Analysis of the Complete Mitochondrial Genome Sequence of the Diploid Cotton Gossypium raimondii by Comparative Genomics Approaches

Cotton is one of the most important economic crops and the primary source of natural fiber and is an important protein source for animal feed. The complete nuclear and chloroplast (cp) genome sequences of G. raimondii are already available but not mitochondria. Here, we assembled the complete mitochondrial (mt) DNA sequence of G. raimondii into a circular genome of length of 676,078 bp and performed comparative analyses with other higher plants. The genome contains 39 protein-coding genes, 6 rRNA genes, and 25 tRNA genes. We also identified four larger repeats (63.9 kb, 10.6 kb, 9.1 kb, and 2.5 kb) in this mt genome, which may be active in intramolecular recombination in the evolution of cotton. Strikingly, nearly all of the G. raimondii mt genome has been transferred to nucleus on Chr1, and the transfer event must be very recent. Phylogenetic analysis reveals that G. raimondii, as a member of Malvaceae, is much closer to another cotton (G. barbadense) than other rosids, and the clade formed by two Gossypium species is sister to Brassicales. The G. raimondii mt genome may provide a crucial foundation for evolutionary analysis, molecular biology, and cytoplasmic male sterility in cotton and other higher plants.


Introduction
Eukaryotes have three genomes, one nucleus and two organelles. The nuclear genome carries the overwhelming majority of information, but the chloroplast and mitochondrial genomes are nonetheless indispensable as well. With the development of next-sequencing technology, a dramatic increase in sequences of complete organelle genomes has been witnessed in the past several years. Due to their smaller sizes and conserved structures, the complete cp genome sequences were determined more frequently than mt genomes. In contrast to the conserved structures of cp genomes, mt genomes are specific to each plant in their variable sizes [1,2], complex structures [3][4][5], multiple RNAediting processes [6,7], frequent reorganizations, and gene loss during evolution [8][9][10][11].
Plant mt genomes, encoding the main manufacturers of cellular ATP, play vital roles in the regulation of cellular metabolism [12]. In seed plants, the size of the mt genome is highly variable, ranging from 208 Kb in Brassica hirta [13] to 11.3 Mb in Silene conica [14]. However, the functional genes in these genomes are quite conservative [15]. The variable sizes of plant mt genomes may be due to expansion of noncoding sequence and duplication of a large segment [16]. The structures of plant mt genomes can be quite complex. Most mt genomes are a major circular molecule but Cucumis sativus has three circular molecules [17]. Additionally, some mt genomes are linear molecules, such as Oryza sativa ssp. japonica [18]. While plant cp genomes have very conservative gene order, mt genomes often do not. For example, the cp genomes of rice, maize, and wheat share almost the same gene orders but their mt genomes are completely different [18][19][20], further reflecting the complexity of plant mt genomes.
Plant mt genomes are rich in repeats, including tandem repeats, short repeats, and large repeats [9,21]. Many experiments showed that mitochondrial repeats contain a lot of genetic information and are also vital components of intramolecular recombination [11]. Methylated sites in maize mt genomes are related to tandem repeats [19]. Short repeats, usually ranging from dozens to several hundred bp, play vital roles in the evolution of plant mt genomes and may be responsible for structural variations and variable sizes in higher plant mt genomes [11]. The mt genome of Cucurbita pepo contains tens of thousands of short repeats (total 371 kb), resulting in a nearly 1 Mb genome [21]. The large repeats (>1 kb) can be divided into direct repeats and inverted repeats, the former accounting for a larger proportion of the total. Genes in these larger repeats or short repeats (100-1000 bp) may have multiple copies. In wheat, atp6, atp8, rrn5, rrn18, rrn26, trnQ, trnK, trnfM, trnD, and trnP have multiple copies because they are in the 11 repeats of wheat mt genome (7 large and 4 short) [20]. Additionally, atp1, nad1, rps3, and nad2 in maize [19], atp6 in Arabidopsis [22], rrn26 and trnfM in beet [23], cox2 in rapeseed [6], and sdh3, trnQ, and trnG in watermelon [21] also have multiple copies due to their mt repeats. Further, recombination across large direct repeats may separate the mt genome into pairs of subgenomic molecules, whereas large inverted repeats may generate some isomeric circles [6,11,17,19,20,22].
RNA-editing is a posttranscriptional process, which occurs in mt and cp genomes of higher plants [24,25]. Most RNA-editing sites may convert hydrophilic amino acids to hydrophobic amino acids, contributing to better folding of protein. Plant mt genomes are known to contain 441 RNAediting sites in 36 Arabidopsis genes [22], 427 in 34 rapeseed genes [6], 357 in 31 beet genes [23], 491 in 34 rice genes [18], 463 in 37 watermelon genes [21], and 444 in 37 Cucurbita pepo genes [21]. Closely related species share more RNA-editing sites than more distant species [25]. For example, Arabidopsis and rapeseed share 84% of RNA-editing sites; watermelon and Cucurbita pepo share over 90%. Most of RNA-editing process occurs in C to U residues, which may generate initiation, internal, or termination codon with unpredictable function [6]. Unpredictable function of mt genomes may be associated with cytoplasmic male sterility (CMS).
