Deleterious mitochondrial heteroplasmies exhibit increased longitudinal change in variant allele fraction

Summary

A common feature of human aging is the acquisition of somatic mutations, and mitochondria are particularly prone to mutation, leading to a state of mitochondrial DNA heteroplasmy. Cross-sectional studies have demonstrated that detection of heteroplasmy increases with participant age, a phenomenon that has been attributed to genetic drift. In this large-scale longitudinal study, we measured heteroplasmy in two prospective cohorts (combined n = 1404) at two time points (mean time between visits, 8.6 years), demonstrating that deleterious heteroplasmies were more likely to increase in variant allele fraction (VAF). We further demonstrated that increase in VAF was associated with increased risk of overall mortality. These results challenge the claim that somatic mtDNA mutations arise mainly due to genetic drift, instead suggesting a role for positive selection for a subset of predicted deleterious mutations at the cellular level, despite a negative impact of these mutations on overall mortality.

Graphical abstract

Highlights

• •Deleterious mitochondrial heteroplasmies preferentially increase with age
• •Heteroplasmies with faster rates of increase are associated with mortality
• •Positive selection may play a role in the rise of somatic mtDNA mutations

Genetics; Cell biology.

Introduction

Mitochondria are involved in essential cellular processes, such as ATP and iron-sulfur cluster synthesis, cell calcium homeostasis, and apoptosis.¹ They possess their own unique genome, with each cell containing hundreds to thousands of mitochondrial DNA (mtDNA) copies^1,2 and mtDNA is highly mutable, with an estimated 2- to 20-fold higher mutation rate than nuclear DNA.^3,4,5 The elevated mutation rate contributes to the high levels of inherited mtDNA diversity and the generation of somatic mtDNA mutations. Inherited mutations in mtDNA can lead to mitochondrial dysfunction and they have been associated with aging-related phenotypes such as longevity,⁶ cancer,⁷ and neurodegenerative diseases.⁸ Somatic mtDNA mutations, acquired throughout an individual’s lifetime, often result in mitochondrial heteroplasmy, a state of mutations being present in a subset of mtDNA molecules within a cell or tissue. The mtDNA bottleneck effect, which causes a random selection of mtDNA during transmission from mother to child, can also contribute to the high levels of somatic heteroplasmies by way of tissue-specific bottlenecks that can occur throughout the lifespan.⁹ These mutations are generally believed to be a product of random genetic drift.^10,11,12,13 However, the presence of mtDNA heteroplasmy, measured cross-sectionally, has been associated with all-cause mortality and cancer in ∼200,000 participants in the UK Biobank,¹⁴ raising the question of whether somatic heteroplasmies are under selection. Cross-sectional studies have observed both purifying selection for a known pathogenic variant with increasing age¹⁵ and positive selection at specific positions in specific tissues,¹⁶ and single-cell sequencing of younger individuals (age < 40 years) found purifying selection against pathogenic variants.¹⁷ Indeed, in the absence of longitudinal studies, our understanding of the dynamics of mtDNA heteroplasmy in the general population remains limited. We, therefore, assessed mitochondrial heteroplasmy in two large population-based studies across time, with two primary objectives: (1) to determine whether somatic heteroplasmy arises solely as a function of genetic drift or exhibits evidence for selection; and (2) to determine whether change in heteroplasmic variant allele fraction (VAF) is associated with health outcomes.

Results

Study population characteristics and heteroplasmy calling

To address these questions, we measured mtDNA heteroplasmy at two time points in DNA derived from blood in participants from two population-based prospective cohorts: the Multi-Ethnic Study of Atherosclerosis (MESA) (n = 1031; median baseline age = 58.5 years; median time between measurements of 9.4 years; buffy coat derived DNA)¹⁸ and the Rotterdam Study (RS) (n = 373; median baseline age = 64.8 years; median time between measurements of 6.4 years; whole blood-derived DNA).¹⁹ Study population characteristics are shown in Table S1. We set an initial heteroplasmy variant allele frequency (VAF) threshold of 5%–95%, chosen a priori to maximize sensitivity to true heteroplasmies, while minimizing false-positives due to nuclear-encoded mitochondrial sequences (NUMTs).¹⁴ When a heteroplasmy was only detected (VAF 5%–95%) at a single visit, we determined whether a heteroplasmy was present below the threshold in the alternate visit (see STAR Methods), and used the observed VAF if > 1%, or set it to a minimum VAF = 1%. Detected heteroplasmies at the first visit that dropped below the detection threshold at the subsequent visit were classified as “lost”, and correspondingly, when the heteroplasmy was detected in the second visit, but below the threshold in the first visit, it was classified as “de novo”. This criterion held even in cases of variants with low read depth, ensuring a reliable assessment of heteroplasmy changes between visits. The annualized rate of change in VAF (deltaVAF), which captures the normalized difference in VAF between visits, was used as the main metric in our analyses (formula in STAR Methods). If the absolute value of deltaVAF was < 0.05, we considered the heteroplasmy to be stable.

