Mitochondria are the site for the citric acid cycle and oxidative phosphorylation (OXPHOS), the final steps of ATP synthesis via cellular respiration. Each mitochondrion contains its own genome; in vertebrates, this is a small, circular DNA molecule that encodes 13 subunits of the multiprotein OXPHOS electron transport complexes. Vertebrate lineages vary dramatically in metabolic rates; thus, functional constraints on mitochondrial-encoded proteins likely differ, potentially impacting mitochondrial genome evolution. Here, we examine mitochondrial genome evolution in salamanders, which have the lowest metabolic requirements among tetrapods. We show that salamanders experience weaker purifying selection on protein-coding sequences than do frogs, a comparable amphibian clade with higher metabolic rates. In contrast, we find no evidence for weaker selection against mitochondrial genome expansion in salamanders. Together, these results suggest that different aspects of mitochondrial genome evolution (i.e., nucleotide substitution, accumulation of noncoding sequences) are differently affected by metabolic variation across tetrapod lineages.

Mitochondria are the site for the citric acid cycle and oxidative phosphorylation (OXPHOS), the final steps of ATP synthesis via cellular respiration. In addition to their critical role in meeting the cell's continuous energy demands, mitochondria also perform a variety of other functions associated with cellular Ca2+ signaling, apoptosis, cell transport, thermoregulation, and immunity (Bonawitz et al. 2006a,b; Detmer and Chan 2007; Scheffler 2008). Each mitochondrion contains multiple copies of its own genome, generally a small circular DNA molecule (∼17 kb in vertebrates) (Boore 1999; Scheffler 2008). Although mitochondria themselves perform diverse tasks, the mitochondrial genome encodes proteins associated with only one key task—OXPHOS. The mitochondrial genome usually comprises 13 genes that encode subunits of the protein complexes that perform electron transport and ATP synthesis, as well as 22 tRNAs and two rRNAs for transcription and translation of these genes within the mitochondrion (Boore 1999; Scheffler 2008). Nuclear genes encode the remaining 76 protein subunits required for OXPHOS (Ryan and Hoogenraad 2007; Scheffler 2008). Thus, metabolic requirements constrain substitution patterns of the entire mitochondrial genome, as well as a subset of the nuclear genome, to maintain the functional integrity of OXPHOS machinery.

Mitochondria exist not as discrete organelles, but as part of a dynamic network characterized by fission, fusion, and intracellular transportation events triggered in response to cellular stimuli (Detmer and Chan 2007). Mitochondrial network organization impacts all mitochondrial functions. Thus, functional OXPHOS proteins are necessary, but not sufficient, to provide cells with ATP; dynamic control of the mitochondrial network in response to cellular energy demands is also required (Benard et al. 2007). Such control results from the coordinated expression of ∼1500 proteins that make up the mitochondrial proteome (Gibson 2005; Scheffler 2008). The vast majority of these genes are encoded by the nuclear genome, translated in the cytosol, and imported into the mitochondria (Neupert and Herrmann 2007). However, regulation of both mitochondrial transcription and mitochondrial DNA replication is also a critical component of this coordinated gene expression and network control (Fernandez-Silva et al. 2003; Bonawitz et al. 2006a). Mitochondrial genes are transcribed as polycistronic transcripts that span almost the entire length of the genome, and mitochondrial DNA replication proceeds from a single replication origin per strand. Thus, both RNA transcription and DNA replication rates are likely impacted by mitochondrial genome size, which varies across the Tree of Life (Rand 1993). Constraints imposed by mitochondrial network organization, critical for both OXPHOS and non-OXPHOS functions, may therefore shape the evolution of mitochondrial genome size, although this hypothesis remains untested.

Among metazoans, average rates of substitution are higher in the mitochondrial genome than in the nuclear genome (Brown et al. 1979), reflecting the highly mutagenic mitochondrial environment (Ballard and Dean 2001; Scheffler 2008; Santos 2012). Coalescent times are also shorter because of the smaller haploid effective population size (Moritz et al. 1987; Ballard and Rand 2005). Mitochondrial sequences have therefore proven to be a powerful tool for evolutionary analyses. Although such studies have often assumed that mitochondrial genome evolution is effectively neutral across lineages (Ballard and Kreitman 1995; Blier et al. 2001), the mitochondrial genome's central role in ATP synthesis, coupled with the large-scale differences in aerobic metabolic rates across taxa, suggests that this assumption may be invalid (Ballard and Kreitman 1995; Bazin et al. 2006; Galtier et al. 2009b; Rand 2011). Consistent with this prediction, recent studies have begun to examine how organismal differences in key traits linked to mitochondrial biology (particularly aerobic metabolic rate, but also temperature) are reflected in different selective regimes acting on the mitochondrial genome (Galtier et al. 2009a; Sun et al. 2011). For example, studies in birds and mammals have tested for differences in selective constraints on mitochondrial DNA sequences across groups that differ in locomotive ability, a proxy for metabolic demand (Shen et al. 2009). These studies show that more weakly locomotive organisms experience weaker purifying selection on metabolic genes. Additionally, Shen et al. (2010) identified positive selection on mitochondrial genes along the lineage leading to bats, suggesting adaptive evolution to meet the increased energy demands of flight. Finally, Castoe et al. (2008) identified extensive positive selection on mitochondrial genes in the ancestral snake lineage, suggesting adaptive evolution for extreme metabolic flexibility. Convergent with snakes, a similar pattern of molecular evolution in mitochondrial genes was detected in agamid lizards (Castoe et al. 2009). These studies provide the first steps toward exploring how differences in metabolic demand across tetrapod lineages impact the strength and type of selection acting on the mitochondrial genome; however, the full range of metabolic demands that characterize different lineages remains unexplored.

