INTRODUCTION

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION METHODS REFERENCES

Mitochondria provide the primary source of cellular ATP in eukaryotes via the process of oxidative phosphorylation (Saraste 1999). In animals, extranuclear mitochondrial genomes are typically circular, and with few exceptions, code for 13 subunits of the oxidative phosphorylation machinery as well as genes for two rRNA subunits and 22 tRNAs (Boore 1999). Mutations in mitochondrial DNA (mtDNA) have a number of known deleterious effects. At least 50 base substitutions and hundreds of insertion/deletion mutations have been identified in human mtDNA (MITOMAP 2000), with effects ranging from degenerative diseases (Wallace 1999) to aging (Michikawa et al. 1999) to cancer (Polyak et al. 1998; Fliss et al. 2000). In addition to their role as the powerhouse of the cell, mitochondria are also involved in regulating programed cell death (apoptosis) (Green and Reed 1998; Susin et al. 1999), and mutagenic reactive oxygen species are generated in the process of energy production (Croteau and Bohr 1997). Pathologies can result directly from the loss of ATP production in affected tissues, the build-up of oxygen radicals due to downstream blockage of the oxidative phosphorylation pathway, or unregulated apoptosis (for review, see Wallace 1999).

Hundreds of mitochondria and thousands of mtDNAs are inherited maternally through the cytoplasm of the oocyte (Lightowlers et al. 1997). If a zygote receives more than one form of mtDNA (heteroplasmy), different forms can be randomly distributed to daughter cells during cell division and, over many cell generations, can drift to high or low frequencies in various cell lineages (Hauswirth and Laipis 1982). Thus, if one of the mutant forms is deleterious, disease may affect lineages where it reaches sufficiently high frequency (Wallace 1999). Somatic mutations in mtDNA appear to behave similarly and may be a significant source of mitochondrial disease. Given the central role of mitochondria in cell physiology, mutations (either inherited or somatic) are probably responsible for many developmental abnormalities. Although a high frequency of mutant mtDNA molecules is likely to be lethal during embryogenesis, oocytes with moderate to low levels of heteroplasmy occur at detectable levels (Lightowlers et al. 1997). Mitochondrial mutations probably affect a number of both general and tissue-specific developmental processes; however, the role of mitochondria in early development has not been well characterized.

Structurally, most animal mitochondrial genomes contain the same 37 genes, and among vertebrates the gene order is highly conserved (Brown 1985; Boore 1999). Vertebrate mitochondrial genomes are typically ~16 kb and are extremely compact with no introns and few, if any, intergenic spacers. The only significant noncoding sequence is the control region, which is involved in regulating transcription and replication (Clayton 1982; Shadel and Clayton 1997) and is usually <5% of the total genome size. Since its endosymbiotic origin around 1.5 billion yr ago, a substantial fraction of original mitochondrial genes have moved to the nucleus (Gray et al. 1999). The products of many of these genes remain essential for oxidative phosphorylation or housekeeping functions and are selectively transported into the mitochondria after translation in the cytoplasm (Schatz 1996). A result of this evolutionary trend is that some mitochondrial abnormalities are due to mutations in genes that now reside in the nucleus and are inherited in a Mendelian fashion, rather than through the maternal, haploid inheritance of mtDNA (Wallace 1999). Knowledge of gene location (and thus mode of inheritance) is therefore essential for accurate characterization of the developmental-genetic basis of mitochondrial abnormalities.

Mitochondrial genomes from ~100 species have now been sequenced (Boore 1999; Curole and Kocher 1999). Their small size and relative autonomy from the nucleus makes mitochondrial genomes valuable windows on the process of genome evolution and with respect to cytonuclear interactions (Babcock and Asmussen 1996; Gray et al. 1999). Mitochondrial sequences have also proven to be of great utility in molecular phylogenetic studies, and complete genome sequences have provided valuable insights into deeper-level phylogenetic problems (e.g., Zardoya and Meyer 1997; Boore and Brown 1998; Mindell et al. 1998; Naylor and Brown 1998; Rasmussen and Arnason 1999).

