Genome-Wide Linkage Scan for Prostate Cancer Susceptibility in Finland: Evidence for a Novel Locus on 2q37.3 and confirmation of signal on 17q21-q22

Abstract

Genome-wide linkage studies have been used to localize rare and highly penetrant prostate cancer (PRCA) susceptibility genes. Linkage studies performed in different ethnic backgrounds and populations have been somewhat disparate, resulting in multiple, often irreproducible signals because of genetic heterogeneity and high sporadic background of the disease. Our first genome-wide linkage study and subsequent fine-mapping study of Finnish hereditary prostate cancer (HPC) families gave evidence of linkage to one region. Here, we conducted subsequent scans with microsatellites and SNPs in a total of 69 Finnish HPC families. GENEHUNTER-PLUS was used for parametric and non-parametric analyses. Our microsatellite genome-wide linkage study provided evidence of linkage to 17q12-q23, with a heterogeneity LOD (HLOD) score of 3.14 in a total of 54 of the 69 families. Genome-wide SNP analysis of 59 of the 69 families gave a highest HLOD score of 3.40 at 2q37.3 under a dominant high penetrance model. Analyzing all 69 families by combining microsatellite and SNP maps also yielded HLOD scores of > 3.3 in two regions (2q37.3 and 17q12-q21.3). These significant linkage peaks on chromosome 2 and 17 confirm previous linkage evidence of a locus on 17q from other populations and provide a basis for continued research into genetic factors involved in PRCA. Fine-mapping analysis of these regions is ongoing and candidate genes at linked loci are currently under analysis.

INTRODUCTION

Prostate cancer (PRCA) is the most common cancer among men in industrialized countries, including Finland where a total of 4,189 cases were diagnosed in 2007.¹ PRCA is also the second most common cause of cancer death in Finland, surpassed only by lung cancer.

PRCA is a heterogeneous disease, where definitive risk factors include age, ethnic origin and family history, all of which point to genetic factors. Although environmental factors may play a role in the development of PRCA, there remains the suggestion of a significant role for genetic components in PRCA epidemiology (reviewed in Schaid).² Infact, the ~40% heritability component in PRCA risk is the highest reported for a common malignancy.³

Despite the evidence for genetic involvement, localization of high penetrant PRCA susceptibility loci has proven challenging. To date, three susceptibility genes, HPC2/ELAC2, HPC1/RNASEL and MSR1 together with multiple other risk loci, such as HPCX1, have been identified through genetic linkage analysis in high-risk hereditary prostate cancer (HPC) families.² However, the results from the three susceptibility genes do not support a major role for them in the causation of HPC or non-hereditary PRCA in the Finns or other populations studied. In addition to the high-penetrance susceptibility genes and loci, several common, low penetrance polymorphisms seem to have a role in the risk of PRCA. For example, loci on 3q, 8q, 10q, 11q, 17q, 19q and Xp have been detected in genome-wide association studies (GWAS).^{4, 5} Specifically, genome-wide SNP studies revealed that rs5945572 of the gene HPCX2 on Xp11.22, showed genome-wide significant association with PRCA.⁶ Additionally, in a separate GWAS, another SNP close to HPCX2 was found to have an association with PRCA. This SNP, rs5945619, located on chromosome Xp11.23 between NUDT10 and NUDT11, was found to be significantly associated with PRCA in cases either diagnosed at < 60 years or with a family history.⁴ Several independent GWAS of PRCA have identified risk loci in the 8q24 region in several ethnic groups, particularly African Americans.^{4, 5} Despite the evidence of association, the causative genes are still unknown and the identified common PRCA risk variants only account for a fraction of the overall genetic variance for PRCA risk. This suggests that additional risk loci remain to be discovered.