Salamanders have the lowest metabolic requirements among tetrapods (Gatten et al. 1992). However, the impacts of such low metabolic rates on mitochondrial genome evolution remain unexplored. In this study, we show that the salamander clade experiences relatively weak purifying selection on mitochondrial gene sequences across all four OXPHOS functional complexes. In contrast, we find no evidence for relatively weak selection against mitochondrial genome expansion in salamanders. Taken together, these patterns suggest that different aspects of mitochondrial genome evolution (i.e., nucleotide substitution in gene sequences, accumulation of noncoding sequences) are differentially affected by variable metabolic requirements across tetrapod lineages.



We chose to compare salamanders with frogs (their sister taxon within Lissamphibia) because these two ectothermic amphibian clades share similar thermal habitats and experience similar environmentally dictated body temperatures. Although variation in metabolic rate exists among lineages within both clades, frog resting metabolic rates are, on average, 1.5- to 2.5-fold higher than those of salamanders (Fig. 1, File 1 in Supporting Information), and minimal cost of transport is significantly lower in salamanders (Feder 1976; Gatten et al. 1992). These differences suggest that the frog and salamander clades have diverged in factors controlling metabolic rate since their point of common ancestry. The mechanistic explanation for this divergence remains unknown, although hypotheses include (1) differences in muscles (e.g., the energetic cost of muscular force production), as well as (2) salamanders’ enormous nuclear genomes and associated low cell-surface-to-volume ratios (Szarski 1983; Taylor 1985; Kozlowski et al. 2003). Here, we test whether this metabolic divergence between clades (whatever its underlying mechanism) is correlated with a clade-level difference in selective constraints on mitochondrial genome sequences, which are directly associated with meeting an organism's metabolic requirements. We emphasize that we are not evaluating whether metabolic rate and selective constraint on mitochondrial genomes have remained correlated throughout the evolutionary history of anuran and caudate amphibians, a different question that would require regression analyses controlling for phylogenetic nonindependence (e.g., independent contrasts, phylogenetic generalized least squares) (Santos 2012). Rather, we are evaluating whether a historical divergence in metabolism between these two clades has had long-lasting, persistent consequences for molecular evolution, a pattern that would be detectable by comparing characteristics of any subsets of extant lineages between the two clades.

Figure 1.

(A) Relationship of families used in this study and summary of taxa represented in the mitochondrial genome size dataset (first column) and the resting metabolic rate dataset (second column). The color of the box indicates the presence (gray) or absence (white) of data for a given family for each dataset. (B) Summary of published resting metabolic rates for families included in the genetic datasets. Frogs (top) and salamanders (bottom) are separated by the dashed line. Within both the salamander and frog clades, there is substantial overlap among families in metabolic rates; less overlap exists between salamanders and frogs.

We obtained whole mitochondrial genome sequences and partial mitochondrial genomes (i.e., sequences that contain all of the protein-coding genes, but lack a portion of the nonprotein-coding sequence) for 94 salamanders (53 whole, 41 partial) and 34 frogs from GenBank (File 2 in Supporting Information). Our total dataset includes representatives of all 10 salamander families and 12 of 38 frog families that encompass the basal/near-basal split within frogs (Roelants et al. 2007) (Fig. 1).

