The ARID1A gene encoding AT-Rich Interactive Domain-containing protein 1A is known as a member of the switching/sucrose non-fermentable (SWI/SNF) complex involved in chromatin remodeling.1 Mutations in and loss of the ARID1A gene mostly lead to its inactivation and ARID1A protein loss.2 Certain types of cancer, including clear cell ovarian carcinoma (46%–50%), gastric adenocarcinoma (10%–35%), and cholangiocarcinoma (15%–27%), frequently harbor ARID1A alterations.2–4 To date, clinical and preclinical data indicate that ARID1A alterations may sensitize tumors to drugs targeting the ataxia telangiectasia and Rad3-related (ATR) protein, the enhancer of zeste 2 (EZH2), or the phosphatidylinositol-3-kinase (PI3K) pathway,5–10 but no therapies targeting ARID1A alterations have been approved. Importantly, Shen et al demonstrated that ARID1A alterations interact with the mismatch repair (MMR) protein MSH2 and, hence, compromise MMR.3 Tumors formed by an ARID1A-deficient ovarian cancer cell line in syngeneic mice exhibited higher mutation load, as well as increased numbers of tumor-infiltrating lymphocytes and elevated programmed cell death-ligand 1 (PD-L1) expression. Furthermore, administration of anti-PD-L1 antibody decreased cancer burden and extended survival of mice bearing ARID1A-deficient but not ARID1A wild-type ovarian tumors.3 Interestingly, alterations in the polybromo-1 (PBRM1) gene, which is another member of the SWI/SNF complex, have been reported to correlate with salutary effects in cancer patients receiving checkpoint blockade inhibitors, though the clinical evidence remains controversial.11 12 In gastric cancers, ARID1A alterations are associated with Epstein-Barr virus infection, which is in turn associated with checkpoint blockade response.13 Herein, for the first time to our knowledge, we investigated the clinical correlation between ARID1A alterations and treatment benefit after anti-programmed cell death-1 (PD-1)/PD-L1 immunotherapy in the human pan-cancer setting.

Starting with 3403 eligible patients who underwent tissue DNA NGS, we found 1540 patients with nine types of cancer diagnoses that had >5% prevalence of characterized ARID1A alterations in tissue DNA NGS (figure 1A and online supplementary figure 1). Of 161 patients with ≥1 characterized ARID1A alteration in diverse types of cancer, 142 had ARID1A substitution or frameshift alterations, while the remaining 19 had insertions, deletions, allelic loss, rearrangement, or truncation. Endometrial and gastroesophageal cancers were the tumor types in which ARID1A alterations were most frequent—49% and 20% of cases, respectively (figure 1A). The median number of genomic coalterations among tumors with ARID1A alterations was 6 (range, 1–72) (not including ARID1A alterations), which was significantly higher than the median of 4 alterations (range, 0–61) among those cancers with wild-type ARID1A (p<0.001). The rate of MSI-high was significantly higher in tumors with ARID1A alterations than in those with wild-type ARID1A (20% vs 0.9%; p<0.001) and in multiple individual tumor types as well (eg, MSI-high in ARID1A-altered vs wild-type endometrial cancer, 41% vs 0%, p=0.001) (figure 1B). Similarly, TMB-high (≥20 mutations/mb) was more often observed in tumors with ARID1A alterations than in those with wild-type ARID1A (26% vs 8.4%; p<0.001) and in individual tumor types (eg, endometrial cancer, 35% vs 0%, p=0.001) (figure 1C).

Overall, 375 patients (24%) among the 1540 patients with the nine types of cancer with >5% ARID1A alterations received anti-PD-1/PD-L1 immunotherapy in the advanced/metastatic disease setting (see online supplementary figure 1). MSI-high and TMB-high were seen in 4.3% (n=16) and 17% (n=65) of these 375 patients, respectively. As shown in figure 2A, patients with ARID1A-altered tumors showed a significantly longer PFS than those with the wild-type tumors (10.9 months vs 3.9 months, p=0.006) from the start of anti-PD-1/PD-L1 immunotherapy. When PFS was analyzed according to cancer diagnosis (only tumor types with ≥5 patients with ARID1A alterations), similar sensitivity was observed in individual tumor types (eg, colorectal cancer (5.2 months vs 2.1 months, p=0.005); endometrial cancer (4.6 months vs 3.0 months, p=0.02)) (see online supplementary figure 2). Importantly, even when only patients without MSI-high were included to the analysis, ARID1A-altered tumors showed a significantly longer PFS than those with wild-type tumors: HR (95% CI), 0.62 (0.40 to 0.97); p=0.03 (figure 2B). In the same way, when only patients without TMB-high were included to the analysis, patients with ARID1A-altered tumors (vs ARID1A wild-type) showed a trend towards longer PFS: HR (95% CI), 0.69 (0.43 to 1.08) although not statistically significant (p=0.10) (see online supplementary figure 3) (small numbers of patients precluded analysis of patients with MSI-high or TMB-high who had ARID1A alterations vs not). When examining OS in ARID1A-altered versus the wild-type patients, median OS time was longer in the ARID1A-altered group (30.8 months vs 20 months), but this did not reach statistical significance (p=0.13) (see online supplementary figure 4). In order to better determine if the correlation between ARID1A alterations and longer PFS was independent of specific confounding variables, we performed a multivariate analysis (patient characteristics of ARID1A-altered vs wild-type patients are shown in table 1). Our Cox-regression model demonstrated that ARID1A alterations were selected as an independent predictor of better outcome (PFS) after anti-PD-1/PD-L1 immunotherapy (HR (95% CI), 0.61 (0.40 to 0.94); p=0.03) (table 2).

In conclusion, 28% of ARID1A-altered tumors (n=32 of 114 patients whose microsatellite and TMB status were both available) had either MSI-high or TMB-high (or both), and the rate of MSI-high and TMB-high was significantly higher in ARID1A-altered versus wild-type tumors. These findings are consistent with previous reports that ARID1A deficiency is correlated with MMR deficiency.3 19 ARID1A alterations were independently and significantly associated with longer PFS after anti-PD-1/PD-L1 immunotherapy (regardless of microsatellite and TMB status). This study has several limitations such as the small number of patients with each cancer type, which restricted our ability to analyze individual tumor histologies. Nevertheless, the results suggest generalizability across tumor types. Another limitation was that improvement in OS in ARID1A-altered patients (vs wild-type) did not reach statistical significance; larger numbers of patients are needed to validate this endpoint. Therefore, ARID1A alterations may be a genomic marker of checkpoint blockade sensitivity, in addition to other putative markers such as MSI-high and TMB-high.20–22 Our observations indicate that ARID1A alterations warrant further studies with longer follow-up and larger numbers of patients in order to confirm if they can be added to the armamentarium of genomic markers that are exploitable for matching patients to immunotherapy in the pan-cancer setting.23 24