Riboregulation of Enolase 1 activity controls glycolysis and embryonic stem cell differentiation

Differentiating stem cells must coordinate their metabolism and fate trajectories. Here, we report that the catalytic activity of the glycolytic enzyme Enolase 1 (ENO1) is directly regulated by RNAs leading to metabolic rewiring in mouse embryonic stem cells (mESCs). We identify RNA ligands that speciﬁcally inhibit ENO1’s enzymatic activity in vitro and diminish glycolysis in cultured human cells and mESCs. Pharmacological inhibition or RNAi-mediated depletion of the protein deacetylase SIRT2 increases ENO1’s acetylation and en-hances its RNA binding. Similarly, induction of mESC differentiation leads to increased ENO1 acetylation, enhanced RNA binding, and inhibition of glycolysis. Stem cells expressing mutant forms of ENO1 that escape or hyper-activate this regulation display impaired germ layer differentiation. Our ﬁndings uncover acetylation-driven riboregulation of ENO1 as a physiological mechanism of glycolytic control and of the regulation of stem cell differentiation. Riboregulation may represent a more widespread principle of biological control. wild-type or mutant DNA. While ENO1 is expressed under an EF1alpha promoter, the ﬂuorescent protein cerulean is expressed under an additional, separate IRES promoter to facilitate an efﬁcient cell sort. The generated plasmids were nucleofected into mESCs using the Nucleofector 4D system according to the manufacturer’s guidelines (Nucleofector Kit P3, program CG-104, one million cells, 16 m g template vector and 4 m g Cas9 vector per nucleofection). Single cell sorting of double positive cells (RFP and cerulean) was performed 48 hours after nucleofection (FACSAria Fusion Sorter). Upon clonal expansion, successful insertion was tested by PCR to screen for homozygous insertions of the ENO1 wild-type or mutant DNA sequence in the ROSA locus (Phire Tissue Direct PCR Master Mix). The mutants of ENO1 were based on the ﬁndings of RBDmap data (Castello et al., 2016; Horos et al., 2019). antibody (mouse: 1:400; F3165, Sigma) for detection of the protein–RNA signal. To measure ENO1 expression by ﬂuorescence imaging, the same primary antibody concentrations were used. For detection, the Alexa 488-conjugated secondary antibodies were used (Donkey anti-Rabbit IgG (H+L) Alexa Fluor 488; A-21206, Thermo Fisher Scientiﬁc; Donkey anti-Mouse IgG (H+L) Alexa Fluor 488; A-21202, Thermo Fisher Scientiﬁc). of statistical analysis of these performed using Graphpad Prism rotating. The IPs were washed once with RIPA lysis buffer, twice with RIPA high salt buffer (500 mM NaCl, 50 mM Tris-HCl pH 7.5, 1% NP40, 0.1% SDS, 0.5% Na-Deoxycholate, Protease Inhibitor), twice with Wash buffer (10 mM Tris-HCl, 0.2% Tween-20, 1mM EDTA) and once again with RIPA lysis buffer. The samples were eluted with glycine as previously described. 10% of the IP was used for western blotting to assess the immunoprecipitation efﬁciency. RNase inhibitor and 10 ng total C.elegans RNA (spike-in control) were added to the remainder of the IP samples and the inputs. Subse-quently, the samples were digested with pre-warmed (at 37 (cid:4) C for 20 mins) proteinase K mix at 55 (cid:4) C for 30 mins (ﬁnal concentration: 2 m g/ m L enzyme). Acid phenol/chloroform/isoamyl alcohol (pH 6.5) was added to the input and IP samples and they were incubated at 37 (cid:4) C for 5 mins while shaking at 1200 rpm. The samples were then loaded on 2 ml Phase Lock gel Heavy tubes, incubated at 37 (cid:4) C for centrifugation (cid:4) The lysate hand careful pipetting from the (16 fractions collected of approximately 600 m l). For the protein precipitation, 150 m l of 100% Trichloroacetic acid (TCA) was added and left on ice for 30 minutes. The individual fractions were centrifuged at full speed and 4 (cid:4) C for 20 minutes. The TCA supernatant was carefully removed, and the pellets were washed once with 1 ml cold acetone (stored at (cid:3) 20 (cid:4) C). The samples were vortexed, and an additional centrifugation step was performed at full speed and 4 (cid:4) C for 30 minutes. The supernatant was again carefully removed, and the pellet was air-dried. Finally, the pellets were taken up in 1 3 loading buffer containing benzonase and used for SDS-PAGE and immunoblotting. in lysis buffer (20 mM Tris-HCl pH 7.5, 500 mM NaCl, 5 mM b-2ME, 5% glycerol, 40 mM imidazole, 0.01% NP40), supplemented with RNase A, Turbo DNase and Protease Inhibitor to the lysis buffer. For efﬁcient lysis, the lysates were processed with a microﬂuidizer. The His-tagged thioredoxin was used for the puriﬁcation of ENO1 using a HisTrap HP column on an A ¨ kta go protein puriﬁcation system. The protein was eluted with increasing concentrations of imidazole and the protein-containing fractions were veriﬁed on a Coomassie gel. The solubility tag was cleaved by HRV 3C protease during FLAG-tagged ENO1 with and without RNase treatment, endogenous ENO1 was knocked down in HeLa cells and the ENO1 mutant-containing plasmids were transfected as described above. FLAG IPs were performed using pre-coupled Flag M2 magnetic beads (25 m l for 0.5 mg protein lysate) under constant rotation at 4 (cid:4) C for 1 hour. The protein complexes were eluted using 20 m lFLAG peptide (1 mg/ml) by incubation on ice for 1 hour. Half of the eluate was incubated with 1 m l RNase A (ﬁnal concentration 1 mg/ml) or water at 37 (cid:4) C for 15 mins. 2.5 m l were used as input for the enzymatic activity assay, omitting the coupling to the ELISA plate, and following the manufacturer’s instructions. C/minute, (cid:4) in in for metabolites.


INTRODUCTION
Glycolysis represents a central metabolic pathway in all living organisms. Regulation of glycolysis is critical for stem cell differentiation and cancer biology (Gu et al., 2016;Kondoh et al., 2007). Pluripotent stem cells must meet their requirements for continuous self-renewal utilizing glycolysis to provide building blocks for the biosynthesis of nucleotides, peptides, and lipids. While glycolysis is required for stemness, a shift toward oxidative phosphorylation (OXPHOS) accompanies lineage differentiation (Folmes et al., 2011). However, the mechanisms underlying this metabolic rewiring remain to be better understood.
RNA-binding proteins (RBPs) have been recognized for decades as regulators of metabolism (reviewed by Muckenthaler et al., 2017) but have only recently received broader attention (Adeli, 2011). RBPs facilitate short-and long-term metabolic adjustments of cells undergoing cell division and differentiation (reviewed by Esparza-Moltó and Cuezva, 2020) and can themselves integrate metabolic stimuli, for example, through posttranslational modifications (Choudhary et al., 2014), changes in localization (Tischbein et al., 2019) or metabolite availability (Kim and Myong, 2016). Recently, many central metabolic enzymes, including glycolytic enzymes, have been identified to bind RNA in different cell types and organisms (reviewed by Hentze et al., 2018). One of these RNA-binding metabolic enzymes is Enolase 1 (ENO1, Castello et al., 2012;Beckmann et al., 2015;Matia-Gonzá lez et al., 2015), which catalyzes the reversible interconversion between 2-phosphoglycerate (2-PG) and phosphoenolpyruvate (PEP). ENO1 is commonly overexpressed in malignant cells (reviewed by Capello et al., 2011) and has proven to be a promising, clinically relevant therapeutic target, particularly in pancreatic cancer (Cappello et al., 2009;Cappello and Novelli, 2013;Hsiao et al., 2013;Song et al., 2014;Principe et al., 2015). Increased ENO1 expression correlates with cell proliferation (Muller et al., 2012) and cancer progression in vivo (Capello et al., 2016). ENO1 binds RNA in different organisms including bacteria (K€ uhnel and Luisi, 2001;Morita et al., 2004), yeast (Entelis et al., 2006;Baleva et al., 2015Baleva et al., , 2017, and mammals (Herná ndez-Pé rez et al., 2011), but apart from its involvement in bacterial RNA degradation (K€ uhnel and Luisi, 2001;Morita et al., 2004) and its role in tRNA targeting to mitochondria in yeast (Entelis et al., 2006;Baleva et al., 2015Baleva et al., , 2017, relatively limited functional insights have been gained, especially in mammalian systems. Here, we find that human ENO1 binds hundreds of mRNAs of the cellular transcriptome via specific binding regions. Synthetic RNA ligands corresponding to these regions inhibit ENO1's enzymatic activity in vitro, diminish glycolysis in HeLa cells, and specifically alter glycolytic metabolite levels and serine synthesis in pluripotent mouse embryonic stem cells (mESCs). Under physiological conditions, the metabolic shift from glycolysis to OXPHOS during stem cell differentiation occurs concomitantly with an increase in ENO1's RNA binding. However, when mutating ENO1 to be hyper-inhibited by RNA, stem cells are dramatically compromised in their differentiation toward the definitive endoderm, whereas RNA-binding-deficient ENO1 promotes endodermal differentiation. ENO1's RNA binding is activated by acetylation, as shown by pharmacological, RNAi and cell differentiation experiments. RNA-mediated inhibition of enzymatic activity-riboregulation-constitutes a physiologically relevant form of metabolic control, which plays a relevant role during stem cell differentiation.

