Quantifying absolute gene expression profiles reveals distinct regulation of central carbon metabolism genes in yeast

We performed a series of 14 steady-state chemostat cultures of S. cerevisiae, which orthogonally modulated either the growth rate (GR experiments) or the nitrogen metabolism (NM experiments) of yeast (Figure 1A, Supplementary file 1a). Nitrogen metabolism was of particular interest since nitrogen is a critical component of amino acids and nucleotides. The nitrogen sources in the NM experiments were chosen to represent those that are preferred (NH₄ and Glu) and non-preferred (Phe and Ile), as previously defined (Godard et al., 2007). Depending on the growth condition, the total protein content in the dry cell weight ranged between 20.6% and 59.4%, and the total RNA content ranged between 1.8% and 8.9%, (Supplementary file 1a), in agreement with previous observations of total protein and RNA content with changing growth rate and nitrogen source (Verduyn, 1991; Larsson et al., 1993; Yu et al., 2020a). We then profiled the absolute-quantitative transcriptomic and proteomic abundances (mmol/gDW) of these chemostat cultures in biological triplicates, generating a multi-omics dataset containing 3127 transcript–protein pairs across 14 conditions (Supplementary file 1b). In both GR and NM experiments, we found >2,000 genes to be differentially expressed at both the transcript and protein levels with false discovery rate (FDR) < 0.01 (one-way ANOVA comparing all conditions; Figure 1B), indicating that gene expression is subject to extensive control by both growth rate and nitrogen metabolism. A total of 1493 genes (48%) in the GR experiments, and 1959 genes (63%) in the NM experiments, were differentially expressed not only significantly, but by more than twofold. In Figure 1—figure supplement 1, we present the total protein–RNA correlation of these chemostat cultures, with a Pearson r of 0.51 and a Spearman ρ of 0.27, and sample-wise proteome–transcriptome correlations, with a median Pearson r of 0.40 and Spearman ρ of 0.63, agreeing well with previous studies (Yu et al., 2020a; Lahtvee et al., 2017; Schwanhäusser et al., 2011; Marguerat et al., 2012). In Figure 1—figure supplement 2, we present the measured absolute quantities of subunits for several complexes (Yu et al., 2020a; Taggart and Li, 2018), showing that our measured protein abundances agree with the known subunit stoichiometry within 1–2 orders of magnitude, typical of iBAQ-based proteomics quantitation (Yu et al., 2020a; Schwanhäusser et al., 2011; Björkeroth et al., 2020). This variability in the proteomics data reflects a mixed effect of inaccuracies in the proteomics methodology, and proteins that are partially synthesized, partially degraded, or not being part of its designated protein complex. For f1f0 ATP synthase, ATP15 and TIM11 were 4 orders of magnitude lower than the abundance of other subunits (Figure 1—figure supplement 2), likely reflecting the difficulty in extracting and quantifying membrane-embedded proteins.

We further mined absolute-quantitative transcriptomics and proteomics data from Xia, 2019a; Xia, 2019b (XIA experiments), wherein S. cerevisiae were grown at nine different dilution rates with glucose being the limiting nutrient (Figure 1A,B). Of these, one condition (carbon-limited, nitrogen source NH₄, dilution rate 0.10 h⁻¹) was duplicated in the chemostats that we performed (Figure 1A); thus, the number of unique conditions examined in this study is a total of 22. Data from the XIA experiments, containing paired transcript-protein abundance of 2235 genes (Supplementary file 1c), allowed us to validate our findings in the GR experiments in a different nutrient limitation setting.

Leveraging these absolute-quantitative data, we performed a series of exploratory analyses with the goal of describing the control of (1) transcription and (2) translation with respect to growth rate; and (3) transcription and (4) translation with respect to nitrogen metabolism. As transcript and protein abundances span several orders of magnitude, for ease of visualization we present most of our analyses using scaled expression levels; details of the scaling done can be found in the Materials and methods section, and the absolute concentrations can be found in supplementary files as indicated. We also suggest mechanisms and consequences of these observations, by examining the abundance of RNA polymerase II, ribosomes, and intracellular amino acid levels. The lessons learned from these analyses are collected into a conceptual model. We note however that this does not imply direct causation, rather hypotheses which should be tested by further experiments. Finally, we showcase the significance of understanding the effect of growth rate and metabolism on gene expression, by accounting for these confounding factors in a re-analysis of two previously published gene expression profiling datasets.

