PI3K pathway mutation predicts an activated immune microenvironment and better immunotherapeutic efficacy in head and neck squamous cell carcinoma

Somatic mutation data and corresponding clinical information for the three independent HNSC cohorts, including the immunotherapy cohort MSKCC-2019 as well as the non-immunotherapy cohorts MD-Anderson and TCGA-HNSC, were obtained from the cBioPortal (https://​www.​cbioportal.​org/​) and TCGA (http://​portal.​gdc.​cancer.​gov/​) databases [16‐18]. FPKM-normalized RNA-Seq data of the TCGA-HNSC cohort were acquired from the UCSC-Xena database (http://​xena.​ucsc.​edu/​) and further transformed to log2 (TPM + 1).

We obtained 29 PI3K pathway genes from a previous study (Table S1) [19]. The maftools package was utilized to extract the mutation status of PI3K pathway genes in the three cohorts, and the mutation landscape of different mutation types was further visualized using the ComplexHeatmap package [20, 21].

The 129 samples from MSKCC-2019 cohort, 501 samples from TCGA-HNSC cohort, and 40 samples from MD-Anderson cohort were included in the survival analysis, respectively. Referring to previous criteria, we defined samples with at least one PI3K pathway gene mutated as the PI3K pathway mutation group, otherwise wild group [22, 23]. Multivariate Cox regression and Kaplan-Meier survival analysis were further performed to evaluate prognostic differences between the PI3K pathway mutation and wild groups in the immunotherapy cohort and the non-immunotherapy cohorts.

TMB is defined as the total number of base substitutions, insertions, and deletions per million bases in the coding region [24]. The TMB of each sample was calculated by the “tmb” function in the maftools package [21]. In addition, we recruited multiple measures, including neoantigens, SNV neoantigens, nonsilent mutation rate, and TCR Shannon from Thorsson V et al., to assess differences in overall genomic instability and immune status between the two groups [25].

Based on the mutation and copy number alteration (CNA) data from the TCGA-HNSC cohort, we comprehensively analyzed the PI3K pathway mutation and wild groups. The analysis steps are as follows: (1) we first calculated the TMB for each tumor sample and selected the top 15 mutated genes for visualization; (2) referring to the previous study, we extracted the four most common mutational signatures (mutational signatures 1, 2, 4, and 7) in HNSC using the deconstructSigs package and further calculated the proportion of their frequencies [26, 27]; (3) after obtaining CNA data from the Firebrowse website (http://​firebrowse.​org/​), we chose chromosome fragments with broad-level CNA frequencies greater than 33% and six classical genes located on chromosomes 7p11.2, 3p24.1, and 9p21.3 for display. In addition, we calculated copy number gain and loss loads occurring at focal and chromosomal arm levels to evaluate global CNA loads between the PI3K pathway mutation and wild groups.

We performed gene set variation analysis (GSVA) and gene set enrichment analysis (GSEA) to explore potential Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Hallmark pathways associated with PI3K pathway mutation and wild phenotypes in HNSC. Afterward, we selected 10 GSVA pathways to visualize via heatmaps and 5 GSEA pathways to visualize via GSEA plots for the mutation and wild groups, respectively.

We applied CIBERSORT and xCell algorithms to evaluate the abundance of immune cell infiltration in the PI3K pathway mutation and wild groups, respectively [28, 29]. In addition, based on Thorsson V et al., we obtained 69 immunomodulators, including 20 co-stimulatory molecules, 12 co-inhibitory molecules, 18 ligand molecules, and 19 receptor molecules [25]. The expression differences of these immunomodulators were further compared between the PI3K pathway mutation and wild groups.

Using the subclass mapping (SubMap) algorithm [30], an unsupervised method, we compared the similarity of mRNA expression patterns between the PI3K pathway mutation and wild groups in the TCGA-HNSC cohort and immunotherapy-responsive and non-responsive populations from seven immunotherapy cohorts (GSE35640, GSE100797, GSE135222, GSE145996, Nathanson, IMvigor210, and TCGA-SKCM). The tumor immune dysfunction and exclusion (TIDE) algorithm can predict immune checkpoint inhibitors (ICIs) efficacy by modeling the two main mechanisms of tumor immune escape (induction of T cell dysfunction and prevention of T cell infiltration) [31]. We further inferred the response rates to ICIs in the PI3K pathway mutation and wild groups via the TIDE web tool (http://​tide.​dfci.​harvard.​edu/​).

