The tissue used for DNA sequencing was derived from a single male A. limnaeus at 3 months post-hatch. The laboratory stock this individual was sampled from has been maintained since 1995 and originated from near Quisiro, Maracaibo basin, Venezuela. The fish was euthanized by immersion in an ice bath for ~5 min followed by cervical transection. Liver, white muscle (with skin and scales removed), and brain tissue were removed and transferred to DNA extraction buffer [10 mmol l⁻¹ Tris-HCl pH = 8.0, 100 mmol l⁻¹ EDTA, 0.5% SDS, 200 μg ml⁻¹ Proteinase K (Thermo Scientific, Waltham, MA, USA, #EO0491)] at a ratio of 1 mg tissue per 10 μl buffer. Tissues were gently homogenized using a Teflon pestle in 1.5 ml microcentrifuge tubes and incubated for 3 h at 50°C with agitation every 20–30 min. DNA was extracted by briefly vortexing the homogenates with 1 vol phenol (equilibrated with 10 mmol l⁻¹ Tris-HCl, pH = 8.0, 1 mmol l⁻¹ EDTA) followed by centrifugation at 5000 × g for 10 min at 4°C. The aqueous phases were collected using a wide-bore pipette and extracted with phenol as described above. Following the second phenol extraction, the aqueous phases containing the DNA were extracted a third time by gentle mixing with 1 volume of chloroform:phenol (1:1) followed by centrifugation at 5000 × g for 10 min at 4°C. The aqueous phase was collected and the DNA precipitated by the addition of NaCl (final concentration of 200 mmol l⁻¹) and 2 volumes of 100% ethanol (EtOH) followed by mixing by gentle inversion. The DNA was pelleted by centrifugation at 16,000 × g for 10 min at 4°C. The DNA pellet was washed twice with 1 ml 70% EtOH. After each wash the DNA was pelleted by centrifugation at 16,000 × g for 5 min at 4°C. The air-dried DNA pellet was resuspended in 500 μl of RNAse buffer (10 mmol l⁻¹ Tris-HCl, pH = 8, 5 mmol l⁻¹ EDTA) with 100 μg ml⁻¹ DNAse-free RNAse A (Thermo Scientific #EN0531) and incubated at 37°C for 45 min with occasional tube inversion. The samples were extracted twice more as described above with chloroform:phenol (1:1) followed by precipitation using 0.1 vol of 3 mol l⁻¹ sodium acetate pH = 5.2, and 2 vol of 100% EtOH. The DNA was pelleted by centrifugation at 16,000 × g for 10 min at 4°C and the pellet was washed twice with 1 ml 70% EtOH as described above. The DNA pellets were dried briefly at room temperature and resuspended in 100 μl of buffer EB (10 mmol l⁻¹ Tris-HCl pH = 8.5, Qiagen, Hilden, Germany, #19086) at 37°C with occasional agitation until pellets dissolved completely.

Total genomic DNA concentrations and A₂₆₀/A₂₈₀ ratios were determined using the Infinite M200 Pro plate reader equipped with a NanoQuant plate (Tecan, San Jose, CA, USA) using 2 μL of sample and default software settings (i-control software, Tecan). We considered A₂₆₀/A₂₈₀ ratios between 1.9 and 2.1 to be of acceptable DNA purity. DNA integrity was determined by 1% agarose gel electrophoresis of 1 μg of total DNA and observation of high-molecular weight DNA above 20 kb.

DNA sequencing libraries from the mixed tissue sample were prepared and sequenced at the University of Oregon High Throughput DNA Sequencing and Genomics facility. Purified DNA was sonicated to an average size of 170 bp and was prepared for sequencing using the Nextera library prep kit (Illumina, San Diego, CA, USA). Fragment libraries were sequenced on the Illumina Hi-Seq 2500 platform with 101 bp paired-end reads.

