Long-read MitoScope reveals tissue-resolved somatic mitochondrial variation and landscape of nuclear-embedded mitochondrial sequences

Mitochondria play a central role in cellular energy production, metabolic homeostasis, and stress-responsive signaling.1,2 Their dysfunction contributes to diverse human diseases ranging from neurodegeneration and cancer to cardiometabolic syndromes.3–7 The human mitochondrial genome (mtDNA) is a 16.5 kilobase circular molecule, encoding 13 oxidative phosphorylation proteins, 22 tRNAs, and 2 rRNAs.8 Unlike the nuclear genome, mtDNA exists at tens to thousands of copies per cell, where its high mutation rate and multi-copy organization generate extensive intra-cellular and intra-tissue variation (heteroplasmy) that is biologically important and technically challenging to resolve. In most tissues, wild-type and mutant mtDNA molecules coexist, and shifts in heteroplasmy levels can modulate oxidative phosphorylation, influence cellular fitness, and contribute to late-onset phenotypes even in clinically normal donors.3 Somatic mtDNA mutations are expected to accumulate in a tissue- and age-dependent manner,5 yet their prevalence and spectrum in healthy human tissues remain incompletely defined. Beyond mitochondrial genomes, nuclear mitochondrial DNA segments (NUMTs) embedded in the nuclear genome can reshape local chromatin, perturb gene regulation, and confound interpretation of both nuclear and mitochondrial variants in rare disease and cancer genomics.9–11 NUMT integrations have been linked to genomic instability, clonal evolution, and age-associated mitochondrial dysfunction,12 yet most current knowledge derives from germline or population-level catalogs,9 rather than tissue-resolved somatic events. Consequently, the extent, dynamics, and functional impact of NUMT mosaicism across human tissues remain largely undefined, limiting the ability to distinguish benign background integrations from potentially pathogenic events and to recognize NUMT-associated instability as a feature of somatic mosaicism.13,14 Traditional short-read sequencing struggles to resolve the full spectrum of mtDNA variation in the presence of NUMTs as short fragments cannot distinguish reads arising from the mitochondrial versus the nuclear genome. NUMTs can range in size from <50 bp to the full length of mtDNA while sharing high sequence identity with mtDNA.9 With short reads, this often leads to mapping artifacts, misassemblies, false heteroplasmies, and incomplete catalogs of both mtDNA variants and NUMTs. Despite short-read-based surveys of large cohorts having cataloged germline mtDNA polymorphisms and thousands of NUMTs,9 somatic mtDNA single-nucleotide variants (SNVs), indels (insertions/deletions), structural variants, and tissue-mosaic NUMT diversity are underestimated, particularly at low heteroplasmy frequencies relevant for somatic mosaicism. Long-read sequencing technologies can capture full-length circular mtDNA molecules, traverse repetitive segments, and distinguish mtDNA from NUMTs using genome alignment and epigenetic signatures, due to higher 5mC levels in nuclear versus mitochondrial DNA. Most mitochondrial analysis tools are optimized for short-read data,15 and existing long-read tools, like Mitorsaw and Himito,16,17 are restricted in platform support, throughput, or scalability, limiting their use in cohort-scale, tissue-resolved analyses of somatic mtDNA and NUMTs. As a result, there is a critical need for a robust, scalable long-read workflow that can jointly resolve mtDNA copy number (CN), heteroplasmic SNVs and structural variants, and NUMTs across diverse tissues and donors. Here we present MitoScope, a tool for long-read-specific mitochondrial analysis with stringent selection of mtDNA reads, high-fidelity heteroplasmic variant calling, circular mtDNA assembly, comprehensive quality control, and systematic NUMT profiling. Applied to benchmarking HapMap cell mixtures and tissues from the Somatic Mosaicism across Human Tissues (SMaHT) Network,18–20 MitoScope achieves high sensitivity and precision, reveals tissue-specific mtDNA CN, uncovers low-frequency pathogenic heteroplasmies, defines age-associated major-arc deletions enriched for microhomology, and identifies >2-fold additional NUMTs compared with short-read catalogs—exposing pervasive, tissue-resolved somatic mtDNA and NUMT instability in healthy human tissues.

MitoScope: A scalable long-read workflow for mtDNA variant detection and NUMT profiling MitoScope is a long-read-specific workflow for comprehensive mitochondrial genome analysis and comprises five key modules: 1) mtDNA selection, 2) variant calling and annotation, 3) assembly generation, 4) quality control, and 5) optional NUMT profiling (Figure S1). It accepts long-read sequencing data in the form of FASTQ, BAM, or CRAM files either raw or aligned, providing flexibility in preprocessing. From raw sequencing reads or chrM-mapped reads from aligned input, MitoScope initiates alignment-independent, k-mer-based extraction of candidate mtDNA reads. Following k-mer selection, reads are aligned to the mitochondrial genome and filtered using alignment and methylation signatures to retain mtDNA reads and remove any remaining contaminating NUMTs. mtDNA is known to exhibit very low levels of 5mC methylation relative to the nuclear genome,21 providing a useful secondary method for differentiation in addition to alignment. Reads that do not fully align to the mitochondrial genome or display elevated 5mC methylation are discarded (Methods). For variant detection, MitoScope adopts baldur,22 a long-read mitochondrial variant calling tool equipped to handle heteroplasmic SNVs, indels, and large deletions alongside the long-read structural variant (SV) caller, Sniffles2,23 with parameters customized to identify heteroplasmic SVs (Methods). Variants are annotated using Variant Effect Predictor (VEP)24 and MITOMAP25 for associated genes, functional consequence, pathogenicity, and population frequencies. De novo circular mitochondrial genome assemblies are produced by iteratively running meta-flye,26 a long-read assembler optimized for metagenomic contexts. Downsampled reads (n=100) are iteratively assembled until the expected ~16.5 Kbp circular assembly is obtained, reducing computational load and resolving misassemblies that arise with excessive coverage. To define the extent and dynamics of NUMT integration across human tissues, MitoScope offers optional functionality to perform systematic characterization of non-reference NUMTs using long-read data. Nuclear insertions that align to the mitochondrial genome or split reads with supplementary alignments to both the mitochondrial and nuclear genomes are reported as NUMTs with their corresponding nuclear and mitochondrial breakpoint locations, size, and levels of read support (Methods). Built as a fully containerized Nextflow pipeline using Singularity images, MitoScope ensures reproducibility, scalability, and ease of use with customization options. Quality metrics per sample include read counts from each stage of mtDNA selection, length distributions (histograms/median), mtDNA genome coverage calculated using Mosdepth,27 mtDNA CN, and mitochondrial haplogroups assigned by Haplogrep3.28 MitoScope achieves accurate, high-fidelity variant calling in HapMap mixture benchmarking samples The quantitative performance of MitoScope in somatic variant calling was evaluated using genetically diverse cell mixtures constructed from six selected lymphoblast cell lines of the International HapMap project mixed at defined ratios (83.5% HG005, 10% HG02622, 2% HG02486, 2% HG02257, 2% HG002, 0.5% HG00438) (Figure 1A). This publicly available benchmarking resource was specially designed and carried out through the efforts of the SMaHT Network18–20 to generate variant allele fractions from 0.25% to 16.5%, approximating somatic mosaicism in tissues and serving as analytical ground truth for sensitivity, precision, and limit of detection across variant