Heteroplasmy prevalence

In MESA, we observed 388 participants with heteroplasmies at either visit; 331 (32.1%) and 358 (34.7%) at the initial and follow-up visits, respectively. Along with the increase in heteroplasmic individuals, we also observed an increase in participants harboring >1 heteroplasmy at the follow-up visit (81 vs. 58 in 1st visit). In RS, we observed 152 participants with heteroplasmies at either visit; 129 (34.6%) and 142 (38.1%) at the initial and follow-up visits, respectively. In RS, we observed a small increase in the number of participants with > 1 heteroplasmy in the follow-up visit (33 vs. 31 in 1st visit). While VAF of heteroplasmic variants was highly correlated across visits (MESA Pearson’s r² = 0.80, RS Pearson’s r² = 0.79) (Figures 1A and 1B), there was a significant upward shift in VAF in the later visits (MESA p < 0.002, RS p < 0.0003) (Figures 1C and 1D). Of the 110 de novo heteroplasmies observed in MESA and 39 in RS, none was completely de novo (with the site with the lowest number of alternate reads detected in the first visit BAM/CRAM file across both MESA and RS having 12 alternate reads, which is unlikely to be explained by sequencing error [Figure S1]), indicating that the change in heteroplasmy count is largely explained by a change in VAF among mutations that already existed prior to the initial visit.

Figure 1.

Change in variant allele fraction (VAF) over time

(A and B) Correlation between VAF at the first measurement (x axis) and the second measurement (y axis), with the black line indicating perfect correlation, in MESA (top) and RS (bottom).

(C and D) Distribution of the change in VAF (deltaVAF) in MESA (top) and RS (bottom).

Role of selection in VAF changes

To determine whether selection was playing a role in the increased VAF observed with aging, we first stratified heteroplasmies by functional categories (Figures 2A and 2B). Synonymous and D loop heteroplasmies were combined as the reference, as neither category was associated with deltaVAF, and both had comparable null effect estimates (data not shown). Compared to synonymous and D loop mutations, in MESA, we observed a significant increase in deltaVAF in missense mutations (p < 0.003) and RNA (p < 0.0006) heteroplasmies (Figure S2A). Despite a smaller sample size, consistent effect estimates were observed in RS for missense heteroplasmies (p < 0.028), and a meta-analysis across both cohorts showed significantly higher deltaVAF for both missense (p < 0.0003) and combined rRNA/tRNA (RNA) (p < 0.001) heteroplasmies relative to synonymous and D loop variants, suggesting non-random expansion of mitochondrial heteroplasmies (Figure S2A). To further test whether functional heteroplasmies were increasing in VAF, we made use of a modified version of a recently developed score (mMLC score, see STAR Methods) that uses local constraint to estimate the functional consequences of mitochondrial heteroplasmies. The mMLC ranges from 0 to 1, with 0 indicating a benign mutation and 1 corresponding to a nonsense mutation that completely abrogates protein function,¹⁶ and higher scores have been associated with cancer risk and overall mortality.^3,17 Cross-sectionally, we found that higher baseline age was associated with higher mMLC (MESA p < 0.014, RS p < 0.030), consistent with the hypothesis that age-acquired (i.e., somatic) heteroplasmies are under relaxed selection compared to inherited variants. Longitudinally, in both MESA (p < 6.2 × 10⁻⁷) and RS (p < 0.034), we saw a significantly higher deltaVAF associated with higher mMLC score (Figures 2C and 2D), indicating that predicted deleterious variants are more likely to increase in VAF with aging (meta-analysis p < 5.9 × 10⁻⁸, Figure S2B). When putting both functional categories and mMLC into the same regression model, only mMLC was significantly associated with deltaVAF (MESA p < 0.002, RS p = 0.11, META p < 0.00045), indicating that predicted deleteriousness is driving the association of missense and RNA heteroplasmies with deltaVAF (Figure S2C). To address potential confounding due to non-driver (i.e., passenger) mutations, we also limited analyses to participants with only a single heteroplasmy (i.e., by definition do not have mitochondrial DNA passenger mutations), or by choosing a single heteroplasmy per participant based on the highest MLC score, and obtained similar results (Figures S3 and S4). To address potential confounding due to demographic variables, we assessed the association of age, sex, self-identified race, DNA collection center, and smoking status with deltaVAF, and found none was associated (all p > 0.05) in MESA or RS. To address potential confounding due to nuclear variants, clonal hematopoiesis of indeterminate potential (CHIP) and mosaic chromosomal abnormalities (mCA) were called in the MESA samples from whole genome sequencing (WGS) data as previously described.^20,21 Of the 387 individuals in MESA with heteroplasmy, 27 have CHIP (1 sample was missing CHIP status information) at the baseline visit. There were 47 mCA carriers among 372 MESA participants with heteroplasmy (16 were missing mCA status information). When we excluded individuals with CHIP at baseline, we observed a stronger association between mMLC and deltaVAF (beta = 0.104 vs. 0.079, p = 3.9 × 10⁻⁹ vs. 6.5 × 10⁻⁷). Similarly, excluding baseline mCA carriers did not significantly change the association between mMLC and deltaVAF (beta = 0.091 vs. 0.086).