Polymerase chain reaction (PCR) was used to amplify specific regions of the mitochondrial genome. PCR products were generated using total DNA from adult liver as the DNA template. PCR reactions (50 μl total volume) consisted of 50 ng total DNA, 10 pmol of forward and reverse primers (Table S1, Integrated DNA Technologies, Coralville, IA, USA), 0.25 U Taq polymerase (New England BioLabs, Ipswich, MA, USA #M0267L), in 1X ThermoPol buffer (New England BioLabs # M0267L). The reactions were heated for an initial 5 min at 95°C followed by 40 cycles of: 20 s at 95°C, 20 s at 60°C, and 5 min at 68°C. A single final elongation step of 10 min at 68°C followed the initial 40 cycles. PCR products were resolved via 1% agarose electrophoresis and product size was estimated by comparison to a DNA ladder (GeneRuler 1 kb Plus, Thermo Scientific #SM0312) using GelAnalyser (http://www.sequentix.de/). PCR products were then excised and cloned into the pGEM-T vector or sequenced directly (Table S2). Plasmids were heat-shock transformed into E. coli JM109 high-efficiency cells (Promega #L2001, Madison, WI, USA) as recommended by the manufacturer. Plasmids were purified using the Qiagen Miniprep kit and Sanger sequenced at Oregon Health and Science University using the primers indicated in Table S2.

Adapters were removed from Illumina reads using Trimmomatic version 0.36 (Bolger et al., 2014) and the included Illumina adapter list. Error-correction was performed using the Allpaths error correction module (Gnerre et al., 2011). Of the total read pool that passed adapter trimming and error-correction, 100 million forward and reverse reads were randomly chosen for mtgenome assembly (Table S3). Sanger reads were trimmed of vector contamination using the UniVec tool in the Geneious 6.1.8 suite (Biomatters, Auckland, New Zealand) and bases with a greater than 5% chance of error were trimmed (Kearse et al., 2012).

Complete mitochondrial genomes from several other Cyprinodontiform fishes were retrieved from the MitoFish database for comparative analysis (Table 1; Iwasaki et al., 2013). Mitochondrial genes in comparator species were identified using the MitoAnnotator pipeline as done with A. limnaeus. Individual coding genes without stops, rRNAs, and tRNAs were aligned using the MUSCLE alignment tool as part of the Geneious suite ver. 8.1.7 (Edgar, 2004; Kearse et al., 2012). Third codon positions were removed from alignments to avoid saturation bias (Kumazawa and Nishida, 1993). Poorly aligned regions were trimmed using the GBLOCKS webserver (http://molevol.cmima.csic.es/castresana/Gblocks_server.html) with default settings (Castresana, 2000). JModelTest 2 was used to estimate the best model of nucleotide substitution on the concatenated alignment (Darriba et al., 2012). The final alignment contained 10,875 nucleotide positions and phylogenetic relationships were inferred using RAxML-HPC 7.7.2 using settings for GTR + I + G, 1000 bootstrap replications, and partitioning by first and second codon positions, rRNAs, and tRNAs (Stamatakis, 2006). RAxML was run using the rapid-hill climbing algorithm that also returned the best-resulting maximum likelihood (ML) tree in a single step. Oryzias latipes (Order Beloniformes) was specified as an outgroup.

Codon usage tables and nucleotide position composition ratios of coding genes were obtained using CAIcal (Puigbò et al., 2008).

