Glutamine independence is a selectable feature of pluripotent stem cells

Glutamine anaplerosis is reduced in highly self-renewing ESCs

In proliferating mammalian cells in vitro, glutamine is the major source of carbon for tricarboxylic acid (TCA) cycle intermediates¹. Consequently, most cell lines including ESCs depend on exogenous glutamine for growth and proliferation^1,8,10. One notable exception is 2i-cultured mouse ESCs in the ground state of pluripotency, which can sustain proliferation in the absence of exogenous glutamine⁸. Previously, we found that addition of 2i reduces the contribution of glutamine-derived carbons to TCA cycle intermediates while increasing the contribution of glucose-derived carbons, thereby reducing reliance on exogenous glutamine to support TCA cycle anaplerosis⁸. To determine whether the effects of 2i on mouse ESC metabolism are a specific consequence of MEK/GSK3β inhibition or a general feature of the metabolic requirements of self-renewing pluripotent cells, we utilized alternative methods to drive ESCs into the ground state of pluripotency. First, we took advantage of a chimeric LIF receptor engineered to respond to granulocyte colony-stimulating factor (GCSF) and harboring a mutation at tyrosine 118 to impair negative feedback by Socs3²¹. Upon treatment with GCSF, cells expressing this chimeric receptor exhibit elevated and sustained JAK/STAT3 signaling (Supplementary Fig. 1a) and are stabilized in the naïve state of pluripotency regardless of the presence of differentiation-inducing stimuli²².

To assess glucose and glutamine utilization in these cells, we used gas chromatography-mass spectrometry to trace the fate of uniformly ¹³C-labeled glucose or glutamine. In addition to enhancing expression of the Stat3 target gene Tfcp2l1 and key transcription factors associated with naïve pluripotency²³ (Supplementary Fig. 1b), GCSF treatment induced metabolic alterations similar to those triggered by 2i. Specifically, GCSF increased the fraction of TCA cycle metabolites derived from oxidative decarboxylation of glucose-derived pyruvate (m+2 labeled isotopologues) and decreased the fraction of metabolites derived from glutamine catabolism (Fig. 1a-d, Supplementary Table 1). Whereas more than 80% of glutamate is derived from glutamine in control ESCs, in line with other cultured mammalian cells, GCSF reduced the fraction of glutamate derived from glutamine to 60% suggesting that GCSF-cultured ESCs can generate glutamate from sources other than glutamine (Fig. 1d, Supplementary Table 1). Indeed, when deprived of exogenous glutamine, GCSF-cultured ESCs were able to use glucose and other anaplerotic substrates to maintain intracellular pools of glutamate (Supplementary Fig. 1c, Supplementary Table 1).

Both MEK/GSK3β inhibition and JAK/STAT3 activation promote naïve pluripotency by altering signal transduction. To determine whether direct activation of pluripotency gene networks is sufficient to rewire intracellular metabolic pathways, we ectopically expressed pluripotency-associated transcription factors Klf4 and Nanog (Supplementary Fig. 1d). Expression of either transcription factor is sufficient to enhance self-renewal^24-26 and induce expression of target genes associated with naïve pluripotency^27,28 (Supplementary Fig. 1e). Similar to MEK/GSK3β inhibition and JAK/STAT3 activation, overexpression of Klf4 or Nanog increased the fraction of TCA cycle intermediates generated from glucose-derived carbons (Fig. 1e, Supplementary Table 1) while decreasing the fraction of TCA cycle intermediates derived from glutamine (Fig. 1f, Supplementary Table 1). Under conditions of reduced glutamine anaplerosis, pyruvate carboxylase can partially compensate to maintain TCA cycle anaplerosis²⁹. Accordingly, both JAK/STAT3-stimulated and Nanog/Klf4-expressing cells exhibited an increase in glucose-derived anaplerosis through pyruvate carboxylase (m+3 isotopologues), further underscoring the reduced reliance on glutamine anaplerosis in cells with enhanced pluripotency networks (Supplementary Fig. 1f-h, Supplementary Table 1).

Together, these results suggest that interventions that enhance ESC self-renewal alter the balance of glucose and glutamine utilization in cells independently of changes in culture conditions or proliferation rates (Supplementary Fig. 1i,j). We have previously represented this metabolic shift—marked by reduced glutamine anaplerosis coupled with enhanced contribution of glucose-derived carbons to TCA cycle metabolites—as a change in the ratio of α-ketoglutarate to succinate⁸. Both JAK/STAT3 induction and Nanog overexpression increase cellular αKG/succinate ratios, while Klf4, which has a more modest effect on glucose and glutamine utilization under routine culture conditions, had no notable effect on this ratio (Fig. 1g,h, Supplementary Table 1).

