Mitochondrial DNA Content Is Linked to Cardiovascular Disease Patient Phenotypes.

Background We sought to determine whether mitochondrial DNA (mtDNA) content can be used as markers for 12 key phenotypes among cardiovascular disease patients, and whether these markers are valid across patients with diverse ancestries. Methods and Results DNA was collected from the peripheral blood of 996 cardiovascular disease patients at the Cleveland Clinic. The mtDNA copy number and DNA-level variation were assessed from whole-genome sequence. Patients were also ascertained retrospectively for histories of 10 clinical events, as well as for maximum stenosis and extent of disease at baseline. Self-reported race and maternal ancestry inferred from mtDNA sequence were recorded. MtDNA copy number and overall mtDNA rare variant load were significantly lower in patients with histories of various adverse clinical events, and mtDNA copy number was inversely correlated with extent of disease. Strong associations were also found between absence of rare variants in the genes MT-ATP6 and MT-COII and patient histories of hyperlipidemia and myocardial infarction, respectively. Importantly, associations were not ancestry dependent. Conclusions This study provides evidence that mtDNA copy number in circulation is associated with a variety of cardiovascular disease patient phenotypes. Results also suggest a protective role for some rare inherited mtDNA variants. Overall, the study supports the potential of mtDNA content and abundance as biomarkers in heart disease, in a manner that is valid across diverse ancestries.

copy number

extent of disease

mitochondrial DNA

Clinical Perspective

What Is New?

This large retrospective study examines relationships between clinical characteristics of patients with cardiovascular disease and mitochondrial DNA (mtDNA) content in circulation.

Abundance of mtDNA in circulation was strongly inversely associated with histories of many types of adverse events and phenotypes.

Analysis of inherited mtDNA variants uncovered evidence that links rare gene variants to an absence of past myocardial infarction events and hyperlipidemia.

What Are the Clinical Implications?

The mtDNA content in blood may potentially serve as a biomarker for cardiovascular disease‐related phenotypes.

Further study is required to determine whether mtDNA content in circulation reflects its characteristics in cardiomyocytes; if so, there are implications for the pathogenesis of cardiovascular disease.

It is well established that properly functioning mitochondria are crucial to energy requirements in the vascular epithelium and the heart itself. The 13 proteins encoded by mitochondrial DNA (mtDNA) genes are all part of complexes in the electron transport chain. These genes are tightly packed in each ≈16.5‐kb copy of mtDNA, along with 2 rRNA genes, 22 tRNA genes, and noncoding regions that have roles in replication and transcription. Unlike nuclear chromosomes, the number of copies of the mitochondrial chromosome per human cell is quite fluid and differs from tissue to tissue and across developmental time.

Although many studies have queried the nuclear genome for variants that are associated with heart disease, , we hypothesized that mitochondrial DNA may also harbor variants that affect risk for cardiovascular disease (CVD)‐related traits. In addition, recent work from Ashar et al. has indicated that individuals with a lower mtDNA copy number (CN) in circulation have higher incident CVD. The same group also implicated lower mtDNA CN as an indicator of risk for sudden cardiac arrest. We therefore sought to examine CN in circulation, using whole‐genome sequence data, to query for associations with multiple CVD‐related events. To this end, we collected deep sequencing mitochondrial genome data from cases at the Cleveland Clinic. Whole‐genome sequencing has the advantage of allowing comprehensive examination of mtDNA variants, as well as mtDNA CN.

Methods

The authors declare that all supporting data are available within the article and its online supplementary files.

Study Population, Sample Collection, Processing, and Sequencing

We collected high‐coverage whole‐genome sequencing data from the blood samples of 996 patients with CVD at the Cleveland Clinic. These patients were enrolled in the Cleveland Clinic GeneBank, a prospective study from 2001 to 2006. Each patient underwent an elective diagnostic cardiac catheterization procedure without acute coronary syndrome. All participants gave written informed consent approved by the Cleveland Clinic Institutional Review Board, and had well‐documented clinical features at enrollment. Longitudinal outcomes were ascertained over 3 years. As a straightforward measure of the extent of disease, we adopted the coronary artery surgery study‐50 (CASS‐50) system, which is essentially a count of the number of coronary vessels affected (at stenosis >50%). Although simple, the CASS‐50 scoring system has strong agreement with other scoring systems. Details of the system can be found in Table S1. Peripheral blood samples were processed for DNA sequencing as previously described. Patient characteristics are summarized in Table 1, and details for each patient are provided in Table S2. The definitions for clinical features are shown in Table S1.