Illumina RNAseq reads derived from a pooled sample of 24 somite embryos reared at 30°C (SRA #SRR2032241) were used to identify mRNA expression of the A. limnaeus mtgenome in addition to polyadenylation signals. Sequenced reads were filtered on quality scores and trimmed for the presence of adapter sequences using Trimmomatic (Bolger et al., 2014) with the following settings: ILLUMINACLIP:2:30:7:1:true, SLIDINGWINDOW:5:15, LEADING:20, TRAILING:20, and MINLEN: 25. Following trimming, RNAseq reads had an average length of 97 bp, standard deviation of 11 bp, and a read length range of 25–100 bp. Filtered and trimmed reads were initially mapped to the annotated A. limnaeus mtgenome using Geneious software (version 8.1.7; max gap size = 5, max gaps per read 10%, ≥80% identity, map multiple best matches randomly) to estimate overall coverage of the mtgenome by RNAseq. To predict polyadenylation sites, slightly more stringent mapping settings were used that omitted reads with multiple best matches (≥99.9% read mapping confidence, no gaps, ≥80% identity, word length = 10). Similar to previous studies, we set a threshold requiring three or more non-template adenine nucleotides at a variant frequency of >4% to call a polyadenylated region at a putative 3′ transcript end (Marková et al., 2015).

The A. limnaeus mitochondrial origin of light-strand replication (Ori-L) and tRNA-Ser putative secondary structures were predicted using Geneious software ver 8.1.7 with default DNA energy mode settings at 25°C (Mathews et al., 2004 model).

The TOPCONS webserver was used for consensus predictions of A. limnaeus mitochondrial peptide membrane topology with default settings (Tsirigos et al., 2015). Secondary structure predictions were performed using PSIPRED v3.3 with default settings on the PSIPRED Protein Sequence Analysis Workbench webserver (Jones, 1999; Buchan et al., 2013). Bovine (Bos taurus) mitochondrial peptide sequences for ND1 (P03887), ND5 (Q85BD6), and ND6 (P03924) were downloaded from the UniProt database and used for comparison to A. limnaeus homologs (UniProt, 2015).

Embryos were exposed to anoxia as previously described (Podrabsky et al., 2012a). To obtain PDII embryos, A. limnaeus embryos were staged at DII and diapause was experimentally terminated by placing dormant embryos in a 33°C incubator illuminated with 24 h illumination for 2 days before being returned to 25°C in the dark. PDII embryos were staged to one of three developmental stages using a dissecting scope: WS 36 (early organogenesis, 4 dpd), WS 40 (full overgrowth, 12 dpd), or WS 42 (early prehatching, 22 dpd; Wourms, 1972a). Prior to anoxia treatment, embryos were collected in petri dishes (100 × 15 mm plastic dishes) and excess embryo medium was removed using a pipette. Embryos were transferred into a Bactron III anaerobic chamber (Sheldon Manufacturing, Cornelius, OR) and rinsed three times with embryo medium that was previously purged for 30 min with N₂ gas and allowed to equilibrate inside the anoxic chamber overnight. Anoxic conditions were maintained through the use of an anaerobic gas mixture (5% H₂, 5% CO₂, balance N₂) and a palladium catalyst for the duration of anoxia exposure. Embryos at WS 36, 40, and 42 were sampled in normoxia (t = 0), after 24 h of anoxia, and after 24 h of aerobic recovery from anoxia (9 embryos per replicate). Unexposed normoxic embryos at 24 and 48 h were also sampled for parallel comparison. Diapause II embryos were sampled only in normoxia. Embryos were incubated at 25°C for the duration of the experiment in all treatments. Prior to sampling, embryos were collected onto a nylon mesh screen (100 μm mesh) and blotted dry using Kimwipes before being transferred into 1.5 ml microcentrifuge tubes and flash-frozen in liquid N₂. Frozen embryos were stored at −80°C until DNA extraction.

Embryonic DNA was extracted using the DNeasy Blood and Tissue Kit (Qiagen, #69581). Embryos were homogenized on ice in microcentrifuge tubes using a Teflon pestle with 400 μl of kit provided lysis buffer ATL and 2 mg ml⁻¹ of Proteinase K (Thermo Scientific, #EO0491). Samples were incubated at 56°C for 1 h and then centrifuged for 5 min at 300 × g at 4°C to pellet insoluble debris. Supernatants were transferred into new microcentrifuge tubes and 400 μl of buffer AL was added. Samples were vortexed briefly and incubated at 56°C for 10 min. Each sample was then mixed with 400 μl of 100% EtOH, vortexed briefly, and applied to the DNeasy silica columns according to manufacturer instructions. After washing columns according to manufacturer's instructions, DNA was eluted in 100 μl of buffer AE. Quality and quantity of DNA was assessed as described above for the adult tissues.

