We obtained PacBio-HiFi data for Karpatiosorbus bristoliensis (ERR13245294) from the SRA (https://www.ncbi.nlm.nih.gov/sra/?term=ERR13245294). The whole-genome sequencing of K. bristoliensis revealed a total of 50.3G bases. For mitogenome assembly, we employed PMAT v2.1.2 (https://github.com/aiPGAB/PMAT2/releases) (Bi et al., 2024), an efficient plant mitogenome assembly toolkit using low-coverage HiFi sequencing data, to perform de novo assembly of the K. bristoliensis mitogenome with the command “PMAT autoMito -I -o -st -g -m”. For plastid genome assembly, the ptGAUL v1.0.5 pipeline (https://github.com/Bean061/ptgaul) (Zhou et al., 2023) was applied to extract and assemble the plastid genome from PacBio-HiFi long-read data with standard parameters. We employed Bandage (Wick et al., 2015) to visualize the graphical representation of the mitochondrial and plastid genomes produced by the PMAT and ptGAUL pipelines, respectively. Mitogenome annotation was performed via the online program Plant Mitochondrial Genome Annotator (PMGA) (http://47.96.249.172:16084/index.html), with the database set as 319 mitogenomes (Li et al., 2024). This database uses multiple sequence alignment to identify genes. The plastid genome annotation was performed via the online program CPGAVAS2 (http://47.96.249.172:16019/analyzer/home) (Shi et al., 2019), with Sorbus helenae (NC_068536.1) as the reference genome. Manual corrections were performed via Sequin software. We utilized OGDRAW (v.1.3.1) (https://chlorobox.mpimp-golm.mpg.de/ogdraw.html) to generate plastid genome and mitogenome maps.

The prediction of simple sequence repeats (SSRs) was performed using the online program MISA-web (https://www.web-blast.ipk-gatersleben.de/misa/). The parameters of mono-, di-, tri-, tetra-, penta-, and hexanucleotides were set to 10, 5, 4, 3, 3, and 3, respectively. Tandem repeats within the five species were detected using Tandem Repeats Finder v4.09 (Benson, 1999) with default parameters. Dispersed repeats were identified using the online REPuter program (https://bibiserv.cebitec.uni-bielefeld.de/reputer) (Kurtz, 2001) with a minimal repeat size of 30 bp and a Hamming distance of 3.

we employed the online BLAST tool (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to search highly similar sequences between the mitogenome and plastid genomes of K. bristoliensis, M. alnifolia (Siebold & Zucc.) Koehne, T. glaberrima (Gand.) Sennikov & Kurtto, S. aucuparia L., and Sorbus aucuparia subsp. pohuashanensis (Hance) McAll., respectively. The identification parameters for homologous sequences were set to ≥70% minimum similarity and ≤1e-5 E-value threshold.

We employed the online program PMGA (http://47.96.249.172:16084/deepredmt.html) to predict RNA editing events in the mitogenomes of K. bristoliensis, M. alnifolia, T. glaberrima, S. aucuparia, and S. aucuparia subsp. pohuashanensis. This online program uses Deepred-Mt (Edera et al., 2021), a novel deep convolutional neural network, to detect potential RNA editing events. The cut-off value was set to 0.5 for accurate prediction.

For plastid genome analyses, plastid genome sequences from a total of 59 species of Sorbus s.l. were downloaded. We used the command “cpstools Pi -d work_dir” from the CPStools package (Huang et al., 2024b) to calculate Pi values for the analyzed species. This package automatically extracts shared gene regions and intergenic sequences from GenBank-format files in work_dir, performs multiple-sequence alignments, and computes Pi values. For mitogenome analyses, we utilized PhyloSuite v1.2.3 (Zhang et al., 2020) to extract protein-coding genes from the five Sorbus s.l. species using default parameters. A total of 35 shared protein-coding genes were aligned using MAFFT (https://mafft.cbrc.jp/alignment/server/) (Katoh et al., 2018) with the default parameters. The nucleotide diversity (Pi) of each gene was subsequently calculated using DnaSP v6 (Rozas et al., 2017) with the option “compute variance of Pi”.