Figure 2.

DeltaVAF by functional consequence

(A and B) Density plots of deltaVAF stratified by mutation consequences in MESA (top) and RS (bottom).

(C and D) Association between the mMLC score and deltaVAF, with a blue line with confidence intervals illustrating the regression slope in MESA (top) and RS (bottom). RNA = combined tRNA and rRNA.

Association of deltaVAF with overall mortality

Previous work has shown that mitochondrial heteroplasmy burden, measured as the MLC score sum (MSS) across all heteroplasmies in a given individual, is associated with an increased risk of overall mortality.¹⁴ To determine if the change in VAF is likewise associated with overall mortality, for each individual we selected the variant with the largest increase in deltaVAF, and assessed the association with overall mortality (Figure 3). In MESA, there were 53 deaths out of 385 samples with follow-up mortality data, with a median follow-up time of 8.1 years (Q1 = 7.8, Q3 = 8.4). After adjusting for age, sex, collection center, race, and smoking status, a 0.10 increase in deltaVAF was associated with a 1.48-fold (95% confidence interval (CI) 1.06–2.07, p < 0.02) increased risk of overall mortality. Additionally adjusting for mMSS calculated at the follow-up visit only modestly attenuated the effect (HR = 1.39, 95% CI 0.96–2.00). In the RS, there were 51 events among 152 individuals followed for up to 10 years. Highly concordant effect sizes were observed in RS (HR = 1.39, 95% CI 1.05–1.85, p < 0.023), again with only modest attenuation when including MSS into the model (HR = 1.31, 95% CI 0.96–1.79). A meta-analysis demonstrates a highly significant association between deltaVAF and mortality (HR = 1.43, 95% CI 1.15–1.77, p < 0.001).

Figure 3.

DeltaVAF by vital status

Density plots of deltaVAF by vital status in (A) MESA and (B) RS.

Discussion

In this large-scale study of longitudinal mtDNA heteroplasmy measurements in humans from two prospective cohorts, we observed that heteroplasmies are largely stable over time (< 0.05 change in deltaVAF, MESA = 325/504 [64.4%]; RS = 131/211 [62.1%]) (Figures S5A and S5B). However, overall, there is an increase in VAF, largely driven by variants below detection (5% VAF) at the initial visit (MESA = 90/120 [75.0%], RS = 32/59 [54.2%]) (Figures S5C and S5D; Table S2). Variants with large deltaVAF are more likely to be deleterious and are associated with overall mortality. To confirm that these results are not confounded by de novo mutations in our dataset having different mMLC score distributions than extant variants, we re-called heteroplasmies with a VAF cut-off of 2%, and excluded de novo variants (i.e., we only examined heteroplasmies that were detectable at baseline). In both MESA (p < 4.42 × 10⁻¹⁴) and RS (p < 2.79 × 10⁻⁴), higher mMLC score of extant heteroplasmies was associated with greater increase in VAF. These results suggest that selection, where heteroplasmies that are deleterious for the organism (human) nevertheless have a survival advantage leading to clonal expansion, may play a role in mitochondrial heteroplasmy, challenging the prevailing notion that acquired mitochondrial heteroplasmy arises mainly due to random genetic drift. While potentially counter-intuitive that predicted deleterious variants may display a selective advantage at the cellular level, this is consistent with recent results demonstrating that these variants are associated with increased risk of hematological cancers.^14,22 Further, functional studies have demonstrated the effect on cellular fitness of a nonsynonymous mitochondrial mutation is dependent on the cellular environment, and can range from harmful to beneficial.²³ An alternative explanation for the higher deltaVAF of more deleterious heteroplasmies could be that they are passenger mutations indicative of the presence of nuclear mutations that lead to clonal expansion. While removing individuals with detectable CHIP and mCA did not appreciably attenuate our results, we could not rule out other potential nuclear drivers of CH. Future functional studies will be crucial to distinguish between these two hypotheses.

Strengths of the current study include the demonstration that results are robust to potential technical and demographic confounders, and are highly concordant across two large prospective population-based cohorts with longitudinal measures of mitochondrial heteroplasmy. Further, MESA included a majority of non-European ancestry individuals (∼59%), while RS was almost entirely European ancestry, suggesting generalizability of results across genetic ancestry groups.

