ARID1A alterations and their clinical significance in cholangiocarcinoma

Background ARID1A is a member of the SWI/SNF chromatin remodeling complex. It functions as a tumor suppressor and several therapeutic targets in ARID1A-mutated cancers are currently under development, including EZH2. A synthetic lethal relationship between ARID1A and EZH2 has been revealed in several tumor entities. Although genomic alterations of ARID1A have been described in various cancers, no study has examined correlations between ARID1A gene mutation and protein expression with clinicopathologic parameters and prognosis, particularly in liver fluke-related cholangiocarcinoma (Ov-CCA). Here, we investigated the clinical significance of ARID1A mutations and protein expression in CCA tissues and determined whether there is a correlation with EZH2 protein expression. Methods We evaluated ARID1A and EZH2 immunoreactivity using immunohistochemistry in 98 Ov-CCA with a wide range of clinicopathological features. Somatic mutations of ARID1A were analyzed using the ICGC sequencing data in 489 of Ov and non Ov-CCA and assessed prognostic values. Results While detecting a loss or reduction of ARID1A expression in 54 cases (55%) in Ov-CCA, ARID1A expression was associated with ARID1A mutations (p < 0.001, adjusted p-value < 0.001). We observed that 12 of 13 tumors (92%) with loss of ARID1A expression had truncating mutations. There were nine of 13 tumors (69%) with loss of ARID1A expression and 25 of 41 tumors (61%) with low ARID1A expression exhibited distant metastasis (p = 0.028, adjusted p-value = 0.168). ARID1A was predominantly mutated in Ov-CCA compared to non Ov-CCA (24% and 14% in Ov-CCA and non Ov-CCA, respectively, p = 0.027). There were 36 of 72 (50%) and 52 of 79 (66%) tumors with ARID1A mutation showed tumor stage IV and T3/T4, respectively. The significant mutual exclusivity and co-occurrence between ARID1A and TP53/KRAS mutations were not found in ICGC cohort. In addition, high EZH2 expression, a potential synthetic lethal target in ARID1A-mutated tumors, was detected in 49 of 98 Ov-CCA (50%). Importantly, neither ARID1A expression nor ARID1A mutations correlated with EZH2 expression in this cohort. Conclusion We found that ARID1A inactivation, by somatic mutation or by loss of expression, frequently occurs in Ov-CCA. Reduction of ARID1A expression and/or somatic mutation was shown to be associated with CCA progression. These findings suggest that ARID1A may serve as a prognostic biomarker, and thus may be a promising therapeutic target for CCA.


