New Prostate Cancer Risk Variants Identified in Expanded Multi-Ancestry GWAS Meta-Analysis - Anqi Wang

March 12, 2024

Anqi Wang unveils findings from her study published in Nature Genetics. Dr. Wang's research marks a significant advance in understanding the genetic landscape of prostate cancer, identifying a total of 451 independent risk loci, including 187 novel variants, through an expanded multi-ancestry meta-analysis. This study not only enlarges the scope of known genetic risk factors but also emphasizes the necessity for further exploration in non-European populations to enhance the prediction and stratification of prostate cancer risk. Moreover, Dr. Wang discusses the clinical potential of the Genetic Risk Score (GRS) derived from these findings, particularly in tailoring prostate cancer screening and informing personalized prevention strategies. This collaborative effort, involving over 300 researchers worldwide, signifies a leap toward more inclusive and precise genetic insights into prostate cancer risk across diverse ancestries.


Anqi Wang, PhD, MBBS, Postdoctoral Research Fellow, Harvard T.H. Chan School of Public Health, Boston, MA

Andrea K. Miyahira, PhD, Director of Global Research & Scientific Communications, The Prostate Cancer Foundation

Read the Full Video Transcript

Andrea Miyahira: Hi, everyone. I'm Andrea Miyahira at The Prostate Cancer Foundation. Joining me is Dr. Anqi Wang, a postdoctoral fellow at the Harvard T.H. Chan School of Public Health. Dr. Wang will present data from her recent paper, "Characterizing Prostate Cancer Risk Through Multi-Ancestry Genome-Wide Discovery of 187 Novel Risk Variants," that was published in Nature Genetics. Dr. Wang, thank you for sharing this study with us today.

Anqi Wang: Thank you for having me, Andrea. Today, I'm very excited to share our group's recent study on the multi-ancestry meta-analysis of GWAS for prostate cancer. First, I would like to begin with a brief recap of our last multi-ancestry GWAS. Prostate cancer is one of the most common cancers globally, but the established risk factors for prostate cancer are limited and include age, family history, and germline variations.

It is estimated that 57% of the genetic heritability of prostate cancer, making it the most heritable cancer. Moreover, men of African ancestry also show higher incidence and mortality than others. This highlights the importance of identifying genetic risk variants in multi-ancestry samples to understand and predict prostate cancer. Previously, with more than 230,000 multi-ancestry cases and controls, our group reported a total of 269 risk variants for prostate cancer, including 86 novel variants.

These 269 risk variants in total explained about 33 to 43% of the familial relative risk of prostate cancer, implying that more than half of the heritability is beyond the 269. And to examine the predictive performance of the 269, we constructed a polygenic risk score with the 269, and it's estimated to be associated with a three to five-fold increase in the risk of prostate cancer for men in the top decile compared to men in the median PRS category. The relatively lower odds ratio in non-European populations, especially in Africans, suggests a need to incorporate more non-European population samples to improve risk variant discovery and risk prediction.

Therefore, to further our understanding of the genetic architecture of prostate cancer, we conducted an expanded multi-ancestry meta-analysis, especially with non-European population studies. Here in gray shows the studies included in the last GWAS, and we now included 11 new studies from four ancestry groups resulting in a 57% increase in the non-European cases and more than a 128% increase in the effective sample size in each population. By adopting a forward selection using the mJAM statistics, we identified a total of 451 independent risk loci for prostate cancer.

Here shows the Manhattan plot of the GWAS result. 187 variants were not reported previously, highlighted in orange in this plot, and 61 of them were within 800 kb of the known risk variants but showed independency after conditioning on those nearby known risk variants. There are 264 known risk variants highlighted in purple in this plot, and among them, 150 risk variants were replaced by a more significant lead variant in this study. We also ended up dropping 18 previously reported prostate cancer risk variants because they did not reach genome-wide significance when we expanded the GWAS, and they also showed no functional evidence for prostate cancer. You can see that by expanding the multi-ancestry samples, we now not only have larger power to detect more risk variants but also can identify independent variants on a finer scale.

Next, to assess the predictive performance of the identified risk variants, we constructed the genetic risk scores using the sum of risk allele-like dosages weighted by the per-allele odds ratios from the conditional model. We calculated and compared the GRS for the 451 risk variants with the historically reported marker sets. Those are markers reported in previous multi-ancestry GWAS with smaller sample size, and most of the samples are from European populations. I also want to note that here we call it GRS instead of PRS, which refers to the polygenic risk score, to differentiate the score with other types of PRS since we only use the genetically significant independent risk variants to construct the score.

