The effects of ARID1A mutations on colorectal cancer and associations with PD‐L1 expression by stromal cells

Abstract Background ARID1A is a component of the SWI/SNF complex, which controls the accessibility of proteins to DNA. ARID1A mutations are frequently observed in colorectal cancers (CRCs) and have been reported to be associated with high mutational burden and tumor PD‐L1 expression in vitro. Aim To clarify the role of ARID1A mutation in CRC. Method and results We used next generation sequencing (NGS) and immunohistochemistry on clinically obtained samples. A total of 201 CRC tissues from Niigata University and Niigata Center Hospital were processed by NGS using the CANCERPLEX panel. Immunohistochemistry for ARID1A, PD‐L1, MLH1, and MSH2 was performed on 66 propensity‐matched (33 microsatellite instability‐high [MSI‐H] and 33 microsatellite‐stable [MSS]) cases among 499 cases from Kyushu University. TCGA data were downloaded from cBioPortal. NGS showed significantly more mutations in ARID1A mutated CRCs (p = 0.01), and the trend was stronger for right‐sided CRCs than left‐sided. TCGA data confirmed these findings (p < 0.01). BRAF V600E and ATM mutations were also found at higher frequencies. Immunohistochemistry showed that 30% of MSI‐H CRCs had ARID1A loss, while this was true in only 6% of MSS CRCs. In both MSI‐H and MSS, PD‐L1 expression by stromal cells was enhanced in the ARID1A‐mutant groups (90% vs 39% in MSI‐H, 100% vs 26% in MSS). Conclusion We found a higher mutational burden in ARID1A‐mutant CRCs, and IHC study showed that ARID1A loss was correlated with high PD‐L1 expression in stromal cells regardless of MSI status. These data support the idea that mutant ARID1A is a potential biomarker for CRCs.


| INTRODUCTION
Despite recent progress in anticancer therapies, colorectal cancer (CRC) remains one of the leading causes of cancer-related death worldwide. 1 Fluorouracil plus either oxaliplatin or irinotecan with biological agents is standard-of-care for patients with progressive CRC, although it has limited efficacy. 1 Since the emergence of immune checkpoint inhibitors (ICIs) that target the programmed death-1 (PD-1)/programmed death-ligand 1 (PD-L1) axis, there has been growing evidence that patients with DNA mismatch repair (dMMR)/microsatellite instability-high (MSI-H) CRC and those with a high tumor mutational burden (TMB) can greatly benefit from ICI treatment. 2,3 However, the overall response rate to ICIs was reported to be only around 36% in recent clinical trials among MSI-H CRC patients. 4,5 This finding highlights the need for new immunotherapy biomarkers beyond MSI status and TMB.
AT-rich interaction domain 1A (ARID1A) is a component of the SWI/SNF chromatin remodeling complex, which controls the accessibility of proteins to DNA. 6 The ARID1A gene is commonly mutated in many types of cancer and is classified as a tumor suppressor. 7 ARID1A is mutated in approximately 10% of CRC patients and is associated with medullary histology, BRAF V600E mutation, and MSI-H status. [8][9][10] Another study found that ARID1A loss or impaired ARID1A binding to MSH2 (a member of the MMR proteins) reduced MMR activity and increased the mutational load and PD-L1 expression of tumor cells. 11 In mice with ARID1A-deficient ovarian cancer, the therapeutic effect of PD-L1 inhibitors was greatly enhanced, suggesting ICI treatment could be beneficial for ARID1A-impaired patients. 11 Additionally, retrospective studies have reported relationships between ARID1A mutations and mutational load and with the immune environment, implying a clinical benefit from ICI in cancers harboring ARID1A mutations. [12][13][14] These findings suggested that ARID1A mutations may not only cooperate with ICI treatment but could also have predictive value for ICI therapy.
In this study, we investigated the mutational status of ARID1A in CRCs using next generation sequencing (NGS) to reveal other cooccurring cancer-related mutations, the number of other mutations, and evaluate the relationship between PD-L1 and ARID1A expression by immunohistochemistry (IHC) to test whether ARID1A-mutant CRCs are more likely to express PD-L1.

| Case selection
We collected two patient cohorts, one for NGS and the other for IHC analysis (Supplementary Figure 1) 2.2 | Next generation sequencing NGS data collection was performed at Niigata University, and the detailed procedures are explained in a previous report. 16 Briefly, gDNA (50-150 ng) was extracted from formalin fixed paraffin embed-

| Immunohistochemistry
We performed IHC using the universal immunoperoxidase polymer method (Envision-kit; Dako-Japan, Tokyo, Japan) for all available cases.
Formalin-fixed, paraffin-embedded tissues were sectioned into 4-μm slices, and then antigen retrieval was performed by boiling the slides in 10 mM sodium citrate (pH 6.0) or Target Retrieval Solution (Dako-Japan).
The primary antibodies and staining conditions are summarized in Supplementary Table 1. Anti-mouse for MLH1 and MSH2 or anti-rabbit for ARID1A and PD-L1 IgG (DAKO-Japan) were used as secondary antibodies. The stained slides were evaluated by Tomohiro Kamori (Figure 1).
The expression of MMR proteins (MLH1 and MSH2) was judged as "loss" when there was a complete absence of nuclear staining in neoplastic cells, while the surrounding non-neoplastic cells consistently showed preserved nuclear staining ( Figure 1A-D).
In our assessment of ARID1A, clear absence of staining in the nuclei of viable tumor tissue (away from necrotic areas) was considered "loss." There were two types of ARID1A staining loss, total and focal loss. Both types were regarded as loss because at least part of the tumor was presumed to have ARID1A mutation ( Figure 1E

