Background

Horizontal gene transfer (HGT) is a process that transmits genetic material between species or individuals without mating. Plant mitochondria are especially susceptible to HGT as numerous foreign mitochondrial genes have been uncovered in various seed plants [1,2,3,4,5,6,7,8,9,10,11,12]. In contrast, only a few cases of integrated mitochondrial DNA were reported in plastids [13,14,15,16,17,18]. There are three explanations for this striking disparity. First, plant mitochondria have an active transmembrane potential-dependent system that allows uptake of exogenous DNA fragments up to a few kilobase pairs [19]. Second, fusion, fission, and recombination frequently occur between plant mitochondria [20]. Third, plant mitogenomes contain lengthy intergenic regions that provide spaces for foreign DNA integration without disruption of functional genes [21].

Horizontally acquired DNA could exert a profound influence on plant mitogenome evolution. One of the most extreme examples was found in Amborella trichopoda whose enormous, 3.9-Mb, mitogenome contains 197 foreign mitochondrial protein genes acquired from green algae, mosses, and other angiosperms [5]. Large-scale HGT was also discovered in the Lophophytum mitogenome where foreign DNA accounts for more than half of the genome [22]. Acquisition of foreign genes can result in pseudogenization, replacement of native genes, or formation of chimeric genes through recombination [21]. Despite a few exceptions reported in parasitic mitogenomes [6, 23], few foreign sequences discovered in plant mitogenomes are actively expressed but rather become pseudogenes [4, 5, 7, 24]. As a result, most detectable HGT events have occurred recently, since ancient non-functional foreign genes degrade over time and vanish from contemporary mitogenomes.

HGT is thought to occur through direct physical contact and vector-mediated transmission (see review in [25]). Based on regular mitochondrial fusion in green plants, the “fusion compatibility” model was put forward to interpret transfer of massive foreign mitochondrial DNA molecules [5]. This model holds that capture of entire foreign mitochondria represents the first step toward mitochondrion-to-mitochondrion HGT, followed by mitochondrial fusion and genomic recombination to generate chimeric mitogenomes where all foreign DNA, including mitochondrial plastid-derived (MTPT) loci, are eventually acquired [5, 11, 22]. So far, horizontal acquisition of massive amounts of DNA has not been observed in any available gymnosperm mitogenome. Only three studies documented HGT of single mitochondrial loci from angiosperms to gymnosperms [2, 8, 10].

Gnetum, including 25‒35 species [26], is the sole genus in the family Gnetaceae of the order Gnetales (i.e., the gnetophytes). This angiosperm-like gymnosperm genus comprises mostly dioecious lianas bearing broad-bladed and pinnate-veined leaves [27,28,29]. They inhabit tropical and subtropical lowland rainforests of West Africa, South America, and Southeast Asia. The latter is the diversity hotspot [30]. Molecular phylogenetic studies have separated Asian Gnetum into two clades, I and II [2, 31], although it is still not clear whether the clade I members are monophyletic [30]. Won and Renner [2] discovered that some Gnetum species within the Asian clade II possess a mitochondrial nad1 intron 2 copy with sequences nearly identical to its angiosperm homologs. This “angiosperm-type” intron was further interpreted to come from an HGT from an euasterid to Gnetum. Although this study represented the first discovery of HGT in gymnosperms, the following questions remain: (1) Have other HGT events taken place in Gnetum? (2) If yes, to what extent is the foreign DNA present in the contemporary mitogenomes? (3) Which species/genera are the HGT donors? and (4) What is the mechanism underlying Gnetum HGT events?

To gain a comprehensive understanding of HGT in Gnetum, we sequenced and assembled mitogenomes from five species that represent the two Asian clades. Our data suggest that the mitogenomes of these species have experienced multiple rounds of HGT involving large amounts of organellar DNA from diverse angiosperms. We also present evidence of DNA-mediated origins of the transferred fragments.