We first considered the regulation of mRNA abundance by growth rate. We used 25 clustering indices (Charrad et al., 2014) to determine the best clustering scheme for transcript abundance in the GR experiments and found that they fell into an optimal number of three clusters (Figure 2A). The list of genes in each of the clusters can be found in Supplementary file 1b and filtering by the column ‘GR.cluster’. Expression of transcripts in GR.cluster.1 and GR.cluster.3, together including 2895 genes (92%), exhibited strong associations with growth rate (Figure 2B). In contrast, expression of 232 transcripts (7%) in GR.cluster.2 did not increase with increasing growth rate (Figure 2C). These growth rate-independent genes in GR.cluster.2 were enriched in GO-slim terms (Cherry et al., 2012) related to central carbon metabolism (CCM), including ‘carbohydrate metabolic process’ and ‘generation of precursor metabolites and energy’ (Figure 2D). We further observed that genes in GR.cluster.1 and GR.cluster.3 closely followed the protein abundance of RNA polymerase II (Carlberg and Molnár, 2014; Figure 2E), suggesting that transcript abundance of these 92% of genes reflected the availability of RNA polymerase II, consistent with previous studies (Klumpp and Hwa, 2008; Klumpp et al., 2009; Heldt et al., 2018). In contrast, the 232 genes in GR.cluster.2 exhibited slightly decreased abundance with increasing RNA polymerase II expression (Figure 2E), indicating that these genes were regulated by a mechanism independent from growth rate-associated changes in RNA polymerase II.

Figure 2 with 1 supplement see all

Download asset Open asset

To confirm these findings, we performed similar analyses using data from the XIA experiments (Xia, 2019a; Xia, 2019b), where the optimal number of clusters was calculated to be two clusters (Figure 2—figure supplement 1A). The list of genes in each of the clusters in the XIA experiment can be found in Supplementary file 1c and filtering by the column ‘XIA.cluster’. XIA.cluster.2 contained 1964 genes (88%), which exhibited growth rate-dependent transcript abundance (Figure 2—figure supplement 1B), while XIA.cluster.1 contained 271 genes (12%) showing growth rate-independence. Genes in XIA.cluster.1 were enriched in CCM GO-slim terms (Figure 2—figure supplement 1C,D), including ‘carbohydrate metabolic process’ and ‘generation of precursor metabolites and energy’, validating our results in the GR experiments. Of note, GO-slim terms related to CCM (‘carbohydrate metabolic process’; ‘cellular respiration’; and ‘generation of precursor metabolites and energy’) were enriched among the most abundant 10% genes in both datasets (Figure 2—figure supplement 1E–F), demonstrating that the transcript expression of vast majority of genes were controlled by the cell growth rate, but a distinct mechanism regulated a small number of CCM-related transcripts that are highly abundant.

We next examined the relationship between mRNA and protein abundance of each gene in the GR experiments (Figure 3). As the underlying distribution of the mRNA and protein abundances in these analyses were non-normal (p_Shapiro-Wilk < 0.01) for a large portion of genes, we used Spearman correlations in the following analyses. We note that Spearman correlation coefficients (ρ) were closely related to Pearson correlation coefficients (r) (Figure 3—figure supplement 1), and as such our observations can be easily compared to previous studies that have used Pearson correlations. In Supplementary file 1b, we report both Spearman ρ and Pearson r values.

Figure 3 with 1 supplement see all

Download asset Open asset

Several previous studies have suggested that, comparing the mRNA and protein abundances of the same gene across different experimental conditions, generally there is a high correlation between protein and mRNA abundance (Lahtvee et al., 2017; Yu and Nielsen, 2019; Newman et al., 2006; Csárdi et al., 2015). Here, in the GR experiments, we observed high protein–mRNA correlations consistent with these studies, with a median ρ of 0.76 (Figure 3A) and 2379 genes (78%) having ρ > 0.5. We then calculated the protein translation rate for each gene (k_sP, in the unit of protein/mRNA/h) (Lahtvee et al., 2017; Schwanhäusser et al., 2011) by:

where k_sP,j is the protein translation rate for gene j; C_prot,j is the concentration of protein j; k_dP,j is the protein degradation rate for gene j, measured by Lahtvee et al., 2017; μ is the specific growth rate; and C_mRNA,j is the concentration of mRNA j. For those genes where k_dP,j was not determined in Lahtvee et al., 2017, the median of all k_dP in that dataset was used. We found that overall k_sP is increased with increasing growth rate (Figure 3B). Concurrently, we observed an increase in the % (mol/mol) of ribosomal proteins in the total proteome (Figure 3C), consistent with previous reports (Yu and Nielsen, 2019; Metzl-Raz et al., 2017) and suggesting that the increase in k_sP in the GR experiments were linked to the availability of ribosomes. Taken together, our data indicate that increasing growth rate modulated gene expression via coordinated increases in both transcription and translation, by increasing the availability of RNA polymerase II and ribosomes.

In the NM experiments, using the same 25 clustering indices (Charrad et al., 2014), we calculated that transcript expression fell into an optimal number of two clusters (Figure 4A). The list of genes in each of the clusters in the NM experiments can be found in Supplementary file 1b and filtering by the column ‘NM.cluster’. In NM.cluster.1, a total of 3011 genes (96%) exhibited increasing expression levels when cells were grown on preferred nitrogen sources (NH₄ and Gln) compared to non-preferred nitrogen sources (Phe and Ile), under carbon-limiting conditions (Figure 4B). In contrast, 116 transcripts (4%) in NM.cluster.2 did not show the same expression pattern (Figure 4C). These genes were enriched in processes related to CCM, as well as amino acid metabolic processes (Figure 4D).

