Two Independent Plastid accD Transfers to the Nuclear Genome of Gnetum and Other Insights on Acetyl-CoA Carboxylase Evolution in Gymnosperms

Abstract Acetyl-CoA carboxylase (ACCase) is the key regulator of fatty acid biosynthesis. In most plants, ACCase exists in two locations (cytosol and plastids) and in two forms (homomeric and heteromeric). Heteromeric ACCase comprises four subunits, three of them (ACCA–C) are nuclear encoded (nr) and the fourth (ACCD) is usually plastid encoded. Homomeric ACCase is encoded by a single nr-gene (ACC). We investigated the ACCase gene evolution in gymnosperms by examining the transcriptomes of newly sequenced Gnetum ula, combined with 75 transcriptomes and 110 plastomes of other gymnosperms. AccD-coding sequences are elongated through the insertion of repetitive DNA in four out of five cupressophyte families (except Sciadopityaceae) and were functionally transferred to the nucleus of gnetophytes and Sciadopitys. We discovered that, among the three genera of gnetophytes, only Gnetum has two copies of nr-accD. Furthermore, using protoplast transient expression assays, we experimentally verified that the nr-accD precursor proteins in Gnetum and Sciadopitys can be delivered to the plastids. Of the two nr-accD copies of Gnetum, one dually targets plastids and mitochondria, whereas the other potentially targets plastoglobuli. The distinct transit peptides, gene architectures, and flanking sequences between the two Gnetum accDs suggest that they have independent origins. Our findings are the first account of two distinctly targeted nr-accDs of any green plants and the most comprehensive analyses of ACCase evolution in gymnosperms to date.


Introduction
Fatty acid biosynthesis in plants begins with the conversion of acetyl-CoA into malonyl-CoA, catalyzed by acetyl-CoA carboxylase (ACCase). Malonyl-CoA is an essential substrate for fatty acid formation (see reviews by Brown et al. [2010] and Huerlimann and Heimann [2013]). There are generally two types of ACCases in plants: the multisubunit heteromeric ACCase (ACCA-D) in plastids and the singlepolypeptide homomeric ACCase (ACC) in the cytosol (Sasaki and Nagano 2004). The plastid heteromeric ACCase is similar to the prokaryotic ACCase and believed to have originated from cyanobacteria, whereas the cytosol homomeric ACCase is the eukaryotic type ACCase (Nikolau and gymnosperms: gnetophytes ) and Sciadopitys Li et al. 2016). Homomeric ACCase is encoded by a single gene, ACC (Nikolau et al. 2003).
Although both plastid and cytosol ACCases convert acetyl-CoA into malonyl-CoA, the two compartments produce distinct end products. Malonyl-CoA is mainly converted into free fatty acids in the plastids (Brown et al. 2010), whereas in the cytosol it is used to synthesize flavonoids, anthocyanins, very long-chain fatty acids, and for the malonylation of D-amino acids and ethylene precursors (Sasaki and Nagano 2004). Thus, both ACCases play vital roles in plants. The loss of either of these genes is lethal, as was demonstrated in Arabidopsis (Baud et al. 2003) and tobacco (Kode et al. 2005). The only exceptions were reported in grasses (Poaceae) and Silene noctiflora, in which heteromeric ACCase is absent (supplementary fig. S1, Supplementary Material online; Konishi et al. 1996;Rockenbach et al. 2016), but a copy of homomeric ACCase has acquired plastid-targeting transit peptides (TPs) and replaced the function of heteromeric ACCase in plastids (Gornicki et al. 1997;Podkowinski et al. 2003;Rockenbach et al. 2016). In other cases (supplementary fig. S1, Supplementary Material online), both heteromeric and homomeric ACCase coexist in the plastids of Arabidopsis (Babiychuk et al. 2011) and Geraniaceae (Park et al. 2017).