Table 1

Patient Characteristics

			Self‐Reported Maternal Ancestry
	Age, y, Median	Sex, % Women	White	Black	Asian	Other
Overall	56	30.4%	82.1%	14.5%	1.2%	2.2%
History of
Coronary artery bypass graft (13.2%)	60	18.9%	82.6%	14.4%	0.8%	2.3%
Coronary artery disease (47.1%)	58	19.6%	80.4%	15.6%	1.3%	2.8%
Congestive heart failure (14.2%)	57	30.0%	76.6%	21.3%	0.0%	2.1%
Diabetes mellitus (31.9%)	57	29.2%	81.8%	15.1%	1.9%	1.3%
Hypertension (64.4%)	57	30.9%	78.8%	18.4%	1.1%	1.7%
Hyperlipidemia (74.1%)	57	27.0%	85.1%	12.1%	1.1%	1.8%
Myocardial infarction (18.3%)	59	19.8%	77.5%	19.8%	1.1%	1.6%
Percutaneous coronary intervention (16.8%)	59	22.2%	76.6%	18.6%	1.2%	3.6%
Stroke (5.0%)	57	42.0%	76.0%	20.0%	2.0%	2.0%
Ventricular arrhythmias (8.5%)	57	29.4%	87.1%	11.8%	0.0%	1.2%
Maximum stenosis >50% (48.3%)	59	18.7%	80.5%	15.8%	1.2%	2.5%
Extent of disease 0 (51.8%)	54	41.3%	83.7%	13.2%	1.2%	1.9%
Extent of disease 1 (13.7%)	58	20.6%	81.6%	13.2%	1.5%	3.7%
Extent of disease 2 (12.4%)	59	22.6%	73.4%	21.8%	2.4%	2.4%
Extent of disease 3 (22.1%)	59	15.5%	83.6%	14.1%	0.5%	1.8%

Age, sex, and ancestry characteristics are given for the entire cohort (Overall row) and within groups of patients with specific clinical characteristics (other rows). Percentage of patients with each clinical characteristic is provided after row name.

Alignment and Extraction of chrM and chr22 Reads

Whole‐genome sequencing was performed with Illumina HiSeq X10TM platform with 30× mean coverage in 150‐bp pair‐end form, and reads alignment was conducted using the ISAAC pipeline to reference build hg38. Reads mapping to chromosome 22, positions 46 000 000 to 46 100 000, and to the mitochondrial chromosome were extracted from the BAM whole‐genome sequencing files using SAMtools. The mitochondrial reads were realigned to the revised Cambridge Reference Sequence using the Burrows‐Wheeler Alignment tool.

Copy Number Estimation

To estimate the mtDNA CN of each sample, chromosome 22 positions 46 000 000 to 46 100 000 (build hg38) was chosen as a region being representative of 2 copies per cell because it contains only very rare population CN variants. Using read depth as a proxy for DNA CN, the mtDNA CN was estimated as , where denotes the average read depth across the mitochondrial chromosome, and denotes the average read depth of the chromosome 22 two‐copy region.

Variant Calling and Downstream Quality Control

We used the Genome Analysis Toolkit to call single nucleotide variants. To minimize false positives, we omitted all variants in the regions from base positions 302 to 316, 513 to 526, 566 to 573, 8860, and 16181 to 16194, which are known to yield false variant calls. Because heteroplasmic variants generally impact disease phenotypes only at levels around 80%, all variants with heteroplasmy levels less than this were omitted from further consideration. We used SnpEff to categorize the effects of each mtDNA variant. Population frequency for each variant was determined from GenBank. Variants that appeared in fewer than 1% of GenBank sequences covering the variant’s position were deemed rare for the purposes of this study. We used HaploGrep2.1.13 to obtain the haplogroup of each sample.