Results

Characterization of three Asian Gnetum mitogenomes

The G. gnemon and G. parvifolium mitogenomes were assembled using the Unicycler assembler (see “Methods”) into 9 and 16 circular-mapping chromosomes with a total size of 575,501 and 1,394,970 bp, respectively (Fig. 1; Additional File 1: Fig. S1). We used a different assembler (hybridSPAdes [32]) for the G. ula genome because Unicycler was unable to complete this genome without system errors. We obtained 21 linear mitogenomic scaffolds for G. ula with a total length 1,372,030 bp (Additional File 1: Fig. S2). The GC content in these three newly assembled mitogenomes ranges from 47.2 to 48.0% (Table 1). Further, 28‒29 protein genes, 3‒4 rRNAs, and 5‒6 tRNAs are annotated as native. The native rpl10 gene is retained in G. gnemon but lost from G. parvifolium and pseudogenized in G. ula. Repetitive sequences, including dispersed and tandem repeats, make up 5.5‒8.6% of the mitogenomes (Table 1). In addition, four gene clusters (i.e., nad2 exons 3‒5 and ccmC; nad2 exons 1‒2 and nad4L; atp4, cob, and nad9; nad4 and nad5 exons 1‒2) are conserved among the Gnetum mitogenomes (Fig. 1; Additional File 1: Figs. S1‒2). We detected 338‒344 C-to-U RNA editing sites, most of them (78.2‒81.6%) nonsynonymous (Table 1).

Fig. 1
figure 1

The mitogenome map of Gnetum parvifolium. Gray bars represent 16 circular-mapping chromosomes displayed as linear molecules for easy comparison. Genes on the outside of the chromosomes are transcribed in counterclockwise directions, while those on the inside are clockwise. Loci are color-coded depending on their origins. Light-blue histograms denote DNA read depths in log scale. Chr,  chromosome

Full size image
Table 1 Characterization of the three sequenced Gnetum mitogenomes
Full size table

Like other seed plants, Gnetum also has mitochondrial plastid-derived DNA obtained through intracellular transfer, called native MTPTs [33]. These native MTPTs amount to 1,693‒23,210 bp (Table 1) and contain 2‒9 tRNAs that can form a three-leafed clover structure (Additional file 2: Table S1). Using blast searches, we detected numerous foreign sequences homologous to plastid or mitochondrial DNA from diverse angiosperms but not matching any non-Gnetum gymnosperms. According to our blast results, these foreign sequences were further designated as angiosperm plastid-derived (we call them foreign MTPTs) and angiosperm mitochondrion-derived (Table 1). Three lines of evidence indicate that these sequences are not the result of DNA contamination. First, our assembled mitogenomes contain the previously characterized “angiosperm-type nad1 intron 2” copy that was horizontally acquired from an euasterid [2]. Second, our blast searches reveal that independently sequenced mitogenomes from other Gnetum species also contain homologs of most of these foreign sequences. Third, all native and foreign sequences are similar in their DNA read depths, except for the native MTPTs (Fig. 1; Additional File 1: Figs S1‒2). The latter are nearly identical in sequence to their plastid counterparts. Their reads are thus counted together with the plastid copies during read mapping, resulting in artificially higher read depths for native MTPTs than other mitogenomic regions. Collectively, the total amount of the foreign sequence varies from 36,635 to 194,613 bp, accounting for 6.4‒14.2% of the Gnetum mitogenomes (Table 1). This suggests that accumulation of foreign DNA has contributed to size variation among the Asian Gnetum mitogenomes.

Phylogenetic evidence of HGT from eudicots to Asian Gnetum

We identified 20 foreign genes in the Gnetum mitogenomes sequenced in this study (Fig. 2). These foreign genes occupy 2.8‒11.3% of the total foreign sequences. Five of them are foreign MTPTs and the rest are derived from mitochondria. We should be able to trace the origins of these foreign genes because their homologs exist in diverse plants. We employed three strategies to broaden taxon sampling and facilitate computing. First, we sequenced and assembled two more mitogenomes from G. gnemon var. brunonianum and G. pendulum using short read platforms. Second, we found and annotated foreign genes from the publicly available G. hainanense (LC650069‒LC650085) and G. montanum (MW354253‒MW354270) mitogenome scaffolds. Third, all available native homologs of the 20 examined foreign genes were retrieved from ferns, gymnosperms, and angiosperms with one representative per genus.

