Sequencing, assembly and genome organization
The Illumina dataset contained 59.76 million paired reads and 17.93 Gb of sequence. The Q20 and Q30 proportions were 97.21% and 92.20%, respectively and the overall GC content was 35.80%. Long-read sequencing yielded 2.63 million reads and 15.85 Gb, with an N50 read length of 17.69 kb. The mitochondrial assembly was presented as a single 359,147-bp circularized reference sequence with a GC content of 44.67% (Table 1; Fig 1 and 2). The Nanopore read-length distribution was strongly right-skewed, with most reads concentrated in shorter size classes and a tail extending beyond 90 kb. This dataset supplied long-range information for assembly, but physical genome configurations require direct junction-spanning validation.
The annotation comprised 52 intact genes (31 protein-coding, 18 tRNA and three rRNA genes) and two pseudogenes, representing 54 annotated loci in total. The protein-coding complement included five ATP synthase genes, four cytochrome c biogenesis genes, three cytochrome c oxidase genes, nine NADH dehydrogenase genes, six small-subunit ribosomal protein genes,
rpl5,
cob,
matR and
mttB. The
trnM-CAU gene occurred in four copies. Introns were annotated in
ccmFc,
nad1,
nad2,
nad4,
nad5,
nad7,
rps1,
rps3 and
trnL-UAG. Two ribosomal protein genes,
rpl16 and
sdh4, were annotated as pseudogenes. The minimum-102-bp ORF scan returned 1,836 candidates, which were excluded from the curated gene inventory because they lacked independent expression or conservation evidence.
Most protein-coding genes used canonical ATG start codons. The
nad1 and
nad4L annotations contained ACG start codons predicted to be converted to AUG through RNA editing. Several genes terminated with TGA, while
ccmFc contained a genomic CGA codon predicted to become TGA after editing.
The 359.15 kb
O. cyri mitochondrial genome assembly falls within the broad size range reported for Fabaceae, whose mitochondrial genomes show ancient and recent shifts in size and content
(Choi et al., 2019). Its core coding complement is consistent with other legume mitogenomes, whereas pseudogenization of
rpl16 and
sdh4 and the copy number of
trnM may reflect lineage-specific gene loss, transfer and duplication
(Choi et al., 2019; Gualberto and Newton, 2017). The circularized sequence is a convenient reference representation rather than proof of one genome-sized molecule: plant mitochondrial DNA frequently occurs as a dynamic mixture of recombination-derived configurations
(Kozik et al., 2019). Multiple repeat-mediated configurations have also been validated in the closely compared legume
C. spinosa (Zhou et al., 2023).
The hybrid sequencing strategy combined Nanopore long-range information with high-Q30 Illumina reads. Reference-guided read enrichment may under-represent highly divergent or lineage-specific mitochondrial sequences. Current guidance on land-plant mitogenome assembly emphasizes hybrid short- and long-read evidence together with explicit evaluation of alternative configurations (
Storchová and Krüger, 2024). Detailed assembly-graph evidence, coverage summaries and junction-spanning read counts are not presented here. Consequently, the circularized sequence should be interpreted as a reference representation; the physical configurations present in vivo remain to be resolved.
Repeat sequences and structural variation
The genome contained 252 repeat elements, including 99 SSRs, four tandem repeats and 149 dispersed repeats (Table 2). Among SSRs, tetranucleotide motifs were the most frequent class (30), followed by mononucleotide (27), dinucleotide (19), trinucleotide (17), pentanucleotide (4) and hexanucleotide motifs (2). The four tandem repeats had repeat units of 6-36 bp and approximately 1.9-4.2 copies. Dispersed-repeat alignments ranged from 29 to 9,778 bp; the longest was a 9,778-bp direct repeat with 100% identity. Repeat features occurred throughout the circularized reference and several dense dispersed-repeat link bundles identify regions that could participate in repeat-mediated recombination (Fig 3).
