMB, for the mTOR pathway genes involved. We next scanned the mutation information in our discovery dataset of Memorial Sloan-Kettering Cancer Center (MSKCC) study and identified genes with available mutation information for further survival analysis.

We included patients who received at least one dose of ICI treatment for two-stage survival analysis. The ICI treatment including anti-PD1 (e.g., nivolumab and pembrolizumab), anti-PDL1 (e.g., atezolizumab, avelumab, and durvalumab) and anti-CTLA4 (e.g., ipilimumab and tremelimumab) agent. In the discovery, we included 1661 patients from a MSKCC cohort that investigated the association of TMB and ICI treatment [10]. We incorporated the 23 mTOR pathway genes with available mutation information into stepwise Cox proportional hazards regression model. Then, we generated the optimal fitted Cox model and selected the variables/genes iterated out. We integrated the variables/genes as a whole signature to explore the survival association in the comparison of mutant- versus wild-type cancer patients treated with ICI. In the validation, we included other 553 patients treated with ICI from other 6 independent published studies for survival association [20‐25]. To exclude the possibility that mTOR pathway mutations was a prognostic marker regardless of the ICI treatment, we also associated the mutation states with survival of non-ICI treated patients with 2244 patients from the MSKCC study as discovery and 763 advanced cancer patients from TCGA as validation. All survival analysis were visualized by Kaplan–Meier curve with log-rank test. To quantify the death risk, we also perform multivariate Cox regression for survival analysis with adjustment of age, gender, TMB and treatment. We adopted the restricted mean survival time (RMST) method for multivariate survival comparisons in the case of unmet requirements for Cox assumptions. Hazard ratio (HR) and life expectancy ratio (LER) were used to express the adjusted results for Cox and RMST model, respectively. Because the data for objective response rate (ORR) and durable clinical benefit (DCB) was limited, we also combined the two-stage patients available for these data and compare the ORR and DCB based on the mutation states of the signature using chi-square test.

We extracted 48,834 cancer patients from 188 non-redundant studies deposited in cbioportal database, and further identified 44,078 patients with available mutation information in 8-gene signature in mTOR pathway, including FGFR2, PIK3C3, FGFR4, FGFR1, FGF3, AKT1, mTOR, and RPTOR. We calculated the mutation frequency of these 8 genes as well as their integrated mutation frequency across multiple cancer types. The MAF files of TCGA were also used to plot the mutation exclusion pattern among them.

We performed subgroup analysis according to the mutation types to explore its effect on patients’ survival upon immune checkpoint blockade. Because of the limited sample size in some rare types of mutations, patients were divided into those with missense mutation and others without this type of mutation. Then Kaplan–Meier plot with log-rank test, along with multivariate analysis (e.g., Cox regression for discovery and RMST analysis for validation) were used for survival comparison.

We associated the mutation states of the 8-gene signature with TMB of 5646 cancer patients from MSKCC cohort and validated this association in TCGA including 10,163 patients. Wilcoxon rank-sum test was used to compare the means of TMB between two groups because of the abnormal distribution of the data. We also tested the survival association of the mutation states in low- (< 10 mut/Mb) and high-TMB (> 10 mut/Mb) patients treated with ICI in MSKCC cohort. In the validation of independent published studies for survival interaction with TMB, a cutoff of 10mut/Mb was also used in the next-generation sequencing study for TMB classification, and the upper quartile of TMB was used for classification in whole-exon sequencing studies. Because decreased sample size in subgroup division that may introduce statistical bias, we also pooled the results in discovery and validation by meta-analysis with random effects model.

As we know, the activation or inactivation of mTOR pathway influence its target genes. To explore whether the mutations may affect the activation status of mTOR pathway, we associated the 8-gene signature mutation in mTOR pathway with mRNA expression of the well recognized target genes using Wilcoxon rank-sum test.

To explain the relationship between 8-gene signature mutation in mTOR pathway and increased TMB, we asked whether there is an association between the signature mutation with genetic mutations involved in DNA damage response (DDR) pathway. Then, we extracted the transcriptome data to calculate immune phenotype (e.g., immune cells inflammation, antigen presentation, and neoantigen production) of cancer patients according to published TCGA studies [26, 27]. With Wilcoxon rank-sum test, we explored the impact of the signature mutation on immune phenotype using TCGA pan-cancer database. Specially, we listed the genes involved in cancer immune casecade, including antigen presentation, costimulatory/coinhibitory signals, related ligand and receptor, and cell adhesion. We computed the fold change in these genes in the comparison of mutant- versus wild-type patients and visualized it in one heatmap. We also explored the impact of the 8-gene signature mutation on expression of immune checkpoint and chemokine genes using Wilcoxon rank-sum test. Finally, gene set enrichment analysis (GSEA) was carried out to investigate the potential mechanism underlying the observation of immune phenotype in the comparison of mutant- versus wild-type patients. We further validated the GSEA results from TCGA by 275 advanced cancer patients in IMvigor210 clinical trial.