Figure 4

Download asset Open asset

Examining RNA polymerase II levels, we found that changes in transcript abundance was not associated with RNA polymerase II availability in the NM experiments (Figure 4E). As the increased transcripts in the majority of genes when cells were grown on preferred nitrogen sources (Figure 4B) must be supported by increased biosynthesis of nucleotides, we therefore examined the intracellular concentrations of Ser and Gly, which are substrates for purine synthesis (Figure 5A), as well as Gln and Asp, which are upstream of both purine (Figure 5A) and pyrimidine synthesis (Figure 5B). We found that transcript abundance in the NM experiments tracked closely with intracellular Ser and Gly concentrations (Figure 5C), but not with Gln or Asp (Figure 5D). Moreover, in absolute-quantitative terms (Supplementary file 1d), Gln and Asp were present at much higher intracellular concentrations (mean of 166.7 and 12.4 µmol/gDW, respectively, across all NM experiments) compared to Gly and Ser (mean of 1.5 and 2.8 µmol/gDW, respectively, across all NM experiments). Together, this suggests that intracellular Gly and Ser were likely limiting for purine synthesis, which in turn constrained transcript abundance when the nitrogen source was scarce or non-preferred.

Figure 5

Download asset Open asset

In the NM experiments, the gene-specific protein–mRNA correlations were surprisingly poor, with a median ρ of 0.07 (Figure 6A) and most genes with −0.5 < ρ < 0.5 (2763 genes, 88%). To check whether the poor correlations stem from lower range of differential gene expression, we split the total of 3127 genes into quartiles by mRNA differential expression (fold-change of largest to smallest measured value) and found that, even in the fourth quartile where the range of differential expression was >5.3-fold, the protein–mRNA correlations remained very low, with a median ρ of 0.24 (Figure 6—figure supplement 1A). Similar results were found when genes were split into quartiles by protein differential expression (Figure 6—figure supplement 1B), indicating that the poor protein–mRNA correlations observed for most genes in the NM experiments cannot be explained by the range of differential gene expression. Calculating the protein translation rate k_sP, we found that gene-specific k_sP in the NM experiments were generally reduced when cells were grown on preferred nitrogen sources, in both nitrogen-limited cultures and carbon-limited cultures (Figure 6B). These changes in k_sP occurred without any changes in the % of ribosomal proteins in the total proteome (Figure 6C). As the growth rate of the NM experiments was controlled to a constant 0.1 h⁻¹ (Figure 1A), this indicates that the % of ribosomal proteins in the total proteome was a direct reflection of the cell growth rate, while nitrogen metabolism modulated protein translation without modifying the abundance of ribosomes.

Figure 6 with 1 supplement see all

Download asset Open asset

Combining all gene expression measurements in both GR and NM experiments, we found that the protein–mRNA correlations remained poor, with a median ρ of 0.22 (Figure 7A). This suggests that, while the most genes were differentially expressed in response to growth rate and/or metabolism at the transcript level, this does not necessarily translate into concordant changes in protein abundance. We also considered the possibility that relative abundance of proteins and transcripts may be well correlated, while their absolute abundance may be irrelevant. However, the correlation of mRNA relative-abundance (% mol/mol of total mRNA) with protein relative-abundance (% mol/mol of total protein) was also poor, with a median ρ of 0.34 (Figure 7—figure supplement 1). Thus, transcription and translation appear to be two distinct points of gene expression regulation that overall are not concordantly modulated.

Figure 7 with 1 supplement see all

Download asset Open asset

The large range of ρ spanning from around −0.5 to 1 (Figure 7A) nevertheless indicates that for some genes, the correlation between protein and transcript levels can be very high, while for others, this correlation would be especially poor. To examine these relationships, we tested for enrichment of GO-slim terms in 200-gene sliding windows of increasing ρ (Yu et al., 2020a; Marguerat et al., 2012). We found that genes with high correlations between protein and mRNA abundance (median ρ of 0.4–0.7 in 200-gene windows) are enriched in GO-slim terms related to amino acid metabolic processes, and processes related to protein translation, including ribosome biogenesis; rRNA processing; tRNA aminoacylation; and several others (Figure 7B). On the other hand, genes with poor protein–mRNA correlations (median ρ of −0.2 to 0.2 in 200-gene windows) were enriched in GO-slim terms related to CCM (‘carbohydrate metabolic process’, ‘cellular respiration’, and ‘generation of precursor metabolites and energy’; Figure 7B), indicating distinct roles of transcription and translation in regulating CCM gene expression, which reflects the unique and complex mechanisms that regulate this important part of metabolism (Yu et al., 2020a).

We then tested whether genes with different protein–transcript correlations have different proportions of specific amino acids in their protein sequences. Splitting genes into 200-gene brackets of increasing ρ, we found that protein–transcript correlations vary with the sequence proportion of several amino acids (Figure 7C). Of note is that genes with low protein–transcript correlations contain higher proportions of Gln, Glu, Asn, and Asp (Figure 7C), which are central to the metabolism of all other amino acids as well as nucleic acids. Thus, it is possible that the uncoupling of translational regulation from transcriptional regulation in these genes represents a mechanism to allow intracellular amino acids to modulate protein abundance. We also found that the codon usage of genes with low protein–transcript correlation tend to have higher usage of purines (Figure 7D), suggesting that nucleic acid levels may also play a role in controlling the transcript abundance of these grenes and contributing to the uncoupling between transcription and translation.

