Mitochondrial Damage: What Tissue Averages Can Hide

A tissue average can conceal severe local damage, but it cannot tell us how many cells have lost function. Human single-fiber studies link high local deletion burdens with respiratory-chain abnormalities. Public donor data also support an age association in deletion-related measurements. Distribution, copy number, assay coverage and functional measurements still separate those observations; they cannot be collapsed into a single percentage of “mitochondrial aging.”

This article focuses on human skeletal muscle. We reanalyze public donor and deletion-coordinate data, check the sampling denominators behind single-fiber findings, and use spatial-mixture and reserve models to explain why these measurements are not interchangeable. The evidence search extends to September 21, 2026.

01Three distinct questions

Mitochondrial DNA is abbreviated mtDNA. When some copies carry a deletion while others retain the corresponding sequence, their coexistence can be described by the mutant fraction, or heteroplasmy. Our focus is large deletions. Point mutations, mitochondrial abundance, membrane potential and every other form of mitochondrial damage are not treated as one measure.

A muscle fiber is a long, multinucleated cell. One abnormal segment does not make every part of that fiber abnormal. Similarly, mtDNA copies per diploid nucleus are not copies per entire muscle fiber. Single-fiber study Assay study

02Human evidence for sharply localized abnormalities

Bua and colleagues studied autopsy muscle from 12 donors aged 49–93 in 2006. They had no known mitochondrial myopathy, but some had cardiovascular, renal or other diseases. This was not a random sample of healthy people. Original study

The researchers first located regions lacking cytochrome c oxidase activity, or COX, while showing increased succinate dehydrogenase, or SDH, staining. All 48 such selected abnormal fibers from seven donors contained a detected mtDNA deletion. Failure to detect deletions in normal comparison fibers did not establish an absolute frequency of zero.

Quantitative profiles along the fiber came from four fibers in two donors. Deletion fractions could approach 99% in abnormal regions, with lower fractions in transition regions and adjacent normal tissue. This supports local correspondence between DNA burden and enzyme abnormalities. It does not turn 48 selected fibers into 48 independent people or estimate a universal 90% threshold.

The much-studied 4,977-base-pair “common deletion” was detected in the tissue homogenates but not in the 48 selected abnormal fibers. Another deletion, spanning 7,664 base pairs, appeared five times among the abnormal fibers despite a lower level in homogenates. Source 1

This does not make the common deletion harmless. It shows that a variant readily measured across a tissue need not be the variant dominating a selected local lesion. Sampling location, lesion selection and the assay all influence the result.

03The often-quoted 30% figure was extrapolated

The same study is frequently associated with estimates that about 6%, 22% and 31% of fibers were affected in donors of different ages. Those numbers include an extrapolation.

Across approximately 2 mm of serial sections, the investigators found abnormalities in 36 of 18,367 fibers from a 49-year-old donor, 65 of 9,285 from a 67-year-old donor, and 98 of 10,652 from a 92-year-old donor. The observed proportions were approximately 0.20%, 0.70% and 0.92%. The authors then extrapolated to fibers about 65 mm long. Source 1

Multiplying the published count proportions by 65/2 gives 6.37%, 22.75% and 29.90%, not an exact reproduction of the approximate 6%, 22% and 31% in the text. Under a different simplifying assumption of independent sampling windows, the probability of at least one abnormal region would be 1 − (1 − p)^(65/2), giving 6.18%, 20.41% and 25.95%.

Neither calculation validates the actual spatial distribution. Lesions have length, can cross window boundaries, and may cluster or recur within a fiber. The second calculation is not a validated correction to the first; it illustrates dependence on assumptions. These counts came from one donor at each age. Thousands of fibers do not justify narrow population-level confidence intervals for an age trend.

04An age association in public data is still a bulk measurement

In 2023, Vandiver and colleagues studied mtDNA deletions using nanopore long-read sequencing and compared the results with digital PCR. We obtained their supplementary sample table and public deletion-coordinate list, summarized events by donor, and modeled the association between age and event rate. Paper and supplementary data

The muscle table contains 15 male donors aged 26–81. Although the abstract and some methods text say 20–81, the actual table and Results support 26–81. We use the observed sample range.

