For determination of the alteration frequency of TET1 among cancer types, all the genomic data from the curated set of non-redundant studies on the cBioPortal (https://www.cbioportal.org) [15, 16] were collected and processed as shown in Additional file 1: Figure S1. Tumors with nonsynonymous somatic mutations in the coding region of TET1 was defined as TET1-mutant (TET1-MUT), while tumors without as TET1-wildtype (TET1-WT) [7].

To evaluate the predictive function of specific mutated genes to ICI treatment, a discovery cohort with annotated response and mutational data of patients receiving ICI treatment from six published studies [17–22] was collected and consolidated (Fig. 1a). Samples of the first two cohorts [17, 18] (n = 280) were sequenced using MSK-IMPACT panel, a U.S. Food and Drug Administration (FDA) authorized comprehensive genomic profiling panel. While whole-exome sequencing (WES) was applied to samples of the latter four cohorts [19–22] (n = 249), which were previously curated and filtered by excluding records without response data and records without qualified mutational data by Miao et al. [22]. Based on Miao et al.’s efforts, we further excluded records of cancer type with patients less than 10 (n = 3) and patients receiving concurrent therapy besides ICIs (n = 7). In the end, 519 patients from five cancer types were included in the discovery cohort. An expanded ICI-treated cohort from Samstein et al. [23] without response data but with survival data was used as the validation cohort to further validated the predictive function of TET1-MUT to ICI treatment (Fig. 1b). The non-ICI-treated cohort from Samstein et al. [23] was also included to investigate the possibility that the observed survival differences among patients with TET1-MUT and TET1-WT might simply be attributable to a general prognostic benefit of TET1-MUT, unrelated to ICIs. As the non-ICI-treated cohort from Samstein et al. mainly consisted of patients with advanced disease, the Cancer Genome Altas (TCGA) cohort consisting of 20 cancer types with adequate survival information as determined by Liu et al. [24] was additionally employed to investigate the prognostic impact of TET1-MUT.Fig1Fig. 1

Flowchart of the clinical cohort consolidation. a. Consolidation of the discovery cohort from six published studies. b. Consolidation of the validation cohort and the non-ICI-treated cohort from Samstein et al. ICI, immune checkpoint inhibitor; HNSCC, head and neck squamous cell carcinoma; NSCLC, non-small-cell lung cancer; IRB, Institutional Review Board

TMB was defined as the total number of nonsynonymous somatic, coding, base substitution, and indel mutations per megabase (Mb) of genome examined [25]. For samples sequenced by WES, the total number of nonsynonymous mutations were normalized by megabases covered at adequate depth to detect variants with 80% power using MuTect given estimated tumor purity, as determined by Miao et al. [22]. For samples sequenced by MSK-IMPACT panel, the total number of nonsynonymous mutations identified was normalized to the exonic coverage of the MSK-IMPACT panel (0.98, 1.06, and 1.22 Mb in the 341-, 410-, and 468-gene panels, respectively), and mutations in driver oncogenes were not excluded from the analysis as Samstein et al. proposed [23]. As previously described, the cutoff of the top 20% within each histology was used to divided patients into TMB-high and TMB-low groups [23].

The primary clinical outcomes were ORR, DCB, PFS, and OS. ORR was assessed using Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. DCB was classified as durable clinical benefit (DCB; complete response [CR]/partial response [PR] or stable disease [SD] that lasted > 6 months) or no durable benefit (NDB, progression of disease [PD] or SD that lasted ≤6 months) [18]. Patients who had not progressed and were censored before 6 months of follow-up were considered not evaluable (NE). PFS was assessed from the date the patient began immunotherapy to the date of progression or death of any cause. Patients who had not progressed were censored at the date of their last scan. In the discovery cohort and validation cohort, OS was calculated from the start date of ICI treatment, and patients who did not die were censored at the date of last contact. Notably, in the non-ICI-treated cohort from Samstein et al., OS was calculated from the date of first infusional chemotherapy, while in the TCGA cohort, OS was calculated from the date of first diagnosis.

