Advertisement
Research Article

Reproducibility, Performance, and Clinical Utility of a Genetic Risk Prediction Model for Prostate Cancer in Japanese

  • Shusuke Akamatsu,

    Affiliations: Laboratory for Biomarker Development, Center for Genomic Medicine, RIKEN, Tokyo, Japan, Department of Urology, Graduate School of Medicine, Kyoto University, Kyoto, Japan

    X
  • Atsushi Takahashi,

    Affiliation: Laboratory for Statistical Analysis, Center for Genomic Medicine, RIKEN, Tokyo, Japan

    X
  • Ryo Takata,

    Affiliations: Laboratory for Biomarker Development, Center for Genomic Medicine, RIKEN, Tokyo, Japan, Department of Urology, Iwate Medical University, Morioka, Japan

    X
  • Michiaki Kubo,

    Affiliation: Laboratory for Genotyping Development, Center for Genomic Medicine, RIKEN, Yokohama, Japan

    X
  • Takahiro Inoue,

    Affiliation: Department of Urology, Graduate School of Medicine, Kyoto University, Kyoto, Japan

    X
  • Takashi Morizono,

    Affiliation: Laboratory for Medical Informatics, Center for Genomic Medicine, RIKEN, Yokohama, Japan

    X
  • Tatsuhiko Tsunoda,

    Affiliation: Laboratory for Medical Informatics, Center for Genomic Medicine, RIKEN, Yokohama, Japan

    X
  • Naoyuki Kamatani,

    Affiliation: Laboratory for Statistical Analysis, Center for Genomic Medicine, RIKEN, Tokyo, Japan

    X
  • Christopher A. Haiman,

    Affiliation: Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, California, United States of America

    X
  • Peggy Wan,

    Affiliation: Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, California, United States of America

    X
  • Gary K. Chen,

    Affiliation: Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, California, United States of America

    X
  • Loic Le Marchand,

    Affiliation: Epidemiology Program, Cancer Research Centre, University of Hawaii, Honolulu, Hawaii, United States of America

    X
  • Laurence N. Kolonel,

    Affiliation: Epidemiology Program, Cancer Research Centre, University of Hawaii, Honolulu, Hawaii, United States of America

    X
  • Brian E. Henderson,

    Affiliation: Department of Preventive Medicine, Keck School of Medicine, University of Southern California, Los Angeles, California, United States of America

    X
  • Tomoaki Fujioka,

    Affiliation: Laboratory for Statistical Analysis, Center for Genomic Medicine, RIKEN, Tokyo, Japan

    X
  • Tomonori Habuchi,

    Affiliation: Department of Urology, Akita University School of Medicine, Akita, Japan

    X
  • Yusuke Nakamura,

    Affiliation: Laboratory of Molecular Medicine, Human Genome Center, Institute of Medical Science, the University of Tokyo, Tokyo, Japan

    X
  • Osamu Ogawa,

    Affiliation: Department of Urology, Graduate School of Medicine, Kyoto University, Kyoto, Japan

    X
  • Hidewaki Nakagawa mail

    hidewaki@ims.u-tokyo.ac.jp

    Affiliation: Laboratory for Biomarker Development, Center for Genomic Medicine, RIKEN, Tokyo, Japan

    X
  • Published: October 10, 2012
  • DOI: 10.1371/journal.pone.0046454

Abstract

Prostate specific antigen (PSA) is widely used as a diagnostic biomarker for prostate cancer (PC). However, due to its low predictive performance, many patients without PC suffer from the harms of unnecessary prostate needle biopsies. The present study aims to evaluate the reproducibility and performance of a genetic risk prediction model in Japanese and estimate its utility as a diagnostic biomarker in a clinical scenario. We created a logistic regression model incorporating 16 SNPs that were significantly associated with PC in a genome-wide association study of Japanese population using 689 cases and 749 male controls. The model was validated by two independent sets of Japanese samples comprising 3,294 cases and 6,281 male controls. The areas under curve (AUC) of the model were 0.679, 0.655, and 0.661 for the samples used to create the model and those used for validation. The AUCs were not significantly altered in samples with PSA 1–10 ng/ml. 24.2% and 9.7% of the patients had odds ratio <0.5 (low risk) or >2 (high risk) in the model. Assuming the overall positive rate of prostate needle biopsies to be 20%, the positive biopsy rates were 10.7% and 42.4% for the low and high genetic risk groups respectively. Our genetic risk prediction model for PC was highly reproducible, and its predictive performance was not influenced by PSA. The model could have a potential to affect clinical decision when it is applied to patients with gray-zone PSA, which should be confirmed in future clinical studies.

