|INVITED ORIGINAL ARTICLE
|Year : 2016 | Volume
| Issue : 4 | Page : 525-529
Race-specific genetic risk score is more accurate than nonrace-specific genetic risk score for predicting prostate cancer and high-grade diseases
Rong Na1, Dingwei Ye2, Jun Qi3, Fang Liu4, Xiaoling Lin4, Brian T Helfand5, Charles B Brendler5, Carly Conran6, Jian Gong5, Yishuo Wu7, Xu Gao8, Yaqing Chen9, S Lilly Zheng6, Zengnan Mo10, Qiang Ding7, Yinghao Sun8, Jianfeng Xu11
1 Department of Urology, Huashan Hospital, Fudan University, Shanghai, China; Fudan Institute of Urology, Huashan Hospital, Fudan University, Shanghai, China; Program for Personalized Cancer Care, NorthShore University HealthSystem, Evanston, Illinois, USA; Health Communication Institute, School of Public Health, Fudan University, Shanghai, China
2 Department of Urology, Shanghai Cancer Center, Fudan University, Shanghai, China
3 Department of Urology, Xinhua Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China
4 Fudan Institute of Urology, Huashan Hospital, Fudan University, Shanghai, China
5 Division of Urology, NorthShore University HealthSystem, Evanston, Illinois, USA
6 Program for Personalized Cancer Care, NorthShore University HealthSystem, Evanston, Illinois, USA
7 Department of Urology, Huashan Hospital, Fudan University, Shanghai; Fudan Institute of Urology, Huashan Hospital, Fudan University, Shanghai, China
8 Department of Urology, Shanghai Changhai Hospital, The Second Military Medical University, Shanghai, China
9 Department of Medical Ultrasonic, Xinhua Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, China
10 Center for Genomic and Personalized Medicine, Guangxi Medical University, Nanning, China
11 Fudan Institute of Urology, Huashan Hospital, Fudan University, Shanghai, China; Program for Personalized Cancer Care, NorthShore University HealthSystem, Evanston, Illinois, USA
|Date of Web Publication||03-May-2016|
Dr. Jianfeng Xu
Fudan Institute of Urology, Huashan Hospital, Fudan University, Shanghai, China; Program for Personalized Cancer Care, NorthShore University HealthSystem, Evanston, Illinois, USA
Dr. Yinghao Sun
Department of Urology, Shanghai Changhai Hospital, The Second Military Medical University, Shanghai
Source of Support: None, Conflict of Interest: None
Genetic risk score (GRS) based on disease risk-associated single nucleotide polymorphisms (SNPs) is an informative tool that can be used to provide inherited information for specific diseases in addition to family history. However, it is still unknown whether only SNPs that are implicated in a specific racial group should be used when calculating GRSs. The objective of this study is to compare the performance of race-specific GRS and nonrace-specific GRS for predicting prostate cancer (PCa) among 1338 patients underwent prostate biopsy in Shanghai, China. A race-specific GRS was calculated with seven PCa risk-associated SNPs implicated in East Asians (GRS7), and a nonrace-specific GRS was calculated based on 76 PCa risk-associated SNPs implicated in at least one racial group (GRS76). The means of GRS7 and GRS76 were 1.19 and 1.85, respectively, in the study population. Higher GRS7 and GRS76 were independent predictors for PCa and high-grade PCa in univariate and multivariate analyses. GRS7 had a better area under the receiver-operating curve (AUC) than GRS76 for discriminating PCa (0.602 vs 0.573) and high-grade PCa (0.603 vs 0.575) but did not reach statistical significance. GRS7 had a better (up to 13% at different cutoffs) positive predictive value (PPV) than GRS76. In conclusion, a race-specific GRS is more robust and has a better performance when predicting PCa in East Asian men than a GRS calculated using SNPs that are not shown to be associated with East Asians.