Because of the genetic homegeity of the Finnish population, the likelihood of multiple genes clustering in the population is lower than in highly heterogeneous populations, such as in the USA. Earlier extensive linkage disequilibrium findings in the Finnish population⁷ offer another advantage to utilizing the Finns for genome-wide association studies. Our first genome-wide linkage study⁸ of 13 families and subsequent fine-mapping study⁹ of 16 families gave significant evidence of linkage to one region on 3p26. From these families, only 10 met our current definition of HPC (had 3 or more family members with PRCA) and were included in our current study. Additionally, we conducted subsequent genome-wide linkage scans with 413 microsatellite markers in 44 new families. We also conducted SNP genotyping in 59 of the total 69 families (all families except for the 10 previously published families). We found a novel significant linkage signal for a locus on 2q37.2 and strong confirmation of signal on 17q21-q22.

MATERIAL AND METHODS

Families

In our study, 69 multiplex Finnish HPC families were included. All of the families had at least 3 confirmed cases of PRCA and the mean number of affected persons in was 3.91(±1.01). The mean number of genotyped cases in the families was 2.82. These 69 families had 1676 individuals including 918 males and 762 females. The total number of family members in each pedigree ranged from 5 to 56 with the mean of 23.71 where the median was 25 and mode was 20. Even though the range of age at diagnosis was from 46 to 87, the mean age at diagnosis of all 44 families was 65.77±8.68.

The detailed description of the families is given in Supplemental Table 1. A detailed description of our sample collection protocol and confirmation of diagnoses is presented elsewhere.⁸

Microsatellite Genotyping

Forty-four families were included in the microsatellite genotyping study. Altogether 525 DNA samples including 112 affected men and 225 women were genotyped at the National Human Genome Research Institute Genomics Core Laboratory. Unaffected persons were genotyped to impute genotypes of deceased parents, to infer phase and to construct the haplotypes of all individuals. The mean number of all genotyped individuals per family was 11.93 (range 3-35). Genomic DNA was prepared from blood samples using standard techniques as described earlier.⁸ Details regarding the PCR conditions and allele-scoring techniques for the 413 markers are described elsewhere.⁸ The mean marker genotyping success rate was 98.94% and the mean sample genotyping success rate was 99.24%.

SNP Genotyping

Fifty-nine families were included in the SNP genotyping study. Altogether, 784 DNA samples from 59 families were genotyped for 6,008 SNP markers at the Center for Inherited Disease Research, which includes the 44 families from the microsatellite study described above, plus an additional 15 families. This includes 157 affected men and 323 women. The mean number of affected individuals genotyped per family was 2.71 (range 1-6). Mean number of all genotyped individuals per family was 8.71 (range 1-35). Genomic DNA was prepared from blood samples using standard techniques as described earlier.⁸ The genotyping success rate was 98.5% and the genotyping errors rate 0.01%. After quality control, 5,982 markers were retained.

Linkage analyses

Genotyped microsatellite and SNP genotypes, phenotypes and pedigree information were imported into a custom Visual Foxpro database (Microsoft Corp., Redmond, WA) for data management. The genetic position of the markers was determined based on the Decode map (http://www.nature.com/ng/journal/v31/n3/extref/ng917-S13.xls). Family relationships were checked using RELCHECK^{10, 11}, PEDSTATS¹² and Mendelian inconsistencies were checked with the SIBPAIR program (http://www2.qimr.edu.au/davidD/sib-pair.html#Routines). SIBPAIR was also used to estimate the allele frequencies at each marker using founders only.

Because of the high density of SNPs present on the Illumina Linkage IVB array and the resultant intermarker linkage disequilibrium, SNPs were pruned from the maps before analysis, using the PLINK program¹³, using both pairwise and long range linkage disequilibrium measures, a window size of 20 SNPs, a shift of 5 SNPs and a variance inflation factor (VIF) of 1.25 (r² of 0.2). Residual intermarker linkage disequilibrium was checked using Haploview¹⁴ and confirmed to be below an r² value of 0.05 across all chromosomes. Lastly, an additional SNP was dropped manually due to a high D’ value between one pair of SNPs within the 1 LOD drop region on chromosome 2. In the final linkage analysis, 5,982 SNPs were used.

