Avoiding misleading estimates using mtDNA heteroplasmy statistics to study bottleneck size and selection

Abstract

Mitochondrial DNA (mtDNA) heteroplasmy samples can shed light on vital developmental and genetic processes shaping mtDNA populations. The sample mean and sample variance of a set of heteroplasmy observations are typically used both to estimate bottleneck sizes and to perform fits to the theoretical “Kimura” distribution in seeking evidence for mtDNA selection. However, each of these applications raises problems. Sample statistics do not generally provide optimal fits to the Kimura distribution and so can give misleading results in hypothesis testing, including false positive signals of selection. Using sample variance can give misleading results for bottleneck size estimates, particularly for small samples. These issues can and do lead to false positive results for mtDNA mechanisms – all published experimental datasets we re-analysed, reported as displaying departures from the Kimura model, do not in fact give evidence for such departures. Here we outline a maximum likelihood approach that is simple to implement computationally and addresses all of these issues. We advocate the use of maximum likelihood fits and explicit hypothesis tests, not fits and Kolmogorov-Smirnov tests via summary statistics, for ongoing work with mtDNA heteroplasmy.

Multiple distinct evolutionary mechanisms govern the dynamics of selfish mitochondrial genomes

“…To this end, prior work generally consists of either theoretical modeling or empirical study of mitochondrial mutations, 29,43,[53][54][55] making it difficult to integrate theory with experiment. Although some modeling studies incorporate empirical data, 29,[55][56][57] these focus on heteroplasmy dynamics within organisms, or within parent-progeny lineages, making it difficult to combine the levels of selection into a complete evolutionary picture. To address these challenges, we developed a hybrid approach that integrates theoretical modeling with experiments designed to individually probe the levels of selection and the mutant frequency distribution.…”

Single-cell analysis reveals context-dependent, cell-level selection of mtDNA

“…We then asked whether the heteroplasmy shifts that we observed in LHON-edited or SILENT-edited cells can be explained by a simple model of drift based on the ‘Kimura’ distribution 36 , 37 . We found that the maximum likelihood fit of the Kimura distribution to our observed silent heteroplasmy distribution was very good (Extended Data Fig.…”

Stochastic organelle genome segregation through Arabidopsis development and reproduction

“…For numerical convenience, we used a population size of N e = 50 in the numerical simulations. Following the usual parameterisation of the Kimura distribution for mtDNA work (Wonnapinij et al, 2008;Giannakis et al, 2023)…”

“…1a). Importantly, our model considers individual heteroplasmy measurements (rather than fitting using coarse-grained, uncertain summary statistics like heteroplasmy variance), increasing the statistical power of our approach (Giannakis et al, 2023).…”

“…S4). Because the statistical approach considers individual heteroplasmy observations, rather than losing information with summary statistics like the sample variance, substantial statistical power is retained to infer parameters and select mechanisms (Giannakis et al, 2023).…”