Tissue microarray used in this research included a total of 159 GC samples that were purchased from Shanghai Outdo Biotech Co., Ltd. The samples were collected between June 2011 and May 2013, and the follow-up ended on June 2017. After ruling out samples with missing data (3 samples) and distant metastasis (2 samples), 154 samples were included in the present study. Patients received neither chemotherapy nor radiation therapy before surgery. All the samples were validated as gastric adenocarcinoma by the researchers. Clinicopathologic parameters, including sex, age, tumor location, tumor histological grade, tumor size, T stage, N stage, tumor-node-metastasis (TNM) stage, and human epidermal growth factor receptor 2 (HER2) status, were collected. The tumor stage was identified using the eighth version of the TNM staging system of the American Joint Committee on Cancer. The overall survival (OS) time was defined from postoperative to death or to the end of the follow-up. Written informed consents were provided by Shanghai Outdo Biotech Co., Ltd., and the research was conducted in accordance with the recognized ethical guidelines. The studies that involved human participants were reviewed and approved by the Ethics Committee of Shanghai Jiao Tong University School of Medicine Affiliated Ruijin Hospital. The patients provided written informed consent to participate in this study.

Tissue microarray was established, and immunochemistry was conducted following the EnVision two-step procedure of Dako REAL^TM Envision^TM Detection System (Dako, Agilent Technologies, CA, United States) (Ji et al., 2019). In brief, the sections were deparaffinized and rehydrated before being heated in a microwave oven for antigen retrieval. After having been washed with phosphate-buffered saline (PBS) three times, the sections were incubated with 3% hydrogen peroxide to block the endogenous peroxidase activity. Afterward, the sections were incubated with 1 × Animal-Free Blocking Solution (Cell Signaling Technology, United States) for 1 h at room temperature to reduce non-specific staining. After blocking, the sections were incubated with the primary antibodies overnight at 4°C. On the next day, they were incubated with the horseradish peroxidase (HRP)–labeled second antibodies at 37°C for 30 min and then were incubated with diaminobenzidine solution for visualization.

Anti-ARID1A mouse monoclonal antibody (sc-32761; 1:50 dilution; Santa Cruz Biotechnology, United States) was used to detect the expression level of ARID1A protein in GC tissues. ARID1A was located in the cellular nuclear area and was dyed yellow to varying degrees in cancer cells. Using CaseViewer (magnification, × 40; 3DHISTECH Ltd.), the expression status of ARID1A protein was determined by two independent investigators. Loss of ARID1A was identified once the nuclear staining was vacant in more than 10% of cancer cells, and the rest of the cases were defined as “preserved” (Kim et al., 2016; Figures 1A,B).

Anti-CD47 rabbit polyclonal antibody (20305-1-AP; 1:50 dilution; Proteintech Group, United States) was applied to identify CD47 expression level in tissues. The expression of CD47 in GC tissues was evaluated by histochemistry score (H score). H score = Σ(pi × i) = (percentage of weak intensity area × 1) + (percentage of moderate intensity area × 2) + (percentage of strong intensity area × 3), where i is the staining intensity, including negative without staining (i = 0), weak positive (i = 1), medium positive (i = 2), and strong positive (i = 3), and pi is proportion of the positive cells for the corresponding staining intensity (Paschalis et al., 2019). The median of H score was defined as the threshold value to distinguish low and high CD47 expression group. The cutoff value is 179.86 (Figures 1C,D).

GC cell lines (BGC-823, MKN-28, SNU-1, SUN-5, SNU-16, MKN-45, KATO-III, BGC-803, AGS, SGC-7901, and NCI-N87) and GSE-1 cell lines were preserved in the Ruijin Hospital (Shanghai, China) (Wang et al., 2019). Cells were cultured in DMEM (Dulbecco modified eagle medium) or RPMI-1640 medium supplemented with 10% fetal calf serum at 37°C and with 5% CO₂.

