One-carbon metabolic enzymes are regulated during cell division and make distinct contributions to the metabolome and cell cycle progression in Saccharomyces cerevisiae

Abstract Enzymes of one-carbon (1C) metabolism play pivotal roles in proliferating cells. They are involved in the metabolism of amino acids, nucleotides, and lipids and the supply of all cellular methylations. However, there is limited information about how these enzymes are regulated during cell division and how cell cycle kinetics are affected in several loss-of-function mutants of 1C metabolism. Here, we report that the levels of the S. cerevisiae enzymes Ade17p and Cho2p, involved in the de novo synthesis of purines and phosphatidylcholine (PC), respectively, are cell cycle-regulated. Cells lacking Ade17p, Cho2p, or Shm2p (an enzyme that supplies 1C units from serine) have distinct alterations in size homeostasis and cell cycle kinetics. Loss of Ade17p leads to a specific delay at START, when cells commit to a new round of cell division, while loss of Shm2p has broader effects, reducing growth rate. Furthermore, the inability to synthesize PC de novo in cho2Δ cells delays START and reduces the coherence of nuclear elongation late in the cell cycle. Loss of Cho2p also leads to profound metabolite changes. Besides the expected changes in the lipidome, cho2Δ cells have reduced levels of amino acids, resembling cells shifted to poorer media. These results reveal the different ways that 1C metabolism allocates resources to affect cell proliferation at multiple cell cycle transitions.


Introduction
One-carbon (1C) metabolism encompasses the chemical reactions that move and use single-carbon functional groups. At the heart of 1C metabolism is tetrahydrofolate (THF; see Fig. 1). THF carries activated 1C units at various oxidation states (Appling et al. 2019). The most reduced form is methyl-THF, which donates the methyl group only to homocysteine, forming the C-S bond of methionine, and then, through S-adenosylmethionine (AdoMet, SAM), supplying all cellular methylations, including on phosphatidylcholine (PC) and histones (Fig. 1). The 1C unit of methylene-THF (CH 2 -THF) forms new C-C bonds in the synthesis of dTMP and the interconversions involving serine and glycine. Formyl-THF (CHO-THF) carries the most oxidized 1C units, forming C-N bonds in the synthesis of purines.
1C pathways are directly involved in the metabolism of amino acids and the synthesis of purines, thymidylate, and phospholipids (West et al. 1996;Fox and Stover 2008;Ducker and Rabinowitz 2017). As a result, 1C outputs govern vital cellular processes such as genome replication and maintenance (through nucleotide synthesis), response to oxidative stress (through glutathione synthesis), gene expression (through methylation of DNA and histones), among others (Fox and Stover 2008;Ducker and Rabinowitz 2017).
Despite the significance of 1C metabolism for cell proliferation and the vast literature dealing with 1C-based antiproliferative interventions, how the activity of 1C enzymes is regulated in the cell cycle of normal cells is not well understood. Changes in the levels of 1C enzymes are expected to change 1C metabolic outputs because the levels of 1C enzymes exceed by several-fold the folate metabolites, and the enzymes compete with each other for the available folate pools (Fox and Stover 2008;Lan et al. 2018). In mammalian cells, some reports suggested that mRNAs encoding 1C enzymes may be periodic in the cell cycle (summarized in (Lan et al. 2018)). However, based on aggregate transcriptomic datasets, only the levels of DHFRL1 (encoding mitochondrial dihydrofolate reductase) and TYMS (encoding thymidylate synthetase) change significantly in the cell cycle, peaking in the S phase (Santos et al. 2015). Likewise, in the budding yeast Saccharomyces cerevisiae, only the mRNAs encoding thymidylate synthase (CDC21) and a subunit of the mitochondrial glycine decarboxylase complex (GCV2) are periodic in the cell cycle, peaking at the G1/S transition, and the G2 phase, respectively (Santos et al. 2015). In addition to transcription, other possible layers of temporal control of 1C enzymes in the cell cycle include translational, post-translational, and differential subcellular localization. Indeed, in mammalian cells, serine hydroxymethyltransferase isoforms (SHMT2α and SHMT1) translocate in the nucleus during the S phase, presumably to provide 1C units needed for thymidylate synthesis and DNA replication (Woeller et al. 2007;Anderson and Stover 2009;Anderson et al. 2012;Lan et al. 2018). In HeLa cells blocked with hydroxyurea in the S phase, SHMT1 protein levels were elevated without concomitant changes in SHMT1 mRNA levels (Anderson et al. 2012).
In budding yeast, translational control could impose a temporal control of 1C metabolism in the cell cycle. The translational efficiency of mRNAs encoding several 1C enzymes was altered in the cell cycle based on ribosome profiling experiments previously reported in ribosomal protein mutants . These mRNAs included ADE17 (ATIC in humans; encoding both a 5-aminoimidazole-4-carboxamide ribonucleotide transformylase and inosine monophosphate cyclohydrolase activities), SHM2 (SHMT1 in humans; encoding the cytoplasmic serine hydroxymethyltransferase), and CHO2 [PEMT in humans; encoding a phosphatidylethanolamine (PE) methyltransferase].
The above enzymes play critical roles in 1C metabolism. Ade17p controls the use of 1C units, in the form of CHO-THF, for purine synthesis (Tibbetts and Appling 2000). The cytoplasmic serine hydroxymethyltransferase (Shm2p in yeast) is a key metabolic switch in 1C metabolism (Piper et al. 2000;Herbig et al. 2002). Shm2p controls the input of carbon in the pathway from serine. Shm2p also contributes to the cytoplasmic CH 2 -THF pool toward dTMP and methylations, or away from them, to make more serine for phospholipids (from glycine and 1C units) (Piper et al. 2000;Herbig et al. 2002). It is serine, but not glycine, that supports 1C metabolism in cancer cells (Labuschagne et al. 2014). Serine supplementation, through 1C metabolism, also regulates chronological lifespan in budding yeast (Enriquez-Hesles et al. 2021). Although Cho2p is not a folate-dependent enzyme, it catalyzes the first step in the conversion of PE to PC during the methylation pathway of de novo PC biosynthesis. It has been reported that PC synthesis is the major consumer of 1C-derived methyl groups in yeast (Ye et al. 2017) and likely in humans (Stead et al. 2006). Cho2p has been proposed to control the flow toward phospholipid synthesis but away from histone methylation (Ye et al. 2017). However, it is not known if the levels of any of the above enzymes change in the cell cycle of unperturbed cells in any system. Furthermore, despite the numerous ways 1C metabolism could impact cell division, besides DNA replication, if and how specific 1C enzymes might control other cell cycle transitions is unknown.
Here, we report that the levels of Ade17p and Cho2p, and to a lesser extent that of Shm2p, are periodic in the budding yeast cell cycle. We also present an analysis of size homeostasis, cell cycle kinetics, and metabolic outputs in cells lacking any of these enzymes. These parameters were altered in the mutants we examined but in distinct ways in each mutant. Lastly, our data argue that Cho2p and de novo synthesis of PC are required to establish the proper kinetics of nuclear division. Overall, this work provides new and important information on the abundance and roles of 1C enzymes during cell division.

