{"paper_id":"05bdc25a-2ad2-4798-ab3f-2f8ee4f92bd4","body_text":"1\nRefinement of a published gene-physical activity interaction impacting HDL-cholesterol: \nrole of sex and lipoprotein subfractions \n \nKenneth E. Westerman1,2,3† , Tuomas O. Kilpeläinen4,5, Magdalena Sevilla-Gonzalez1,2,3, \nMargery A. Connelly6, Alexis C. Wood7, Michael Y. Tsai8, Kent D. Taylor9, Stephen S. Rich10, \nJerome I. Rotter11, James D. Otvos12, Amy R. Bentley13, Samia Mora14,15,16, Hugues Aschard17,18, \nDC Rao19, Charles Gu19, Daniel I. Chasman15,20,21, Alisa K. Manning1,2,3† , on behalf of the \nCHARGE Gene-Lifestyle Interactions Working Group \n \n1 Clinical and Translational Epidemiology Unit, Massachusetts General Hospital, Boston, MA, USA \n2 Department of Medicine, Harvard Medical School, Boston, MA, USA \n3 Programs in Metabolism and Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, \nMA, USA \n4 Novo Nordisk Foundation Center for Basic Metabolic Research, Faculty of Health and Medical Sciences, \nUniversity of Copenhagen, Copenhagen, DK \n5 Novo Nordisk Foundation Center for Genomic Mechanisms of Disease, Broad Institute of MIT and Harvard, \nCambridge, MA, USA \n6 LabCorp Diagnostics, Morrisville, NC, USA \n7 USDA/ARS Children's Nutrition Center, Baylor College of Medicine, Houston, TX, USA \n8 Department of Laboratory Medicine and Pathology, University of Minnesota, Minneapolis, MN, USA \n9 The Lundquist Institute for Biomedical Innovation at Harbor-UCLA Medical Center, Torrance, CA, USA \n10 Center for Public Health Genomics, University of Virginia, Charlottesville, VA, USA \n11 The Institute for Translational Genomics and Population Sciences, Department of Pediatrics, The Lundquist \nInstitute for Biomedical Innovation at Harbor-UCLA Medical Center, Torrance, CA, USA \n12 Lipoprotein Metabolism Laboratory, Translational Vascular Medicine Branch, National Heart, Lung, and Blood \nInstitute, National Institutes of Health, Bethesda, MD, USA \n13 Center for Research on Genomics and Global Health, National Human Genome Research Institute, National \nInstitutes of Health, Bethesda, MD, USA \n14 Center for Lipid Metabolomics, Brigham and Women's Hospital, Boston, MA, USA \n15 Division of Preventive Medicine, Brigham and Women's Hospital, Boston, MA, USA \n16 Cardiovascular Division, Brigham and Women's Hospital, Boston, MA, USA \n17 Department of Computational Biology, Institut Pasteur, Université de Paris, Paris, FR \n18 Program in Genetic Epidemiology and Statistical Genetics, Harvard T.H. Chan School of Public Health, Boston, \nMA, USA \n19 Division of Biostatistics, Washington University, St. Louis, MO, USA \n20 Division of Genetics, Brigham and Women's Hospital, Boston, MA, USA \n21 Department of Epidemiology, Harvard T.H. Chan School of Public Health, Boston, MA, USA \n \n† Correspondence to: Kenneth E. Westerman (kewesterman@mgh.harvard.edu) and Alisa K. \nManning (akmanning@mgh.harvard.edu) \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \nNOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.\n\n   2\nAbstract \n \nLarge-scale gene-environment interaction (GxE) discovery efforts often involve compromises in \nthe definition of outcomes and choice of covariates for the sake of data harmonization and \nstatistical power. Consequently, refinement of exposures, covariates, outcomes, and population \nsubsets may be helpful to establish often-elusive replication and evaluate potential clinical \nutility. Here, we used additional datasets, an expanded set of statistical models, and interrogation \nof lipoprotein metabolism via nuclear magnetic resonance (NMR)-based lipoprotein subfractions \nto refine a previously discovered GxE modifying the relationship between physical activity (PA) \nand HDL-cholesterol (HDL-C). This GxE was originally identified by Kilpeläinen et al., with the \nstrongest cohort-specific signal coming from the Women’s Genome Health Study (WGHS). We \nthus explored this GxE further in the WGHS (N = 23,294), with follow-up in the UK Biobank \n(UKB; N = 281,380), and the Multi-Ethnic Study of Atherosclerosis (MESA; N = 4,587). Self-\nreported PA (MET-hrs/wk), genotypes at rs295849 (nearest gene: LHX1), and NMR \nmetabolomics data were available in all three cohorts. As originally reported, minor allele \ncarriers of rs295849 in WGHS had a stronger positive association between PA and HDL-C (p\nint \n= 0.002). When testing a range of NMR metabolites (primarily lipoprotein and lipid \nsubfractions) to refine the HDL-C outcome, we found a stronger interaction effect on medium-\nsized HDL particle concentrations (M-HDL-P; pint = 1.0×10-4) than HDL-C. Meta-regression \nrevealed a systematically larger interaction effect in cohorts from the original meta-analysis with \na greater fraction of women (p = 0.018). In the UKB, GxE effects were stronger both in women \nand using M-HDL-P as the outcome. In MESA, the primary interaction for HDL-C showed \nnominal significance (pint = 0.013), but without clear differences by sex and with a greater \nmagnitude using large, rather than medium, HDL-P as an outcome. Towards reconciling these \nobservations, further exploration leveraging NMR platform-specific HDL subfraction diameter \nannotations revealed modest agreement across all cohorts in the interaction affecting medium-to-\nlarge particles. Taken together, our work provides additional insights into a specific known gene-\nPA interaction while illustrating the importance of phenotype and model refinement towards \nunderstanding and replicating GxEs. \n \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   3\nBackground \n \nGene-environment interactions (GxEs), which describe the modification of environmental effects \non a phenotype by a genetic factor (or vice versa), inform attempts to better understand complex \nand polygenic traits such as cardiometabolic diseases and their risk factors. The emergence of \nlarger-scale cohorts and consortia have allowed for hypothesis-free GxE discovery efforts in \ngenome-wide interaction studies (GWIS). Such investigations have explored genetic \nmodification of relationships such as that between fish oil supplementation and plasma lipids in \nthe UK Biobank (UKB) cohort (Francis et al., 2021) or between and smoking and lipids in a \nlarge-scale meta-analysis (Bentley et al., 2019).  \n \nDespite increasing attention to GWIS, robust discovery and translation of their results faces \nmultiple challenges. First, loci uncovered using such hypothesis-free approaches often lack \nevidence of clear biological mechanisms by which the genetic and environmental pathways \noverlap. This is an extension of the more general “variant to function” challenge in the genetics \ncommunity, but with the added obstacle of incorporating an understanding of the interacting \nexposure. Second, replication of GxEs has proven to be a major challenge. Part of this replication \nchallenge can be attributed to low statistical power for identifying interactions (Gauderman et al., \n2017), which are a limiting factor for typical sample sizes used to-date (Westerman et al., 2023). \nPart is also due to heterogeneity across cohorts in the distribution and measurement of exposures \nas well as the complex confounding structure found in the observational datasets that are used for \nmost GxE studies. \n \nBoth of these challenges can be addressed by “refinement” of GxEs originally identified in large-\nscale, hypothesis-free discovery efforts. A more fine-grained understanding of the specific \ninteracting elements (e.g., low- versus high-intensity physical activity [PA]) and relevant sub-\ncohorts (e.g., GxEs that are specific to a given sex or ancestry subgroup) can inform more \neffective replication efforts, including cohort selection and modeling choices. Furthermore, \nomics measurements, such as metabolomics, have been used to interrogate mechanisms for \nmarginal (interaction-free) genetic effects on complex phenotypes (Auwerx et al., 2023; Yin et \nal., 2022) and can further act as mediators of the genetic and/or exposure effects for an \nestablished GxE.  \n \nTo explore the value of GxE refinement, we focused on the strongest GxE identified in a GWIS \nmeta-analysis of PA on HDL-C (Kilpeläinen et al., 2019). We aimed to refine our understanding \nof its specific context using a series of observational datasets, structured modeling with \nadditional covariates, and additional relevant data types (NMR-based lipoprotein measurements). \nIn the Women’s Genome Health Study (WGHS) dataset, for which the interaction was strongest \nin the original meta-analysis, we first validated the interaction and identified the most relevant \nlipoprotein-related quantities affected. In the UKB dataset, we next explored the specificity of \nthe interaction based on sex and physical activity subtypes. Finally, in the Multi-Ethnic Study of \nAtherosclerosis (MESA) dataset, we replicated the interaction while illustrating the potential for \nresidual heterogeneity in comparing lipoprotein quantities across cohorts and measurement \nplatforms. \n \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   4\nResults \n \nOutcome and population refinement in WGHS \n \nWGHS was chosen to begin the investigation since it has (1) the most significant study-specific \nG×PA interactions at rs295849 in the original meta-analysis, and (2) available NMR \nmetabolomics data. Data from 23,018 women were available with appropriate variables for \nanalysis (population summary in Supp. Table S1). PA was associated with HDL-C in main effect \n(genotype-free) models (4% higher in active versus inactive, p = 1.5×10-23). The previously-\nreported interaction between rs295849 and PA (Fig. 1a) was not materially changed by the more \nsubstantial covariate adjustment used here. Sensitivity models with genetic principal components \nand with additional genetic interaction terms for each covariate showed no meaningful change in \nthe primary interaction estimate (Fig. 1b). \n \nNMR-derived lipoprotein and other metabolites measures capture lipoprotein metabolism in \ngreater detail than the clinical HDL-C measure alone. Thus, as alternative outcomes in otherwise \nidentical statistical models, they can allow better resolution in understanding the relevant biology \nof the GxPA interaction. As shown in Fig. 1c and listed in Supp. Table S2, GxE significance was \nhigh for HDL-C but surpassed by medium HDL particle concentrations (M-HDL-P) and H3P (a \nsubset of M-HDL-P). \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   5\n \nFigure 1: Validation and expansion of the known interaction in WGHS. a) Genotype- and PA-\nstratified HDL-C means (in mg/DL) show the interaction. b) Stability of the interaction in the \ncontext of additional covariates, with interaction effect estimates shown across multiple \nadjustment strategies (primary covariates, including genetic principal components, and including \ngene-covariate interaction terms for all covariates). c) Z-statistic of the primary interaction test \nusing a series of alternative NMR-based outcomes (x-axis labels for M-HDL-P and its sub-\ncomponents are bolded). NMR-measured HDL-C is indicated by “NHDLC” and a full list of \nmetabolite abbreviations is available in Supp. Table S2. d) Bubble plot of study- and population-\nspecific interaction estimates (in units of log-transformed HDL-C [mg/dL] / activity status / \nallele) contributing to the meta-regression. Interaction effect estimates are plotted against cohort \nsex proportions (percentage of females), with bubble sizes corresponding to interaction estimate \nprecision (inverse of the effect estimate variance). Study labels are shown for studies with N > \n2,000. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   6\n \nThe fact that the WGHS cohort is entirely female, and that HDL-C levels exhibit dimorphism \naccording to sex, raises the possibility that the interaction may be more pronounced (i.e., a \ngreater difference in PA effect across genotypes) in females compared to males. We performed a \nmeta-regression on sex proportion to test this hypothesis, finding evidence that the interaction \neffect tended to be stronger in cohorts with a larger fraction of women (p = 0.018; I2 = 1.68%; \nFig. 1d). This association was not due solely to the WGHS; in a sensitivity analysis excluding all \nsingle-sex cohorts, which would have outsized leverage affecting the meta-regression fit, the \nrelationship became stronger (p = 0.0007; Supp. Fig. S1). Leave-one-out analysis confirmed that \nno single cohort was responsible for the effect, with the only substantial change being an \nincrease in the meta-regression estimate and significance after removal of the all-male METSIM \nstudy. \n \nExposure refinement and validation in UKB \n \nA summary of the UKB population (N = 281,380 after exclusions and based on relevant data \navailability) can be found in Supp. Table S3. We first explored PA main effects in models \nwithout genotype terms. We verified the expected positive association between PA and HDL-C \nin the European-ancestry subset of UKB (0.10 SD HDL-C / SD PA, p < 10-300), which was \npresent regardless of the questionnaire subsets used for PA estimation (IPAQ versus RPAQ; \nbetween-instrument Pearson correlation of 0.45; see Methods). Nested covariate adjustment sets \nshowed that covariate choice affected the magnitude of the estimates: IPAQ-based PA estimates \nincreased with more adjustment for SES and healthy lifestyle variables (possibly due to its \ninclusion of occupational PA, which often associates negatively with SES), whereas RPAQ-\nbased measures decreased with more adjustment (indicating removal of the confounding effects \nof high SES and healthy lifestyle). For both questionnaire-based estimates, we observed a non-\nmonotonic relationship in which HDL-C was generally positively associated with PA, but began \nto decrease at more extreme PA values (Supp. Fig. S2). We compared various PA \ntransformations to account for this nonlinearity, using PA-HDL-C main effect strength as a \nguide, and chose to winsorize at the 90th percentile for all PA variables moving forward (in both \nUKB and MESA). PA main effects were modestly different by sex (0.09 versus 0.13 SD HDL-C \n/ SD PA in women and men, respectively; p-interaction = 7.8×10-5). \n \nUKB provided an opportunity to explore the replication of the observed interaction across a \nseries of modeling choices: NMR metabolite outcomes (as tested in the WGHS), female-only \nversus the full population (given the meta-regression results), and a more fine-grained set of \npotential PA variables (based on IPAQ and RPAQ). Interaction effects from these models are \npresented in Fig. 2a. In general, IPAQ-based PA estimates produced stronger effect estimates \nthan those from RPAQ (mirroring results from PA-HDL-C main effect models; Supp. Fig. S2d), \nwith no major differences evident using IPAQ subsets corresponding to moderate versus \nvigorous activity. Focusing on the IPAQ estimates, interaction effects on HDL-C (as measured \nby a standard biochemical assay) were somewhat greater in the subset of the population with \navailable NMR data than the entire population. We did not identify any characteristics of the \nsub-population that clearly explained this heterogeneity (among those covariates listed in Supp. \nTable S3). As expected, effects estimated from NMR-measured HDL-C were comparable to \nthose from standard HDL-C in the same sub-population (given the Pearson correlation of 0.94 \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   7\nbetween the two HDL-C measures). Finally, using the M-HDL-P outcome, we found that \ninteraction effect estimates were modestly higher than for HDL-C and biased toward greater \nmagnitudes in women, validating the outcome and sub-population refinements suggested by the \nWGHS analysis. Results for all NMR quantities are shown in Supp. Table S4. Visualization of \nthe interaction using stratified genotypes and PA tertiles revealed a qualitative interaction \nmirroring that from WGHS, such that additional minor alleles amplified the positive PA-HDL \nrelationship without inducing a meaningful marginal genotype effect (Fig. 2c). We note that \ndifferences in HDL subfraction labeling across NMR platforms complicates conclusions about \nconcordance across cohorts (further discussion below). \n \n \nFigure 2: Refinement of the interaction in the UK Biobank. a) Interaction effect sizes \n(standardized to units of [SD outcome / SD PA / allele]) are plotted against various PA \nquestionnaires and subtypes. Panels correspond to various outcomes and UKB sub-populations \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   8\n(full set of unrelated participants versus the subset with available NMR metabolomics data). b) \nHistogram of IPAQ-based PA measures, with the spike at the right reflecting the 90th percentile \nwinsorization. Dotted lines indicate tertile boundaries. c) Visualization of the interaction using \nmean M-HDL-P estimates plotted against allele at rs295849, stratified by tertile of IPAQ-based \nPA. \n \nReplication in MESA \n \nWe used the MESA cohort (N = 4,587) to test for replication of the interaction, including the \nrefinements established in WGHS and UKB, noting that interaction effects in MESA population \nsubgroups were directionally consistent but nonsignificant in the original meta-analysis. A \nsummary of the MESA population can be found in Supp. Table S5. PA and HDL-C were \npositively associated (0.03 SD log(HDL-C) / SD PA, p = 0.01; Supp. Fig. S3), though less \nstrongly than in the other cohorts, with no evidence of heterogeneity by sex. Despite a \nsubstantially smaller sample size than UKB and WGHS, limiting the statistical power for \nreplication, MESA interaction models using NMR-measured outcomes showed nominally \nsignificant replication of the primary interaction influencing HDL-C (p = 0.01), but little \nindication of a female bias (Fig. 3a). Despite a consistent effect direction, the interaction did not \naffect M-HDL-P at a nominal significance level, rather showing an increasing interaction \nmagnitude moving from small to medium to large HDL-P subfractions. After testing more \ngranular HDL subfractions (seven categories based on average particle diameter), it appeared \nthat the overall signal was attributable primarily to the largest particle subsets (H5P – H7P). \nSensitivity analyses confirmed minimal heterogeneity in interaction effects when using PA \nsubtypes (Fig. 3b) and when stratifying by self-reported race/ethnicity (Fig. 3c). \n \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   9\nFigure 3: MESA replication results. a) Interaction effect sizes (standardized to units of [SD \noutcome / SD MVPA / allele]) are plotted against NMR-measured quantities including HDL-C, \nlabeled HDL subfractions (small, medium, and large), and more granular HDL subfractions (in \norder of increasing diameter). b) As in (a), but plotting effects on HDL-C across PA types. b) As \nin (a), but plotting effects on HDL-C across PA types. c) As in (a), but plotting interactions \ninvolving MVPA and HDL-C across the four race/ethnicity subgroups comprising the MESA \npopulation. \n \nAs described above, we observed heterogeneity across cohorts in terms of the specific HDL-P \nsubfractions showing the strongest signal. Given the difference in NMR measurement platform \nbetween UKB (Nightingale) and the others (Labcorp), we further explored the relationships \nbetween these subfractions and in the context of their specific particle size annotations. In \nMESA, larger HDL particles showed lower concentrations but higher correlations with HDL-C \ncompared to smaller HDL particles, as expected (Fig. 4a). Importantly, we note major \ndifferences in subfraction labeling across platforms, such that the M-HDL-P subfraction from \nNightingale platform corresponds more closely to the large, rather than medium, HDL-P \nsubfraction reported by Labcorp (Fig. 4b). Integrating these objective particle size-related \nfindings with our modeling results, we found that there was in fact closer agreement as to the \nmost relevant particle size between UKB and MESA (around 11nm diameter particles), \ncompared to WGHS (around 9nm). Integrating the results from all three cohorts, the interaction \nappeared to be more pronounced for a general group of medium-to-large HDL particles, with \ndiameters between roughly 9 and 12nm. \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   10\n \nFigure 4: Exploration of HDL subfraction labels and sizes across platforms. a) Empirical \nrelationships between Labcorp-labeled HDL subfractions in MESA, including fractions of \noverall HDL-P concentration (top panel) and Pearson correlations with NMR-measured HDL-C \n(bottom panel). b) Platform-specific particle diameters associated with HDL subfraction labels. \nSize-based annotations are available as ranges for Labcorp and as average diameters for \nNightingale. c) Interaction testing z-statistics are shown as a function of annotated HDL size \n(rather than label) for each primary subfraction (small, medium, and large; top panel) and \ngranular subfraction (H1P-H7P as reported by Labcorp only; bottom panel). Because WGHS and \nMESA were both measured using the Labcorp platform, labels are shown for MESA only. \nAverage diameters for labeled Labcorp subfractions were approximated as the midpoint of the \nprovided diameter range. \n \nDiscussion \n \nOur study was motivated by the recognition that even significant findings in large-scale, multi-\nstudy meta-analyses of GxE effects may provide only limited insight into mechanism. We chose \none specific gene-PA interaction to explore, finding that increased attention to each piece of the \ninteraction resulted in overall greater understanding and agreement among studies. Using WGHS \nand other cohorts from the original meta-analysis, we found that the interaction was stronger in \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   11\nwomen and impacted medium-sized HDL subfractions to a greater degree than HDL-C, \nconclusions that were supported by results in the much larger UKB study. MESA results \nreplicated the primary interaction (with greater significance than its results in the original meta-\nanalysis) and reinforced the tendency of the interaction to impact larger HDL particles, although \nwithout indication of a stronger effect in females as in the other cohorts. \n \nOverall, we saw consistency in the impact of this interaction on HDL subgroups, with medium-\nto-large subfractions most strongly impacted, but there was some heterogeneity in the specific \nparticle diameters affected. This is likely due to some combination of assay heterogeneity and \nbiological variability between cohorts. Though both WGHS and MESA NMR data came from \nthe Labcorp LP4 platform, effects were biased toward medium-sized particles (diameters \nbetween roughly 8-10 nm) in WGHS, versus larger particles (diameters greater than 10 nm) in \nMESA. UKB measurements came from the Nightingale platform, which uses a different set of \ndiameter cutoffs and includes an “extra-large” subfraction. UKB results showed greater impacts \non a diameter subset consistent with the L-HDL-P subfraction from MESA (average diameter of \n10.9 nm). Taken together, these results indicate effects most strongly concentrated on medium-\nto-large HDL particles (approximately 9-11 nm in diameter), but with meaningful heterogeneity \nin this estimate across cohorts. \n \nMedium and large HDL-P (as defined for the Labcorp panel, including relatively lower-diameter \nspecies) are more strongly associated with cardiovascular protection than smaller subfractions. \nM-HDL-P was the strongest predictor of reduced coronary events in statin-treated subjects from \nthe MRC/BHF Heart Protection Study (Parish et al., 2012), the best predictor of incident CVD in \nCKD patients (Shao et al., 2023), and predicted risk of CHD death in the MRFIT trial (Kuller et \nal., 2007). This is unlikely to be due to improved cholesterol efflux capacity (CEC): Shao and \ncolleagues note that S-HDL-P is more efficient in promoting cholesterol efflux, and that M-\nHDL-P is associated with protection from vascular complications of type 1 diabetes without a \ncorresponding difference in measured CEC (Shao et al., 2023). Other studies have shown inverse \nassociations of L-HDL-P with incident CVD (Mora et al., 2009) and both M-HDL-P and L-\nHDL-P with