To characterize the tumor immune microenvironment of TET1-MUT tumors, we further compared the TMB, neoantigen load, tumor-infiltrating leukocytes, immune signatures and immune-related gene expressions between TET1-MUT and TET1-WT tumors in the TCGA dataset. Somatic mutational data from Ellrott et al. [26], neoantigen data from Thorsson et al. [27], and RNA-seq data from UCSC Xena data portal (https://xenabrowser.net) were collected and processed as shown in Additional file 2: Figure S2. TMB was retained as the total number of somatic nonsynonymous mutation count without normalization, and neoantigen load was defined as the total predicted neoantigen count as determined by Thorsson et al. [27]. The R package MCP-counter was used to estimate the abundance of tumor-infiltrating leukocytes [28]. Seven classical immune signatures were adopted from Rooney et al. [29], and the R package GSVA was used to determine the single sample gene set enrichment (ssGSEA) scores of each immune signature [30]. Immune-related genes and their functional classification were obtained from Thorsson et al. [27], whose expression level was quantified as FPKM (Fragments Per Kilobase of exon model per Million mapped fragments) and log2-transformed.

Assessment of enrichment of specific mutated genes with response (CR/PR versus PD/SD) was done with fisher’s exact test and the P values were controlled for false discovery rate (FDR). The association between TET1 status and ORR or DCB were examined using fisher’s exact test. The progression-free and overall survival probability of TET1-MUT and TET1-WT patients were analyzed by Kaplan–Meier method, log-rank test, and Cox proportional hazards regression analysis, which were adjusted for available confounding factors, including 1) age, sex, cancer type, and TMB in the discovery cohort; 2) age, sex, cancer type, TMB status, and MSI status in the validation cohort; 3) sex, cancer type, TMB status in the non-ICI-treated cohort from Samstein et al.; and 4) age, sex, ethnicity, cancer type, histology grade, tumor stage, TMB, and year of first diagnosis in the TCGA cohort. Interactions between the TET1 status and the following factors were assessed, including age (≤ 60 versus > 60 years), sex (male versus female), cancer type (melanoma, bladder cancer, NSCLC versus other cancers), TMB status (low versus high) and drug class (monotherapy versus combination therapy). The differences of TMB, neoantigen load, tumor-infiltrating leukocytes, immune signatures, or immune-related gene expressions between TET1-MUT and TET1-WT tumors were examined using Mann-Whitney U test. The nominal level of significance was set at 0.05 and all statistical tests were two-sided. Statistical analyses were performed using R v. 3.5.2 (http://www.r-project.org).

As shown in Fig. 1a, mutational data with annotated response data were pooled from six publicly available studies to form the discovery cohort, which included 519 patients across five cancer types: bladder cancer (n = 27), esophagogastric cancer (n = 40), head and neck squamous cell carcinoma (HNSCC) (n = 10), melanoma (n = 148), and NSCLC (n = 294). Patient characteristics of the discovery cohort were summarized in Table 1. Particularly, more than half (61.7%) of patients were treated with PD-(L)1 monotherapy, representing its predominant role in immunotherapy. Twenty-one key genes involving in the regulation of DNA methylation, including DNA methyltransferase DNMT1, DNMT3A, DMNT3B, and DNA demethylase TET1, TET2, TET3, and other mediators, were manually collected from previous studies [13, 14] (Additional file 3: Table S1) and investigated. Within these genes, TET1-MUT was significantly enriched in patients responding to ICI treatment (Fig 2a) (P = 0.003), indicating that TET1-MUT may be a potential predictive biomarker for ICI treatment.Tab1