classes. Using MitoScope, we processed sequencing data (n=7) from Pacific Biosciences (PacBio) and Oxford Nanopore Technologies (ONT) generated across five SMaHT genome characterization centers (GCCs: Baylor College of Medicine (BCM), University of Washington and Seatle Children Research Institute (UW-SCRI), Washington University at St. Louis and Van Andell Institute (WashU-VAI), Broad Institute (Broad), and New York Genome Center (NYGC)) and determined coverage, median read length, and copy number (Table S1). MtDNA SNV and indel variant calls were evaluated against a mitochondrial variant truth set (98 SNVs, 13 indels) from the pangenome graph-derived variant calls of the Human Pangenome Reference Consortium (HPRC).29 The variant calling performance of MitoScope was evaluated alongside Himito and Mitorsaw, two long-read-specific mitochondrial variant calling tools, as well as high coverage short-read data processed by GATK’s Mutect230 in mitochondrial mode (Figure 1B–C, Table S2). Across the entire range of truth set variants, MitoScope performed well in terms of both sensitivity (0.76–0.87) and precision (0.77–0.81) with high F1-scores (0.78–0.84) consistently using both PacBio and ONT data. MitoScope’s performance on long reads was comparable to Mutect2’s performance on short reads in terms of sensitivity (0.73–0.82), while obtaining better precision (0.68–0.75) and F1-scores (0.74–0.78). MitoScope also outperformed Mitorsaw, a PacBio-specific long-read mitochondrial tool, with higher sensitivity (0.40–0.42) and F1-scores (0.53–0.56). Between MitoScope and Himito, MitoScope showed particularly better performance on ONT long-read data with consistently higher F1-scores (0.51–0.79). Notably, mtDNA abundance inferred from the long-read data was highly variable, ranging from 18 to 157 copies per diploid nuclear genome and considerably lower than the mtDNA CN inferred from short-read data (CN: 328–1071). mtDNA CN inferred from long-read datasets displayed an overall reverse correlation with median read length (Figure S2A) and were associated with variation in the library preparation protocols deployed by different GCCs. Long-read sequencing data generated by protocols optimized for longer read lengths (median read length >=20 Kbp) with size selection of the sheared DNA severely underestimated the mtDNA CN. mtDNA abundance is, therefore, underestimated in the standard long-read datasets as mtDNA is preferentially lost during DNA shearing and long fragment selection. Furthermore, because PacBio Fiber-seq31 deploys nuclei preparation which depletes mitochondria not physically attached to the nuclear membrane, protocols using Fiber-seq have reduced recoverable mtDNA in long-read libraries. Despite the variability in copy number, heteroplasmy levels of long-read MitoScope variant calls were highly concordant with the short-read Mutect2 variant calls (Figure 1D, Table S3). Variants were subset to those in common between Mutect2 and MitoScope (n=116) and mean heteroplasmy levels across GCCs were compared for each variant in common. Heteroplasmy levels were highly concordant between the PacBio and ONT calls made by MitoScope (Pearson correlation coefficient R>0.99) and between the Mutect2 short-read calls against both the MitoScope PacBio and ONT calls (R>0.99). Two discordant calls were identified (8860A>G, 4769A>G) where MitoScope called variants homoplasmic on both PacBio and ONT data, while Mutect2 called the same variants heteroplasmic with frequencies between 0.7–0.8. Manual inspection of short and long reads using pileup data supported the homoplasmic calls for the discordant variants (Figure S2B). Despite variant presence in the population at high frequencies (>97% in MITOMAP) and across haplogroups, both variants have previously been reported as being miscalled by Mutect2,32 corroborating the importance of the high accuracy calls from MitoScope. Somatic mtDNA mosaicism across human tissues shows age and tissue dependent patterns We next profiled mitochondrial genome variation across four phenotypically normal postmortem donors (ST001-ST004) collected as part of benchmarking efforts in the SMaHT Network, examining somatic mosaicism in four tissues (i.e., lung, liver, colon, and brain) (Figure 2A). Donors were male, spanning young (70 years; ST002 and ST004) age groups. We processed PacBio and ONT sequencing data generated at each of the five GCCs using MitoScope and determined coverage, read count, length, and CN (Table S4). mtDNA CN showed striking tissue specificity—brain and liver exhibited ~10-fold higher levels than lung and colon (Figure 2B, Figure S3), mirroring the known mitochondrial enrichment in higher-energy demanding tissues such as brain, liver, and muscle33 with no evident age effect. To explore SNV and indel variation across donors and tissues, we classified all variants (n=368) as homoplasmic (heteroplasmy frequency, HF≥0.95; n=166) or heteroplasmic (HF<0.95; n=202) and further as high- or low-confidence. High-confidence variants were defined as those detected by at least two GCCs, in at least two independent samples (tissues or donors), or by both long-read platforms (PacBio and ONT), whereas variants observed only once were retained in the low-confidence set. This strategy enabled robust inclusion of low-frequency heteroplasmies (HF<1%) that would otherwise be difficult to distinguish from technical noise (Figure S4A–B). Most of the SNVs were high-confidence (n=270), including 100% of the homoplasmic (n=166) and 69% of heteroplasmic (n=104) SNVs. No homoplasmic indels were identified and detailed examination of the high-confidence heteroplasmic indels (n=34) largely revealed localization within homopolymer tracts or blacklisted sites, suggesting potential sequencing or alignment artifacts; therefore, downstream analyses focused on the 270 high-confidence SNVs. Aggregated high-confidence SNV calls across donors and tissues revealed a distinct bimodal distribution, predominantly homoplasmic (HF>95%; 166 total), with a second peak of heteroplasmic variants (HF<10%; 104 total), of which 73 were ultra-low frequency (<1%), 30 at 1–5%, and 1 at 5.7% (Figure 2C, Table S5). Of the 65 unique heteroplasmic SNVs detected in our limited survey, eight have not been captured in MITOMAP, a human mitochondrial genome database with an extensive collection of mitochondrial SNVs (n=19,973), reinforcing the value of somatic tissue-based analysis to uncover somatic mosaicism in mtDNA. These heteroplasmic variants exhibited HFs ranging from 0.25% to 2.7% and were largely missense variants in gene regions encoding subunits of NADH dehydrogenase (ND1, ND2, and ND4) and cytochrome c oxidase. One variant (m.7515A>C) was universally found in all four donors with HFs between 0.38–1%, a noncoding polymorphism lacking documented functional significance that suggests many common heteroplasmic variants remain underexplored. Homoplasmic SNVs were common among tissues from the same donor, indicating their germline nature while the numbers and identities of heteroplasmic SNVs varied substantially across donors and tissues, suggesting they were mostly somatic. Although we were limited in the number of donors within each age group and tissue type, a higher overall mtDNA SNV burden was seen in aged donors, with ST002 Lung and ST004 Brain exhibiting more total variants than matched tissues from younger donors, ST001 Lung and ST003 Brain (Figure 2D). This supports the progressive accumulation of mtDNA mutations over the lifespan, consistent with replication-linked errors and chronic oxidative stress in aging tissues. Notably, the single highest heteroplasmic load was observed in lung from the young donor ST001 (<25 years; n=24), exceeding heteroplasmic counts in ST001 Liver and in ST002 Lung. Base substitutions were largely confined to transitions over transversions and showed strand bias with a higher frequency of heavy strand variants consistently across tissues (Figure 2E). The heavy strand (H-strand) spends more time single-stranded