Primers for qPCR were designed using default settings in the IDT PrimerQuest tool (http://www.idtdna.com/Primerquest/) with the mitochondrial gene NADH dehydrogenase 4 (ND4) or nuclear insulin-like growth factor-1 receptor (IGFR1, NCBI Reference Sequence: XM_014003519.1) genes as inputs (Table S4). The amplification specificity of each gene/primer set was verified by cloning amplicons generated from A. limnaeus embryo tissues into the pGEM-T Easy Vector (Promega, Madison, WI, #A1360), transformation into JM109 High-efficiency cells (Promega, #L2001), and Sanger DNA sequencing the plasmid inserts as previously described (Wagner and Podrabsky, 2015). Purified plasmids with verified PCR inserts for either ND4 or IGFR1 were used as standards for qPCR assays. qPCR reactions (20 μl total volume) were set up in triplicate using SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, Hercules, CA, USA, #172-5270) and consisted of 25 ng total DNA, 10 μl of 2X SsoFast Probes Supermix, and 500 nmol l⁻¹ of forward and reverse primers. Assays were set up in clear 96-well semi-skirted PCR plates (Hard-Shell High-Profile PCR plates, Bio-Rad, #HSS-9601) with optical flat caps (Bio-Rad, #TCS-0803). All qPCR reactions were carried out in the Stratagene Mx3005P thermocycler (Agilent Technologies, Santa Clara, CA). Standard curves were generated using a serial 1:10 dilution of amplicon-specific plasmid copies (1 ng–1e⁻⁷ ng for ND4 and 1e⁻¹–1e⁻⁷ for IGFR1). Reactions were initially heated at 95°C for 5 min to activate the DNA polymerase and subsequently thermocycled for 35 cycles by denaturation at 95°C for 30 s and annealing/elongation at 60°C for 30 s. Fluorescence readings (excitation 492 nm, emission 516 nm) were taken at the end of each elongation step. Quantification cycle (C_q) thresholds were set automatically in Stratagene MxPro software (v.4.10, 2007, Agilent Technologies) using adaptive baseline, moving average, and amplification-based threshold settings. Although rarely necessary, thresholds were manually adjusted to improve standard curve best-fit regressions. Reaction efficiencies of ND4 or IGFR1 (E_gene) were estimated using the standard curve slopes such that E_gene = 10^(−1/slope) (Pfaffl, 2001). Assays were validated by visualizing qPCR end products via gel electrophoresis in addition to observation of a single melting dissociation curve.

The complete A. limnaeus mitochondrial genome contains 21,039 bases, 37.3% GC content, and is circular. MitoAnnotator identified two rRNAs, 21 unique tRNAs, 13 protein-coding genes, and two D-loops (D-loop 1 and D-loop 2) separated by a partial 16S rRNA unit and duplicated tRNA-Leu2 sequence (Figure 1, Table 2). The initial assembly of the A. limnaeus mtgenome using only MitoBIM and Illumina reads incorrectly assembled the duplicated D-loop region, as discovered by post-assembly read mapping and identification of breakpoints characteristic of misassembly. Cloned PCR products amplified from around the D-loop region identified the correct sequence order (Figure 2). The correct sequence order was confirmed by mapping of Illumina reads from both forward and reverse libraries with a final average coverage of 9059X (range 1579X–19,520X). A total of 1,705,832 reads mapped with 85% of bases mapping with quality >Q20. The resulting mtgenome assembly has been deposited into NCBI Genbank (KX371089).

