Variants in Notch signalling pathway genes, PSEN1 and MAML2, predict overall survival in Chinese patients with epithelial ovarian cancer

Abstract To identify genetic variants in Notch signalling pathway genes that may predict survival of Han Chinese patients with epithelial ovarian cancer (EOC), we analysed a total of 1273 single nucleotide polymorphisms (SNPs) within 75 Notch genes in 480 patients from a published EOC genomewide association study (GWAS). We found that PSEN1 rs165934 and MAML2 rs76032516 were associated with overall survival (OS) of patients by multivariate Cox proportional hazards regression analysis. Specifically, the PSEN1 rs165934 AA genotype was associated with a poorer survival (adjusted hazards ratio [adjHR] = 1.41, 95% CI = 1.07‐1.84, and P = .014), compared with the CC + CA genotype, while MAML2 rs76032516 AA + AC genotypes were associated with a poorer survival (adjHR = 1.58, 95% CI = 1.16‐2.14, P = .004), compared with the CC genotype. The combined analysis of these two SNPs revealed that the death risk increased as the number of unfavourable genotypes increased in a dose‐dependent manner (P trend < .001). Additionally, the expression quantitative trait loci analysis revealed that the SNP rs165932 in the rs165934 LD block (r 2 = .946) was associated with expression levels of PSEN1, which might be responsible for the observed association with SNP rs165934. The associations of PSEN1 rs165934 and MAML2 rs76032516 of the Notch signalling pathway genes with OS in Chinese EOC patients are novel findings, which need to be validated in other large and independent studies.

with OS in Chinese EOC patients are novel findings, which need to be validated in other large and independent studies.  4,5 The genomewide association study (GWAS) approach has also been used to identify susceptibility loci associated with clinical outcomes in European EOC patients. 6 Recently, the availability of published GWAS data has promoted the association analysis for risk and outcomes of cancer using a hypothesis-driven pathway approach. 7,8 The evolutionarily conserved Notch signalling pathway has been identified to be critical in regulating cell differentiation, proliferation, apoptosis and cell-cell communication. 9,10 The Notch signalling pathway consists of four Notch receptors (Notch1-4) and five ligands (Delta-like1, Delta-like3, Delta-like4, Jagged1 and Jagged2) in mammals. [11][12][13][14] The initiation of signalling is the combination of Notch ligands with their receptors, which could in turn affect downstream effectors through releasing the intracellular domain of the receptor that is activated by a cascade of proteolytic cleavages mediated by γ-secretase. For example, it has been reported that NOTCH3 participates in the pathogenesis of EOC recurrence through enhancing carboplatin resistance of cancer cells. 15 A high NOTCH4 mRNA expression level was revealed to be significantly correlated with a favourable overall survival (OS) of EOC patients and thus was regarded as a prognostic factor. 16 As the γ-secretase inhibitors in the Notch signalling pathway, DAPT and MK-0752 have been reported to be highly promising therapeutic drug targets for treatment of EOC patients. 17,18 Recently, a 10-gene signature of the Notch signalling pathway, including FZD3, HES1, PSEN2, JAG2, PPARG, FOS, HEY1, CDC16, MFNG and EP300, has been identified to be associated with a higher risk of recurrence of EOC. 19 Therefore, the Notch signalling pathway has been suggested to have biological significance and important value in the prognosis of EOC.
The effects of genetic variants in Notch signalling pathway genes on survival of cancer patients have been identified in several cancer types, including cutaneous melanoma, 20 non-small-cell lung cancer 7 and hepatocellular cancer. 21 In addition, the deregulation of Notch signalling pathway genes is also involved in the development of platinum resistance and recurrence of EOC. 22 Therefore, we hypothesize that genetic variants in Notch signalling pathway genes contribute to death risk of EOC. To test the hypothesis, we investigated the role of genetic variants of 76 genes in the Notch signalling pathway on OS of EOC patients using available genotyping data from a previously published GWAS data set in a single cancer centre. 23

| Study population
The recruitment of participants has been described previously. 23

| Genotyping data and quality control
According to the databases of Molecular Signatures (MsigDB, http:// www.broadinstitute.org/gsea/msigdb/search.jsp), 76 Notch signalling pathway genes located on the autosomes with their ±2-Kb flanking regions (hg19.) were selected for the analysis. As a result, a total of 1740 single nucleotide polymorphisms (SNPs) were genotyped and included in the analysis.
The genotyped data of the GWAS study were generated by the Illumina HumanOmni Zhonghua-8 BreadChip, and the detailed genotyping information was described previously. 21 Systematic quality control (QC) was applied to the raw genotyping data before the analysis, and the exclusion criteria of loci were as follows: (i) without mapping to autosomal chromosomes; (ii) with a low call rate in GWAS samples (<95%); (iii) with MAF < 0.05; and (iv) with Hardy-Weinberg equilibrium P < 1 × 10 −5 . After QC, a total of 1273 SNPs within these 75 genes were available from the GWAS genotyping data for the final analysis (Table S1).