Strains and media
All the strains used in this study are shown in the Reagent Table. For most experiments, the cells were cultivated in the standard, rich, undefined medium YPD (1% w / v yeast extract, 2% w / v peptone, and 2% w / v dextrose), at 30° (Kaiser et al. 1994). To generate the cho2Δ haploid strain, we sporulated and dissected the commercially available homozygous diploid cho2Δ/cho2Δ strain (see Reagent Table). For the experiments with cho2Δ cells, we also used synthetic minimal media (SMM), containing 0.17% w / v yeast nitrogen base without amino acids, 0.5% w / v ammonium sulfate, 2% w / v glucose, the required amino acid auxotrophies at standard concentrations (Kaiser et al. 1994) and, if indicated, 1 mM choline chloride.
Single-gene homozygous deletion strains, lacking ADE17, SHM2, or CHO2, were commercially available (see Reagent  Table). Their genotype was validated by PCR to confirm that the gene of interest was absent and replaced by the appropriate marker. To construct the ADE17-TAP strain (see Reagent Table), a PCR product was generated using plasmid pBS1539 as a template The pathway for the de novo synthesis of phosphatidylcholine, the major consumer of 1C units in the cell, and the step catalyzed by Cho2p are also indicated. (Puig et al. 2001), with primers Ade17-TAP-FWD and Ade17-TAP-REV, and used to transform strain BY4741. SHM2-TAP and CHO2-TAP strains were commercially available (see Reagent  Table).

Multiple correspondence analysis
Data collection and analyses were done as described previously (Bermudez et al. 2020;Polymenis 2020). All the relevant data are given in Supplementary File 1.

Immunoblot analysis
For protein surveillance, protein extracts were made as described previously (Amberg et al. 2006) and resolved on 12% Tris-Glycine SDS-PAGE gels, unless indicated otherwise. To detect TAP-tagged proteins with the PAP reagent (used at 1:4,000 dilution), we used immunoblots from the extracts of the indicated strains, as described previously (Blank et al. 2017Maitra et al. 2020Maitra et al. , 2022. Loading was measured with an anti-Pgk1p primary antibody (at 1:2,000; abcam, Cat#:ab38007), followed by a secondary antibody (at 1:1,500; Jackson Immunoresearch Laboratories, Alexa Fluor 488 AffiniPure Goat Anti-Mouse IgG (H + L)). Imaging and quantification were done as described previously (Blank et al. 2017Maitra et al. 2020Maitra et al. , 2022.

Centrifugal elutriation, cell size, and DNA content measurements
All methods have been described previously (Hoose et al. 2012;Soma et al. 2014). Briefly, after early G1 cells were collected, they were monitored at regular time intervals for cell size, budding, or DNA content. Samples were also assayed in downstream procedures, such as nuclear staining, as described in the relevant sections.

