Introduction

Endometrial cancer [EC] is the sixth most common cancer in women worldwide and the fourth most common cancer in women in high-income countries, with an increasing incidence in recent years [1]. There are well-known life-style risk factors that contribute to EC development and increase in incidence [2]. There are also well-known inherited genetic factors that contribute significantly to EC risk [3]. EC risk is especially high in women with inherited loss-of-function variants in specific genes [3]. Lynch syndrome is the most common high-risk EC syndrome, responsible for approximately 2-5% of all EC cases [3]. After excluding known EC germline variant carriers, the relative risk of EC in first-degree relatives is two-fold compared to the general female population [4]. This indicates that unexplained heritable factors are involved in EC susceptibility. Recent genome-wide association studies [GWAS] of single nucleotide polymorphisms [SNPs] have identified several low-risk loci for EC, however these explain less than 10% of the familial risk of EC [5–7].

Haplotypes are a specific consecutive combination of SNPs located close to each other on a chromosome and are usually preserved through many generations because they are rarely affected by larger relocations of the genome during homologous recombination cycles. Studying haplotypes instead of individual SNPs in cancer susceptibility and other complex diseases could increase the power of statistical analysis and could define a risk locus more precisely, as shown previously [8–10]. Thus, identified unique risk haplotypes define the borders of a DNA region holding an assumed risk variant between the first and last marker.

We hypothesized that haplotype analysis could be more powerful than SNP-only-analysis in identifying novel EC susceptibility loci with higher risk. To test this hypothesis, we conducted a two stage GWAS in EC cases and controls. Previous GWAS on EC risk have been performed in heterogeneous populations [5–7]. Since Sweden has a more homogenous population, making it more beneficial for haplotype GWAS, we conducted the discovery GWAS using only Swedish cases and controls from the Endometrial Cancer Association Consortium [ECAC] study [5]. For validation, a replication GWAS of the Swedish novel loci were performed in two small European cohorts from genetically similar populations (Belgian and German) together with the same Swedish cohort to enable haplotype comparison and to identify potential founder mutations [5].

Materials and methods

Results

To examine the association between a specific combination of close SNPs (a haplotype) and EC risk a two stage GWAS was performed. In the discovery Swedish dataset of 1,116 EC cases and 5,021 controls 332,906 SNPs were analyzed which resulted in 8,315,750 sliding windows. Significant SNPs or haplotypes in a delimited region is defined as one susceptibility locus and typically represented by several haplotypes of different lengths, ORs and statistical significance. The discovery haplotype GWAS identified 15 novel susceptibility loci located on chromosomes 2q31.1, 4p16.1, 4p15.31, 6q13, 7p21.1, 9p13.3, 10q26.3, 11q21, 12q13.11, 13q12.11, 15q13.3, 16q24.3, 19q13.32, 20p12.3 and 22q13.2 (Table 1, S1–23 Tables, S3 and S4 Figs). The loci had a frequency ranging from 1% to 8%, ORs from 1.6 to 3.3 and p-values from 4.25x10⁻⁸ to 9.86x10⁻¹⁵ (Table 1). Three loci spanned more than one gene, 11 loci spanned one gene, and one locus spanned no known gene (Table 1). The detailed haplotypes for each locus are listed in S24 Table. One significant SNP was identified on 16q24.3 (OR 1.58, p 3.36x10⁻⁸) (Table 1, S25 Table and S4 Fig). For the other novel loci, the lowest p-value for one SNP ranged from 3.8x10⁻⁶ to 0.19 (S25 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Haplotypes with lowest P value in 15 endometrial cancer risk loci.

https://doi.org/10.1371/journal.pone.0316086.t001

All suggested 15 susceptibility loci were novel, and validation was therefore warranted. Two European cohorts were used for replication. Theoretically, a haplotype defines the borders of a susceptibility region that contains a possible underlying mutation. Therefore, we aimed to replicate the 15 haplotype regions from the discovery GWAS (Table 1) from the first to the last SNP at each locus. The significant SNP at 16q24.3 was sorted out during the QC for the replication analysis. Therefore the haplotype with a SNP before and after is presented for comparison in the replication analysis. To identify potential founder mutations identical SNPs are required for comparison. This was assured with a new QC with combined samples from the Swedish, Belgian and German cohorts. After QC, a separate haplotype analysis with a window size of 1-25 was performed in the two replication cohorts and the enlarged Swedish cohort (with 6 new cases and 1,059 new controls).