We next asked whether the αKG/succinate ratio varies in correlation with the inherent self-renewal potential of ESCs. The significant population heterogeneity of ESCs cultured under conventional serum/LIF (S/L) conditions provided an opportunity to determine whether the functional heterogeneity of ESCs is accompanied by metabolic heterogeneity. To this end, we utilized ESCs harboring a GFP reporter at the endogenous Nanog locus³⁰ to sort out cells with the lowest and highest expression of Nanog (Fig. 1i, left). These Nanog-GFP cells have previously been used to illustrate that “Nanog Low” cells are more prone to differentiate than their “Nanog High” counterparts^30,31. Because of the inherent metastability of S/L-cultured ESCs, within three days of sorting both “Nanog Low” and “Nanog High” sorted populations had begun to re-establish the variable Nanog expression that characterizes S/L-cultured ESCs (Fig. 1i, right). Nevertheless, the ratio of αKG/succinate was significantly different in the two populations, with Nanog High cells characterized by an elevated αKG/succinate ratio consistent with their enhanced capacity for self-renewal (Fig. 1j, Supplementary Table 1). Further supporting a tight correlation between the αKG/succinate ratio and ESC self-renewal, the αKG/succinate ratio increased progressively from Nanog-low to Nanog-intermediate and Nanog-high populations (Supplementary Fig. 1k, Supplementary Table 1).

ESCs with enhanced self-renewal exhibit reduced dependence on exogenous glutamine

We next probed the functional outcome of reduced glutamine anaplerosis in order to further test whether altered proliferative metabolism is an inherent feature of cells with enhanced capacity for self-renewal. Glutamine is required to maintain proliferation, viability and self-renewal of ESCs in traditional S/L culture, and restoring anaplerosis with cell-permeable αKG is sufficient to compensate for glutamine withdrawal (Fig. 2a and Supplementary Fig. 2a,b). Therefore, we reasoned that decreased reliance on glutamine anaplerosis would enable cells to better tolerate withdrawal of exogenous glutamine. Indeed, Nanog-high, Nanog-intermediate, and Nanog-low cells were progressively more sensitive to glutamine withdrawal (Supplementary Fig. 2c). Even after several days in culture, “Nanog High” cells were significantly more resistant to apoptosis triggered by glutamine deprivation than their “Nanog Low” counterparts (Fig. 2b and Supplementary Fig. 2d). As a control, we included cells cultured in the presence of 2i, which is sufficient to enable glutamine-independent proliferation⁸ and blocked apoptosis induced by glutamine withdrawal (Fig. 2b and Supplementary Fig. 2d). Similarly, while only 40% of control ESCs remained viable after 48 h of glutamine withdrawal, over 60% remained viable in ESCs with constitutive JAK/STAT3 activation (Fig. 2c). Furthermore, cells with enhanced self-renewal mediated by either JAK/STAT3 activation or Klf4/Nanog overexpression were able to increase the number of viable cells in the culture despite the absence of exogenous glutamine (Fig. 2d,e). Conversely, pharmacologic inhibition of JAK/STAT3 signaling sensitized ESCs to glutamine deprivation (Fig. 2f).

In order to determine which metabolic substrates are limiting for survival under conditions of glutamine deprivation, we cultured cells with cell-permeable analogs of pyruvate, αKG and succinate. Only αKG, the direct substrate for de novo glutamine biosynthesis, was capable of rescuing survival and proliferation in the absence of glutamine, and this rescue was contingent upon the ability of cells to use αKG to engage in de novo glutamine biosynthesis (Fig. 2g,h, Supplementary Fig. 2e,f). Therefore, we asked whether cells with enhanced self-renewal are better able to maintain αKG pools during conditions of glutamine withdrawal. In complete medium, even in cells with enhanced self-renewal, glutamine provides the dominant source of carbon for TCA cycle anaplerosis (Fig. 1d,f). Glutamine withdrawal profoundly reduced steady-state levels of TCA cycle metabolites in all ESC lines tested (Supplementary Fig. 2g,h, Supplementary Table 1). However, both JAK/STAT3-activated and Klf4/Nanog-overexpressing cells were better than their control counterparts at sustaining intracellular pools of αKG, but not downstream TCA cycle metabolites (Supplementary Fig. 2g-j, Supplementary Table 1). As a result, JAK/STAT3 and Klf4/Nanog-overexpressing cells were better able to maintain an elevated αKG/succinate ratio relative to control ESCs in the absence of exogenous glutamine (Fig. 2i,j, Supplementary Table 1).