The zebrafish, Danio rerio, (Actinopterygii, Cyprinidae) has become a prominent vertebrate genetic and developmental model system (Detrich et al. 1998). Here we describe the zebrafish mitochondrial genome with respect to gene content and organization, codon usage, nucleotide composition, and putative functional motifs. We also use a phylogeny based on the mitochondrial genomes of 36 vertebrate species to characterize the zebrafish relative to other vertebrates. We test evolutionary hypotheses pertaining to nucleotide and amino acid composition, genome-wide patterns of variability, and evolutionary rates among major vertebrate lineages. The wild-type zebrafish mitochondrial genome and its relation to other vertebrates may provide an important baseline for studies of development, disease, and mitochondrial function. It will also complement the sequence of the zebrafish nuclear genome that may be completed in the near future.

	RESULTS AND DISCUSSION

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION METHODS REFERENCES

The total length of the zebrafish mitochondrial genome was 16,596 bp, within 100 bp of the length of other teleost fishes. It is particularly close to two other cyprinid fishes, only 18 bp longer than the goldfish (Carassius auratus) and 21 bp longer than the carp (Cyprinus carpio). The zebrafish gene order and content (Table 1) is identical to the common vertebrate form. The genome contains 13 protein-coding genes: seven subunits of the NADH ubiquinone oxidoreductase complex (ND), one subunit of the ubiquinol cytochrome c oxidoreductase complex (Cytb), three subunits of the cytochrome c oxidase complex (CO), and two subunits of ATP synthase (ATP). Also contained are small (12S) subunit and large (16S) subunit ribosomal RNA genes and 22 tRNA genes. A noncoding control region contains the origin of heavy strand replication, and the origin of light strand replication is found within a complex of five tRNA genes. Sequence data provided no evidence of length or nucleotide site heteroplasmy in the individual zebrafish studied.

Protein-Coding Genes

Salient features of zebrafish mitochondrial genes are listed in Table 1. All but one of the protein-coding genes begin with the orthodox ATG start codon; the lone exception is COI, which begins with GTG. Stop codons include seven TAA and three TAG. The COII, ND4, and Cytb genes do not possess proper stop codons but do show a terminal T or TA. This condition is not uncommon among vertebrate mitochondrial genes, and it appears that TAA stop codons are created via posttranscriptional polyadenylation (Ojala et al. 1981). Reading frames of two pairs of genes, ATP8-ATP6 and ND4L-ND4, each overlap by seven nucleotides, and ND5-ND6 overlap by four nucleotides. This is also a common vertebrate feature, although in mammals ATP8 and ATP6 overlap by 40-46 bp. A number of other genes share one or two nucleotides in common with adjacent tRNA genes (Table 1).

Nucleotide composition (Table 2) reflects a vertebrate bias against G on the light strand. (For all protein genes except ND6, the heavy strand serves as the template for transcription; however, we discuss gene sequences in terms of the light strand, which is the sense strand with respect to mRNA. Only for ND6 is the heavy strand the sense strand). The bias against G is particularly marked at third codon positions, where only 7% of sites are G. A complementary bias is maintained in ND6, in which C is found at only 12% of all sites and only 3% of third positions.

Codon usage patterns in zebrafish are shown in Table 3. Three other vertebrates (another fish, a frog, and human) are shown for comparison. For amino acids with fourfold degenerate third positions, codons ending in A are always the most frequent in zebrafish, followed in frequency by codons ending in T or C. Among twofold degenerate codons, C appears to be used somewhat more than T. Consistent with the overall bias against G, G is the least common third position nucleotide in all categories except for arginine and glycine codons (where G is similar in frequency to C and T but still much less than A). These patterns are generally similar across vertebrate groups, although the frog, Xenopus, shows a tendency to use T more frequently than C (Roe et al. 1985).