The mitochondrial phylogenomic tree was reconstructed via 14 mitochondrial genomes, including those of five Sorbus s.l. species (Sorbus aucuparia subsp. pohuashanensis ON478177, Sorbus aucuparia MT648825, Micromeles alnifolia PP506330, Torminalis glaberrima MT610102, and Karpatiosorbus bristoliensis), with Rosa rugosa (PQ474155) designated as the outgroup. We employed PhyloSuite v1.2.3 (Zhang et al., 2020) to extract protein-coding genes that were shared by the 14 species. Multiple sequence alignment was subsequently performed using MAFFT (https://mafft.cbrc.jp/alignment/server/) (Katoh et al., 2018) with default parameters. The aligned sequence trimming was performed using the trimAl program within the PhyloSuite v1.2.3 (Zhang et al., 2020) with the default (-automated1) parameter. We performed phylogenetic reconstruction using maximum likelihood method in IQ-TREE v2 (Minh et al., 2020), with 1,000 bootstrap replicates (-bb 1000) and 1,000 replicates for the SH-like approximate likelihood ratio test (SH-aLRT; -alrt 1000). The optimal nucleotide substitution model was determined using ModelFinder (Kalyaanamoorthy et al., 2017) in IQ-TREE v2, which identified K3Pu+F+R4 as the best-fit model based on the Bayesian Information Criterion (BIC).

A total of 65 plastomes, including those of 59 Sorbus s.l. species, were utilized for phylogenetic analyses ( Supplementary Table S1 ). The whole plastid genomes were aligned with MAFFT (https://mafft.cbrc.jp/alignment/server/) (Katoh et al., 2018) with default parameter. Phylogenetic reconstruction employed identical software and parameters as above, with the TIM+F+R4 nucleotide substitution model specified according to the Bayesian Information Criterion (BIC)

We performed pairwise sequence alignments among five species of the Sorbus s.l genus using MUMmer v4.0.1 (Marçais et al., 2018), with the parameters set as: nucmer –maxmatch -c 100 -b 500 -l 50. Filtering of aligned sequences was performed with the following parameters: a minimum alignment identity of 90% and a minimum alignment length of 50 bp. Identification of collinear sequences and structural rearrangement events among mitochondrial genomes was conducted using SyRI v1.7.0 (Goel et al., 2019). Genome structural visualization was generated using Plotsr v1.1.0 (Goel and Schneeberger, 2022).

We employed the autoMito program in PMAT to assemble the mitochondrial genome of K. bristoliensis. The average depth of the assembled results was 1408.7x ( Supplementary Figure S1 ). The mitochondrial genome of K. bristoliensis was 386,757 bp in size. It encoded 55 unique genes, including 34 protein-coding genes, 18 tRNA genes and 3 rRNA genes ( Figure 1A ). The GC content was 45.3%. Three genes (ccmFC, trnE-UUC, trnS-GCU) had one intron, one gene (nad4) had three introns, and four genes (nad1, nad2, nad5, nad7) had four introns ( Supplementary Tables S2 , S3 ).

The K. bristoliensis plastid genome has structures identical to those of most angiosperm plants, consisting of one large single copy (LSC), one small single copy (SSC), and two inverted repeats (IRs) ( Figure 1B ). The total length of the plastid genome was 160,322 bp, including 88,182 bp, 19,306 bp and 26,417 bp in the LSC, SSC and IR regions, respectively. The GC content was 36.50%. The plastid genome encodes 75 protein-coding genes, 29 tRNA genes, four rRNA genes, and one pseudogene (ycf1Ψ) for a total of 108 unique genes. Among these genes, eight genes (rps16, atpF, rpoC1, petB, petD, ndhB, rpl2, ndhA) have one intron, and two genes (ycf3, clpP) possess two introns ( Supplementary Table S4 ).

We detected 52, 53, 52, 48, and 51 simple sequence repeats (SSRs) in the mitochondrial genomes of K. bristoliensis, M. alnifolia, S. aucuparia subsp. pohuashanensis, S. aucuparia, and T. glaberrima, respectively ( Figure 2A ). All five species possess three types of SSRs: monomeric, dimeric, and trimeric repeats, with monomeric repeats being the most abundant. Notably, most SSRs were distributed in intergenic spacer regions, with fewer detected in protein-coding regions ( Supplementary Table S5 ). Within protein-coding regions, SSRs primarily occurred in nad1, nad2, and nad3 genes ( Figure 2B ). We identified 50 dispersed repeats in each species, consisting of forward and palindromic repeats ( Figure 2C ). Repeat sequence lengths ranged from 91–428 bp (K. bristoliensis), 88–2745 bp (M. alnifolia), 76–26,130 bp (S. aucuparia subsp. pohuashanensis), 86–6452 bp (S. aucuparia), and 80–438 bp (T. glaberrima) ( Figure 2C ). Notably, S. aucuparia contained two large palindromic repeats (26,130 bp and 24,730 bp; Supplementary Table S6 ). A total of 24, 15, 22, 22, and 24 tandem repeats were identified in the mitochondrial genomes of K. bristoliensis, M. alnifolia, S. aucuparia subsp. pohuashanensis, S. aucuparia, and T. glaberrima, respectively ( Figure 2D ) ( Supplementary Table S7 ).

