The mitochondrial genome of L. suprareticulata is a single circular molecule 364,645 bp in length, with a base composition of A (26.72%), T (27.08%), C (22.93%), and G (23.28%). The GC content of the entire mitochondrial genome is 46.20% (Figure 1). Both genome size and GC content are similar to those of Punica granatum mitochondrial genome, which belongs to the same family. A total of 62 genes were located and annotated in this mitochondrial genome, with the vast majority of the sequence consisting of intergenic regions, and gene sequences accounting for only a small proportion. Among the 62 functional genes, there are 3 rRNA genes (4.84%), 21 tRNA genes (33.87%), and 38 protein-coding genes (61.29%). These 38 protein-coding genes can be further classified into 10 functional categories. With the exception of the sdh4 gene (two copies), all other genes in the L.suprareticulata mitochondrial genome were present in single copy. Furthermore, 386 open reading frames (ORFs) were identified, comprising 191 on the forward strand and 195 on the reverse strand. Among these, 50 ORFs are longer than 300 bp. Repeat event analysis of the L.suprareticulata mitochondrial genome identified 60 repeat sequences longer than 100 bp, with the two long repeat over 500 bp, being 888 bp and 889 bp, two middle repeat, being 463 bp, respectively (Figure 2).

The simple sequence repeats (mtSSRs) in the mitochondrial genomes of five species within Lythraceae were identified and compared (Figure 3). The results showed that the mtSSRs were dominated by mono-nucleotide and tetra-nucleotide repeats. This characteristic was also observed in most of the analyzed mitochondrial genomes. Specifically, the number of mono-nucleotide repeats was 42, which was only lower than that of P. granatum but much higher than those of species in the genus Trapa and the congeneric Lagerstroemia indica. The number of tetra-nucleotide repeats was 36, which was comparable to that of P. granatum and significantly higher than those of Trapa species and L. indica. Its dinucleotide repeats were comparable to those of most species, while the abundances of penta-nucleotide and hexa-nucleotide repeats were extremely low, this result consisted with that in other species.

Using the default settings of Deepred-Mt software, a total of 480 potential RNA editing sites were identified in the mitochondrial genome of L. suprareticulata(Figure 4). All edits were of the C-to-U type and were distributed in 36 mitochondrial protein-coding genes. Among them, the nad4 gene contained the largest number of RNA editing sites (41), followed by the ccmB gene with 30 sites. Notably, no predicted RNA editing sites were detected in the mitochondrial rpl2 gene. Analysis of the codon usage bias of mitochondrial protein-coding genes revealed that only the relative synonymous codon usage (RSCU) values of AUG (methionine) and UGG (tryptophan) were 1, while the RSCU values of all other codons were greater than 1. Among these, GCU had the highest RSCU value (1.62), followed by CAU (1.53). In contrast, the RSCU values of GCG and CAG were both less than 0.5, indicating their relatively low codon usage frequencies.

The Ka/Ks values were calculated based on 15 PCGs shared by all selected plants. As shown (Figure 5), the Ka/Ks ratios of the ccmB and rps4 genes in four species of the Lythraceae family (Punica granatum, Trapa incisa, Trapa bicornis, and Lagerstroemia indica) exceeded 1. The Ka/Ks ratios of most other genes were relatively low (approximately 0.2 to 1.0), they might have undergone a purifying selection that the ccmB and rps4 genes may exhibit unique stress resistance when facing selective pressure. On the contrary, the average Ka/Ks values of atp8 and cox1 were the lowest among all genes (about 0.2), indicating a most obvious purifying selection for these genes. They have shown extremely high conservation during the evolution of PCGs in the plant mitochondrial genome.

Phylogenetic trees were constructed for the mitochondrial and chloroplast genomes of 38 species representing diverse taxonomic positions, with Aglaia odorata set as the outgroup (Figure 6). The resulting tree indicated a close evolutionary relationship between L. suprareticulata and L. indica, both belonging to the same genus. The phylogenies inferred from chloroplast and mitochondrial genomes were largely congruent and aligned with the established evolutionary relationships among angiosperm species. Nevertheless, subtle discrepancies in the internal relationships of Brassicaceae species were found between the trees inferred from mitochondrial and chloroplast data. These inconsistencies may stem from incomplete lineage sorting, where differences in evolutionary rate and sequence conservation across genomic regions lead to inferred phylogenies that do not fully reflect the actual evolutionary relationships.

