Disentangling Positive Selection from Relaxed Selection in Animal Mitochondrial Genomes

Abstract

Disentangling different types of selection is a common goal in molecular evolution. Elevated d_N/d_S ratios (the ratio of nonsynonymous to synonymous substitution rates) in focal lineages are often interpreted as signs of positive selection. Paradoxically, relaxed purifying selection can also result in elevated d_N/d_S ratios, but tests to distinguish these two causes are seldomly implemented. Here, we reevaluated seven case studies describing elevated d_N/d_S ratios in animal mitochondrial DNA (mtDNA) and their accompanying hypotheses regarding selection. They included flightless lineages versus flighted lineages in birds, bats, and insects and physiological adaptations in snakes, two groups of electric fishes, and primates. We found that elevated d_N/d_S ratios were often not caused by the predicted mechanism, and we sometimes found strong support for the opposite mechanism. We discuss reasons why energetic hypotheses may be confounded by other selective forces acting on mtDNA and caution against overinterpreting singular molecular signals, including elevated d_N/d_S ratios.

Introduction

Detecting different patterns of selection using sequence data is a key goal of molecular evolution. A popular, simple tool in such studies is the d^N/d^S ratio: the ratio of nonsynonymous to synonymous nucleotide substitution rates. Nonsynonymous or missense substitutions alter the amino acid sequence of the resulting protein and are expected to be under selection, while synonymous or silent substitutions do not and are more likely to be selectively neutral. A relative overabundance of nonsynonymous changes (d_N/d_S < 1) suggests positive selection for a gene to undergo adaptive amino acid replacement, while d_N/d_S ratios near 1 reflect neutral evolution, and d_N/d_S ratios <1 indicate purifying selection (Hughes and Nei 1989). However, this is conservative, especially applied across an entire gene (Hughes and Nei 1989). Alternatively, researchers sometimes estimate d_N/d_S ratios across a phylogeny and compare focal lineages with reference lineages (Yang and Nielsen 1998). Higher d_N/d_S ratios in focal lineages are often interpreted as intensified positive selection. Paradoxically, elevated d_N/d_S ratios can also result from relaxed purifying selection owing to slightly deleterious amino acid changes being more tolerated or unable to be purged because of inefficient selection. However, studies rarely explicitly disentangle relaxed selection from positive selection as causes of elevated d_N/d_S ratios.

Specifically, elevated d_N/d_S ratios in mitochondrial DNA (mtDNA) are often used as evidence for positive selection on energetics. Despite being originally assumed to be a strictly neutral marker (Brown et al. 1979; Ballard and Kreitman 1995; Lynch 1996), mtDNA variation has been convincingly linked to adaptation across different types of environments (James and Ballard 2003; Ballard and Whitlock 2004; Cheviron and Brumfield 2009; Ballard and Melvin 2010; Dobler et al. 2014; Camus et al. 2017; Hill 2019; Greenway et al. 2020), and up to 26% of nonsynonymous substitutions in animal mtDNA were likely fixed by adaptive evolution (James et al. 2016). Given the myriad roles mitochondria play in cellular metabolism, mtDNA variation may be widely important in animal adaptive evolution (Havird et al. 2019; Hill et al. 2019). Specific hypotheses can be readily tested because of the availability of complete metazoan mtDNA sequences (fig. 1), and complete mtDNA sequences can be directly extracted from next-generation sequencing datasets (Allio et al. 2020).

Figure 1:

Number of complete animal mitochondrial (mt) DNA genomes available in the National Center for Biotechnology Information organelle database, both per year and cumulative (accessed August 5, 2022).

Studies of selection on mtDNA offer a useful example to demonstrate the importance of disentangling positive selection from relaxed selection when interpreting d_N/d_S results. It is tempting to speculate that an increase in the d_N/d_S ratio is indicative of positive selection on mtDNA in any animal lineage that occupies an energetically interesting ecological or metabolic niche. However, mtDNA d_N/d_S ratios are also influenced by other factors. Because of its uniparental inheritance, effective haploidy, and lack of recombination, mtDNA should have a relatively small effective population size (N_e), making it prone to accumulate slightly deleterious mutations leading to higher d_N/d_S ratios (Lynch 1996; Lynch and Blanchard 1998; Neiman and Taylor 2009). Therefore, it is crucial to explicitly distinguish positive selection from relaxed selection. Increases in mtDNA d_N/d_S ratios have been interpreted as positive selection for low echolocation frequencies in bats (Zhang et al. 2021) and environmental adaptation in gentoo penguins (Noll et al. 2022), but as relaxed selection in salamanders with low metabolic rates (Chong and Mueller 2013). Some studies have specifically discriminated positive selection from relaxed selection and found consistent, global signatures of relaxed selection along with elevated d_N/d_S ratios in eusocial shrimps (Chak et al. 2020) and bivalves with doubly uniparental inheritance of mtDNA (Maeda et al. 2021).