A locus was considered replicated if the replication GWAS presented any SNP or haplotype in the discovery region with a p-value <  0.05. Eight of the novel Swedish loci from the discovery GWAS were replicated in both the German and Belgian cohorts (4p15.31, 6q13, 9p13.3, 10q26.3, 11q21, 12q13.11, 13q12.11 and 22q13.2); whereas two loci (19q13.32 and 20p12.3) were replicated in the Belgian cohort and one locus (15q13.3) was replicated in the German cohort (Fig 1) (S26 Table). The loci at 2q31.1, 4p16.1, 7p21.1 and 16q24.3 were not replicated in either cohort.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Replication of the 15 novel endometrial cancer risk loci.

Replication GWAS in a German (221 cases; 251 controls) and a Belgian cohort (528 cases; 1,266 controls) along with the Swedish cohort (1,139 cases; 6,080 controls). Each locus presented with first (SNP1) and last SNP (SNP2), genes in the area (if any). The haplotype with the lowest p-value in each cohort is presented within the discovery locus (marked with a box) with corresponding frequency (F), odds ratio (OR), p-value (P) and indication of replication (P < 0.05). Shared identical haplotypes between the cohorts are marked in dark grey, indicating possible founder variants. The discovery SNP on 16q24.3 (marked with a box) was not analysed in the replication analysis but situated between the two SNPs in the replication haplotype.

https://doi.org/10.1371/journal.pone.0316086.g001

Shared identical haplotypes between cohorts may indicate a possible underlying founder mutation. Possible European founders were suggested on 4p15.31 and 12q13.11, where identical haplotypes between the Swedish, German, and Belgian cohorts were identified. Two possible founders were suggested on 22q13.2, one Swedish-Belgian and one Belgian-German. An indication of a Swedish-Belgian founder was seen on 6q13, 13q12.11 and 15q13.3, a Swedish-German founder on 9p13.3 and 10q26.3, and a Belgian-German founder on 4p16.1 and 11q21 (Fig 1).

Discussion

In the search for founder haplotypes of endometrial cancer, we reanalyzed genotype data from Swedish samples included in ECAC by performing a discovery haplotype GWAS of 1,116 invasive EC cases and 5,021 controls. This haplotype GWAS suggested 15 novel risk loci for EC, of which eight were replicated in two separate small European cohorts and an additional three in any of the two cohorts. Each haplotype conferred a two- to three-fold increased risk of EC. Our study demonstrated that haplotype analysis was able to identify novel risk loci with higher ORs than previous GWAS based on single variants [5–7].

Two European cohorts were available for validation analysis. Both were however small, which limited the ability to achieve statistical significance in this analysis. Nonetheless, the consistent presence of positive ORs for the haplotypes, similar to those observed in the discovery analysis, indicated that risk alleles are likely located within these haplotypes and even suggested the possibility of European founders. Identification of potential founder mutations required a new QC including all the three cohorts to involve identical SNPs for comparison. This means that the Swedish replication result was not identical to the discovery result. This was expected, as GWAS is highly dependent on the markers used in the analysis. Importantly, all loci had positive ORs in the replication for all three cohorts. All loci which showed compelling associations in the Swedish discovery cohort did not reach the same level of statistical significance in the replication sets. Nevertheless, this could be explained by discrepancies between markers in the discovery and the replication analyses. Using different markers in two GWAS conducted on the same data set may not identify the same haplotypes. Moreover, due to genetic population heterogeneity between the Swedish, German and Belgian cohorts, we cannot expect all loci to be replicated.