The protein-coding genes included NADH-ubiquinone oxidoreductase (complex I) subunits 1, 2, 3, 4, 5, and 6 (ND1, ND2, ND3, ND4, ND5, and ND6), cytochrome c oxidase subunits 1, 2, and 3 (COI, COII, COIII), cytochrome b (Cytb), and F₁F_o-ATP synthase subunits (ATPase-6 and ATPase-8). Identical to the mtgenomes of other species within the Order Cyprinodontiformes used in this study, most identified genes and RNAs were determined to be encoded on the heavy (H, purine rich) strand, aside from the ND6 gene and eight tRNAs which are encoded on the light (L, pyrimidine rich) strand (Iwasaki et al., 2013). As expected, TOPCONS was able to predict transmembrane domains for all putative coding genes based on translated structure (Table 2).

The overall gene nucleotide composition of the A. limnaeus mtgenome is similar to other Cyprinodontiform fishes examined (Table 1). Also similar to the other Cyprinodontiformes, the A. limnaeus coding genes that are transcribed on the H strand are depleted of guanine at the third codon position (Figure S1A). This bias against guanine at the third codon positions is not as apparent in ND6 (Figure S1B), which is transcribed on the L strand. Overall, the A. limnaeus coding genes show negative skew for AT [skew: (A−T)/(A+T)] and GC [skew: (G−C)/(G+C)], indicating a bias toward the pyrimidines T and C in all coding genes except for ND6 (Figure S2). The top five most frequent codons for the aplocheloid fish are listed in Table S5. These five species share three of the top five most frequently used codons in their respective mtgenomes.

Five SNPs were identified at 10% frequency or higher, with four occurring in the 16S rRNA and one predicted to cause a synonymous amino acid change in ND4L (Table 3). The four SNPs occurring in the 16S rRNA appear to occur as a single haplotype, as suggested by nearly identical frequency of occurrence of the SNPs at this site, as well as simultaneous presence of all four SNPs in the reads that map to the region. The shared frequency between the 16S rRNA SNPs and the ND4L SNP indirectly suggest that there may be an alternative mitochondrial haplotype occurring at ~25% in the read pool that contains the five identified SNPs.

Although the total length of the A. limnaeus mtgenome is larger than other species used in this study, the lengths of coding genes are comparable, with exceptions for ND1, ND5, ND6, and ATPase-6 (Figure 3). The A. limnaeus ND1 gene is shorter near the 5′ by 27 nucleotides (alignment positions 13–39) and also differs by having a “GTG” start codon rather than “ATG.” ND5 is longer near the 5′ end by 33 nucleotides (alignment positions 7–39) and has an “ATA” start codon as opposed to the overall consensus of “ATG.” ND6 is 5′ truncated by 54 nucleotides (alignment positions 1–54) and has a “GTG” start codon instead of “ATG.” Finally, ATPase-6 differs at both 5′ and 3′ ends, with a small insertion of 6 nucleotides at the 5′ (alignments positions 5–10) and an extension of 34 nucleotides at the 3′ end compared to other species. NCBI BLASTx or BLASTn searches of the extended sequence in ATPase-6 did not return significant results. These unique gene features do not have clear implications for the mitochondrial physiology of A. limnaeus, but the number of differences, and the clustering of the differences in complex I and V suggest the potential for functional significance.

Seven mitochondrial-encoded subunits form the hydrophobic core of complex I, which participates in essential functions such as forming a channel for ubiquinone to move out of the lipid bilayer and into the protein complex and proton transfer (Hirst, 2013). Importantly, the majority of ROS production from ETC activity is thought to occur in complex I (Stowe and Camara, 2009; Quinlan et al., 2012). PSIPRED secondary structure predictions of A. limnaeus ND1 and ND5 did not reveal major structural changes when compared to their bovine homologs. However, the first transmembrane helix (ND6-TMH1) of A. limnaeus ND6 appears to be incomplete or absent when compared to bovine ND6 (Figure 4). Bovine ND6-TMH1 is believed to be closely associated with other transmembrane helices in ND6 and ND4L. Currently there are no other known instances of a vertebrate missing ND6-TMH1, but the possible functional relevance of these missing residues is unknown (Vinothkumar et al., 2014; Judy Hirst, pers. comm.).