Keywords: Chinese; genetic risk score; genome-wide association study; prostate cancer; single nucleotide polymorphism
|How to cite this article:|
Na R, Ye D, Qi J, Liu F, Lin X, Helfand BT, Brendler CB, Conran C, Gong J, Wu Y, Gao X, Chen Y, Zheng S L, Mo Z, Ding Q, Sun Y, Xu J. Race-specific genetic risk score is more accurate than nonrace-specific genetic risk score for predicting prostate cancer and high-grade diseases. Asian J Androl 2016;18:525-9
|How to cite this URL:|
Na R, Ye D, Qi J, Liu F, Lin X, Helfand BT, Brendler CB, Conran C, Gong J, Wu Y, Gao X, Chen Y, Zheng S L, Mo Z, Ding Q, Sun Y, Xu J. Race-specific genetic risk score is more accurate than nonrace-specific genetic risk score for predicting prostate cancer and high-grade diseases. Asian J Androl [serial online] 2016 [cited 2021 Aug 3];18:525-9. Available from: https://www.ajandrology.com/text.asp?2016/18/4/525/179857 - DOI: 10.4103/1008-682X.179857
Rong Na, Dingwei Ye, Jun Qi
These authors contributed equally to this study
| Introduction|| |
Prostate cancer (PCa) is one of the most common cancers worldwide.  In China, the incidence of PCa is relatively low; however, it has increased rapidly in recent decades. , Inherited risk of developing the disease is one of the most important risk factors for determining the pathogenesis of PCa. Up to 42% of the disease risk could be explained by heritable factors.  Positive family history has been shown to be strongly associated with PCa  and is widely used in clinical practice for risk assessment of PCa. However, family history information may be influenced by family size, age, survival status of patient's relatives, family communication, recall abilities, etc. In addition, family history must be continually assessed as family history status may change.
Genetic risk scores (GRSs) as measures of inherited risk have been repeatedly shown to provide additional information to family history when assessing one's risk of developing PCa. ,, These GRSs are calculated based on the genotypes of PCa risk-associated single nucleotide polymorphisms (SNPs) implicated from genome-wide association studies (GWASs). To date, GWASs have found more than 100 SNPs associated with PCa; however, most of these studies were conducted primarily in Caucasian men.  As such, the vast majority of these PCa risk-associated SNPs were not found to be significantly associated with PCa in Chinese men.  A recent study demonstrated the predictive performance of GRS in several racial groups using all established PCa risk-associated SNPs.  Although these nonrace-specific GRSs could predict PCa risk, using nonrace-specific SNPs has the potential to lead to over- or under-estimates of disease risk. Whether or not a race-specific GRS (calculated only using SNPs that were significantly associated with the disease in a defined population) is more accurate for predicting disease risks remains unclear. In this study, we compared the performances of two GRSs for predicting PCa and high-grade PCa in a prostate biopsy cohort. The two GRSs were based on (a) East Asian population-specific (Chinese and Japanese), disease-associated SNPs and (b) disease-associated SNPs regardless of race information.
| Materials and Methods|| |
Study population and study design
This was a multicenter study of a biopsy cohort from four tertiary medical centers in Shanghai (Huashan Hospital, Fudan University; Shanghai Cancer Center, Fudan University; Xinhua Hospital, Shanghai Jiao Tong University School of Medicine; Changhai Hospital, the Second Military Medical University), China. Consecutive patients (n = 1 617) undergoing initial prostate biopsy at these four centers were enrolled from August 2013 to December 2014. Written informed consent was obtained from each patient. The study was approved by the Institutional Review Board of each medical center.
Indications for prostate biopsy included (a) tPSA >10.0 ng ml−1 , (b) tPSA >4.0 ng ml−1 (with a confirmation after 2-3 months), (c) %fPSA <0.16 when patients had a suspicious total PSA level (>4.0 ng ml−1 ), and (d) suspicious lesions detected by digital rectal examination (DRE) or ultrasound at any level of tPSA. Demographic and clinical information were documented before biopsy, including age, total prostate-specific antigen (tPSA) level, and free PSA (fPSA) level. Biopsy specimens were analyzed in the Department of Pathology at each hospital. Prostate cancer diagnosis and high-grade disease (defined as Gleason Score ≥7) were recorded. Patients were excluded from the study analyses if (a) records of pathological diagnosis were missing or (b) tPSA, fPSA, or p2PSA were unable to be tested due to poor blood sample quality. Thus, 79 patients were excluded, and 1538 patients were included for further analyses.
