Independent degradation in genes of the plastid ndh gene family in species of the orchid genus Cymbidium (Orchidaceae; Epidendroideae)

In this paper, we compare ndh genes in the plastid genome of many Cymbidium species and three closely related taxa in Orchidaceae looking for evidence of ndh gene degradation. Among the 11 ndh genes, there were frequently large deletions in directly repeated or AT-rich regions. Variation in these degraded ndh genes occurs between individual plants, apparently at population levels in these Cymbidium species. It is likely that ndh gene transfers from the plastome to mitochondrial genome (chondriome) occurred independently in Orchidaceae and that ndh genes in the chondriome were also relatively recently transferred between distantly related species in Orchidaceae. Four variants of the ycf1-rpl32 region, which normally includes the ndhF genes in the plastome, were identified, and some Cymbidium species contained at least two copies of that region in their organellar genomes. The four ycf1-rpl32 variants seem to have a clear pattern of close relationships. Patterns of ndh degradation between closely related taxa and translocation of ndh genes to the chondriome in Cymbidium suggest that there have been multiple bidirectional intracellular gene transfers between two organellar genomes, which have produced different levels of ndh gene degradation among even closely related species.


Introduction
The first two plastid genomes (plastomes) sequenced included the entire ndh 11-gene family, which is analogous to complex I in the mitochondrial genome (chondriome) [1,2]. Subsequently, the function of the ndh plastome genes has been described in many studies. The Ndh complex codes for an NADH-specific dehydrogenase with low levels of expression [3,4], and the family is involved in cyclic electron flow and chlororespiration [4,5]. Recently, Yamori et al. [6] investigated the function of Ndh complex in low light. However, in spite of this role, the Ndh complex is dispensable for plant growth under optimal conditions [4], and an alternative cyclic electron transport pathway has been reported [7,8]. Therefore that ndh-lacking species in which at least one of ndh genes is non-functional may be able to use the alternative pathway for cyclic electron transport [9]. When the loss of the 11 ndh genes in Pinus thunbergii was reported [10], this striking feature was considered unique because ndhF had been found to be present in all other major sequenced vascular plant clades [11]. However, losses of ndh gene function have subsequently been reported in various clades of land plants. In bryophytes, the 11 ndh genes in the parasitic liverwort, Aneura mirabilis (synonym, Cryptothallis mirabilis), were partially or completely deleted [12], and ndhF of the leafy liverwort, Ptilidium pulcherrimum, was found to be a pseudogene [13]. In the fern clade, some leptosporangiate ferns had internal stop codons in ndh genes, but this seemed to be related RNA editing [14][15][16]. In gymnosperms, ndh gene losses have been reported in Pinaceae [10,[17][18][19] and Gnetales [20,21]. Parasitic angiosperms have lost the function of ndh genes as well as other photosynthesis-related genes [22][23][24][25], but some autotrophs also lack the ndh gene [26][27][28][29].
Degradation of ndh in Orchidaceae is noteworthy from the perspective of the 11 ndh genes found in 743 angiosperm plastomes (Fig 1) (S1 Table). All 11 ndh genes had been coded into four classes [30], and different coding ndh gene patterns have been in each order based on the extent to which ndh genes were variously degraded. Reported plastome sequences of rosids comprise 32.5% of the 743 plastid genomes, but only the rosid order Geraniales have degraded ndh genes [28,31]. With the exception of internal stop codons caused by 1-bp insertions or deletions (indels) in Asterales [32,33], ndh gene degradation in the asterids is restricted to parasitic taxa in Lamiales and Solanales [23,24,[34][35][36][37]. In monocots, the number of sequenced Poales is 21.4% of angiosperms, but only ndhA in some species seems to be a pseudogene caused by short indels.
Degradation of ndh genes among genera in Orchidaceae seems to be independent [9,30], but the scale of variation among closely related species level has yet to be investigated. The plastomes of the two Phalaenopsis species sequenced had similar ndh gene degradation patterns [41], which was observed as well as in the plastome of Phalaenopsis hybrids [30]. Most ndh genes in the eight species of Cymbidium sequenced were full-length, although some of them had frame-shift mutations that render them functionless [43]. Degradation of ndh in subtribe Oncidiinae varied slightly among genera [40]. However, 15 of the reported Oncidiinae were complex hybrids, and it was difficult to determine the ancestral character status of ndh gene degradation among these. Comparative analysis of ten species of coralroot orchids [48] and two species of a distantly related genus, Epipogium [49], all of which are holomycoheterotrophic, indicated ndh genes had become pseudogenes or were completely deleted in each of their common ancestors. However, recently submitted plastome sequences of Cymbidium in GenBank showed different ndh gene deletions among individuals within species. Therefore, it seems that ndh genes in Cymbidium may be being actively degraded and that an investigation of ndh gene status will help us understand broader patterns of ndh gene degradation in Orchidaceae.
