For the acquisition of the chloroplast genome and mitochondrial genome of A. dahurica, fresh foliar samples were harvested from Suining, Sichuan, China. The extraction of total genomic DNA was performed utilizing the cetyltrimethylammonium bromide (CTAB) technique (Aboul-Maaty and Oraby, 2019). This was followed by the construction of a DNA library, with an insert size of 300 bp. In parallel, Oxford Nanopore sequencing was employed on identical plant specimens used for next-generation sequencing (NGS). This involved long-read sequencing of high-quality DNA from fresh samples, adhering to the protocols specified in the SQK-LSK109 genomic sequencing kit provided by ONT, Oxford, UK.

Total RNA extraction was meticulously performed utilizing the RNAsimple Total RNA Extraction Kit (DP419) from TIANGEN. In the process of constructing the lncRNA library, transcriptase in conjunction with random hexamer primers. The cDNA underwent end repair, adaptor ligation, and precise size selection employing the AMPure XP system. The final sequencing was executed on the advanced Illumina Novaseq 6000 platform, which facilitated the generation of 150 bp paired-end reads.

Initially, the assembly of the A. dahurica mitochondrial genome was conducted using long-read data, employing the default settings of the Flye software, which produced graphical outputs in Genome Fragment Assembly (GFA) format (Kolmogorov et al., 2019). Long-read sequencing data were subjected to de novo assembly employing Flye software (version 2.9.2) with its default settings. The resultant assembly encompassed sequences from the nuclear, chloroplast, and mitochondrial genomes. Following this, a contig library, essential for subsequent analyses, was constructed utilizing the makeblastdb utility. To identify contigs harboring mitochondrial DNA segments, we leveraged the BLASTn algorithm (Blast N G, 2019), referencing Arabidopsis thaliana mitogenome (NC_037304) and applying stringent parameters: "-evalue 1e-5 -outfmt 6 -max_hsps 10 -word_size 7 -task blastn-short" (Chen et al., 2015). This procedure facilitated the selective extraction of mitochondrial-associated contigs. In the subsequent phase, the GetOrganelle tool was employed to procure short reads specific to the mitochondrial genome, which were then utilized to construct a graphical representation of the genome using the SPAdes software. To integrate and corroborate these findings, we applied the Unicycler software, incorporating the BWA tool, to align the graphical genome-derived contigs against the mitochondrial contigs obtained from long-read assembly (Li and Durbin, 2010). This process enabled the reconstruction of the A. dahurica mitochondrial genome by amalgamating short and long reads, culminating in the successful retrieval of the complete mitochondrial genome (Li and Durbin, 2010). Visualization of the reconstructed genome was accomplished using Bandage software (Wick et al., 2015).

Annotation of the mitochondrial genome was then conducted, referencing the previously published mitochondrial genomes of A. thaliana (NC_037304.1), utilizing Geseq software for this purpose (Tillich et al., 2017)(https://www.ncbi.nlm.nih.gov/nuccore/NC_037304.1). tRNA genes within the mitochondrial genome were annotated through the application of tRNAscan-SE software (Chan et al., 2021), while rRNA genes were annotated via BLASTn analysis (Chen et al., 2015). Corrections to any inaccuracies in the annotations of each mitochondrial gene were made using Apollo software (Lewis et al., 2002).

The differential codon usage rates across diverse organisms are postulated to be a consequence of evolutionary equilibrium established through prolonged selective processes. Analysis of Relative Synonymous Codon Usage (RSCU) typically involves extracting protein-coding sequences from the mitochondrial genome utilizing PhyloSuite software (Zhang et al., 2020). This is followed by the computation of RSCU values ( Supplementary Table S1 ), which is accomplished using MEGA software (version 7.0) (Kumar et al., 2016). For the identification of repetitive sequences within the genome, microsatellites, tandem repeats, and dispersed repeats was conducted using MISA, Tandem Repeats Finder (TRF), and REPuter, respectively (Benson, 1999; Kurtz et al., 2001; Beier et al., 2017). Visualization of these findings was achieved through the use of Microsoft Excel and the Circos visualization package (Zhang et al., 2013).

