Metabolic regulation of species-specific developmental rates

Animals display significant inter-specific variation in the rate of embryonic development despite broad conservation of the overall sequence of developmental events. Differences in biochemical reaction speeds, including the rates of protein production and degradation, are thought to be responsible for distinct species-specific rates of development [1–3]. However, the cause of differential biochemical reaction speeds between species remains unknown. Using pluripotent stem cells, we have established an in vitro system that recapitulates the two-fold difference in developmental rate between early mouse and human embryos. This system provides a quantitative measure of developmental speed as revealed by the period of the segmentation clock, a molecular oscillator associated with the rhythmic production of vertebral precursors. Using this system, we showed that mass-specific metabolic rates scale with developmental rate and are therefore elevated in mouse cells compared to human cells. We further showed that reducing these metabolic rates by pharmacologically inhibiting the electron transport chain slows down the segmentation clock. The effect of the electron transport chain on the segmentation clock is mediated by the cellular NAD+/NADH redox balance independent of ATP production and, further downstream, by the global rate of protein synthesis. These findings represent a starting point for the manipulation of developmental rate, which would find multiple translational applications including the acceleration of human pluripotent stem cell differentiation for disease modeling and cell-based therapies.

gestation in mammals ranges from 20 days in mice to 645 days in African elephants [7,8]. Differences in developmental rate are best illustrated at early stages of embryogenesis when different species are most morphologically and molecularly similar. At 37°C, mouse embryos start gastrulation at day 7 post-fertilization (PF), the first somite forms at day 8, and the forelimb bud is first visible at day 9; whereas in human embryos gastrulation begins at day 16 PF, the first somite forms at day 20, and the forelimb bud is first seen at day 30 [9]. Thus, even though early mouse and human embryos undergo the same series of developmental steps and share a similar overall size, human embryos do so at rates 2-3 times slower [9].
Such differences in developmental rate between mouse and human embryos can be recapitulated in vitro by studying the differentiation kinetics of pluripotent stem cells (PSCs) [1,2,10,11]. We previously reported the establishment of PSC-derived models of the mouse and human segmentation clocks, which displayed approximately two-fold slower oscillations in human cells compared to mouse cells [12]. The segmentation clock is a molecular oscillator that operates in presomitic mesoderm (PSM) cells and controls the periodicity of somite formation in vertebrate embryos [13]. This clock represents an ideal model for developmental rate because its period is species-specific, temperature-sensitive, and scales with the speed of development [1,14,15]. We thus adopted our in vitro segmentation clock system as a platform to directly compare developmental rate between mouse and human cells.
To mitigate the influence of potential confounding factors in the comparison of mouse and human cells, we first developed an updated protocol that allowed mouse and human PSCs to be differentiated towards PSM fate under identical conditions (Fig. 1a). This protocol still relies on Wnt activation coupled with BMP inhibition to induce PSM, but the basal media was standardized for both species. The differentiation efficiency, as measured by induction of a specific posterior PSM fate marker (MSGN1-Venus), was remarkably high at 78.3 ± 1% for mouse cells and 96.5 ± 1.5% for human cells (Fig. 1b, Extended Data Fig. 1a). Under these conditions, we observed differences in developmental rate between mouse and human cells at several levels. First, mouse cells activated MSGN1-Venus with accelerated kinetics compared to human cells (1-2 days for mouse; 2-3 days for human; Fig. 1b). Second, the cell cycle duration was shorter in mouse (13.9 ± 2 hours) than in human (21.9 ± 3.6 hours) PSM cells (Fig.  1c). Lastly, as previously reported, oscillations of the segmentation clock reporter HES7-Achilles were faster in mouse (2.6 ± 0.3 hours) compared to human (4.9 ± 0.3 hours) PSM cells (Fig. 1de) [12,[16][17][18]. Together, these observations indicated an approximately two-fold difference in developmental rate between mouse and human PSM cells differentiating in vitro.