Metabolite profiling
The untargeted, primary metabolite, biogenic amine, and complex lipid analyses were done at the NIH-funded West Coast Metabolomics Center at the University of California at Davis, according to their established mass spectrometry protocols, as described previously Maitra et al. 2020Maitra et al. , 2022. The analysis was done from six independent samples in each case. The raw data for the primary metabolite measurements are in Supplementary File 3. The raw data for the primary amine measurements are in Supplementary File 4. The raw data for the complex lipid measurements are in Supplementary File 5. To identify significant differences in the comparisons among the different strains and media, based on peak intensities from each metabolite, we used the robust bootstrap ANOVA, as described below. Detected species that could not be assigned to any compound were excluded from the analysis. In the complex lipid analyses, we also excluded species that are not present in yeast, such as those containing polyunsaturated fatty acid side chains.

Fluorescence microscopy
To visualize the nuclear envelope, we followed the same procedures described previously (Maitra et al. 2022). Briefly, when the nucleus is near-spherical, the ratio of the long to short nuclear axes is 1. However, as the nucleus starts to expand, the ratio increases, signifying the elongation of the nuclear envelope (Fig. 6).

Histone methylation
Overnight cultures in YPD or SMM media of the indicated strains were diluted 1:100. To supplement SMM with choline, choline chloride was added at 1 mM, 2 hours after dilution. After several hours, cells were harvested at 1E + 07 cells/ml. Protein extracts were prepared and analyzed by SDS-PAGE and immunoblotting as described above. Primary antibodies were used to detect histone H3 (at 1:1,000 dilution; Abcam, Cat#ab1719) and H3K4me3 (at 1:2,000 dilution; Epicypher, Cat#13-0041), followed by an antirabbit secondary antibody (used at 1:2,000 dilution). Pgk1p was detected as described above. Protein bands were quantified with ImageJ. H3 and H3K4me3 signals were normalized to the respective Pgk1p bands. Methylation levels were then calculated by the ratio of the normalized H3K4me3 levels over the normalized H3 levels. The quantification is in Supplementary Fig. 3 and all the immunoblot source data are in Supplementary Fig. 4.

Statistical analysis, sample size, and replicates
For sample-size estimation, no explicit power analysis was used. All the replicates in every experiment shown were biological ones, from independent cultures. A minimum of three biological replicates was analyzed in each case, as indicated in each corresponding figure's legends. The robust bootstrap ANOVA was used to compare different populations via the t1waybt function, and the post hoc tests via the mcppb20 function, of the WRS2 R language package (Wilcox 2011;Mair and Wilcox 2020). We also used nonparametric statistical methods, as indicated in each case. The Kruskal-Wallis and posthoc Nemenyi tests were done with the posthoc.kruskal.nemenyi.test function of the PMCMR R language package. No data or outliers were excluded from any analysis.

Classification of existing cell cycle phenotypes of folate-dependent enzymes
Besides cdc21 (encoding thymidylate synthase), identified in the classic cell division cycle screen and shown to arrest in the S phase (Hartwell et al. 1974), there is limited information about cell cycle-related phenotypes of other folate-dependent enzymes. We generated a complete matrix of all the loss-of-function phenotypes associated with the 15 genes encoding enzymes that catalyze folate-dependent reactions (shown in bold in Fig. 1; Supplementary File 1) from the available data on the Saccharomyces Genome Database (Cherry et al. 2012). Sixty-six loss-of-function phenotypes are associated with the 15 folatedependent enzymes (Supplementary File 1/Sheet1). The most common phenotypes, reported for at least 6 of the 15 genes, are rather generic, reflecting altered competitive fitness in some conditions and chemical resistance. In contrast, only three genes have been associated with altered cell cycle progression: CDC21 (arrest in S phase), DFR1 (abnormal S phase), and MIS1 (G1 delay). To identify any patterns that may underlie the observed phenotypes of 1C mutants, we applied multiple correspondence analysis (analogous to principal component analysis, but for categorical data) to define and group those phenotypes and the genes that may be primarily responsible for them (Bermudez et al. 2020;Polymenis 2020).
The 66 different phenotypes can be reduced to 14 dimensions, with just 3 accounting for 68% of the observed variance ( Fig. 2a; Supplementary File 1/Sheet2). The major phenotypes contributing to Dimension 1 were abnormal cell morphology, meiotic arrest, lack of sporulation, cell cycle arrest, increased chitin deposition, decreased resistance to toxins, and petite colony formation (Supplementary File 1/Sheet3). By far, the primary gene driving this grouping was CDC21 (75% contribution; see Fig. 2b and Supplementary File 1/Sheet4). The major phenotypes contributing to Dimension 2 appear unrelated to cell cycle progression (e.g. abnormal endoplasmic reticulum morphology, increased RNA accumulation, and stress resistance; see Supplementary File 1/Sheet3). The classification was driven chiefly by gcv3 phenotypes (73% contribution; see Fig. 2b and Supplementary File 1/ Sheet4). The primary phenotypes contributing to Dimension 3 were also rather generic (e.g. increased compound excretion, decreased resistance to hyperosmotic stress, and absent or reduced utilization of nitrogen sources; see Supplementary File 1/Sheet3), driven by ade8 and ade3 phenotypes (making a combined 73% contribution).
Overall, we conclude that the phenotypic classification of folate-dependent enzyme mutants is dominated primarily by the phenotypes of just two, cdc21 and gcv3. As we mentioned earlier, nearly all information regarding cell cycle progression are from studies with cdc21, which was discovered decades ago in the classic cdc screen for cell cycle mutants (Hartwell et al. 1974). Given the importance of 1C pathways in cell proliferation and our previous results identifying ADE17, SHM2, and CHO2 among transcripts with altered translational efficiency in the cell cycle in ribosomal protein mutants , we decided to focus on their abundance in the cell cycle and their cell cycle-related phenotypes.