In this paper, 11 ndh loci among 23 Cymbidium species including hybrids and three closely related taxa are analyzed for ndh gene degradation. Except for ndhF, we tried to investigate all ndh genes. The ndhF gene was completely deleted in some species in Cymbidium or contained a number of internal homopolymer regions, which we assume indicates non-functional genes. Therefore, we confirmed only the presence of ndhF in each plastome. Additionally, we analyzed NGS data to determine if ndh genes had been translocated to the chondriome [9] because we found multiple copies of some ndh genes in Cymbidium species in our investigations.

Results
Ten ndh loci among 23 Cymbidium species and three closely related taxa Four regions (ndhB, ndhJ-K-C, ndhD, ndhE-G-I-A-H) that included ten ndh genes from 23 Cymbidium species and three outgroups were amplified by PCR and sequenced (Table 1). However, some intergenic or coding regions could not be sequenced because they contained homopolymers and polyA/T-polyG/C or problematic secondary structure (inverted repeats). To identify indels in ten ndh genes among 23 Cymbidium species and three closely related taxa, the fully intact (functional) ndh genes of Masdevallia coccinea were used as reference sequence.
Except for C. tigrinum in which only half of exon1 is present and C. mastersii in which the 5 0 region failed to produce sequence, all Cymbidium species were documented to contain a full-length ndhB gene (S1A Fig). A 1-bp insertion at 37 bp downstream of the 5 0 end of ndhB results in a frame-shift mutation in ndhB in reported plastome sequences of Cymbidium, and this was also identified in all Cymbidium species studied here and the closely related Acriopsis and Thecostele accessions (subtribe Cymbidiinae) [51]. A large deletion including exon1, intron and exon2 was detected in ndhB of Acriopsis.
The ndhJ-K-C region was more variable than that of ndhB (S1B Fig). A 12-bp direct repeat was distributed 63 bp downstream of the 5 0 end of ndhC and 69~82 bp downstream of 3 0 end of ndhJ in most Cymbidium species. However, the sequence between the direct repeats was only deleted in C. goeringii, a result that conflicts with the complete plastome sequence of same species in GenBank (NC_028524), but this was based on a different individual of that species. Deletions caused by direct repeat sequences were also found in the 5 0 region of ndhJ in three Cymbidium species (C. floribundum, C. erythrostylum, and C. tigrinum), Acriopsis and Thecostele. Unexpectedly, two copies of ndhJ-K-C region were detected in C. atropurpureum. Type I was similar to other Cymbidium sequences, whereas type II contained a 87-bp insertion 39 bp downstream of the 5 0 end of ndhK. This 87 bp insertion is not present in any other of the 743 angiosperm plastomes in GenBank. Only C. madidum, C. finlaysonianum and the mt copy of ndhK in all Cymbidium species contained sequences of this same type.
The ndhD regions of Cymbidium were relatively conserved (S2A Fig). Large deletions were located in the 3 0 region of the gene. Some of these occurred between direct repeat sequences.
The largest deletion of ndh genes in Cymbidium was identified in the ndhE-G-I-A-H region (S2B Fig), the end points of which were commonly located in an extremely AT-rich region. In particular, deletion of ndhA exon1 and ndhH in C. goeringii corresponded to those occurring in the plastomes of C. ensifolium, C. kanran, C. lancifolium and C. macrorhizon even though the plastome of different individuals of C. ensifolium (NC_028525) and C. goeringii (NC_028524) contained full length pt-ndhA and ndhH.

Different types of the ycf1-rpl32 region in Cymbidium
The ycf1-rpl32 region of the sequenced plastomes of Cymbidium was subdivided into two different types in comparison with that of M. coccinea (Fig 2A). Type A ycf1-rpl32 was similar to the reference, whereas 420 bp of 3 0 region of ndhF was replaced with ycf1 sequence in type B ycf1-rpl32.