Introduction

Prostate cancer (PC) is the most frequent non-cutaneous malignancy in the Western countries [1], and its incidence has also been rapidly increasing in Asian countries including Japan [2]. Although environmental factors such as fat-rich or high-calorie diet intake may play some important roles in PC carcinogenesis, the only definitive risk factors for PC are age, family history, and ethnicity [3], [4]. Prostate specific antigen (PSA) is widely used as a diagnostic biomarker for PC. Patients with PSA above the cutoff value of 2–4 ng/ml are suspected to have PC and undergo invasive transrectal or transperineal prostate needle biopsies. However, the predictive performance of PSA is recognized to be unfavorable, especially for patients with ‘gray-zone’ PSA of less than 10 ng/ml. Only 20–25% of the patients with gray-zone PSA are histologically diagnosed with PC at biopsy [5], [6], [7]. Hence, many patients without PC suffer from the harms of unnecessary prostate biopsies such as pain, hematuria, rectal bleeding, prostatitis, and sepsis [8]. The usefulness of compensatory indexes such as PSA velocity, PSA density and free to total PSA ratio are limited due to their dependence on PSA [5], [9], and performances of other biomarkers such as PCA3 are still under investigation [10]. Therefore, novel biomarkers that can risk stratify patients at gray-zone PSA to decide who should be recommended to undergo prostate biopsies are urgently needed.

Recently, more than forty genomic loci associated with susceptibility to PC have been identified by genome-wide association studies (GWAS) [11], [12], [13], [14], [15], [16], [17], [18], [19], [20]. Besides identifying new genes or pathways to clarify the etiology of diseases, information gained from GWAS can also be utilized to estimate predispositions to developing diseases [21]. However, the effect size, or odds ratio (OR) of common susceptibility variants identified by GWAS are generally small (1.1–2.0), and the utility of each variant in estimating disease susceptibility is limited. Therefore, efforts have been made to develop genetic risk prediction models incorporating multiple susceptibility variants [22].

We have previously identified five novel loci associated with PC in the GWAS of Japanese, and also reported that although 19 of the 31 then reported loci were replicated (P-value<0.05) in the Japanese, 12 were not, confirming the presence of ethnic heterogeneity in genetic susceptibility to PC [20]. Some of the five novel loci we have identified were later replicated in a Caucasian population at lower odds ratios and some were not [23], [24]. These data suggest that genetic risk prediction models should be individualized to each ethnic group. In the present study, we have created a genetic risk prediction model for PC based on 16 genetic variants that were associated with PC in the GWAS of Japanese, and confirmed its reproducibility in two independent sets of Japanese samples. We have also tested the performance of the model in a clinical scenario and estimated whether the model could be clinically and practically useful to risk-stratify patients at gray-zone PSA.

Materials and Methods

Genetic Variant selection and creation of a genetic risk prediction model

Of the 31 PC susceptibility germline variants reported by July 2010, 15 variants passed the threshold of significance after Bonferroni correction (P-value<1.6×10−3) in our GWAS stage 1, including multiple variants at 8q24 (Table S1) [20]. There are five independent regions associated with PC in 8q24 [25]. The variants that showed the strongest association in the Japanese GWAS for each of the five regions were selected to be included in the model. We have previously fine-mapped so-called Region 2 of 8q24 [26], and rs1456315, which showed the strongest association with PC (P = 2.00×10−24, OR = 1.74) in the Japanese in this region was chosen for Region 2 of 8q24. Five novel variants we have identified in the Japanese GWAS [20] were also included, and the total of 16 variants were incorporated into a genetic risk prediction model for PC as explanatory variables in a logistic regression model (Table 1).No variants were within the same linkage disequilibrium block, and all the variants selected were not correlated with (r2<0.2) each other. For creation of a risk prediction model, each sample was scored for each of the 16 variants with the number of risk alleles (0, 1, and 2). A genetic risk prediction model was created by unconditional logistic regression incorporating all 16 SNPs. Odds ratio (OR) was estimated for each sample based on the model from the following formula:
where are the regression coefficients of the each SNP and are the number of the risk alleles at the each SNP locus. We also evaluated our genetic prediction model by two independent Japanese sample sets not used for the model construction.

thumbnail

Table 1. 16 SNPs incorporated in the risk prediction model and their regression coefficients in the model.

