Association of alteration (ALT) with outcome measures in all tumors not treated with ICIs in the MSK-MetTropism cohort

Of 25,775 patients enrolled in the MSK-MetTropism cohort, 1992 (7.7%) were CDKN2A-DEL, 1412 (5.5%) were CDKN2A-MUT and 22,280 (86.4%) were CDKN2A-WT (Fig. 2A). This cohort included 27 cancer types, and Skin Cancer (Non − Melanoma) had the highest CDKN2A ALT frequency (46.7% were CDKN2A-DEL and 4.8% were CDKN2A-MUT), followed by melanoma, mesothelioma, pancreatic cancer, head and neck cancer, bladder cancer and gastrointestinal stromal tumor (Fig. 2B). Of 24,503 patients with complete OS data, CDKN2A-DEL patients had the most FGA, MSI score and largest number of patients having metastatic disease, while those CDKN2A-DELMUT patients had the biggest TMB and mutation count (Table S1). Log-rank test showed that CDKN2A-ALT (hazard ratio [HR], 1.76; 95% CI, 1.67–1.85; Fig. 2C), CDKN2A-DEL (HR, 1.81; 95% CI, 1.69–1.93; Fig. 2D), CDKN2A-MUT (HR, 1.69; 95% CI, 1.57–1.83; Fig. 2D) patients had poorer OS than CDKN2A-WT patients. Moreover, after adjusting FGA, MSI, TMB, age, metastasis status and gender using multivariable Cox regression, CDKN2A-DEL, CDKN2A-MUT, CDKN2A-Other ALT still could predict OS of participants (HR, 1.38; 95% CI, 1.29–1.48; 1.22; 95% CI, 1.12–1.33; 1.16; 95% CI, 0.85–1.58; Fig. 2E).

Fig. 2

CDKN2A alterations in the MSK-MetTropism cohort. (A) The distribution of CDKN2A-DEL, CDKN2A-MUT and CDKN2A-WT patients. CDKN2A-MUT: CDKN2A point mutation, CDKN2A-DEL: CDKN2A deletion, CDKN2A-WT: CDKN2A-wide type. (B) Frequency of CDKN2A DEL, CDKN2A MUT and CDKN2A other ALT in each cancer type. (C) Over survival (OS) in CDKN2A-WT and CDKN2A-alteration (ALT) patients. (D) OS in prespecified subgroups including CDKN2A-DEL, CDKN2A-MUT, CDKN2A-WT and CDKN2A other-ALT groups. HR: hazard ratio, CI: confidence interval. (E) Multivariable Cox regression analysis presented the association between CDKN2A DEL, CDKN2A MUT, CDKN2A other ALT, various types of cancer and OS after controlling for other factors including FGA, MSI score, TMB, age, and metastasis status. The median values for FGA, MSI Score, TMB, Age, and Mutation Count are 0.15, 0.24, 4.32, 63.9, and 5.0, respectively. Patients with values below the median were assigned to the reference group, while those equal to or above the median were assigned to the treatment group. Forest plot of association between CDKN2A ALT and OS of all patient (F) and patients diagnosed with primary cancer (G)

Subsequently, a pan-cancer survival analysis uncovered a consistent association between CDKN2A ALT and inferior OS in various cancer types (Fig. 2F). Likewise, a comparable association was observed in patients with CDKN2A-DEL (Figure S1E). However, this particular observation was limited to patients diagnosed with six specific types of cancer and having CDKN2A-MUT (Figure S1B). Significantly, a meta-analysis confirmed the positive correlation between CDKN2A ALT, CDKN2A MUT, CDKN2A DEL, and unfavorable prognosis in pan-cancers (Fig. 2F, Figure S1B and E).