The finding that predicted deleterious mitochondrial heteroplasmies are more likely to increase in VAF over time compared to benign variants has important implications in light of the accumulating evidence that these variants are associated with aging-related disease, including cancer.^14,22 Our results demonstrating only modest attenuation of the association of deltaVAF with mortality when adjusting for the mMSS, suggest that monitoring the deltaVAF provides important information not captured by the predicted deleteriousness of the heteroplasmy, and may further aid in identifying individuals at high-risk for developing disease. Second, if the higher deltaVAF for predicted deleterious heteroplasmies is indeed indicative of selection for some of these deleterious variants, understanding the biological mechanisms that lead to selection for these variants may provide important insights into disease mechanisms and identify novel interventional targets to prevent or delay progression of disease.

Limitations of the study

As limitations, we note that despite being collectively common, with ∼1/3 of MESA and RS participants carrying 1 or more heteroplasmies, the vast majority of individual heteroplasmies are exceedingly rare, with ∼90% observed at frequencies of <1/10,000.^14,22 Thus, it is not surprising that, despite an overall trend toward positive selection, specific deleterious variants may be selected against, as has been observed for m.3243A>G.¹⁵ Second, both MESA and RS only included individuals > 40 years of age, precluding assessment of changes in heteroplasmy VAF during development and/or early life. Third, while prior studies demonstrated strong tissue-related heteroplasmic variants,¹⁶ we assessed bulk DNA isolated from either buffy coat or whole blood, and thus our results reflect average changes across cell types that contribute to these tissues. Thus, we could not assess the potential contribution of immune cell expansion to our results. We did not observe mtDNA heteroplasmic hotspots with high deltaVAF in our study (maximum mean deltaVAF < 0.050 for heteroplasmies appearing > 3 times, Table S3), and the study of longitudinal mtDNA heteroplasmy across tissues in the same individual would be important to shed light on the differences in mtDNA heteroplasmy dynamics across tissues. Finally, we only measured mtDNA heteroplasmy at two time points, not allowing for the observation of any non-linear changes in VAF, and the ∼3 years shorter follow-up time in RS that likely contributed to the proportionally fewer de novo variants detected than in MESA (18% vs. 22%) (Table S2).

Resource availability

Lead contact

Further information and requests should be directed to and will be fulfilled by the lead contact, Dan E. Arking (arking@jhmi.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Multi-Ethnic Study of Atherosclerosis: MESA WGS sequencing data are available through dbGaP accession number phs001416.v3.p1. Twist mtDNA sequencing data have been deposited with the parent study, and is available upon request from the study coordinating center (https://www.mesa-nhlbi.org/). Due to restrictions based on privacy regulations and informed consent of the participants, data cannot be made freely available in a public repository. Rotterdam Study: Rotterdam Study data can be obtained upon request. Requests should be directed toward the management team of the Rotterdam Study (secretariat.epi@erasmusmc.nl), which has a protocol for approving data requests. Due to restrictions based on privacy regulations and informed consent of the participants, data cannot be made freely available in a public repository.
• •Mitochondrial genetic variation was called using the MitoHPC pipeline: https://github.com/ArkingLab/MitoHPC. Genetic ancestry was called in the Rotterdam Study using ADMIXTURE: https://dalexander.github.io/admixture/index.html.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work is supported by the National Heart, Lung, and Blood Institute (NHLBI) (R01HL1445609, R01HL131573) and National Institute on Aging (P30AG021334).

Whole genome sequencing (WGS) for the Trans-Omics in Precision Medicine (TOPMed) program was supported by the National Heart, Lung, and Blood Institute (NHLBI). WGS for “NHLBI TOPMed: Multi-Ethnic Study of Atherosclerosis (MESA)” (phs001416.v3.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 QC, and general study coordination, were provided by the TOPMed Data Coordinating Center (3R01HL-120393-02S1), and TOPMed MESA Multi-Omics (HHSN2682015000031/HSN26800004). The MESA projects are conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with MESA investigators. Support for the Multi-Ethnic Study of Atherosclerosis (MESA) projects are conducted and supported by the National Heart, Lung, and Blood Institute (NHLBI) in collaboration with MESA investigators. Support for MESA is provided by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 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, DK063491, and R01HL105756. The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutes can be found at http://www.mesa-nhlbi.org.

The Rotterdam study is funded by Erasmus Medical Center and Erasmus University, Rotterdam, Netherlands Organization for the Health Research and Development (ZonMw), the Research Institute for Diseases in the Elderly (RIDE), the Ministry of Education, Culture and Science, the Ministry for Health, Welfare and Sports, the European Commission (DG XII), and the Municipality of Rotterdam. The current study was supported by VOILA (ZonMW 457001001). The authors are grateful to the study participants, the staff from the Rotterdam study and the participating general practitioners and pharmacists.