Human Enolase 1 is a bona fide RNA-binding protein
We first confirmed that human ENO1 binds RNA in HeLa cells ( Figure 1A). To this end, we used the T4 polynucleotide kinase (PNK) assay (Richardson, 1965), which relies on UV crosslinking to establish a covalent bond between protein and RNA, combined with cell lysis and limited RNase treatment. RNA molecules crosslinked to ENO1 are immunoprecipitated and used as the input for PNK-mediated 5 0 end RNA labeling with radioactive ATP. We observed that with an increasing RNase I concentration, the radioactive signal collapses to a sharp band just slightly exceeding the molecular mass of ENO1 (signal above 48 kDa in 32 P-RNA autoradiograph; Figure 1A), indicating that ENO1 associates with RNA in HeLa cells.
ENO1 is among the top 100 most abundant proteins in the cell (Nagaraj et al., 2011). For the enzyme-RNA interaction to play a role, a considerable fraction of ENO1 needs to be associated with RNA. We addressed this question by exposing RNasetreated or untreated lysates to sucrose density gradient centrifugation and found around 10% of HeLa cell ENO1 to be sensitive to RNase treatment ( Figure 1B), indicating that in highly glycolytic cells only a small fraction of ENO1 is bound by RNA. This raises the question of whether the fraction of RNA-bound ENO1 changes in response to physiological stimuli. We investigate this question further in Figure 4.
We next determined ENO1's RNA ligands by applying the enhanced crosslinking and immunoprecipitation (eCLIP) protocol (Figures S1A-S1C, Van Nostrand et al., 2016). We ascertained that ENO1 interacts with a wide range of RNAs in HeLa cells with a preference toward the 5 0 untranslated region (5 0 UTR), 3 0 UTR and coding region of spliced mRNAs ( Figure S1D). Based on the DEWseq analysis (Sahadevan et al., 2022) of the RNA crosslink sites, we identified approximately two thousand direct ENO1-binding regions across the transcriptome . Interestingly, the two top-scoring linear sequence motifs ( Figure S1E) jointly account for only $22% of all ENO1-binding sites. When assessing the identity of the proteins encoded by the ENO1 ligand mRNAs, we found that many encoded proteins are involved in biological processes and have molecular functions relevant for the RNA lifecycle such as splicing and translational initiation ( Figure S1F and S1G). Principally, ENO1 interacts with a large pool of mRNAs without a prevalent linear RNA sequence motif and with a tendency for mRNAs that encode proteins involved in RNA-related processes. This led us to investigate whether the interactions between ENO1 and RNA are specific and subject to regulation.
Enolase 1 binds specific RNA ligands in vitro and in vivo First, we set out to assess the specificity of ENO1's RNA binding. To this end, we validated six ENO1 ligand RNAs, identified by eCLIP, in RNA immunoprecipitation (RIP) coupled with quantitative reverse transcription polymerase chain reaction (qRT-PCR) experiments, confirming their enrichment over a control IP with IgG antibody of the same isotype ( Figures 1D and S1H). By contrast, RNAs of similar or higher abundance that were not previously identified by eCLIP are not significantly enriched, suggesting that ENO1 recognizes specific cellular RNAs.
Next, we synthesized RNAs of 35 nucleotides in length that either correspond to ENO1-binding regions or GC contentmatched controls, derived from the same mRNAs (schematic in Figure 1E; sequences are given in Table S1). We used these in an orthogonal assay to validate the eCLIP results and further assess the specificity of binding. Representative ligand and control RNAs, derived from the PABPC1 5 0 UTR, were analyzed in a competition electromobility shift assay (EMSA) using recombinant human ENO1 (Figures 1F and 1G;K i(target) : 27 ± 19 nM; K i(control) : 2,587 ± 9 nM, Figure S1I). Highly consistent results were obtained for two additional ligand and control pairs derived from the FTH1 and PTP4A1 mRNAs, respectively (Figures 1G, S1J, and S1K). Using NMR, we observed RNA-induced chemical shift perturbations and line broadening of ENO1 resonances in 1 H, 15 N-HSQC spectra, confirming direct RNA binding in vitro Figures 1H and S1L). Taken together, ENO1 binds RNA at numerous transcriptomic sites in human cells with two orders of magnitude difference between specific and non-specific interactions.
To explore the importance of ENO1's RNA binding, we knocked down endogenous ENO1 by RNAi, and collected comparative RNA sequencing data to identify possible effects of ENO1 on its RNA targets ( Figure S2A). Despite an efficient knockdown of ENO1 ( Figures S2B and S2C), we could not obtain convincing evidence for a regulatory effect of ENO1 on the transcriptome in general or the ENO1 targets determined by eCLIP in particular ( Figures S2A, S2D, and S2E). We cannot presently exclude that such effects may have been missed due to the limited engagement of ENO1 with cellular RNAs ( Figure 1B) or ENO1 primarily affecting mRNA function without changes in abundance. Nonetheless, recent findings highlighted the regulatory potential of RNA on protein function, including vault RNA-mediated regulation of p62's role in autophagy (Horos et al., 2019), the inhibition of SHMT1's activity by SHMT2's 5 0 UTR (Guiducci et al., 2019), and classic work on the control of E. coli RNA polymerase by 6S RNA (Wassarman and Storz, 2000). We therefore tested whether RNA binding might affect ENO1's enzymatic activity.

RNA ligands inhibit ENO1's activity in vitro
To test the impact of RNA on ENO1's activity, we purified recombinant human ENO1 and assayed its enzymatic activity in vitro (Figures 2A, 2B  and Nucleolin of the same experiment. (B) Top: HeLa cell lysates were treated with RNase I/A/T1 or left untreated, subjected to a sucrose density gradient (5%-25%), centrifugation and fractionation. A RNase-sensitive protein moves from the heavy to the lighter fractions (light). Bottom: quantification of the biological replicates for ENO1 for the experimental setup in the top panel (SD, n = 3). (C) Volcano plot of differential crosslinking site occurrences as determined by the Differential Expressed Window sequencing (DEWseq) tool; each dot corresponds to a window of the genomic region (50 nucleotides), gray coloring indicates significant enrichment in ENO1 IPs (independent hypothesis weighting (IHW)-adjusted p value < 0.1, log 2 FC > 0.5). The data are based on five independent biological replicates and were normalized to the background. Only the positive enrichment is displayed. Green circles indicate ENO1's target sites selected for further experiments, and dark gray circles represent three additional significantly enriched genes selected for experiments in (D). See also Figures S1A-S1G. (D) Formaldehyde (0.1%)-crosslinked RNA-IP (RIP) of ENO1 and an isotype-matched IgG from HeLa cells. qRT-PCR for six genes previously identified in eCLIP ( Figure 1A) and three control genes. The enrichment is calculated relative to a spike-in control and normalized to the mean of the IgG enrichment. Statistically significant differences were detected using two-way ANOVA and Tukey-correction for multiple comparison testing (n = 3). Statistical significance is represented as follows: p value of 0.0001 to 0.001, ***p value of 0.001 to 0.01, **p value of 0.01 to 0.05, * and p value of R0.05, not indicated. See also Figure S1H. (E) Schematic of ENO1's binding profile as identified by eCLIP relative to the input control used for the identification of target and control sites. Every line represents an accumulation of crosslink sites at an individual nucleotide. (F) Competition electromobility shift assay (EMSA) for target and control RNA using radioactively labeled PABPC1 target 35-mer as a probe and unlabeled competitor RNA from either the target or control region as indicated in (E). (G) Inhibition constants (K i ) for target and control RNAs for the binding to ENO1wt as determined by EMSA (n = 3). The K i was calculated using a non-linear fitting with least squares regression. See also Figures S1I-S1K. (H) Overlay of 1 H, 15 N-HSQC spectra of free ENO1 (red) and ENO1 incubated with 2-fold excess of RNA (FTH1 18-mer start, black; Figure S1L). Full titration points (in 1:0.1/0.2/0.4/0.8/1.2/2 ratios) of some residues are shown in insets.
with ENO1wt as measured by competition EMSA (Figures 1F,  1G, and S1I-S1K). Importantly, in contrast to its wild-type counterpart ENO1down displays no discernible alteration in its enzymatic activity with any of the RNAs tested ( Figures 2D, S2K, and S2L). Thus, both the use of control RNAs and evaluation of the ENO1down mutant corroborate that RNA specifically riboregulates ENO1's activity in vitro.
ENO1's glycolytic substrates compete with RNA in vitro Next, we tested whether ENO1's enzymatic substrates and RNA compete for ENO binding by performing EMSA experiments using the two ENO1 substrates of the forward and backward reaction, respectively, 2-PG or PEP (Lebioda et al., 1993) as competitors. For these experiments, we used metabolite concentrations that are close to the expected concentrations in cells (Park et al., 2016). While both ENO1 substrates compete with RNA for ENO1 binding, 3-phosphoglycerate (3-PG), the immediate precursor of 2-PG with an identical molecular mass used as a specificity control, fails to compete ( Figures 2E and 2F). Thus, we observe specific competition between substrates and an RNA ligand for binding to ENO1.

ENO1's RNA binding affects glycolysis in HeLa cells
In addition to the ENO1down mutant, we next generated a version of ENO1 with increased RNA-binding capacity in cellulo, referred to here as ENO1up. The design of ENO1up was also guided by RBDmap data (Castello et al., 2016) and entails the change of three lysine residues (K89A/ K92A/K105A).  (B) Representative image of a proximity ligation assay (RNA-PLA) for ENO1wt with its mRNA target FTH1 (red dots). ENO1 is being detected using an anti-FLAG antibody and the endogenous ENO1 is knocked down using siRNAs. The red PLA signal is presented as a maximal projection of a z-stack (10 pictures for 10-mm stack). Pictures for the DAPI (nuclear staining) and cellular outline (gray) are taken in one plane. The scale bars represent 20 mm. (H) Immunoprecipitates of transiently expressed ENO1wt in siENO1-treated HeLa cells were exposed to RNase A or left untreated. The FLAG-HA tagged ENO1wt was eluted with FLAG peptide and used as input for ENO1 activity assays (SD, n = 3). The two-tailed Student's t test is used to detect statistically significant differences. Statistical significance is represented as follows: p value of 0.01 to 0.05, * and p value of R0.05, not indicated. See also Figure S3J.