Mitochondrial genome analyses revealed 14, 13, 12, 12, and 14 chloroplast-derived DNA fragments in K. bristoliensis, M. alnifolia, T. glaberrima, S. aucuparia, and S. aucuparia subsp. pohuashanensis, respectively ( Supplementary Table S8 ). The lengths of these transferred sequences exhibited limited variation, ranging from 3,846 bp (K. bristoliensis), 3,834 bp (M. alnifolia), 3,830 bp (T. glaberrima), 3,755 bp (S. aucuparia), and 3,021 bp (S. aucuparia subsp. pohuashanensis). Notably, the largest individual transferred fragment spanned 1,874 bp. Notably, K. bristoliensis, T. glaberrima, and S. aucuparia subsp. pohuashanensis each contained five fully transferred tRNA genes (trnW-CCA, trnD-GUC, trnH-GUG, trnN-GUU, and trnM-CAU). In contrast, M. alnifolia and S. aucuparia exhibited four completely transferred tRNA genes (trnW-CCA, trnH-GUG, trnD-GUC, and trnN-GTT). Furthermore, partial transfer events involving protein-coding genes were detected, including fragments of psbE, psbC, psaA, psbA, and atpA.

A total of 442, 403, 474, 463, and 352 RNA editing sites were identified in the mitochondrial genomes of K. bristoliensis, M. alnifolia, S. aucuparia, S. aucuparia subsp. pohuashanensis, and T. glaberrima, respectively ( Figure 3A ). Comparative analysis revealed that three taxa (K. bristoliensis, S. aucuparia, and S. aucuparia subsp. pohuashanensis) each contained RNA editing modifications in 32 protein-coding genes, while M. alnifolia and T. glaberrima possessed 31 and 28 edited genes, respectively. Notably, nad4L exhibited the lowest frequency of RNA editing events among all protein-coding genes across the five species, each containing a single editing site. The nad4 gene demonstrated the highest RNA editing activity in four mitochondrial genomes (M. alnifolia, S. aucuparia, S. aucuparia subsp. pohuashanensis, and T. glaberrima), with 39 editing sites identified in M. alnifolia. In contrast, the K. bristoliensis mitochondrial genome showed unique editing patterns, where ccmB and ccmC collectively contained the highest number of RNA editing sites (31 total), representing a distinct regulatory feature in this species. All RNA editing sites belong to C to U. Notably, the start codons of cox1, nad1, rpl16, and rps4 in the mitochondrial genome of K. bristoliensis, through RNA editing, have changed from ACG to AUG. Additionally, RNA editing has contributed to stop codons for atp6 in the mitochondrial genomes of K. bristoliensis, S. aucuparia, and S. aucuparia subsp. pohuashanensis ( Supplementary Table S9 ). The most frequent amino acid substitution was serine-to-leucine, except in M. alnifolia where proline-to-leucine predominated ( Figure 3B ).

Nucleotide diversity analysis was performed using DnaSP v6.0. For mitochondrial genomes, the Pi values ranged from 0 to 0.01808, with the nad4 gene exhibiting the highest diversity (0.01808) ( Figure 4A ). In plastid genomes, the Pi values showed greater conservation, spanning from 0 to 0.02672 (petG-trnW-CCA) ( Figures 4B–D ). Comparative analysis identified seven hypervariable regions as potential molecular markers. Seven regions with relatively high Pi values may be designed for DNA barcodes, including trnH-GUG-psbA, trnT-GUG-trnL-UAA, petG-trnW-CCA, ndhC-trnV-UAC-exon2, trnW-CCA-trnP-UGG, rpl33-rps18.

We recovered a plastome-based phylogeny of Sorbus s.l., which included an ingroup of 59 accessions ( Figure 5 ). Our plastome-based phylogeny yielded strong bootstrap values (> 90%) for the vast majority of branches. With robust node support, we recovered eight major clades (A, B, C, D, E, F, G, H). Among these clades, Sorbus s.l. was dispersed in six clades. Clade A contains four genera (Chamaemespilus, Aria, Torminalis, and Karpatiosorbus) of Sorbus s.l. Karpatiosorbus, which are closely phylogenetically related to Torminalis. The genus Aria was found to be nonmonophyletic, with one species embedded in the genus Chamaemespilus. Clade C contains one species, Cormus domestica. Clade E corresponds to the genera Micromeles and Hedlundia. Micromeles are not monophyletic, with Hedlundia embedded in them. Clades F-H correspond to Sorbus s.s. and are expected to include the hybrid-origin genus Hedlundia and Scandosorbus.

