Blindspot Genetics — The mitochondrial genome

If you arrived here by searching for a gene such as MT-ND1, MT-ND4, MT-ND4L or MT-ND6 — the genes a Leber hereditary optic neuropathy panel targets — the short answer is that there is no gene card for them and there is not going to be one until the underlying analysis exists. The longer answer is worth reading, because “no card” here does not mean “nothing to worry about”. It means the opposite.

What we hold for chrM

Nothing. Measured, not assumed.

chrM rows in refseq_select_transcripts

chrM rows in gene_callability

chrM rows in the GIAB stratifications

chrM rows in our ClinVar load

chromosomes on the panel — chr1–22, X, Y

mitochondrial genes for which we do hold ClinGen curation

The panel is built on nuclear genes with a RefSeq Select transcript. Mitochondrial genes do not have one, so the contig is excluded by the panel’s own definition rather than being queued behind it. Of the 164 HGNC protein-coding genes we do not cover (measured 2 October 2026, matched by HGNC id), 13 are mitochondrial and 151 are nuclear. If the search had told you that Blindspot covers 19,133 of 19,297 protein-coding genes and left it there, you would reasonably have concluded your gene was in a small unlucky remainder that will be filled in next quarter. It is not, and that sentence has been changed.

Why the gene-card framework does not transfer

Every class on this site — Reliable through Refractory — is computed from benchmark confidence on GRCh38 and T2T-CHM13, difficult-region overlap, and the share of coding sequence the benchmark actually scored. None of those inputs exists for chrM. A mitochondrial gene card would therefore have to display a class derived from nothing, which is the single failure this product exists to prevent. We would rather show no card than a confident one.

That is not the whole story either, because the mitochondrial genome is not merely unmeasured by us — it fails differently. Short-read variant calling on chrM is not diploid genotyping. It is allele-fraction estimation on a high-copy circular genome whose sequence also exists, in fragments, inside the nuclear genome. The three findings below are the ones a reader should take away, and none of them is ours.

The blind spots that do apply

Nuclear look-alike sequence limits where low-level heteroplasmy can be called at all — and it depends on your read length.

Segments of mitochondrial sequence also sit in the nuclear genome (NUMTs). A read drawn from one can align to chrM and appear as a low-abundance heteroplasmy that is not there. Albayrak and colleagues mapped where nuclear and mitochondrial sequence are identical or near-identical, and found the proportion of the mitochondrial genome in which nuclear sequence does not interfere rises steeply with read length:

Single-end read	Genome free of nDNA interference
36 bases	20.4%
76 bases	44.5%
150 bases	67.9%
417 bases	100%

At 150-base single-end reads — a common clinical configuration — roughly a third of the mitochondrial genome still cannot support a low-abundance heteroplasmy call on sequence alone. The authors also report up to a 25-fold variation in the length of the longest shared subsequence across the genome, so the problem is strongly positional rather than uniform.

Albayrak et al., BMC Genomics 2016;17(1):1017 · 10.1186/s12864-016-3375-x · via PubMed

In the largest population reference, NUMT-derived false positives could account for most apparent heteroplasmies — so it reports none below 10%.

Building mtDNA frequencies for 56,434 genomes, the gnomAD team reported that mtDNA copy number per cell varied across cohorts and drove the fraction of NUMT-derived false-positive calls, which “can account for the majority of putative heteroplasmies”. They excluded samples prone to NUMT misalignment and report only variants at heteroplasmy of 10% or above. 98% of the calls that survive are homoplasmic. They also found that about 1 in 250 individuals carries a pathogenic mtDNA variant at heteroplasmy above 10%. Two consequences for a report in front of you: a population frequency for a low-level heteroplasmy usually does not exist, and a pathogenic mtDNA variant is not a rare thing to encounter.

Laricchia et al., Genome Research 2022;32(3):569–582 · 10.1101/gr.276013.121 · via PubMed

Specific positions are known to be untrustworthy, and they are named.

gnomAD’s v3.1 mtDNA release applies an artifact_prone_site filter that removes all calls at chrM:301, 302, 310, 316, 3107 and 16182, where sequence context makes true variants hard to separate from technical artifact, and flags variants as common_low_heteroplasmy where they are “likely to be enriched for sequencing errors and NUMT misalignments”. Two homopolymer stretches are called out by name, chrM:300–317 and chrM:16180–16193. This is the same shape as the per-exon difficult regions the gene cards show — a named, positional, reproducible reason to distrust a call — which is precisely why it belongs on a future mitochondrial card rather than in a paragraph.

gnomAD v3.1 mitochondrial DNA release notes, Broad Institute · gnomad.broadinstitute.org

And mtDNA variants are not interpreted by the ordinary rules.

The ClinGen Mitochondrial Disease Variant Curation Expert Panel published a specification of the ACMG/AMP standards for mtDNA, because maternal inheritance, heteroplasmy and its threshold effect, the absence of splicing, and haplogroup context all change how evidence is weighed. A laboratory applying the unmodified 2015 criteria to a mitochondrial variant is not following current practice.