We then identified potential drugs for the PI3K pathway wild group using the following three steps: (1) the expression data of cancer cell lines (CCLs) were obtained from the Cancer Cell Line Encyclopedia (CCLE) database. The drug-sensitive data with AUC values were obtained from the Cancer Therapeutics Response Portal (CTRP) and PRISM repurposing databases; (2) afterward, we further estimated the drug sensitivity of each sample in the TCGA-HNSC cohort according to the pRRophetic package [32]; (3) finally, based on differential drug response analysis, we identified potential drugs sensitive to patients in the PI3K pathway wild group using the following thresholds: log2FC > 0 and P < 0.01 for CTRP and PRISM data.

All data processing and analysis for this study were performed using the R 4.1.1 software. Continuous variables between the PI3K pathway mutation and wild groups were analyzed by Student’s t-test or Wilcoxon rank-sum test; categorical variables were analyzed by chi-square test or Fisher’s exact test. Differences in Kaplan-Meier survival curves between the two groups were evaluated using the log-rank test. All P values were two-sided, and P < 0.05 was considered statistically significant.

The workflow for this study is shown in Fig. S1. A total of 13, 17, and 3 PI3K pathway mutated genes were identified in the MSKCC-2019, TCGA-HNSC, and MD-Anderson cohorts (Figs. 1A, B and S2A). Multivariate Cox regression analysis showed that the hazard ratio (HR) of PI3K pathway mutation phenotype in the immunotherapy cohort MSKCC-2019 was 0.533 (95% CI: 0.313–0.910; p = 0.021), whereas the HRs in the non-immunotherapy cohorts TCGA-HNSC and MD-Anderson were 0.888 (95% CI: 0.636–1.241; p = 0.487) and 1.939 (95% CI: 0.483–7.781; p = 0.351), respectively (Figs. 1C, D and S2B). In addition, Kaplan-Meier survival analysis was further performed to determine whether the PI3K pathway mutation phenotype was associated with OS and DFS (disease-free survival). Similarly, patients with PI3K pathway mutation had significantly prolonged OS in the immunotherapy cohort MSKCC-2019 (n = 129), but there were not statistically significantly associated with OS and DFS in the non-immunotherapy cohorts TCGA-HNSC (n = 501) and MD-Anderson (n = 40; Figs. 1E, F and S2C-D).

After stratifying HNSC patients into the PI3K pathway mutation and wild groups, the results of all three cohorts exhibited significantly higher TMB in the PI3K pathway mutation group than in the wild group (Fig. 2B–D). In addition, we chose the top 15 mutated genes of the TCGA-HNSC cohort further to explore the mutation landscape of the two groups. The results showed that the mutations in tumor suppressor gene CDKN2A and oncogenes TTN, MUC16, and NOTCH1 were more common in the mutation group (Fig. 2A). In addition, CNA analysis at focal and chromosomal arm levels exhibited a higher frequency of copy number deletions in the wild group (P < 0.05), while copy number amplifications failed to reach statistical significance, although there was a corresponding trend (P > 0.05; Fig. 2A and E–H). These results were also confirmed at the gene level by high-frequency amplifications of oncogenes (EGFR and PSPH) and high-frequency deletions of tumor suppressor genes (TGFBR2, EOMES, CDKN2A, and CDKN2B) in the PI3K pathway wild group (Fig. 2A).

GSVA enrichment analysis integrating GO, KEGG, and Hallmark pathways indicated that immune-related pathways were significantly enriched in the PI3K pathway mutation group, while the PI3K pathway wild group was closely related to metabolism-related pathways (Fig. 3A). Furthermore, as shown in Fig. 3B, the GSEA analysis result also demonstrated that the PI3K pathway mutation group was prominently enriched for immune pathways, such as B cell-mediated immunity (NES = 2.334, FDR < 0.01), MHC class II protein complex (NES = 2.200, FDR < 0.01), MHC protein binding (NES = 1.791, FDR < 0.01), T cell receptor complex (NES = 1.503, FDR < 0.01), and positive regulation of T cell proliferation (NES = 1.329, FDR < 0.01), while the PI3K pathway wild group was significantly correlated with extracellular matrix (ECM) receptor interaction (NES = − 2.064, FDR < 0.01), angiogenesis (NES = − 1.773, FDR < 0.01), positive regulation of mitochondrial fission (NES = − 1.509, FDR < 0.01), ribose phosphate biosynthetic process (NES = − 1.437, FDR < 0.01), and fatty acid beta oxidation (NES = − 1.428, FDR < 0.01; Fig. 3C).