The ARID1A lentiviral short hairpin RNAs (shRNAs) were supplied by GenePharma (Shanghai, China). The target sequences are listed in Supplementary Table 1. The shRNAs were cloned into PGMLV-SB3 (PGMLV-hU6-MCS-CMV-Puro-WPRE) vector, and an empty vector was used as the control vector. For lentivirus transduction, cells were transduced following the manufacturer’s instruction and were selected by puromycin (cat. Ant-pr-5; Invivogen) with the concentration of 2 μg/mL. The expression of ARID1A in BGC-823 and MKN-28 was confirmed by western blot and quantitative reverse transcriptase–polymerase chain reaction (qRT-PCR).

Proteins were fractionated by 10% sodium dodecyl sulfate–polyacrylamide gel electrophoresis gels and were transferred to polyvinylidene fluoride (PVDF) membranes. The primary antibodies included ARID1A (12354S; 1:1,000 dilution; Cell Signaling Technology, United States), CD47 (20305-1-AP; 1:1,000 dilution; Proteintech Group, United States), and GAPDH (51332S; 1:1,000, Cell Signaling Technology, United States). PVDF membranes were blocked by 5% skim milk for 2 h at room temperature and then were incubated with primary antibodies overnight at 4°C. The membranes were then incubated with HRP-conjugated secondary antibodies (7074, 7076; 1:2,500, Cell Signaling Technology, United States). ECL western blotting substrate (cat. P10300, NCM Biotech, China) and infrared imaging system (LI-COR Biosciences) were used to visualize the protein bands.

RNA was extracted by Trizol reagent method. Reverse transcription in 20-μL system was performed following the protocol of Fasting gDNA Dispelling RT SuperMix kit (KR118-03; TIANGEN, China). Primers for qRT-PCR were ARID1A: 5′-GCATCCTTCCATGAACCAAT-3′ (forward), 5′-CCCATGCCTGTGTGTATCTG-3′(reverse);GAPDH:5′-GGAC CTGACCTGCCGTCTAG-3′ (forward), 5′-GTAGCCCAGGA TGCCCTTGA-3′(reverse);CD47:5′-AGAAGGTGAAACGATC ATCGAGC-3′ (forward), 5′-CTCATCCATACCACCGGATCT-3′ (reverse). Relative mRNA expression was normalized to GAPDH and calculated by 2^–△CT method. Each experiment was performed in triplicate.

Surface expression of CD47 in GC cells was measured by flow cytometry. In brief, equal amounts of cells were harvested, were Fc-blocked, and were incubated with anti-CD47 antibody (PE) (cat. 12283-MM07-P; SinoBiological) or isotype control for 30 min at room temperature. Afterward, the cells were washed three times and were resuspended in 1% fetal bovine serum/PBS. The fluorescence intensity of CD47 in GC cells was collected on a BD Calibur flow cytometer and was analyzed by FlowJo V10 software.

Chromatin immunoprecipitation (ChIP) assays were performed as previously described (Yu et al., 2018). Briefly, BGC-823 and MKN28 cells were harvested and crosslinked in 1% formaldehyde at room temperature, and the fixation reactions were quenched by adding glycine to a final concentration of 0.125 M. After sonication, we separately incubated the soluble chromatins with the antibodies anti-ARID1A (12354S, Cell Signaling Technology) or control IgG (2729, Cell Signaling Technology). Afterward, chromatin immunocomplexes were precipitated with protein A (Millipore, 16-661). The immunoprecipitated complex was washed, and DNA was extracted and purified by QIAquick PCR Purification Kit (Qiagen). ChIP DNA was analyzed by qPCR using specific primers, and the data were normalized by input DNA. The primers for ChIP-qPCR are the following: human CD47 (5′-AAAGAAGGGGATCCCTAGCA-3′, 5′-CCATCTCCAAATGCACACAC-3′).