Patients with CDKN2A ALT had poor prognosis in as many as 12 primary cancer types (Fig. 2G). Furthermore, when it came to metastatic cancers, we discovered that patients with CDKN2A ALT had even worse prognostic outcomes in 10 different cancer categories (Figure S1A). Additionally, our study enabled us to identify a specific subset of cancers that consistently displayed these distinctive prognostic characteristics. Examples of such cancers included pancreatic cancer, bladder cancer, thyroid cancer, renal cell carcinoma, and gastrointestinal stromal tumor (Fig. 2G and Figure S1A). Patients with CDKN2A DEL had similar outcomes (Figure S1F and G). CDKN2A MU were associated with poor prognosis in as many as six primary tumor types (Figure S1C). Furthermore, when it came to metastatic cancers, patients with CDKN2A MU had even worse prognostic outcomes in four different tumor categories (Figure S1D). Additionally, stratified survival analysis suggested that CDKN2A-ALT patients were shorter OS than CDKN2A-WT patients across multiple levels of CNA, MSI, TMB, age, mutation count, gender, tumor purity and race (Figure S1H). To determine the association between CDKN2A ALT and response to ICIs among pan-cancer patients treated with immunotherapy (CTLA-4 blockade, PD-1/PD-L1 blockade or combination), tumor-specific ORRs were summarized according to some cancer immunotherapy studies. Of note, with the frequency of CDKN2A ALT extracted from the MSK-MetTropism cohort, we found a significant correlation between CDKN2A ALT frequency and ORRs (r, 0.58; P = 0.002; Figure S1I), suggesting that cancer patients with CDKN2A ALT might benefit more from immunotherapy.

Association of CDKN2A ALT with outcome measures in all tumors treated with ICIs in the MSK-IMPACT cohort

We used 1661 patients from the MSK-IMPACT cohort treated with ICIs to assess changing effect of CDKN2A ALT on clinical outcomes of some cancer types after immunotherapy. Association of CDKN2A ALT with prolonged OS reached statistical significance in all patients (HR, 0.77; 95% CI, 0.59–0.99; Figure S2A). Particularly noteworthy, CDKN2A-MUT patients outlived CDKN2A-WT patients after immunotherapy (HR, 0.65; 95% CI, 0.47–0.88; Fig. 3A), whereas patients with CDKN2A DEL did not (HR, 1.27; 95% CI, 0.65–2.45; Fig. 3A). Moreover, multivariable Cox regression showed that CDKN2A DEL, CDKN2A MUT, CDKN2A other ALT could not predict the OS of participants after immunotherapy (HR, 1.54; 95% CI, 0.80–2.98; 0.88; 95% CI, 0.64–1.21; 1.77, 95% CI, 0.94–3.32; Fig. 3B). Next, in the MSK-IMPACT cohort with ICIs treatment, significant associations of CDKN2A MUT, DEL, and ALT with OS were not found in esophagogastric cancer, breast cancer, melanoma, and NSCLC except for bladder cancer (Figure S2B; Fig. 3C and D), in contrast to the MSK-MetTropism cohort (Fig. 2F, Figure S1B and E). These disparities may be due to immunotherapy used in the MSK-IMPACT cohort’s 1661 patients. Similarly, a random effects meta-analysis of studies found no significant association between CDKN2A MUT, DEL, ALT, and OS in pan-cancer patients (Fig. 3C and D, Figure S2B).

Fig. 3

Associations of CDKN2A alterations with OS and objective response rates to ICIs in the MSK-IMPACT cohort. (A) OS in prespecified subgroups including CDKN2A-DEL, CDKN2A-MUT, CDKN2A-WT and other ALT groups. (B) Multivariable Cox regression analysis presented the association between CDKN2A-DEL, CDKN2A-MUT and CDKN2A-other ALT and OS after controlling for other factors including drug type, sex, TMB, age, and metastasis status. Forest plot of association between CDKN2A-MUT (C) and CDKN2A-DEL (D) and OS. OS of CDKN2A-DEL, CDKN2A-MUT, CDKN2A-WT and other ALT groups with PD-1 (E) and CTLA-4 (F) inhibitors treatment, primary cancer (G) or metastasis (H)