To validate that this in vitro model accurately recapitulated developmental rate, we directly compared the cell cycle time and segmentation clock period of PSC-derived mouse PSM cells with primary explants of mouse PSM taken from E9.5 embryos [19]. We found that these parameters did not significantly differ between PSC-derived and primary PSM cells (Extended Data Fig. 1b-c). These results confirmed that PSM cells generated in vitro can precisely reflect the endogenous pace of development.
Differences in developmental rate can be cell autonomous or context dependent. For instance, the neural differentiation rate of human PSCs can be accelerated by co-culturing them with an excess of mouse cells [20]. In contrast, human PSC derivatives maintain their slow developmental rate during teratoma formation in a mouse host or when grafted as cortical neurons into the mouse brain [10,21]. The segmentation clock retains its species-specific period in isolated cells, thus suggesting that it is a cell autonomous feature that does not require tissue-level coupling [1,12,19]. To further determine whether the period of the segmentation clock is indeed cell autonomous, we co-cultured HES7-Achilles human PSM cells with control PSM cells of either human or mouse origin at a ratio of 1:100 (Fig. 1f). The segmentation clock period was the same when co-cultured with either human (4.71 ± 0.94 hours) or mouse (4.71 ±0. 88 hours) PSM cells (Fig. 1g). This result indicated that the segmentation clock period is cell autonomous even in chimeric conditions. The cell cycle has been proposed to function as a clock that controls developmental speed [2,22]. Our observation that cell cycle duration scales with developmental rate in mouse and human PSM cells motivated us to test this hypothesis. We blocked cell division in human PSM cells with the replication inhibitor aphidicolin. However, complete arrest of the cell cycle at early S phase did not impact the oscillatory period ( Fig. Extended Data Fig. 1d-e). We thus conclude that the cell cycle does not contribute to the regulation of the segmentation clock period.
In general, the mass-specific metabolic rate of smaller animals is accelerated compared to that of larger animals [23,24]. Kleiber's law, a well-established allometric relationship between body mass and metabolic rate, applies to most animals (Kleiber's law) [23]. Whether this reflects scaling of the intrinsic cellular metabolic rates or constraints in energy delivery by the vascular system in adults is controversial [24,25]. The cellular mass-specific metabolic rate has not been compared across species during development and thus whether metabolic rate scales with developmental rate remains unknown. Such differences in basal metabolism could provide an attractive hypothesis to explain the accelerated biochemical kinetics observed in mouse cells compared to human cells [1,2].
Having established that our in vitro segmentation clock system represented a reliable and cell-autonomous model of developmental rate, we set out to compare the rates of metabolism of mouse and human PSM cells directly. We first measured the oxygen consumption rate (OCR) and glycolytic proton efflux rate (glycoPER). When an identical number of cells were seeded per well, no significant difference was observed in basal glycoPER between species, and basal OCR was slightly elevated in human cells (Extended Data Fig. 2ab). However, we noted that human PSM cells were approximately twice as large in volume as mouse cells (2060 ± 524 fL vs. 885.9 ± 157.9 fL, respectively) ( Fig. 2a; Extended Data Fig. 2c).
To compare how total cell mass differs between human and mouse cells, we measured the total mass, volume and density of individual cells by repeatedly weighing the cells in different density fluids using the suspended microchannel resonator [26,27]. The total mass of human cells was also approximately twice that of mouse cells (2002 ± 71 pg vs. 1066 ± 56 pg) (Fig. 2b), such that the overall density was similar for both species (1.057 ± 0.003 pg/fL vs. 1.061 ± 0.002 pg/fL; Fig. 2c). PSC-derived human PSM cells were thus two-fold larger and heavier than mouse PSM cells. Therefore, the metabolic rate per cell could not be directly compared between mouse and human PSM cells. Instead, we compared the mass-specific metabolic rates and found that mass-specific OCR and glycoPER were twice as fast in mouse cells as in human cells (Fig. 2de).
