Proteomic and Transcriptomic Patterns during Lipid Remodeling in Nannochloropsis gaditana

Nutrient limited conditions are common in natural phytoplankton communities and are often used to increase the yield of lipids from industrial microalgae cultivations. Here we studied the effects of bioavailable nitrogen (N) and phosphorus (P) deprivation on the proteome and transcriptome of the oleaginous marine microalga Nannochloropsis gaditana. Turbidostat cultures were used to selectively apply either N or P deprivation, controlling for variables including the light intensity. Global (cell-wide) changes in the proteome were measured using Tandem Mass Tag (TMT) and LC-MS/MS, whilst gene transcript expression of the same samples was quantified by Illumina RNA-sequencing. We detected 3423 proteins, where 1543 and 113 proteins showed significant changes in abundance in N and P treatments, respectively. The analysis includes the global correlation between proteomic and transcriptomic data, the regulation of subcellular proteomes in different compartments, gene/protein functional groups, and metabolic pathways. The results show that triacylglycerol (TAG) accumulation under nitrogen deprivation was associated with substantial downregulation of protein synthesis and photosynthetic activity. Oil accumulation was also accompanied by a diverse set of responses including the upregulation of diacylglycerol acyltransferase (DGAT), lipase, and lipid body associated proteins. Deprivation of phosphorus had comparatively fewer, weaker effects, some of which were linked to the remodeling of respiratory metabolism.


Introduction
Bioavailable nitrogen and phosphorus are essential macronutrients required by microalgae for optimal, balanced growth. In the oceans, the effects of nitrogen (N) and phosphorus (P) supply on phytoplankton physiology and elemental stoichiometry are well recognized [1,2], where nutrient abundance often controls primary production, community structure, and ultimately the flux of matter and energy through ecosystems [3,4]. Many species of microalgae also have applications in biotechnology, where modulating the nutrient supply to intensive cell cultures is a common technique used to induce the accumulation of triacylglycerol (TAG) and secondary carotenoids [5,6]. Understanding how microalgae respond to changes in nutrient availability, especially the supply of N and P, is therefore valuable for characterizing their behavior in natural and industrial settings.
Protein accounts for a large share of cellular N, but nitrogen is also a component of nucleic acids (RNA and DNA) and chlorophyll. Phosphorus is required in lower amounts, but is nevertheless coupled with high biomass turnover and dilution with fresh medium (Figure 1a). Nitrate-starved cultures showed a gradual increase in cell density toward the end of the experiment, a result of maintaining constant turbidity whilst the cells experienced chlorosis (loss of pigmentation). The Ncultures experienced an immediate reduction in growth rate to 0.11 ± 0.02 d −1 at day 3 and 0.05 ± 0.02 d −1 at day 5. In comparison the onset of P-conditions was more dampened with the growth rate 0.49 ± 0.06 d −1 at day 3 and 0.44 ± 0.07 d −1 at day 5. Analysis of fatty-acids showed a substantial increase in TAG comprised primarily of C16:0 and C16:1 fatty-acids in the N-treatments (Figure 1b). After 5 days in N-conditions, fatty-acids in TAG comprised 21.4% of the cell dry weight but remained at only 1.0% and 2.2% of the dry weight in the control (C) and P-treatments, respectively. The longchain PUFAs eicosapentanoic acid (EPA, C20:5n-3) and arachidonic acid (ARA, C20:4n-6) were mostly present in the polar lipids. At day 5 the EPA accounted for 26.5% and ARA for 2.5% of total fatty acids (TFA) in control cultures. In N-cultures the EPA content was reduced substantially to 6.3% TFA after 5 days, due to the reduction of polar lipids and the accumulation of fatty acids in TAG.
(a) (b) Figure 1. (a) Image of the flat-plate photobioreactors operated as turbidostats including measurement of pH, temperature, CO2 concentration in the sparging gas, and turbidity. The growth rate (d -1 ) and the cell density (g L −1 ) are shown with the changes in the dissolved extracellular nitrate (NO3 -) and phosphate (PO4 3-) concentrations (mean ± sd, n = 4). (b) Lipid analysis including the fatty-acid profiles (left) of polar and neutral lipids (TAG) in control, N-, and P-treatments after 3 and 5 days of the experiment, as fatty-acid methyl-esters-FAME (mg/g dry weight). The total FAMEs in control, N-, and P-treatments after 3 and 5 days of the experiment (right). Data are the mean ± sd of n = 4 experimental replicates (except n = 3 for N-treatments at day 5). profiles (left) of polar and neutral lipids (TAG) in control, N-, and P-treatments after 3 and 5 days of the experiment, as fatty-acid methyl-esters-FAME (mg/g dry weight). The total FAMEs in control, N-, and P-treatments after 3 and 5 days of the experiment (right). Data are the mean ± sd of n = 4 experimental replicates (except n = 3 for N-treatments at day 5).

Identification and Differential Expression of Proteins and Their Transcripts
In total 3423 proteins were identified across all of the tested conditions. After 3 days of N-deprivation 1543 of these proteins were significantly differentially regulated, whilst in P-treatments only 113 proteins were significantly differentially regulated ( Figure 2). Transcriptome analysis showed that after 3 days of N-treatment, 1448 of the 10,496 genes in the B31 genome were differentially expressed, where 528 transcripts were upregulated and 920 were downregulated. After 5 days of N-treatment, the number of differentially expressed genes (DEGs) increased to 2371, where 859 were upregulated and 1512 were downregulated. Phosphorus depletion resulted in far fewer DEGs, where only 52 genes were upregulated and two were downregulated after 3 days, increasing to a total of 122 DEGs after 5 days. Principal components analysis showed that in the protein dataset there was distinct clustering of N-samples, but much weaker demarcation between P-and control treatments ( Figure S3). Principal components analysis of the transcriptomic data indicated clear divergence between each of the treatments after 3 days, strengthening further after 5 days.