Article
After reducing endogenous ENO1 levels in HeLa cells by RNAi and exogenously expressing the respective FLAG-tagged ENO1 variants, ENO1up displays increased RNA binding compared with ENO1wt (PNK assay; Figures 3A and 3F), while the recombinant ENO1up has a comparable K i value for target and control PABPC1 to ENO1wt (EMSA; Figures 1G and S3D). We further investigate this difference between the in cellulo and in vitro RNA-binding of ENO1up in Figure 4.
In line with the in vitro data ( Figures 2C and S2H-S2J), ENO1down displays substantially decreased RNA binding in HeLa cells relative to ENO1wt ( Figures 3A and 3F). We independently confirmed the differential RNA binding of the ENO1up and ENO1down mutants using an immunofluorescence-based, UV crosslinking-independent RNA proximity ligation assay (RNA-PLA; Zhang et al., 2016) that enables the in situ detection of endogenous or tagged proteins with their RNA targets (see STAR Methods). ENO1's association with the FTH1 mRNA ligand was validated by the combination of an antisense probe hybridizing close to ENO1's FTH1 mRNA-interaction region and an antibody specifically recognizing the FLAG-tagged ENO1 variants. Using this orthogonal assay, we validated the differential RNA binding of ENO1wt ( Figures 3B and 3E), ENO1down ( Figures 3C and 3E), and ENO1up ( Figures 3D and 3E) in HeLa cells, ensuring that the expression levels and localization of the ENO1 variants were comparable ( Figure S3A) and that the PLA signal is specific ( Figures S3B and S3C).
When we tested these mutants for their ability to rescue glycolysis (lactate accumulation in the medium) in HeLa cells after knockdown of the endogenous ENO1, ENO1wt expectedly rescued lactate production ( Figure 3F). ENO1down, the RNAbinding-deficient mutant, has a comparable activity to the wild-type protein in HeLa cells as expected when only 10% of endogenous ENO1 is RNA bound ( Figure 1B), showing that the K343A mutation does not incapacitate the enzyme. In contrast, ENO1up fails to rescue the knockdown-induced inhibition of lactate accumulation ( Figure 3F), although it is fully active when tested in the absence of RNA in vitro (Figures 3G,S3H,and S3I). The activity measurements in vitro were controlled with a mutant lacking enzymatic activity (ENO1as, E295A/D320A/ K394A, Figures S3H and S3I, Kang et al., 2008).
We next probed whether RNA directly interferes with ENO1's activity in HeLa cells and whether differential RNA binding of ENO1 is the cause of the differential capacity of the ENO1 variants to rescue lactate production. We immunoprecipitated the three ENO1 variants from lysates of cells lacking the endogenous protein. We then treated these immunoprecipitates with RNase or left them untreated, eluted the ENO1 variants with FLAG peptide, and assayed their enzymatic activity . While the activity of ENO1down is irresponsive to the RNase treatment, ENO1wt is more active after RNase treatment. Furthermore, ENO1up also responds to RNase treatment and displays a tendency to be affected by the RNase treatment even more strongly. These experiments strongly support the concept that endogenous cellular RNAs interfere with ENO1's activity and that the ENO1 variants represent a powerful tool to assess the importance of these regulatory interactions.
We complemented these findings by nucleofection experiments with synthetic 35-mer ligand and control RNAs into HeLa cells ( Figures S3K and S3L). The results unambiguously confirm specific riboregulation of lactate production by ENO1's RNA ligands and fully support the notion that RNA binding interferes with ENO1's enzymatic activity in cells.

Acetylation augments ENO1's RNA binding
What may explain the difference between the enhanced RNA binding of the ENO1up mutant in cellulo (Figures 3E and 3F) and the normal RNA binding of the recombinant protein in vitro ( Figure S3D)? Considering that the ENO1up mutant represents a change of three lysine residues to alanine, we hypothesized that a post-translational lysine modification such as ubiquitination or acetylation in cellulo could activate ENO1's RNA binding. Multiple experimental interrogations yielded no evidence for relevant ubiquitination (data not shown). By contrast, treatment of HeLa cells with sodium butyrate, an inhibitor of protein deacetylases, profoundly induced ENO1's acetylation and RNA binding ( Figure 4A). When we subjected ENO1's IPs from HeLa cells treated with sodium butyrate to mass spectrometry and compared them to untreated samples, we identified several acetylated lysine residues. While the expected general trend of increased lysine acetylation of ENO1 following sodium butyrate treatment is apparent ( Figures 4B-4D), only the peptide encompassing K81 and K89 was found to be significantly more acetylated ( Figure 4D). Interestingly, K89 is one of the amino acids mutated in ENO1up (K89A/K92A/K105A).
SIRT2 had previously been implicated in ENO1's deacetylation (Cha et al., 2017;Hamaidi et al., 2020). Thus, we knocked down SIRT2's expression with siRNAs ( Figures S4A and S4B) and assessed the consequences on the RNA binding of (E) Experimental setup as in (B) for the peptide of acetylated lysine 103/105. (F) Representative PNK for FLAG-tagged ENO1wt, ENO1up and ENO1down in HeLa cells after siRNA-mediated knockdown of SIRT2. Western blotting for the FLAG-tagged ENO1 variants, SIRT2 to assess the knockdown efficiency, and Nucleolin as a loading control. RNA binding as detected by the autoradiograph was normalized to the FLAG IP efficiency. Statistically significant differences were detected using two-way ANOVA and Sidak-correction for multiple comparison testing (SD, n = 3). See also Figures S4A and S4B. (G) SIRT2 siRNA-treated or control siRNA-treated lysates of ENOwt, ENO1up, or ENO1down expressing HeLa cells were treated with RNase I/A/T1 or left untreated, subjected to sucrose density gradient (5%-25%) centrifugation and fractionation. Quantification of biological replicates for ENO1 immunoblots are given on the right (SD, n = 3). The two-tailed Student's t test is used to detect statistically significant differences. (H) Representative PNK for FLAG-tagged ENO1wt, ENO1up, ENO1down, and an acetylation-mimicking version of the ENO1up mutant (ENO1KtoQ) in HeLa cells. Western blotting was performed for the FLAG-tagged ENO1 variants and nucleolin as a loading control. RNA binding as detected by the autoradiograph was normalized to the FLAG IP efficiency. Statistically significant differences were detected using two-way ANOVA and Sidak-correction for multiple comparison testing (SD, n = 3). Statistical significance is represented as follows: p value of <0.0001, ****p value of 0.0001 to 0.001, ***p value of 0.001 to 0.01, **p value of 0.01 to 0.05, * and p value of R0.05, not significant. (I) Experimental setup as in (H) for a charge-mimicking version of the ENO1up mutant (ENO1KtoR) in HeLa cells. Article wild-type ENO1 and our ENO1 mutants. RNA binding of ENO1wt and ENO1down increased when knocking down ENO1's putative deacetylase, while ENO1up remained unaffected ( Figure 4F). This result implicates SIRT2 as the relevant deacetylase involved in the regulation of ENO1's RNA binding and suggests that the ENO1up mutation mimics acetylated ENO1. Sucrose density gradient centrifugation analysis (experimental setup as in Figure 1B) was then used to further quantify the impact of the ENO1 mutations on RNA binding and to assess the role of acetylation for ENO1's RNA binding in cells (Figure 4G). To this end, we digested lysates from HeLa cells treated with control siRNAs or siRNAs targeting SIRT2 mRNA and expressing ENO1wt, ENO1up, or ENO1down with RNase or left them undigested. We then assessed the differences by western blotting and detected that ENO1wt is more sensitive to RNase treatment than ENO1down, as expected ( Figure 4G, top and middle panel). By contrast, ENO1up is nearly 4 times more sensitive to RNase treatment, confirming that these mutations enhance RNA binding ( Figure 4G, bottom panel). Modulating the acetylation status of these ENO1 mutants by knocking down SIRT2 confirms the PNK results ( Figure 4F) and shows that increased acetylation profoundly augments ENO1wt's RNA binding (from 11% ± 6% to 34% ± 4%; Figure 4G, top panel), and ENO1down's by $3.5-fold (from 0% ± 7% to 26% ± 3%; Figure 4G, middle panel). As predicted, ENO1up fails to increase its RNA binding upon SIRT2 KD ( Figure 4G, bottom panel), in agreement with our PNK results ( Figure 4F). These data support the hypothesis that one (or more) of the amino acids K89, K92, and K105 are controlled by SIRT2 activity.
To exclude the possibility that the increase in RNA binding upon SIRT2 KD is caused by off-target effects, we validated our findings using two different pharmacological inhibitors of SIRT2, SirReal2 (Rumpf et al., 2015), and Thiomyristoyl (Jing et al., 2016;Figures S4C and S4D). Treatment of HeLa cells with either of these two inhibitors consistently increases ENO1's RNA binding. Thus, three different modes of SIRT2 modulation confirm SIRT2 as an important modulator of ENO1's RNA binding.
To further test the concept that ENO1up mimics the acetylated state of ENO1, we mutated the lysines that are changed to alanine in ENO1up (K89, K92, and K105) to the more conventionally used acetylation-mimic, glutamine (ENO1KtoQ). In support of the above-mentioned results and interpretation, ENO1KtoQ shows the same enhancement of RNA binding as ENO1up compared to ENO1wt ( Figure 4H). To examine the effect of a K89, K92, and K105 mutation that maintains the positive charge but alters the identity of the amino acids, we replaced the three lysines with arginines, generating ENO1KtoR. We tested its RNA binding in comparison with the other ENO1 variants ( Figure 4I). ENO1KtoR phenocopies ENO1wt, indicating that ENO1up mimics the acetylated form of the protein by loss of the positive charge.

