This preprint reports the first large-scale cross-trait and multi-polytranscriptomic score (PTS) analysis applied to Parkinson's disease (PD). Using whole-blood RNA-sequencing from three independent clinical cohorts within the Accelerating Medicines Partnership Parkinson's Disease (AMP-PD) programme (N = 2,741; cases = 1,644), the authors computed approximately 550 PTS representing gene expression signals for 100 phenotypes. They tested these PTS for cross-trait association with PD and found 26 Bonferroni-significant associations (p < 9×10-5) across 18 phenotypes and 11 trait categories, including neurodegenerative disease, respiratory function, sleep and cardiovascular traits. Incorporating multiple PTS into machine learning multi-PTS models improved PD case/control prediction beyond age, sex and a PD polygenic score (PGS), with a sparse model adding seven PTS achieving an AUC of 0.74 (0.70–0.78) on external validation and improving AUC by 0.09.
The analysis used whole-blood RNA-seq data from three AMP-PD clinical cohorts; the combined sample size was N = 2,741 with 1,644 PD cases. PTS were defined as the sum of an individual's observed gene expression values weighted by transcriptome-wide association z-scores derived from transcriptome-wide association study methods. Approximately 550 PTS were calculated, covering 100 different phenotypes. The design aims to combine the relative stability of inherited genetic effects with the dynamic, measured biology of gene expression to detect associations not captured by genetics alone.
Cross-trait testing of PTS against PD status identified 26 associations that passed a Bonferroni threshold (p < 9×10-5). These associations spanned 18 phenotypes across 11 trait categories. The implicated categories included neurodegenerative diseases, respiratory measures, sleep-related phenotypes and cardiovascular traits, among others. The manuscript highlights that these associations indicate transcriptomic overlap between PD and a diverse set of clinically relevant traits.
The authors compared PTS-derived associations with those obtained from corresponding polygenic scores (PGS). They report that only one of the 26 Bonferroni-significant cross-trait PTS associations was also observed using PGS. This contrast is presented as evidence that integrating directly measured transcriptomic data can reveal relationships missed by genetics-only approaches such as PGS.
Multiple PTS were combined in machine learning models to predict PD case/control status. Models included baseline covariates of age and sex and incorporated a PD-PGS. A sparse multi-PTS model that added seven PTS to the baseline covariates and PD-PGS produced improved discrimination on external validation, achieving an AUC of 0.74 with 95% interval 0.70–0.78. The authors report this as a 0.09-point gain in AUC compared with the model excluding the additional PTS, illustrating potential predictive benefit from measured expression-derived scores.
The findings suggest measurable blood transcriptomic overlap between PD and multiple trait domains, including other neurodegenerative signals, respiratory function, sleep and cardiovascular traits. The authors argue that combining genetic stability with observed gene expression through PTS can uncover novel trait relationships and improve risk stratification models. They emphasize that these results may provide novel biological insights and potential to improve disease prediction, while noting that the work remains exploratory and reported as a preprint.
Data used in this project are available to approved researchers through the AMP-PD platform, with Tier 1 and Tier 2 access procedures described by AMP-PD. Software and derived resources cited in the manuscript include FUSION for TWAS analysis, SMR for summary-based Mendelian randomization, and GCTB for SBayesRC analyses and LD/reference resources. Full TWAS and SMR results, and eQTL FUSION-format weights and validation metrics are reported as deposited on Zenodo. The manuscript lists publicly accessible URLs for these software packages and data deposits.
This manuscript is a preprint and has not undergone peer review; the authors explicitly state it should not be used to guide clinical practice. The study used AMP-PD data with documented participant consent and institutional approvals, and the authors confirm adherence to relevant ethical guidelines. One author (OP) reports consultancy work for UCB; all other authors declare no competing interests.
Funding sources acknowledged in the manuscript include Parkinson's UK, Aligning Science Across Parkinson's, the Michael J. Fox Foundation and the Wellcome Trust (grant referenced). The preprint lists these funders and states the copyright terms under a CC-BY 4.0 International license. The authors provide details on data access and code availability to support reproducibility.
Note: All statements here are drawn from the preprint text. The original article is a preprint posted August 26, 2026, and specific analytic details, full results and supplementary materials are available in the manuscript and linked data deposits cited by the authors.