In addition to serving as an obligate substrate for de novo glutamine biosynthesis, intracellular αKG promotes Nanog expression and increases self-renewal^8,12,14. Accordingly, while control cells rapidly lost Nanog expression upon glutamine deprivation such that both the overall Nanog expression and the proportion of cells in the Nanog-high population decreased, cells with JAK/STAT3 activation were able to retain Nanog expression and the fraction of cells in the Nanog-high population despite absence of exogenous glutamine (Fig. 2k, Supplementary Fig. 2k). Together, these results indicate that stem cells with strengthened pluripotency networks are not only better able to survive glutamine deprivation but also able to withstand the destabilizing effect of loss of glutamine on markers of self-renewal.

The relative glutamine independence of cells with heightened self-renewal led us to speculate that we could exploit this metabolic property to select for cells with the highest potential for self-renewal. To test this idea, we performed competition assays in which GFP-tracked cells expressing empty vector, Klf4 or Nanog were mixed with parental ESCs and the proportion of GFP+ cells was assessed following 48 h culture in glutamine-replete or glutamine-free medium. Expression of Nanog or Klf4 resulted in a notable selective advantage in the absence of exogenous glutamine, such that the proportion of GFP+ cells increased by 41% (Nanog) or 69% (Klf4) relative to cells cultured in the continuous presence of glutamine (Fig. 2l).

We next asked whether glutamine depletion could select for cells with endogenously strengthened self-renewal potential from within the heterogeneous population characteristic of ESCs. In order to assess population heterogeneity, we developed a quantitative immunofluorescence (IF)-based assay that allowed us to measure the expression levels of pluripotency-associated transcription factors in individual cells. Consistent with previous reports, IF analyses demonstrated that S/L-cultured ESCs exhibit highly variable Nanog expression and relatively homogenous, unimodal Oct4 expression (Supplementary Fig. 3a)^15,32,33. Addition of 2i increased both homogeneity and overall level of Nanog expression while having minimal effects on mean Oct4 expression (Supplementary Fig. 3a). Of note, 2i eliminated a long tail of “Oct4-low” cells representing approximately 10% of S/L-cultured ESCs that have previously been reported to represent differentiated cells readily apparent in traditional S/L ESC cultures³² (Supplementary Fig. 3a). As Oct4 is required for pluripotency³⁴, Oct4-low cells represent the most committed cells within a heterogenous population. Therefore, we asked whether these cells were the most susceptible to glutamine withdrawal. Indeed, within 8 hours of glutamine withdrawal, the proportion of cells in the Oct4-low fraction declined by 50% and 24 h of glutamine deprivation almost entirely eliminated the population of Oct4-low cells (Fig. 3a). Similarly, 24 h of glutamine deprivation eliminated the majority of Nanog-low cells; however, consistent with the observation that Nanog expression is sensitive to glutamine availability (Fig. 2k), glutamine deprivation also reduced the fraction of cells exhibiting the highest Nanog expression (Supplementary Fig. 3b,c).

Transient withdrawal of exogenous glutamine selects for ESCs with enhanced self-renewal

While Nanog expression is metastable and Nanog-low cells can remain undifferentiated and regenerate Nanog-high cells, cells with very low Oct4 represent differentiated cells that cannot self-renew³². As these Oct4-low cells were sensitive to glutamine deprivation, we hypothesized that transient glutamine deprivation would eliminate the most committed cells and thereby improve the overall self-renewal potential of a population. This simple procedure entailed subjecting regularly cultured ESCs to glutamine free medium for 24 h (“pulse”) and then recovering the cells in complete medium before seeding for follow-up experiments (Fig. 3b). Immunofluorescence assays confirmed that ESCs subjected to a 24 h period of glutamine withdrawal had higher Nanog and Oct4 on a per-cell basis, reflecting elimination of the most committed cells and the restoration of Nanog levels following glutamine re-addition (Fig. 3c). Similarly, Nanog-GFP reporter cells demonstrated that “pulsed” ESCs were more likely to fall into the Nanog-high population than their conventionally cultured control counterparts (Fig. 3d, Supplementary Fig. 3c). In line with their enhanced expression of pluripotency-associated transcription factors, pulsed cells also exhibited enhanced self-renewal. Colony formation assays analyzed more than one week after the initial pulse demonstrated that pulsed cells were more likely to give rise to undifferentiated colonies and less likely to give rise to differentiated colonies (Fig. 3e,f). Importantly, ESCs subjected to transient glutamine withdrawal remained competent for multi-linage differentiation, giving rise to all three germ layers in vivo during teratoma formation (Supplementary Fig. 3d).