Public data from The Cancer Genome Atlas (TCGA), including that of 375 GC patients was included in our analysis. Other cancer data from TCGA including colon adenocarcinoma (COAD), lung adenocarcinoma (LUAD), breast invasive carcinoma (BRCA) and kidney renal clear cell carcinoma (KIRC) was also included in this analysis. TCGA data were downloaded from the Genomic Data Commons data portal,¹ and ARID1A mutation information was acquired from the cBioPortal.² The cutoff value of CD47 in TCGA data sets was determined as median. Tumor-infiltrating immune cells were estimated on the TCGA data by quanTIseq³ (Finotello et al., 2019) and EPIC⁴ (Racle et al., 2017).

Multispectral immunofluorescence staining was performed following the manufacturer’s instructions of Opal 4-Color Manual IHC Kit (NEL794001KT; PerkinElmer, Waltham, MA, United States). The Opal kit uses tyramine signal amplification conjugated fluorophores to detect targets within an immunofluorescence assay. Initially, the slides were baked in the oven at 65°C for 1 h, dewaxed with xylene, and rehydrated through a graded series of ethanol solutions. After rehydration, slides were fixed in 10% neutral buffered formalin for 20 min. The slides were then placed in microwave for 45 s at 100% power followed by additional 15 min at 20% power in AR buffer. The slides were blocked with blocking buffer (PerkinElmer) and incubated in a humidified chamber for 10 min at room temperature. After removal of the blocking buffer, the slides were incubated with primary antibodies overnight at 4°C. Primary antibodies including CD11c (Abcam, ab11029, 1:75), CD163 (Abcam, ab182422, 1:800), and Foxp3 (Abclonal, A12051, 1:50). The slides were sequentially incubated with Opal Polymer HRP MS + Rb (PerkinElmer) and Opal Fluorophore Working Solution (pal 520, Opal 570, Opal 670; 1:50 dilution) for 10 min at room temperature, respectively. Finally, the slides were incubated with DAPI solution (PerkinElmer) for 5 min at room temperature and were mounted with ProLong Diamond Antifade Mountant (Invitrogen). Imaging was performed using Vectra Quantitative Pathology Imaging Systems (PerkinElmer), and image analysis was performed using the InForm Advanced Image Analysis software (inForm 3.0; PerkinElmer).

The data were analyzed using GraphPad Prism (version 8.0.1), statistical product and service solution (SPSS) (version 22), R (version 4.1.0), and Rstudio (version 4.0.4) software. A χ²-test was used to analyze the relationship between clinicopathologic parameters and ARID1A status and CD47 expression level. Survival analysis was performed using the Kaplan–Meier method, log-rank test, and Cox regression analysis under various conditions. Analysis of variance and post hoc test were used to calculate the difference in the means of multiple groups. Non-parametric test was used to analyze the data, which did not follow normal distribution. p < 0.05 was considered statistically significant.

The clinicopathological characteristics of 154 GC patients enrolled in this study are shown in Table 1. In the registered patients, 66.2% of patients were males, and 33.8% were females. In line with a previous study, the incidence rate ratio of male to female was approximately 2:1 in China (Sung et al., 2021). Most patients were 60 years or older (119 cases, 77.3%). For the histological grading, the samples of 66.2% of the patients were poorly differentiated. Regarding T stage, we found that most of the patients were at T4 stage, indicative of the fact that most of the patients were diagnosed at an advanced stage. For the classification according to HER2 status, 116 of patients (75.3%) were found to be negative/weak positive (1 +), which assigned them as HER2 negative, and 9.1% of them were strong positive (3 +), which assigned them as HER2 positive.