ENO1's RNA association increases during differentiation
To explore physiological functions of the ENO1-RNA interaction, we chose mESCs. Similar to cancer cells, mESCs utilize glucose as a major energy source in the undifferentiated state (Gu et al., 2016;Kondoh et al., 2007). To directly test the inhibitory effect of RNA on ENO1's enzymatic activity in mESCs and to explore how ENO1's riboregulation alters cellular metabolism and metabolites, we nucleofected control or ligand PABPC1 RNA. As previously seen in HeLa cells ( Figures S3K and S3L), nucleofected ligand RNAs specifically inhibit lactate accumulation in the medium ( Figure 5D). 13 C glucose tracing experiments show that glucose progresses through the first steps of glycolysis and its branching pathways without significant differences imposed by ENO1's ligand RNA (see levels of fructose-1,6-bisphosphate, ribose-5-phosphate, and glycerol-3-phosphate, Figure 5A). However, metabolites downstream of ENO1 are significantly diminished when the specific ENO1 RNA ligand was used (see levels of PEP, pyruvate, and lactate, Figure 5A), demonstrating Figure 5. Metabolic rewiring upon riboregulation and time-dependent changes in ENO1's RNA binding during mESC differentiation (A) mESCs were incubated with 13 C-containing glucose for 60 min following nucleofection of either control or ligand PABPC1 RNAs. Metabolites in bold were directly measured, with black indicating no change and red a downregulation of the metabolite upon exposure to the RNA ligand. The two-tailed Student's t test is used to detect statistically significant differences (SD, n = 3). Statistical significance is represented as follows: p value of <0.0001, ****p value of 0.0001 to 0.001, ***p value of 0.001 to 0.01, **p value of 0.01 to 0.05, * and p value of R0.05, not significant. (B) Levels of non-essential amino acids determined in cell lysates prepared 60 min after nucleofection of control or ligand PABPC1 into mESCs. The two-tailed Student's t test is used to detect statistically significant differences (SD, n = 3). Statistical significance is represented as follows: p value of 0.01 to 0.05, * and p value of R0.05, not indicated. (C) Same as (B), except that control or ligand FTH1 RNA was used. Statistical significance is represented as follows: p value of <0.0001, ****. (D) Three distinct sets of target and control RNAs (5 mM) were nucleofected into mESCs. Upon nucleofection of the control or target RNAs, the lactate accumulation in the medium was measured after 30, 60, and 90 min and used to estimate the accumulation rate by calculating the slope. The R 2 value was used as a quality control. The two-tailed Student's t test is used to detect statistically significant differences (SD, n = 3). (E) Lactate accumulation rate of pluripotent mESCs after LIF withdrawal for a period of 7 days (SD, n = 3). The two-tailed Student's t test is used to detect statistically significant differences. Statistical significance is represented as follows: p value of 0.01 to 0.05, *. (F) Oxygen consumption rate of cells was measured with the Oxytherm System (SD, n = 3). The two-tailed Student's t test is used to detect statistically significant differences. Statistical significance is represented as follows: p value of 0.01 to 0.05, *. See also Figure S5A. (G) Formaldehyde (0.1%)-crosslinked RNA-IP of ENO1 or an isotype-matched IgG from pluripotent mESCs, cells differentiated for 3, 5, and 7 days (ÀLIF). qRT-PCR for five specific and two control mRNAs. The RNA enrichment is calculated relative to the mean of the IgG samples and normalized to the input. Statistically significant differences were determined using two-way ANOVA and Sidak-correction for multiple comparison testing. Statistical significance is represented as follows: p value of <0.0001, ****p value of 0.0001 to 0.001, ***p value of 0.001 to 0.01, **p value of 0.01 to 0.05, * and p value of R0.05, not significant.
(H) ENO1's IPs from pluripotent mESCs, cells differentiated for 3, 5, and 7 days (ÀLIF), followed by immunoblotting of acetylated lysine. The ratio of acetyl(K) signal in comparison to ENO1 staining was calculated for three biological replicates. Statistically significant differences were determined using an unpaired Student's t test and highlighted in blue, relative to the ratio in pluripotent cells (SD, n = 3). See also Figure S5B. a clear, direct, and specific effect of riboregulation on the metabolome of mESCs. Interestingly, we also detected that 3-PG is reduced upstream of the ENO1 ''block,'' suggesting that ENO1 inhibition may serve to supply serine biosynthesis with 3-PG, which branches off from glycolysis just upstream of ENO1. To investigate this further, we measured serine levels after nucleofection with either of two ENO1 ligands (PABPC1, Figure 5B; FTH1, Figure 5C) and found that in both instances serine levels are increased following ENO1's riboregulation while other, nonessential amino acids are not impacted. These data represent direct evidence for ENO1's ligand RNAs rewiring mESCs' carbon metabolism.
Removal of the leukemia inhibitory factor (LIF) from the culture medium of mESCs induces differentiation, accompanied by a decrease in glycolysis and increased respiration (Figures 5E and 5F, Williams et al., 1988;Gu et al., 2016). Of note, the decrease in glycolysis is accompanied by increased RNA binding after LIF withdrawal for 7 days (PNK; Figure S5A). To temporally resolve the increase in RNA binding over the course of mESC differentiation, we performed RIP-qRT-PCR experiments for the ligand and control mRNAs previously validated in HeLa cells ( Figure 1D). We detect only minimal changes in ENO1's RNA association during the first 3 days, a modest increase in its binding to some of the ligands after 5 days ( Figure 5G, FTH1, PABPC1, and PPIA), and a pronounced, significant enrichment for four of the five ligands after 7 days without LIF ( Figure 5G). This rise in RNA binding during mESC differentiation is accompanied and may be caused by an increase in ENO1's acetylation ( Figures 5H and S5B).
Unfortunately, the RNA nucleofection protocol is incompatible with meaningful mESC differentiation analyses. To test the importance of ENO1's riboregulation in mESC differentiation, we therefore first used unperturbed mESCs (Sladitschek and Neveu, 2019), withdrew LIF for a period of 7 days, and sorted cells that were positive for the expression of Brachyury (blue fluorescent protein [BFP]-positive), which is primarily found in cells differentiating toward the primitive streak (Rivera-Pé rez and Magnuson, 2005), or Eomes (mCherry-positive), which is predominantly expressed in the definitive endoderm (Arnold et al., 2008). We detected that lactate accumulation in the medium of Eomes + cells significantly exceeds that of Brachyury + cells ( Figure S5C), suggesting that the differentiation to the definitive endoderm may require sustained glycolysis in comparison to the primitive streak. Of note, ENO1 binding to RNA correlates inversely (compare Figure S5D). Sorted Eomes-expressing cells show decreased RNA binding and trend toward decreased ENO1 acetylation in comparison to the Brachyury-expressing cells ( Figures S5E and S5F).
Differential RNA association of ENO1 alters stem cell differentiation To examine whether the correlation between ENO1's RNA binding, its acetylation state and lactate accumulation in cells reflects a causal requirement for riboregulation of ENO1 during ESC differentiation, we introduced murine versions of the ENO1 variants characterized before (Figure 3) into both alleles of the Rosa26 locus by CRISPR-Cas9 genome editing and subsequently knocked out endogenous ENO1. The different heterolo-gous forms of ENO1 are expressed at similar levels to each other ( Figure S6A). The lactate accumulation in the medium of ENO1wt cells is somewhat less than seen in control cells, indicating potential differences in expression levels between the endogenous ENO1 protein and the protein expressed from a different locus and promoter ( Figure 6B). As previously observed in HeLa cells, ENO1up displays increased RNA binding leading to a decreased enzymatic activity as reflected by a reduction in lactate production; likewise, ENO1down shows a decrease in RNA binding compared with ENO1wt ( Figure 6A, compare Figure 3F).
Independent clones of these cell lines were subjected to LIF withdrawal and analyzed for differentiation into the different germ layers. Engineered mESCs expressing ENO1wt differentiated normally into the distinct germ layers, as assessed by qRT-PCR analysis of the expression of respective marker genes ( Figures 6C and  S6B). By contrast, ENO1up-expressing cells fail profoundly in their differentiation to definitive endoderm and neuroectoderm (Figure 6E), while the expression of primitive streak and mesodermal markers was quite variable and statistically not significantly affected. We also noticed that ENO1down cells, where ENO1's activity escapes riboregulation, conversely show increased differentiation toward the definitive endoderm ( Figure 6D).
To corroborate that the phenotypic changes of ENO1up-expressing cells is a consequence of diminished ENO1 activity, we fused an auxin-inducible degron tag to the C terminus of endogenous ENO1 of both alleles in mESCs carrying the OsTir1 receptor in the TIGRE locus. We then triggered ENO1 degradation by the addition of auxin for 48 h at the previously determined critical point of differentiation of 4 days, where we detected an increase in ENO1's RNA association ( Figure 5G). When depleting ENO1 from differentiating mESCs at this point, we observe specific, defective differentiation toward neuroectoderm and definitive endoderm, phenocopying cells expressing ENO1up ( Figures 6D and 6E). This experiment shows that the expression of ENO1up, which is strongly inhibited by RNA ( Figure 3J), phenocopies a loss of ENO1 at a critical time point during differentiation.
Taken together, our experiments demonstrate the physiological importance of ENO1's riboregulation for mESC differentiation, and especially for the formation of the endodermal germ layer. As such, they uncover a central form of regulated stem cell differentiation.

