Dunaliella viridis strain dumsii was grown in a modified Ben-Amotz medium (pH 7.5) as reported previously (Murphree et al., 2017). Experiments were performed in a Percival E-36L growth chamber under continuous cool white fluorescent light (125 μmol photons m²/s) at a temperature of about 23°C. A preliminary growth experiment was performed as previously (Murphree et al., 2017) to determine a GLN concentration suitable for sufficient growth and lipid production. Concentrations of 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, and 20 mM GLN were screened and 5 mM GLN (the original concentration from Murphree et al., 2017) was chosen for more detailed analysis (Supplementary Figure S1). Four experimental media (pH 7.5) were created with nitrogen sources of 5 mM KNO₃, 5 mM L-glutamine (Sigma-Aldrich), 5 mM L-glutamate (Sigma-Aldrich), or no addition of KNO₃ (Supplementary Figure S2). The L-glutamate condition was used to control for nitrogen starvation in the presence of an unused and stable amino acid (Murphree et al., 2017). Cultures for inoculation were grown for a week in the KNO₃ experimental condition and were harvested by centrifugation (3500 × g, 6 min.) in mid-logarithmic phase. These cells were suspended in 450 ml of the proper experimental media in 1 l narrow-necked Erlenmeyer flasks to a concentration of 1 million cells/ml. Each of the four experimental cultures was duplicated to give two technical replicates and the entire experiment was repeated three times to provide three biological replicates.

Samples were taken in 24 h intervals starting at 3 h after inoculation (3, 27, and 51 h). Cultures were given 3 h to adjust to the new media to attempt to avoid effects from transfer shock in the data. In addition to the samples taken from the experimental cultures, half volume samples were taken in duplicate from the inoculum culture (INOC) right before inoculation. Sample collection consisted of aliquots taken for cell density/diameter/pH measurements (1 ml), RNA extraction (30 ml), neutral lipid fluorescence assay (5 ml), carbohydrate detection (100 ml), soluble protein assay (5 ml), and chlorophyll content measurement (1 ml) (Supplementary Figure S2). The aliquots for RNA, carbohydrate, and protein measurements were pelleted and flash frozen in liquid nitrogen for bulk processing. Cell density/diameter/pH, neutral lipid fluorescence assay, and chlorophyll content samples were processed immediately using the following protocols.

Cell density and diameter were determined using a TC10 Automated Cell Counter (Bio-Rad). Cell diameters were calculated from histogram data provided from the counter. Cells were grouped into 2 μm wide bins and an approximate average cell diameter was calculated from those data. The pH was measured with a Glass Semi-Micro Combination pH Electrode (Beckman Coulter) directly on the 1 ml cell suspension used for cell density/diameter measurements.

Chlorophyll content was quantified as described in Srirangan et al. (2015). A 1 ml aliquot of cells was pelleted (3500 × g) and 0.75 ml of supernatant was removed. 1 ml of 100% ethanol was added to the 250 μl pellet and the samples were vortexed. After incubating at room temperature for 10 min, the samples were vortexed again and centrifuged for 10 min at 13000 rpm. The absorbance of the supernatant (1 ml) was read at 652 nm in a BioSpectrometer basic (Eppendorf). The total chlorophyll concentration (in mg/ml) was calculated as A_{652 nm}/36, where 36 is the extinction coefficient for chlorophyll A and B in ethanol (Arnon, 1949; Winter, 1993).

5 ml frozen pellets of cells were suspended in 1 ml of deionized water and soluble protein was extracted as described previously (Murphree et al., 2017). Briefly, frozen cells suspended in water were pelleted and the supernatant removed. The pellet was re-extracted with 1 ml of 0.1 N NaOH, incubated at room temperature with intermittent vortexing for 60 min, centrifuged (3500 × g), and the supernatants combined. Protein was quantified using the Micro BCA^TM Protein Assay Kit (ThermoFisher) on a plate reader (Biotek Synergy HT) (Srirangan et al., 2015). Protein concentrations were calculated using linear regression analysis against a bovine serum albumin standard curve ranging from 0 to 200 μg/ml.

100 ml culture were pelleted (3500 × g) and the supernatant discarded. The pellets were frozen until processing. Total carbohydrates were quantified by resuspending the pellets in 1 ml of water. A 1:100 dilution of the suspension with water was used for carbohydrate extraction and quantified using a microplate-adapted version of the Dubois method (Dubois et al., 1956). Briefly, 40 μl of cell dilutions were placed in 96-well polystyrene plates (Olympus Plastics) and then 40 μl of 5% w/v crystalline phenol/H₂O was added and mixed. Plates were incubated for 15 min and then 200 μl of 95–98% H₂SO₄ was mixed into the samples. Absorption of the samples at 490 nm was detected in a plate reader (Biotek Synergy HT). All samples were prepared in triplicate and a standard curve of 5 to 500 μg/ml sucrose in H₂O was used to calculate total carbohydrate content.