Identification and Differential Expression of Proteins and Their Transcripts
In total 3423 proteins were identified across all of the tested conditions. After 3 days of Ndeprivation 1543 of these proteins were significantly differentially regulated, whilst in P-treatments only 113 proteins were significantly differentially regulated ( Figure 2). Transcriptome analysis showed that after 3 days of N-treatment, 1448 of the 10,496 genes in the B31 genome were differentially expressed, where 528 transcripts were upregulated and 920 were downregulated. After 5 days of N-treatment, the number of differentially expressed genes (DEGs) increased to 2371, where 859 were upregulated and 1512 were downregulated. Phosphorus depletion resulted in far fewer DEGs, where only 52 genes were upregulated and two were downregulated after 3 days, increasing to a total of 122 DEGs after 5 days. Principal components analysis showed that in the protein dataset there was distinct clustering of N-samples, but much weaker demarcation between P-and control treatments ( Figure S3). Principal components analysis of the transcriptomic data indicated clear divergence between each of the treatments after 3 days, strengthening further after 5 days. scale on the y-axis, and for transcripts these are the adjusted p-values from the DESeq2 methodology. Proteins determined significantly differently regulated at corrected thresholds p < 0.022 (N-/C treatments) or p < 0.002 (P-/C treatments) are indicated in the uppermost segment. Proteins differentially expressed at p < 0.050, but not reaching the adjusted threshold, are indicated in the central segment.

Correlation Between the Nannochloropsis Proteome and Transcriptome
The global patterns in protein and mRNA abundance were examined using three complimentary approaches. First, the correlation between the log2 fold changes (L2fc) of mRNA transcripts and their corresponding proteins was performed (Figure 3a). The N-/C treatment yielded moderate correlation (R 2 = 0.25), whilst the correlation in P-/C treatments was much weaker (R 2 = 0.08). Our second method combined data for all observations (C, N-, and P-treatments) together, and a linear mixed-effects model was used to describe the relationship between mRNA abundance (log (RPKM)) and protein abundance (log (Mol%)) across all gene/protein accessions (Figure 3b). For Figure 2. Volcano plots showing the differential expression of proteins and transcripts in the nitrogen starved (N-) and phosphorus starved (P-) treatments, vs. controls. The x-axis displays the log 2 fold change (L 2 fc) of protein or transcript expression, where positive values indicate upregulated proteins and negative values correspond to downregulated proteins. The p-values are presented on -Log 10 scale on the y-axis, and for transcripts these are the adjusted p-values from the DESeq2 methodology. Proteins determined significantly differently regulated at corrected thresholds p < 0.022 (N-/C treatments) or p < 0.002 (P-/C treatments) are indicated in the uppermost segment. Proteins differentially expressed at p < 0.050, but not reaching the adjusted threshold, are indicated in the central segment.

Correlation between the Nannochloropsis Proteome and Transcriptome
The global patterns in protein and mRNA abundance were examined using three complimentary approaches. First, the correlation between the log 2 fold changes (L 2 fc) of mRNA transcripts and their corresponding proteins was performed (Figure 3a). The N-/C treatment yielded moderate correlation (R 2 = 0.25), whilst the correlation in P-/C treatments was much weaker (R 2 = 0.08). Our second method combined data for all observations (C, N-, and P-treatments) together, and a linear mixed-effects model was used to describe the relationship between mRNA abundance (log (RPKM)) and protein abundance (log (Mol%)) across all gene/protein accessions (Figure 3b). For comparative purposes, a conventional Pearson's R 2 of 0.31 was also calculated for the same data, indicating moderate positive correlation between transcript and protein abundance. Our third method fitted individual linear regression models to each gene/protein pair, yielding 2576 regression models. The distribution of R 2 values from these linear models are presented in Figure 3c (upper panel), and for only the subset of proteins which showed significant differential expression (Figure 3c lower panel). The median R 2 for all accessions was 0.29, but increased substantially to R 2 = 0.58, with a shoulder at R 2~0 .8, when only the significantly differentially expressed proteins were included. For those significantly differentially expressed proteins, 79% of the gene/protein correlation slopes were positive, the remaining 21% were negative (Figure 3d). Together, these three alternative approaches characterize a moderate but detectible cell-wide association between mRNA and protein expression in these data. comparative purposes, a conventional Pearson's R 2 of 0.31 was also calculated for the same data, indicating moderate positive correlation between transcript and protein abundance. Our third method fitted individual linear regression models to each gene/protein pair, yielding 2576 regression models. The distribution of R 2 values from these linear models are presented in Figure 3c (upper panel), and for only the subset of proteins which showed significant differential expression ( Figure  3c lower panel). The median R 2 for all accessions was 0.29, but increased substantially to R 2 = 0.58, with a shoulder at R 2~0 .8, when only the significantly differentially expressed proteins were included. For those significantly differentially expressed proteins, 79% of the gene/protein correlation slopes were positive, the remaining 21% were negative ( Figure 3d). Together, these three alternative approaches characterize a moderate but detectible cell-wide association between mRNA and protein expression in these data.

The Effect of Nitrogen and Phosphorus Stress on Subcellular Proteome Remodeling
To investigate large-scale changes in subcellular proteomes under N-and P-conditions, we examined the overall fold changes of proteins after grouping them into their respective cellular locations. For most compartments, N-treatments exhibited greater variance in protein abundance than P-treatments ( Figure 4). Proteins associated with the plastid were mostly downregulated under nitrogen deprivation, with a median L 2 fc of −0.42. Proteins localized to the mitochondrion, membranes and the endoplasmic reticulum (ER) also displayed variation in L 2 fc, but their median fold changes each remained around zero (L 2 fc 0.00, 0.02, and −0.08, respectively). The data indicate that under N-conditions the plastid proteome shrank, whilst the ER, mitochondrial and membrane proteins were remodeled but did not substantially change overall size. In P-treatments there were no substantial shifts in expression of any of the subcellular proteomes, and variation in L 2 fc was much lower than those in N-treatments, indicating only limited remodeling.

The Effect of Nitrogen and Phosphorus Stress on Subcellular Proteome Remodeling
To investigate large-scale changes in subcellular proteomes under N-and P-conditions, we examined the overall fold changes of proteins after grouping them into their respective cellular locations. For most compartments, N-treatments exhibited greater variance in protein abundance than P-treatments ( Figure 4). Proteins associated with the plastid were mostly downregulated under nitrogen deprivation, with a median L2fc of −0.42. Proteins localized to the mitochondrion, membranes and the endoplasmic reticulum (ER) also displayed variation in L2fc, but their median fold changes each remained around zero (L2fc 0.00, 0.02, and −0.08, respectively). The data indicate that under N-conditions the plastid proteome shrank, whilst the ER, mitochondrial and membrane proteins were remodeled but did not substantially change overall size. In P-treatments there were no substantial shifts in expression of any of the subcellular proteomes, and variation in L2fc was much lower than those in N-treatments, indicating only limited remodeling.