We interpreted the released coordinates using explicit circular-genome arithmetic, counted spans longer than 2 kb, and divided by each donor's mapped mtDNA reads. An unweighted donor-level regression of the base-10 logarithm of that rate, using HC3 covariance and t intervals with 13 degrees of freedom, gave:

• A fitted rate ratio of 2.26 per 10-year age difference, with a 95% HC3 interval of 1.81–2.83.
• An in-sample R² of approximately 0.844, close to the reported 0.84; this is not validation in new donors.
• Rate ratios of approximately 2.19–2.45 when omitting one donor at a time.

The 2.26-fold estimate compares different donors of different ages. It does not show a person accumulating 2.26 times as many deletions over ten years. The small, all-male sample and its sampling and health context limit generalization. The original study also compared deletion-length thresholds in these same donors, so a high R² does not establish age-prediction performance.

The two assays on the same donors are not independent replications. The median ratio of our nanopore event rate to the ddPCR deletion fraction was about 10.9, ranging from approximately 1.3 to 51.5. The assays differ in the deletions they capture, primer or cut sites, read length and calling algorithms. A larger number does not automatically measure ten times the true damage.

During self-review we added an exploratory check. Mean mapped mtDNA read length could affect detection of large deletions, and its logarithm correlated with age at approximately 0.53. Including age and log mean read length together gave an age-associated rate ratio of about 2.46 per decade, with a 95% HC3 interval of 2.03–2.99, still positive. This conditional model in only 15 donors does not remove other technical or biological confounding and cannot be described as correcting all detection bias.

An earlier study, published online in 2020, also reported an age association in 14 male muscle samples. Its often-quoted approximately 98-fold difference is obtained by evaluating a regression at ages 20 and 80, not by following individuals for 60 years. The supplement we inspected contains primers, probes and restriction-digest validation, rather than complete individual data for refitting those 14 samples. Possible overlap with the 2023 specimens remains unresolved. Source 2

05The public file does not fully reconcile with the report

The 2023 public file contains 12,621 muscle deletion records, matching the reported total. However, its coordinates yield 2,492 spans longer than 2 kb with circular decoding, or 2,490 using plain end minus start. The author's code contains the corresponding count of 2,390; the paper's description of 100 common-deletion-like events representing 4.18% also implies that denominator. Source 3 Author code

The size range also differs. Released coordinates include zero spans, spans below 100 bp, and spans of 100–137 bp, whereas the manuscript describes a minimum of 138 bp. We preserve those rows and their original coordinates. We do not invent missing intermediate files, or claim to have repeated alignment and calling from raw sequencing reads.

The age association retained its direction under the specified sensitivity analyses: using linear spans, restricting to endpoints within the reference, or excluding conflicts with the reported size range. It also persisted when all 37 zero or sub-100-bp records were assigned to the large-deletion category as an extreme scenario. These checks cannot cover every possible calling error.

At the 3- and 5-kb thresholds, one donor has no eligible events, making the unmodified logarithm undefined. We did not silently discard that donor. We show a consistently applied one-event pseudocount at every threshold and retain the undefined no-pseudocount fits in the results.

The brain release contains only 224 records already filtered for larger deletions. It cannot be directly compared with the paper's 1,945 events across the full reported size range. There are only three brain donors in each age group, and placental deletion calls are absent from this file. Neither adds to the muscle sample size. These checks support stability of the age association in the released file under specified choices; they do not revalidate every original experiment or establish calling accuracy.

06Identical averages can conceal different local distributions

A conditional model makes the weighting problem explicit. Divide tissue into equal-volume segments. Let p be the fraction of high-burden segments, h₁ their mutant fraction, and h₀ the fraction in other segments. Let their mtDNA copy density per unit volume be r times that of other segments. An ideal, unbiased bulk mutant fraction would be:

m = [p × r × h₁ + (1 − p) × h₀] / [p × r + (1 − p)]

Here m is an ideal molecular fraction. An observed number of deletion events per read cannot simply be substituted without assay calibration. Reads may contain multiple events, and ascertainment is incomplete.

A bulk fraction of 1% could arise from 1% in every segment; from about 1.05% of segments at 95% burden with zero elsewhere; or from only about 0.35% of segments at 95% if their copy density is three times higher. If a local abnormality threshold of 90% is additionally assumed, the fraction exceeding it differs across the three scenarios.

These are checkable mathematical scenarios, not estimates of human disease prevalence or evidence for a universal 90% threshold. A long muscle fiber adds another layer: a 1-mm lesion occupies only a small part of a 65-mm fiber. The right panel assumes that 31% of equally sized fibers each contain one lesion and illustrates changes with lesion length and copy density. The 31% remains a scenario; the model does not turn it into directly measured prevalence.

