A genome wide SNP genotyping study in the Tunisian population: specific reporting on a subset of common breast cancer risk loci

Breast cancer is a complex disease with a strong heritable component. Numerous genetic, hormonal and external environmental factors are involved in breast cancer etiology [1].

Currently, half of the inherited susceptibility to breast cancer is explained by a combination of high, intermediate, and low-risk alleles [2]. Rare high risk alleles such as BRCA1 [3, 4], BRCA2 [5, 6], TP53 [7], STK11, PTEN and CDH1 explain approximately 20–30% of the inherited susceptibility, intermediate-risk alleles in genes such as CHEK2, ATM, BRIP1(FANCJ) [8‐10] and PALB2 (FANCN) [11‐15] explain an additional 5%, while common lower penetrance alleles explain approximately 16% of the breast cancer risk [16‐23]. Thus, many additional loci remain to be identified [24‐26].

In North Africa, breast cancer is the most common cancer among women, representing 25 to 35% of all female cancers [27]. Compared to Western countries, the incidence of breast cancer in North-African countries (Tunisia, Algeria, Morocco, Libya and Egypt) is lower [28, 29]. However, it is reported that breast cancer is more aggressive in North Africa than in Western countries with notably large proportions of young patients [30‐32].

These large differences are mainly explained by different genetic architectures between populations. Initiatives such as the international HapMap and the 1000 Genomes projects [33, 34] provided an unprecedented opportunity to systematically analyze genetic differences and similarities between human populations from European, Asian, and Sub-Saharan origins. It is well known that any given haplotype frequency may differ from one population to another because of different Linkage Disequilibrium (LD) structures and variable genetic variant frequencies among populations [33]. Analysing differences in genetic loci patterns between ethnicities may help to decode the biological mechanisms by which the risk associated variant at a susceptibility locus causes breast cancer [34‐37]. Indeed, the high level of haplotype diversity in African populations may facilitate fine-mapping of causal variants that underlie disease associations [38‐40]. For instance, fine mapping in the African-American population contributed to the localization of a causal variant in FGFR2, a low penetrance breast cancer susceptibility locus [41]. Therefore, studying other African populations may bring new insights on breast cancer genetic knowledge.

The Tunisian population (TUN), is a heterogeneous and admixed population from African and European origins [42]. The particular structure of the Tunisian population is also due to a relatively high rate of consanguinity that has an impact on the incidence of monogenic diseases such as those predisposing to cancer [43]. Despite its genetic wealth, few breast cancer genetic studies have been performed in the TUN population such as those that focused on the identification of the mutational spectrum of BRCA1 and BRCA2 [44, 45]. Thus, little is known about the involvement of other genes of high and moderate penetrance and much less for common variants regarding their frequencies, correlation coefficients and association with breast cancer risk in Tunisia. Indeed, a study investigating breast cancer risk in Tunisia associated with 9 risk SNPs has been performed [46]. Because of the limited number of breast cancer risk SNPs that have been investigated so far in the Tunisian population, this current study aim to extend the number of common genetic variants investigated in the general healthy Tunisian population and to compare their characteristics between different ethnic groups.

In the present study, we investigated the breast cancer genetic architecture in a set of 135 subjects from the general Tunisian population by analyzing several previously identified breast cancer associated variations. Additional file 1: Table S1. summarizes the 90 breast cancer susceptibility loci that have been investigated in this current study. The studied genomic regions have been selected as follow: for the high and moderate penetrance genes we selected a genomic region that covers the LD block containing each gene in order to perform haplotype analysis. For low penetrance loci (SNPs) we studied a genomic region that covers 20 kb upstream and downstream of each SNP. We selected 20 kb on both sides in order to avoid overlap between variants that are located on the same locus. Thus, a 7 Mb genomic region have been spanned for each of the 135 studied subjects.

One thousand five hundred eighty-seven variants directly genotyped on the array and located on these 90 selected loci have been included in this study. Among them, 28 SNPs are known to be associated with overall breast cancer risk (Table 1) including 3 SNPs already identified to be associated with breast cancer risk in the Tunisian population namely rs1219648, rs2981582 in FGFR2 and rs8051542 in TOX3 [46].

