Identifying breast cancer risk loci by global differential allele-specific expression (DASE) analysis in mammary epithelial transcriptome

Breast cancer is the most common cancer and the second most common cause of cancer-related death in women. It is estimated that one out of eight American women will develop breast cancer some time in their lifespan and 3.0% will die from this disease [1]. For the year 2012, about 226,870 new invasive breast cancer diagnoses and 39,970 breast cancer related deaths are expected in the United States [2]. Due to this high prevalence and severe consequences, genetic factors contributing to breast cancer risk have been intensively studied.

Family history is known to be associated with 20% - 30% of breast cancer incidence in the United States [3]. Pedigree analysis of clustered familial cases followed by positional cloning in the 1990s led to the discovery of tumor suppressor genes, BRCA1[4] and BRCA2[5], two major breast cancer susceptibility loci. Deleterious mutations in these genes increase the risk of developing breast cancer by more than 10 fold and overall account for 15% - 30% of observed risks in familial breast cancer cases [6]. To discover breast cancer susceptibility alleles that constitute the remainder of genetic risk, genes associated with BRCA1/2 pathways were investigated in BRCA1/2 mutation negative familial cases. Such candidate gene approaches revealed that germline mutations of TP53[7], PTEN[8], ATM[9], CHEK2[10], BRIP1[11], PALB2[12], NBS1[13] and RAD50[14] are correlated with breast cancer risk, but to a much more moderate extent than BRCA1 and BRCA2. Therefore, new unbiased genomic approaches are needed for identifying genetic factors that influence breast cancer susceptibility.

Over the last decade, advances in array technologies have resulted in the ability to evaluate the expression of thousands of genes simultaneously. These platforms offer a powerful tool to test multiple biomarkers for breast cancer tumorigenesis and prognosis, as well as targeted breast cancer therapy [15]. However, gene expression assessed by current techniques represents the total level of transcripts produced by both parental alleles. The absolute transcript level failed to resolve potential imbalances in relative allelic contribution to the total expression. This perspective is particularly important for familial breast cancers, where an individual inherits a germline mutation on one parental allele, followed by a somatic mutation of the second allele in the tumor cells. Previously, we have reported that mutant BRCA1 transcripts containing premature stop codons were eliminated or destabilized by nonsense-mediated mRNA decay (NMD) [16] and could lead to a state of haploinsufficiency. As a result, the ratio between the expressions from the wild-type allele and the corresponding mutant allele was significantly increased, resulting in what we coined differential allele-specific expression (DASE) or allelic imbalance (AI) [17].

DASE is a common phenomenon in human tissues [18]. Although its contribution to breast cancer susceptibility has been implicated [17], it has not been studied on a transcriptome-wide scale in breast cancer precursor tissues. Since the phenomenon of DASE at a locus may help identify nearby cis-acting transcriptional and epigenetic regulatory sites as well as mutations resulting in non-mediated RNA decay [16, 18], we propose that DASE is a sensitive functional index for genetic variants, and can be used as a novel approach to identify risk alleles for breast tumorigenesis. The main objectives of this study are to identify genes with DASE by comparing the allele-specific expression (ASE) and to demonstrate that global DASE analysis could be a powerful new approach to identify breast cancer risk alleles.

In our current study, we demonstrated transcriptome-wide DASE analysis as a novel approach to identify breast cancer susceptibility loci. The HumanOmni1-Quad BeadChip we utilized has state of the art coverage of common SNPs on the human transcriptome. Our study identified 60 candidate DASE loci by both SNP- and gene-based methods (Table 1 and Figure 2). Pathway analysis reveals one major DASE gene network which is likely associated with breast tumorigenesis (Figure 3). Using PCR and Sanger sequencing, we successfully validated the DASE predictions in this breast cancer-related network, which includes cancer causative genes ZNF331 and USP6, and breast cancer risk associated gene DMBT1 (Table 2). By analyzing the 5′UTR region of DMBT1, we successfully identified rs2981745 as the causal variant for the DASE in DMBT1 (Figure 4). Therefore, we presented an example supporting our original expectation that DASE analysis may lead to the discovery of functional DASE-causing variants. DASE in ZNF331 has possibly resulted from genomic imprinting as indicated from studies in human extraembryonic tissues [26], but we are not able to verify such speculation for the lack of genetic material from parents. Nevertheless, we reported for the first time about such phenomenon of ZNF331 in primary cultures of adult human tissue. We did not investigate further with the 26 non-coding DASE candidate loci in current studies largely because very little information could be obtained to help interpret their roles. However, there have been a few studies denoting the importance of non-coding RNAs in cellular processes during recent years [27, 28]. Considering the high percentage (nearly 50%) of such loci in our candidate list, it is an important next step for us to validate non-coding candidate DASE loci and study their likely roles in breast tumorigenesis.