Author contributions

Conceptualization, J.B.J.v.M. and D.E.A.; methodology, L.M.K., J.S.B, E.G., and D.E.A.; software, W.S. and D.P.; formal analysis, L.M.K., W.S., J.A.M.V., P.A., J.X., and D.E.A.; investigation, L.M.K, W.S., J.A.V., Y.S.H., L.B., J.X., C.N., J.B.J.v.M., and D.E.A; writing – original draft, L.M.K. and D.E.A.; writing – review and editing, W.S., J.A.M.V., Y.S.H., P.A., D.P., L.B., J.X., C.N., S.S.R., K.D.T., J.I.R., J.S.B., E.G., and J.B.J.v.M.; funding acquisition, S.S.R., K.D.T., J.I.R., J.S.B., E.G., J.B.J.v.M., and D.E.A.; supervision, J.B.J.v.M. and D.E.A; project administration, C.N.

Declaration of interests

J.S.B. is an equity holder and advisor of Opentrons Labworks Inc and of Dextera Biosciences Inc. S.S.R. is a consultant for Westat, the administrative coordinating center for the NHLBI Trans-Omics for Precision Medicine (TOPMed) program.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
MESA DNA samples	MESA Coordinating Center	Not applicable
Rotterdam Study DNA Samples	Rotterdam Study Coordinating Center	Not applicable
Critical commercial assays
TWIST Mitochondrial Panel	Twist Bioscience	102040
Library Preparation Enzymatic Fragmentation (EF) Kit 2.0	Twist Bioscience	104207
Twist HT Universal Adapter System	Twist Bioscience	106390 … 106408
NovaSeq 6000 SP Reagent Kit v1.5 (300 cycles)	Illumina Inc.	20028400
High Sensitivity Large Fragment 50kb kit	Agilent	# DNF-464-0500
Deposited data
MESA WGS data	dbGaP	phs001416.v3.p1
MESA Twist sequencing data	deposited with MESA coordinating center (https://www.mesa-nhlbi.org/)	N/A
Software and algorithms
MitoHPC pipeline	https://github.com/ArkingLab/MitoHPC	N/A
ADMIXTURE	https://dalexander.github.io/admixture/index.html	N/A
R (version 4.4.1)	https://cran.r-project.org/src/base/R-4/	N/A
stats (version 4.4.1)	https://www.R-project.org/	N/A
lme4 (version 1.1.35.5)	https://github.com/lme4/lme4/	N/A
ggplot2 (version 3.5.1)	https://ggplot2.tidyverse.org	N/A
splines (version (4.4.1)	https://www.R-project.org/	N/A
survival (version 3.7.0)	https://doi.org/10.18637/jss.v036.i03	N/A
metafor (version 4.6.0)	https://doi.org/10.18637/jss.v036.i03	N/A

Experimental model and study participant details

Multi-Ethnic Study of Atherosclerosis

MESA is a prospective cohort of 6,452 individuals aged 45–84 who were free of clinical cardiovascular disease at time of recruitment from six communities across the United States. Participants had examinations at field centers with initial examination in 2000–2002 and exam 5 in 2010–2011.¹⁸ Information collected at exams includes age, sex, race, weight, height, blood pressure, smoking status, medication, lifestyle, family history, and medical history. Blood is drawn from participants at each exam and DNA is extracted from the buffy coat of the first examination and the fifth examination (N = 4475). Annual follow-up were conducted through 2019 with mail, electronic mail, and telephone interviews; classification of cardiovascular events and mortality were collected from death certificates, hospital records, and interviews with physicians, relatives, or friends.¹⁸ For the current study, 1105 participants were selected randomly from participants with DNA available from visits 1 and 5 and who did not have a cardiovascular event prior to visit 5. Descriptive statistics of the study population are available in Table S1.

Studies involving human participants were reviewed and approved by all field centers and core labs of the MESA study including: Columbia University, New York; Johns Hopkins University, Baltimore; Northwestern University, Chicago; UCLA, Los Angeles; University of Minnesota, Twin Cities; Wake Forest University, Winston Salem; University of Washington Coordinating Center; University of Vermont Laboratory; UCLA Medical Center Research and Education Institute; New England Medical Center; The Lundquist Institute for Biomedical Innovation at Harbor-UCLA Medical Center; University of Virginia; Broad Institute of MIT and Harvard; National Heart, Lung, and Blood Institute. The patients/participants provided their written informed consent to participate in this study.