DISCUSSION
Here, we elucidate a physiological role of the RNA-binding activity of mammalian ENO1. Through multiple lines of evidence, both in vitro and in cell systems, we show that RNA specifically controls ENO1's activity. We also provide direct experimental evidence for a physiological role of ENO1's riboregulation during mESC differentiation, especially for endoderm formation. Future experiments will directly examine how the resulting alterations in energy metabolism and metabolites affect cell differentiation.
Regulation of glycolysis by the collective mRNA transcriptome (''crowd control'') Our biochemical studies and the data using the ENO1 mutants show that RNA directly interacts with ENO1 ( Figure 1) in a way that is mutually exclusive with substrate binding (Figure 2 The conventional function of the mRNA transcriptome is based on its protein-encoding potential. Our data show that many, possibly thousands of sites within mammalian cells' transcriptomes display specific ENO1 binding. The PABPC1, FTH1, and PTP4A1 ligands-derived from eCLIP targets selected from a cloud of hundreds of similar ones ( Figure 1C), largely mRNAs ( Figure S1D)-riboregulate ENO1 activity . Riboregulation of ENO1 affects mESC differentiation (A) mESC lines were established for endogenous ENO1 knockout and constitutive transgenic expression of FLAG-HA-tagged ENO1wt, ENO1up, and ENO1down from the Rosa26 locus. RNA binding was measured for the indicated ENO1 variants by FLAG IP and PNK after UV crosslinking in pluripotent mESCs (SD, n = 3). The statistically significant differences were detected using the two-tailed Student's t test when comparing to ENO1wt. Statistical significance is represented as follows: p value of 0.01 to 0.05, * and p value of R0.05, not indicated. See also Figure S6A.
(B) Measurement of the lactate accumulation in the culture medium for three independent clones of ENO1wt, ENO1up, and ENO1down in pluripotent mESCs (SD, n = 3), and the statistically significant differences were detected using the two-tailed Student's t test when comparing to ENO1wt. Statistical significance is represented as follows: p value of 0.01 to 0.05, * and p value of R0.05, not indicated. (C-E) Differentiation of mESCs and qRT-PCRs of lineage marker for ENO1wt (C), ENO1up (D), and ENO1down (E) after 7 days of LIF withdrawal for three independent clones from (A) and (B) (SD, n = 3). The statistically significant differences were detected using two-way ANOVA with Tukey-corrected multiple comparison testing when comparing to ENO1wt. Statistical significance is represented as follows: p value of <0.0001, ****p value of 0.0001 to 0.001, ***p value of 0.001 to 0.01, **p value of 0.01 to 0.05, * and p value of R0.05, not indicated. (F and G) Cell lines were established for a transient ENO1 depletion using the auxin-inducible degron system. 4 days after withdrawing LIF from the mESC medium, DMSO (F) or auxin (G) was added for 48 h. The cells were differentiated for a total of 7 days, and qRT-PCRs were performed for differentiation and lineage marker. Cell viability was monitored regularly by microscopic inspection, and no adverse effects of treatment noticed. This experiment was performed with two independent clones in three biological replicates (SD). The statistically significant differences were detected using two-way ANOVA with Sidak-corrected multiple comparison testing when comparing to the DMSO-treated samples. Statistical significance is represented as follows: p value of <0.0001, ****p value of 0.0001 to 0.001, ***p value of 0.001 to 0.01, **p value of 0.01 to 0.05, * and p value of R0.05, not indicated. See also Figures S6D and S6E. in vitro ( Figure 2) and in cellulo ( Figures 5D, S3K, and S3L). These data suggest an additional function that is shared between these ENO1-binding mRNAs, and hence a collective function of the mRNA transcriptome, namely ''crowd-controlled'' riboregulation. We do not know formally whether all ENO1 eCLIP targets function in the same way, but the data presented here would not favor an opposing view.  Figures 4F and 4G) also identify a deacetylase able to regulate ENO1's RNA association. Future experiments will identify the ENO1 acetylases and deacetylases involved in the regulation of mESC differentiation. Interestingly, recent findings show that the glycolytic flux is rerouted in T cells depleted for SIRT2 (Hamaidi et al., 2020), which in turn promotes serine production and nucleotide synthesis. We find that riboregulation of ENO1, which is supported by increased ENO1 acetylation, also alters glycolytic metabolites and increases cellular serine levels ( Figure 5). How does acetylation activate ENO1's RNA binding, especially since the loss of a positive charge more often diminishes protein interactions with negatively charged nucleic acids (reviewed by Blee et al., 2015)? We envisage that acetylation of the lysines altered in ENO1up induces a conformational change to a more ''open state'' that facilitates RNA binding and thus promotes inhibition of enzyme function. While such a model is plausible, alternative scenarios cannot be excluded at this stage, and structural investigations should help to resolve this question.
Riboregulation as a path to different forms of regulatory drugs for glycolysis? For the past decade, research groups have set out to identify ENO1 inhibitors for their use in the treatment of cancer, type 2 diabetes, and infectious diseases (Cho et al., 2017;Jung et al., 2013;Leonard et al., 2016;Pietkiewicz et al., 2009;Satani et al., 2016). Many of the inhibitors interact with the magnesium ion in the active site of ENO1 or chelate the metal ion rendering the enzyme inactive (Jung et al., 2013;Lebioda et al., 1993;Poyner and Reed, 1992). Our findings could pave the way toward a different class of compounds that exploit the cells' endogenous riboregulatory mechanisms for therapeutic intervention (Yu et al., 2018).
Limitations of the study Although many metabolic enzymes and nearly all glycolytic enzymes have recently been found to bind RNA, this study is focused on ENO1. Future work will have to examine other enzyme RBPs to identify common features and distinctions. We describe a specific function of ENO1's RNA-binding activity, riboregulation, and we do not exclude alternative or additional functions of ENO1's RNA-binding activity. Since metabolic remodeling and regulation of glycolysis represents a common feature of cell differentiation and in cancer biology, it will be important to explore the riboregulation of ENO1, described here, in these other biological and pathophysiological contexts.
The structural basis of the specific interaction between ENO1's ligand RNAs and the enzyme also remains to be defined to better understand how RNA binding inhibits the catalytic function of the enzyme and to determine how acetylation controls ENO1's RNA binding. In addition to K89, it remains to be determined whether acetylation of other amino acids or other posttranslational modifications of ENO1 contribute to the regulation of ENO1's RNA binding.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We thank S. Sahadevan for his expert support with the data analysis; M. B€ uscher for advice on the knockout and knockin experiments; and A. Chatterjee for the RIP protocol and advice on the related data. We also thank current and former members of the Hentze laboratory for their feedback; K.

DECLARATION OF INTERESTS
The authors declare the following competing interests: the European Molecular Biology Laboratory with the inventors M.W. Hentze and I. Huppertz applied for a patent (WO2022074066) on the use the ENO1-RNA interaction as a screening method for the identification of novel therapeutic compounds.

Materials availability
Plasmids and cell lines generated in this study are available upon request from the lead contact.
Date and code availability d The RNA sequencing and eCLIP data have been deposited at EMBL-EBI's ArrayExpress. The metabolomics data, raw imaging, and blotting data have been deposited at Mendeley Data. Accession numbers are listed in the key resources table and all data are publicly available as of the date of publication. d This paper does not report original code. d Any additional information required to reanalyse the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
HeLa cells (human female origin) were cultured in high-glucose (10 mM) Dulbecco's Modified Eagle Medium (DMEM), supplemented with 10% heat-inactivated fetal bovine serum (FBS), 2 mM L-glutamine and 100 U/ml Pen/Strep. For the transfection of HeLa cells with siRNAs, cells were grown until 70%-80% confluent, trypsinized, counted and 200,000 cells were seeded per well in a 6-well dish for reverse transfection of siRNAs (ENO1 pool, SIRT2 pool or control pool, Dharmacon) following the manufacturer's instructions (Lipofectamine RNAiMAX Transfection Reagent). For the transfection of plasmids, 5 mL of Lipofectamine 2000 reagent were used combined with 1 mg of plasmid DNA. In case of combined knock-down and rescue experiments, the cells were seeded, and siRNA transfection was performed. After 24 hours, the plasmid was transfected followed by an additional 24-hour incubation. The mouse embryonic stem cells (R1) were a donation from the group of Alexander Aulehla. Cell culture dishes were coated with 0.1% gelatine in PBS for 15 minutes prior to the addition of mESCs. The mESCs were cultured in high-glucose (10 mM) DMEM, supplemented with 15% FBS (EmbryoMax ES Cell Qualified), 2 mM L-glutamine, 100 U/ml Pen/Strep, 100 mM MEM non-essential amino acid solution, 1 mM sodium pyruvate and 0.1 mM b-mercaptoethanol. If the mESCs were maintained to remain pluripotent, the leukaemia inhibitory factor (10 3 units/ml) was added to the medium. The triple knock-in cell line was a donation from the group of Pierre Neveu (Sladitschek and Neveu, 2019) and was cultured under the same conditions. The cell lines were not authenticated.

METHOD DETAILS CRISPR/Cas9 gene insertion of ENO1 mutants
Single guide RNAs targeting the ROSA locus were predicted using the CRISPR online tool (Haeussler et al., 2016;Ran et al., 2013), ordered from Merck, annealed and cloned into the pSpCas9(BB)-2A-RFP (kindly provided by Kyung-Min Noh, EMBL) using the BbsI restriction sites (kindly cloned by David Kuster). The template plasmids for the knock-in of murine ENO1 mutants were prepared by introducing homology arms complementary to the ROSA locus upstream and downstream of the ENO1 wild-type or mutant DNA. While ENO1 is expressed under an EF1alpha promoter, the fluorescent protein cerulean is expressed under an additional, separate IRES promoter to facilitate an efficient cell sort. The generated plasmids were nucleofected into mESCs using the Nucleofector 4D system according to the manufacturer's guidelines (Nucleofector Kit P3, program CG-104, one million cells, 16 mg template vector and 4 mg Cas9 vector per nucleofection). Single cell sorting of double positive cells (RFP and cerulean) was performed 48 hours after nucleofection (FACSAria Fusion Sorter). Upon clonal expansion, successful insertion was tested by PCR to screen for homozygous insertions of the ENO1 wild-type or mutant DNA sequence in the ROSA locus (Phire Tissue Direct PCR Master Mix). The mutants of ENO1 were based on the findings of RBDmap data (Castello et al., 2016;Horos et al., 2019).

