Background: Variants in the gene encoding the macrophage scavenger receptor 1 (MSR14) protein have been identified in men with prostate cancer, and several small studies have suggested that the 999C>T (R293X) protein-truncating mutation may be associated with an increased risk for this disease.
Methods: Using large case-control, cohort, and prostate cancer family studies conducted in several Western countries, we tested for the 999C>T mutation in 2,943 men with invasive prostate carcinoma, including 401 males from multiple-case families, 1,982 cases unselected for age, and 575 men diagnosed before the age of 56 years, and in 2,870 male controls. Risk ratios were estimated by unconditional logistic regression adjusting for country and by a modified segregation analysis. A meta-analysis was conducted pooling our data with published data.
Results: The prevalence of MSR1*999C>T mutation carriers was 0.027 (SE, 0.003) in cases and 0.022 (SE, 0.002) in controls, and did not differ by country, ethnicity, or source. The adjusted risk ratio for prostate cancer associated with being a 999C>T carrier was 1.31 [95% confidence interval (CI), 0.93-1.84; P = 0.16]. The modified segregation analysis estimated the risk ratio to be 1.20 (95% CI, 0.87-1.66; P = 0.16). The risk ratio estimated from the meta-analysis was 1.34 (95% CI, 0.94-1.89; P = 0.10).
Conclusion: Our large-scale analysis of case and controls from several countries found no evidence that the 999C>T mutation is associated with increased risk of prostate cancer. The meta-analysis suggests it is unlikely that this mutation confers more than a 2-fold increased risk.
Many studies have shown that prostate cancer exhibits familial aggregation, but the genetic and shared nongenetic factors underlying the familial risk are unknown (1-3). Many studies have claimed linkage of prostate cancer to different genomic regions, but as yet no genes that are mutated in a significant proportion of multiple-case prostate cancer families have been conclusively identified (4, 5). It is possible that many loci with a wide spectrum of risks and different modes of inheritance are involved in genetic susceptibility to prostate cancer (2, 6).
Linkage to a region on chromosome 8p has been reported from analysis of 159 multiple-case families (heterogeneity logarithm of odds score, 1.84; P = 0.004; ref. 7) and has also been reported in a Swedish study (8). Subsequently, it was reported that variants in the gene encoding the macrophage scavenger receptor 1 (MSR1), situated within the linked region on 8p, were more frequent in men with prostate cancer than in controls (9). These variants included six rare missense mutations and one nonsense mutation, 999C>T. The latter variant, numbered according to GenBank accession number NM_138715, is predicted to result in a truncated protein (R293X) in such a fashion that the region that is critical for ligand binding would be lost. Initially, this protein-truncating mutation was detected in 8 of 317 (0.025) men with prostate cancer not meeting the clinical criteria for hereditary prostate cancer, but in only 1 (0.004) of 256 unaffected men, yielding a crude odds ratio of 6.6 [95% confidence interval (CI), 0.9-294; P = 0.08]. Among the hereditary prostate cancer families, the heterogeneity logarithm of odds for linkage to this region was 1.40 in families carrying a variant in MSR1 compared with 0.05 in those without, although families carrying the 999C>T mutation did not show formal evidence of linkage (P = 0.3; ref. 9). An increased prevalence of variants in cases was also found in a study of African Americans (10). Since then, a study of 438 affected males from families with multiple-case prostate cancer, 492 unselected prostate cancer cases, and 488 controls found the frequency of 999C>T mutation carriers to be 0.017, 0.028, and 0.033, respectively, providing no evidence for an increased risk (11). A further study from Finland also found no association between the R293X mutation and risk of prostate in cancer in either a hereditary or sporadic setting (12). To definitively confirm or refute the proposed association, we have genotyped the 999C>T (R293X) variant in large case-control, cohort, and prostate cancer family studies conducted in several Western countries.
Australia: Case-Control Study. Eligible cases were residents of Melbourne, Sydney, and Perth and identified from the population cancer registries, with histopathologically confirmed prostate cancer, excluding tumors with Gleason scores of <5, diagnosed before the age of 70 years with sampling stratified by age at diagnosis. Eligible controls were males identified through government electoral rolls and frequency matched to the age distribution of the cases (i.e., there is no statistical difference in the ages of cases and controls; refs. 13-15). For this study, 827 of the 832 cases and all of the 735 controls from whom blood was sampled were genotyped. The self-reported ethnicity of the cases and controls was 97% Caucasian.