Blood samples were collected for extracting DNA. DNA samples were genotyped using Illumina BeadXpress platform with the Golden Gate SNP genotyping assay for 80 SNPs (Supplementary Table 1). All of the candidate SNPs were found to be significantly associated with PCa in Caucasian, of which seven SNPs were significantly associated (reached genome-wide significant level of P < 5 × 10−8 ) with PCa in East Asian (Chinese and Japanese) populations. Two-hundred samples (13.0%) and four SNPs failed to be genotyped because of DNA quality and assay design. The remaining 1338 samples reached the SNP call rate >95%.[Additional file 1]
Calculation of GRSs
Two GRSs based on 76 SNPs (GRS76; using all the SNPs regardless of race) and seven SNPs (GRS7; using race-specific SNPs) were calculated. Briefly, the allelic OR of each SNP was first obtained from external studies.  Second, a genotypic OR of each SNP was calculated from the allelic OR based on a multiplicative model (carrying two risk alleles at one locus, RR: OR  ; carrying one risk allele at the locus, RN: OR; and carrying two nonrisk alleles at the locus, NN: 1). Third, the risk relative to the average risk in the population was calculated based on genotypic OR and risk allele frequency (1000 Genome Project, CHB population) for each SNP in the Chinese populations; the final GRS was calculated by multiplying the risk of each SNP. Theoretically, individuals with GRS of 1.0 are considered to be at average risk of developing a disease compared to other members of their race at large. Individuals with a GRS lower or higher than 1.0 are considered to be at decreased or increased risk of developing the disease, respectively, compared to the general population (defined by race).
In univariate analyses, Student's t-tests and Mann-Whitney U-tests were used to compare different variables among groups for normal distribution variables and nonnormal distribution variables, respectively. GRSs were adjusted by age, tPSA, and fPSA using logistic regression in multivariate analyses. Areas under the receiver operating characteristic curve (AUC of ROC) analyses were used to evaluate the predictive values of GRS7 and GRS76. The AUCs were compared by Z-test. The differences between GRS7 and GRS76 at specific cutoff values were described using positive predictive values (PPVs). Net reclassification improvement analyses (NRIs) were used to evaluate the improvement of GRS7 from GRS76. All statistical analyses were performed using SPSS 22.0 (IBM, North Castle, NY, USA). Two-tailed P < 0.05 was considered statistically significant.
| Results|| |
Characteristics of the study population and univariate analyses between each group are shown in [Table 1] . Age at diagnosis (mean: 67.59 vs 64.44, P = 8.51 × 10−6 ), tPSA (median: 25.61 vs 8.80, P = 1.73 × 10−73 ), GRS76 (median: 1.90 vs 1.50, P = 7.76 × 10−6 ), and GRS7 (median: 1.22 vs 0.99, P = 3.47 × 10−10 ) were significantly higher in PCa group than in non-PCa group while %fPSA was significantly lower in PCa group (median: 0.11 vs 0.16, P = 2.50 × 10−37 ). Similar results were found between high-grade PCa group and others (low-grade PCa and non-PCa). In multivariate analyses, both GRS76 (odds ratio, OR = 1.10 for PCa, P = 4.02 × 10−4 ; OR = 1.07 for high-grade PCa, P = 0.014) and GRS7 (OR = 1.45 for PCa, P = 9.84 × 10−6 ; OR = 1.34 for high-grade PCa, P = 0.001) were significant independent risk factors of PCa as well as high-grade PCa when adjusting for age, tPSA, and fPSA ([Table 2]).
|Table 1: Characteristics of study population and the univariate analysis of each variable between PCa group and non-PCa group |
Click here to view
|Table 2: Multivariate logistic regression analyses of GRSs adjusting for different variables |
Click here to view
To assess whether GRS of 1.0 represents average risk in the cohort, we calculated means of GRSs after excluding the extreme values (GRS above 75 th percentile + 1.5 interquartile range, and GRS below 25 th percentile - 1.5 interquartile range). For GRS7, the means were 1.19 in the entire population, 1.32 in PCa group, 1.11 in non-PCa group, 1.36 in high-grade PCa group, and 1.12 in nonhigh-grade PCa group. For GRS76, the means were 1.85 in the entire population, 1.99 in PCa group, 1.77 in non-PCa group, 2.04 in high-grade PCa group, and 1.77 in nonhigh-grade PCa group.