Cymbidium dayanum in subg. Cymbidium and nine species of subg. Cyperorchis contained type A ycf1-rpl32, which was highly conserved (Fig 2B). In contrast to type A ycf1-rpl32, type B ycf1-rpl32 of Cymbidium had number of indels in 3 0 region of ndhF ( Fig 2C). The type B ycf1-rpl32 of C. sinense sequenced in this paper was only 87% similar to that of C. sinense plastome owing to many indels. Type B ycf1-rpl32 was also found in three Cymbidium species in which plastid ndhF was completely deleted. In comparison to type B ycf1-rpl32, type C ycf1-rpl32 had large deletion in the 3 0 region of ndhF, and the end point of the deletion corresponded to the end point of the replaced ycf1 region (Fig 2C and 2D).
Type D ycf1-rpl32 in which ndhF was completely deleted was found in half of the Cymbidium species examined and the three closely related taxa with a high level of similarity among them ( Fig 2E). In comparison with type A ycf1-rpl32, two large deletions occurred in type D ycf1-rpl32; one was the complete deletion of ndhF and the other was an intergenic deletion between ndhF and rpl32.

Multiple copies of ndh genes in Orchidaceae
The 38 ndh partial sequences were detected from 15 contigs using four sets of NGS data from Orchidaceae ( Table 2). With the exception of one contig in C. lancifolium, the ratio of the depth of mt-ndh genes to the depth of plastome in 15 contigs was 5.5~14.5, and BLAST results confirmed that they were derived from the chondriome.
The contig that contained the ndhJ-K-C region in C. lancifolium was present in relatively lower depth and did not contain a mitochondrial region, but there were only two SNPs and one indel that differed among the mt-ndhJ-K-C region in C. lancifolium and C. macrorhizon. Consequently, we concluded all 16 contigs have been translocated from the plastome to the chondriome.
Two Cymbidium species in section Pachyrhizanthe. All 11 ndh genes have been found in the chondriome of two Cymbidium species, and most of them do not differ in these two species. The mt-ndhB gene lacked 44 bp of exon1 and contained a 132-bp deletion in exon2 ( Fig  3A). Similarities of the ndhB genes in the same genome among different species were 99.0 and 99.5%. However, those in the genomes of two accessions of same species were only 91.1 and 91.9% similar. Mt-ndhJ and ndhK contained a large deletion and insertion, respectively ( Fig  3B). The length variation of insertion in mt-ndhK between two Cymbidium species was due to tandem repeats of 28 bp sequence. Even though plastid ndhF was completely deleted, two copies of mt-ndhF were found in two Cymbidium species (Fig 3C). One copy of these was similar to ndhF in type B ycf1-rpl32, and the other was similar to ndhF in type C ycf1-rpl32. In comparison with their plastome sequence, mt-ndhD was truncated and mt-ndhA and ndhH genes were almost full length ( Fig 3D). In addition, another mt-ndhD (773 bp) was found in C. lancifolium. Dendrobium catenatum. The nine mt-ndh genes were found in four large contigs ( Table 2). Among them, three contigs could form subgenomic circles [52]. Because a number of pt-ndh genes of D. catenatum have been deleted [53], we used a completely intact set of ptndh genes as a reference sequence, in this case Sobralia.
The region of mt-ndhJ-K-C was similar to the reference sequence in length with the exception of a large deletion in mt-ndhK, whereas pt-ndhK and ndhC were completely absent ( Fig  4A). Mt-ndhF was longer than pt-ndhF, but both of them were highly truncated ( Fig 4B). The regions between 194 bp downstream of rpl32 and 317 bp downstream of the 5 0 end of ndhG were relatively conserved between pt-and mt-ndh genes, but the 3 0 region of ndhD had a large deletion in both genomes (Fig 4C). The regions with pt-ndhI and ndhA exon2 were deleted [53], whereas these genes were found in chondriome but with a large inversion upstream of 5 0 end of ndhG and downstream of the 5 0 end of ndhA (Fig 4D). The mt-ndhJ-K-C region contained a large deletion in and a large insertion in mt-ndhK. The length variation between two large insertions in mt-ndhK was caused by 28 bp tandem repeats. C) In contrast with deleted plastid ndhF, two types of mt-ndhF were found in both species. D) Both ndhA exon1 and ndhH were deleted in the plastome, whereas they were found in the chondriome of both species. https://doi.org/10.1371/journal.pone.0187318.g003 Epipogium aphyllum. We found mt-ndhI and ndhA genes in achlorophyllous (holomycotrophic) E. aphyllum, but all pt-ndh genes in this species were completely deleted [49]. Unexpectedly, there was also an inversion mutation like that found in mt-ndhI-A of D. catenatum (Fig 4E).