Functional Enrichment Analysis of Differentially Expressed Proteins and Transcripts
To capture the main patterns in gene expression and protein abundance, gene ontology (GO) and KEGG pathway ontology (KO) terms were examined ( Figures 5 and 6). Under N-conditions changes in the proteome and transcriptome were mostly concordant, where downregulation of proteins and mRNA transcripts was observed in protein translation processes (GO:0006412), proteinchromophore linkage (GO:0018298), and light-independent chlorophyll biosynthesis (GO:0036068), together with photosynthesis (GO:0015979) and its light-dependent (GO:0009765) and lightindependent reactions (GO:0019685). Fewer gene and protein GO categories were significantly upregulated in N-treatments, but genes and proteins with roles in amine metabolism (GO:0009308), the tricarboxylic acid cycle (GO:0006099), and nucleotide catabolism (GO:0009166) were increased.

Functional Enrichment Analysis of Differentially Expressed Proteins and Transcripts
To capture the main patterns in gene expression and protein abundance, gene ontology (GO) and KEGG pathway ontology (KO) terms were examined ( Figures 5 and 6). Under N-conditions changes in the proteome and transcriptome were mostly concordant, where downregulation of proteins and mRNA transcripts was observed in protein translation processes (GO:0006412), protein-chromophore linkage (GO:0018298), and light-independent chlorophyll biosynthesis (GO:0036068), together with photosynthesis (GO:0015979) and its light-dependent (GO:0009765) and light-independent reactions (GO:0019685). Fewer gene and protein GO categories were significantly upregulated in N-treatments, but genes and proteins with roles in amine metabolism (GO:0009308), the tricarboxylic acid cycle (GO:0006099), and nucleotide catabolism (GO:0009166) were increased.  In P-treatments, over-represented GO terms for proteomic and transcriptomic data were less concordant. The downregulation of proteins involved in translation (GO:0006412), protein stabilization (GO:0050821), D-ribose catabolic process (GO:0019303), and carbohydrate transport (GO:0008643), together with the upregulation of tricarboxylic acid cycle (GO:0006099) and glycolytic process (GO:0006096), was not echoed by the transcriptome ( Figure 5). After 5 days of phosphorus starvation, gene expression associated with tRNA (GO:0006418) and rRNA processing (GO:0006364) were also lowered, together with reductions in ribosome biogenesis (GO:0042254), ribosome assembly (GO:0000028), protein refolding (GO:0042026), and amino-acid biosynthesis (GO:0008652). Transcripts associated with amine metabolism (GO:0009308) were also downregulated after 5 days of P deprivation, contrasting with the upregulation of the same group during N deprivation. Upregulated gene clusters in P-treatments included increases in phosphate-ion transport (GO:0006817) and increases in transcripts associated with lipid catabolism (GO:0016042), ATP synthesis (GO:0015986, GO:0042773), and oxidative phosphorylation (GO:0006119).

Translation, Nitrogen Acquisition, and Metabolism
Under N-conditions, 12 of the 30 most downregulated proteins were ribosomal (Table 1), mostly 30S and 50S that are plastid-associated. The L2fc of all ribosomal proteins were examined, and we found that both plastidic ribosomes and ribosomal proteins of eukaryotic origin (40S and 60S) were downregulated after 3 days of N-conditions ( Figure S6). In P-treatments the expression of ribosomal proteins and their transcripts was not substantially changed. Both nitrate and nitrite reductase were among the most downregulated proteins in the N-treatments, highlighting the reduced investments in N acquisition from the extracellular environment. In P-treatments, over-represented GO terms for proteomic and transcriptomic data were less concordant. The downregulation of proteins involved in translation (GO:0006412), protein stabilization (GO:0050821), D-ribose catabolic process (GO:0019303), and carbohydrate transport (GO:0008643), together with the upregulation of tricarboxylic acid cycle (GO:0006099) and glycolytic process (GO:0006096), was not echoed by the transcriptome ( Figure 5). After 5 days of phosphorus starvation, gene expression associated with tRNA (GO:0006418) and rRNA processing (GO:0006364) were also lowered, together with reductions in ribosome biogenesis (GO:0042254), ribosome assembly (GO:0000028), protein refolding (GO:0042026), and amino-acid biosynthesis (GO:0008652). Transcripts associated with amine metabolism (GO:0009308) were also downregulated after 5 days of P deprivation, contrasting with the upregulation of the same group during N deprivation. Upregulated gene clusters in P-treatments included increases in phosphate-ion transport (GO:0006817) and increases in transcripts associated with lipid catabolism (GO:0016042), ATP synthesis (GO:0015986, GO:0042773), and oxidative phosphorylation (GO:0006119).

Translation, Nitrogen Acquisition, and Metabolism
Under N-conditions, 12 of the 30 most downregulated proteins were ribosomal (Table 1), mostly 30S and 50S that are plastid-associated. The L 2 fc of all ribosomal proteins were examined, and we found that both plastidic ribosomes and ribosomal proteins of eukaryotic origin (40S and 60S) were downregulated after 3 days of N-conditions ( Figure S6). In P-treatments the expression of ribosomal proteins and their transcripts was not substantially changed. Both nitrate and nitrite reductase were among the most downregulated proteins in the N-treatments, highlighting the reduced investments in N acquisition from the extracellular environment. Table 1. The 30 proteins with largest fold increase and 30 proteins with the largest fold decrease in the N-treatments (n = 4), relative to the controls (n = 4). Proteins annotated as "uncharacterized" were omitted and the p-values are from permutation tests. The suffix string of the Accession Number "9STRA" or "NANGC" refers to the B31 or CCMP526 N. gaditana reference proteomes, respectively. The reduced plastid proteome and diminished photosynthetic capacity associated with N starvation led us to hypothesize that enzymes involved with protein/amino-acid catabolism, nitrogen recycling, and recovery could be upregulated. Consistent with increases in amine metabolic processes (GO:0009308, Figure 5), an amine oxidase (W7TFN3_9STRA) was the second-most upregulated protein under N-conditions with an L 2 fc of +1.38 (Table 1). In P-treatments, the same protein was significantly downregulated (L 2 fc −0.32, p < 0.001). Further searching through the proteome revealed an additional six proteins annotated as amine oxidases, and of these a further two were significantly upregulated under N-conditions (Table S5). Additional proteins associated with amine metabolism were also significantly upregulated in N-treatments, including an amine dehydrogenase (W7TI92_9STRA) with an L 2 fc of +0.77.

