Finding. Existing whole-exome and whole-genome files already contain mitochondrial DNA reads. Schecter, Lee, Davis and colleagues ran a Mitoverse Fusion workflow on 54,151 Mount Sinai participants and rebuilt the same logic under All of Us Workbench constraints for 197,361 more. The combined call set is 12.9 million variant observations from 251,512 people. No one was resequenced.
Why this paper matters
Population genetics has a mitochondrial blind spot that is logistical, not conceptual. gnomAD shipped without mtDNA until a dedicated pipeline was written. UK Biobank and other giants have shown that haplogroup, heteroplasmy, and copy number associate with common disease and mortality, but most institutional and national resources still hand investigators a nuclear VCF and leave chromosome M on the cutting-room floor.
That is wasteful. Mitochondrial DNA is 16,569 bases, maternally inherited, present in hundreds to thousands of copies, and frequently heteroplasmic. Pathogenic variants are more common in the general population than clinically recognized mitochondrial disease, which means incomplete penetrance is hiding in biobanks that already paid for the sequencing. The barrier is that nuclear germline callers are the wrong tool: the genome is circular, nuclear mitochondrial sequences (NUMTs) generate false positives, and the variants you care about often sit at a few percent heteroplasmy.
This preprint does not invent a caller. It shows that the existing Mitoverse mtDNA-Server 2 Fusion stack (GATK Mutect2 plus the mitochondria-specific caller mutserve2) can be made to travel: native Nextflow on an institutional cluster, and a from-scratch cloud rewrite when a national trusted research environment forbids Docker. That portability is the product.
What they actually measured
Mount Sinai Million: whole-exome off-target mitochondrial reads from 54,151 people, processed in six batches of about 4,300 on Minerva. Output: 3,497,139 observations, 16,128 unique variants, mean 64.6 calls per participant, median 58 (IQR 37–84). Median mitochondrial depth 410× (IQR 219–874×). 92% of observations passed quality filters. Ancestry projection onto HapMap3 gave AFR 15,644, AMR 8,096, EAS 2,535, EUR 21,059, SAS 2,312, plus 4,505 unassigned.
All of Us Controlled Tier v8: whole-genome CRAMs from 197,361 participants in the three largest genetically inferred groups (EUR 125,845, AMR 33,166, AFR 38,350), chosen around a SNOMED-wide phenotypic net. Because Researcher Workbench will not run the containerized Nextflow, the authors streamed chrM with samtools, called with Mutect2 and mutserve2 against slightly different references (GRCh38 chrM versus rCRS), normalized with bcftools, kept PASS variants, took indels from Mutect2 only, and emitted Mitoverse-shaped annotated tables. Batches of about 20,000 on 96 CPUs / 360 GB RAM took 60–72 hours and about US$300 each. Output: 9,461,434 observations, 34,142 unique variants, mean 47.9 per person, median depth 2,401×.
Together: 12,958,573 observations. In both cohorts roughly 61% of unique variants appear in five or fewer people; 88% (MSM) and 94% (AoU) have cohort frequency below 0.1%. Most observations are near-homoplasmic, 87.7% above 95% heteroplasmy in the exomes versus 67.4% in the genomes. The authors correctly refuse to call that a biological difference. Deeper WGS simply sees more low-level heteroplasmy.
How to read the score
High importance, high confidence, methods paper. The numbers are large, the environments are the two that matter (institutional HPC versus a locked national workbench), and the scientific claim is modest in the right way: they did not discover a new mitochondrial gene. They removed the reason those genes stay unanalyzed.
What they did not do is equally important. No shared sample was run through both implementations, so we do not have a concordance table. WES depth is off-target and kit-dependent. The 3% floor is a choice, not a law. NUMTs are mitigated, not abolished. The genotypes stay behind access control. Score this as infrastructure that unblocks PheWAS and rare-disease reanalysis, not as a map of mitochondrial pathogenesis.
What to do with it
If you sit on WES or WGS and do not ship an mtDNA call set, this is now the paper you either implement or cite as the gap. Copy Fusion, keep the 3% floor unless you have a reason, emit haplogroup and contamination fields, and do not meta-analyze MSM against AoU heteroplasmy as if they were the same assay. The scientific prize is downstream: how many undiagnosed mitochondrial-disease alleles, and which common-disease PheWAS hits, are sitting in these 251,512 people now that chromosome M is no longer a discarded contig.