Plant materials, plastome dataset, assembly, and annotation

The present study investigated plastome evolution within the subfamily Aurantioideae of the family Rutaceae. A total of 103 plastome accessions were analyzed, comprising 88 Aurantioideae accessions and 15 outgroup taxa. The dataset included 29 accessions for which sequencing data had been generated in our previous studies [3,9], 15 accessions represented by publicly available raw sequencing datasets, and 59 accessions represented by publicly available complete plastome sequences. Thus, no new sequencing data were generated specifically for the present study. A complete list of the taxa, data sources, and corresponding sequence accession numbers is provided in S1 Table.

Different subsets of this dataset were used for specific analyses. Pseudogenization patterns were examined in 45 Aurantioideae taxa together with 14 additional Rutaceae accessions for comparative purposes. Plastome-based classification within Citrus was conducted using 56 Citrus accessions. Divergence-time estimation focused primarily on Aurantioideae and included four additional Rutaceae accessions, with Ailanthus altissima designated as the outgroup.

For the 29 accessions generated in our previous studies [3,9], DNA originated from the same extraction batches used in those studies. DNA quality was evaluated by electrophoresis on a 1% agarose gel, and DNA concentration was determined using a DNA BR Assay Kit (Thermo Fisher Scientific, USA). Sequencing libraries were prepared from total genomic DNA by Novogene (Singapore) using the NEBNext Ultra DNA Library Prep Kit for Illumina (New England Biolabs, USA) and sequenced on an Illumina NovaSeq 6000 platform to generate 150-bp paired-end reads. In addition, raw sequencing data for 15 accessions were obtained from public databases. Accession numbers and data sources for all raw sequencing datasets are provided in S1 Table. Raw reads used for plastome assembly were preprocessed using fastp v0.23.2 with default settings [18]. Summary statistics for the filtered reads, including the number of read pairs, total filtered bases, average read length, Q20 and Q30 percentages, and GC content, were calculated using a custom Python script (S2 Table).

De novo plastome assembly and annotation were performed following the procedures described in our previous study [9]. Plastomes were assembled from the 44 raw-read datasets using GetOrganelle v1.7.5 [19] with the parameters -R 15, -k 21,45,65,85,105, and -F embplant_pt. Assembly quality was evaluated by mapping the quality-filtered reads back to the corresponding assembled plastomes using Bowtie2 v2.4.4 [20]. The resulting alignments were processed using SAMtools v1.14 [21] and visually inspected in Integrative Genomics Viewer (IGV) to assess read coverage and assembly consistency. Qualimap v2.1.1 [22] was used to obtain quantitative mapping statistics, including the percentage of reads mapped to the assembled plastome, mean sequencing coverage, and mean mapping quality (S2 Table).

Plastome annotation was performed using GeSeq v2.03 [23], with protein-coding sequences predicted using Chloë v0.1.0, tRNA genes identified using tRNAscan-SE v2.0.7 [24], and rRNA genes identified using HMMER. Following automated annotation, all plastome annotations were manually inspected and corrected where necessary to ensure annotation accuracy and consistency. Final circular plastome maps were generated using OGDRAW v1.3.1 [25].

Accession numbers for the source raw sequencing reads and previously deposited plastome sequences, together with the corresponding data sources and information on plant materials processed by our group, are provided in S1 Table. The 15 plastomes assembled de novo from additional public raw-read datasets and the re-annotated plastome files derived from 59 public complete plastome sequences, manually curated where necessary, are available in the Supplementary Dataset deposited in Zenodo (https://doi.org/10.5281/zenodo.22743534).

Comparative plastome sequence and LSC–IR boundary analyses

To assess sequence conservation and divergence among representative Aurantioideae plastomes, whole-plastome sequence comparisons were performed using mVISTA [26] with the Shuffle-LAGAN alignment mode [27], with the Citrus sinensis plastome used as the reference. Representative taxa included Aegle marmelos, Atalantia bilocularis, A. ceylanica, A. roxburghiana, Citropsis gabunensis, Clausena anisata, Citrus glauca, Limonia acidissima, Glycosmis citrifolia, Naringi crenulata, Murraya caloxylon, Micromelum minutum, Murraya koenigii, Murraya paniculata, Paramignya lobata, Swinglea glutinosa, and Triphasia trifolia. The resulting alignments were examined to evaluate patterns of sequence conservation and divergence across the plastomes.