The model also cautions against treating higher total mtDNA as automatically better reserve. If deleted copies account for the increase, functional capacity need not rise. Bua's abnormal segments showed local expansion of deleted copies. Source 1

07A threshold depends on more than the mutant fraction

For total copy number N and mutant fraction h, the number of full-length copies is N × (1 − h). At 90% heteroplasmy, 1,000 total copies leave 100 full-length copies; 3,000 leave 300. The same fraction need not leave the same number.

If function is assumed to require at least K effective copies, a simplified boundary for N at least as large as K is h = 1 − K/N. It depends on copy number and demand. The figure holds an assumed requirement constant while changing total copy number.

Real systems are more complex. Full-length DNA may contain other variants, while expression, translation, complex assembly, mitochondrial networks and local demand affect function. These curves do not predict ATP production, and K was not fitted to human data. They demonstrate why a mutant fraction alone, without copy number, deletion properties and functional conditions, cannot serve as a universal diagnostic threshold.

08Expansion has competing mechanistic explanations

“Deleted DNA is shorter, so it must replicate faster” is not a sufficient explanation for every local expansion. In 2022, Insalata and colleagues presented a spatial stochastic model in which greater local carrying density, random birth and death, and exchange between regions allow mutants to spread without assigning each mutant a faster replication rate. Spatial stochastic model

This is a mechanism under defined conditions. Some human spatial profiles were reused from existing single-fiber studies, not generated in a new independent cohort. Comparison with one replicative-advantage model does not eliminate every possible selection mechanism.

In 2026, Kowald and Kirkwood proposed inferring accumulation times from cross-sectional single-cell data across ages and validated the approach using synthetic data with known truth. Parameter recovery can be tested under those assumptions, but RNA-to-DNA inference, selection, observation windows and identifiability remain conditional. Methodological study

We do not present those recovered simulated times as observed human accumulation speeds. Nor do 15 bulk tissue samples constitute single-cell age trajectories. Distinguishing new mutation input, expansion of existing clones and loss of affected segments requires spatial information and perturbations that discriminate between mechanisms.

09DNA measurements do not automatically establish functional decline

A 2026 human study measured muscle mtDNA, respiration and redox responses to acute exercise in 12 younger and 10 older adults selected for health. The older group had lower mtDNA copy number and higher deletion frequency, but baseline respiration measurements did not detect a clear group difference; the reported comparison of the maximal measured ADP-supported respiration had p=0.123. Human functional study

This does not establish equivalence. The sample was small, and respiration was measured under particular substrate and ADP-supported conditions rather than across every aspect of mitochondrial function. A whole-biopsy result also cannot rule out local lesions.

Only 15 participants contributed to the redox analyses. The unadjusted association between PRDX3 change and deletion frequency was relatively strong, with reported r=0.72 and p=0.002. After including age, the deletion term had p=0.151. Candidate proteins came from exploratory screening without multiple-testing correction and were not independently validated. This does not establish that deletions independently impair exercise adaptation; the adjusted nonsignificant result does not prove no effect either.

There was only one post-exercise sampling time, and the older participants performed less absolute work. This was not a randomized trial of long-term training or slower aging. Its useful lesson here is that DNA, local enzyme staining, whole-muscle respiration and exercise performance must be measured separately and connected with evidence.

10What the evidence supports

Stronger studies would link spatial location, deletion identity, absolute copy numbers and local function within the same tissue, then determine how those changes affect overall performance. Mechanistic studies also need perturbations and time series that distinguish competing explanations.

Averages remain useful for describing burden under a defined assay. To identify where damage occurs, how severe it is, and whether it affects the person, distribution and function must be brought back into the analysis.

11Data and reproduction

Download numerical extracts, analyses, models and the five-figure reproduction package. It preserves donor metadata, extracted public coordinates, source discrepancies, sensitivity analyses and model assumptions. All figures were drawn for this article.

We independently parsed 161 original XLSX numeric cells and donor identities, checked all 12,845 released deletion rows and the table extraction for 48 selected fibers, and used a separate R implementation to check the principal statistical and model calculations. Raw sequencing alignment, human experiments and clinical functional models were not revalidated. Actual reading scope, unavailable material, conflicts and the same-author review are recorded below.