Statistical Procedures

All statistical computations were performed using R version 3.6.1. Associations among the 11 binary clinical features (history of clinical events and maximum stenosis at least 50%) were computed as odds ratios, with corresponding P values computed using the Fisher exact test. Associations between mtDNA CN and binary clinical features were tested using the following logistic regression model:where P values were extracted from the model fit with the ANOVA (analysis of variance) function, using the χ2 test argument. To compute the age‐ and sex‐adjusted odds ratio for a binary clinical outcome associated with 1 standard deviation decrease in CN, CN was first converted to a Z score, which was used in place of the log(CN) term in Equation 1 above. The odds ratio was computed as exp(‐), where is the CN coefficient estimate in the model, and the 95% CI computed as exp(‐β ± 1.96 × standard error). Odds ratio for first versus fourth quartiles was computed using only patients with CN in these quartiles, and the log(CN) term in Equation 1 above was instead coded as an indicator variable for the quartile 1 CN. Odds ratio and 95% CIs were computed as above. To test for association with the ordinal variable extent of disease (EOD), P trend was computed with proportional odds logistic regression, using the polr() function in the MASS package. The 2‐sided P value was computed using the normal‐distribution quantile of the t value corresponding to the CN’s coefficient. We also performed these regression analyses while adjusting for maternal race/ancestry, self‐reported and inferred from mitochondrial haplogroup. Specifically, we added terms

andto the models, where here I(∙) denotes the indicator function:Rare variant association analyses were performed using a similar model,Here, rare variant is either the numerical variable log (total number of rare mtDNA variants), or a binary variable indicating presence or absence of a rare variant in the mtDNA feature of interest (e.g., specific mRNA genes, any tRNA). P values were computed as above. To correct for multiple tests, we converted P values to q values using the R package qvalue.

Results

Patient Cohort

Cleveland Clinic patients were ascertained for history of prior events/disease (Table 1). Maximum stenosis was also assessed. Table S1 provides the working definitions for each of these events/histories. Unsurprisingly, there was substantial correlation among many of these clinical features (Figure 1). Patients also reported ancestries of both parents in broad terms (Asian, Black, White). Because mtDNA is the subject of our study, we focused solely on maternal ancestry. As a complementary, objective assessment of maternal ancestry, we determined haplogroups using each patient’s mtDNA sequence. As expected, the reported maternal ancestry tracked closely with mtDNA continental haplogroups (Figure S1).

Figure 1

Correlations among clinical features of the cardiovascular disease cohort.

Each pixel is shaded according to the OR between the pairs of clinical features in the corresponding row and column, with darker shades indicating stronger correlation (see scale). Phenotype abbreviations are explained in Table S1. Fisher exact test P values are provided for each significant pair tested. OR indicates odds ratio.

mtDNA CN Is Lower in Patients With Specific Clinical Histories

Prior studies have found that low mtDNA CN in blood is associated with risk of CVD and sudden cardiac death, , demonstrating the utility of mtDNA CN as a predictor of risk. We reasoned that mtDNA CN may also be associated with other specific patient histories. To this end, we tested the 10 clinical histories and maximum stenosis for association with mtDNA CN.

For all of these 11 clinical variables, those with the corresponding history/feature had lower median mtDNA CN than those without. Seven of the 11 attained nominal statistical significance (Figure 2). The relationships between mtDNA CN and history of coronary artery disease (P=3.7×10−4, after adjusting for age and sex), history of myocardial infarction (P=9.7×10−5), history of coronary artery bypass graft (P=1.2×10−5), and maximum stenosis >50% (P=3.5×10−5) were all significant after correction for multiple testing. One standard deviation decrease in CN was associated with odds ratios of 1.29 (95% CI, 1.21–1.39), 1.43 (95% CI, 1.13–1.57), 1.58 (95% CI, 1.42–1.76), and 1.35 (1.18–1.55), respectively, for these phenotypes after adjusting for age and sex (Table 2). Correspondingly, we observed significantly more prevalent history of these events among individuals with mtDNA CN in the first quartile (Table 2).

Figure 2

Clinical features associated with lower mitochondrial DNA (mtDNA) copy number (CN).

Each panel represents a different clinical feature, given at the bottom of the panel. The 2 groups in each panel represent individuals that do (YES) or do not (NO) have the clinical feature, with each point representing a patient and the vertical axis indicating that patient’s mtDNA CN. The overlaid boxplots give the 25th, 50th, and 75th percentiles, and the whiskers extended either to the maximum/minimum values or 1.5×interquartile range (IQR) from the box (where IQR is the distance between the first and third quartiles), whichever is less extreme. P values were computed using a logistic regression model that adjusts for age and sex (see Methods section for details). Phenotype abbreviations are explained in Table S1.