Furthermore, survival analysis stratified by immunotherapy types suggested that CDKN2A-MUT patients treated with PD-1/ PD-L1 inhibitors (HR, 0.65; 95% CI, 0.44–0.95; Fig. 3E) but not CTLA-4 inhibitors (HR, 0.53, 95%CI, 0.24–1.18; Fig. 3F) or a combination of the two (HR, 1.15, 95%CI, 0.56–2.40; Figure S2C) had a better OS than CDKN2A-WT patients. Next, CDKN2A DEL was not related to OS in either primary or metastasis cancers (Fig. 3G and H; HR, 1.23; 95% CI, 0.55–2.77; metastasis, HR, 1.20; 95% CI, 0.38–3.74). Conversely, CDKN2A-MUT patients exhibited longer OS than CDKN2A-WT patients with metastatic disease (HR, 0.59; 95% CI, 0.40–0.87; Fig. 3H). Patients with primary cancers did not show this difference (HR, 0.81; 95% CI, 0.47–1.38; Fig. 3G). We combined the five cancer types with the highest CDKN2A ALT frequency in the MSK-MetTropism cohort and the TCGA cohort. Three cancer types were eventually identified and survival analyses were performed, and it was found that CDKN2A-ALT had a worse OSin PAAD and MESO except HNSC (Figure S2D-F). Based on this, we found that CDKN2A-ALT has a poor prognostic effect in some tumors, so pan-cancer analysis was subsequently performed.

Associations of CDKN2A ALT with clinical outcomes in TCGA cohort

The TCGA cohort (n = 10,953) was used to assess the effect of CDKN2A ALT on clinical outcomes. The CDKN2A ALT was associated with poorer OS (HR, 2.04; 95% CI, 1.89–2.19; Fig. 4A) and progression-free survival (PFS; HR, 1.87; 95% CI, 1.76–1.99; Fig. 4B). A pan-cancer survival analysis revealed that CDKN2A-ALT patients had a shorter OS and PFS than CDKN2A-WT patients in multiple cancer types, including LGG, KIRC, MESO, kidney renal papillary cell carcinoma (KIRP), pancreatic adenocarcinoma (PAAD), head and neck squamous cell carcinoma (HNSC), glioblastoma multiforme (GBM), lung adenocarcinoma (LUSC, Fig. 4F and G). In addition, 399 (3.6%) of all patients were CDKN2A-MUT, 1378 (12.6%) were CDKN2A-DEL, and 8673 (79.2%) were CDKN2A-WT (Figure S3A). In most tumor types, the frequency of CDKN2A-DEL was greater than that of CDKN2A MUT (Fig. 4E).

Fig. 4

Associations of CDKN2A alterations with clinical outcomes in the TCGA cohort. OS (A) and PFS (B) in CDKN2A-WT and CDKN2A-ALT patients. OS (C) and PFS (D) in CDKN2A-DEL, CDKN2A-MUT, CDKN2A-WT and CDKN2A-other ALT patients. (E) Frequency of CDKN2A-DEL, CDKN2A-MUT and CDKN2A-other ALT in each cancer type. Forest plot of association between CDKN2A ALT and OS (F) and PFS (G). Forest plot of association between CDKN2A MUT(H) and DEL(I) and OS

Next, all patients with CDKN2A MUT (OS, HR, 1.59, 95%CI: 1.36–1.86, Fig. 4C; PFS, 1.56, 95%CI: 1.36–1.79, Fig. 4D) or CDKN2A DEL (OS: HR, 2.67, 95%CI: 2.46–2.89, Fig. 4C; PFS, 2.37, 95%CI: 2.21–2.55, Fig. 4D) had a worse prognosis in TCGA cohort. Meta-analysis suggested that patients with CDKN2A DEL had poor OS (HR, 1.88, 95%CI: 1.41–2.51, Fig. 4I) and PFS (HR, 1.59, 95%CI: 1.26-2.00, Figure S3C), while those with CDKN2A MUT did not (OS: HR, 1.12, 95%CI: 0.94–1.34, Fig. 4H and PFS, 1.18, 95%CI: 0.95–1.46, Figure S3B). However, after extracting tumor-specific ORRs from some cancer immunotherapy studies, the correlation of CDKN2A ALT frequency with ORRs did not reach statistical significance (r, 0.2; p = 0.28) in TCGA cohort (Figure S3D). In conclusion, CDKN2A ALT was associated to a worse OS in patients who were not treated with ICIs; the addition of ICIs also might improve clinical outcomes in pan-cancer patients.