Tricarboxylic Acid (TCA) Cycle, Glycolytic Processes, and Oxidative Phosphorylation
Evidence from Figures 4-6 indicated that remodeling of mitochondrial or respiratory activity took place under both N-and P-conditions. To establish which proteins and transcripts were differentially expressed, and how regulatory activity potentially differed under N-and P-conditions, the L 2 fc of respiratory-associated proteins were examined together with their transcripts (Figure 7). In N-conditions, most proteins and transcripts associated with the TCA cycle were upregulated, but those associated with glycolytic processes were both up-and downregulated. Two glycolytic enzymes, glyceraldehyde-3-phosphate dehydrogenase and phosphoglycerate kinase included multiple copies that were not coregulated with one another, with different accessions showing divergent patterns of regulation (e.g., W7U208_9STRA vs. W7T2R0_9STRA). In P-conditions, most TCA cycle and glycolytic proteins and transcripts were weakly upregulated.
The reduced plastid proteome and diminished photosynthetic capacity associated with N starvation led us to hypothesize that enzymes involved with protein/amino-acid catabolism, nitrogen recycling, and recovery could be upregulated. Consistent with increases in amine metabolic processes (GO:0009308, Figure 5), an amine oxidase (W7TFN3_9STRA) was the second-most upregulated protein under N-conditions with an L2fc of +1.38 (Table 1). In P-treatments, the same protein was significantly downregulated (L2fc −0.32, p < 0.001). Further searching through the proteome revealed an additional six proteins annotated as amine oxidases, and of these a further two were significantly upregulated under N-conditions (Table S5). Additional proteins associated with amine metabolism were also significantly upregulated in N-treatments, including an amine dehydrogenase (W7TI92_9STRA) with an L2fc of +0.77.

Tricarboxylic Acid (TCA) Cycle, Glycolytic Processes, and Oxidative Phosphorylation
Evidence from Figures 4-6 indicated that remodeling of mitochondrial or respiratory activity took place under both N-and P-conditions. To establish which proteins and transcripts were differentially expressed, and how regulatory activity potentially differed under N-and P-conditions, the L2fc of respiratory-associated proteins were examined together with their transcripts (Figure 7). In N-conditions, most proteins and transcripts associated with the TCA cycle were upregulated, but those associated with glycolytic processes were both up-and downregulated. Two glycolytic enzymes, glyceraldehyde-3-phosphate dehydrogenase and phosphoglycerate kinase included multiple copies that were not coregulated with one another, with different accessions showing divergent patterns of regulation (e.g., W7U208_9STRA vs. W7T2R0_9STRA). In P-conditions, most TCA cycle and glycolytic proteins and transcripts were weakly upregulated.

Polyunsaturated Fatty Acid (PUFA) Metabolism
The primary route to medium and long-chain polyunsaturated fatty-acid biosynthesis in microalgae is via a series of steps involving desaturase and elongase enzymes. A ∆5 desaturase (K8YSX2_NANGC) was amongst the most downregulated proteins in N-treatments (Table 1). Six other desaturase enzymes were also significantly downregulated during N-conditions (Table S6), including a ∆12 ω-6 desaturase (K8YR13_NANGC) and a glycerolipid ω-3 desaturase (I2CR09_NANGC), with L 2 fc of −0.37 and −0.53 respectively (p ≤ 0.005). Under P-conditions the abundance of the same ∆5, ∆12, and glycerolipid desaturases did not significantly change.

Proteins Associated with TAG Biosynthesis and Storage in Oil Bodies
The most upregulated protein in N-treatments with an L 2 fc of +1.93 (p < 0.001) was a lipid droplet surface protein (W7TWF7_9STRA), which is concordant with the substantial increases in TAG observed in the same samples (Table 1, Figure 1). Although the N. gaditana genome is reported to encode 11 copies of DGAT2, only one diacylglycerol acyltransferase (DGAT) family protein (W7U9S5_9STRA) was identified. This protein was significantly upregulated under N-conditions (L 2 fc +0.30, p = 0.004), but not under P-conditions (L 2 fc −0.14, p = 0.420). In comparison, the transcript data quantified the expression of eight different genes annotated as DGAT or DGAT2, where three were significantly upregulated under N-conditions and two were significantly downregulated (Table S7). Further upstream in lipid biosynthesis, Lysophosphatidylglycerol acyltransferase (LPAT) catalyzes the conversion of lysophosphatidic acid to phosphatidic acid. We identified a single LPAT protein (K8YP17_NANGC), that did not respond significantly in either N-or P-conditions.

Glycerolipid and Phospholipid Biosynthesis
A single protein annotated as monogalactosyldiacylglycerol synthase (MGDG synthase, W7TN13_9STRA) was not significantly differently expressed in either N-or P-conditions (L 2 fc < 0.07, p > 0.130). A choline/ethanolamine kinase family protein (K8YV04_NANGC) was significantly upregulated (L 2 fc +0.28, p = 0.001) in P-conditions, but was not significantly changed in N-conditions (L 2 fc +0.13, p = 0.072). The proteomics data also identified a Udp-sulfoquinovose synthase (W7TMH8_9STRA) that was significantly downregulated in N-conditions (L 2 fc −0.2, p < 0.001), but significantly upregulated in P-conditions (L 2 fc +0.24, p < 0.001). In P-conditions an Acid sphingomyelinase-like phosphodiesterase 3b (W7TQ09_9STRA) was amongst the most upregulated proteins with an L 2 fc of 0.68 (p = 0.011) ( Table 2). Table 2. The 30 proteins with largest fold increase and 30 proteins with the largest fold decrease in P-treatments (n = 2), relative to the controls (n = 4). Proteins annotated as "uncharacterized" were omitted and the p-values are from permutation tests. The suffix string of the Accession Number "9STRA" or "NANGC" refers to the B31 or CCMP526 N. gaditana reference proteomes, respectively.

Lipase Activity and Lipid Catabolism
In P-conditions a single lipase (W7TUB0_9STRA) was significantly downregulated (L 2 fc −0.32, p = 0.001). The same accession was substantially upregulated under N-conditions (L 2 fc +1.06, p < 0.001), in addition to the significant upregulation of five other lipase family proteins, including two lysophospholipases (Table S8).