Neutral lipids were detected in whole cells using a Nile Red dye-based assay as described in Murphree et al. (2017). Briefly, 5 ml culture samples were pelleted and suspended in 600 μl of fresh modified Ben-Amotz media. Samples were divided into 200 μl triplicates in 96-well polystyrene plates (Olympus Plastics) and Nile Red (9-diethylamino-5H-benzo(α)phenoxazine-5-one, Sigma-Aldrich) was added to a final concentration of 0.26 μM. Samples were incubated at room temperature in the dark for 15 min and then excited at 485 nm and detected at 590 nm in a plate reader (Biotek Synergy HT). Coconut oil standards of 5–100 μg/ml were used to calculate neutral lipid concentrations (Murphree et al., 2017).

Original transcriptome annotations for D. viridis strain dumsii were from a 2012 Blast2GO annotation run (Srirangan et al., 2015). Due to the advancements in the number of genomes sequenced and the expansion of annotation databases, the reference transcriptome was out-of-date. The newly released Dunaliella salina CCAP 19/18 genome, available through Phytozome (Polle et al., 2017), was used to re-annotate based upon similarity of D. viridis transcripts to D. salina proteins. The D. salina proteome was obtained from Phytozome on June 11, 2017 and converted to a Blast database within CLC Genomics Workbench version 10.0¹. Blastx was used to extract the best protein hit for each D. viridis transcript using the following parameters: maximum number of blast hits = 1; e-value cutoff = 10⁻¹; low complexity filter = off. D. salina annotations were then ascribed to the D. viridis transcripts. Previous manual annotations were kept as they had already been confirmed. This rough annotation increased the number of annotated genes in D. viridis strain dumsii from 9400 to 13273. Genes highlighted in results figures were all checked for conflicts between the original and the new annotation, and conflicts were resolved by Blastx against the NCBI nonredundant (NR) protein database. Annotations are available in Supplementary Data Sheet S1.

Total RNA was extracted from the 30 ml frozen pellets of the 39 samples using the RNeasy Plant Mini Kit (Qiagen) following standard procedures for use with a QIAcube (QIAgen, batch V1, serial #14666). DNA was removed from total RNA using RNase-Free DNase (Qiagen), and rRNA was depleted from the samples using the Ribo-Zero^TM Magnetic Kit (Plant Leaf) (Epicenter). After each processing step, RNA quality was confirmed using a Bioanalyzer RNA Pico Chip (Agilent). Strand specific cDNA libraries with 300–450 bp inserts were generated with the NEBNext^® Ultra^TM Directional RNA Library Prep Kit for Illumina^® (New England BioLabs). Sample libraries were barcoded using the TruSeq DNA LT Sample Prep Kit, Indexed Adapter Sequences (Illumina). Final quality of the libraries was confirmed using a Bioanalyzer DNA high sensitivity chip (Agilent). Libraries were split into 2 groups and submitted for single end 125 bp Illumina HiSeq 2500 sequencing runs in 2 flow cells at the North Carolina State University Genomic Sciences Laboratory (GSL). Quality control, filtering, and trimming were performed using FastQC and the FAST-X toolkit and resulted in an average loss of 3% of reads.

Reads from each sample were mapped to the previously de novo assembled reference transcriptome of D. viridis strain dumsii (Srirangan et al., 2015) using the RNA-Seq Analysis module of the CLC Genomics Workbench version 10.0¹. The reference transcriptome was modified to eliminate seven contaminating transcripts from tomato, and six transcripts were trimmed or eliminated due to previous adapter/index contamination in the original assembly. The RNA-Seq Analysis module, which uses an expectation-maximization (EM) algorithm like RSEM (Li and Dewey, 2011), was run with default parameters (mismatch cost = 2; insertion cost = 3; deletion cost = 3; length fraction = 0.8; similarity fraction = 0.8), and strand specificity was ignored. Since putative alternate splice transcripts are represented individually in the reference transcriptome, the “maximum number of hits for a read” was increased to 30 so that reads that matched the multiple transcripts were not discarded. As these putative transcripts are only predicted and not verified for D. viridis, gene counts are preferable to transcript differences. To obtain gene specific data, raw counts for each transcript were exported from the CLC Genomics Workbench, and the counts for each putative transcript of each gene were summed to give a total raw read count per gene. The total gene counts were then imported back into CLC Genomics Workbench for differential expression analysis using the Microarray Tools in CLC.