INTRODUCTION
Cholangiocarcinoma (CCA) is the second most common liver cancer that develops along the epithelial bile duct, accounting for 10% to 20% of primary liver cancer (Banales et al., 2016). The incidence and mortality rates of CCA have been rising worldwide in the past decade (Saha et al., 2016). The incidence rate of intrahepatic CCA (ICC) reported in the US increased from 0.44 in 1973 to 1.18 in 2012 cases per 100,000 (Saha et al., 2016). In Europe, ICC has increased by 9% from 1996 to 2008, while mortality from ICC increased by around 9% from 1990 to 2008 (Patel, 2001;Bertuccio et al., 2013). Major risk factors of CCA include liver fluke infection, primary sclerosing cholangitis, hepatolithiasis, and choledochal cysts which result in chronic inflammation along the epithelial of bile ducts (Khan, Toledano & Taylor-Robinson, 2008). Based on the endemic area, Opisthorchis viverri-associated CCA (Ov-CCA) has been associated with infestation of Ov. The highest incidence rates of Ov-CCA are in South-East Asia, where endemic areas of liver flukes occur (Sripa & Pairojkul, 2008;Banales et al., 2016), especially in countries lining the Mekong River such as Thailand, Vietnam, and Laos (Sripa et al., 2007). In contrast, the major risk factors of non-liver fluke associated CCA (non Ov-CCA) include primary sclerosing cholangitis and cirrhosis. Currently, surgical resection is the only regular option for treatment. Current 5-year survival rates for CCA after surgery and chemotherapy is around 5% to 15% (Pattanathien et al., 2013;Thunyaharn et al., 2013;Luo et al., 2014;Khuntikeo et al., 2015).
Clinical trials evaluating targeted therapies in unselected CCA populations have shown minimal benefits (Chen et al., 2015). The ASCO guidelines have recommended adjuvant capecitabine as the standard of care for a period of six months following curative resection of biliary tract cancers (Shroff et al., 2019). Recently, cisplatin and gemcitabine have become the recognized reference regimen for first-line treatment in patients with advanced biliary tract cancers (Valle et al., 2017). The median survival of standard chemotherapy using gemcitabine and cisplatin combination remains less than one year (Valle et al., 2010). Thus, finding molecular biomarkers that can be used as targets of therapy and/or predict prognosis in CCA are essential to improve disease management and assist in appropriate therapy.
Growing evidence from molecular genetic studies of CCA has increased our understanding of CCA and has initiated a significant shift towards a more precision medicine-based approach. Previous studies reported relatively high frequencies of potentially actionable mutations in CCA (Ross et al., 2014;Nakamura et al., 2015;Jusakul et al., 2017;Montal et al., 2020). Of note, in the high frequency of somatic mutations in genes associated with chromatin remodeling occurring in CCA (Chan-On et al., 2013;Jiao et al., 2013;Simbolo et al., 2014;Jusakul et al., 2017). Among of these genes, genetic alterations in the ARID1A were detected in 7% to 36% of ICC (Chan-On et al., 2013;Jiao et al., 2013;Simbolo et al., 2014;Churi et al., 2014;Jusakul et al., 2017) and 5% to 12% of extrahepatic CCA (ECC) cases (Chan-On et al., 2013;Simbolo et al., 2014;Churi et al., 2014;Nakamura et al., 2015). ARID1A (also known as BAF250A) encodes a nuclear protein involved in chromatin remodeling. Inactivating mutations in ARID1A have been identified in a wide variety of malignancies (Wiegand et al., 2010;Guichard et al., 2012;Wang et al., 2015), suggesting that it functions as a tumor suppressor. Inactivation of ARID1A is thought to activate cell cycle progression, thereby contributing to uncontrolled cellular proliferation in cancer cells (Ho & Crabtree, 2010). Interestingly, there is interest in developing therapeutic targets in ARID 1 A-mutated cancers, including enhancer of zeste homolog 2 (EZH2) (Alldredge & Eskander, 2017). EZH2 is a histone methyltransferase subunit of a polycomb repressor complex. EZH2 inhibition in ARID1A mutated tumors acts in a synthetically lethal manner to suppress cell growth and promote apoptosis, revealing a unique new therapeutic opportunity (Bitler et al., 2015). Clinical trials of EZH2 inhibitors for advanced solid tumors are ongoing and have shown promise in ARID 1 A-mutated gastric cancer (Alldredge & Eskander, 2017). Thus, ARID1A mutational status or its expression might be a surrogate prognostic predictive biomarker of EZH2 inhibitors.
Although genomic alterations of ARID1A have been described in CCA, no study has determined whether there are correlations between ARID1A gene mutation and protein expression with clinicopathologic parameters and prognosis, particularly in Ov-CCA. In the present study, we analyzed sequencing data from the International Cancer Genome Consortium (ICGC) of 489 tumors and performed immunohistochemical staining for ARID1A in 98 Ov-CCA which were sequenced in ICGC cohort. We evaluated whether ARID1A expression and mutational status could be a prognostic biomarker for Ov-CCA. To study if ARID1A could be a surrogate biomarker for EZH2 inhibitors in CCA, we evaluated EZH2 protein expression in Ov-CCA. The correlation between alterations of ARID1A and EZH2 expression in matched CCA tissues was evaluated for the first time in this study.

CCA tissue and mutational data
Ninety-eight paraffin embedded human CCA tissues and clinical data were obtained from Cholangiocarcinoma Research Institute, Khon Kaen University, Khon Kaen, Thailand. All patients signed consent forms. The study was approved by the Ethic Committee for Human Research, Khon Kaen University (HE611195). The primary tumor at the time of resection was staged according to the 7th AJCC.
For mutational analysis, a total of 489 mutational data of bile duct tumors were obtained from the ICGC data portal (Jusakul et al., 2017).