In Figure 8, we collect the lessons learned from all of the above analyses into a conceptual model. Information flow in the central dogma and a simplified flow of nitrogenous metabolites are given as reference points. Analysis of data in the GR experiment revealed that changes in the abundance of transcripts and proteins with respect to the cell growth rate are highly correlated, and both are closely associated with levels of RNA polymerase II and ribosomes in the cell (Figure 8, green text and arrows). This general rule however does not apply to a small group of genes enriched in CCM (Figure 8, note a). In contrast, nitrogen metabolism influences gene expression via a mechanism that is independent of RNA polymerase II and ribosome levels, with transcript and protein abundance being well correlated only in a subset of genes (Figure 8, blue dashed arrow). Here, transcript expression is limited by purine availability, which in turn is constrained by intracellular Gly and Ser abundances; while protein expression may be controlled by nitrogen availability centered around Glu, Gln, Asp, and Asn concentrations in the cell (Figure 8, blue text and arrows). The protein abundance of genes with poor protein–transcript correlations, again enriched in CCM enzymes, are especially affected by this control mechanism (Figure 8, note b) because of the high proportion of these specific amino acids in their sequences.

Figure 8

Download asset Open asset

With an understanding of the effects of the cell growth rate and nitrogen metabolism on gene expression, these effects can be factored into gene expression analyses to allow the extraction of gene expression programs related specifically to the perturbation of interest. Here we showcase two such applications of our findings, first using a dataset in a distantly related yeast, S. pombe, and second using a dataset from a human cancer cell model.

Marguerat et al., 2012 have shown that in S. pombe, gene expression in quiescent cells was characterized by a drastic downregulation of the total transcriptome compared to proliferating cells, with >85.3% transcripts being lowered by >2-fold. However, there is an inherent difference in growth rate between proliferating and quiescent cells, and quiescence was induced experimentally by nitrogen starvation (Marguerat et al., 2012). Together, we expect these differences in growth rate and nitrogen availability to influence the abundance of >90% of transcripts. What, then, are the gene expression changes induced by quiescence per se, without the confounding factors of growth rate and nitrogen limitation? To answer this question, we took the transcriptome abundance and protein translation rates of proliferating S. pombe and calculated the theoretical transcriptome abundance and protein translation rates if reduced growth rate and limitation of the nitrogen source (NH₄) were the only regulatory signals controlling differential gene expression. Comparisons between the measured and calculated transcriptome abundance (Figure 9A) and protein translation rates (Figure 9B) can then inform us of how gene expression is regulated by quiescence per se. As the expression of CCM genes generally do not follow the gene expression controls by growth rate and nitrogen metabolism, in the calculations of the theoretical transcriptome abundance and protein translation rates, we therefore considered them to be at a constant level for these genes. In effect this means that, for CCM genes (identified by the PomBase GO-slim term ‘carbohydrate metabolic process’) (Lock et al., 2019), the comparisons made in Figure 9A,B are equal to directly comparing the transcriptome abundance and protein translation rates between proliferating cells and quiescent cells. Our results showed that for most transcripts, the difference in abundance between proliferating and quiescent cells was largely in line with the difference in growth rate (Figure 9A). Quiescence per se induced a specific gene expression program by upregulating 707 transcripts (14%) and downregulating 574 transcripts (11%). In contrast, protein translation dynamics was extensively remodeled by quiescence, with 90% of k_sP being upregulated (Figure 9B), showing that protein translation plays a surprisingly large role in regulating gene expression in response to quiescence in S. pombe, which would be interesting to explore in future studies.