Differential expression analysis was used to determine significant transcriptional differences between the four nitrogen sources, nitrate (NO3), nitrogen starvation (STAR), glutamate (GLU), and glutamine (GLN) at each of the three time points (3, 27, and 51 h) (Supplementary Figure S2). Comparisons were also made between the time points within each nitrogen source to determine transcriptional differences over time for a single nitrogen source. All samples were also compared to the INOC to provide a common comparison point (Supplementary Figure S2). DE analysis was performed using the “Empirical analysis of DGE” tool in CLC, which utilizes the edgeR version 3.4.0 implementation of Fisher’s Exact Test for two-group comparisons (Robinson et al., 2010). Only genes that had a total number of read counts > 5 within each comparison group were used to estimate tagwise dispersions, and the counts from biological replicates were averaged. Initial analysis showed all comparisons between the GLU and STAR conditions were not statistically different, and thus all comparisons were regenerated without GLU data. DE was considered significant based on the following parameters: False Discovery Rate (FDR) corrected p-value ≤ 0.05, log₂ fold change ≥|1|, and the total read count for the gene within the comparison was > 75. Raw read files and total read counts are available through the Gene Expression Omnibus accession, GSE111548 (Dums et al., 2018). Log₂ fold change values and FDR p-values for all comparisons are available in Supplementary Data Sheet 1.

The genes for lipid synthesis, triacylglycerol synthesis, and starch metabolism pathways were extracted from previous work (Srirangan et al., 2015). Genes for nitrogen assimilation, the Calvin-Benson Cycle, and the Krebs Cycle were extracted using KEGG Orthology (KO) terms for those respective pathways (Kanehisa et al., 2017). Missing genes were manually obtained from the D. viridis transcriptome (Srirangan et al., 2015) using Chlamydomonas reinhardtii orthologs from the KO Database (November 27, 2017). Putative D. viridis mRNA sequences were manually translated with ExPASy translate tool (Artimo et al., 2012) and submitted for KO annotation using BlastKOALA (Kanehisa et al., 2016) to confirm that they were appropriate orthologs.

Cell density and diameter were measured to determine changes in cell division rates and cell volumes under each nitrogen treatment. The GLU and STAR treatments showed identical responses, with a 90% increase in cell density from 3 to 27 h (Figure 1A). GLN and NO3 supplementation show a 110% increase in cell density from 3 to 27 h which is significantly different from the nitrogen starved (GLU, STAR) conditions. Cells in both nitrogen replete treatments (GLN, NO3) continued to divide until 51 h, a 140% increase, while the nitrogen starved treatments (GLU, STAR) showed only a 15% increase (Figure 1A). Cultures containing GLN or NO3 showed an increase in cell size at 27 h over nitrogen-starved (GLU, STAR) cells; however, at 51 h cells provided with GLN were the same size as starved (GLU, STAR) cells, while the NO3-grown cells were larger (Figure 1B).

Medium pH of the growing cultures was monitored to determine whether the nitrogen source accelerated or attenuated biological medium alkalization over time. All cultures experienced an initial decrease in pH as the INOC was at a higher pH than the fresh media. Each culture condition resulted in increased pH over time. At 51 h the nitrogen-starved conditions (GLU, STAR) had lower pH levels than the nitrogen-sufficient conditions (NO3, GLN) with NO3 showing increased alkalization beyond that of the GLN treatment (Figure 1C).

Dunaliella viridis cells supplemented with GLN showed an intermediate amount of chlorophyll per cell in comparison to NO3 and nitrogen starved (GLU, STAR) conditions. The nitrogen-starvation conditions (GLU, STAR) show a trend of decreasing cellular chlorophyll content whereas the nitrogen-sufficient (NO3, GLN) conditions show greater chlorophyll content per cell over the nitrogen starved (GLU, STAR) conditions (Figure 1D). At 27 h, GLN, and NO3 supplemented cultures have ∼67% more chlorophyll per cell than the starvation conditions (GLU, STAR). At 51 h, cells in the NO3 culture maintain their chlorophyll content while cells growing in the GLN culture show a decrease in chlorophyll content that is similar to the chlorophyll content the cells had at 3 h after inoculation. GLN-grown cells still maintain more chlorophyll than nitrogen-starved cells (GLU, STAR) (Figure 1D).

