(C) PLOS One
This story was originally published by PLOS One and is unaltered.
 . . . . . . . . . .



Identifying shared polygenic risk across cancers [1]

['Jiaqi Hu', 'Department Of Chronic Disease Epidemiology', 'Yale School Of Public Health', 'New Haven', 'Connecticut United States Of America', 'Maiyier Muheyati', 'Department Of Biostatistics', 'Connecticut', 'United States Of America', 'Leqi Xu']

Date: 2026-09 

Our findings demonstrate that although shared genetic susceptibility across cancers is limited at the genome-wide level, it becomes evident when examined at regional and polygenic scales. Integrating local genetic correlation and PRS decomposition analyses reveals structured patterns of shared genetic risk, providing a framework for investigating cross-cancer polygenic susceptibility.

Genome-wide genetic correlation analyses identified 20 significantly correlated cancer pairs, whereas local analyses revealed 82 regions with shared genetic signals across 66 cancer pairs. Five regions exhibited mutually correlated cancer clusters, with enrichment in functional domains such as inflammatory functions. Cross-cancer PRS analyses identified five cancer pairs with shared polygenic risk. Decomposition of PRSs indicated that these cross-cancer associations were enriched in specific pleiotropy groups and immune-related pathways rather than reflecting diffuse genome-wide overlap.

We estimated pairwise local genetic correlations across 16 specific cancers and one pan-cancer phenotype using SUPERGNOVA. Genome regions harboring multiple cancers with mutually correlated local genetic effects were annotated using genetic correlations with non-cancer phenotypes and associations from the GWAS Catalog. In parallel, cross-cancer PRS associations were evaluated, and significant cancer pairs were identified. Genome-wide PRSs for selected pairs were further decomposed into pleiotropy-informed and pathway-specific components to assess functional enrichment of shared polygenic risk.

Shared genetic susceptibility across cancers has been reported but is generally modest at the genome-wide level. Whether such shared polygenic risk exhibits structured convergence at regional or functional levels remains unclear. We investigated shared genetic risk across cancers by integrating local genetic correlation analyses with cross-cancer polygenic risk score (PRS) associations.

Many cancers are known to share inherited genetic risk, but it remains unclear where this genetic overlap occurs and what biological mechanisms underlie it. Identifying these shared genetic influences may reveal common processes involved in cancer development and could ultimately improve cancer risk prediction and prevention. In this study, we analyzed genetic data from 24 specific cancer types and a pan-cancer phenotype to investigate shared inherited cancer risk. In addition to assessing genetic overlap across the genome, we examined specific genomic regions and groups of genes involved in related biological functions. We found that most cancers showed only limited shared genetic risk at the genome-wide level. However, many pairs of cancers shared genetic influences within particular genomic regions and biological pathways. These findings suggest that different cancers may arise through overlapping genetic mechanisms, even when their overall genetic similarity is low. Overall, this study provides a framework for identifying shared genetic susceptibility across cancer types and offers new insight into biological processes that contribute to cancer development, with potential implications for future cancer prevention and risk prediction.

Funding: This was supported in part by National Institute of Health (NIH; https://www.nih.gov/) grant R01 HG012735 and P50 CA196530 to HZ. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability: The individual genotype and phenotype data underlying this article were provided by the UK Biobank by permission (ref: 29900), and the instructions to apply for the data can be found at https://www.ukbiobank.ac.uk/enable-your-research/apply-for-access . The GWAS summary statistics were downloaded from publicly available databases, and the information on related articles was available in Methods. The summary-level data (e.g. PRS weights) are available on Zenodo ( https://zenodo.org/records/22694018) .

Copyright: © 2026 Hu et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

In this study, we aimed to characterize shared genetic susceptibility across cancers by integrating regional correlation analyses with polygenic risk evaluation. We decomposed the genome-wide sharing into local signals through local genetic correlations and PRS decompositions. Our results confirmed that the shared genetic architecture across cancers on the genome-wide scale is modest, but the overlapped genetic risk is enriched in certain functions, such as immune-metabolic pathways.