Rotterdam Study

The Rotterdam Study is a population-based cohort with four sub-cohorts located in Ommoord, a suburb of Rotterdam, the Netherlands.¹⁹ The participants of the first sub-cohort (RS-I) consisted of individuals aged ≥ 55 years (n = 7983, response rate 78%) and they visited the research center for the first time between 1990 and 1993. In 2000–2001, individuals who had reached 55 years of age or moved into the study area (n = 3011, response rate 67%) were invited to join the study and the second sub-cohort RS-II was initiated. Between 2006 and 2008, a third sub-cohort (RS-III) was established enrolling new individuals aged ≥ 45 years (n = 3932, response rate 65%). The fourth cohort started between 2016 and 2018, consisting of all residents of Ommoord aged 40 years and over who had not been invited previously (n = 3005, response rate 45%). Participant information was collected every 3–5 years through home interviews, questionnaires, and examinations at our dedicated research center located in the Ommoord district. The home interviews collected information such as sex, smoking status, and medical history. Blood was drawn when participants visited the research center, and DNA was extracted from whole blood. Information on the vital status of participants was obtained biweekly via municipal population registries and databases of general practitioners and hospitals.¹⁹ We determined genetic ancestry based on information from all participants of the RS cohort with available genetic information. The cleaned genotype data from all samples was merged with HapMap CEU release 22 (build 36)²⁴ (to provide a backbone population set). The genotype data were pruned to only keep variants in linkage equilibrium and then analyzed using ADMIXTURE (default parameters used).²⁵ We performed cross-validation for 1 to 8 ancestral populations recognized in HapMap. Finally, ancestral groups were delineated based on containing at least 50% genetic material from a specific ancestral group. In participants where genetic ancestry information was unavailable, we assigned them to an ancestry if at least 3 of their grandparents were born in a region. For the current study, 392 participants from the first (RS-I-1) and third (RS-I-3) visit of the first subcohort of the Rotterdam Study with DNA availability for mtDNA sequencing from two time points were included. The samples were randomly selected from the full cohort. Descriptive statistics of the study population are available in Table S1.

The Rotterdam Study has been approved by the Medical Ethics Committee of the Erasmus MC (registration number MEC 02.1015) and by the Dutch Ministry of Health, Welfare and Sport (Population Screening Act WBO, license number 1071272-159521-PG). The Rotterdam Study Personal Registration Data collection is filed with the Erasmus MC Data Protection Officer under registration number EMC1712001. The Rotterdam Study has been entered into the Dutch Trial Register (NTR; https://onderzoekmetmensen.nl) and into the WHO International Clinical Trials Registry Platform (ICTRP https://www.who.int/clinical-trials-registry-platform, search portal https://trialsearch.who.int/) under shared catalog number NL6645 / NTR6831. The Rotterdam Study project persistant identifier is https://ror.org/02ac58f22. All participants provided written informed consent to participate in the study and to have their information obtained from treating physicians.

The current study was approved by the The Johns Hopkins Medicine Institutional Review Board.

Method details

WGS data

For MESA, WGS data from Trans-Omics for Precision Medicine (TOPMed) Freeze 9 data release was available for all first visit samples. The study provides ∼30X coverage WGS data sequenced using Illumina’s next-generation sequencing technology and passed all quality control metrics.²⁶ MESA samples were sequenced in one batch at a single sequencing center (TOPMed phase 2). WGS for the MESA samples was generated by the Broad Institute using an Illumina HiSeq X Ten. Average read length was 151 bp, mean mtDNA coverage of 4978 and median was 4727.

mtDNA sequencing

Rotterdam Study

The Twist Library Preparation EF 2.0 protocol was performed on a PerkinElmer Sciclone robot with 100 ng DNA and 7 PCR cycles. In total, 1536 different indexes were used. After the library preparation, the DNA concentrations were measured with nanodrop and 16 samples with 200 ng DNA of each sample were pooled for the hybridization step. The pooled DNA was dried using vacuum pressure. To each dried sample, 4 μl of the cmtDNA probe panel (Twist) was added and hybridized in a PCR machine at 70°C for 17 h. The capture of the mtDNA was performed on a PerkinElmer robot with the Twist Target enrichment standard hybridization v2 protocol. For each pool of samples, 15 PCR cycles were performed. DNA concentrations were measured with a fluorescence-based assay and size distribution was checked with the Tapestation system. Sequencing was performed at 2 × 150 bp on an Illumina Novaseq6000 system at up to 10,000x per sample. Samples were sequenced in two batches, with samples from both visits always sequenced in the same batch.

