Leucine Biosynthesis Is Involved in Regulating High Lipid Accumulation in Yarrowia lipolytica

Phenotypic changes during chemostat cultivations.A DGA1-overexpressing strain and a control strain were both exposed to nitrogen or carbon limitation conditions using chemostat cultivations operated at a dilution rate of about 0.05 h⁻¹, which is roughly 25% of the maximum growth rate (Table 1). For each condition, we performed biological experiments in triplicate.

Both the genetic and environmental factors, i.e., nutrient limitation and DGA1 overexpression, had major effects on the physiology of the yeast as determined by its lipid content (see Fig. S1 in the supplemental material), while other physiological parameters, e.g., the specific glucose uptake rate, remained largely unchanged (Table 1). Total lipid accumulation was not significantly induced by DGA1 overexpression alone or by nitrogen limitation alone; such induction was seen only when the two factors occurred together (Fig. S1A). Several different lipid classes contributed to the total lipid accumulation, albeit DGA1 catalyzes only the last step of triacylglycerol biosynthesis (Fig. S1B). Regardless, this supports previous findings where overexpression of DGA1 had been identified as a promising strategy to increase lipid production in Y. lipolytica (2).

Transcriptional changes.To establish the relative levels of the transcripts, samples were taken from the chemostat cultivations and analyzed by RNAseq, and normalized counts were used for relative quantifications. The multifactorial design of the study allowed separation of the contributions of each individual factor to the expression level. A general linear model was constructed describing the expression level based on the following factors: (i) baseline expression; (ii) expression due to DGA1 overexpression; (iii) expression due to nitrogen limitation; and (iv) expression due to an interaction between DGA1 overexpression and nitrogen limitation (DGA1 × N-lim). While nitrogen limitation had the largest effect on the differential expression of transcripts, with almost a sixth of the genes being differentially expressed (adjusted P < 0.05), the overexpression of DGA1 and DGA1 × N-lim interaction also resulted in large transcriptional changes (Fig. 1A; see also Data Set S1 in the supplemental material).

• Open in new tab
• Download powerpoint

FIG 1 

Consensus gene-set analysis using differential RNA expression according to a general linear model. (A) Overview of number of genes whose expression values were significantly (adjusted P < 0.05) influenced by DGA1 overexpression (DGA1), nitrogen limitation (N-lim), or the DGA1 × N-lim interaction. (B) GO term enrichment analysis. For each significantly changed GO term (rank score of ≤5), the direction and significance of the changes in RNA levels of their constitutive genes are shown, together with the total number of genes within each GO term. ox-red, oxidation-reduction; polII, polymerase II; TF, transcription factor; transp. act., transporter activity. (C) Expression profiles of ILV3, involved in leucine biosynthesis, and ATG8, involved in autophagy, exemplifying the factorial effects. acyl-CoA, acyl-coenzyme A.

Gene-set enrichment analysis was performed to identify the systemic response to each of the three factors. The effect of nitrogen limitation was dominated by a downregulation of beta-oxidation, transmembrane transporter, and transcription factors (Fig. 1B), with many of the genes downregulated significantly (adjusted P < 0.05). These results are shown to be more coherent by examination from a different perspective: glucose limitation results in upregulation of the aforementioned gene sets. The restricted availability of carbon triggers the cells to increase expression of transmembrane glucose transporters and to scavenge carbon from storage lipids by upregulating beta-oxidation. The upregulation of ergosterol biosynthesis upon nitrogen limitation is striking. In Y. lipolytica, lipids accumulate in lipid droplets that contain triacylglycerol and sterol esters; however, the highest lipid accumulation was observed only when nitrogen limitation occurred in combination with DGA1 overexpression (Fig. S1).