Polyketide Synthase, Fatty Acid Synthase, and Lipoxygenase Expression
Six proteins annotated as polyketide synthases (PKS) were detected in the proteomics data, but none responded significantly in either the nitrogen-starved or phosphorus-starved treatments (Table S9). A single fatty acid synthase (FAS1) domain protein (W7TBQ5_9STRA) was significantly downregulated in nitrogen-starved conditions (L 2 fc −0.47, p < 0.001) but not phosphorus-starved conditions (L 2 fc −0.10, p = 0.091). An Arachidonate 5-lipoxygenase (K8Z8I5_NANGC) was also amongst the most upregulated proteins with an L 2 fc of +0.71 (Table 1), whilst a manganese lipoxygenase protein (W7TYD4_9STRA) was also significantly upregulated under N-conditions, providing evidence for the upregulation of oxylipin pathways during nitrogen starvation.

Discussion
The 3423 proteins identified in this study represent a third of the gene models in the N. gaditana genome [20,21] providing deep profiling of the Nannochloropsis proteome. The data also offers the opportunity to compare the expression of proteins with their mRNA transcripts.

Global Correlation of Nannochloropsis Protein and Transcript Expression
Integrating different 'omics datasets is a challenge but offers the chance to ask valuable questions. On one hand, transcriptome sequencing provides high-throughput measurements of global responses to physiological stress and has been widely adopted. Nevertheless, the abundance and activity of proteins in cells, which ultimately determines the phenotype, is regulated by numerous mechanisms beyond mRNA expression alone [18]. Our proteomic and transcriptomic data presented here are concordant with studies on other organisms, where generally only weak-moderate associations have been observed at the whole-cell level. Whether the unexplained residual variation is due to post-transcriptional mechanisms or to methodological sensitivity, is not always clear [15].
Correlating the L 2 fc (Figure 3a) is a straightforward method of associating transcript and protein data that relies only on relative changes in expression. Here N starvation produced a stronger correlation than P starvation, likely due to larger changes in protein and transcript abundance under N stress. However, our additional correlation methods help to provide a more complete picture. In Figure 3b we used measures of protein and transcript abundance, rather than their relative fold changes, and obtained an R 2 = 0.31. This value is comparable to observations in the model plant Arabidopsis thaliana (R 2 = 0.27-0.46) and bacteria (R 2 = 0.20-0.47), but lower than yeasts (R 2 = 0.34-0.87) [16]. When individual linear models were fitted separately to data from each protein/transcript, we were able to show the heterogeneity of correlations across different genes (Figure 3c). Proteins that were significantly differentially expressed often exhibited higher correlation with their transcripts, providing support for the role of effect-size in determining the strength of gene-protein correlations. Nevertheless, a proportion of significantly regulated proteins remained only weakly correlated with their transcripts. Like other eukaryotes, microalgae employ a multitude of post-transcriptional systems, but to what extent ncRNAs, splicing, post-translational modifications, and protein turnover [19,[26][27][28][29] impact transcript/protein/metabolome relations in oleaginous microalgae, is not yet very clear. The effect of N, but not P deprivation, on reducing ribosomal protein abundance illustrates that ribosome density varies with certain stress responses, representing a further layer of regulation between transcription and translation. Lastly, the dynamic nature of gene-protein regulatory circuits may be a critical variable [30]. Our turbidostat cultures controlled for the light intensity, but during the experimental treatments the cultures remained non-steady-state systems, where there may be overshoot in the transcriptional control of protein abundance [30,31]. Future studies can address this aspect by using alternative bioreactor control strategies.

N and P Deprivation Remodels Organelle Proteomes and Energy Metabolism
Eukaryotic cells are highly compartmentalized and the size, spatial arrangement, and contacting of subcellular compartments is re-optimized under stress conditions. In our data the dampened onset of P-stress contrasted with the rapid reduction in growth and changes in protein/gene expression observed in N-conditions. These differences can be reconciled by the way phosphorus is utilized inside the cell. Under nutrient-replete conditions luxury phosphorus uptake takes place and cells can accumulate excess reserves of intracellular phosphorus which acts as a short-term buffer during P-conditions [32]. Secondarily, certain classes of phosphorus-containing compounds can be functionally replaced by phosphorus-free alternatives (see Section 3.3), which reduces the impact of P-conditions on metabolism.
Our analysis showed that the plastid proteome was downregulated under N-conditions, consistent with the nitrogen-starved phenotypes (chlorosis and reduced polar lipid content) and the downregulation of mRNAs and proteins associated with photosynthesis. Despite the reduced photosynthetic capacity, mitochondrial proteins remained on average at comparable abundance to the control treatments, but there was evidence of reorganization, The changes in TCA cycle and glycolytic proteins under N-and, weakly, under P-conditions, highlights the active role played by respiratory processes during macronutrient stress. Previous research has indicated increased expression of glycolytic enzymes including glyceraldehyde-3-P dehydrogenase during N-conditions [33]. Our data indicates that these proteins, which are present in multiple copies, can show opposing patterns of regulation and therefore more information e.g., on cellular localization and targeting is required before their roles can be fully understood. In plants, phosphate deprivation is associated with regulation of alternative pathways in glycolysis and oxidative phosphorylation [34], and evidence from the proteome of the diatom Phaeodactylum [35] also indicates upregulation of TCA cycle activity under N-limited conditions. As mitochondrial activity is central to pathways in energy metabolism and amino acid cycling, alternative configurations of the mitoproteome play a central role in acclimation to protracted macronutrient deficits and further research is needed on mitochondrial metabolic flux under nutrient stress.