during asymmetric replication, making it vulnerable to deamination and oxidative damage—explaining the consistent H-strand bias and transition/transversion excess. This mutational signature, preserved across tissues, reflects endogenous replication-coupled mutagenesis as the major source of mtDNA variation. Across regions in the mitochondrial genome, the control region showed significant enrichment (~5.2-fold, P < 0.001) of SNVs, protein-coding genes showed significant depletion (~1.5-fold, P < 0.001), and rRNA/tRNA genes were stringently constrained, clustering near the expected frequency across all tissues (Figure 2F). This pattern indicates that mtDNA variation is preferentially tolerated in the noncoding regulatory D-loop as previously reported,34 while purifying selection constrains mutations in protein-coding and especially RNA genes critical for oxidative phosphorylation and mitochondrial translation, reinforcing the functional importance of these elements for mitochondrial fitness and cellular energy homeostasis. Using both homoplasmic and heteroplasmic SNVs, we next quantified unique and shared variants across donors and tissues. Homoplasmic variants, presumed germline origin, were predominantly donor-specific (82%), with only a small subset shared between donors, consistent with a limited pool of common mtDNA polymorphisms (Figure 3A). In contrast, 66% of heteroplasmic variants were unique in individual tissues and donors, indicating extensive, individualized somatic mosaicism across the mitochondrial genome (Figure 3B) with rare heteroplasmic variants shared between tissues of the same donor likely representing either maternally inherited heteroplasmies or early somatic events that arose before tissue divergence. Shared heteroplasmic variants between donors with the same tissue type (ST001 and ST002 Lung, ST003 and ST004 Brain) were also rare. Among lung and colon in ST002, frequencies for shared heteroplasmic variants were comparable overall while shared heteroplasmic variant frequencies for ST001 were 4–10-fold lower in liver than lung with overall lower HF levels in the liver (Figure 3C). Shared variants in ST001 with lower HF levels in liver included 16274G>A and 5149C>A, located within the D-loop and ND2 oxidoreductase of the electron transport chain, critical for mitochondrial replication and ATP production, respectively. These observations are consistent with stronger purifying constraints on mtDNA variation in high-energy-demand tissues, favoring maintenance of mitochondrial function and genome stability, while recognizing alternative contributing factors such as cell turnover and replication rates. Across tissues, multiple mtDNA SNVs mapped to MITOMAP entries with reported or confirmed disease associations (Figure 3D). MITOMAP classifies variants as “reported,” with “conflicting reports,” or as “confirmed” (Cfrm) disease variants, where the latter is further annotated according to the ClinGen Pathogenicity Ratings (VUS - Variant of Uncertain Significance; LP - Likely Pathogenic; P - Pathogenic), reflecting the cumulative strength and consistency of clinical and functional evidence. Notably, each tissue harbored several reported variants (published in literature with limited evidence). Among these, we detected the canonical pathogenic m.3243A>G substitution in MT-TL1 (MELAS-associated) at 2.5% heteroplasmy in ST004 Brain and a likely pathogenic m.13042G>A missense variant in MT-ND5 (linked to optic neuropathy and retinopathy) at 0.5% heteroplasmy in ST002 Lung (Figure S4C–D). The presence of such low-frequency, disease-associated heteroplasmies in clinically unremarkable postmortem tissues underscores that pathogenic mtDNA mutations can arise and persist somatically with age despite purifying selection, remaining functionally silent at low levels or contributing to subclinical mitochondrial dysfunction and late-onset disease risk at functional thresholds. Mitochondrial deletion landscapes reveal age-biased, tissue-specific instability driven by microhomology-mediated end joining Beyond somatic SNVs, the mitochondrial genome also harbors extensive large-scale deletions that accumulate with age in post-mitotic tissues and are implicated in primary mitochondrial disease, neurodegenerative disorders, and multisystem syndromes.3,5,7 Deletions in mtDNA have been most intensively surveyed in disease contexts; here we systematically mapped the deletion landscape across benchmarking tissues. We observed widespread deletions spanning hundreds to thousands of base pairs, with the vast majority localizing to the major arc, which sits between the origins of replication of the light and heavy strand, and contains most protein-coding genes, indicating that this region is particularly prone to deletions (Figure 4A). Despite platform- and center-specific variation in absolute deletion frequencies, deletions were consistently most abundant in brain, with the aged brain (ST004) showing >2-fold higher total deletion heteroplasmy than the younger brain (ST003; HF 1.22% vs 0.45%), and a similar trend in aged versus young lung (ST002 vs ST001; HF 0.26% vs 0.19%) (Figure 4B). The canonical 4,977 bp “common deletion,” which removes five tRNA genes and seven protein-coding genes and is known to compromise respiratory chain function at high heteroplasmy levels,35 was detected exclusively in brain at low levels (HF ~0.01%) and was essentially absent in other tissues (Figure 4A–B), consistent with a minor allele confined to subsets of cells. Except in liver, 88–97% of deletions mapped to the major arc, whereas a 50 bp control-region deletion in the D-loop (m.298_347del50) predominated in liver (Figure 4A–B). This deletion has previously been reported in the germline of healthy Chinese cohorts36 and as a recurrent, clonally expanded somatic lesion in gastric adenocarcinoma37 and aging human putamen;34 its presence as a major deletion class in liver but absence in matched lung from the same donor suggests a mosaic, tissue-restricted event. Together, these observations indicate that although structural mtDNA deletions are widespread and can be tolerated in both germline and somatic compartments, they can clonally accumulate in energetically demanding tissues, supporting a model in which mtDNA instability contributes to age-related vulnerability38 and tissue-specific mitochondrial dysfunction. Deletion breakpoints were frequently flanked by short direct repeats or microhomology, a configuration compatible with formation by slipped-strand mispairing and/or microhomology-mediated end joining (MMEJ). The 4,977 bp “common deletion” is flanked by a 13 bp perfect direct repeat and m.298_347del50 by a 9 bp perfect direct repeat. Whereas slipped-strand mispairing reflects polymerase misalignment at repeats during replication, MMEJ acts on double-strand breaks by aligning 2–25 bp microhomology tracts on opposite ends. To infer the predominant mechanism, we examined sequence features at deletion junctions (n=3,638) and identified short direct repeats (4–15 bp) across 87% of unique breakpoints. The observed distribution of microhomology was significantly higher than expected by chance (Wilcoxon rank-sum, P 6 bp microhomology were ~4-fold more frequent than expected (Figure 4C). The combination of large deletion sizes (hundreds to thousands of bp) and 4–15 bp microhomology directly juxtaposed at the junction, often outside perfect tandem arrays, is most consistent with MMEJ-like repair as a major pathway generating sporadic mtDNA deletions. Long-read detection of NUMTs reveals tissue-mosaic events, underreported and previously unidentified NUMTs, and trinucleotide biases in insertion contexts NUMTs have been linked to carcinogenesis and to age-related processes involving genomic instability and mitochondrial dysfunction,9 but whether they predominantly act as causal drivers or as byproducts of cellular stress remains unresolved. Additionally, NUMTs’ genomic positions, allele frequencies, and tissue distributions can profoundly affect