The abundance of dispersed repeats provides a substrate for homologous recombination and structural isoforms, a major source of plant mitochondrial genome dynamism
(Alverson et al., 2011; Gualberto and Newton, 2017;
Kozik et al., 2019). The 99 SSRs also define candidate mitochondrial marker loci, although organellar specificity must be established against the nuclear and chloroplast genomes before these loci are used in marker-assisted genetic improvement programmes.
The repeat map shows connections distributed across the reference sequence, including several dense bundles. These regions identify candidates for targeted examination of alternative configurations. Repeat occurrence alone does not demonstrate active recombination or establish the relative abundance of physical isoforms.
Codon usage
Codon usage was biased toward several A- or U-ending codons. The highest RSCU values among synonymous sense codons included UUA for leucine (1.5494), GCU for alanine (1.5380), UAU for tyrosine (1.5369), CAU for histidine (1.4825) and CAA for glutamine (1.4786). Codon-level values grouped by amino-acid family are shown in Fig 4.
Codon usage in
O. cyri favored several A- and U-ending synonymous codons, consistent with an influence of nucleotide composition on mitochondrial coding sequences. RSCU describes relative use within synonymous families; it does not by itself distinguish mutation pressure from selection or demonstrate translational optimization.
The RSCU profile shows broad within-family usage differences for synonymous codons of arginine, leucine and serine, whereas the one-codon families Met and Trp carry no information about synonymous preference. Because RNA editing can alter codons after transcription, these values describe genomic codon usage rather than edited transcripts.
Predicted RNA editing
A total of 474 RNA-editing sites were predicted in 30 protein-coding genes. The largest numbers occurred in
nad4 (41 sites),
ccmFn (35),
nad7 (30),
ccmB and
nad2 (29 each) and
ccmC (28). Edits predicted to convert hydrophilic amino acids to hydrophobic amino acids formed the largest functional class (227 sites; 47.89%). Hydrophobic-to-hydrophobic changes accounted for 150 sites (31.65%), hydrophilic-to-hydrophilic changes for 62 (13.08%) and hydrophobic-to-hydrophilic changes for 34 (7.17%). One predicted edit converted CGA to the stop codon UGA. The distribution among genes is shown in Fig 5.
The 474 predicted editing sites are consistent with extensive post-transcriptional modification of mitochondrial transcripts in flowering plants
(Takenaka et al., 2013). NADH dehydrogenase and cytochrome c biogenesis genes carried particularly high editing loads and many changes increased amino-acid hydrophobicity, consistent with restoration of conserved residues in membrane-associated respiratory proteins. The predicted stop codon in
ccmFc and ACG-to-AUG start-codon changes in
nad1 and
nad4L are functionally notable; without matched RNA-sequencing or cDNA data, they remain computational predictions.
Interspecific nucleotide variability
Interspecific nucleotide variability differed among the 34 evaluated loci (mean Pi = 0.01706). The
rps7 locus had the highest Pi value (0.17157), followed by
nad9 (0.04445),
atp6 (0.03532),
atp1 (0.03434) and
atp8 (0.02795) (Table 3; Fig 6). In contrast,
atp9 was the most conserved locus (Pi = 0.00198), followed by
nad7,
nad4,
mttB and
nad4L. These estimates describe divergence among sampled species and should not be interpreted as within-population diversity.
The interspecific Pi analysis identified
rps7,
nad9,
atp6,
atp1 and
atp8 as the most variable loci in the sampled taxa. The exceptional
rps7 value may be influenced by its short alignment and by gene absence or divergence in some plant lineages. The more moderately variable
nad9,
atp6,
atp1 and
atp8 loci are candidates for comparative marker development, but their utility for population-level studies requires validation in multiple
O. cyri individuals.
Whole-genome similarity and rearrangement
Whole-genome alignments showed that 35.48-51.63% of the
O. cyri mitochondrial reference was represented in homologous regions with the eight comparison genomes. The greatest reference coverage was observed with
G. glabra (51.63%) and
G. uralensis (51.62%), followed by
C. spinosa (49.99%) and
V. faba (47.89%). The lowest coverage occurred with
P. vulgaris (35.48%), illustrating extensive divergence in non-coding and structurally variable regions despite conservation of core mitochondrial genes. The pairwise dot plots contained many short, discontinuous matches in both orientations rather than a single genome-length diagonal (Fig 7), consistent with rearrangement and lineage-specific intergenic turnover.