doi:10.1371/journal.pone.0046454.t001

Samples and genotyping

The basic characteristics of the study participants are summarized in Table 2. The samples that were used to create the model (AKY) were drawn from Akita-Kyoto Cohort (AKYC) including 732 cases and 957 controls [27]. The case samples were recruited from Japanese PC patients at Kyoto University and Akita University. All case samples were histologically diagnosed by local pathologists, and all clinical data were collected by local urologists. The control samples of AKYC included Japanese samples collected from the male patients visiting the urology clinics at the two university hospitals with diseases other than PC and other malignancies, and from healthy male volunteers visiting for medical checkups. The control patients had at least one PSA tests every 1–2 years along with digital rectal exams. Although the exact PSA data for each sample is missing, patients with PSA>4 ng/mL were not included in the current study controls except those who received prostate needle biopsy along with surgery due to benign prostatic hyperplasia, and had no malignancy in the pathological specimen. AKYC samples were genotyped by multiplex PCR-based invader assays [28]. Sample call rates were >99%, and there was no deviation from Hardy-Weinberg equilibrium for all the 16 SNPs among the controls. After genotyping, 689 cases with full genotype and age data and 749 controls that were matched by age groups of 5-year intervals were drawn to create the genetic risk prediction model. The samples used for the 1st validation study (BBJ) were drawn from stage 2 of our GWAS of Japanese [20]. All PC cases and male controls were obtained from BioBank Japan at the Institute of Medical Science, the University of Tokyo [29]. This project was started in 2003 to collect a total of 300,000 cases that have at least one of the 47 diseases by a collaborative network of 66 hospitals located at all areas of Japan. Of the 3,001 case and 5,415 male control samples used for the replication study in the GWAS, full genotype data was available for the 16 SNPs in 2,950 cases and 5,235 controls. The 2nd validation study (BBJ2) is a nested case-control study using sample drawn from BioBank Japan. The 344 cases of BBJ2 had been enrolled in BioBank Japan as patients with diseases other than PC during 2003–2007 and were subsequently diagnosed with PC during follow-up periods (at most 7 years). The 1,045 controls of BBJ2 were drawn from 14,541 male samples used in GWAS for 18 diseases other than PC. The controls were matched by age groups of 5-year intervals and by the follow-up years with the cases: control ratio of 1:3. The BBJ2 case samples were genotyped for the 16 SNPs by multiplex PCR-based invader assays. The BBJ2 control samples were genotyped using Illumina HumanHap550v3 or Human 610-Quad Bead Chip. For the BBJ2 case and control samples, baseline-PSA levels were measured using their sera pooled at the time of the registration to BioBank Japan by chemiluminescent enzyme immunoassay (CLEIA).

thumbnail

Table 2. Clinical characteristics of the cases and the controls.

doi:10.1371/journal.pone.0046454.t002

Ethics Statement

The research project and the sample collection were approved by the ethical committees in the Institute of Medical Science, the University of Tokyo, Yokohama Institute of RIKEN, Kyoto University, Akita University, University of Southern California, and University of Hawaii. Written informed consent was obtained from all individuals.

Receiver operating characteristic (ROC) analysis

We used R statistical environment version 2.61 for analysis. R package Epi and pROC were used to draw ROC plots and calculate areas under curve (AUCs). AUCs were statistically compared by paired or unpaired DeLong's test.

Calculation of positive prostate needle biopsy rates at given cutoffs

Assuming the positive prostate needle biopsy rate before genetic risk prediction to be 20% [5], [6], [7], positive prostate biopsy rate for the risk population, or positive predictive value is (0.2 SE)/{0.2 SE+0.8 (1-SP)} and positive prostate biopsy rate for the non-risk population, or (1-negative predictive value) is 0.2 (1-SE)/{0.2 (1-SE)+0.8 SP}, where SE and SP are sensitivity and specificity of the genetic prediction model at an arbitrary cutoff value.

Results

Reproducibility and discriminatory potential of the genetic risk prediction model

The regression coefficients of the risk prediction model were positive for all the 16 variants except rs620861, indicating that the risk alleles of the variants selected based on stage1 of our GWAS are positive risk factors in AKY samples except at rs620861 (Table 1). rs620861 is in the so-called Block 3 of 8q24, and was first reported as a PC-susceptibility locus in a GWAS of European ancestry, and was confirmed to be associated with PC in stage 1 of our GWAS of the Japanese. Although this variant showed negative association with PC in AKY samples, the association was mildly positive in BBJ and BBJ2 as well as in the meta-analysis of the three sample sets used in the present study (Table S2), indicating that the negative association of this variant with PC risk in AKY samples may be due to sample collection bias.

The AUC and the 95% confidence interval (CI) of the risk prediction model was 0.679 (0.651–0.706) for the model constructing sample set AKY. We applied this PC risk prediction model to two other independent Japanese sample sets, BBJ and BBJ2, and the AUCs were 0.655 (0.643–0.668) for BBJ, 0.661 (0.628–0.693) for BBJ2, and 0.659 (0.649–0.67) when all the three sample sets were combined (Figure 1). There were no statistically significant differences in AUCs between the sample sets, indicating high reproducibility of the predictive performance of this PC risk prediction model in the three independent sample sets. The risk prediction model also showed similar reproducibility (AUC 0.655, 95%CI 0.631–0.679) when it was applied to a sample set of Japanese from the Multiethnic Cohort in Hawaii and California (26) (MEC, 980 cases and 1,005 controls) (Figure S1).

thumbnail

Figure 1. ROC curves for (A) the samples used to create the genetic risk prediction model (AKY), (B)(C) the two validation studies (BBJ and BBJ2), and (D) when all the samples were combined.

The AUCs of the model and 95% confidence intervals are indicated. The AUCs of BBJ and BBJ2 are statistically compared with that of AKY, and P-values are reported. (x: the logarithm of odds ratio at the best cut off, Sens: sensitivity of the model at the best cutoff, Spec: specificity of the model at the best cutoff.)

doi:10.1371/journal.pone.0046454.g001

When we examined the distribution of ORs estimated from the risk prediction model in all the samples used in the present study (BBJ, BBJ2, and AKY), there was almost ten-fold difference in ORs between those in the top and bottom 5 percentile of the population (Figure S2A). The ORs of the case samples were modestly but significantly higher than the controls (Figure S2B).