variant interpretation and nuclear–mitochondrial crosstalk. Building on this, we systematically profiled NUMTs in the nuclear genomes of the benchmarking tissues. We applied a confidence framework to NUMTs analogous to that used for SNVs and indels. High-confidence NUMTs required support in at least two independent contexts (distinct GCCs, technologies, donors, or tissues), whereas events observed in only a single sample replicate across GCCs and platforms were classified as low-confidence and excluded from downstream analyses. This strategy yielded 63 NUMT calls, of which 57 met high-confidence criteria and collapsed into 17 distinct NUMTs (Figure 5A), indicating a modest number of NUMTs shared across donors and tissues with strong technical support. These NUMTs ranged from 32 bp to 870 bp in length (median 66 bp, mean 153 bp, s.d. 203 bp), with insertion breakpoints distributed across the genome but most frequently mapping to chromosome 11 (Figure 5C), suggesting a nonuniform chromosomal landscape for mitochondrial insertions. Across benchmarking tissues, each sample harbored 7 to 11 NUMTs (mean 9.5; s.d. = 1.4) (Figure 5B), approximately two-fold higher burden than the average NUMT count per individual reported in a recent short-read study of 66,083 genomes from the 100,000 Genomes Project,9 which identified 1,637 NUMTs in total. This enrichment underscores the substantial reservoir of non-reference NUMTs that remain undetected by short-read sequencing and highlights long-read sequencing as a more sensitive modality for capturing these insertion events. For donors with two profiled tissues (ST001 and ST002), all but one NUMT were shared between tissues (Figure 5A), consistent with predominantly germline or early embryonic insertions that are widely distributed across somatic lineages. In contrast, a single NUMT at chr21:9,676,568 (NUMT-17) was detected only in the lung of donor ST002 with variant allele frequencies 0.25–0.32 and absent from the matched colon, indicating a tissue-restricted mosaic NUMT and illustrating how NUMT insertions can also arise later in development. Using the population frequencies from the 100,000 Genomes Project short-read study, we classified the 17 high-confidence NUMTs into 10 common (allele frequency ≥1%), two rare (<1% and ≥0.1%), one ultra-rare (<0.1%), and four previously unreported NUMTs that were absent from the 66,083-genome survey (Figure 5D). Surprisingly, all three of the NUMTs categorized as rare or ultra-rare according to the short-read study were present in at least two or more of the four benchmarking donors, implying that many of these ‘rare/ultra-rare’ NUMTs may be considerably more common than previously reported in short-read based surveys. To expand NUMT discovery beyond the benchmarking tissues and incorporate a more comprehensive long-read NUMT catalog, we leveraged long-read whole-genome data from 359 individuals enrolled in the SeqFirst-neo project.39 Across the long-read cohort, we identified 168 unique non-reference NUMTs, of which 82 (49%) were absent in the existing short-read catalog (Figure 5E, Table S6). Notably, 59 of these new NUMTs were rare (allele frequency <1%) among the SeqFirst cohort. These findings demonstrate that long-read sequencing uncovers a large, previously hidden layer of both common and low-frequency NUMT variation and emphasizes the need for more comprehensive, publicly accessible long-read-based NUMT catalogs to improve variant interpretation, mitigate NUMT-mediated artifacts, and enable more accurate studies of nuclear–mitochondrial genome interactions. To examine sequence motifs that may promote mtDNA insertions into the nuclear genome, we analyzed the trinucleotide context surrounding NUMT breakpoints in both the nuclear and mitochondrial loci of the SeqFirst cohort (Figure 5F). The observed trinucleotide distribution at NUMT breakpoints deviated markedly from the expected genome-wide frequencies in both the nuclear and mitochondrial compartments (P < 0.001). Specifically, breakpoints were enriched for nCC/CCn motifs and depleted for nTT/TTn motifs across both nuclear and mitochondrial loci, suggesting that local sequence features—such as potential microhomology or secondary structure propensity—facilitate the non-homologous end joining or microhomology-mediated end joining that drives NUMT formation as has previously been proposed.9 In nuclear breakpoints, nCG trinucleotides were also underrepresented, while mitochondrial breakpoints exhibited the strongest depletion for GCG (>3-fold reduction relative to background), highlighting biases in DNA fragility or mtDNA packaging constraints. These nonrandom sequence preferences imply that NUMT insertion sites are not stochastic but biased toward specific fragile motifs, with broader implications for predicting NUMT hotspots and understanding mitochondrial genome instability in aging and disease contexts.

In this work, we present MitoScope, a scalable long-read workflow, to enable high-fidelity detection of somatic mtDNA variants—including low-frequency heteroplasmies, age- and tissue-biased deletions via MMEJ, and unidentified NUMTs with nonrandom trinucleotide insertion motifs—across SMaHT benchmarking tissues and expanded cohorts. MitoScope outperforms prior tools and reveals substantial improvements over short-read methods with high sensitivity and precision for variant calls on the HapMap mixture cell lines. Our integrated analysis reveals pervasive mtDNA instability across human tissues, driven by replication stress, purifying selection, and repair pathway biases such as microhomology-mediated end joining. Heteroplasmic variants were predominantly tissue- or donor-specific, with low-frequency pathogenic heteroplasmies persisting somatically in clinically normal tissues, suggesting subclinical contributions to late-onset disease risk. Heteroplasmic variant and NUMT discovery emphasize the need for large-scale long-read studies to capture this hidden somatic reservoir. A major challenge in long-read mtDNA analysis is the marked variability in inferred copy number, which significantly underestimates true abundance compared to short-read estimates due to protocol-specific biases in high molecular weight DNA preparation. Size-selected libraries favoring long fragments preferentially lose short mtDNA molecules during shearing and selection, as these circular ~16.5 Kbp genomes fragment inefficiently and are discarded, systematically biasing all long-read datasets analyzed here. This underrepresentation not only depresses apparent mtDNA coverage but also risks uncovering low-frequency somatic variants and large mtDNA deletions, as reduced input read depth amplifies stochastic sampling effects and diminishes statistical power for rare alleles. Standardized, mtDNA-inclusive library prep protocols or separate enrichment strategies will be essential to mitigate these artifacts and ensure accurate quantification of mtDNA variation in future long-read studies. mtDNA play a central role in human biology, powering oxidative phosphorylation and influencing metabolism, aging, and disease across tissues. In our limited analyses, the single highest heteroplasmic load was observed in lung from a young donor (ST001) with documented pulmonary embolism and aspiration pneumonia, suggesting acute or tissue-specific stressors can drive substantial somatic mtDNA variation even in early adulthood. Initiatives like the SMaHT Network, with its standardized multi-center long-read benchmarking across diverse post-mortem tissues, enable unprecedented systematic surveys of mtDNA mosaicism using robust tools like MitoScope and optimized protocols. Such expanded studies will facilitate critical phenotype-genotype associations to elucidate these relationships. These efforts are expected to illuminate variation patterns; uncover hidden heteroplasmies, deletions, and NUMTs; and reveal their contributions to tissue-specific instabilities, late-onset disorders, variant evolution, and nuclear-mitochondrial crosstalk.