To further investigate structural variation at plastome junctions, boundary regions were manually examined and compared among representative Aurantioideae plastomes, with particular attention to the junctions between the LSC and the IRs. The analysis focused especially on the LSC–IR boundary regions containing rpl22 and rps19. Structural differences, including gene fragmentation, pseudogenization, and local gene loss, were identified manually by comparing gene order, gene orientation, and junction positions among the annotated plastomes.

Whole-plastome alignment and phylogenetic analyses

Whole-plastome alignment and phylogenetic analyses were performed as in the previous study [9]. Complete plastome sequences were aligned using HomBlocks (v1.0) [28] with default settings. The best-fit evolutionary model was identified using ModelTest-NG (v0.1.7) [29]. Maximum-likelihood (ML) trees were constructed using RAxML (v8.2.12) [30] with 1,000 bootstrap replicates. Bayesian inference (BI) trees were generated using BEAST (v1.10.4) [31] with Markov chain Monte Carlo sampling for 10,000,000 generations, sampling every 10,000 generations and discarding the first 10% as burn-in. ML and BI trees were visualized using Dendroscope (v3.8.8) and FigTree (v1.4.4), respectively. In the present study, this framework was applied to the Citrus plastome dataset and the Aurantioideae-centered datasets used for comparative analyses.

Identification and phylogenetic analysis of pseudogenes

Protein sequences were extracted from annotation files using genbank_to (v0.42) [32]. Orthologous protein groups were identified using OrthoFinder (v2.5.4) [33], and amino acid sequences were aligned using MAFFT (v7.515) [34]. The resulting alignments were inspected in MEGA11 (v11.0.13) [35] for comparative analysis. Pseudogenes were identified on the basis of frameshift-causing insertions or deletions, premature stop codons, or loss of recognizable start codons, with comparisons made against intact homologous sequences. This strategy was applied to analyses of rpl22, ycf68, orf56, cemA, ycf15, rps4, and ycf1. Where raw sequencing reads were available, candidate pseudogenization events were further validated by visual inspection of read alignments in IGV. Read coverage across the affected regions was examined to confirm that the sequence variants responsible for pseudogenization were supported by the underlying raw reads rather than resulting from assembly or annotation errors.

To assess whether similar pseudogenization states corresponded to particular phylogenetic branches, pseudogenization patterns were mapped onto an Aurantioideae-centered plastome phylogeny that included 14 additional Rutaceae accessions for comparison. We focused on the distribution of repeated pseudogenization involving rpl22, ycf68, and orf56, and separately examined lineage-specific events involving cemA, ycf15, rps4, and ycf1. The phylogenetic interpretation of these events was based on the whole-plastome trees described above.

Divergence-time estimation

Divergence times were estimated using MCMCtree in PAML v4.10.10 [36] with the approximate-likelihood method [37]. The ML topology inferred from the plastome alignment using IQ-TREE v2.3.6 [38] was fixed for molecular dating and rooted with Ailanthus altissima as the outgroup. The analysis included 47 taxa and a 77,661-bp nucleotide alignment. Branch lengths, gradient vectors, and the Hessian matrix required for the approximate-likelihood calculation were estimated using BASEML under the HKY85 + Γ model with five discrete gamma rate categories and subsequently used in MCMCtree [37,39,40].

Four temporal constraints were applied following published fossil assessments and divergence estimates. The root separating A. altissima from the sampled Rutaceae was assigned a secondary soft bound of 100–117 Ma [B (1.00, 1.17)] [41]. Fossil-based soft minimum constraints were assigned to stem Zanthoxylum at 47.8 Ma [L (0.478)] [41], the stem Clausena + Glycosmis lineage at 27.23 Ma [L (0.2723)] [42], and stem Citrus at 11.6 Ma [L (0.116)] [43]. A prior-only analysis (usedata = 0) was first performed to evaluate the effective time priors. Posterior dating was then conducted using the approximate likelihood (usedata = 2) under an independent-rates relaxed clock (clock = 2) with a birth–death-sampling prior (BDparas = 1 1 0.1 m), rgene_gamma = 2 20 1, and sigma2_gamma = 1 10 1. Two independent MCMC chains were run, each with 200,000 burn-in iterations followed by 20,000 retained samples collected every 100 iterations. Convergence was assessed using rank-normalized R-hat and bulk and tail effective sample sizes [44]. Calibration sensitivity was assessed by replacing the 47.8-Ma minimum-age constraint for Zanthoxylum with a 66-Ma minimum-age constraint based on the fossil Rutaspermum biornatum [41], while keeping all other analytical settings unchanged.

