Written by: Isabela LandsteinerLindsey K StolzeTess E PetersonAndrew PerryPhillip LinQuanhu ShengShilin ZhaoKaushik AmancherlaMichael Y MiTakenori IkomaLaura P CohenIlya GivertsJoseph CampainJoao A C LimaWendy S PostKent D TaylorRobert GersztenStephen S RichRussell TracyPeter DurdaAlexis C WoodRajeev MalhotraJerome I RotterClary ClishMatthew NayorEric R GamazonRavi V ShahGregory D Lewis
The flow of our study design is shown in Figure 1. Data, analytical methods, and study materials will be made available on reasonable request by the corresponding authors. Of note, our methods description below is adapted from previous work by our group17–20 with minimal alteration to promote scientific rigor and reproducibility. Scientific attribution is provided by this statement.Graphical abstract and study scheme. Overview of the study design linking invasive cardiopulmonary exercise testing (iCPET), metabolomics, and genomics. In the MGH-ExS (Massachusetts General Hospital Exercise Study) iCPET cohort of patients with heart failure (HF) with preserved ejection fraction (HFpEF; n=820), we quantified 7 organ-specific exercise physiological deficits and peak Vo2, derived metabolite signatures for each deficit, and related these signatures to clinical outcomes. The relation of these signatures with clinical-demographic features, cardiac magnetic resonance imaging, and incident HF was measured in MESA (Multi-Ethnic Study of Atherosclerosis; n=6345). We also mapped deficit-implicated metabolites to tissue-specific genetic variation in ≈2 million individuals with HF and in the largest genome-wide association studies of HFpEF comorbidities to evaluate shared metabolic mechanisms of HFpEF pathophysiology.The MGH-ExS (Massachusetts General Hospital Exercise Study) cohort included consecutive patients who underwent clinically indicated iCPET for the evaluation of exertional dyspnea between 2006 and 2023. In this study, we selected patients with symptomatic dyspnea and New York Heart Association class II to IV symptoms with confirmed diagnosis of HFpEF based on criteria delineated in the universal definition of HF,8 requiring symptoms or signs of HF, left ventricular (LV) ejection fraction (LVEF) ≥50%, and at least 1 additional criterion among structural cardiac abnormalities, elevated natriuretic peptides, hemodynamic evidence of elevated filling pressures, or a prior HF event (Table S1).Clinical characteristics and risk factor assessment were obtained on the same day the iCPET was performed. Medical records were reviewed, with the following clinical outcomes adjudicated by a panel of cardiologists from the time of cardiopulmonary exercise testing (CPET) through November 2023. Our primary outcome assessed was the composite of first cardiovascular event (defined as any hospitalization for a principal cardiovascular cause, including arrhythmia, acute coronary syndrome, coronary revascularization, stroke, and HF) and all-cause mortality. Our secondary outcome was a composite of first HF hospitalization or all-cause mortality. Events were ascertained from hospital records in conjunction with the Social Security Death Master File. The study was approved by the Mass General Brigham Institutional Review Board. All participants provided written informed consent.The MESA is a prospective cohort study initiated in 2000 to study the characteristics and progression of subclinical cardiovascular disease among a diverse community-based population enrolled across 6 geographic areas in the United States.21 At the time of enrollment, participants were 45 to 84 years of age and had no history of clinical cardiovascular disease, including coronary artery disease, peripheral vascular disease, cerebrovascular disease, or HF. The study protocol was approved by the Institutional Review Boards of participating institutions, and all participants signed informed consent.Information on hospital admissions, outpatient diagnoses, and deaths was ascertained at follow-up telephone interviews, and medical records and International Classification of Diseases diagnosis codes were obtained for events. HF criteria included symptomatic, physician-diagnosed HF and ≥1 of the following: (1) pulmonary edema/congestion by chest x-ray, (2) dilated LV or poor LV function by echocardiography or ventriculography, or (3) evidence of LV diastolic dysfunction. HFpEF was defined as meeting these criteria with an LVEF ≥50%, HF with reduced LVEF ≤40%, and HF with mildly reduced LVEF of 40-49% as identified on echocardiogram or imaging studies at the time of HF hospitalization. Cases for which no LVEF data were available for classification were included in analyses for total HF events but not HF subtype events. Records were independently adjudicated by 2 physicians for HF classification and event date assignment.A subset of participants also underwent a standardized cardiovascular magnetic resonance imaging (CMR) imaging protocol at baseline with 1.5-T scanners (Magnetom Avanto and Magnetom Espree, Siemens Medical Systems, Erlangen, Germany) with 6-channel anterior and posterior phased-array torso coil elements, described in detail previously.22 Briefly, LV function, mass, and dimensions were estimated with a steady-state free precession pulse sequence to acquire 1 cine horizontal long-axis section, at least 12 cine short-axis sections from the atria to the cardiac apex, and 1 cine vertical long-axis section. Additional details can be found in the Supplemental Methods.We performed iCPET to define central, pulmonary, and peripheral deficits contributing to EI.23 In brief, participants underwent pulmonary function testing followed by pulmonary arterial catheterization through the internal jugular vein and systemic arterial catheterization through the radial artery. Resting end-expiratory and mean right atrial pressure, pulmonary artery pressure, and PCWP were recorded in the supine position in the catheterization laboratory. Participants then performed maximal incremental exercise on an upright cycle ergometer, which involved an initial period of at least 3 minutes of rest followed by 3 minutes of unloaded exercise and a continuous ramp protocol until maximal volitional effort. An individualized ramp protocol was used according to the participant’s estimated fitness level (10–25 W/min). After peak exercise was achieved, the ergometer resistance was removed, and patients were instructed to continue unloaded pedaling for 1 minute followed by 2 minutes of passive upright rest. Upright hemodynamic measurements were obtained at rest and minute by minute during exercise and averaged across ≥3 respiratory cycles. Upright hemodynamic measurements were recorded after setting the zero-reference level using the fourth anterior intercostal space in the transverse plane, the midsagittal plane, and the midthoracic level in the frontal plane.24Breath-to-breath respiratory gas exchange was continuously measured with a metabolic cart interfaced with the cycle ergometer (Medical Graphics Corp, St. Paul, MN) to assess minute ventilation, oxygen uptake (Vo2), and CO2 output (Vco2). The peak respiratory exchange ratio