Parametric linkage analyses were performed using affecteds-only models, in which all unaffected men irrespective of age and all women were treated as unknown. In all analyses, a biallelic major locus was assumed and only individuals with a verified diagnosis of PRCA were considered affected. One liability class was used to specify lifetime penetrances for each of the three possible trait locus genotypes for the affected men. In the parametric analyses, three models were used; a dominant affecteds-only model with high penetrance (penetrance was specified to be 1.0 and 0.001 for genotypes DD/Dd and dd), a reduced penetrance dominant affecteds-only model (penetrance was specified to be 0.5 and 0.05 for genotypes DD/Dd and dd) and a recessive model with reduced penetrance (penetrance was specified to be 0.5, and 0.05 for genotypes DD, and Dd/dd). The frequency of the rare disease susceptibility allele was set to 0.01 in all three models.

GENEHUNTER-PLUS ^{15, 16} was used for parametric and non-parametric analyses. The X chromosome version of GENEHUNTER-PLUS (version 1.3) was used in X chromosome analyses. Two families had to be excluded from the analyses of the chromosome X markers because they exhibited very strong evidence of male-to-male transmission of PRCA. The non-parametric affecteds-only linkage analyses included NPL scores from GENEHUNTER-PLUS using the “all” option and allele sharing LOD scores as developed by Kong and Cox¹⁵ (performed by the ASM program in conjunction with GENEHUNTER-PLUS).

Merging Data

We have previously reported a genome-wide scan in 13 multiplex Finnish families.⁸ We merged the microsatellite and SNP data on the 59 new families with the microsatellite data on the HPC families from our previously published linkage study.⁸ In the previously published study, only 10 of the 13 families met our current definition of HPC (had 3 or more family members with PRCA) and it is those 10 families that were merged with the 59 new families. Details of these methods are provided in the Supplemental section.

Of note, the 10 previously published families had only microsatellite data, but of the 59 newly collected families, 44 had both microsatellite and SNP data and 15 had only SNP markers genotyped because these families were collected and genotyped at different times. Thus, the analyses of the combined microsatellite and SNP data use all 69 families. We did not remove any additional markers from the merged map since intermarker LD between SNP and microsatellite markers is considered very unusual, even in the Finnish population, due to the frequent occurrence of new mutations at the microsatellite markers (see e.g. Hastbacka et al).¹⁷

Significance Threshold

Lander and Kruglyak suggested that parametric LOD scores of 3.3 and 1.9 should be used as criteria for declaring genome-wide significant and suggestive evidence of linkage, respectively.¹⁸ If we conservatively correct for testing multiple models¹⁹ we would add 0.5 to the Lander and Kruglyak¹⁸ significance threshold of 3.3 for a new threshold of 3.8. However, Hodge et al²⁰ showed that as sample size, number of informative meioses and marker map density increases, the overall Type I error rates of multipoint linkage studies that test several penetrance models rapidly decrease to close to the nominal level expected assuming that the test follows a χ² distribution with 1d.f. Thus, we retain the Lander and Kruglyak thresholds in this study given the large number of extended pedigrees and very dense marker map used here.

RESULTS

Linkage results

Merged Microsatellite and SNP Data

Merging the SNP and microsatellite data and analyzing all 69 families together resulted in genome-wide significant evidence of linkage of PRCA risk to regions on 2q and 17q, with HLOD scores greater than 2.0 for four regions under the high penetrance dominant model (Figure 1a) and five regions for the reduced penetrance dominant model (Figure 1b). No evidence of linkage was observed under the recessive model (data not shown). The most-strongly linked markers with respective chromosomal locations and HLOD scores are detailed in Table 1. The highest HLOD score was 3.44 (α= 0.71, p= 0.0008) at 17q12-q21.3 (Figure 2) under the reduced penetrance model with an NPL score of 3.30 (p=0.0009) at this location. The second significant HLOD was 3.32 (α= 0.71, p= 0.0001) at 2q37.3 under the high penetrance model (Figure 3) with an NPL score of 3.70 (p= 0.0002) at this location. The other regions with a positive HLOD score greater than 2.0 in the whole genome under the high or low penetrance models are 6p21.1, 12q21-q24, and 13q13.3 (Figure 1). We also analyzed the microsatellite and SNP data separately. Details of these analyses are in the Supplemental results.