The predictive performance of the genetic risk model is not influenced by serum PSA level

We next evaluated the predictive performance of the model in the samples with serum PSA 1–10 ng/ml, where decisions to proceed to prostate needle biopsies are often difficult to make. Since serum PSA data of the control samples were not available in AKY and BBJ, the predictive performance was analyzed using the case samples with PSA 1–10 ng/ml and all the controls. In the BBJ2 nested case-control study samples, serum PSA level was measured for all the case and control samples, and those with PSA 1–10 ng/ml were used for analysis. As a result, AUCs of the risk prediction model were 0.676 (0.642–0.71) for AKY, 0.643 (0.617–0.669) for BBJ, and 0.655 (0.608–0.703) for BBJ2 (Figure 2). These data were not statistically different from the AUCs observed when the samples were not confined to those with PSA 1–10 ng/ml, indicating that the predictive performance of the risk model is not significantly influenced by serum PSA level. The predictive performance of the risk model was also not influenced by serum PSA level in MEC (AUC 0.683, 95% CI 0.631–0.734 for the samples with PSA 1–10 ng/ml) (Figure S1). Furthermore, there was no correlation between PSA and the OR calculated from our risk prediction model (r2 = 0.02) (Figure S3)

thumbnail

Figure 2. ROC curves of the genetic risk prediction model when the samples were confined to those with PSA 1–10 ng/ml.

In (A) AKY and (B) BBJ, case samples with PSA 1–10 ng/ml and all the control samples are used for analysis. In (C) BBJ2, samples are confined to those with PSA 1–10 ng/ml in both case and control samples. The AUCs of the model and 95% confidence intervals are indicated. In each of the sample sets, the AUCs are statistically compared to the AUCs when serum PSA level is not confined (as shown in Figure 1), and P-values are reported.

doi:10.1371/journal.pone.0046454.g002

Application of the genetic prediction model in a clinical scenario

Clinical and practical utilities of biomarkers cannot be assessed only by sensitivity, specificity, or AUC. Prevalence of the disease to be tested or pre-test probability also needs to be considered. To assess the clinical utility of the genetic risk prediction model in individuals with PSA gray-zone, we have calculated the probabilities of positive prostate needle biopsies in genetically high and low risk groups at various cut-off OR values assuming that the overall probability of positive prostate needle biopsies is 20%, a typical value in patients with gray-zone PSA [5], [6], [7]. As a result, there was more than two fold difference in positive prostate biopsy rates between the high and low risk groups at any cutoff OR value (Table 3). At the cutoff OR of 0.5, 24.2% of the population was classified as low risk, and the probability of a positive prostate biopsy was 10.7%. On the other hand, at the cutoff OR of 2.0, 9.7% of the population was classified as high risk, and the probability of a positive biopsy was 42.4%. These estimations indicate that although the AUC of the genetic prediction model is not high as a biomarker, it can be clinically and practically useful when it is applied in a limited situation where pre-test probability is relatively high, such as in PSA gray-zone.

thumbnail

Table 3. Probability of positive prostate biopsy in high and low risk patients grouped by the genetic risk model at various cutoffs.

doi:10.1371/journal.pone.0046454.t003

Discussions

PSA is a protein secreted specifically from the prostate gland, and has been widely accepted as a serum biomarker for PC. However, other medical conditions, such as benign prostatic hypertrophy and inflammation can cause serum PSA elevation [30]. Hence, the diagnostic specificity of PSA is quite low, especially at boarder-line levels of PSA, or ‘gray-zone’. Patients suspected to have PC by PSA screening usually undergo prostate needle biopsy, which is an invasive procedure that accompany complications, some of which are severe. In addition, recent randomized controlled trials have shown no or little benefit of PSA screening in extending cancer-specific survival [31], [32]. Economic burden of prostate needle biopsies, followed by overdiagnosis and overtreatment for PC, is another serious issue since it is estimated that each year, more than one million patients undergo prostate needle biopsies in the US, a procedure which costs $500–1,000 for each [33], [34]. Therefore, there is a world-wide controversy over PSA screening, and additional biomarkers which can better identify the patients that need prostate needle biopsies are definitely required.

The risk of PC is 2.5 times higher in the patients with a positive family history of PC in their first degree relatives [35]. However, only a small proportion of patients have known positive family history of PC in Asians including the Japanese [36], [37], and collection of detailed family history is often problematic. In our GWAS of the Japanese which includes 4,584 PC, only 6.6% of the patients had positive family history of PC, and data was missing in 15.5% of the patients. Contrarily, risk prediction models based on easily accessible genetic information can be applied to the general population. Furthermore, unlike other biomarkers which show some degree of fluctuation that could affect their reproducibility, genetic risk scores are stable in each individual. So far, few studies have studied the reproducibility of genetic risk prediction models using independent sets of samples. Although reproducibility of genetic risk prediction models could be affected by sample collection bias, using independent sets of samples we have shown that our model is highly reproducible in a single ethnic group. In fact, our model showed similar reproducibility when it was applied to a sample set of Japanese from the Multiethnic Cohort in Hawaii and California (MEC). These data warrant similar predictive performance of our model in other Japanese population or East Asian population as well.