was derived from the highest achieved ratio of Vco2 to Vo2 during exercise. Fick cardiac output (CO) was determined at the same intervals at rest and during each minute throughout exercise by measuring Vo2 and simultaneous radial arterial and mixed venous O2 content to calculate the arteriovenous O2 content difference (Cavo2). Peak Vo2 was defined as the highest O2 uptake averaged over 30 seconds during the last 90 seconds of symptom-limited exercise. PCWP/CO slope was calculated from upright matched PCWP and CO measurements obtained at rest and each minute of exercise (mean, 10.2±2.5 repeated measurements per test).We defined 7 exercise physiological deficits applicable to HFpEF based on iCPET-derived measurements, spanning cardiac, vascular, pulmonary, and peripheral limitations to functional tolerance based on prior published work by our group and others23,25–32: (1) elevated exercise LV filling pressure, defined as PCWP/CO slope >2 mm Hg·L−1·min−123,25; (2) decreased stroke volume, defined as percentage of predicted peak exercise stroke volume <80%26; (3) decreased heart rate augmentation, defined as percentage of predicted peak exercise heart rate <80%31,32; (4) increased pulmonary vascular resistance (PVR), defined as PVR at peak exercise >1.5 Wood units27,28; (5) decreased breathing reserve, defined by indexing peak minute ventilation to maximum voluntary ventilation >80%29; (6) increased metabolic cost of initiating exercise, defined as internal work >30 W10; and (7) decreased peripheral oxygen extraction, defined as percentage of predicted peak exercise oxygen extraction Cavo2 <80%30 (Figure 2A). We also included peak Vo2 as a global measure of integrated cardiopulmonary reserve relevant to clinical decision-making in HF.Heterogeneity in cardiac, vascular, pulmonary, and peripheral mechanisms quantified during exercise in HFpEF. A, The 7 invasive cardiopulmonary exercise testing (iCPET)–derived exercise physiological deficits with abnormal thresholds for each deficit as indicated. B, Joint prevalence of physiological deficits in the heart failure with preserved ejection fraction (HFpEF) iCPET cohort. Each participant was first classified as having or not having each of the 7 exercise deficits based on the abnormal thresholds shown in A, and these binary classifications were then used to construct the UpSet plot. Vertical bars represent the number of participants exhibiting each specific combination of deficits (shown only for combinations occurring in ≥10 individuals). Horizontal bars indicate the total number of participants meeting criteria for each individual deficit, and colored dots denote which deficits are present within each combination. C, Heat map in which each row represents 1 patient with HFpEF and each column represents 1 of the 7 exercise deficits. For every patient, the measured value of each deficit was converted into a percentage of the predicted normal value so that all deficits could be shown on the same scale. Darker blue indicates worse impairment (lower percent of predicted), as illustrated by the raw physiological values shown in the scale in the top right corner of the figure. Patients are sorted from top to bottom based on their pulmonary capillary wedge pressure/cardiac output (PCWP/CO) slope, with those having the highest PCWP/CO slope at the top and those with the lowest at the bottom, allowing readers to visualize how the other 6 deficits change across the full spectrum of central hemodynamic limitation. Cavo2 indicates arteriovenous oxygen content difference; HR, heart rate; IW, internal work; MVV, maximum voluntary ventilation; pred., predicted; PVR, pulmonary vascular resistance; SV, stroke volume; VE, minute ventilation; and WU, Wood units.Given the strong links between exercise physiology, HF, and metabolism,11,12,20,33–36 we performed metabolite profiling using methods described by our group (Broad Institute of Harvard and Massachusetts Institute of Technology; methods reproduced with attribution by this statement and minimal modification from the reference20 for scientific rigor and reproducibility). Fasting plasma samples were collected at rest before exercise, centrifuged, and stored at −80° C without freeze-thaw before assay. Metabolite profiling was performed with 2 liquid chromatography-mass spectrometry methods: (1) amino acids, amino acid metabolites, acylcarnitines, dipeptides, and other cationic polar metabolites (hydrophobic interaction liquid chromatography in the positive ionization mode) and (2) sugars, organic acids, purines, pyrimidines, and other anionic polar metabolites (hydrophobic interaction liquid chromatography in the negative ionization mode). Analyses used <100 μL of plasma per sample and were performed in a single laboratory (no interlaboratory variability) using non–absolute quantification assays refined within the Broad Institute Metabolite Profiling Platform. MESA metabolite profiling was performed in the same laboratory, as described elsewhere.37Group comparisons were performed with unpaired t tests for 2-group comparisons and 1-way ANOVA for comparisons across multiple groups. We estimated Cox models for each of the exercise deficits (and peak Vo2) for association with our primary composite outcome of incident cardiovascular event or all-cause mortality. Models were adjusted for age, sex, and body mass index (BMI). As a sensitivity analysis, we also estimated hazards for incident HF or all-cause mortality. For 2 exercise deficits (peak stroke volume and peak heart rate), the survival analysis was also adjusted for the resting value to avoid regression to the mean effects.Of 471 metabolites assayed, we removed metabolites with missing data on >40% of study participants and drug metabolites, leaving 425 metabolites in our final analytical set. Missing values were imputed at 50% of the lowest detected level for that metabolite across participants. We log-transformed and standardized metabolite levels for modeling. In addition, 3 of the 7 CPET-defined physiological deficits (PCWP/CO slope, internal work, and PVR) were log-transformed to address skew, and all deficits were subsequently standardized. We estimated linear regression models adjusted for age and sex to measure the association of each of the 7 exercise deficits and peak Vo2 with metabolites (1 model per metabolite), with a 5% Benjamini-Hochberg false discovery rate to address multiplicity. Of note, for 2 deficits (peak heart rate and peak stroke volume), the value at rest was included as a covariate to account for mean regression effects. These metabolites were entered into genetic analyses (see the Human Genetics section).Next, we constructed a multimetabolite signature of each of the 7 deficits (and peak Vo2) through least absolute shrinkage and selection operator (LASSO) regression with cross-validation. As with linear models, for peak heart rate and stroke volume, the corresponding resting value was included with metabolites in LASSO modeling. However, the coefficients for these resting values were not included in calculating the final metabolic signature. Given our desire to validate