Clinical outcome and molecular landscape of patients with ARID1A‐loss gastric cancer

Abstract

Chromatin remodelers are commonly altered in human cancer. The mutation of AT‐rich interactive domain 1A (ARID1A) in gastric cancer (GC), a component of the SWI/SNF chromatin remodeling complex, was proven associated with treatment response in our previous study. However, ARID1A loss of function was caused not only by mutations but also copy number deletions. The clinicopathologic, genomic, and immunophenotypic correlates of ARID1A loss is largely uncharacterized in GC. Here, 819 patients with clinicopathological information and sequencing data or formalin‐fixed paraffin‐embedded tissues from four cohorts, Zhongshan Hospital (ZSHS) cohort (n = 375), The Cancer Genome Atlas (TCGA) cohort (n = 371), Samsung Medical Center (SMC) cohort (n = 53), and ZSHS immunotherapy cohort (n = 20), were enrolled. ARID1A loss was defined by genome sequencing or deficient ARID1A expression by immunohistochemistry. We found that ARID1A mutation and copy number deletion were enriched in GC with microsatellite instability (MSI) and chromosomal‐instability (CIN), respectively. In the TCGA and ZSHS cohorts, only CIN GC with ARID1A loss could benefit from fluorouracil‐based adjuvant chemotherapy. In the SMC and ZSHS immunotherapy cohorts, ARID1A loss exhibited a tendency of superior responsiveness and indicated favorable overall survival after anti‐PD‐1 immunotherapy. ARID1A‐loss tumors demonstrated elevated mutation burden, neoantigen load, and interferon gamma pathway activation. Moreover, in CIN GC, ARID1A loss was correlated with higher homologous recombination deficiency. ARID1A loss defines a distinct subtype of GC characterized by high levels of genome instability, neoantigen formation, and immune activation. These tumors show sensitivity to both chemotherapy and anti‐PD‐1 immunotherapy. This study provides valuable insights for precision treatment strategies in GC.

Our data indicate that ARID1A loss predicted superior benefit from adjuvant chemotherapy (ACT) and anti‐PD‐1 immunotherapy in gastric cancer (GC), especially in chromosomal instability (CIN) subtype, while in non‐CIN subtype, we did not observe any survival benefit from ACT regardless of ARID1A status. ARID1A‐loss GC is associated with elevated mutation burden, neoantigen load, homologous recombination deficiency, and interferon gamma pathway activation. This study may help clinicians stratify patients according to ARID1A status and develop personalized treatment for each molecular subtype in future.

Abbreviations

ACT

adjuvant chemotherapy

ARID1A

AT‐rich interactive domain 1A

CIN

chromosomal instability

CNV

copy number variation

CPS

combined positive score

CR

complete response

DCR

disease control rate

EBV

Epstein–Barr virus

EMT

epithelial–mesenchymal transition

GC

gastric cancer

GS

gnomical stability

HRD

homologous recombination deficiency

ICB

immune checkpoint blockade

IFN

interferon

IHC

immunohistochemistry

mGC

metastatic gastric cancer

MSI

microsatellite instability

ORR

overall response rate

OS

overall survival

Pan‐F TBRS

Pan fibroblast TGFβ response characteristics

PD

progressive disease

PR

partial response

SD

stable disease

SMC

Samsung Medical Center

ssGSEA

single‐sample gene set enrichment analysis

STAD

stomach adenocarcinoma

TCGA

The Cancer Genome Atlas

TIME

tumor immune microenvironment

TMB

tumor mutation burden

ZSHS

Zhongshan Hospital

1. INTRODUCTION

Gastric cancer (GC) is the fifth most common cancer and the third leading cause of cancer‐related death. ¹ Over the past few decades, the clinical and functional significance of major driver genes, including the activation of oncogenes (e.g., PIK3CA and ERBB2) ² , ³ , ⁴ and the loss of function of tumor suppressor genes (e.g., TP53 and CDH1), ⁵ , ⁶ have been well characterized. However, most GC patients lack clear driver mutations, progress rapidly, and face limited treatment options. For better management of GC patients, it is necessary to better understand the clinical and molecular characteristics of GC with specific genetic alterations. ⁷