Patient characteristics of the discovery cohort

There were 23 TET1-MUT patients, accounting for 4.4% of the population in the discovery cohort (Table 1). We further investigated the alteration frequency of TET1 across multiple cancer types with genomic data collected from cBioportal. After data assembling, 32,568 patients from 39 cancer types were included in the analysis (Additional file 1: Figure S1). The somatic mutations of TET1 were evenly distributed (Fig. 2b), without any annotated functional hotspot mutations from 3D Hotspots [31] (https://www.3dhotspots.org). The average alteration frequency of TET1 was 2.4% among these 39 cancer types, 22 of which had an alteration frequency above 1%. Skin, lung, gastrointestinal tract and urogenital system were among the most frequently affected organs (Fig. 2b).Fig2Fig. 2

TET1-MUT was enriched in patients responding to ICI treatment. a. Response-associated mutations in CR/PR versus PD/SD (two-tailed Fisher’s exact test, n = 126 patients with CR/PR, n = 389 patients with PD/SD). The dashed grey line indicated false discovery rate adjusted P = 0.05 (Fisher’s exact test). b. The proportion of TET1-MUT tumors identified for each cancer type with alteration frequency above 1%. Numbers above the barplot indicated the alteration frequency, numbers close to cancer names indicated the number of TET1-MUT patients and the total number of patients. ‘CNS tumor’ denoted tumor in the central nervous system. ‘Truncating mutations’ included nonsense, nonstop, splice site mutations, and frameshift insertion and deletion; ‘non-truncating mutations’ included missense mutations and inframe insertion and deletion

The baseline patient characteristics according to TET1 status were shown in Additional file 4: Table S2, and no significant differences were observed between TET1-MUT and TET1-WT patients except for TMB. According to RECIST version 1.1, the best overall response of four patients was not evaluable. In the remaining 515 patients, 23 patients were TET1-MUT while 492 patients were TET1-WT. The ORR of patients with TET1-MUT was more than 2.5-fold higher than that of patients with TET1-WT (Fig. 3a, 60.9% versus 22.8%, odds ratio = 5.26 [95% confidence interval (CI), 2.06 to 14.16], P < 0.001). As for DCB, 71.4% (15/21) of patients with TET1-MUT derived DCB from ICI treatment; while only 31.6% (150/474) of patients with TET1-WT did (Fig. 3b, odds ratio = 5.38 [95% CI, 1.93 to 17.27], P < 0.001). As expected, longer PFS, adjusted by age, sex, cancer type and TMB, of patients with TET1-MUT was observed (Fig. 3c, hazard ratio [HR] = 0.46 [95% CI, 0.25 to 0.82], adjusted P = 0.008). The median PFS was 13.32 months (95% CI, 9.10 to not reached) in the TET1-MUT group versus 3.49 months (95% CI, 2.99 to 4.05) in the TET1-WT group. The OS benefit from ICI treatment was also more prominent in the TET1-MUT group than that in the TET1-WT group (Fig. 3d, median OS, 26.4 months [95% CI, 20.3 to not reached] in the TET1-MUT group versus 15.0 months [95% CI, 13.0 to 18.2] in the TET1-WT group). However, after adjusted for age, sex, cancer type, and TMB, there was only numerically significant OS benefit (HR = 0.54 [95% CI, 0.27 to 1.11], adjusted P = 0.095), probably due to the limited sample size of the TET1-MUT group (n = 22).Fig3Fig. 3

Association of TET1 status and clinical outcomes in the discovery cohort. a. Histogram depicting proportions of patients achieved objective response (ORR) in TET1-MUT and TET1-WT groups. (n = 515; 4 patients with best overall response not evaluable). b. Histogram depicting proportions of patients derived durable clinical benefit (DCB) in TET1-MUT and TET1-WT groups. (n = 495; 24 patients with durable clinical benefit not evaluable). c. Kaplan-Meier estimates of progression-free survival (PFS) in the discovery cohort comparing patients with TET1-MUT with their respective WT counterparts. (n = 519). d. Kaplan-Meier estimates of overall survival (OS) in the discovery cohort comparing patients with TET1-MUT with their respective WT counterparts. (n = 490; 29 patients with no OS information available)