Genetic risk prediction of PC was first reported using only five common susceptibility variants [38]. The model was established by simply counting the number of risk alleles. Subsequently, models incorporated increasing number of variants, and logistic regression models were adapted to account for the effect size of each variant [39], [40], [41]. In the present study, we have also created a model based on step-wise logistic regression analysis. When compared to the model incorporating 9 of the 16 variants that remained significant in a step-wise model, the predictive performance of the model including all the 16 variants was modestly but statistically significantly superior (Figure S4). Over forty PC susceptibility variants have been reported so far, and still more remains to be identified. Inclusion of novel variants newly identified to be associated with PC is expected to further improve the predictive performance of the model, although careful selection of the variants based on association studies in each ancestry group is necessary. Of note, even when we have included only 16 PC susceptibility variants identified to be associated with PC in GWAS of Japanese, one variant showed negative association with PC in the samples used to create the risk prediction model, suggesting that a variant with very mild effect would show either positive or negative association in different sample sets even in a single ancestry group, which may lower the predictive performance of the model.

There is still a large debate over the clinical utility of genetic risk prediction models. The overall predictive performances of genetic risk prediction models as assessed by ROC analysis are usually modest, since the distribution of the ORs between the case and controls largely overlap. However, it has been implicated in breast cancer that genetic risk prediction models could be clinically useful among a subset of high risk patients [42]. In case of PC, patients can be risk-stratified using PSA, and genetic risk prediction models can be a useful compensatory marker at gray-zone PSA, where patients have relatively high risk of PC, and the diagnostic ability of PSA is the lowest. Furthermore, PCs are generally slow growing, and even if patients with PC are false negatively classified as low risk by a genetic prediction model, they can still be followed with serial PSA measurements, and can have prostate biopsy with increasing PSA before reaching advanced stages except in rare cases of very aggressive tumor. Identification of aggressive PCs is another important issue in PC diagnosis. Most of the PC susceptibility variants identified by GWAS have fallen short of discriminating aggressive from non-aggressive PCs [43], and there was no significant difference in the distribution of ORs between the aggressive and non-aggressive PCs in our genetic risk prediction model as well (Figure S5). Additional biomarkers that can discriminate aggressive and indolent PCs should be explored.

The circumstance where biomarkers are applied is also important in assessing their clinical values. When the prevalence of a disease is very low, even a marker with very high sensitivity and specificity shows low positive predictive value. Since the overall predictive value of genetic risk prediction models are low, it is important to utilize the model in a situation where disease prevalence or pre-test probability is relatively high. We have demonstrated that the predictive performance of the genetic risk prediction model is not affected by serum PSA level, and that the model can be utilized in the patients with gray-zone PSA, where there is usually 20–25% probability of a positive prostate needle biopsy. Recently, a few common variants have been reported to be associated with serum PSA level [44], and two of them, rs10993994 at 10q11 and rs7501939 at 17q12 were included in our genetic risk prediction model. It is presumed that the small effect size of these variants on PSA level did not significantly affect overall predictive performance of our risk prediction model at gray-zone PSA. Some other studies that have evaluated a model incorporating PSA and genetics have argued that there was only minor increase in AUC when genetic information was added [45]. However, PSA fluctuation is problematic in PSA gray-zone, and it is possible that the true effect of combining PSA with genetic risk information is confounded by these variances in PSA.

We have shown that while the genetic risk model may not be helpful clinically in all the patients with gray-zone PSA, it may largely influence decision making in a portion of patients. In our clinical simulation, 24.2% of the patients had OR<0.5, and these patients had 10.7% chance of being positive after prostate needle biopsy. Considering the complications of prostate needle biopsies, these patients might chose serial PSA follow-up rather than immediate prostate needle biopsy. On the other hand, 9.7% of the patients with OR>2, who have more than 42.4% chance of being positive for prostate cancer, may choose to undergo immediate prostate needle biopsy. Although the genetic risk prediction model should further be evaluated prospectively in clinics, our data suggests that it can be an additional biomarker that can risk-stratify individuals at gray-zone PSA in Japanese, leading to personalized medicine.

Conclusions

We have created a genetic risk prediction model based on 16 common variants shown to be associated with PC in Japanese GWAS. The model was highly reproducible and its predictive performance was not affected by PSA. The model has a potential to be a clinically useful biomarker that can risk stratify patients at gray-zone PSA, which should be confirmed in future prospective studies.

Supporting Information

Table S1.

Summary results of the previously reported SNPs associated with prostate cancer susceptibility in GWAS of the Japanese.

doi:10.1371/journal.pone.0046454.s001

(DOCX)

Table S2.

The results of association study for rs620861.

doi:10.1371/journal.pone.0046454.s002

(DOCX)

Figure S1.

