Refinement of the 22q12-q13 Breast Cancer–Associated Region: Evidence of TMPRSS6 as a Candidate Gene in an Eastern Finnish Population Free

Breast cancer is the most common of cancers among women in western countries. Susceptibility genes thus far identified (e.g., BRCA1, BRCA2, ATM and CHEK2) explain ∼20% of the familial aggregation of breast cancer and the yet unidentified genes presumably are numerous and confer a moderate risk (1). Such low-penetrance susceptibility genes are likely to interact with environmental and life-style factors as well as with other genetic factors to cause disease. Linkage disequilibrium (LD) based genetic association studies are suitable tools for detecting these genes, and young, rapidly grown isolated populations may provide more help by reducing the genetic heterogeneity. For example, single nucleotide polymorphism (SNP) haplotypes in the Finnish population have been successfully used in the detection of susceptibility genes for complex disease (2). The population history makes the Eastern Finns in the late settlement region especially suitable for LD analysis and association studies (3, 4).

We have previously reported an autosome-wide microsatellite scan for LD-based association with breast cancer in this Eastern Finnish population where we found three chromosomal regions as candidate locations for genetic breast cancer risk factors (5). One of them locates on 22q12-q13, in which three microsatellites defined a 550-kb breast cancer–associated region. To refine this association in the present study, we genotyped 10 SNPs in a 516-kb region in the vicinity of the three microsatellites in a set of 497 Eastern Finnish breast cancer cases and 458 controls. On chromosome 22q12-q13, several studies have reported allelic imbalance (AI) in different tumor tissues, e.g., breast, ovary, and colon, and it has been suggested as a possible location for a tumor suppressor gene (6–10). Therefore, to find further evidence on whether there is a connection between the described AI and the association that we observed, we analyzed 45 breast cancer tumors for AI. Here, we report the results of the SNP association and AI analysis of the 22q12-q13 region as evidence of a breast cancer risk–associated gene located in this region.

Samples. The entire sample is a set of 497 breast cancer cases and 458 controls from the province of Northern Savo in Eastern Finland. The cases represent 96% of the total of 516 breast cancers diagnosed in Kuopio University Hospital between April 1990 and December 1995, and the 458 age and long-term area-of-residence matched controls were selected from the National Population Register during the same time period. The sample material is described in more detail in refs. (5, 11). Genomic DNA was extracted from peripheral blood lymphocytes of both cases and controls using standard procedures (12).

The entire sample set (497 cases and 458 controls) is a population sample that also includes women born outside of the Northern Savo region. In order to get as genetically homogeneous a subset of the population as possible, we selected a stratified subset of the entire sample. In the stratified sample, we included only cases born in the province of Northern Savo and their age and long-term area-of-residence matched controls. Altogether, 280 of the 497 cases and 257 of the 458 controls from the entire sample were included in the stratified subsample. The cases between the two sample sets did not differ in body mass index, tumor histology, histologic grade, stage, lymph node status, estrogen receptor status, or progesterone receptor status. Mean age at diagnosis in the entire sample was 58.9 years, and 54.3 years (median 56.3 and 53.2, respectively) in the subsample.

For the AI analysis, we used paraffin-embedded samples of breast cancer tissue from 45 of the original 49 cases included in the initial autosome scan (5). Samples were microdissected by an experienced pathologist in carefully matched tumor areas containing at least 70% tumor cells (13). For each specimen, DNA was extracted from five 10-μm-thick consecutive sections according to standard protocols (14). The Kuopio Breast Cancer Project has been approved by the joint ethics committee of Kuopio University and Kuopio University Hospital.

SNP validating. We searched SNPs in the genes on 22q12-q13 from databases (http://www.sanger.ac.uk, http://www.ncbi.nlm.nih.gov/SNP, http://snp.cshl.org/db/snp) and confirmed the frequencies of 26 suggested SNPs on denaturing high-pressure liquid chromatography (Transgenomics, Crewe, United Kingdom) according to the manufacturer's instructions. We also screened the coding, 5′-untranslated region and 3′-untranslated region sequences of one of the genes in the associated region (RABL4) for sequence variations using denaturing high-pressure liquid chromatography as no coding SNPs for this gene were found in the databases. For validation, we used a set of 48 British and a set of 48 Eastern Finnish samples. The variants from denaturing high-pressure liquid chromatography analyses were re-amplified and sequenced using AbiPrism 377 DNA sequencer and Sequencing Analysis 3.3 software (Applied Biosystems, Foster City, CA) according to the manufacturer's instructions. All reaction conditions, PCR programs and primer sequences are available on request.