Lipid Metabolism and Remodeling
The regulation of lipid metabolism in oleaginous microalgae has been the subject of substantial scientific and commercial attention, yet the underlying mechanisms are still not completely resolved [36]. Transcriptome sequencing studies have shown that the genes involved in lipid biosynthesis are actively regulated during nutrient-induced stress [12], yet attempts to increase oil yields by overexpression of key genes have yielded mixed results [37], indicating that lipid biosynthetic enzymes are not necessarily rate-limiting. In Nannochloropsis, nitrogen starvation is primarily associated with TAG production and lipid storage in oil droplets, but surprisingly our GO and KEGG enrichment analysis (Figures 5 and 6) did not prioritize lipid-related protein or gene families during oil accumulation. However, several lipid-related proteins were strongly upregulated under N starvation, including a lipid droplet surface protein (LDSP) with the highest fold change in the whole dataset. Similar proteins have been characterized from Nannochloropsis oceanica, Chlamydomonas, and Phaeodactylum [38,39]. These proteins play a structural role in oil bodies, and so their abundance scales with neutral lipid accumulation [39].
The N. gaditana genome is reported to encode 11 DGAT2 genes, but we were only able to distinguish one diacylglycerol acyltransferase protein, although the expression of eight different DGAT2 genes were counted in the transcript data. The upregulation of DGAT under N-stress, but not under P-stress, indicates a regulatory role in TAG accumulation, and the same accession (corresponding to gene Naga_100006g86) also responds to changing light conditions in this species [40]. We identified a single LPAT protein that was unresponsive to either N-or P-conditions. However, several LPAT orthologs are present in the Nannochloropsis genome and their subcellular localization and functional role is not shared equally among them [36]. The protein Acetyl-CoA carboxylase, which drives lipid biosynthesis in the plastid [41], was downregulated under N-deprived conditions indicating reduced de-novo fatty acyl chain biosynthesis.
Macronutrient deprivation not only induces accumulation of TAG, but the remodeling of membrane (polar) lipids.
Nitrogen deprivation especially induces the degradation of plastidic glycerolipids, especially phosphatidylglycerol (PG), monogalactosyldiacylglycerol (MGDG), and digalactosyldiacylglycerol (DGDG) that that contain the majority of the EPA [42]. The fate of PUFAs under nutrient stress has important consequences for the lipid and fatty acid composition of the cell, and different processes including de-novo PUFA synthesis, translocation, and degradation/oxidation of fatty acids together contribute to the overall lipid profile. Recent evidence indicates that limited de-novo synthesis of LC-PUFAs does occur during nutrient deprivation [43], but the degradation of polar lipids and the translocation of PUFAs into TAG are significant processes that can affect the nutritional properties of microalgae. We found that PUFA biosynthesis was strongly downregulated in N-conditions, with major reductions in desaturase activity. In Nannochloropsis, ∆5 desaturase activity is associated with ARA and EPA biosynthesis [23], and together the proteomic data and fatty-acid profiles indicate that de-novo LC-PUFA biosynthesis probably plays only a minor role in lipid composition under N starvation.
Lipid-class remodeling has been associated with phosphorus starvation, where specific classes of P-containing membrane lipids are substituted with nonphospholipids [10]. Phospholipid remodeling in plants and microalgae involves acyltransferase and phospholipase activity [44]. Whilst various proteins annotated as phospholipases were identified in our data, none were significantly upregulated under P-conditions. Instead, increased lipase activity was a signature of oil-accumulating cells under N-starvation. However, we found that a choline/ethanolamine kinase was upregulated under P-conditions, which could indicate attempts to maintain phospholipid (phosphatidylcholine, PC and phosphatidylethanolamine, PE) production in these conditions. We also identified a Udp-sulfoquinovose synthase protein that was significantly downregulated in N-conditions, but significantly upregulated in P-treatments. This enzyme is associated with the synthesis of sulfoquinovosyldiacylglycerol (SQDG), a thylakoid lipid that can potentially replace and compensate for loss of phospholipids, especially PG, during phosphorus-scarce conditions [11].
An interesting feature of our data was the upregulation of two putative lipoxygenase (LOX) proteins under N-stress. Lipoxygenases provide the enzymatic route to oxylipin production where PUFAs, primarily C18 and C20 series, are converted to various oxidized lipid derivatives [45]. Oxylipins have roles in cell signaling and stress response and, although LOX activity has not been widely investigated in different microalgae species, oxylipin production has been measured in Nannochloropsis [46], and hydroxylated EPA was abundant in the metabolome of the diatom Phaeodactylum tricornitum under similar experimental conditions [35].

Cultivation
Nannochloropsis gaditana (CCMP 526, National Center for Marine Algae and Microbiota, East Boothbay, ME, USA) was cultivated in 400 mL flat plate photobiorectors (Algaemist-S, Wageningen UR, The Netherlands) using f/2 medium (Guillard and Ryther, 1962). The nutrient concentrations were increased proportionally to support high cell density, equivalent to 3.0 g L −1 NaNO 3 . Cultures were maintained as turbidostats (constant optical density) by automatically adding fresh medium and collecting the overflowing broth. Turbidostat cultures provide a high level of experimental control by eliminating variables such as changes in internal irradiance that typically occur in batch or flask cultures. The temperature (25.0 ± 0.2 • C) was maintained by internal heating/external cooling modules and a constant irradiance of 350 µmol m −2 s −1 was provided by warm-white light emitting diodes. These conditions ensured high cell density and rapid biomass turnover. Before experimental treatments the cultures were maintained for several days, where they reached a constant growth/dilution rate. Control (C) treatments were subsequently maintained at the same steady-state, whilst nitrogen (N-) and phosphorus (P-) stress treatments were selectively applied by omitting either nitrate or phosphate from the feed medium. The high biomass turnover ensured cells in stress treatments were subjected to a rapid, natural depletion of either N or P. Since there were two photobioreactor units, the cultivation sequence was designed to avoid treatment bias (Table S1), and in total there were n = 4 independent replicate cultures for C, N-, and P-conditions. Conditions inside the photobioreactors were recorded by a program written in Python v2.7, running on a Raspberry Pi single-board computer (Raspberry Pi foundation, UK). The maximum duration of our experiment was 5 days, by which time growth in N-treatments had nearly ceased and the limit of turbidity control was reached. Based on the cultivation data in Figure 1 we selected day 3 for proteomics and transcriptomics analysis, because it represented the mid-point in the onset of stress conditions, allowing sufficient time to detect metabolic and molecular changes in the cells.

Sample Collection
Samples for proteomic and transcriptomic analysis were each collected into 2.0 mL tubes. Cells were immediately pelleted by centrifugation (5000 rcf, 2 min) and quenched in liquid nitrogen, then stored at −80 • C. Samples for metabolite analysis were collected in 2.0 mL tubes and additionally desalted by washing with isotonic ammonium formate, then stored at −20 • C. The sample supernatant was retained for analysis of nitrate and phosphate. The sample time points selected for molecular characterization are shown in Table 3. Our experiment comprised 12 turbidostat cultivations, but only 10 TMT labels were available for proteomic analysis. Thus, control and N-proteome treatments each have four biological replicates, whilst P-treatments have two replicates for the proteome. Statistical analysis accounted for the degrees of freedom and multiple comparisons. Table 3. Summary of the experimental samples used for proteomic and transcriptomic analysis. The "No. Cultivations" is the total number of replicate turbidostat cultures available for each treatment. Ten proteome samples were obtained after 3 days of C, N-, or P-treatment. Twelve RNA samples were obtained after both 3 and 5 days and are repeated measurements from the same experimental units.