Chromatin remodeling plays a crucial role in the regulation of gene expression, and aberrations in this process have emerged as possible mechanisms in cancer initiation and development. ⁸ , ⁹ AT‐rich interaction domain 1A (ARID1A), a gene encoding a large nuclear protein member belonging to the SWI/SNF chromatin remodeling complex, is the most prevalent altered chromatin remodeler gene in cancers. ¹⁰ As a recognized tumor suppressor gene, ARID1A plays a crucial role in regulating gene expression that drives oncogenesis or tumor suppression, and loss of ARID1A increased the risk of cancer progression in several cancers. ¹¹ However, ARID1A loss also induced multiple molecular events in cancer, including DNA damage repair, ¹² cell cycle regulation, ¹³ epithelial–mesenchymal transition (EMT), ¹⁴ and so on, which indicated therapeutic vulnerability for these patients. ¹⁵

In GC, our previous study revealed that somatic ARID1A mutation could predict superior response to fluorouracil‐based adjuvant chemotherapy (ACT) and anti‐PD1 immunotherapy. ¹⁶ However, due to the presence of haploinsufficiency, ARID1A loss of function could also be a result of copy number deletion. ¹⁷ , ¹⁸ This indicated that GC patients with either ARID1A point mutation or copy number deletion might refer to the same entity. Thus, in this study, we defined ARID1A loss as either point mutation or copy number deletion and analyzed the clinicopathologic, genomic, and immunophenotypic correlates of ARID1A‐loss GC, which might give us a more comprehensive understanding for the impact of ARID1A alteration on GC and guide precision treatment for these patients.

2. MATERIALS AND METHODS

2.1. Study cohorts

This study recruited overall 819 patients from four independent cohorts, The Cancer Genome Atlas (TCGA) cohort (n = 371), Zhongshan Hospital (ZSHS) cohort (n = 375), Samsung Medical Center (SMC) cohort (n = 53), and ZSHS immunotherapy cohort (n = 20). Clinical and pathological features are presented in Tables S1–S3.

2.2. Patient inclusion

The ZSHS cohort originally consisted of 496 GC patients under the approvement of the Clinical Research Ethics Committee of Zhongshan Hospital, Fudan University (Shanghai, China; approval number: Y2015‐054). In this study, dots on the tissue microarrays for 55 patients were lost during immunohistochemistry (IHC), whereas other 66 patients were excluded for unavailable molecular subtype information, incomplete clinical information, or unknown therapeutic status. Ultimately, 375 patients with basic clinical data and available molecular subtype information were eligible for further analysis. All data were gathered retrospectively, and the survival periods were defined as months after surgery. Patients with ACT were defined as patients with tumor‐node‐metastasis (TNM) II/III advanced tumors who received 5‐fluorouracil‐based ACT. In these patients, 5‐fluorouracil‐based ACT, including combination drug regimens with 5‐fluorouracil (mainly 5‐fluorouracil combined with platinum) or 5‐fluorouracil single‐drug regimen, was administered according to the Chinese Society of Clinical Oncology (CSCO) and National Comprehensive Cancer Network (NCCN) guidelines for GC and patients' wills. ZSHS immunotherapy cohort enrolled 20 GC patients from ZSHS, who were diagnosed with unresectable metastatic GC and received anti‐PD‐1 immunotherapy ± chemotherapy during 2020 and 2022.

2.3. Public data sets

In the TCGA cohort, 371 GC patients were included with clinical information as well as genomic and transcriptomic data downloaded from http://www.cbioportal.org. A total of 69 patients were finally excluded due to lack of ARID1A mutation and copy number variation (CNV) information (n = 6), TCGA molecular subtype information (n = 58), and survival information (n = 5). The SMC cohort was from a single‐center, phase 2 trial, in which 61 patients with metastatic gastric cancer (mGC) were treated with pembrolizumab monotherapy (ClinicalTrials.gov, NCT#02589496). In this study, eight patients were excluded due to incomplete clinical information or unknown genomic data.