Here, we reanalyzed seven case studies where positive or relaxed selection on mtDNA was concluded according to elevated d_N/d_S ratios. Three cases concerned the evolution/loss of flight, and four investigated metabolic innovations. We expanded taxon sampling, calculated branch-specific d_N/d_S ratios, and used RELAX (Wertheim et al. 2014) to distinguish between positive selection and relaxed selection on mtDNA. Our analyses suggest that assuming that elevated d_N/d_S ratios fit a particular narrative without explicitly testing for their causes may lead to erroneous conclusions (Gould et al. 1979).

Material and Methods

Positive Selection versus Relaxed Selection in Flighted and Flightless Lineages

The first three case studies we reexamined explored the general hypothesis that mitochondrial function is under positive selection in flighted lineages compared with flightless lineages because flight is energetically intensive. First, mtDNA in bats was hypothesized to be under positive selection compared with flightless mammals (Shen et al. 2010). We therefore compared 77 bats with 58 flightless mammals (including insectivores, carnivores, cetaceans, and ungulates, similar to Shen et al. 2010). For the second and third case studies, mtDNA in flightless birds and insects was hypothesized to be under relatively relaxed selection compared with flighted lineages (Shen et al. 2009; Mitterboeck and Adamowicz 2013). We therefore compared 49 flightless birds with 47 closely related flighted birds and 50 flightless insects with 56 closely related flighted insects. For birds, species were classified as flighted if they were at least weakly flighted (e.g., tinamous were classified as flighted). Because flight can be more plastic in insects, we categorized ambiguous species as flighted if they produced flighted, reproductive females, as mitochondria are transmitted only by females. We note that the use of explicit tests for positive selection versus relaxed selection is essential when considering flight, as hypotheses predict both gaining flight and losing flight will be associated with elevated d_N/d_S ratios, but for different reasons.

Positive Selection for Metabolic Innovations

The four other case studies hypothesized positive selection on mtDNA in lineages with energetic innovations. First, physiological redesign associated with lung volume reduction, consumption of infrequent/large meals, and venom production has been proposed to impose positive selection on mtDNA in snakes compared with lizards (Castoe et al. 2008). We therefore compared 21 snakes with 31 closely related lizards. The next two case studies hypothesized positive selection on mtDNA in two independent origins of electric fishes due to their ability to generate electric fields for communication and prey detection (Elbassiouny et al. 2020). Therefore, for South American gymnotiform electric fishes, we compared 44 gymnotiforms with 43 closely related characiforms. For the African mormyroid electric fishes, we compared 39 mormyroids with 79 closely related osteoglossiforms/clupeiforms. Finally, in the last case study, we reexamined the brain-energy hypothesis, which states that accelerated evolution of mtDNA in primates reflects adaptive evolution for enhanced brain function (Goldberg et al. 2003; Grossman et al. 2004; Osada and Akashi 2012). We therefore compared 10 primates with 20 rodents, ungulates, and carnivores. We note that in the bats, snakes, electric fishes, and primate cases, there is only a single monophylectic lineage of interest, limiting the phylogenetic power of the analyses compared with the bird and insect cases, where many independent flightless lineages were examined. Our analyses may be relatively underpowered in cases with single origins of the phenotype of interest.

Sequence Curation

For each case study, we searched the National Center for Biotechnology Information (NCBI) organelle genome database for complete mtDNA sequences from relevant species. In some datasets, we used other resources, including the EFISH Genomics 2.0 database for electric fishes (https://efishgenomics.integrativebiology.msu.edu/data/) and Campagna et al. (2019) for flightless birds, where transcriptomes/genomes were searched for mitochondrial genes using BLAST (Altschul et al. 1997). For electric fishes and flightless birds, we de novo assembled and annotated complete (or nearly complete) mitogenomes from publicly available Illumina sequencing data on NCBI’s Sequence Read Archive using MitoFinder version 1.4 with default settings (Allio et al. 2020). We therefore report new mitogenomes from six birds, 26 gymnotiforms, and nine mormyroids (GenBank accessions: BK061684–BK061730; table S1).