To explore collinearity across species from different phylogenetic positions, we also performed a comparative analysis of mitochondrial genomes from nine species, including L. suprareticulata (Figure 7). The dataset included Trapa incisa, L. indica, Trapa bicornis, Punica granatum, and L. suprareticulata from Lythraceae; Rhodomyrtus tomentosa from Myrtaceae; Gossypium arboreum from Malvaceae; Vatica mangachapoi from Dipterocarpaceae; and Sapindus mukorossi from Sapindaceae. The analysis revealed that L. suprareticulata exhibited collinear segments with all the selected species, but these blocks were mostly short. Among these species, L. suprareticulata displayed the strongest collinearity with L. indica, a species from the same genus, with the most collinear blocks identified. Fewer collinear fragments were detected between L. suprareticulata and the more distantly related species. The decreasing number of collinear segments with increasing evolutionary distance indicates a high level of mitochondrial genome variation across species.

Mitochondrial genome structures exhibit substantial divergence across species, resulting in widespread gene rearrangement phenomena. We compared the gene arrangement pattern of the L. suprareticulata mitochondrial genome with four species from the Lythraceae family (P. granatum, T. incisa, T. bicornis, L. indica) and four other Magnoliopsida plants (S. mukorossi, V. mangachapoi, G. arboreum, and R. tomentosa) (Figure 8). The results reveal significant interspecific variation in the arrangement of the COX3-ATP8 segment. This region is relatively conserved in S. mukorossi, V. mangachapoi, P. granatum, and T. bicornis, whereas in the genus Lagerstroemia, a distinct NAD4-ATP8-COX3-NAD4L arrangement pattern is observed, accompanied by an inversion in two species. The CYTB-NAD3-ATP6 arrangement pattern remains relatively consistent among the closely related species T. incisa, T. bicornis, L. suprareticulata, and L. indica, although inversion or translocation events have occurred during evolution. These differences indicate that species-specific gene rearrangements have occurred in the mitochondrial genomes of Lythraceae plants during evolution. Furthermore, the detailed variations in gene arrangement patterns among different species within the family (such as between L. suprareticulata and L. indica) further reflect the dynamic and diverse nature of mitochondrial genome evolution.

When comparing the mitochondrial and chloroplast genomes of L. suprareticulata, we identified 34 homologous sequences (Figure 9A), with a total length of 16,866 bp, accounting for 4.63% of the mitochondrial genome. Among these, homologous sequences of lower than 100 bp were the most abundant, totaling 14, followed by sequences of 1001–2000 bp, which amounted to eight (Figure 9B). The longest sequence was 1,485 bp, while the shortest was 33 bp. In addition, a total of seven homologous sequences were detected in the mitochondrial genome. Through comparison between the mitochondrial and nuclear genomes, we identified 2,182 homologous sequences, with a total length of 230,716 bp (Figure 9C). Among these, sequences of 51–100 bp were the most numerous, totaling 943, followed by sequences of 30–50 bp, which amounted to 829 (Figure 9D).

L. suprareticulata were collected and planted in Nanning City, Guangxi, China (Lat. 22.921932° N, Lon.108.353726° E), the process of plant collection and propagation strictly complies with local government regulations, and no destructive collection has been made on the plants themselves. All the voucher specimens were kept by the local laboratory, and can be obtained with the accession number (Holotype: 2025LS0001, isotype: 2025LS0002). Fresh L. suprareticulata samples including leaves stems and flowers were collected from a propagate three-year-old seedlings by cuttings. (May 2022 to October 2023). All the sequenced materials, tender leaves, were surface-cleaned by 70% alcohol to remove dirt and debris. Then they were rapidly frozen in liquid nitrogen for next experiments. High-molecular-weight genomic DNA was isolated from target samples using a modified CTAB method, with residual RNA removed by RNase A digestion. The quantity and integrity of purified DNA were assessed using both of the Qubit fluorometer and agarose gel electrophoresis. Only DNA samples around 10kb were used for subsequent library preparation. SMRTbell template libraries were constructed following the manufacturer’s standard protocol for PacBio HiFi sequencing. After damage repair, end-prep and adapter ligation, the libraries were subjected to primer annealing and polymerase binding. Sequencing was conducted on a PacBio Sequel II/IIe system, and circular consensus sequencing (CCS) was applied to generate highly accurate HiFi reads. Raw sequencing data were processed using official PacBio bioinformatics pipelines to remove low-quality reads and adapters, yielding clean HiFi reads for further bioinformatic analysis.

