Differences in the molecular profile of endometrial cancers from British White and British South Asian women

Objectives To identify differences in the mutational profile of endometrial tumours between British White (BW) and South Asian (BSA) women. Methods We analysed primary tumours from matched cohorts of British White (BW) and British South Asian (BSA) women resident in Leicestershire diagnosed with EC. Next Generation Sequencing was performed to investigate mutational differences in a panel of 10 genes previously identified as being commonly mutated in EC. The presence of somatic Mismatch Repair (MMR) gene deficiencies was determined by immunohistochemistry. Results In total, 57 tumours (27 BSA and 30 BW) were sequenced. There was no significant difference in the overall mutation frequency of the 10 genes analysed; however, numerous differences were observed between the groups. There was a positive association between PIK3CA and PTEN mutations in the BSA group, with 78% of PIK3CA-mutant tumours harbouring a PTEN mutation, whereas only 11% of PIK3CA wild-type (wt) tumours were PTEN mutant positive (p = 0.0012). In BW women, 90% of ARID1A mutant tumours had co-existent PI3K pathway mutations versus 50% of wild-type (wt) ARID1A patients (p = 0.0485). This trend was not significant in the BSA group (p = 0.66). The age at diagnosis was significantly higher in the BW group with a somatic MMR gene deficiency compared to those with no deficiency (72.8 years versus 59.6 years, p = 0.007), whereas this difference was not seen in the BSA group (64 years versus 60 years, p = 0.37). Conclusion We have identified differences in the mutational profile of primary EC tumours from BW and BSA women. Further research is needed to confirm these findings and to explore their potential implications for early detection, treatment response and prognosis.


Introduction
Internationally endometrial cancer (EC) is the second most common gynaecological malignancy with an estimated 382,069 new cases and 89,929 deaths worldwide in 2018 [1]. The incidence of endometrial cancer (EC) is typically higher in high-income countries, as compared to low-income countries however, the picture is changing with India in particular having the highest annual increase in EC incidence internationally between 2005 and 2007, a rise of 13% [2].
The issue of investigating ancestry and genetics in research is challenging due to the concept of race being confused or influenced by the interaction of environment and culture, as well as heterogeneity within populations [3]. Although it has been proposed that the use of race as a surrogate marker for measurable genetic differences should be avoided [4], it is acknowledged that there is utility when investigating the interplay of genes and environmental factors [5] and with a "biological correlate relevant to the disease" [6]. Much research has already been conducted in to the racial/ethnic differences in EC, predominantly from the USA, focusing on Black or African American (BoAA) and Caucasian women. A lower incidence of EC in BoAA women has been reported, as compared to Caucasian women [7][8][9], however, with a significantly worse prognosis [10]. Although, inequality in healthcare has been proposed as one reason for this difference it does not explain why BoAA women have greater propensity for developing serous subtypes as compared to other racial groups. The TCGA database has been utilised to look into this further and distinct molecular groupings have been identified in the EC tumours from BoAA and Caucasian women [11]. This supports the view that the clinical differences seen between BoAA and Caucasian women are due to underlying genetic differences, with a higher rate of TP53 mutations and ERBB2 amplification in tumours from BoAA women compared to tumours from Caucasian women [12,13], whereas the opposite is true for the frequency of PTEN mutations [14].
Limited evidence regarding the mutational landscape of EC exists for other geographic populations, with one of the least studied groups being Asian women. The categorisation of 'Asian' heritage in the medical literature is fraught with challenges and misconceptions, with Asia being the descriptive term for a large geographical area composed of many different cultural and environmental situations, which may share very few similarities. Therefore using the collective term Asian for women for the purposes of analysis could give misleading results and emphasises the need to be highly descriptive and avoid broad categories [15]. It could help to explain reported differences between data from the USA where the incidence of EC in Asian residents was found to be 40% less compared to the Caucasian population [8], whereas a UK study comparing White British with South Asian women living in the UK showed no difference (incidence rate ratio 0.90 vs 1, CI 0.81-1.01) [16]. One finding consistently reported however, is that the age at diagnosis is significantly lower in Asians as compared to Caucasian populations [17,18].
The aim of our study was to address this lack of evidence by investigating the mutational profile of genes commonly associated in the pathogenesis of EC in primary tumour specimens from women from two groups resident in Leicestershire: British White (BW) and Asian/Asian British-Indian/Pakistani, which we will refer to collectively as British South Asian (BSA). BSA is an accepted term that refers to a person whose ancestry originates in the Indian subcontinent, and who identifies with, or is identified with, their host country [19].