We then compared the performances of GRS76 and GRS7 for discriminating PCa and high-grade diseases (Gleason score ≥7). GRS7 (AUC = 0.602) had a better discriminative ability for PCa than GRS76 (AUC = 0.573) did; however, this difference did not reach a statistically significant level (P = 0.20). Similarly, GRS7 (AUC = 0.603) performed better for discriminating high-grade PCa than GRS76 (AUC = 0.575) did; however, no statistical significance was observed ([Table 3]). In NRI analyses, results showed that GRS7 slightly improved the predictive ability of PCa and high-grade diseases from GRS76 but did not reach statistical significance ([Table 4]). For instance, at the cutoff value of 1.5, the NRIs from GRS76 to GRS 7 were 0.033 for predicting PCa and 0.031 for high-grade PCa, indicating that using GRS7, there would be ~3% net improvement of predictive ability from GRS76.
|Table 3: AUCs of receiver operating curve analyses of each GRS for predicting PCa and high-grade PCa |
Click here to view
Using PPV, we found that GRS7 had a better performance than GRS76 for predicting PCa and high-grade diseases at various cutoff values ([Figure 1] and [Figure 2]). For example, at a cutoff of 1.5, 29% of men were classified as higher risk using GRS7, with a PPV of 48.2% for predicting PCa. In comparison, 53% of men were classified as higher risk using GRS76, with a PPV of only 41.8% ([Figure 1]). The difference in PPV between the two GRSs was 6.4%. At a cutoff value of 2.0, 14% of men were classified as higher risk using GRS7, with a PPV of 56.4% for predicting PCa. In comparison, 40% of men were classified as higher risk using GRS76, with a PPV of only 43.9%. The difference in PPV between the two GRSs was even larger, at 12.5%, when using a cutoff of 2.0. Similar findings were observed for predicting high-grade PCa ([Figure 2]). At a cutoff of 1.5, the PPV of GRS7 was 40.5% for predicting high-grade PCa whereas the PPV of GRS76 was 34.5%. At a cutoff of 2.0, the PPV was 48.4% for GRS7 and 36.9% for GRS76.
|Figure 1: Distribution of PCa (red) and non-PCa (blue) at different GRS levels. Dash lines highlight the cutoff values of 1.0, 1.5, and 2.0, respectively. n: the number of study population with GRS over the cutoffs; PPV: positive predictive value. ( a ) GRS76; ( b ) GRS7.|
Click here to view
|Figure 2: Distribution of high-grade PCa (red) and others (nonhigh-grade PCa, blue) at different GRS levels. Dash lines highlight the cutoff values of 1.0, 1.5, and 2.0, respectively. n: the number of study population with GRS over the cutoffs; PPV: positive predictive value. ( a ) GRS76; ( b ) GRS7|
Click here to view
| Discussion|| |
In this study, we used genotyping data from a Chinese biopsy cohort to calculate two GRSs for each subject: one race-specific GRS (using seven SNPs previously shown to be significantly associated with East Asian men at P < 5 × 10−8 ) and one GRS that was based on all 76 SNPs previously reported to be significantly associated with PCa in any races. We found that while both GRS7 and GRS76 were significant predictors of PCa and high-grade PCa, GRS7 had slightly better, though nonstatistically significant, AUCs than GRS76 for discriminating PCa from non-PCa, and for discriminating high-grade PCa from nonhigh-grade PCa. More importantly, we found that GRS7 had considerably better PPVs than GRS76 for predicting PCa and high-grade PCa. Finally, we found that the mean of GRS7 (1.19) of the study population was closer to the expected value of 1.0 than GRS76 (1.85), a factor that is critical for risk assessment at an individual level to define higher or lower risk for PCa.
Multiple measurements can be used to assess whether a biomarker is a predictor of PCa risk at a population level, including a test for different means of a biomarker between cases and controls, AUC for discriminating cases from controls, and PPV for predicting probability of PCa among individuals classified as higher risk. These measurements in this study support that both race-specific GRS (GRS7) and nonrace-specific GRS (GRS76) can be used as predictors of PCa risk at a population level. However, based on the results of PPV, the race-specific GRS is a better choice. PPV is a more relevant measurement than AUC for the purpose of risk assessment for identifying high-risk men. Compared to GRS76, GRS7 identified fewer men at higher risk, but a higher proportion of them developed PCa and high-grade PCa, suggesting that GRS7 is more effective in identifying high-risk men for targeted intervention.
