Prostate Cancer Susceptibility Loci: Finding the Genes

Introduction

In 2006 alone nearly 234,460 men in the United States will be diagnosed with prostate cancer. In addition, about 27,000 will die from, rather than with, their disease (1). Although there is considerable variability in disease incidence by age, race, and family history, about 70% of all cases are ≥65 years (yrs) at diagnosis, and the median age at diagnosis in the United States is 68 yrs (2). The disease is more frequent among African American than Caucasians; Asian men have the lowest reported incidence (3, 4).

Segregation analysis were the first to suggest a strong hereditary component for prostate cancer, particularly among younger men. Two similar studies based on ascertainment of family history through probands treated with radical prostatectomy showed evidence for the dominant transmission of a rare high-risk allele/s (population prevalence of 0.3%–0.6%), with carriers having an 88%–89% risk of getting prostate cancer by age 85, compared to 3%–5% in non-carriers (5, 6). Carter et al., have suggested that the cumulative proportion of prostate cancer cases within the U.S. population that is attributable to high-risk susceptibility alleles is 43% for men diagnosed ≤55 years, 34% for men ≤70 years, and 9% for men ≤85 years (5). By comparison, population-based studies from Sweden (6) and Australia (7, 8), estimate a higher population prevalence of carriers (1.1%–1.67%) and a lower lifetime incidence (63%–79%). The latter studies also suggest that 23% of all prostate cancer cases diagnosed <65 years may be due to inherited mutations in susceptibility genes (6).

In addition to evidence for autosomal dominant Mendelian inheritance, several studies also support an X-linked or recessive model with higher relative risks for prostate cancer in men with affected siblings compared to affected fathers. A segregation analysis from Australia found that the best fitting models included a dominantly inherited increased risk that was greater at younger ages (penetrance of 70% by age 80) and a recessively inherited or X-linked increased risk that was greater at older ages of diagnosis (8). The latter study also found that all two-locus models gave better fit than single-locus models, suggesting that multiple loci are responsible for the disease.

Several types of epidemiological studies also present compelling evidence for the existence of prostate cancer susceptibility loci. Both case-control and cohort studies show that having a first-degree relative with prostate cancer increases a man’s risk of being diagnosed with the disease by two- to three-fold relative to those without a family history (9). If the relative is diagnosed before age 65 (RR=5.9) or if there are ≥3 affected first-degree relatives (RR=10.9), the risk is increases significantly (9–11). Twin studies report higher concordance rates for monozygotic (19%–27%) than dizygotic (4%–7%) twins (7, 12, 13), with the largest study reporting a relative risk (R.R.) for prostate cancer of 12.3 (95% CI 8.4–18.1) in monozygotic twins (13).

Efforts to identify susceptibility loci for hereditary prostate cancer (HPC) have been ongoing for several years (14–30). At the heart of the problem is the extreme locus and disease heterogeneity that are hallmarks of the disease (31–35). With over a dozen genome scans completed to date, suggestive evidence for loci has been described on nearly every chromosome, and efforts to replicate results using seemingly similar data sets has been challenging (31–35). Yet, in recent years, it appears that progress is finally being made. Strategies for overcoming the problem of locus heterogeneity and ultimately identifying prostate cancer loci are the focus of this discussion.

Genome Wide Scans for Prostate Cancer Susceptibility

In 1993, Carter et al. (36) provided a definition for hereditary prostate cancer (HPC) based on families meeting at least one of the following criteria: 1) prostate cancer in ≥3 first-degree relatives; 2) prostate cancer in three successive generations of the maternal or paternal lineages; or 3) two first-degree relatives affected at age ≤55 years. We, like other researchers, have used these criteria to ascertain families for genome-wide scans aimed at finding prostate cancer susceptibility loci. In aggregate, the community has published the results from over a dozen genome wide scans (37–39). This has allowed the community to make several predictions regarding the genetic nature of prostate cancer in high-risk families. These are as follows:

1. Prostate cancer susceptibility is caused by mutations in multiple genes

2. Different genes are likely associated with distinct population frequencies in various ethnic groups; and

3. Distinct models of Mendelian inheritance and varying levels of penetrance are associated with different loci.

Loci Mapped to Date

Among data sets of hereditary families, several genes have been suggested as causative. However there has been extreme difficulty in both confirming linkage results, as well as results derived from candidate genes analysis of selected populations. Among the most frequently debated genes are HPC2/ELAC2 (40), HPC1/RNASEL (41), (42), and BRCA2 (43). For each, both confirmatory and lack of replication studies are reported. In some cases, such as that of HPC1/RNASEL, replication studies have focused exclusively on isolated populations such as Ashkenazi Jewish patients or Finnish men, in an effort to reduce problems associated with heterogeneity (44–46).

In late 2003, eight genome-wide scans for prostate cancer susceptibility, including our own of 255 families were published (21). The aggregate results are summarized in a review by Easton and colleagues (33, 38) and shown in Figure 1. The eight scans include 1293 families with multiple cases of prostate cancer. Across all studies, eleven peaks with Lod scores > 2.0 were observed, identifying regions on chromosomes 2, 3, 4, 5, 6, 7, 9, 16, 17, 19, and 20 as containing putative susceptibility loci. However other than a replication by the group describing the initial results on chromosome 20, none of these peaks coincided precisely with any of the previously reported linkage peaks. The most promising were the results on chromosome 19 by Wiklund et al. (24) which is close to the peak on chromosome 19 originally reported by reported by Hsieh et al. (27).

Figure 1.

Prostate Cancer Linkage Peaks

Of even greater surprise was the fact that no chromosomal region was reported with a significance level of 2.0 or greater by more than one study. In addition, only one study demonstrated a Lod score ≥3.0—that of Cunningham et al. (19), which reported a Lod score of 4.77 on chromosome 20q13. The same group had originally mapped HPC20 and the data in the replication study included a subset of the original mapping families. The highest reported Lod score by any other group on chromosome 20, was Lod = 1.02.

The results from other candidate loci were similarly discouraging. A total of 24 additional Lod scores >1.0 were reported, identifying every chromosome as carrying a putative locus except 12, 13, 21 and 22. No Lod scores >1.0 were observed at the HPC2 locus on 17p or the HPCX locus on the X chromosome. Indeed, Lod scores in excess of 1.0 were noted only by Xu et al. (25) and Cunningham et al. (19). Again, both data sets contained families from the original study. Scores in excess of 1.0 were observed on chromosome 16q only by Witte et al., (47) which was the hypothesis generating group, and one other group. In addition to Xu et al. (25) who originally reported linkage to 8p (25), only one other group noted a Lod score >1.0. In summary, none of the “candidate loci” received true statistical confirmation by an independent group and only the locus at 19p proposed by Hsieh et al. (27) was replicated by an independent group with a Lod score of ≥2.0.

These facts speak to the enormous level of both phenotypic heterogeneity and locus heterogeneity observed with prostate cancer. Clearly there are many genes contributing to the disease with varying levels of penetrance. The introduction of the prostate specific antigen blood test (PSA) contributes to the overall phenotypic variability, as man are now diagnosed earlier in life then they might have been previously. Indeed, most data sets in the literature today are mixed and reflect a subset of men diagnosed before and after PSA came into common use, which is estimated to have occurred in the late 1980s. To try and overcome these difficulties and find prostate cancer loci, the prostate cancer research community at large has employed four strategies, each of which are discussed in turn below.

Meta Analysis and the International Consortium of Prostate Cancer Genetics

The prostate cancer mapping community has formed an international working group termed the International Consortium of Prostate Cancer Genetics (ICPCG). The goal of the ICPCG is to work together, often sharing data prior to publication, to generate large meta data sets with increased power for tackling problems related to prostate cancer susceptibility. The advantage of this strategy is that is allows extensive sub classification and stratification, while still retaining sufficient numbers in each strata such that statistically meaningful analysis can be done. Appropriate corrections for multiple corrections can be made, and accurate results still achieved.