our metabolic signatures in the MESA study, we included only metabolites that were measured across both cohorts (iCPET and MESA, matched with Human Metabolome Database identifier or name; 298 identifiable metabolites). Given the uniqueness of our data (iCPET in a large HFpEF sample), we randomly split our data into training (70%) and testing (30%) subsets. The relation between signatures was estimated by Pearson correlation. Each signature was regressed in Cox models against both our primary and secondary outcomes (with adjustments for age, sex, and BMI).We studied 6345 MESA participants with plasma metabolite profiling at baseline examination 1 (2000–2002) and follow-up observed through December 31, 2021 (median follow-up, 18.6 years). As in the iCPET discovery, missing values in MESA were imputed at 50% of the lowest detected level for a given metabolite. We log-transformed and standardized metabolite levels, calculated deficit-specific metabolic scores (hereafter referred to as signatures) using coefficients from LASSO models derived in the iCPET study, and standardized signatures for comparability in downstream modeling.We first estimated cross-sectional associations between metabolic signatures and demographic and clinical characteristics, including cardiac structural and functional metrics predictive of incident HF in MESA, among the subset of participants who underwent CMR.22,38 Basic demographic and clinical characteristics were analyzed as predictors of metabolic signatures, with adjustment for study site. We quantified the proportion of the variance in metabolite scores explained by key demographic and clinical risk factors using type I sums of squares from multivariable regression models. Metabolic signatures were analyzed as predictors of continuous CMR metrics (LVEF, LV mass index, LV mass/volume, and left atrial volume index, with adjustment for study site, age, sex, and self-reported race and ethnicity.We then estimated hazard ratios (HRs) separately for time to incident adjudicated all-cause HF and HFpEF per 1-SD increment in metabolite score using Cox proportional hazards. Models were first adjusted for study site, age, sex, and self-reported race and ethnicity (model 1) and then further adjusted for traditional risk factors, including BMI, current smoking, diabetes, systolic blood pressure, antihypertensive medication use, total and high-density lipoprotein cholesterol, and estimated glomerular filtration rate (2021 Chronic Kidney Disease Epidemiology Collaboration; model 2). We further adjusted for log-transformed plasma NT-proBNP (N-terminal pro-B-type natriuretic peptide) among 5399 participants with available measurements (model 3). Time at risk was calculated as time from the date of MESA examination 1 to incident event, death, loss to follow-up, or end of follow-up (December 31, 2021), whichever occurred first. We also calculated C statistics and continuous net reclassification improvement for HF prediction with the addition of each metabolite score (separately and combined) compared with our model with traditional risk factors, with bootstrapped 95% CIs calculated over 1000 resamples.Statistical analyses were performed with R. A 2-tailed value of P<0.05 was considered statistically significant, with type 1 error handling as noted. Proportional hazards assumptions were assessed with Schoenfeld residual plot visualization or time-by-covariate interactions.Identifying metabolites associated with a physiological deficit does not necessarily provide support for the potentially causal involvement of the underlying metabolic pathways in HF. We mapped 136 metabolites (by human metabolome database identifier) associated with any of the 7 deficits and peak Vo2 in the HFpEF sample (of 304 unique human metabolome database identifiers at 5% false discovery rate) to genes statistically or biologically implicated in their metabolism using curated references from our group (GeneMAP).39,40 We then performed a TWAS using the largest genome-wide association study (GWAS) of HF,41 including ≈2 million individuals (≈153 000 with HF), to identify functional gene-metabolite pairs. TWAS models were developed after our previous work from the Genotype-Tissue Expression Project42 using the MR-JTI approach,42 which incorporates causal inference into traditional TWAS. We prioritized HF as an outcome (not HFpEF) in GWASs to maximize statistical power (with a secondary analysis on an HFpEF subset involving ≈3800 individuals), and we prioritized tissues central to cardiometabolic-kidney processes in HFpEF or HFpEF-related risk5 (skeletal muscle, visceral adipose tissue, LV, kidney [cortex], liver, and lung) for TWAS. For visualization, we curated top targets across tissues defined by magnitude of the product of statistical significance and effect size (−log10 false discovery rate×z score). Multiple hypothesis testing was controlled in TWAS with Bonferroni correction (5% threshold for statistical significance).To estimate enrichment for TWAS signals among a prioritized set of genes (eg, genes implicated by our deficit metabolome through GeneMAP), we tested for a significant effect (at 0.05 adjusted for the 6 tissues evaluated) of gene membership in the prioritized set on the TWAS Z score while accounting for the correlation in gene expression. We thus fit the following mixed-effect model:Here, ZTWAS is the n-dimensional vector of gene-level TWAS z scores, β is a vector of fixed effects (in this case, involving the covariates gene length, number of single-nucleotide polymorphism features in the TWAS gene model, and gene set membership), and X is the incidence matrix. The n-dimensional random vectors u (with incidence matrix Z) and ε (residual errors) are multivariate normal:where Σ = ω2G is an n×n covariance matrix of predicted gene expression and Inxn is the identity matrix. For genes that lie on different chromosomes or are separated by >5 MB on the same chromosomes, this correlation was set to zero. We used the lme4qtl package.Enrichment for signals within TWAS was visualized by quantile-quantile plots. Last, using the same gene targets, we performed TWASs of estimated glomerular filtration rate (based on serum creatinine, N=1 004 04043,44), BMI (N=419 163 in UK Biobank45), and type 2 diabetes (N=2 535 601 individuals, 428 452 cases46) to assess whether the HFpEF deficit metabolome could identify a shared genetic predisposition across HFpEF and these linked conditions.
Results