To date, the only characterization of ACCase genes in gymnosperms is limited to the loss of plastid-encoded accD from gnetophytes ) and Sciadopitys Li et al. 2016). Whether the lost plastid accD has been transferred to the nuclear genomes of these two lineages has not been verified. As a previous study identified a partial ACCD transcript in Sciadopitys (Li et al. 2016), we hypothesized that its accD was transferred to the nucleus. However, no information is available for the gnetophytes; it is unknown whether they retained heteromeric ACCase or if it was replaced with homomeric ACCase (as in grasses). In addition, the plastid accD sequences of cupressophyte genera (Hirao et al. 2008;Yi et al. 2013;Li et al. 2018) and a Pinaceae genus, Tsuga (Sudianto et al. 2016), are much longer than their homologs from other gymnosperms. This elongation may have accelerated accD nucleotide substitution rates in both Tsuga (Sudianto et al. 2016) and cupressophytes (Li et al. 2018). Whether the accelerated rates of ACCD subunit in these taxa influence the other ACCase subunits remains uncertain.
Here, we report the characterization of genes and transcripts encoding heteromeric and homomeric ACCases in Gnetum ula and 75 additional gymnosperm transcriptomes sampled from public repository data. We also assess if both ACCase forms are present in gymnosperms. Furthermore, we investigate the evolutionary fate of accD across the five major groups of gymnosperms, including the accD that was lost from gnetophytes and Sciadopitys plastomes. Finally, we perform a protoplast transient expression assay to identify the localization target of nuclear-encoded accD from gnetophytes and Sciadopitys.

Isolation of Nucleic Acids and cDNA Synthesis
We collected G. ula (voucher Chaw 1569) and Sciadopitys verticillata (voucher Chaw 1496) from the living plants in the green house of Academia Sinica and Floriculture Experimenter Center, Taipei, respectively. Both specimens were deposited in the Herbarium of Academia Sinica, Taipei (HAST). Total RNA was extracted from both tissues following the protocol of . First strand cDNA was synthesized using RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fischer Scientific, Waltham) using random hexamer primers.

Sequence Retrieval
Coding regions of the plastid accD gene, including three nonvascular plants and 116 species of gymnosperms, were collected from GenBank and transcriptome data (supplementary table S1, Supplementary Material online). For the 76 transcriptomes analyzed in this study, we downloaded 74 assembled gymnosperm transcriptomes from oneKP (Matasci et al. 2014) and NCBI TSA databases (https://www.ncbi.nlm.nih. gov/genbank/tsa/; Last accessed on 10 November 2018), including 54 cupressophytes, 11 Pinaceae, 4 gnetophytes, 4 cycads, and 1 ginkgo (supplementary table S2, Supplementary Material online). We de novo assembled the transcriptomes of the two remaining Gnetum species-G. parvifolium and G. ula. Gnetum parvifolium reads were obtained from NCBI SRA (SRX1133345). RNA Sequencing, Transcriptome Assembly, and Identification of ACCase Transcripts RNA-Seq of G. ula was sequenced using an Illumina HiSeq2500 platform, yielding $100 million paired-end reads (2Â 100-bp length). Trimmomatic (Bolger et al. 2014) was used to remove low-quality reads and adapters from both the newly sequenced G. ula and retrieved G. parvifolium reads. Transcriptomes of both species were de novo assembled using SOAPdenovo-Trans 1.03 (Xie et al. 2014). TransDecoder 4.0.0 (Haas et al. 2013) was used to identify candidate coding regions. We used BlastP (E-value <1e À5 ) to identify the potential ACCase proteins from the gymnosperm transcriptome assemblies. Protein sequences of four heteromeric ACCase (accA, accB, accC, and accD) and two homologs of homomeric ACCase (ACC1 and ACC2) from Arabidopsis were used as queries.

Sequence Alignment and Phylogenetic Tree Reconstruction
ACCase protein sequences were aligned using MAFFT v7 (Katoh and Standley 2013) in Auto setting. Sequences with <50% total protein length and duplicated sequences with 100% similarity were removed from further analyses. ProtTest3 (Darriba et al. 2011) was used to find the best-fit model for phylogenetic tree reconstruction. Maximum likelihood (ML) trees were reconstructed using RAxML 8.2.10 (Stamatakis 2014) in the recommended JTTGAMMAI model with 1,000 bootstrap replications. Tajima's relative rate tests were performed using MEGA 7 (Kumar et al. 2016).