To better understand the mitochondrial genome evolution of Sorbus s.l., we also utilized the maximum likelihood method to construct mitochondrial genome-based phylogenetic trees based on 35 shared protein-coding genes ( Figure 6 ). The Prunus genus was used as the outgroup. Most nodes of the phylogenetic tree presented relatively lower bootstrap values than those of the plastome-based phylogenetic tree did. Consistent with the plastome-based phylogeny, the genus Karpatiosorbus is a sister to Torminalis.

Analysis of mitochondrial genome sequences and structures across five Sorbus species revealed extensive collinear regions, encompassing structural homology, translocations, and inversions ( Figure 7 ). The longest sequence-similarity segments between species pairs were: 84,781 bp (M. alnifolia vs S. aucuparia), 97,089 bp (S. aucuparia vs S. aucuparia subsp. pohuashanensis), 229,635 bp (K. bristoliensis vs T. glaberrima), and 101,625 bp (T. glaberrima vs M. alnifolia). The highest degree of syntenic segment was observed between K. bristoliensis and T. glaberrima (310,482 bp), accounting for 80.28% of the K. bristoliensis mitochondrial genome length. Additionally, a 76,099 bp inversion structure was identified in the K. bristoliensis and T. glaberrima ( Supplementary Table S10 ).

This study reports the first complete mitochondrial and plastid genome assemblies for the endangered tree species K. bristoliensis. The circular mitochondrial genome spans 386,757 bp, demonstrating near-identical size conservation (1 bp difference) with its maternal progenitor Torminalis glaberrima (386,756 bp), consistent with its hybrid origin (Aria × Torminalis). The chloroplast genome of K. bristoliensis is 68 bp smaller than that of Torminalis glaberrima. Compared with those of other species within Sorbus s.l., the mitochondrial genome of K. bristoliensis was smaller than those of M. alnifolia (45,5361 bp) and S. aucuparia subsp. pohuashanensis (396,857 bp) but larger than that of Sorbus aucuparia (384,977 bp). Previous studies have reported that the most minor mitochondrial genome in Rosaceae is Taihangia rupestris var. ciliata T.T.Yu & C.L.Li, with a size of 265,633 bp (Li et al., 2025), and the largest one is Prunus mume (Siebold) Siebold & Zucc. (535,727 bp) (Sun et al., 2022). The mitochondrial genome size of Sorbus s.s. falls within the range of mitochondrial genome size variation observed in Rosaceae. The mitochondrial genome of K. bristoliensis exhibited a GC content of 45.3%, aligning closely with the conserved genomic composition observed across Rosaceae mitogenomes (range: 43.31%–45.62%); (Lu et al., 2023).

SSRs exhibit high polymorphism and codominant inheritance, thus frequently used in population genetics studies (Hu et al., 2019). This study identified a large number of SSRs within the mitochondrial genome of Sorbus s.l., which exhibit potential utility for population genetic investigations. The number of SSRs in the mitochondrial genome of Sorbus s.l. is comparable to that observed in other Rosaceae species such as Rubus idaeus L (Zhang et al., 2025), yet demonstrates a lower frequency compared to members of the Malus genus within the same family (Wang et al., 2024). Consistent with previous findings, monomeric repeats predominated among the three SSR motif categories, exhibiting the highest numerical abundance (Niu et al., 2024; Sun et al., 2024). Tandem repeats may contribute to mitochondrial genome size expansion, particularly in species with exceptionally large genomes (Wynn and Christensen, 2019). Some tandem repeats exhibit species-specific amplification, serving as molecular markers for phylogenetic studies. A total of 107 tandem repeats were detected among the five species, which will be helpful for future studies. Dispersed repeats are nonadjacent homologous sequences ranging from tens to thousands of base pairs and are often distributed throughout the mitochondrial genome. These repeats facilitate homologous recombination, a key driver of structural rearrangements such as inversions, duplications, and the generation of subgenomic circles (Wu and Sloan, 2019). Previous studies have demonstrated prevalent mitochondrial genome rearrangements in Rosaceae species, as exemplified by members of the genus Malus (Wang et al., 2024) and Fragaria (Fan et al., 2022). Notably, one large palindromic repeat (26,130 bp) and forward repeat (24,730 bp) were detected in the mitochondrial genome of S. aucuparia. In addition, the S. aucuparia subsp. pohuashanensis mitogenome also presented one forward repeat exceeding 6,000 bp (6452 bp). Whether these large repeats are involved in mitochondrial genome rearrangement and other structural evolution remains to be further studied.