We reported one case with advanced lung squamous cell carcinoma in our center with AKT1 mutation, who was received two cycle treatment of 200 mg pembrolizumab upon initial diagnosis. The patient has signed a written informed consent to denote related clinical data for the purpose of scientific research. We got the computed tomography images before and after the pembrolizumab treatment, in order to visualize the treatment response.

By extraction of sequencing data from blood TMB study [28], we investigated whether there was a survival association of the cfDNA mutations in mTOR pathway with ICI treatment outcome. We also explored the clinical outcome in the comparison of docetaxel versus atezolizumab treatment in mutant-type patients.

We conducted a two-stage analysis (Additional file 2: Table S1) to explore and validate the association of mutations in mTOR pathway genes with survival of patients treated with ICI. The patients’ characteristics were presented in Additional file 2: Tables S2, S3. We found that compared with the PGDx elio Tissue Complete and Foundation One NGS panel, the MSK-IMPACT panel had the most genes involved in mTOR pathway. We identified 23 mTOR pathway genes (Additional file 2: Table S4) with available mutation information in 1661 patients from MSKCC study of TMB and immunology. By incorporating mutations in these genes in stepwise Cox regression, we further identified 8 genes, including FGFR2, PIK3C3, FGFR4, FGFR1, FGF3, AKT1, mTOR, and RPTOR, that fitted an optimal survival model for patients treated with ICI (Fig. 2A). The stepwise Cox analysis considered both forward and backward direction and resulted in decreased akaike information criterion (AIC) and variables as the increase in iteration (Fig. 2B). Of note, the 8 genes were found to be located at the key position of the mTOR pathway, including the ligand and receptor, and genes involved in midstream phosphorylase kinase (Fig. 2C). We next asked the survival association of mutation states for this 8-gene signature by integrated analysis. We observed prolonged survival by mutations of this 8-gene signature in mTOR pathway, with death risk decreased by 36% (HR = 0.64, 95%CI = 0.50–0.83, P = 5.17 × 10^–4) upon multivariate adjustment for age, sex, TMB, and treatment (Fig. 2D). Because of unmet assumptions for Cox model, we used RMST method and successfully validated this survival association in other 553 patients from 6 published trials who received ICI (LER = 1.70, 95%CI = 1.08–2.70, P = 0.023, Fig. 2F). In 2244 patients who did not receive ICI in the MSKCC study, we did not find any association between the mutations and patients’ survival (Fig. 2E). The lack of survival prediction for patients with non-ICI treatment was also validated in 763 advanced cancer patients from TCGA (Fig. 2G), indicating the specific predictive ability of the mutations in mTOR pathway genes for ICI treatment efficacy.

Using 44,078 cancer patients from 188 non-redundant studies deposited in cbioportal database, we found that the mutant frequency for the 8 genes ranged from 1.5 to 3% (Fig. 3A). Endometrial cancer had the leading mutation frequency by integrated analysis of the 8-gene signature in mTOR pathway, followed by breast cancer and melanoma (Fig. 3B). By analysis of MAF files of TCGA pan-cancer mutations, we found that most of them displayed significant co-mutation (Fig. 3C).

In both discovery (Additional file 1: Fig. S1A, B) and validation (Additional file 1: Fig. S1C, D), we found that the impact of mTOR pathway mutations on better survival of patients with ICI treatment was mainly attributed to the missense mutations, and we did not find any association of other mutations with patients’ survival. We also visualized the multivariate results in forest plot, in order to present the impact of overall, missense and other mutations, respectively, on patients’ survival upon ICI treatment (Additional file 1: Fig. S1E, F).

Mutations in the 8-gene signature was found to be associated with increased TMB in 5646 patients from the MSKCC cohort (Fig. 4A), which was further demonstrated in 10,163 patients from TCGA database (Fig. 4B). To explain the potential reason, we questioned whether there was a correlation between the mutations in mTOR pathway and DDR pathway. As a result, in both MSKCC and TCGA cohort, we found that the mutation frequency of most genes involved in DDR pathway were up-regulated in presence of mutations in the 8-gene signature in mTOR pathway (Additional file 1: Fig. S2). Specially, these up-regulations were involved in DNA double-strand breaks repair (DSBR), nucleotide excision repair (NER), and base excision repair (BER), including ATM, ERCC3 and 4, and BRCA1 and 2 genes, indicating that the abnormal DNA repair might be the inducement to the increased TMB in the patients with mutant-type signature in mTOR pathway (Additional file 2: Tables S5, S6). In MSKCC study, we found that TMB classification did not influence the trend towards better survival for mutant-type patients with adjustment for age, gender, TMB and treatment (HR = 0.67, 95%CI = 0.47–0.96, P = 0.028 for low-TMB patients, Fig. 4C; and HR = 0.66, 95%CI = 0.48–0.92, P = 0.013 for high-TMB patients, Fig. 4D). Using multivariate RMST analysis, we observed similar trend with borderline significance in the validation (Fig. 4E, F). We also calculated adjusted RMST results in discovery for polling with validation using meta-analysis method. Polling results demonstrated a better survival in presence of mTOR pathway mutations in low-TMB patients (Fig. 4G). In high-TMB patients, the results achieved a borderline trend similar to that in discovery (Fig. 4H). Because there are limited patients with available treatment response data, we also combined the samples in the two stages to extract the response data and found that mutant-type patients had higher treatment response rate as well as DCB rate compared with wild-type patients (Fig. 4I, J).

