Phospholipid: diacylglycerol acyltransferase contributes to the conversion of membrane lipids into triacylglycerol in Myrmecia incisa during the nitrogen starvation stress

In addition to the Kennedy pathway for de novo biosynthesis, triacylglycerol (TAG), the most important stock for microalgae-based biodiesel production, can be synthesized by phospholipid: diacylglycerol acyltransferase (PDAT) that transfers an acyl group from phospholipids (PLs) to diacylglycerol (DAG). This study presents a novel gene that encodes PDAT from the green microalga Myrmecia incisa Reisigl H4301 (designated MiPDAT ). MiPDAT is localized on the plasma membrane (PM) via the agroinfiltration of tobacco leaves with a green fluorescent protein-fused construct. MiPDAT synthesizes TAG based on functional complementary experiments in the mutant yeast strain H1246 and the membrane lipid phosphatidylcholine (PC) is preferentially used as substrates as revealed by in vitro enzyme activity assay. The gradually increased transcription levels of MiPDAT in M. incisa during the cultivation under nitrogen starvation conditions is proposed to be responsible for the decrease and increase of the PC and TAG levels, respectively, as detected by liquid chromatography-mass spectrometry after 4 d of nitrogen starvation. In addition, the mechanism by which MiPDAT in this microalga uses PC to yield TAG is discussed. Accordingly, it is concluded that this PM-located PDAT contributes to the conversion of membrane lipids into TAG in M. incisa during the nitrogen starvation stress.

and reduced size of chloroplasts in M. incisa during the nitrogen starvation stress 18 , we determined whether the enzyme PDAT contributes to TAG accumulation.
In this study, a full-length complementary DNA (cDNA) encoding PDAT (designated MiPDAT) was cloned from M. incisa, and its function was identified by a complementary experiment in the TAG-deficient strain H1246 of Saccharomyces cerevisiae. It was shown that MiPDAT used phosphatidylcholine (PC) to synthesize TAG indirectly by feeding the transgenic yeast with various fatty acids (FAs) as substrates and directly by in vitro enzyme activity assay. The in vivo evidence for the function of this gene in M. incisa was provided by estimating the variation of PLs and MiPDAT transcription levels using liquid chromatography-mass spectrometry (LC-MS) and quantitative real-time PCR (Q-RT-PCR), respectively, in this microalga during the nitrogen starvation stress. Furthermore, to determine the subcellular localization of MiPDAT, this gene was fused with the green fluorescent protein (GFP) gene to construct the vector p1300-MiPDAT-GFP. This vector was then infiltrated into the tobacco leaf mediated by Agrobacterium tumefaciens GV3101. It is concluded from these data that MiPDAT contributes to TAG accumulation by converting membrane PC in M. incisa grown under nitrogen starvation stress.