MESA

The Twist Library Preparation EF 2.0 protocol was performed manually on 50 ng DNA per sample with enzymatic fragmentation at 37°C for 20 min. Samples were amplified for 8 PCR cycles. In total 384 different indexes were used. After the library preparation, the DNA concentration for 25% of the samples was measured with Qubit dsDNA BR assay. The concentrations were then averaged to determine the volume used from each sample to pool approximately 200 ng of DNA in sets of 8. To each pooled set of samples, 4 μl of the Twist mtDNA probe panel was added along with universal blockers and blocking solution, and then dried using vacuum pressure. The dried samples were then stored overnight at −20C. The following day, the capture of the mtDNA was performed manually following the Twist Target Enrichment Fast Hybridization protocol. Samples were hybridized in a thermocycler at 60C for 2 h. For each pool of samples 15 PCR cycles were performed. DNA concentrations of the completed libraries were measured with Qubit dsDNA HS assay and size distribution was checked with the Agilent 5200 Fragment Analyzer and the High Sensitivity Large Fragment 50kb kit by the Johns Hopkins Single Cell and Transcriptomics Core. Completed libraries were pooled together at 4 nM, and sequencing was performed at 2 × 150 bp on the Illumina NovaSeq 6000 system in a single lane of a NovaSeq SP Flowcell by the Johns Hopkins Genetics Resources Core Facility High Throughput Sequencing Center. This protocol was used to sequence all follow-up visit samples, with samples sequenced in 3 batches. In addition, 406/1031 first visit MESA samples were sequenced to confirm the comparability of WGS and Twist sequencing. Across 183 heteroplasmies called by either WGS or Twist, the correlation of VAF was high (Pearson’s r² = 0.91), with no excess of de novo variants seen in the Twist data relative to the WGS data (Figure S6; Table S4).

Detection of mtDNA heteroplasmy

MitoHPC (20240306 version) was used for mtDNA heteroplasmy detection from WGS and Twist sequencing data.²⁷ MitoHPC utilizes GATK Mutect2²⁸ twice for variant identification and SAMtools²⁹ for generating read count and coverage used for mtDNA-CN calculation. Prior to analyses, we downsampled each sample to 2000X at each mitochondrial genome position using the samtools view command. We implemented a heteroplasmy allele frequency cut-off of 5%, where variants with allele frequency 5%–95% were considered heteroplasmic, and <5% or >95% homoplasmic. Documentation for MitoHPC can be found at: https://github.com/ArkingLab/MitoHPC. We only included variants at sites with minimal read depth in both visits of ≥300 to avoid false-negative calls. We also removed INDELs and variants in the polyC homopolymer regions, and any variants with ‘base|slippage|weak|position|strand’ flag in the FILTER column of the output VCF. For multi-allelic heteroplasmy at a single position, the heteroplasmy with higher MLC score is retained. When a heteroplasmy was initially detected at only a single visit (VAF≥0.05), we determined whether a heteroplasmy was present below the detection threshold in the corresponding visit by running MitoHPC with the freebayes³⁰ variant detector, which has higher sensitivity than Mutect2,³¹ and used the observed VAF if > 0.01, or set it to a minimum VAF = 0.01. We excluded a group of heteroplasmies that occurred together at similar VAF (positions 3097, 3098, and 3100), and were seen only in visit 5 in 4 MESA participants. These variants were not seen in ∼180,000 TOPMed samples nor in ∼490,000 UK Biobank samples, suggesting that they are false-positives introduced during TWIST sequencing. We also excluded position 319, which was “lost” at follow-up in 4 samples, and not seen in a Twist baseline sample despite being observed in the matching WGS data, suggesting a systematic false-negative in Twist sequencing. In the RS, we excluded a group of heteroplasmies that occurred together at similar VAF (positions 12684, 12705) that were only seen in the first visit, present in the BAM files but not detected by MitoHPC, pointing toward possible contamination.

Sample QC

In MESA, we removed 69 samples that had DNA isolated from visit 1 using an alternate protocol than the rest of the MESA samples which was associated with significantly lower mtDNA-CN and a significant excess of heteroplasmy calls. Based on prior analyses in ∼200,000 samples from the UK Biobank,¹⁴ we removed individuals with >5 heteroplasmies (MESA, N = 3; RS, N = 0). We removed 2 additional samples from MESA, 1 with homoplasmic variants mismatch between initial and follow-up visit, and 1 identified as likely contaminated by MitoHPC (Haplocheck contamination score >0.03).^27,32 In RS, we remove 18 samples with haplogroup mismatches between visits. Furthermore, we excluded 1 additional sample with 5 heteroplasmic variants that were all present in the first visit but in the second visit only seen in the BAM files, pointing toward possible contamination.