The levels of Ade17p and Cho2p peak late in the cell cycle
The steady-state levels of ADE17, SHM2, or CHO2 transcripts do not change in the cell cycle (Spellman et al. 1998;Santos et al. 2015;Blank et al. 2017Blank et al. , 2020. However, ribosome profiling experiments suggested that the translational efficiency of those transcripts changes in the dividing cells of ribosomal protein mutants . To monitor the levels of the corresponding proteins in the cell cycle, we used otherwise wild-type, haploid strains carrying alleles encoding C-terminal, TAP-tagged versions of these enzymes (Puig et al. 2001), expressed from their endogenous locations in the genome. To maintain the physiological coupling between cell growth and division, we used centrifugal elutriation to prepare growing, synchronous cultures (Aramayo and Polymenis 2017). In S.cerevisiae, budding marks the initiation of cell division, and in daughter cells, the cell size is a proxy for cell cycle position (Hartwell and Unger 1977;Johnston et al. 1977). Newborn daughter cells in early G1 were sampled at regular intervals as they progressed in the cell cycle, recording their size and budding (Fig. 3).
The levels of Ade17p-TAP increased markedly (∼15-fold) from the beginning to the end of the cell cycle (Fig. 3, left panels). Likewise, Cho2p-TAP levels were higher (∼4-fold) late in the cell cycle, when the cells were large and budded (Fig. 3, right panels). In contrast, Shm2p-TAP levels were the highest in the early G1 phase and then declined slightly (<2-fold) as the cells progressed in the cell cycle (Fig. 3, middle panels). These results show that the levels of Ade17p and Cho2p are periodic in the cell cycle. They are also consistent with the notion that translational control of these gene products contributes to their oscillation in dividing cells.

Loss of Shm2p or Ade17p delays START
Because cell size changes are often accompanied by an altered timing of cell cycle transitions, we first looked at the cell size distributions of homozygous diploid, asynchronous ade17Δ or shm2Δ cultures (Fig. 4). Cells lacking Ade17p or Shm2p are born at normal size ( Fig. 4a, middle panel), but they have a larger mean size (Fig. 4a, left panel), suggesting a delay at some later step in the cell cycle. We note that it was previously reported that cells carrying a catalytically dead SHM2 allele were bigger (Yang and Meier 2003), consistent with our observations. Next, we measured the kinetics of cell cycle progression from synchronous, elutriated cultures. These experiments allowed us to determine the size at which half the cells in the culture are budded (a.k.a. critical size), marking passage through START in late G1, when cells commit to a new round of cell division and initiation of DNA replication. Loss of either Shm2p or Ade17p increased the critical size for START (Fig. 4a, right panel). The cells grew larger before starting to bud compared with wild-type, consistent with the notion that they delay exiting from the G1 phase into the S. From the same experiments, we also measured the specific rate at which cells increase in size (k, Fig. 4b), as a measure of cell growth. Cells lacking Shm2p have a lower k, indicating a growth defect. Hence, shm2Δ cells have a longer G1 phase because they have a larger critical size (Fig. 4a, right) and reach that size at a slower rate (Fig. 4b). On the other hand, ade17Δ cells increase in size at a normal rate (Fig. 4b). These data point to a more specific role of Ade17p at START.