Lipid Analysis
Polar and neutral lipids were separated by solid phase extraction and the fatty acids were analyzed with a Gas Chromatograph and Flame Ionization Detector (GC-FID). Approximately 8 mg lyophilized samples were weighed with a precision balance (Mettler Toledo, Columbus, OH, USA, MX5) and transferred into 2.0 mL tubes containing 300 µL of 0.1 mm glass beads. Cell disruption was performed by adding 1.0 mL chloroform:methanol (2:2.5) spiked with C15:0 TAG (tripentadecanoin) internal standard, before bead-milling. The homogenate was transferred to a 10 mL glass tube with the addition of another 3.0 mL chloroform:methanol. Phase separation was used to recover the chloroform fraction, which was then dried under a stream of N 2 to recover total lipids. Polar and neutral lipid extracts were then prepared using solid-phase columns (Waters Sep-Pak 6cc/1g silica) and derivatized to fatty-acid methyl-esters (FAMEs) by adding 3.0 mL of 12% H 2 SO 4 in methanol, then heating at 70 • C for 3 h. FAMEs were separated and quantitated using a Scion 436 GC-FID (Bruker, USA) fitted with a splitless injector and a 30 m CP-WAX column (Agilent Technologies, USA). Supelco 37-component standards (Sigma-Aldrich, Oslo, Norway) were used for identification and quantitation of the FAMEs with five-point calibrations. Blanks were included throughout extraction and derivatization, to eliminate trace background peaks.

Nutrient Analysis
The concentration of nitrate in the broth was measured with standard colorimetric reagents using a miniaturized microplate method and NADH:nitrate reductase [25]. The absorbance was measured at 540 nm with a Tecan Sunrise microplate reader. Seven-point calibrations were included in each plate (R 2 > 0.995). Phosphate was analyzed with the ammonium molybdate/ascorbic acid method, and the absorbance was measured at 650 nm with a 1.0 cm cell.

Proteomics
Protein was extracted by resuspending cell pellets in 1.0 mL of extraction buffer (phosphate buffered saline +0.03% Triton X-100 + protease inhibitor cocktail) on ice, and homogenized briefly with a bead mill (Precellys, Bertin Instruments, Montigny-le-Bretonneux, France, 0.1 mm glass beads, 6500 rpm, 15 s). The suspension was centrifuged (20,000 rcf, 15 min, 4 • C) and the supernatant transferred to new tubes. Proteins were then precipitated by adding five volumes of ice-cold acetone, followed by centrifugation (20,000 rcf, 15 min, 4 • C). The supernatant was removed, and the protein pellets were allowed to air dry for 2 min at room temperature. Protein pellets were suspended in Laemlli buffer and the protein concentration of each sample was measured in duplicate with a BCA protein assay kit (Microplate BCA™ Protein Assay Kit-Reducing Agent Compatible, Thermo Scientific, Waltham, MA, USA). A seven-point calibration was used (R 2 > 0.999) and samples were blank-corrected using the sample buffer ( Figure S1). A standardized 95.1 µg of protein from each sample was loaded to an SDS-PAGE gel and trapped for analysis.
Analysis and database searching was performed by University of York metabolomics and proteomics facility (York, UK) using 10-plex Tandem Mass Tags (Thermo Scientific, TMT10plex™). In-gel tryptic digestion was performed after reduction with dithioerythritol and S-carbamidomethylation with iodoacetamide. Digests were incubated overnight at 37 • C, then peptides were extracted with 50% aqueous acetonitrile containing 0.1% trifluoroacetic acid, before drying in a vacuum concentrator and reconstituting in aqueous 0.1% trifluoroacetic acid. Peptides were buffer exchanged into aqueous 50 mM triethylammonium bicarbonate using Strata C 18 -E cartridges before TMT labelling (Table S2 for label-sample assignments). Labelled samples were combined together, loaded onto a conditioned reversed-phase C 18 spin column (Pierce) and subject to centrifugation at 5000 rcf for 2 min before washing with 300 µL of LC-MS grade water. Peptides were eluted from columns into eight fractions using increasing concentrations of acetonitrile in aqueous triethylamine. Fractions were dried in a vacuum concentrator before reconstituting in aqueous 0.1% trifluoroacetic acid. Fractions were analyzed over 4 h acquisitions with elution from a 50 cm C 18 EasyNano PepMap nanocapillary column using an UltiMate 3000 RSLCnano HPLC system (Thermo) interfaced with an Orbitrap Fusion hybrid mass spectrometer (Thermo). Positive ESI-MS, MS 2 and MS 3 spectra were acquired with multi-notch synchronous precursor selection using Xcalibur software (version 4.0, Thermo). Mascot Daemon (version 2.5.1, Matrix Science) was used to search against the Nannochloropsis gaditana subset of the UniProt database. To maximize the number of identified proteins, the search was conducted on a database containing concatenated data from the B31 and CCMP526 proteomes (15,363 sequences; 5,747,225 residues). The Mascot 0.dat result file was imported into Scaffold Q+ (version 4.7.5, Proteome Software) and a second search run against the same database using X!Tandem. Protein identifications were filtered to require a maximum protein and peptide false discovery rate of 3% [47] with a minimum of two unique peptide identifications per protein. Protein probabilities were assigned by the Protein Prophet algorithm [48]. Relative quantitation of protein abundance was calculated from the TMT reporter ion intensities using Scaffold Q+. TMT isotope correction factors were applied according to the manufacturer. Differentially expressed proteins were determined by applying Permutation Tests with significance levels (p-values) adjusted with the Benjamini-Hochberg method. TMT labelling provides sensitive measurements of differential expression of individual proteins in multiplexed samples. However, the effect of peptide length and composition means that the reporter ion responses across different proteins are only semi-quantitative estimates of abundance, i.e., different peptides/proteins have different response factors. To more accurately estimate protein quantities, the "protein abundance in multiplexed samples" (PAMUS) method [49] was applied, which is based on the empirical linear relationship between the protein abundance index (PAI) and the logarithm of absolute protein abundance [50]. The exponentially modified PAI (emPAI) for each protein was first obtained from Scaffold Q+ to estimate the relative amount of each protein in the multiplexed sample. Then for each protein, the TMT reporter ion intensities were used to quantify the proportion of emPAI attributed to each individual sample/label. The abundance of the proteins in the individual samples was then expressed in Mol% [50]. The location of mature proteins in the cell was annotated based on the "Subcellular location" field of the UniProtKB database (www.uniprot.org). Complete mass spectrometry data sets are open-access and available to download from MassIVE (MSV000085294) and ProteomeXchange (PXD018605) (doi:10.25345/C5GQ50).