Horizontal gene transfer (HGT) from chloroplasts to mitochondrial genomes represents a notable phenomenon in plant organelle evolution, reflecting complex inter-organellar genomic interactions (Yang et al., 2022). Although chloroplast-to-mitochondrion gene transfers occur less frequently than chloroplast-to-nuclear transfers do, accumulating evidence reveals their evolutionary significance in mitogenome evolution (Hong et al., 2021). In this study, we identified some chloroplast fragments transferred to the mitochondrial genome, with sizes ranging from 3,021 bp to 3,856 bp. Compared with previous studies in other Rosaceae, such as Rubus idaeus L. (46,456 bp transferred from chloroplasts to mitochondria) (Zhang et al., 2025), we found that homologous fragment transfer between the plastome and mitogenome in Sorbus s.l. was relatively rare. These findings indicate that the mitochondrial and chloroplast genomes of Sorbus s.l. are relatively highly conserved. Consistent with previous studies (Wang et al., 2007), we also found that the most frequently transferred genes were tRNAs, with a total of five tRNA genes (trnW-CCA, trnD-GUC, trnH-GUG, trnN-GUU, and trnM-CAU) shared by the plastid genome and the mitochondrial genome.

RNA editing is frequently reported in plant mitochondrial genomes (Liu et al., 2024). It is a crucial posttranscriptional modification process that plays a crucial role in controlling gene expression and functionality (Liu et al., 2020). At present, the common type of RNA editing is C–U editing, and G-to-C and T-to-A editing can also occur (Liu et al., 2024; Ichinose and Sugita, 2016; Yang et al., 2022). In this study, all RNA editing sites in Sorbus s.l. were associated with C–U editing. The number of RNA editing sites differed greatly in Sorbus s.l., ranging from 352 (T. glaberrima) to 463 (S. aucuparia subsp. pohuashanensis). Compared with those in other Rosaceae species, the number of RNA editing sites in Sorbus s.l. species was greater than that in Prunus pedunculata (262) (Liu et al., 2023) but less than that in Rubus chingii var. suavissimus (492) (Shi et al., 2024) and Taihangia rupestris var. ciliata (470) (Li et al., 2025). Overall, RNA editing sites exhibit significant variations across different species and even within the same genus, such as S. aucuparia subsp. pohuashanensis and S. aucuparia. Notably, the RNA editing sites appear to exhibit a gene preference, with editing events occurring most frequently within the NADH dehydrogenase genes. Similar results have been reported in other species, such as Prunus pedunculata (Pall.) Maxim. (Liu et al., 2023) and Gelsemium elegans (Gardner & Champ.) Benth (You et al., 2023). We also observed that RNA editing influenced the start and termination codons of protein-coding genes. For example, the start codons of cox1, nad1, rpl16, and rps4 changed from ACG to AUG through RNA editing, whereas RNA editing led to the termination codons of atp6. RNA editing events that alter initiation and termination codons are frequently observed, as exemplified by modifications to the nad1 initiation codon and ccmFc termination codon in the mitochondrial genome of Garcinia mangostana L (Wee et al., 2022).

The mitochondrial and chloroplast genomes harbor a substantial number of polymorphic sequences that can be developed as DNA barcodes for phylogenetic and population genetic studies (Li and Wei, 2022). This study analyzed the mitochondrial genomes of five Sorbus s.l. species and revealed that the nad4 gene presented the greatest number of nucleotide polymorphisms. These findings suggest its potential utility as an effective DNA barcode for investigating phylogenetic relationships within Sorbus s.l. Through comparative analysis of chloroplast genomes, we identified six plastid gene regions exhibiting high genetic polymorphism, which could serve as chloroplast DNA barcodes for investigating phylogenetic relationships within Sorbus s.l. Furthermore, the ycf1 gene, conventionally employed as a universal barcode for angiosperms (Dong et al., 2015), demonstrated lower nucleotide polymorphism in this study.