We therefore performed haplotype analysis by constructing LD blocks and phasing haplotypes on the 90 loci. We calculated correlation coefficients and allelic frequencies and characterized haplotype tagging SNPs (htSNPs). This constitutes a genetic database for use in further breast cancer association studies in the Tunisian population.

We determined the frequencies, in the Tunisian population, of at risk haplotype that carry at risk alleles of the 28 variants known to be associated with breast cancer risk (Fig. 1). Our results suggest that at risk haplotypes on 2p24, 4q21, 6q25, 9q31, 10q26,11p15, 11q13 and 14q32 loci are considerably frequent in the Tunisian population (haplotype frequency > 20%), however, at risk haplotypes of the 2q31; 4q34, 7q35 and 22q13 loci seem to be rare (frequency < 5%).

Moreover, we explored the functional role of the 28 selected variants. We performed analysis using RegulomeDB to identify DNA features and regulatory elements overlapping this set of variants and by applying a heuristic RegulomeDB score to prioritize candidate functional variants prior to further investigation (for a description of the RegulomeDB scoring scheme and referenced datatypes refer to http://​www.​regulomedb.​org). High RegulomeDB scores were attributed to four SNPs: rs11099601, rs1494961, rs9911630 and rs4808801 (Table 1). Further functional annotations for each of these SNPs are shown in Additional file 1: Table S2.

The highly ranked score “1b” was assigned to the BRCA1-rs9911630 variant. The “1b” RegulomeDB score suggests that this variant is a putative functional SNP that may be associated to eQTL evidences. Then, eQTL associations have been assessed for the 28 variants using data from the GTEx database. In breast mammary tissues, significant eQTL associations have been observed for rs720475, rs3903072, rs9911630, rs799916 and rs4808801 (Table 1). The most strongly expression-associated variants were rs9911630 and rs799916, two strongly correlated variants in the BRCA1 genomic region. Both variants were associated with expression levels of NBR2, CTD-3199 J23.6 and LINC00854 genes with a highly significant eQTL evidence for NBR2, a BRCA1 neighbor gene (p = 1.2 × 10^− 23 and p = 2.1 × 10^− 25 for rs9911630 and rs799916, respectively) (Additional file 2: Figure S1).

Based on its interesting RegulomeDB score and the significant eQTL associations, we undertook further analysis for rs9911630. Using the Genevar platform, we assessed eQTL associations for rs9911630 in 8 populations. Data showed that rs9911630 is significantly associated with the expression level of BRCA1 gene (ILMN_173827-ENSG00000012048-BRCA1) in non-African populations (Asian (JPT and CHB) and Caucasian population (CEU)) but not in Africans. (Fig. 2a and b). Consequently, we compared the allelic frequencies of rs9911630 between these different populations.

Consistently with eQTL results, the frequency of rs9911630 is significantly different between Africans and non-African populations. Indeed, the “G” minor allele of rs9911630 in European and Asians shifted to a major allele in Africans (Additional file 1: Table S3). The allelic frequency of rs9911630 in Tunisia is significantly different from Africans and Asians and not different from Europeans and haplotype analysis using Haploview tagger tool showed that rs9911630 is a haplotype tagging SNP in the Tunisian population (Fig. 2c).

In silico predictions of micro RNA binding sites of rs9911630, as a BRCA1–3′ variant, have been also performed using mirBase. In silico analysis predicts that rs9911630 alters the binding sites of three non human microRNAs by the gain of mse-miR-2766 binding site and the loss of bmo-miR3287 and ssa-miR-19d-5p binding sites (Additional file 2: Figure S2; Additional file 1: Table S4).

Based on the different allelic frequencies of rs9911630 (A > G) in the different ethnic groups investigated in this study, we suggested that the effect of rs9911630 on the microRNAs binding sites will differ from one population to another. Indeed, in European and Asian populations, rs9911630 would cause a gain of miR-2766 binding and a loss of bmo-miR3287 and ssa-miR-19d-5p Micro-RNAs. However, in Africans, rs9911630 would act in a different way by promoting the binding of bmo-miR3287 and ssa-miR-19d-5p and loosing the binding site of miR-2766.