Fig. 2
figure 2

Summary of the 20 detected foreign genes and their close phylogenetic relatives. Foreign genes are listed on the left-hand side, while their close relatives are shown on the right-hand side. The color-coded rectangles in the middle denote the status of these foreign genes defined in the box of the right panel. G. gnemon var. brunonianum is abbreviated to “G. gnemon var.”. The ML tree shown on the upper panel was inferred from concatenating 28 native mitochondrial genes under a 50% majority rule. Supported values along branches were estimated from 1000 bootstrap replicates

Full size image

Our maximum likelihood (ML) trees (Additional File 1: Figs. S3‒21) placed all examined foreign genes within eudicots rather than gymnosperms, strengthening the inference of horizontal acquisition of multiple genes during the Asian Gnetum evolution. For example, native and foreign ccmB genes coexist across the Asian Gnetum clade II (Fig. 2) and our tree suggests that all foreign ccmB copies form a clade that not only deviates from gymnosperms but also strongly affiliates to a Rubiaceae genus Coptosapelta (BS = 100%; Additional File 1: Fig. S3). Such strong affiliations were also recovered in ten other foreign genes restricted to the clade II of Asian Gnetum, including foreign ccmFc, ccmFn, matR, nad1 exons 2‒3, nad1 exons 4‒5, nad5 exons 4‒5, rps1, rps7, ndhB exon 1 and intron, and rps13 (all BS > 90%; Additional File 1: Figs. S4‒13). These results indicate that Coptosapelta is a source of ample genetic materials for Gnetum prior to species diversification within the Asian clade II.

Furthermore, our ML trees support multiple rather than a single origin for seven other foreign genes: cox2, nad6, psaA, psaB, rpl10, rps3, and rps4, though they also are confined to the Asian Gnetum clade II (Fig. 2). A distinctive clade was recovered for each of these seven foreign genes (Additional File 1: Figs. S14‒20), clearly indicating that their associated HGT events occurred prior to the origin of the Asian clade II. Although foreign nad6 and rps4 are next to each other (Fig. 1), they differ in their donors: Balanops for the former (BS = 98%; Additional File 1: Fig. S15), but a clade comprising Garcinia, Populus, and Salix for the latter (BS = 99%; Additional File 1: Fig. S20). In contrast, two neighboring foreign MTPT genes, psaA and psaB, are of a congeneric origin from Hydnocarpus (Both BS > 95%; Additional File 1: Figs. S16‒17), indicating a co-transfer event. Foreign cox2, rpl10, and rps3 genes reside on different chromosomes and have independent origins (Fig. 2): the clade comprising Ricinus, Vernica, Hevea, and Manihot for foreign cox2 (BS = 86%; Additional File 1: Fig. S14), only Garcinia for foreign rpl10 (BS = 91%; Additional File 1: Fig. S18), and the Garcinia-Bruguiera clade for foreign rps3 (BS = 83%; Additional File 1: Fig. S19). Notably, the abovementioned ten genera belong to Malpighiales (Fig. 2), highlighting frequent HGT from this eudicot order to Asian Gnetum.

The foreign atp4 copies in G. ula and G. pendulum were not placed into any sister clade with greater than 50% confidence (Fig. 2; Additional File 1: Fig. S21). A foreign MTPT that includes ndhB intron and its downstream exon (i.e., ndhB intron and exon 2) is uniquely present in G. gnemon and its variety within the Asian clade I (Fig. 2). This G. gnemon-specific locus has a close relationship to Averrhoa of Oxalidaceae (BS = 73%; Additional File 1: Fig. S12) but differs from its Coptosapelta-derived homolog in G. ula and G. montanum within the Asian clade II (Fig. 2). Therefore, there were at least two independent transfers of ndhB from different eudicots to Asian Gnetum in the past. Taken together, our phylogenetic results provide solid evidence that multiple rounds of independent HGT events have significantly shaped the mitogenome complexity in Asian Gnetum.