Associations of CDKN2A ALT with clinical outcomes in the merged cohort

Considering that the MSK-IMPACT cohort was a cohort after ICIs treatment, while the OrigiMed 2022 cohort lacked survival outcomes, we only merged the MSK-MetTropism cohort and TCGA cohort to explore the impact of CDKN2A ALT on patient survival. In the merged cohort, we found that CDKN2A ALT was associated with poorer OS (HR, 1.95; 95% CI, 1.86–2.03; Figure S3E). A pan-cancer survival analysis revealed that CDKN2A-ALT patients had a shorter OS than CDKN2A-WT patients in 23 cancer types, including thymic epithelial tumor, thyroid cancer, renal non clear cell carcinoma, salivary gland cancer, gastrointestinal stromal tumor, adrenocortical carcinoma and so on. Among the 37 types of cancers in the merged cohorts, 23 cancer types with CDKN2A ALT exhibited shorter OS compared to CDKN2A-WT patients (Figure S3F). These cancer types include some of the most common malignancies such as breast cancer, glioma, pancreatic cancer, prostate cancer, head and neck cancer, non-small cell lung cancer, gastrointestinal stromal tumor, bladder cancer, colorectal cancer, and hepatobiliary cancer. Additionally, a meta-analysis indicated that patients with CDKN2A ALT had poor OS (HR, 1.89, 95%CI: 1.54–2.32, Figure S3F).

Landscape of CDKN2A ALT in a China pan-cancer cohort (OrigiMed 2022)

We explored the prevalence of CDKN2A ALT among 10,194 patients across 25 cancer types in the OrigiMed 2022 cohort with clinical data (Fig. 5A). In total, 1220 patients (11.97%) had CDKN2A-ALT, 605 patients (5.93%) had CDKN2A-MUT, and 421 patients (4.13%) had CDKN2A-DEL (Figure S4A). Mutations diagram circles were colored with respect to the corresponding ALT types and showed that the H83Y was the most frequent somatic mutation site (Fig. 5B). Among nine genes germline pathogenic variants including APC, ATM, BRCA1, BRCA2, CDH1, MLH1, MSH2, MSH6, and PALB2 being associated with the risk of gastric cancer [43], CDKN2A had co-occurrence with MLH1, MSH6, BRCA1 and BRCA2 (Fig. 5C). To explore the immunotherapy response, we found the correlation between CDKN2A ALT frequency and ORRs in the OrigiMed 2022 cohort did not reach statistical significance (r, 0.29; p = 0.16; Fig. 5D). The highest frequency of CDKN2A ALT was observed in patients with esophagogastric cancer, followed by thymic tumor, pancreatic cancer, gallbladder carcinoma and melanoma (Fig. 5E). Patients with CDKN2A MUT had more elevated TMB, mutation count and age than those with CDKN2A WT (Fig. 5F, G and Figure S4B). Patients with CDKN2A ALT tended to have metastatic disease, be male and stage IV (Fig. 5H, and Figure S4C, D). Patients with CDKN2A MUT showed similar results. Those CDKN2A DEL patients were more likely to be recurrent and older than CDKN2A WT patients (Fig. 5H and Figure S4B). Furthermore, the treatment distribution differed significantly among four CDKN2A VAR subtypes (Fig. 4SE).

Fig. 5

Clinical and CDKN2A alterations characteristics in the OrigiMed cohort. (A) Landscape of somatic CDKN2A alterations across 25 cancer types. (B) The alterations sites of CDKN2A in the OrigiMed cohort. (C) Heatmap showing the pattern of co-occurrence/mutual exclusivity of CDKN2A, APC, ATM, BRCA1, BRCA2, CDH1, MLH1, MSH2, MSH6, and PALB2. (D) Correlation between CDKN2A ALT frequency and objective response rates to ICIs. (E) Frequency of CDKN2A-DEL, CDKN2A-MUT and CDKN2A-other ALT in each cancer type. Comparison of TMB (F), mutation count (G) and metastasis (H) among four subgroups