Figure 9 with 1 supplement see all

Download asset Open asset

We then performed a similar re-analysis of gene expression profiling data in a human cancer cell model, to showcase that changes in gene expression caused by oncogenic perturbation can also be uncoupled from those that are induced by physiological parameters such as increased growth rate and altered metabolism. We mined gene expression datasets in the B-cell line P493–6 (Lin et al., 2012; Feist et al., 2018), which carries a conditional c-Myc allele that can be experimentally turned on (MYC^high), mimicking overexpression of the MYC transcription factor that is the driving oncogenic event in Burkitt’s lymphoma (Schuhmacher et al., 1999). Lin et al., 2012 quantified the absolute abundance of 1263 transcripts in the P493–6 cell system and found that 707 transcripts (56%) were upregulated by >2-fold in MYC^high cells. Similarly, Feist et al., 2018 quantified the absolute abundance of 1662 proteins and found that 871 proteins (52%) were upregulated by >2-fold in MYC^high P493–6 cells. However, MYC overexpression in P493–6 also leads to faster cell growth (Schuhmacher et al., 1999) and Gln dependency (Feist et al., 2018), both of which could influence gene expression. We therefore sought to uncouple the effects of these physiological differences from the specific effects of MYC overexpression, which would help to elucidate direct MYC targets. Taking a similar approach as we have done in the re-analysis of the S. pombe dataset, we took the transcriptome abundance and protein translation rates of MYC^low P493–6 cells and calculated the theoretical transcriptome abundance and protein translation rates if increased growth rate and usage of Gln as the nitrogen sources were the only regulatory signals controlling differential gene expression. Comparisons between the measured and calculated transcriptome abundance (Figure 9C) and protein translation rates (Figure 9D) can then inform us of how gene expression is regulated by the MYC oncogene, without the secondary effects of increased growth rate and Gln usage. As before, we also considered all CCM genes (identified by AmiGO 2 term ‘cellular carbohydrate metabolic process’) (AmiGO Hub et al., 2009) to be expressed constantly; thus, for CCM genes, the comparisons made in Figure 9C,D are equal to directly comparing the transcriptome abundance and protein translation rates between MYC^highand MYC^low P493–6 cells. Results indicated that the majority of the upregulated transcripts could be accounted for by changes in the cell growth rate and Gln usage (Figure 9C), while MYC overexpression specifically upregulated 309 transcripts (24%) and downregulated 167 transcripts (13%). Similarly, protein translation dynamics in MYC^high P493–6 cells were also largely accounted for by changes in the cell growth rate and Gln usage (Figure 9D), with MYC overexpression causing the up- and downregulation of k_sP for 101 (6%) and 212 (13%) proteins, respectively. Genes with upregulated mRNA abundance generally had lowered k_sP, and vice versa (Figure 9—figure supplement 1), indicating that protein translation plays a buffering role in MYC-overexpressing cancer cells, producing dampened effects on protein abundance compared to changes in transcript abundance, which will be an important consideration in efforts to develop MYC-targeting therapeutics.

The yeast S. cerevisiae CEN.PK113-7D (MATa, MAL2-8c, SUC2) was used for all experiments. Cells were stored in aliquoted glycerol stocks at −80°C. Chemostat experiments were carried out in DASGIP 1L bioreactors (Jülich, Germany) equipped with off-gas analysis, pH, temperature, and dissolved oxygen sensors. Chemostat experiments were carried out at 30°C, pH 5, working volume 0.5 L, aeration 1 vvm, pO₂ > 30%, and agitation speed 800 rpm. The glucose and nitrogen source concentrations are in Supplementary file 1a. Additional media components are as follows: KH₂PO₄, 3 g L⁻¹; MgSO₄·7H₂O, 0.5 g L⁻¹; trace metals solution, 1 mL L⁻¹; vitamin solution, 1 mL L⁻¹; and antifoam, 0.1 mL L⁻¹. The trace metal solution contained: EDTA (sodium salt), 15.0 g L⁻¹; ZnSO₄⋅7H₂O, 4.5 g L⁻¹; MnCl₂⋅2H₂O, 0.84 g L⁻¹; CoCl₂⋅6H₂O, 0.3 g L⁻¹; CuSO₄⋅5H₂O, 0.3 g L⁻¹; Na₂MoO₄⋅2H₂O, 0.4 g L⁻¹; CaCl₂⋅2H₂O, 4.5 g L⁻¹; FeSO₄⋅7H₂O, 3.0 g L⁻¹; H₃BO₃, 1.0 g L⁻¹; and KI, 0.10 g L⁻¹. The vitamin solution contained: biotin, 0.05 g L⁻¹; p-amino benzoic acid, 0.2 g L⁻¹; nicotinic acid, 1 g L⁻¹; Ca-pantothenate, 1 g L⁻¹; pyridoxine-HCl, 1 g L⁻¹; thiamine-HCl, 1 g L⁻¹; and myo-inositol, 25 g L⁻¹. The measured nutrient fluxes in all chemostats are given in Supplementary file 1e.

The dead volume was collected with a syringe and discarded. For transcriptome sampling, biomass was collected from the reactor with a syringe and injected into chilled 50 mL Falcon tubes filled with 35 mL crushed ice. Samples were centrifuged for 4 min at 3000 × g at 4°C; cell pellets were washed once with 1 mL of chilled water, transferred into Eppendorf tubes, flash frozen in liquid nitrogen, and stored at −80°C until analysis. For proteome sampling, biomass was collected from the reactor with a syringe and injected into 50 mL Falcon tubes chilled on ice. Samples were centrifuged for 4 min at 3000 × g at 4°C; cell pellets were washed once with 20 mL of chilled dH₂O, washed again with 1 mL of chilled water, transferred into Eppendorf tubes, flash frozen in liquid nitrogen, and stored at −80°C until analysis. For intracellular amino acid measurements, biomass was collected from the reactor with a syringe and injected into −80°C MetOH at 10 times the volume of the sample. Samples were centrifuged to 4 min at 3000 × g at −20°C, and pellets were flash frozen in liquid nitrogen and stored at −80°C until analysis. Biomass determination was done by filtration of the culture broth on pre-weighed filter paper, drying in a microwave at 360 W for 20 min, and desiccating in a desiccator for >3 days.