Foreign genes in Asian Gnetum mitogenomes are non-functional

Among the 15 foreign mitochondrion-derived genes (Fig. 2), 12 coexist with functional native homologs and the rest are either alone or present with a pseudogenized native copy (e.g., rpl10 of G. ula). This prompts us to raise the question: Are foreign genes transcribable to functionally complement the lost native homologs?

Combining the mapped RNA reads from both strands covers more than 91% of the three mitogenomes we sequenced. Focusing on G. parvifolium and G. ula because their mitogenomes contain many foreign genes, we calculated transcripts per million (TPM) values for all genes in their mitogenomes. TPM values of the foreign genes range from 2.7 to 614.3 in G. parvifolium and 36.3 to 561.9 in G. ula (Additional File 1: Fig. S22). No C-to-U edited sites were found in the RNA reads mapped to these foreign genes. In contrast, native genes exhibit much higher TPM values, ranging between 3438.6 and 97,600.5 in G. parvifolium and 2619.8 to 142,232.2 in G. ula, a statistically significant difference (two-tailed Mann–Whitney U test, P < 0.01 for both G. parvifolium and G. ula). Given that >91% of the Gnetum mitogenomes are transcribable, some level of transcription is expected for non-functional loci. Coupled with the presence of premature termination codons and frame-shifting indels in these genes, we consider it likely that the foreign copies are not functional.

DNA-mediated HGT in Asian Gnetum mitochondria

It is worth noting that 8 of the 20 foreign genes contain introns, including ccmFc, nad1 exons 2‒3, nad1 exons 4‒5, nad5 exons 4‒5, ndhB exon 1 and intron, cox2, rps3, and ndhB intron and exon 2 (Figs. 1‒2; Additional File 1: Figs. S1‒2). The foreign nad1 exons 4‒5 gene contains foreign matR within its intron (Fig. 1). Their congeneric origin is phylogenetically supported (Fig. 2). Apparently, these eight foreign intron-containing genes and matR have originated from DNA-mediated transfer. The clustering of foreign genes is a hallmark of DNA-mediated transfer when all foreign genes within a cluster have the same phylogenetic origin [7]. Such traits are observed in three foreign gene clusters in G. parvifolium (i.e., nad1 exons 2‒3 and rps13; ccmFn and nad1 exons 4‒5; rps1 and ccmFc) and two in G. ula (i.e., psaA and psaB; nad5 exons 4‒5, rps7, and ndhB exon 1 and intron).

The prevalence of C-to-U RNA editing in seed plant mitochondria provides traceable information to distinguish retroprocessing from direct integration of genomic DNA [4]. Four foreign genes, ccmB, rpl10, nad6, and rps4, were examined because the former two lack introns and the latter two have different origins despite forming a gene cluster (Figs. 1 and 2). If integration was mediated by mRNA, we would expect to see thymidines at the conserved editing sites of foreign genes. However, we observe mostly unedited cytidines at these loci (Fig. 3), making the cDNA integration origin unlikely. Moreover, we find thymidines at positions 28 and 176 of the ccmB alignment in both Gnetum and Coptosapelta but cytidines in other Rubiaceae genera (Fig. 3). This finding reinforces the transfer of ccmB from Coptosapelta to Gnetum after losses of these two editing sites through genomic C-to-T substitutions in Coptosapelta.