To further validate the predictive function of TET1-MUT on OS benefit, an expanded ICI-treated cohort (n = 1395) was gathered (Fig. 1b). In this validation cohort, the OS benefit was still more prominent in the TET1-MUT group than that in the TET1-WT group (Fig. 4a, the median OS was not reached in the TET1-MUT group versus 19.0 months [95% CI, 16.0 to 22.0] in the TET1-WT group). Even after adjusted for confounding factors, including age, sex, cancer type, MSI status and TMB status, TET1-MUT still independently predicted favorable OS outcomes (Fig. 4a, HR = 0.47 [95% CI, 0.25 to 0.88], adjusted P = 0.019). In patients with known MSI status (n = 1172), 47 of them were MSI-H while 40 were TET1-MUT, and only 7 patients were both MSI-H and TET1-MUT (Fig. 4b). Notably, in patients with low microsatellite instability (MSI-L) or microsatellite stable (MSS), TET1-MUT could still identify patients whose OS was significantly longer than that of TET1-WT patients (Fig. 4c, HR = 0.43 [95% CI, 0.20 to 0.92], adjusted P = 0.030), and nearly equal to that of MSI-H patients (Fig. 4c), indicating that TET1-MUT was compatible and comparable with MSI-H as predictive biomarkers. The favorable clinical outcomes for TET1-MUT versus TET1-WT were also prominent and consistent across subgroups of age, sex, cancer type, TMB status and drug class (Fig. 5, all P _interaction > 0.05).Fig4Fig. 4

Validation of the predictive function of TET1-MUT. a. Kaplan-Meier curves comparing the overall survival (OS) of patients with TET1-MUT and patients with TET1-WT in the validation cohort. b. Venn diagram showing the concomitant presence of MSI-H and TET1-MUT within patients with known MSI status of the validation cohort. c. Kaplan-Meier curves comparing the OS in the MSI-H, MSI-L/MSS and TET1-MUT, and MSI-L/MSS and TET1-WT groups in the validation cohort. d. Kaplan-Meier curves investigating the prognostic impact of TET1-MUT in the non-ICI-treated cohort from Samstein et al.

To characterize the tumor immune microenvironment of TET1-MUT tumors, we compared the tumor immunogenicity and anti-tumor immunity between TET1-MUT and TET1-WT tumors. The TMB level was significantly higher in TET1-MUT tumors compared with that in the TET1-WT tumors, both in the Samstein’s cohort (Fig. 6a, left panel, P < 0.001) and in the TCGA cohort (Fig. 6a, right panel, P < 0.001). Accordantly, the neoantigen load was also significantly higher in TET1-MUT tumors (Fig. 6b, P < 0.001), indicating that TET1-MUT was associated with enhanced tumor immunogenicity.Fig6Fig. 6

TET1-MUT was associated with enhanced tumor immunogenicity and activated anti-tumor immunity. a. Comparison of tumor mutational burden between the TET1-MUT and TET1-WT tumors in the Samstein’s cohort (left panel) and the TCGA cohort (right panel). b. Comparison of neoantigen load between the TET1-MUT and TET1-WT tumors in the TCGA cohort. c. Heatmap depicting the log2-transformed fold change in the mean tumor-infiltrating leukocytes MCP-counter scores of the TET1-MUT tumors compared to TET1-WT tumors across different cancer types. d. Heatmap depicting the log2-transformed fold change in the mean immune signature ssGSEA scores of the TET1-MUT tumors compared to TET1-WT tumors across different cancer types. e. Heatmap depicting the mean differences in immune-related gene mRNA expressions between TET1-MUT and TET1-WT tumors across different cancer types. The x-axis of the heatmap indicated different cancer types and the y-axis indicated tumor-infiltrating leukocytes, immune signatures, or gene names. Each square represented the fold change or difference of each indicated tumor-infiltrating leukocyte, immune signature, or immune-related gene between TET1-MUT and TET1-WT tumors in each cancer type. Red indicated up-regulation while blue indicated down-regulation