De novo synthesis of phosphatidylcholine is required for cell growth and size control
Exogenous choline can be salvaged through the CDP-choline pathway to generate PC (Dowd et al. 2001). Hence, to properly query the requirement for CHO2 and de novo synthesis of PC, we evaluated haploid cho2Δ cells not only in the standard, rich undefined medium (YPD) as above but also in synthetic minimal medium (SMM), without or with choline (at 1 mM) supplementation. Asynchronous cells lacking Cho2p had a larger overall size in all media tested (Fig. 5a, left panels), including in minimal media supplemented with choline (Fig. 5a, left, middle panel), and they were also born slightly bigger (Fig. 5a, middle panels). Note that these cell size values for the wild-type cells were much lower than the corresponding ones in the previous experiments (see Fig. 4) because we used haploid cells here.
To gauge cell cycle kinetics in all these conditions, we examined synchronous elutriated cultures, as described above (see Fig. 4). Remarkably, cho2Δ cells had a significantly bigger critical size, albeit only in minimal media (Fig. 5a, right panels). The latter effect was evident regardless of exogenous choline supplementation (Fig. 5a, right, middle, and bottom panels). Lastly, in all media tested, cho2Δ cells had a lower specific rate of size increase ( Fig. 5b; P < 0.05 based on the same tests as in Fig. 4), consistent with growth defects. These data suggest that Cho2p and the de novo pathway for PC synthesis are required for optimal cell growth and size homeostasis even when exogenous choline is available. To ensure that the budding index accurately reflects the position of cho2Δ cells in the cell cycle, in a separate elutriation experiment, we also measured the DNA content of CHO2 + and cho2Δ cells in SMM with or without exogenous supplementation with choline ( Supplementary Fig. 2). The same conclusions were reachedthat cho2Δ cells grow in size slower and they are larger when they initiate DNA replication.

Loss of de novo phosphatidylcholine synthesis dampens the kinetics of nuclear elongation late in the cell cycle
We had previously reported that increased lipogenesis promotes nuclear elongation and division (Maitra et al. 2022). To test if de novo synthesis of PC impinges on these late mitotic events, we analyzed the nuclear morphology of cho2Δ cells and their otherwise wild-type CHO2 counterparts from synchronous elutriated cultures. Early G1 daughter cells were isolated as in Fig. 5 and allowed to progress in the cell cycle in minimal media with or without choline. At regular intervals, the cells were fixed for immunofluorescence against the nucleoporin Nsp1p. Note that the nuclear envelope remains intact during mitosis in S. cerevisiae. Before metaphase, the nucleus elongates across the bud neck, between the mother cell and the large bud. The ratio of the long: short axes of the nucleus provides a metric of nuclear elongation (Maitra et al. 2022). A ratio of one corresponds to a spherical nucleus, while values greater than 1 reflect nuclear elongation. In minimal media without choline, shortly after wild-type cells reach their critical size (∼41 fL, see Fig. 6a, bottom), their nuclei start elongating, reaching their maximum length at ∼46 fL (Fig. 6b, bottom). Cells lacking Cho2p have a larger critical size (∼45 fL, see Fig. 6a, top), and their nuclei are maximally elongated at ∼48 fL (Fig. 6b, top). However, instead of a well-formed, sharp peak, cho2Δ cells elongate their nuclei more gradually than wildtype cells (Fig. 6b, top). Choline supplementation did not suppress this behavior (if anything, it might have delayed somewhat the full elongation of the nuclei of cho2Δ cells; compare the top panels in Fig. 6b). Importantly, as we will describe next, we confirmed that the cells take up the exogenous choline because cells lacking Cho2p have ∼15-fold lower choline levels in SMM medium compared with wild-type cells, restored to wild-type levels upon exogenous choline supplementation. These results suggest that cho2Δ cells elongate their nuclei less coherently or synchronously than wild-type cells, implicating the de novo pathway of PC synthesis in nuclear elongation and division.