TME in TCGA pan-cancer cohort

Pan-cancer immunogenicity can be influenced by genomic mutation. In the TCGA pan-cancer cohort, we used CIBERSORT to assess immune cell infiltration of tumor tissue. NK cells activated, monocytes, Tregs, B cells naive, T cells CD8, T cells CD4 memory resting, and macrophages M2 were lower in CDKN2A-MUT or DEL patients’ tumor tissue than in CDKN2A-WT patients, implying an immune cold tumor immune microenvironment in TCGA pan-cancer with CDKN2A VAR (Fig. 6A). Similarly, CDKN2A-MUT or DEL patients had lower expression of critical immune function genes in most tumor types, which could influence immune regulators in the tumor microenvironment, such as immunoinhibitory molecules, immune stimulators, MHC molecules, chemokines, and chemokine receptors, than CDKN2A-WT patients (Fig. 6B, Figure S5 A and B). These findings suggested that CDKN2A MUT or DEL could result in a cold tumor immune microenvironment and primary resistance to immune checkpoint therapy.

Pathways associated with CDKN2A ALT in TCGA pan-cancer cohort

We evaluated at eight cancer types with high frequencies of CDKN2A VAR and a significant influence on patient outcomes: ESCA (Esophageal carcinoma), KIRC (Kidney renal clear cell carcinoma), KIRP (Kidney renal papillary cell carcinoma), LUAD (Lung adenocarcinoma), MESO (Mesothelioma), pancreatic (PAAD), SARC (Sarcoma) and Stomach adenocarcinoma (STAD). To further assess pathways associated with CDKN2A mutational status, GSEA were implemented on gene sets for all patients with and without CDKN2A ALT on in TCGA above eight cancer types-cohorts. Several pathways of tumorigenesis, development and metastasis, including TNFA signaling via NFKB, MYC targets, mTORC1 signaling, KRAS signaling, G2M checkpoint, Estrogen response, P53 pathway, EMT and E2F targets were upregulated in CDKN2A-MUT or DEL tumors (Fig. 6C). Of note, inflammatory response and IL6 JAK STAT3 signaling were not significantly altered among four groups, while IL2 STAT5 signaling were downregulated in CDKN2A-DEL tumors, and IFN ALPHA response were upregulated in CDKN2A-MUT and DEL tumors (Fig. 6C). Expectedly, hypoxia was significantly downregulated in CDKN2A-DEL tumors than in CDKN2A-WT tumors. Hypoxia can attenuate the function of cytotoxic T cells and attract regulatory T cells to reduce tumor immunogenicity [44].

Fig. 6

Assessment of immune infiltration, immune signatures, and pathway enrichment in CDKN2A-mutated pan-cancer patients in the TCGA cohort. Associations of CDKN2A ALT with immune cells infiltration (A), immune-related genes (B), signaling pathways (C) in TCGA Cohort (D) OS in CDKN2A-DEL, CDKN2A-MUT, CDKN2A-WT and CDKN2A-other ALT patients from Smita Sihag esophagogastric cancer cohort. (E) Protein comparison of p14 ARF (p14) and p16INK4a (p16) in CDKN2A-mutated STAD patients from the TCGA cohort. (F) Western blotting showed the expression of p16 and p14 in stomach cancer cell lines after CDKN2A knockdown

Three pan-cancer cohorts (the MSK MetTropism, TCGA, and OrigiMed 2020 cohorts) found high CDKN2A ALT rates and significant survival effects in esophageal/stomach cancer, and Smita Sihag et al. found that CDKN2A had strong association with poor OS in 487 esophageal/stomach cancer patients with targeted sequencing [45]. Furthermore, utilizing the latter cohort, we found that all patients with CDKN2A MUT (OS: HR, 1.76, 95%CI: 1.31–2.36, Fig. 6D) or CDKN2A DEL (OS: HR, 1.61, 95%CI: 1.08–2.39, Fig. 6D) tended to have poor OS than those with WT. We used the TCGA gastric cancer dataset to compare the expression levels of the p14 ARF (p14) and p16INK4a (p16) proteins in the CDKN2A mutant group and found that the p16 protein was considerably less expressed than the p14 protein (Fig. 6E) [46]. Next, western blotting revealed a downregulation of p16 but no change in p14 expression in CDKN2A knockdown gastric cancer cell lines (AGS and HGC-27, Fig. 6F). Our findings demonstrated that CDKN2A VAR promoted cell proliferation by reducing expression of p16 with function as a regulator of the cell cycle of inhibiting CDK4 and CDK6 in gastric cancer.