We next performed a series of experiments to clarify how transient glutamine withdrawal improves the self-renewal potential of a population of ESCs. We first compared pulsed glutamine withdrawal, which eliminates the most committed cells, with interventions that increase ESC self-renewal. In contrast to pulsed glutamine deprivation, pulsed treatment with 2i or αKG—interventions that transiently increase Nanog-GFP expression (Supplementary Fig. 3e,f)—had no durable effect on the self-renewal capacity of a population of ESC cells (Fig. 3g, Supplementary Fig. 3g). These findings suggest that interventions that increase expression of pluripotency factors may exert different population effects compared to interventions that select against the most committed cells from within a population. To confirm that selective elimination of cells that cannot survive glutamine deprivation is required for the benefit of the pulsed glutamine withdrawal, we supplemented cells with cell-permeable αKG at the time of glutamine withdrawal. Blocking cell death during glutamine deprivation (Fig. 2g) eliminated the selective advantage of glutamine deprivation and abrogated the benefit of the pulse (Fig. 3h). Further supporting the notion that the glutamine withdrawal pulse selects for cells that have the endogenous metabolic capacity to sustain de novo glutamine synthesis, inhibition of glutamine synthetase concurrent with glutamine withdrawal eliminated surviving cells and completely blocked the ability of transient glutamine withdrawal to improve the population self-renewal capacity (Fig. 3h, Supplementary Fig. 3h).

Enhanced self-renewal is associated with the ability to sustain intracellular αKG in the absence of exogenous glutamine (Supplementary Fig. 2i,j). In addition to enabling glutamine synthesis, intracellular αKG can also promote the activity of αKG-dependent chromatin modifying enzymes. The ability of surviving ESCs to preserve intracellular αKG pools to maintain αKG-dependent demethylation reactions may also contribute to the beneficial effect of transient glutamine withdrawal, as addition of the histone H3 trimethylated lysine 27 (H3K27me3) demethylase inhibitor GSK-J4 impaired ESC self-renewal both when administered during transient glutamine withdrawal and in the presence of exogenous glutamine (Supplementary Fig. 3h,i). In contrast to glutamine, glucose is required for the viability of ESCs regardless of self-renewal capacity⁸. Accordingly, pulsed glucose withdrawal strongly reduced the colony-formation capacity of mouse ESCs (Fig. 3i, Supplementary Fig. 3j). Moreover, in line with previous reports that combined glucose and glutamine withdrawal eliminates undifferentiated stem cells¹⁰, pulsed withdrawal of both glucose and glutamine decimated the ability of ESCS to form colonies (Fig. 3i). Together, these findings support a model wherein glutamine deprivation selectively eliminates the most committed cells within a population thereby durably enhancing the overall self-renewal capacity of the remaining cells.

Transient glutamine withdrawal enhances mouse somatic cell reprogramming to pluripotency

Reprogramming of somatic cells to pluripotency represents a major area in which stem cell heterogeneity poses a significant experimental hurdle. Reprogramming is an inefficient process hampered by low efficacy and the persistence of incompletely reprogrammed cells³⁵. Interventions that consolidate the pluripotency network enhance reprogramming efficiency: for example, adding 2i to partially reprogrammed cells efficiently promotes the formation of fully reprogrammed cells³⁶. Therefore, we tested whether glutamine withdrawal, which selects for cells with strengthened pluripotency gene networks, improves reprogramming efficiency. First, we utilized mouse embryonic fibroblasts (MEFs) harboring a polycistronic cassette enabling doxycycline (dox)-inducible expression of Oct4, Klf4, Sox2 and c-Myc (OKSM) (Fig. 4a)³⁷. After eight days of dox-induced OKSM expression, dox removal forces cells to rely on reactivated endogenous pluripotency networks in order to sustain proliferation and ESC-like features including reactivity to alkaline phosphatase (AP). Consequently, cells that were never exposed to dox are fully AP-negative while control cells exposed to dox exhibit heterogeneous AP staining with numerous variably stained regions punctuated with discrete, well-stained colonies reminiscent of undifferentiated ESC colonies (Supplementary Fig. 4a). As expected, sustained 2i treatment over the last four days of reprogramming, which helps cells induce and consolidate pluripotency transcriptional networks³⁸ and blocks spontaneous differentiation¹⁷, resulted in a marked increase in the number of discrete, ESC-like colonies (Fig. 4b,c). Strikingly, just 24 h of glutamine deprivation two days after dox withdrawal was likewise sufficient to increase the overall number of ESC-like AP+ colonies (Fig. 4b,c). Conversely, there was an overall reduction in flat, intermediately stained regions, consistent with selective elimination of cells with weak activation of pluripotency-associated gene networks (Fig. 4b). Notably, transient glutamine withdrawal was just as effective at enhancing reprogramming efficiency as transient treatment with 2i (Fig. 4b,c, Supplementary Table 2).