Although prior studies have quantified the generally modest genetic overlap across cancers and highlighted that shared signals may be confined to specific genomic regions, the underlying biological mechanisms of this overlap remain insufficiently understood. In particular, it is unclear whether the variants or loci shared across cancers converge on common oncogenic pathways or reflect distinct, context-dependent mechanisms. Moreover, most existing analyses have examined cancer pairs in isolation, limiting the ability to characterize broader patterns of shared genetic architecture. As a result, a comprehensive, network-level understanding of how multiple cancers are genetically interconnected has yet to be fully established.

Efforts to quantify shared genetic susceptibility across cancers—typically measured through genetic correlations— have consistently shown that genome-wide overlap is generally modest. Sampson et al. (2015) evaluated pairwise genetic correlations among 13 cancers and identified significant correlations for only a few pairs, such as bladder and lung cancers, indicating limited shared heritable risk at the genome-wide level [ 6 ]. Subsequent studies leveraging larger GWAS datasets have reported similarly low genetic correlations [ 7 , 8 ]. Nevertheless, some cancer pairs exhibit local signals despite weak or nonsignificant genome-wide correlations. For instance, shared risk at the chromosome 9p21 region has been observed across several cancers [ 7 ], suggesting that genetic overlap may be concentrated within specific genomic regions rather than distributed broadly across the genome. Furthermore, polygenic risk scores (PRSs), which aggregate effects of risk variants, have revealed cross-cancer associations that are not captured by the global genetic correlation estimate [ 9 ], pointing to additional cross-cancer relationships.

Genetic factors contribute substantially to cancer susceptibility. Evidence from a large Nordic twin cohort demonstrates that approximately one-third of the variance in overall cancer risk is attributable to heritable factors (heritability [h 2 ] = 33%, 95% CI: 30%-37%) [ 1 ]. Genome-wide association studies (GWASs) have identified numerous common variants associated with risk across diverse cancer types [ 2 – 4 ]. Notably, some variants exhibit pleiotropy, i.e., they are independently associated with two or more cancers. For example, rs78378222 in TP53 has been implicated in the risk of breast cancer, skin cancer, and several other malignancies [ 2 , 5 ]. These observations suggest shared genetic architecture across cancers and highlight the importance of systematically characterizing cross-cancer genetic overlap.

Second, we decomposed PRSs into pathway-specific PRSs (PS-PRSs), which assigned SNPs to genes and then labelled with pre-defined pathways, to evaluate shared biological mechanisms between cancers. A total of 8,004 autosomal genes from 364 KEGG pathways [ 26 ] were mapped to genetic variants using positional mapping with a ± 10-kilobase window. For each pathway, a PS-PRS was calculated using pathway-specific variant sets and SDPR-derived weights, implemented in PLINK 1.9 [ 23 ]. Across-cancer associations for PS-PRSs were evaluated using logistic regression models analogous to those described above.

Nagelkerke's R 2 was calculated for each model. The for excluding PD-PRS was calculated as the difference between R 2 of equation (1) and of equation (3) , and the for excluding remaining PD-PRS was the difference between R 2 of equation (1) and of equation (2) . The attenuation in R 2 was computed as the difference between the values for PD-PRS and for the remaining PD-PRS. To test the null hypothesis that the across-cancer association cannot be attributed to the PD-PRS, a permutation test was conducted with 1,000 permutations while preserving the associations between the outcome, the PRS for cancer i, and covariates. Specifically, we first fitted the null model including the PRS for cancer i and all covariates and estimated the corresponding outcome probabilities. Permuted outcomes were then generated by resampling from these fitted probabilities while preserving the overall number of cases. The association between the PD-PRS and the permuted outcomes was subsequently evaluated to obtain the empirical null distribution.