McCormick et al., Human Mutation 2020;41(12):2028–2057 · 10.1002/humu.24107 · via PubMed

The one we cannot measure, and will not assert.

The question a clinician actually wants answered is whether the test in front of them analyzed chrM at all, at what heteroplasmy floor, and in which tissue. Those are properties of a laboratory’s pipeline and specimen, not of a gene or a reference genome, and Blindspot holds no data on them. We are not going to estimate them. If a report is silent on mitochondrial analysis, that silence is the thing to ask about — it is not evidence that nothing was found.

Blindspot Genetics — stated limitation, nothing measured

All 37 mitochondrial genes

What we can measure on chrM, and what it does not tell you.

Read this before reading the table

These are not the gene-card axes, and there is no callability class here on purpose. The nuclear categories earned their place empirically: GIAB measured which regions short reads actually get wrong against a truth set, at roughly 30× coverage. chrM is not sequenced at 30× — mtDNA copy number per cell puts a whole-genome run at hundreds to thousands of ×, and depth is precisely the variable those categories were calibrated against. GC content in particular predicts much less here than it does on a nuclear card. Homopolymer transfers best, because polymerase slippage is systematic error and depth does not wash it out. The table is deliberately not styled like a gene card so that a reader calibrated on those cards does not carry that calibration across.

The result worth knowing before the table. gnomAD names six chrM positions it filters outright and two homopolymer stretches where sequencing errors concentrate. Mapped onto the annotation, both named homopolymer runs and five of the six positions fall outside every one of the 37 genes — they are in the control region. Exactly one, chrM:3107, lies inside a gene: MT-RNR2, which is also the gene carrying the rCRS placeholder base.

That is reassuring, and it is not the whole story. The longest homopolymer run in the entire mitochondrial genome — A×8 at chrM:12418–12425 — sits inside MT-ND5, which ClinGen rates Definitive for both Leigh syndrome and mitochondrial disease. gnomAD does not name it, because gnomAD names the positions where it saw trouble across 56,434 genomes rather than every run that could cause it. So: the known artifact hotspots avoid the genes; the worst run by sequence does not.

Gene	Span on rCRSRefSeq annotation	Strand	bp	GC%computed · weak predictor here	Longest homopolymercomputed · transfers best	Runs ≥5computed	Tandem bpcomputed	ClinGen gene–disease validityour stored curation
13 protein-coding genes
MT-ND1	3307–4262	+	956	47.7	6	7	180	Definitive Leigh Definitive mito disease
MT-ND2	4470–5511	+	1042	43.0	7	9	209	Limited Leigh Moderate mito disease
MT-CO1	5904–7445	+	1542	46.2	7	5	263	Limited Leigh Definitive mito disease
MT-CO2	7586–8269	+	684	46.2	5	5	133	Limited Leigh Definitive mito disease
MT-ATP8	8366–8572	+	207	39.6	6	5	71	Limited mito disease
MT-ATP6	8527–9207	+	681	44.2	5	4	109	Definitive Leigh Definitive mito disease
MT-CO3	9207–9990	+	784	46.6	7	6	180	Limited Leigh Definitive mito disease
MT-ND3	10059–10404	+	346	40.2	5	3	87	Definitive Leigh Definitive mito disease
MT-ND4L	10470–10766	+	297	43.1	4	0	44	Limited mito disease
MT-ND4	10760–12137	+	1378	44.3	7	11	288	Definitive Leigh Definitive mito disease
MT-ND5	12337–14148	+	1812	44.9	8	10	377	Definitive Leigh Definitive mito disease
MT-ND6	14149–14673	-	525	42.7	7	7	157	Definitive Leigh Definitive mito disease
MT-CYB	14747–15887	+	1141	46.3	5	2	214	Definitive mito disease
2 ribosomal RNA genes
MT-RNR1	648–1601	+	954	45.5	6	3	246	Definitive mito disease
MT-RNR2	1671–3229	+	1559	42.8	7	9	361	Limited mito disease
22 transfer RNA genes
MT-TF	577–647	+	71	40.9	4	0	26	Definitive mito disease
MT-TV	1602–1670	+	69	42.0	0	0	9	Moderate Leigh Definitive mito disease
MT-TL1	3230–3304	+	75	38.7	4	0	14	Limited Leigh Definitive mito disease
MT-TI	4263–4331	+	69	29.0	5	1	24	Moderate HCM Limited Leigh Definitive mito disease
MT-TQ	4329–4400	-	72	48.6	4	0	17	Limited mito disease
MT-TM	4402–4469	+	68	45.6	4	0	26	Definitive mito disease
MT-TW	5512–5579	+	68	35.3	0	0	18	Limited Leigh Definitive mito disease
MT-TA	5587–5655	-	69	43.5	4	0	17	Definitive mito disease
MT-TN	5657–5729	-	73	41.1	0	0	15	Definitive mito disease
MT-TC	5761–5826	-	66	45.5	4	0	14	Limited mito disease
MT-TY	5826–5891	-	66	42.4	5	1	19	Definitive mito disease
MT-TS1	7446–7514	-	69	49.3	6	3	27	Definitive mito disease
MT-TD	7518–7585	+	68	23.5	5	2	31	Moderate mito disease
MT-TK	8295–8364	+	70	32.9	4	0	19	Definitive Leigh Definitive mito disease
MT-TG	9991–10058	+	68	26.5	6	1	17	Moderate mito disease
MT-TR	10405–10469	+	65	23.1	4	0	28	Moderate mito disease
MT-TH	12138–12206	+	69	36.2	4	0	17	Definitive mito disease
MT-TS2	12207–12265	+	59	45.8	5	1	11	Definitive mito disease
MT-TL2	12266–12336	+	71	35.2	5	1	27	Disputed Leigh Definitive mito disease
MT-TE	14674–14742	-	69	39.1	5	1	8	Definitive mito disease
MT-TT	15888–15953	+	66	37.9	5	1	18	Moderate mito disease
MT-TP	15956–16023	-	68	33.8	5	1	20	Definitive mito disease