Fig. 3
figure 3

Nucleotide states of RNA editing sites in the foreign genes. Cytidines in native genes are bolded if their transcripts are predicted to be C-to-U edited. Only conserved editing sites on the gene alignment are shown

Full size image

Ancient HGT residues in Asian Gnetum mitogenomes

The 20 foreign genes mentioned above are relatively young since they are either Asian clade I- or II-specific (Fig. 2). To investigate ancient HGT residues, we looked for angiosperm mitochondrion-derived sequences shared by the two Asian Gnetum clades. We identified 544 bases shared by G. gnemon and G. ula, 5280‒5304 by G. gnemon and G. parvifolium, and 11,594‒11,941 by all three species (Fig. 4a). These shared bases constitute 45 separate non-coding sequences in G. gnemon, varying from 61 to 1526 bp in length (Additional file 2: Table S2). Blast searches of these 45 sequences reveal that (1) they are not homologous to any available mitogenomic sequences from non-Gnetum gymnosperms and that (2) except for Gnetum itself, the best matches are from mitogenomes of diverse angiosperms, such as the ANITA group, monocots, and eudicots (Fig. 4b). Despite the absence of phylogenetic evidence, these 45 foreign sequences can be regarded as ancient HGT residues gained before the split of the two Asian clades.

Fig. 4
figure 4

Ancient angiosperm mitochondrion-derived sequences. a Quantitative Venn diagrams indicate the number of unique and shared angiosperm mitochondrion-derived sequences among G. gnemon, G. parvifolium, and G. ula. b Word clouds show the best matched taxa with the font size reflecting their frequencies in the blast searches of the 52 ancient angiosperm mitochondrion-derived sequences (also see Additional file 2: Table S2)

Full size image

Discussion

Previously, exploration of foreign genes in gymnosperm mitogenomes has mainly relied on PCR [2, 8, 10]. Despite easy management, PCR methods have limitations in evaluating foreign genes. First, it is difficult to design appropriate primers when the donors of targeted foreign genes are uncertain. Second, most foreign genes are pseudogenized and degraded over time, impeding the design of universal primers across a taxonomic range of interest. Third, the exact cellular compartments where foreign genes of PCR targets reside need verification. Thus, PCR-based methods can underestimate the foreign (especially non-coding) DNA content in a genome.

In this study, we sequenced and assembled the mitogenomes from five Gnetum species across two Asian clades. Our genome-scale survey reveals that Asian Gnetum has received tremendous amount of foreign DNA, leading to great variation in their mitogenome size. Despite highly variable, the numbers of foreign genes found in the Asian Gnetum clade II are unprecedented among the so far elucidated gymnosperm mitogenomes. This observation provides evidence that gymnosperm mitogenomes can carry a large number of exogenous DNA/genes, like some angiosperms, such as Amborella [5], Geranium [7], and several parasitic eudicots [6, 9, 11, 22].

All 20 foreign genes in the examined Gnetum mitogenomes are pseudogenized and expressed at significantly lower levels than native loci without RNA editing. Together with the presence of foreign introns and the clustering of foreign genes of the same origin, these observations indicate DNA-mediated transfer from diverse angiosperms to Asian Gnetum as a consequence of silent HGT [21]. We see no evidence of RNA-mediated HGT during the Gnetum mitogenome evolution. This agrees well with several earlier studies showing that transfer of foreign DNA fragments is overwhelmingly common among seed plant mitochondria [4, 5, 7, 11, 22]. Furthermore, we detected many ancient angiosperm mitochondrion-derived sequences shared by both Asia Gnetum clades but not by any non-Gnetum gymnosperms. Two alternative hypotheses can explain their origins: (1) they were originally ancient angiosperm mitochondrial DNA fragments independently inserted into the common ancestor’s mitogenomes of the two Asia Gnetum clades or (2) they were initially seed plant descendants but had been lost from all gymnosperms except Gnetum. The second hypothesis is unlikely because Gnetum mitogenomes have drastically elevated rates of nucleotide substitutions with frequent gene loss [34, 35].