The assembly of the chloroplast genome was executed via the GetOrganelle toolkit with default parameters (Jin et al., 2020), followed by an enhancement of annotations through the application of CPGAVAS2 software (Shi et al., 2019). Subsequent analysis of homologous sequences ( Supplementary Table S2 ) was conducted employing the BLASTn (Chen et al., 2015), with graphical representation facilitated by the Circos visualization tool (Zhang et al., 2013). Comparative analysis of mitochondrial genomes was undertaken utilizing the BLASTn (Chen et al., 2015). This was accompanied by the identification of conserved collinear segments ( Supplementary Table S3 ), specifically isolating homologous sequences exceeding a threshold of 500 base pairs in length (Wang et al., 2012). The visualization of these segments was accomplished through the use of the Multiple Synteny Plot system (Baek et al., 2016).

According to the taxonomic relationship, the mitochondrial genome of the closely related species ( Supplementary Table S4 ) was selected for further investigation. Our methodology began with the protein-coding gene (PCG), incorporating adjacent regions spanning 100 base pairs as referential sequences. We aligned strand-specific RNA-seq reads to these sequences via HISAT2 (version 2.2.1) (Kim et al., 2019), adhering to parameters “–rna-strandness RF –sensitive –no-mixed –no-discordant”. Subsequently, we engaged REDItools (version 2.0) (Picardi and Pesole, 2013) to pinpoint RNA editing sites, setting the detection threshold at a minimum coverage of 5 and a frequency of 0.1 or greater ( Supplementary Table S5 ). Subsequently, the DNA sequences of the 16 protein-coding genes (PCGs) shared among these ten mitogenomes were extracted ( Table 1 ). These sequences were aligned with MAFFT (v7.450) (Rozewicki et al., 2019), and a phylogenetic tree was constructed using Phylosuite with the maximum likelihood (ML) method based on the alignment. The credibility of the phylogenetic tree was assessed by performing bootstrap testing with 1,000 replications. Finally, the resulting maximum-likelihood tree was visualized using iTOL (https://itol.embl.de/) (Letunic and Bork, 2021).

In accordance with our predictive analyses, we conducted a verification of the loci where start and stop codons underwent modifications as a consequence of RNA editing. This process entailed the amplification of genomic DNA (gDNA) and complementary DNA (cDNA) via polymerase chain reaction (PCR), employing a system that has been documented in prior studies (Jiang et al., 2023). Detailed descriptions of the polymerase chain reaction (PCR) amplification system and the corresponding conditions are provided in Supplementary Table S6 . Subsequently, the PCR products were subjected to Sanger sequencing. The sequencing outputs were then meticulously visualized and cross-verified manually utilizing the SnapGene software.

From the untig graph, we extracted sequences near the double bifurcation structures. Each structure aligned with four unique sequence patterns, sharing the same central repeat sequences but varying in adjacent sequences. This variation hinted at recombination driven by repetitive sequences. For different conformation analysis, we extracted fasta files using Bandage and aligned them against nanopore sequencing data with minimap2 (Li, 2018). Long-read supported sequence configurations were validated via PCR and Sanger sequencing. Primer design for these assays was focused on sequences flanking the repetitive regions ( Supplementary Table S7 ), following the methodology described in subsection 2.5.

The assembly graph of the A. dahurica genome is branched with a total length of 228,315 bp and a GC content of 45.06% ( Table 1 ). Following the long reads of ONT data analysis to assess the repetitive regions, we successfully assembled 12 circular contigs ( Figure 1 ). The longest chromosomes 1 was 26,966 bp in length and the shortest chromosomes 12 was 9,875 bp, respectively. A total of 34 unique protein-coding genes were annotated from the mitochondria, comprising of 24 core genes and 10 non-core genes ( Table 2 ), along with 23 tRNA genes and three rRNA genes. The core gene set includes five ATPase genes, 9 NADPH dehydrogenase genes, four cytochrome C genes, three cytochrome C oxidase genes, one membrane transport protein gene, one mature enzyme gene and one panthenol-cytochrome C reductase gene. Non-core gene set consists of four ribosomal large subunit genes, five ribosomal small subunits and two succinate dehydrogenase genes.