Protoplast Transient Expression of nr-ACCDs TPs
The putative TP sequences from G. ula (both nr-accD1 and nr-accD2) and Sciadopitys were amplified from cDNA of respective species with specific primers (see supplementary table S3, Supplementary Material online, for primers used). Amplified fragments were each cloned into the p326-GFP vector at XbaI and BamHI restriction sites. These constructs were transfected into Arabidopsis protoplasts following polyethylene glycolmediated transformation described by Lee and Hwang (2011). In brief, Arabidopsis protoplasts were isolated from the leaves of 3-week-old plants as previously described ) and then transformed with 10-20 mg of plasmid DNA from three GFP constructs described above. Plasmid DNA from p326-GFP vector was also used as a control to specify cytosol localization. Protoplasts and plasmids were incubated in the dark for 16 h at room temperature. The transformed protoplast images were captured using a Zeiss LSM780 ELYRA PS1 confocal microscope system.

Molecular Dating and Estimation of Absolute Nucleotide Substitution Rates
We sampled 34 gymnosperm genera in which all 5 ACCase sequences were available from the plastome and transcriptome data. Both synonymous (d S ) and nonsynonymous (d N ) substitution rates of these ACCase subunits were estimated using PAMLX (Xu and Yang 2013). We used the program CODEML with the following parameters: runmode ¼ 0, seqtype ¼ 1, CodonFreq ¼ 2, estFreq ¼ 0, model ¼ 1, and cleandata ¼ 1. We concatenated the accA-C sequences to calculate the d S and d N of nr-heteromeric ACCase. The constraint tree topology (supplementary fig. S2, Supplementary Material online) was reconstructed using 29 plastid-encoded photosynthetic genes. The relative divergence times were estimated using RelTime (Tamura et al. 2012) in MEGA 7.0. Seven estimated points from TimeTree (Kumar et al. 2017) were used as the calibration points (supplementary fig. S3, Supplementary Material online). The absolute synonymous (R S ) and nonsynonymous substitution rates (R N ) were derived by dividing the d N and d S branch lengths by their respective divergence times. We compared the absolute branch lengths of 1) accD versus ACC, 2) accD versus concatenated accA-C (nr-heteromeric ACCase), and 3) nr-heteromeric ACCase versus ACC to determine their correlations. Only terminal branches were considered in our analyses.

Gymnosperms Have Both Heteromeric and Homomeric ACCases
Transcripts encoding heteromeric and homomeric ACCases were identified in most gymnosperms. Transcripts of accA-D were present in all examined species, except Cephalotaxus and Sundacarpus, whose accB transcript was absent, possibly due to fragmented transcriptome assemblies ( fig. 1 ). All accA, accB, and accC transcripts carried plastid-targeting TPs (supplementary table S4, Supplementary Material online). Two to three unique transcripts of accA and accB were detected in ginkgo, cycads, gnetophytes, and cupressophytes ( fig. 1). In Cupressaceae, some duplicated accA and accB transcripts contained mitochondria-targeting TPs (supplementary table S4, Supplementary Material online). No duplicated accC transcript was detected, suggesting that there is only a single copy of accC in gymnosperms. We detected two copies of accD transcripts in Gnetum and one copy in Ephedra, Welwitschia, and Sciadopitys, though this gene is absent from their plastid genomes (plastomes). A single copy of homomeric ACCase (ACC) transcripts was present in all examined gymnosperms, FIG. 1.-The gymnosperms contain both heteromeric and homomeric ACCase transcripts. The ACCase complex is differentiated into two groups: heteromeric (including accA, accB, accC, and accD) and homomeric (ACC). The color-filled boxes indicate the location and completeness of the transcript. The number inside the boxes indicates the presence of two or three transcript copies in some species. Phylogenetic relationships between the groups were derived from Li et al. (2017). Sci., Sciadopityaceae; gne., gnetophytes; and gink., ginkgo. and no gymnosperms had plastid-targeting TPs (supplementary table S4, Supplementary Material online).
All sampled cupressophyte ACCDs bear insertions with little to no similarities to those of Pinaceae, ginkgo, and cycads (supplementary fig. S4, Supplementary Material online). These insertions contain various types of TRs that differ among closely related lineages. The number of TR types ranges from one to six for each cupressophyte genus. For example, 13 genera in the Araucariales order (including Araucariaceae and Podocarpaceae) share one TR type (see red font in table 1). Other than order-specific TRs, familyspecific TRs were also identified in Araucariaceae (3 TRs, blue), Podocarpaceae (1 TR, purple), and Cupressaceae (1 TR, green). In addition, genera of the subfamily Cupressoideae share one subfamily-specific TR (brown). No common TR type was found in Taxaceae.