Many genes originally found in mt have been lost during the evolution of plant mt genomes, and the loss is related to function and evolution [12]. For example, sdh2 was lost in the early evolution of plant mt genomes; rps9, rps11, and rps16 were lost in the differentiation between higher plants and lower plants [18]; rps12, sdh3, and sdh4 were lost in monocots and rps2 was lost in dicots [26]. Because of their very low nucleotide mutation rate [5,27], seed plant mitochondrial genes are often utilized in plant evolutionary analysis, especially phylogenetic analysis [28][29][30]. The phylogenetic tree constructed by some conserved mt genes could be beneficial to illustrate the evolution of the constituent genes in mt genomes of seed plants [11].
Cotton, native to tropical and subtropical regions, is one of the most important economic crops and is a primary source of natural textile fiber [31]. Its seed oil and also byproducts of cotton processing are potential raw materials for the production of biofuels. Apart from its economic value, cotton is also an excellent model system for studying polyploidization, cell elongation, and cell wall biosynthesis [32][33][34]. Recently, two Gossypium mt genomes (G. barbadense and G. hirsutum) have been published [35,36], and they are both tetraploid cottons, accounting for about 94% of commercial cotton production.
Here, we report the complete mt genome of G. raimondii (GenBank accession number: NC 029998.1), one of the putative progenitor species of tetraploid cotton [37]. The three previously published Gossypium mt genomes, with rather similar structures to G. raimondii, could be used as references for the G. raimondii mt genome. In this study, the complete mitochondrial DNA sequence of G. raimondii was assembled into a circular genome of length of 676,078 bp and then further studied for its gene content, frequent reorganizations, cp-derived tRNA genes, RNA-editing, and similarities and differences with other higher plant mt genomes. The G. raimondii mt genome may provide a crucial foundation for evolutionary analysis, molecular biology, and cytoplasmic male sterility in cotton and other higher plants.

Plant Materials and Genome
Sequencing. The raw reads used in this study were available in the NCBI Sequence Read Archive (SRA) under accessions SRX027534 and SRX027436 [37]. The raw reads were sequenced with a combination of Roche/454 GS FLX sequencing (14.95x linear and 3.1x nonredundant pairs assembled coverage) and Illumina based short reads (primarily to correct 454 sequencing errors) at the Joint Genome Institute (http://www.jgi.doe.gov/sequencing/ protocols/prots production.html) [37].

Mitochondrial Genome
Assembly. The procedure for assembling this mt genome was similar to the method described by Zhang et al. [38]. First of all, we assembled all the raw reads using Newbler (version 3.0) with the following parameters: -cpu 20, -het, -sio, -m, -urt, -large, and -s 100. According to the Newbler manual, the contigs generated by Newbler are constructed with the trimmed reads and there are nearly no overlaps among these contigs. The original contigs were a mixture of nuclear and organellar DNA. In order to isolate mt contigs from the whole genome contigs, we used NCBI-BLASTn with default parameters to search against the reference mt genome sequences of 368 plants from NCBI Organelle Genome Resources (including three cottons, G. harknessii, G. barbadense, and G. hirsutum). The read depths of most similar mt contigs were between 15 and 50. Therefore, we set the coverage between 15 and 50 to filter these similar mt contigs out. To visualize the connections among similar mt contigs, we used Perl scripts and a file named "454AllContigGraph.txt," which was generated from Newbler. Moreover, we removed false links and forks in the assembly graph and used all the raw reads to fill the gaps (usually allowing zero or one mismatched base pair) presented in two connected contigs. After assembling the whole mt genome by 454 GS FLX sequencing reads, we used Illumina sequencing data to validate the final genome assembly by BWA [39] and SAMtools [40] with default parameters.

Genome Annotation.
The mt genome of G. raimondii was annotated using the method described by Alverson et al. [21]. A local database was built with the known mt genomes of angiosperms, which contained almost all the protein and ribosomal RNA (rRNA) genes of previously sequenced plant mt genomes [35]. The mt genome of G. raimondii was used as a query sequence against the database by using local NCBI-BLASTn to identify protein-coding and rRNA genes. The tRNAscan-SE was used to annotate transfer RNA (tRNA) genes with default settings [41]. We manually revised the start and stop codons of protein-coding genes by using local NCBI-BLASTn against homologous genes from other sequenced mt genomes. All these genes were confirmed manually to verify the results. The GC content was analyzed by a local Perl script. The physical circular map of the mt genome was drawn using OrganellarGenomeDRAW (OGDRAW) [42].

Analysis of Repeat Sequences.
Tandem repeats in the mt genome of G. raimondii were identified using Tandem Repeats Finder version 4.07 with default settings [43]. The MISA Perl script [44] was used to detect simple sequence repeats (SSRs) with a motif size of one to six nucleotides and thresholds of eight, four, four, three, three, and three, respectively. As described in other plant mt genomes [9,21,35], AB-blast was used to further identify repeat sequences in G. raimondii by searching against itself with the following parameters: = 1, = −3, = 3, and = 3, kap, span, = 1 × 10 9 , and = 7. All BLAST hits withvalue ≤ 1 were considered repeats. All repeats identified with the various programs were manually verified and nested or redundant results were removed.