Use of generative AI

ChatGPT (https://chat.openai.com/) was used to rewrite the manuscript into more appropriate English and proofread the English text. Subsequently, the authors verified the correctness of the revisions.

Aurantioideae plastomes were highly conserved with localized genomic and sequence variation

A total of 103 plastome accessions were analyzed, comprising 88 Aurantioideae accessions and 15 outgroup taxa (S3 Table). Among the Aurantioideae accessions, plastome size ranged from 158,779 bp in Naringi crenulata to 161,172 bp in Clausena excavata. Overall GC content varied by only 0.39 percentage points, ranging from 38.28% in C. excavata to 38.67% in Clausena harmandiana and Clausena lansium. Gene content was also highly uniform across the subfamily. All 88 Aurantioideae plastomes contained 30 unique tRNA genes and four unique rRNA genes. Most accessions (85 of 88) contained 83 unique protein-coding genes, corresponding to a total of 117 unique genes per plastome (S3 Table).

Whole-plastome sequence conservation was further evaluated among representative Aurantioideae taxa using mVISTA, with Citrus sinensis as the reference plastome (S1 Fig). The comparison included representatives from multiple genera across the subfamily, including Aegle, Atalantia, Citropsis, Clausena, Citrus, Glycosmis, Micromelum, Murraya, Naringi, Paramignya, Swinglea, and Triphasia. Coding regions were generally more conserved than non-coding regions, whereas greater sequence divergence was observed mainly in intergenic and other non-coding regions (S1 Fig). The inverted-repeat regions, including the regions containing the rRNA genes, were highly conserved across the sampled taxa. In contrast, localized reductions in sequence similarity were evident at several positions within the large single-copy and small single-copy regions. Despite these localized differences, the overall alignment profiles were highly similar among the representative plastomes (S1 Fig).

Structural variation at the LSC–IR boundaries and repeated putative pseudogenization across Aurantioideae plastomes

Despite the overall sequence conservation observed among Aurantioideae plastomes, structural variation was detected at the LSC–IR boundaries (Fig 1). Comparative analysis revealed pronounced variation in the regions containing rpl22 and rps19, particularly at the LSC–IRB and IRA–LSC junctions (Fig 1). These differences included gene fragmentation, putative pseudogenization, and local gene loss, indicating substantial structural variation at the LSC–IR boundaries among the examined taxa.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Structural variation at plastome boundary regions in representative Aurantioideae species.

The structures of the LSC–IRB, IRB–SSC, SSC–IRA, and IRA–LSC junctions are compared among representative Aurantioideae plastomes. Gene positions and orientations near the boundaries are shown together with the lengths of adjacent regions. Fragmentation, pseudogenization, and local loss involving rpl22 and rps19 are evident in several lineages. These comparisons illustrate that plastome boundary regions are structurally variable within Aurantioideae, especially around the LSC–IR boundaries. Asterisks indicate sequences assembled de novo.

https://doi.org/10.1371/journal.pone.0349882.g001

The rpl22 region showed considerable variation among the examined species. Two rpl22-associated sequences were identified near the LSC–IR junctions (Fig 1). The sequence located at the LSC–IRB boundary generally retained the expected gene length with identifiable start and stop codons, whereas the corresponding sequence at the IRA–LSC boundary was truncated or fragmented relative to the intact gene (Fig 1). A representative configuration is shown for Citrus japonica in Fig 1. In several lineages, the LSC–IRB-associated rpl22 copy was inferred to be pseudogenized, primarily through the occurrence of internal premature stop codons (Figs 1 and 2). Notably, rpl22 was absent from the corresponding boundary region in Glycosmis citrifolia, whereas Glycosmis pentaphylla retained a putatively pseudogenized copy at this position (Figs 1 and 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Distribution of rpl22, ycf68, and orf56 pseudogenization across Aurantioideae and additional Rutaceae accessions.