The fragmented forward- and reverse-orientation matches in the collinearity plots provide structural context for the 35.48-51.63% aligned proportions. Conserved mitochondrial blocks have been reordered against a background of rapidly changing intergenic DNA, allowing core coding genes to retain a coherent phylogenetic signal despite poor whole-genome collinearity. The dot plots describe the resulting structural differences but do not resolve a unique rearrangement pathway.
Putative plastid-derived mitochondrial sequences
The
O. cyri chloroplast genome used for comparison was 121,932 bp. Homologous chloroplast regions totaled 5,208 bp (4.27% of the chloroplast genome), corresponding to 5,173 bp or 1.44% of the mitochondrial assembly (Fig 8). Five tRNA genes-
trnW-CCA,
trnD-GUC,
trnN-GUU,
trnH-GUG and
trnM-CAU-were completely located within homologous chloroplast regions. Partial chloroplast-gene matches included
psaA,
psaB,
trnI-GAU,
trnA-UGC,
rrn16,
petG,
psaJ and
rrn23. On the mitochondrial side, the corresponding homologous regions included complete copies of the five tRNA genes and partial matches involving
rrn18 and
rrn26. These similarities identify putative plastid-derived regions but do not by themselves establish transfer direction or functionality.
Approximately 1.44% of the
O. cyri mitochondrial assembly was homologous to chloroplast DNA. Complete plastid-like
trnW,
trnD,
trnN,
trnH and
trnM genes are consistent with the persistence of plastid-derived tRNA genes in seed-plant mitochondria (
Gandini and Sanchez-Puerta, 2017;
Stern and Palmer, 1984;
Wang et al., 2007). Partial matches to
psaA and
psaB are more consistent with non-functional remnants than intact mitochondrial genes. At candidate transfer boundaries, reliable organellar assignment requires long reads anchored uniquely in mitochondrial flanking sequence; sequence similarity alone does not identify transfer direction.
Phylogenetic relationships
The maximum-likelihood analysis recovered the major relationships represented by the sampled Fabaceae mitogenomes (Fig 9).
O. cyri was placed as sister to
C. spinosa with 77% bootstrap support and this pair was associated with
Phyllolobium and
Oxytropis lineages within a strongly supported broader clade. The 77% support identifies the
O. cyri-
C. spinosa relationship as moderately supported within the current taxon sample.
The placement of
O. cyri with
C. spinosa is supported by the sampled mitochondrial coding sequences. The 77% bootstrap value and limited representation of close relatives define this relationship as provisional rather than a definitive genus-level taxonomic result.
Exploratory selective-constraint analysis
The Ka/Ks dataset comprised 908 pairwise gene estimates. The mean ratio was 0.544 and 795 estimates (87.6%) did not exceed one, consistent with predominantly constrained evolution in this descriptive dataset. A total of 113 estimates exceeded one, most frequently in
nad9 (29 comparisons),
ccmB (14),
ccmFn and
ccmFc (13 each),
rps1 (10) and
atp6 (7). The largest value was 4.23084 for
cox2. Values above one were treated as candidate signals because estimates can be inflated by low Ks values, short sequences or local alignment uncertainty (Fig 10).
Most pairwise Ka/Ks ratios did not exceed one, a pattern consistent with broad functional constraint on mitochondrial protein-coding genes. Ratios above one in
nad9, ccm genes,
rps1 and
atp6 identify candidate deviations from this background, but pairwise ratios alone do not establish adaptive evolution and remain sensitive to small Ks values, short alignments, non-independence among comparisons and alignment uncertainty.