For each selected cancer pair, logistic regression models were fitted with cancer (excluding individuals with cancer ) as the outcome and each PD-PRS for cancer as the exposure, adjusting for the PRS of cancer , age at recruitment, sex (except for sex-specific cancers), and the top 10 genetic PCs. To quantify the contribution of each pleiotropy group to the across-cancer association, we evaluated attenuation in explained variance using Nagelkerke’s R 2 . Specifically, a full model including the PRS for cancer (Equation (1)) was compared with reduced models substituting either the PD-PRS ( Equation (2) ) or the remaining PD-PRS constructed using variants not included in the PD-PRS ( Equation (3) ).

First, we applied our recently developed pleiotropy-decomposed PRS (PD-PRS) framework, which decomposed the genome-wide PRS into pleiotropy-informed components based on local genetic correlations [ 24 ]. Briefly, the genome was partitioned into approximately 2,000 independent LD blocks. Block-specific genetic correlations between cancers and 47 non-cancer phenotypes—grouped into 12 pleiotropy groups [ 25 ]—were estimated using SUPERGNOVA [ 16 ]. These groups were defined using domain knowledge of phenotype relationships, with their biological relevance further supported by significant genetic correlations among grouped phenotypes. Variants located within blocks showing significant correlations (p-value < 0.05) were assigned to the corresponding pleiotropy groups. For each pleiotropy group, a PD-PRS was constructed using variants assigned to that group and SDPR-derived genome-wide PRS weights, implemented with PLINK v1.9 [ 23 ]. An additional PD-PRS was defined using variants not assigned to any pleiotropy groups. In total, 13 PD-PRSs were generated for each cancer.

To identify cancer pairs sharing polygenic risk, we conducted pairwise association analyses. For each cancer pair ( , ), cancer was modeled as the outcome, and its association with the PRS for cancer was evaluated using logistic regression, adjusting for the PRS for cancer , age at recruitment, sex (except for sex-specific cancers), and the top 10 genetic principal components (PCs). The inclusion of PRS for cancer was intended to evaluate whether the PRS for cancer provided information on the risk of cancer beyond that explained by the PRS for cancer . This procedure was repeated across all cancer pairs, and statistically significant associations were identified. Because cancer comorbidity may be influenced by factors beyond shared common polygenic susceptibility, including rare variants, somatic alterations, and environmental exposures, we conducted sensitivity analyses excluding individuals diagnosed with both cancers to assess the robustness of the observed associations. Specifically, analyses were repeated among participants diagnosed with cancer but not cancer . Cancer pairs that remained significant after exclusion of comorbid cases were considered to share polygenic risk independent of clinical co-occurrence.

To evaluate cross-cancer polygenic associations, we constructed genome-wide PRSs for each specific cancer. PRS weights were inferred from cleaned GWAS summary statistics using Summary statistics-based Dirichlet Process Regression (SDPR) [ 21 ], a Bayesian shrinkage method that does not require parameter tuning and has demonstrated stable predictive performance across multiple cancers [ 22 ]. SDPR employs a Dirichlet process prior on SNP effect-size variances, resulting in continuous shrinkage of effect estimates toward zero without explicitly assuming a fixed proportion of causal variants. Individual-level PRSs were calculated in the UKB participants as the weighted sum of risk alleles using PLINK 1.9 [ 23 ]. All PRSs were standardized to have a mean of zero and a standard deviation of one. Predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC), and the statistical significance was assessed accordingly. Associations between PRSs and cancer outcomes were summarized using odds ratios (ORs).

Identified regions were annotated to characterize broader biological relevance in two complementary ways. First, we evaluated correlations with non-cancer phenotypes. GWAS summary statistics for 3,950 non-cancer traits, restricted to individuals of genetically inferred European ancestry, were obtained from the Neale Lab UKB analyses [ 18 ]. For each selected region, phenotypes with enriched local heritability (p-value < 0.05) were identified using LAVA [ 19 ]. Local genetic correlations between cancers in the region and the selected non-cancer phenotypes were then estimated using SUPERGNOVA, with significance assessed using Bonferroni correction. Second, variants within the regions of interest were extracted and annotated using the SNP2GENE function in FUMA [ 20 ] with the default positional mapping settings. Mapped genes were further examined using the GENE2FUNC module to identify previously reported associations in the GWAS Catalog. Through this approach, regional genetic correlations were linked to known variant-phenotype associations.