The phylogenetic tree shows the distribution of pseudogenization states of rpl22, ycf68, and orf56 across Aurantioideae, with 14 additional Rutaceae accessions included for comparison. Intact reference sequences were defined from Citrus australasica (rpl22), Citrus aurantium (ycf68), and Citrus sinensis (orf56), and all other sequences were evaluated relative to these references. Colored symbols indicate pseudogenization states or gene loss in each lineage. Representative mutation types are summarized on the right: premature stop codons in rpl22 (green symbols with internal asterisks), C insertion–induced frameshifts in orf56 (yellow dashed symbols), and multiple patterns in ycf68, including poly-T deletions (dark blue symbols with internal T counts), associated frameshifts (light blue dashed symbols with internal T counts), and A insertions (blue symbols). Gene loss is indicated by gray lines. These results demonstrate that similar pseudogenization states occurred in multiple phylogenetically separated lineages. Asterisks indicate sequences assembled de novo.

https://doi.org/10.1371/journal.pone.0349882.g002

Variation was also observed in rps19 at the plastome boundaries. In Afraegle paniculata, the rps19 sequence located at the IRA–LSC boundary was inferred to be pseudogenized, whereas the corresponding gene remained intact in Aegle marmelos (Fig 1). In addition, rps19 was absent from the LSC–IRB boundary of Glycosmis citrifolia (Fig 1).

Beyond these boundary-associated changes, putative plastid gene pseudogenization was assessed across Aurantioideae together with 14 additional Rutaceae accessions (Fig 2). The infA gene was inferred to be pseudogenized in all examined taxa, whereas putative pseudogenization of ycf68 and orf56 occurred in multiple lineages across the phylogeny (Fig 2). Sequence comparisons showed that orf56 pseudogenization was associated with the insertion of a cytosine nucleotide, resulting in a frameshift mutation (Fig 2). Similarly, ycf68 pseudogenization was associated with either the deletion of thymine residues within a poly-T region or the insertion of adenine residues, both of which resulted in frameshift mutations (Fig 2). Representative IGV read alignments showed that the mutations associated with putative pseudogenization of rpl22, orf56 and ycf68 were consistently supported by mapped sequencing reads (S2 Fig for rpl22 and orf56; S3 Fig for ycf68).

Additional lineage-specific putative pseudogenization events were detected in cemA, ycf15, rps4, and ycf1

In addition to the more broadly distributed pseudogenization patterns described above, several lineage-specific events were identified at the sequence level (S4–S6 Figs). In Afraegle paniculata, cemA was inferred to be pseudogenized relative to the intact sequence in Aegle marmelos, with a 5-bp insertion of adenine residues disrupting the coding sequence (S4 and S5 Figs). In Aegle marmelos, ycf15 lacked a recognizable start codon compared with the intact sequence in Citrus sinensis (S4 and S5 Figs). Putative pseudogenization of rps4 was also detected in C. sinensis (S6 Fig).

A further lineage-specific event was detected in Atalantia kwangtungensis, in which ycf1 was inferred to be pseudogenized relative to the intact gene in Atalantia buxifolia (S6 Fig). This disruption was associated with deletion of a thymine nucleotide, resulting in disruption of the reading frame.

Whole-plastome phylogeny resolved detailed relationships within Citrus

ML analysis was conducted using complete plastome sequences from 56 Citrus accessions with Atalantia buxifolia designated as the outgroup (Fig 3). The final alignment comprised 88,414 nucleotide positions, including 1,672 parsimony-informative sites. The resulting topology showed high overall resolution, with most major and several species-level relationships receiving strong bootstrap support. Eight major plastome-defined groups were recovered: the Pummelo, Papeda, Ichang Papeda/Yuzu, Mandarin, Poncirus, Fortunella, Citron, and Australian clusters (Fig 3).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. ML phylogenetic tree of Citrus based on complete plastome sequences.

The tree was constructed using complete plastome sequences from 56 Citrus accessions with Atalantia buxifolia used as the outgroup. For this analysis, the optimal evolutionary model (GTRGAMMAIX) was implemented in RAxML. The alignment used for tree construction contained 88,414 nucleotide positions, including 1,672 parsimony-informative sites. Bootstrap values are shown at the nodes. Superscript numbers distinguish different accessions assigned to the same species name. The eight major plastome clusters recognized in this study are indicated by different colors. The scale bar represents genetic divergence in substitutions per site. Sequences assembled de novo are denoted by asterisks.

https://doi.org/10.1371/journal.pone.0349882.g003

The Pummelo Cluster comprised the largest assemblage of accessions and included C. maxima, C. sinensis, C. aurantium, C. limon, C. × paradisi, and several other cultivated taxa (Fig 3). Notably, the two wild species C. hongheensis and C. latipes were also placed within the Pummelo Cluster and formed a strongly supported sister pair (BS = 100) (Fig 3). The sampled C. maxima accessions occurred within the same broader plastome lineage as grapefruit, sweet orange, sour orange, and lemon accessions. The Papeda Cluster comprised C. micrantha and two C. aurantiifolia accessions and received strong bootstrap support (BS = 100) (Fig 3). Citrus cavaleriei and C. junos formed a separate, strongly supported Ichang Papeda/Yuzu Cluster (Fig 3).