Eleven of the 20 foreign genes we identified were phylogenetically inferred to be horizontally transferred from Coptosapelta to the ancestors of the Asian Gnetum clade II (Fig. 2). This inference is supported by the emergence of Coptosapelta during the Upper Cretaceous [36], much earlier than the split of the two Asian Gnetum clades dating around 10‒39 MYA [31]. Coptosapelta is found in the southeast of Asia where species of the Asian Gnetum clade II also grow. Such an overlapping biogeographic distribution creates opportunities for HGT between these phylogenetically remote seed plant lineages. Hence, the most parsimonious HGT paradigm is that a large piece of Coptosapelta mitochondrial DNA invaded the ancestral mitogenomes of the Asian Gnetum clade II. The subsequent mitogenomic rearrangement and species-specific degradation explain the patchy distribution of the Coptosapelta-derived genes in the Asian Gnetum clade II ([2]; this study). This co-transfer paradigm also holds for the acquisition of the Coptosapelta-derived rps7 and ndhB exon 1 and intron cluster via mitochondrion-to-mitochondrion HGT after intracellular plastid-to-mitochondrion transfer of the associated genes in Coptosapelta. Indeed, a strong affinity of Gnetum to Rubiaceae (to which Coptosapelta belongs) is revealed in the phylogenetic tree based on the flanking regions of this Coptosapelta-derived MTPT gene cluster and its mitochondrial homologs (Additional File 1: Fig. S23).

In contrast, several Malpighiales genera appear to be donors of seven other foreign genes also specific to the Asian Gnetum clade II. However, the Malpighiales ancestry of these foreign genes cannot be supported because the crown groups of Malpighiales diverged approximately 90 MYA [37]. It is also possible that the origin of these foreign genes was mis-inferred in our trees due to a biased taxon sampling resulting from under-representation of some angiosperm lineages in GenBank. Nonetheless, the discrete chromosomal locations coupled with distinct phylogenetic origins lead us to believe that these Malpighiales-derived genes initially came from independent HGT events, especially since they also include a foreign MTPT gene cluster psaA-psaB. Although the flanking region sequences proved uninformative (Additional File 1: Fig. S24), we found homologs of this psaA-psaB cluster in the mitogenomes of some Malpighiales species, such as Populus tremula (NC028096), Hevea pauciflora (NC080334), and Kandelia obovata (NC06922). Therefore, it is reasonable to conclude that integration of the Malpighiales-derived psaA-psaB cluster in Asian Gnetum has resulted from mitochondrion-to-mitochondrion HGT, resembling the majority of foreign MTPTs in angiosperms [33]. Unfortunately, we could not clarify the transfer route for the G. gnemon-specific foreign MTPT (i.e., ndhB intron and exon 2) because homologous sequences of this MTPT’s flanking region are not present in GenBank.

Why dose Asian Gnetum stand out for the unusually large amounts of foreign genes in their mitogenomes? Strikingly, some conifer mitogenomes are huge with lengthy intergenic regions, offering more spaces for foreign gene integration. The Larix sibirica mitogenome is 8‒20-fold larger than Gnetum, yet no foreign genes have been reported in this conifer species [38]. Insect-mediated HGT during pollination was previously suggested as a possible mechanism for uptake of angiosperm DNA in Gnetum mitogenomes [2]. However, illegitimate pollination is more likely when two plant species are closely related [25].

Rice et al. [5] proposed a “wounding-HGT model” and maintained that wounds facilitate capture of foreign mitochondria whose genomic DNA is then integrated into the recipient mitogenome via mitochondrial fusion. Cell-to-cell movement of mitochondria and the subsequent fusion also can be responsible for horizontal mitochondrial DNA transfer between grafted plants [39]. Formation of channels that allow intercellular exchange of organelles was observed in direct contact between callus cells generated from plant grafted junctions [40]. Gnetum is the only lianas among gymnosperms. Their stems and those of sympatric angiosperms are often closely entwined (Fig. 5). Due to frequent friction from the entwined stems between Gnetum and angiosperms, wounds can be easily formed, creating opportunities for interspecific cell-to-cell contact and exchange of mitochondria. The large-scale integration of angiosperm mitochondrial DNA in the Asian Gnetum clade II mitogenomes, including foreign MTPTs, can be explained by mitochondrial fusion after capture of foreign mitochondria. Vertical transfer of the integrated foreign DNA could be achieved if the transgenic cell is incorporated into meristems that later differentiate into reproductive tissues [25, 41]. Future work will have to clarify whether the ancestry of the Asian Gnetum clade II had encountered strong genetic bottlenecks to fix the integrated foreign DNA in its populations.