The Plastid accDs of Gnetophytes and Sciadopitys Were Transferred to the Nucleus and Gnetum Retains Two Copies of nr-ACCD We detected ACCD transcripts in the transcriptome assemblies of gnetophytes and Sciadopitys, with predicted protein sequences ranging from 312 aa (Sciadopitys) to 370 aa (Gnetum) despite the absence of accD from their plastomes. This finding suggests that accD was transferred to the nuclear genomes of both lineages. The nuclear-encoded (nr) ACCD of Sciadopitys discovered in this study is 100 aa longer than previously reported by Li et al. (2016). We identified two homologous ACCD sequences in the three Gnetum transcriptomes (G. ula, G. montanum, and G. parvifolium; fig. 3). By contrast, the two Ephedra species sampled in this study (E. sinica and E. trifurca) and Welwitschia only contain one copy of ACCD ( fig. 3). The nr-ACCD of both Sciadopitys and gnetophytes contain 24-51 aa upstream of the usual start codon ( fig. 3a), which were predicted to be leader sequences that target plastids (supplementary table S5, Supplementary Material online). The nr-ACCD1 of Gnetum was also predicted to target mitochondria, albeit with lower scores. Because the identified nr-ACCD transcript of Welwitschia only contains partial TP, its localization target was not clearly predicted. TPs in gnetophytes were estimated to be 29 aa (nr-ACCD2 of Gnetum) to 50 aa (Ephedra) in length. Pairwise sequence identity of orthologous TPs is >90% among congeneric species. In contrast, the predicted TPs of nr-ACCD1 and nr-ACCD2 of Gnetum spp. share low sequence similarity (15.6-18.2%).
A phylogenetic tree inferred from 66 ACCD sequences places the newly identified nr-ACCD of Sciadopitys within the cupressophytes ( fig. 3b), consistent with the species tree. Using Tajima's relative rate test, we found that the ACCD sequences of cupressophytes evolve faster than those of other gymnosperms except Gnetum (supplementary table S6, Supplementary Material online). The two nr-ACCD sequences from each of the three studied Gnetum species form a clade with nr-ACCD sequences of Ephedra and Welwitschia ( fig. 3b). However, the two clades of Gnetum nr-ACCD were not sister to each other. The nr-ACCD2 clade of the three Gnetum species was placed as the outgroup to gnetophytes' nr-ACCD1 clade ( fig. 3b), suggesting that the common ancestor of gnetophytes had two nr-ACCDs.
By mapping the two nr-ACCD transcripts of G. ula to its draft genome, we discovered that the gene architectures of nr-accD1 and nr-accD2 are distinct ( fig. 4a). The former contains a single exon of 1,113 bp, whereas the latter contains two exons (36 and 1,023 bp) and one intron (170 bp) with a total length of 1,229 bp ( fig. 4a). In the nr-accD1 scaffold, two short fragments (1,170 and 247 bp) reside at 40-kb upstream and 4.5-kb downstream of the coding region, respectively, were identified as pseudogenes of plastome origins. Similarly, in nr-accD2 scaffold, a 547-bp fragment at 7-kb downstream of the stop codon is composed of five plastid pseudogenes ( fig. 4a). None of these nuclear plastid sequences (NUPTs) contain full-length and/or in-frame coding sequences. Despite their high sequence similarities to their corresponding genes in the plastome (>85% identity), these NUPTs (33-284 bp) are not syntenic with the G. ula plastome (Hsu, Wu, Surveswaran, et al. 2016), implying that they were rearranged either during transfer or subsequently (Hazkani-Covo and Martin 2017).