incident coronary heart disease (Akinkuolie et al., 2014). Given the above, the \nstronger impact of this study’s focal interaction on larger HDL subfractions increases its clinical \nrelevance beyond the previously recognized effect on HDL-C. Notably, women have higher \nHDL-C on average, as well as a larger mean particle size (Franczyk et al., 2023), linking to the \nobserved sex dependence in the interaction effect. The potential for a common causal factor \nunderlying these observations is an intriguing direction for further study. \n \nWe did not find evidence that the interaction explored here is specific to a certain intensity or \ndomain of PA. Interaction effects using moderate versus vigorous PA were similar, and in UKB, \nPA estimates based on the IPAQ questionnaire (which additionally includes occupational PA) \nproduced stronger interactions. PA has well-established positive associations with HDL \ncholesterol and function (Franczyk et al., 2023), and accumulating evidence points to a positive \nimpact on larger HDL subclasses (Franczyk et al., 2023; Sarzynski et al., 2015), with directional \nconsistency across exercise intensities (Slentz et al., 2007). These relationships are notably \ncomplex, however. While PA tends to increase L-HDL-P, it is consistently associated with \nmodestly lower M-HDL-P (Sarzynski et al., 2015). Furthermore, baseline HDL characteristics \nmay affect response to exercise; in individuals with prediabetes, M-HDL-P predicts the impact of \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   12\nexercise on HDL functionality, with implications for downstream cardiovascular risk (Sarzynski \net al., 2018). \n \nMeta-regression, motivated by the strength of the interaction in the all-female WGHS cohort, \nsuggested that this interaction is more prominent in women. This may link to the HDL \nsubfraction findings discussed above; women have higher HDL-C on average, as well as a larger \nmean particle size (Franczyk et al., 2023), which we confirmed in UKB. Though two-way \ninteractions are already complex, these results indicate the importance of exploring three-way \ninteractions for important potential effect modifiers such as sex. We note that the observation of \nfemale-specific effects allowed us to uncover interaction effects with larger magnitude in UKB \nthan if we had assumed homogeneity of these interaction effects. \n \nWhile interrogation of genetic mechanisms at the intergenic rs295849 was not the primary focus \nof this investigation, we present additional notes here. As originally reported by Kilpeläinen et \nal. (Kilpeläinen et al., 2019), the nearest gene is LHX1, encoding a transcription factor without \nclear relevance to lipoprotein metabolism. They suggest the nearby acetyl-CoA carboxylase \n(ACACA), encoding a key enzyme in fatty acid biosynthesis and metabolism pathways, as a \nbiologically plausible effector gene. In support of this hypothesis, we find that a chromatin \ninteraction was reported between one chromatin anchor harboring rs295849 and another at \nACACA (Jin et al., 2013; Pan et al., 2021), suggesting that the observed statistical interaction \nmay reflect co-regulation events involving ACACA. Despite rs295849 sitting outside a clear \ngenic or enhancer region, its RegulomeDB score of 0.6 indicates potential regulatory activity, \nbased partially on its binding by the transcription factor and spliceosome component RBM25 \n(Dong et al., 2023). RBM25 affects lipid metabolism; the overexpression of three rare RBM25 \nmutants in Huh-7 hepatocytes resulted in decreased LDL uptake (Zanoni et al., 2022). \nFurthermore, RBM25 gene expression tends to be higher in females, providing a possible link to \nthe sex-biased interaction effect observed here (Zhang 2021). Functional experiments exploring \ncontext-specific expression of potential effector genes (such as ACACA) and activity of RBM25 \nmay help to clarify the molecular mechanisms at play. \n \nThis study’s major strength lies in the accumulation of evidence and refinement across multiple \ncohorts and data types, helping to expand and reinforce conclusions from each other and the \noriginally reported interaction. Despite this fact and our attention to harmonization of PA and \noutcome variables, residual heterogeneity across cohorts remains a notable limitation. \nQuestionnaire items informing calculated PA differed between the three cohorts here, which may \nhave contributed to the differing shapes of the PA-HDL main effect relationship. As discussed, \nNMR platforms differed between UKB and the other studies, with imperfect ability to harmonize \nHDL subfractions across platforms using average particle diameters. Furthermore, in this study \nwe did not investigate downstream links between HDL particles and risk for cardiovascular and \nother diseases; ongoing research into the effectiveness of HDL subfractions as predictive \nbiomarkers may help to strengthen this link.  \n \nIn conclusion, this study makes two key contributions. First, we provide substantially greater \ninsight into a previously discovered gene-PA interaction, showing how rs295849 interacts with \nPA behaviors to impact lipoprotein profiles associated with cardiovascular risk. Second, we \ndemonstrate that careful attention to phenotype refinement and harmonization, covariate \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   13\nadjustment, and relevant population subgroups can account for some sources of heterogeneity \nacross studies and thus allow for improved GxE replication. This approach provides a template \nfor future studies to characterize and strengthen GxE findings, which will be a critical ingredient \nin building a robust evidence base of gene-lifestyle interactions to inform precision medicine. \n \nMethods \n \nWomen’s Genome Health Study \n \nThe WGHS is a prospective US-based cohort of healthy adult females with baseline age 45 years \nor older at enrollment in 1992-1994, and includes a majority subset with European ancestry \nverified by genetics that was used here and in the original analysis (Ridker et al., \n2008,Kilpeläinen et al., 2019). The WGHS component of this work was conducted under \napproval from the Mass General Brigham IRB (protocol 2006P001259). \n \nHDL-C was determined enzymatically from baseline plasma, and log transformed for analysis. \nChemically-measured (rather than NMR-measured) HDL-C was used for analysis unless \notherwise noted.  Physical activity (PA) was ascertained by self-report questionnaire and \ndichotomized (a threshold of 225 MET-mins/wk moderate-to-vigorous PA) (refer to Kilpeläinen \net al., 2019).  We note that the original analysis coded physical activity as active (0) compared to \ninactive (1). Here, the coding in the WGHS is the opposite (active=1, inactive=0) so that the \ninterpretation matches the rest of the current analysis. Genotype dosages for rs295849 were \nbased on microarray genotyping followed by imputation to the TOPMed reference panel (August \n26, 2019). Sensitivity models included one with five genetic principal components and another \nwith genotype-covariate product terms for each covariate. \n \nCovariates for the analysis included age at baseline, alcohol (four intake categories and other), \nsmoking (current, former, never, or other), educational attainment (categorical as L.P.N. or \nL.V.N., 2-year R.N., 3-year R.N., Bachelor’s, Master’s, Doctoral, or other), income (categorical \nin thresholds $10,000, $20,000, $30,000, $40,000, $50,000, $100,000, and other), and a diet \nquality score based on the Alternative Healthy Eating Index. We note that this covariate set \nincludes additional variables related to socioeconomic status (SES) and healthy lifestyle \ncompared to those from the original meta-analysis.  \n \n45 metabolites measured using nuclear magnetic resonance (NMR) were available from the \nLipoScience/Labcorp Vantera\n® Clinical Analyzer platform (subsequently referred to as \n“Labcorp” in this manuscript; using the LP4 algorithm) as previously described (Ahmad et al., \n2018), reporting the concentrations of lipoprotein subfraction particles according to class (LDL, \nHDL, and triglyceride-rich lipoprotein particles [TRLP]) and physical diameter (Huffman et al., \n2022). \n \nUK Biobank \n \nUKB is a large prospective cohort with both deep phenotyping and molecular data, including \ngenome-wide genotyping, on over 500,000 individuals of age 40-69 living throughout the UK \nbetween 2006-2010 (Sudlow et al., 2015). The UKB component of this work was conducted \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   14\nunder a Not Human Subjects Research determination for UKB data analysis (NHSR-4298 at the \nBroad Institute of MIT and Harvard) and UK Biobank application 27892. We used a sample set \nincluding individuals of European ancestry (determined by the Pan-UKBB project (Pan-UKB \nteam, 2020)) that had not withdrawn consent by the time of analysis. Additionally, we limited \nanalysis to a subset of unrelated individuals (by including only those that were used for genetic \nprincipal components analysis during central genetic data preprocessing) and removed \nparticipants who were pregnant or had diabetes, coronary heart disease, liver cirrhosis, or cancer.  \n \nThe outcome trait, HDL-C, was originally measured using enzyme immunoinhibition analysis in \nplasma samples and was log-transformed prior to analysis. Multiple PA exposures were derived \nfrom questionnaire responses. International physical activity questionnaire (IPAQ)-based PA \nestimates (MET-min/wk) were derived from touchscreen questions about the frequency and \nduration of moderate and vigorous intensity physical activity and walking (e.g., “Number of \ndays/week of vigorous physical activity 10+ minutes”). Recent physical activity questionnaire \n(RPAQ)-based PA estimates (MET-min/wk) were derived from touchscreen questions about the \nfrequency and duration of leisure-time physical activity in various specific categories (e.g., \n“Duration of strenuous sports”). Genotyping, imputation, and initial quality control for the UKB \ngenetic dataset have been described previously (Bycroft et al., 2018). Variant rs295849 was \ngenotyped directly and was retrieved from genetic data release version 3. \n \nCovariates included genetically-determined sex, age, age2, a sex-by-age product term, 5 genetic \nprincipal components (calculated centrally using the entire UKB population), smoking \n(categorical: never, past, or current), alcohol intake (categorical: weekly frequency estimates), \nquantitative diet variables from a food frequency questionnaire (cooked vegetables, raw \nvegetables, fresh fruit, oily fish, non-oily fish, processed meat), a categorical diet variable (bread \ntype), and a multiple deprivation index (one for each of England, Scotland, and Wales, with each \nvariable containing a quantitative value for participants living in that country and zero \notherwise). For variables coded as categorical, ambiguous categories such as “do not know” or \n“prefer not to answer” were left as non-missing to allow them to constitute an independent \ncategory for adjustment. All continuous variables, including blood biomarkers, PA variables, and \ncontinuous covariates, were winsorized at 5 standard deviations from the mean after all other \npreprocessing steps. \n \nNMR metabolites were available from the Nightingale platform for 66,870 participants (having \nother necessary variables and passing described exclusion criteria) and were preprocessed using \nthe ukbnmr R package (Ritchie et al., 2023), which includes imputation of zero values, log-\ntransformation, adjustment for key batch variables such as shipment plate and time between \nsample preparation and measurement, and transformation back into absolute concentrations. \nAfter preprocessing, 325 metabolites were available for analysis, covering a similar set of \nbiological quantities (primarily lipoprotein lipid and particle concentrations) and including \nadditional derived variables: 107 non-derived variables (e.g., M-HDL-P concentration), 61 \ncomposite variables (e.g., total lipids in medium HDL), 135 percentages (e.g., cholesterol as a \npercentage of total lipids in medium HDL), and 22 ratios (e.g., free cholesterol to cholesteryl \nesters in HDL). \n \nMulti-Ethnic Study of Atherosclerosis \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   15\n \nThe Multi-Ethnic Study of Atherosclerosis (MESA) is a study of subclinical cardiovascular \ndisease and the risk factors that predict progression to clinically overt cardiovascular disease or \nprogression of the subclinical disease (Bild, 2002). MESA consisted of a diverse, population-\nbased sample of an initial 6,814 asymptomatic men and women aged 45-84, with the first \nexamination (analyzed here) taking place between July 2000 and July 2002. 38 percent of the \nrecruited participants were white, 28 percent African American, 22 percent Hispanic, and 12 \npercent Asian, predominantly of Chinese descent. The MESA component of this work was \nconducted under approval from the Mass General Brigham IRB (protocol 2017P000531). \n \nAs in WGHS, chemically-measured (using the cholesterol oxidase method [Roche Diagnostics, \nIndianapolis, Indiana]; rather than NMR-measured) HDL-C was used for analysis unless \notherwise noted. The primary PA measure used in MESA was moderate and vigorous PA \n(MVPA), derived as previously described (Bertoni et al., 2009). Briefly, a series of questionnaire \nitems (28 questions including household chores, lawn/yard/garden/farm, care of children/adults, \ntransportation, walking (not at work), dancing and sport activities, conditioning activities, leisure \nactivities, and occupational and volunteer activities) were used to generate weekly activity levels \nin three categories (light, moderate, and vigorous), with MVPA calculated as the simple sum of \nall moderate and vigorous activity. Genotypes in MESA came from whole-genome sequencing \n(WGS) data generated through the NHLBI TOPMed program (Freeze 9b data release). Details \non WGS data generation and preprocessing are available at: \nhttps://topmed.nhlbi.nih.gov/topmed-whole-genome-sequencing-methods-freeze-9\n. Variant \nrs295849 was extracted for MESA participants from TOPMed-wide .gds files. \n \nNMR-based lipoprotein subclasses were measured in 2012 using the LipoScience/Labcorp \nVantera\n® Clinical Analyzer platform (LP4 algorithm, matching that of WGHS) as previously \ndescribed (Huffman et al., 2022). Missing covariate values were imputed to preserve sample \nsize, using the median value for continuous variables, a missing indicator for categorical income, \nand “never” for smoking. \n \nStatistical modeling \n \nFor all cohorts, the primary interaction model used the following form: \n \n/g1834/g1830/g1838 -/g1829  ~  /g1859/g3397/g1842 /g1827/g3397/g1859/g1499/g1842 /g1827/g3397/g1829  \n \nwhere g represents the genotype, PA represents physical activity, and C represents covariates.  \nHDL-C was log-transformed in WGHS and MESA and treated as a continuous variable (log-\ntransformation had minimal effect on model estimates). Covariates differed slightly between \ncohorts due to data availability, but aimed to adjust for age, age\n2, sex, healthy lifestyle, and \nsocioeconomic status. The directed acyclic graph guiding these choices can be found in Supp. \nFig. S4. PA main effect models omitted the genetic main effect and interaction product terms. \nWhen modeling NMR metabolites, the statistical model remained the same, with metabolite \nquantities replacing HDL-C as the outcome. Unless otherwise noted, data analysis was \nconducted using R version 4.2.2 (R Core Team, 2022).  \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   16\nMeta-regression of interaction summary statistics from Kilpelaïnen et al. was performed to test \nwhether there was a systematic difference in gene-PA interaction effect estimates according to \nsex. Meta-regression used the meta package for R (metareg function), based on a generic inverse \nvariance meta-analysis (metagen function) with variance estimation by restricted maximum \nlikelihood and Wald-type confidence intervals. Sex proportions (between 0 [all male] and 1 [all \nfemale]) for each cohort-ancestry combination were not reported by Kilpeläinen et al., so were \nretrieved primarily from a prior study using these same cohorts (De Vries et al., 2019) or using \nliterature-based reports for the few remaining cohorts otherwise.  