2.4. Immunohistochemistry and sequencing data analysis

Immunohistochemistry staining for ARID1A and ARID1A expression evaluation were carried out according to previous protocols. ¹⁶ For the TCGA, SMC, and ZSHS immunotherapy cohorts, ARID1A‐loss GC was defined as patients with ARID1A mutation (missense mutation, nonsense mutation, frame shift deletion, frame shift insert, splice site) or copy number deletion (hemizygous deletion or homozygous deletion). For the ZSHS cohort lacking relevant genomic information, as we previously reported, positive intranuclear staining of ARID1A in‐tumor cells was regarded as normal ARID1A expression (ARID1A‐non‐loss GC), while negative intratumoral ARID1A expression with positive staining of stromal cells was defined as deficient ARID1A expression (ARID1A‐loss GC).

For genomic analyses, driver genes alteration of the oncogenic pathways was computed according to driver landscape of GC described by Yasushi Totoki and colleagues ¹⁹ (genes are listed in Table S4). The OncoKB ²⁰ database was used to determine the potential actionability of gene alteration observed in each sample. Each genomic alteration was stratified into one of four levels of clinical actionability. For transcriptomic analyses, the RNA‐seq data from the TCGA cohort were normalized using the formula log2 (FPKM+1). The infiltration proportion of 28 types of immune cells, immune checkpoint signature, T effector and IFN‐γ pathway signature, antigen processing machinery signature, CD8 T cell effector signature, EMT signature, angiogenesis characteristics signature, pan‐fibroblast TGFβ response characteristics (Pan‐F TBRS) signature were calculated by the single‐sample gene set enrichment analysis (ssGSEA) algorithms based on related gene expression as previously reported ²¹ , ²² (genes are listed in Table S5).

2.5. Statistical analysis

Kaplan–Meier analysis and log‐rank test were conducted for survival analysis. Univariate analyses and multivariate analyses were performed by Cox proportional hazards regression, and hazard ratio and 95% confidence intervals were reported. Pearson's chi‐squared test and Fisher's exact test were applied to compare categorical variables. Statistical p‐values were computed using the Mann–Whitney U test, and detailed statistical tests were described in corresponding figure legends. The statistical analysis was two‐tailed, and p < 0.05 was considered statistically significant. All analyses were conducted using IBM SPSS Statistics v20.0 and R 4.1.2 software.

3. RESULTS

3.1. ARID1A loss associated with prognosis in GC patients

Considering the consensus of the TCGA molecular classification system in GC, ²³ we first aimed to investigate the distribution of ARID1A gene alteration across molecular subtypes. Interestingly, we found distinct aggregation features between ARID1A mutation and copy number deletion. In detail, ARID1A mutation was mainly observed in the microsatellite instability (MSI) subtype (Figure 1A). However, copy number deletion dominantly occurred in the chromosomal instability (CIN) subtype (Figure 1B). When we combined ARID1A mutation and ARID1A copy number deletion into ARID1A loss for analysis, it was found that only the gnomically stable (GS) subtype was largely absent in ARID1A‐loss GC (Figure 1C).

FIGURE 1.

The distribution of ARID1A alteration across molecular subtypes and its association with survival in GC patients. (A) ARID1A mutation types in chromosomal instability (CIN) (n = 220 patients), Epstein–Barr virus (EBV) (n = 30 patients), gnomical stability (GS) (n = 49 patients) and microsatellite instability (MSI) (n = 72 patients) subtypes. ARID: ARID/BRIGHT DNA binding domain (1017–1104), BAF250_C: SWI/SNF‐like complex subunit BAF250/Osa (1975–2231). (B) Bar plots summarizing ARID1A copy number deletion proportion in CIN, EBV, GS, and MSI subtypes. (C) Stacked bar chart illustrating the distribution of the ARID1A loss (n = 202 patients) and ARID1A non‐loss (n = 169 patients) in CIN, EBV, GS, and MSI subtypes. (D, E) Kaplan–Meier curves representing OS of patients with ARID1A loss and ARID1A non‐loss in the TCGA cohort (D) and ZSHS cohort (E); p‐values were obtained using the log‐rank test. p < 0.05 was considered statistically significant. (F) Univariate Cox regression was used to analyze the prognostic value of the ARID1A status across molecular subtypes; the HRs and 95% CIs are reported. p‐values were estimated by means of the two‐sided log‐rank test in the univariate analysis. CI, confidence interval; HR, hazard ratio; TCGA, The Cancer Genome Atlas; ZSHS, Zhongshan Hospital.