Our first attempt to perform transient expression using the length of predicted TPs (52-60 aa) failed to detect the plastid localization in either Gnetum or Sciadopitys. By increasing the length of cloned TPs to >80 aa (extending further into the protein sequence), we concluded that the two accDs of Gnetum target distinct compartments. The TP of nr-ACCD1 dually targets plastid stroma and mitochondria, whereas TP of nr-ACCD2 potentially targets plastoglobuli, a microcompartment within the plastids ( fig. 4b). Meanwhile, the nr-ACCD of Sciadopitys mainly targets cytosol with some putative plastid localization (supplementary fig. S5, Supplementary Material online). Overall, we experimentally demonstrated that the nr-ACCDs of gnetophytes and Sciadopitys have made up for the loss of accD from the plastomes of both lineages. denote the sequences that were derived only from transcriptomes (i.e., no plastome available). Phylogenetic relationships between the groups were derived from Li et al. (2017). Taxa., Taxaceae; Sci., Sciadopityaceae; Ara., Araucariaceae; Podo., Podocarpaceae; gne., gnetophytes; and gink., ginkgo.
Nuclear Genes Encoding Two Heteromeric ACCase Subunits (accA and accB) Are Duplicated in Various Lineages of Gymnosperms Figure 5a suggests accA duplication occurred in the common ancestor of ginkgo and cycads. Although some gnetophyte and cupressophyte genera have at least two copies of accA (figs. 1 and 5a), we could not infer the time of their duplications in the gene tree. In the accB phylogeny ( fig. 5b), however, there are two likely scenarios of gene duplication at the nodes leading to some Podocarpaceous genera and Cupressaceae. We propose that independent duplications have taken place in Cupressaceae and Podocarpaceae, as most genera of the former and some of the latter family contain two accB copies ( fig. 5b). By mapping  The green box and blue line denote the exon and intron, respectively. Orange and red boxes correspond to the distinct TPs of the two nr-accDs in G. ula. Sequences with high similarity to G. ula plastid genes are designated as pink triangles. The intron donor and splice sites are labeled with GT and AG, respectively. Figures are not drawn to scale. (b) Transient expression assays using nr-accD TPs of G. ula indicate that the nr-accD1 dually targets plastids and putative mitochondria, but nr-accD2 likely targets plastoglobuli. Red and green signals in the first two columns represent chlorophyll autofluorescence and GFP, respectively. "Merged" column shows combined chlorophyll and GFP signals, whereas "bright" indicates bright field picture of the protoplasts. The yellow and white arrows in G. ula nr-accD1 merged column indicate plastids and mitochondria localization, respectively. Scale bar ¼ 5 mm. Pt, plastid; mt, mitochondria.
Pinus, except for accC. The accC in Ginkgo has one more exon than in Gnetum and Pinus (supplementary fig. S6, Supplementary Material online). However, intron lengths are highly variable between species, ranging from 75 bp to in the accB of Pinus is close to the longest intron (318,524 bp) ever identified in Pinus taeda, the genome of which contains 108 introns with length of >100 kb (Wegrzyn et al. 2014).
Nucleotide Substitution Rates of Plastid accD and nr-Heteromeric ACCase Are Not Coelevated We calculated the nucleotide substitution rates of ACCase genes in the five major gymnosperm groups (cycads, ginkgo, pines, cupressophytes, and gnetophytes). Gnetophytes and Sciadopitys were excluded from our further analyses as their plastid accD genes were transferred to the nucleus and their nr-accDs have much higher substitution rates than others (supplementary fig. S7, Supplementary Material online). We only found weak correlation (R 2 ¼ 0.12, P ¼ 0.059) between the R S of plastid accD and the transcripts encoding nrheteromeric ACCase, but not R N (R 2 ¼ 0.055, P ¼ 0.203) ( fig. 6). No significant correlation was observed between plastid accD (or nr-heteromeric ACCase) and ACC at both R S ( fig. 6a) and R N sites ( fig. 6b). These results indicate that in gymnosperms, mutations in the plastid accD sequences have little to no effect on their nr-heteromeric ACCase.