Disentangling Positive Selection from Relaxed Selection on mtDNA

We extracted nucleotide sequences from the 13 mitochondrial protein-coding genes in each mitogenome (as well as out-group crustacean sequences for the insect case). For each gene, we translated and aligned resulting amino acid sequences for each of the seven datasets outlined above using MUSCLE (Edgar 2004) as implemented in MEGA X (Kumar et al. 2018). Resulting alignments were manually corrected by eye and concatenated to make a mitogenome dataset representing all 13 genes. Phylogenies for the aligned, concatenated sequences were generated using RAxML (ver. 8.2.12) with the gamma WAG model of amino acid substitution and 100 rapid bootstrap replicates (-f a -# 100 -m PROTGAMMAWAG; Stamatakis et al. 2008; Stamatakis 2014). Resulting topologies were rooted according to the original publications referenced above to guide selection analyses. For flightless insects, we used a constrained topology at the level of orders based on Misof et al. (2014).

We first used branch-specific models in PAML (ver. 4.8; Yang 2007) to fit different d_N/d_S ratios on phylogenetic branches in each dataset (“model p = 2”): one for the focal/test lineage(s) (red branches in figs. 2, 3, including terminal and consensus internal branches) and one for reference lineages. Internal branches with descendants in both character states (e.g., flighted and flightless) were coded with the ancestral condition by default (e.g., flighted in insects). We compared the resulting likelihood scores from this model with scores from a model where all branches were fitted with a single d_N/d_S ratio (“model = 0”) using a likelihood ratio test. This was performed for each gene individually and the mitogenome dataset in each case study.

Figure 2:

Selection on mitochondrial genes in flightless lineages and flighted lineages. A, Trees showing focal (red; from left to right: bats, flightless birds, and flightless insects) and reference (black) lineages. B, Comparison of d_N/d_S ratios in focal lineages and reference taxa. Data are presented as the natural log of the response ratio (RR), with numbers above zero indicating higher d_N/d_S ratios in focal lineages. C, Natural log of the k parameter from the RELAX analysis, where ln k < 0 indicates relaxed selection and ln k > 0 indicates positive selection in focal lineages. Data are presented for individual mitochondrial genes and the concatenated set of all 13 mitochondrial protein-coding genes. P < .05 is shown with black points, and P > .05 is shown with gray points.

Figure 3:

Selection on mitochondrial genes in lineages with energetic innovations. A, Topology showing the number and distribution of focal (red; from left to right: snakes, mormyroids, gymnotiforms, and primates) and reference (black) lineages. B, Comparison of d_N/d_S ratios in focal lineages and reference taxa. Data are presented as the natural log of the response ratio (RR), with numbers above zero indicating higher d_N/d_S ratios in focal lineages. C, Natural log of the k parameter from the RELAX analysis, where ln k < 0 indicates relaxed selection and ln k > 0 indicates positive selection in focal lineages. Data are presented for individual mitochondrial genes and the concatenated set of all 13 mitochondrial protein-coding genes. P < .05 is shown with black points, and P > .05 is shown with gray points.

To distinguish between positive selection and relaxed selection in focal lineages, we performed analyses using RELAX (Wertheim et al. 2014), both locally within the HyPhy package (Pond et al. 2005) and on the Datamonkey web server (Weaver et al. 2018). Phylogenetic branches were coded as either test or reference as above (with reference indicating the ancestral condition). Theoretically, when selection is “intensified,” some sites experience stronger positive selection (d_N/d_S > 1), while others experience stronger purifying selection (d_N/d_S < 1), and the distribution of d_N/d_S ratios diverges. When selection is relaxed, the distribution of d_N/d_S ratios converges toward one. Briefly, RELAX compares the distribution of d_N/d_S ratios across sites in test branches versus reference branches, summarized by a k parameter, with k < 1 suggesting relaxed selection and k > 1 suggesting intensified/positive selection in test branches. The statistical significance of the k parameter is assessed by comparison to a model where a single distribution of d_N/d_S ratios is applied across all branches. RELAX was chosen over methods such as the McDonald-Kreitman (1991) test that rely on polymorphism data gathered from many individuals within a species, which are generally unavailable (i.e, the RELAX method is more analogous to testing for elevated d_N/d_S ratios in focal lineages). We performed RELAX analyses on each gene individually and the whole mitogenome dataset in each case study.