Similar to the structural changes in ND6, the effect of the extended N- and C-termini of the ATPase-6 gene when compared to other fish species is not immediately clear. ATPase-6 comprises the a subunit of complex V (the F_o subunit of the F₁F_oATP synthase), which is the complex primarily known for its role as an ion channel that couples proton flux from the inter-membrane space into the matrix with ATP synthesis through interactions with the F₁ subunit of the ATP synthase complex (Jonckheere et al., 2012). Tight regulation of ATP synthesis, and prevention of ATP consumption during dormancy associated with diapause or anoxia is clearly important (St-Pierre et al., 2000). Making inferences on the possible functional implications of these sequence changes is difficult due to a lack of detailed information on the crystal structure of the F_o complex (DeLeon-Rangel et al., 2013). The role of the rather long extension of the C-terminus in ATPase-6 is intriguing and potentially offers the possibility of altered interactions with other subunits of the complex or with proteins in the intermembrane space.

Because complexes I and V both contain several essential and accessory subunits encoded by the nucleus, it will be interesting to see if these other subunits also contain structural changes that are divergent from other fish species. Additionally, confirmation of these complex I and complex V subunit features in other populations of A. limnaeus and exploration of the functional consequences of these features may be a fruitful area for future studies.

Sequence alignment of the two A. limnaeus D-loops (Alim-Dloop 1 and Alim-Dloop 2) identified a 1497 bp region with 100% pairwise identity with the remaining 3′ sections appearing to be evolving independently (Figure 5). Both Alim-D-loop 1 and Alim-D-loop 2 contain a high proportion of AT repeats, the most frequent of which is the motif “TAA,” significantly clustered in the unshared portion of D-loop sequences (Figure S3). After alignment of D-loops from several species, two putative CSBs were identified in A. limnaeus (CSB-1 and CSB-2), and these CSBs were discovered in both Alim-D-loop 1 and Alim-D-loop 2 based on the conserved motif previously identified by Broughton et al. (2001) (Figure 6). Aside from the three aplocheloid species (Nothobranchius furzeri, Kryptolebias marmoratus and A. limnaeus) a single D-loop is found between tRNA-Phe and tRNA-Pro, suggesting that this may be the ancestral positioning of the D-loop. In A. limnaeus, the non-coding region between tRNA-Phe and tRNA-Pro has been reduced to 59 bp (referred to as pseudoDloop from this point forward), several hundred bps smaller than the other species examined. NCBI blast of this short region did not yield significant results, nor did the sequence have high identity to any other region in the A. limnaeus mtgenome. Although sequence and length of the D-loop region can be highly variable between species, possibly due to relaxed selection (Lee et al., 1995), the presence of large D-loop translocations or duplications appears to be uncommon in fishes. Duplicate D-loops have been previously described in snakes (Kumazawa et al., 1996, 1998; Dubey et al., 2012), birds (Abbott et al., 2005; Morris-Pocock et al., 2010), and fish (Lee et al., 2001; Li et al., 2015), but the physiological implications of having repositioned and/or duplicated D-loops are unclear. In snakes, there is evidence that both regions are able to function in replication and transcription processes (Jiang et al., 2007). Similar to the D-loop duplication found in K. marmoratus (Lee et al., 2001), the snake duplicated D-loop regions flank both the 12S and 16S rRNAs. Jiang et al. (2007) proposed that having dual regulatory regions might allow for differential regulation of mitochondrial rRNA and mRNA, but this hypothesis has yet to be tested. In typical mammalian mitochondria, different transcriptional binding sites allow for differential transcription of rRNA and mRNAs, with one polycistronic molecule transcribing almost the entire H-strand, and the other stopping after transcribing through the 12S and 16S rRNAs (Fernández-Silva et al., 2003). Fernández-Silva et al. (2003) showed that the incomplete polycistronic RNA containing the 16S and 12S rRNA is transcribed ~20 times more frequently than the polycistronic RNA covering nearly the entire H strand, suggesting that rRNA abundance is an important factor in mitochondrial function. Interestingly, H-strand transcription starting from the repositioned D-loop in A. limnaeus would cause the rRNAs to be transcribed near the end of the polycistronic molecule. This implies that either a new transcriptional start site has evolved upstream of the rRNAs, or that all of the H-strand protein coding and tRNA genes would be transcribed along with rRNAs after every round of transcription. Coordinated expression of the entire H-strand may be advantageous to A. limnaeus by allowing for rapid and simultaneous expression, or depression, of tRNA, rRNA, and mRNA. The presence of CSB-2, thought to be important in mitochondrial RNA transcription stabilization (Xu and Clayton, 1995; Fernández-Silva et al., 2003; Pham et al., 2006), in both D-loops suggests that H-strand RNA polymerase binding sites may also be duplicated, perhaps allowing for increased transcription rates of the H-strand. Loss of a canonical D-loop between tRNA-Phe and tRNA-Pro has been previously described in the tongue sole, Cynoglossus semilaevis, whereby the D-loop occurs in between the ND1 and tRNA-Gln genes. Thus, it appears fish mitochondrial genomes are able to tolerate diverse translocations of the D-loop while still retaining function.