The changes in cellular soluble protein content during growth in the different media shows a very similar pattern to that of the chlorophyll content. Nitrogen starvation conditions (GLU, STAR) resulted in a decrease in cellular protein content over time, while nitrogen sufficiency (NO3, GLN) showed increased protein content. However, unlike chlorophyll, the protein content in GLN grown cells at 51 h was reduced to the equivalent level of the nitrogen starved cells (Figure 1E).

In microalgae, changes in carbon storage molecules are common under variable nitrogen conditions, especially the increase in carbon storage under nitrogen starvation (Schmollinger et al., 2014; Goncalves et al., 2016b). A similar response pattern was observed at 27 h after inoculation for total carbohydrate (Figure 1F) and neutral lipids (Figure 1G), where all carbon storage components showed a doubled or higher quantity when the cells were grown under nitrogen starvation (GLU, STAR).

After 51 h in the respective culture media, the cells show differences in the accumulation of carbon storage components. The total carbohydrate content of GLN-grown cells increased while the NO3-grown cells showed a decrease in total carbohydrate levels (Figure 1F). The GLN-grown cells accumulated an intermediate carbohydrate quantity between nitrogen starvation and nitrate conditions similar to chlorophyll content (Figure 1D). For neutral lipid content, there were slight decreases in neutral lipid content from 27 to 51 h in the nitrogen-replete conditions (NO3, GLN), while the cells grown under nitrogen-starvation conditions (GLU, STAR) continue to accumulate neutral lipids. Neutral lipid content in the cells of the NO3 and GLN supplemented cultures was not different at either 27 or 51 h (Figure 1G).

We compared the different transcriptome profiles over time (INOC, 3, 27, and 51 h) (Supplementary Figure S3 and Supplementary Table S2) and between treatments at each time point (NO3, GLN, and GLU) (Figure 2). Overall, 7019 transcripts or 38% of the D. viridis transcriptome was differentially expressed (DE) in at least one comparison. At 3 h there were 64 uniquely DE genes between the three nitrogen treatments which increased to 3927 DE genes by 27 h (Figure 2). At 51 h 21.9% of the transcriptome or 4048 genes were DE. At 3 h the NO3 and STAR transcriptomes were significantly different from the GLN transcriptome, but not from each other. This changes at 27 h when there are 3562 DE gene between NO3 and STAR. At 27 h, only 20 DE genes were identified between GLN and NO3 treatments with 6 DE genes involved in nitrogen metabolism. However, the 2748 DE genes between STAR and GLN largely overlapped the NO3 vs. STAR comparison (Figure 2). At 51 h, NO3, and STAR maintained a large amount of DE genes at 3719. However, at 51 h DE genes between GLN and NO3 increased to 694 genes and the DE genes between GLN and STAR decreased to 1330 genes (Figure 2).

When analyzing the genes involved in nitrogen assimilation, DE was determined by comparison of all transcript profiles to INOC so that the magnitude of response could be compared within nitrogen sources and across time. Overall, GLN and STAR show more dynamic responses over time than NO3, and since the INOC was grown with NO3 this response was expected.

The STAR condition at 3 h resulted in upregulation of several nitrate/nitrite porters (NNP), all ammonium transporters (AMTs), nitrate assimilation related 2 (NAR2), and the rhythm of chloroplast 40 (ROC40) transcription factor (Figure 3). At 27 h, all the NNP genes were downregulated in comparison to 3 h, while the AMTs, cytosolic glutamine synthetase (GS), NADH-dependent GLN oxoglutarate aminotransferase (GOGAT), and several regulatory proteins and transcription factor genes were upregulated. The upregulation of these genes was either maintained or increased at the 51 h time point, while the expression of the downregulated genes was maintained below the INOC expression levels (Figure 3).

The GLN condition which represents glutamine-derived NH₄⁺ or potentially glutamine resulted in downregulation of several NNPs, formate/nitrite transporter (FNT) 2, nitrate reductase, nitrite reductase, NAR2, and AMT1.1 at 3 h (Figure 3). The only gene upregulated was ROC40. In GLN at 27 h, most of the transcriptional repression was released, and several genes returned up to INOC levels or were upregulated above INOC. ROC40 was downregulated back to the inoculum level which was maintained through 51 h. For genes upregulated at 27 h, most were downregulated at 51 h, which represented a return to expression levels that were similar to their levels at 3 h. However, AMT1.1 was upregulated at 51 h along with other NH₄⁺ metabolism genes (Figure 3).