Migration of Protein-Coding Genes and tRNA Genes.
Protein-coding genes transferred from the cp genome to the mt genome were identified with NCBI-BLASTn search of protein-coding genes in the G. raimondii annotated cp genome against the mt genome (identity ≥ 80%; -value ≤ 1 −10; and coverage ≥ 50%). The G. raimondii mt genome was searched against all tRNA genes of its cp genome to detect cpderived tRNA genes using NCBI-BLASTn (identity ≥ 80%; E-value ≤ 1 − 10; and coverage ≥ 50%). Protein-coding and tRNA genes transferred from the mt genome to its nuclear genome were identified with the same method utilized above. The NCBI-BLASTn version is 2.2.30+.

Analysis of RNA-Editing and Substitution Rate.
Possible RNA-editing sites in protein-coding genes of the G. raimondii mt genome were identified using the online program Predictive RNA Editor for Plants (PREP) sites, based on the evolutionary principle that editing increases protein conservation among species (http://prep.unl.edu/) [45]. For this analysis, the cut-off value was set as = 0.6 to achieve accurate prediction. Protein-coding genes from land plant mt genomes in the PREP-mt program were used as references for inferring RNA-editing sites in G. raimondii mt genome.
To analyze the synonymous ( s ) and nonsynonymous ( a ) substitution rates of G. raimondii mt protein-coding genes in comparison to other higher plants, C. papaya and P. tremula were selected as references. Corresponding protein-coding genes of the three mt genomes were extracted and aligned separately with ClustalW in MacVector version 14.0.4 (37). Synonymous ( s ) and nonsynonymous ( a ) substitution rates for each gene were estimated in DnaSP v5. 10.01 with default settings [40]. All RNA-editing sites and substitution rates were confirmed manually.  [47] implemented in MEGA 6.0 with default parameters. As described in other plant mt genomes [9,11,21,35], the Maximum Likelihood (ML) tree based on the General Reversible Mitochondrial model was then constructed with the result of phylogenetic analyses in MEGA 6.0. Moreover, a discrete Gamma distribution was used to model the evolutionary rate differences among sites (5 categories). The bootstrap index value (%) of each branch was calculated by 1000 replications. Meanwhile, the neighborjoining (NJ) tree was also constructed with 1000 bootstraps of replications. M. polymorpha was set as the outgroup. All positions containing gaps and missing data were eliminated.

Genome Assembly and Correction.
The raw sequencing reads of G. raimondii were sequenced by the Roche/454 GS FLX platform. A total of 1,649,158 reads covering 1.23 Gb were generated, with average length of 744 bp (Table 1). After assembling the raw reads using Newbler software, a total of 140,540 contigs (longest length: 120,130 bp; average length: 989 bp) of total length of ∼85 Mb were generated, with contig N50 value of 1,073 bp. There were also some contigs with length > 5 kb and high coverage, which were separated with other shorter contigs. We selected 4,358 contigs with read depths between 15 and 50 as candidates for assembling the complete mt genome. These candidates were then aligned to the reference mt genomes of 368 other plants from NCBI Organelle Genome Resources using NCBI-BLASTn. Next, we selected 15 contigs (>5 kb) to construct the initial mt contig graph with Perl scripts. In combination with the file "454ContigGraph.txt" generated from Newbler, we finally obtained 21 contigs to construct the complete graph for assembling the G. raimondii mt genome. As shown in Table S1 in Supplementary Material available online at http://dx.doi.org/10.1155/2016/5040598, two contigs (contig01360 and contig18936) resemble cp-like sequences with their high coverages. However, the two contigs are essential for assembling the complete mt genome with the connection observed in "454ContigGraph.txt," indicating that the two contigs may have been transferred from their cp genome. Among the 21 selected contigs, 7 (contig00004, contig00007, contig00008, contig00018, contig00019, con-tig00021, and contig90952) were assembled into the complete mt genome twice or more, while the other 14 were assembled only once (Table S1). That indicated that the high coverage of contigs was related to the multiple copies in the genome. After connecting the 21 contigs, there were still some gaps (usually zero or one mismatched base pair) between two connected contigs. These gaps were filled by the following method: first, we mapped the raw reads onto both ends (3-60 bp) of the assembled contigs and then extended the contigs by joining the reads, which partly overlapped (identity ≥ 95% and -value ≤ 1 − 30). Raw data generated from the Roche/454 GS FLX platform might contain a lot of sequencing mistakes because of the complex repeats, especially in simple sequence repeats. To validate the genome assembly, we used two Illumina (Illumina Genome Analyzer II) runs (12.7 Gb) from a standard 400-base pair (bp) fragment library. Using BWA and SAMtools, we mapped the Illumina sequencing data onto the complete mt genome and filtered the matched reads with Perl scripts. After reassembling these matched reads with Newbler software, a total of 100 contigs (longest length: 65,661 bp; average length: 9,591 bp) with total length of 588,612 bp were generated, and the contig N50 value was 18,050 bp. Next, we mapped the 100 contigs onto the original mt genome in MacVector version 14.0.4 to correct the mistakes (mainly in A/T enriched regions). Finally, the complete G. raimondii mt genome was finished (http://www.ncbi.nlm.nih.gov/nuccore/NC 029998.1), comprising 676,078 bp with GC content of 44.95%.  (Table S2).