| Statistical analysis
We assessed statistical differences between groups using the Mann-Whitney U-test, the chi-squared test, or Fisher's exact test. All calculations were performed using JMP software v13.0 (SAS Institute, Cary, North Carolina). p-Values <0.05 were considered significant.  Table 1. There were 22 patients with ARID1A mutations (10%) similar to previous reports. 9, 19 We did not find any meaningful differences in age, sex, sidedness, tumor differentiation, lymphatic or venous invasion status, or pathological stage between the ARID1A-mutated and wild-type groups. Recently, the laterality of CRCs has been reported to determine their mutational features 24 ; therefore, we divided the patients according to sidedness (Supplementary Tables 2 and 3). This revealed that tumor histological grade was significantly correlated with ARID1A mutation status in patients with right-sided CRC, while this difference was not observed in left-sided CRC patients. The mutational status of ARID1A and mutations that were observed more frequently in ARID1A-mutant cases are listed in Supplementary Tables 4 and 5. ARID1A mutations were scattered throughout the ARID1A locus and there was no hotspot. Patients with ARID1A mutations were likely to have ATM (25%) or BRAF V600E (24%) mutations, which was similar to a previous report. 25 All cases with both BRAF  sequenced by CANCERPLEX. This trend was also confirmed with TCGA database, which included 276 cases of whole exome sequencing ( Figure 2B). In TCGA cohort, the mean number of mutations in

ARID1A-mutant CRCs was over 4-fold greater than the number in
ARID1A wild-type CRCs. This feature was also affected by sidedness, as shown in Figure 2C,D. Only right-sided CRCs with ARID1A mutations showed high mutational load compared with wild-type CRCs.
Left-sided CRCs with ARID1A mutations also showed a slight tendency to have more mutations than wild-type CRCs, although the difference was not significant. In summary, NGS analysis showed that right-sided CRCs with ARID1A mutations were likely to have a distinct mutational signature (BRAF V600E and ATM mutations) and a higher TMB, which was compatible with previous results from in vitro studies.

| DISCUSSION
We used two different cohorts to explore the impacts of ARID1A mutations on CRC and found that ARID1A-mutated CRCs tended to have higher mutational loads and that ARID1A-deficient CRCs were more likely to be accompanied by enhanced PD-L1 expression by stromal cells. We used the CANCERPLEX panel, which can detect as many as 415 cancer-related genes, to analyze the NGS group. This revealed two primary conclusions. One was that CRCs with ARID1A mutations were likely to co-occur with BRAF V600E and ATM mutations, which are both clinically targetable with molecular therapies. 25 The association between ARID1A mutations and the BRAF V600E mutation has already been reported, 8  ATM is also a well-known oncogene that regulates cell proliferation and DNA double strand break repair. 26 The second conclusion from this analysis was that ARID1Amutant CRCs had a greater number of mutations compared with ARID1A-wild type CRCs, and this relationship was stronger in rightsided CRCs. We examined the number of co-occurring mutations rather than mutational burden due to the lack of mutational burden data in our cohort. This is not a standard evaluation, and is one of the limitations of our study. However, we confirmed the same trend in TCGA cohort, and there are other reports in a large cohort study with over 40 000 cases that ARID1A-mutated cancers including CRCs had higher mutational burdens than cancers without ARID1A mutations. 12,14 These data also reinforced the finding of higher mutation numbers in the ARID1A-mutant CRCs in our cohort. High TMB is a good predictor for tumors that express high amounts of neoantigens, which recruit inflammatory cells including CD8-positive T lymphocytes into the tumor microenvironment. 3 These lymphocytes interact with the tumor cells via secreting IFNγ, and the tumor cells express PD-L1 in turn to deactivate the lymphocytes and escape immune reactions. 27 These findings were so critical to our study that we further determined an IHC evaluation of ARID1A and PD-L1 expression

ACKNOWLEDGMENTS
We thank all of the technical staff of the Department of Pathology and the Department of Surgery and Science, Kyushu University for their assistance. We also thank James P. Mahaffey, PhD, from Edanz Group (https://en-author-services.edanzgroup.com/) for editing a draft of this manuscript.

DATA AVAILABILITY STATEMENT
The result of Figure 2B is in whole based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.
The other data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

ETHICS STATEMENT
This study was approved by the institutional review board of Kyushu University Fukuoka and Niigata University Niigata, Japan (permission number: 30-367) and conformed to the tenets of the Declaration of Helsinki. Informed consent was waived by the institutional review board owing to the retrospective nature of the study.