Results

Inconsistent Selection on mtDNA in Flighted Lineages versus Flightless Lineages

As previously reported, we found elevated d_N/d_S ratios in bats compared with flightless mammals (0.44 vs. 0.40 for the mitogenome, P < 001; fig. 2B; Shen et al. 2010). However, the signal was not consistent: five genes showed significantly elevated d_N/d_S ratios, one (ATP6) had a significantly lower d_N/d_S ratio, and the rest showed nonsignificant differences. Contrary to the original hypothesis, RELAX analyses indicated that this pattern was likely due to relaxed, not positive, selection in bats (k = 0.93 for the mitogenome, P < .001; fig. 2C). No individual gene showed the predicted pattern of an elevated d_N/d_S ratio associated with an elevated k value (fig. 2).

For birds, flightless lineages did not have elevated d_N/d_S ratios, as reported previously (Shen et al. 2009; 0.04 vs. 0.04 for the mitogenome, P = 717; fig. 2B). RELAX analyses indicated a weak but significant trend toward intensified, not relaxed, selection in flightless lineages (k = .1.04 for the mitogenome, P = .034; fig. 2C). However, only ND2 showed a significant pattern of elevated d_N/d_S ratios in flightless taxa (0.06 vs. 0.05, P < .001; fig. 2B) consistent with relaxed selection (k = 0:91, P = .043; fig. 2C).

Contrary to previous results (Mitterboeck and Adamowicz 2013), reduced, not elevated, d_N/d_S ratios were observed in flightless insect lineages compared with flighted insect lineages (0.03 vs. 0.04 for the mitogenome, P < .001; fig. 2B), which was fairly consistent across genes (fig. 2B). RELAX analyses were largely statistically nonsignificant (fig. 2C), although k values in general suggested relaxed selection in flightless lineages. Three individual genes (ATP6, ATP8, and CYTB) did show the expected patterns of elevated d_N/d_S ratios associated with signals of relaxed selection.

Positive and Relaxed Selection on mtDNA in Lineages with Energetic Innovations

For snakes, we confirmed previously reported elevated d_N/d_S ratios relative to lizards (0.07 vs. 0.06 for the mitogenome, P = .002; fig. 3B; Castoe et al. 2008), which were consistent across genes (fig. 3B). As hypothesized, this was likely due to positive selection (k = 1.1, P < .001 for the mitogenome; fig. 3C). However, individual genes were inconsistent. The only two genes with k values significantly greater than one (ND1 and ND4) did not have significantly elevated d_N/d_S ratios. Three individual genes with significantly elevated d_N/d_S ratios showed k values significantly less than one (ATP6, ND5, and ND5), suggesting relaxed, not positive, selection on these genes.

For both gymnotiform and mormyroid electric fishes, there was a clear and consistent trend of elevated d_N/d_S ratios in electric fishes compared with nonelectric fishes, both for the mitogenome (difference in d_N/d_S ratios of 0.004–0.008, P < .001 for mormyroids, P = .182 for gymnotiforms; fig. 3B) and for individual genes (fig. 3B), as previously reported (Elbassiouny et al. 2020). However, positive selection was not consistently supported via RELAX. For mormyroids, there was a clear and consistent pattern of relaxed, not positive, selection (k = 0.83, P < .001 for the mitogenome, similar trends for individual genes; fig. 3C). For gymnotiforms, the pattern was inconsistent, with similar numbers of genes indicating relaxed selection and positive selection (the mitogenome had a small but statistically significant signature of positive selection: k = 1.02, P < .024; fig. 3C).

As suggested previously, primates did show significantly elevated d_N/d_S ratios across the mitogenome (0.09 vs. 0.04, P < .001) and in nearly all individual genes (up to fivefold for COX1; fig. 3B). Contradicting the brain-energy hypothesis (Grossman et al. 2004), this was consistent with relaxed, not positive, selection in primates, both across the mitogenome (k = 0.51, P < .001) and in nearly all individual genes (ATP8 was the most extreme, being under 5.5 times less intense selection in primates; fig. 3C).

Discussion

Our results suggest that some previous conclusions may have been based on misinterpreting the underlying causes of elevated d_N/d_S ratios in mtDNA. While increased taxon sampling may have allowed us to describe d_N/d_S ratios more thoroughly in some focal lineages (e.g., insects), we argue that explicitly distinguishing between relaxed selection and intensified positive selection is critical. Of the seven case studies we examined, none showed elevated d_N/d_S ratios that were consistently attributed to the predicted mechanism, and sometimes the opposite mechanism was convincingly supported when using RELAX.