The MGH-ExS iCPET sample included 820 subjects who met diagnostic criteria for HFpEF with a mean age of 61.7±14.1 years, 54% female, and BMI of 30.6±6.7 kg/m2 (Table 1). Distribution of exercise deficits (Figure 2A) is summarized in Table 2. In contrast to a traditional “cardiocentric” view of HFpEF, we observed a broad range of cardiac and extracardiac exercise deficits among individuals with HFpEF. When analyzed as dichotomous or continuous measures (as percentages of predicted values; Figure 2B and 2C), the severity and pattern of exercise physiological deficits were heterogeneous in HFpEF, with most patients demonstrating concurrent abnormalities across multiple organ systems. Notably, exercise deficits beyond PCWP/CO slope were not proportionate to PCWP/CO slope, suggesting that they did not arise secondarily to central hemodynamic abnormalities.Characteristics of the iCPET HFpEF CohortPrevalence of Physiological Deficits Across Diagnostic Groups in HFpEFAcross the 7 physiological deficits (and peak Vo2), we observed strong, directionally plausible associations between deficits and adverse prognosis among individuals with HFpEF at a median 3.9-year follow-up (25th–75th percentile, 1.7–7.3 years; 241 HF/mortality events, 352 cardiovascular/mortality events; Figure 3A; full regression results in Supplemental Data File 1). When we stratified patients with HFpEF into 6 groups by the total number of deficits (0, 1, 2, 3, 4, ≥5 physiological deficits, exclusive of peak Vo2), we observed a monotonic decline in peak Vo2 in patients with HFpEF with increasing number of deficits (P<0.0001 across groups; Figure 3B). A greater number of deficits was also associated with a progressively higher hazard for our combined end point of cardiovascular hospitalization or all-cause mortality: Individuals with ≥5 exercise deficits had a nearly 4-fold higher hazard of the composite outcome relative to those with no deficit (HR, 3.90 [95% CI, 1.74–8.75]; P<0.0001; Figure 3C), and the hazard per number of deficits was 1.28 (95% CI, 1.17–1.39; P<0.0001). Although we observed some nonproportionality of HRs by age in fully adjusted models, inclusion of interaction of time with age in adjusted models did not attenuate the significance of the increased hazard related to number of deficits (HR, 1.28 [95% CI, 1.17–1.40]; P<0.0001) or with ≥5 exercise deficits compared with no deficits (HR, 3.93 [95% CI, 1.75–8.80]; P<0.0001).Physiological deficits are linked to progressive decline in functional capacity and clinical outcomes in HFpEF. A, Association of each exercise physiological deficit with two adjudicated outcomes: cardiovascular hospitalization or all-cause mortality, and heart failure hospitalization or all-cause mortality among participants with heart failure (H) with preserved ejection fraction (HFpEF) in the invasive cardiopulmonary exercise testing (iCPET) cohort. Hazard ratios (HRs) were estimated from Cox models adjusted for age, sex, and body mass index (BMI). B, Each patient with HFpEF had the number of exercise physiological deficits quantified across all 7 deficits, and was then assigned to a single group based on the total deficit count (0, 1, 2, 3, 4, or ≥5). Deficits were determined from the abnormal thresholds shown in Figure 2A. The plot displays peak oxygen consumption (Vo2) for patients within each deficit-count group, with each dot representing an individual patient. Boxplots display median, interquartile range, and overall distribution. One-way ANOVA suggested significant differences across all groups (P<0.0001). C, Prognostic association of compound physiological deficits in HFpEF using Kaplan-Meier curves for the composite outcome of all-cause mortality or cardiovascular events. As in B, each patient with HFpEF was classified according to the total number of exercise physiological deficits and assigned to 1 of the deficit-count groups (0, 1, 2, 3, 4, or ≥5 deficits). Curves illustrate survival differences across these groups, with the reference group consisting of patients who had none of the 7 deficits. Inset HRs are derived from Cox proportional hazards models adjusted for age, sex, and BMI. Cavo2 indicates arteriovenous oxygen content difference; CO, cardiac output; CV, cardiovascular; MVV, maximum voluntary ventilation; PCWP, pulmonary capillary wedge pressure; PVR, pulmonary vascular resistance; and VE, minute ventilation.Patients with HFpEF with 1, 2, and 3 exercise cardiac deficits (decreased HR or stroke volume or elevated PCWP/CO slope) compared with no exercise cardiac deficits demonstrated progressively greater impairment in exercise capacity and incrementally higher hazards for the composite end point (Figure S3A and S3B). Extracardiac deficits (decreased breathing reserve, decreased peripheral O2 extraction, increased PVR, and increased internal work) were associated with lower peak Vo2, but their relationship with outcomes was not statistically significant except in the presence of ≥3 deficits (Figure S4A and S4B).Given the emerging recognition of a role for metabolic dysfunction as integral to HFpEF pathogenesis, we next measured the relation of a circulating metabolome with the set of prognostic physiological deficits by iCPET (the deficit metabolome). We observed a broad set of significant, shared metabolic associations in HFpEF (Figure 4A), with concordance in direction of metabolite association across deficits (Figure 4B displays metabolites associated with at least 4 exercise measures; full regression summaries in Supplementary Data File 1). As a central “integrator” of EI, peak Vo2 was broadly associated with metabolites linked to global dysmetabolic states, including liver steatosis and decreased cardiorespiratory fitness (dimethylguanidino valeric acid,47,48 glutamate), lipid and oxidative metabolism (carnitines, Krebs cycle intermediates [eg, fumarate, malate, cis-aconitate]), nitric oxide signaling (symmetric dimethylarginine), and anti-inflammatory microbial metabolites (cinnamoylglycine, imidazolepropionic acid49), among others.Metabolic correlates of physiological deficits in HFpEF. A, Joint metabolite-deficit associations across the 7 deficits from linear regression models built in invasive cardiopulmonary exercise testing (iCPET) participants with heart failure with preserved ejection fraction (HFpEF) adjusted for age and sex. B, This result shown graphically through a heat map of estimated regression coefficients of metabolite-deficit associations, subset to metabolites significantly associated with ≥4 clinical phenotypes. HN and HP in metabolite name are hydrophobic interaction liquid chromatography (HILIC) negative and HILIC positive, respectively. False discovery rate (FDR) <0.05. Full regression results are available in Supplementary Data File 1*. C**, Metabolites (FDR <0.05) related to peripheral (arteriovenous oxygen content difference [Cavo2]) and central (pulmonary capillary wedge pressure/cardiac output [PCWP/CO] slope) deficits. MVV indicates maximum voluntary ventilation; PVR, pulmonary vascular resistance; and VE, minute ventilation.In general, we observed a distinct metabolome for central and peripheral deficits, captured by PCWP/CO slope and peripheral extraction, respectively (Figure 4C). Metabolites associated with peripheral oxygen extraction (Cavo2) included pathways central to muscle metabolism, including inflammation and altered redox balance (cys-gly oxidized50), insulin resistance/signaling mechanisms (branched-chain amino acids and catabolites,51,52 aminoadipic acid,53 imidazolepropionic acid54), and markers of glycolysis (eg, sn-Glycero-3-phosphate; among others). In contrast, central cardiopulmonary reserve measures (eg, PCWP/CO slope) were associated with metabolites implicated in mitochondrial β-oxidation (carnitines55), vascular inflammation and hypertension (glutamate56), and urea cycle catabolites (homocitrulline, uric acid57–59), all implicated in metabolic function or cardiac dysfunction in HF. We also observed clustering of some metabolites with more favorable metabolic responses to exercise (Figure 4B). For example, higher circulating levels of long-chain acyl lysophosphatidylcholines have been associated with higher mitochondrial oxidative capacity in the Baltimore Longitudinal Study of Aging,60 and we observed higher peak Vo2, Cavo2, and heart rate and more efficient exercise initiation with higher levels of lysophosphatidylcholines 19:0, 20:0, and 24:0 (Figure 4B). Circulating levels of several phosphatidylethanolamines were also associated with favorable metabolic responses to exercise, consistent with their known role in promoting aerobic capacity in muscle mitochondria.61 Last, in addition to characterized pathways, we observed a large number of significant associations not classically implicated in HFpEF physiology (eg, polyamines62) but with a potential role in cellular growth and metabolic processes.To prioritize genetic pathways of metabolism linked to HF, we next mapped functional gene-metabolite pairs (from the application of GeneMAP to the deficit metabolome) to the findings from a TWAS of HF using the largest GWAS of the phenotype to date (N≈2 million; see Methods and Figure 5; all results given in Supplementary Data File 1). Overall, we found strong enrichment in genes implicated by the exercise deficit–associated metabolome in the TWAS of HF (quantile-quantile plot, Figure 5A). We prioritized genes by significance and effect size in each tissue (filtered by 5% false discovery rate and within top 20 displayed in Figure 5B), providing evidence of genes with cross-tissue effects on HF risk with putative function in HF (eg, ILRUN, PGAP3, HSD17B12, NFATC2IP, KAT8) and more tissue-specific associations (eg, LPA in liver only; Figure 5C). This focused approach yielded both canonical and emerging pathways in HFpEF (Table 3) implicated by our deficit-associated metabolome, including (1) inflammation (ILRUN63); (2) hypertrophy and fibrosis (TNKS,75 LOXL4,79 NFATC2IP,67,68 SH2B186); (3) metabolism and energetics, including mitochondrial function and mitophagy (HSD17B12,89 KAT8,69 CA12,72 TUFM,76,77 LPL, MRPS16,85 PCCB71); and (4) vascular reactivity at rest and with exercise stimulus (NOS3,87,88 PNMT93), among others. Particularly important were potential mediators of HF identified by our approach that have not been extensively studied previously, including genes and long noncoding RNAs implicated in cancer (eg, MAB21L4, INCA1, DARS-AS1), metabolite transport (eg, SLC5A11), mitochondrial function (STARD3), transcriptional control (ZBTB4, ZNF789), innate immunity (NCR3), drug metabolism (NAT2), and lysosomal metabolism (ATP6V1G2), among several others without curated function. Indeed, several had plausible direct metabolic connections to HFpEF, including STARD3 (involved in cellular cholesterol transport, in which higher STARD3 expression may disfavorably affect mitochondrial lipid accumulation and cellular function90) and SLC5A11 (SGLT family of transporters, involved in myo-inositol shuttling, more highly expressed in muscle and kidney than heart91).Deficit Metabolome-Implicated Genes Highly Enriched in HF GWAS (≈2 Million Participants)Genes implicated by the deficit metabolome map to genome-wide signals in HF and HFpEF comorbidities by TWAS. A, Quantile-quantile plot of observed and expected P values for each included tissue (see Methods) in transcriptome-wide association studies (TWASs) of the largest genetic study in heart failure (HF; ≈2 million participants). B, Heat map of the TWAS effect size from the HF TWAS (more positive=higher HF risk) for genes that were significant at 5% false discovery rate in at least 1 tissue (FDR; per tissue) and within top 20 for each tissue (ranked by product of significance level and effect size; see Methods). An effect size of 0 was entered for those genes for which a TWAS model was not able to be constructed to facilitate cluster visualization. C, Intersects between significant genes (at 5% FDR) across tissues in the HF TWAS. Red arrow denotes the intersect of all tissues, with the 7 corresponding genes labeled (Table 3 provides ontology). D through F, Quantile-quantile plots for genes implicated by the deficit metabolome in diabetes (D), creatinine (E), and body mass index (F), demonstrating enriched signals for these HF with preserved ejection fraction (HFpEF) comorbidities among genes implicated by the deficit metabolome. G, Venn diagram displaying intersection of genes across the HF and HFpEF comorbidity TWAS, with 25 overlapping across the conditions. LV indicates left ventricle; and VAT, visceral adipose tissue.Notably, genes linked to the deficit metabolome were strongly enriched in large GWASs of cardiometabolic-kidney HFpEF comorbidity (Figure 5D), implicating a shared genetic architecture revealed by metabolic exercise responses across HF and HFpEF comorbidities. Of note, despite a dramatically smaller sample size in the HFpEF sample in the largest GWAS of HF to date,41 we found highly significant (Spearman) correlations in gene-level (TWAS) effect size between HF and HFpEF (a smaller subset of ≈3800 individuals) for the genes linked to the deficit metabolome in each tissue (LV: ρ=0.24, P=5.6×10−14; visceral adipose tissue: ρ=0.24, P=6.4×10−16; skeletal muscle: ρ=0.22, P=9.5× 10−12; kidney: ρ=0.20, P=9.5× 10−5; liver: ρ=0.27, P= 4.2×10−14), including some genes that we identified in Table 3 (eg, SLC22A3,70 MRPS1685) in HFpEF.Modern stratification algorithms in individuals both at risk for and with established HFpEF rely on measures of metabolism and cardiac function that may not capture metabolic heterogeneity underlying HFpEF-defining, precision physiological deficits.92 We next turned to the construction of multimetabolite models of our precision deficits. We estimated L1-penalized models for each of the 7 deficits (and peak Vo2) in our HFpEF sample (coefficients in Supplementary Data File 1). Model fits in a hold-out 30% sample (see Methods) are shown in Figure S5, with best fits for peak Vo2. We observed a strong association (adjusted for age, sex, and BMI) for our composite outcome of HF or cardiovascular hospitalization and all-cause mortality (Figure 6; full Cox model results in Supplementary Data File 1). Metabolite models in the HFpEF sample of central hemodynamic (PCWP/CO slope: HR, 1.31 [95% CI, 1.16–1.47]; exercise PVR: HR, 1.29 [95% CI, 1.15–1.45]), obesity-related exercise inefficiency (internal work: HR, 1.19 [95% CI, 1.05–1.35]), and global metabolic response to exercise (peak Vo2: HR, 0.79 [95% CI, 0.70–0.90]) were all associated with clinical outcomes in an expected direction, consistent with observations from the parent iCPET measures (Figure 3A).Survival analysis of metabolic signatures of physiological deficits in HFpEF. Forest plots of association hazards from Cox models comparing the least absolute shrinkage and selection operator–based metabolite