Occurring between D-loop 1 and D-loop 2 is a region with partial homology to the 16S rRNA, including a 178 bp (95.2% identity) and 431 bp (88.5% identity) region, collectively referred to as pseudo-16S (Figure 7). The occurrence of the region between Cytb and the 12S rRNA and the occurrence of both the 16S and pseudo-16S regions were confirmed in multiple A. limnaeus adult individuals (Figure S4). The putative origin of light-strand replication (Ori-L) is a 36 bp region found between tRNA-Asn and tRNA-Cys and is predicted to fold into a stable hairpin structure (Figure 8; Seutin et al., 1994). The Ori-L also contains the conserved motif 3′-GGCCG-5′ near the 5′ end of the hairpin structure, a sequence thought to interact with replication enzymes to direct light-strand replication (Hixson et al., 1986; Hurst et al., 1999).

Although nearly all tRNAs were identified and folded into predicted structures (Figure 9), neither MitoAnnotator nor MITOS annotation pipelines were able to predict tRNA-Ser (anticodon AGN) in A. limnaeus. Mapping of the N. furzeri tRNA-Ser sequence to the assembled A. limnaeus mtgenome identified a putitative A. limnaeus tRNA-Ser (51.5% pairwise identity). The putative A. limnaeus tRNA-Ser has the potential to fold into a cloverleaf structure, as predicted using Geneious. Similar to the previous report for the A. charrua mtgenome, the A. limnaeus tRNA-Cys is predicted to be the same length as that of A. charrua at 57 bp and to lack the D-arm (Figure 9). This feature of tRNA-Cys missing the D-arm [tRNA-Cys(−D)] found in the two South American annual killifishes is not found in any of the species examined here, all of which carry a complete tRNA-Cys [tRNA-Cys(+D)] of typical tRNA length. The A. limnaeus mtgenome also contains a duplicated tRNA-Leu2. Although uncommon, duplicated mitochondrial tRNAs have been previously reported, including the Cyprinodontiform Xenoteca eiseni (duplicated tRNA-Met; Setiamarga et al., 2008). Initially, the N. furzeri mtgenome was thought to contain a duplicated tRNA-Glu but more recently this putative tRNA was suggested to be a pseudogene (Hartmann et al., 2011; Tatarenkov et al., 2015).