We next explored the impact of ARID1A loss on the prognosis of GC. In the TCGA cohort, there was no difference between the ARID1A‐loss and ARID1A‐non‐loss subgroups for overall survival (OS) (p = 0.73; Figure 1D). However, ARID1A loss indicated favorable OS in the ZSHS cohort (p = 0.015; Figure 1E). In view of the differential distribution of ARID1A loss across molecular subtypes, we further analyzed the prognostic value of ARID1A loss in patients with specific GC subtypes. Nevertheless, the presence of ARID1A loss no longer implied a trend toward improved OS in any GC subtypes (Figure 1F).

3.2. ARID1A loss predicted superior adjuvant chemotherapeutic benefit

Although ACT has been the standard treatment for advanced GC after surgery, not all these patients gained survival benefit from adjuvant GC. ²⁴ Therefore, we sought to determine whether ARID1A status was associated with survival benefit from ACT in TNM stage II/III patients. In the TCGA and ZSHS cohorts, it was shown that ACT application had a significantly positive prognostic effect on OS in the ARID1A‐loss subgroups (p < 0.001 and p = 0.017, respectively; Figure 2A,B). Considering that ACT was significantly associated with patient age and tumor stages (Table S6), multivariate Cox proportional hazards regression analysis with combined clinicopathologic variables was carried out. As shown, the OS difference between ARID1A‐loss GC patients with ACT and ARID1A‐loss GC patients without ACT remained highly significant after adjusting for age, TNM classification, grade, and TCGA subtype (TCGA: HR, 0.39, 95% CI, 0.22–0.71, p = 0.002, Table S7; ZSHS: HR, 0.07, 95% CI, 0.02–0.21, p < 0.001, Table S8). However, ACT could no longer improve OS in the ARID1A‐non‐loss subgroups (p = 0.15 and p = 0.064, respectively; Figure 2A,B), which was further validated by multivariate Cox regression (TCGA: HR, 0.60, 95% CI, 0.30–1.19, p = 0.14, Table S7; ZSHS: HR, 0.68, 95% CI, 0.46–1.004, p = 0.053, Table S8). We next classified tumor samples according to four molecular subtypes to further analyze the predictive value of ARID1A loss for survival benefit from ACT in different subtypes of GC. Notably, ARID1A loss could still predict superior survival benefit from ACT in CIN GC, as evidenced significantly by prolonged OS by ACT only in ARID1A‐loss tumors (p = 0.004 and p = 0.015; Figure 2C) but not in ARID1A‐non‐loss tumors (p = 0.37 and p = 0.28; Figure 2C). However, in the non‐CIN subtype, we did not observe any survival benefit from ACT regardless of ARID1A status (Figure 2D,E; Figure S1).

FIGURE 2.

ARID1A loss is associated with superior response to adjuvant chemotherapy (ACT). (A, B) Kaplan–Meier analyses of OS among patients treated with ACT or not in the ARID1A‐loss and ‐non‐loss group in (A) The Cancer Genome Atlas (TCGA) cohort and (B) Zhongshan Hospital, Fudan University (ZSHS) cohort. (C, D) Kaplan–Meier analyses of OS among patients with the chromosomal instability (CIN) subtype treated with ACT or not in the TCGA cohort (C) and ZSHS cohort (D). Kaplan–Meier analyses of OS among patients with the non‐CIN subtype treated with ACT or not in the TCGA cohort (E) and ZSHS cohort (F). The response to ACT analysis was conducted in patients with TNM II/III stage diseases. p‐values were obtained using the log‐rank test.

Taken together, we concluded that ARID1A loss predicted superior benefit from ACT in GC, especially in the CIN subtype. Given the small number of patients in our cohort, the result should be interpreted cautiously, and further studies are needed to validate this finding.