Thus far the group has focused on replication studies on chromosome 1 and 20 (48, 49). In the case of HPC1, the community has shown that only very large families with no evidence for linkage to the X chromosome likely can attribute their disease to mutations at HPC1. On chromosome 20 the consortium finds little evidence for replication in a dataset of over 1200 families, calling into question the original finding (49). More recently, the ICPCG has also undertaken studies focused on families with an excess of aggressive disease (50).

Clinical Features of Disease

Several studies have focused exclusively on aggressive disease (Reviewed in (51)). While many men will die with prostate cancer, it is those that die of prostate cancer that are the most clinically interesting and the ones that the research community most wishes to study. Initial studies focused on using Gleason score as a measure of aggressiveness. Gleason score is an assessment of tumor grade (52). To obtain a Gleason score, several regions of the tumor are independently scored by a pathologist and assigned a number of 1–5, representing well to poorly differentiated patterns. The higher the score the worst the prognosis (53). The two predominant scores are then added together to give a summary score between 2 and 10, with most tumors falling in the range of Gleason 5–7.

Some studies have treated Gleason score as a quantitative trait for outcome of disease aggressiveness, since it is reported to be a good predictor of survival (53). Others have treated the Gleason score as a covariate, using it to explain locus heterogeneity. For example, Witte et al. analyzed grade as a quantitative trait on 513 men from concordant sibships. They reported evidence for linkage on chromosomes 5q31-33, 7q32, and 19q12-13.11 (54). Using the same data set as Witte (54), Goddard et al. (18) used a variety of factors as covariates including sum of the sib-pair Gleason scores, mean family age at diagnosis, existence of male-to-male transmission (which argues against X linkage), and the number of affected first-degree relatives. In doing so, they detected linkage at three previously reported loci (1q24-25, 1q42.2-43, and 4q) and found linkage at Xq12-13 (LOD score 3.06, P= 0.00053), adjacent to the androgen receptor. In addition, they identified five other loci with Lod scores ≥2.5.

Interestingly, the loci at 1q24, Xq12 and chromosome 5 were evident only when Gleason score was considered as one of the covariates. In the absence of the covariates, results were weak to non existence. Others have found this approach to be similarly informative (55–57). Some of the strongest data using Gleason score as a measure of tumor aggressiveness have come from Slager et al. in a genome scan of 161 sibpairs (58). They not only strongly confirm the linkage results for chromosome 19q (P< 0.00001), but report evidence for linkage on chromosome 4 (P = 0.00012). In a subsequent and independent study, involving 175 brother pairs from 103 families the same group found evidence for linkage at 6q23 (P = 0.0009), 1p13-q21 and 5p13-q11 (59).

As more studies have been undertaken, it has become clear that a precise definition of aggressiveness is needed. Most recent studies have used a definition of aggressive disease (60–62) that includes families in which at least two genotyped men had at least one of the following disease features: regional- or distant-stage disease (based on pathology if a radical prostatectomy has been done, including T3, T4, N1 or M1, otherwise data from clinical staging is accepted); a Gleason score at diagnosis of ≥7 (poorly differentiated grade if no Gleason score is available); a pre-treatment prostate-specific antigen (PSA) score of 20 ng/ml or higher; and if deceased, death from metastatic prostate cancer at <65 years.

Using the above criteria, we reported suggestive evidence for linkage on chromosome 22 (dominant HLOD = 2.18) (62). We utilized clinical data from 784 affected men from 248 HPC families for whom a genomic screen had been previously performed (21). Disease characteristics described above were used to classify affected men into categories of clinically insignificant, moderate, or aggressive prostate cancer. Only men with aggressive disease were coded as affected in the linkage analysis. Suggestive linkage was observed at chromosome 22q11.1 (dominant HLOD = 2.75) and at 22q12.3-q13.1 using a recessive model of inheritance (HLOD = 1.90). Other studies have reported at least nominal evidence for linkage on chromosome 22. For instance, Lange et al. reported a Lod score of 1.87 at 45cM in a set of 16 African American families and a Lod = 1.87 at 51 cM in 79 families with four or more affected men (22). A subset of younger-age-at-onset families from Utah (HLOD = 2.42) also gave evidence for linkage in this region (29).