We confirmed that mass-specific metabolic rates are faster in mouse cells than in human cells using multiple independent approaches. We measured glucose consumption and lactate secretion over the course of six hours for both species. Both glucose consumption and lactate secretion per unit mass were significantly higher in mouse cells compared to human cells ( Fig.  2f-g). The mass-specific glutamine consumption, which fuels the TCA cycle through glutaminolysis, was also significantly elevated in mouse PSM cells (Extended Data Fig. 2d). We performed stable isotope tracing with uniformly labeled U-13 C6-Glucose and U-13 C5-Glutamine in both mouse and human PSM cells. At isotopic steady state, we observed that glucose tracing led to high levels of labeling for pyruvate and lactate, as well as partial labeling of TCA intermediates (Extended Data Fig. 2e-j). Glutamine tracing showed intermediate levels of labelling for glutamate and TCA metabolites, but not pyruvate nor lactate supporting an anaplerotic role in the TCA cycle (Extended Data Fig. 2k-p). These data are consistent with the Warburg-like metabolism of PSM cells, wherein active glycolysis producing large amounts of lactate coexist with aerobic respiration [28,29]. Importantly, stable isotope labeling patterns were almost identical between mouse and human PSM cells, suggesting that glucose and glutamine are broken down by metabolic pathways in the same way in both species and that only the rate of these processes differs between them. Together, these data confirmed that mouse PSM cells display faster mass-specific metabolic rates than human PSM cells.
To determine whether a higher mass-specific metabolic rate in mouse cells relative to human cells is observed in other cell types, we compared human and mouse neural progenitors differentiated in vitro from pluripotent stem cells. Simultaneous inhibition of the TGFβ and BMP pathways [30] led to the induction of PAX6+ neural progenitors after 5 days of treatment for mouse and 7 days for human PSCs (Extended Data Fig. 3a-bf). We did not detect a significant difference in volume between mouse and human PSCs (Extended Data Fig. 3c), indicating that it is not generally true that human cells are larger than their mouse equivalents. However, the mass-specific OCR and extracellular acidification rate (ECAR) were significantly higher in mouse cells than in human cells (Extended Data Fig. 3d-ei). Thus mass-specific metabolic rates are higher in different mouse cell types compared to human.
Next, we performed the mitochondrial stress test on the Seahorse instrument to measure more detailed aspects of cellular respiration beyond the basal rate. Mouse and human PSM cells differed in their maximal respiration rate and their spare respiratory capacity, both of which were significantly higher in mouse cells (Extended Data Fig. 3f-g). Respiration was less coupled to ATP production in mouse cells than human cells, thus reflecting a higher level of proton leak in mouse (Extended Data Fig. 3f,h) [31]. One potential explanation for the increased basal and maximal respiration rates of mouse cells could be a higher mitochondrial density. Indeed, comparison of the mitochondrial content between mouse and human cells using the MitoTracker Deep Red dye revealed that mouse cells contain a moderately but significantly higher number of mitochondria per unit volume (Fig. 2h). The total mitochondrial content per cell was nonetheless approximately 60% higher in human cells (Extended Data Fig. 3i), so that the oxygen consumption per mitochondrion was elevated in mouse cells relative to human cells (Extended data Fig. 3j). This was suggestive of inherent differences in mitochondrial properties between these species [32]. We thus measured the inner mitochondrial membrane potential (ΔΨm) with the dye tetramethylrhodamine methyl ester (TMRM), which displays increased fluorescence intensity at higher ΔΨm values. The TMRM signal was normalized by MitoTracker signal to account for differences in mitochondrial content between the species. This experiment revealed that human cells display a slightly higher TMRM/MitoTracker signal (Fig. 2i), suggesting that the higher OCR of mouse compared to human cells is not linked to a higher ΔΨm . The abundance and functional properties of mitochondria thus differed significantly between mouse and human PSM cells.