3.3. ARID1A loss as a potential biomarker for immunotherapy response

Immunotherapy has changed the direction of cancer care, and GC is no exception. ²⁵ Thus, we sought to evaluate how ARID1A correlated with immunotherapy responses in GC. In the SMC cohort, we depicted a landscape of the ARID1A status and clinicopathological features in GC patients treated with anti‐PD‐1 immunotherapy (Figure 3A). In ARID1A‐loss tumors, complete response (CR)/confirmed partial response (PR) was achieved in seven patients (41.2%) and five patients (29.4%) had stable disease (SD), resulting in an overall response rate (ORR) of 41.2% and a disease control rate (DCR) of 70.6%. In ARID1A‐non‐loss tumors, six patients achieved (17.6%) CR/PR and thirteen patients (38.2%) had SD, resulting in an ORR of 17.6% and a DCR of 55.8%. The ORR of ARID1A‐loss tumors exhibited a tendency to be higher compared with ARID1A‐non‐loss tumors (41.2% vs. 17.6%, respectively, p = 0.069; Figure 3B). Moreover, ARID1A loss was significantly associated with increased OS (p = 0.04; Figure 3C). As shown in Table S9, ARID1A loss remained an independent determinant of survival after immunotherapy (HR, 0.39; 95% CI, 0.15–0.98; p = 0.044) in a multivariable model adjusting for age, gender, MSI, EBV, and mutational load.

FIGURE 3.

ARID1A loss as a potential predictor of immunotherapy efficacy. (A) The heatmap depicted clinicopathologic characteristics of patients in the SMC cohort (n = 53). (B) Proportion of patients responding to PD‐1 inhibitor immunotherapy: complete response (CR)/partial response (PR) and stable disease (SD)/progressive disease (PD): 41.2%/58.8% in the ARID1A‐loss group and 17.6%/82.4% in the ARID1A‐non‐loss group (p = 0.069, χ ² test). (C) Kaplan–Meier survival curves of OS comparing ARID1A‐loss (n = 17) versus ARID1A‐non‐loss (n = 36) tumors in the SMC cohort. (D, E) Swimmer plot showing clinical response, duration of therapy, and PD‐L1 CPS in the ZSHS immunotherapy cohort. Each lane represents one patient. # Sample size too small to conduct analysis across molecular subtype. (F) Waterfall plot of response to anti‐PD‐1 immunotherapy ± chemotherapy in ARID1A‐loss and ARID1A‐non‐loss gastric cancer (GC). Each bar indicates the percentage of maximum tumor volume change in the sum of target tumor measurement from baseline according to RECIST 1.1 criteria. # P12 was excluded due to a new lesion developed. SMC, Samsung Medical Center; mGC, metastatic gastric cancer.

To further validate these findings, we enrolled another 20 patients with mGC treated by immunotherapy ± chemotherapy between 2020 and 2022 in ZSHS. Consistently, we also found that the DCR of patients with ARID1A loss tumors had a trend to be higher than that of patients with ARID1A‐non‐loss tumors (83.3% vs. 42.9%, respectively) (Figure 3D–F). Unfortunately, the response of immune checkpoint blockade (ICB) could not be assessed for patients with the specific molecular subtype due to the small sample size of patients in each subtype. In conclusion, ARID1A loss might be able to serve as a potential biomarker for better response to immunotherapy in GC.

3.4. Association between ARID1A loss and genomic features

Given the comprehensive impact on therapeutic responsiveness, we next tried to clarify the molecular characteristics in patients with different ARID1A status in GC. Tumor mutational burden (TMB) and MSI have been approved to predict response to checkpoint immunotherapy in several cancers. We found that TMB and MSI differed between the ARID1A‐loss and ARID1A‐non‐loss group, with TMB and MSI being significantly higher in the ARID1A‐loss group (p < 0.001; Figure 4A; p = 0.001; Figure 4B). Compared with the ARID1A‐non‐loss tumors, ARID1A‐loss tumors had a greatly elevated level of DNA damage‐related variables including aneuploidy score and homologous recombination deficiency (HRD) score only in the CIN subtype but not in other subtypes (Figure 4C,D). HRD is characterized by a defect in DNA double‐strand break repair, which is thought to render particularly sensitive to platinum‐based chemotherapies. ARID1A‐loss tumors in the CIN subtype showed the highest HRD score, which implied survival benefit from ACT in these patients.