Because there are only 22 tRNAs in the mitochondria, there is only one specific tRNA species for most amino acids. The exceptions are leucine and serine, which have two tRNAs serving six possible codons each. Thus, of the possible codons for any amino acid, only one (occasionally two) will be perfectly complementary to the available tRNA anticodon, and translation of the others must involve nonspecific (wobble) base pairing. For each amino acid, only codons ending in A or C will be perfectly matched by a complementary tRNA anticodon (Table 3). The matched nucleotide is A for fourfold codons and twofold purine codons, and C for twofold pyrimidine codons (except methionine, AUG, which also serves as a start codon). Thus in zebrafish, the most frequently used codons are those with matching tRNAs except for isoleucine, methionine, and phenylalanine. This indicates that there may be some advantage to matched codons and anticodons in protein translation, although, as seen in Xenopus, the phenomenon is not universal.

Codon bias may be associated with strand-specific nucleotide bias in mtDNA. It is noteworthy that, for ND6, 16 of the 20 amino acids are preferentially specified by codons that do not match the available tRNA. This is almost exclusively due to the high number of codons ending in T. Strand-specific substitution bias favoring A over T on the light strand and thus T over A on the complementary heavy strand appears to be the explanation. This bias is particularly evident at third codon positions, where substitutions are less likely to alter encoded amino acids and nucleotides are freer to vary. For all protein genes on the light strand (= sense strand), the frequency of A at third positions is 44% and the frequency of T is 27%. Conversely, for ND6 (sense strand = heavy strand) the frequency of A at third positions is 24% and the frequency of T is 52%. It therefore appears that nucleotide bias is a strand-specific phenomenon rather than being strictly a result of codon preference for translation efficiency. The preference for A may also be related to transcription efficiency, as ATP is generally the most common ribonucleotide available in mitochondria (Xia 1996). The substantial bias against G may also be due in part to selection against less stable G nucleotides on the light strand, which is exposed as a single strand for a considerable length of time during the asymmetrical replication of mtDNA (Clayton 1982).

Transfer and Ribosomal RNA Genes

All zebrafish mitochondrial tRNA genes possess anticodons that match the vertebrate mitochondrial genetic code. Each tRNA sequence may be folded into a cloverleaf structure with 7 bp in the aminoacyl stem, 5 bp in the TC and anticodon stems, and 4 bp in the DHU stem. tRNA stem regions include numerous noncomplementary and T-G base pairings, several of which are shared with carp. Such mutations appear to accumulate in mitochondrial genes, in part because mtDNA is not subject to the process of recombination, which may facilitate elimination of deleterious mutations (Lynch 1997). As in other vertebrates, it appears that CCA nucleotides are added posttranscriptionally to the 3' ends to form mature, functional species (Roe et al. 1985).

Both ribosomal gene sequences may be folded into secondary structures. Potential secondary structures have free energies of 220.3 kcal/mole for the 12S subunit and 335.6 kcal/mole for the 16S subunit. Stem regions appear to be conserved, whereas loop regions are somewhat more variable relative to other vertebrate sequences. The functional requirement for specific base pairing appears to constrain the evolution of stems relative to some portions of loops. This pattern is consistent with phylogenetic studies of a number of vertebrate groups (e.g., Sullivan et al. 1995).

Noncoding Sequences

The major noncoding region (control region) in mtDNA regulates replication and transcription (Clayton 1982,1991; Shadel and Clayton 1997). The primary sequence of much of the control region does not appear to be particularly important for regulatory function, as this region shows extensive variability across taxonomic groups and even among closely related species. The 950-bp zebrafish control region was much less similar to other fishes than were the coding sequences, with numerous nucleotide substitutions and insertions and deletions. However, several important regulatory elements are present. Conserved sequence blocks (CSBs) 1-3, found in the 3' end of the control region, appear to be involved in positioning RNA polymerase both for transcription and for priming replication (Clayton 1991; Shadel and Clayton 1997). All three of these elements are identifiable in zebrafish and they show strong similarity to CSBs identified by authors of other vertebrate sequences (Roe et al. 1985; Foran et al. 1988) (Table 4). Another relatively conserved element is the termination associated sequence (TAS) located at the 5' end of the control region. This sequence appears to act as a signal for termination of D-loop strand (7S DNA) synthesis. A sequence present in zebrafish (tacataaaatgcat) shows limited similarity with TAS sequences identified in other vertebrates. Regulatory elements in the zebrafish also show strong similarity to a North American cyprinid fish, the spotfin shiner (Cyprinella spiloptera; Broughton and Dowling 1994). Between tRNA^asn and tRNA^cys of zebrafish is a 32-bp noncoding sequence similar to the origin of light strand replication (O_L) in other vertebrates. Secondary structures at O_L appear to act as initiation signals for light strand replication (Wong and Clayton 1985). The zebrafish sequence may potentially form such a secondary structure consisting of a perfect 11-bp stem and a 14-bp loop. All of these features reflect strong conservation of vertebrate mitochondrial regulatory elements.