In seed plants, the size of the mt genome varies a lot, but functional genes were quite conservative. As shown in Table S2      G. raimondii, 44.95%). The GC content of angiosperm mt genomes is smaller than that of gymnosperms but larger than that of bryophytes, indicating that GC content changed a lot during plant evolution. Further, the GC content of proteincoding genes was smaller compared to other regions (Table 3) and the distinction may be associated with the regulation of gene expression and gene mutation.

Gene Clusters and Repeat
Sequences of the G. raimondii mt Genome. In plant mt genomes, the sequences of proteincoding genes were highly conserved, whereas the relative order of these genes varied due to frequent rearrangement during evolution, with only a few highly conserved gene clusters preserved. In this research, we found thirteen conserved gene clusters in the G. raimondii mt genome (Table 4). Comparing the results with that in G. hirsutum [35], we found that the two cotton species shared almost the same gene clusters; only a few differences were found in the interval lengths between two genes. Additionally, two clusters (rrn5-rrn18 and mttB-nad9) contain two copies because they located in two large repeats of G. raimondii mt genome. Only one cluster (rpl2-rpl5-nad5c; c indicates exon3) consists of three different genes, while others consist of two genes (sdh4-cox3, cox1- nad5ab-atp9, and nad2abc-sdh3; a, b, c, d, and e indicate exon1, exon2, exon3, exon4, and exon5, resp.). Short intergenic regions or even partially overlapping sequences in coding genes, ranging from 28 bp to 1353 bp, are usually found between two clustered genes. Comparing the interval lengths with those in G. hirsutum, we found that the intergenic regions were extremely conserved in the similar species [6,19,21,35,48]. What is more, comparing the type of gene clusters with those of other higher plants [6,19,21,35,48], rrn5-rrn18, nad3-rps12, and rps3-rpl16 are conserved in most of seed plant mt genomes. There are also gene clusters that are specifically conserved in dicots (nad1d-matR, cob-rps14, adh4-cox3, and atp4-nad4L) and monocots (ccmFn-rps1-matR-nad1e), respectively. The atp4-nad4L gene cluster exists in most of dicots, except for G. raimondii and other cottons [49], suggesting lineage specific disruption of this gene cluster in cotton.
Research into gene clusters in plant mt genomes is important because genes in the same cluster may share the same promoter and may be transcribed from the same strand. For example, the gene cluster of rps3-rpl16-nad3-rps12 in rice shares a common promoter and functions in a coordinated manner [50]. Moreover, these clusters may also be important for predicting coexpressed genes or potential function of clustered genes in angiosperms [48]. Boldface indicates interval length between two cluster genes. f Gene clusters contain two copies. a, b, c, d, and e followed with nad represent exon1, exon2, exon3, exon4, and exon5, respectively. Repetitive sequences, including simple sequence repeats (SSR), tandem repeats, short repeats, and large repeats, were analyzed. A total of 674 SSRs were identified in the G. raimondii mt genome, among which nearly 80% belong to monomers (44.5%) and dimers (35.5%), whereas trimers, tetramers, pentamers, and hexamers occurred with lower frequency. Of the monomers, A/T sequence (90.67%) occupied the main proportion, while G/C was only 9.33%. Further, the presence of trimers was 3.41%, while that of tetramers and pentamers was 14.54% and 1.78%, respectively. Only two hexamers (ACCAAT and TTCTCT) were observed in the mt genome. The specific size and location of pentamers and hexamers are shown in Table 5. Among the 14 polymers shown in Table 5, only one was localized in a coding region (matR), two were in cis-introns (nad4-intron3 ×2), and the others were all in intergenic spacers. As shown in Table 6, a total of six tandem repeats with lengths ranging from 15 bp to 42 bp and 100% sequence identity were also identified in the G. raimondii mt genome. All of the tandem repeats were identified in intergenic spacers of nad4-exon1/trnD-GUC (×2), nad7-exon5/ccmB, nad2-exon5/rpl2, rps14/rps4, and sdh3/atp4. SSRs and tandem repeats in the G. raimondii mt genome are unevenly distributed, being heavily concentrated in intergenic spacers, indicating that they mainly appear in noncoding regions.