Local genetic correlations were further evaluated using SUPERGNOVA [ 16 ], a method that provides stable estimates of regional genetic correlations [ 17 ]. The genome was partitioned into more than 2,000 approximately independent linkage disequilibrium (LD) blocks, and block-specific genetic correlations were computed for each cancer pair. Statistical significance was assessed using Bonferroni correction within each cancer pair. To identify shared correlated regions, we hypothesized that regions exhibiting local genetic correlations may contribute to shared risk across multiple cancers in a coordinated manner. Accordingly, we identified genomic regions having correlated cancer clusters involving at least three distinct cancer types. These regions were retained for downstream analyses.

Analyses were restricted to unrelated participants of genetically inferred European ancestry [ 14 ]. We used phase III UKB genotype data, in which participants were genotyped using either the UK BiLEVE Axiom Array or the UK Biobank Axiom Array, covering approximately 820,000 variants. Genotypes were centrally imputed using the 1000 Genomes Project and Haplotype Reference Consortium (HRC) reference panels, resulting in approximately 93 million variants per individual. Standard quality control procedures were applied, retaining autosomal variants with an imputation quality score > 0.3, a Hardy-Weinberg equilibrium p-value > 1 × 10 -5 , and an MAF > 0.05.

Individual-level data from the UKB were used for PRS analyses. UKB is a large prospective cohort study that has enrolled over 500,000 participants aged 40–69 years across the United Kingdom [ 13 ]. Cancer case status was ascertained using a combination of self-reported diagnoses and linked hospital records, incorporating International Classification of Diseases, Ninth and Tenth Revisions (ICD-9 and ICD-10), as well as Office for National Statistics Classification of Interventions and Procedures (OPCS) codes. The specific codes used to define cancer outcomes are provided in S2 Table . Individuals with any cancer diagnosis were classified as cases, and controls were defined as participants without a recorded diagnosis of any cancers.

All summary statistics underwent quality control procedures. We excluded non-autosomal variants, insertion-deletion polymorphisms, and variants not present in the UKB genotype. For multi-allelic variants, a single allele was retained based on the smallest association p-value. The cleaned summary statistics were then harmonized and processed using LD score regression (LDSC) [ 12 ] to estimate SNP-based heritability, restricting analyses to variants with minor allele frequency (MAF) ≥ 0.001. Cancers with heritability not significantly larger than 0 were excluded from downstream analyses.

We downloaded GWAS summary statistics for 24 specific cancers and one pan-cancer phenotype, restricting analyses to studies of European ancestry and excluding datasets that included UKB participants [ 2 – 4 , 10 , 11 ]. Sample sizes and study details for all cancer phenotypes are provided in S1 Table . Briefly, the breast cancer GWAS included 133,384 cases and 113,789 controls [ 2 ], the prostate cancer GWAS included 79,194 cases and 61,112 controls [ 3 ], the lung cancer GWAS comprised 44,069 cases and 68,712 controls [ 4 ], and the ovarian cancer GWAS consisted of 25,509 cases and 40,941 controls [ 10 ]. The remaining 21 cancer GWASs were derived from the FinnGen R12 release [ 11 ], which used a shared set of cancer-free controls (except for sex-specific cancers including cervix uteri and testis cancers) and included between 152 cases for nasopharyngeal cancer and 121,495 cases for the pan-cancer phenotype. The pan-cancer was defined as diagnosis of any malignant neoplasm. FinnGen is a large-scale national research initiative integrating genotype data from over 500,000 Finnish biobank participants with comprehensive longitudinal health registry data to study disease susceptibility.

We next examined pairwise polygenic overlap using PRSs in UK Biobank (UKB) participants. Genome-wide PRSs were used to identify cancer pairs exhibiting significant cross-cancer associations. To further investigate the potential sources of shared polygenic risk, PRSs were decomposed into components with different functions, enabling attribution of cross-cancer associations to pleiotropy groups and biological pathways.