signatures of each deficit against our 2 coprimary outcomes: composite cardiovascular event or all-cause mortality and composite heart failure (HF) event or all-cause mortality. Cavo2 indicates arteriovenous oxygen content difference; CO, cardiac output; CV, cardiovascular; HFpEF, heart failure with preserved ejection fraction; HR, hazard ratio; MVV, maximum voluntary ventilation; PCWP, pulmonary capillary wedge pressure; PVR, pulmonary vascular resistance; VE, minute ventilation; and Vo2, oxygen consumption.It is important to note that for all primary exposure variables (number of deficits and metabolite-based scores), time-by-covariate interaction terms were nonsignificant, indicating stable relative effects over follow-up despite mild nonproportionality in age.To assess whether these findings were predictive of incident HFpEF in the general population, we studied 6345 MESA participants with metabolite profiling (methods similar to the iCPET cohort). Demographic and clinical characteristics of MESA participants are summarized in Table 4. The overall age and sex distributions were similar to those in our iCPET cohort (age, 62±10 years; 52% female), with a similar prevalence of diabetes (12%) and lower prevalence of hypertension (45%). On average, LVEF was preserved (69±7%), with only 2% of individuals with an LVEF <50%. Reconstructed metabolic signatures in MESA exhibited low to moderate correlation with each other (Figure S6) and expected sex and age relations based on the parent iCPET phenotype (eg, greater peak Vo2 metabolite signature at younger age in men; Figure S7). Of note, although clinical risk factors thought to account for HFpEF risk (eg, BMI, age, sex, diabetes) were associated with metabolic signatures, they did not explain a majority proportion of the variation in any of them (Figure 7).Characteristics of 6345 MESA Participants at the Time of Metabolite Profiling (Baseline Examination 1)Relation of metabolic signatures of physiological deficits derived in HFpEF iCPET sample to key clinical-demographic risk factors for HFpEF in MESA. Stacked bar plot (top) depicts variance in each metabolite score explained by key demographic and clinical risk factors, estimated by type I sums of squares from multivariable regression models. Heat map (bottom) depicts associations between metabolic signatures and these risk factors as mean standardized differences in metabolic signatures comparing presence and absence of dichotomous characteristics or per 1-SD increment in continuous characteristics, estimated with linear regression and adjusted for study site. Cavo2 indicates arteriovenous oxygen content difference; CO, cardiac output; HFpEF, heart failure with preserved ejection fraction; iCPET, invasive cardiopulmonary exercise testing; MESA, Multi-Ethnic Study of Atherosclerosis; MVV, maximum voluntary ventilation; PCWP, pulmonary capillary wedge pressure; PVR, pulmonary vascular resistance; VE, minute ventilation; and Vo2, oxygen consumption. P<0.05. P<0.01. ***P*<0.001.We next examined the relation of each signature with phenotypic markers of subclinical HF and incident HF. Among MESA participants who underwent CMR at baseline, deficit metabolic signatures were strongly associated with LV size and geometry with clinically consistent directionality, largely robust to multiple testing (Bonferroni; Figure 8A and Supplemental Data File 1). Specifically, imaging traits linked to HFpEF susceptibility (LV mass, concentricity, and left atrium volumes) but not systolic function were related to metabolic signatures of lower peak Vo2 and more adverse central hemodynamic (PCWP/CO slope, exercise PVR, stroke volume) deficits after adjustment for study site, age, sex, race, and ethnicity. Of 441 incident HF cases over a median of 18.6 years (25th–75th percentile, 12.0–19.5 years) in MESA, 408 had LVEF data available for HF subtype classification (196 as HFpEF; 212 as HF with reduced LVEF and HF with moderately reduced LVEF). Nearly all metabolic signatures from the iCPET were associated with incident overall HF in MESA, with adjustment for study site, age, sex, race, and ethnicity (with varying degrees of attenuation on further adjustment for traditional risk factors, Figure 8B; full models in Supplemental Data File 1). Consistent with prior studies in healthy populations,36,93 the metabolic signature for peak Vo2 was strongly associated with HF. It is important to note that 2 metabolic signatures of deficits accessible only through invasive techniques (PCWP/CO slope, exercise PVR) characteristic of HFpEF exhibited a strong, risk factor–independent association with HF (including after adjustment for plasma NT-proBNP and CMR-derived LV mass index; results in Supplementary Data File 1). Metabolic signatures of higher PCWP/CO slope, higher exercise PVR, and lower peak exercise heart rate—all reflective of hemodynamic or autonomic compensation during exercise—were strongly associated with incident HFpEF (eg, PCWP/CO slope HR, 1.43 per 1-SD increment [95% CI, 1.20–1.71]; P=7.18×10−5 for HFpEF-derived signatures, all Cox results in Supplementary Data File 1). We noted some mild violations of proportionality for adjustment variables (sex and diabetes) but not for metabolite-based scores.Clinical-structural associations of metabolic signatures of physiological deficits in MESA. A, Standardized mean difference in cardiac characteristics per 1-SD increment in metabolic signatures, estimated with linear regression and adjusted for study site, age, sex, and self-reported race and ethnicity. Sample sizes varied by cardiovascular magnetic resonance imaging (CMR) parameter: n=4684 for left ventricular (LV) ejection fraction (LVEF), LV mass indexed by body surface area (LVMi), and LV mass-to-volume ratio; and n=4007 for maximum left atrial volume indexed by body surface area (LAVi); B, Results of survival analysis for all-cause HF and HF with preserved ejection fraction (HFpEF)–specific hazards. Standardized hazard ratios (HRs) were estimated with Cox models in 6345 participants in MESA (Multi-Ethnic Study of Atherosclerosis) with 441 incident HF cases (any subtype) adjudicated over a median of 18.6 years. Forest plot shows results for analyses adjusted for study site, age, sex, and self-reported race and ethnicity (model 1) and further adjusted for traditional cardiovascular risk factors (model 2; see Supplementary Data File 1). Complete results of these analyses are presented in Supplementary Data File 1. Cavo2 indicates arteriovenous oxygen content difference; CO, cardiac output; MVV, maximum voluntary ventilation; PCWP, pulmonary capillary wedge pressure; PVR, pulmonary vascular resistance; VE, minute ventilation; and Vo2, oxygen consumption. P<0.05. P<0.01. ***P*<0.001.Our adjusted models demonstrate the complementary nature of metabolite profiles added to traditional risk factors in predicting incident HF. In MESA, the addition of metabolic signatures yielded a 38% continuous net reclassification improvement over traditional risk factors, which remained near ≈20% after inclusion of plasma NT-proBNP (included in the Supplementary Data File 1).