Table 2

Clinical Features' Associations With 1 Standard Deviation Decrease in Mitochondrial DNA Copy Number and for First Quartile Mitochondrial DNA Versus Fourth

		Overall, n=996	White Mother, n=818	Black Mother, n=144	Haplogroup H, n=365	Haplogroup L, n=137	Haplogroup U, n=112
HxCabg OR (95% CI)	1 SD CN decrease	1.58 (1.29–1.76)	1.53 (1.22–1.91)	1.73 (1.01–2.99)	1.81 (1.23–2.66)	2.04 (1.08–3.85)	2.13 (1.08–4.20)
	1st quartile vs. 4th	5.16 (2.66–10.01)	3.97 (2.01–7.85)	4.37 (0.80–23.90)	5.09 (1.62–15.97)	6.07 (0.64–57.61)	NA
HxCad OR (95% CI)	1 SD CN decrease	1.29 (1.13–1.48)	1.24 (1.07–1.45)	1.51 (1.06–2.15)	1.37 (1.08–1.74)	1.43 (1.00 – 2.04)	1.32 (0.87–2.00)
	1st quartile vs. 4th	1.97 (1.36–2.86)	1.74 (1.14–2.67)	2.77 (1.06–7.21)	2.56 (1.31–4.97)	2.14 (0.82–5.60)	2.34 (0.63–8.74)
HxMI OR (95% CI)	1 SD CN decrease	1.43 (1.20–1.71)	1.38 (1.14–1.69)	2.06 (1.30–3.25)	1.35 (1.00–1.83)	2.25 (1.38 – 3.68)	1.43 (0.78–2.63)
	1st quartile vs. 4th	2.51 (1.54–4.12)	2.14 (1.23–3.71)	6.31 (1.74–22.83)	2.01 (0.90–4.47)	6.40 (1.78–22.95)	2.22 (0.35–13.94)
MXSTEN50 OR (95% CI)	1 SD CN decrease	1.35 (1.18–1.55)	1.37 (1.17–1.60)	1.33 (1.08–1.62)	1.48 (1.16–1.88)	1.39 (0.97–1.98)	1.52 (0.98–2.34)
	1st quartile vs. 4th	2.25 (1.54–3.28)	2.28 (1.49–3.50)	2.83 (1.07–7.47)	3.14 (1.61–6.16)	2.43 (0.92–6.46)	3.51 (0.96–12.81)

ORs and CIs adjusted for age and sex are shown overall and stratified by patient ancestry (self‐reported and assessed by haplogroup). CN indicates copy number; HxCabg, history of coronary artery bypass graft; HxCad, history of coronary artery disease; HxMI, history of myocardial infarction; MXSTEN50, maximum stenosis at least 50%; NA, insufficient numbers to calculate; and OR, odds ratio.

Race is a potential confounder in testing for association between event history and CN, as rates of specific clinical histories are not uniform across self‐reported ancestries (Table S3). However, the association between event histories and mtDNA CN remain significant when we additionally control for maternal ancestry, both self‐reported and inferred from mtDNA haplogroup (Table S4). The associations also remain largely significant when we restrict to subsets of patients within different maternal ancestral groups (self‐reported, as well as the haplogroups [H, L, and U] that have at least 10% frequency in the cohort), and in all cases the associations are in the same direction (Table 2, Figures S2‐S6). We conclude that the observed association between mtDNA CN and clinical features is independent of ancestry.

Extent of Disease Is Negatively Correlated With mtDNA CN

For each patient, severity of CVD (EOD) was scored as 0, 1, 2, or 3, on the basis of the number of diseased vessels (see Table S1 for details). There was a statistically significant trend toward lower mtDNA CN for more severe EOD (P trend=1.8×10−8). Correspondingly, individuals in higher quartiles of the mtDNA CN were significantly more likely to have less severe EOD (Figure 3).

Figure 3

Extent of disease (EOD) is inversely associated with mitochondrial DNA (mtDNA) copy number (CN). The mtDNA CN for each patient is plotted, stratified by EOD (left panel).

Here, P trend is computed using a proportional odds logistic regression model that adjusts for age and sex. In the right panel, the proportion of patients with each EOD score is shown for each mtDNA CN quartile. Higher mtDNA CN quartiles have fewer patients with more severe EOD.

As with event histories, EOD’s association with mtDNA CN was not confounded by ancestry, as the statistical relationship holds when we control for self‐reported ancestry (P trend=8.2×10−6) and mtDNA haplogroups (P trend=7.3×10−6). The association also holds within the various self‐reported ancestries and haplogroups (Figure S7).