of the study Our study provides a crucial glimpse into the underexplored areas of mitochondrial genome analysis using the available donors and tissues sequenced as part of the SMaHT Network benchmarking set. As the benchmarking set, the number of individual tissues and donors included were limited in number. Future work will aim to incorporate a larger dataset of donors and tissue types in addition to mother/child duos to better extrapolate the true origin of germline or somatic variants and investigate age and tissue related variant associations on a greater scale. We also recognize that mtDNA methylation analysis represents an important and exciting frontier for understanding epigenetic regulation of mitochondrial function and nuclear-mitochondrial interactions. While MitoScope can generate single-read 5mC methylation calls across mtDNA CpG sites from long-read data and enable nucleotide-resolution epigenetic profiling alongside variant detection, we did not extensively analyze mtDNA methylation due to its controversial nature and pronounced technical discordance between platforms—PacBio calls exhibited higher noise and elevated 5mC levels inconsistent with reports of very low mitochondrial methylation,40 while ONT data aligned better with expected near-background patterns. This variability likely reflects platform-specific base calling artifacts or limited CpG contexts in the compact mtDNA, underscoring the need for orthogonal validation in future studies to resolve whether trace methylation plays a functional role in mtDNA regulation. RESOURCE AVAILABILITY Lead contact Further information and requests for resources and tools should be directed to and will be fulfilled by the Lead Contact, Chia-Lin Wei ([email protected]). Data and code availability This study was conducted as part of the NIH Common Fund consortium initiative, Somatic Mosaicism across Human Tissues (SMaHT). The benchmark datasets described in this study are available through dbGaP (http://www.ncbi.nlm.nih.gov/gap) under the study accession number phs004193. The data used in this work was provided by the SMaHT Data Analysis Center (DAC) [1UM1DA058230] on behalf of the SMaHT Network. More information about the SMaHT Network is available online at https://smaht.org/, about the SMaHT Data Portal at https://data.smaht.org/, and types of data generated by the Network at https://data.smaht.org/about/consortium/data. The HPRC pangenome-graph derived variant calls used for the mitochondrial truth set are available from the HPRC data portal (https://data.humanpangenome.org/alignments). The PacBio and ONT SeqFirst-neo data used for NUMT analysis are available through the GREGoR Consortium: Genomics Research to Elucidate the Genetics of Rare Disease dbGaP study (accession no. phs003047) in NHGRI’s Analysis Visualization and Informatics Lab-space (AnVIL) platform (https://explore.anvilproject.org/datasets). MitoScope code has been deposited in GitHub at https://github.com/czakarian/mitoscope and is publicly available as of the date of publication. Analysis scripts used to create and summarize results have been deposited in GitHub at https://github.com/czakarian/smaht_mitoscope_analysis and are publicly available as of the date of publication. Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request.

MitoScope mtDNA selection BAM/CRAM files are converted to FASTQs prior to k-mer selection. A k-mer count database of mtDNA k-mer sequences (k= 29) is generated using the human mitochondrial reference sequence (GenBank: NC_012920.1) and the k-mer tool kit, KMC3 (v3.2.1).41 The k-mer count database is used to selectively retain reads in the FASTQ file with >2,500 matching k-mers for retention of reads with sufficient length and sequence similarity to mtDNA. After k-mer selection, reads are aligned to the mitochondrial genome using Minimap2 (v2.24)42 and evaluated for the removal of contaminating NUMT reads. Thresholds are enforced for maximum soft-clipped sequence in a read that does not align back to the mitochondrial genome as well as the fraction of 5mC methylated CpG sites in a read. Up to 200 bp of unaligned soft-clipping and a maximum of 50% methylated CpG sites are used for the removal of NUMTs to avoid compromising mtDNA read counts. A stricter threshold of 200 bp of unaligned sequence alongside a more lenient threshold of 50% methylated CpG sites is used to prioritize removal of reads with significant levels of unaligned sequence while also discarding reads with good alignment but deviating methylation profiles. A methylation likelihood probability threshold of 0.5 is used to classify sites as methylated or unmethylated. Reads failing either filtration threshold are filtered out and written to a separate discard BAM file. Foldback artifacts, where a molecule folds back on itself and appears as a split read aligning to opposite strands, are also removed to avoid compromising assembly accuracy. MitoScope variant calling MitoScope uses baldur22 (git-commit: 73cb26e) to call heteroplasmic SNVs, indels, and large deletions (options: -P 5 -q 20 -Q 20 -I 30 --output-deletions --small-deletion-limit 25 –large-deletion-limit 25 --indel-thresholds 0.1 0.1 --snv-thresholds 0.0005 0.0025). Structural variants are called using Sniffles2,23 (v2.6.2) with parameters customized to identify heteroplasmic SVs (options: --qc-output-all --minsvlen 5 --minsupport 4). Variant Effect Predictor (VEP)24 (v115.0) is run on the resulting SNV/indel VCF for variant annotations and MITOMAP25 annotations are integrated using BCFtools (v1.21)43 for pathogenicity and population frequencies. MitoScope assembly generation Mitochondrial genome assemblies are generated using meta-flye26 (v2.9.6, options: --meta -m 2500). Meta-flye is iteratively run on downsampled reads (n=100) generated by seqtk (v1.5) until the expected ~16.5 Kbp circular contig is obtained or a maximum number of iterations is reached (maximum of 10 iterations). A tolerance of 100 bp relative to the length of the mitochondrial genome (16,569 bp) is allowed for an assembly to be considered the valid length. For low-coverage samples where a circular assembly of the expected length cannot be achieved, the retained assembly output is from the final iteration. MitoScope copy number calculation mtDNA copy number (CN) is estimated using coverage of the mitochondrial genome and the nuclear genome (chr1–22,X,Y). Coverage values are generated using mosdepth27 and the relative mtDNA CN per cell, assuming a diploid nuclear genome, is calculated as (mtDNA coverage / nuclear coverage * 2). MitoScope NUMT characterization Non-reference NUMTs are called based on insertion calls or split read-derived calls. For insertion-derived NUMTs, structural variants are called using Sniffles2,23 (v2.6.2, options: --minsvlen 20) on aligned bam/cram inputs. Sequences for called insertions are written to a FASTA file that is mapped against the mitochondrial genome using BLAST44 (v2.16.0, options: -word_size 25 -perc_identity 95). Insertions that share >95% sequence identity with mtDNA are run through BLAST a second time using the GRCh38 reference to retain only those insertions with higher sequence similarity to the mitochondrial genome than to the nuclear genome. Split read-derived NUMTs are called based on reads with supplementary alignments to both the mitochondrial and nuclear genomes with sufficient breakpoint support (>=4 reads). Final reported NUMTs are categorized according to their nuclear and mitochondrial breakpoint locations, levels of read support, and variant allele frequencies. SMaHT benchmarking data The sequencing data for the HapMap mixture cell line and the benchmarking tissues (ST001-ST004) was generated by the Somatic Mosaicism across Human Tissues (SMaHT) benchmarking effort.18–20 Long-read (PacBio and ONT) and short-read (Illumina) DNA