Reproducibility of the model in MEC Japanese samples. ROC curves of the risk prediction model in a set of Japanese samples from the Multiethnic Cohort in Hawaii and California (MEC). Right panel shows the ROC curve when all the samples were used. Left panel shows the ROC curve for the samples with PSA 1–10 ng/ml (only the samples whose serum PSA levels were known before genotyping are included). The AUCs of the model and 95% confidence intervals are indicated.

doi:10.1371/journal.pone.0046454.s003

(TIF)

Figure S2.

Distribution of odds ratio estimated from the prediction model. (A) Distribution of odds ratio in all the case and control samples used for analysis in the present study. Odds ratio at each percentile of population is described. (B) Distribution of odds ratio in case (red) and control (blue) samples are plotted. The difference in odds ratio between case and controls are statistically compared by Student's t-test, and the P-value is reported.

doi:10.1371/journal.pone.0046454.s004

(TIF)

Figure S3.

Correction between serum PSA level and OR. Serum PSA level and OR are plotted for all the case and control samples with PSA 1–10 ng/ml, and the correlation coefficient (R2) is indicated.

doi:10.1371/journal.pone.0046454.s005

(TIF)

Figure S4.

A risk prediction model created by step-wise logistic regression. ROC analysis of a risk prediction model created by step-wise logistic regression. 9 variants that remained significant were incorporated into the model. The ROC plot is based on the combined data of all three sample sets used in the present study. The AUC of the model and 95% confidence intervals are indicated. The AUC of the model is statistically compared with that of the model created by unconditional logistic regression based on 16 SNPs by DeLong's test, and the P-value is reported.

doi:10.1371/journal.pone.0046454.s006

(TIF)

Figure S5.

The distribution of odds ratio in high and non-high risk tumors. Distribution of odds ratio in AKY are plotted separately in high (red) and non-high (blue) risk tumors. High risk tumors fulfill either of the following criteria; PSA≥20, or GS≥8, or clinical stage≥C. Odds ratio between high and non-high risk tumors are statistically compared by Student's t-test, and the P-value is reported. Y-axis shows cumulative frequency.

doi:10.1371/journal.pone.0046454.s007

(TIF)

Acknowledgments

We thank Ms.Yoko Mitobe in Department of Urology, Akita University and Ms. Megumi Kishida in Department of Urology, Kyoto University for sample management and clinical data collection. We are also grateful to the members of the BioBank Japan project and the Rotary Club of Osaka-Midosuji District 2660 Rotary International in Japan for supporting our study.

Author Contributions

Conceived and designed the experiments: SA RT YN OO HN. Performed the experiments: SA AT TI CAH. Analyzed the data: SA AT RT TM TT NK CAH PW GKC. Contributed reagents/materials/analysis tools: TI MK TM TT NK CAH LLM LNK BEH TF TH YN OO. Wrote the paper: SA AT CAH HN.