FIGURE 4.

Comparison of the genomic features of the ARID1A‐loss and ‐non‐loss groups in the TCGA cohort. (A, B) Comparison of tumor mutation burden (TMB) and microsatellite instability (MSI) between ARID1A loss (n = 202) and non‐loss (n = 169). (C, D) Boxplot illustrating the differences in aneuploidy score and homologous recombination defects across the two subgroups and TCGA molecular subtype. (A–D) Statistical analysis by the Mann–Whitney U test. (E) Comparison of gene alteration frequencies (presented as percentages of all samples) between the ARID1A‐loss (n = 202) and ARID1A‐non‐loss (n = 169) groups. Data were analyzed by Pearson's chi‐square test. CNV, copy number variation.

We evaluated the altered frequencies of GC driver genes in eleven oncogenic signaling pathways according to ARID1A status (Figure 4E; Table S4). We found an increased mutation rate in the ARID1A‐loss group compared with ARID1A‐non‐loss tumors. For example, the ARID1A‐loss group harbored 23.8% KMT2D mutation, compared with 9.5% in the ARID1A‐non‐loss group. A similar increased mutation rate was seen in EGFR, ERBB3, KRAS, PTEN, KDM6A, and several other driver genes.

Analysis of CNV in ARID1A‐loss tumors revealed an elevated CNV frequency relative to ARID1A‐non‐loss tumors. Among deleted loci are TP53 (deleted in 45.0% of ARID1A loss vs. 33% of ARID1A non‐loss), SMAD4 (52.0% vs. 40.8%), KDM6A (20.3% vs. 11.2%), CDH1, RNF43, BRCA1, and several other tumor suppressors. Among the amplified loci are those harboring PIK3CA (amplified in 45.0% vs. 24.8%), CCND1 (30.7% vs. 19.5%), ERBB3 (29.2% vs. 19.5%), and RHOA. Overall, CNV perturbing oncogene and tumor suppressor loci were more frequent in ARID1A‐loss tumors than in ARID1A‐non‐loss tumors.

Besides, the alteration frequencies of 12 oncogenic signaling pathways based on ARID1A status were evaluated. Compared with the ARID1A‐non‐loss group, oncogenic signaling pathway alterations in the ARID1A‐loss group were enriched in the receptor tyrosine kinase (RTK)/RAS/phosphatidylinositol 3‐kinase (PI3K)/mitogen‐activated protein kinase (MAPK), WNT, transforming growth factor‐beta (TGFβ), cell cycle and immune evasion pathways (Figure S2A). We further explored the druggability of gene alteration defined in the Oncology Knowledge Base (OncoKB) database. Patients with actionable genes were classified into the ARID1A‐loss and ARID1A‐non‐loss group. Of the total patients, actionable gene alteration was detected more frequently in the ARID1A‐loss group (88.1%) than in the ARID1A‐non‐loss group (61.5%) (Figure S2B).

3.5. ARID1A loss induced an active tumor immune microenvironment (TIME)

In order to better understand the relationship between ARID1A status and composition of the TIME, the infiltration level of 28 different kinds of immune cells in ARID1A‐loss and ARID1A‐non‐loss tumors was analyzed. The results showed that most cells had significant differences between ARID1A‐loss and ‐non‐loss tumors. Abundant infiltration of CD4+ T cell and Th17 cell, were discovered in ARID1A‐loss tumors (p < 0.001 and p < 0.05; Figure 5A). The higher frequency of C2 (IFN‐γ‐dominant) immune phenotype ²⁶ in the ARID1A‐loss subgroup further reinforced immune‐activated tumor microenvironment (TME) in ARID1A‐loss patients (p < 0.001; Figure 5B). Along with the increase of neoantigen burden, ARID1A‐loss patients also presented an enhanced antigen‐presenting ability (p < 0.001; Figure 5C; p < 0.05; Figure 5F). The expression of immune checkpoint‐related signature and T effector and IFN‐γ pathway signature which was associated with response to ICIs was significantly increased in ARID1A‐loss patients (p < 0.001; Figure 5D; p = 0.013; Figure 5E). Moreover, we assessed a set of genes related to specific biological processes identified by Mariathasan et al. ²² In this analysis, high expression of EMT markers, angiogenesis characteristics, and Pan‐F TBRS were found in the ARID1A‐non‐loss group. The CD8 effector and antigen presentation signatures were significantly highly expressed in the ARID1A‐loss group (Figure 5F).