Figure 1.

Whole genome graphical images for linkage results in 69 Finnish prostate cancer families using the combined SNPs and microsatellite data. HLOD linkage results are shown for a) for a high penetrance dominant affected-only model (penetrance specified as 1.0 and 0.001 for genotypes DD/Dd and dd) and b) a reduced penetrance dominant affected-only model (penetrance specified as 0.5 and 0.05 for genotypes DD/Dd and dd) using GENEHUNTER-PLUS.

Table 1.

Maximum HLOD scores in descending order and associated NPL and KC-LOD scores for the combined microsatellite and SNP data in the 69 families (10 families only genotyped with microsatellites, 44 families genotyped with both SNPs and microsatellites and 15 families only genotyped with SNPs)

Chromosome	Position (cM)	Maximum HLOD (p value)‡	NPL Score (p value) ‡	KC- LOD	Proximal Flanking Marker*	Distal Flanking Marker*
High Penetrance Model
2	258.72	3.32 (0.0001)	3.70 (0.0002)	3.24	rs6751336	rs3821280
17	74.06	2.81 (0.0003)	3.30 (0.0008)	2.77	rs758408	rs1526189
13	106.26	2.45 (0.0008)	2.81 (0.003)	2.15	rs714668	rs1894758
12	100.98	2.34 (0.0010)	2.39 (0.01)	1.44	rs1163016	rs9143
Reduced Penetrance Model
17	74.06	3.44 (0.0001)	3.30 (0.0009)	2.77	rs758408	rs1063647
2	258.72	3.20 (0.0001)	3.71 (0.0002)	3.24	rs6751336	rs3821280
12	111.98	2.15 (0.0017)	2.35 (0.01)	1.43	rs1433251	rs9143
6	80.89	2.12 (0.0018)	2.08 (0.02)	1.10	rs875142	rs199630
13	106.26	2.02 (0.0023)	2.81 (0.003)	2.15	D13S158	rs1894758

^‡Asymptotic p-values are given for both maximum HLOD and NPL scores.

^*Proximal and distal flanking markers denote the boundaries of the 1-LOD drop region.

Figure 2.

Individual HLOD plot for chromosome 17 from the linkage analysis results for 69 Finnish prostate cancer families using the combined SNPs and microsatellite data. HLOD linkage result of 3.44 (p=0.0001) is shown for a reduced penetrance dominant affected-only model (penetrance specified as 0.5 and 0.05 for genotypes DD/Dd and dd) using GENEHUNTER-PLUS. The dashed lines denote the 1-LOD drop region.

Figure 3.

Individual HLOD plot for chromosome 2 from the linkage analysis results for 69 Finnish prostate cancer families using the combined SNPs and microsatellite data. HLOD linkage result of 3.32 (p=0.0001) is shown for a high penetrance dominant affected-only model (penetrance specified as 1.0 and 0.001 for genotypes DD/Dd and dd) using GENEHUNTER-PLUS. The dashed lines denote the 1-LOD drop region.

DISCUSSION

The present study of the 59 new multiplex PRCA families was undertaken to localize new genomic areas linked to PRCA susceptibility and to confirm previously reported loci seen from GWAS in other populations.

The signal on 17q12-21 represents strong confirmation of multiple other reports of linkage to this region. Under the most conservative significance thresholds, the linkage signal on chromosome 2q37 is “suggestive” rather than significant. However, Hodge et al. have shown that as the size, number of informative meioses and marker map density increases, the overall Type I error rates of multipoint linkage studies that test several penetrance models rapidly decrease to the nominal level. Therefore, we believe that this linkage signal also represents important novel evidence of linkage. In an evidential framework²¹, there is strong evidence of linkage on 2q37 since the likelihood ratio (LR) is > 2000 and θ₁ is quite small (0 in multipoint analysis). The possibility that this strong evidence for linkage is “misleading” is best controlled by replication. Indeed, other studies have observed suggestive linkage in this region. In addition to the strong linkage evidence on 2q37 and 17q12-21, we also found suggestive linkage (HLOD > 1.9) in four other regions; 6p21, 12q21-q24, 13q13, and 13q31-q34. Finally, we had weak evidence of linkage (HLOD > 1) in regions: 3q26, 10q26, 14q32, 15q26, 16p13 and Xq26.