References

  1. 1. Grönberg H (2003) Prostate cancer epidemiology. Lancet 361: 859–864. doi: 10.1016/s0140-6736(03)12713-4
  2. 2. Matsuda T, Saika K (2009) Comparison of time trends in prostate cancer incidence (1973–2002) in Asia, from cancer incidence in five continents, Vols IV–IX. Jpn J Clin Oncol 39: 468–469. doi: 10.1093/jjco/hyp077
  3. 3. Bostwick DG, Burke HB, Djakiew D, Euling S, Ho SM, et al. (2004) Human prostate cancer risk factors. Cancer 101: 2371–2490. doi: 10.1002/cncr.20408
  4. 4. Pienta KJ, Esper PS (1993) Risk factors for prostate cancer. Ann Intern Med 118: 793–803. doi: 10.7326/0003-4819-118-10-199305150-00007
  5. 5. Loeb S, Catalona WJ (2007) Prostate-specific antigen in clinical practice. Cancer Lett 249: 30–39. doi: 10.1016/j.canlet.2006.12.022
  6. 6. Ito K, Ichinose Y, Kubota Y, Imai K, Yamanaka H (1997) Clinicopathological features of prostate cancer detected by transrectal ultrasonography-guided systematic six-sextant biopsy. Int J Urol 4: 474–479. doi: 10.1111/j.1442-2042.1997.tb00288.x
  7. 7. Ito K, Ohi M, Yamamoto T, Miyamoto S, Kurokawa K, et al. (2002) The diagnostic accuracy of the age-adjusted and prostate volume-adjusted biopsy method in males with prostate specific antigen levels of 4.1–10.0 ng/mL. Cancer 95: 2112–2119. doi: 10.1002/cncr.10941
  8. 8. Wolf AM, Wender RC, Etzioni RB, Thompson IM, D'Amico AV, et al. (2010) American Cancer Society guideline for the early detection of prostate cancer: update 2010. CA Cancer J Clin 60: 70–98. doi: 10.3322/caac.20066
  9. 9. Vickers AJ, Savage C, O'Brien MF, Lilja H (2009) Systematic review of pretreatment prostate-specific antigen velocity and doubling time as predictors for prostate cancer. J Clin Oncol 27: 398–403. doi: 10.1200/jco.2008.18.1685
  10. 10. Auprich M, Augustin H, Budäus L, Kluth L, Mannweiler S, et al. (2011) A comparative performance analysis of total prostate-specific antigen, percentage free prostate-specific antigen, prostate-specific antigen velocity and urinary prostate cancer gene 3 in the first, second and third repeat prostate biopsy. BJU Int 109: 1627–1635. doi: 10.1111/j.1464-410x.2011.10584.x
  11. 11. Amundadottir L, Sulem P, Gudmundsson J, Helgason A, Baker A, et al. (2006) A common variant associated with prostate cancer in European and African populations. Nat Genet 38: 652–658. doi: 10.1038/ng1808
  12. 12. Haiman CA, Chen GK, Blot WJ, Strom SS, Berndt SI, et al. (2011) Genome-wide association study of prostate cancer in men of African ancestry identifies a susceptibility locus at 17q21. Nat Genet 43: 570–573. doi: 10.1038/ng.839
  13. 13. Yeager M, Chatterjee N, Ciampa J, Jacobs KB, Gonzalez-Bosquet J, et al. (2009) Identification of a new prostate cancer susceptibility locus on chromosome 8q24. Nat Genet 41: 1055–1057. doi: 10.1038/ng.444
  14. 14. Gudmundsson J, Sulem P, Gudbjartsson D, Blondal T, Gylfason A, et al. (2009) Genome-wide association and replication studies identify four variants associated with prostate cancer susceptibility. Nat Genet 41: 1122–1126. doi: 10.1038/ng.448
  15. 15. Thomas G, Jacobs K, Yeager M, Kraft P, Wacholder S, et al. (2008) Multiple loci identified in a genome-wide association study of prostate cancer. Nat Genet 40: 310–315. doi: 10.1038/ng.91
  16. 16. Eeles R, Kote-Jarai Z, Al Olama A, Giles G, Guy M, et al. (2009) Identification of seven new prostate cancer susceptibility loci through a genome-wide association study. Nat Genet 41: 1116–1121. doi: 10.1038/ng.450
  17. 17. Eeles R, Kote-Jarai Z, Giles G, Olama A, Guy M, et al. (2008) Multiple newly identified loci associated with prostate cancer susceptibility. Nat Genet 40: 316–321. doi: 10.1038/ng.90
  18. 18. Kote-Jarai Z, Olama AA, Giles GG, Severi G, Schleutker J, et al. (2011) Seven prostate cancer susceptibility loci identified by a multi-stage genome-wide association study. Nat Genet 43: 785–791. doi: 10.1038/ng.882
  19. 19. Gudmundsson J, Sulem P, Rafnar T, Bergthorsson J, Manolescu A, et al. (2008) Common sequence variants on 2p15 and Xp11.22 confer susceptibility to prostate cancer. Nat Genet 40: 281–283. doi: 10.1038/ng.89
  20. 20. Takata R, Akamatsu S, Kubo M, Takahashi A, Hosono N, et al. (2010) Genome-wide association study identifies five new susceptibility loci for prostate cancer in the Japanese population. Nat Genet 42: 751–754. doi: 10.1038/ng.635
  21. 21. Nakagawa H, Akamatsu S, Takata R, Takahashi A, Kubo M, et al. (2012) Prostate cancer genomics, biology, and risk assessment through genome-wide association studies. Cancer Sci 103: 607–613. doi: 10.1111/j.1349-7006.2011.02193.x
  22. 22. Janssens AC, van Duijn CM (2008) Genome-based prediction of common diseases: advances and prospects. Hum Mol Genet 17: R166–173. doi: 10.1093/hmg/ddn250
  23. 23. Batra J, Lose F, Chambers S, Gardiner RA, Aitken J, et al. (2011) A replication study examining novel common single nucleotide polymorphisms identified through a prostate cancer genome-wide association study in a Japanese population. Am J Epidemiol 174: 1391–1395. doi: 10.1093/aje/kwr271