Overexpression of DGA1 resulted in an upregulation of genes associated with various GO terms related to ribosomes and translation (Fig. 1B), although the enrichment of these GO terms occurs mainly through the general upward drift of their constituent genes and, to a lesser extent, as a consequence of highly significantly changing transcript levels. While these changes hint at an increased level of activity of TORC1 (12), this effect is not very strong, and it is unclear how DGA1 overexpression would activate TORC1. In nitrogen-limited shake-flask cultivations of wild-type (WT) Y. lipolytica (6), the ribosome was identified as significantly downregulated, and inhibition of translation with cycloheximide resulted in moderately increased lipid accumulation. In our experiments, however, it was not DGA1 overexpression itself that provoked the largest phenotypic change but, rather, the combination of DGA1 overexpression and nitrogen restriction that resulted in the highest lipid accumulation (Fig. S1A).

The transcriptional changes due to the DGA1 × N-lim factor were dominated by an upregulation of autophagy and a downregulation of amino acid biosynthesis (Fig. 1B). The strong upregulation of autophagy genes such as ATG8 (Fig. 1C), together with an upregulation of genes involved in allantoin degradation (Data Set S1), suggests activation by the Gln3 transcription factor (13). During nitrogen limitation, amino acid biosynthesis is upregulated, while additional overexpression of DGA1 results in repression (Fig. 1C; Fig. S2), indicating that Gcn4 is involved in its regulation (14).

Identification of transcription factor binding motifs.The expression dynamics described above raised questions concerning whether common transcription factors are involved in regulating the observed gene expression. To rationally elucidate this, the relative contributions of the three factors were ranked and the differentially expressed genes (adjusted P < 0.01) with the same order of factor contributions were grouped together (Fig. 2A).

• Open in new tab
• Download powerpoint

FIG 2 

Enrichments of GO terms and transcription-factor binding motifs from genes with similar regulatory responses. (A) For each differentially expressed gene (adjusted P < 0.01), the relative contributions of the three factors (D, DGA1 overexpression; N, nitrogen limitation; D×N, interaction of DGA1 overexpression and nitrogen limitation) were ranked. The genes with the same order of factor contributions were subsequently clustered together (the number of genes per cluster is indicated). For example, for genes in group 4 (e.g., ILV3; see Fig. 1C), nitrogen limitation positively affects transcript levels, while D×N has a negative effect. DGA1 overexpression by itself has an effect that is between those of the other two factors. (B) For each cluster, (i) the GO term enrichment (redundant terms removed by manual curation) and (ii) the motifs found by DREME (15), their E value, and S. cerevisiae transcription factors with similar binding motifs, as identified with Tomtom (31), are indicated.

The promoter regions of each group of genes were queried for overrepresentation of motifs using DREME (15). From the same genes, overrepresentations of GO terms were identified with gene-set enrichment analysis. This revealed the involvement of several transcription factors orchestrating the expressional changes (Fig. 2B). For instance, motif A is seemingly involved in the upregulation of genes upon nitrogen limitation, while a further upregulation is observed when DGA1 is overexpressed (group 1). Genes that show this behavior are enriched for involvement in autophagy. Notably, Mig1-like motifs (D and H) were identified in genes strongly downregulated upon nitrogen limitation. These genes are, among others, enriched for beta-oxidation, corroborating previous identifications of this motif from batch fermentations (6). Downregulation of amino acid biosynthesis under conditions of the DGA1 × N-lim interaction (Fig. 1C; Fig. 2A, group 4) was potentially regulated through transcription factors with motifs (E and G) that bear resemblance to S. cerevisiae Gcn4 and Leu3 motifs. Contrastingly, Gln3-like motifs B and F were also found in group 2, where the DGA1 × N-lim interaction had very little effect on the transcript levels, while nitrogen limitation had a strong positive effect. While this agrees with the strong upregulation of genes involved in autophagy and allantoin degradation, motif A has an additional positive impact on genes involved in autophagy upon DGA1 × N-lim interaction (Fig. 1C).