Metabolic profiling of cells lacking Ade17p, Shm2p, or Cho2p
Because 1C enzymes control several key metabolic outputs (Fig. 1), we sought to place the cell cycle phenotypes we uncovered in the context of possible metabolic changes. First, we asked if there was any change in histone methylation. One might expect that reduced PC synthesis in cho2Δ cells might lead to an increased use of 1C units for histone methylation (Ye et al. 2017). In the same strains and media that we described above (see Figs. 4 and 5), we measured the levels of trimethylated histone H3 on Lys4 (H3K4me3) compared with the total levels of histone H3 (see Materials and Methods). We found no significant changes (P > 0.05 in the Kruskal-Wallis test) in the ratio of H3K4me3:H3 levels among the different media or strains (ade17Δ, shm2Δ, or cho2Δ cells) we examined (Supplementary Fig. 3).
To gain a broader overview of metabolic alterations in the mutants that we examined, we measured the distribution of the steady-state levels of metabolites and lipids by mass spectrometry (see Materials and Methods). Different analytical platforms were used to separate and identify primary metabolites (Supplementary File 3), biogenic amines (Supplementary File 4),  File 5). We combined the identified primary metabolite and biogenic amine datasets, while the complex lipid datasets were kept separate. To identify metabolite and metabolic pathway enrichment, we used the MetaboAnalyst platform (Chong et al. 2019). Pairwise comparisons of the metabolome between wild-type vs ade17Δ and wild-type vs shm2Δ cells growing in undefined, rich YPD medium revealed few significant changes (Fig. 7a, top), and these changes were very similar in cells lacking Ade17p or Shm2p. No group was significantly enriched among the metabolites with lower levels (1.5-fold change and P < 0.05) in ade17Δ or shm2Δ cells. On the other hand, among metabolites with significantly higher levels in shm2Δ cells (19 compounds), the only group that was significantly enriched were metabolites associated with gluconeogenesis (P = 0.0458 with Holm-Bonferroni correction). The same trend was evident in ade17Δ cells (P = 0.0646 with Holm-Bonferroni correction from the 23 metabolites with significantly higher levels). Although there were few metabolite changes associated with the loss of either Shm2p or Ade17p (Fig. 7a, top), these mutants seemed to acquire gluconeogenic signatures, even though they were growing in a glucose-replete, rich medium. Lastly, while in both shm2Δ and ade17Δ cells there was a reduction in the levels of some complex lipids compared with wild-type cells ( Supplementary Fig. 6a,  top), the lipids that changed in abundance were different. Compared with wild-type cells, shm2Δ cells had lower levels of triacylglycerols (P = 1.79E−05, with Holm-Bonferroni correction). On the other hand, ade17Δ cells had lower levels of lyso-PC (P = 4.22E−10, with Holm-Bonferroni correction), PC (P = 3.98E −06, with Holm-Bonferroni correction), and monoacylglycerophosphocholines (P = 0.0124, with Holm-Bonferroni correction). These data suggest that the loss of Shm2p reduces triacylglycerols, while the loss of Ade17p may direct lipid synthesis away from PC, resembling qualitatively cells lacking Cho2p, as described below. Nonetheless, despite these statistically significant differences, the overall number of lipid species with altered levels and their fold-change in abundance in ade17Δ and shm2Δ cells compared with wild-type cells was relatively small (Supplementary Fig. 6a, top).
The metabolite changes were profound in CHO2 vs cho2Δ cells, with >200 primary metabolites and biogenic amines displaying significantly altered steady-state levels (see Fig. 7a, bottom), in all three different media we tested (YPD, SMM, SMM + choline). The compounds with reduced levels in cho2Δ cells were significantly enriched for groups associated with amino acid metabolism and the overall pathway of "aminoacyl-tRNA biosynthesis" (P = 0.0049 with Holm-Bonferroni correction). In other words, amino acid pools were depleted in cho2Δ cells cultured in the rich YPD medium. The same pattern was observed in the minimal (SMM) medium, comparing CHO2 vs cho2Δ (P = 0.0321 with Holm-Bonferroni correction; see Fig. 7b, left panel). Strikingly, compared with the metabolite repertoire in wild-type cells in YPD vs SMM media (see Supplementary Fig. 5a), which queries nutrient effects, metabolites associated with the overall pathway of "aminoacyl-tRNA biosynthesis" were also significantly enriched (P = 8.08E−05 with Holm-Bonferroni correction). Intracellular amino acid levels were higher in wild-type cells cultured in the rich YPD medium vs the minimal medium ( Supplementary Fig. 5b, left panel). These results suggest that the loss of de novo PC synthesis in cho2Δ cells in rich undefined media leads to metabolic changes typically associated with nutrient shifts to poorer, minimal media. We note that, as expected, choline levels were dramatically reduced (∼25-fold) in cho2Δ cells in the minimal SMM media (Fig. 7b, left panel). But the methylation potential of these cells remained high, having elevated AdoMet (S-adenosylmethionine) levels (Fig. 7b, left panel), consistent with the unchanged histone methylation levels that we reported above ( Supplementary Fig. 3). Regarding changes in the lipidome, in both YPD and SMM media, loss of Cho2p significantly reduces the levels of >60 complex lipids ( Supplementary Fig. 6a, bottom panels). As expected, cho2Δ cells had lower levels of lyso-PC, PC, and diacylglycerophosphocholines (P < 3E−06, with Holm-Bonferroni correction).
These conclusions in CHO2 vs cho2Δ cells were reinforced with the metabolite measurements in the minimal SMM medium supplemented with 1 mM choline. Wild-type cells showed no significant changes (see Supplementary Fig. 5a, left panel). But the changes in cho2Δ cells in SMM medium were reversed by adding Fig. 6. Loss of de novo phosphatidylcholine synthesis in cells lacking CHO2 dampens the kinetics of nuclear elongation late in the cell cycle. a) Synchronous cultures of the indicated strains and conditions were obtained by elutriation (see Materials and Methods), from which the percentage of budded cells (y-axis) is shown against the mean cell size (in fL; x-axis). b) Cells were processed for fluorescence microscopy to visualize the nucleus from the same samples as in a, as described in Materials and Methods. Nuclear axes ratio (y-axis) is shown against the mean cell size (in fL; x-axis), from >2,000 cells/strain. Loess curves and the SE at a 0.95 level are shown. The values used to generate the graphs are in Supplementary File 2/Sheet4. 1 mM choline. The pathway of "aminoacyl-tRNA biosynthesis" became significantly enriched in cho2Δ cells compared with wild-type (P = 3.18E−06 with Holm-Bonferroni correction), as reflected in their higher amino acid levels (Fig. 7b, left panel). In wild-type cells, the addition of choline did not cause any changes in amino acid levels ( Supplementary Fig. 5b, right panel). Taken together, these results strongly suggest that the inability to carry out de novo PC synthesis in cho2Δ cells leads to profound changes in amino acid homeostasis. Lastly, the lipid changes mentioned above between wild-type and cho2Δ cells were also completely reversed by supplementation with 1 mM choline ( Supplementary Fig. 6a, bottom right panel), with cells lacking Cho2p even having higher levels of lyso-PC and diacylglycerophosphocholines (P < 0.012, with Holm-Bonferroni correction).