Phylogenetic relationships between pt-and mt-ndh genes in Orchidaceae
In most ndh-gene trees (S3 Fig), the mt-ndh genes of Cymbidium formed a clade. It was noteworthy that the clustering of mt-ndhD, ndhE and ndhG from the NGS data and direct sequencing was strongly supported. However, the mt-ndhH genes of section Pachyrhizanthe formed a clade with the pt-ndhH genes of previously sequenced Cymbidium plastomes [43], whereas all pt-ndhH genes of Cymbidium sequenced in this study formed a strongly supported cluster. In addition, the ndhJ, ndhK and ndhC genes of C. madidum, C. finlaysonianum and type II C. catenatum has a mt-ndhI-A exon2 region that is inverted relative to the reference, whereas this region was completely deleted in the plastome. E) Plastome of E. aphyllum has completely deleted all 11 ndh genes, whereas its chondriome has retained an ndhI-A region; there was an inversion between 3 0 region of ndhI and upstream of 5 0 end ndhA exon2.
https://doi.org/10.1371/journal.pone.0187318.g004 atropurpureum formed a cluster with mt-ndhJ, ndhK and ndhC of section Pachyrhizanthe. The second copy of mt-ndhD in C. lancifolium clustered with the mt-ndhD of Oncidium, and they formed a strongly supported group with other orchid mt-ndhD genes. The clustering of the pt-ndhG of C. ensifolium (NC_028525) and mt-ndhG from other species of Cymbidium was strongly supported, whereas another pt-ndhG from C. ensifolium (KU179434) formed a group with pt-ndhG in Cymbidium.
Multiple copies of the mt-ndh genes from Erycina pusilla (subtribe Oncidiinae) formed a unique cluster with the exception of one copy of mt-ndhD (246 bp), which was relatively shorter than other mt-ndhD genes (480~1078 bp) in E. pusilla. Furthermore, these mt-ndh genes clustered with their pt-counterparts with the exception of pt-ndhA, ndhI and ndhE, which were truncated or missing from the plastome of E. pusilla.
The mt-ndhA, ndhD, ndhE, ndhG, ndhH, ndhI and ndhJ genes in Masdevallia picturata were most closely related to the pt-ndh genes of Masdevallia, and almost all mt-ndh genes in Paphiopedilum also formed clusters with the pt-ndh genes of these species.

Patterns of ndh degradation in Cymbidium
Function of ndh genes has been independently lost in some orchid clades [9,30]. With the exception of the directly sequenced plastomes of Goodyera, ndh-missing/non-intact species and ndh-intact species have not been so far found in same genus of Orchidaceae [41,43,48], in contrast to the situation in Erodium [27, 28]. Therefore, loss of function in the ndh complex seems to have occurred in the common ancestor of the ndh-missing/non-intact species within those genera rather than independently at the species level. The situation for ndhB in Cymbidium indirectly supports this scenario. With the exception of inverted repeat (IR)-deleted species, this gene is normally located in the IR, which position seems to play a role in its structural stability [54]. Substitution rates of the IR are also lower than those of single copy regions [55][56][57][58][59]. Therefore, ndhB is structurally more conserved than other ndh genes that are located in the single copy regions. In Cymbidium species, a 1-bp insertion at 37 bp downstream of the 5 0 end of ndhB has been found in all species with the exception of the species that contain a truncated copy of ndhB. Therefore, at least, the ancestor of all Cymbidium species is likely to have lacked a functional ndh complex.
The first sequenced plastomes of Cymbidium [43] and directly uploaded sequences (NC_028525 and NC_028524) contained full-length ndh genes even though most of them were pseudogenes due to frameshift mutations. However, recently a sequenced plastome of Cymbidium lacked pt-ndhF, ndhH and ndhA exon1. As a result, there are two plastomes of C. ensifolium with different ndh gene content. With the exception of technical errors (misidentification at the time of collection or laboratory errors), which is difficult to determine in this study, our results support the hypothesis that Cymbidium species have undergone dynamic and recent ndh gene degradation. Because the common ancestor of all Cymbidium species seems to have lacked ndh function, many different substitutions and indels may have accumulated in the various species due to relaxed selection. The large deletions that caused ndh degradation should be shared between closely related taxa if ndh gene degradation had occurred in an ancestral pseudogene further in the past. However, most of the large deletions detected are unique in each accession.