Correlating transcript and protein levels.Changes in transcript levels can be interpreted as representing attempts to adapt to changing conditions, especially if these changes are orchestrated through shared transcription factors. However, changes in transcripts are not necessarily translated into changes in the levels of proteins. We therefore performed relative quantifications of proteins using the accurate mass and time (AMT) tag proteomics approach (16). Normalized peptide abundances obtained from the proteomics analysis were subsequently compared with normalized read counts obtained from RNAseq. To increase the confidence of the correlation between protein and transcripts, we focused only on the genes for which data were obtained from all 12 samples by both RNAseq and proteomics. This subset consisted of 1,516 genes, corresponding to 21.1% of the 7,170 genes annotated in the Y. lipolytica genome.

The correlation between all transcript-protein pairs was low (Pearson’s r, 0.426) (Fig. 3A). This would be expected due to the differences in the translation rates for each protein (reviewed in reference 17), while low correlations are also expected for non-differentially expressed genes. Instead, when each transcript was correlated with its corresponding protein across conditions, the correlation over all 12 experiments was moderately enriched for positive correlations between transcript and protein (Fig. 3B). Separating 444 differentially expressed transcript-protein pairs (selected from the general linear model at the transcript level, adjusted P < 0.01; Data Set S1) resulted in a strong enrichment for positive correlation (Fig. 3B; Data Set S2). Additionally, there seems to be a trend toward stronger correlations for genes with a greater level of significant differential expression (Fig. 3C). It is worth noting that autophagy was upregulated through the DGA1 × N-lim factor, while ribosomes were (moderately) upregulated due to DGA1 overexpression (Fig. 1). These changes are likely to affect protein half-lives and translation rates, modulating the protein concentration by means other than transcript levels. Nonetheless, the proteomics data supported the changes identified from the general linear model, as a hypergeometric gene-set enrichment analysis of highly correlated genes (P < 0.01) unveiled similar GO terms (Fig. 3D).

• Open in new tab
• Download powerpoint

FIG 3 

Correlation of log-normalized RNAseq read counts with log-normalized protein counts. (A) All measured RNA and protein counts combined in one comparison, with poor correlation observed. (B) Density plot of Pearson’s r for all RNA-protein pairs (dotted line) and for only those RNA-protein pairs that showed differential expression at the level of RNA (adjusted P < 0.01; solid line). (C) Representation of the correlations of both the genes that were not significantly changed (in red box) and the genes whose transcripts were found to have changed in comparisons of all chemostats (adjusted P < 0.01). (D) Network plot of enriched GO terms in the highly correlated gene pairs (P < 0.01). The size of the node represents the number of genes within the GO term. The node’s color represents the adjusted P value. The thickness of the edge indicates the number of genes overlapping two GO terms. m.p., metabolic process; b.p., biosynthetic process.

Control of fluxes.The combination of transcriptomics, proteomics, and binding-motif searches corroborates that amino acid biosynthesis and beta-oxidation are transcriptionally regulated, giving rise to the issue of whether these changes are also reflected in changes of metabolic fluxes. We therefore calculated metabolic fluxes using a genome-scale metabolic model by constraining measured exchange fluxes (Fig. S1; Table 1) and using random sampling of the solution space (18). This analysis was performed through four cross-comparisons: nitrogen versus carbon restriction in WT and DGA1 strains and DGA1 versus WT strains during nitrogen and carbon limitation. For each flux-gene pair, correlations were calculated from Z scores indicating the observed changes in metabolic flux and transcript and protein levels.

The number of strongly correlating fluxes differed drastically between the different comparisons (see Table S1 in the supplemental material). No large changes are expected in fluxes when phenotypic changes are only moderate. With only a few exchange fluxes constraining the intracellular fluxes, the random sampling approach gives a relatively high average relative standard deviation (RSD) for all flux-carrying reactions in the four conditions, and the fluxes therefore must change drastically for a significant Z score (typically >2) to occur. This approach is therefore likely to give an underrepresentation of fluxes that are correlated with changes in transcripts and protein levels.