Materials and methods
Ethical approval for the study was granted by Yorkshire and The Humber-Bradford Leeds Research Ethics Committee (REC reference number: 15/YH/0510). Women diagnosed with EC at the University Hospitals of Leicester between February 2016 and February 2018 were recruited. Informed written consent was obtained prior to participation and all underwent primary surgery, including hysterectomy. Participants' ethnicity was self-designated using the National Office of Statistics for the decennial UK National Census classification.
Sixty of the patients recruited (30 BW and 30 BSA) who had confirmed histological diagnosis of EC were selected after being matched in terms of risk for adjuvant therapy using the ESGO-ESMO-ESTRO classification [20]. Based on this consensus document, patients with EC were categorised based on their risk of developing a recurrence to low, intermediate, highintermediate, high, advanced and metastatic groups using FIGO stage, grade, histology, myometrial invasion, lymphovascular space involvement and residual disease following surgery. In total, 15 'low-risk' and 15 intermediate/high-risk tumours were analysed from both the BW and BSA groups.

DNA extraction
Formalin-Fixed, Paraffin Embedded (FFPE) blocks of the primary tumour were obtained. Following Haematoxylin-Eosin staining (H&E), a Consultant Pathologist marked a high density area of tumour and a 1.5mm core biopsy was performed. DNA was extracted using the QIAamp Generead DNA FFPE Tissue Kit (Qiagen) as per manufacturer's instructions.

Targeted Next Generation Sequencing (tNGS)
The Ion AmpliSeq designer software (https://www.ampliseq.com) was used to generate a custom 190 amplicon Ion AmpliSeq panel for analysis of hotspot regions in 10 genes commonly mutated in EC (ARID1A, CTNNB1, FBXW7, KRAS, PIK3CA, PIK3R1, POLE, PTEN, PPP2R1A, TP53) based on publicly available databases (https://portal.gdc.cancer.gov/) [21]. The panel was designed to contain two pools, against FFPE tissue, with amplicons ranging from 125-175 bp. Initial library preparation was performed using 20 ng of FFPE DNA. Samples were sequenced on a 318 chip using the Ion PGM Sequencer according to manufacturers' instructions. Sequencing data was accessed and analysed through the Torrent Suite v5.6.0. All mutations with a quality score below 100 were omitted and all variants detected in the first or last 10 bases of an amplicon were omitted as likely mispriming events [22]. ANNOVAR [23] was used to annotate all mutations with refGene ID, functional consequence (e.g., non-synonymous), and functional predictions using SIFT [24], Polyphen-2 [25] and MutationTaster [26]. Known germline variants without pathological consequence were omitted from the analysis. All variants detected were also manually confirmed across all samples using the Integrated Genomics Viewer 2.3 [27,28]. The mutant allele frequency (termed AF) was calculated as the proportion of total reads at a site, which contained the variant allele (e.g. if you have 200 reads in total and 8 of them have the variant, then the AF is (8/200) x100 = 4%). TIER 1 and TIER2 drivers were identified using the cancer genome interpreter for SNVs, INDELs, and SCNA, with potential treatments analysed using the web-based analysis function at https://www. cancergenomeinterpreter.org.