Finally, we used the chi-square statistical test to compare the genetic characteristics of the selected 28 SNPs between TUN and other populations (Additional file 1: Table S3). Interestingly, the frequency of rs13329835 seems to differ significantly between the Tunisian population and all other HapMap populations. In addition, three SNPs (rs2046210, rs941764 and rs3803662) showed significant difference in their frequencies between Tunisian and European (CEU + TSI), Asian (CHB + JPT) and African populations (MKK, LWK and YRI). However, no significant differences have been observed between rs3803662 frequencies in Tunisian and in ASW and MEX populations that are considered as admixed populations. For the remaining 24 SNPs, significant differences either with Africans, Europeans or Asians were observed. In addition, in order to investigate the breast cancer genetic architecture in different ethnic groups, we studied the distribution of the studied population using a principle component analysis based on the frequencies of the 28 genotyped variants. PCA revealed clear distinction of the breast cancer architecture among the three human geographic origins (Europe, Asia and Africa) (Fig. 3). Admixed populations such as MEX and GIH seem to cluster in an intermediate position. However, TUN population seems to cluster close to Europeans populations.

Moreover, the first principal component (Fig. 3a) showed that the largest inter-ethnic variability is found between African and non-African populations (49.5%). A set of 15 variants that contributed significantly to the observed genetic variability between African and non-African populations (Dim1) have been characterized (Additional file 2: Figure S3a). rs4808801 seems to be the variant that contributed more to this inter-ethnic variability. rs9911630 also figure in the list of variants that contribute very significantly to the genetic variability between African and non-African populations. PCA results also showed that the variability between European populations (CEU and TSI) and non-European ones, is about 27.8% and a list of 11 variants have been also identified as the most contributors to the genetic variability between these populations (Dim2) (Additional file 2: Figure S3B).

Consistently with the PCA results, we also showed that YRI, LWK and JPT populations are the most contributors to variability between African and non-Africans (Dim 1), and TUN, MEX and GIH are those who contributed less to this variability (Additional file 2: Figure S4A). This represents an additional argument toward the admixed and intermediate classification of these populations.

However, the Tunisian population altogether with the TSI and CHB populations are the most contributors to diversity between Europeans and non-Europeans, which confirms the close genetic relationship between Tunisians and population from European ancestry (Additional file 2: Figure S4B).

In order to explore the genetic relationship between North African and European populations, we performed another PCA using a new matrix that integrates values from additional populations from North Africa (Tunisian Berbers, Moroccans and Egyptians) and Southern Europe mainly from Spain. This second analysis confirmed the close genetic relationship between North African populations and Europeans (Fig. 3b).

Several differences have been observed in breast cancer epidemiological features between populations [57]. This is mainly due to different demographic, environmental and lifestyle factors, but also result from differences in genetic architecture from one population to another. Indeed, it has been shown that American women from African origins (Afro-Americans) are three times more likely than Caucasian Americans to develop the highly aggressive triple-negative and inflammatory breast cancer forms [58]. Moreover, several studies showed that high rate and long history of consanguinity, commonly observed in developing countries, decrease breast cancer incidence rate by decreasing the frequency of mutations in high penetrant breast cancer genes such as BRCA1 and BRCA2 [59, 60]. In Tunisia, mutation prevalence of these two genes is considered lower than in Europe with 19.7 and 7.5% for BRCA1 and BRCA2 respectively [61]. Medimegh et al. [62] suggested that in the absence of deleterious mutations on BRCA1, wild type alleles of BRCA1 genetic variants seem to decrease the expression level of the BRCA1 in 50% of familial breast cancer cases through their interaction with micro-RNAs (miRNAs) that are increasingly recognized as mediators in a variety of biological processes including breast cancer [62].