| False-positive report probability
False-positive report probability (FPRP) is the probability of falsepositive association between genetic variants and disease under the given statistically significant findings. In brief, three factors could account for the value of FPRP: the assumed prior probability, an observed P value, and statistical power to detect the hazards ratio (HR) of the alternative hypothesis at the given P value. For the results of all the selected SNPs, we assigned a prior probability of .01 to detect an HR of 2.0 for an association with genotypes of each SNP. In addition, BFDP (the Bayesian false-discovery probability) was used with a cut-off value of 0.75 as recommended, 24 because BFDP shares the ease of calculation of the recently proposed falsepositive report probability (FPRP) but uses more information, has a noteworthy threshold defined naturally in terms of the costs of false-discovery and non-discovery, and has a sound methodological foundation. In a multiple testing situation, BFDP is straightforward to estimate the expected numbers of false discoveries and false nondiscoveries. The difference between FPRP and BFDP is that FPRP uses the observed significance region, whereas BFDP used the q value that has a fixed region, which allows the false-discovery rate (FDR) to be controlled, a property not inherited by FPRP. 24 Therefore, we assumed that SNPs with FPRP values <0.2 or BFDP values <0.75 were considered statistically noteworthy for the identified significant associations.

| Statistical analysis
The main analysis of the data was to evaluate the associations between genetic variants of 75 genes in the Notch signalling pathway and OS, and the latter was defined as the time interval between the date of histological diagnosis and the date of last follow-up or time of death. The analysis of OS was conducted for patients who were known to have a minimum of cytoreductive surgery for the first-line treatment followed by chemotherapy. For the subgroups of patients known to have received standard chemotherapy (≥4 cycles of paclitaxel and platinum at 3-weekly intervals) after surgery, they were hereafter referred to as having the "standard chemotherapy"; and for the subgroups of patients known to have received chemotherapy without definite cycles or less than 4 cycles, they were hereafter referred to as having the "general chemotherapy." The strength of associations between genetic variants and OS of   All reported P values were two-sided, and P < .05 was considered statistically significant.

| Patients' characteristics and associations with survivals
A total of 480 EOC patients were eligible and included in the final analysis after removal of those without survival data. The study analysis flow chart is shown in Figure 1. in an additive genetic model (Table S2). Finally, two SNPs (PSEN1 rs165934 and MAML2 rs76032516) remained noteworthy with FPRP < 0.2 and BFDP < 0.75 (Table S2), and the identified associations were visualized in a Manhattan plot ( Figure S1). However, the limited findings are likely due to the limited sample size and possibly by chance.
Specifically, in the multivariable analysis ( than those without unfavourable genotype; and the trend test was statistically significant (P < .001). The classification performance of risk genotypes of PSEN1 rs165934 (Figure 2A,C) and MAML2 rs76032516 ( Figure 2B,D), and their combined effects ( Figure 2E) on OS was illustrated by Kaplan-Meier curves. Obtained in an additive model. The results were in bold, if P < .05. SNP, single nucleotide polymorphisms; OS, overall survival; EOC, epithelial ovarian cancer; HR, hazard ratio; CI, confidence interval. *P, chi-square test for genotype distribution between the two groups; Unfavourable genotypes: PSEN1 rs165934 AA; MAML2 rs76032516 AC/CC. Furthermore, we performed the stepwise multivariate Cox regression analysis to select the optimal predictors of OS in EOC patients, with the clinical variables listed in Table 1

| Stratified analysis between unfavourable genotypes and EOC survival
We performed stratified analysis to assess differential effects of clinical variables on death risk associated with unfavourable genotype groups (Table 4). Overall, patients carrying two unfavourable genotypes tended to have an evidently increased death risk in subgroups of patients ≥50 year at diagnosis, high tumour grade, serous EOC, FIGO stage III-IV, residue disease as well as both subgroups of chemotherapies, compared with that of patients carrying 0-1 unfavourable genotype (P < 0.05 for all).