Immunohistochemistry
The expression and localization of ARID1A and EZH2 in CCA tissues were determined by immunohistochemistry (IHC). The primary antibodies include rabbit polyclonal anti-ARID1A (HPA005456, Sigma-Aldrich, Dorset, UK) and rabbit anti-EZH2 antibody (36-6300, Invitrogen, CA, USA). IHC was performed as previously described (Thanan et al., 2020). Briefly, the paraffin-embedded tissues were de-paraffinized in xylene and rehydrated through descending series of ethanol. Sections were treated with 0.01 M sodium citrate, pH 6.0 for 3 mins in pressure cooker for antigen retrieval. After blocking with 0.3% (v/v) hydrogen peroxide in phosphate buffered saline, the sections were incubated with primary antibody: anti-ARID1A (1:250) or anti-EZH2 antibody (1:250) at 4 • C overnight. The sections were incubated with peroxidase-conjugated Envision TM secondary antibody (DAKO, Glostrup, Denmark) at room temperature for 1 h. The reaction products were visualized using 3, 3 -diaminobenzidine tetrahydrochloride substrate kit (Vector, Laboratories, Inc., Burlingame, CA, USA). The sections were counterstained with Mayer's hematoxylin.
The expression of ARID1A and EZH2 was examined in only bile duct cells in CCA. Localization of cytoplasmic and nuclear staining was scored separately. Tumors were scored positive if tumor cells showed definite nuclear staining and negative if tumor nuclei had no immunoreactivity but other nontumor cells from the same samples showed immunoreactivity. Sections were evaluated using the Immunoreactive score (IR score) and was scored by multiplying of the intensity and frequency of DAB-staining results (Halvorsen et al., 2007). The intensity scored as 0 (negative), 1 (weak), 2 (moderate) to 3 (strong) and proportion of positively stained cells expressed as a percentage categorized as 0 = 0%,1+ = 1 − 10%,2+ = 11 − 50% and 3+ = >50%. The intensity and proportion of stained cells were multiplied to produce the final score between 0 and 9. The median of IR score was used to divide CCA patients into two groups as low and high expression. The cut-off values of nuclear ARID1A and EZH2 expression were 3.7 and 3.5, respectively. The cut-off values of cytoplasmic ARID1A and EZH2 expression were 2.8 and 2.7, respectively.

Statistical analysis
The statistical analysis was carried out using SPSS software (version 19.0). The association between mutational data, protein expression profile and the clinicopathological features of CCA patients were performed using Chi-squared or Fisher's exact test. Adjusted pvalues were calculated using Benjamini-Hochberg correction. The survival analysis was determined using Kaplan-Meier estimate with log-rank test. Statistical significance was considered at p < 0.05.

Decreasing nuclear expression of ARID1A in CCA and its correlation with clinicopathological features
A total of 98 Ov-CCA included in the ICGC cohort (Jusakul et al., 2017), were evaluated for ARID1A expression using IHC. Clinicopathological features are summarized in Table 1. The representative IHC staining of ARID1A in CCA tissue samples is shown in Fig. 1A. Of the total number of cases, 13 (13%) had loss of ARID1A expression, 41 (41%) had low ARID1A expression, and 44 (44%) cases had high ARID1A expression in nuclei. In this study, nuclear expression of ARID1A was not significantly different from matched adjacent normal bile duct. The correlations of nuclear ARID1A protein expression with clinicopathological parameters are shown in Table 2. ARID1A expression tended to associate with distant metastasis. There were 9 of 13 tumors (69%) with loss of ARID1A expression and 25 of 41 tumors (61%) with low ARID1A expression exhibited distant metastasis (p = 0.028, adjusted p-value = 0.168), suggesting that ARID1A may play a role in CCA progression. Of note, ARID1A expression was associated with ARID1A mutations (p < 0.001, adjusted p-value < 0.001, Table 2 and Fig. S1). We observed that 12 of 13 tumors (92%) with loss of ARID1A expression had truncating mutations (nonsense and frameshift insertion/deletion) ( Table 2). In contrast to nuclear expression, there was no correlation between cytoplasmic ARID1A expression and clinicopathological parameters (Table S1). The characteristic of tumors with ARID1A truncating mutations is shown in Table S2. Kaplan-Meier survival (Fig. 1B)