FIGURE 5.

Biological characterization of tumor microenvironment (TME) in the ARID1A‐loss group. (A) Heatmap showing the differential expression of 28 immune cell types in patients of the ARID1A‐loss (n = 202) and ‐non‐loss groups (n = 169). Data were analyzed by the Mann–Whitney U test. (B) Stacked bar chart illustrating the distribution of the immune subtypes in the TCGA cohort and in the two subgroups (the C5 subtype is not represented in TCGA‐STAD). p‐value was derived from the Pearson's chi‐squared test. (C) Boxplot showed an elevation in predicted neoantigens (NeoAgs) in ARID1A‐loss gastric cancer (GC) tumors. (D, E) Boxplot showed representative functional signatures including immune checkpoint signature and T effector and IFN‐γ signature on ARID1A status. C–E Statistical analysis by the Mann–Whitney U test. (F) Differences in stroma‐activated pathways and immune‐related pathways, including the Pan‐F TBRS, EMT, CD8 T effector, antigen presentation, and angiogenesis pathways, between the ARID1A‐loss and ‐non‐loss groups (Wilcoxon test).

4. DISCUSSION

ARID1A loss plays a significant role in GC, as it was a frequently observed mutation in approximately 27% of patients in the TCGA cohort. ²⁷ , ²⁸ Existing research has linked ARID1A deficiency in GC to a hypermutated phenotype and elevated levels of PD‐L1 expression, ²⁹ , ³⁰ potentially influencing patient treatment response. Hence, our study aimed to investigate the clinicopathological, genomic, and immunophenotypic correlations of ARID1A loss in GC and its potential implications for precision treatment.

Our previous investigation has documented ARID1A‐mutant GC as a molecularly distinct subtype exhibiting an immunogenic TME and a wide spectrum of therapeutic targets. ¹⁶ Nevertheless, we have also detected the presence of background noise arising from haploid functionality. In fact, apart from somatic mutations, DNA CNVs have the capacity to influence the expression of both protein‐coding and non‐coding genes. ³¹ , ³² , ³³ This underscores the potential of evaluating copy number deletions as a supplementary approach to gene mutation status assessment, thus enabling the identification of a more clinically meaningful patient cohort. Consequently, we have incorporated CNV detection to more accurately identify GC patients experiencing functional loss of ARID1A. We observed that ARID1A mutations and copy number deletions were enriched in MSI and CIN GC subtypes, respectively, suggesting distinct mechanisms of ARID1A loss in different molecular subtypes of GC. Additionally, ARID1A‐loss GC patients were characterized by a favorable prognosis.

Following the initial investigations, ³⁴ , ³⁵ our subsequent focus was on assessing the impact of ARID1A loss on treatment responses. Notably, within the TCGA and ZSHS cohorts, a significant finding emerged: Patients with CIN GC and ARID1A‐loss exhibited a marked improvement in response to fluorouracil‐based ACT. Compared with our previous interesting finding that ARID1A mutation was associated with an increased clinical benefit from ACT, ¹⁶ this study identified a specific subtype of patients with ARID1A loss that might benefit more from ACT. Additionally, an analysis of the SMC and ZSHS immunotherapy cohorts unveiled the correlation between ARID1A loss and superior responsiveness to anti‐PD‐1 immunotherapy, and patients with ARID1A‐loss tumors appeared to have better OS after immunotherapy. Besides, in a multivariable model adjusting for age, gender, MSI, EBV, and mutational load, ARID1A loss remained an independent determinant of favorable OS, while our previous study did not conduct multivariate analysis to adjust MSI, EBV, and mutational load as confounders. These findings indicate the potential of ARID1A loss as a valuable predictive biomarker for chemotherapy and immunotherapy responses in GC patients.