Using GWAS summary statistics for 24 specific cancers and one pan-cancer phenotype, we first characterized shared genetic susceptibility across cancers by estimating genetic correlations at both genome-wide and regional scales. To provide functional context for regional genetic sharing, we further annotated correlated regions by evaluating their genetic correlations with non-cancer phenotypes and by linking variants within these regions to previously reported associations in the GWAS Catalog.

Results

An overview of the study design and analytic workflow is shown in Fig 1.

PNG larger image

TIFF original image Download: Fig 1. Overview of the study design. This figure summarizes the analytic framework used to characterize shared polygenic risk across cancers. Panel A shows the genetic correlation analyses based on GWAS summary statistics. Genome-wide genetic correlations were estimated pairwise using GNOVA, and local genetic correlations were calculated using SUPERGNOVA to identify genomic regions harboring correlated cancer clusters. Regions with shared signals were further annotated using genetic correlations with non-cancer traits and variant mappings from the GWAS Catalog. Panel B illustrates the PRS analyses. Genome-wide cancer PRSs were constructed using SDPR and evaluated in cross-cancer association analyses to identify cancer pairs with shared polygenic risk. For selected pairs, PRSs were further decomposed into pleiotropy-informed and pathway-specific components to assess functional enrichment of shared genetic risk. https://doi.org/10.1371/journal.pgen.1012308.g001

GWAS summary statistics We downloaded and cleaned GWAS summary statistics for 24 specific cancers and one pan-cancer phenotype. SNP-based heritability was estimated using LDSC [12]. Seventeen cancers showed significantly enriched heritability except for testis cancer, stomach cancer, oral cavity cancer, cervix uteri cancer, acute myeloid leukemia, nasopharyngeal cancer, chronic myeloid leukemia, and oropharyngeal cancer. Among the cancers with significant heritability estimates, prostate and breast cancers showed the highest estimates, h2 = 0.13 ± 0.02; and h2 = 0.13 ± 0.01, respectively, whereas acute lymphocytic leukemia (ALL) had the lowest (h2 = 0.002 ± 0.001) (S1 Table).

PRS analyses PRS weights were inferred for 16 specific cancers using SDPR, and individual-level PRSs were calculated for UKB participants of European ancestry. Discriminative performance was evaluated using AUC, and relative risk was summarized using ORs. Results are shown in S9 Table. The number of cases ranged from 124 for ALL to 25,666 for NMSC. Fifteen PRSs except ALL PRS showed AUCs and ORs significantly different from the null hypothesis (p-value < 0.05) and were retained for subsequent analysis. Pairwise correlations among the 15 PRSs are shown in S1A Fig. Most PRSs were significantly correlated (p-value < 4.76 × 10-4), although the magnitudes of correlation were modest. Statistical significance was likely influenced by the large number of shared controls, whereas the small effect sizes reflected limited genome-wide genetic overlap across cancers. We further examined overlap among individuals in the top 10% of each cancer (S1B Fig). Consistent with the modest correlation coefficients, the proportion of shared high-risk individuals was approximately 10% for most cancer pairs. An exception was observed for NMSC and MSC, for which nearly 20% overlap was detected. Overall, these results indicate that genome-wide polygenic overlap across cancers is present but generally limited in magnitude. We next assessed associations between 15 cancer-specific PRSs and specific cancer outcomes in the UKB using a two-step approach. In the first step, we tested cross-cancer associations by modeling each cancer outcome as a function of the PRS for another cancer, while adjusting for the target cancer's PRS. Among the 345 cancer-PRS pairs examined, 18 showed significant associations after Bonferroni correction (p-value < 1.45 × 10-4; Table 2). In the second step, we assessed whether these associations could be explained by clinical comorbidity by excluding individuals diagnosed with both cancers. After Bonferroni correction (p-value < 2.78 × 10-3), seven cancer-PRS pairs remained statistically significant (Table 2). These included associations between bladder cancer risk and the lung cancer PRS, as well as associations between lung cancer risk and the bladder cancer PRS. To further evaluate whether the identified shared polygenic risk can be fully attributed to smoking, we conducted stratified analyses for the bladder–lung cancer pair by smoking status. The association between bladder cancer and the lung cancer PRS remained significant among both ever and never smokers. In contrast, associations between lung cancer and the bladder cancer PRS were no longer significant in either stratum, but the effect estimates were similar to those observed in the pooled analysis (S10 Table), indicating potential loss of power in stratified analyses. Our results suggest that smoking-related mechanisms alone are unlikely to account for the observed genetic overlap between bladder and lung cancers. Based on these results, subsequent analyses focused on six cancer-PRS pairs: bladder cancer-lung cancer PRS, lung cancer-bladder cancer PRS, breast cancer-NMSC PRS, MSC-NMSC PRS, NMSC-CLL PRS, and NMSC-lung cancer PRS. Because CLL is a clinical subtype of non-Hodgkin lymphoma, the NMSC-CLL pair was excluded from further analyses. PNG larger image