After constructing the score, first, we categorized individuals' risk based on the GRS quintiles in the controls. This Sankey diagram shows the movement flow of the assignment of controls across different GRS categories and over time we can see that with the discovery of additional risk variants, there's greater stability in the assignment of controls to the GRS categories. If we focus on the top or bottom quintiles, we see that 58% of men in the top or bottom quintile remain in the same quintile between the earliest GRS 100 and GRS 181, whereas about 70% of them remain between the GRS 269 and the current GRS 451. After the assignment, and for the assignment of cases, we still see movement across GRS categories, but the percentage of cases has increased within the top GRS quintile from 40% to 50% and decreased within the bottom GRS quintile from 8% to 4%.

We also evaluated the association between prostate cancer and the standardized GRS in a logistic regression model. In the GWAS discovery sample, where we have the largest sample sizes and larger power, the odds ratio per standard deviation has increased over time in all populations, and this improvement was consistently observed in the replication samples. In addition, for the most current GRS 451, the estimated odds ratio is still the highest among European ancestry and the lowest in men of African ancestry, but the odds ratio in African ancestry showed the largest percent of improvement from the last GWAS in both discovery and replication samples.

Next, we looked at the effect of the GRS by age of diagnosis. Most prostate cancer cases were diagnosed among men aged older than 55 years old, and the early onset of prostate cancer is often related to poorer prognosis. Here we saw age to modify the association of GRS 451 with prostate cancer risk. There were significantly greater associations of GRS among men aged 55 or younger for both men of European and African ancestry. We observed similar trends in Asian and Hispanic populations, but the P-heterogeneity is not significant for those populations, most likely because our sample size for these two populations is too small.

We also looked at the relationship between the current GRS and disease aggressiveness, which is defined by the criteria here. GRS 451 was associated equally with the risk of aggressive versus non-aggressive prostate cancer in European, African, and Hispanic men. However, for men of African ancestry, GRS was significantly more associated with the risk of aggressive disease than non-aggressive disease. Lastly, to show the accumulated risk of developing prostate cancer for men with different genetic risks, we calculated their absolute risk for a given age based on the actuarial estimation in each population. The density plot here shows the GRS distribution of African and European populations. About 16% of African and 20% of European men would have more than a two-fold risk of prostate cancer compared to those men who have an average genetic risk. And the solid line here illustrates the cumulative absolute risk for individuals who have more than a two-fold risk compared to average men in African and European ancestry, respectively.

For a man of African ancestry with an average genetic risk, their risk of having a prostate cancer diagnosis is 11.6% by the age of 85. Well, for those men on this curve, they would achieve this 11.6% risk at the age of 66 years, which is 19 years earlier than those with an average risk. And similarly, for men of European ancestry, those 20% of men would achieve an absolute risk of 7.8% at the age of 69 years old, which is 16 years earlier than those who have an average GRS.

This absolute risk estimation implies the importance of a personalized screening strategy for men with different GRS. For example, men with higher genetic risk should start their screening much earlier than the average man to detect their prostate cancer at an earlier stage. Lastly, I would like to acknowledge that the study was conducted under the guidance of my Ph.D. advisors, Dr. Chris Haiman and Dr. David Conti at USC. I also want to highlight that this work is a collaborative effort involving over 300 researchers from 26 different countries, as well as many funding sources, including the Prostate Cancer Foundation. The collective contributions have been pivotal in achieving the result we present today. Thank you.

Andrea Miyahira: Thank you so much for sharing this. So, how many prostate cancer risk genes or SNPs do you think remain yet to be identified, particularly in understudied populations?

Anqi Wang: Yes, our study estimated that the current 451 risk variants account for about 30% to 50% of the familial relative risk associated with prostate cancer, which indicates that over half of the disease heritability remains unexplained, and this missing heritability could be attributed to other genetic or genomic factors that are not covered by GWAS studies, such as germline rare variants, DNA copy number variations, or epigenetic modifications, and so on. But at the same time, I think there are still plenty of common risk SNPs for prostate cancer that remain undiscovered, especially for non-European populations.

In this current study, compared to our last multi-ancestry GWAS, we have successfully demonstrated that the inclusion of more multi-ancestry samples has not only led us to detect more novel risk variants for all populations but also refine our understanding of previously identified risk regions. However, it is also important to acknowledge that our sample is still predominantly represented by men of European ancestry, with cases from European ancestry outnumbering those from African ancestry by more than six-fold, and even more so for Asian ancestry and Hispanic ethnicity. It's clear that there's substantial room for ancestry-specific variants for those non-European populations, particularly for populations of African ancestry, which have a more complex linkage disequilibrium structure in their genome. A larger sample size is necessary to adequately capture the genetic diversity and identify specific risk variants.

Andrea Miyahira: So, how have the additional risk SNPs increased the ability to identify patients at risk for aggressive or lethal prostate cancer specifically?