In‐depth analyses encompassing genomics and immunophenotyping revealed a heightened accumulation of mutations within ARID1A‐loss tumors. Additionally, these analyses unveiled an elevated load of neoantigens and the activation of the IFN‐γ pathway in such tumors. Collectively, these discoveries provide vital insights into the underlying mechanisms that drive the potentiated immune response observed in GC characterized by ARID1A loss.

Moreover, in the context of CIN GC, it was observed that ARID1A loss correlated with an increased aneuploidy score and a deficiency in HR. This suggests that ARID1A‐loss tumors within the CIN subtype may display heightened sensitivity to platinum‐based chemotherapeutic drugs owing to their HRD. These findings offer a plausible rationale for the previously mentioned treatment responsiveness‐related results. ²³ , ³⁶ , ³⁷ They also underscore the potential of ARID1A loss as a predictive marker for treatment response and indicate therapeutic vulnerabilities that could be targeted in GC patients.

Given that a certain proportion of patients with hemizygous deletions of ARID1A cannot be well clarified in the protein level, ³⁸ the inconsistent frequency of ARID1A loss in the TCGA cohort based on genomic sequencing and ZSHS cohort based on immunostaining is understandable. Moreover, it is necessary to further investigate the correlation between ARID1A expression and genetic alterations, including mutations and CNVs, in GC patients.

In conclusion, our study characterizes ARID1A loss as a distinct class of GC with high levels of genome instability, neoantigen presentation, and immune activation. These tumors demonstrate sensitivity to chemotherapy and anti‐PD‐1 immunotherapy. The findings from this study have implications for precision treatment in GC and provide a better understanding of the clinical and molecular characteristics associated with ARID1A alteration. Further validation and exploration of the underlying mechanisms are necessary to translate these findings into clinical practice and improve patient outcomes in GC.

AUTHOR CONTRIBUTIONS

Mengyao Sun: Data curation; investigation; methodology; validation; visualization; writing – original draft; writing – review and editing. Yun Gu: Data curation; formal analysis; investigation; methodology; validation; visualization; writing – original draft; writing – review and editing. Hanji Fang: Methodology; resources. Fei Shao: Methodology; resources. Chao Lin: Methodology; resources. Heng Zhang: Methodology; resources. He Li: Methodology; resources. Hongyong He: Methodology; resources. Ruochen Li: Conceptualization; data curation; funding acquisition; project administration; supervision; validation; visualization; writing – original draft; writing – review and editing. Jieti Wang: Conceptualization; data curation; funding acquisition; investigation; project administration; supervision; validation; visualization; writing – original draft; writing – review and editing. Hao Liu: Conceptualization; data curation; funding acquisition; investigation; project administration; supervision; validation; visualization; writing – original draft; writing – review and editing. Jiejie Xu: Conceptualization; data curation; funding acquisition; investigation; project administration; supervision; validation; visualization; writing – original draft; writing – review and editing.

FUNDING INFORMATION

This study was funded by grants from the National Natural Science Foundation of China (81972219, 82003019, 82103313, 82203201, 82272786, 82273192, 82303966, 82373417), Shanghai Municipal Natural Science Foundation (23ZR1409900), Shanghai Rising‐Star Program (22QA1401700), and Shanghai Sailing Program (21YF1407600). All these study sponsors have no roles in the study design and in the collection, analysis, and interpretation of data.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS STATEMENT

Approval of the research protocol by an Institutional Review Board: The study was approved by the Clinical Research Ethics Committee of Zhongshan Hospital, Fudan University, with the approval number Y2015‐054.

Informed Consent: Written informed consent was obtained from each patient included, and this study was performed in accordance with the Declaration of Helsinki.

Registry and the Registration No. of the study/trial: N/A.

Animal studies: N/A.

Supporting information

Appendix S1

Sun M, Gu Y, Fang H, et al. Clinical outcome and molecular landscape of patients with ARID1A‐loss gastric cancer. Cancer Sci. 2024;115:905‐915. doi: 10.1111/cas.16057

DATA AVAILABILITY STATEMENT

Data and materials generated that are relevant to the results are included in this article. Other data are available from the corresponding author Prof. Xu upon reasonable request.