As the Illumina Human Omni1-Quad Array is originally designed for targeting genomic DNA sequences instead of cDNA sequences, intronic SNPs and many SNPs at the exon/intron boundary are left out for DASE profiling. Although the number of “usable” SNPs covers the majority of transcripts, the coverage could be further increased with newly developed platforms with additional SNP markers. Furthermore, the size of the probes on the Illumina SNP array is 50mer, and it is likely difficult for them to pick up small size transcripts, such as miRNAs. Therefore, it would be logical to consider designing customized BeadChips covering pre-selected probes to improve both cost-effectiveness and specificity, and to reduce the data processing load for global DASE profiling as well. In addition, we observed fluorescent interference between X and Y channels during the data analysis due to the initial probe design [29], and it was able to be alleviated by canonical data normalization. Despite these limitations, Infinium-based BeadChip is still a practical platform for whole transcriptome DASE analysis as indicated by the results of successful validation using Sanger sequencing (Table 2).

As mentioned in the Introduction, currently known high and moderate penetrance risk alleles help only to explain a fraction of familial breast cancer incidents and the existence of more susceptibility genes are likely to be very rare. Thus, it is plausible that the remaining cases could be complemented by the synergistic effects of multiple low penetrance alleles, each conferring an elevated risk of <1.5 fold [30]. The completion of the Human Genome Project and fast development of SNP array technologies have made it practical to perform genome-wide association studies (GWAS) to identify genetic factors that account for breast tumorigenesis. Since the first wave of GWAS to search for such alleles, 22 loci have been reported to significantly associate with breast cancer risk by 16 studies [31–34]. Among those candidate risk loci, nearly half (2q35, 3p24.1, 5p12, 5q11.2, 6q25.1, 8q24.21, 10q26.13, 11p15.5, 16q21.1-21.2) were reproducible in multiple independent studies, which denotes the reliability of GWAS prediction. Despite these successes, the mechanisms of how these GWA variants affect breast cancer pathogenesis are often unknown, because in most cases it is not clear which gene(s) are associated with GWA signals [35, 36]. Therefore, understanding the function of these breast cancer associated variants and the mechanisms of how they contribute to breast cancer is a logical next step to validate GWAS findings. Until now, most of them are merely SNP tags for yet unknown breast cancer causal variants [37]. A couple of exceptions so far were rs2981582- A [31] and rs1219648- G [32] on FGF2R locus. The discovery of these two common low risk variants eventually pinpointed rs2981576 and rs2981578 as causal variants by genetic mapping and CHIP assay [38]. Evaluation of the functional impact of putative causal variants identified by GWAS could be very challenging as many GWA signals (e.g., 8q24) map some distance from the nearest coding regions, and are likely to mediate disease predisposition through remote regulatory effects on transcription [39]. In addition, the causal alleles involved in breast cancer susceptibility are likely to have moderate molecular and cellular effects, and the measurable effects of an allele in a given functional assay may not exhibit a causal role in breast cancer pathogenesis by itself. Therefore, novel strategies are required to functionally characterize the multiple GWA signals in a genome-wide manner for convincing genetic evidence. Based on previous reports [24, 40], our study has pinpointed a likely causal variant for DASE in DMBT1 (Figure 4). In addition, genetic mapping in mice has indicated that DMBT1 is a candidate modifier of mammary tumors and breast cancer risk [41], and our results further support that allelic loss of expression in DMBT1 could contribute to breast cancer development, which is consistent with the reports from previous studies [42, 43]. These findings support the idea that global DASE profiling mapping could be a powerful approach to validate GWAS findings when the DASE loci map is overlapped with existing data from GWAS [44].

In our study, we have compared top DASE loci identified in our genome-wide ASE studies with the current cancer genome database provided by Cancer Gene Census (Sanger Institute) and have identified two DASE loci (USP6 and ZNF331) listed as cancerous genes [45]. Somatic mutations in USP6 (p.V678A and p.R475Q) and in ZNF331 (p.G193E) have been reported in 1% of primary breast cancer and 2% of ovarian cancer, respectively [45]. These findings suggest that global DASE profiling could be a powerful tool in combination with currently available cancer genome databases to identify novel breast cancer “driver” genes. Furthermore, on-going cancer genome sequencing projects (e.g. Cancer Genome Atlas by NCI) have identified thousands of so-called variants of unknown/uncertain significance (VUS), including variations typically characterized by a single base change, or a change to several intronic bases. This large amount of VUS data produced by the next generation sequencing-based cancer genome projects has been termed “overkill” from a clinical perspective [46]. Currently, the consequence of most VUS in oncogenesis has yet to be established. As DASE is a sensitive functional index for pathogenic mutations, it can be applied for validating these VUS at a genome-wide scale as well.

We have demonstrated for the first time that global DASE analysis is a novel approach to identify breast cancer risk alleles. The results from our study are very promising and we expect our strategy will help validate the functional variants identified by the GWAS and Cancer Genome Projects. Importantly, the research strategy developed here could be easily applied to investigating susceptibility for many other types of cancers.

This work was kindly supported by the Susan G. Komen for the Cure (KG100274 to X.C.), the Eileen Stein Jacoby Fund, and the Risk Assessment and Presentation Keystone Program at Fox Chase Cancer Center.