To determine whether glutamine deprivation indeed increased the proportion of cells with activated endogenous pluripotency gene networks, we utilized a second reprogramming system. Here, we infected MEFs harboring a GFP reporter knocked into the endogenous Oct4 locus³⁹ with viruses carrying dox-inducible OKSM. The Oct4-GFP reporter is helpful in distinguishing fully reprogrammed iPSCs from partially reprogrammed “pre-iPSCs” which, despite having ESC-like morphology, do not activate endogenous pluripotency genes³⁶ and thus cannot ultimately maintain stable Oct4-GFP expression. Once again, we subjected cells to sustained 2i (7 days) or a 24 h pulse of either glutamine deprivation (“Pulse −Q”) or 2i treatment (“Pulse 2i”) beginning 2 days after dox withdrawal (Fig. 4d). Consistent with the observation that glutamine withdrawal eliminates the most committed, Oct4-low cells, the number of Oct4-GFP+ cells decreased transiently during glutamine withdrawal but rebounded within 24 h of recovery in glutamine-replete medium (Supplementary Fig. 4b). By the end of the experiment, all interventions significantly increased the proportion of cells expressing Oct4-GFP (Fig. 4e) and increased generation of tight, strongly AP+ ESC-like colonies (Supplementary Fig. 4c). Once again, 24 h of glutamine withdrawal was as effective as 24 h of 2i treatment in enhancing generation of Oct4+iPSCs (Fig. 4e, Supplementary Fig. 4c). Consistent with being fully reprogrammed iPSCs, cells isolated from colonies with compact, ESC-like morphology sustained Oct4-GFP expression for multiple passages in vitro and were able to generate colonies from single cells (Supplemental Fig. 4d,e) while remaining sensitive to LIF withdrawal (Supplemental Fig. 4d,f). These data suggest that transient glutamine deprivation is sufficient to increase the fraction of somatic cells that undergo successful reprogramming and further reveal that a metabolic intervention is as effective as well-described alterations in signal transduction pathways at improving reprogramming efficiency.

Transient glutamine withdrawal increases markers of pluripotency in human ESCs

Finally, we asked whether glutamine withdrawal exerted similar effects in human pluripotent stem cells despite the fact that human ESCs are cultured with dramatically different growth factors and represent a more committed, post-implantation stage of development⁴⁰. As with mouse ESCs, pulsed glutamine withdrawal eliminated a sub-population of cells with low expression of OCT4 (Fig. 4f). Moreover, pulsed glutamine withdrawal resulted in overall enhanced expression of key pluripotency factors SOX2 and OCT4 (Fig. 4f,g). Thus, transient glutamine deprivation represents a general method to enhance expression of key pluripotency markers in both mouse and human pluripotent stem cells under a variety of culture conditions.

Cell Culture.

Mouse ESC lines (ESC-1, ESC-2) were previously generated from C57BL/6 × 129S4/SvJae F1 male embryos⁸. Nanog-GFP reporter ESCs were a gift from R. Jaenisch (MIT). Nanog-GFP lines expressing the chimeric LIF receptor (a gift from A. Smith)²¹ and ESC-1 lines overexpressing Nanog or Klf4 were previously generated⁵⁶. ESC-1 cells were used for all experiments unless otherwise noted. ESCs were maintained on gelatin-coated plates in serum/LIF (S/L) medium containing Knockout DMEM (Life 10829–018) supplemented with 10% FBS (Gemini), 0.1 mM 2-mercaptoethanol, 2 mM L-glutamine and 1000 U/mL LIF (Gemini). For culture in 2i (S/L+2i), S/L medium was supplemented with 3 µM CHIR99021 (Stemgent) and 1 µM PD0325901 (Stemgent). Cells were adapted to 2i or GCSF (Gemini) by passaging cells in S/L+2i or S/L+GCSF medium at least three times prior to use in experiments. S/L+2i-adapted cells were maintained for a maximum of nine passages. All cells were routinely tested for mycoplasma. For human embryonic stem cells (hESC) culture, an H1 hESC line (NIHhESC-10–0043) with an inducible Cas9 insertion was used⁵⁷. This line was maintained in chemically defined, serum-free E8 conditions (Thermo Fisher Scientific, A1517001) on tissue culture treated polystyrene plates coated with vitronectin (Thermo Fisher Scientific, A14700). hESCs were split with 0.5mM EDTA at a 1:10–1:20 split ratio every 3–5 days. Cells have been confirmed to be mycoplasma-free by the MSKCC Antibody and Bioresource Core Facility. All experiments were approved by the Tri-SCI Embryonic Stem Cell Research Oversight Committee (ESCRO). Further details on cell lines can be found in the Reporting Summary.