Discussion
Our data reveal that the levels of some 1C enzymes are dynamic in the cell cycle, and the corresponding loss-of function mutations lead to distinct cell-cycle phenotypes and metabolic changes. We place these findings in the context of the existing literature and discuss possible implications.
1C pathways receive enormous attention in diseases, especially cancer (Locasale 2013;Labuschagne et al. 2014;Rosenzweig et al. 2018;Reina-Campos et al. 2019). For example, during dTMP synthesis by thymidylate synthase (Cdc21p in S. cerevisiae), the methylene-THF coenzyme is the 1C unit source and the reducing power, getting oxidized in the process to DHF. To maintain the THF levels needed in the cells, dihydrofolate reductase (Dfr1p in S. cerevisiae) reduces DHF back to THF (Fig. 1). Methotrexate is one of the oldest chemotherapeutics and a potent and specific inhibitor of dihydrofolate reductase (Williams et al. 1979), leading to THF deficiency, accounting for its antiproliferative properties. In a recent extensive transcriptomic profiling of 1,454 metabolic enzymes across 1,981 tumors among 19 cancer types, 1C enzymes had the highest score for being consistently overexpressed compared with noncancerous tissue samples (Nilsson et al. 2014).
In humans, loss-of-function mutations have been described for ATIC (ADE17 ortholog), SHMT1 (SHM2 ortholog), and PEMT (CHO2 ortholog). Patients with ATIC mutations accumulate AICAR (5-aminoimidazole-4-carboxamide-1-β-D-ribofuranoside) and display a wide range of developmental abnormalities (Ramond et al. 2020). At the cellular level, in patient-derived fibroblasts, a reduced ability to form purinosomes and drive purine biosynthesis correlates with the clinical phenotypes of individual patients Fig. 7. Changes in primary and biogenic amine metabolites in cells lacking Ade17p, Shm2p, or Cho2p. a) Metabolites whose levels changed in the indicated pairwise comparisons and media (shown in parenthesis above each panel) were identified from the magnitude of the difference (x-axis; Log2-fold change) and statistical significance (y-axis), indicated by the red lines. The analytical and statistical approaches are described in Materials and Methods. The values used to generate the graphs are in Supplementary File 2/Sheet6. Note that the lowest calculated P-values from the robust ANOVA were at the 0.0001 level. b) Boxplots of the levels of the amino acids (x-axis) detected in the biogenic amine dataset (see Materials and Methods), shown as Log2-transformed relative abundance between wild-type (CHO2) and cho2Δ cells (y-axis), from six independent samples in each case. The levels of choline and AdoMet (S-adenosylmethionine) from the same measurements are also shown. The values used to generate the graphs are given in Supplementary File 2/Sheet7. (Baresova et al. 2012). Interestingly, however, moderately reducing ATIC activity in adult mice with a specific inhibitor is beneficial, leading to an increase in AICAR and AMP-activated kinase (AMPK) activity, ameliorating metabolic syndrome phenotypes (Asby et al. 2015). How does one reconcile these observations? It is reasonable to expect that during embryogenesis and early adult life, when cell proliferation underpins developmental milestones in animals, loss-of-function perturbations in 1C metabolism would lead to profound phenotypes, but perhaps not so later in life. This is also the basis for the selective effects of antifolates in chemotherapy. Indeed, the LD 50 for methotrexate given to 5-week-old mice is 59 mg/kg, whereas that for 16-week-old mice it is 284 mg/kg (Freeman-Narrod and Narrod 1977). In yeast, perhaps analogous to a proliferating animal tissue, a complete loss of ATIC, in double ade16,17Δ mutants, leads to inviability unless the cells are supplemented with exogenous adenine (Tibbetts and Appling 2000). The single ade17Δ cells we evaluated here allowed us to reveal rather specific effects at START (Fig. 4). Even though the growth rate is unaffected, the larger critical size of ade17Δ mutants is consistent with the interpretation that cells may monitor some aspect of purine biosynthesis before committing to a new round of cell division. In this regard, having the levels of Ade17p under translational control, rising as cells exit G1 (Fig.  3), provides another way for the cell to link protein synthesis and overall biosynthetic capacity with metabolic pathways needed for cell division, contributing to the general task of coupling cell growth with cell division.
