Table 1. Key results.

	Key results presented in this paper.
$\mathbb{I}$	If only one mtDNA species is controlled the variance of the controlled species reaches a constant value. When both species are controlled with equal strength their variances increase at identical rates, and, in general, the more tightly controlled species has a more slowly increasing variance (Fig 1). [D]
$\mathbb{\text{II}}$	The mean energetic cost of maintaining a tissue can increase over time due to the nonlinear influence of mtDNA variance, even if the energetic demand on the tissue stays the same and mean levels of mtDNA are constant (Eq 6). [D]
$\mathbb{\text{III}}$	Intermediate heteroplasmy states can be more expensive than states homoplasmic in either mutant or wildtype. [C]
$\mathbb{\text{IV}}$	A control lacking any mutant contribution can show an exponentially increasing cost, and the effects of particular cellular control strategies are more pronounced in low copy number cells (Fig 3A and 3B). [D, C]
$\mathbb{\text{V}}$	Control strategies based on the energy status of the cell can often outperform control based on mtDNA copy number or sensing mtDNA mass (which would work well for deficient deletion mutants, but would be suboptimal for deficient point mutations) (Fig 3C). [D,C]
$\mathbb{\text{VI}}$	Even for pathological mutants, reduction of mutant mtDNA alone is not always the optimal control strategy for a cell to adopt (Fig 4). [C]
$\mathbb{\text{VII}}$	Tissues with high mean heteroplasmy levels will generally be harder to treat with mitochondrially targeted endonucleases if the heteroplasmy variance is high, especially if this high mean level is caused by a small percentage of cells (Fig 6A and 6B). [D,T]
$\mathbb{\text{VIII}}$	Weak long-term rather than short intense endonuclease treatments are more likely to beneficially impact mtDNA populations (Fig 6D and 6E). [D,T]

Here we present key results of this paper, which hold under the assumptions used in our models (see text and Discussion). We place in square brackets the models we invoke for each part: D—our model for mitochondrial dynamics; C—a particular illustrative family of cost functions; T—a model for gene therapy.