EZH2 expression in CCA and its correlation with clinicopathological features
Several therapeutic targets in ARID1A mutated cancers are in development, including EZH2 inhibitors. EZH2 inhibition in ARID1A mutated tumors acts in a synthetically lethal manner to inhibit cancer progression, revealing a therapeutic opportunity. Since the response to EZH2 inhibitors correlates with EZH2 overexpression, we then investigated protein expression of EZH2 by IHC staining in tumor cells of Ov-CCA (Fig. 3A). Of the total number of cases, 49 of 98 (50%) had low and 49 (50%) cases had high nuclear expression of EZH2. The significant correlation between nuclear and cytoplasmic EZH2 expression and clinicopathological features was not found in Ov-CCA (Table 4 and Table  S1). Additionally, there was no significant difference between groups for low and high expression of EZH2 in overall survival (HR = 0.750, 95% CI [0.491-1.145]) (Fig. 3B). To address if ARID1A alterations could be a predictive biomarker of EZH2 inhibitor, we studied the correlation between EZH2 expression, ARID1A expression and mutation. Neither ARID1A expression nor ARID1A mutations associated with EZH2 expression in our cohort (Table 4).

DISCUSSION
ARID1A is one of the most frequently mutated tumor suppressor genes in various types of cancer (Wei et al., 2014). It has been suggested that ARID1A mutations and its expression carry prognostic significance (Zhang et al., 2018;Simbolo et al., 2018;Bi et al., 2019). Recently, ARID1A has emerged from whole exome and genome studies as one of the significantly mutated gene in CCA (Chan-On et al., 2013;Jusakul et al., 2017). Of note, ARID1A mutations were enriched in liver fluke related CCA. However, the prognostic significance of ARID1A mutation and its expression has yet been explored in this subgroup of CCA. To date, correlation between ARID1A mutations and protein expression in CCA has been explored in the independent studies (Yang et al., 2016;Simbolo et al., 2018;Bi et al., 2019). In this study, we evaluated the clinicopathologic significance of ARID1A protein expression and somatic mutations in the same cohort of Ov-CCA. More importantly, we investigated the correlation between ARID1A expression and ARID1A mutations. We found that decreasing ARID1A immunoreactivity in Ov-CCA and loss of ARID1A was associated with tumor metastasis. Our results suggested the somatic mutations of ARID1A were associated with immunoreactivity. Additionally, expression of EZH2, a potential synthetic lethal target in ARID1A mutated tumors, was also detected in this study.
Regarding the high frequency of ARID1A mutations in Ov-CCA, we then evaluated ARID1A expression specifically in Ov-CCA tumors. Here, there were 50% of Ov-CCA exhibited low expression of ARID1A. In this study, there were 9 of 13 CCA (69%) with loss of ARID1A expression and 25 of 41 CCA (61%) with low ARID1A expression exhibited distant metastasis. Similar correlation between ARID1A expression with invasion was reported in ICC (Yang et al., 2016), suggesting that lower ARID1A protein expression is highly correlated with the invasion and metastasis of CCA. Several reports have demonstrated association between loss of ARID1A protein expression and cancer progression in various cancer (Wang et al., 2012;Wei et al., 2014;Zhang et al., 2018). Reduced ARID1A expression was associated with lymph node metastasis, tumor infiltration, and poor prognosis in patients with gastric carcinoma (Wang et al., 2012;Ashizawa et al., 2019). Similarly, ARID1A protein expression was decreased in patient-derived HCC tumor tissues, and that decreased expression was significantly correlated with lymph node and distant metastasis, and poor prognosis (He et al., 2015).
Previous studies have revealed the relevance of ARID1A mutation or protein loss to survival in several carcinomas (Ashizawa et al., 2019), although the findings were varied. In this study, ARID1A mutation and protein expression were not significantly associated with overall survival of both Ov-CCA and non Ov-CCA. To date, overall survival affected by mutation in ARID1A has been shown in ICC (Simbolo et al., 2018). Low expression of ARID1A protein and mRNA were associated with poor prognosis in 57 ICC (Yang et al., 2016). These finding suggest a prognostic role of ARID1A in ICC, unfortunately the correlation between ARID1A mutation and protein expression was not determined in that studies. In contrast to ICC, there was no correlation between loss of ARID1A expression and overall survival in ECC (Sasaki et al., 2016). We speculate that these different results may arise from the hypothesis that CCA with different etiologies and anatomical sites display profound differences in their major-driven molecular profiles that drive carcinogenesis (Jusakul et al., 2017). In contrast to several studies, Bi et al. (2019) reported that ARID1A was highly expressed in ICC tumor tissues and increased expression of ARID1A was associated with a higher risk of mortality and disease recurrence in ICC patients. The dual roles in both oncogenicity and tumor suppression of ARID1A were demonstrated in several studies (Otto & Kadoch, 2017;Sun et al., 2017) and may contribute to the difference results between studies. Thus, the functional roles of ARID1A needs to be further investigated.