When assessing the performance of GRS for discriminating and predicting PCa, a mean GRS that closes to the expected value of 1.0 is critical for defining higher or lower risk for an individual. An advantage of the GRS used in this study is that it is population standardized, and the mean GRS is expected to be 1.0. This theoretical property is confirmed in a simulation study, in which up to 100 different risk-associated SNPs were simulated to be associated with disease risk (Yu et al. in the same issue). The study indicated when the OR and risk allele frequency of each SNP used in the calculation of GRS are correct (i.e., they are the same as simulated values), the mean GRS in the cohort was close to 1.0 regardless of the number of SNPs (top 30, 50, and 100 SNPs with highest ORs) used in the calculations of GRSs. The mean GRS, however, could deviate from 1.0 if ORs and risk allele frequencies used in the GRS calculation were different from the true simulated values. Therefore, observation of a mean GRS in the cohort close to 1.0 is important to ensure that the ORs and allele frequencies used in GRS calculation are appropriate. From this consideration, GRS7 is better than GRS76 because its mean is closer to 1.0.
It should be noted that although the mean of GRS7 in the cohort was close to 1.0, it was slightly higher than 1.0 (mean = 1.19). Two possible reasons may account for this deviation. First, the estimates of ORs and risk allele frequencies of the seven SNPs used in GRS calculations may be over- or under-estimated. The estimated ORs of these SNPs are likely to be reliable because they were obtained from a large meta-analysis in East Asians.  The allele frequencies of these SNPs were obtained from a small sample (82 subjects) of the CHB (Chinese Han Beijing) population in the 1000 Genomes Project. These frequencies differed from the risk allele frequencies in this cohort, which ranged from 0.6% to 7.9% (Supplementary Table 1 ). The impact of over- and under-estimated OR and risk allele frequency is stronger on the GRS76 than on the GRS7 and may explain the larger deviation of its mean (1.85) from 1.0. The allele frequencies of these 76 SNPs were also from the CHB population of the 1000 Genomes Project. However, the differences in allele frequencies between the CHB population and our cohort ranged from 0.2% to 42% ( Supplementary Table 1 ). Second, the higher mean GRS may indicate that men in this cohort have higher average genetic risk for PCa. This is plausible because a considerable proportion of men in this biopsy cohort have PCa. Furthermore, a subset of patients with negative biopsy results may also have PCa that could have been missed on initial needle biopsy.
A limitation of this study should be noticed. We only had genotypes of 80 SNPs rather than the ~100 PCa risk-associated SNPs that have now been identified. However, because the impact of any of these SNPs on GRS is small due to their modest ORs, missing a subset of these SNPs is unlikely to change the main conclusion of the study.
| Conclusions|| |
A race-specific GRS is more robust and accurate than a nonrace-specific GRS for assessing an individual's risk of developing PCa. Only risk-associated SNPs that have been previously confirmed in a specific racial group (P value reaching GWAS significant level) should be used for calculating GRS for a population of interest.
| Author Contributions|| |
RN, DY, and JQ participated in statistical analyses and drafted the manuscript; FL, YW, XG, YC, and JG participated in acquisition of data and gave technical support; BTH, CBB, and CC participated in critical revision of the manuscript; SLZ and ZM participated in administrative, technical and material support; QD participated in study design; YS and JX participated in study design and supervision of the study.
All authors have read and approved the final version of the manuscript and agree with the order of presentation of the authors.
| Competing Interests|| |
There is no conflict of interests in this paper.
| Acknowledgments|| |
This work was in part supported by grants from the Key Project of the National Science Foundation of China to Jianfeng Xu (81130047), the National Key Basic Research Program Grant 973 of China to Jianfeng Xu (2012CB518301), the National Natural Science Foundation of China (Grant No. 81402339) to Rong Na, the intramural grants from Huashan Hospital Fudan University to Rong Na. This study is also partially supported by the Ellrodt-Schweighauser Family Chair of Cancer Genomic Research of NorthShore University HealthSystem to JX. Finally, We would like to thank all the subjects included in this study.