Under the high-penetrance dominant model, the locus on 2q37.3 gave the strongest signal with a maximum HLOD of 3.32. Previously, in an analysis of 504 brothers with PRCA, Suarez et al²² saw nominally positive evidence for linkage (Zlr score >1.645) at a broad region from 2q32.1-2q37.3. Suggestive linkage to this area (2q37.2-q37.3; KC LOD = 1.01; P = 0.02) has also been seen in 12 American HPC families with co-occurrence of adenocarcinoma of the pancreas.²³ A number of interesting genes lie under this 2q peak, the most notable of which is APO7. This gene, also known as NGEP (New Gene in Prostate) and TMEM16G (transmembrane protein 16G), is primarily expressed in prostate²⁴ and is significantly associated with expression of the prostate-cancer associated genes PSA (KLK3, MIM:176820), PAP (ACPP, MIM:171790) and KLK2 (MIM:147960).²⁵

Another genes of interest in the 2q37.3 region is histone deacetylase 4 (HDAC4, MIM:605314). Transcription can be activated through acetylation of histone lysine residues, which destabilizes nucleosomes and induces conformational changes. This allows transcription factors to access the DNA. Deacetylation reverses this process and can therefore repress transcription. The repression of cell growth and differentiation genes by HDACs have been linked to a number of cancers.² HDAC4 can be localized in the cytoplasm or the nucleus, and has been observed to be more present in the nucleus in more aggressive PRCA and that androgen is involved in this localization.²⁶

Other noteworthy genes in the 2q37 region are, an RNA binding motif protein (RBM44) and RAMP1 (MIM:605153), a receptor activity-modifying protein 1 precursor. RAMPs are required to transport calcitonin-receptor-like receptor (CALCRL, MIM:114190) to the plasma membrane. In the presence of RAMP1, CALCRL functions as a calcitonin-gene-related peptide (CGRP, MIM:114130) receptor. RAMP1 is involved in terminal glycosylation, maturation, and presentation of the CGRP receptor to the cell surface. CGRP is a neuroendocrine marker and serum CGRP levels in untreated PRCA patients have been reported to reflect tumor volume and aggressiveness.

Under the reduced penetrance dominant model in our combined SNP plus microsatellite analysis, the locus on 17q21-q22 gave the strongest linkage signal (maximum HLOD =3.44). Previous studies have implicated the chromosome 17q region to be associated with cancer.^{27, 28} Our results on 17q21-q22 reconfirm the linkage peak first seen by Lange ²⁹ in 175 HPC pedigrees from the University of Michigan Prostate Cancer Genetics Project (PCGP). Most of the evidence in the Lange et al study came from a subset of pedigrees with four or more affected individuals (LOD = 3.27). Recently, Lange et al ³⁰ reported a fine-mapping study of the 17q21-q22 region where 95 new multiplex families and 9 additional microsatellite markers were analyzed. A novel subset of 147 families, combination of the old and new sets, that had four or more PRCA cases and an average age of PRCA diagnosis ≤ 65 years, resulted in a maximum LOD score of 5.49 at 78 cM with a 1-LOD support interval of 10 cM. However, a new set of 131 families from PCGP revealed only modest evidence for PRCA linkage at 17q21 (LOD=0.28).³¹ Michigan is one of the areas where many Scandinavians, including the Finnish, emigrated at the beginning of the 19^th century. Therefore, it is possible that the current 17q21-q22 linkage signal can have its origin in Northern European populations. Besides the linkage evidence, SNP association signals have been reported on 17q12 and 17q24.3 that have been suggested to have a cumulative and significant association with PRCA.^{32, 33} However, Levin et al³⁴ found that SNPs in this region did not account for their prior evidence of PRCA risk linkage to chromosome 17.