Other Cancers

While the above criteria have proven useful for dealing with the locus heterogeneity problem, they don’t fully solve the problem. One additional strategy is that of examining high-risk prostate cancer families for an excess of other diseases and then using the resulting group of families in an isolated analysis.

Johannesson et al. have done that with high risk prostate-kidney cancer families (63). An association between these prostate and kidney cancer has been suggested by at least two studies. In a Swedish study, Grönberg and colleagues examined 1,364 relatives of 62 HPC families for the incidence of other cancers and found a significant association for kidney cancer (Standardized Incidence Ratio, SIR = 2.51; 95% CI 1.15–4.77) (64). A second Swedish study utilizing the nation-wide Family Cancer Database reported a familial relative risk of 1.3 (95% CI 1.0–1.76) for kidney cancer in offspring of prostate cancer cases (65).

We selected a set of 15 prostate-kidney families from among the 154 families on which we had completed a genome wide scan using 441 microsatellite markers (21) The 15 families all reported both primary prostate and primary kidney cancer. All kidney cancer cases were confirmed by either death certificate or medical records. To be eligible for the study, the kidney case had to be either a prostate cancer case himself or a first-degree relative of a prostate cancer case. There are 191 (99 genotyped) individuals, and all families were of Caucasian ancestry. Ten of the kidney cancer cases also had prostate cancer, and the other five kidney cancer cases were first-degree relatives of prostate cancer cases.

While we found no statistically significant evidence for linkage in the initial analysis, we found two regions of suggestive linkage at 11q12 and 4q21, with HLOD scores of 2.59 and 2.10, respectively. The primary result on chromosome 11 was strengthened after excluding two families with members with transitional cell carcinoma (TCC). This was a valid strategy, as 11 TCC is a very different and much more rare disease then in renal cell carcinoma, which was reported by the majority of families.

The non parametric analysis revealed a Kong and Cox p-value of 0.004 for marker D11S1290 at 11p11.2. The 8 cM region between 11p11.2 and 11q12.2 was refined by the addition of 16 additional markers. The subset of HPC families with a median age of diagnosis >65 years demonstrated the strongest evidence for linkage (HLOD = 2.50). The p-values from non-parametric analysis ranged from 0.004 to 0.05 across five contiguous markers. There are some provocative candidate genes in the area, including the gene for Prostate-Specific Membrane Antigen (PSMA). However mutation screening of that gene is made difficult by the presence of another gene, termed PSMA-Like, that is located at 11q14.3 and shares 98% homology with the coding sequence of the PSMA gene itself. Studies pursing this interesting question continue in our lab and others.

Studies of Isolated Populations

Studies of isolated populations have frequently been undertaken in cancer genetics as a way to both deal with locus heterogeneity as well as identify founder mutations for specific tumor suppressor genes. In the case of prostate cancer, studies to date have focused on populations from Iceland (66–68), Finland (69) and Ashkenazi Jewish men (70–74).

In our collaborative group, Friedrichsen et al. have examined a genome wide scan in a set of 36 families of Ashkenazi Jewish origin (75). The 36 Jewish families represent a combined dataset of 17 Jewish families from the Fred Hutchinson Cancer Research Center (FHCRC)-based PROGRESS dataset, and 19 Ashkenazi Jewish families collected at Johns Hopkins University (JHU). All available family members, including 94 affected men, were genotyped using a set of microsatellite markers distributed across the genome at an average of about 8–10 cM density. To combine the two datasets, only markers present in the UCSC genome browser April 2003 assembly (http://genome.ucsc.edu/) were used (PROGRESS 421 markers, JHU 398 markers) and map order and distance between markers were taken from the UCSC map.