Quantification and statistical analysis

Calculation of change in heteroplasmy variant allele fraction (deltaVAF)

Given that the VAF is for a specific allele, to avoid a potential bias due to using the rCRS reference, in instances where the VAF in the first visit exceeded 0.8 or when the VAF was between 0.5 and 0.8 and a homoplasmy on that position was detected in more than 2 UK Biobank participants (based on MitoHPC calls in ∼490,000 WGS samples),¹⁴ we used the complement of the VAF (1 - original VAF) in both visits as the VAF in our analyses. This approach was taken as a high initial VAF likely reflects heteroplasmy of the alternate allele. Subsequently, we calculated the change in heteroplasmy variant allele fraction (deltaVAF) as: $\frac{\mathit{log}\left(\frac{\left(VAFvisit2\right)/\left(1-VAFvisit2\right)}{\left(VAFvisit1\right)/\left(1-VAFvisit1\right)}\right)}{timeinyearsbetweenvisits}$.

Association of deltaVAF with demographic variables

To assess whether deltaVAF was associated with demographic variables (self-identified race, age, sex, DNA collection center, and smoking status) we fitted linear mixed models using the R-package lme4, allowing for multiple heteroplasmies per individual. Specifically, sample ID was included as a random effect, and models were compared with and without the demographic variable using a likelihood ratio test. p < 0.05 was considered significant. Age was incorporated with a natural spline with 2 degrees of freedom.

Association of deltaVAF with predicted variant function

We fitted generalized linear models using the R-package glm to determine the associations of the consequence of the mutation and the MLC score³³ with deltaVAF. The MLC score, a measure of local tolerance to base or amino acid substitutions, is calculated for every possible mtDNA single nucleotide variant (SNV), as previously described in detail.³³ Briefly, for each base, a 30 bp window is used to calculate the observed:expected (oe) ratio of substitutions and its 90% confidence interval (CI) using the gnomAD database.³⁴ The window start position is then shifted by 1 bp, and this process is repeated to capture all 16,569 positions. For protein-coding genes, calculations are limited to missense variants, under the assumption that synonymous changes are tolerated, while for all other positions (e.g., D loop, tRNA, rRNA), all base substitutions are used. For each position in the mtDNA, the mean of oe ratio 90% CI upper bound fraction (OEUF) is calculated using all overlapping windows, and then percentile ranked to generate a positional score between 0 and 1, where 1 is most constrained and 0 is least constrained. An MLC score is assigned for every mtDNA SNV as follows: non-coding, RNA, and missense variants are assigned their positional score; synonymous, stop gain, and start/stop lost being assigned scores of 0.0, 1.0, and 0.70, respectively, which reflect scores based on the OEUF value of the variant class. Note that MESA had a single nonsense heteroplasmy, which was included in the missense category for analyses. Additionally, the original MLC score, which was developed solely on heteorplasmies, was modified to account for homoplasmies (mMLC), under the assumption that common homoplasmies are unlikely to be deleterious, as follows:

$$\text{mMLC}=\text{MLC}/(\log 10(\text{homoplasmy}\_\text{count}+1)+1)\text{.}$$

Homoplasmy count was derived from ∼490,000 UK Biobank participants called with MitoHPC.^14,22 In both cohorts, we also included a covariate for sequencing batch. We performed sensitivity analyses analyzing the subset of individuals with only a single heteroplasmy, or by including only a single heteroplasmy per participant selected based on the highest mMLC score. All results were robust to potential outliers, demonstrated by negligible changes when analyses were run with rank-transformed deltaVAF (data not shown). Similarly, exclusion of heteroplasmies in regions corresponding to NUMT sequences, as annotated by MitoHPC,²⁷ did not significantly alter results (Figure S7). Meta-analyses were performed using fixed-effects inverse variance weighting as implemented in the R-package metafor.

Association of deltaVAF with mortality

We performed Cox Proportional Hazards models using the R survival package to determine the association between deltaVAF and all-cause mortality, with follow-up time starting from the time of second DNA collection. A single heteroplasmy, selected based on the largest deltaVAF, was chosen per individual to minimize potential confounding due to non-driver mutations. Analyses were adjusted for age using a natural spline with 4 degrees of freedom, sex, race, and smoking status (current, previous, never). DeltaVAF was multiplied by a factor of 10 so that hazard ratios are reported per 0.10 change in deltaVAF. Analyses for MESA also included DNA collection center and sequencing batch. Analyses in RS were truncated ten years after time of second DNA collection to enhance comparability to MESA. Secondary analyses also included mMSS calculated at the second visit from variants with VAF ≥5% to assess whether the association of deltaVAF with mortality provides additional prognostic information beyond the overall heteroplasmic burden. We performed sensitivity analyses in which we used the rank-transformed deltaVAF, demonstrating that results were not driven by extreme values (Table S5). Meta-analyses were performed as described above. All p values were calculated using a two-sided test.

Statistics and reproducibility

Participant selection from MESA and RS are described in the Study populations section above. No statistical method was used to predetermine the sample size. The Wilcoxon signed-rank test with a two-wide exact p value was used to determine the significance of the right skew of deltaVAF.

Published: May 6, 2025