Results
Cloning and characterization of a PDAT gene from M. incisa. Seven contigs, contigs 1820, 4542, 7311, 7384, 8397, 13344, and 15617, from the transcriptome database of M. incisa 24 were determined to be homologous to the gene PDAT. From the manual assembly of these contigs, a 2,939-bp fragment was obtained and verified by PCR amplification. Based on this fragment sequence, two pairs of primers (GSP5-1 and NGSP5-1, GSP3-1 and NGSP3-1, Supplementary Table S1) were designed, and then, a 1,162-bp 3′ -rapid-amplification of cDNA ends (RACE) product and a 467-bp 5′ -RACE product (Fig. 1A) were amplified and sequenced. These amplified products shared overlapping regions at their corresponding ends of the 2,939-bp assembled fragment. After manual assembly of these sequences, a 3,079-bp full-length cDNA consisting of a 2,076-bp open reading frame (ORF), 218-bp 5′ -untranslated region (UTR), and 785-bp 3′ -UTR was obtained and designated MiPDAT. The ORF of MiPDAT was predicted to encode a protein consisting of 691 amino acids with a molecular weight of 77.27 kD. This putative protein MiPDAT had 50%, 32%, and 34% similarities with the PDATs from Chlamydomonas reinhardtii (GenBank accession No. AFB73928), Saccharomyces cerevisiae (GenBank accession No. DAA10549), and Arabidopsis thaliana (GenBank accession No. AED91921), respectively.
To characterize the gene structure of MiPDAT, the corresponding genomic DNA of MiPDAT was amplified with one pair of designed primers (MiPDAT-OS and MiPDAT-OA, Supplementary Table S1). After cloning and comparison of the corresponding cDNA and DNA sequences, it was revealed that MiPDAT was separated by 12 introns with lengths of 361 bp, 136 bp, 215 bp, 239 bp, 266 bp, 384 bp, 176 bp, 199 bp, 211 bp, 246 bp, 227 bp, and 209 bp from the 5′ -terminus (Fig. 1B). All of the introns had splicing consensus GT-AG borders. Both the cDNA and DNA sequences of MiPDAT were deposited in GenBank under accession Nos KU851950 and KU871014, respectively.
Homologous alignment and phylogenetic inference of MiPDAT. By searching InterPro, a lecithin: cholesterol acyltransferase (LCAT, EC 2.3.1.43) domain (Phe128-Asp690) in the putative MiPDAT was identified, suggesting that it belonged to the LCAT family (Pfam: 02450). Therefore, two LCAT and ten other PDAT homologs were included in the multiple sequence alignment ( Supplementary Fig. S1), showing at least 7 reserved domains. Of these domains, Domain II, designated the lid domain by Peelman et al. 25 , was closed by a disulphide bridge, as detected between the two nearly neighbouring residues Cys171 and Cys196 in MiPDAT, corresponding to Cys74 and Cys98 in human LCAT. This bridge was highly conserved and covered a hydrophobic active site in all of the PDATs. The residue Trp in this lid domain was also conserved in MiPDAT and others ( Supplementary Fig. S1), and it was predicted to bind cleaved FAs in the active site of these enzymes 26 .
The highly conserved Domain III, which contains a salt bridge between Asp258 and Arg260, may be involved in PL recognition 12 . A catalytic triad of Ser205, Asp369, and His401 distributed in Domains IV, VI, and VII, respectively, of MiPDAT was also conserved in both the LCATs and PDATs. This triad is a part of the catalytic domain of the LCAT enzymes, in which a FA is transesterified from PC to cholesterol to yield a cholesterol ester 12,25 . There was a consensus sequence, GHSXG, which is part of a conserved lipase motif, in Domain IV of all LCATs 27 . The first Gly in the domain of these LCATs was, however, different from that in the plant PDATs, and it was replaced by Ser in algal PDATs or Pro in higher plants ( Supplementary Fig. S1). Compared to the domains discussed above, the roles of the relatively conserved Domains I and V are poorly understood in PDATs.
The neighbour-joining phylogenetic tree ( Supplementary Fig. S2) inferred from MiPDAT and other LCAT-like family proteins supported the conclusion 12,28 that the LCAT-like family proteins from higher plants, animals, fungi, and algae could be divided into four major groups. This phylogenetic tree ( Supplementary Fig. S2) also showed that MiPDAT was grouped into PDATs, which was supported by a bootstrap value of 99%, and that it was closer to the microalgal PDATs, including a function-identified Chlamydomonas PDAT 12 , than the higher plant PDATs.
Based upon these characteristics and the phylogenetic analysis, it was suggested that the cloned gene MiPDAT from M. incisa should function as a PDAT to yield TAG by transferring an acyl group from PLs.

Functional expression of MiPDAT in Saccharomyces cerevisiae H1246.
To identify the function of the protein encoded by MiPDAT, the ORF of this gene was used to generate a recombinant plasmid, pY-MiPDAT ( Supplementary Fig. S3), for complementary experiments in the TAG-deficient mutant yeast strain H1246 29 . TAG, in transgenic or wild-type lines, was detected by thin layer chromatography (TLC) analysis, and oil bodies stained with BODIPY were observed. As shown in the TLC profile ( Fig. 2A), a prominent spot corresponding to a TAG standard occurred in the recombinant line with MiPDAT comparable to the wild type, whereas there was no TAG formed in the mutant strain H1246 or the negative control only carrying empty vector. This result demonstrated that the TAG-deficient mutant recovered TAG synthesis after MiPDAT was introduced, indicating that MiPDAT can synthesize TAG.
TAG is usually present in the form of oil bodies in both yeast and plants [19][20][21][22] . Therefore, observations on the formation of oil bodies in yeast may be beneficial for the understanding of MiPDAT function. Using the unique staining agent BODIPY 23 , oil bodies were observed in wild-type Scy62 (the original strain of H1246 29 ) cells and the recombinant line with MiPDAT, but they were absent in the negative control and mutant strain H1246 (Fig. 2B). Apparently, this introduced gene led to the formation oil bodies in the mutant yeast cells, confirming the TAG synthesis function of MiPDAT, similar to the recently reported DGATs in M. incisa 23 .