CRISPR/Cas9 gene deletion of ENO1
Using the ENO1 wildtype or mutant-expressing cell lines, we knocked out ENO1 in a subsequent step. Guide RNAs targeting the ENO1 locus were predicted using the CRISPR online tool (Haeussler et al., 2016;Ran et al., 2013), ordered from Merck, annealed and cloned into the pSpCas9(BB)-2A-GFP (PX458), or pSpCas9(BB)-2A-RFP using the BbsI restriction sites. The generated plasmids were nucleofected into mESCs using the Nucleofector 4D system according to the manufacturer's guidelines (Nucleofector Kit P3, program CG-104, 1 million cells and 5 mg of each of the guide RNA-containing plasmids). Single cell sorting of double positive cells (RFP and GFP) was performed 48 hours after nucleofection (FACSAria Fusion Sorter). Upon clonal expansion, successful insertion was tested by PCR to screen for homozygous deletion of ENO1 (Phire Tissue Direct PCR Master Mix). Furthermore, we tested the expression of ENO1 on a Western blot.
CRISPR/Cas9 gene insertion of the OsTIR1 receptor and the AID tag Single guide RNAs targeting the TIGRE locus for insertion of the OsTIR1 gene were used as described in Dossin et al. (2020). Furthermore, constructs for insertion of the AID tag with either a blasticidin or neomycin resistance gene were provided by Franç ois Dossin (Group of Edith Heard). The ENO1 homology arms were introduced into both constructs using restriction-free cloning. The generated plasmids were nucleofected into mESCs using the Nucleofector 4D system according to the manufacturer's guidelines in two rounds to achieve homozygous insertion (Nucleofector Kit P3, program CG-104, 1 million cells and 5 mg of each of the guide RNA-containing plasmids). Cells were selected with 6 mg/mL blasticidin and 200 mg/mL neomycin.

Differentiation and Fluorescence-activated cell sorting
To induce differentiation, LIF was removed from the medium for a period of seven days before harvesting (15 cm dishes). The cells were then trypsinized using 5 ml TrypLE Express at 37 C for five minutes and sorted using a FACSAria Fusion Sorter. Cell populations of BFP-positive and mCherry-positive cells were respectively plated on gelatine-coated 10 cm dishes using LIF-free medium and cultured for five additional days before performing further experiments.
Polynucleotide kinase assay Prior to harvesting cells, ENO1 antibody (ab112994 for HeLa cells or ab155102 for mouse embryonic stem cells, Abcam) was coupled with SureBeads Protein G magnetic beads for one hour at room temperature. As a control, rabbit IgG isotype control (Invitrogen, 10500C), normal rabbit IgG (Cell Signaling, 2729) or normal mouse IgG (Santa Cruz, sc-2025) was coupled to the beads at the same concentration. Typically, 1 mg was coupled with 30 ml of bead slurry. After coupling, the magnetic beads were washed twice with lysis buffer (100 mM NaCl; 50 mM Tris-HCl pH 7.5; 0.1% SDS; 1 mM MgCl 2 ; 0.1 mM CaCl 2 ; 1% NP-40; 0.5% sodium deoxycholate; protease inhibitors). Alternatively, pre-coupled FLAG M2 magnetic beads were washed twice with lysis buffer (25 mL per IP). HeLa cells or mESCs were washed twice with ice-cold PBS. Subsequently, the cells were UV-crosslinked at 150 mJ/cm 2 on ice and lysed in lysis buffer (Stratalinker 1800 UV Crosslinker). The lysates were sonicated (BioruptorPico) and treated with 0.1-8 U/ml RNase I and 2 U/ml Turbo DNase for five minutes at 37 C. The lysates were centrifuged full speed at 4 C for 15 minutes and 15 ml were saved as inputs for a Western blot. The remainder of the lysates was then utilized for IPs at 4 C for two hours. After the IP and three times three-minute washes with lysis buffer at room temperature, beads were washed additionally twice with PNK buffer (50 mM NaCl; 50 mM Tris-HCl, pH 7.5; 10 mM MgCl 2 ; 0.5% NP-40; protease inhibitors). The magnetic beads were resuspended in PNK buffer containing 0.1 mCi/ml [g-32 P] ATP, 1 U/ml T4 PNK, 1mM DTT and labelled for 15 minutes at 37 C. After four washes with PNK buffer, proteins were eluted at a low pH (0.1 M glycine, pH 2.0), neutralized with 1.5 M Tris-HCl, pH 8.5, and mixed with 43sample loading buffer. The samples were heated to 95 C for five minutes and quickly spun down. The samples were resolved by SDS-PAGE and blotted on a nitrocellulose membrane using the Trans-Blot Turbo Transfer System. The membrane was exposed overnight to a phosphorimaging screen and scanned with a Typoon FLA 9500, followed by immunoblotting. The files were processed with the Image Lab or ImageJ software. eCLIP eCLIP was performed as previously published (Van Nostrand et al., 2016), with minor adaptations. 1-3 mg of ENO1 antibody (ab112994, Abcam) or appropriate control IgG (rabbit) was coupled for 1 hour at RT to 30 ml of Protein G coupled magnetic beads. The cell lysates were treated with 0.1 U/ml RNase I for five minutes. One ml of lysate was used for each IP (concentration of 2 mg/ml) for two hours at 4 C. Complexes were eluted at low pH (0.1 M glycine, pH 2.0) and neutralized with 1.5 M Tris-HCl pH 8.5 before loading them on 4-12% XT Bis-Tris Protein Gels. The transfer onto the nitrocellulose membrane was performed using the Trans-Blot Turbo Transfer System. cDNA libraries composed of the IP (16 cycles) and the size-matched input controls (9 cycles) were multiplexed and sequenced using paired-end sequencing (PE125) on the Illumina HiSeq2000 platform.

Data processing for eCLIP and statistical analysis
The quality check of the eCLIP data was performed using fastqc (Andrews, 2010). The Unique Molecular Identifier (UMI) barcodes that are attached during library preparation were extracted and appended to the read name using umi-tools extract (Smith et al., 2017). The ligated adapters were trimmed and the reads shorter than 18 nucleotides were discarded using the cutadapt tool (Martin, 2011). The adapter-trimmed reads were aligned to the human genome (GRCh38.v23 from GENCODE) with STAR (Dobin et al., 2013), a splice-aware aligner, with the end-to-end alignment mode. The PCR duplicated reads were filtered from the mapped reads, based on the UMI barcodes appended to the read name, using umi-tools dedup with default parameters.
The GENCODE annotation of GRCh38.v23 genome was extended by including coordinates for tRNAs provided by tRNAscan (Lowe and Chan, 2016), and the resulting annotation was pre-processed with the htseq-clip suite (available at https://htseq-clip. readthedocs.io/en/latest/overview.html; accessed 21 May 2020) into overlapping windows of 50 nucleotides in size with a step size of 20 nucleotides. The truncation site (position -1 relative to the start site of a read, also called crosslink site) was extracted and quantified using htseq-clip. We used the R/Bioconductor DEWSeq package (Sahadevan et al., 2022) to detect significantly enriched windows in IP samples over the corresponding size-matched input control samples (log 2 fold change >0.5, p-adjusted <0.1). IHW (Ignatiadis et al., 2016) was used for multiple hypothesis correction. Overlapping significant windows were merged to binding regions.

ENO1 Motif analysis
The previously obtained significant binding regions of ENO1 were filtered to retain regions with a minimum average of 30 normalized crosslink counts in the IP samples of protein-coding genes. We retained 472 protein coding exon regions, which were used for motif analysis. DREME(v4.11.3) (Bailey, 2011) from the MEME suite package (Machanick and Bailey, 2011) was used to identify the de novo motifs with parameters maxk set to 20 and norc.

Proximity Ligation Assay
A probe was designed based on ENO1's identified and validated eCLIP binding site in the FTH1 mRNA. The probe is composed of a 20mer of DNA complementary to the RNA region proximal to ENO1's binding site. Furthermore, the probe has a 3'biotinylation tag [BtnTg] to be recognized with an anti-biotin antibody (rabbit: ab234284 and mouse: ab201341).
Cells were grown on cover slips and fixed with 100% ice-cold acetone at room temperature for 20 minutes. The fixed cells were washed with PBS and permeabilized with PBS containing 1% BSA and 0.1% Triton X-100. To prepare the cells for the hybridization with the probe, the cells were washed with 0.1M Triethanolamine containing acetic anhydride. After washing twice with PBS-T (0.02% Tween20), the cells were washed twice with hybridization buffer (1x Denhardt's solution, 0.1% (v/v) Tween20, 0.1% (w/v) CHAPS, 5 mM EDTA, 1 mg/ml RNase-free tRNA, 100 mg/ml heparin) and then incubated in hybridization buffer containing 100 nM of the relevant probe in a wet chamber at 37 C overnight. The probe was boiled for five minutes at 95 C prior to addition. Subsequently, Duolink in situ red starter kit rabbit (Sigma) was followed using the anti-ENO1 antibody (rabbit: 1:400; 11204-1-AP, Proteintech) or anti-FLAG antibody (mouse: 1:400; F3165, Sigma) for detection of the protein-RNA signal. To measure ENO1 expression by fluorescence imaging, the same primary antibody concentrations were used. For detection, the Alexa 488-conjugated secondary antibodies were used (Donkey anti-Rabbit IgG (H+L) Alexa Fluor 488; A-21206, Thermo Fisher Scientific; Donkey anti-Mouse IgG (H+L) Alexa Fluor 488; A-21202, Thermo Fisher Scientific).
After the last wash of the Duolink PLA manufacturer's protocol, antibody diluent was mixed with DAPI (final concentration of 0.1 mg/ ml) and the cover slips were incubated at room temperature for 45 minutes. The cover slips were washed once with PBS-T and a glass cover was mounted using ProLong Diamond Antifade Mountant. The cover slips were stored at 4 C until fluorescence microscopy.
The microscopy was performed using a LSM 780 Laser Scanning Microscope (Zeiss) equipped with an AxioCam camera and 63x/ 1.4 objective with immersion oil (Immersol). The microscope was operated using the ZEN 2012 software (Zeiss). The DAPI signal was recorded in one plane to act as a reference for counting the number of cells. The PLA (Alexa 594) signal was recorded as a Z-stack (10 pictures for 10 mm stack). Three images were taken per condition. The images were acquired as .lsm files using the same settings (gain, laser power, pinhole and offset) for the same protein and analysed using the Fiji/ImageJ software. The .lsm files were split into individual channels, the Z-stacks for the PLA signal was projected into a single plane and the brightness was set to the same level in all images of the same channel to enable the comparison of the results. The individual channels were ultimately saved as .tif files.
To count the PLA signals per cell for each of the different conditions, the CellProfiler Software version 4.1.3 was used. The range of the signal spot size was set to 8 to 20 pixels and the range of the nuclear size (DAPI signal) was set to 100 to 400 pixels. In both instances, the global threshold strategy minimum cross-entropy was used. The threshold smoothing factor was 1.3488 for the PLA signals and 20 for the nuclei. The cells were exposed to high laser power in the 488 channel to retrieve an outline of the cells. Clumped objects were separated by intensity and objects touching the border of the images were discarded. The PLA signal per cell were counted by combining the information of the cellular outline and the DAPI signal. The statistical analysis of these results was performed using Graphpad Prism version 9.

