Heteroplasmy is the presence of more than one mitochondrial sequence state within the biological material being examined. The definition is simple; the measurement is not. A reported value is shaped by tissue sampling, the number of original molecules, extraction and amplification, sequencing chemistry, alignment, error correction, and caller thresholds. It should be read as an estimate produced by a specified assay, not as a fixed property of an entire person or organism.
This distinction matters at low fractions, where technical artifacts can be as numerous as genuine minority molecules, and at high fractions, where a consensus caller may hide a substantial minor state by reporting only the majority base.
Heteroplasmy exists at several scales
A tissue sample pools cells. Each cell can contain many mitochondria, and each mitochondrion can contain multiple mtDNA molecules. “20% heteroplasmy” could describe a roughly even minority fraction in many cells or a mixture of cell populations with very different fractions. Bulk sequencing usually cannot distinguish those distributions.
Terms should be tied to the level observed:
- Molecular heteroplasmy refers to sequence states among mtDNA molecules.
- Cellular heterogeneity refers to differences among cells.
- Tissue heterogeneity refers to differences among sampled tissues or regions.
- Length heteroplasmy describes coexisting length states, often in repetitive tracts.
- Point heteroplasmy describes nucleotide substitutions at a position.
Homoplasmy is also operational rather than absolute. An assay that observes no alternate state above a 5% detection limit cannot establish that every molecule is identical; it establishes that no alternate state met the assay’s reporting rule.
Variant fraction is an estimator
With $k$ alternate-supporting reads among $n$ usable reads, a naive estimate is
$$\hat{f} = \frac{k}{n}.$$If reads were independent, unbiased observations with negligible error, binomial uncertainty would shrink as $n$ increases. Real data violate those assumptions. PCR duplicates can descend from the same original molecule, strand and sequence-context biases can change allele balance, and the error rate may be similar to the fraction being measured. Ten alternate reads among 1,000 are not compelling if they all occur on one strand, share one start site, or match a known platform artifact.
The relevant denominator is the number of usable observations after predefined filters, not headline depth. Reports should distinguish raw depth, filtered depth, unique molecular depth if UMIs are used, and the counts supporting each state on each strand. Confidence intervals quantify sampling uncertainty under a model; they do not absorb systematic bias or contamination.
Biology makes the fraction contextual
mtDNA populations can shift through replicative segregation, genetic bottlenecks, drift, and selection. Fractions can differ among tissues and can change over time. The same variant may be readily observed in one specimen type and undetectable in another. Sample age, storage, extraction, and cell composition can further alter what enters the assay.
Accordingly, longitudinal or cross-tissue comparisons require matched methods. A numerical difference between two laboratories, tissues, or platforms may reflect a changed limit of detection or amplification bias rather than a biological shift. Reports should not generalize from one tissue to unsampled tissues.
Technical signals that mimic heteroplasmy
Sequencing and base-calling error. Error is context- and platform-dependent. A fixed 1% threshold is not universally conservative.
PCR error and jackpotting. An early polymerase error can be amplified into many descendant reads. Low template count increases this risk.
Strand bias. Alternate support confined largely to one orientation often indicates chemistry, alignment, or filtering artifacts.
NUMTs. Nuclear mitochondrial segments can map to the mitochondrial reference and introduce consistent apparent alternate alleles. High mtDNA copy number does not eliminate this risk, especially in depleted or degraded material.
Contamination or sample mixture. A second person’s mtDNA can create multiple linked minor alleles that resemble a coherent haplotype. Haplogroup-informative positions and controls can help identify this pattern, but mixture analysis requires an appropriate method.
Index misassignment and carryover. Multiplexed runs and prior libraries can contribute low-level reads from other samples.
Alignment around indels and homopolymers. Alternative representations can redistribute support among nearby positions. Length heteroplasmy in control-region tracts is particularly sensitive to alignment and sequencing method.