The mitogenome was assembled by two approaches to guarantee accuracy. Initially, the PMAT was used for first mitogenome assembly, with the settings as ‘-st HiFi -g 300M’ (Bi et al., 2024). The second assembly was performed using Canu with default parameters (Koren et al., 2017). The two data were compared by minimap2, the final assembly was confirmed through comparison. For the convenience of description, we processed mitogenome into a linear molecule by utilizing bandage v 0.8.1 (Wick et al., 2015) to expand the common snippets. The mitogenomes were annotated using the web-based tool PMGA (www.lkmpg.cn/mgavas). Subsequently, the tRNA and rRNA genes were further examined using BLASTn, respectively. We also assembled the chloroplast genome of L. suprareticulata and annotated it with CPGAVAS2 (http://47.96.249.172:16019/analyzer/home) to obtain gbf files (Shi et al., 2019). The final graphs were manually checked.

Simple sequence repeats (SSRs) were fast calculated by the MISA online platform (https://webblast.ipk-gatersleben.de/misa/) (Beier et al., 2017). The specified parameters were same as the previous research (Zhou et al., 2023c), encompassing mono-, di-, tri-, tetra-, penta- and hexa-nucleotides with a minimum occurrence of 10, 5, 4, 3, 3 and 3, respectively. Dispersed repeat was calculated based on a local BLAST (Altschul et al., 1990), and a cutoff of e-value 1-e^-5 was applied. Links for repeat sequences were visualized in TBtools functions (Chen et al., 2023).

The synonymous codons (RSCU) were calculated by Phylosuite function (Zhang et al., 2020) by default parameters. PREPACT3.0 was used to predict the RNA editing sites of PCGs in the mitochondrial genomes. The threshold was established at 0.8 to minimize the similarity score in the plagiarism detector. Using TBtools Ka/Ks calculator function, we calculated the synonymous mutation rates, non-synonymous mutation rates, and their ratios (Ka/Ks).

The other mitochondrial genomes except L. suprareticulata were retrieved and downloaded from NCBI for further comparison, data were obtained by April 2024 (37 species including 11 in Myrtales: Trapa incisa NC_086691, Lagerstroemia indica NC_035616, Trapa bicornis NC_086690, Punica granatum NC_071229, Rhodomyrtus tomentosa NC_071968, Melaleuca alternifolia PP533606, Eucalyptus camaldulensis NC_085240, Eucalyptus grandis NC_040010, Syzygium samarangense OQ701348, Psidium guajava PP934632, Eucalyptus rudis PP920092; 6 in Malvales: Tilia amurensis PQ072837, Gossypium arboreum KR736342, Vatica mangachapoi PP861159, Microcos paniculata NC_086687, Gossypium trilobum NC_035076, Hibiscus cannabinus NC_035549; 8 in Sapindales: Sapindus mukorossi NC_050850, Xanthoceras sorbifolium MK333231, Aglaia odorata NC_084341, Melicope pteleifolia PQ221923, Citrus medica PQ636878, Toona fargesii PQ287271, Nephelium lappaceum PP916047, Litchi chinensis PP932631, Limonia acidissima NC_086955; 8 in Brassicales: Brassica rapa NC_049892, Lepidium apetalum NC_088489, Crucihimalaya lasiocarpa NC_085700, Malcolmia africana OQ784930, Arabidopsis lyrata OQ852786, Brassica nigra NC_029182, Camelina sativa PQ165104, Rorippa indica, PP780168; 4 in Fagales: Castanopsis carlesii PP853255, Fagus sylvatica MW771358, Morella rubra PP533608, Lithocarpus litseifolius NC_065018). To ascertain the phylogenetic location and conflict of the PCGs of selected species were extracted and conserved mitochondrial and chloroplast PCGs were screened for tree building, respectively. After alignment and trimming, a maximum likelihood (ML) method was chosen in IQ-TREE to make the basic tree structure (Minh et al., 2020), based on a default nucleotide replacement model with 1000 self-developing values. The final visualization of the phylogenetic tree was achieved through iTOL software (Letunic and Bork, 2024).