Australia: Cohort Study. Cases and controls were identified from a prospective cohort study of 17,154 men ages 40 to 69 years at recruitment in 1990-1994 (16). For this study, all 469 incident cases identified up to June 2002 and 1,637 randomly selected controls were genotyped. The self-reported ethnicity of the cases and controls was 98% Caucasian (mean age, 61.8 years; range, 40-69 years). The controls were a random sample of all men in the cohort (mean age, 54.8 years; range, 40-69 years).
Australia: Early-Onset Case Series. Eligible cases were all men with histopathologically confirmed prostate cancer diagnosed before the age of 56 years identified from the Victorian Cancer Registry since 1999. For this study, 318 of 357 cases were genotyped. The self-reported ethnicity of the cases was 98% Caucasian.
United Kingdom: Unselected Case Series. Eligible cases were all UK-born men diagnosed with prostate cancer who attended the Urology Unit of the Royal Marsden NHS Trust, London, United Kingdom. For this study, 631 of 635 cases were genotyped. Twenty-one of these cases were also enrolled in the early-onset case-control study and are included in that study for the purposes of this analysis. The analysis was therefore based on 610 cases. Cases with known self-reported Afro-Caribbean origins were excluded because the UK controls were largely or entirely of White ethnic origins. The mean age of the cases was 69.0 (SD, 6.40).
United Kingdom: Early-Onset Case-Control Study. Eligible cases were all men with histopathologically confirmed prostate cancer diagnosed before the age of 56 years referred to a national study through collaborating general practitioners (mean age, 50.7 years; SD 3.90; ref. 17). Male controls were identified through the case's general practitioner, but not all cases could be matched. The controls were matched to the cases on age, and the only exclusion criterion was a previous diagnosis of prostate cancer. For this study, 259 of 267 cases diagnosed before the age of 56 and 186 of 189 controls available were genotyped. Two affected males also included in the ACTANE consortium set were also included in the latter study for the purposes of analysis.
United Kingdom: Spouse Controls. Eligible controls were the UK-born male spouses (n = 147) of cases enrolled in a UK population–based study of colorectal cancer (principal investigator, RSH). For this study, 141 males were genotyped. No ethnicity data were available. The mean age of these controls was 53.0 years (SD, 8.51).
Canada: Ashkenazi Jewish Case-Control Study. Eligible cases were prevalent invasive prostate cancers diagnosed in Ashkenazi Jewish men attending McGill University teaching hospitals in Montreal, Canada (18). For this study, 133 of 145 cases were genotyped. Eligible controls were men of Ashkenazi Jewish origin obtained anonymously from a genetic study of Quebec populations. For this study, 133 men were genotyped. The mean age at diagnosis of cases was 67.6 years (SD, 7.38). The ages of the controls are unavailable as they were anonymous at ascertainment.
The ACTANE Consortium: Multiple-Case Prostate Cancer Families
Multiple-Case Familial Prostate Cancer Series. Eligible subjects were males whose DNA samples were obtained because they belonged to families that had three or more cases of prostate cancer diagnosed at any age. The ascertainment of families with multiple cases of prostate cancer by an international consortium has been described previously in the context of a genome-wide search (19), wherein 65 families with the highest e-logarithm of odds scores were included. For this study, 180 affected (cases) and 34 unaffected males from 65 prostate cancer families were genotyped. The countries of origin of these families are shown in Table 1.
Familial Prostate Cancer Series. Eligible subjects, ascertained in Montreal, were males whose DNA samples were analyzed from Canadian and U.S. families that had multiple cases of prostate cancer but were not included in the above-mentioned genome-wide search. For this analysis, 229 affected males (cases) and 4 unaffected males (controls) from 156 families were genotyped. The ethnic origin of these cases was mixed White European.