Numerous genes were found within a 1 LOD drop of the 17q21-q22 peak. The following selected genes have known functions suggesting a possible implication in the development and progression of PRCA.

Like chromosome 2, the linkage peak for chromosome 17 also includes a member of the RAMP gene family. RAMP2 (MIM:605154), is a member of the receptor calcitonin activity modifying proteins (RAMP) which transports the calcitonin-receptor-like receptor (CRLR) to the cell surface where it functions as an adrenomedullin receptor at the cell surface. Adrenomedullin has been shown to stimulate prostate cancer tumor cells.³⁵

The STAT genes represent a family of proteins involved in cell growth and apoptosis. When phosphorylated due to cytokines and growth factors, STAT genes translocate to the cell nucleus where they activate transcription. There are three STAT genes present within the linkage peak of chromosome 17. These STAT genes are STAT3 (MIM: 102582), STAT5A (MIM:601511), and STAT5B (MIM:604260). Persistent activation of STAT3 is oncogenic and has been implicated in solid and hematologic tumors, including PRCA.^{36, 37}

BRCA1 (MIM:113705) is a nucleolar phosphoprotein involved in genomic stability (e.g. DNA repair) and tumor suppression. Mutations in this gene have previously been implicated in breast and ovarian cancer. More recent evidence suggests that BRCA1 may play a role in the risk of developing PRCA.³⁸ In patients with BRCA1 and BRCA2 mutations, PRCA risk was increased for founders with BRCA2 but not BRCA1 mutations and the BRCA1-185delAG mutation was associated with high Gleason score tumors.³⁹

Our results showed three regions with suggestive evidence for linkage: 12q22-23, 6p21 and 13q13.3. Additionally, 10q26, 14q32, 15q26, 16p13 and Xq26 exhibited weak evidence for linkage (1.0 < HLOD < 2.0). We discuss the relevance of these suggestive and weak linkage regions in the Supplemental materials.

Of our three regions of suggestive evidence, the area on 12q was detected also in our first published GWL analysis⁸ with a LOD score of 0.99 (θ = 0.1) for marker D12S326 on 12q22-23, which overlaps with the present region at 12q21.2-q23.2 showing HLOD > 2.0. Another suggestive region, the area of 6p21, was recently re-identified using novel linkage analysis methods⁴⁰ and was among the seven regions initially seen in the ICPCG consortium pooled data analysis or the subset of aggressive prostate cancer pedigrees.⁴¹ The 6p21 area was also one of the nine regions of potential chromosomal deletion seen in androgen-independent tumors.⁴² The 6p21 region was identified as a fragile site⁴³ which holds implications for increased risk of genetic damage and concomitant increased cancer risk. The third area at 13q13.3 has not been seen in linkage analysis prior to our results.

A few regions for which linkage to HPC has been previously reported exhibited weak (1.0 < HLOD < 2.0) linkage. For example, the locus on 10q26 has been suggested to harbor a prostate cancer susceptibility and tumor aggressiveness gene.⁴⁴ The locus on 14q32 has been previously seen in aggressive Utah HPC families (dominant HLOD =2.09 at D14S1426)⁴⁵ and the locus on 15q26 was recently published in a dense genome-wide SNP linkage scan in 201 HPC families with more aggressive prostate cancer phenotypes (HLOD 1.99).⁴⁶ The locus of 16p13 was one of the five suggestive linkage loci seen in 269 families with at least five affected members, analyzed by the ICPCG consortium⁴¹ and the site of Xq26 is one of the first HPC linkage hits, also seen and analysed in the Finnish HPC families.⁴⁷

Our significant linkage peak regions on chromosomes 2 and 17 warrant further investigation. BRCA1 is a candidate gene in the chromosome 17 region. However, the preponderance of evidence suggests that one or more additional high penetrance loci may contribute to PRCA risk in this region. Though the recruitment of multiplex HPC families is arduous, the data presented here strongly support the value of this type of family data in studying the genetic determinants of PRCA. Family studies will most likely return to importance in human genetics after a few years of silence because they offer different information than can be obtained from GWAS concerning high impact, rare variants.