Since no segregation analysis has been done exclusively on Ashkenazi Jewish men, and models of inheritance were thus hard to predict, the data were analyzed primarily using nonparametric multipoint methods. The strongest signal in this data set was a significant linkage peak at 7q11-21, associated with a nonparametric linkage (NPL) score of 3.01 (p = 0.0013). Simulation analysis indicate that this corresponds to a genome-wide empirical p = 0.006. Empirical p values were calculated using the computer program Merlin (76), which was used to generate and analyze 1,000 replicates of the entire genome from the original dataset of 36 Jewish families.

After genotyping additional markers within the 7q11-21 peak, the NPL score increased to 3.35 (p = 0.0004) at marker D7S634 with an allele-sharing LOD of 3.12 (p = 0.00007). Detailed SNP analysis is underway in an attempt to find a shared haplotype that is over represented among affected versus unaffected men within the Jewish families. Within that haplotype should lie the susceptibility gene and variant of interest.

We had noted a minor signal on chromosome 7q in our initial genome scan (21). We were thus curious as to the degree to which Jewish families accounted for that result. In the 254 PROGRESS families we previously reported an HLOD of 2.25 (LOD = 1.55) at marker D7S2212 on 7q21 using a recessive parametric model. In a nonparametric analysis we reported an NPL score of 1.79 (P = 0.038) in the same region. Analysis of the 237 non-Jewish families from the PROGRESS dataset yielded an NPL score of 1.11 (P = 0.134), revealing no evidence for linkage. This suggested clearly that the majority of result in the original genome-wide scan for the PROGRESS families was due to the presence of a modest number of Jewish families.

Similar conclusions were reached by investigators at JHU. In the genome-wide study of 188 JHU families by Xu et al., the strongest result on 7q was an allele-sharing LOD of 1.63 with marker D7S486 which is adjacent to the region of interest (25). When 17 of the 19 JHU Ashkenazi Jewish families were analyzed using D7S486, the allele-sharing LOD was only 0.04, suggesting that the Johns Hopkins-collected Ashkenazi Jewish families do not contribute significantly to the results previously reported for 7q22.

Our analysis of 36 Jewish families also highlighted regions on chromosomes 1q31-32, 2p11, 3q27-28, 14q12, and 20q11 with p values of 0.02 to 0.06. The strongest of these was at 14q12 (P = 0.02). Other minor peaks with an NPL p value ≤ 0.05 included 3q27-28 (P = 0.03) and 20q11 (p = 0.04). While these may represent other loci that contribute to prostate cancer in Jewish, and perhaps non Jewish families, further investigation is clearly needed to draw definitive conclusions.

Summary

Studies to date suggest that prostate cancer is a genetically very heterogeneous disease. High-risk families, in which multiple men are affected likely reflect the contributions of a number of genes, some that are rare and highly penetrant, while others are more common and weakly penetrant. In this review we have discussed only the first type of loci, and found the identification of such genomic regions be a formidable problem. Replication between seemingly similar data sets is weak, likely reflecting the older age of onset associated with the disease, the inability to collect affected individuals from more the two generations in a family, and the variation seen in disease presentation, in addition to the underlying locus heterogeneity. Indeed, the definition of prostate cancer is ever changing, as diagnostic criteria and tools for pinpointing early lesions improve.

Are we making progress? Clearly the answer is yes. The ability to divide large data sets into homogenous subset of families likely to share common genetic underpinnings has improved power to identify loci and reproducibility between loci is now more common. Indeed, several groups report linkage to loci on chromosomes 1, 17, 19 and 22. Key to our continued success is our ever increasing ability to understand the disease. Identifying the subset of men who are likely to get clinically significant disease is the goal of genetic studies like these, and identifying the underlying loci is key for developing diagnostics. The willingness of the community to share data prior to publication, combine datasets, and to work together has been an important factor in the successes the community has enjoyed to date, and will likely be as important as we move forward to untangle the genetics of this complex and common disorder.