The three new complete snake mt genomes presented here, together with previously existing vertebrate mt genomes, provide a preliminary perspective on a complex history of potentially adaptive mt genomic change in snakes. Unusual changes in gene size and nucleotide substitution rates are associated with changes in mt genomic architecture (Figure 4). Nevertheless, the changes in substitution dynamics cannot be directly explained by the changes in mt genome architecture. Snake mt genome evolution is most consistent with some type of broad selective pressure on the efficiency and function of oxidative metabolism in snakes early in their evolutionary history.
In mt genomes (particularly in vertebrates), the processes of replication and transcription are not entirely functionally independent, and genome structural organization plays a prominent role in both processes. The CR acts as the origin of heavy strand replication, in addition to its role as the promoter for both heavy and light strand transcription . Genome replication also depends on the processing of light strand transcripts to produce short primers required for heavy strand initiation of genome replication (originating from the CR ). The regular distribution of the tRNA genes throughout the mt genome is functionally significant, and these play an important role in RNA processing of polycistrons to yield mature RNAs, transcription initiation and termination, as well as initiation of light strand replication . Collectively, many functional ramifications are linked tightly to mt genome architecture in vertebrates.
Mitochondrial genome size reduction due to gene shortening in alethinophidians is more than offset by the retention of their duplicate control regions. If size reduction is caused by selective pressure, the long term retention of dual CRs suggests that having both copies provide some selective advantage. Although the duplicate control region appears to function in heavy strand replication in at least some snakes, there is considerable variation in CR usage across snake lineages (Table 1). Thus, if the duplication has been maintained by selection, control of replication may not be the singular or primary selective driving force.
The possession of two functional control regions in most snake mt genomes might be advantageous by increasing the rate at which genome replication proceeds, and/or increasing the overall number of genome copies per mitochondrion. Since the dual CRs essentially flank the rRNA genes, they (along with adjacent tRNAs) could also plausibly function to independently control rates of protein-coding and rRNA gene transcription. Across snake species, variation in the tRNAs flanking the CRs includes the translocation of tRNALeu (3' of CR2) and the duplication/translocation/truncation of tRNAPro. In vertebrates, tRNALeu has been shown to decouple rates of rRNA and mRNA transcription by acting as a terminator of ~95% of heavy strand transcripts (leading to ~20-fold higher rRNA vs. mRNA levels; ). Considering the ectothermy of snakes, transcriptional decoupling via independent control regions could provide a more direct means of countering thermodynamic depression of enzymatic rates at low temperatures.
Independent CR duplications have also been identified in eels , frogs , birds [22, 23], and lizards [24, 25]. Our results (and additional unpublished data) suggest that the dramatic shifts in rates and patterns of molecular evolution in snakes represent a unique phenomenon that we do not expect to be necessarily associated with CR duplication, but rather more likely associated with selection for mitochondrial function. Nevertheless, these independent duplications may be useful to test the consequences of duplication on mutational processes.
Concerted evolution in and around the duplicate control regions
The two control regions clearly undergo concerted evolution to maintain reciprocal homogeneity between control regions within a genome [7–9], presumably through gene conversion. Interestingly, an apparently nonfunctional partial (or pseudo) proline tRNA (Ψ-tRNAPro) in colubrid mt genomes also appears to be maintained by concerted evolution (Figure 1). The gene conversion process that homogenizes the control region may also occasionally pick up extra DNA, making tRNAPro, or part of it, prone to duplication at this location. The existence of a duplicate tRNAPhe between CR2 and tRNALeu in the viperid O. okinavensis  suggests that frequent gene duplication adjacent to the CRs may occur (these two tRNAPhe differ by only 3 of 64 bp; implying either concerted evolution or recent duplication). The concerted evolution of these tRNAs could be explained by a tendency for gene conversion events involving the duplicate control regions to extend into the homologous flanking tRNA regions.
Another point of interest concerning gene conversion that arises from this study is a preliminary indication of differential evolutionary processes operating on the CRs within versus between species. Vertebrate mitochondrial control regions typically evolve very rapidly, and this is the case in a comparison of the two viperid species (O. okinavensis and A. piscivorus) in which CRs from these species are (on average) approximately as divergent as the fastest evolving positions within the mt genome, third codon positions (Figure 2B). In contrast, the two A. pisvicorus genomes, Api1 and Api2, have surprisingly similar CRs between individuals (Figure 2A; Additional file 2), comparable to the similarity between rRNA genes, among the slowest evolving regions in the mt genome. A previous study on viperid snakes also showed slow within-species CR evolutionary rates , and other studies have demonstrated particularly slow intra-species rates and differential rates of CR evolution operating within versus between species in birds  and fish .
In this study we have found a great deal of rate heterogeneity among genes, so it is certainly possible that the normally unconserved control regions have become suddenly critical and conserved in A. piscivorus. Alternatively, it is plausible that the complex (and poorly understood) process of gene conversion of CRs within a genome may also alter rates of CR evolution within species through a yet unknown process of gene conversion that may involve intragenomic (or even intergenomic) recombination.
Comparative rates of molecular evolution
Previous studies have suggested that snake mt genomes have an accelerated rate of evolution [7, 8]. Our results suggest this general conclusion is an oversimplification of a much more complex scenario, and that rates of snake mt genome evolution incorporate broad temporal (branch-specific) and spatial (gene and gene region-specific) dynamics. Branches early in snake evolution appear to be associated with dramatically elevated evolutionary rates and extreme relative rate dynamics across the mt genome (Figure 4). In contrast, terminal branches appear to have patterns of mt genome evolution that are strikingly similar to other (non-snake) vertebrates.
In support of a hypothesis involving selection for overall oxidative metabolic function, the accelerated rates of molecular evolution in snakes appear to depend greatly on gene function, with most ND subunits accelerating only slightly and occasionally, while COX, ATP, CytB, and rRNA evolutionary accelerations are dramatic and punctuated. The roles of these proteins (and the mitochondria in general) in energetics via oxidative phosphorylation are well known, and it may be that a single causative agent accompanying the diversification of snakes that dramatically altered metabolic demand, or led to a fluctuation in metabolic demand, was responsible for large-scale changes in selective pressure on these proteins.