Stephen S. Rich

Background: Socioeconomic status (SES) is strongly associated with cardiovascular disease burden and risk. Plasma proteins may elucidate a biological mechanism through which low SES increases the risk of cardiovascular disease. Methods: Participants from the following 3 cohorts were included: the Multi‐Ethnic Study of Atherosclerosis (n=2139), Jackson Heart Study (n=967), and Women's Health Initiative (n=1104). SES was approximated using a cumulative score (range 0–3) across education, income, and employment status. A meta‐analysis of Cox proportional hazards ratios was used to assess the association between SES and incident events for coronary heart disease (CHD) and stroke. Linear regression was used within each cohort to determine associations between SES and Olink proteins (n=1270 proteins) and the results were meta‐analyzed across cohorts. Regression‐based mediation analysis was used to determine whether the identified proteins were statistical mediators of the association of SES and incident events. Results: Higher SES scores were associated with a lower risk of CHD (hazard ratio [HR], 0.74 [95% CI, 0.63–0.87]) and stroke (HR, 0.81 [95% CI, 0.66–1.00]). After correcting for multiple testing ( P −5 ), 233 proteins were associated with SES in the meta‐analysis. Of these 233 proteins, 21 had a significant indirect effect on SES and CHD. Because the SES‐stroke association was not statistically significant, proteins were not tested as mediators. Conclusions: Our analysis of multiple cohorts identified evidence of different proteomic pathways associated with SES and cardiovascular disease. This work suggests that exogenous social factors (eg, SES) may influence cardiovascular disease via proteomic profiles.

Replication of scientific findings across independent studies is essential, yet it remains challenging in high-dimensional biological datasets. A critical aspect of true replication requires detecting not only the repeated presence but also the consistent direction of an effect. Traditional methods for evaluating replication often ignore effect directionality or lack adequate statistical power when adjusting for multiple comparisons. To address this, we extend the directional MaxP test (DMT) and our screening-integrated procedure, ScreenDMT, to evaluate directional replication across an arbitrary number of independent studies. We mathematically prove ScreenDMT’s validity for an arbitrary number of studies when the studies’ sample sizes grow unbounded. Through extensive simulation studies, we demonstrate that ScreenDMT effectively controls error rates, such as the false discovery rate (FDR), in finite sample sizes and yields higher statistical power than competing methods especially for three studies. We applied the extended ScreenDMT to untargeted blood metabolomics profiles across three diverse human cohorts—ARIC (N=1,337), MESA (N=1,874), and HCHS/SOL (N=6,180)—to discover replicable biomarkers for dichotomous insomnia, long sleep, short sleep, and excessive daytime sleepiness (EDS). We found low levels of cortisol to be replicably associated with EDS, and high levels of n-formylmethionine to be associated with long sleep. Furthermore, directional pathway analysis revealed that low levels of xanthine metabolites are replicably associated with insomnia. These convergent findings validate ScreenDMT as a powerful methodology for simultaneous multi-cohort replication and offer insights into the metabolic signatures underlying sleep disorders across diverse populations.

Objective: Anti–β ₂ ‐glycoprotein I (anti‐β ₂ GPI) antibodies are central to antiphospholipid syndrome. The APOH locus has been associated with anti‐β ₂ GPI, but the causal variant and thrombotic risk implications remain unclear. Methods: We performed a multiancestry genome‐wide association study of quantitative total anti‐β ₂ GPI antibody levels in 5,969 participants from the Multi‐Ethnic Study of Atherosclerosis (MESA). The relationship between genetically determined anti‐β ₂ GPI levels and venous thromboembolism (VTE) risk was evaluated using two‐sample Mendelian randomization and Bayesian colocalization. Candidate causal variants were prioritized by fine‐mapping and integrative functional genomic analyses. Molecular dynamics simulations were used to investigate the structural effects. Results: We identified a genome‐wide significant association at the APOH locus (lead single‐nucleotide polymorphism rs1801690‐G, β = 0.21, P = 1.08 × 10 ⁻¹⁰ ). Paradoxically, genetic variation at the APOH locus associated with higher anti‐β ₂ GPI levels was associated with reduced VTE risk (β = −0.25, P = 3.95 × 10 ⁻⁶ ; posterior probability for hypothesis 4 (PP4) = 0.97), likely reflecting horizontal pleiotropy rather than a true protective antibody effect. Fine‐mapping and functional genomics prioritized the missense variant rs1801690 (W335S) as the most likely causal variant. Molecular dynamics simulations support a dual‐effect mechanism whereby W335S impairs phospholipid binding in domain V while increasing epitope exposure in domains I and II. Conclusion: The APOH genetic signal associated with higher anti‐β ₂ GPI levels paradoxically reduced VTE risk, likely mediated by W335S via epitope accessibility and disrupted phospholipid binding. These findings provide human genetic evidence that phospholipid binding in domain V is essential for anti‐β ₂ GPI–mediated thrombosis. The clinical utility of the APOH genotype warrants further investigation.