Since nitrogen starvation is a known inducer of lipid accumulation in algae, transcriptional regulation of TAG synthesis genes was analyzed for DE at all time points in STAR vs. NO3 samples. Genes for fatty acid synthesis were downregulated over time or remained unchanged in STAR cultures with malonyl-CoA ACP transacylase and ketoacyl-ACP synthase III being the most downregulated at 51 h (Figure 4). Genes involved in TAG metabolism showed a mixed response with 4 downregulated genes and 7 upregulated genes at 51 h (Figure 5). The fatty acid desaturase genes FAD6-2 and FAD7, as well as UDP-sulfoquinovose synthase and glycerol-3-P acyltransferase were reduced to half their transcript abundance. The upregulated genes encode digalactosyldiacylglycerol synthase, major lipid droplet protein, phosphatidate phosphatase, diacylglycerol kinase, and three lipases. The phosphatidate phosphatase gene PAP2 shows a large increase in transcript abundance at 27 h of about 180-fold and at 51 h of 128-fold (Figure 5), and is the most DE gene in the fatty acid/TAG metabolism pathway although its overall expression is low.

The regulation of carbon metabolism under nitrogen starvation is a salient consequence of a changing nitrogen environment. Therefore DE at the 51 h time point was compared between STAR and NO3 treatments. Expression levels between 27 and 51 h were nearly identical, so the 51 h time point was analyzed (Supplementary Data Sheet 1). The Calvin-Benson cycle shows mixed regulation with 9 genes being significantly DE. Ribose 5-phosphate isomerase 1, aldolase 1, and glyceraldehyde 3-phosphate dehydrogenase 1 are the most highly DE genes of the Calvin-Benson cycle with Log₂FC values of, respectively, −2.4, 3.1, and 4.8 (0.2−, 8.6−, and 28-fold) changes in STAR over NO3 (Figure 6).

Regarding the Krebs cycle, most DE genes encoded enzymes involved in phosphoenolpyruvate (PEP), pyruvate, and oxaloacetate conversion (Figure 6). Downregulation of pyruvate kinase 2, mitochondrial malate dehydrogenase 1, and malic enzyme 3 was observed, although other isoforms of the same genes showed no change. More total genes were upregulated with pyruvate phosphate dikinase, PEP carboxylase, PEP carboxykinase, and mitochondrial and cytosolic malate dehydrogenase (M-MDH3, and C-MDH1) showing increases in transcription (Figure 6).

Lipid and starch metabolism branch off the Calvin and Krebs Cycles. The entry point for lipid metabolism, ACCase, shows downregulation in STAR (Figures 4, 6). Both starch synthesis and degradation are upregulated in STAR with the synthesis enzymes, starch synthase, and starch branching enzyme being more highly upregulated than the degradation enzymes α-amylase and starch phosphorylase (Supplementary Data Sheet 1).

Carbonic anhydrase (CA) transcription was also examined over time in comparison to INOC in both NO3 and STAR. Universally α-CAs and θ-CAs are downregulated between inoculation to the 3 h time point (Figure 6). For α-CAs, downregulation continued under STAR while in NO3 the α-CAs were upregulated at 27 and 51 h. The largest changes were observed for α-CA1 at 51 h which reached a Log₂FC of −6.4 (0.01-fold) in STAR, and a Log₂FC of 1.9 (3.7-fold) in NO3. The θ-CAs showed different patterns of expression, but in general by 51 h, expression of θ-CAs was higher in NO3 than STAR cultures (Figure 6).

From 27 to 51 h, the number of DE genes between GLN and NO3 increased from 20 to 694 genes (Table 1). At this junction metabolic shifts were observed, warranting a closer look at patterns of expression of the 694 genes to discover what might be causing these shifts. 350 genes changed in response to GLN, whereas only 34 changed for NO3 (Table 1). Of the 34 DE NO3 genes, 23 were upregulated in NO3 and did not change in GLN. These genes consisted of heavy metal transporters and chaperones as well as carbonic anhydrases (CAs). 11 of the 34 genes were DE in both NO3 and GLN and regardless of expression pattern, all 11 genes had lower expression in GLN than NO3. Of the 339 genes that were only DE in GLN, 113 were downregulated and 226 were upregulated. The downregulated genes consisted of some nitrogen assimilation genes, protein synthesis and degradation, stress tolerance genes, porphyrin adjacent proteins, and lipid transport. The upregulated genes consisted of nitrogen assimilation genes, arginine biosynthesis, glycolysis, and protein degradation (Table 1).