Three of the seven case studies we reexamined linked flight to mitochondrial function. However, this and related hypotheses may not be falsifiable on the basis of d_N/d_S ratios alone. For example, flighted lineages should show higher d_N/d_S ratios due to positive selection, but elevated d_N/d_S ratios in flightless taxa due to relaxed selection also support this hypothesis. Because finding increased d_N/d_S ratios in either flightless lineages or flighted lineages would support the hypothesis, it is only by distinguishing the cause of increased d_N/d_S ratios that this hypothesis can be properly tested. Additionally, flightlessness may result in positive selection, as it is possible to positively select for a lower metabolic rate.

While positive selection on mitochondrial genes is an important part of environmental adaptation (Hill 2019), other forces also shape mtDNA evolution. Differences in N_e may shape relative rates of mutational accumulation in mtDNA, as purifying selection is a weaker force in small populations (Bazin et al. 2006; Meiklejohn et al. 2007). For example, parasitic lineages often show accelerated mtDNA evolution compared with nonparasitic lineages, possibly owning to reduced N_e (Castro et al. 2002; Oliveira et al. 2008; Jakovlić et al. 2021). Because flightless lineages may have lower dispersal abilities, they may also have lower N_e (McCulloch et al. 2009; Ikeda et al. 2012) and be under relatively relaxed selection and have higher d_N/d_S ratios, but not because they lead a less energetic lifestyle. Many flightless taxa (especially among birds) are also associated with islands and inherently low N_e (Woolfit and Bromham 2005). High d_N/d_S ratios in mormyroid electric fishes were consistent with relaxed selection (fig. 3) and may stem from relatively low N_e compared with reference taxa such as Clupeiformes. Primates also had high d_N/d_S ratios due to relaxed selection, contrary to the brain-energy hypothesis (Goldberg et al. 2003; Grossman et al. 2004) but consistent with lower N_e in primates compared with reference taxa such as rodents. Longevity (Nabholz et al. 2008; Galtier et al. 2009; Hua et al. 2015), generation time (Thomas et al. 2010), and other cytoplasmic endosymbionts (Hurst and Jiggins 2005) also play roles in mtDNA evolution, possibly confounding a role of energetics.

Multiple factors may also shape energetics in complementary or contradictory ways in specific lineages. For example, thermic habit (i.e., endothermy vs. ectothermy) may predict patterns of mtDNA evolution (Rand 1993, 1994), and several of our flightless birds were penguins, which may need efficient mitochondrial process for thermoregulation (fig. 2A). In these taxa, flight loss is confounded with other high-energy adaptations, which may explain why overall d_N/d_S ratios were similar between flighted taxa and flightless taxa (fig. 2B). Some studies do attempt to correct for confounding factors. For example, Mitterboeck and Adamowicz (2013) purposely excluded any cases of insect flight loss that were confounded with transitions in life history or population sizes. Importantly, some previous authors also considered alternative interpretations of elevated d_N/d_S ratios, but their overall conclusions were driven by biological observations (Elbassiouny et al. 2020) or additional tests of molecular evolution (Shen et al. 2010).

Overall, we suggest a cautionary approach when interpreting molecular signatures like d_N/d_S ratios, especially as data become increasingly available (fig. 1). This applies to all genetic loci, not just mtDNA. For example, adaptive hypotheses of rapid molecular evolution in reproductive genes and genes involved in phenotypic plasticity were later overturned in favor of relaxed selection (Hunt et al. 2011; Dapper and Wade 2020; Marinić and Lynch 2020; Patlar et al. 2021). Authors should be skeptical of adaptive just-so stories of genetic selection. Unfortunately, tools to explicitly disentangle relaxed selection versus positive selection are limited (but see, e.g., Crotty et al. 2020). Selective hypotheses stemming from molecular analyses should ultimately be evaluated through further functional experiments to avoid overinterpretation of molecular signals in a reductionist, adaptationist framework (Gould et al. 1979).

Data and Code Availability

All data analyzed here are publicly available, and accession numbers are provided in table S1. Relevant tree, alignment, and raw data files from this project are available via figshare (https://doi.org/10.6084/m9.figshare.21277536.v2; Havird 2022).