Because of the lack of large studies on common breast cancer genetic variation in the Tunisian population, we undertook this genome wide study with a focus on 90 breast cancer loci. We investigated differences and similarities between these loci in terms of allelic frequencies in more than 11 different populations, we also investigated linkage disequilibrium, correlation ratios and haplotype structure of breast cancer loci in Tunisia. Moreover, in silico functional assessment of the most relevant SNPs have been performed. This study represent the largest and the most complete study on common breast cancer variants in the Tunisian and North African populations.

Genome wide-PCA performed in this current study (data not shown) showed the admixed and intermediate genetic architecture of the Tunisian population. Although it’s geographic belonging to the African continent, the Tunisian genetic characteristics of breast cancer common variants seems to be closer to Europeans than to Africans. The close genetic relationship between Tunisian population and Europeans may be explained by migratory waves that happened in the Mediterranean region since the Paleolithic period. Different studies using mitochondrial DNA, Y chromosome and SNP genotyping in Tunisia also showed the admixed and intermediate genetic architecture of the Tunisian population [63‐67].

Haplotype analysis of the selected breast cancer susceptibility loci, showed that at risk haplotypes on 2p24, 4q21, 6q25, 9q31, 10q26,11p15, 11q13 and 14q32 loci are considerably frequent in the Tunisian population (> 20%). When comparing the allele frequencies of several common variants between Tunisian and other ethnic groups, some variants found on these loci showed significant differences between populations. Indeed, 6q25-rs2046210 and 14q32-rs941764, differ significantly between Tunisians and all other population expect ASW and MEX populations, respectively. This represents an argument of the admixed nature of the Tunisian population.

The 6q25 represents an ambiguous locus in breast cancer association studies in different ethnic groups [21]. Indeed, rs2046210 was found to be associated with breast cancer risk in European and East Asian populations but not in African populations. However, subsequent studies showed that other variants on the 6q25 region are associated with breast cancer risk in Africans namely: rs9397435 and rs2046211 [68]. Interestingly, the 6q25 locus contains the ESR1 gene that encodes for the estrogen receptor protein which is strongly involved in breast carcinogenesis. Thus, association studies of the 6q25 SNPs in the North African population may help to explain the development of specific breast cancer phenotypes in North African population such as the triple negative breast cancer subtype.

Moreover, two variants (rs1219648 and rs2981582) found on the 10q26 locus have been recently identified as associated with breast cancer risk in the Tunisian population [54]. These two variants seem to be in a strong correlation (r² = 0.8, Additional file 2: Figure S5.) meaning that they are carried by the same haplotype. Therefore, further analysis on the 10q26 locus in the TUN population should be prioritized.

In addition, the frequency of rs13329835 seems to differ significantly between Tunisians and all other HapMap populations. Recent studies showed that this variant is associated with breast cancer risk in Europeans and Afro Americans [24, 69]. Because, the allele frequency for this SNP differ considerably across ethnic groups, association studies in the Tunisian population are required to assess its association with breast cancer risk.

We also performed in silico functional analysis in order to identify SNPs more or less likely to have functional effects. High RegulomeDB scores were assigned to variants on the 4q21, 19p13 and 17q21 loci.

The functional impact of rs1494961 and rs11099601 at 4q21 locus have been previously investigated by our group using eQTL analysis and ENCODE data. Significant eQTL associations and interactions with different biofeatures such as histones and enhancer elements have been identified for these two SNPs [70].

Fine-scale mapping studies have been also performed on the 19p13 locus in order to identify causal variants at this breast cancer susceptibility locus [71]. In this current study, 19p13-rs4808801 was identified as a putative functional SNP with interesting functional evidence (eQtl association with SSBP4 gene, p = 1.6 × 10^− 5). Our data also showed that rs4808801 is the variant that contributed more to the inter-ethnic genetic diversity between African and non-African-populations.

However, no functional studies have been reported for 17q21-rs9911630 except one of our previous studies that mentioned the potential association of this variant with differential allelic expression [62].

Here, we identified rs9911630-BRCA1 variant as a haplogroup tagging SNP in the Tunisian population. Statistical analysis also showed that the frequency of this variant varies significantly between ancient populations (Africans) and more recent ones (Europeans and Asians). rs9911630 frequency in Tunisia seems to be between that of Europeans and Africans which would be expected in this admixed population with European and African heritage. eQTL analysis indicated that rs9911630 is significantly associated with the expression levels of NBR2 and BRCA1 genes. Significant eQTls were also observed between rs9911630 and other genes namely: CTD-3199 J23.6 and LINC00854.