Rare mtDNA Variant Burden Is Negatively Associated With Past Clinical Histories

Deep sequencing of 996 mitochondrial genomes allowed us to interrogate the impact of mtDNA variants on patient phenotypes. Overall, the cohort had 26 805 nonreference variants (2089 unique variants), among which 4959 (1746 unique) were deemed rare (population frequency <1%; Table S5). We chose to focus on rare variants for 2 reasons. First, they are more likely to exert a large effect than common variants. Second, common variants often track very closely with, or even define, ancestral mtDNA haplogroups, making it difficult to distinguish common variant associations with disease from associations with ancestry.

We tested for association of rare variants with each of the 11 aforementioned clinical features. By definition, each specific variant would be expected to appear in at most a handful of cases. As such, and because there are large numbers of rare variants, for any particular variant we are substantially underpowered to detect statistical enrichment that would withstand multiple‐test correction. We therefore collapsed rare variants by mtDNA features (all tRNAs combined, all rRNAs combined, control region, and nonsynonymous variants in individual mRNA genes). We tested for association between presence/absence of rare variants in each of these 16 mitochondrial genome entities and each of the 11 clinical features. We also tested overall rare variant burden for association with each clinical feature. Overall, this resulted in 187 (11×17) statistical tests, necessitating adjustment for multiple testing. We corrected for multiple tests by computing q values for each P value to control for the false discovery rate.

As shown in Figure 4A, the P values largely follow the expected null distribution, with the 3 most significant (q<0.15) deviating from expectation. Intriguingly, all of these associations are consistent with rare mtDNA variants being protective against various clinical histories. Individuals with history of hyperlipidemia have a significantly higher overall rare variant burden than patients without the history (P=2.3×10−3; q=0.14; Figure 4B). None of the 27 individuals with rare nonsynonymous variants in the MT‐COII gene have history of myocardial infarction (P=7.6×10−4; q=0.071; all variants shown in Figure 4C, colored by patient haplogroup). Presence of rare nonsynonymous variants in the MT‐ATP6 gene is significantly less common in individuals with history of hyperlipidemia (P=7.6×10−4; q = 0.071). These analyses were not adjusted for age or sex because mtDNA variants are maternally inherited at conception and are therefore influenced by neither age nor sex. On the other hand, mtDNA variants are strongly associated with race/ancestry. However, as shown in Figure 4D, the associations are all in the same direction when stratified by maternal ancestry.

Figure 4

Significant associations between rare mitochondrial DNA (mtDNA) variants and clinical features.

A, A quantile‐quantile (QQ) plot of 187 association tests shows 3 (in red) with q values below 0.15, including HxHyperlipidemia being associated with lower rare variant burden (B), HxMI associated with absence of rare NS MT‐COII variants (all 27 rare MT‐COII variants shown appear in individuals without HxMI) (C), and HxHyperlipidemia being associated with absence of rare NS MT‐ATP6 variants (D). Here the P values are computed using a logistic regression model (see Methods section), and q values are computed using the qvalue package in R. Note that (D) shows that the association holds across maternal ancestries, with H, L, and U indicating haplogroup. Disease abbreviations are explained in Table S1. NS indicates nonsynonymous; and OR, odds ratio.

Discussion

Here we have presented results from the first study, to our knowledge, to investigate the relationship between phenotypes of patients with CVD and mtDNA content. Although genome‐wide association studies number in the thousands for the nuclear genome, there are few corresponding studies focusing on the mitochondrial genome.

We were able to expand upon the results of a prior study that showed an inverse association between mtDNA CN and CVD. In our retrospective study, patients were ascertained for clinical histories/features at the time of the sample collection that was used for mtDNA analysis. Although the associations reported here are quite robust, it is impossible from our data to determine whether the lower mtDNA CN portends these events, or is instead the result of their occurrence. Making such a determination would require a prospective longitudinal study of patients with CVD. We also considered single nucleotide variants and their relationship with patient clinical characteristics. MtDNA variants are fairly stable over space and time, and as such, homoplasmic or near‐homoplasmic variants detected in blood are likely to have been present in nearly all of the individual’s tissue, including the heart, since birth. Our examination of the relationship between mtDNA variants and patient characteristics implicated the mitochondria‐encoded genes MT‐ATP6 and MT‐COII. Rare variants in MT‐ATP6 were significantly less common in individuals with a history of hyperlipidemia, and none of the 182 individuals with a history of myocardial infarction harbor were among the 27 individuals with rare variants in MT‐COII. We were initially surprised to find evidence that some rare germline variants may be protective against hyperlipidemia and myocardial infarction. However, from an evolutionary perspective, this is perhaps not so surprising. Because the examined traits generally have late age of onset, any selective advantage conferred by the variants needs not manifest in higher allele frequencies in the population. Furthermore, multiple studies have found protective associations between similar phenotypes and minor allele variants in both mtDNA , and nuclear DNA. In the mitochondrial realm, a Japanese study found a specific mitochondrial haplotype to be protective against myocardial infarction. Another study reported 2 rare variants in MT‐ATP6 as being potentially protective against Leigh syndrome, a genetic disorder associated with hypertrophic cardiomyopathy. Interestingly, one of that study’s reported variants, A8795G (H90R), affects the same amino acid as an MT‐ATP6 variant (C8794T [H90Y]) found in 2 patients in our study.