Discussion
HF has traditionally been viewed as a disease of the myocardium, with measures that capture elevated cardiac filling pressures as a prime diagnostic arbiter.7–9 Emerging recognition of HFpEF as a systemic disorder, integrating inputs from cardiac, pulmonary, autonomic, vascular, and peripheral muscular sources (among others), has challenged this view.2,3,10,94–97 Here, we present results from the largest sample to the best of our knowledge of iCPET with HFpEF to phenotype HFpEF susceptibility across multiple organ domains linked to molecular states. Individuals with HFpEF exhibited heterogeneity across all domains, with compound deficits across domains proving to be functionally and prognostically significant. The deficit-associated metabolome reflected pathways of inflammation and oxidative stress, insulin resistance, oxidative metabolism, vascular inflammation, nitric oxide signaling, and microbial metabolites, all putatively related to HFpEF. These results were bolstered by genetic studies (≈2 million individuals) in which tissue-specific TWAS approaches using genes nominated by the deficit metabolome demonstrated 2 major findings: (1) striking enrichment of deficit metabolome-nominated genes for gene-level associations with HF (including concordance with HFpEF effect sizes), identifying potential genetic targets, and (2) a shared genetic architecture of HF and HFpEF-driving comorbidity (obesity, renal disease, diabetes), linking lifelong genetic liability of comorbidity and HF. Last, deficit metabolic signatures were strongly related to HFpEF and relevant outcomes in our iCPET cohort and 6345 at-risk individuals from MESA, with significant reclassification and discrimination over clinical risk (over ≈2 decades). Collectively, this study provides the broadest to-date resource linking deep exercise responses across multiple organ systems to genetics, metabolism, and outcomes across the spectrum of clinical risk, offering insight into personalized, precision risk assessment in HFpEF and potential therapeutic targets grounded in diagnostic pathophenotypes.The dramatic positive clinical results of glucagon-like peptide-1 receptor agonist therapy98 and sodium-glucose cotransporter 2 inhibition99,100 have vaulted metabolic dysfunction to center stage in HFpEF pathogenesis. In response, modern algorithms incorporate BMI along with cardiocentric traits (echocardiography, atrial fibrillation) to capture this role of metabolism in HFpEF.92 Although these accessible measures may identify susceptibility,101 they are less effective at resolving heterogeneity within HFpEF, which presents with both central and peripheral deficits.2,30,102–104 Here, we found that nearly every patient exhibited a combination of abnormalities spanning multiple physiological systems, with exercise deficits co-occurring in mixed combinations. Two markers of peripheral limitations in HFpEF not quantifiable at rest—metabolic cost of exercise initiation10 and decreased peripheral oxygen extraction30,105—were frequent in participants with HFpEF (45% and 44%, respectively), whereas central hemodynamic abnormalities during exercise were not uniform in patients meeting resting diagnostic criteria for HFpEF. The phenotypic specificity of defining and rank-ordering physiological deficits unmasked during exercise has potential therapeutic relevance: for example, exercise training and iron repletion may be particularly effective in individuals with poor peripheral oxygen extraction,106,107 whereas altering medications that control exercise heart rate (eg, β-blockers108,109) or ventricular-vascular coupling (eg, angiotensin receptor blocker/neprilysin inhibitors) may particularly benefit individuals with autonomic or hemodynamic limitation. Our finding that individuals with compound deficits have greater functional limitations, which culminate in peak Vo2 levels below those requisite to perform routine activities of daily living, and poorer outcomes highlights the need to quantify and target multiple deficits in HFpEF.Although these results reaffirm exercise testing as part of the diagnosis in EI and HFpEF,110 performing a CPET—especially with hemodynamic phenotyping—is not feasible for routine widespread use in most clinical settings. The tradeoff between precision and facile phenotyping has led to the search for surrogates for these more complex, difficult-to-obtain exercise-based phenotypes in HFpEF. Metabolite profiling has taken center stage in this quest for reproducible diagnostic/prognostic biomarkers of HFpEF.111–117 Most studies have compared across subtypes of HF (HFpEF versus HF with reduced LVEF), demonstrating differences in mitochondrial β-oxidation pathway, branched-chain amino acids, glutamine-glutamate balance, and metabolites of nitric oxide signaling and oxidative stress, many similar to our findings here. Although population-based studies examining the predictive ability of metabolomics have demonstrated single metabolite association with a clinical diagnosis of HFpEF (with limited event rates),117,118 our CPET-based construct builds composite metabolite models against deep HFpEF phenotypes (exercise physiological deficits)—as opposed to the heterogeneous clinical diagnosis—demonstrating strong association, discrimination, and reclassification of HFpEF by signatures that capture central hemodynamic stress (PCWP/CO slope, exercise PVR) and a composite multisignature approach. It is important to note that these results were achieved with a small number of metabolites per model (eg, <50 metabolites for PCWP/CO slope or exercise PVR) with excellent reproducibility across 2 cohorts spanning the spectrum of clinical risk (established versus incident disease). These results provide strong evidence in support of a deficit metabolome signature for HFpEF diagnosis and prognosis.A final innovation here was in integrating the deep iCPET phenotype and metabolome with human genetics.39 Through the combination of functional gene-metabolite mapping and TWAS approaches in the largest human GWAS studies, we were able to identify specific gene targets linked to HF pathogenesis. Our findings provide human genetic support for a multiorgan paradigm for HFpEF through the lens of metabolic responses to exercise. We observed a striking enrichment of genes implicated by our deficit metabolome in gene-level associations with HF across organ systems (heart, kidney, muscle, liver, lung, fat) beyond the heart alone. These genes implicated key HF pathways consistent with our metabolomic findings—inflammation, mitochondrial dysfunction, fibrosis, and hypertrophy—and a host of novel but physiologically plausible targets not yet investigated. Second, we observed enrichment for prioritized genes in noncardiac organs, both in the HF GWAS and in related cardiometabolic-kidney GWASs clinically linked to HFpEF (obesity, diabetes, renal function). We recognize that this approach was performed in the largest all-cause HF sample41 as opposed to within HFpEF GWAS, an inherent limitation given the necessarily small GWAS sample sizes for HFpEF. However, we noted a concordance in the effect size for the iCPET deficit metabolome-implicated genes between all HF and the smaller subset with HFpEF, suggesting that our findings are applicable to HFpEF. These results support a multiorgan paradigm of HFpEF pathogenesis realized through exercise phenotype, consistent with a potential for organ systems outside the heart to affect HFpEF.119 Last, given that genetic effects are lifelong, the fact that the deficit metabolome-nominated genes were strongly enriched across HFpEF metabolic comorbidities underscores support for addressing metabolic risk at early stages for HFpEF prevention.This report comprehensively integrates invasive hemodynamic CPET with metabolomic, phenotypic, prognostic, and genetic integration, highlighting the power of leveraging heterogeneous HFpEF phenotypes in risk prediction and target discovery. Several caveats warrant mention. We prescribed an accepted composite definition of HFpEF, although we recognize that these definitions of HFpEF vary across studies.120 Nevertheless, the prognostic association across MESA and our iCPET cohort is a testament to external validity. Although our GWAS integration in HF relied on all-cause HF to maximize power (versus a smaller subset with HFpEF), the deficit metabolome derivation that drove selection of genes was from a population with normal LV systolic function (MGH-ExS iCPET cohort). Our approach relied on examining each iCPET deficit individually to capture its prognostic and metabolic correlation separately. Advanced approaches to provide metabolic signatures across HFpEF phenotypes (eg, integrative artificial intelligence/machine learning) and our results in MESA suggest that additional discrimination with the addition of all metabolic signatures in a single model is supportive. In the MESA cohort, 8% of events lacked classification, reflecting in part the inherent challenges in the classification of HFpEF in particular. We elected to focus on deeply phenotyped patients in our derivation iCPET cohort to ensure precise phenotypes for the derivation of metabolite signatures while taking advantage of the large sample size and implications for generalizability with the MESA validation cohort. In addition, the iCPET cohort was enrolled over a long period and has some distinct characteristics from large clinical trials of HFpEF (eg, lower BMI). Despite this, baseline characteristics in our iCPET referral cohort showed commonality with other HFpEF cohorts on the basis of prevalent hypertension, obesity, and atrial fibrillation. Furthermore, changes in medical therapy across this period may have influenced our results. However, validation in MESA is strongly supportive of external generalizability. We also acknowledge that patients making up this cohort, who were able to perform maximum exercise, tend to be younger with milder HF severity than frequently hospitalized patients with HFpEF, yet multidomain deficits were common across the spectrum of hemodynamic impairment, as assessed by PCWP/CO slopes (Figure 2C). Conceptually stratified according to the presence or absence of exercise hemodynamic abnormalities (PCWP/CO slope ≥2 mm Hg·L−1·min−1), the overall distribution, patterning, and co-occurrence of multiorgan physiological deficits remain broadly similar between groups. Last, we observed evidence of nonproportionality in our survival analyses, suggesting the potential for a time-varying hazard. However, to avoid overstepping our data, we did not stratify models or assess this time-dependent hazard. Reported hazards represent a mean hazard across time epochs over the period of follow-up.121 Further work in large populations of HFpEF will be helpful in parsing time-specific effects.