The overall observation is that transcriptionally regulated fluxes are downregulated. In each comparison, the reference condition had a lower lipid yield, and increased lipid accumulation was not associated with an upregulation of lipid biosynthetic enzymes. The highest numbers of transcriptionally regulated fluxes were seen in the comparison of nitrogen limitation to carbon limitation with the DGA1 overexpression strain, a result which can be expected from the changes in environmental conditions and lipid phenotype. In the comparison of the DGA1 overexpression strain to the control strain during nitrogen limitation, only a few reactions were transcriptionally regulated. Interestingly, most transcriptionally regulated genes were those involved in leucine biosynthesis (Fig. 4A), correlating the observed transcriptional changes with not only protein levels but also changes in metabolic fluxes.

• Open in new tab
• Download powerpoint

FIG 4 

Regulation of Y. lipolytica metabolism. (A) Transcriptional downregulation of leucine biosynthesis. The boxes indicate z scores, incorporating both fold change and significance values. Regulation scores indicate correlations between all three levels and are calculated as detailed in Materials and Methods. (B) Relative intensity levels of two metabolites as measured by exometabolomics (also represented in Fig. S3 in the supplemental material). While most carbohydrates (e.g., d-mannitol) are excreted during nitrogen limitation in both WT and DGA1 strains, 2-isopropylmalate (IPM) is no longer excreted when nitrogen limitation co-occurs with DGA1 overexpression.

Metabolome analysis.To obtain further insight into whether flux changes are associated with changes in metabolite levels, we performed untargeted metabolomics analysis of intracellular and extracellular metabolites. Linear modeling was used to separate the changes in relative metabolite levels, and this analysis showed that overexpression of DGA1 had limited effects on metabolite levels (Fig. S3 to S5). The main differences in extracellular metabolites were the excretion of sugars, sugar alcohols, and disaccharides under nitrogen restriction conditions. During batch stage, Y. lipolytica excretes disaccharides as an overflow metabolite when glucose is provided in excess, while the disaccharides are subsequently consumed upon glucose depletion (19). One of the metabolites excreted upon nitrogen restriction is the leucine biosynthetic pathway intermediate 2-isopropylmalate (IPM). However, the simultaneous overexpression of DGA1 resulted in decreased excretion of this metabolite (Fig. 4B), even while the DGA1 × N-lim interaction did not significantly influence the excretion of the other carbohydrates (Fig. S3). This lack of IPM excretion is consistent with the transcriptional downregulation of the leucine biosynthetic pathway under the conditions with the highest lipid accumulation, likely through binding of a Leu3-like transcription factor to the leucine biosynthetic genes (cf. group 4 in Fig. 2A; Fig. 4A).

The changes in intracellular metabolite levels were more difficult to decipher. As seen with the external metabolites, nitrogen limitation resulted in increased levels of carbohydrates and organic acids such as glycolysis and tricarboxylic acid (TCA) cycle intermediates (Fig. S3). Additional DGA1 overexpression increases the levels of many of the same metabolites. The increase in the levels of many amino acids resulting from DGA1 × N-lim interaction, even though amino acid biosynthetic pathways were downregulated, is surprising. The levels of leucine and valine also increased, while branched-chain amino acid biosynthesis was strongly downregulated at the level of transcription. As autophagy is strongly upregulated through DGA1 × N-lim interactions, it is possible that the increased amino acid levels were a consequence of protein degradation occurring through autophagy rather than as a consequence of increased de novo biosynthesis. These elevated levels could result in downregulation of amino acid biosynthesis, as particularly seen for the branched-chain amino acid biosynthesis. A cell undergoing nitrogen limitation would potentially benefit from increased expression of branched-chain amino acid transaminases (BCAT), as a means of shuffling amino groups between the products of protein degradation. Nonetheless, BCAT genes BAT1 and BAT2 respond differently to nitrogen limitation (Fig. S6). While cytosolic BAT2 is upregulated irrespective of whether or not DGA1 overexpression is present, mitochondrial BAT1 instead shows an expression response similar to that seen with, e.g., leucine biosynthesis. This suggests that the primary function of BAT1 is de novo biosynthesis, while BAT2 is instead involved in reshuffling amino groups within the cytosolic branched-chain amino acid pool. The observed amino acid levels, however, do not provide a clear picture supporting this remodeling of amino acid metabolism.