RNA immunoprecipitation
The medium was removed from 15 cm dishes of 80-90 % confluent HeLa or mESCs and PBS with a final concentration of 0.1% formaldehyde was added to the cells for 9 minutes (room temperature). The cells were washed twice with excess of PBS and the crosslinking reaction was quenched by adding 125 mM Glycine to the cells for 5 mins (room temperature). Afterwards, the cells were washed 3x with ice-cold PBS for 1-2 mins. The lysis was performed using RIPA lysis buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.5, 1% NP40, 0.1% SDS, 0.5% Na-Deoxycholate, Protease Inhibitor). The cells were sonicated with a Bioruptor(Pico) set to 'High' for 10 cycles with 30s ON and 30s OFF setting at 4 C. The lysates were centrifuged at 10,000g and 4 C for 15 mins and the protein concentration was measured. For the RNA immunoprecipitation experiment, 300 mg lysate and 3 mg Antibody (ab112994 for HeLa cells or ab155102 for mouse embryonic stem cells, Abcam). The antibody was coupled to 20 ml Protein G magnetic beads for 1 hour at room temperature. 2% of the lysate was used as input material and the lysate was incubated with the antibody-conjugated beads for 1 hour at 4 C while constantly rotating. The IPs were washed once with RIPA lysis buffer, twice with RIPA high salt buffer (500 mM NaCl, 50 mM Tris-HCl pH 7.5, 1% NP40, 0.1% SDS, 0.5% Na-Deoxycholate, Protease Inhibitor), twice with Wash buffer (10 mM Tris-HCl, 0.2% Tween-20, 1mM EDTA) and once again with RIPA lysis buffer. The samples were eluted with glycine as previously described. 10% of the IP was used for western blotting to assess the immunoprecipitation efficiency. RNase inhibitor and 10 ng total C.elegans RNA (spike-in control) were added to the remainder of the IP samples and the inputs. Subsequently, the samples were digested with pre-warmed (at 37 C for 20 mins) proteinase K mix at 55 C for 30 mins (final concentration: 2 mg/mL enzyme). Acid phenol/chloroform/isoamyl alcohol (pH 6.5) was added to the input and IP samples and they were incubated at 37 C for 5 mins while shaking at 1200 rpm. The samples were then loaded on 2 ml Phase Lock gel Heavy tubes, incubated at 37 C for 5 mins again and then centrifuged at 13,000g for 15 mins at room temperature. The aqueous layer was transferred to 5 ml Eppendorf tubes and the RNA Clean & Concentrator-5 kit was used to purify the RNA, following the manufacturer's instructions. The RNA eluates were treated with 2 mL Turbo DNase, 2.5 mL 10 x Turbo DNase Buffer in a final 25 mL reaction volume. The mixture was incubated at 37 C for 30 mins. Subsequently, the reaction was stopped by adding 2.4 mL inhibition resin. The reaction mix was incubated for 5 mins at room temperature and then centrifuged at 10,000g for 2 mins. 20 mL of the supernatant was retrieved and all of it was used as input for the SuperScript IV reverse transcription (1x SSIV, 0.5 mM dNTP, 5 mM DTT, 5U/ mL Super Script IV Reverse Transcriptase, 2.5 mM random hexamers), following the manufacturer's instructions. These samples were then used for subsequent qPCR experiments.
RT-PCR and qPCR RNA was isolated using the Direct-zol RNA Miniprep kit, as recommended by the manufacturer. 1.0 mg total RNA was used to synthesize cDNA using the Maxima First Strand cDNA Synthesis Kit for quantitative reverse transcription polymerase chain reaction (RT-qPCR). RT-qPCR was performed using the SYBR-Green qPCR Master mix on a QuantStudio 6 Flex Real-Time-PCR-System. Gene expression values were normalized using ACTB and are shown as a relative fold change to the value of control samples. All experiments were performed in biological triplicates and error bars indicate ± standard deviation as assayed by the DDCt method. Statistically significant differences were detected using two-way ANOVA with multiple comparisons correction (Tukey). All RT-qPCR primers are listed in Table S1.
RNAseq library preparation RNA was isolated using the Direct-zol RNA Miniprep kit, as recommended by the manufacturer. 1.0 mg total RNA was used as input for the library preparation. The TruSeq Stranded mRNA kit was used for the library preparation. The samples were multiplexed and sequenced on a Hiseq2000.

Sucrose density centrifugation and fractionation
For the preparation of lysates for the testing of RNA dependency through ultracentrifugation, previously published protocols were used as a basis (Caudron-Herger et al., 2019;Hö ck et al., 2007). HeLa cells were cultured on a 15 cm dish and lysed in 300 ml lysis buffer (25 mM Tris-HCl, pH 7.4, 150 mM KCl, 0.5% NP-40, 2 mM EDTA, 1 mM NaF, 0.5 mM DTT, protease inhibitor). A pre-clearing step was performed by centrifugation at 10,000g for 10 minutes at 4 C. The lysates were then treated with a combination of RNase I, RNase T1/A or no RNase was added at 37 C for 15 minutes. For the fractionation, gradients from 5% (w/v) to 25% (w/v) sucrose in 150 mM KCl, 25 mM Tris-HCl (pH 7.4) and 2 mM EDTA were prepared using a gradient maker (settings: 1:52, angle 81.5, speed: 15). Lysates were separated by centrifugation at 30,000 rpm and 4 C in an SW40 rotor for 18 hours. The lysate fractions were collected by hand through careful pipetting from the top (16 fractions were collected of approximately 600 ml). For the protein precipitation, 150 ml of 100% Trichloroacetic acid (TCA) was added and left on ice for 30 minutes. The individual fractions were centrifuged at full speed and 4 C for 20 minutes. The TCA supernatant was carefully removed, and the pellets were washed once with 1 ml cold acetone (stored at À20 C). The samples were vortexed, and an additional centrifugation step was performed at full speed and 4 C for 30 minutes. The supernatant was again carefully removed, and the pellet was air-dried. Finally, the pellets were taken up in 13 loading buffer containing benzonase and used for SDS-PAGE and immunoblotting.

Protein Purification
The ENO1 wildtype and mutant DNA sequence were cloned into the pETM-22 plasmid (generously provided by the Protein Expression and Purification core facility at EMBL) using the NcoI and XhoI site. Through the expression of the ENO1 variants, the proteins are tagged with the protein Thioredoxin to enable a clear distinction between the cleaved and uncleaved protein, which can be cleaved off using the HRV 3C Protease and removed using a reverse Ni-NTA column.
For the protein purification, the ENO1 variants were expressed by transforming the plasmid into E.coli BL21(DE3) CodonPlus-RIL competent cells and plated on Kanamycin-containing LB plates. One colony was picked and used for protein expression in a 400 ml flask. Once the OD600 was reached, the culture was cooled down to 18 C and IPTG was added to reach a final concentration of 1 mM. The bacteria were pelleted and lysed in lysis buffer (20 mM Tris-HCl pH 7.5, 500 mM NaCl, 5 mM b-2ME, 5% glycerol, 40 mM imidazole, 0.01% NP40), supplemented with RNase A, Turbo DNase and Protease Inhibitor to the lysis buffer. For efficient lysis, the lysates were processed with a microfluidizer. The His-tagged thioredoxin was used for the purification of ENO1 using a HisTrap HP column on an Ä kta go protein purification system. The protein was eluted with increasing concentrations of imidazole and the protein-containing fractions were verified on a Coomassie gel. The solubility tag was cleaved by HRV 3C protease during overnight dialysis using the POR tubing Spectra/Por 2. The following day, a reverse nickel column was performed using again a HisTrap HP column on an Ä kta go protein purification system. The protein was eluted with the HRV 3C protease and Thioredoxin being retained due to their remaining His tag.