Immunohistochemical (IHC) staining was carried out on GC tissue microarrays to evaluate ARID1A (Figures 1A,B) and CD47 expression (Figures 1C,D). According to the evaluation criteria, loss of ARID1A expression was identified in 36 cases of GC tissues (23.4%). The remaining 118 cases (76.6%) exhibited preserved expression of ARID1A. Regarding CD47 expression, 68 of GC tissues (44.2%) were identified as low expression tissues, and 86 of them (55.8%) were identified as high-expression ones. Next, we investigated the correlation between the expression of ARID1A and CD47 in GC tissues. More specifically, in GC tissues, the H score of CD47 was higher in the ARID1A-loss group than in the ARID1A-preserved group (Figure 1E). Previous studies have reported that most ARID1A mutations are inactivated non-sense mutations or frameshift mutations, which result in the loss of ARID1A protein expression (Wang et al., 2004; Wiegand et al., 2010; Wu et al., 2014). Thus, we examined the ARID1A-mutated expression as a surrogate of ARID1A-loss expression in TCGA data set. An analogous trend was also identified in CD47 expression at transcriptional level through analysis of the data from the TCGA. More specifically, the expression of CD47 in ARID1A-mutated group was found to be higher than that of the ARID1A wild-type group (Figure 1F). Furthermore, we found that the mRNA expression level of CD47 was higher in ARID1A-mutated samples than ARID1A wild-type samples in COAD, LUAD, BRCA, and KIRC, respectively (Supplementary Figure 1). Putting side-by-side the information obtained so far, we concluded that loss of ARID1A was significantly correlated with high CD47 expression in GC.

The relationships between ARID1A status, CD47 expression, and clinicopathologic features of GC patients are shown in Table 2. As shown in Table 2, ARID1A status and CD47 expression level were significantly associated with T stage. Notably, in early T stage (T1 and T2) cases, the expression of ARID1A was always preserved, and in the majority of instances, the expression of CD47 was low. These findings in early T stage samples might be explained by the tumor suppressor effect of ARID1A and the lack of CD47-mediated antiphagocytosis effect (Wu and Roberts, 2013; Morrissey et al., 2020). As for TNM stage, the proportion of ARID1A-loss cases increased with the progression of the TNM stage, and the proportion of high CD47 cases appeared to show a similar trend. These might indicate that ARID1A is often lost at the advanced stages of GC. Moreover, we also found that ARID1A status was correlated with tumor size. No significance difference in the expression of ARID1A was found between groups with age, gender, tumor location, histological grade, and N stage. Besides, no significant difference in the expression of CD47 was found between groups with age, gender, tumor location, histological grade, tumor size, N stage, and TNM stage.

To validate the relationship between ARID1A and CD47, we examined the expression of ARID1A in GC cell lines using western blot and qRT-PCR (Figures 2A,B). We found that the expression of ARID1A was depleted in SNU-16, SNU-5, and SNU-1, and these cell lines were poorly differentiated. Cell lines MKN-28 and AGS were more differentiated, and the expression of ARID1A was relatively higher in them. To study the loss of ARID1A in GC cell lines, we chose BGC-823 and MKN28 cells for their relative abundant expression and found that CD47 expression level was increased at the protein and mRNA level after ARID1A knockdown (Figures 2C,D). Besides, flow cytometry analysis showed that the expression of membranous CD47 was also significantly increased after ARID1A knockdown (Figure 2E). Analogous findings were observed using another shRNA. Notably, the extent of the correlation between expressions of ARID1A and CD47 in GC cell lines was the same as that observed in GC tissues, which is indicative of a strong association between loss of ARID1A and a high expression of CD47. To investigate how ARID1A regulates CD47 expression in GC, ChIP-qPCR assay was performed in BGC-823 and MKN28 cell lines, and the results revealed that CD47 was a direct downstream target gene of ARID1A in GC (Figures 2F,G).