The primary meta-regression \nmodel included an intercept to avoid the strong assumption that the interaction effect is precisely \nzero in all-male cohorts. Sensitivity analyses included (1) exclusion of single-sex cohorts that \nwill tend to exert high leverage on the regression fit, and (2) leave-one-out analyses that \nsystematically excluded each study-population combination. \n \nData and code availability \n \nCode supporting the analyses described here can be found at \nhttps://github.com/kwesterman/gxpa-nmr\n. Access to WGHS data is restricted by the institutional \nreview board, but analysis may be performed through collaboration; please contact Daniel \nChasman (dchasman@bwh.harvard.edu\n). The UK Biobank data can be obtained through \napplication at https://www.ukbiobank.ac.uk/. MESA data can be accessed through the TOPMed \nprogram via the NCBI Database of Genotypes and Phenotypes (dbGaP).  \n \nAcknowledgments \n \nThis investigation was supported by two grants from the U.S. National Heart, Lung, and Blood \nInstitute (NHLBI), the National Institutes of Health, R01HL118305 and R01HL156991. KEW \nwas supported by K01DK133637. TOK was supported by the Novo Nordisk Foundation \n(NNF18CC0034900, NNF21SA0072102). ARB was supported by the Intramural Research \nProgram of the National Human Genome Research Institute of the National Institutes of Health \nthrough the Center for Research on Genomics and Global Health (CRGGH). SM was supported \nby HL160799, HL117861, and K24 HL136852.  \n \nThe WGHS is supported by the National Heart, Lung, and Blood Institute (HL043851 and \nHL080467) and the National Cancer Institute (CA047988 and UM1CA182913), with funding for \ngenotyping provided by Amgen and funding for NMR assays by the American Heart \nAssociation. \n \nWhole genome sequencing (WGS) for the Trans-Omics in Precision Medicine (TOPMed) \nprogram was supported by the National Heart, Lung and Blood Institute (NHLBI). WGS for \n“NHLBI TOPMed: Multi-Ethnic Study of Atherosclerosis (MESA)” (phs001416.v3.p1) was \nperformed at the Broad Institute of MIT and Harvard (3U54HG003067-13S1). Centralized read \nmapping and genotype calling, along with variant quality metrics and filtering were provided by \nthe TOPMed Informatics Research Center (3R01HL-117626-02S1). Phenotype harmonization, \ndata management, sample-identity QC, and general study coordination, were provided by the \nTOPMed Data Coordinating Center (3R01HL-120393-02S1), and TOPMed MESA Multi-Omics \n(HHSN2682015000031/HSN26800004). The MESA projects are conducted and supported by \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   17\nthe National Heart, Lung, and Blood Institute (NHLBI) in collaboration with MESA \ninvestigators. Support for the Multi-Ethnic Study of Atherosclerosis (MESA) projects are \nconducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in \ncollaboration with MESA investigators. Support for MESA is provided by contracts \n75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, \n75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, \nN01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-\nHC-95166, N01-HC-95167, N01-HC-95168, N01-HC-95169, UL1-TR-000040, UL1-TR-\n001079, UL1-TR-001420, UL1TR001881, DK063491, and R01HL105756. The authors thank \nthe other investigators, the staff, and the participants of the MESA study for their valuable \ncontributions.  A full list of participating MESA investigators and institutes can be found \nat http://www.mesa-nhlbi.org\n. \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   18\nReferences \n \nAhmad S, Mora S, Franks PW, Orho-Melander M, Ridker PM, Hu FB, Chasman DI. 2018. \nAdiposity and Genetic Factors in Relation to Triglycerides and Triglyceride-Rich \nLipoproteins in the Women’s Genome Health Study. Clin Chem 64:231–241. \ndoi:10.1373/clinchem.2017.280545 \nAkinkuolie AO, Paynter NP, Padmanabhan L, Mora S. 2014. High-density lipoprotein particle \nsubclass heterogeneity and incident coronary heart disease. Circ Cardiovasc Qual \nOutcomes 7. doi:10.1161/CIRCOUTCOMES.113.000675 \nAuwerx C, Sadler MC, Woh T, Reymond A, Kutalik Z, Porcu E. 2023. Exploiting the mediating \nrole of the metabolome to unravel transcript-to-phenotype associations. Elife 12. \ndoi:10.7554/elife.81097 \nBentley AR, Sung YJ, Brown MR, Winkler TW, Kraja AT, Ntalla I, Schwander K, Chasman DI, \nLim E, Deng X, Guo X, Liu Jingmin, Lu Y, Cheng C-Y, Sim X, Vojinovic D, Huffman JE, \nMusani SK, Li C, Feitosa MF, Richard MA, Noordam R, Baker J, Chen G, Aschard H, \nBartz TM, Ding J, Dorajoo R, Manning AK, Rankinen T, Smith A V., Tajuddin SM, Zhao \nW, Graff M, Alver M, Boissel M, Chai JF, Chen X, Divers J, Evangelou E, Gao C, Goel A, \nHagemeijer Y, Harris SE, Hartwig FP, He M, Horimoto ARVR, Hsu F-C, Hung Y-J, \nJackson AU, Kasturiratne A, Komulainen P, Kühnel B, Leander K, Lin K-H, Luan J, \nLyytikäinen L-P, Matoba N, Nolte IM, Pietzner M, Prins B, Riaz M, Robino A, Said MA, \nSchupf N, Scott RA, Sofer T, Stancáková A, Takeuchi F, Tayo BO, van der Most PJ, Varga \nT V., Wang T-D, Wang Y, Ware EB, Wen W, Xiang Y-B, Yanek LR, Zhang W, Zhao JH, \nAdeyemo A, Afaq S, Amin N, Amini M, Arking DE, Arzumanyan Z, Aung T, Ballantyne \nC, Barr RG, Bielak LF, Boerwinkle E, Bottinger EP, Broeckel U, Brown M, Cade BE, \nCampbell A, Canouil M, Charumathi S, Chen Y-DI, Christensen K, Concas MP, Connell \nJM, de las Fuentes L, de Silva HJ, de Vries PS, Doumatey A, Duan Q, Eaton CB, Eppinga \nRN, Faul JD, Floyd JS, Forouhi NG, Forrester T, Friedlander Y, Gandin I, Gao H, Ghanbari \nM, Gharib SA, Gigante B, Giulianini F, Grabe HJ, Gu CC, Harris TB, Heikkinen S, Heng \nC-K, Hirata M, Hixson JE, Ikram MA, Jia Y, Joehanes R, Johnson C, Jonas JB, Justice AE, \nKatsuya T, Khor CC, Kilpeläinen TO, Koh W-P, Kolcic I, Kooperberg C, Krieger JE, \nKritchevsky SB, Kubo M, Kuusisto J, Lakka TA, Langefeld CD, Langenberg C, Launer LJ, \nLehne B, Lewis CE, Li Y, Liang J, Lin S, Liu C-T, Liu Jianjun, Liu K, Loh M, Lohman \nKK, Louie T, Luzzi A, Mägi R, Mahajan A, Manichaikul AW, McKenzie CA, Meitinger T, \nMetspalu A, Milaneschi Y, Milani L, Mohlke KL, Momozawa Y, Morris AP, Murray AD, \nNalls MA, Nauck M, Nelson CP, North KE, O’Connell JR, Palmer ND, Papanicolau GJ, \nPedersen NL, Peters A, Peyser PA, Polasek O, Poulter N, Raitakari OT, Reiner AP, \nRenström F, Rice TK, Rich SS, Robinson JG, Rose LM, Rosendaal FR, Rudan I, Schmidt \nCO, Schreiner PJ, Scott WR, Sever P, Shi Y, Sidney S, Sims M, Smith JA, Snieder H, Starr \nJM, Strauch K, Stringham HM, Tan NYQ, Tang H, Taylor KD, Teo YY, Tham YC, \nTiemeier H, Turner ST, Uitterlinden AG, van Heemst D, Waldenberger M, Wang H, Wang \nLan, Wang Lihua, Wei W Bin, Williams CA, Wilson G, Wojczynski MK, Yao J, Young K, \nYu C, Yuan J-M, Zhou J, Zonderman AB, Becker DM, Boehnke M, Bowden DW, \nChambers JC, Cooper RS, de Faire U, Deary IJ, Elliott P, Esko T, Farrall M, Franks PW, \nFreedman BI, Froguel P, Gasparini P, Gieger C, Horta BL, Juang J-MJ, Kamatani Y, \nKammerer CM, Kato N, Kooner JS, Laakso M, Laurie CC, Lee I-T, Lehtimäki T, \nMagnusson PKE, Oldehinkel AJ, Penninx BWJH, Pereira AC, Rauramaa R, Redline S, \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   19\nSamani NJ, Scott J, Shu X-O, van der Harst P, Wagenknecht LE, Wang J-S, Wang YX, \nWareham NJ, Watkins H, Weir DR, Wickremasinghe AR, Wu T, Zeggini E, Zheng W, \nBouchard C, Evans MK, Gudnason V, Kardia SLR, Liu Y, Psaty BM, Ridker PM, van Dam \nRM, Mook-Kanamori DO, Fornage M, Province MA, Kelly TN, Fox ER, Hayward C, van \nDuijn CM, Tai ES, Wong TY, Loos RJF, Franceschini N, Rotter JI, Zhu X, Bierut LJ, \nGauderman WJ, Rice K, Munroe PB, Morrison AC, Rao DC, Rotimi CN, Cupples LA. \n2019. Multi-ancestry genome-wide gene–smoking interaction study of 387,272 individuals \nidentifies new loci associated with serum lipids. Nat Genet 51:636–648. \ndoi:10.1038/s41588-019-0378-y \nBertoni AG, Whitt-Glover MC, Chung H, Le KY, Barr RG, Mahesh M, Jenny NS, Burke GL, \nJacobs DR. 2009. The association