Electromobility shift assay (EMSA)
The synthesized RNA molecules (Merck) used for testing ENO1's binding were radioactively end-labelled using the T4 PNK enzyme and [g-32 P] ATP. For all EMSA experiments, a 20 cm-long 5% acrylamide (40% acrylamide (19:1), TBE, glycerol, APS -freshly prepared, TEMED) native gel was poured for the separation of the free radioactive probe and the formed complex of ENO1 and RNA. The reaction was performed in a 20 ml reaction mix, containing 1 mM ENO1 protein and 10 nM RNA in reaction buffer (1 mg/ml of BSA, 10 mg/ml RNAsin, 5 mM DTT, 0.5 mM PMSF, 2.5 mM MgCl 2 , 100 mM KCl; 20 mM HEPES, pH 7.9; 0.2 mM EDTA and 20% glycerol). Reactions were incubated at room temperature for 10 minutes. After the reaction samples were loaded on the native gel, it was run overnight at 55V. The next day, the gel was dried for 1 hour at 80 C and exposed to a phosphorimaging screen for four hours or overnight. The images were quantified using the ImageLab software and the modelling of binding curves of target and control RNAs was done using GraphPad Prism 8.

ENO1 activity assay
For measuring the activity of ENO1, the ENO1 activity assay kit has been utilized according to the manufacturer's instructions. For measuring the impact of RNA on the activity of ENO1, synthesized RNAs (concentrations indicated in the respective figures) were added to the coupling step of the recombinant ENO1 to the ELISA plate. The plate was washed several times according to the manual and the reaction mix added.
For the activity measurements of the FLAG-tagged ENO1 with and without RNase treatment, endogenous ENO1 was knocked down in HeLa cells and the ENO1 mutant-containing plasmids were transfected as described above. FLAG IPs were performed using pre-coupled Flag M2 magnetic beads (25 ml for 0.5 mg protein lysate) under constant rotation at 4 C for 1 hour. The protein complexes were eluted using 20 ml FLAG peptide (1 mg/ml) by incubation on ice for 1 hour. Half of the eluate was incubated with 1ml RNase A (final concentration 1 mg/ml) or water at 37 C for 15 mins. 2.5 ml were used as input for the enzymatic activity assay, omitting the coupling to the ELISA plate, and following the manufacturer's instructions.

ENO1 isotope labelling and protein purification
To produce 15 N-labelled, deuterated ENO1, BL21(DE3) CodonPlus-RIL cells expressing pETM-22-TRX-1-His6-tagged human ENO1 were pre-cultured in M9 minimal media supplemented with 15 NH 4 Cl overnight at 37 C. The pre-culture was diluted in pre-warmed D 2 O M9 media to a final OD600 of 0.1. This second pre-culture was grown at 37 C to an OD600 of 0.8 and added to the final D 2 O M9 minimal media volume in a 1/10 dilution. The final culture was then grown to an OD600 of 0.9 at 37 C and human ENO1 expression was induced by adding 0.2 mM IPTG and grown overnight at 18 C.
The cells were harvested and resuspended in lysis buffer (20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 40 mM imidazole, 5% glycerol, 0.01% NP-40, 5 mM b-mercaptoethanol, protease inhibitor cocktail), followed by lysis with a microfluidizer (Microfluidics). The cleared lysate was applied to a 5 ml His trap FF column, washed, and eluted with an imidazole buffer gradient (20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 600 mM imidazole, 5% glycerol). The TRX-His6 tag was removed by addition of HRV 3C protease. After dilution of the protein to lower the imidazole concentration (20 mM Tris-HCl, pH 7.5, 500 mM NaCl, 40 mM imidazole) the sample was applied to a second His trap column to remove the cleaved tag and the protease. The flow-through fraction was collected, concentrated, and loaded onto a Superdex 200 16/600 size exclusion column equilibrated with NMR buffer (20 mM NaPi, pH 6, 150 mM NaCl, 20 mM MgCl 2 , 0.5 mM TCEP).
NMR data acquisition and titration 1 H, 15 N-HSQC spectra of deuterated ENO1 were recorded using an Avance III Bruker NMR spectrometer, which operates at a field strength, corresponding to a proton Larmor frequency of 800 MHz, equipped with a cryogenic triple resonance gradient probe head. Measurements were performed at 298 K, using 0.2 mM ENO1 in NMR buffer, supplemented with 5% D 2 O for the deuterium lock. This sample was titrated with increasing concentrations of 18-mer synthetic RNA (IBA) (0.02, 0.04, 0.05, 0.16, 0.24 and 0.4 mM) and at each titration step another 1 H , 15 N-HSQC was recorded. Data were processed and analysed using NMRPipe and Sparky.

Nucleofection of target and control RNAs
The synthesized specific and unspecific RNAs were nucleofected into HeLa cells or mESCs using the Nucleofector 4D system according to the manufacturer's guidelines (HeLa cells: SE Cell Line 4D-Nucleofector X Kit, programme CN-114, 1 million cells; mESCs: P3 Primary Cell 4D-Nucleofector X Kit, programme CG-104, 1 million cells). The cells were then cultured for an additional 90 minutes in DMEM, free of phenol red and FBS, to allow the collection of lactate-containing medium.

13
C-glucose tracing experiment and GC-mass spectrometry Mouse embryonic stem cells were exposed to specific or unspecific RNAs through nucleofection. Afterwards, the cells were transferred to 5 ml Eppendorf tubes containing 4 ml warm DMEM free of phenol red and FBS containing D-Glucose (U-13 C6). The cells were exposed to the labelling medium for 60 minutes and then they were collected by centrifugation at 500g (4 C) for 3 minutes.
The cells were re-suspended in ice-cold PBS and the centrifugation step was repeated. The PBS was removed, and the cell pellet was re-suspended in 5 ml of fresh ice-cold PBS. The cells were transferred into a 2 ml Eppendorf tube, and centrifuge again at 500g for 3 minutes. Finally, the PBS was removed and 500 ml of very cold HPLC-grade methanol was added, and the suspension was mixed well. The samples were mixed and kept on dry ice until they were stored at -80 C. On the day of the extraction, all tubes were placed on ice, vortexed and 5 ml of adonitol (50 mg/ml) was added. The samples were incubated at 72 C for 15 minutes in a water bath. 500 ml MilliQ H 2 O was added to the extract. The samples were vortexed and placed at -80 C for one hour. After the incubation, the samples were thawed on ice, thoroughly vortexed and centrifuged at 15,000g at 4 C for 12 minutes. The supernatants were transferred to high recovery vials and vacuum was applied for drying the samples (Genevac EZ-2 Plus evaporator; program, hplc fraction; temperature, 30 C). The dried polar metabolites were derivatized with 50 mL of 20 mg/mL methoxyamine hydrochloride solution in pyridine for 90 minutes at 37 C, followed by reaction with 100 mL N-methyl-N-(trimethylsilyl)trifluoroacetamide for 10 hours at room temperature, as justified in Kanani and Klapa (2007). GC-MS analysis was performed using a Shimadzu TQ8040 GC-(triple quadrupole) MS system (Shimadzu Corp.) equipped with a 30 m x 0.25 mm x 0.25 mm ZB-50 capillary column (Phenomenex, 7HG-G004-11). One mL of sample was injected in split mode (split ratio 1:5) at 250 C using helium as a carrier gas with a flow rate of 1 mL/ minute. GC oven temperature was held at 100 C for 4 minutes followed by an increase to 320 C with a rate of 10 C/minute, and a final constant temperature period at 320 C for 11 minutes. The interface and the ion source were held at 280 C and 230 C, respectively. The detector was operated both in scanning mode recording in the range of 50-600 m/z, as well as in MRM mode for specified metabolites. For peak annotation, the GCMSsolution software was utilized. The metabolite identification was based on an in-house database with analytical standards utilized to define the retention time, the mass spectrum, and the quantifying ion fragment for each specified metabolite. The ratio of the different mass isotopologues for each metabolite was determined by integrating the area under the curve (AUC) of the quantifying ion fragments followed by correction for the presence of natural abundant isotopes with the Isotope Correction Toolbox (ICT) (Jungreuthmayer et al., 2016). All peak integrations were manually checked.

Lactate accumulation measurement
For the determination of lactate levels in the supernatant of HeLa and mESCs, the supernatant was collected 30, 60 and 90 minutes after the replacement with DMEM, free of phenol red and FBS. The lactate colorimetric/fluorometric assay kit for the determination of lactate levels was utilized according to the manufacturer's instructions.
Lactate accumulation and oxygen consumption of pluripotent and differentiated cells mESC were differentiated in 15 cm dishes for seven days in the absence of LIF or maintained for two days in the pluripotent state. Seeding of cells was done asynchronously so that pluripotent and differentiated cells could be processed in parallel on the same day. For determining oxygen consumption, cells were washed with PBS, trypsinized, pelleted by centrifugation (5 minutes at 230g) and resuspended in mESC medium (+ or À LIF as appropriate). Live cell number was counted in a TC20 automated cell counter using trypan blue staining. Immediately afterwards, the oxygen consumption rate of 300 ml of cell suspension containing $one million living cells was determined in an Oxytherm System at 37 C using the Oxygraph Plus data acquisition software. For determining lactate secretion, the medium of the 15 cm dishes with pluripotent/differentiated cells was exchanged with 11 ml of mESC medium without serum and without phenol red. After 30 min of incubation, an aliquot of the conditioned medium was taken and stored at À80 C. Cells were lysed in RIPA buffer and total protein quantified using the Protein Assay Dye Reagent Concentrate. Lactate concentration was measured in 3 ml of conditioned medium with the Lactate Assay kit following manufacturer's instructions. Lactate measurements were normalized by the protein amount on the dish.

QUANTIFICATION AND STATISTICAL ANALYSIS
Statistical analysis of results was performed using two-tailed Student's t-test (parametric), or analysis of variance (ANOVA), followed by Sidak's or Tukey's multiple comparison tests, as stated in the figure legends. All the analyses were done using GraphPad Prism, version 8.4.2. Statistical significance is represented in all figures, as follows: p-value of <0.0001: ****, p-value of 0.0001 to 0.001: ***, p-value of 0.001 to 0.01: **, p-value of 0.01 to 0.05: * and p-value of R0.05: not significant.