To analyze the prognostic value of ARID1A and CD47 in GC, Kaplan–Meier survival analysis and log-rank test were conducted. The endpoint of survival analysis was OS time. As shown in Figure 3A, the low expression of CD47 was correlated with a better outcome for GC patients. Meanwhile, loss of expression of ARID1A was associated with a poorer outcome in such patients (Figure 3B). The mutation of tumor suppressor gene and a change in the immune microenvironment contribute, to a large extent, to tumor progression. Thus, we combined ARID1A status with CD47 expression to investigate the impact on OS of GC patients. Among all cases, cases in the ARID1A^preservedCD47^low expression subgroup had the longest OS time (median OS = 53.9 months) than any other subgroups, whereas the outcome of cases in ARID1A^lossCD47^high expression subgroup was the worst (median OS = 20.4 moths) (Figure 3C). It has been found that HER2 is overexpressed in approximately 20% of GC patients, and anti-HER2 targeted therapy significantly improves the prognosis of HER2-positive GC patients (Oh and Bang, 2020). To evaluate the therapeutic value of ARID1A and CD47, HER2 positivity–stratified analysis was conducted. In HER2-negative cases, loss of ARID1A expression or high CD47 expression indicated a poor outcome (Figures 3F,G). However, analysis did not reveal a statistically significant difference in the outcome of HER2-positive cases (Figures 3D,E). Thus, these results indicate that GC patients with ARID1A^preservedCD47^low might have a better clinical outcome.

Cox regression analysis was performed to find out independent prognostic factors for the OS of GC. Factors including gender, age, tumor size, histological grade, tumor location, TNM stage, ARID1A status, and CD47 expression were taken into multivariate survival analysis. The results showed that CD47 [multivariate analysis: hazard ratio (HR) = 1.705; 95% confidence interval (CI) = 1.118–2.601; p < 0.01] and ARID1A (multivariate analysis: HR = 0.604; 95% CI = 0.374–0.976; p < 0.05) were independent prognostic factors for the OS of GC patients (Figure 4). In addition, histological grade was also identified as an independent prognostic factor for GC patients.

To understand the survival difference in different subgroups stratified by CD47 expression and ARID1A status, we further investigated the tumor-infiltrating immune cell infiltration difference in different subgroups stratified by low/high CD47 expression and mutated/wild-type ARID1A expression by using quanTIseq and EPIC algorithms in the analysis of the data from TCGA. As shown in Figure 5A, higher proportion of M1-polarized macrophages were found in the ARID1A^mutatedCD47^high subgroup than any other subgroups. A previous study showed that high M1-polarized macrophages correlated with a better outcome in GC patients (Zhang et al., 2015). However, a recent study indicated that M1 macrophages have an attenuated prognostic value in CD47^high GC patients (Shi et al., 2021). Thus, the status of M1 macrophage might lose its prognostic value under high CD47 expression condition in GC. Lower proportions of immunosuppressive regulatory T cells (Tregs) were found in the ARID1A^wildtypeCD47^low subgroup than any other. Besides, the infiltration proportion of M2-polarized macrophages in ARID1A^mutatedCD47^high subgroup was significantly higher than that in the ARID1A^wildtypeCD47^low subgroup. In the ARID1A^wildtypeCD47^low subgroup, the proportion of CD8⁺ T cells was higher compared with any other subgroups. No difference was found in CD4⁺ T-cell and B-cell infiltration proportions. To further validate the immune cell infiltration implicated by the computational analyses, we carried out multispectral immunofluorescence assay to measure the infiltration of M1 macrophages, M2 macrophages, and Tregs in GC tissues with ARID1A^lossCD47^high or ARID1A^preservedCD47^low expression. Consistent with the computational analyses, multispectral immunofluorescence assays showed that higher proportions of M1-polarized macrophages, M2-polarized macrophages, and Tregs were found in the ARID1A^lossCD47^high subgroup than in the ARID1A^preservedCD47^low subgroup (Figures 5B,C). Putting side-by-side this information, we concluded that better prognosis of GC patients in ARID1A^preservedCD47^low group might correlate with low Tregs and low M2 macrophages cell infiltration.