  24. 24. Lindström S, Schumacher FR, Campa D, Albanes D, Andriole G, et al. (2012) Replication of Five Prostate Cancer Loci Identified in an Asian Population–Results from the NCI Breast and Prostate Cancer Cohort Consortium (BPC3). Cancer Epidemiol Biomarkers Prev 21: 212–216. doi: 10.1158/1055-9965.epi-11-0870-t
  25. 25. Ghoussaini M, Song H, Koessler T, Al Olama AA, Kote-Jarai Z, et al. (2008) Multiple loci with different cancer specificities within the 8q24 gene desert. J Natl Cancer Inst 100: 962–966. doi: 10.1093/jnci/djn190
  26. 26. Chung S, Nakagawa H, Uemura M, Piao L, Ashikawa K, et al. (2011) Association of a novel long non-coding RNA in 8q24 with prostate cancer susceptibility. Cancer Sci 102: 245–252. doi: 10.1111/j.1349-7006.2010.01737.x
  27. 27. Akamatsu S, Takata R, Haiman CA, Takahashi A, Inoue T, et al. (2012) Common variants at 11q12, 10q26 and 3p11.2 are associated with prostate cancer susceptibility in Japanese. Nat Genet 44: 426–429. doi: 10.1038/ng.1104
  28. 28. Ohnishi Y, Tanaka T, Ozaki K, Yamada R, Suzuki H, et al. (2001) A high-throughput SNP typing system for genome-wide association studies. J Hum Genet 46: 471–477. doi: 10.1007/s100380170047
  29. 29. Nakamura Y (2007) The BioBank Japan Project. Clin Adv Hematol Oncol 5: 696–697.
  30. 30. Strope SA, Andriole GL (2010) Prostate cancer screening: current status and future perspectives. Nat Rev Urol 7: 487–493. doi: 10.1038/nrurol.2010.120
  31. 31. Schröder FH, Hugosson J, Roobol MJ, Tammela TL, Ciatto S, et al. (2009) Screening and prostate-cancer mortality in a randomized European study. N Engl J Med 360: 1320–1328. doi: 10.1056/nejmoa0810084
  32. 32. Andriole GL, Crawford ED, Grubb RL, Buys SS, Chia D, et al. (2009) Mortality results from a randomized prostate-cancer screening trial. N Engl J Med 360: 1310–1319. doi: 10.1056/nejmoa0810696
  33. 33. Littrup PJ, Goodman AC, Mettlin CJ, Murphy GP (1994) Cost analyses of prostate cancer screening: frameworks for discussion. Investigators of the American Cancer Society-National Prostate Cancer Detection Project. J Urol 152: 1873–1877. doi: 10.3322/canjclin.43.3.134
  34. 34. Ekwueme DU, Stroud LA, Chen Y (2007) Cost analysis of screening for, diagnosing, and staging prostate cancer based on a systematic review of published studies. Prev Chronic Dis 4: A100.
  35. 35. Schaid DJ (2004) The complex genetic epidemiology of prostate cancer. Hum Mol Genet 13 Spec No 1: R103–121.
  36. 36. Kawakami S, Koga F, Fujii Y, Saito K, Yamamoto S, et al. (2008) History of malignancy is a predictor of prostate cancer detection: incorporation into a pre-biopsy nomogram. Int J Urol 15: 1055–1060. doi: 10.1111/j.1442-2042.2008.02183.x
  37. 37. Zheng J, Liu F, Lin X, Wang X, Ding Q, et al. (2012) Predictive performance of prostate cancer risk in Chinese men using 33 reported prostate cancer risk-associated SNPs. Prostate 72: 577–583. doi: 10.1002/pros.21462
  38. 38. Zheng SL, Sun J, Wiklund F, Smith S, Stattin P, et al. (2008) Cumulative association of five genetic variants with prostate cancer. N Engl J Med 358: 910–919. doi: 10.1056/nejmoa075819
  39. 39. Zheng SL, Sun J, Wiklund F, Gao Z, Stattin P, et al. (2009) Genetic variants and family history predict prostate cancer similar to prostate-specific antigen. Clin Cancer Res 15: 1105–1111. doi: 10.1158/1078-0432.ccr-08-1743
  40. 40. Aly M, Wiklund F, Xu J, Isaacs WB, Eklund M, et al. (2011) Polygenic risk score improves prostate cancer risk prediction: results from the Stockholm-1 cohort study. Eur Urol 60: 21–28. doi: 10.1016/j.eururo.2011.01.017
  41. 41. Kote-Jarai Z, Olama AA, Giles GG, Severi G, Schleutker J, et al. (2011) Seven prostate cancer susceptibility loci identified by a multi-stage genome-wide association study. Nat Genet 43: 785–791. doi: 10.1038/ng.882
  42. 42. Antoniou AC, Beesley J, McGuffog L, Sinilnikova OM, Healey S, et al. (2010) Common breast cancer susceptibility alleles and the risk of breast cancer for BRCA1 and BRCA2 mutation carriers: implications for risk prediction. Cancer Res 70: 9742–9754. doi: 10.1016/s1359-6349(08)71913-x
  43. 43. Catalona WJ, Bailey-Wilson JE, Camp NJ, Chanock SJ, Cooney KA, et al. (2011) National Cancer Institute Prostate Cancer Genetics Workshop. Cancer Res 71: 3442–3446. doi: 10.1158/0008-5472.can-11-0314
  44. 44. Gudmundsson J, Besenbacher S, Sulem P, Gudbjartsson DF, Olafsson I, et al. (2010) Genetic correction of PSA values using sequence variants associated with PSA levels. Sci Transl Med 2: 62ra92. doi: 10.1126/scitranslmed.3001513
  45. 45. Klein RJ, Hallden C, Gupta A, Savage CJ, Dahlin A, et al. (2011) Evaluation of multiple risk-associated single nucleotide polymorphisms versus prostate-specific antigen at baseline to predict prostate cancer in unscreened men. Eur Urol 61: 471–477. doi: 10.1016/j.eururo.2011.10.047