The results of xCell and CIBERSORT consistently demonstrated that the PI3K pathway mutation group possessed superior immune effector and immunosuppressive cells infiltration, including B cell, memory B cell, memory CD4 T cell, central memory CD8 T cell, and regulatory T cell (Treg). However, stromal components were more enriched in the PI3K pathway wild group (such as endothelial cell; Figs. 4A and S2E-F). In addition to abundant immune cells, the mutation group also exhibited higher levels of co-stimulatory molecules (such as HAVCR1 and TNFRSF13C) and co-inhibitory molecules (such as CD274, LAG3, TIGIT) expression (Fig. 4B, C). We have also observed corresponding trends in other molecules that play a key role in anti-tumor immunity and immunotherapy, such as ligand molecules and receptor molecules (Figs. S2G and S3A). Furthermore, several critical indicators for evaluating tumor immunogenicity and antigen presentation efficiency collected from the study of Thorsson et al. also indicated that the PI3K pathway mutation group displayed significantly higher neoantigen, SNV neoantigens, nonsilent mutation rate, and TCR Shannon (Fig. 4D–G) [25].

The SubMap analysis demonstrated that the PI3K pathway mutation group was genetically more similar to patients responding to immunotherapy, particularly anti-PD-L1 therapy (P < 0.05; Fig. 5A–G). In parallel, TIDE also revealed a significantly higher response rate to immunotherapy in the PI3K pathway mutation group (P < 0.05; Fig. 5H). In addition, using drug-sensitive data from the CTRP and PRISM databases combined with CCLs expression data from the CCLE database, we further developed five potential therapeutic drugs for the PI3K pathway wild group, including “BMS-536924,” “linsitinib,” “NVP-TAE684,” “PLX-4720,” and “clonazepam” (P < 0.01; Fig. 5I–J).

Given the accurate predictive power of the PI3K pathway mutation for immunotherapy in HNSC, we further explored its significance in other common tumors based on pan-cancer genomic data and immunotherapy information from the MSKCC-2019 cohort (esophagogastric cancer and colorectal cancer) and MSKCC-2014 cohort (melanoma). As shown in Fig. 5K–L, the results of Kaplan-Meier survival analysis revealed a significantly prolonged OS in immunotherapy cohorts of esophagogastric cancer (P = 0.031) and colorectal cancer (P = 0.0022). It is worth pointing out that since the MSKCC-2014 cohort had only 64 melanoma samples, although their Kaplan-Meier curves prominently separated, the result did not reach statistical significance (P = 0.18; Fig. 5M).

Given that numerous studies have demonstrated that human papillomavirus (HPV) infection status is a favorable prognostic factor in HNSC patients [33, 34], based on HPV status information from the TCGA-HNSC cohort, we further explored the relationship between the PI3K pathway mutation and HPV status and patient prognosis. As exhibited in Fig. S3B-C, we calculated the proportion of HPV status in the PI3K pathway wild and mutation group and the proportion of PI3K pathway mutation in patients with different HPV status, and the differences did not reach statistical significance. On the other hand, the PI3K pathway mutation status was unable to observably stratify OS in HNSC patients (P = 0.68); however, HPV-positive status was associated with significantly prolonged OS (P = 0.04; Fig. 1F and S3D). In addition, we performed Kaplan-Meier survival analysis of PI3K pathway mutation in HPV-negative patients (among the 19 HPV-positive HNSC patients, only 1 of 12 PI3K wild-type patients died, while 7 PI3K pathway mutated patients did not even die, and further subgroup analysis could not be performed). The results demonstrated that when we stratified patients according to HPV status, there was no significant difference in the effect of PI3K pathway mutation on OS, and the differences did not reach statistical significance (P = 0.62; Fig. S3E). Finally, we performed Kaplan-Meier survival analysis of HPV status in patients with PI3K pathway mutation and wild group separately (Fig. S3F-G). Consistent with Fig. S3D, HPV-positive patients owned prolonged OS in both PI3K pathway mutation and wild groups, but the differences did not reach statistical significance limited to their small sample size. Overall, these results certified that HPV status can predict a better prognosis in HNSC patients and is not affected by PI3K pathway mutation status.