To allow for comparisons between the mitochondrial genome and nuclear (i.e., nonmetabolic) sequences, we also obtained sequences for four nuclear genes from diverse salamander and frog lineages chosen to maximize taxonomic overlap with the mitochondrial dataset: BDNF for 46 salamanders and 30 frogs, NCX1 for 23 salamanders and 33 frogs, POMC for 48 salamanders and 33 frogs, and RAG1 for 54 salamanders and 40 frogs (File 2 in Supporting Information). Of the 52 genera represented in our salamander mitochondrial dataset, 27 genera are represented by one or more species of the same genus in our BDNF data, 18 genera are represented in our NCX1 data, 28 genera are represented in our POMC data, and 34 genera are represented in our RAG1 data. Of the 21 genera of frog represented in our mitochondrial dataset, 15 genera are represented by one or more species of the same genus in our BDNF data, 14 genera are represented in our NCX1 data, 15 genera are represented in our POMC data, and 19 genera are represented in our RAG1 data.


We performed multiple sequence alignments based on amino acid sequences for each gene (13 mitochondrial and four nuclear) using MUSCLE version 3.8 (Edgar 2004). We excluded regions of ambiguous alignment, based on the presence of indels, from further analysis. We then estimated 10 maximum likelihood (ML) trees: (1) the concatenated mitochondrial tree for salamanders, (2) the concatenated mitochondrial tree for frogs, (3–4) the BDNF trees for salamanders and for frogs, (5–6) the NCX1 trees for salamanders and for frogs, (7–8) the POMC trees for salamanders and for frogs, and (9–10) the RAG1 trees for salamanders and for frogs. All of these trees were estimated using RAxML version 7.2 (Stamatakis 2006), partitioning the data by gene and codon position and specifying the GTR +Γ model of nucleotide substitution for each partition. These 10 trees serve as the basis for our selection analyses.


To test whether salamanders experience weaker purifying selection on their mitochondrial genes than do frogs, we compared the ratio of nonsynonymous (dN) to synonymous (dS) substitution rates (ω=dN/dS) between the two clades. ω is commonly used to measure the strength of selection: a small ω value indicates strong purifying selection and a large ω value (where ω is still less than 1) indicates weak, or relaxed, purifying selection.

For each mitochondrial gene, we first estimated dN, dS, and ω for both salamanders and frogs assuming a single value for each parameter across all branches (Model 0 in Codeml, implemented in PAML version 4.4 [Yang 2007]); topologies were fixed to the ML trees estimated from the concatenated mitochondrial sequences. We then used these estimates as starting values to estimate dN, dS, and ω under a model that allows these parameter values to vary for each branch on a tree (Model 1 in Codeml). Although this analysis yields ω estimates for all branches (both internal and tip), we restricted one round of comparisons to tip taxon estimates because these are less likely to be impacted by uncounted multiple substitutions at any given site. We then repeated our analyses including both internal and tip branches to ensure that any patterns detected were not restricted to tip lineages. For each gene, we tested for differences in ω values between salamanders and frogs using a Mann–Whitney test implemented in R, as the data were nonnormally distributed (Shapiro–Wilks test for normality, P < 0.0001). Additionally, we performed 10 replicate subanalyses under equal taxon sampling (i.e., randomly selecting 34 of our 94 total salamanders) for each gene to test whether any differences detected in ω between the two clades reflected bias introduced by having more salamander than frog sequences in our dataset. Finally, we performed the same set of analyses on all four nuclear genes to exclude the possibility that any difference in ω between salamander and frog mitochondrial genes reflects differences in the strength of genetic drift between the two clades. Stronger genetic drift, as a result of smaller effective population size, would lead to higher ω values for genes encoded by both genomes, irrespective of metabolic function.


If salamanders experience weaker selection for DNA replicational and RNA transcriptional efficiency than do frogs, we would predict a pattern of larger mitochondrial genomes (i.e., the sequence length of complete genomes) in salamanders than in frogs (Selosse et al. 2001). We tested for a difference in overall mitochondrial genome size between the two clades using a Mann–Whitney test implemented in R, as the data were nonnormally distributed (Shapiro–Wilks test for normality, P < 0.0001).

Results and Discussion


ω is significantly larger in salamanders than in frogs for 10 of the 13 mitochondrial protein-coding genes (two-way Mann–Whitney test; P < 0.0001 for all 10 genes) (Fig. 2); no difference in ω between the two clades exists for the remaining three mitochondrial genes (ATP8, COX1, and COX2; P > 0.31 for all three genes). This pattern holds, both for our complete dataset, as well as the subsampled datasets controlling for unequal taxon sampling (two-way Mann–Whitney with Bonferroni correction for multiple comparisons). In contrast, ω for BDNF, POMC, and RAG1, nuclear-encoded genes not associated with metabolic function, are not significantly different between frogs and salamanders (two-way Mann–Whitney test, P > 0.25 for all three genes), and ω for the nonmetabolic nuclear NCX1 gene is actually larger in frogs (two-way Mann–Whitney test, P < 0.02). Inclusion of only tip branches (Fig. 2), or both internal and tip branches (File 2 in Supporting Information), yields similar results. These estimates of ω for the nuclear genes suggest that the differences in ω between frog and salamander mitochondrial genes do not reflect varying strengths of genetic drift between the two clades. Taken together, our results are consistent with weaker purifying selection on mitochondrial genes in salamanders than in frogs, as predicted by clade-level differences in metabolic requirements. Our results are in agreement with similar analyses performed in birds and mammals; Shen et al. (2009) demonstrate that mitochondrial ω values are lower in “strongly locomotive” lineages, which suggests stronger purifying selection associated with higher metabolic requirements. Thus, similar patterns emerge from comparisons involving highly aerobic, endothermic vertebrates (birds and mammals) as well as less aerobic, ectothermic vertebrates (salamanders and frogs).