TIFF original image Download: Table 2. Significant cross-cancer associations for genome-wide PRSs. https://doi.org/10.1371/journal.pgen.1012308.t002 PRSs involved in the six selected cancer pairs—lung cancer PRS, bladder cancer PRS, NMSC PRS, and CLL PRS—were involved for the following decompositions. Specifically, we applied two decomposition frameworks to the four selected genome-wide PRSs to further characterize sources of shared polygenic risk. Specifically, PRSs were decomposed into pleiotropy-informed PD-PRSs and pathway-specific PS-PRSs, and cross-cancer associations were evaluated using these component scores.

PD-PRSs The genome-wide PRSs for bladder cancer, lung cancer, NMSC, and CLL were decomposed into 14 PD-PRSs, with the number of SNPs in each component summarized in S2 Fig. For all three cancers, the largest proportion of variants was assigned to the other PD-PRS, which comprised variants located in genomic regions showing no detectable genetic correlation with the 47 non-cancer phenotypes. Associations between PD-PRSs and their corresponding cancers were evaluated using logistic regression models with covariate adjustment (S3 Fig). All PD-PRSs for lung cancer and NMSC were significantly associated with their respective cancers after Bonferroni correction (p-value < 8.93 × 10-4; S3C-S3D Fig). For bladder cancer, associations with alcohol consumption- and hypertension-related PD-PRSs failed to reach significance (S3A Fig). For CLL, two PD-PRSs—alcohol consumption and chronic kidney disease—were not significantly associated and were excluded from the following analyses (S3B Fig). Despite containing the largest number of variants, the other PD-PRS did not consistently yield the strongest associations with cancer risk. Instead, higher ORs were often observed for PD-PRSs corresponding to specific pleiotropy groups, suggesting that cancer-associated genetic risk may be enriched invariants shared with non-cancer phenotypes. For bladder cancer, the neuropsychiatric disease PD-PRS showed the strongest association (OR = 1.29, 95% CI: 1.26-1.33), implicating enrichment of related function in the development of bladder cancer. For CLL, the autoimmune disease PD-PRS showed the strongest association (OR = 1.45, 95% CI: 1.38-1.54), supporting an immune-related component of genetic susceptibility. For lung cancer, the diabetes PD-PRS exhibited the largest effect size (OR = 1.27, 95% CI: 1.23-1.30), consistent with evidence linking metabolic dysfunction, insulin resistance, and diabetes-related pathways to cancer development and progression [33].. For NMSC, the neuropsychiatric disease PD-PRS showed the strongest association (OR = 1.46, 95% CI: 1.44–1.48), suggesting that genetic components enriched for neuropsychiatric disease may also contribute to NMSC susceptibility. Although the underlying mechanisms remain unclear, previous studies have reported shared genetic architecture between neurological disorders such as Parkinson’s disease and skin cancers, potentially reflecting overlap in pigmentation biology, neural crest development, DNA damage response, and immune regulation [34]. Further, we analyzed cross-cancer associations (excluding comorbidity) using PD-PRSs for the six cancer pairs identified in prior analyses. Among 82 tested associations, 30 were statistically significant (p-value < 0.05; Fig 3 and S11 Table). For the bladder-lung cancer pair, the diabetes PD-PRS for lung cancer showed the strongest association with bladder cancer risk (OR = 1.06, 95% CI: 1.03-1.09), consistent with its prominent role in lung cancer susceptibility. While lung cancer-bladder cancer PRS pair, the diabetes PD-PRS for bladder cancer was not associated with lung cancer risk (OR = 1.02, 95% CI: 0.99-1.04), and the lipids PD-PRS showed the strongest association (OR = 1.04, 95% CI: 1.01-1.07). For the breast-NMSC pair, four PD-PRSs showed significant associations with comparable effect sizes. Notably, the neuropsychiatric PD-PRS, which showed the strongest association with NMSC, was not associated with breast cancer but instead showed its strongest association with MSC, suggesting cancer-specific patterns of pleiotropy. For the NMSC-CLL pair, two CLL PD-PRSs—others and smoking PD-PRSs—were significantly associated with NMSC, although the association with the smoking PD-PRS was inverse (OR = 0.99, 95% CI: 0.97-1). In addition, seven lung cancer PD-PRSs, including diabetes PD-PRS, were significantly associated with NMSC risk. Together, these findings suggested that cross-cancer PRS associations are driven by specific pleiotropic components rather than by uniform genome-wide effects. PNG larger image