RNA was extracted using Qiagen RNeasy Mini Kit (Qiagen, Hilden, Germany) according to manufacturer’s protocol. RNA integrity was examined using a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). RNA concentration was determined using a Qubit RNA HS Assay Kit (Thermo Fisher, Waltham, MA). The Illumina TruSeq Stranded mRNA Library Prep Kit (Illumina, San Diego, CA) was used to prepare mRNA samples for sequencing. Paired-end sequencing (MID Output 2 × 75 bp) was performed on an Illumina NextSeq 500 (Illumina, San Diego, CA). Reads were quality controlled, mapped to the S. cerevisiae reference genome (Ensembl R64-1-1), and counted using the nf-core RNAseq pipeline (SciLifeLab, Stockholm, Sweden), available at https://nf-co.re/rnaseq.

Liquid chromatography–mass spectrometry (LC–MS)-based proteomic analysis was performed largely as described in our previous publication (Yu et al., 2020a). Yeast cell pellets were suspended in the lysis buffer (2% sodium dodecyl sulfate, 50 mM triethylammonium bicarbonate [TEAB]) and homogenized using a FastPrep−24 instrument (Matrix C, 1 mm silica spheres; MP Biomedicals, Akron, OH) for 5 repeated 40 s cycles at 6.5 m/s, with 30–60 s breaks in-between. The lysate was centrifuged at 21,100 x g for 20 min, the supernatant was transferred to new tubes and diluted five times with the lysis buffer, and protein concentration was determined using Pierce BCA Protein Assay Kit (Thermo Fischer Scientific, Waltham, MA) and the Benchmark Plus microplate reader (Bio-Rad Laboratories, Hercules, CA), with bovine serum albumin solutions as standards. Pooled reference sample (‘Reference’) has been prepared by mixing equal amounts (according to BCA (bicinchoninic acid) measurement) from each supernatant.

For the TMT-based relative quantification, aliquots containing 25 µg of protein were taken from each experimental sample and from the pooled reference sample; an aliquot of 50 µg of the Reference sample was spiked with 10.6 µg of the UPS2 Proteomics Dynamic Range Standard (Sigma–Aldrich, Saint-Louis, MO) for the IBAQ (Schwanhäusser et al., 2011) quantification. Each sample was reduced by the addition of 2 M dl-dithiothreitol to a final concentration of 100 mM and incubated at 56°C for 30 min. The samples were processed according to the modified filter-aided sample preparation method (Wiśniewski et al., 2009). In short, reduced samples were diluted to 300 µL by addition of 8 M urea, transferred onto Nanosep 30 k Omega filters (Pall Corporation, Port Washington, NY), and washed two times with 200 µL of 8 M urea. The free cysteine side chains were alkylated with 10 mM methyl methanethiosulfonate solution in digestion buffer (1% sodium deoxycholate [SDC], 50 mM TEAB) for 30 min at room temperature, and the filters were then repeatedly washed with digestion buffer. Trypsin in digestion buffer was added (250 ng for TMT samples/500 ng for the IBAQ sample), and the sample was incubated at 37°C for 3 hr, then another aliquot of trypsin (250/500 ng) was added and incubated overnight. Digested peptides were collected by centrifugation at 9500 × g for 20 min, followed by a wash with 20 µL of the digestion buffer and centrifugation at 9500 × g for 20 min. Peptide samples for relative quantification were labeled using the five sets TMT 10plex isobaric reagents according to the manufacturer’s instructions (Thermo Scientific), combined into five pooled TMT samples, concentrated using vacuum centrifugation, and SDC was removed by acidification with 10% TFA (Trifluoroacetic acid) and subsequent centrifugation at 16,000 × g for 10 min. The digested sample for IBAQ label-free quantification was acidified with 10% TFA and centrifuged at 16,000 × g for 10 min to remove the precipitated deoxycholic acid.

The TMT sets were fractionated into 44 primary fractions by basic reversed-phase chromatography using a Dionex Ultimate 3000 UPLC system (Thermo Fischer Scientific, Waltham, MA). Peptide separations were performed on a reversed-phase XBridge BEH C18 column (3.5 μm, 3.0 × 150 mm, Waters Corporation) and a linear gradient from 3% to 40% solvent B over 17 min followed by an increase to 100% B over 5 min. Solvent A was 10 mM aqueous ammonium formate at pH 10.0, and solvent B was 90% acetonitrile, 10% 10 mM ammonium formate at pH 10.00. The primary fractions were concatenated into final 20 fractions (1+21+41, 2+22+42, … 20+40), evaporated, and reconstituted in 15 μL of 3% acetonitrile, 0.2% formic acid for nLC–MS analysis.

The pooled sample for the label-free quantification was fractionated into 44 primary fractions that were concatenated into 10 final fractions (1+11+21+31+41, 2+12+22+32+42, … 10+20+30+40). The final fractions were evaporated and reconstituted in 15 μL of 3% acetonitrile, 0.2% formic acid for nLC–MS analysis.