Nutrient deprivation experiments.

For glutamine deprivation experiments in mouse ESCs, cells were initially plated in standard S/L medium as described above. The following day, cells were washed with PBS and then cultured in experimental medium containing a 1:1 mix of glutamine-free DMEM (Gibco 11960–051) and glutamine-free Neurobasal medium (Gibco 21103–049) including 10% dialyzed FBS, 2-mercaptoethanol, and LIF as described above and containing (“+Q”) or lacking (“−Q”) L-glutamine (2 mM) as indicated. When indicated, dimethyl-α-ketoglutarate (Sigma 349631) dissolved in DMSO to 1 M was added to a final concentration of 4 mM. For glucose and glutamine deprivation experiments, cells were cultured in medium containing a 1:1 mix of glutamine and glucose-free DMEM (Gibco A14430–01) and glutamine and glucose-free Neurobasal-A medium (Gibco A24775–01) including 10% dialyzed FBS and all supplements as described above and containing or lacking glucose or glutamine as indicated.

Embryonic stem cell competition assays.

GFP-negative parental ESCs were mixed with GFP-positive vector or Klf4/Nanog-overexpressing transgenic ESCs and seeded at a concentration of 30,000 total cells per well of a 12-well plate in triplicate. The following day, cells were washed with PBS and then changed to experimental medium containing a 1:1 mix of glutamine-free DMEM and glutamine-free Neurobasal medium including 10% dialyzed FBS, 2-mercaptoethanol, LIF, and containing (“+Q”) or lacking (“−Q”) L-glutamine as indicated. After 48 h, cells were trypsinized for flow cytometry analysis. Cells were evaluated for GFP and DAPI on either a LSRFortessa or LSR-II machine (Becton Dickinson). Analysis of DAPI exclusion and GFP mean fluorescence intensity was performed using FlowJo v9.0.

Glutamine pulse experiments.

For transient glutamine withdrawal (“pulse”) experiments, cells were initially plated in standard S/L medium as described above. The following day, cells were washed with PBS and then changed to experimental medium containing a 1:1 mix of glutamine-free DMEM and glutamine-free Neurobasal medium including 10% dialyzed FBS, 2-mercaptoethanol, LIF, and containing (“Ctrl”) or lacking (“Pulse −Q”) L-glutamine as indicated. 24 h later, cells were washed with PBS and then returned to glutamine-replete medium (“Recover”). 24 h later, cells were subjected to either image analysis, flow cytometry, or plated for colony formation assays as indicated. For glutamine pulse experiments in human ESCs, cells were initially plated at a density of 300,000 cells/well in a tissue-culture treated polystyrene 12-well plate coated with vitronectin (Thermo Fisher Scientific, A14700) in E8 medium (Thermo Fisher Scientific, A1517001) containing 10 μM ROCK inhibitor Y-27632 (Selleck Chemicals S1049). 24 h after plating, medium was changed to modified E8 medium containing: DMEM high glucose without glutamine (Thermo Fisher 11960044), 10.7 mg/L Transferrin (Sigma T0665), 64 mg/L L-Ascorbic Acid (Sigma A890), 14 μg/L Sodium Selenite (Sigma S5261), 543 mg/L Sodium Bicarbonate (Research Products International 144558), 19.4 mg/L insulin (Sigma I9278), 100 μg/L bFGF (EMD Millipore GF003AF), 2 μg/L TGFβ1 (Peprotech 10021), and 2 mM L-glutamine. After 24 h of culture in modified E8 medium, medium was changed to modified E8 medium containing 2 mM or 0 mM L-glutamine for 24 h. All cells were then changed to modified E8 medium containing 2 mM L-glutamine and cultured for 24 h before harvest for analysis.

Growth curves.