Anqi Wang: Yeah, this is an interesting question because when we look at the association between GRS 451 and the disease's aggressiveness, we actually observed no significant association between GRS for aggressive versus non-aggressive prostate cancer, with the exception of the African ancestry population. But interestingly, when we removed some of the SNPs that are associated with PSA levels, we actually observed a positive association between the GRS and disease aggressiveness in all populations. This leads to the question of potential detection bias due to PSA screening. A large portion of the prostate cancer cases included in this study, as well as in the real world, are localized prostate cancer without any symptoms. They're often detected due to elevated PSA levels from PSA screening. Given the widespread use of PSA screening, non-aggressive cases may be overrepresented in our GWAS samples, and this, in turn, may lead us to identify PSA-related SNPs rather than the true prostate cancer SNPs.

To address this potential PSA screening bias, we compared our 451 risk variants with the 128 PSA risk variants reported from previous GWAS on PSA levels, and we found 51 variants are overlapped or highly correlated. Those variants could be related to prostate cancer risk directly or their ability to increase PSA and relate to prostate cancer biopsy and diagnosis. If that's the case, then those SNPs would be more likely to be associated with non-aggressive disease. Therefore, when we remove these overlapping PSA variants and reconstructed the GRS with the 400 non-PSA related SNPs, we found that GRS 400 is now significantly associated with aggressive prostate cancer in the multi-ancestry samples. Specifically, we see the positive association in European and Hispanic populations, and also Asian ancestry, which were not observed with GRS 451, and also the association in the African ancestry is now stronger.

Andrea Miyahira: That's really interesting. So, do you think the GRS 451 score is ready for clinical application, or does more work need to be done, for instance, removing certain SNPs or identifying additional risk SNPs in other understudied populations?

Anqi Wang: While our study has demonstrated the potential of GRS in effectively stratifying men with higher risk of prostate cancer from the general population, I think there are several critical questions that need to be further investigated before its clinical application. One of the foremost challenges in the clinical application is to determine a universally applicable cutoff point of the score. In this study, we showed that based on the quintile characterization, about 50% of prostate cancer cases fall into the highest risk quintile, and this implicates a question of balance between sensitivity and specificity. If we use the top quintile as a cutoff point, we would potentially overlook half of the cases that fall into lower quintiles. Conversely, if we use a more lenient cutoff point to include a larger proportion of men with higher genetic risk, we would end up having a large number of men without prostate cancer, and that would dilute the specificity of screening.

The other issue is to identify a method to standardize the GRS calculation to ensure its robustness and reliability across varying genotyping platforms. Not all of the 451 risk variants identified in our research are covered across different genotyping platforms, which could lead to discrepancies in GRS calculation. To minimize this discrepancy, we have provided two versions of GRS risk variants on the PGS catalog. One is based on genome build 37, one is based on genome build 38, both of which showed high consistency and comparable performance in our studies, but still, some of the genotyping platforms may not catch the exact SNPs that we provided in the 451, or the risk variants could have lower genotyping quality, and this makes the application of GRS across different settings challenging.

Another critical issue, as we've noted, is the concern about the performance of the current GRS in non-European populations, which may not be as robust as it is in the European ancestry populations. And this again underscores the importance of including more diverse populations to identify additional risk variants and regions that are specific to other ancestry populations. It's also important not only to understand the genetic risk among those populations but also to ensure that the benefit of the genetic risk score can be equally and equitably accessed by all populations.

Andrea Miyahira: Thanks, and then my last question is, how do you envision the GRS score could be used in the clinic?

Anqi Wang: This GRS score is currently primarily used in research settings to estimate germline susceptibility to prostate cancer. And in terms of applying the GRS in clinical settings, one of the most interesting and promising areas, I believe, is to incorporate it into prostate cancer screening, especially combined with other currently used tests like PSA screening and imaging. There are some ongoing clinical trials, and we're looking forward to seeing the results. Beyond screening and for those men diagnosed with prostate cancer, another important area is to identify those patients who are at higher risk of disease progression and poor prognosis.

We're currently investigating the possibility of combining the GRS with tumor-specific factors, such as molecular expressions and the tumor microenvironment, to enhance the accuracy of prostate cancer prognosis and predictions. In addition to using GRS as a predictive tool, it can also be used to inform clinical recommendations. Studies have demonstrated that adopting a healthier lifestyle, such as avoiding smoking, maintaining a healthy weight, and engaging in more active physical activity, can mitigate the risk of developing lethal prostate cancer. And this might seem like general healthy tips, but it also demonstrates that men with higher genetic risk are more likely to benefit from those lifestyles. This also illustrates the value of incorporating the GRS score into routine patient care for a more personalized approach to managing prostate cancer.

Andrea Miyahira: Well, thank you so much for sharing this study with us today.

Anqi Wang: Thank you for inviting me.