Conclusions
In a large population of individuals with HFpEF, we characterized cardiac, vascular, pulmonary, and peripheral deficits through comprehensive iCPET, metabolomics, clinical prognostic studies, and human genetics. Our results demonstrate heterogeneity in EI mechanisms within HFpEF, with compound deficits linked to EI severity and prognosis. Genomic studies and metabolomic studies over nearly 2 decades demonstrated biologically plausible and prognostic links between the metabolomics of precision exercise deficits, metabolic dysfunction, and HFpEF, pinpointing putative and emerging pathways of HFpEF. These studies provide the largest to-date resource of phenotypic and metabolic profiles of HFpEF-defining exercise deficits, providing feasible metabolic biomarkers of HFpEF susceptibility across the spectrum of clinical risk for clinical translation.The prognostic significance of iCPET-defined physiological deficits and their multiplicity frames the need to develop and refine noninvasive surrogate measures in the form of both metabolite biomarkers and noninvasive exercise measures. Three of the iCPET-derived deficits can be assessed with noninvasive CPET (heart rate, pulmonary mechanical limitations, and internal work). For additional central cardiac measures, recent advances in exercise echocardiography hold promise for the quantification of pressure-flow relationships122 and stroke volume, as well as reciprocal derivation of peripheral oxygen extraction from simultaneous assessment of Vo2 and echocardiographic CO.In addition, the complexity of iCPET served as one of our motivations to fit metabolomic measurements to these physiological phenotypes derived from iCPET. The performance of plasma metabolite signatures in the distinct validation cohort of MESA, in whom iCPET was not performed, supports the generalizability and “portability” of the signatures beyond iCPET referral cohorts at specialized centers. Potential hurdles to clinical adoption of these metabolomic signatures include the cost and requisite analytical expertise of applying mass spectrometry–based metabolomic profiling that strives to measure a large number of metabolites with low variance. However, the use of small volumes of human plasma (<100 μL) to derive metabolic signatures, coupled with future efforts to define parsimonious signatures of each deficit with fewer prioritized metabolites, will promote the feasibility of incorporating metabolite profiling into clinical practice. Recent advances in cross-platform harmonization further support a translational pathway in which models derived from relative quantification can be applied to targeted, absolutely quantified metabolites.
Article Information
The authors thank the MGH-ExS and MESA participants, investigators, and staff for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org.Dr Lewis is supported by National Heart, Lung and Blood Institute R01 HL151841 and HL159514 for this work and National Heart, Lung and Blood Institute U01 5U01HL160278-05 and an Americal Heart Association Strategically Focused Research Network Award for related work. Dr Lewis has been a consultant for and received research support from American Regent, Applied Therapeutics, AskBio, Bayer, Cytokinetics, and Edwards and received research support from Amgen, AstraZeneca, Pfizer, Alexion, Novartis, and Rivus. Drs Shah and Post are supported by grants from the National Institutes of Health. Dr Shah has equity ownership in and is a consultant for Thryv Therapeutics. Dr Shah is a coinventor on pending patents or disclosures on molecular biomarkers of fitness, lung disease, cardiovascular diseases and phenotypes, and metabolic health; use of RNAs (including spatial) as therapeutics and diagnostic biomarkers in disease; and methods in metabolomics. Dr Mi is supported by the National Institutes of Health (K23HL171855). Dr Nayor is supported by National Institutes of Health (grants R01HL156975, R01HL131029, and R21AG086679). Dr Gamazon is supported by National Institutes of Health (grants U01DK140952 and R01HG011138) and a grant (P30DK020593) from the Vanderbilt Diabetes Center (Functional Multi-Omics of Diabetes and Metabolism Core). Dr Gamazon has performed consulting for Thryv Therapeutics and has pending patents or disclosures on molecular biomarkers of fitness, cardiovascular diseases and phenotypes, and metabolic health; use of RNAs as therapeutics and diagnostic biomarkers in disease; and methods in metabolomics. Dr Malhotra was supported by the National Heart, Lung, and Blood Institute (R01HL159514 and R01HL162928), the Leducq Foundation, the American Heart Association (22TPA969625), and the Wild Family Foundation. Dr Malhotra receives research funding from Angea Biotherapeutics and Amgen and serves as a consultant for Pharmacosmos, Myokardia/BMS, Renovacor, Rocket Pharmaceuticals, Epizon Pharma, and Third Pole. Dr Malhotra is a scientific cofounder of Angea Biotherapeutics, receives royalties from UpToDate for scientific content authorship, and has performed in speaker bureaus through Vox Media. Dr Landsteiner is supported by an American Heart Association Strategically Focused Research Network Award and the MGH Heart Failure Research Innovation Fund. Dr Ikoma is supported by the Japanese College of Cardiology Overseas Research and the MGH Heart Failure Research Innovation Fund. Whole-genome sequencing for the Trans-Omics in Precision Medicine (TOPMed) program was supported by the National Heart, Lung and Blood Institute. Whole-genome sequencing for the NHLBI TOPMed: Multi-Ethnic Study of Atherosclerosis (MESA; phs001416.v1.p1) was performed at the Broad Institute of MIT and Harvard (3U54HG003067-13S1). Centralized read mapping and genotype calling, along with variant quality metrics and filtering, were provided by the TOPMed Informatics Research Center (3R01HL-117626-02S1). Phenotype harmonization, data management, sample-identity quality control, and general study coordination were provided by the TOPMed Data Coordinating Center (3R01HL-120393-02S1). The MESA projects are conducted and supported by the National Heart, Lung, and Blood Institute in collaboration with MESA investigators. Support for MESA is provided by contracts 75N92025D00022, 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92025D00026, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92025D00024, 75N92020D00003, N01-HC-95162, 75N92025D00027, 75N92020D00006, N01-HC-95163, 75N92025D00025, 75N92020D00004, N01-HC-95164, 75N92025D00028, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, N01-HC-95169, UL1-TR-000040, UL1-TR-001079, UL1-TR-001420, UL1TR001881, and R01HL105756.Tables S1–S4Figures S1–S8Excel File S1
RM
Dr. Rohit Malhotra
Cardiology · Cardiac ElectrophysiologyUniversity of Virginia Health Systemrohit