Besides SSRs and tandem repeats, there are also 487 repeats (total length: 234,389 bp; 34.67% of the genome) identified in the G. raimondii mt genome. As shown in Figure 2, most of repeats are 20 bp to 39 bp long (242 repeats, 49.97%), and about 10.68% (52 repeats) are longer than 100 bp (Table 7), with only four longer than 1 kb (R1: 63,905 bp; R2: 10,624 bp; R3: 9,130 bp; and R4: 2532 bp). Large repeats (>1 kb) are notable because they are associated with reversible genomic structural changes. Pairwise large direct repeats and inverted repeats may produce two small subgenomic circles or isomeric circles, respectively. R1, R2, and R4 repeats share the same orientation, while R3 has opposite orientation. Some genes appearing in repeats have multiple copies. For example,  rrn5, rrn18, mttB, nad9, nad4, and trnD-GUC in R1 have an intact copy; trnM-CAU-2 and rrn26 in R3 have an inverted copy; R16 is responsible for the third inverted copy of trnD-GUC; R30 is responsible for the intact copy of cp-derived trnM-CAU; R c is responsible for the inverted copy of trnW-CCA. Additionally, most of these repeats, larger than 100 bp, have two copies (34 repeats), while 14 repeats have three (Table 7). Large repeats are commonly found in many plant mt genomes, such as Arabidopsis (6.5 and 4.2 kb), rapeseed (2.4 kb), and sugar beet (6.2 kb), and have been found to be active in intramolecular recombination in the evolution of higher plants. Different species show different DNA sequences associated with intramolecular recombination, contributing to the complexity of plant mt genomes. Genes present in larger repeats may generate more copies. Comparing the mt genomes of G. raimondii to Arabidopsis, rapeseed, and sugar beet, the genes present in large repeat sequences are completely different from each other. In Arabidopsis, rapeseed, and sugar beet, the genes present in large repeats are atp6 and orf139, cox2, and rrn26, respectively. Because of its large R1 repeats, G. raimondii generates more gene copies (nad4, nad9, mttB, rrn5, rrn18, rrn26, trnW-CCA, trnM-CAU, and trnD-GUC). R1 repeats (∼64 kb) are also found in another cotton, G. barbadense, resulting in seven gene copies (atp1, mttB, nad4, nad9, rrn5, rrn18, and trnD-GUC). 3.4. DNA Transfer between Mitochondria and Nucleus of G. raimondii. As described by Chang et al. in soybean mt genome, DNA transfers from nucleus to mitochondria are more difficult than the reverse transfers [11], perhaps because the mt genome is much smaller than the nuclear genome, so that the smaller genome cannot get enough locus to accept transfers. Moreover, DNA transfers from nucleus may not survive in mt genome because of the negative effect on energy metabolism played by the transferred DNA.
In our study, nearly all G. raimondii mt protein-coding genes shared a high similarity (most identity > 99%) with sequences in its nuclear genome, except nad3 and rps12 (Table  S4). These genes, presumably transferred from the mt to the nuclear genome, are referred to as mt-like nuclear genes. Almost all of these protein-coding genes have a complete intact copy in the nuclear genome except nad7, for which only the first two of its five exons are found in the nuclear genome (Table S4). Most of sequence identities between corresponding genes in the mt and nuclear genomes are larger than 95%, indicating recent migration of these mt-like nuclear genes. Seven of the 13 G. raimondii chromosomes have mt-like nuclear genes. Some mt-like nuclear genes have more than one copy in nuclear genomes, and the multiple copies may appear on different chromosomes. For example, ccmFc has three copies, one each on chromosomes (Chr) 1, 2, and 13, respectively; ccmFn, cox1, cox3, and rps10 have two copies on Chr1 and 13; matR has two copies on Chr1 and Chr5; rpl5 has two copies on Chr1 and Chr8. Additionally, most of mt-like nuclear genes appeared on Chr1 (72.86%, 51 genes or exons) and Chr13 (14.29%, 10 genes or exons), whereas only 12.86% (9 genes or exons) appeared on the other five chromosomes (Chr2, Chr5, Chr8, Chr9, and Chr10).
Previous study had identified a Chr1 region which included many genes closely resembling mt homologs in G. raimondii [37]. Here, we not only confirmed it on the basis of the enrichment of mt-like nuclear genes on Chr1 but also identified all these genes clearly, including complex I genes (nad1, nad2, nad4, nad4L, nad5, nad6, nad7-exon1, nad7-exon2, and nad9), complex II genes (sdh3 and sdh4), complex III genes (cob), complex IV genes (cox1, cox2, and cox3), complex V genes (atp4, atp6, atp8, and atp9), cytochrome c biogenesis genes (ccmB, ccmC, ccmFc, and ccmFn), Maturase gene (matR), transport membrane protein (mttB), and ribosomal proteins (rpl2, rpl5, rpl10, rpl16, rps3, rps4, rps7,  rps10, and rps14). All of these genes are distributed in a 655.3 kb region (23,164,259-23,819,545 nt) on Chr1. Further,  ccmFc, cox1, cox3, and rps10). To compare the Chr1 and Chr13 regions, we used NCBI-BLASTn to align the two chromosomes and did find similar regions. Note that gene orders in plant mt genomes differ tremendously during evolution. However, by comparing Figure 1 and Table S4, we found that the gene orders between Chr1 and mt genomes were relatively conserved, and the clusters of conserved gene orders are as follows: rpl2-ccmFN (13 genes), ccmFc-rps10 (5 genes), atp9-atp8 (5 genes), ccmB-nad2 (6 genes), nad6-atp4 (5 genes), and mttB-nad4 (3 genes). The mt-like nuclear genes on Chr13 (nad1-cox3, 6 genes) also share the same order after transfer from the G. raimondii mt genome. What is more, as shown in Table S4, the identity between mt and mt-like nuclear genes is very high on Chr1. Further analysis found that there were only 38 coding SNPs out of 28440 bp between mt and mt-like nuclear genes. Large transferred sequences are not under purifying selection, since their genes likely have not acquired nuclear promoters. The mt-like nuclear genes on Chr1 and Chr13 have the entirely similar orders and high identities (>99.9%) compared to that in the G. raimondii mt genome, suggesting that the transfer event of mt to nuclear genome might be recent in the evolution of plants.