All samples were analyzed on an Orbitrap Fusion Tribrid mass spectrometer interfaced with Easy-nLC1200 liquid chromatography system (both Thermo Fisher Scientific). Peptides were trapped on an Acclaim Pepmap 100 C18 trap column (100 μm × 2 cm, particle size 5 μm, Thermo Fischer Scientific) and separated on an in-house packed analytical column (75 μm × 30 cm, particle size 3 μm, Reprosil-Pur C18, Dr. Maisch) using the linear gradients with 0.2% formic acid in water as a solvent A and 80% acetonitrile, 0.2% formic acid as solvent B with the flow of 300 nL/min.

Each TMT-labeled fraction was separated on a 90 min gradient from 5% to 35% B over 75 min, from 35% to 100% B over 5 min, and 100% B for 10 min. Each fraction for the label-free IBAQ quantification was injected three times and analyzed using the same elution gradient.

For the TMT-based relative quantification, MS scans were performed at 120,000 resolution, m/z range 380–1380 with the wide quadrupole isolation and AGC target 4e5; the most abundant precursors with charges 2–7 were selected for fragmentation over the 3 s cycle time with the dynamic exclusion duration of 60 s. Precursors were isolated with a 0.7 Da window, fragmented by collision-induced dissociation (CID) at 35% collision energy with a maximum injection time of 50 ms and AGC target 1e4, and the MS² spectra were detected in the ion trap followed by the synchronous isolation of the 10 most abundant MS² fragment ions within m/z range of 400–1400 and fragmentation by higher-energy collision dissociation at 65%; the resulting MS³ spectra were detected in the Orbitrap at 50,000 resolution with m/z range 100–500, maximum injection time 105 ms and AGC target 1e5.

For label-free quantification IBAQ experiment, MS scans were performed at 120,000 resolution, m/z range 380–1380 with the wide quadrupole isolation and AGC target 2e5; the most abundant precursors with charges 2–7 were selected for fragmentation over the 1 s cycle time with the dynamic exclusion duration of 45 s. Precursors were isolated with a 1.2 Da window, fragmented by CID at 35% collision energy with a maximum injection time of 50 ms and AGC target 1e4, and the MS² spectra were detected in the ion trap.

Peptide and protein identification and quantification was performed using Proteome Discoverer version 2.2 (Thermo Fisher Scientific) with Mascot 2.5.1 (Matrix Science, London, UK) as a database search engine. The Baker’s yeast (S. cerevisiae ATCC 204508/S288c) reference proteome database was downloaded from Uniprot (February 2018, 6049 sequences) and used for the database search on the TMT-based relative quantification files; the concatenated database containing the yeast sequences and the 48 UPS protein sequences was used for the processing of the UPS2-spiked files.

Precursor masses have been re-calibrated with the initial mass tolerance of 20 ppm prior to the main database search. For the TMT relative quantification data and the label-free IBAQ data, trypsin with one missed cleavage was used as a cleavage rule, MS peptide tolerance was set to 5 ppm, and MS² tolerance for identification was set to 600 milli mass units (mmu). Variable modifications of methionine oxidation and fixed modifications of cysteine methylthiolation were used for both sub-data sets. TMT-6 label on lysine and peptide N-termini was set as a fixed modification for the TMT data. Percolator was used for the peptide-spectrum match validation with the strict FDR threshold of 1%.

Precursor ion quantification was accomplished via the Minora feature detection node in Proteome Discoverer 2.2, with the maximum peak intensity values used for quantification. Abundance values of all unique and shared peptides were used for the IBAQ calculation. Abundances from the three technical replicates were averaged and divided by the number of theoretically observable peptides for a protein to yield the IBAQ intensity (the number of observable peptides being calculated using an in-house Python script). The known absolute amount values of the UPS2 standard proteins were used to scale the log-transformed IBAQ intensity values versus log-transformed protein concentration. Only UPS2 proteins with at least two identified unique peptides were used for scaling.

The TMT reporter ions were identified in the MS³ HCD spectra with a mass tolerance of 3 mmu, and the signal-to-noise (S/N) abundances of the reporter ions of all unique and shared peptides were used for relative quantification with the minimal average reporter S/N threshold set at 19 and the co-isolation threshold at 100. The resulting reporter abundance values for each sample were normalized within Proteome Discoverer 2.2 on the total peptide amount.

To extract intracellular amino acid, 1–2 mL of boiling 75% ethanol was poured directly onto the cell pellets. Samples were vortexed for 1 min, boiled for 3 min, and placed on ice until cool. Samples were centrifuged for 15 min at 13,000 × g at 4°C. Supernatants containing intracellular amino acids were collected and stored at −20°C until labeling and analysis. Amino acids were labeled using the SCIEX aTRAQ Reagents Application Kit (Danaher, Washington, DC) with aTRAQ Reagent Δ8 and mixed with internal standards pre-labeled with aTRAQ Reagent Δ0, according to manufacturer’s protocol. Samples were analyzed using a SCIEX QTRAP 6500+ system (Danaher, Washington, DC) with a Nexera UHPLC system (Shimadzu, Japan), on a SCIEX AAA column (150 × 4.6 mm) at an oven temperature of 50°C. Mobile phase A contained 0.1% formic acid and 0.01% heptafluorobutyric acid in water; mobile phase B contained 0.1% formic acid and 0.01% heptafluorobutyric acid in methanol. The flow rate was 0.8 mL min⁻¹, with a gradient profile was as follows: 0 min, 2% B; 6 min, 40% B; 10 min, 40% B; 11 min, 90% B; 12 min, 90% B; 13 min, 2% B; 18 min, 2% B. The retention time, precursor (Q1 m/z), and product (Q3 m/z) for each amino acid and internal standard were as described by the manufacturer’s protocol. The mass spectrometer was set to monitor the transitions with the following ion source parameters: CUR (curtain gas) 30, CAD (collision activated dissociation) MED, IS (ionization voltage) 5500, TEM (temperature) 500, GS1 (source gas 1) 60 and GS2 (source gas 2) 50, and with compound parameters: DP (declustering potential) 30, EP (entrance potential) 10, CE (collision energy) 30, and CXP (collision cell exit potential) 5.