TIFF original image Download: Fig 3. Cross-cancer associations of PD-PRS. ORs and 95% confidence intervals for associations between PD-PRSs and cancer outcomes after excluding individuals with comorbid cancers are shown for six cancer pairs of interest. Each panel corresponds to a target cancer outcome, and PD-PRSs were derived from the genome-wide PRS of the paired cancer. Asterisks indicate nominally significant associations (p-value < 0.05). Across pairs, only a subset of PD-PRS components showed significant associations, indicating that cross-cancer polygenic overlap is concentrated in specific pleiotropy-defined trait domains rather than uniformly distributed across the genome. https://doi.org/10.1371/journal.pgen.1012308.g003 To formally assess whether cross-cancer associations could be attributed to specific PD-PRSs, we evaluated attenuation in explained variance using differences in Nagelkerke’s . Among the 30 significant associations, 26 showed significantly different values between the PD-PRS and the remaining PRS (permutation p-value < 0.05; Fig 4). In 23 of these associations, the PD-PRS explained a larger proportion of the variance than the remaining PRS, suggesting enrichment of shared genetic risk within specific pleiotropy components. For bladder cancer risk, lung cancer PD-PRSs related to atherosclerotic disease, chronic kidney disease, diabetes, and neuropsychiatric disease showed significantly greater , while for lung cancer, all five associated bladder cancer PD-PRSs related to atherosclerotic disease, autoimmune disease, chronic kidney disease, lipids, and neuropsychiatric disease showed significantly greater All four PD-PRSs for NMSC associated with breast cancer, as well as all seven PD-PRSs for NMSC associated with MSC, demonstrated larger values. In contrast, for the NMSC-CLL pair, only the other PD-PRS showed greater variance explained, suggesting that additional or uncharacterized mechanisms may contribute to this association. Overall, decomposing genome-wide cancer PRSs into pleiotropy-informed components enabled quantification of relative contributions of specific phenotype groups to cross-cancer associations. PNG larger image

TIFF original image Download: Fig 4. Attenuation of explained variance ( Attenuation of explained variance () for PD-PRSs. For each selected cancer pair, the proportion of variance explained by the PD-PRS (red) was compared with that explained by the corresponding remaining PRS (blue). Each panel represents one selected cancer-PRS pair. Asterisks indicate PD-PRS components for which the difference in between the PD-PRS and remaining PRS was statistically significant based on the permutation testing (1,000 permutations). Although the overall variance explained by cross-cancer PRSs was modest, selected PD-PRS components accounted for a larger share of the explained variance, indicating enrichment of shared polygenic risk within specific functions. https://doi.org/10.1371/journal.pgen.1012308.g004

[END]
---
[1] Url: https://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1012308

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/