Figure 2.

Comparison of dN/dS (ω) values for 13 mitochondrial protein-coding genes and four nuclear genes (BDNF, NCX1, POMC, and RAG1). Frogs are on the left; salamanders are on the right. Dashed lines separate mitochondrial genes belonging to different OXPHOS functional complexes, and the solid line separates mitochondrial and nuclear genes. For all mitochondrial genes except COX1, COX2, and ATP8, salamander ω values are significantly larger than frog values.

The three genes that show no difference in ω between salamanders and frogs (ATP8, COX1, and COX2) fall on the opposite ends of the range of nonsynonymous substitution rates (dN) exhibited by mitochondrial genes. Across diverse vertebrate lineages, including those examined in this study, COX1 and COX2 have the lowest dN, whereas ATP8 has the highest (Pesole et al. 1999; Shen et al. 2009). COX1, COX2, and COX3, in combination with 10 nuclear-encoded protein subunits, form Complex IV (cytochorome c oxidase) of the electron transport chain; ATP6 and ATP8, along with 14 nuclear-encoded subunits, form Complex V (ATP synthase). Our results suggest that the difference in functional constraint experienced by frog and salamander mitochondrial genes varies among the OXPHOS functional complexes. For the most constrained complex (Complex IV), the lower metabolic demands of salamanders may not translate into weaker purifying selection across all three genes; the ATP requirements of salamanders likely still impose the same level of selective constraint on COX1 and COX2 sequence evolution. For the least constrained mitochondrial protein (ATP8 of Complex V), high levels of dN (relative to other mitochondrial genes) may be permitted by the metabolic requirements of either clade. For the other two complexes (Complex I, composed of ND1–6 and ND4L and 36 nuclear-encoded proteins, and Complex III, composed of CYTB and 10 nuclear-encoded proteins), the metabolic differences between salamanders and frogs likely translate into different degrees of purifying selection on all mitochondrial-encoded subunits. Similar analyses comparing strongly and weakly locomotive birds showed significant differences in ω for only four out of 13 mitochondrial genes, although analyses binning genes by functional complex showed differences in both Complex I and V (Shen et al. 2009). Thus, studies in amphibians and birds both suggest heterogeneous responses to variable metabolic constraint at the level of individual mitochondrial genes and functional complexes. Future studies will show whether such patterns hold for nuclear-encoded OXPHOS proteins, as well as whether coevolution between the two genomes is correlated with metabolic requirements.


Many of the larger genomes in both salamanders and frogs result from repetitive sequences in the control region (the noncoding sequence that regulates replication and transcription) and/or the presence of a duplicate control region (e.g., Mueller and Boore 2005; Kurabayashi et al. 2010). Gene rearrangements resulting from tandem duplication and random loss are also present in both clades (e.g., Mueller and Boore 2005; Kurabayashi et al. 2006; Kurabayashi et al. 2008). Our results show that mitochondrial genomes are not larger in salamanders than in frogs; in fact, frog genomes are larger (frog mean = 18.2 kb; salamander mean = 17.4 kb) (two-way Mann–Whitney test, P < 0.0001). Thus, we do not detect the pattern predicted by weaker selection against genomic expansion accompanying lower metabolic requirements in salamanders (Selosse et al. 2001). Instead, nonadaptive processes including genetic drift and/or a replicative advantage of genomes with duplicate control regions (irrespective of any consequences on organismal phenotype) may determine mitochondrial genome size variation in frogs and salamanders, as they do in other taxa (Kumazawa et al. 1998; Yokobori et al. 2004; Lynch et al. 2006; Boussau et al. 2011).

Associate Editor: C. Peichel


This work was supported by the National Science Foundation [NSF-DEB 1021489 to RLM] and Colorado State University and used the CSU ISTeC HPC System [NSF CNS-0923886]. J. López-Arriaza, C. Sun, and anonymous reviewers provided helpful comments. R. Thomson and R. Schwartz provided helpful discussion. R. Bonett provided access to an unpublished Eurycea tynerensis mitochondrial genome sequence for reference.