Substrate preference analysis of MiPDAT. To better understand whether MiPDAT has a preference for
FAs for the synthesis of TAG, the recombinant yeast carrying pY-MiPDAT was fed various FAs selected primarily based on their presence in M. incisa 17 . These exogenous FAs were separately added as substrates to the SC-uracil induction medium for yeast cultivation. TAG extracted from the recombinant line carrying pY-MiPDAT was detected by TLC analysis (Fig. 3A). The intensity of the coloured spots corresponding to the TAG (Fig. 3A) differed in each sample cultivated with different substrates, suggesting that this recombinant line had a substrate preference. Quantitative analysis by gas chromatography-mass spectrometry (GC-MS) indicated that this recombinant line used γ -linolenic acid (GLA) preferentially to synthesize TAG because the level of GLA in the TAG (Fig. 3B) from the same weight of lyophilized yeasts was significantly (P < 0.01) higher than that of other FAs. Additionally, MiPDAT had a significant (P < 0.01) preference for linoleic acid (LA), arachidonic acid (ArA), and α -linolenic acid (ALA) (Fig. 3B). This finding was consistent with the preference analysis of Linum usitatissimum PDAT 11 .
In vitro enzyme activity assays could directly reveal the preference for PLs used by MiPDAT in TAG synthesis. As shown in Fig. 4, there was one prominent spot corresponding to the TAG standard occurred when PC+ DAG or phosphatidylethanolamine (PE)+ DAG were mixed and incubated with transgenic yeast crude extracts. The abundance of TAG synthesized from PC seemed to be more than that from PE (Fig. 4). By contrast, there was no formed TAG while either one of two lipid donors or acceptor was mixed with the crude extracts. Neither was there while incubating with the crude extracts alone (Fig. 4). This result indicated that MiPDAT could use PC and PE to synthesize TAG, and MiPDAT had a preference for PC over PE.
Effect of nitrogen starvation on the content of microalgal phospholipids. PLs usually decrease in microalgae, particularly when grown under nitrogen starvation 2,13-15 . A culture during the nitrogen starvation stress was therefore performed to investigate the variation of PLs as detected by LC-MS in M. incisa. A principle component analysis (PCA) model with two components was constructed and showed that the lipid samples designated 4 d group (culture under nitrogen starvation for 4 d) were clearly separated from those designated 0 d group (culture at the onset of nitrogen starvation) ( Supplementary Fig. S4A). In addition, the discriminant analysis, orthogonal projection to latent structures with discriminant analysis (OPLS-DA) [R 2 Y(cum) = 0.991 and Q 2 (cum) = 0.948] model with high R 2 Y(cum) and Q 2 (cum) values also provided reliable support for the  Fig. S4B). It was concluded that M. incisa varied significantly in its lipid profile after cultivation under nitrogen starvation for 4 d.
To identify the lipid that contributed to the separation of the nitrogen starvation cultivation group for 4 d from the 0 d group, 84 potential lipid biomarkers [variable importance in the projection (VIP) ranging from 3.57 to 1.00] were selected according to the VIP values as well as the corresponding 95% confidence intervals based on a jack-knife procedure and the absolute value of p(corr) 30,31 . Of these biomarkers, the plasma-membrane lipids, PE and PC, accounted for 54.76% of the total (46 to 84) and possessed relatively high VIPs (ranging from 1.01 to 3.40), indicating that these two PLs were the main contributors to this separation.
Furthermore, the PC and PE contents were compared between the 0 d and 4 d groups of M. incisa cultured under nitrogen starvation stress. As shown in Fig. 5A, the PC level decreased by 13% in the microalga after 4 day nitrogen starvation cultivation and the PE level decreased by only 1%. These results suggest that PC rather than PE should be converted to other lipids when M. incisa is cultured during the nitrogen starvation stress.