Supplementary information is linked to the online version of the paper on the Asian Journal of Andrology website.
| References|| |
Jemal A, Bray F, Center MM, Ferlay J, Ward E, et al.
Global cancer statistics. CA Cancer J Clin
2011; 61: 69-90.
Qi D, Wu C, Liu F, Gu K, Shi Z, et al.
Trends of prostate cancer incidence and mortality in Shanghai, China from 1973 to 2009. Prostate
2015; 75: 1662-8.
Zhou M, Wang H, Zhu J, Chen W, Wang L, et al.
Cause-specific mortality for 240 causes in China during 1990-2013: a systematic subnational analysis for the global burden of disease study 2013. Lancet
2016; 387: 251-72.
Lichtenstein P, Holm NV, Verkasalo PK, Iliadou A, Kaprio J, et al.
Environmental and heritable factors in the causation of cancer - Analyses of cohorts of twins from Sweden, Denmark, and Finland. N Engl J Med
2000; 343: 78-85.
Johns LE, Houlston RS. A systematic review and meta-analysis of familial prostate cancer risk. BJU Int
2003; 91: 789-94.
Aly M, Wiklund F, Xu J, Isaacs WB, Eklund M, et al.
Polygenic risk score improves prostate cancer risk prediction: results from the Stockholm-1 cohort study. Eur Urol
2011; 60: 21-8.
Kader AK, Sun J, Reck BH, Newcombe PJ, Kim ST, et al.
Potential impact of adding genetic markers to clinical parameters in predicting prostate biopsy outcomes in men following an initial negative biopsy: findings from the REDUCE trial. Eur Urol
2012; 62: 953-61.
Sun J, Na R, Hsu FC, Zheng SL, Wiklund F, et al.
Genetic score is an objective and better measurement of inherited risk of prostate cancer than family history. Eur Urol
2013; 63: 585-7.
Wang M, Takahashi A, Liu F, Ye D, Ding Q, et al.
Large-scale association analysis in Asians identifies new susceptibility loci for prostate cancer. Nat Commun
2015; 6: 8469.
Na R, Liu F, Zhang P, Ye D, Xu C, et al.
Evaluation of reported prostate cancer risk-associated SNPs from genome-wide association studies of various racial populations in Chinese men. Prostate
2013; 73: 1623-35.
Hoffmann TJ, Van Den Eeden SK, Sakoda LC, Jorgenson E, Habel LA, et al
. A large multiethnic genome-wide association study of prostate cancer identifies novel risk variants and substantial ethnic differences. Cancer Discov
2015; 5: 878-91.
[Figure 1], [Figure 2]
[Table 1], [Table 2], [Table 3], [Table 4]
|This article has been cited by|
||Polygenic Risk Scores in Prostate Cancer Risk Assessment and Screening
| ||Lindsey Byrne,Amanda Ewart Toland |
| ||Urologic Clinics of North America. 2021; |
|[Pubmed] | [DOI]|
||Epidemiology and genomics of prostate cancer in Asian men
| ||Yao Zhu,Miao Mo,Yu Wei,Junlong Wu,Jian Pan,Stephen J. Freedland,Ying Zheng,Dingwei Ye |
| ||Nature Reviews Urology. 2021; |
|[Pubmed] | [DOI]|
||The emerging field of polygenic risk scores and perspective for use in clinical care
| ||Tatiane Yanes,Aideen M McInerney-Leo,Matthew H Law,Shelly Cummings |
| ||Human Molecular Genetics. 2020; |
|[Pubmed] | [DOI]|
||Validation of the novel susceptibility loci for prostate cancer in a Chinese population
| ||Yishuo Wu,Haitao Chen,Ying Ji,Rong Na,Zengnan Mo,Dingwei Ye,Meilin Wang,Jun Qi,Xiaoling Lin,Qiang Ding,Jianfeng Xu,S. Zheng,Yinghao Sun,Wei Meng |
| ||Oncology Letters. 2017; |
|[Pubmed] | [DOI]|