Abstract
Most polygenic risk scores (PRS) are derived and validated using genetic data from European populations. However, European-based PRS perform poorly in individuals of non-European ancestry, and their transferability to the admixed Brazilian population remains unknown. PRS derived from the United Kingdom Biobank (UKB) were selected from the PGS Catalog, including 33 scores for body mass index (BMI), 36 for systolic blood pressure (SBP), and 33 for diastolic blood pressure (DBP). PRS were applied to 4,758 participants from two geographically distinct Brazilian cohorts (São Paulo and North Minas Gerais) and a UKB sample. Performance was evaluated across self-identified racial subgroups and in Brazilian individuals genetically similar to the UK sample, as determined by genomic clustering techniques (Uniform Manifold Approximation and Projection, UMAP; and Principal Component Analysis, PCA). Effect sizes were compared using multivariable mixed-effects models. Most BMI PRS were validated in São Paulo (96.7%) and North Minas Gerais (90.9%), whereas blood pressure PRS showed lower validation rates (SBP: 66.7 and 38.9%; DBP: 69.7 and 54.5%, respectively). Validated PRS consistently exhibited lower effect sizes in Brazilian cohorts compared to the UKB (p < 0.001). PRS calibration for obesity and hypertension using cohort-specific quintiles and precision-recall F1 scores did not improve performance. BMI PRS effects were slightly higher in São Paulo than in North Minas Gerais (+0.39 kg/m2 per SD, p < 0.001), whereas SBP and DBP effects did not differ significantly between regions. BMI PRS effects were higher in Brazilian Whites than Non-Whites (+0.60 kg/m2 per SD, p < 0.001), but blood pressure PRS effects were similar in the racial subgroups. Across all traits, PRS effects were higher in the UKB than in Brazilian individuals clustering with the British by UMAP/PCA (p < 0.001), and no differences were observed between Brazilian UMAP/PCA subgroups. European-derived PRS are not directly transferable to the Brazilian population without prior empirical validation. Regional origin, self-identified race, and genetic similarity to the UKB do not reliably identify Brazilian subgroups with differential PRS performance. These findings highlight the urgent need for polygenic scores derived and trained on Brazilian genomic data.</p>