A distinct phylogenetic position was observed for C. mangshanensis, which occurred outside the defined Mandarin Cluster rather than grouping with the core mandarin accessions (Fig 3). The Mandarin Cluster comprised C. depressa, C. tachibana, C. reshni, C. limonia, C. tangerina, C. sunki, C. unshiu, C. reticulata, and C. clementina, with strong support for several internal relationships. Within this cluster, the two C. reticulata accessions grouped with C. clementina, whereas C. tangerina, C. sunki, and the C. unshiu accessions formed a separate sublineage (Fig 3).

The Poncirus Cluster contained two C. trifoliata accessions together with C. polytrifolia, whereas the Fortunella Cluster comprised C. japonica, C. hindsii, C. japonica var. margaritia, and C. madurensis (Fig 3). All sampled C. medica accessions were confined to a highly supported Citron Cluster, including Mac Veu Mountain Citron, Buddha’s Hand Citron, Humpang Citron, Corsica Citron, and C. medica (Fig 3).

The Australian Cluster comprised Clymenia polyandra, C. glauca, C. australis, and C. australasica (Fig 3). Within this lineage, C. polyandra was resolved as the earliest-diverging taxon, followed by C. glauca, whereas C. australis and the two C. australasica accessions occupied more derived positions (Fig 3). The two C. australasica accessions formed a strongly supported sister pair (BS = 100) (Fig 3).

BI based on the same plastome dataset produced a topology highly congruent with the ML analysis (S7 Fig), recovering the same eight major plastome clusters and principal interspecific relationships, with most nodes receiving high posterior probabilities.

Divergence-time analysis suggested two comparable Asia–Africa splits within Aurantioideae

To place plastome diversification in a temporal framework, we estimated divergence times using MCMCtree for Aurantioideae and four additional Rutaceae accessions, with Ailanthus altissima as the outgroup (Fig 4). Calibration sensitivity was assessed by replacing the Zanthoxylum-based minimum constraint with an alternative Rutaspermum-based constraint. This analysis produced a comparable overall temporal pattern and retained the relative ordering of the principal divergence events, indicating that these conclusions were not strongly affected by the alternative calibration tested. Subgroup names followed our previous study [9], in which Aurantioideae classifications were based on conserved nuclear genes and plastome data.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Divergence-time tree of Aurantioideae inferred from complete plastome sequences.

Divergence times were estimated using MCMCtree based on the fixed maximum-likelihood topology inferred with IQ-TREE. The analysis included 47 taxa, with Ailanthus altissima as the outgroup. Four temporal constraints were applied: a secondary root calibration of 100–117 Ma and fossil-based minimum-age constraints of 47.8 Ma for stem Zanthoxylum, 27.23 Ma for the stem Clausena + Glycosmis lineage, and 11.6 Ma for stem Citrus. Node bars represent 95% highest posterior density (HPD) intervals, and the scale indicates divergence time in millions of years ago (Ma). The two Asia–Africa lineage splits discussed in the text are indicated for Citropsis–Naringi and Aegle–Afraegle. Asterisks indicate plastomes assembled de novo from raw sequence data.

https://doi.org/10.1371/journal.pone.0349882.g004

Two biogeographically parallel lineage separations were identified within Aurantioideae (Fig 4). In the Thanaka group, African Citropsis and Asian Naringi diverged at an estimated 19.51 Ma (95% HPD: 11.85–28.40 Ma). In the Bael fruit group, African Afraegle and Asian Aegle diverged at 18.14 Ma (95% HPD: 9.43–28.08 Ma) (Fig 4). Their similar posterior mean estimates and broadly overlapping HPD intervals indicate comparable early Miocene divergence times.

Divergences involving Australian Citrus had younger posterior mean estimates, although their HPD intervals overlapped those of the two Asia–Africa separations (Fig 4). The lineage comprising Australian Citrus, Clymenia polyandra, and Citrus medica diverged from the remaining sampled true citrus fruit trees at 14.71 Ma (95% HPD: 9.47–20.69 Ma) (Fig 4). Within this lineage, C. medica separated from Clymenia polyandra and Australian Citrus at 11.81 Ma (95% HPD: 7.47–16.88 Ma), followed by the separation of Clymenia polyandra from Australian Citrus at 10.57 Ma (95% HPD: 6.58–15.22 Ma; Fig 4).