The 999C>T mutation was analyzed either by restriction digestion analysis or by direct sequencing. One of four sets of primers was used to amplify the exon 6 region surrounding the variant (details of the primer sequences and conditions used are available from the corresponding author). When using restriction analysis as an analytic method, the PCR product was digested with the NlaIII restriction endonuclease (New England Biolabs, Beverly, MA) according to the manufacturer's instructions to generate smaller digested fragments in the presence of the 999C>T mutation. Digested products were visualized by agarose gel electrophoresis. Mutation carriers from the UK and Australian sets were confirmed by metaphor agarose gel electrophoresis (BioWhittaker, Walkersville, MD). A selection of UK and Australian samples heterozygous and homozygous for the wild-type alleles (as well as the single Australian sample that was homozygous for the rare allele) was also confirmed by sequencing. A subset of samples including all genotypes observed was analyzed with both methods for concordance: there were no discordant results. Comparative resequencing was done on PCR products purified with Multiscreen PCR 96-well plates (Millipore, Billerica, MA; Montreal samples). Sequencing reactions were done using the ABI Prism dRhodamine Terminator Cycle Sequencing Ready Reaction Kit (Applied Biosystems, Foster City, CA; UK samples) or the BigDye Terminator Cycle sequencing ready reactions kit (version 2.0, Montreal samples; Applied Biosystems, Foster City, CA). The products were analyzed on an ABI 377 Genetic Analyzer (UK samples) or on ABI 3700 automated DNA sequencers (Montreal samples; Applied BioSystems). The run files were processed using Sequencing Analysis software (version 3.6) and then aligned and compared using either Autoassembler 2.1 (Applied BioSystems) or PHRED and PHRAP (Montreal samples).
The association between the MSR1*999C>T mutation and prostate cancer was assessed using two methods. First, the estimated prostate cancer risk ratio associated with being a MSR1*999C>T carrier was estimated using logistic regression. To allow for potential differences in allele frequency among populations, the analysis was stratified by country (United Kingdom, Australia, Canada, Norway, and United States). Analyses were done using STATA version 7.0. To account for the fact that some individuals were related, 95% CIs for the risk ratio were computed using the Huber-White sandwich estimator, with the robust option in Stata.
Although the logistic regression analysis provides a valid test of the null hypothesis of no association, the estimate of the risk ratio is biased away from 1 because some cases were ascertained on the basis of having a family history. To adjust for this bias, the data were also analyzed by a modified segregation analysis (20) under a model described in terms of the risk ratio r and the allele frequencies pk (k = 1, …, 3). The incidence rates in country k (Australia, United Kingdom, or other) at age t were assumed to be λ0(t,c) in noncarriers and rλ0(t,c) in carriers. The rates λ0(t,c) were chosen at each stage so that the overall prostate cancer age-specific incidence rates agreed with national rates for the period 1988-1992 (21). This model was implemented using the pedigree analysis program MENDEL (22, 23).
A meta-analysis of our data combined with all other published data on the MSR1*999C>T mutation and risk of prostate cancer was conducted. Studies were identified by Medline searches using the terms MSR1, macrophage scavenger receptor, and prostate cancer. Where possible, we divided publications into their component substudies. Thus, there are three case-control series from one study (9) and three from the current study. The meta-analysis was conducted using the S-plus 6.1 statistical software package (Insightful Corp., Seattle, WA), using standard methods for combining the crude estimates of odds ratios based on the weighted sum of the log estimates with the inverse of the variance of the estimate as weight (24). The odds ratios for the different studies, with exact 95% CIs and P values, were calculated by using the StatXact statistical software package v.4.0.1 (Cytel Software Corp., Cambridge, MA).
Homogeneity in risk ratios across studies was evaluated by calculating the weighted (inverse of variance) sum of the squared differences between the log risk ratio estimates and the log of the pooled risk ratio estimate and assuming that this statistic follows a χ2 distribution with n − 1 degrees of freedom (where n = number of studies). Homogeneity in the genotype frequencies across studies was tested separately for cases and controls by using a χ2 test.
The overall prevalence of 999C>T mutation carriers was 0.027 (SE, 0.003) in cases and 0.022 (SE, 0.002) in controls. These prevalences did not differ by country (P = 0.6 and 1.0, respectively), ethnicity (P = 0.7 and 0.9, respectively), or source of subjects (P = 0.9 and 1.0, respectively). Moreover, the prevalence was similar in the UK cases <56 years old and spouse controls (P = 0.5) and in the controls from the Australian cohort and case-control studies (P = 0.9; Table 1). After adjusting for country, the estimated risk ratio from the logistic regression analysis was 1.31 (95% CI, 0.93-1.84; P = 0.12). The modified segregation analysis estimated the risk ratio to be 1.20 (95% CI, 0.87-1.66; P = 0.27; Table 2).
The logistic regression analysis showed no support for an association between the 999C>T mutation and prostate cancer within either the UK or Australian samples (Table 2). Using the modified segregation analysis in which the country-specific population rates are taken into consideration, the risk ratio estimated from the Australian data was 1.48 (95% CI, 0.98-2.21; P = 0.06), but this was not different from the combined estimate from the other studies of 0.88 (95% CI, 0.46-1.67; P = 0.7).