A pattern emerged from the metabolic data: at 3 h all treatments were the same, at 27 h GLN and NO3 were equal and separate from STAR, and at 51 h NO3 and STAR maintained separation, but GLN shifted from the equal of NO3 toward the STAR condition. This was reflected in the cell diameter, pH, chlorophyll, protein, and carbohydrate data. Cell density and neutral lipid data maintained the separation of the 27 h time point. Similar results were demonstrated previously at 144 h after inoculation in D. viridis (Murphree et al., 2017). While much of the metabolic data showed a starvation response at 51 h for GLN, the equal cell density between GLN and NO3 suggested that NH₄⁺ (GLN) grown cells had altered metabolism.

NH₄⁺ has been shown to alter the metabolism of D. salina (Giordano and Bowes, 1997) and the transcriptional data at first suggested that this was true for D. viridis. However, over time the transcriptome profiles of GLN-grown cultures resembled less the patterns of the NO3-grown cultures and more like the STAR cultures (Figure 2). For example, the down regulation of the photosynthetic apparatus and chlorophyll biosynthesis is a hallmark of nitrogen deficiency (Schmollinger et al., 2014; Park et al., 2015; Tan et al., 2016) and is represented in the STAR and GLN conditions but not in NO3 (Table 1). This is seen in the metabolic data with chlorophyll content and protein content decreasing in GLN in comparison to NO3 (Figures 1D,E). The upregulation of arginine biosynthesis genes in NH₄⁺ starved C. reinhardtii (Park et al., 2015) was comparably detected in the NO₃⁻ starved (STAR) and NH₄⁺ starved (GLN) conditions of D. viridis (Table 1 and Supplementary Data Sheet 1). Modulation of the nitrogen assimilation genes accented the presence of NH₄⁺ in the media as downregulation of genes at 3 h, release of the repression by 27 h, and a small downregulation at 51 h was observed (Figure 3). Repression of nitrogen assimilation genes in the presence of NH₄⁺ is observable from cultures of C. reinhardtii (Camargo et al., 2007; Giordano and Raven, 2014; Sanz-Luque et al., 2015) and the modulation of these genes in D. viridis suggests that most or all NH₄⁺ released from GLN was likely consumed by the cells by 27 h. This hypothesis is supported by the observation that GLN and STAR conditions had comparable transcriptional and metabolic phenotypes at 51 h. The evidence suggests that the release of NH₄⁺ from GLN was too slow to maintain growth for 51 h and that the metabolic effects seen in the GLN condition are the consequence of nitrogen starvation rather than NH₄⁺, the presence of residual GLN, or pyroglutamate.

Some of the most transcriptionally active sets of genes under these variable nitrogen conditions are involved in nitrogen assimilation. Modulation of nitrogen assimilation genes by NH₄⁺ and nitrogen starvation is a well-characterized phenomenon in algae, especially in C. reinhardtii (Camargo et al., 2007; Giordano and Raven, 2014; Sanz-Luque et al., 2015). However, the transcriptional regulation of the nitrogen assimilation genes in Dunaliella spp. is little described except for qPCR data (Li et al., 2007; Song et al., 2011). As a nitrogen source, NH₄⁺ requires less energy than NO₃⁻ to stably incorporate into a carbon scaffold (Arnold et al., 2015); however, NH₃ is a potent electrochemical gradient uncoupler when present in cells, and in sufficiently high amounts is cytotoxic.

Dunaliella viridis responds strongly to the presence of glutamine-derived NH₄⁺ by attenuating or increasing the expression of genes involved in NH₄⁺ transport and assimilation. The short-term upregulation of NNPs was most likely an attempt to scavenge NO₃⁻/NO₂⁻, but with no additional NO₃⁻/NO₂⁻, they were downregulated in favor of the NH₄⁺ uptake and metabolism genes (Schmollinger et al., 2014; Arnold et al., 2015). This is not unexpected, as NH₄⁺ is less costly for the cells to incorporate than NO₃⁻. It is plausible that the NIT2 transcription factor regulates these genes under nitrogen starvation as shown in C. reinhardtii (Camargo et al., 2007), although it should be noted that D. viridis NIT2 is not repressed by NH₄⁺ like in C. reinhardtii (Camargo et al., 2007).

NH₄⁺ released by GLN was associated with downregulation of NNPs, FNT nitrite transporters, NR, NiR, NAR2, and AMT1.1 (Figure 3). The downregulation of NO₃⁻/NO₂⁻ assimilation genes in GLN suggests that D. viridis has a mechanism to prioritize assimilation of NH₄⁺ over NO₃⁻. The downregulation of AMT1.1 in the presence NH₄⁺ potentially confounds this explanation; however, the toxicity of NH₄⁺/NH₃ may explain this response. In the presence of NH₄⁺, D. viridis may in fact limit the uptake of NH₄⁺ in order to minimize the cytotoxic effects of electrochemical decoupling, while allowing enough NH₄⁺/NH₃ into the cell that it can be assimilated into amino acids via the action of GS/GOGAT.