ESCs were seeded at a density of 30,000–40,000 cells per well of a 12-well plate. The following day, three wells of each line were counted to determine the starting cell number. The remaining cells were washed with PBS and cultured in medium containing a 1:1 mix of glutamine-free DMEM and glutamine-free Neurobasal medium including 10% dialyzed FBS, 2-mercaptoethanol, LIF, and containing or lacking L-glutamine or glucose as indicated and with or without the addition of supplements as indicated: dimethyl-α-ketoglutarate (4 mM), dimethyl-succinate (4 mM), methyl-pyruvate (2 mM), ruxolitinib (500 nM), methionine sulfoximine (1 mM). Cells were counted on the indicated days thereafter using a Beckman Multisizer 4e with a cell volume gate of 400 – 10,000 fL. Cell counts were normalized to starting cell number. All curves were performed at least two independent times.

Colony formation assays.

Cells were seeded at 200 or 500 cells per well in six-well plates in standard S/L medium. Medium was refreshed every 2–3 days. Six to seven days after initial seeding, wells were fixed with citrate/acetone/3% formaldehyde for 30 seconds and stained using the Leukocyte Alkaline Phosphatase Kit (Sigma) according to manufacturer instructions. Colonies were scored manually by two independent reviewers.

Fluorescence activated cell sorting.

For evaluation of cell viability and Nanog-GFP expression, Nanog-GFP ESCs³⁰ were seeded at a concentration of 40,000 cells per well of a 12-well plate. The next day, cells were washed with PBS and medium was changed to experimental medium containing a 1:1 mix of glutamine-free DMEM and glutamine-free Neurobasal medium including 10% dialyzed FBS, 2-mercaptoethanol, LIF, and containing (“+Q”) or lacking (“−Q”) L-glutamine as indicated. On the day of analysis, cells were trypsinized and resuspended in FACS buffer (PBS + 2% FBS + 1 mM EDTA) containing DAPI (1 μg/mL). Cells were evaluated for GFP and DAPI on either a LSRFortessa or LSR-II machine and FACSDiva software (Beckman Dickinson). Viable cells were those excluding DAPI (100-fold less than DAPI-positive cells). Nanog-GFP expression was measured by GFP mean fluorescence intensity and quantified using FlowJo v9.0. All experiments were performed at least two independent times.

For sorting of Nanog High, Nanog Medium, and Nanog Low populations, Nanog-GFP ESCs that were cultured either in S/L medium or adapted to S/L+2i medium as described above were resuspended in sterile FACS buffer containing DAPI. DAPI-excluding cells were evaluated for Nanog-GFP expression on a BD FACSAria III cell sorter (Becton Dickinson). “Nanog High,” “Nanog Medium,” and “Nanog Low” populations were sorted based on Nanog-GFP expression levels in the highest 10%, median 10%, and lowest 10% of the population, respectively. Following sorting, cells were washed 2 times with PBS to remove any residual FACS buffer and plated in 6-well gelatin-coated plates in standard S/L medium with the addition of penicillin/streptomycin (Life technologies). Cells were used 24 to 72 h later for experiments.

Evaluation of apoptosis was performed using an Annexin V Apoptosis Detection kit (BD Biosciences BDB556570). Nanog-GFP ESCs that had been sorted based on Nanog-GFP expression 24 h earlier as described above were plated in standard S/L medium. The next day, cells were washed with PBS and cultured in experimental medium containing a 1:1 mix of glutamine-free DMEM and glutamine-free Neurobasal medium including 10% dialyzed FBS, 2-mercaptoethanol, LIF, and containing (“+Q”) or lacking (“−Q”) L-glutamine as indicated. 24 h later, cells were trypsinized and resuspended in Annexin V binding buffer containing FITC-conjugated Annexin V and propidium iodide (PI) for 15 min at room temperature. Subsequently, excess binding buffer was added and both FITC and PI fluorescence was assessed on an LSRFortessa machine (Becton Dickinson). Apoptosis was quantified as cells positive for Annexin V based on a 100-fold increase in fluorescence as compared to negative cells.

For evaluation of OCT4 and SOX2 expression in human ESCs, cells were dissociated using TryPLE-Select (Thermo Fisher 12563029) and resuspended in FACS buffer (5% FBS and 5 mM EDTA in FBS). Cells were first stained with LIVE/DEAD Violet (Molecular Probe, L34955, 1:1,000) for 30 min at room temperature. Cells were fixed and permeabilized in 1X fix/perm buffer (eBioscience, 00–5523-00) for 1 h at room temperature. Cells were then stained with fluorophore conjugated antibodies OCT4-APC (eBioscience 50–5841- 82, 1:25) and SOX2-Alexa488 (eBioscience 53–9811-82, 1:100) in permeabilization buffer (Thermo Fisher 00–8333-56) for 30 min at room temperature. Cells were washed by addition of FACS buffer and centrifugation between all steps. Analysis was performed after resuspension in FACS buffer using an LSRFortessa machine (Becton Dickinson).