sequencing data was downloaded from the SMaHT data portal (https://data.smaht.org) in the form of aligned and indexed CRAM files generated using the GRCh38 reference. The HapMap mixture sample and all tissue samples were each sequenced by at least three different GCCs (BCM, UW-SCRI, WashU-VAI, Broad, NYGC). Intact tissue or homogenates were used as starting input material. Further details on sequencing and library preparation of the HapMap and benchmarking samples can be found in the SMaHT data portal, the SMaHT HapMap mixtures paper, and the SMaHT flagship paper.19,20 Benchmarking tool performance using the HapMap mixture cell line data The mitochondrial truth set used to evaluate variant calling performance was derived from the HPRC Version 2 release of the pangenome-graph derived variant calls. The truth set was generated by subsetting the VCF to variants located on chrM and present in at least one of the HapMap mixture samples (HG005, HG02622, HG02486, HG02257, HG002, HG00438). As the variant calls made on the HapMap mixture samples were indistinguishable based on sample origin, we generated a collapsed set of HapMap variant calls (98 SNVs, 13 indels) from the truth set. Genotype columns in the mitochondrial truth set VCF corresponding to each HapMap sample in the mixture were collapsed to a single column indicating variant presence. Variant calling performance was evaluated across Mitoscope (v0.2.1), Himito16 (git-commit: 95edeb1), and Mitorsaw17 (v0.2.4) using default parameters. GATK Mutect245 (v4.5.0.0) was run in mitochondrial mode (--mitochondria-mode) with otherwise default parameters followed by FilterMutectCalls to retain passing variants. Sensitivity, precision, and F1 scores were calculated using RTG tools (v3.12.1).46 HapMap variants in homopolymer regions prone to sequencing artifacts (m.309–311,16179–16183) were excluded from sensitivity, precision, and F1 calculations. SNV/indel analysis SNV and indel variant calls were evaluated for high/low confidence status by counting the number of replicates across GCCs, sequencing technologies, donors, and tissues. Variants called in at least two different instances were retained in the high-confidence set for further analysis, while variants seen only once were retained in the low-confidence set. Variants called in specific homopolymer regions (m.309–311,16179–16183) prone to sequencing artifacts were excluded. Heavy and light strand base substitutions were assigned based on the reference base in the mitochondrial reference sequence (revised Cambridge Reference Sequence) which is that of the light strand. SNVs with reference pyrimidine bases (C,T) were counted as light strand variants while purines (A,G) were collapsed to the pyrimidine view and counted as heavy strand. Binomial tests were performed to assess whether observed variant counts were significantly enriched or depleted across mitochondrial gene regions (control region, protein coding, rRNA, tRNA) using expected counts proportional to the length of each region. Variant counts were aggregated per region for testing, and p-values were adjusted for multiple testing correction using the Benjamini–Hochberg false discovery rate method. For visualization, variant counts were normalized by region length to obtain variant frequencies. Deletion analysis Frequency and breakpoint spanning deletion analysis was performed using output from baldur with Sniffles2 calls used for further validation of the m.298_347del50 control region deletion and the 4,977 bp ‘common’ deletion. Only deletions >45 bp were kept for quantification. Major arc deletions were quantified by counting deletions with breakpoints falling within m.5576–15976. The “common deletion” was quantified by counting deletions whose start/end breakpoints fell within m.8460–8480 and m.13430–13455, respectively. Control region deletions were quantified by counting deletions with breakpoints within m.16024–16569 or m.1–576. The m.298_347del50 control region deletion was quantified for deletions with start/end breakpoints within m.287–307 and m.336–356. Frequencies for each deletion group were calculated by normalizing read counts with deletions by total mtDNA read counts. Deletion breakpoint microhomology analysis was performed by identifying perfect direct repeats between the first 20 bp of deletion sequence and the 20 bp of flanking sequence at the end of the deletion using BLAST (options: -word_size 4 -perc_identity 100). For each reported deletion with unique breakpoints across the benchmarking tissues, control data was simulated with matching deletion length and randomized breakpoints with five simulated deletions per observed deletion (3,638 observed deletions, 18,190 simulated deletions). Wilcoxon-rank sum test was used to test for statistical significance between observed and expected distributions. NUMT analysis NUMTs included in the analysis of the benchmarking tissues and SeqFirst samples were insertion-derived. Split read-derived NUMTs were not included to avoid the complication of comparing NUMTs with potentially only one mitochondrial breakpoint. Precise nuclear and mitochondrial NUMT coordinates for benchmarking tissue samples with the potential to identify individuals were not specified for participant privacy. For comparison of NUMTs across different samples, a tolerance of 100 bp was used to cluster NUMTs by nuclear and mitochondrial breakpoints. SeqFirst samples with both PacBio and ONT data (n=359) were used to derive the long-read NUMT catalog. NUMTs that were called across either or both technologies were included in the set of NUMTs and classified the same way as in the benchmarking tissues using nuclear and mitochondrial breakpoints. The short-read NUMT catalog of 1,637 NUMTs for the 100,000 Genomes Project in England was downloaded from the Wei et. al. (2022)9 publication. NUMT comparisons between the different catalogs was performed using 100 bp tolerance surrounding nuclear breakpoints. Trinucleotide breakpoint analysis was performed using the directly flanking trinucleotide sequences surrounding the nuclear and mitochondrial breakpoints of NUMTs. Breakpoints directly at the beginning or end (1 or 16,569) that could pose issues with genome wrap-around were excluded. Genome-wide trinucleotide distributions were used for the control set by calculating the proportion of each trinucleotide across the mitochondrial genome and the nuclear genome (chr1–22,X,Y). Trinucleotides sequences were collapsed and counted together with their corresponding reverse complements. Chi-square goodness of fit test was used to test for significance between observed and expected trinucleotide distributions. Supplementary Material ACKNOWLEDGMENTS This research is supported by the NIH Common Fund, through the Office of Strategic Coordination/Office of the NIH Director under awards U24 MH133204, U24 NS132103, UG3 NS132024, UG3 NS132061, UG3 NS132084, UG3 NS132105, UG3 NS132127, UG3 NS132128, UG3 NS132132, UG3 NS132134, UG3 NS132135, UG3 NS132136, UG3 NS132138, UG3 NS132139, UG3 NS132144, UG3 NS132146, UM1 DA058219, UM1 DA058220, UM1 DA058229, UM1 DA058230, UM1 DA058235, and UM1 DA058236. Research reported in this publication was supported, in part, by the National Human Genome Research Institute of the National Institutes of Health (NIH) under NHGRI Award Number 1U01HG011744 (to C.L.W and E.E.E.) and NIH Common Fund UM1DA058220 (to J.T.B., A.B.S., E.E.E., and C.L.W.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. E.E.E. is an investigator of the Howard Hughes Medical Institute. Footnotes DECLARATION OF INTERESTS A.B.S. is a co-inventor on a patent related to the Fiber-seq and DAF-seq methods. J.T.B. is a consultant for Mosaica Medicines. E.E.E. is a scientific advisory board (SAB) member of Variant Bio, Inc. All other authors declare no competing interests.