The transcription of MiPDAT in M. incisa grown under nitrogen starvation and replenishment
conditions. If MiPDAT is related to TAG accumulation and/or PLs degradation in M. incisa grown during the shift from nitrogen starvation to rich in BG-11 medium, the transcription of this gene should vary correspondingly. To determine this variation, the gene transcription of MiPDAT was detected using Q-RT-PCR. As shown in Fig. 5B, with nitrogen starvation, the abundance of transcripts for MiPDAT in M. incisa was increased with the duration of this culture condition and the level increased until reaching a maximal level at 96 h, when it was approximately 3-fold higher than at 0 h (P < 0.01). After the onset of nitrogen replenishment, the transcription of this gene decreased gradually, and the level of transcripts at 168 h was only 43% of the level at 96 h (P < 0.01). This result indicated that MiPDAT was up-and down-regulated in M. incisa by nitrogen starvation and nitrogen replenishment, respectively, and this regular alternation showed the same trend as the TAG level as reported by Chen et al. 23 , suggesting that the transcription of MiPDAT is positively related to TAG accumulation.

Subcellular localization of MiPDAT in tobacco.
MiPDAT may link TAG accumulation with PL degradation in M. incisa during the course of nitrogen starvation; therefore, it was hypothesized that MiPDAT is localized to the membrane to easily use PLs. The subcellular localization of MiPDAT was predicted to determine this. A transmembrane domain (Phe70-Ala92) was predicted by TMHMM analysis, suggesting that MiPDAT is a membrane-bound protein. Neither signal peptide nor transit peptide was predicted by the SignalP 4.1 Server, TargetP 1.1 Server, or the ChloroP 1.1 Server, indicating that this PDAT is not a plastid-localized PDAT, which is different from the PDAT in C. reinhardtii 12 .
To clarify the subcellular localization of MiPDAT, the complete ORF of this gene was fused with GFP to generate a binary vector (Supplementary Fig. S5) and then infiltrated into the lower epidermal cells of tobacco leaves via Agrobacterium tumefaciens GV3101. The green fluorescence from the GFP was dispersed thoroughly in the protoplasts of the infected cells (Fig. 6). When the construct p1300-GFP was used as a control, the fluorescent signal on the plasma membrane (PM) of these infected cells, as seen in the two guard cells shown in the middle row of Fig. 6, was not distinct from the cytoplasm. By contrast, after MiPDAT was introduced, the green fluorescent signal was mainly visualized on the PM of infected cells so that the PM was clearly distinguished from the cytoplasm (Fig. 6).
These results indicate that this PDAT from M. incisa is located on the PM of infected tobacco leaf epidermal cells.