The tRNA genes are transferred from G. raimondii mt to nuclear genomes frequently, creating mt-like nuclear tRNA genes. Nearly all G. raimondii mt tRNA genes had one copy (identity > 90%) in the nuclear genome (Table S5). They are distributed on 10 chromosomes, being absent from Chr3, Chr6, and Chr12, and are most abundant on Chr1 (23 tRNA genes, 52.27%), Chr7 (4 tRNA genes), and Chr13 (5 tRNA genes), while other chromosomes only have 12 mt-like nuclear tRNA genes. Interestingly, the mt-like nuclear tRNA genes on Chr1 are also located in the 655.3 kb region rich in mt-like nuclear genes (Table S5), suggesting that proteincoding genes and tRNA genes may have been transferred to the nucleus at the same time. Apart from protein-coding and tRNA genes, the intergenic DNA sequences of G. raimondii mt genomes were also transferred to nucleus during evolution. As illustrated in Table S6, a total of 16 long syntenous blocks were identified with high identities (>99.5%) between G. raimondii mt and nuclear genomes. The results mentioned above indicated that nearly all of the G. raimondii mt genome has been transferred to nucleus on Chr1, and the transfer event must be very recent because the identities of syntenous blocks are very high and the gene orders are extremely similar between mt and nuclear genomes.

Protein-Coding and tRNA Genes Transfer from cp to mt
Genome. We only found some fragmentary sequences of G. raimondii cp protein-coding genes transferred to the mt genome. These fragments are so small that none cover even 50% of their complete genes. In contrast to protein-coding genes, tRNA genes were transferred from cp to mt genomes frequently. We identified 15 native (i.e., of mitochondrial origin) G. raimondii tRNA genes and 10 cp-derived (plastid originated; 40% of all tRNA genes) tRNA genes in the G. raimondii mt genome (Table 8). These cp-derived tRNA genes include trnD-GUC (×3), trnH-GUG, trnN-GUU, trnS-GGA, trnW-CCA (×2), and trnM-CAU-cp (×2). Noting the total of 25 tRNA genes identified in the G. raimondii mt genome, some tRNA genes (trnA, trnL, trnR, and trnT) have been lost during the evolution of higher plants, indicating that the role of these missing tRNA genes may be supplied by either cp or nuclear genomes.
We also identified native and cp-derived tRNA genes in six other plants, including one bryophyte (M. polymorpha) [51], one gymnosperm (C. taitungensis) [52], two monocots (O. sativa and T. aestivum), and two dicots (A. thaliana and N. tabacum) (Table 8) [53,54]. Six cp-derived tRNA genes (trnH-GUG, trnM-CAU, trnN-GUU, trnP-UGG, trnS-GGA, and trnW-CCA) were common to the mt genomes of all angiosperms. The cp-derived tRNA gene (trnD-GUC) is only common to dicots. Figure 3 shows that the proportions of cp-derived tRNA genes in bryophyte and gymnosperm mt genomes (M. polymorpha and C. taitungensis) are all below 20% [51,52], whereas the proportions in other angiosperm mt genomes range from 24% (T. aestivum) to 50% (O. sativa) [6,18,53,54]. This implies that the transfer of tRNA genes from cp to mt is more frequent in angiosperms than in bryophytes and gymnosperms. Indeed, the loss of cp-derived tRNA genes in the bryophyte mt genome (M. polymorpha) indicates that cp-derived tRNA genes may only occur in flowering plants.

Introns and RNA-Editing Sites in the G. raimondii mt
Genome. In the G. raimondii mt genome, twenty-one cisspliced introns (length = 35,710 bp; 5.28% of the mt genome) were identified in ten genes (nad1, nad2, nad4 (×2), nad5, nad7, ccmFc, rps3, cox2, and rps10) and five trans-spiced introns were identified in the three NADH dehydrogenase genes (nad1, nad2, and nad5). The numbers and positions of introns are conserved in seed plant mt genomes. As shown in Table 9, 19 cis-spliced and 5 trans-spliced introns were found in B. napus [6], 17 cis-spliced and 6 trans-spliced introns were found in N. tabacum [54], and 18 cis-spliced introns and 6 trans-spliced introns were found in T. aestivum [53]. Compared with the other two dicots (B. napus and N. tabacum), a cis-spliced intron of the rpl2 gene is lost in the G. raimondii mt genome and is also lost in most of higher plants.