We have previously demonstrated the ability to use haplotype analysis, in addition to SNP analysis, when identifying novel risk loci for cancer in general as well as for colorectal and breast cancer in a Swedish population [8–10]. We suggest that haplotype analysis may define the spanned locations for disease-causing variants. SNP analysis results in an exact position for a risk-associated SNP; however, the distance to the actual disease-associated variant in SNP analysis is often unknown. In contrast, a risk haplotype is assumed to define the borders of the region containing the disease-causing variant. However, more homogenous populations are required to enable this strategy to identify founder risk haplotypes. Further genotyping studies could search for founder haplotypes.

Haplotype analysis also demonstrated the complexity of the genetic contribution of low- to moderate-risk loci, depending on the combined effect of more than one disease-causing variant. This was supported by the present study, in which three risk haplotypes covered more than one gene (Table 1), a finding that was also seen in our previous studies [8–10]. The odds ratios for the risk haplotypes were higher than those previously reported in published SNP GWAS [5–7]. The higher ORs could be explained by the superior ability of a haplotype to identify a population at risk, compared to a single variant. Therefore, haplotype analysis resulted in higher ORs than single variant analysis at the same locus (Table 1 and S26 Table). This is consistent with our previous findings from haplotype GWAS [8–10]. Single variant analysis showed similar ORs and frequencies of known EC susceptibility SNP´s in the Swedish and previous ECAC analyses for the SNP´s included in the present analysis. However, a statistically significant level was not reached, probably because of the smaller sample size (S27 Table) [5–7].

The P-value criteria for a sliding window haplotype analysis are complicated and can be discussed as multiple testing is performed. Numerous overlapping haplotypes were generated in the sliding window analysis. We assumed that the various haplotypes in each delimited region (S1–S23 Tables) reflect the same genetic risk and thus could represent one risk locus. Therefore, no correction for multiple testing was performed and p < 5x10⁻⁸ was used for the discovery analysis. The cohorts used for replication in this study were small and further studies are required to replicate the Swedish results in other European populations.

Haplotype analysis revealed one locus on chromosome 16. The haplotype with the lowest p-value in this region was a single variant, suggesting that the actual risk variant may be very close to this variant (Table 1). In the replication analysis, this SNP was not analyzed, as it was sorted out during the QC of the replication sets. There are three overlapping elements in this small region: the gene ANKRD11, a coding protein AC137932.1-201 and a ribosomal DNA, RP11-46C24.5. ANKRD11 is a chromatin regulator and p53 activator that has been studied in several cancer types and has recently been suggested to be involved in predisposition to ovarian cancer [20,21].

Several genes located in these 15 loci have been previously discussed in relation to cancer, although many have not been previously reported to be involved in EC. GREM1, located at the novel locus 15q13.3, is known for its role in hereditary mixed colorectal cancer syndrome [22]. The potassium channel-interacting protein 4 gene (KCNIP4) at 4p15.31 has been proposed as a candidate gene for renal cell carcinoma and was recently identified as a lung cancer risk gene in a familial GWAS [23,24]. The potassium channel voltage-gated KQT-like subfamily member 5 (KCNQ5) at 6q13 is considered a hypermethylation marker in colorectal cancer [25]. The ATP-binding cassette member 5 gene (ABCB5) at 7p21.1, which is upregulated in cancers, such as skin-, breast-, colon cancer and melanoma, is a multidrug resistance mediator in diverse malignancies [26,27]. TTC40 at 10q26.3 was found to be amplified in HER2-negative gastric cancer in one study and a nearby suggested tumor suppressor gene at locus 10q26.1, the cancer susceptibility candidate 2 gene (CASC2), is commonly lost in EC [28,29]. The locus on 12q13.11 holds no gene but is located close to SLC38A1, which has been studied in many types of cancer and specifically suggested to have a role in EC, where an upregulation of the gene was observed and a downregulation of the gene resulted in decreased EC cell proliferation [30]. TMX4 at 20p12.3 promotes palmitoylation of cysteine residues adjacent to the membrane-spanning domain of the transmembrane thioredoxin family (TMX) [31]. Palmitoylation is essential for proteins encoded by both oncogenes and tumor suppressor genes. Therefore, activation of TMX4 may be a risk factor for cancer [32]. The locus on 22q13.2 showed two possible founder variants, one Swedish-Belgian and one German-Belgian. The Belgian haplotype, including both founder regions presented higher OR than the single founder Swedish and German haplotypes, consistent with an increased risk resulting from the two variants. The signal peptide-, cub domain- and EGF-like domains-containing protein 1 (SCUBE1) at the 22q13.2 locus is a promising biomarker in renal cancer, and possibly also in breast cancer [33,34].