Discussion
This study presents a novel microalgal PDAT cloned from M. incisa and different from the reported C. reinhardtii PDAT (designated CrPDAT) because the MiPDAT is located on the PM (Fig. 6). The CrPDAT, however, is proposed to be located on chloroplasts based on the presence of a chloroplast transit peptide 12 , although there is no direct evidence for its subcellular localization. Except for this difference, MiPDAT is similar to CrPDAT and others because they possess seven characteristic conserved domains ( Supplementary Fig. S1), and this protein is grouped into the PDAT clade, including the function-identified CrPDAT, which is supported by a 99% bootstrap value in the reconstructed phylogenetic inference (Supplementary Fig. S2). Furthermore, the ability of MiPDAT to synthesize TAG is shown by complementary experiments in yeast (Fig. 2). These data indicate that MiPDAT is a member of the PDAT family. In this case, which PL is used by MiPDAT to synthesize TAG?
PDAT transfers an acyl group from the sn-2 position of PLs, such as PC, to the sn-3 position of diacylglycerol (DAG) to yield TAG 9,10,12,32 . By feeding yeast exogenous FAs for the functional identification of PDAT in Linum usitatissimum (designated LuPDAT), Pan et al. 11 found that these added FAs were transformed to lyso-PC by acyl-CoA: lysophosphatidylcholine acyltransferase (LPCAT EC 2.3.1.23) 33 to yield FAs-PC. As a result, LuPDAT can use the FA-PCs to synthesize TAG in yeast 11 . A similar feeding experiment was performed for this cloned MiPDAT, and the preference for GLA (Fig. 3) by the introduced MiPDAT was determined, which is similar to that described by Pan et al. 11 Therefore, MiPDAT may use the same FA-PCs as LuPDAT to synthesize TAG. In vitro activity assay of the enzyme extracted from the transgenic yeast with MiPDAT (Fig. 4) confirms this deduction evidently and MiPDAT has a preference for PC to PE for TAG synthesis. In this case, the level of PC in M. incisa may be reduced, whereas the level of TAG increases 23 during the nitrogen starvation culture. This hypothesis is verified by a lipidomics analysis showing that PC is a main contributor to the significantly changed PLs and that the level of PCs significantly decreases (Fig. 5A) in M. incisa grown under nitrogen starvation stress for 4 d while comparing to that at the onset of nitrogen starvation. Consistent with the changed levels of PC (Fig. 5A) and TAG (Fig. 4 in the ref. 23), the transcription of MiPDAT is up-and down-regulated (Fig. 5B) by the onset of nitrogen starvation and replenishment, respectively. Taken together, this study provides temporal evidence that MiPDAT is involved in PC degradation and TAG accumulation in this microalga grown under nitrogen starvation, although the MiPDAT-knockout or -knockdown M. incisa remains to be completed.
In this study, we have observed that MiPDAT is primarily located on the PM of tobacco leaf epidermis cells (Fig. 6) by co-expression with GFP. As discussed above, MiPDAT uses PC to synthesize TAG. There may be a bridge that connects the PM with the endoplasmic reticulum (ER) because the ER is the principal site for PC biosynthesis and TAG accumulation 34,35 . The ER is an interconnected network in eukaryotic cells, including microalgae, and membrane contact sites (MCSs) are present between the ER and other organelles [35][36][37] . The MCSs display a wide variety of functions including, but not limited to, lipid exchange between two compartments 36 . The ER-PM MCSs, therefore, are proposed to provide a possible mechanism by which the PM-located MiPDAT uses the ER-synthesized PC to accumulate TAG in the ER to form oil bodies in M. incisa. From the transcriptome database of M. incisa 24  In addition, the GFP fluorescent signal has seemingly been detected on the transgenic tobacco leaf chloroplasts, although it cannot be distinguished clearly from the negative control (Fig. 6). For the absence of a transit peptide in MiPDAT as predicted above, and because PC is one component of the outer envelope of chloroplasts 38 , MiPDAT may also be located on this outer envelope. In this case, MiPDAT can synthesize TAG by consuming the chloroplast PC, possibly at the ER-chloroplast MCSs 34,35 . Two contigs (Supplementary Data 3) coding for coatomer proteins that act on the ER-chloroplast MCSs 34,37 have also been found in the transcriptome database of M. incisa 24 . However, the chloroplast localization of MiPDAT is not certain, unless new direct evidence using, for example, gold immunoelectron microscopy, is provided.
The temporal and spatial evidence for MiPDAT contributing to the conversion of membrane lipids into TAG in M. incisa during nitrogen starvation stress is provided. There are many ways that microalgae utilize membrane lipids to synthesize TAG. For example, PC can also be degraded by phospholipases, and the resulting acyl group can be sequentially used to synthesize TAG along the Kennedy pathway 33 . Additional studies are necessary to further understand the physiological roles of PDAT in microalgal cells grown under stress conditions. Because the ER is the general location of TAG accumulation, this PM-located MiPDAT is more significant than the chloroplast-located one for microalgae-based biodiesel production.