We next used oscillations of the segmentation clock as a proxy to investigate which aspect of metabolism controls embryonic time. In the chicken embryo, inhibition of respiration but not glycolysis alters segmentation clock oscillations [28]. We thus partially impaired the electron transport chain (ETC) with a set of small molecule inhibitors at sub-lethal concentrations in human PSM cells and measured the effect on the period of the segmentation clock. Inhibitors of the ETC complexes I (rotenone), III (antimycin a) and IV (sodium azide), which pump protons across the inner mitochondrial membrane and transfer electrons (Fig. 3a), all resulted in a significant lengthening of the segmentation clock period (Fig. 3b). Surprisingly, inhibiting the ATP synthase with oligomycin did not alter the oscillatory period ( Fig. 3a-b). All these treatments resulted in reduced OCR and caused dampened oscillatory dynamics with a premature arrest of oscillations, suggesting that the segmentation clock cannot continue operating long-term when oxidative phosphorylation is impaired (Extended Data Fig. 4a-e). As ETC impairment leads to cell cycle arrest [33], we could not measure the cell cycle length under these conditions. Uncoupling the ETC from ATP production by treatment with the ionophore FCCP did not change the period either ( Fig. 3b; Extended Data Fig. 4f-g), suggesting that it is the activity of the ETC rather than oxidative phosphorylation that is relevant to the segmentation clock period. These results are consistent with the phenotype of ETC mutants in Caenorhabditis elegans, where mutants with impaired complex I-IV function display slow larval growth, whereas ATP synthase mutants grow at normal rates [34]. It thus seems that complexes I-IV of the ETC play a conserved role in regulating developmental rate.
The observation that ATP synthase inhibition did not significantly change the segmentation clock period suggested that cellular ATP levels are not involved in its regulation. In fact, the ADP/ATP ratio was higher in mouse PSM cells than in human cells (Fig. 2j), suggesting that an increased cellular energy charge does not mediate the accelerated developmental rate of mouse cells. To further test this hypothesis, we cultured human PSM cells under conditions that increased the cellular ATP concentration. First, we supplemented the media with succinate to directly feed the ETC at complex II. Second, we replaced glucose by galactose in the medium to increase reliance on oxidative phosphorylation for ATP production. Third, we treated the cells with the pyruvate dehydrogenase kinase (PDK) inhibitor dichloroacetate (DCA), which increases the conversion of pyruvate to acetyl-CoA for consumption by the TCA cycle (Fig. 3c). Under all three conditions, ATP levels were increased (Fig. 3d) and the OCR/ECAR ratio was increased (Extended Data Fig. 5a). We then measured the segmentation clock period and found that not only was the period not shortened, but it was moderately lengthened ( Fig. 3e; Extended Data Fig. 5b-d). Under these conditions, the cell cycle length was also increased, reflecting slower proliferation (Fig. 3f). Together, these results confirmed that higher ATP concentrations do not promote faster oscillations or proliferation.
In cancer cells, PDK inhibition also diminishes the cellular proliferation rate [35]. Upon PDK inhibition, pyruvate is shunted away from lactate dehydrogenase (LDH), leading to a lower NAD + /NADH ratio (Fig. 3c). At the same time, complex I of the ETC cannot sufficiently regenerate cellular NAD + because the inner mitochondrial membrane becomes hyperpolarized and opposes the pumping of additional protons. Supplementing cells with pyruvate, which can be rapidly reduced by LDH to generate NAD + , has been shown to rescue the proliferation rate of cells treated with a PDK inhibitor by restoring the NAD + /NADH ratio [35]. Similarly, restoring ETC activity by dissipating the elevated ΔΨm with uncoupling agents like FCCP has also been shown to rescue the NAD + /NADH ratio and proliferation rates [35]. We reasoned that a similar mechanism may be responsible for the increased segmentation clock period in PSM cells subjected to PDK inhibition. Indeed, we observed a significantly lower total NAD + /NADH ratio (Fig. 3g) and increased ΔΨm (Extended Data Fig. 5e) in PSM cells treated with DCA. The total NAD + /NADH ratio could be restored to control levels by pyruvate supplementation or FCCP treatment (Fig. 3g). These total NAD + /NADH ratio measurements included both the mitochondrial and cytoplasmic NAD(H) pools. We generated a human PSC line carrying the fluorescent sensor Peredox, which displays higher fluorescence when cytoplasmic -but not mitochondrial-levels of NADH are increased [36] (Extended Data Fig. 5f-g). Using this reporter line, we could confirm the changes in cytosolic NAD + /NADH ratio upon DCA treatment and its restoration by pyruvate and FCCP (Fig. 3h). Importantly, the period of the segmentation clock in cells treated with DCA was fully rescued by both pyruvate and FCCP (Fig. 3i). Together, these data suggest that the segmentation clock period depends on NAD+ availability rather than ATP supply.