Codon preference denotes a phenomenon in which specific codons are utilized more prevalently than others in the DNA or RNA sequences of certain organisms, often indicating a propensity that can influence gene expression and protein assembly (Parvathy et al., 2022). An examination of codon selection was conducted for 34 mitochondrial protein-encoding genes (PCGs), detailing the occurrence of each amino acid codon in Supplementary Table S1 . Codons demonstrating a Relative Synonymous Codon Usage (RSCU) value above 1 were recognized as favored by amino acids. Apart from the RSCU values of the starting codon AUG and the codon for tryptophan (UGG), both precisely 1, a significant pattern in codon usage preference is evident within mitochondrial PCGs ( Figure 2 ). Additionally, we conducted a comparative analysis of codon usage preferences between A. dahurica and ten closely related species, revealing similar frequencies with higher usage rates for Arginine, Leucine, and Serine ( Supplementary Figures S1-10 ).

Simple Sequence Repeats (SSRs), commonly known as microsatellites, constitute brief, yet significant, DNA segments replicated in a sequential manner within the genome (Wang et al., 2021). These SSRs are extensively utilized as biomarkers in genetic research, owing to their pronounced variability among distinct individuals. Within the A. dahurica mitochondrial genome, these SSRs manifest as microsatellite repeat sequences characterized by tandemly repeated motifs, each composed of 1-6 nucleotides. Notably, Chromosome 1 harbors eight instances of these SSRs, underscoring their prevalence in the genome ( Supplementary Figure S11 ).

Furthermore, tandem repeats, pivotal in genetic research and forensic science, are DNA sequences where nucleotides are duplicated in a consecutive fashion, exhibiting variability in length (Tørresen et al., 2019). Classified often as satellite DNA, these tandem repeats comprise contiguous sequences of units, each spanning a range of 7 to 200 nucleotides. However, a distinct feature of Chromosome 1 is its absence of such tandem repeats, as elucidated in Figure 3 . Additionally, the genome encompasses a diverse set of 10 repetitive sequences, each extending over 30 base pairs (bp). This array includes four pairs of forward repetitive sequences and six pairs of palindromic repetitive sequences, thus highlighting a rich complexity of repetitive elements within the genomic structure.

In the context of plant mitochondrial genomes, homologous recombination, involving the interchange of genetic material between analogous or identical DNA sequences, markedly enhances the diversity and evolutionary dynamics of these genomes (Yang et al., 2022). This mechanism is pivotal in preserving the structural and functional wholeness of mitochondrial DNA across diverse plant species. During the assembly of the A. dahurica mitochondrial genome facilitated by long-read sequencing techniques, two distinct repetitive sequences were identified as potential mediators of homologous recombination: R1 (ctg13) and R2 (ctg14), measuring 4,733 base pairs (bp) and 1,119 bp, respectively. These sequences, characterized as direct repeats, have specific roles: R1 (ctg13) associates with ctg12 and ctg2 to form chromosomes 1 and 2, while R2 (ctg14) aligns with ctg11 and ctg7 to generate chromosomes 3 and 4, as delineated in Figure 4 . Additionally, these repeat sequences facilitate the formation of circular chromosomal molecules for chromosomes 1 and 2, and 3 and 4.

To substantiate the role of these repeat sequences in orchestrating homologous recombination, rigorous experimental approaches were employed, encompassing Polymerase Chain Reaction (PCR) amplification and Sanger sequencing. The methodology for primer design is exhaustively described in Supplementary Figures S12, S13 , with a focus on primers adept at amplifying the repeat sequences. The PCR products obtained were instrumental in verifying a range of genomic configurations, corroborating the insights gained from the long-read sequencing analysis. These findings support the hypothesis of intricate homologous recombination mechanisms within the A. dahurica mitochondrial genome. Consequently, it is postulated that these repetitive sequences facilitate chromosomal recombination, culminating in the emergence of 12 unique circular chromosomal conformations, as illustrated in Supplementary Figure S14.