No Homomeric ACCase Was Found in the Plastids of Any Gymnosperms
Previous studies indicate that the plastids of some angiosperms contain homomeric ACCase, either from substitutions or coexisting with the heteromeric ACCase. For instance, grasses (Poaceae) and Silene noctiflora have completely lost heteromeric ACCase from their plastids (supplementary fig. S1, Supplementary Material online; Konishi et al. 1996;Rockenbach et al. 2016). Meanwhile, in Brassicaceae and Geraniaceae, both ACCase forms coexist in the plastids (supplementary fig. S1, Supplementary Material online; Rousseau-Gueutin et al. 2013;Park et al. 2017). Similarly, in some algal groups, such as Prasinophyceae (green algae), haptophytes, and heterokonts (red algae), the plastid heteromeric ACCase is replaced by plastid-targeted homomeric ACCase (Huerlimann and Heimann 2013;Huerlimann et al. 2015). AccD is absent from the plastomes of chlamydomonadalean algae but present in their nuclear genomes (Smith et al. 2013;Smith and Lee 2014). The presence of homomeric ACCase in angiosperm plastids sometimes coincides with the loss of plastid accD (e.g., grasses) or accD elongation (e.g., Geraniaceae).
This study shows that the accDs of some gymnosperms have been lost (e.g., gnetophytes, Sciadopitys) or elongated (e.g., cupressophytes). However, transcripts encoding homomeric ACCase (ACC) are present as a single copy in all sampled gymnosperms (figs. 1 and 5d) and none of them possessed TPs (supplementary table S4, Supplementary Material online). We also demonstrate that nr-ACCDs of gnetophytes and Sciadopitys are transcribed (figs. 3 and 4). Prediction-and experimental-based assays ( fig. 4b and supplementary fig. S5 and supplementary table S5, Supplementary Material online) confirmed that their encoded products can be targeted to plastids to compensate for the loss of plastid-encoded accD in both lineages. Thus, we found no evidence of homomeric ACCase replacing or coexisting with heteromeric ACCase in the plastids of any sampled gymnosperm.

Insertions of TR into Plastid accD Do Not Affect nr-Heteromeric ACCase Evolution
We verified that accDs are elongated by in-frame, lineagespecific TR insertions in four of the five cupressophyte families (excluding Sciadopityaceae) (table 1 and (Park et al. 2017), and Passiflora (Rabah et al. 2019). Thus, elongation of accD appears to have occurred repeatedly during seed plant evolution and coincides with elevated nucleotide substitution rates. It has been hypothesized that repetitive elements in the inserted sequences promoted accD sequence variability (Li et al. 2018). Length polymorphism in the accD is likely the result of "replication slippage" as reported in the Oenothera plastomes (Massouh et al. 2016). Recent finding suggests that these length variations may account for the differences in competitiveness among the four plastid genotypes of Oenothera (Sobanski et al. 2019).
Although ACCD subunit has been known to directly interact with the heteromeric ACCA subunit (Sasaki and Nagano 2004), we did not detect significant evidence of coevolution between the plastid and nuclear genes ( fig. 6). TR insertions that elongate the accD of cupressophytes (and Tsuga) mostly occur in the middle of the sequence (supplementary fig. S4, Supplementary Material online). However, catalytic sites of ACCD, which interact with the ACCA subunit, are located in the C-terminal region (Lee et al. 2004) and highly conserved among gymnosperms (supplementary fig. S4, Supplementary Material online). Thus, TR insertions do not affect plastidnuclear interaction in the heteromeric ACCase of gymnosperms. A similar finding was previously reported in Silene species, where protein structural analyses indicate that large insertions in their ACCD subunit did not involve functionally important residues in protein-protein interactions (Rockenbach et al. 2016).  6.-Comparison of absolute nucleotide substitution rates between accD, nr-htACCase, and ACC of gymnosperms, except gnetophytes and Sciadopitys. Scatterplot and regression analyses of accD versus ACC, accD versus nr-htACCase, and nr-htACCase versus ACC for (a) absolute synonymous rates (R S ) and (b) absolute nonsynonymous rates (R N ), respectively. The points in each plot represent the 31 gymnosperm genera included in this analysis. Nr-htACCase, nuclear-encoded heteromeric ACCase; SSB, substitutions per site per billion years.