Impaired ETC activity also decreases the NAD + /NADH ratio because the regeneration of NAD + by complex I is altered. The complex I, III and IV inhibitors that cause slow segmentation clock oscillations also cause low total NAD + /NADH ratio and increased Peredox fluorescence (Fig. 3j-k). Pyruvate supplementation of cells treated with the complex IV inhibitor sodium azide partially restored the NAD + /NADH ratio [37] (Fig. 3j-k) and also partially rescued the segmentation clock period ( Fig. 3l; Extended Data Fig. 5h). As an orthogonal approach, we supplemented azide-treated cells with duroquinone, which can mediate NAD + regeneration from NADH via the quinone oxidase NQO1 [38]. Duroquinone treatment led to a modest recovery of the NAD + /NADH ratio (Fig. 3k) and slightly but significantly accelerated segmentation clock oscillations relative to cells treated with azide alone (Fig. 3l; Extended Data Fig. 5i). This suggests that ETC inhibition slowed the segmentation clock by impairing NAD redox homeostasis.
The cytoplasmic NAD + /NADH ratio can be manipulated by modifying the amount of pyruvate and lactate in the cell culture medium, as these substrates will shift the balance of the reaction catalyzed by LDH towards NAD + or NADH production, respectively. In the presence of high glucose, pyruvate or lactate supplementation did not change the NAD + /NADH ratio nor did it impact the segmentation clock period (Extended Data Fig. 6a-d). Under conditions of glucose and glutamine starvation, the period of the segmentation clock was severely increased (Extended Data Fig. 6c, 6e). However, starved cells supplemented with pyruvate exhibited a higher NAD+/NADH ratio and oscillated faster than those supplemented with lactate (Extended Data Fig. 6a-c, 6e). Similarly, inhibiting LDH with sodium oxamate to prevent NAD + regeneration by this enzyme led to a decreased NAD+/NADH ratio and an increased segmentation clock period (Extended Data Fig. 6f-i). Thus, the tempo of the segmentation clock could be manipulated by changing the cytosolic NAD + /NADH ratio.
These results suggested that differences in NAD redox balance may underlie the differences in developmental rate between mouse and human cells. NAD + is used as an electron acceptor in many metabolic reactions and is required for key steps in central carbon metabolism, nucleotide synthesis, amino acid metabolism and lipid metabolism [39]. Faster developmental rates might therefore demand higher NAD + availability to sustain accelerated growth and differentiation. Surprisingly, the total NAD + /NADH ratio was significantly lower in mouse cells compared to human cells (Fig. 2k). However, mouse PSM cells contained overall higher levels of both NAD + and NADH than human cells (Extended Data Fig. 3k), suggesting that NAD + and NADH concentrations may be more important than the NAD + /NADH ratio per se for the regulation of oscillatory period between species. The related dinucleotide, NADP(H), plays a central role in the defense against oxidative stress and serves as an electron donor for the synthesis of nucleotides, lipids and amino acids [40]. We therefore measured the total NADPH/NADP + ratio in mouse and human PSM cells but found that it did not differ significantly between the species (Extended Data Fig. 3l).