scRNA-seq of gastric cancer cells identifies cold immune microenvironment in CDKN2A ALT tissues

Previous studies in gastric cancer have reported that CDKN2A loss confers a cold tumor immune microenvironment [12]. We analyzed the scRNA-seq profiles of 21 primary gastric cancers (n = 59,594 cells), 20 of which were CDKN2A WT and one were CDKN2A-mutant (Missense Mutation) based on WES [34], to determine if CDKN2A ALT in gastric cancer was significantly associated with decreased inflammation. To shed light on cell populations present in gastric tumors, we clustered the 59,594 cells and identified 15 subclusters (Figure S6A) which were subsequently determined to be eight cell subgroups including B cells (MS4A1, CD79A), CD4 T − cells (CD3D, IL7R), CD8 T − cells (GZMA, GZMB, CD3E), endothelial cells (PLVAP, NOTCH3, ENG), epithelial cells (MUC1, KRT8), fibroblasts cells (COL1A1, COL1A2), myeloid cells (CD14, S100A9) and plasma cells (SDC1, XBP1) (Fig. 7A and C). A high proportion of CD8 T cells was observed in patients with CDKN2A WT (39.33%) while a low proportion of CD8 T cells was observed in patients with CDKN2A ALT (1.32%) (Fig. 7B and D). Those CDKN2A other ALT samples from TCGA cohort used in this study were shown to display more CD8 T cells (Fig. 6A). Similarly, three immune-related pathways including interferon alpha response, inflammatory response and interferon gamma response were inhibited in the CDKN2A ALT group (Fig. 7E-G). The mTOR signaling, MYC targets v1 signaling, and TNFA signaling via NFKB signaling were low enriched in the CDKN2A ALT group (Fig. 7H-J).

Fig. 7

Single-cell transcriptional profiling of human GC. (A) scRNA-seq tSNE projection of 59,594 single cells from 21 primary GCs samples. (B) tSNE plot of 59,594 single cells from 21 primary GCs samples, color-coded by group. (C) Dot plot showing the gene expression patterns of cell-type marker genes in the scRNA-seq data. (D) Proportions of cell types in individual samples (above) and in different groups (bottom). scRNA-seq, single-cell RNA sequencing. (E-J) The GSEA plot shows enrichment of several pathways in CDKN2A ALT group. FDR < 0.05 is considered as significantly enriched. (K) tSNE plot of CD8 + T cells from 21 primary GCs samples. (L) The proportion of each CD8+ T cell cluster in CDKN2A-WT and CDKN2A- ALT patients

To understand the characteristics for the tumor cells, we performed re-clustering of CD8 T cells demonstrating nine sub-cell population including cell cycle cells (TUBA1B, HIST1H4C, KIAA0101, STMN1 and TUBB), effector cells (NKG7), effector memory cells (GZMB, ANXA1 and CREM), exhausted cells (CTLA4 and LAG3), intraepithelial cells (LIPF), macrophages cells (HSPA1A, HSPA1B and CXCL8), naïve cells (IL7R), resident memory cells (ANXA1 and GZMB) and Treg cells (IL2RA) (Fig. 7K and Figure S6B). The examination of the gastric cancer scRNA-seq data revealed that the tumors from CDKN2A-ALT patients had more exhausted CD8 T cells than those from CDKN2A WT patients (Fig. 7L). The results demonstrated the presence of a cold immune microenvironment in CDKN2A-ALT tumors, which may account for the resistance to ICIs. Our results highlighted the potential of scRNA-seq data to illuminate previously unknown characteristics of the immune microenvironment in cancers and promote the development of more effective immunotherapies.