The Two nr-accDs of Gnetum Were the Product of Independent Transfers Rather than Gene Duplication Plastid accD genes of gnetophytes and Sciadopitys have been transferred to the nucleus, as nr-accD transcripts from the two lineages along with their plastid-targeting TPs ( fig. 3a) attest. Based on our dated tree (supplementary fig. S3, Supplementary Material online), the accD transfer in gnetophytes took place after its common ancestor split from Pinaceae (ca. 245 Ma) and before the diversification of three gnetophyte genera (ca. 145 Ma), and the transfer occurred less than 253 Ma in Sciadopitys. Gnetum has two copies of nr-accD (figs. 1, 3, and 4), which is so far unique among green plants. This was validated in the transcriptomes of three sampled Gnetum species ( fig. 3) and the draft genome of G. ula (fig. 4). The presence of two nr-accDs in Gnetum could have been caused either by 1) two independent plastid-to-nucleus transfer events or 2) the duplication of nr-accD after transfer to the nuclear genome. The former scenario appears to be favored because the two nr-accDs are distinctive in their TPs ( fig. 3a), gene architectures and flanking NUPTs ( fig. 4a). We propose that, in the common ancestor of gnetophytes, two copies of accD were independently transferred from the plastid to the nucleus. However, we could not detect any nr-ACCD2 from the transcriptomes of Ephedra and Welwitschia ( fig. 3). This absence may be due to the loss of nr-accD2 gene from both genera or, alternatively, the lack of nr-accD2 gene expression in the isolated RNA. The nuclear genomes of Ephedra and Welwitschia need to be sequenced in order to confirm if the nr-accD2 was indeed lost from both genera.
The Two nr-ACCDs of Gnetum Target Different Sites Using protoplast transient expression assays, we verified that the two nr-accDs of Gnetum are targeted to different subplastidic structures ( fig. 4b). The nr-accD1 TP directed GFP to the plastid stroma (and to a lesser extent to mitochondria), whereas nr-accD2 TP likely targeted GFP to the plastoglobuli. The speckled pattern we observed in the nr-accD2 construct ( fig. 4b) closely resembles the phytoene synthases (PSYs) of maize and rice that are delivered to the plastoglobuli (Shumskaya et al. 2012). The distinct targeting of the two nr-accD genes potentially suggests neo-and subfunctionalization of nr-ACCD1 and nr-ACCD2, respectively.
We were surprised to observe that nr-accD1 of Gnetum likely targets the mitochondria. Regulation of ACCase in plant mitochondria is less well characterized than its counterparts in plastids or cytosols. ACCase is reportedly absent from eudicot mitochondria, and their malonyl-CoA is synthesized using alternative pathways (Gueguen et al. 2000). However, in grasses, malonyl-CoA is generated by a homomeric ACCase that dually targets plastids and mitochondria (Focke et al. 2003). To date, little is known about ACCase in the gymnosperm mitochondria, especially whether the gymnosperm mitochondria produce malonyl-CoA via ACCase (as in grasses) or malonyl-CoA synthetase (as in eudicots) remains unclear. Nr-accD1 function in the mitochondria of Gnetum requires further investigation, as none of the other heteromeric ACCase subunits (accA-C) were predicted to be localized to mitochondria (supplementary table S4, Supplementary Material online). Mitochondrial localization might also just be a case of evolutionary noise in subcellular targeting evolution (Martin 2010).
Similarly, it is unusual that nr-accD2 targets putative plastoglobuli. Plastoglobuli are plastid lipid microcompartments that aid in plastid metabolism (reviewed in Van Wijk and Kessler [2017]). A number of nr-proteins have been reported to target plastoglobuli (Shumskaya et al. 2012;Delfosse et al. 2015). Although plastoglobuli play important roles in lipid storage and metabolism (Br eh elin and Kessler 2008), no ACCase metabolic pathway has been elucidated in the microcompartment. The only fatty acid-related enzyme identified in plastoglobuli so far is phytol ester synthase (PES; Van Wijk and Kessler 2017).
In summary, our study sheds light on the ACCase complex and its evolutionary history in seed plants. We found little evidence for coevolution between accD and its counterparts in heteromeric ACCase and the possible effects of TR insertions on its enzymatic function remain elusive. To date, we still rely on the bacterial ACCase structure to interpret the function and interaction of plants' heteromeric ACCase subunits. Elucidation of the plant-specific heteromeric ACCase structure will be critical to decipher the plastid-nuclear subunit interactions in heteromeric ACCase. Moreover, further studies on the functions of the two accDs in Gnetum will be necessary to identify why the genus has two copies of the gene.

Supplementary Material
Supplementary data are available at Genome Biology and Evolution online.