Like Dunaliella tertiolecta (Tan et al., 2016; Pick and Avidan, 2017), nitrogen starvation leads to an increase in both carbohydrate and neutral lipid content in D. viridis with a preference toward carbohydrate over lipid. It also appears that carbohydrate storage is a quicker or larger response in D. viridis since the partially starved GLN culture accumulated more carbohydrates than it did neutral lipids (Figures 1F,G). However, a time-course with higher temporal resolution would be necessary to identify the rapidity and magnitude of the metabolic shifts caused by these different treatments.

The mechanisms governing the ability of green algae species to control the partitioning of carbon storage under nitrogen starvation remain unclear (Goncalves et al., 2016b). C. reinhardtii and D. tertiolecta are the most closely related species to D. viridis for which there are salient investigations of carbon accumulation under nitrogen starvation. Comparable transcriptional patterns have been observed between these three algae, however, notable differences exist.

Accumulation of TAG in D. viridis is associated with transcriptional upregulation of the major lipid drop protein and the two transcription factors NRR1 and ROC40. This is a similar response as that of C. reinhardtii (Moellering and Benning, 2010; Boyle et al., 2012; Goncalves et al., 2016a). However, under nitrogen starvation C. reinhardtii upregulates several fatty acid synthesis genes while D. tertiolecta and D. viridis downregulate fatty acid synthesis genes (Figure 4; Yao et al., 2017). This observation leads to the erroneous prediction that less fatty acids would be synthesized and available for TAG synthesis in D. tertiolecta and D. viridis. Instead, the initial increase in TAG in Dunaliella spp. is suggested to come from recycling membrane lipids into TAG (Tan et al., 2016). Under nitrogen starvation, some algae are known to break down chloroplast membranes (Preininger et al., 2015), which are a source of fatty acids for TAG synthesis. These reactions are catalyzed by diacylglycerol acyltransferase genes (DGTT, DGAT, PDAT, and PGD) but no D. viridis homologs for these genes showed any change in expression. However, there were expression increases in putative lipase genes, which could be mobilizing fatty acids for incorporation into TAG (Boyle et al., 2012). Unfortunately, since only neutral lipid levels were measured in D. viridis, the difference between membrane recycling and de novo lipid synthesis cannot be distinguished, although it is certainly possible that both pathways occur.

Carbohydrates have been reported to be the major carbon storage compound in Dunaliella spp. (Slocombe et al., 2015; Tan et al., 2016; Figure 1F). The maintenance of accumulated carbohydrates in STAR seems to correlate with the upregulation of the starch anabolic enzymes, starch synthase and 1,4-α-glucan branching enzyme, which are slightly more upregulated than the catabolic enzymes, α-amylase and starch phosphorylase. In D. tertiolecta starch is primarily synthesized in the beginning of nitrogen starvation and then is degraded for fatty acid synthesis and TAG production under continuing nitrogen starvation (Pick and Avidan, 2017). Major turnover between starch and lipids took place after 72 h (Pick and Avidan, 2017) when all starch metabolism genes were upregulated (Tan et al., 2016). At 51 h, D. viridis both up and down regulates starch metabolism (Supplementary Data Sheet 1). This difference may be a consequence of the earlier time point used in this study, but could also stem from the use of continuous light as a growth condition, which has been shown to be disruptive of the typical diurnal starch metabolism regulation in D. viridis (Srirangan et al., 2015).

The Krebs Cycle is a common regulatory hub for carbon partitioning as it serves to provide most carbon skeletons for amino acid synthesis and degradation. In D. tertiolecta and C. reinhardtii, all enzymes of the Krebs Cycle are upregulated under nitrogen starvation (Park et al., 2015; Tan et al., 2016), whereas D. viridis shows a mixture of up and down-regulation of the Krebs Cycle. It is unclear as to where this mixed regulation originates from although differences in experimental design could be a factor. Under continuous light, D. viridis shows highly modified metabolism as continuous light eliminates diurnal cycles and also desynchronizes the cell cycle (Srirangan et al., 2015), so an interaction between light and nitrogen stress could cause the differences in Krebs Cycle regulation.