Provenance, column by column. Span, length and strand are the RefSeq NC_012920.1 annotation. GC, homopolymer and tandem-repeat columns are computed here from the rCRS by scripts/mito_sequence_metrics.py, circular-aware, and checked before use: all 13 coding sequences translate under the vertebrate mitochondrial code with zero internal stop codons, so the sequence and the coordinates agree. ClinGen is our stored curation. Nothing in this table was typed by hand.

Two things the table cannot do, and one it must not

You cannot add these rows up. The mitochondrial genome has genuinely overlapping reading frames: MT-ATP8 and MT-ATP6 share 46 bp, MT-ND4L and MT-ND4 share 7 bp, and three tRNA pairs share 1–3 bp. The gene lengths sum to 15,416 bp but cover only 15,358 distinct bases, so 58 bp belong to two rows at once. Any “percent of the mitochondrial genome that is X” computed by summing this table is wrong by construction. 92.7% of the molecule is inside a gene, which is nothing like the nuclear picture.

The coordinates wrap. The control region is annotated 16024–17145 on a 16,569 bp molecule — it runs off the end and returns to the origin, because the genome is circular. A linear containment test reports positions 301–316 as intergenic when they are in the control region; that is how the first version of this analysis got it wrong, and every span operation behind this table now wraps.

And the column that matters most is not here. Nothing in this table measures NUMT interference, mappability, or benchmark confidence, because none of those can be computed from a 16,569 bp sequence alone — the first two are relationships to the nuclear genome and the third does not exist for chrM. The read-length effect quoted further up this page is a whole-genome average from a publication, not a per-gene figure of ours. Treating the columns below as a substitute for it would be the error this page exists to prevent.

A note on counting: the RefSeq annotation of NC_012920.1 carries 37 genes — 13 protein-coding, 2 rRNA, 22 tRNA — which is the set below. HGNC additionally lists MT-LIPCAR, a mitochondrially encoded long non-coding RNA that the annotation does not carry as a gene feature and ClinGen has not curated, so it has no coordinates here and no row.

What covering chrM would actually take

Stated plainly so that the gap is auditable rather than vague. A mitochondrial tier is not a matter of loading 37 more rows; it needs a different set of axes and at least three sources we do not currently hold: a positional NUMT map, so the read-length effect above can be shown per gene rather than as a whole-genome average; the gnomAD mtDNA frequency and filter data, which is public and is the single most useful missing input; and a benchmark for chrM, which the GIAB stratifications we use do not provide. Until those exist here, a mitochondrial gene card would be decoration.

Sources

Albayrak L, Khanipov K, Pimenova M, Golovko G, Rojas M, Pavlidis I, Chumakov S, Aguilar G, Chávez A, Widger WR, Fofanov Y. The ability of human nuclear DNA to cause false positive low-abundance heteroplasmy calls varies across the mitochondrial genome. BMC Genomics 2016;17(1):1017. doi:10.1186/s12864-016-3375-x. Retrieved via PubMed.

Laricchia KM, Lake NJ, Watts NA, et al. Mitochondrial DNA variation across 56,434 individuals in gnomAD. Genome Research 2022;32(3):569–582. doi:10.1101/gr.276013.121. Retrieved via PubMed.

McCormick EM, Lott MT, Dulik MC, et al. Specifications of the ACMG/AMP standards and guidelines for mitochondrial DNA variant interpretation. Human Mutation 2020;41(12):2028–2057. doi:10.1002/humu.24107. Retrieved via PubMed.

gnomAD v3.1 Mitochondrial DNA Variants. Broad Institute, November 2020. gnomad.broadinstitute.org.

Counts attributed to Blindspot on this page were measured against the live database on 19 September 2026 and are reproducible from tests/mito_clingen.sql. Figures attributed to a publication are that publication’s, not ours.