In addition, one of the main factors involved in ndh gene degradation is likely to be intracellular recombination. A number of deletions have been found between direct repeat sequences or extremely AT-rich (homopolymer) regions. These patterns have been known to relate to intramolecular recombination [60,61] and illegitimate recombination [62], respectively. These results suggest that the plastomes in Cymbidium species have undergone independent ndh gene degradation, probably after they speciated. The different levels of plastid ndh gene degradations in different individuals of C. ensifolium and C. goeringii also support a hypothesis of recent ndh gene degradation in Cymbidium.
However, we cannot suggest a clear explanation for why there appears to be a recent burst in this activity in the extant species of Cymbidium. In contrast, the ndh-lacking genera of photosynthetic orchids, i.e. Phalaenopsis [41], Oncidium, Paphiopedilum [30], Dendrobium and Bletilla, have retained similar ndh gene degradation patterns among their species. In general, with the exception of extremely reduced mycoheterotrophic orchids [45,49], a number of pseudogenes have been retained in the plastomes of Orchidaceae [46][47][48]. In particular, the closely related green and non-green coralroot orchids (Corallorhiza), which have lost some ndh genes, are similar in plastid genome size [48]. Therefore, the plastome of Orchidaceae may be prone to retain its size due to some selective constraints.
Barrett et al. [47] hypothesised that non-functional genes in mycoheterotrophic plants may have undergone point mutations and frame-shift mutations under relaxed selective pressure over time, and large deletions occur rarely after purifying selection on non-functional genes ceases. Unlike other genera in Orchidaceae, the most recent common ancestor (MRCA) of Cymbidium seems to have been under selective genome size constraint even though ndh function had been lost. However, structural mutations like bidirectional homologous recombination between the two organellar genomes or gene conversion in ndhF after splitting of populations or speciation might have led the plastome to be under relaxed selective constraints. As a result, it is likely that dynamic ndh gene degradation has occurred among Cymbidium species, perhaps even among populations.

Diverse ndhF genes result from gene conversion and indels
The first five Cymbidium species studied previously had full-length plastid ndhF genes [43], but ndhF deletions occurred in four recently submitted sequences. As we reported for the ndhA-H region, the deleted pt-ndhF genes of C. lancifolium and C. macrorhizon were transferred to chondriome (Fig 3C). As a result, C. sinense contains type B ycf1-rpl32 in its plastome and type D ycf1-rpl32 in its chondriome, whereas C. kanran, C. ensifolium, C. macrorhizon and C. lancifolium contain type D ycf1-rpl32 in their plastomes and type B ycf1-rpl32 in their chondriomes. Other Cymbidium species also contain different types of ycf1-rpl32 in their organellar DNAs, but we do not know in which genomes these are located. Species that have the same type of the ycf1-rpl32 region are not related to each other (i.e. they belong to different clades in the Cymbidium phylogenetic tree). Nevertheless, four types of the ycf1-rpl32 region seem to be related each other.
Type A ycf1-rpl32 is similar to that of other Orchidaceae, whereas 420 bp of the 3'region of ndhF in type B ycf1-rpl32 is similar to the ycf1 region and contained a number of indels. The ndhF sequence near IR B /SSC was replaced with ycf1 near SSC/IR A . This replacement might result from IR expansion via gene conversion [63] (S4 Fig). First, recombination was initiated within the IR. Then, a Holliday junction on the IR was moved to SSC, creating heteroduplex DNAs. These heteroduplex DNAs were repaired using the complementary strand as the model. Finally, base substitutions and indels occurred in the ycf1 like region in ndhF. Significantly, an end point for deletion of ndhF in Acriopsis and Thecostele was identical to that of a ycf1-like region in ndhF of C. tortisepalum (Fig 2C and 2D). Therefore, it is possible that type C ycf1-rpl32 was derived from type B ycf1-rpl32 due to deletion of a chimeric region.
Kim et al.