- 1.Chen W., Zhao H., and Li Y. (2023). Mitochondrial dynamics in health and disease: mechanisms and potential targets. Signal Transduct. Target. Ther. 8, 333. 10.1038/s41392-023-01547-9. [DOI] [PMC free article] [PubMed] [Google Scholar] - 2.Tábara L.-C., Segawa M., and Prudent J. (2025). Molecular mechanisms of mitochondrial dynamics. Nat. Rev. Mol. Cell Biol. 26, 123–146. 10.1038/s41580-024-00785-1. [DOI] [PubMed] [Google Scholar] - 3.Alston C.L., Rocha M.C., Lax N.Z., Turnbull D.M., and Taylor R.W. (2017). The genetics and pathology of mitochondrial disease. J. Pathol. 241, 236–250. 10.1002/path.4809. [DOI] [PMC free article] [PubMed] [Google Scholar] - 4.Wallace D.C. (2012). Mitochondria and cancer. Nat. Rev. Cancer 12, 685–698. 10.1038/nrc3365. [DOI] [PMC free article] [PubMed] [Google Scholar] - 5.Wang Z., Li Z., Liu H., Yang C., and Li X. (2025). Mitochondrial clonal mosaicism encodes a biphasic molecular clock of aging. Nat. Aging 5, 1637–1651. 10.1038/s43587-025-00890-6. [DOI] [PMC free article] [PubMed] [Google Scholar] - 6.Chinnery P.F., and Hudson G. (2013). Mitochondrial genetics. Br. Med. Bull. 106, 135–159. 10.1093/bmb/ldt017. [DOI] [PMC free article] [PubMed] [Google Scholar] - 7.Klemmensen M.M., Borrowman S.H., Pearce C., Pyles B., and Chandra B. (2024). Mitochondrial dysfunction in neurodegenerative disorders. Neurotherapeutics 21, e00292. 10.1016/j.neurot.2023.10.002. [DOI] [PMC free article] [PubMed] [Google Scholar] - 8.Bernardino Gomes T.M., Vincent A.E., Menger K.E., Stewart J.B., and Nicholls T.J. (2024). Mechanisms and pathologies of human mitochondrial DNA replication and deletion formation. Biochem. J. 481, 683–715. 10.1042/BCJ20230262. [DOI] [PMC free article] [PubMed] [Google Scholar] - 9.Wei W., Schon K.R., Elgar G., Orioli A., Tanguy M., Giess A., Tischkowitz M., Caulfield M.J., and Chinnery P.F. (2022). Nuclear-embedded mitochondrial DNA sequences in 66,083 human genomes. Nature 611, 105–114. 10.1038/s41586-022-05288-7. [DOI] [PMC free article] [PubMed] [Google Scholar] - 10.Yuan Y., Ju Y.S., Kim Y., Li J., Wang Y., Yoon C.J., Yang Y., Martincorena I., Creighton C.J., Weinstein J.N., et al. (2020). Comprehensive molecular characterization of mitochondrial genomes in human cancers. Nat. Genet. 52, 342–352. 10.1038/s41588-019-0557-x. [DOI] [PMC free article] [PubMed] [Google Scholar] - 11.Ju Y.S., Tubio J.M.C., Mifsud W., Fu B., Davies H.R., Ramakrishna M., Li Y., Yates L., Gundem G., Tarpey P.S., et al. (2015). Frequent somatic transfer of mitochondrial DNA into the nuclear genome of human cancer cells. Genome Res. 25, 814–824. 10.1101/gr.190470.115. [DOI] [PMC free article] [PubMed] [Google Scholar] - 12.Marzetti E., Di Lorenzo R., Calvani R., Coelho-Junior H.J., Pesce V., Landi F., Leeuwenburgh C., and Picca A. (2025). Nuclear mitochondrial DNA transfer revisited: From genomic noise to hallmark of aging. Ageing Res. Rev. 112, 102892. 10.1016/j.arr.2025.102892. [DOI] [PubMed] [Google Scholar] - 13.Willett-Brozick J.E., Savul S.A., Richey L.E., and Baysal B.E. (2001). Germ line insertion of mtDNA at the breakpoint junction of a reciprocal constitutional translocation. Hum. Genet. 109, 216–223. 10.1007/s004390100564. [DOI] [PubMed] [Google Scholar] - 14.Turner C., Killoran C., Thomas N.S.T., Rosenberg M., Chuzhanova N.A., Johnston J., Kemel Y., Cooper D.N., and Biesecker L.G. (2003). Human genetic disease caused by de novo mitochondrial-nuclear DNA transfer. Hum. Genet. 112, 303–309. 10.1007/s00439-002-0892-2. [DOI] [PubMed] [Google Scholar] - 15.Ip E.K.K., Troup M., Xu C., Winlaw D.S., Dunwoodie S.L., and Giannoulatou E. (2022). Benchmarking the effectiveness and accuracy of multiple mitochondrial DNA variant callers: Practical implications for clinical application. Front. Genet. 13, 692257. 10.3389/fgene.2022.692257. [DOI] [PMC free article] [PubMed] [Google Scholar] - 16.Su H., Huang Y., Durham T., Kong N., Casey E., Benjamin D., Jin S.C., and Garimella K.V. (2025). Himito : A graph-based toolkit for mitochondrial genome analysis using long reads. bioRxiv, 2025.11.03.686348. 10.1101/2025.11.03.686348. [DOI] [Google Scholar] - 17.mitorsaw (Github; ). [Google Scholar] - 18.Coorens T.H.H., Oh J.W., Choi Y.A., Lim N.S., Zhao B., Voshall A., Abyzov A., Antonacci-Fulton L., Aparicio S., Ardlie K.G., et al. (2025). The somatic mosaicism across human tissues network. Nature 643, 47–59. 10.1038/s41586-025-09096-7. [DOI] [PMC free article] [PubMed] [Google Scholar] - 19.Somatic Mosaicism across Human Tissues Network, and Abyzov A. (2025). Comprehensive benchmarking of somatic mutation detection by the SMaHT Network. bioRxivorg. 10.1101/2025.10.09.678885. [DOI] [Google Scholar] - 20.Kong N., Tang Z., Ruttenberg A., Macias-Velasco J.F., Li Z., Zhang W., Miao B., Xin Z., Fu Q., Park H., et al. (2025). A pangenomic method for establishing a somatic variant detection resource in HapMap mixtures. bioRxivorg. 10.1101/2025.09.29.679336. [DOI] [Google Scholar] - 21.Liu B., Du Q., Chen L., Fu G., Li S., Fu L., Zhang X., Ma C., and Bin C. (2016). CpG methylation patterns of human mitochondrial DNA. Sci. Rep. 6, 23421. 10.1038/srep23421. [DOI] [PMC free article] [PubMed] [Google Scholar] - 22.Keraite I., Becker P., Canevazzi D., Frias-López C., Dabad M., Tonda-Hernandez R., Paramonov I., Ingham M.J., Brun-Heath I., Leno J., et al. (2022). A method for multiplexed full-length single-molecule sequencing of the human mitochondrial genome. Nat. Commun. 13, 5902. 10.1038/s41467-022-33530-3. [DOI] [PMC free article] [PubMed] [Google Scholar] - 23.Smolka M., Paulin L.F., Grochowski C.M., Horner D.W., Mahmoud M., Behera S., Kalef-Ezra E., Gandhi M., Hong K., Pehlivan D., et al. (2024). Detection of mosaic and population-level structural variants with Sniffles2. Nat. Biotechnol. 10.1038/s41587-023-02024-y. [DOI] [Google Scholar] - 24.McLaren W., Gil L., Hunt S.E., Riat H.S., Ritchie G.R.S., Thormann A., Flicek P., and Cunningham F. (2016). The Ensembl Variant Effect Predictor. Genome Biol. 17, 122. 10.1186/s13059-016-0974-4. [DOI] [PMC free article] [PubMed] [Google Scholar] - 25.Brandon M.C., Lott M.T., Nguyen K.C., Spolim S., Navathe S.B., Baldi P., and Wallace D.C. (2005). MITOMAP: a human mitochondrial genome database−-2004 update. Nucleic Acids Res. 33, D611–3. 10.1093/nar/gki079. [DOI] [PMC free article] [PubMed] [Google Scholar] - 26.Kolmogorov M., Bickhart D.M., Behsaz B., Gurevich A., Rayko M., Shin S.B., Kuhn K., Yuan J., Polevikov E., Smith T.P.L., et al. (2020). metaFlye: scalable long-read metagenome assembly using repeat graphs. Nat. Methods 17, 1103–1110. 