Immunohistochemistry
Immunohistochemistry was conducted using a Dako Autostainer in order to determine the presence of a somatic mutation within the four mismatch repair (MMR) genes associated with EC (MSH2, MSH6, MLH1 and PMS2). The 5μM sections were incubated with the following primary antibodies; Monoclonal Mouse MLH1 (clone ES05), Monoclonal Mouse MSH2 (clone FE11), Monoclonal Rabbit MSH6 (clone EP49) and Monoclonal Rabbit PMS2 (clone EP51), all obtained from Dako (Santa Clara, CA 95051, USA). Antigen retrieval was performed in a Dako PTlink using a high pH target retrieval solution. Following a 30 minutes incubation with the primary antibodies, visualisation was carried out using the Dako enVision Flex + detection system for MSH6 and PMS2, with a mouse linker for MLH1 and MSH2. On-slide control tissue was used, along with internal control of background endometrium. Results were independently assessed by two pathologists.

Statistical analysis
Chi-square and Fisher's exact tests were used to compare categorical variables. Analysis was two-sided and differences were considered to be statistically significant at p<0.05. All analyses were performed using GraphPad Prism software, version 7.0 (GraphPad Software, Inc, CA, USA).

Results
The mean body mass index (BMI) was lower in the BSA compared to the BW groups (32.5kg/ m 2 versus 37.8kg/m 2 , p = 0.0253), however the incidence of diabetes was more than double, 44% versus 20% respectively (p = 0.08) (

Gene mutations
There were a total of 129 non-synonymous mutations in the 10 genes analysed, a mean of 2.3 per case (Table 2). For full details of each mutation per patient, see S1 Table. A few associations were seen across the whole cohort, in particular the presence of TP53 mutations and older age at diagnosis (72.5 vs 60.5 years, p = 0.008). A similar association was seen between PIK3CA mutations and age (65.9 vs 59.3 years, p = 0.0245). No differences were identified in the incidence of non-synonymous mutations in these ten genes between the two groups although differences were seen in associations. In the BW cases, ARID1A mutations co-existed with PI3K pathway mutations (defined as mutations in either PIK3CA, PIK3R1 or PTEN), as 90% (9/10) of ARID1A mutation carriers had a concurrent PI3K pathway mutation versus 50% (10/20) of wild-type (wt) ARID1A patients (p = 0.0485). This was not seen in the BSA cases (67% versus 52%, p = 0.66). Additional analysis was performed, including logistic regression, to adjust for BMI, grade and histology and this did not alter the result.
In the BSA cases, somatic CTNNB1 mutations tended to be younger at diagnosis (mean age 55.6 years compared to 63.5 for their wild type counterparts, p = 0.077), a finding not seen in the BW cohort (63.3 years for somatic CTNNB1 mutation carriers compared to 62.1 for wt-CTNNB1, p = 0.85). 92% of the CTNNB1 mutations (12/13) identified in both groups were found in exon 3, from codon 32 to 41 and all of them were missense. The majority of these alterations (67%, 8/12) affected Serine or Threonine residues (S33, S37 and T41). Ninety-three percent (13/14) of FBXW7 mutations detected were missense as were all six KRAS mutations, with three due to a substitution of glutamine to histidine (Q61H), a mutation mainly reported in colorectal malignancies [29]. Interestingly, all four KRAS-mutant tumours in the BW group had co-existent PI3K pathway mutations versus only 38% (10/26) of wt-KRAS cases (p = 0.0365). Only one BSA case contained a KRAS mutation, and therefore no comparison was possible.
Across the whole cohort, there was a total of 30 PIK3CA mutations, all missense, of which 10 were located in the p85-binding domain of the gene, with 4 of the 10 being a gain-of-function R88Q mutation (all identified in BW group). Another common mutation was the E545 in the helical domain (13% of the total, n = 4, two BW and two BSA). There was a strong, positive association between PIK3CA and PTEN mutations in the BSA, but not BW, group in which 78% (7/9) of PIK3CA-mutant tumours were also carrying a PTEN mutation versus only 11% (2/18) for wt-PIK3CA tumours (p = 0.0012). Interestingly, PIK3CA-mutant patients in the BW group were significantly older at diagnosis (mean age 68.7 years versus 58 for wt-PIK3CA, p = 0.0077) but this was not seen in BSA group.
PTEN mutations in the BW group were more likely to localize in the phosphatase catalytic domain of the gene (53%, 8/15), whereas, in the BSA group the majority of PTEN mutations was observed in the C2 domain of the gene (64%, 7/11). These differences however did not reach significance (p = 0.168).
As BMI was the only variable significantly different between the BW and BSA groups further analysis was performed dividing the whole cohort into BMI <35kg/m 2 and >35kg/m 2 . There was no significant difference in the number of mutations in the ten genes tested nor in the rates of MMR deficiency in these two groups. The only association identified was PIK3CA mutations co-existed more frequently with PTEN mutations in the <35kg/m 2 BMI group (9/ 13, 69% of PIK3CA-mutant patients had co-existent PTEN mutations compared to 4/16, 25% of wt-PIK3CA, p = 0.027), which was not the case in the >35 kg/m 2 BMI cohort (4/8, 50% versus 5/20, 25% respectively, p = 0.37).