The trans-spliced introns were only found in nad1, nad2, and nad5 genes in all observed species, indicating that only the three NADH dehydrogenase genes undergo the transsplicing process in most of higher plant mt genomes.
Using the PREP-mt program, we predicted 479 RNAediting sites in the 39 protein-coding genes (including three multicopied genes) of the G. raimondii mt genome and 100% of these were C to U RNA-editing (Figure 4). Three genes (ccmFn, ccmB, and nad4) encoded most of RNA-editing sites (29), and three genes (rps7, atp8, and sdh3) encoded the fewest sites (2). Additionally, 34.45% (165 sites) of these sites occurred in the first base position of the codon, whereas 65.55% (314 sites) were in the second position and none were in the third position. Therefore, the amino acid would be changed due to nucleotide substitution in the codon, which may contribute to the diversity of start and stop codons of these protein-coding genes. Among the 479 amino acids, 112 (23.38%) were converted from Proline to Leucine and 106 (22.13%) were converted from Serine to Leucine. The other Table 8: Comparison of tRNA genes in seven higher plant mt genomes.  [54]. b Data from [53]. c Data from [52]. d Data from [51].
261 amino acids are distributed in other RNA-editing types, such as Ala to Val, His to Tyr, Leu to Phe, Pro to Phe, Pro to Ser, Gln to X, Arg to Cys, Arg to Trp, Ser to Phe, Thr to Ile, Thr to Met, and Arg to X (X = stop codon). RNA-editing is a posttranscriptional process, which has been shown to exist in mt and cp genomes of higher plants [24,25,[55][56][57]. RNA-editing may lead to critical alternations in transcription on the basis of the change from C to U residues, and RNA-editing of C to U substitution has been identified in most of angiosperms [1]. This process may generate an initiation or termination codon but more often generates an internal codon with strong functional relevance [6]. The study of RNA-editing sites may be also beneficial toward understanding mt gene expression in plants. As shown in Table S3, rps10, nad1, and nad4L use ACG as the start codon, which is altered to the normal AUG by this RNAediting process. RNA-editing of C to U substitution (ACG to AUG) has been reported in many protein-coding genes of different plant mt genomes, such as atp6, cox1, rps10, and nad4L of C. papaya, atp6, nad1, and nad4L of B. vulgaris, cox1, rps4, and rps10 of N. tabacum, and nad1 and nad4L of O. sativa and S. bicolor. Genes with the unknown start codons (mttB and rpl16 of the G. raimondii mt genome) were also reported in matR, mttB, and rpl16 of B. napus; matR, mttB (orfX), and ccb203 of A. thaliana; mttB of C. papaya and B. vulgaris; and matR of O. sativa and S. bicolor. Moreover, rps10 uses CGA as its stop codon, which is also altered to UGA by this process. RNA-editing of C to U substitution (CGA to UGA) also has been reported in three other species of Gossypium (G. barbadense, G. harknessii, and G. hirsutum).
Previous studies showed that closely related taxa generally shared more RNA-editing sites [25]. For example, 463 and 444 RNA-editing sites in 37 genes were found in C. lanatus and C. pepo mt genomes, and 394 sites are shared between    45) was less than the number in mt genomes due to the smaller genome size [24].

Synonymous ( s ) and Nonsynonymous ( a ) Substitution
Rates in the G. raimondii mt Genome. All of the 36 proteincoding genes in G. raimondii mt genome were used to analyze synonymous and nonsynonymous substitution rates against C. papaya and P. tremula. As shown in Figure 5, most of genes have a / s ratios between C. papaya and G. raimondii of less than 1,  Figure 6 and Figure S2, both the ML tree and the NJ tree   atp4  atp6  atp8  atp9  ccmB  ccmC  ccmFc  ccmFn  cob  cox1  cox2  cox3  matR  mttB  nad1  nad2  nad3  nad4  nad4L  nad5  nad6  nad7  nad9  rpl16  rpl10  rpl5  rpl2  rps10  rps4  rps7  rps3  rps12  rps14  sdh3   strongly supported separation of the one bryophyte and two gymnosperms from the 27 angiosperms. In addition, the trees also supported the separation of 22 eudicots from 5 monocots and the separation of 6 asterids from 16 rosids. As illustrated in Figure 6, two Gossypium species were classified into a clade (Malvales) with a high bootstrap value (100%) and the clade is evolutionarily close to the Brassicales (C. papaya, B. maritima, A. thaliana, and B. napus) with a high bootstrap value of 95%.