[30] described the important role of ndhF in the instability of the IR/SSC junction in Orchidaceae. Retention of full-length ndhF seems to be related to the selective constraints that maintain the IR/SSC boundary. The ndhF of the type B ycf1-rpl32 region is similar to ndhF in type A ycf1-rpl32 in length, but in its content is similar to the truncated version of ndhF due to the replacement of 3'end region of ndhF. As a result, it seems that gene conversion leads to relaxed selective constraint of the IR/SSC junction, after which truncated ndhF versions in type B and type C ycf1-rpl32 may be followed by ndhF deletion as in type D ycf1-rpl32.

Intracellular gene transfers between organellar DNA
Chang et al. [39] confirmed the in-frame sequences of ndhA, ndhF and ndhH that are completely deleted in the plastome of Phalaenopsis aphrodite and suggested that they were transferred to nuclear genome. However, in the recently published whole genome of P. equestris [64], it was shown that there was also no intact ndh gene [30]. Subsequently, mt-ndh genes were found in many unrelated clades of Orchidaceae [9], and we also found mt-ndh genes in several distantly related species. Therefore, intact ndh genes that are deleted from the plastome of Phalaenopsis are likely to be found in its chondriome. However, this is not surprising because such transfers are known to occur widely in seed plants [65][66][67][68].
To evaluate relationships between plastid and mitochondrial copies of ndh genes in Orchidaceae, we constructed gene trees (S3 Fig), which gave us information about ndh gene transfer, although some nodes are not well resolved. First, it is likely that the transfers of ndh genes from plastome to chondriome have usually occurred in the MRCA of the species in each genus. As there is limited ndh gene information at the species level, especially for mt-ndh genes, it is impossible to infer a time for these transfers. However, many of the pt-and mt-ndh genes from a given genus cluster together. For instance, mt-ndhC, ndhD, ndhG, ndhH and ndhJ of Erycina pusilla (subtribe Oncidiinae) were transferred after Erycina diverged from its common ancestor with Oncidium (subtribe Oncidiinae). The mt-ndh genes in Masdevallia picturata (subtribe Laeliinae, subfamily Epidendroideae) and Paphiopedium (subfamily Cypripedioideae) were also sister to pt-ndh genes of species within each genus, respectively.
In the ndh tree of Cymbidium, most mt-ndh genes are distantly located from their pt-ndh counterparts, and the entire mt-ndhD-E-G-I-A-H region can be assembled from NGS data for two species, which we confirmed by PCR of the mt-ndhD-E-G region in six Cymbidium species. These mt-ndh genes clustered uniquely with strong support. Although the combined ndh gene tree for ten species of Cymbidium had a different topology from that of combined ITS +matK [69], it is clear that the transfer of the ndh genes in the single-copy region dates back at least to the common ancestor of these Cymbidium species.
Secondly, transfers between the chondriome of photosynthetic orchids have occurred more than once. The mt-ndhD genes of Cymbidium (Cymbidiinae) and Erycina (Oncidiinae) were divided into two groups. The mt-ndhD genes (from mt-ndhD-E-G region) of Cymbidium and Erycina clustered with mt-ndhD genes in same genus. However, another copy of mt-ndhD gene in C. lancifolium and Erycina formed a strongly supported cluster with the mt-ndhD genes from Oncidesa Gower Ramsey (a complex hybrid between species in Oncidium and Gomesa, most likely with the plastid genome of the former) and a member of another subfamily Goodyera fumata (tribe Cranichidae, subfamily Orchidoideae). These four mt-ndhD genes clustered with mt-ndhD gene of D. catenatum (tribe Malaxidae, subfamily Epidendroidae), to which the plastid ndhD of Dendrobium was an outlier with moderate support. It is therefore likely that mt-ndhD of Dendrobium has been directly transferred independently to the other four species [70]. In addition, mt-ndhE of Oncidesa Gower Ramsey (subfamily Epidendroideae) and V. planfolia (subfamily Vanilloideae) are identical. Although the substitution rate of the chondriome is slower than in plastid DNA [52], it is unlikely that mt-ndhE of two species originated in their common ancestor because of the long time, before the end of the Cretaceous, since the members of these orchid subfamilies diverged [70]. Consequently, our results suggest a recent transfer of mt-ndh gene between distantly related taxa in Orchidaceae. Horizontal gene transfer (HGT) between photosynthetic orchids has not been reported so far. However, multiple mt-genes from different lineages have been transferred into the chondriome of Geraniaceae [71]. Because there is little information of the chondriome of Orchidaceae, it is difficult to figure out how and when this HGT might have occurred.