Table 3 and Figure 1 show the point estimates for the prevalence of mutation carriers in cases and controls and the subsequent risk ratio estimates for association with prostate cancer, based on combining our data with that from others identified within published studies. The pooled estimate was 1.34 (95% CI, 0.94-1.89; P = 0.10). There was no evidence for heterogeneity in risk ratios (P = 0.3), although there was evidence for heterogeneity of the frequency of mutation carriers among controls (P = 0.0004) but not among cases (P = 0.2). The meta-analysis does not suggest that there is a statistically significant excess of MSR1*999C>T carriers in cases compared with controls. Moreover, when the study of Xu et al. (9) that generated the initial hypothesis is excluded, the risk ratio decreases to 1.25 (95% CI, 0.88-1.78).
We have attempted to replicate the findings of the initial, and much smaller, study that found evidence for a strong association between the MSR1*999C>T mutation and risk of prostate cancer (9). We conducted an analysis on nearly 3,000 cases and over 2,800 controls from several countries and did a pooled analysis that combined our data with that identified in published studies to date. To our knowledge, this is the largest single allelic association study of prostate cancer. We found that there was no significant difference in the frequency of the MSR1*999C>T mutation between cases and controls, although the results from the Australian study almost achieve statistical significance (risk ratio, 1.48; 95% CI, 0.98-2.21; P = 0.06). The meta-analysis did not indicate that the MSR1*999C>T mutation is more prevalent in prostate cancer cases compared with controls. Overall, both our new analyses and those of the meta-analysis suggest that if the mutation is associated with an increased risk of prostate cancer, its effect on risk is not large and is unlikely to be more than 2-fold, as judged by the upper 95% limits of the confidence intervals.
Although it provides a valid test of the association, the risk ratio estimates from the standard case-control analysis using logistic regression are theoretically biased away from unity because some cases (but not controls) were selected on the basis of having a family history of prostate cancer. This bias, which would affect other similar studies, can be avoided by use of a segregation analysis approach that allows the family history of prostate cancer and genetic relationships between individuals to be taken into account naturally. In the analyses presented here, we used a simplified version that ignores other causes of familial aggregation of prostate cancer. This simplifying assumption is approximately equivalent to assuming that MSR1*999C>T and other risk factors combine multiplicatively (20). The risk ratio estimate from this approach (1.20) was slightly lower than that from the logistic regression approach (1.31).
The MSR1 gene was initially studied in the context of prostate cancer because of suggestive linkage to the region in which this gene lies on chromosome 8p22 (7). Evidence of linkage to this region has also been found by another group (8), although not by other linkage searches (4). Initially it was suggested that the 999C>T mutation, as well as other rare variants, may be implicated in prostate cancer risk (9). More recently, it has been suggested that common haplotypes at MSR1 may also be associated with risk (25). In this study, we focused only on the 999C>T mutation because it is predicted to truncate the protein and would likely result in loss of several important MSR1 protein domains. Our analysis does not, therefore, address the question as to whether other MSR1 variants are associated with prostate cancer risk. It does suggest that the magnitude of any real risk associated with the 999C>T mutation is smaller than originally thought and, importantly, that the contribution of this mutation to overall prostate cancer susceptibility is minimal given its rarity and the low upper bound of its confidence interval. If we assume a risk ratio of 1.34 and a population prevalence of 0.022, then the population attributable risk percent for prostate cancer in association with this mutation would be 0.74%. Moreover, this mutation cannot account for the positive logarithm of odds scores noted in the region around this gene. The most recent studies of this mutation in MSR1 are in broad agreement with our findings (10, 11).
One potential explanation for the differences in results of association studies in prostate cancer is the variation in prevalence of prostate-specific antigen testing, which results in a much higher incidence of disease with low Gleason score. In our study, we included series of cases and controls from countries where the prevalence of prostate-specific antigen screening was high (United States, Canada, where most cases are screen detected), low (United Kingdom, Norway) and intermediate (Australia). We found no evidence of any differences in MSR1 genotype frequencies in cases or risk ratios between countries, suggesting that prostate-specific antigen screening is not an explanation for the differences.
In the Australian case-cohort series, the controls were, on average, 7 years younger than the cases. A similar age difference was seen for the UK unselected cases and controls. If some of these controls were later destined to become cases within the next 10 years or so, it is possible that we may have underestimated the magnitude of the association between the 999C>T variant and prostate cancer risk in this series. However, the Australian case-control series was frequency matched on age, and 3.3% of the cases and 2.3% of the controls carried the variant. Moreover, because the risk of prostate cancer between the ages of 55 and 65 is approximately 1% in the populations we studied, we can assume that not more than 1% of the controls are misclassified. This would result in an underestimation of the odds ratio by <0.01, and such an error would not show in two decimal places.