We next asked whether these differences in respiration between mouse and human PSM cells could be responsible for the differences in protein production and degradation rates that were suggested to underlie species-specific developmental timing [1,2]. Indeed, increased mitochondrial activity is correlated with faster transcription, faster translation, and accelerated growth in human cell lines [41-43]. We first compared translation rates by pulsing PSM cells of both species with puromycin for one hour. We subsequently detected the amount of puromycilated peptides produced during this time, a surrogate measure of the rate of total protein translation, using either antibodies or click chemistry (Fig. 4a). These experiments indicated that the mass-specific translation rate is almost twice as fast in mouse PSM cells compared to human cells and therefore scales with developmental rate ( Next, we asked if the ETC inhibitors that slow down the segmentation clock can impact the translation rate. Indeed, rotenone, antimycin A and sodium azide all decreased the translation rate as measured by puromycin incorporation (Fig. 4e). Strikingly, the magnitude of the effect on the segmentation clock period scaled with the magnitude of translation rate reduction for these inhibitors. Supplementing azide-treated cultures with pyruvate to partially restore the NAD + /NADH ratio also partially rescued the translation rate (Fig. 4e). Cultures treated with sodium azide for only one hour already displayed reduced translation, indicating rapid downregulation of protein synthesis upon ETC inhibition (Extended Data Fig. 7d). These results suggested that ETC inhibition and NAD + depletion slow down the segmentation clock at least in part by reducing the protein production rate. To confirm this epistatic relationship, we performed Seahorse experiments in human PSM cells treated with cycloheximide. All aspects of respiration, including the basal and maximal rates, the proton leak and the non-mitochondrial respiration, were indistinguishable between cycloheximide-treated and control cells ( Fig. 4f; Extended Data Fig. 7e-f). We thus concluded that mitochondrial activity acts upstream of the translation rate to control the segmentation clock period.
Differences in global protein stability also correlate with species-specific developmental rates, as the proteome half-life is 2-fold shorter in mouse neural progenitors compared to human [2]. In PSM cells, the half-life of puromycilated peptides was indeed shorter in mouse PSM cells compared to human cells (Fig. 4g). Given that the proteasome is responsible for the degradation of most cellular proteins, we compared the activity of the proteasome between the two species. As expected, the mass-specific proteasome activity was significantly elevated in mouse PSM cells (Fig. 4h), suggesting that the reduced protein stability observed in this species is caused by higher proteasome activity.
We next tested whether manipulating global protein stability could affect the segmentation clock period. High levels of proteasome inhibition led to an arrest of segmentation clock oscillations as negative feedback loops are disrupted [44]. Treating human PSM cells with low doses of the proteasome inhibitors bortezomib or lactacystin caused premature arrest of oscillations but did not change the oscillatory period despite significant inhibition of proteasome activity ( Fig. 4i; Extended Data Fig. 7g-h). The cell cycle duration could not be assessed under these conditions as proteasome inhibitors induce cell cycle arrest [45]. In addition, azide treatment, which slowed down oscillations, did not reduce proteasome activity or increase the stability of puromycilated peptides (Extended Data Fig. 7i-j). To assess the degradation profile of the full-length proteome, we performed pulse-chase experiments with the methionine analog L-azidohomoalanine (AHA). Unlike puromycin, AHA incorporation into growing peptides does not induce chain termination and thus labels full-length proteins. AHA labeling resulted in indistinguishable decay profiles between control and azide-treated cells over the timeframe relevant to segmentation clock oscillations (Extended Data Figure 7k). However, protein degradation seemed to stall after 18 hours of azide treatment, suggesting that a lack of protein turnover may contribute to premature arrest of oscillations upon ETC inhibition (Extended Data Figure 7k). Together, these experiments suggest that the segmentation clock period is more sensitive to the rate of protein production than the rate of protein degradation.
In summary, we found that mass-specific respiration rates scale with and regulate the segmentation clock period by modulating NAD redox balance and, more downstream, the global translation rate. Given that the segmentation clock period can be used as a proxy for developmental rate [1], our results may explain at least in part the differences in developmental rate observed at early stages of mouse and human embryonic development. Such a mechanism may also be modulated locally in the embryo to generate heterochronic changes in segmentation, as for instance in the acceleration of segmentation clock speed relative to developmental rate in snakes [46]. However, additional studies in other embryonic cell types and other mammalian species are still required to assess the generality of our findings. In particular, it remains unclear whether cells not undergoing Warburg-like metabolism would similarly rely on NAD + over ATP [47]. Moreover, distinct mechanisms may control developmental speed at fetal stages of development, when the difference in both size and developmental rate quickly increases between mouse and human [9].