Unidirectional vs bidirectional IGT
The most remarkable feature of ndh genes in Cymbidium is the presence of multiple copies in their organellar genomes. For example, C. sinense has a type B ycf1-rpl32 in its plastome and type D ycf1-rpl32 in its chondriome, whereas C. kanran, C. ensifolium, C. macrorhizon and C. lancifolium have type D ycf1-rpl32 in their plastomes and type B ycf1-rpl32 in their mt-DNA. Some species also have other types, e.g. ycf1-rpl32 types A and D. It is highly perplexing that Cymbidium species can have different types of the ycf1-rpl32 region in one genome (plastome or chondriome) and the same type of ycf1-rpl32 region in different genomes. We have two hypotheses that could explain this phenomenon: C. sinense and C. macrorhizon represent nonfunctional ndhF (type A, B and C) and completely ndhF-deleted species (type D), respectively.
The first hypothesis is unidirectional transfer (Fig 5A). The ycf1-rpl32 region containing ndhF (ancestral type) was transferred to its chondriome. Subsequently, the mt-ndhF (C. sinense) and pt-ndhF (C. macrorhizon) were independently deleted. The second hypothesis is bidirectional transfer (Fig 5B). In this scenario, the ycf1-rpl32 region containing plastid ndhF was transferred to chondriome in the ancestor of Cymbidium and closely related genera of subtribe Cymbidiinae. After this transfer, the mt-ndhF copy was eliminated by gene rearrangements or gene deletion (as in C. sinense). Some species then underwent homologous recombination between the two ycf1-rpl32 copies in their plastomes and chondriomes (e.g. C. macrorhizon).
Type D ycf1-rpl32 among Cymbidium and three closely related taxa is highly conserved and shares two large deletions (Fig 2). The first hypothesis therefore must assume that two deletions in ycf1-rpl32 in both the plastome and chondriome have occurred at exactly the same position in all Cymbidium species and closely related taxa. However, the second hypothesis more easily explains this high level of similarity of the type D ycf1-rpl32 region among these genera because it originated in their common ancestor and mt-DNA has low substitution rate [52]. Similarly, because the plastid ndhH genes of previously sequenced Cymbidium plastomes have been re-transferred from chondriome, it is likely that they should cluster with the mt-ndhH genes of Cymbidium section Pachyrhizanthe.
In relative terms, the plastid genome is ten times more abundant that the mitochondrial genome of D. catenatum. This means that plastid regions are easier to amplify than mt-region even if the mt-region had exactly the same primer binding sites as the plastid copy. With the exception of C. atropurpureum, only one PCR product of the plastid ndhJ-K-C region was produced from all Cymbidium species and three related species studied here, and the plastid copies of ndhJ, ndhK and ndhC all clustered as expected with the exception C. finlaysonianum and C. madidum, making it likely that the ndhJ-K-C region of these two species was from their plastome.
In contrast, the type II ndhK found in C. atropurpureum was in mitochondrial genome of C. lancifolium and C. macrorhizon, so it is likely that type II ndhJ-K-C region of C. atropurpureum was located in the chondriome. Considering the phylogenetic relationship between C. atropurpureum and C. macrorhizon [69,72], the plastid ndhJ-K-C region might have been transferred to chondriome in the ancestor of Cymbidium. It also seems that the mt-ndhJ-K-C Two hypotheses for multiple copies of ycf1-rpl32 region in Cymbidium species. C. sinense illustrates the ndhF-containing types (type A, B, C), and C. macrorhizon the ndhF-deleted type (type D) in plastome. Green and red boxes indicate plastome and chondriome, respectively. A) The ycf1-rpl32 region containing the ndhF (ancestral type) was transferred to the chondriome, and then mt-ndhF (C. sinense) and plastid ndhF (C. macrorhizon) were independently deleted. B) The ycf1-rpl32 region containing ndhF were region of C. finlaysonianum and C. madidum was replaced with its plastid counterpart via recent homologous recombination. As a result, reimported plastid ndh genes are derived from the mt-ndh copies. The clustering of ndhG and ndhH among the two organellar genomes in some Cymbidium species also supports the hypothesis that their plastid ndh genes were relatively recently reimported from chondriome, probably via homologous recombination.