| Correlations between the loci identified and mRNA expression levels
From the online SNPinfo website with genotyping data for Chinese Han Beijing (CHB) subjects, three loci were in high LD with PSEN1 rs165934 (for rs165932 with r 2 = .946; rs165935 with r 2 = .973; and rs177415 with r 2 = .945) ( Figure S2A). We further searched for the functional relevance in the RegulomDB database for PSEN1 rs165934 and MAML2 rs76032516, as well as the three high LD loci (rs165932, rs165935 and rs177415), and found that PSEN1 and 42 AA carriers), and we found that PSEN1 expression levels increased with a borderline significant trend (P = .060) ( Figure 4B).
Taken together, it is biologically plausible that the associations between PSEN1 rs165934 A allele (in LD with) and survival might be explained by mRNA expression that is regulated by the rs165932 variant locus. However, the difference in MAML2 expression levels among the three genotypes for the locus rs76032516 was not statistically evident, which might be because of the limited sample sizes ( Figure S2B).

| DISCUSSION
To our knowledge, this is the first study to evaluate the associations of genetic variants of Notch pathway genes with OS in Han Chinese EOC patients. We found that PSEN1 rs165934 A > C and MAML2 rs76032516 A > C were associated with survival of patients, and the underlying mechanism of the PSEN1 rs165934 A > C on survival is likely to result from expression regulation by the change of rs165932 A to C that within the same LD block.
Therefore, our results suggest that the observed effects of PSEN1 rs165934 in Notch signalling pathway on survival of EOC patients are biologically plausible. However, the possible mechanism of the MAML2 rs76032516 A > C on the survival remains to be unravelled.
Notch signalling is an evolutionarily conserved mechanism to control cells' response to intrinsic or extrinsic developmental cues that are obligatory to unfold specific developmental programs. 9 The Notch activity involves in the process of differentiation, proliferation, and apoptotic programs in all stages of development of the organisms. There is evidence that aberrant activation of the Notch signalling pathway plays a crucial role in the process of ovarian F I G U R E 4 The mRNA expression levels by PSEN1 rs165934 (A) and rs165932 (B) in ovaries from the expression quantitative trait loci analysis from the GTEx database carcinogenesis and chemoresistance in ovarian cancer patients. 26,27 A previous GWAS-based pathway analysis found that NCOR2, NCSTN and MAML2 variants in the Notch signalling pathway predicted survival in patients with cutaneous melanoma. 20 Besides, variants in ADAM12 and TLE1 were also found to be associated with a poor survival in non-small-cell lung cancer patients. 7 In the present gene-set analysis using the available GWAS data set in Han Chinese women, we observed that variants in PSEN1 and MAML2 were associated with OS of EOC patients.
The OS-associated SNP PSEN1 rs165934 is located in the intron between the exons 8 and 9. The present study showed that carriers of the PSEN1 rs165934 A allele had a poorer OS than C allele carriers and that there was a non-significant trend of decreased PSEN1 Further functional studies are needed to reveal the underlying biological mechanisms of PSEN1 on survival.
MAML2 (mastermind-like transcriptional coactivator 2) rs76032516 is located in the third intron of the gene on 11q21, which is likely to be functional as predicted by RegulomeDB.
The MAML2 protein contains a conserved basic domain that could bind to the ankyrin repeat domain of the intracellular domain of the Notch receptors (ICN1-4) in their N-terminus. The protein binds to an extended groove that is formed by the interaction of CBF1, suppressor of Hairless, LAG-1 with ICN, which could positively regulate the Notch signalling pathway. 34,35 In fact, the translocation of MAML2 could create a fusion oncogene MECT1/MAML2, which could be involved in the process of disrupting the normal cell cycle, differentiation and tumour development, exerting an oncogenic role in mucoepidermoid carcinoma. 36 Recently, the oncogenic CRCT3-MAML2 fusion was also identified to be associated with a better survival of patients with mucoepidermoid carcinoma. 37 Although MAML2 rs76032516 was found to be significantly associated with OS of EOC patients in the present study, the mechanism of MAML2 underlying the effect on the survival was not identified in the present study, nor has not been reported in the literature. Thus, our findings need to be further validated for its functional relevance through in vivo and in vitro studies of MAML2 variants and the gene in the future.
However, the present study has several limitations. First, an inherent limitation was the observational design of the study with limited information on available clinical outcomes, such as chemotherapy-related outcomes, progress free survival (PFS) and objective response rate; second, we were not able to explore the exact mechanism of the loci that were identified to have an effect on survival of EOC; third, we did not have the mutation profile of the tumours that may have impact on survival; fourth, we did not have an access to suitable and comparable target tissue samples from the subjects included in the present study to explore the possible molecular mechanism underlying the observed associations; and finally, the results of the present study may be a chance finding by a limited sample size without additional validation. Therefore, further large and independent studies are required to validate our findings to reduce the positive results by chance.