Ten of the validated SNPs were selected for genotyping in 497 cases and 458 controls. The criteria for SNPs to be selected were rare allele frequency of 10% or more and suitability to be detected by TaqMan assays. Coding SNPs were preferred, as well as location and distance between adjacent SNPs were considered so that the region would be covered as evenly as possible. The selected SNPs cover a 516-kb region in chromosome 22q12-q13 and adjacent SNPs are 0.9 to 148 kb apart (average distance, 57 kb; Table 1).

SNP genotyping. Genotyping was carried out using the 5′ nuclease assay (TaqMan, Applied Biosystems) in 15 μL reactions in 96-well format as previously described (15). Primer and probe sequences and annealing temperatures are shown in Table 1.

AI analysis. For the AI studies, we used 13 microsatellite markers from the initial autosome-wide scan (5), spanning ∼6.8 Mb on chromosome 22q12-q13 (Fig. 1). Intermarker distances and order were obtained at http://www.sanger.ac.uk and Human Genome Browser Gateway at the University of California at Santa Cruz (http://genome.ucsc.edu). The AI analysis was conducted as previously described (13).

Statistical analyses. The significance levels for comparisons of the SNP allele and genotype frequencies between cases and controls were computed using Fisher's exact test and Monte Carlo approximation implemented in SPSS v 11.5. The consistency of the genotypes with the Hardy-Weinberg equilibrium (HWE) was calculated using the standard χ² test. Breast cancer–associated risks for the SNP genotypes and for rare/common allele carriers were estimated as odds ratios (OR) with 95% confidence intervals (CI) using cross-tabulation in SPSS v 11.5. Haplotype frequencies for multiple loci were estimated using the expectation-maximization algorithm (16) implemented in SNPAlyze v4.0 software. SNPAlyze was also used for testing the association of the estimated haplotypes and breast cancer. This software implements a global test for haplotype frequency differences between cases and controls. Moreover, SNPAlyze provides haplotype-specific tests which allow evaluation of all individual haplotypes. It also provides permutated (empirical) P values for the global test and individual haplotypes. All permutated P values were reached using 10,000 replicates. P ≤ 0.05 were considered significant in all analyses.

LD between two SNPs was estimated by calculating the D′ values for all 45 possible pairs of the 10 SNPs using the combined set of cases and controls (altogether, 891-931 samples for each SNP; ref. 17).

Power estimations for the SNP association studies were calculated using the Genetic Power Calculator, case-control for discrete traits at http://statgen.iop.kcl.ac.uk/gpc/cc2.html (18). In these estimations, we used α = 0.05 and breast cancer prevalence of 0.8% (19). D′ was set at 1 and the allele frequencies were assumed equal for the risk SNP and the marker SNP. Also the risks were assumed equal (1.5 or 2) for the homozygous and heterozygous high-risk allele carrying genotypes.

SNP genotyping. Breast cancer association was tested by comparing the allele and genotype frequencies of the genotyped SNPs between cases and controls separately in the entire sample and in the subsample. Genotype- and allele-specific risks were also calculated for both sample sets. In the entire study sample, no significant difference in the allele frequencies between cases and controls were observed with any of the studied 10 SNPs (data not shown). With rs733655, the difference in genotype frequencies between cases and controls in the entire sample was significant with P = 0.044 (Table 2). With the same SNP, a near significant OR, 1.28 (95% CI, 0.99-1.50) was observed for the rare allele C carriers (TT genotype versus TC and CC combined, not shown) and the genotype-specific risk for heterozygous genotype TC was significant (OR, 1.39; 95% CI, 1.06-1.83; Table 2). Differences in genotype frequencies between cases and controls with the other SNPs were not significant in the entire sample.

In the stratified subsample (which included only cases born in the province of Northern Savo and their age and long-term area-of-residence matched controls), the most significant association was also detected at SNP rs733655 in the matriptase-2 gene (TMPRSS6). The allele frequencies differ between cases and controls with P = 0.009 (data not shown) and genotype frequencies with P = 0.0003 (Table 3). With rs733655, a significant OR of 1.90 (95% CI, 1.34-2.69) was observed for the rare allele C carriers (TT versus TC and CC; data not shown) and also the heterozygous genotype TC was observed to be the risk genotype with OR, 2.11 (95% CI, 1.46-3.05; Table 3). In this subsample, the allele frequencies of SNP rs7285064 in the CSF2RB gene also differed between cases and controls with P = 0.034 (data not shown) and the genotype-specific risk for the heterozygous genotype CT was significant, indicating a protective effect (OR, 0.62; 95% CI, 0.41-0.96; Table 3). Differences in genotype and allele frequencies between cases and controls with the other SNPs were not significant in the subsample.