DNA extraction, sequencing, annotation
Fresh leaves of C. finlaysonianum, C. devonianum and Grammatophyllum speciosum were collected from the orchid collection at the Royal Botanic Gardens, Kew, and Ratcliffe Orchids, Ltd. (Hampshire, UK). Total DNA was extracted by the CTAB method [73]. Except for these three, all other genomic DNAs were taken from DNA Bank at the Royal Botanic Gardens, Kew (Table 3; http://apps.kew.org/dnabank/introduction.html). Vouchers are deposited in the spirit collection at the Royal Botanic Gardens, Kew.
Four regions including all 11 ndh genes (ndhB, ndhJ-K-C, ndhF, and ndhD-E-G-I-A-H) were assembled from the plastomes of Cymbidium [43]. Except for the ndhF region, primers were designed for three regions to sequence the full length of each region. In the ndhF region, there were a number of homopolymers near both ends. According to previous studies [43] and submitted sequences, this gene was completely deleted in some accessions of Cymbidium. Therefore, primers were designed just to confirm absence/presence of ndhF in each accession.
The four regions in each species sampled were amplified as follows: 95˚C 5min, (95˚C 30 sec-50~55˚C 30sec-65~72˚C 2min) × 31 cycles, 65~72˚C 2min using TaKaRa Premix Taq. PCR products were purified with Qiagen kits using the protocol of the manufacturer and were sequenced using Big-Dye chemistry on an ABI3730XL sequencer following the protocols of the manufacturer. All sequences were assembled by taxon and region using Geneious [74]. We annotated 11 ndh genes in each Cymbidium and three closely related taxa using complete sequenced plastome sequences in Orchidaceae.

Detecting ndh genes in chondriome
We used the data set from the Sequence Read Archive [75] and Cymbidium data generated by Kim (not published) to confirm if ndh genes had been translocated to the chondriome ( Table 2). We slightly modified the assembly method of Kim et al. [30] (Fig 6). Read ends were trimmed with an error probability limit of 0.01, and then reads under 40 bp and their counterpart reads were removed from data set. Each data set was aligned to the chondriome sequence of Phoenix dactylifera [65] under the medium sensitivity option in Geneious [74]. Then, the reads assembled with the reference were extracted and re-assembled using de novo assembly in Geneious with zero mismatch and gaps [74]. Several contigs were generated, and reads were re-aligned to them with zero mismatch and gaps with 25 iterations. We generated consensus contigs and aligned them by de novo assembly. The resulting contigs were re-used as reference sequences.
Whenever this process was repeated, the number of contigs was reduced, and lengths of resulting contigs extended, and this cycle was repeated until the contigs produced were not transferred to chondriome in the ancestor of the extant species of Cymbidium and closely related genera. Then, the mt-ndhF was removed from ycf1-rpl32 via gene rearrangements or gene deletion (C. sinense). In addition, homologous recombination between two ycf1-rpl32 regions of the plastome and chondriome occurred in some taxa or populations. As a result, ndhF was found not in the plastome but in the chondriome (e.g. in C. macrorhizon).
extended. To prevent misassembled contigs, only paired reads that matched and upstream or downstream sequence were used throughout the assembly process.
All contigs were investigated for similarity to chondriome sequences using BLAST [76]. Thereafter, mitochondrial contigs were annotated in comparison with their own plastomes. To distinguish the location of genes, genes in the plastome are prefixed with pt-and those in chondriome are prefixed with mt-. Information on mt-ndh genes is described in Table 2.

Phylogenetic analysis of ndh genes in both organellar genomes in Orchidaceae
The pt-and mt-ndh genes in Cymbidium and three closely related taxa were sequenced in this paper. In addition, 55 plastomes (S2 Table) and 38 chondriome sequences (S3 Table) were downloaded from NCBI. The three Phalaenopsis plastomes and Vanilla planifolia have a 768 3 bp inversion upstream of the 3 0 end of ndhB. Each ndh gene set was aligned via MAFFT alignment [77].
The ndhF gene was excluded from phylogenetic analysis because many species contained two types of ndhF genes, and it was difficult to determine where they were located in the organellar genomes. Introns in ndhA and ndhB were also removed from data set. The best-fit substitution model for each data set was determined using jModeltest2 [78]. Bayesian analysis was Supporting information S1 Fig. Alignment of ndh genes of 23 Cymbidium species and three closely related genera. Masdevallia coccinea ndh genes were used as reference. A) ndhB region. B) ndhJ-K-C region. Grey and black in the alignment indicate agreement and disagreement with the consensus sequence, respectively. Red in the alignment indicates ambiguous sites. Black bars at