The common target genes of mTOR pathway were shown in Additional file 1: Fig. S3A. We observed up-regulation in mRNA expression of EIF4E (EIF4EBP1), AKT1, HIF-1a (mRNA transcription effector of mTOR pathway), S6K (RPS6KA1 and RPS6KB1), but down-regulation in PKC (PRKCA), SGK1, TFEB and ULK1 by the mTOR pathway mutations (Additional file 1: Fig. S3B). Although most have direction towards the activation of mTOR pathway, the deep regulation in protein function, phosphorylation modification for example, are still unclear.

Using transcriptome data extracted from TCGA, we found that mutations in the 8-gene signature in mTOR pathway induced higher overall immune cells infiltration, T cells infiltration, antigen presentation, and neoantigen production (Fig. 5A). Specially, we observed the increased infiltration of CD8 + T cells, activated B (plasma) cells, activated mast cells, activated CD4 + memory T cells, M1 macrophages, but decreased M2 macrophages (Fig. 5C). Further TCGA analysis also resulted in the tendency towards up-regulation in mutant-type patients for the mRNA expression involved in different immune phenotype (Fig. 5B), especially for chemokines (Fig. 5D) and immune checkpoints (Fig. 5E), including PD1 and PDL1 (also termed CD274, Fig. 5E) that were closely associated with immune cells recruitment and ICI treatment efficacy. To reveal the potential mechanism underlying our observation in TCGA, we performed pathway enrichment analysis using GSEA method. There was an obvious immune pathway enrichment, such as T cell receptor signaling, NK cell mediated cytotoxicity, antigen presentation (Fig. 6A, B), and interferon response pathway (Fig. 6B), in the comparison of mutant-type versus wild-type patients. As expected, the mTORC1 and metabolism pathway was also enriched (Fig. 6B), because of the function in metabolism reprogramming by mTOR pathway mutation. DNA repair and protein folding, two pathways that caused production of neoantigen, were also enriched by the signature mutation in mTOR pathway (Fig. 6A, B). We repeated the GSEA analysis in transcriptome data from IMvigor210 clinical trial, and validated the enriched pathways mainly overlapped in immunology (Additional file 1: Fig. S4). The results for GSEA analysis in TCGA and IMvigor210 in detail were presented in Additional file 2: Tables S7 and S8, respectively. Taken together, our observations provided the clues for the association between mTOR pathway mutations and “hot” microenvironment in tumors (Fig. 6C). We finally exemplified an 82-year-old patient with advanced lung squamous cell carcinoma in our center, who had AKT1 mutation in the tumor and good response to pembrolizumab. The tumor has shrunk by approximately two thirds after 2 cycle treatment of 200 mg pembrolizumab (Additional file 1: Fig. S5).

Blood-based NGS was performed in OAK and POPLAR trial with comparison of ICI treatment versus chemotherapy, which was extracted for the association analysis between the mTOR mutations and ICI treatment response. We identified 853 patients with cfDNA mutation data, of which 424 patients received docetaxel and 429 patients received atezolizumab. Among these patients, 157 had mutations in the 8-gene signature in mTOR pathways, including 85 patients treated with docetaxel and 72 patients treated with atezolizumab. We did not find any significant association of the mutations in mTOR pathway signature with efficacy of atezolizumab treatment (data not shown). However, compared with docetaxel, we found that mutant-type patients had prolonged overall survival (OS) (Additional file 1: Fig. S6A), progression-free survival (PFS) (Additional file 1: Fig. S6B), and increased ORR and DCB (Additional file 1: Fig. S6C, D) in the comparison of atezolizumab versus docetaxel treatment. In the subgroup analysis by TMB, survival remained favorable in mutant-type patients treated with atezolizumab in comparison with docetaxel (Additional file 1: Fig. S7), except for PFS in high-TMB patients that only reached insignificant trend in multivariate RMST analysis (Additional file 1: Fig. S7D).