The deviation of the genotype frequencies from the HWE was tested separately for the cases and controls in the entire sample and in the subsample. In the controls, the rs733655 genotypes deviated slightly from HWE in the entire sample and significantly in the controls of the subsample (P = 0.041 and 0.008, respectively), whereas the cases were in equilibrium. Other SNPs were in equilibrium in cases and controls in both sample sets (Tables 2 and 3).

LD analysis. LD between SNPs was estimated in the entire sample by calculating pairwise D′ values for all 45 possible SNP pairs. In 6 of the 45 pairs, D′ ≥ 0.5 were observed (Table 4). As expected, D′ was detected to decline when the distance between SNPs increased. The highest D′ = 0.99 was obtained for a pair of SNPs that are in the same gene (rs228941 and rs228942) and are separated by 898 bp (Table 4). However, we observed moderate LD between SNPs that are separated by up to 236 kb (D′ = 0.49-0.61).

Haplotypes. Two-marker haplotypes were estimated for adjacent SNPs that were in LD with each other (D′ ≥ 0.5). Altogether, five SNP pairs were tested (Table 5). Of these, the pair rs25095 and rs733655 (61 kb apart), showed significant difference in haplotype frequencies between cases and controls with empirical global P = 0.041 in the subsample (Table 5). An individual haplotype AC of this SNP combination is significantly more frequent in cases (empirical P = 0.003). In the entire sample, this haplotype AC is also more frequent in cases with borderline significance, P = 0.055 (empirical P = 0.073), but the global P for the rs25095 and rs733655 haplotypes is not significant (Table 5). With other tested two-marker haplotypes, the global empirical P for difference between cases and controls was not significant.

Allelic imbalance. AI was detected in 82% (37 of 45) of the tumors at 1 or more of the 13 microsatellite marker loci studied on chromosome 22q12-q13. Altogether, five microsatellite markers, D22S1142, D22S924, D22S1177, D22S445, and D22S279, showed significant AI (Fig. 1). Two of these markers, D22S1177 and D22S445, locate in the breast cancer–associated 550-kb region (5). However, marker IL2RB being the closest marker to TMPRSS6 gene, where the significant SNP association was detected, did not exhibit significant AI.

Here, we report the results of a further association study using 10 SNPs to refine a 516-kb region in a previously detected breast cancer–associated 550-kb region on chromosome 22q12-q13 (5) and AI analysis for a larger region on 22q12-q13 for the detection of a possible tumor suppressor gene(s). A SNP in matriptase-2 gene (TMPRSS6) was detected to associate with breast cancer risk. Our results suggest that the matriptase-2 gene is a candidate for breast cancer risk factor in this Eastern Finnish population.

Significant breast cancer association was detected with SNP rs733655. This SNP locates in the intronic sequence of TMPRSS6 gene which encodes a membrane-bound serine proteinase 6 called matriptase-2. Matriptase-2 is a member of a family of type II transmembrane serine proteinases which have a possible role in cancer development. Matriptase-2 has the ability to degrade extracellular matrix components, suggesting that it may participate in some of the matrix-degrading processes occurring in both normal and pathologic conditions, including cancer progression (20). Because matriptase-2 has only recently been discovered, little is known about its physiologic function(s). Matriptase-2 is expressed in normal breast tissue and the expression is elevated in breast cancer (invasive ductal carcinoma; ref. 21), which points out that matriptase-2 is not expected to be a tumor suppressor. Our AI analysis provided further support for this because AI was not detected in the vicinity of the TMPRSS6 gene. This also indicates that the detected AI seems to be a separate event from the original breast cancer association (Fig. 1), i.e., hereditary risk factor, whereas AI may indicate the involvement of a nonhereditary factor. However, in line with other studies (6–10), the observed AI on the 22q12-q13 region does not exclude the existence of a tumor suppressor, other than TMPRSS6.