Two key metabolites derived from glycolysis, phosphoenolpyruvate (PEP) and pyruvate (PYR), are used to create acetyl-CoA, which can feed into the Krebs Cycle or fatty acid synthesis. As such, they are important regulatory molecules in partitioning of carbon into carbohydrates or lipids (Polle et al., 2014). Under nitrogen starvation, D. tertiolecta and C. reinhardtii both show an upregulation of the pyruvate dehydrogenase complex which creates acetyl-CoA from PYR and allows for the synthesis of fatty acids (Park et al., 2015; Tan et al., 2016). However, D. viridis does not show this regulatory pattern and the pyruvate dehydrogenase complex shows no change in transcription (Figure 6). When looking at the transcriptional program around PEP and PYR, it appears that D. viridis favors starch production by avoiding the build-up of PYR which can be converted into acetyl-CoA and eventually fatty acids. This avoidance of PYR comes in several different strategies. The first is stopping PYR production by downregulation of pyruvate kinase and malic enzymes. The upregulation of pyruvate phosphate dikinase suggests that if PYR is created it is converted into PEP to avoid conversion into acetyl-CoA. Another strategy is the upregulation of PEP carboxylase to convert PEP into oxaloacetate for the Krebs Cycle. The concurrent upregulation of PEP carboxykinase suggests that oxaloacetate being converted to PEP could be used to prompt gluconeogenesis and eventual starch synthesis. Although PYR is a critical metabolite, the transcriptional data suggest that D. viridis attempts to limit the pool of PYR by limiting production or converting PYR into alternate metabolites other than acetyl-CoA.

A disadvantage of batch culture is that conditions in the culture flask change over time as the organism interacts with the environment. Dunaliella alkalizes the environment as part of the dissolved inorganic carbon uptake mechanism (Shiraiwa et al., 1993). This creates a trend of increasing medium pH as cell density increases. Within the subset of DE genes at 51 h between GLN and NO3, several genes appeared to be regulated by the difference in pH between cultures, as they were DE in all nitrogen conditions from inoculum to 3 h when the pH fell from 9.2 to 8.2. They were also DE in the 51 h over 27 h comparison of the NO3 and GLN conditions when the pH between NO3 and GLN is significantly different (Table 1).

Carbonic anhydrases are major constituents of the carbon concentrating mechanism (CCM) in algae (DiMario et al., 2018). In C. reinhardtii, CAs are upregulated under low CO₂ conditions (Mackinder et al., 2017). In Dunaliella spp., low CO₂ and high salt induce CA expression (Jeon et al., 2016) and in D. viridis continuous light induces α-CA expression (Srirangan et al., 2015). Nitrogen starvation also appears to be a regulating factor in D. viridis as the α-CA and θ-CA genes were all downregulated by nitrogen starvation. θ-CAs are homologs of the C. reinhardtii LCIB/LCIC proteins which are chloroplast stroma-localized and predicted to convert escaped CO₂ from the pyrenoid back into HCO₃⁻ under low CO₂ levels (Wang et al., 2015; Mackinder et al., 2017). It is likely that the θ-CAs are responding to pH because both GLN and NO3 had high pH values even though GLN had entered nitrogen starvation by 51 h. Interestingly, this increase in CA expression over time suggests CO₂ starvation, even though the cells only grew for 51 h in a media with 50 mM HCO₃⁻. It is possible that as the pH rises D. viridis interprets the shift in carbon species from H₂CO₃ to HCO₃⁻ to CO₃²⁻ as CO₂ starvation.

The CCM also requires the transport of inorganic carbon across multiple membranes. Recently, two thylakoid-localized bestrophin chloride channels were found to interact with LCIB/LCIC (θ-CAs) (Mackinder et al., 2017). Since human bestrophins are permeable to HCO₃⁻ and Cl⁻ (Qu and Hartzell, 2008), Mackinder predicted that the bestrophins could function as the unknown HCO₃⁻ transporter from the thylakoid into the pyrenoid (Mackinder et al., 2017). Three D. viridis bestrophin homologs were upregulated over time as the pH increased in conjunction with the θ-CA homologs suggesting that this association and mechanism may also exist for the D. viridis CCM (Supplementary Data Sheet 1). In C. reinhardtii, the HLA3/LCI1 complex and LCIA reside, respectively, on the plasma and chloroplast membranes and are critical for proper transport of HCO₃⁻ (Mariscal et al., 2006; Yamano et al., 2015; Mackinder et al., 2017). However, none of the D. viridis homologs are regulated by increasing pH and the LCIA homolog (FNT2) appears to be solely reactive to NH₄⁺ repression. Other putative bicarbonate transporters, Cl⁻/HCO₃⁻ antiporters (Bonar and Casey, 2008) and a CIA8 homolog (Machingura et al., 2017) show no difference in transcription in all conditions. This suggests that increasing pH may not be interpreted as CO₂ limitation by D. viridis or that CCM regulation is different between C. reinhardtii and D. viridis.