Key Points: An integrative risk framework with genomic, epigenomic, and clinical factors was developed to predict risk of new-onset kidney failure in a study. Addition of polygenic and methylation information to a clinical risk score modestly improved the prediction of kidney failure. Polygenic and methylation risk scores alone were not superior to a clinical risk algorithm. Background: Kidney failure affects approximately 800,000 adults in the United States, and African American individuals carry a disproportionate burden of disease. Robust algorithms for clinical risk prediction, such as the kidney failure risk equation (KFRE), have previously been validated, whereas the clinical utility of polygenic risk scores (PRS) and methylation risk scores (MRS) have not been as well characterized. Methods: In this study, we evaluated the capacity of PRS and MRS for CKD to predict multiple clinical end points, including incident kidney failure and decline in kidney function. Leveraging genomic, epigenomic, and clinical data from African American participants of the Hypertension Genetic Epidemiology Network (HyperGEN), Multi-Ethnic Study of Atherosclerosis (MESA), and Electronic Medical Records and Genomics Network, we evaluated the performance of these risk scores in comparison with and in conjunction with the widely validated KFRE. Results: In HyperGEN, a higher PRS was associated with a higher risk of kidney failure. Similarly, the MRS was associated with incident kidney failure in both HyperGEN and MESA. Neither PRS nor MRS were associated with the rate of decline in kidney function in Electronic Medical Records and Genomics or MESA. The KFRE outperformed both PRS and MRS as a single predictor. However, multiscore models suggested a synergistic effect of PRS and MRS to the predictive accuracy of the KFRE, with the greatest risk of kidney failure among individuals in high-risk strata of KFRE, PRS, and MRS. Conclusions: Results from this exploratory analysis suggest that, although PRS and MRS may not be superior to clinical algorithms, inclusion of -omics information may improve disease risk prediction. Future studies should seek to increase sample size to better evaluate these risk models and improve their capacity.

Circulating diabetes biomarkers have been identified with proteomics measured at a single time point, but what is additionally provided by longitudinal repeated measurements in the same individuals, over many years, is unknown. We studied participants in the Multi-Ethnic Study of Atherosclerosis (MESA; n = 5,322; mean baseline age, 61.7 years) at exams 1 (2000–2002), 5 (2010–2012), and 6 (2016–2018) profiled with the Olink Explore (3 K) platform. Associations with incident diabetes, mostly of type 2, were modeled using Cox proportional hazards with exam 1 proteins (i.e., single time point) and time-updating Cox with proteins from all three exams (i.e., longitudinal repeated) adjusted for clinical risk factors. We identified 27 novel single time point associations and up to a fourfold increase in longitudinal associations (false discovery rate < 0.05), with a proportional increase in the number consistent with causality via cis-Mendelian randomization (∼5%) and a smaller overlap in MESA longitudinal versus single associations with UK Biobank single time point findings (42% vs. 87%, respectively). Compared with small-molecule metabolism pathway enrichment among the shared proteins, proteins unique to the longitudinal analyses enriched for protein and cellular processing pathways. We conclude that longitudinal repeated measurements identify a number of distinct disease biomarkers, in part by revealing the progression of relevant biological processes within an individual closer to a clinical diabetes diagnosis versus single measurements. Article Highlights: There is up to a fourfold increase in diabetes biomarkers identified with longitudinal repeated versus single time point proteomic measurements. The increase in biomarkers identified with longitudinal repeated measurements is supported by a similar proportion being nominated as causal for type 2 diabetes with Mendelian randomization. Proteins unique to the longitudinal repeated analyses highlighted biological pathways (e.g., posttranslational protein modification and cellular structure and cycle regulation) that were distinct from pathways enriched among the shared proteins (e.g., small-molecule metabolic and catabolic processes). Longitudinal protein measurements identify additional novel disease biomarkers and disparate biological pathways compared with single measurement analyses.