There are no reported SHMT1 mutations causing human pathologies. In mice, however, Shmt1 knockout animals are a genetic model of folate deficiencies (MacFarlane et al. 2008). SHMT1 has been shown to directly interfere with folate metabolism and its loss leads to folate deficiency, sensitizing embryos to neural tube defects (Beaudin et al. 2011). Adult Shmt1 −/− mice, however, are fertile and healthy. Their methionine plasma levels were lower (at 43 μM), irrespective of whether they were on a folate-replete or folate-limited diet (MacFarlane et al. 2008). Wild-type mice fed a folate/choline-deficient diet for 5 weeks had reduced methionine plasma levels too (by ∼25%, from 60 μM to 45 μM), while serine and glycine levels did not change (MacFarlane et al. 2008). Yet, significant Shmt1-dependent changes to methylation capacity, gene expression, and purine synthesis were not observed (Macfarlane et al. 2011). Overall, it appears that loss of the cytoplasmic serine hydroxymethyltransferase in mammals leads to minimal, if any, adverse phenotypes in adults. In the constantly dividing S. cerevisiae cells, on the other hand, the lower growth rate and larger critical size of shm2Δ cells (Fig. 4) likely reflect the dependency on cytoplasmic hydroxymethyltransferase for optimal cell proliferation.
Mammals rely mainly on the salvage PC synthesis pathway, incorporating dietary choline into PC, except for the liver, where Pemt catalyzes significant de novo PC synthesis (Walkey et al. 1997). The corresponding knockout mice were normal, including in their hepatocyte morphology, bile composition, or plasma lipid levels (Walkey et al. 1997), arguing for minimal contributions, if any, for the de novo PC pathway in mammals. In yeast, Cki1p encodes choline kinase, which catalyzes the first step in PC synthesis in the choline salvage pathway. We previously showed that cki1Δ cells have a slightly smaller critical size in rich, YPD medium (Maitra et al. 2022). Here, we found that cho2Δ cells have a significantly larger critical size in minimal media, regardless of whether exogenous choline becomes available (Fig. 5). They also have a reduced specific rate of size increase in all media, even in YPD (Fig. 5b), suggestive of rather broad impacts on cellular physiology. Overall, unlike the situation in animal cells, our results point to important roles in proliferating S. cerevisiae cells for de novo PC synthesis. Our data suggest that nuclear elongation late in the cell cycle is one such role.
Nuclear division places heavy demands for new membrane material in the relatively short time it takes for mitosis. This is true not only in fungi, which undergo closed mitosis but in animal cells as well, when the fragmented nuclear membrane must reform around each of the nuclei in telophase. In S. pombe and S. cerevisiae, loss-of-function mutations in lipid synthesis change the size and shape of the nucleus (Santos-Rosa et al. 2005;Witkin et al. 2012;Walters et al. 2012;Siniossoglou 2013;Kume et al. 2017;Zach and Prevorovsky 2018). In animal and yeast cells, inactivating acetyl-CoA carboxylase, the rate-limiting lipogenic enzyme, blocks cells in mitosis, and exogenous fatty acids cannot rescue the cell cycle arrest (Schneiter et al. 1996;Al-Feel et al. 2003;Scaglia et al. 2014). Conversely, we have shown that increased lipogenesis promotes nuclear division (Maitra et al. 2022). In this context, it might appear unsurprising that disrupting PC synthesis in cho2Δ cells alters the kinetics of nuclear elongation (Fig. 6). However, we also showed that nuclear elongation kinetics were unaffected in cki1Δ cells with impaired PC synthesis from the salvage pathway (Maitra et al. 2022). In contrast, the data in this report highlight the need for de novo PC synthesis during nuclear division. Our data also revealed unexpected metabolome changes in cells lacking Cho2p (Fig. 7). To our knowledge, this is the first time that such a link between de novo PC synthesis and amino acid levels has been reported. Astonishingly, even in rich media, cho2Δ cells behave as if they are in minimal media (Figs. 7 and Supplementary Fig. 5). How the inability to synthesize PC de novo also leads to a lowering of amino acid levels is unclear.
In summary, the presented results revealed unappreciated contributions of 1C metabolism to cell cycle progression. 1C metabolism is an excellent platform for integrating various metabolic inputs and outputs with cell division. This study adds to effort to decipher how specific cell cycle processes are linked to the enzymes of 1C metabolism. The results we presented could also inform research that may not be focused on cell cycle progression, especially in the context of several recent studies linking 1C metabolism with longevity mechanisms in multiple systems Enriquez-Hesles et al. 2021;Annibal et al. 2021;Lionaki et al. 2022).

Data availability
Strains and plasmids are available upon request. The authors affirm that all data necessary for confirming the conclusions of the article are present within the article, figures, and tables. The Supplementary Material is available through figshare (https:// doi.org/10.6084/m9.figshare.21816270).

Conflicts of interest statement
The authors declare no conflict of interest that might raise any questions of bias in the work and in the article's conclusions, implications, or opinions.
Supplemental material available at G3 online.