Methods
Myrmecia incisa and culture conditions. The microalga Myrmecia incisa Reisigl H4301 16 was commercially provided by the Culture Collection Algae of Charles University of Prague (CAPU). This alga was cultivated, washed, and harvested as described previously 24 . For the Q-RT-PCR assay, M. incisa was cultivated, washed, and harvested as described by Yu et al. 39 . For lipid analysis, approximately 60 mL of an algal suspension was sampled every 4 d during the cultivation course of nitrogen starvation. The collected microalgal pellets were washed with distilled water, lyophilized, and stored at − 80 °C until use.
Complementary DNA (cDNA) and DNA cloning of MiPDAT. Total RNA was extracted from M. incisa using TRIzol reagent according to the manufacturer's instructions (Invitrogen, USA). Complementary DNA was synthesized from the total RNA using the PrimeScript ™ RT Reagent Kit (TaKaRa, Japan). The full-length cDNA of MiPDAT was amplified with a SMART ™ RACE cDNA Amplification Kit (Clontech, USA) with two designed pairs primers based on the screened sequences from a high throughput transcriptome of M. incisa 24 . One pair of primers (GSP5-1 and NGSP5-1) was designed for the 5′ -RACE reaction, and the other pair of primers (GSP3-1 and NGSP3-1) was designed for the 3′ -RACE reaction (Supplementary Table S1). The amplified products were resolved on a 1.0% low-melting-point agarose gel for DNA recovery. The target product was recovered using agarose gel purification and an extraction kit (Aidlab, China) and was ligated to a pMD19-T vector (TaKaRa, Japan). The constructed vector was subsequently transformed into Escherichia coli DH5α competent cells (Tiangen, China), and positive clones were sent to Sangon (China) for sequencing. The DNA was extracted using a modified cetyltrimethylammonium bromide (CTAB) method 40 Table S1) designed with Hind III/BamH I digestion sites and then ligated into the pYES2 vector (Invitrogen, USA) to generate the construct pY-MiPDAT. This recombinant vector and the empty plasmid pYES2, the negative control, were separately introduced into the mutant strain H1246 by electroporation (Bio-Rad, USA). Transformants were selected on SC-uracil agar plates and then inoculated in SC-uracil medium with 2% glucose on a shaker (220 rpm) at 30 °C. Following incubation for 24 h, the transformed line was collected by centrifugation and resuspended in SC-uracil medium with 2% galactose as an inducer for another 48-h incubation at 16 °C.
BODIPY staining of yeast cells. BODIPY 505/515 (Life Technology, USA), a specific dye for the assay of neutral lipids, was used to stain yeast oil bodies as described previously 23 . The fluorescence image was captured by a confocal laser scanning microscope (Carl Zeiss, Germany) with a 488 nm excitation filter.

Extraction of lipids from yeasts and TLC analysis. Total lipids were extracted from yeasts according to
Blight and Dyer 41 . The sample was resolved in 30 μ L of chloroform and pipetted by capillary onto TLC Silica Gel 60 F254 plates (Merck, Germany) and developed with hexane/diethyl ether/acetic acid (80:20:1, v/v/v). The plates were visualized by spraying with a solution containing 8% phosphoric acid (v/v) and 10% CuSO 4 ·5H 2 O (w/v) and subsequent heating at 140 °C for 10 min.
Feeding experiment for substrate preference and GC-MS analysis. To test whether MiPDAT had a substrate preference, LA, ALA, GLA, and ArA, the main FAs in M. incisa 17 , were selected in this feeding experiment. The transgenic yeast was incubated with 200 μ M FAs and 0.01% (v/v) tyloxapol (Sigma-Aldrich, Germany) as a surfactant in the induction medium described above and harvested at an optical density of 3.5. Approximately 150 mg of lyophilized yeast cells were used for lipid extraction and TLC analysis as described above. The recovery of TAG-containing fractions and the preparation of FA methyl esters (FAMEs) of the recovered TAG were according to the previously described method 23 . Fifty micrograms of heptadecanoic acid (C17:0, Sigma-Aldrich, USA) was added as an internal standard.
GC-MS analysis was performed with an Agilent 7890-5975b system (Agilent Technologies, USA). The detailed GC-MS conditions are provided in the Supplemental Methods. The mole number of FAMEs was calculated with the peak area of the internal standard.

Preparation of yeast crude extracts and in vitro
MiPDAT activity assay. The crude extracts of transgenic yeast were prepared as described by Dahlqvist et al. 9 . The extracts were resuspended in the storage buffer (50 mM potassium phosphate, pH 7.2, 10% glycerol) to a protein concentration of 5 μ g μ L −1 . The in vitro assay of MiPDAT activity was performed as described by Liu et al. 4 . The reaction mixture containing 250 μ M lipid donors (PC or PE), 250 μ M lipid acceptor (DAG), and the resuspended crude extracts (40 μ g protein in total) with 50 mM potassium phosphate (pH 7.2) in a final volume of 200 μ L. Reactions were incubated at 30 °C for 1 h and the lipids were extracted and detected as described above. The crude extracts alone and crude extracts with either lipid donors or acceptor were used as controls.