Plasma proteomics is increasingly used for biomarker discovery and predictive modeling, yet diurnal protein trajectories remain insufficiently characterized. In our review of recent proteomic biomarker studies, 43% of the identified biomarkers had previously been reported to display 24-h rhythmicity. We demonstrate that ignoring these short-term dynamic effects compromises the robustness of reported models predicting health outcomes. We integrated a population-scale multi-ethnic longitudinal cohort with repeated measures over 10 years, with two cohorts of healthy adults undergoing frequent plasma sampling across days under controlled circadian, sleep and food-intake conditions. This design enabled estimation of short-term intraindividual variability (ST), long-term intraindividual variability (LT), population-level variability (POP) and genetic effects (GEN) across 7,289 protein targets. ST, LT, POP, and GEN define diverse protein trajectories, including rapid dynamics, long-term change, and individual-specific signatures. Using and generalizing this framework will facilitate covariate selection, study design, biomarker prioritization, and variability-aware modeling by users of proteomic data. Graphical

Background: Psychosocial stress is a key risk factor for coronary heart disease (CHD), particularly in postmenopausal women who face both a high stress burden and elevated cardiovascular risk. DNA methylation (DNAm), a critical epigenetic modification bridging environment and health, remains understudied as a contributor to stress-related CHD. Methods: We conducted an epigenome-wide association study (EWAS) of stress in the Women's Health Initiative (WHI), an ancestrally diverse cohort of postmenopausal women (n=3,857). At screening visit, participants completed a questionnaire assessing stressful life events and provided whole blood for DNAm. Incident CHD was then longitudinally ascertained (follow-up mean/SD: 16.7/8.4 years), and DNAm signatures were evaluated as CHD predictors using Cox regression. Predictive models were independently validated in the Jackson Heart Study (JHS; n=3,053) and Multi-Ethnic Study of Atherosclerosis (MESA; n=870). The bulk-level DNAm associations were computationally deconvolved at the cell-type-specific level using tensor composition analysis (TCA). Results: The EWAS in WHI identified 841 stress-related DNAm sites (99 hypermethylated, 742 hypomethylated with stress) after FDR correction, with 13 significant after Bonferroni correction, including sites located on immune and CHD-related genes (e.g., TNF, ALDH2). Methylation risk scores (MRSs) integrating the 841 FDR-significant sites (MRS ₈₄₁ ) and 13 Bonferroni-significant sites (MRS ₁₃ ) predicted incident CHD (HR=1.33-1.37; p≤0.0008) and mediated 16.5-17.7% of the association between stress and CHD. In JHS and MESA, MRS ₁₃ independently predicted CHD (HR=1.34; p=0.036), whereas MRS ₈₄₁ was suggestively associated with CHD (HR=1.27; p=0.087). TCA indicated that the greatest number of stress-related sites predictive of CHD was specifically in monocytes (133 total), with directions consistent with bulk-level associations (9 hypermethylated, 124 hypomethylated with stress). Conclusion: Our study supports methylation risk scores as novel biomarkers of stress-related CHD and uncovers epigenetic regulation in monocytes as a potential underlying mechanism. These findings highlight biological pathways linking stress and disease and may promote personalized interventions in high-risk populations.