For transcriptomics, the absolute concentrations of 31 transcripts with >10 FPKM, and covering the entire dynamic expression range, were measured using lysates of S. cerevisiae CEN.PK 113-7D cells from the J. Nielsen lab (Chalmers University of Technology, Göteborg, Sweden). Linear regression between the absolute concentrations of these mRNAs and their corresponding FPKM values from RNAseq was performed to obtain the slope and y-intercept, which were used to quantify all mRNA in this study. The calculated mRNA abundance was then scaled to the total RNA content measured by Qubit RNA HS Assay Kit (Thermo Fisher, Waltham, MA). For proteomics, linear regression was calculated between the known log10 UPS2 concentrations and the log10 IBAQ abundances for the corresponding proteins. The resulting regression coefficients were used to estimate the absolute concentrations of all detected yeast proteins.

Protein translation rate k_sP (Lahtvee et al., 2017; Schwanhäusser et al., 2011) for each protein j was calculated by:

The gene-specific protein degradation rate (k_dP) for S. cerevisiae was mined from Lahtvee et al., 2017. Protein degradation rate for S. pombe was mined from Christiano et al., 2014. Protein degradation rate for a cancer cell line (HeLa) was mined from Boisvert et al., 2012. Where the k_dP,j was unavailable in the mined datasets, the median of all k_dP in the dataset was used.

As transcript and protein abundance as well protein translation rates span several orders of magnitude, for ease of visualization we present most of our analyses using scaled expression levels. In the GR dataset, the scaling is done as follows: for each gene, linear regression is performed between the data being scaled (transcript abundance, protein abundance, protein translation rate, or amino acid abundance) and the growth rate. The y-intercept is obtained, and the data being scaled is divided by the y-intercept. This scaled data is then plotted on a log2 scale, which can be interpreted intuitively as the fold-change from when the cells are growing at a theoretical growth rate of 0. In the NM dataset, the scaling is done as follows: for each gene, the data being scaled (transcript abundance, protein abundance, protein translation rate, or amino acid abundance) is divided by the mean of the data in nitrogen-limited Phe and Ile as nitrogen source conditions (samples 34, 35, 36, 40, 41, and 42, in Supplementary file 1a). This scaled data is then plotted on a log2 scale, and here the interpretation is the fold-change from when cells are growing with severe nitrogen restriction (i.e. nitrogen is limiting and the nitrogen source is non-preferred).

Processed quantitative transcriptomics and proteomics data are in Supplementary files 1b,c. Processed intracellular amino acid concentrations are in Supplementary files 1d. Raw RNAseq data are available at ArrayExpress, accession E-MTAB-9117 (Yu et al., 2020b). The mass spectrometry proteomics data are deposited to the Proteome Xchange Consortium via the PRIDE (Perez-Riverol et al., 2019) partner repository with dataset identifier PXD021218 (Vorontsov and Nielsen, 2020). GO-slim terms for S. cerevisiae genes are available from the Saccharomyces Genome Database (Cherry et al., 2012; The Gene Ontology Consortium, 2019). GO-slim terms for S. pombe genes are available from PomBase (Lock et al., 2019; The Gene Ontology Consortium, 2019). GO terms for human genes are available from AmiGO 2 (AmiGO Hub et al., 2009; The Gene Ontology Consortium, 2019). Custom code are available in the GitHub repository https://github.com/SysBioChalmers/YeastAbsQuantForCCMRegulation (Yu, 2021).

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank the National Bioinformatics Infrastructure Sweden (NBIS) for discussions. We thank Pannipa Pornpitakpong and Alexandra Hoffmeyer (Technical University of Denmark) for conducting RNA sequencing. The Proteomics Core Facility of the University of Gothenburg is grateful to the Inga-Britt and Arne Lundbergs Forskningsstiftlese for the donation of the Orbitrap Fusion Tribrid MS instrument. This research was supported by funding from the Novo Nordisk Foundation (grant number NNF10CC1016517) and the Knut and Alice Wallenberg Foundation.

1. Received: December 13, 2020
2. Accepted: March 13, 2021
3. Accepted Manuscript published: March 15, 2021 (version 1)
4. Version of Record published: April 1, 2021 (version 2)

© 2021, Yu et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.