Lipidomics analysis of Myrmecia incisa.
Chlorophylls in the total extracted lipids described above were removed by solid phase extraction (SPE) with CNWBOND Carbon-GCB SPE columns (CNW, Germany). Reversed-phase analysis of lipids following lipid extraction was performed on a Dionex Ultimate 3000 High Performance Liquid Chromatography (HPLC) system (Thermo Fisher Scientific, USA) by injection of 4 μ L of sample onto a Kinetex C18 column (100 × 2.1 mm, particle size 1.9 μ m; Phenomenex, USA). The detailed LC-MS conditions are provided in the Supplemental Methods.
Raw LC-MS data for all samples were initially processed using the Thermo SIEVE 2.1 Qualitative Analysis Software (Thermo Fisher Scientific, USA). The data from each sample were then normalized to the total area, and all data with peak numbers [based on the retention time and mass-to-charge ratio (m/z)], sample names, and normalized peak intensities were imported into the software SIMCA-P+ 12.0 (Umetrics, Sweden). Multivariate analyses, such as PCA and OPLS-DA, were applied for the classification of lipid samples. The quality of the model OPLS-DA was evaluated by two parameters, R 2 Y(cum) and Q 2 (cum), according to Wiklund 42 . If the value of the latter was higher than 0.9, the model was considered excellent 42 .
Identification of lipid metabolites. To select potential lipid biomarkers that contributed to the separation between the 0 d and 4 d groups grown under nitrogen starvation stress, three criteria, specifically, the high Scientific RepoRts | 6:26610 | DOI: 10.1038/srep26610 variable importance in the projection (VIP) and jack-knifed confidence interval (CIJF JK ) excluding zero and an absolute value of p(corr) higher than 0.4, were used 29,30 . The lipid metabolites were identified using a commercial MS/MS database, Lipid Search (Thermo Fisher Scientific, USA). Changes in PC and PE were expressed as the relative abundance (%), which was the ratio of the peak intensity of the test group (4 d) to that of the control group (0 d) with the same dry weight, and the relative abundances (%) of the control was expressed as 100% 43-45 . MiPDAT transcription levels estimated by Q-RT-PCR. Quantitative RT-PCR was used to detect the MiPDAT transcription levels as described previously 39 . The cDNA was synthesized using the Reverse Transcribed Kit II (TaKaRa, China). The Myrmecia incisa β -actin gene (GenBank accession No. FJ548973) served as an internal standard. The primers used in the Q-RT-PCR were shown in Supplementary Table S1. The Q-RT-PCR amplification was performed in an ABI 7500 Real Time PCR system (Thermo Fisher Scientific, USA) with the SYBR ® RT-PCR Kit (TaKaRa, Japan). The relative gene transcription data from triplicate reactions performed for each incubation time were expressed as the means ± standard error using the 2 −ΔΔCT method 46 .

Subcellular localization by agroinfiltration of tobacco leaves.
To investigate the subcellular localization of MiPDAT, the complete ORF of MiPDAT was fused with GFP into a binary vector and then infiltrated into tobacco leaves using Agrobacterium tumefaciens. The PCR products for the MiPDAT ORF amplified with the primers pKpnF and pXbaR (Supplementary Table S1, containing Kpn I/Xba I digestion sites) were ligated into a modified pCAMBIA1300 binary vector containing triple GFP molecules (from Prof. Z.-N. Yang) in tandem (designated p1300-MiPDAT-GFP). This recombinant vector and an empty plasmid, p1300-GFP, as the negative control were separately introduced into A. tumefaciens GV3101 by electroporation (Bio-Rad, USA) and then infiltrated into the abaxial side of leaves of 4-to 5-week-old Nicotiana benthamiana plants 47 . After agroinfiltration, the transgenic line was placed in the dark for 48 h. The epidermis cells of the infiltrated tobacco leaves were observed under a confocal laser scanning microscope (Carl Zeiss, Germany). GFP fluorescence was monitored with a 500-to 550-nm band pass emission filter excited at 488 nm, and the autofluorescence of chlorophyll was examined with the same excitation filter as GFP, but with a 650 nm to 750 nm emission filter.