The rs733655 allele frequencies that we observed among controls are similar to those expected in the National Center for Biotechnology Information RefSNP database (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?CMD=Display&DB=snp). In the RefSNP European test population sample, the HWE probability is 0.05, indicating that in normal situations, the genotype frequencies deviate from those expected under HWE. This deviation was also detected in our controls group. The allele frequencies among our cases differ from those expected (RefSNP) as genotypes that we observed do not deviate from HWE. If the number of controls by genotype is replaced by the numbers expected under HWE, the difference in genotype frequencies between cases and controls becomes nonsignificant in the entire sample (P = 0.338), but still remains significant in the subsample (which includes only cases born in the province of Northern Savo and their age and long-term area-of-residence matched controls; P = 0.013). Also, the risk for the heterozygous genotype TC becomes nonsignificant in the entire sample but remains significant in the subsample, although reducing from 2.11 to 1.67 (CI, 1.17-2.39). Therefore, we consider that the detected difference between the case and control populations is real. The rare allele C carriers had increased risk of breast cancer in the subsample (OR, 1.90; 95% CI, 1.34-2.69) and nearly significant risk in the entire sample (OR, 1.28; 95% CI, 0.99-1.50) but the rare homozygous genotype CC did not associate significantly with breast cancer. This may be partly explained by too small number of observed rare homozygotes alone to show significant increase in risk (Tables 2 and 3). When validating the SNPs for this study, we sequenced 20 samples for rs733655 and no sequence variation other than rs733655 was observed in the region of the designed TaqMan primer and probe sequences. Genotyping assays were optimized and allele calling was unambiguous as allele controls were used. Also, 141 samples from altogether six plates were re-genotyped (original genotype was confirmed) and thus, a genotyping error is also excluded. Although it is common to exclude markers which deviate from HWE from association studies, it is possible to miss true risk factors by doing so. A recent report described such a polymorphism in the urokinase-type plasminogen activator gene associating with Alzheimer's disease (22).

A common way to enhance the power of genetic case-control association studies is to combine the statistical power of associated markers by estimating haplotypes (23). Haplotypes are dependent on the distance and number of meiotic recombinations between studied markers. In our study, we constructed haplotypes for the five marker pairs that were in LD with D′ ≥ 0.5. One pair involving the most associated SNP, rs733655 in the matriptase-2 gene, showed significant difference in haplotype frequencies between cases and controls, enhancing the association with breast cancer in this chromosomal region. Haplotype analysis was not feasible across the whole 550 kb region as the SNPs were not linked tightly enough to allow identification of common multimarker haplotype. However, the finding of an associated haplotype increases the interest on matriptase-2 gene and the next step is to type additional SNPs covering the TMPRSS6 gene aiming to identify a haplotype and further locate the possible functional change in the gene. To resolve the global importance of TMPRSS6 in genetic risk of breast cancer, the association also has to be tested in other populations. Another SNP, rs7285064 in the CSF2RB gene, showed a moderate association with breast cancer and further evidence concerning the importance of this gene in breast cancer is necessary.

The power to detect risk affecting alterations in LD association studies depends on the allele frequencies of the markers and the strength and length of the LD between the alteration and studied marker. According to the power calculations, our sample set of 497 cases and 458 controls has >95% power to detect a risk allele that is in perfect LD with the marker allele (D′ = 1) and has a relative risk of 2. The power to detect a risk allele that is not in perfect LD with the marker allele (D′ = 0.5) and has a relative risk of 2 varies between 44% and 74% with the 10 SNPs. In the stratified sample, the power for the detected risk (2.11 for the TC and 1.16 for the TT genotype) with SNP rs733655 is 83%. Within a short distance (1 kb), we detected highly significant LD (D′ = 1) and even at distances >200 kb, D′ is ≥0.5 at some instances. It is presumable that LD does not remain at the constant level across the whole 516 kb region and some associations may therefore have been missed. Furthermore, our negative results do not rule out association involving other nearby SNPs and positive results do not necessarily indicate the discovery of the causal SNP but a marker in LD with a true causal SNP located some distance away. Of the genes located in this 516 kb region, matriptase-2 is the most promising candidate for a breast cancer risk gene. Other genes in the studied region have not been reported to associate with (breast) cancer.

In conclusion, the initial autosome-wide scan (5), as well as the present SNP association analysis, has shown that a breast cancer–associated risk factor locates on 22q12-q13. The results here are one step forward in the process of identifying the causative gene and a functional variant within and imply that TMPRSS6 is the gene of interest in the studied Eastern Finnish population. Thus, the possibility of matriptase-2 involvement in cancer progression is further supported and a more specific investigation for the association between matriptase-2 and breast cancer, as well as the identification of the causal genetic variant is needed.