between physical activity and subclinical atherosclerosis: \nThe multi-ethnic study of atherosclerosis. Am J Epidemiol 169. doi:10.1093/aje/kwn350 \nBild DE. 2002. Multi-Ethnic Study of Atherosclerosis: Objectives and Design. Am J Epidemiol \n156:871–881. doi:10.1093/aje/kwf113 \nBycroft C, Freeman C, Petkova D, Band G, Elliott LT, Sharp K, Motyer A, Vukcevic D, \nDelaneau O, O’Connell J, Cortes A, Welsh S, Young A, Effingham M, McVean G, Leslie \nS, Allen N, Donnelly P, Marchini J. 2018. The UK Biobank resource with deep \nphenotyping and genomic data. Nature 562:203–209. doi:10.1038/s41586-018-0579-z \nDe Vries PS, Brown MR, Bentley AR, Sung YJ, Winkler TW, Ntalla I, Schwander K, Kraja AT, \nGuo X, Franceschini N, Cheng CY, Sim X, Vojinovic D, Huffman JE, Musani SK, Li C, \nFeitosa MF, Richard MA, Noordam R, Aschard H, Bartz TM, Bielak LF, Deng X, Dorajoo \nR, Lohman KK, Manning AK, Rankinen T, Smith A V., Tajuddin SM, Evangelou E, Graff \nM, Alver M, Boissel M, Chai JF, Chen X, Divers J, Gandin I, Gao C, Goel A, Hagemeijer \nY, Harris SE, Hartwig FP, He M, Horimoto ARVR, Hsu FC, Jackson AU, Kasturiratne A, \nKomulainen P, Kühnel B, Laguzzi F, Lee JH, Luan J, Lyytikäinen LP, Matoba N, Nolte IM, \nPietzner M, Riaz M, Said MA, Scott RA, Sofer T, Stan\nč áková A, Takeuchi F, Tayo BO, \nVan Der Most PJ, Varga T V., Wang Yajuan, Ware EB, Wen W, Yanek LR, Zhang W, \nZhao JH, Afaq S, Amin N, Amini M, Arking DE, Aung T, Ballantyne C, Boerwinkle E, \nBroeckel U, Campbell A, Canouil M, Charumathi S, Chen YDI, Connell JM, De Faire U, \nDe Las Fuentes L, De Mutsert R, De Silva HJ, Ding J, Dominiczak AF, Duan Q, Eaton CB, \nEppinga RN, Faul JD, Fisher V, Forrester T, Franco OH, Friedlander Y, Ghanbari M, \nGiulianini F, Grabe HJ, Grove ML, Gu CC, Harris TB, Heikkinen S, Heng CK, Hirata M, \nHixson JE, Howard B V., Ikram MA, Jacobs DR, Johnson C, Jonas JB, Kammerer CM, \nKatsuya T, Khor CC, Kilpeläinen TO, Koh WP, Koistinen HA, Kolcic I, Kooperberg C, \nKrieger JE, Kritchevsky SB, Kubo M, Kuusisto J, Lakka TA, Langefeld CD, Langenberg C, \nLauner LJ, Lehne B, Lemaitre RN, Li Y, Liang J, Liu Jianjun, Liu K, Loh M, Louie T, \nMägi R, Manichaikul AW, McKenzie CA, Meitinger T, Metspalu A, Milaneschi Y, Milani \nL, Mohlke KL, Mosley TH, Mukamal KJ, Nalls MA, Nauck M, Nelson CP, Sotoodehnia N, \nO’Connell JR, Palmer ND, Pazoki R, Pedersen NL, Peters A, Peyser PA, Polasek O, Poulter \nN, Raffel LJ, Raitakari OT, Reiner AP, Rice TK, Rich SS, Robino A, Robinson JG, Rose \nLM, Rudan I, Schmidt CO, Schreiner PJ, Scott WR, Sever P, Shi Y, Sidney S, Sims M, \nSmith BH, Smith JA, Snieder H, Starr JM, Strauch K, Tan N, Taylor KD, Teo YY, Tham \nYC, Uitterlinden AG, Van Heemst D, Vuckovic D, Waldenberger M, Wang L, Wang Yujie, \nWang Z, Wei W Bin, Williams C, Wilson G, Wojczynski MK, Yao J, Yu B, Yu C, Yuan \nJM, Zhao W, Zonderman AB, Becker DM, Boehnke M, Bowden DW, Chambers JC, Deary \nIJ, Esko T, Farrall M, Franks PW, Freedman BI, Froguel P, Gasparini P, Gieger C, Horta \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   20\nBL, Kamatani Y, Kato N, Kooner JS, Laakso M, Leander K, Lehtimäki T, Magnusson PKE, \nPenninx B, Pereira AC, Rauramaa R, Samani NJ, Scott J, Shu XO, Van Der Harst P, \nWagenknecht LE, Wang YX, Wareham NJ, Watkins H, Weir DR, Wickremasinghe AR, \nZheng W, Elliott P, North KE, Bouchard C, Evans MK, Gudnason V, Liu CT, Liu Y, Psaty \nBM, Ridker PM, Van Dam RM, Kardia SLR, Zhu X, Rotimi CN, Mook-Kanamori DO, \nFornage M, Kelly TN, Fox ER, Hayward C, Van Duijn CM, Tai ES, Wong TY, Liu \nJingmin, Rotter JI, Gauderman WJ, Province MA, Munroe PB, Rice K, Chasman DI, \nCupples LA, Rao DC, Morrison AC. 2019. Multiancestry Genome-Wide Association Study \nof Lipid Levels Incorporating Gene-Alcohol Interactions. Am J Epidemiol 188. \ndoi:10.1093/aje/kwz005 \nDong S, Zhao N, Spragins E, Kagda MS, Li M, Assis P, Jolanki O, Luo Y, Cherry JM, Boyle \nAP, Hitz BC. 2023. Annotating and prioritizing human non-coding variants with \nRegulomeDB v.2. Nat Genet. doi:10.1038/s41588-023-01365-3 \nFrancis M, Li C, Sun Y, Zhou J, Li X, Brenna JT, Ye K. 2021. Genome-wide association study \nof fish oil supplementation on lipid traits in 81,246 individuals reveals new gene-diet \ninteraction loci. PLoS Genet 17:e1009431. doi:10.1371/journal.pgen.1009431 \nFranczyk B, Gluba-Brzózka A, Ciałkowska-Rysz A, Ławi\nń ski J, Rysz J. 2023. The Impact of \nAerobic Exercise on HDL Quantity and Quality: A Narrative Review. Int J Mol Sci. \ndoi:10.3390/ijms24054653 \nGauderman WJ, Mukherjee B, Aschard H, Hsu L, Lewinger JP, Patel CJ, Witte JS, Amos C, Tai \nCG, Conti D, Torgerson DG, Lee S, Chatterjee N. 2017. Update on the State of the Science \nfor Analytical Methods for Gene-Environment Interactions. Am J Epidemiol 186:762–770. \ndoi:10.1093/aje/kwx228 \nHuffman KM, Parker DC, Bhapkar M, Racette SB, Martin CK, Redman LM, Das SK, Connelly \nMA, Pieper CF, Orenduff M, Ross LM, Ramaker ME, Dorling JL, Rosen CJ, Shalaurova I, \nOtvos JD, Kraus VB, Kraus WE. 2022. Calorie restriction improves lipid-related emerging \ncardiometabolic risk factors in healthy adults without obesity: Distinct influences of BMI \nand sex from CALERIE\nTM a multicentre, phase 2, randomised controlled trial. \nEClinicalMedicine 43. doi:10.1016/j.eclinm.2021.101261 \nJin F, Li Y, Dixon JR, Selvaraj S, Ye Z, Lee AY, Yen CA, Schmitt AD, Espinoza CA, Ren B. \n2013. A high-resolution map of the three-dimensional chromatin interactome in human \ncells. Nature 503. doi:10.1038/nature12644 \nKilpeläinen TO, Bentley AR, Noordam R, Sung YJ, Schwander K, Winkler TW, Jakupović  H, \nChasman DI, Manning A, Ntalla I, Aschard H, Brown MR, de las Fuentes L, Franceschini \nN, Guo X, Vojinovic D, Aslibekyan S, Feitosa MF, Kho M, Musani SK, Richard M, Wang \nH, Wang Z, Bartz TM, Bielak LF, Campbell A, Dorajoo R, Fisher V, Hartwig FP, Horimoto \nARVR, Li C, Lohman KK, Marten J, Sim X, Smith A V., Tajuddin SM, Alver M, Amini M, \nBoissel M, Chai JF, Chen X, Divers J, Evangelou E, Gao C, Graff M, Harris SE, He M, Hsu \nFC, Jackson AU, Zhao JH, Kraja AT, Kühnel B, Laguzzi F, Lyytikäinen LP, Nolte IM, \nRauramaa R, Riaz M, Robino A, Rueedi R, Stringham HM, Takeuchi F, van der Most PJ, \nVarga T V., Verweij N, Ware EB, Wen W, Li X, Yanek LR, Amin N, Arnett DK, \nBoerwinkle E, Brumat M, Cade B, Canouil M, Chen YDI, Concas MP, Connell J, de \nMutsert R, de Silva HJ, de Vries PS, Demirkan A, Ding J, Eaton CB, Faul JD, Friedlander \nY, Gabriel KP, Ghanbari M, Giulianini F, Gu CC, Gu D, Harris TB, He J, Heikkinen S, \nHeng CK, Hunt SC, Ikram MA, Jonas JB, Koh WP, Komulainen P, Krieger JE, Kritchevsky \nSB, Kutalik Z, Kuusisto J, Langefeld CD, Langenberg C, Launer LJ, Leander K, Lemaitre \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   21\nRN, Lewis CE, Liang J, Alizadeh BZ, Boezen HM, Franke L, Navis G, Rots M, Swertz M, \nWolffenbuttel BHR, Wijmenga C, Liu J, Mägi R, Manichaikul A, Meitinger T, Metspalu A, \nMilaneschi Y, Mohlke KL, Mosley TH, Murray AD, Nalls MA, Nang EEK, Nelson CP, \nNona S, Norris JM, Nwuba CV, O’Connell J, Palmer ND, Papanicolau GJ, Pazoki R, \nPedersen NL, Peters A, Peyser PA, Polasek O, Porteous DJ, Poveda A, Raitakari OT, Rich \nSS, Risch N, Robinson JG, Rose LM, Rudan I, Schreiner PJ, Scott RA, Sidney SS, Sims M, \nSmith JA, Snieder H, Sofer T, Starr JM, Sternfeld B, Strauch K, Tang H, Taylor KD, Tsai \nMY, Tuomilehto J, Uitterlinden AG, van der Ende MY, van Heemst D, Voortman T, \nWaldenberger M, Wennberg P, Wilson G, Xiang YB, Yao J, Yu C, Yuan JM, Zhao W, \nZonderman AB, Becker DM, Boehnke M, Bowden DW, de Faire U, Deary IJ, Elliott P, \nEsko T, Freedman BI, Froguel P, Gasparini P, Gieger C, Kato N, Laakso M, Lakka TA, \nLehtimäki T, Magnusson PKE, Oldehinkel AJ, Penninx BWJH, Samani NJ, Shu XO, van \nder Harst P, Van Vliet-Ostaptchouk J V., Vollenweider P, Wagenknecht LE, Wang YX, \nWareham NJ, Weir DR, Wu T, Zheng W, Zhu X, Evans MK, Franks PW, Gudnason V, \nHayward C, Horta BL, Kelly TN, Liu Y, North KE, Pereira AC, Ridker PM, Tai ES, van \nDam RM, Fox ER, Kardia SLR, Liu CT, Mook-Kanamori DO, Province MA, Redline S, \nvan Duijn CM, Rotter JI, Kooperberg CB, Gauderman WJ, Psaty BM, Rice K, Munroe PB, \nFornage M, Cupples