Fig. 5
figure 5

Photos showing closely twisted Gnetum and angiosperms. White, orange, and red arrows indicate Gnetum, angiosperms, and healed wounds, respectively

Full size image

Conclusions

Two decades after the discovery of the angiosperm-type nad1 intron 2 [2] in Gnetum, we provide the first evidence that horizontal transfer of massive amounts of angiosperm mitochondrial DNA has had a great influence on Asian Gnetum mitogenome evolution. The high prevalence of foreign DNA makes Asian Gnetum an excellent system for investigating HGT between gymnosperms and other seed plants. The detection of multiple HGT events suggests that there was an active mechanism allowing for the frequent transfer of mitochondrial DNA from diverse angiosperms to the Asian Gnetum ancestry. The DNA-mediated HGT of multiple mitochondrial genes suggests that interspecific mitochondrial fusions may account for the high prevalence of angiosperm mitochondrion-derived DNA in the Asian Gnetum mitogenomes, given that Gnetum and angiosperms are often closely entwined to facilitate direct cell-to-cell contact between species. Deciphering the mitogenomes from African and South American Gnetum and other species from the Asian Gnetum clade I will help reconstruct a more comprehensive HGT history in Gnetum.

Methods

Taxon sampling, DNA and RNA extraction, and sequencing

Fresh leaves were harvested from G. gnemon, G. parvifolium, and G. ula individuals grown in the Academia Sinica greenhouse. Genomic DNAs were extracted from harvested leaves using the CTAB method described in [42]. Short DNA fragments were removed using Short Fragment Eliminator Kits (Circulomics, MD), followed by library preparation for MinION sequencers (FLO-MIN106, Oxford Nanopore Technologies: ONT) based on the recommend protocol with native barcoding genomic DNAs (EXP-NBD104), 1D sequencing kit, and SQK-LSK109 ligation kit. Approximately 13‒15 million long reads were generated for each species. Over 4 million pairs of 150 bp pair-end (PE) reads per species were also obtained from an Illumina NovaSeq 6000 platform and DNA libraries constructed using Celero™ DNA-Seq Library Preparation Kits (TECAN, Switzerland). Genomic DNA of G. gnemon var. brunonianum and G. pendulum collected in the South China Botanical Garden of the Chinese Academy of Sciences was extracted using DNeasy Plant Pro and Plant Kits (QIAGEN, Germany) and sequenced on a HiSeq X Ten system to yield over 8 million pairs of 100 bp PE reads per species. The sampled specimens are deposited in the herbarium with voucher numbers shown within parentheses: G. gnemon (Chaw1605), G. parvifolium (Chaw1607), and G. ula (Chaw1608) in Academia Sinica; G. pendulum (YN2024001) and G. gnemon var. brunonianum (YN2024002) in the South China Botanical Garden of Chinese Academy of Science. Total RNAs were isolated from fresh young G. gnemon, G. parvifolium, and G. ula leaves using Plant Total RNA Purification Kits (GeneMark, Georgia). After DNase I treatment, the extracted RNA was subjected to rRNA depletion and strand-specific library construction using Ovation RNA-Seq Systems 1‒16 for Model Organisms‒Arabidopsis (NuGEN, CA). The RNA libraries were sequenced on a NovaSeq 6000 platform to produce approximately 4.1‒5.5 million pairs of 150 bp PE reads per species.