Our findings also suggest that mass-specific metabolic rates for embryonic cell types may scale with adult body mass as predicted by Kleiber's law, even under uniform culture conditions [31]. Interestingly, different cell types may achieve such scaling through opposing strategies: whereas PSM cells scaled their mass, neural progenitor cells scaled their metabolism [48]. Our studies also revealed a striking parallel between the metabolic requirements of cancer cell proliferation and those of the segmentation clock [49]. In both cases, NAD+ redox homeostasis is more important than ATP availability to maintain normal growth and oscillation rates, respectively [35]. Given that PSM cells exhibit Warburg-like metabolism with high levels of aerobic glycolysis [28], these similarities further strengthen the notion that cancer cells resemble embryonic progenitors. Moreover, this connection suggests that understanding how developmental rate is regulated will have potential applications in halting tumor growth. Similarly, the finding that the segmentation clock period is sensitive to NAD + levels draws a parallel between developmental rate and aging. NAD + decreases progressively with age and restoring NAD + levels can ameliorate aging-associated phenotypes [50]. The aging process and developmental rate may share some regulatory mechanisms, especially given that lifespan and gestation period are positively correlated [51]. Future work should focus on the identification of factors that underlie increased mass-specific respiration rates in fast-developing species [31]. Ultimately, interspecific differences in developmental rate must be traceable to genetic causes. Lastly, the implication of translation rate downstream of mitochondrial activity strongly suggests that the integrated stress response may play a role in determining developmental rate [47,52]. Continued research in this area will reveal how developmental time can be manipulated externally, with important applications in human stem cell therapy and in vitro disease modeling. . The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank the NeuroTechnology Studio at Brigham and Women's Hospital for providing access to the Agilent Seahorse XF96 extracellular flux analyzer and Zeiss LSM880 confocal microscope, and for consultation on data acquisition and data analysis. We also thank the Center for Neurologic Diseases (ARCND) Flow Cytometry Core Facility at Brigham and Women's Hospital for access to the BD Fortessa flow cytometer. In addition, we thank the Seahorse core facility at Brigham and Women's hospital for training and access to the Agilent Seahorse XFe24 extracellular flux analyzer. We also express our gratitude towards Dr. Seungeun Oh and the Kirschner laboratory at Harvard Medical School for generously providing access to the Moxi Go II for cell volume measurements.       Extended Data Table 1 List of small molecule inhibitors and metabolite supplements . Whenever the effects of chemical inhibitors or culture conditions on the segmentation clock were tested, human HES7-Achilles or HES7-Achilles; AAVS1-CAG-H2B-mCherry cells were differentiated under serum-free conditions as previously described [2] to avoid confounding factors from FBS composition.

Neural Progenitor Differentiation
Neural progenitor induction relied on dual Smad inhibition and was adapted from previously described protocols [6]. Mouse ESCs were seeded on fibronectin-coated dishes at a density of 1× 10 4 cells/cm 2 and pre-differentiated to epiblast state as described above. Human iPSCs were seeded on matrigel-coated plates at a density of 3.5× 10 4 cells per cm 2 in mTeSR1 with 10μM Rocki. At 24 hours, the medium was replaced by mTeSR1 without Rocki. Two days after initial seeding, both mouse and human cells were switched to NDiff227 supplemented with 1% FBS, 0.1 μM LDN193189, and 10 μM SB431542 (Selleck Chemicals cat. no. S1067). The media was refreshed daily. Mouse cells were cultured for 5 days and human cells for 7 days. Neural progenitor fate was confirmed by PAX6 immunofluorescence as described below.

Generation, Validation and Imaging of AAVS1-CAG-Peredox-mCherry-NLS
CAG-Peredox-mCherry-NLS was inserted into the AAVS1 safe harbor locus using the approach previously described by Oceguera-Yanez and colleagues [7]. Briefly, we cloned the CAG-Peredox-mCherry-NLS sequence from pcDNA3.1-Peredox-mCherry-NLS (Addgene cat. no. 32384) into the pAAVS1-P-CAG-DEST vector (Addgene cat. no. 80490) by Gibson assembly and co-transfected it along with the pXAT2 vector (Addgene cat. no. 80494) into NCRM1 cells. Two days after transfection, we selected positive clones by supplementing mTeSR1 with puromycin (0.5 μg/mL, Sigma-Aldrich cat. no. P7255) for a total of 10 days. To enhance single cell survival, we added CloneR (StemCell Technologies cat. no. 05888) to the media during the first 2 days of selection. We obtained several positive clones and confirmed the homozygous insertion of the sensor by PCR as previously described [7].