Mitochondrial Genome Evolution

Phylogeny

View larger version (23K): [in this window] [in a new window]   Figure 1   Maximum likelihood estimate of vertebrate phylogeny (ln likelihood = 239855.997). Likelihood analysis used the GTR + I +  model with parameter values estimated from the maximum parsimony tree. This model was significantly better by likelihood ratio test than the Tamura-Nei + I +  model (X2 = 15,792, d.f. = 3, P < 0.001). The tree shown had significantly greater likelihood for the present data than the generally accepted vertebrate tree, the maximum parsimony tree, and the LogDet distance tree indicated by the Shimodaira-Hasegawa (Shimodaira and Hasegawa 1999) test (P < 0.001 in each case). Scale bar indicates the probability of change per nucleotide site over the branch length interval shown. Lamprey was the designated outgroup.

Evolutionary Genomic Patterns

RATES OF EVOLUTION ACROSS THE MITOCHONDRIAL GENOME

View larger version (42K): [in this window] [in a new window]   Figure 2   Variability of nucleotides, amino acids, and hydropathy across the mitochondrial genome. The sequence is of the concatenated protein-coding genes and the bar at top indicates the limits of each gene. The sequence for ND6 is the reverse complement to maintain a consistent reading frame. Plots are the mean values for all 36 vertebrate taxa with a sliding window of 10 for amino acids and third-position nucleotides or 20 for (combined) first- and second-position nucleotides.

RATES OF EVOLUTION AMONG LINEAGES

NUCLEOTIDE AND AMINO ACID FREQUENCIES

Conclusions

Despite extensive variation in mitochondrial genome structure among animal phyla, gene order and content are identical among zebrafish and the majority of vertebrates. Of 56 known vertebrate mitochondrial genomes, 44 have this same gene order, including all 30 placental mammals and 12 fishes (all but the sea lamprey) (Boore 1999). Gene order variants exist in some birds (Desjardins and Morais 1990), crocodilians (Janke and Arnason 1997), and snakes (Kumazawa et al. 1998); however, their taxonomic scope tends to be limited. The patterns of strand-specific nucleotide bias and unequal codon usage seen in the zebrafish are also conserved among vertebrates. In contrast, there are substantial differences in evolutionary rates and nucleotide and amino acid composition among vertebrate groups. The most striking finding is that rates of sequence evolution vary widely both within and among major groups, whereas nucleotide composition tends to be conserved within groups but varies substantially between mammals and all other taxa. This trend is also shown at the amino acid level where lower GC percent in mammals (particularly at first and second positions) results in amino acids with A + T-containing codons being more frequent than those with G + C-containing codons. All other taxa show higher GC percent and corresponding differences in amino acid composition. Whether this is based on differences in nucleotide bias at the genome level or on adaptive significance of proteins with different amino acid composition is not clear. However, these differences appear to be independent of selection on specific genes and metabolic rate or thermal habit.