Utilizing BLASTn analysis for comparative genomics between the mitochondrial and plastid genomes (accession number PP049083, which was assembled during the course of this study), nine homologous sequences were identified, collectively spanning 1,732 bp ( Figure 5 ). This constitutes 0.76% of the total mitochondrial genome, as detailed in Supplementary Table S2 . The most extensive of these sequences is designated as MTPT9, measuring 382 bp in length. Within these homologous fragments, nine complete genes have been identified, comprising one protein-coding gene (petG) and five transfer RNA (tRNA) genes, specifically trnD-GUG, trnH-GUG, trnN-GUU, trnI-CAU, and trnW-CCA. Specifically, MTPT3 and MTPT6 merit attention due to their alignment with homologous regions situated in the Inverted Repeat segments of the chloroplast genome, leading to the duplication of these sequences in the chloroplast. Such loci are hypothesized to be critical hotspots for sequence migration, indicating a significant role in genomic structural dynamics.

To elucidate the evolutionary lineage of the A. dahurica mitochondrial genome, a phylogenetic analysis was conducted. This involved the construction of a phylogenetic tree based on the DNA sequences of 24 conserved mitochondrial protein-coding genes (PCGs) from 29 distinct species, as illustrated in Figures 6 , 7 . For this analysis, two mitochondrial genomes from the order Solanales were utilized as outgroups. The resulting phylogenetic topology aligns with the contemporary classification system of the Angiosperm Phylogeny Group (APG). Within this framework, A. dahurica is classified under the family Apiaceae, exhibiting a closer evolutionary relationship to Saposhnikovia divaricata.

In our genomic synteny study, we have identified remarkably homologous and collinear blocks within the genome of A. dahurica, interconnected through complex ribbons, as detailed in Supplementary Table S3 . Figure 7 highlights the extensive rearrangements these collinear blocks undergo across various plant orders, illustrating significant genomic variability. Moreover, a comprehensive comparative genomic analysis contrasts A. dahurica with its closely related species, particularly S. divaricata (NC058846.1). Phylogenetic analysis of a genus is an important means of homology analysis, such as plastid, mitochondrial and geographical distribution analysis of the entire genus Lauraceae (Liu et al., 2022; Tan et al., 2023; Yang et al., 2023a; Yang et al., 2023b). In this comparison, we discovered ten fragments exceeding 5000 bp in length, indicative of long segments with high homology to S. divaricata, thus reinforcing their close phylogenetic relationship. However, despite these similarities, the genomic structure of A. dahurica demonstrates a notable lack of conservation compared to its closely related counterparts, thereby underscoring its unique structural properties.

RNA editing phenomena within 34 protein-coding genes (PCGs) of the A. dahurica mitochondrial genome were investigated utilizing Deepred-mt, applying a threshold value of 0.9, as depicted in Figure 8 . This analysis revealed a total of 615 putative RNA editing sites across these PCGs, as cataloged in Supplementary Table S5 . Each of these sites involved cytidine-to-uridine (C-to-U) base modifications. Notably, the nad4 gene exhibited the highest frequency of potential RNA editing, with 48 sites, surpassing all other mitochondrial genes. This was closely followed by the mttB gene, which presented 41 RNA editing events.

To further substantiate the occurrence of RNA editing events, the cox1, atp6, and atp9 genes were selected for detailed examination through Polymerase Chain Reaction (PCR) amplification and subsequent Sanger sequencing. The Sanger sequencing analyses were executed using SnapGene software. The results confirmed the presence of RNA editing at specific sites, including cox1-2, atp6-718, and rps9-223. Among these, the cox1 gene demonstrated a particularly high RNA editing efficiency, as illustrated in Figure 9 .