Some genes involved in the 15 novel EC loci have been shown to play a role in the tumor microenvironment. Integrin gene alpha-6 (ITGA6) at locus 2q31.1 has been linked to a more metastatic cell phenotype in cancer [35]. Integrins play an important role in cell adhesion through cooperation with different cell surface receptors, including growth factor and G protein-coupled receptors [36]. Many studies have provided evidence for the role of laminin-binding integrins in tumorigenesis, and both tumor-promoting and tumor-suppressive activities have been identified [37].

A genome-wide association with a trait can be either a direct genetic effect or an indirect effect through an associated risk factor. Interestingly, the gene TUBA3C at 13q12.11 has been suggested to play a role in obesity, which is a known risk factor for EC [38]. KIAA0232, at 4p16.1, has been associated with platelet counts [39], and has been reported to affect cancer development and progression [40].

Three loci suggest haplotypes spanning more than one gene. The 9p13.3 locus spans six genes. A Swedish-German founder haplotype confirmed this locus, which was further supported by a SNP in the Belgian cohort. It is possible, however unlikely, that all genes contribute to a risk variant of this haplotype. Nevertheless, the large Swedish and German haplotype in this region indicated that more than one risk variant could be involved. RUSC2 is associated with cancer therapy pathways, indicating its role in cancer development [41]. Other genes have also been reported to be involved in cancer. Fam166B has been found to be upregulated in adipose tissue, possibly with an effect on cancer risk [42]. CD72 is constantly expressed in chronic lymphocytic leukemia and other B-cell lymphoproliferative disorders [43]. TESK1 has been suggested to be involved in cancer development [44]. SIT1 has been implicated in B-cell proliferation [45], and RMRP has been suggested to have an oncogenic role in hepatocellular carcinoma [46]. In the locus on 11q21 three possible genes may be involved. HEPHL1 and PANX1 were proposed to have a founder mutation in Belgium and Germany, while the Swedish analysis suggested only an association with FOLR4. It is still possible that the other variants were missed in the replication analysis depending on the other SNPs used in the analysis. Both HEPHL1 and PANX1 have been previously described in relation to cancer, although this is the first implication in EC. Three SNPs of HEPHL1 are associated with dietary iron intake and colorectal adenoma, and PANX1 appears to facilitate tumor growth and metastasis in tumor cells [47,48]. Finally, the locus on 19q13.32 presented a haplotype including three genes, PVRL2, TOMM40 and APOE in the first analysis. In the replication, both the German and Belgian analysis resulted in OR > 1, but p-value below 0.05 only in the Belgian analysis. In a recent study, co-expression of PVRL2 was observed in multiple tumor types, interestingly with the highest co-expression observed in EC [49]. APOE has also been suggested to be associated with EC, and increased APOE expression has been associated with poorly differentiated endometrial tumors [50,51]. TOMM40 overlaps the APOE.

Conclusion

In conclusion, a discovery haplotype analysis identified 15 novel candidate loci associated with increased EC risk in a Swedish cohort. A second replication study supported most of the loci with positive ORs and presented identical genotypes between populations at some loci. These results support the hypothesis that haplotype GWAS could be powerful in identifying risk loci with higher risks than previous GWAS. The results suggested that more than one risk gene/variant could contribute to the cancer risk at one locus. Further studies are warranted to replicate this study and how to incorporate the results in clinical risk prediction.

Supporting information

Acknowledgments

The authors thank all individuals who participated in the RENDOCAS, CAHRES, KARMA, LES, LMBC, BECS, and BBCC studies as well as all clinicians, technicians and administrative staff involved in the five studies.