The loss of protein-coding and tRNA genes occurred frequently during the evolution of higher plant mt genomes. Pseudogenes are also regarded as lost genes. The results of phylogenetic analyses provided a backdrop for further analysis of gene loss of mt genomes during the evolution of higher plants. The protein-coding genes of the G. raimondii mt genome are compared with those of twenty-two other higher plants (shown in Table S2). The mt ribosomal proteins of higher plants are encoded partly by mt native genes and partly by nuclear genes [19,22,58]. During the course of evolution from bryophytes to the common ancestor of angiosperms, rpl6 and rps8 genes were lost. Gymnosperms C. taitungensis and G. biloba inherited the most ribosomal proteins (rpl2, rpl5, rpl16, rps1, rps2, rps3, rps4, rps7, rps10,  rps11, rps12, rps13, rps14, and rps19) from the mt genome progenitor of seed plants [59]. In contrast, A. reptans, belonging to the Lamiales family, encodes only five ribosomal proteins (rpl2, rps3, rps4, rps12, and rps13). The ribosomal protein rps11 gene exists in gymnosperms but was lost in both dicots and monocots, perhaps during the divergence of angiosperms and gymnosperms. After the divergence of  Figure 6: Maximum likelihood tree based on 23 conserved protein-coding genes of 30 representative higher plant mt genomes. Numbers on each node are bootstrap support values. Marchantia polymorpha was used as outgroup. The yellow, red, blue, green, and baby blue circles represent the asterid, rosid, monocot, Gymnospermae, and Bryophyta classes, respectively. The black circle indicated G. raimondii, belonging to rosids. monocotyledonous and dicotyledonous plants, four genes (sdh3, sdh4, rps10, and rps14) were lost in monocots, while rps2 was lost in dicots. The two corresponding mt genes (rps2 and rps11), missing in dicots, probably have been transferred to nuclear genomes. That means that the loss of proteincoding genes may be associated with functional transfer to nuclear genomes [60,61]. Compared with other dicots, Fabales (G. max and M. truncatula) have lost rpl2, rpl10, rps7, rps13, rps19, sdh3, and sdh4 genes. Additionally, Brassicales (A. thaliana and B. napus) have lost rpl10, rps1, rps10, rps13, rps14, rps19, sdh3, and sdh4 genes. C. lanatus, belonging to Cucurbitales, has conserved the maximum number of mt genes in angiosperms, having lost only rps2 and rps11, which were also lost in other dicots. The process of protein synthesis requires 21 kinds of tRNA genes in mt genomes, including trnA, trnC, trnD, trnE, trnF, trnG, trnH, trnI, trnK, trnL, trnM, trnN, trnP, trnQ, trnR, trnS, trnT, trnV, trnW, trnY, and trnfM. Table 8 compares the tRNA genes of the G. raimondii mt genomes with those of A. thaliana, N. tabacum, O. sativa, T. aestivum, C. taitungensis, and M. polymorpha. Three tRNA genes (trnI-GAU, trnT-UGU, and trnV-GAC) were lost in all observed plants, indicating that the loss probably occurred before the formation of bryophytes. During evolution from bryophytes (M. polymorpha) to gymnosperms (C. taitungensis), six tRNA genes (trnA-UGC, trnG-UCC, trnL-UAA, trnR-ACG, trnR-UCG, and trnT-GGU) were lost. Four tRNA genes (trnL-CAA, trnL-UAG, trnR-UCU, and trnV-UAC) were lost during the evolution from gymnosperms to angiosperms. After the divergence of monocots and dicots, trnG-GCC was lost in monocots.
Based on comparison of protein-coding and tRNA genes with the representative gymnosperm C. taitungensis, the G. raimondii mt genome has lost five protein-coding genes (rps1, rps2, rps11, rps13, and rps19) and six tRNA genes (trnG-GCC, trnL-CAA, trnL-UAG, trnfM-CAU, trnR-UCU, and trnV-UAC). Similarly, the 11 genes also are absent in other cottons (G. barbadense, G. harknessii, and G. hirsutum), suggesting that these genes may not be essential for cotton or their functions can be served by some other genes in cotton.

Conclusion
In summary, the G. raimondii mt genome shares many common characteristics with those of other higher plants.
The GC content of protein-coding genes was less than other non-coding regions in the G. raimondii mt genome and this distinction may be associated with gene expression and gene mutation. Four large repeats identified in the G. raimondii mt genome provide important information for analyzing intramolecular recombination of Gossypium mt genomes. Thirteen gene clusters were found in this mt genome, which can be beneficial for predicting coexpression or potential interrelated function of clustered genes. The complete mt genome of G. raimondii verified a prediction of the enrichment of mt-like nuclear genes on chromosome 1 [37]. Here, we not only identified all of these mt-like nuclear genes clearly but also found that the total mt genome might have been transferred to Chr1 at a very recent time. The C to U conversions of G. raimondii mt protein-coding genes may generate initiation, termination, or internal codons with completely unpredictable function. a / s analysis presented here shows that purifying selection influences complex III, complex IV, and complex V genes in the evolution of higher plant mt genomes. Phylogenetic analysis of thirty higher plants shows that two Gossypium species (G. raimondii and G. barbadense) were classified into a clade (Malvales) and are evolutionarily close to the Brassicales (C. papaya, B. maritima, A. thaliana, and B. napus). Overall, the results in the diploid cotton (G. raimondii) mt genome provide a crucial foundation for evolutionary analysis, molecular biology, and CMS of higher plant mt genomes, especially for cotton.