Here, we have focused on mtDNA CN and germline point substitutions, but a natural extension of our work would be to query whole‐genome sequence for large deletions and low‐heteroplasmy somatic mtDNA mutations. Studies have reported elevated levels of large mtDNA deletions in the heart tissue and blood of patients with coronary artery disease, and such deletions can be detected in whole‐genome sequencing data. Additionally, clonal hematopoiesis, the clonal expansion of hematopoietic stem cells harboring age‐related somatic mutations, has been shown to be associated with CVD. Whether somatic mtDNA mutations also play a role in CVD has not been investigated. Additional studies using deep sequencing of mtDNA in heart tissues and blood would yield further insight into the impact of germline and somatic mutations in mitochondrial function and its role in CVD.

Our study does have certain limitations. Interrogating mtDNA in peripheral blood is, at best, an indirect measure of the mtDNA content in cardiomyocytes, the primary tissue of interest. Although this may be an issue when analyzing CN, it should not be a problem when examining mtDNA sequence content, which should largely be the same in all tissues. Another limitation is a sample size (n=996), which left us underpowered to detect associations with individual rare variants. We also did not interrogate all classes of mtDNA variants.

Despite these limitations, our study design does have several strengths. Our patient cohort comprises a fairly large, well‐annotated collection of disease cases. The use of whole‐genome sequence yields extremely high depth coverage of the mitochondrial genome, which enables comprehensive and sensitive detection of all mtDNA substitutions. Imputation, which is known to be error prone, is unnecessary here. Because all patients were ascertained at the same center, and sequencing protocols were performed in a uniform manner, batch effects are not likely to be an issue, and meta‐analysis is not needed. This overall uniformity is probably one of the reasons that we were able to uncover significant associations in a sample size below 1000.

In summary, we have presented an analysis of mtDNA content in the peripheral blood of patients with CVD. We detected robust associations between the patients’ clinical characteristics and both mtDNA CN and rare mtDNA variants. All significant associations held true across different patient ancestries. Future studies are needed to validate these findings in other cohorts. Incorporating mtDNA analysis of cardiomyocytes would shed light on the implication of lower mtDNA content in the blood of more severely affected individuals.

Sources of Funding

Drs Tang and Hazen have been partially supported by grants from the National Institutes of Health and the Office of Dietary Supplements (R01HL103866, R01DK106000, R01HL126827). Dr LaFramboise has been partially supported by grants from the National Institutes of Health (R01LM013067, R21CA249138). The GeneBank study has been supported by National Institutes of Health grants (P01HL076491, P01HL098055, R01HL103931) and the Cleveland Clinic Clinical Research Unit of the Case Western Reserve University CTSA (UL1TR000439). Whole‐exome sequencing was provided by Human Longevity Inc. (San Diego, CA).

Disclosures

Dr Tang has served as consultant to Sequana Medical AG, and has received honoraria from Springer Nature and American Board of Internal Medicine for authorship/editorship. Dr Hazen is named as co‐inventor on pending and issued patents held by the Cleveland Clinic relating to cardiovascular diagnostics and therapeutics, is a paid consultant for Proctor & Gamble, has received research funds from Proctor & Gamble and Roche Diagnostics, and is eligible to receive royalty payments for inventions or discoveries related to cardiovascular diagnostics or therapeutics from Quest Diagnostics/Cleveland HeartLab or Proctor & Gamble. The remaining authors have no disclosures to report.

Tables S1–S5

Figures S1–S7

Click here for additional data file.