Loss of ARID1A protein correlated with the presence of ARID1A mutations was previously reported in ovarian and uterine endometrioid carcinoma (Wiegand et al., 2010;Guan et al., 2011). All mutations in endometrioid carcinomas were nonsense or insertion/deletion mutations, and there was 73% and 50% of ovarian clear-cell carcinoma and endometrioid carcinoma, respectively, with an ARID1A mutation showed a loss of ARID1A expression (Wiegand et al., 2010). In this study, we found the association between ARID1A expression and mutations. Most of CCAs with loss of ARID1A expression had truncating mutation of ARID1A gene. Of note, we observed half of CCA with ARID1A truncating mutations showed positive protein staining. Our findings are similar to those reported in ovarian clear-cell and endometrioid carcinomas (Wiegand et al., 2010;Guan et al., 2011). It is likely that ARID1A mutation occurred in clones of cells within the tumor, resulting in a heterogeneous staining pattern of ARID1A expression. Moreover, the presence of ARID1A immunoreactivity in tumors positive for ARID1A mutation may indicate that haploinsufficiency is pathogenic, as has been reported in mice (Gao et al., 2008).
Interestingly, we found that there was 50% and 66% of CCA with ARID1A mutation showed tumor stage IV and T3/T4, respectively. Moreover, ARID1A was predominantly mutated in Ov-CCA. These data suggest that ARID1A mutation may involve in CCA progression and the different etiology may be one of the underlying factors that drives CCA heterogeneity. Sasaki et al. (2016) reported that there was no biliary carcinoma harboring both ARID1A and KRAS mutations. However, mutually exclusivity between mutations of ARID1A, TP53 and KRAS was not found in this study.  Novel ways of treating patients with ARID1A mutations have focused largely on using synthetic-lethal approaches. Bitler et al. (2015) highlighted the potential of targeting the antagonistic activity between SWI/SNF and EZH2 methyltransferase with the EZH2 inhibitor, which triggered apoptosis in ARID1A-mutated cells. EZH2 was overexpressed in many solid cancers, suggesting the promise of therapeutic potential of EZH2 inhibitors for cancers (Kim & Roberts, 2016). The response to EZH2 inhibitors often correlates with EZH2 overexpression. There were 50% of Ov-CCA had high expression of EZH2, but the levels of expression did not correlate with patient prognosis. In contrast to our finding, high EZH2 expression was significantly associated with short overall survival in CCA (Wasenang et al., 2019). Neither ARID1A expression nor mutation was correlated with EZH2 expression in our study. Similar to our finding, there was no correlation between ARID1A expression and EZH2 or H3K27me3 amounts in bladder carcinomas. An in vitro study showed that ARID1A-depletion did neither increase EZH2 protein or trimethylated H3K27 levels (Garczyk et al., 2018). These finding do not support ARID1A deficiency as predictive biomarker for EZH2-inhibitor treatment response. Future studies should be conducted to validate these preliminary observations by including ARID1A-mutated and wildtype CCA cells.

CONCLUSIONS
In conclusion, this is the first investigation that showed the correlation between mutations and expressions of ARID1A within the same Ov-CCA cohort. Based on ARID1A protein expression and mutational analysis, we found that ARID1A inactivation, by somatic mutation or by loss of expression, frequently occurs in Ov-CCA. CCA with distant metastasis had lower ARID1A expression than those without distant metastasis. ARID1A mutation may involve in CCA progression and predominantly in CCA tumors with high tumor stage. Importantly, ARID1A protein expression was also correlated with ARID1A mutation, suggesting that loss of ARID1A immunoreactivity might be used as a surrogate marker to detect ARID1A mutations in tissues. To expand the therapeutic portfolio for CCA patients, EZH2 expression was investigated in Ov-CCA. Neither ARID1A mutation nor protein expression correlated with EZH2 expression. Further studies are necessary to determine the role of ARID1A-deficiency in response to EZH2 inhibitor in CCA.