Tumour Mismatch Repair (MMR) protein deficiency
A quarter of tumours (14/57) exhibited a deficiency in at least one of the MMR genes (Table 3). There was no difference in the presence of mutations between the BW and BSA group (20% vs 30%, p = 0.54), with the most common mutation MLH1/PMS2 being identified in 4 cases (67%) in the BW group and 7 cases (88%) in the BSA group. MSH6 mutations were identified in one BW and one BSA case and a MSH2/MSH6 mutation was only seen in one BW case. The age at diagnosis was significantly higher in the BW group with a somatic MMR gene deficiency compared to those with no deficiency (72.8 years versus 59.6 years, p = 0.007), whereas this difference was not seen in the BSA group (64 years versus 60 years, p = 0.37). Examining associations between MMR gene deficiencies and the non-synonymous mutations described previously showed that within the BSA group, there was a negative association between MMR deficiency and CTNNB1 mutations as none of the eight CTNNB1-mutant

Discussion
In this study we have explored the mutational landscape of EC in a previously under investigated population. Leicester is an ethnically diverse region with 28.3% of the Leicester population identifying as 'Indian' in the 2011 National census (n = 93,335) [30]. By selecting a population living in the same region, being treated within the 'free at the point of need' NHS we have attempted to address the differences that may result from access to health care and geographical location, although it was not possible to determine the duration of residence in the UK of the women in the BSA group. The BSA population is of interest since the leading risk factors for EC are obesity and diabetes, both of which have been shown to have different thresholds in the BSA as compared to White Europeans [31]. BSA women have lower obesity cut-off points for the development of dysglycemia, 21.5kg/m 2 compared to 30kg/m 2 for White European women, which could impact on prevention strategies and identifying women at risk of EC. Unlike previous epidemiological studies on EC in Asian women, our BSA cohort was very homogeneous, the majority of women were of Guajarati Indian heritage [32], and despite the sample size being relatively small, it is larger than the paradigm shifting TCGA EC study [21], which included only 20 Asian patients, with no other geographical definitions. The 10 genes of interest were selected based on their frequency in EC, for which there is a wealth of evidence [33], and their potential significance in guiding personalised therapy. We have shown a strong co-existence of PIK3CA and PTEN mutations in the BSA group, which was not seen in the BW group. The correlation between PIK3CA and PTEN has been reported previously in EC in a cohort of cases from Japan [34] and might have clinical implications since a large study of a variety of solid tumours (mainly colorectal and ovarian) showed that patients with PIK3CA and/or PTEN mutations had a significantly greater partial response to PI3K/mTOR inhibition [35] compared to wt-tumours of these two genes. Our results therefore raise the possibility of greater efficacy of these treatments in BSA, as compared to BW, populations.
Previous studies have shown an association between ARID1A and mutations in the PI3K pathway with ARID1A mutations independently increasing the downstream activation of PI3K pathway [36]. We have confirmed this finding in the BW but not in the BSA group, with BW cases being more likely to 'need' ARID1A suppression for PI3K pathway activation whereas cases in the BSA group were more likely to have an activated PI3K pathway due to coexistence of PIK3CA and PTEN mutations. Also, it has recently been shown that ARID1A mutations can render cancer cells sensitive to glutathione and glutamate-cysteine ligase synthetase catalytic subunit (GCLC) inhibitors [37], while they also increase the sensitivity of cancer cells to PI3K inhibitors [38]. It can therefore be proposed that combining PI3K with GCLC inhibitors could be an effective strategy for the 30% of endometrial cancers carrying an ARID1A mutation. Our data therefore support the need for a personalised approach to identify the presence of mutations in different populations before embarking on trials investigating novel drug combinations.