Some groups of prostate cancer cases may be particularly likely to carry the 999C>T mutation, but it is not possible to identify whether such groups exist from inspection of the data provided here and in the meta-analysis. In particular, there is no evidence that either multiple-case families or early-onset cases are more likely to harbor this mutation (Tables 1 and 3).
Our analysis illustrates a problem inherent in studying so-called “low-penetrance” variants. By definition, they are not associated with large risks. If the population frequency of the supposedly at-risk genotype is very low (e.g., <2%) then very large studies will be required to detect such risks or do so with much precision. As has been shown in a recent meta-analysis of the Ser217Leu and Ala541Thr polymorphisms in the putative prostate cancer susceptibility gene ELAC2, it is often the studies with the largest sample sizes that find risk estimates closest to unity (15). This observation is supported by a formal analysis of 55 meta-analyses (26). In most meta-analyses studied, the largest study within each meta-analysis provided more conservative findings than were suggested by the overall result of the meta-analysis. This was true even if the result of the meta-analysis was itself statistically significant. Indeed, the first study published may be inherently likely to overestimate the effect size, sometimes by a considerable margin. This seems to be the case for both linkage (27) and association studies (28). These observations have limited the widespread acceptance of results from smaller series. Notably, as stated by Göring et al. (27), “joint estimation of locus position and effect simply does not work on the same data set, at least when power is … low….” As has been proposed by the International Collaborative Group for Prostate Cancer Genetics, large meta-analyses of existing data sets may be required to resolve these issues.
Progress in the genetics of prostate cancer susceptibility has been limited by the lack of universally accepted risk-associated genes. Thus far, only mutations in BRCA2 can be said to be clearly associated with a marked increase in prostate cancer risk (17), although occasional families with apparently disease-associated mutations in other genes such RNASEL and ELAC2 have been reported (5). There have been numerous genome-wide screens completed, with multiple regions of interest identified, but, thus far, none have led to the clear identification of a prostate cancer susceptibility locus (4). This is probably a consequence of considerable genetic heterogeneity, given that recent segregation analyses of prostate cancer favor multigene rather than a single gene model for susceptibility (6).
In conclusion, this very large analysis of case and control series from several countries found no evidence that the MSR1*999C>T mutation is associated with increased risk of prostate cancer. When combined with data from all published studies, a meta-analysis found, at best, marginal evidence that the mutation is associated with any risk and that it is unlikely this mutation confers more than a 2-fold increased risk of prostate cancer.
Grant support: U.S. Army grant DAMD17-00-1-0033 and the Canadian Genetic Diseases Network (W.D. Foulkes); NIH grant U01 CA89600 (W.D. Foulkes, R.A. Eeles, G.G. Giles, D.F. Easton, and J.L. Hopper), The Prostate Cancer Charitable Trust, Times Christmas Appeal, Institute of Cancer Research, Cancer Research UK, BREAKTHROUGH Breast Cancer, and a legacy of the late Marion Silcock (R.A.Eeles); National Health and Medical Research Council grant 126402 (J.L. Hopper); Tattersall's and the Whitten Foundation, the Cancer Council Victoria, National Health and Medical Research Council grant 20905, VicHealth grant 1999-0227 (J.L. Hopper and G.G. Giles); and Fonds de la recherche en Santé du Québéc Network of Applied Genetic Medicine (D. Sinnett and J. Simard).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Q. Hope, S. Bullock, and C. Evans contributed equally to this work. J. Simard holds a Canada Research Chair in Oncogenetics and D.F. Easton is a Principal Research Fellow of Cancer Research UK. K. Heimdal is currently in the Department of Medical Genetics, Rikshopitalet, University Hospital, Oslo, Norway. The list of Cancer Research UK/British Association of Urological Surgeons' Section of Oncology Collaborators are available on request.
We thank all the patients and their families for their participation in this study, the individuals who donated blood (controls; these individuals were identified by Penny Kelham and Artitaya Lophatananon, University of Nottingham), Sheila Seal and Anita Hall for kindly storing the samples that were provided, and Margaret Stevens for database management. We thank Kimberly Kotar and Mcgill University urologists for recruiting cases in Montreal.