Further details on flow cytometry can be found in the Reporting Summary.

Immunofluorescence and microscopy.

For mouse ESCs, cells were seeded on 12-well MatTek glass-bottom dishes (P12G-1.0–10-F*) coated in laminin (Sigma, 10 μg/mL in PBS containing Ca²⁺ and Mg²⁺). Cells were fixed in 2% paraformaldehyde for 10 min and then permeabilized in 0.1% tween. Cells were washed with PBS and blocked for 1 h in 2.5% BSA in PBS. After blocking, cells were incubated overnight with primary antibodies diluted in blocking solution. The following antibodies were used: Oct¾ (Santa Cruz Biotechnologies, sc-5279 at 1:100) and Nanog (eBioscience, 145761–80 at 1:125). The next day, cells were washed with PBS and incubated with secondary antibodies (AlexaFluor 488 or 594 or 647, Molecular Probes) diluted 1:500 in blocking solution for 1 h. For nuclear counterstaining, Hoechst 33342 (Molecular Probes, H3570 at 1 µg/mL) was added to the same secondary solution. After washing with PBS, cells were stored in the dark and imaged within 1 or 2 days. Cells were imaged using an AxioObserver.Z1 epifluorescence inverted microscope with a motorized stage. A CCD attached camera allowed digital image acquisition (Hammamatsu, Orca II). For multi-well and multidimensional microscopy, definite focus was used and the microscope was programmed to image consecutive image fields (typically 60 per condition). These fields were stitched together using the built-in Axiovision function and exported as raw 16-bit TIFF files without further processing. Typically at least 10,000 cells per well at 200x magnification were imaged.

Image analysis required three steps: cell detection, nuclear segmentation and fluorescence detection in a per cell basis. These steps were implemented on custom-made Matlab (MathWorks) routines. First, cells were detected by adapting a Matlab implementation of the IDL particle tracking code developed by David Grier, John Crocker, and Eric Weeks (http://physics.georgetown.edu/matlab/). This algorithm finds cells as peaks in a Fourier space rather than by thresholding. This approach is less susceptible to problems that typically arise when segmenting large images such as autofluorescent and bright speckles or day-to-day variability in imaging conditions. Cell detection allowed us to count, identify and get the spatial coordinates (centroid) for each cell. Second, nuclear segmentation was achieved by a combination of regular thresholding together with a watershed process based on the distance of cell centroids determined in the previous step. Obtained nuclear regions were then used as masks to quantify pixel intensities for all the fluorescent channels (that reported levels of different proteins) on a per cell basis. We used cumulative values, which were then normalized to the Hoechst staining to correct for area, cell location along the Z-axis and DNA condensation differences. After image analysis, data was processed and plotted also with Matlab. Raw data, image analysis, and data processing routines are available upon request. Further details on staining and analysis can be found in the Reporting Summary.

Teratomas.

ESCs were initially plated in standard S/L medium as described above. The following day, cells were washed with PBS and then changed to experimental medium containing a 1:1 mix of glutamine-free DMEM and glutamine-free Neurobasal medium including 10% dialyzed FBS, 2-mercaptoethanol, LIF, and containing (“Ctrl”) or lacking (“Pulse −Q”) L-glutamine as indicated. 24 h later, cells were washed with PBS and then returned to glutamine-replete medium (“Recover”). 24 h later, 1 × 10⁶ cells per replicate were harvested from each group and mixed 1:1 with medium plus Matrigel Basement Membrane Matrix (BD) and injected into the flanks of recipient female SCID littermate mice aged 8–12 weeks (NOD scid gamma JAX 005557 purchased from Jackson Laboratories). All conditions produced tumors in 4–8 weeks. Mice were euthanized before tumor size exceeded 1.5 cm in diameter. Tumors were excised and fixed in 4% paraformaldehyde overnight at 4 °C. Tumors were paraffin-embedded and sections were stained with haematoxylin and eosin according to standard procedures by HistoWiz. All experiments were performed in accordance with a protocol approved by the Memorial Sloan Kettering Institutional Animal Care and Use Committee.

Statistical analyses.

GraphPad PRISM 7 software was used for statistical analyses except for IF data. Error bars, P values and statistical tests are reported in figure legends. Statistical analyses on images were performed using Matlab. We set the threshold to define “Oct4-low” cells as one standard deviation below the mean values of the control population (typically S/L in the presence of glutamine).

Data Availability.

The data and MatLab code that support the findings of this study are available from the corresponding author upon reasonable request. Source data for all GC-MS data are provided in Supplementary Table 1.