10.1038/s41592-020-00971-x. [DOI] [PMC free article] [PubMed] [Google Scholar] - 27.Pedersen B.S., and Quinlan A.R. (2018). Mosdepth: quick coverage calculation for genomes and exomes. Bioinformatics 34, 867–868. 10.1093/bioinformatics/btx699. [DOI] [PMC free article] [PubMed] [Google Scholar] - 28.Schönherr S., Weissensteiner H., Kronenberg F., and Forer L. (2023). Haplogrep 3 - an interactive haplogroup classification and analysis platform. Nucleic Acids Res. 51, W263–W268. 10.1093/nar/gkad284. [DOI] [PMC free article] [PubMed] [Google Scholar] - 29.Porubsky D., Vollger M.R., Harvey W.T., Rozanski A.N., Ebert P., Hickey G., Hasenfeld P., Sanders A.D., Stober C., Human Pangenome Reference Consortium, et al. (2023). Gaps and complex structurally variant loci in phased genome assemblies. Genome Res. 33, 496–510. 10.1101/gr.277334.122. [DOI] [PMC free article] [PubMed] [Google Scholar] - 30.Cibulskis K., Lawrence M.S., Carter S.L., Sivachenko A., Jaffe D., Sougnez C., Gabriel S., Meyerson M., Lander E.S., and Getz G. (2013). Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219. 10.1038/nbt.2514. [DOI] [PMC free article] [PubMed] [Google Scholar] - 31.Stergachis A.B., Debo B.M., Haugen E., Churchman L.S., and Stamatoyannopoulos J.A. (2020). Single-molecule regulatory architectures captured by chromatin fiber sequencing. Science 368, 1449–1454. 10.1126/science.aaz1646. [DOI] [PubMed] [Google Scholar] - 32.Diroma M.A., Modi A., Lari M., Sineo L., Caramelli D., and Vai S. (2021). New insights into mitochondrial DNA reconstruction and variant detection in ancient samples. Front. Genet. 12, 619950. 10.3389/fgene.2021.619950. [DOI] [PMC free article] [PubMed] [Google Scholar] - 33.Abd Radzak S.M., Mohd Khair S.Z.N., Ahmad F., Patar A., Idris Z., and Mohamed Yusoff A.A. (2022). Insights regarding mitochondrial DNA copy number alterations in human cancer (Review). Int. J. Mol. Med. 50, 104. 10.3892/ijmm.2022.5160. [DOI] [PMC free article] [PubMed] [Google Scholar] - 34.Williams S.L., Mash D.C., Züchner S., and Moraes C.T. (2013). Somatic mtDNA mutation spectra in the aging human putamen. PLoS Genet. 9, e1003990. 10.1371/journal.pgen.1003990. [DOI] [PMC free article] [PubMed] [Google Scholar] - 35.Yusoff A.A.M., Abdullah W.S.W., Khair S.Z.N.M., and Radzak S.M.A. (2019). A comprehensive overview of mitochondrial DNA 4977-bp deletion in cancer studies. Oncol. Rev. 13, 409. 10.4081/oncol.2019.409. [DOI] [PMC free article] [PubMed] [Google Scholar] - 36.Bi R., Zhang A.-M., Zhang W., Kong Q.-P., Wu B.-L., Yang X.-H., Wang D., Zou Y., Zhang Y.-P., and Yao Y.-G. (2010). The acquisition of an inheritable 50-bp deletion in the human mtDNA control region does not affect the mtDNA copy number in peripheral blood cells. Hum. Mutat. 31, 538–543. 10.1002/humu.21220. [DOI] [PubMed] [Google Scholar] - 37.Burgart L.J., Zheng J., Shu Q., Strickler J.G., and Shibata D. (1995). Somatic mitochondrial mutation in gastric cancer. Am. J. Pathol. 147, 1105–1111. [PMC free article] [PubMed] [Google Scholar] - 38.Sprason C., Tucker T., and Clancy D. (2024). MtDNA deletions and aging. Front. Aging 5, 1359638. 10.3389/fragi.2024.1359638. [DOI] [PMC free article] [PubMed] [Google Scholar] - 39.Wenger T.L., Scott A., Kruidenier L., Sikes M., Keefe A., Buckingham K.J., Marvin C.T., Shively K.M., Bacus T., Sommerland O.M., et al. (2024). SeqFirst: Building equity access to a precise genetic diagnosis in critically ill newborns. medRxiv, 2024.09.30.24314516. 10.1101/2024.09.30.24314516. [DOI] [Google Scholar] - 40.Sigurpalsdottir B.D., Stefansson O.A., Holley G., Beyter D., Zink F., Hardarson M.Þ., Sverrisson S.Þ., Kristinsdottir N., Magnusdottir D.N., Magnusson O.Þ., et al. (2024). A comparison of methods for detecting DNA methylation from long-read sequencing of human genomes. Genome Biol. 25, 69. 10.1186/s13059-024-03207-9. [DOI] [PMC free article] [PubMed] [Google Scholar] - 41.Kokot M., Dlugosz M., and Deorowicz S. (2017). KMC 3: counting and manipulating k-mer statistics. Bioinformatics 33, 2759–2761. 10.1093/bioinformatics/btx304. [DOI] [PubMed] [Google Scholar] - 42.Li H. (2018). Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics 34, 3094–3100. 10.1093/bioinformatics/bty191. [DOI] [PMC free article] [PubMed] [Google Scholar] - 43.Danecek P., Bonfield J.K., Liddle J., Marshall J., Ohan V., Pollard M.O., Whitwham A., Keane T., McCarthy S.A., Davies R.M., et al. (2021). Twelve years of SAMtools and BCFtools. Gigascience 10. 10.1093/gigascience/giab008. [DOI] [Google Scholar] - 44.Camacho C., Coulouris G., Avagyan V., Ma N., Papadopoulos J., Bealer K., and Madden T.L. (2009). BLAST+: architecture and applications. BMC Bioinformatics 10, 421. 10.1186/1471-2105-10-421. [DOI] [PMC free article] [PubMed] [Google Scholar] - 45.Benjamin D., Sato T., Cibulskis K., Getz G., Stewart C., and Lichtenstein L. (2019). Calling Somatic SNVs and Indels with Mutect2. Bioinformatics. [Google Scholar] - 46.rtg-tools: RTG Tools: Utilities for accurate VCF comparison and manipulation (Github; ). [Google Scholar] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Data Availability Statement This study was conducted as part of the NIH Common Fund consortium initiative, Somatic Mosaicism across Human Tissues (SMaHT). The benchmark datasets described in this study are available through dbGaP (http://www.ncbi.nlm.nih.gov/gap) under the study accession number phs004193. The data used in this work was provided by the SMaHT Data Analysis Center (DAC) [1UM1DA058230] on behalf of the SMaHT Network. More information about the SMaHT Network is available online at https://smaht.org/, about the SMaHT Data Portal at https://data.smaht.org/, and types of data generated by the Network at https://data.smaht.org/about/consortium/data. The HPRC pangenome-graph derived variant calls used for the mitochondrial truth set are available from the HPRC data portal (https://data.humanpangenome.org/alignments). The PacBio and ONT SeqFirst-neo data used for NUMT analysis are available through the GREGoR Consortium: Genomics Research to Elucidate the Genetics of Rare Disease dbGaP study (accession no. phs003047) in NHGRI’s Analysis Visualization and Informatics Lab-space (AnVIL) platform (https://explore.anvilproject.org/datasets). MitoScope code has been deposited in GitHub at https://github.com/czakarian/mitoscope and is publicly available as of the date of publication. Analysis scripts used to create and summarize results have been deposited in GitHub at https://github.com/czakarian/smaht_mitoscope_analysis and are publicly available as of the date of publication. Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request.