Sanger peak artifacts. Dye blobs, compression, baseline noise, incomplete extension, and poor read ends can produce secondary peaks. Peak-height ratios are not automatically molecular fractions because dye response and local context are unequal.
Detection limit is not one number
A method should distinguish at least:
- Limit of blank: the signal distribution in samples expected to contain no alternate state.
- Limit of detection: the lowest fraction detected with a specified probability under defined conditions.
- Limit of quantification: the lowest fraction measured with acceptable precision and bias.
- Reporting threshold: the operational threshold chosen for the study, which may be above the analytical detection limit.
These limits are variant- and context-sensitive. Validation should include mixtures across fractions, DNA inputs, relevant sequence contexts, batches, and operators, with negative controls and replicates. A threshold copied from another platform or publication does not validate a local assay.
A defensible research workflow
- Predefine scope. State whether the study targets point or length heteroplasmy, the coordinates considered, and whether homopolymer tracts are excluded or handled separately.
- Preserve primary evidence. Keep raw reads or chromatograms, sample metadata, extraction and library batch, and controls.
- Control molecules and mappings. Record input quantity, amplification strategy, duplicate handling, mapping reference, circular-genome handling, NUMT strategy, and base/mapping-quality filters.
- Measure balanced support. Retain reference and alternate counts, total usable depth, strand balance, read-position bias, and local alignment context.
- Evaluate contamination. Review negative controls, sample-to-sample patterns, and linked minor alleles; use a method appropriate to the study design.
- Confirm according to validation. Independent extraction, amplification, strand, or orthogonal technology may be required, especially near the reporting threshold.
- Separate observation from interpretation. Report the assay estimate and uncertainty before any functional, population, or clinical discussion.
Common interpretation mistakes
Treating an IUPAC ambiguity code as a fraction. R records A/G ambiguity but does not say whether the states are 50:50, 90:10, or a low-quality unresolved call.
Calling every secondary Sanger peak heteroplasmy. The trace must meet a validated, context-aware mixture policy and usually requires confirmation.
Equating depth with sensitivity. More reads cannot rescue few original molecules or systematic error.
Comparing fractions across tissues without qualification. The specimens represent different cellular populations.
Using a population database as proof of artifact or effect. Frequency and phylogenetic context inform review but do not replace sample-level evidence.
Interpreting below the validated range. Software may emit a number more precise than the assay can support.
Review and reporting checklist
| Domain | Minimum report |
|---|---|
| Specimen | Tissue/material, collection time, storage, extraction, DNA input |
| Assay | Enrichment or primers, platform, read structure, controls, replicates |
| Reference | Sequence/build, circular mapping strategy, NUMT mitigation |
| Caller | Software/version, filters, duplicate policy, excluded regions |
| Evidence | Reference/alternate counts, usable depth, strand support, quality |
| Performance | Limit of blank, detection, quantification, and reporting threshold |
| Confirmation | Independent method or replicate policy and result |
| Result | Variant notation, fraction, uncertainty, no-call rules, tissue scope |
| Interpretation | Explicit distinction between observation and biological inference |
Where GeneFlow fits, and where it does not
GeneFlow can preserve source AB1, FASTA, FASTQ, and VCF files with sample and project metadata. For AB1 data it stores the four chromatogram channels, called bases, and Phred scores; its QC summaries and flagged regions can direct a reviewer back to weak trace intervals. It also keeps consensus sequences, HaploGrep3 results, and generated reports associated with the sample.
GeneFlow’s current shipped analysis does not detect, quantify, or validate heteroplasmy. Its FASTQ support is for sequence records with qualities, not a raw-read mtDNA variant-calling pipeline. Its VCF path passes an existing variant profile to HaploGrep3 rather than recalculating allele fractions. AB1 trace display can support manual review, but GeneFlow does not turn secondary peak heights into validated heteroplasmy fractions. Heteroplasmy claims therefore require an external assay-specific pipeline, validation evidence, and expert interpretation; GeneFlow can hold the resulting files and provenance, not substitute for that method.