The mitochondrial genetic bottleneck

Advanced

Genetic drift, selection or both?

A central question of mtDNA genetics was posed long ago: are pathogenic variants transmitted like neutral ones, by random drift alone, or are they subject to selection (Chinnery et al., 2000)? Current evidence indicates that the answer depends on the variant and the developmental stage (Xie et al., 2026). The bottleneck hypothesis itself arose from observations in Holstein cows, in which a newly identified mtDNA variant shifted from heteroplasmy to homoplasmy within a few generations; for neutral variants this genetic bottleneck merely accelerates random drift, whereas for deleterious variants the distribution in offspring is skewed, pointing to selection (Xie et al., 2026).

Effective size of the segregating pool

Population-genetic models compare the variance of heteroplasmy among offspring with p₀(1 − p₀), where p₀ is the starting level, and use the ratio to estimate the effective number of segregating units. Otten and colleagues described segregation in oocytes, zygotes and blastomeres of carriers with the Kimura distribution and obtained bottleneck-size estimates ranging from 10 to 145 between carriers (Otten et al., 2018). This means that even if two mothers have the same heteroplasmy level, their risk of passing disease to a child may differ substantially. The effective size summarises many processes — the number of segregating units and the number of rounds of division and replication — and is not equal to the physical copy number.

Physical mtDNA copy number

Floros and colleagues showed that primordial germ cells of human embryos at CS12 contain about 1,000-fold less mtDNA than oocytes (median 1,425 copies per cell versus a mean of 1.22×10⁶ in oocytes), with approximately five mtDNA molecules in each mitochondrion; rare variants reached higher heteroplasmy levels after completion of migration, consistent with a bottleneck (Floros et al., 2018). Copy number in human primordial germ cells was lower than in mice at corresponding stages; the authors infer a “tighter” bottleneck and faster segregation in humans (Floros et al., 2018). In mice, Cree and colleagues explained rapid segregation by a reduction in the number of molecules during embryogenesis (Cree et al., 2008), whereas Wai and colleagues concluded that the bottleneck arises during postnatal folliculogenesis through replication of a subpopulation of genomes (Wai et al., 2008). The disagreement on timing and mechanism illustrates that no single model fully resolves the mechanism. According to the review by Xie and colleagues, mathematical modelling with the measured copy number and number of cell divisions showed that the reduction in copy number alone is sufficient to explain the spread of heteroplasmy in offspring, even without selection; the biological meaning of the bottleneck is still debated — it may make harmful mutations susceptible to natural selection (Xie et al., 2026).

A possible role for replication and selection

• Selection in primordial germ cells. Floros and colleagues found signatures of selection against non-synonymous, tRNA-gene and D-loop variants, coinciding with a shift to oxidative metabolism (Floros et al., 2018).
• Preferential propagation of mutant mtDNA in the oocyte. In mice carrying the m.5024C>T mutation, the proportion of mutant mtDNA increased as oocytes matured (without cell division), but the accumulation of critically high levels was blocked by purifying selection; a similar pattern — preferential transmission at low and selection at high load — has been described in pedigrees with m.3243A>G, m.8344A>G and m.8993T>G/C (Zhang et al., 2021).
• Purifying selection and age. In mice carrying mutations in complex I genes generated by base editing, purifying selection occurred predominantly during postnatal oocyte maturation and intensified as the mother aged (Ru et al., 2024). These are mouse data; in human pedigree studies, by contrast, the number of new mtDNA mutations in children correlated positively with maternal age, and direct observations are needed before clinical conclusions can be drawn (Xie et al., 2026).
• Human oocytes. Duplex sequencing of mtDNA in individual oocytes found no age-related increase in new mutations and pointed to frequency-dependent purifying selection (Arbeithuber et al., 2025); a preprint re-analysing the same data reaches a different conclusion: positive selection predominates on average, for non-coding control-region mutations and, according to the authors, even coding mutations (Franco et al., 2026, preprint).
• Different windows of selection. In mice, elimination of defective mtDNA is tied to different developmental stages depending on the specific mutation. Selection may begin after the primordial germ cell and oocyte stages (m.3875delC), during follicle maturation (m.2820G>A, m.12918G>A), at the secondary follicle stage (m.3177G>A), or in early embryogenesis through mitophagy (m.5019A>G). It is not yet clear how much of these differences reflects real biology and how much reflects the research methods used (Xie et al., 2026).
• Levels at which selection acts. Mechanisms have been described at the level of cells (follicular atresia), organelles (mitochondrial fission, mitophagy) and molecules (preferential replication); most data come from mice, Drosophila and nematodes (Xie et al., 2026).
• Population data. Pathogenic mtDNA variants with low heteroplasmy levels occur in roughly 1 in 200 people, whereas mtDNA disease affects roughly 1 in 5,000–10,000; this discrepancy points to selection against transmission of deleterious variants. Human data are less consistent than mouse data: large deletions are rarely inherited; in a large study of mother–child pairs, mutations in rRNA genes were transmitted to offspring less often than expected, while variants in the D-loop region were inherited neutrally or with an advantage (Xie et al., 2026).
• Variant specificity in carriers. For m.3243A>G the data were better described by a Kimura distribution truncated at 76.5% (no high loads); for m.8993T>G values were predominantly extreme and, in pedigree data, offspring levels were more often higher than maternal levels; for m.14487T>C no signs of selection were found (Otten et al., 2018).

Differences between oocytes of one woman

A mother’s somatic heteroplasmy level cannot predict the heteroplasmy level in a particular oocyte. In the study by Spath and colleagues, women with detectable somatic heteroplasmy had a higher proportion of embryos carrying the variant, yet they still produced embryos without a detectable variant; m.8993T>G showed skewed segregation (Spath et al., 2021). Hellebrekers and colleagues noted that most carriers of point variants will have some oocytes with loads below the 18% reference point (Hellebrekers et al., 2012).

Variant-specific distributions in embryos

The published embryo values compiled in the Between different embryos section show mostly extreme values for m.8993T>G and a substantial proportion of intermediate values for m.3243A>G. These distributions are observed data from individual families and cycles; they are not population probabilities and do not account for differences between families.

Implications for PGT-mt

• Whether PGT-mt is feasible depends on the spread of heteroplasmy among a particular patient’s embryos, not on the average for a specific pathogenic mtDNA variant.
• Mathematical models estimate risks and chances of success before a treatment cycle using maternal heteroplasmy as an input; the model was developed using pedigree data (Ji et al., 2023); these estimates rest on assumptions about bottleneck size and thresholds.
• In the Newcastle programme, five patients undergoing PGT obtained no embryo with heteroplasmy below 30% (Hyslop et al., 2025). This observation shows that PGT-mt does not always yield an embryo that meets the chosen criterion.

No single mechanism — genetic drift, selection or restricted replication — has been established as a universal explanation of the bottleneck across all variants and developmental stages. The relative contribution of each mechanism has not been measured in humans and probably differs between variants and developmental stages.