The observed eQTLs associated between rs9911630 and BRCA1 derived from analysis performed in fibroblast, lymphoblastoid cell line and T cell. However, eQTL associations between this variant and NBR2, CTD-3199 J23.6 and LINC00854 derived from breast mammary cell lines. As regulatory effects are often tissue-specific this may explain the different eQTL associations observed in these different cell types. CTD-3199 J23.6 is a transcribed processed pseudogene, no further information on it is function or its involvement in carcinogenesis have been reported. LINC00854 is an RNA gene, and is affiliated with long non-coding RNA class (lncRNAs). NBR2 is also a non-protein coding gene that encodes a long non-coding RNA and suppresses tumor development through regulation of adenosine monophosphate–activated protein kinase (AMPK) activation [72]. It resides adjacent to the tumor suppressor gene BRCA1. Given the close proximity of the human BRCA1 and NBR2 genes, it has been suggested that these 2 genes may be coordinately expressed. Since many lncRNAs regulate the transcription of neighboring genes, the hypothesis that NBR2 regulates BRCA1 transcription is also plausible [73].

lncRNAs is still a poorly understood class of non-coding RNAs that refer to a classified group of RNA transcripts longer than 200 nucleotides and have no apparent protein-coding potential. Accumulating recent evidence links long non-coding RNAs to cancer metabolism by regulating various aspects of cancer metabolism through their cross-talk with other macromolecules [74, 75]. In addition, recent studies showed that lncRNAs can cross-talk with other non-coding RNAs such as microRNAs through competing endogenous RNA (ceRNA) mechanisms [76]. Interestingly, our results indicated that rs9911630 alters the binding sites of three non-human miRNAs (miR-2766, bmo-miR3287 and ssa-miR-19d-5p). The regulation of gene expression by cross-species microRNAs has been previously reported and their link to cancer development or prevention have been also explored [77]. Indeed, non-human microRNA such as plant and animal derived microRNAs have been detected in human blood in a large nutrigenomics study cohort [77]. Moreover, the role of food-related microRNAs in regulating the expression of key human cancer-related genes was highly debated [78, 79]. Thus, long and small non coding RNAs (lncRNAs and microRNAs), altogether with protein-coding RNAs may form complex regulatory networks involved in various aspects of cancer biology.

Interestingly, significant eQTL associations observed between rs9911630 and BRCA1 have been observed only in European and Asian individuals but not in Africans and the allele frequencies of this variants also differ significantly between African and non-African populations. We hypothesized that differences in BRCA1 expression levels between African and non African populations may be explained by difference in the allele frequency of this variants and by differences in the expression level of some exogenous microRNAs mainly due to different environmental and lifestyle factors such as food intake habits. Moreover, in a separate study, we identified rs16942-BRCA1 as a modifier variant of breast cancer risk in BRCA1 mutation carriers [80]. rs16942 seems to be in complete LD with rs9911630 in populations from European origins but not in African populations. Thus, we suggest that rs9911630 is a potential functional variant that may be associated with breast cancer risk in an ethnic specific manner by altering the expression level of key tumor suppressor genes such as BRCA1 and NBR2. However, the association of this variant with breast cancer risk in different ethnic groups warrants further association and functional investigations.

The goal of this research was to explore the genetic architecture of a number of breast cancer risk SNPs in the Tunisian population comparing their frequency to 11 different ethnic groups. The observed discordance in the genetic background between populations highlights the necessity for researchers to establish a specific genotype profile for each population. Therefore, caution should be exercised in applying any genetic risk prediction model based on tagSNPs outside of the ancestry group in which it was derived. To our best knowledge, this study is the largest investigation of breast cancer common loci in the general Tunisian population. Further investigations are in progress by our group regarding the association of the potential functional variants identified in this study and breast cancer risk in the Tunisian population.