Studies in ovarian clear cell carcinoma (OCCC) have shown higher prevalence of the disease in East Asia compared to North America and Europe [39]. Since one of the most frequent mutations in OCCC is ARID1A, higher rates of such genetic alterations in the BSA cohort could be expected. On the contrary, we have shown no difference in the rate of ARID1A mutations between the two groups, which is consistent with our previous analysis of TCGA cohort (11). One possible explanation is that many epidemiological studies in OCCC were conducted in Japan, a very different phylogenetic and environmental population as compared to South Asia.
Identifying the MMR status is becoming increasingly important in EC since it guides screening for potential Lynch Syndrome cases but also drug therapy options, for example checkpoint inhibitors such as Pembrolizumab. Our reported rate of somatic MMR mutations (25%) is very similar to recent reports in the literature [40,41]. There was no difference in the prevalence of MMR deficiency in the two groups however, the higher age at diagnosis in the BW group supported the view that MMR testing should be universal rather than focused on the younger EC population. The association between older age at diagnosis and MMR deficiency has been previously shown in studies on colorectal cancer and confirms our findings [42,43]. Our study reports a trend associating obesity with decreased rates of MMR deficiency (mean BMI 36.5 for MMR proficient tumours compared to 31.5 for MMR deficient, p = 0.066), and although this trend is not significant, this association has been reported in both colorectal and endometrial cancer studies [42,44]. Reasons for this association are not clear, although previous studies in colorectal cancer had shown that hyperestrogenemia reduces the risk of MMR deficiency, and this may explain the higher incidence of MSI in thin and elderly patients [45]. With aging, estrogen receptors (ER) in colonic cells hypermethylate, leading to reduced ER expression, which subsequently increases the carcinogenic potential of the colonic cells [46]. It is unknown as to whether the same mechanism occurs in endometrial cells and could be the focus of future studies.
There was a negative association between CTNNB1 mutations and MMR deficiency in our cohort, which supports evidence from previous larger study that showed that CTNNB1 mutations were less frequent in MSI positive tumours [47], although the mechanism accounting for this negative correlation has yet to be elucidated.
The main limitation of our study was the relatively small number of cases analysed. The sample size meant that it was not possible to adjust for all clinico-pathological factors and it is possible that significant associations have not been identified due to lack of statistical power. The likelihood of the relatedness of the BSA, although possible, is unlikely given Leicester's large South Asian population and a similar possibility could be raised for the BW population.
In summary, we have identified differences in the molecular profile of ten genes associated with EC in tumours from BW and BSA women. Further research is needed to confirm these findings in a larger population and to explore their potential implications for treatment response and prognosis.
Supporting information S1 Table. (XLSX) supported this study and the Centre for Black and Minority Ethnic (BME) Health, University of Leicester for reviewing the manuscript.