Transcriptomics
Total RNA was extracted from cell pellets by adding 1.0 mL QIAzol (Qiagen) followed by lysis with a bead-beater (Precellys, Bertin Instruments, Montigny-le-Bretonneux, France, 0.1 mm glass beads, 6500 rpm, 15 s). After adding 0.2 mL chloroform, the sample was centrifuged (20,000 rcf, 15 min, 4 • C) and the aqueous supernatant was added directly to RNA Clean and Concentrator columns (Zymo Research, Irvine, CA, USA) and prepared according to the manufacturer instructions. The cleaned RNA was eluted from the columns using molecular grade water and quality and quantity checked using a 2200 TapeStation instrument (Agilent Genomics, Santa Clara, CA, USA) and Nanodrop Spectrophotometer (Thermo Fisher Scientific). Libraries were prepared using Poly(A) selection to enrich for mRNA and a NEBNext Ultra Directional RNA Library Prep kit for Illumina (New England Biolabs Inc., Ipswich, MA, USA) according the manufacturer protocols. Barcoded sample libraries were pooled in equal amount and sequenced on an Illumina NextSeq 500 platform using High Output Kit v2. A total of 443 million 150 bp paired-end reads were obtained and archived at NCBI web portal under Bioproject PRJNA589063.
The quality of reads was assessed with FastQC (Babraham Bioinformatics, Cambridge, UK) and gentle adapter and quality trimming (Q > 20, L > 50) was applied using cutadapt v1.13 [51]. The annotated reference genomes of N. gaditana were downloaded for strains CCMP526 (assembly ASM24072v1) and B31 (assembly NagaB31_1.0) and assessed. Although we used strain CCMP526 in our study (verified genetically, Figure S2), the more recent reference genome for strain B31 provided more unique mapped reads in our data (for reference comparisons see Table S3). Our analysis therefore uses reads that were aligned to the B31 reference genome using the splice-aware aligner STAR 2.5.3a [52], with the annotation aware option. The PCR duplication rate was assessed using the Bioconductor package "dupRADAR" [53] in R v. 3.3.3 and was found to be low (<0.1%). Counts of reads for gene-level quantification were extracted using "featureCounts" [54] supplied with annotation information and strands of reads. Raw counts were imported into the Bioconductor package "DESeq2" v 1.14.1 [55] and differential expression analysis was performed with independent filtering enabled and alpha = 0.05. Genes that had an FDR p-adjusted value < 0.05 and L 2 fc > 1.0 (fold change of > 2) were chosen as the differentially expressed genes. Taking into account our design (four replicates in each group and fold change > 2.0) we reached more than 90% statistical power to detect differentially expressed genes [56].

Gene Ontology and KEGG Pathway Gene Set Enrichment Analysis
Gene ontology (GO) terms were obtained from the UniProtKB database (http://www.uniprot.org). Annotation of genes for KEGG Orthology (KO) numbers was performed using GhostKOALA [57].
Gene ontology and KEGG pathway enrichment analyses were performed for both transcriptome and proteome data sets. Gene set enrichment analysis implemented in Babelomics 5.0 suite [58], was used to detect GO functional sets of genes and proteins significantly affected by nutrient deprivation. The logistic model using the L 2 fc of all genes or proteins was employed with significance cut-off FDR-adjusted p-value of 0.01. GOs with log-odds ratio (LOR) <0.0 were taken to be over-represented for downregulated genes/proteins, and LOR > 0.0 were over-represented for upregulated genes/proteins. The gene set approach was also used to identify the most perturbed KEGG pathways with unidirectional changes of gene and protein expression. The analysis was performed using the Bioconductor package GAGE 2.24 [59] with L 2 fc values as per gene statistics, q < 0.05 and only pathways with more than five annotated KO numbers. GO enrichment analysis was performed separately for up-and downregulated genes using classic Fisher's exact test in R package topGO v2.26 [60] with FDR correction at 0.05 and pruning the GO hierarchy from terms which have less than five annotated genes. To identify the most perturbated KEGG pathways with unidirectional or bidirectional changes of gene expression the gene set approach was used. The analysis was performed using the Bioconductor package GAGE 2.24 [59] with L 2 fc values as per gene statistics and only pathways with more than five annotated KO numbers.

Data Analysis
The protein and transcript data were associated together using their unique ID (gene, UniProt) numbers. Data was analyzed using the R programming language, and the package "nlme" [61], was used to fit a linear mixed-effects model ( Figure 3b, Table S4, Figure S4). The mixed-model fixed effects were (log RPKM~log Mol%) with the random effects formula (~1 + logMol%|replicate) following nlme notation, where "log RPKM" is the natural logarithm of transcript counts in units RPKM and "log Mol%" is the natural logarithm of protein abundance in Mol%. The "replicate" term is the individual turbidostat cultivation (n = 10). Correlation coefficients, summary statistics, and linear regression models were implemented in base R.

Conclusions
This study provides new insights into global protein and gene expression in the oleaginous microalga Nannochloropsis gaditana. Both proteomic and transcript sequencing methods each tended to capture the major patterns in expression, but at the whole-cell level protein and transcript associations were characteristically noisy. In Nannochloropsis macronutrient stress is associated with lipid remodeling and oleaginous phenotypes, but lipid metabolic processes were not highly enriched in our GO and KEGG analyses. We did however find major changes in several lipid-related proteins, including increased expression of DGAT and lipid body proteins under N-starved conditions. Pathways in lipid remodeling, fatty-acid oxidation and signaling could be prioritized for future studies, as these are key processes that determine the fate of valuable long-chain polyunsaturated fatty acids. Adjustments in respiratory/mitochondrial activity featured in our data, with shifts in TCA cycle activity and glycolytic processes providing metabolic compensation under stress. The active reshaping of organelle (compartment) proteomes and the control of inter-organelle metabolic flux are therefore important research areas. Finally, our data raises the topic of post-transcriptional mechanisms, which may in part explain the observed patterns of gene/protein/metabolite correlations.