In this study, we provide a high-resolution maternal genetic framework for Sorbus s.l. Aria pannonica (Kárpáti) Sennikov & Kurtto, which is a hybrid-origin species with a ploidy level of triploid (Somlyay and Sennikov, 2015). A. pannonica originated from intragenus hybridization because of its morphological similarity to A. nivea Host and A. graeca (Spach) M.Roem (Somlyay and Sennikov, 2015). However, A. pannonica is more closely related to the genus Chamaemespilus in our plastome-based phylogeny. Therefore, we inferred that the female parent of this hybrid species may have originated from Chamaemespilus. This fact indicated that Aria pannonica should treat as the member of Majovskya. The genus Karpatiosorbus is thought to originate from the hybridization of Aria and Torminalis (Pellicer et al., 2012). Both our mitogenome- and platome-based phylogenetic data supported that Karpatiosorbus and Torminalis clustered into one branch. These findings indicate that the female parent of this genus is Torminalis. The genus Hedlundia contains one sexual hybrid and 39 apomictic species and originates from crosses between various species of Aria and Sorbus s.l (Sennikov and Kurtto, 2017). The genus Hedlundia is separated into two clades: one clade is closely related to Micromeles, and the other clade is embedded in Sorbus s.s, indicating the complex origin of the hybrid genus. Sennikov and Kurtto (2017) proposed the establishment of the new nothogenus Sorbomeles to accommodate hybrids of Sorbus and Micromeles origin. In the present study, molecular evidence suggests Micromeles likely served as the maternal progenitor of Hedlundia austriaca (Beck) Sennikov & Kurtto and H. persica (Hedl.) Mezhenskyj. Consequently, these two taxa are hereby formally proposed for reclassification into the hybrid genus ×Sorbomeles. A previous study indicated that hybridization between Aria and Sorbus s.s. contributed to the genus Micromeles. Both our results and those of previous plastome-based studies support that Micromeles is a sister to Sorbus s.s., whereas this is inconsistent with nuclear phylogenetic analyses (Zhang et al., 2023). Therefore, hybridization occurs between Micromeles and Sorbus s.s. Scandosorbus intermedia (Ehrh.) Sennikov originates from hybridization among Sorbus s.s., Aria, and Torminalis (Pellicer et al., 2012). It involves a complex species formation process. In this study, S. intermedia was a sister to S. dumosa. These findings indicate that the female parent of this species is S. dumosa House or a related species.

Phylogenetic analyses have consistently recovered Sorbus s.s. as a monophyletic clade distinct from other simple-leaved genera and the compound-leaved genus Cormus in previous studies (Campbell et al., 2007; Zheng and Zhang, 2007; Potter et al., 2007; Lo and Donoghue, 2012). This limitation arises from inadequate sampling in prior studies, which excluded phylogenetically critical hybrid-origin taxa. However, our current phylogenomic reconstruction reveals that Hedlundia and Scandosorbus (simple-leaved hybrid-origin genera) nests within Sorbus s.s., thereby rendering the latter non-monophyletic. This phylogenetic incongruence results from the hybrid origin of Hedlundia and Scandosorbus, which inherited its maternal genome from ancestral Sorbus s.s. lineages. Sorbus s.s. is one of the most typical examples of taxonomic complexity arising from the combined effects of hybridization, polyploidy and apomixis in the Rosaceae (Robertson et al., 2010). Previous research has demonstrated that both insect-pollinated and bird-pollinated are present in the genus Sorbus s.s.; moreover, mammals also play an important role in the spread of Sorbus s.s. seeds (Raspé et al., 2000; Bednorz et al., 2006). Some species of this genus are self-incompatible (Raspé et al., 2000). These factors strongly increase the frequency of interspecific hybridization in Sorbus s.s. as well as hybrid with other genera. Therefore, elucidating taxonomic challenges in Sorbus s.s requires resolving hybridization patterns and clarifying phylogenetic placements of hybrid-origin taxa within the genus.

Mitochondrial genome structural complexity arises from extensive repetitive sequences and frequent structural rearrangements. Such mitochondrial genome rearrangements have also been frequently reported in other genera of Rosaceae, notably Prunus (Zhai et al., 2025) and Malus (Wang et al., 2024). This study revealed that K. bristoliensis shares substantial homologous segments with T. glaberrima (species from its maternal genus), while exhibiting limited homology with phylogenetically distant genera. This pattern aligns with their close phylogenetic origin and the hybrid origin of Karpatiosorbus. Despite the extensive shared homology, a 76,099 bp inversion structure was identified between their mitochondrial genomes, suggesting incipient mitogenomic divergence between the two genera.