LA, Rotimi CN, Morrison AC, Rao DC, Loos RJF. 2019. Multi-\nancestry study of blood lipid levels identifies four loci interacting with physical activity. Nat \nCommun 10. doi:10.1038/s41467-018-08008-w \nKuller LH, Grandits G, Cohen JD, Neaton JD, Prineas R. 2007. Lipoprotein particles, insulin, \nadiponectin, C-reactive protein and risk of coronary heart disease among men with \nmetabolic syndrome. Atherosclerosis 195. doi:10.1016/j.atherosclerosis.2006.09.001 \nMora S, Otvos JD, Rifai N, Rosenson RS, Buring JE, Ridker PM. 2009. Lipoprotein particle \nprofiles by nuclear magnetic resonance compared with standard lipids and apolipoproteins \nin predicting incident cardiovascular disease in women. Circulation 119. \ndoi:10.1161/CIRCULATIONAHA.108.816181 \nPan Q, Liu YJ, Bai XF, Han X Le, Jiang Y, Ai B, Shi SS, Wang F, Xu MC, Wang YZ, Zhao J, \nChen JX, Zhang J, Li XC, Zhu J, Zhang GR, Wang QY, Li CQ. 2021. VARAdb: A \ncomprehensive variation annotation database for human. Nucleic Acids Res 49. \ndoi:10.1093/nar/gkaa922 \nPan-UKB team. 2020. Pan-UK Biobank. \nParish S, Offer A, Clarke R, Hopewell JC, Hill MR, Otvos JD, Armitage J, Collins R. 2012. \nLipids and lipoproteins and risk of different vascular events in the MRC/BHF heart \nprotection study. Circulation 125. doi:10.1161/CIRCULATIONAHA.111.073684 \nR Core Team. 2022. R: A Language and Environment for Statistical Computing. \nRidker PM, Chasman DI, Zee RYL, Parker A, Rose L, Cook NR, Buring JE. 2008. Rationale, \nDesign, and Methodology of the Women’s Genome Health Study: A Genome-Wide \nAssociation Study of More Than 25 000 Initially Healthy American Women. Clin Chem \n54:249–255. doi:10.1373/clinchem.2007.099366 \nRitchie SC, Surendran P, Karthikeyan S, Lambert SA, Bolton T, Pennells L, Danesh J, Di \nAngelantonio E, Butterworth AS, Inouye M. 2023. Quality control and removal of technical \nvariation of NMR metabolic biomarker data in ~120,000 UK Biobank participants. Sci Data \n10. doi:10.1038/s41597-023-01949-y \nSarzynski MA, Burton J, Rankinen T, Blair SN, Church TS, Després JP, Hagberg JM, Landers-\nRamos R, Leon AS, Mikus CR, Rao DC, Seip RL, Skinner JS, Slentz CA, Thompson PD, \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   22\nWilund KR, Kraus WE, Bouchard C. 2015. The effects of exercise on the lipoprotein \nsubclass profile: A meta-analysis of 10 interventions. Atherosclerosis 243. \ndoi:10.1016/j.atherosclerosis.2015.10.018 \nSarzynski MA, Ruiz-Ramie JJ, Barber JL, Slentz CA, Apolzan JW, McGarrah RW, Harris MN, \nChurch TS, Borja MS, He Y, Oda MN, Martin CK, Kraus WE, Rohatgi A. 2018. Effects of \nIncreasing Exercise Intensity and Dose on Multiple Measures of HDL (High-Density \nLipoprotein) Function. Arterioscler Thromb Vasc Biol 38. \ndoi:10.1161/ATVBAHA.117.310307 \nShao B, Afshinnia F, Mathew A V., Ronsein GE, Thornock C, Irwin AD, Kansal M, Rao PS, \nDobre M, Al-Kindi S, Weir MR, Go A, He J, Chen J, Feldman H, Bornfeldt KE, Pennathur \nS. 2023. Low concentrations of medium-sized HDL particles predict incident CVD in \nchronic kidney disease patients. J Lipid Res 64. doi:10.1016/j.jlr.2023.100381 \nSlentz CA, Houmard JA, Johnson JL, Bateman LA, Tanner CJ, McCartney JS, Duscha BD, \nKraus WE. 2007. Inactivity, exercise training and detraining, and plasma lipoproteins. \nSTRRIDE: A randomized, controlled study of exercise intensity and amount. J Appl Physiol \n103. doi:10.1152/japplphysiol.01314.2006 \nSudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, Downey P, Elliott P, Green J, \nLandray M, Liu B, Matthews P, Ong G, Pell J, Silman A, Young A, Sprosen T, Peakman T, \nCollins R. 2015. UK Biobank: An Open Access Resource for Identifying the Causes of a \nWide Range of Complex Diseases of Middle and Old Age. PLoS Med 12:e1001779. \ndoi:10.1371/journal.pmed.1001779 \nWesterman KE, Walker ME, Gaynor SM, Wessel J, Dicorpo D, Ma J, Alonso A, Aslibekyan S, \nBaldridge AS, Bertoni AG, Biggs ML, Brody JA, Chen YDI, Dupuis J, Goodarzi MO, Guo \nX, Hasbani NR, Heath A, Hidalgo B, Irvin MR, Johnson WC, Kalyani RR, Lange L, \nLemaitre RN, Liu CT, Liu S, Moon JY, Nassir R, Pankow JS, Pettinger M, Raffield LM, \nRasmussen-Torvik LJ, Selvin E, Senn MK, Shadyab AH, Smith A V., Smith NL, Steffen L, \nTalegakwar S, Taylor KD, de Vries PS, Wilson JG, Wood AC, Yanek LR, Yao J, Zheng Y, \nBoerwinkle E, Morrison AC, Fornage M, Russell TP, Psaty BM, Levy D, Heard-Costa NL, \nRamachandran VS, Mathias RA, Arnett DK, Kaplan R, North KE, Correa A, Carson A, \nRotter JI, Rich SS, Manson JE, Reiner AP, Kooperberg C, Florez JC, Meigs JB, Merino J, \nTobias DK, Chen H, Manning AK. 2023. Investigating Gene–Diet Interactions Impacting \nthe Association Between Macronutrient Intake and Glycemic Traits. Diabetes 72. \ndoi:10.2337/db22-0851 \nYin X, Bose D, Kwon A, Hanks SC, Jackson AU, Stringham HM, Welch R, Oravilahti A, \nFernandes Silva L, Locke AE, Fuchsberger C, Service SK, Erdos MR, Bonnycastle LL, \nKuusisto J, Stitziel NO, Hall IM, Morrison J, Ripatti S, Palotie A, Freimer NB, Collins FS, \nMohlke KL, Scott LJ, Fauman EB, Burant C, Boehnke M, Laakso M, Wen X. 2022. \nIntegrating transcriptomics, metabolomics, and GWAS helps reveal molecular mechanisms \nfor metabolite levels and disease risk. Am J Hum Genet 109. doi:10.1016/j.ajhg.2022.08.007 \nZanoni P, Panteloglou G, Othman A, Haas JT, Meier R, Rimbert A, Futema M, Khalil YA, \nNorrelykke SF, Rzepiela AJ, Stoma S, Stebler M, Van Dijk F, Wijers M, Wolters JC, Dalila \nN, Huijkman NCA, Smit M, Gallo A, Carreau V, Philippi A, Rabès JP, Boileau C, Visentin \nM, Vonghia L, Weyler J, Francque S, Verrijken A, Verhaegen A, Van Gaal L, Van Der \nGraaf A, Van Rosmalen B V., Robert J, Velagapudi S, Yalcinkaya M, Keel M, \nRadosavljevic S, Geier A, Tybjaerg-Hansen A, Varret M, Rohrer L, Humphries SE, Staels \nB, Van De Sluis B, Kuivenhoven JA, Von Eckardstein A. 2022. Posttranscriptional \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   23\nRegulation of the Human LDL Receptor by the U2-Spliceosome. Circ Res 130. \ndoi:10.1161/CIRCRESAHA.120.318141 \n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   24\nSupplementary Figures \n \n \nSupplementary Figure S1: Bubble plot describing the meta-regression of interaction summary \nstatistics from Kilpeläinen et al. a) Interaction effect sizes from each cohort are plotted against \nthe corresponding sex proportions for the primary meta-regression. Bubble size is proportional to \nthe inverse variance of the interaction effect estimate. Study labels are shown for studies with N \n> 1,000. b) As in (a), but excluding all single-sex studies. c) Leave-one-out sensitivity analysis \nplots show meta-regression estimates and confidence intervals after excluding each study-\npopulation combination. \n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   25\n \nSupplementary Figure S2: PA-HDL-C main effects in UKB. a) Histogram of PA measured \nusing questions based on the IPAQ (left) or RPAQ (right) questionnaires. b) Shrunken cubic \nspline fits of the relationship between PA and unadjusted mean HDL-C, with shadows \ncorresponding to 95% CIs. c) As in (b), but using PA distributions winsorized at the 90th \npercentile. d) Adjusted main effect z-statistics according to PA transformation. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n  2 6\nSupplementary Figure S3: PA-HDL-C main effects in MESA. a) Histogram of questionnaire-\nbased moderate-to-vigorous PA estimates. b) Shrunken cubic spline fit of the relationship \nbetween PA and unadjusted mean HDL-C, with shadows corresponding to 95% CIs. \n \n \n \nSupplementary Figure S4: Directed acyclic graph informing covariate selection based on PA-\nHDL-C main effect. \n \n \n 6\n \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   27\n  \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint \n\n   28\nSupplementary Tables \n \nSee accompanying Excel document. \n . CC-BY 4.0 International licenseIt is made available under a \n is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)\nThe copyright holder for this preprint this version posted January 24, 2024. ; https://doi.org/10.1101/2024.01.23.24301689doi: medRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}