Mitogenome assembly and annotation

Before de novo genome assembly, ONT reads shorter than 5 kb were discarded to decrease the computational burden. We used the Unicycler v0.5.0 hybrid assembler [43] that incorporates PE and ONT reads to assemble the G. gnemon and G. parvifolium gnomes. After short read assembly with a wide range of k-mer sizes (min_kmer_frac = 0.6; kmer_count = 5), Unicycler evaluated each of these assembly graphs and then used long reads to build bridges. The nature of multiple genome copies per cell enables distinguishing mitochondrial scaffolds from most of the nuclear ones in genome skimming data [44]. We set the options “min_fasta_length = 2000” and “--spades_options --cov-cutoff 2” to remove potential nuclear contigs and to reduce computational complexity. Another hybrid assembler, hybridSPAdes [32], was adopted for G. ula because system errors repeatedly occurred when application of Unicycler to this species. PE reads from G. gnemon var. brunonianum and G. pendulum were assembled using SPAdes v3.13 [45] with options of “careful” and k-mer sizes 21, 33, 55, 77, and 89. Mitochondrial chromosomes/scaffolds were searched by blast against protein-coding genes and rRNAs retrieved from Cycas (NC010303) and Ginkgo (NC027976) mitogenomes under an expected threshold = 10-10. The identified mitochondrial chromosomes/scaffolds were polished using their associated PE reads in Pilon v1.24 [46] with two to three rounds of iterations. Genome annotations were conducted in Geneious Prime [47] with the Ginkgo mitogenome as the reference. tRNAs were predicted using tRNAscan-SE v2.0 [48].

Exploration of MTPT and foreign sequences

To explore MTPT and foreign sequences, mitochondrial chromosomes/scaffolds were clipped into 1-kb non-overlapping fragments using the “getfasta” program implemented in Bedtools v2.31.0 [49]. These 1-kb fragments were subsequently megablasted against the NCBI nr/nt database with the parameters: max targets = 100, max E-value = 10-10, and word size = 28. Queries were recognized as native MTPTs if they matched published Gnetum plastomes well. MTPTs were considered foreign if they matched only non-gymnosperm plastomes. We considered other mitochondrial loci foreign if the sequences they matched were not included in mitogenomes of non-Gnetum gymnosperms. To identify shared foreign loci in the Gnetum mitogenomes, we first searched syntenic regions among G. gnemon, G. parvifolium, and G. ula using MUMmer 3.0 [50], and then compared foreign loci within syntenic regions between species.

Read mapping, RNA editing site detection, and TPM calculation

DNA read mapping analyses were performed using Bowtie2 v2.5.1 [51] with the default settings. Read depths were counted in Geneious Prime, followed by transformation into log scales: Log (coverage + 1) / log (maximum coverage + 1). For detection of RNA editing sites and calculation of gene expression levels, strand-specific RNA reads were mapped to the assembled Gnetum mitogenomes using TopHat v2.1.1 [52] with the parameters: library-type = fr-secondstrand, read-mismatches = 15, read-gap-length = 0, and read-edit-dist = 15. Samtools 1.9 [53] was used to filter, sort, and combine the mapped RNA reads into BAM files. These files were then used to identify RNA editing sites using the method described in Wu and Chaw [35]. Transcripts generated from native and foreign genes were also counted in Geneious Prime and then normalized into TPM values. We used PREPACT v3.12 [54] to predict RNA editing sites in angiosperm mitochondrial genes homologous to the foreign genes identified in Gnetum with the options of “Type of analyses = BLASTX prediction” and “Protein reference = Organelle: mitochondrion”, such that all angiosperm references were selected.

Identification of HGT origins for foreign genes

To trace HGT origins, foreign genes were extracted from the identified foreign sequences. These foreign genes were aligned with their native homologs, including MTPTs, using MAFFT v7.490 [55], followed by manual adjustments. Gene trees were estimated using IQ-TREE v2.2.0 [56] with 5000 ultrafast bootstraps and the “MFP” option of automatic model selection. Possible HGT donors were recognized if the examined foreign genes were placed as a sister to a particular angiosperm clade with >50% bootstrap support. Trees were visualized in Figtree v1.4.4 [57].

Mitogenome visualization

Mitogenome maps were drawn using Circos v0.67 [58].