	METHODS

TOP ABSTRACT INTRODUCTION RESULTS AND DISCUSSION METHODS REFERENCES

Mature specimens of Danio rerio, strain ABC, were obtained from the University of Oregon Zebrafish Facility. Total DNA was isolated from ~25 mg of muscle tissue from a single individual. The preparation used the QIAamp tissue kit (QIAGEN) with DNA resuspended in 10 mM Tris (pH 8.0). Generation of sequencing templates used shotgun cloning of mtDNA. The mitochondrial genome was amplified by long PCR in two overlapping fragments of ~9 kb each. Reactions used Herculase DNA polymerase blend (Stratagene) according to supplier recommendations. One primer set consisted of 16S-AH (5' atgtttttggtaaacaggcg 3') located in the 16S rRNA gene and Arg-BL (5' caagacccttgatttcggctca 3') in the arginine tRNA gene. The other primer set consisted of 16S-BL2 (5' tggtgcagccgctattaagg 3') in the 16S rRNA gene and NAP-2H (5' tggagcttctacgtgrgcttt 3') in the ND4 gene. The two fragments overlapped over ND4L and part of ND4. The small gap between the divergent 16S primers was amplified (and sequenced separately) with primers 16S-ArL (5' ctcggcaacacaagcctcgc 3') and 16S-BrH (5' tyayagatagaaactgacctgg 3'). Products of long PCR were purified with QIAquick PCR purification columns (QIAGEN) and randomly fragmented using a nebulizer. Broken ends of DNA strands were repaired with Klenow polymerase and phosphorylated with T4 polynucleotide kinase. Fragments in the range of 600-3000 bp were selectively recovered from a 1.0% agarose gel. These fragments were blunt-end ligated into (SmaI cut, dephosphorylated) pUC19 at 4°C overnight and transformed into Escherichia coli JM109 cells. White colonies were picked and grown in 96-deep well blocks (1.75 mL Luria Bertani medium plus ampicillin per well).

Plasmids from four blocks (= 384 clones) were purified via alkaline lysis methods using a Hydra microdispenser (Robbins Scientific). Sequencing used universal primers, BigDye terminator chemistry, and ABI 377 instruments. Individual sequences were assembled and analyzed with phred/phrap/CONSED (Ewing et al. 1998; Gordon et al. 1998). A few small gaps were closed via sequencing with specific primers on whole-genome templates generated via long-PCR using primers 16S-AH and 16S-BL2. Sequence coverage was sufficient to achieve a CONSED error rate of 0.01%. The zebrafish sequence was manipulated and aligned with Sequencher (GeneCodes). Genes were identified by alignment and comparison with sequences from other vertebrate taxa, particularly the common carp (Cyprinus carpio; Chang et al. 1994) and goldfish (Carassius auratus; Murakami et al. 1998). The complete sequence is available under GenBank accession no. AC024175.

Protein-coding genes from 35 vertebrate species obtained from GenBank were concatenated and aligned using Sequencher and corrected by eye to preserve reading frame. Regions of gene overlap and regions of unique insertions (gap in all but one sequence) were excluded, yielding a total alignment of 11,592 nucleotides. Of these, 8,189 were variable and 7,354 were parsimony informative (variants occurring in two or more sequences). Phylogenetic analyses using maximum parsimony, maximum likelihood, and neighbor-joining were performed with the computer program PAUP* ver. 4b8 (Swofford 2001). Parsimony searches used 100 random addition sequence replicates with TBR branch swapping and equal weighting of all characters/character-states. Maximum likelihood analysis used empirical mean nucleotide frequencies and empirical transition:transversion ratios. Other model parameters for likelihood were obtained by optimization on trees obtained either from maximum parsimony or neighbor-joining with LogDet distances. These parameters included the shape parameter (?) for a discrete (four-class) gamma approximation of among site rate variation, the estimated proportion of invariant sites, and values for nucleotide change-rate matrices. The choice of likelihood rate model was based on model performance using the likelihood ratio test. Given the large numbers of taxa and characters, maximum likelihood analyses were computationally intensive and only a few rearrangements (<200) were evaluated for each search. However, all likelihood searches using the GTR model quickly converged on the same tree, regardless of starting parameters. Calculation of codon usage and nucleotide and amino acid frequencies used the computer program DAMBE ver. 4.0.43 (Xia 2000) and relative rate tests were conducted with MEGA ver. 2.0 (Kumar et al. 2001).