To validate that the Peredox NADH/NAD+ sensor worked as expected in the newly generated AAVS1-CAG-Peredox-mCherry-NLS line, we first performed fluorescence lifetime imaging (FLIM) under different conditions (Extended Data Fig. 4f) . Peredox emission was recorded in the green channel. The photodetector signals and laser sync signals were preamplified and then digitized at 1.25 gigasamples per second using a field programmable gate array board (PC720 with FMC125 and FMC122 modules, 4DSP, Austin, TX). Microscope control and image acquisition, as well as laboratorybuilt firmware and software for fluorescence lifetime determination, can be found elsewhere [8].
To further confirm that similar results could be obtained by ratiometric measurements, we imaged AAVS1-CAG-Peredox-mCherry-NLS cells with an LSM880 confocal microscope using a 20X/0.8 objective (Extended Data Fig. 4g) PSM cells were cultured in a balanced salt solution supplemented with 2mM, 0.2mM or 0mM glucose, 10mM lactate, 10mM pyruvate or a mixture of lactate to pyruvate at a ratio of 30:1 as described above. Images were thresholded based on mCherry-NLS fluorescence and nuclei were automatically segmented in Fiji [9]. A ratio of the mean Peredox: mCherry fluorescence intensity was calculated for each cell. All cells within an image were averaged to obtain the mean ratio for each sample. This ratiometric imaging approach was used in all experiments.

Mouse and Human PSM Co-Culture
Mouse and human PSCs were differentiated to PSM fate separately as described above. On day 2 of differentiation, PSM cells were dissociated with TrypLE. Mixed cell suspensions consisting of HES7-Achilles; AAVS1-CAG-H2B-mCherry human cells combined with either human NCRM1 or mouse E14 cells at a ratio of 1:100 were prepared and reseeded on fibronectin-coated 24 well glass-bottom plates at a density of 2.5 x 10 6 cells per cm 2 . Cells were then allowed to attach for one hour and subjected to timelapse imaging. AAVS1-CAG-H2B-mCherry+ cells were automatically segmented and tracked as previously described [2] to obtain single-cell HES7-Achilles fluorescence intensity profiles.

Primary Mouse PSM Explant culture
Explant culture was performed as described by Hubaud and colleagues [5].

Timelapse Microscopy
Time lapse-imaging of mouse and human PSM cells was performed on a Zeiss LSM 780 pointscanning confocal inverted microscope fitted with a large temperature incubation chamber and a CO2 module. An Argon laser at 514 nm and 7.5% or 2% power was used to excite Achilles or Venus fluorescent proteins, respectively. A DPSS 561 laser at 561nm and 2% laser power was used to excite mCherry. In all cases, a 20X Plan Apo (N.A. 0.8) objective was used to acquire images with an interval of 18 or 25 minutes in the case of human samples and 12 minutes for mouse samples, for a total of 24-48 hours. A 3x3 tile of 800x800 pixels per tile with a single zslice of 18 μm thickness and 16-bit resolution was acquired per position. Multiple positions, with at least two positions per sample, were imaged simultaneously using a motorized stage. Mouse explant imaging was performed on a Zeiss LSM780 microscope using a 20X/0.8 objective.
Single section (~19.6μm thick) with tiling (3x3) of a 512x512 pixels field was acquired every 7.5 minutes at 8-bit resolution.

Oscillation analysis
Time lapse movies of HES7-Achilles were first stitched and separated into subsets by position in the Zen program (Zeiss). Then, background subtraction and Gaussian blur filtering were performed in Fiji [9] to enhance image quality. A small region of interest (ROI) was drawn and the mean fluorescence intensity over time was calculated. Intensity is presented in arbitrary units. For figure 1e, intensity profiles were smoothened in GraphPad Prism using 6 neighboring data points and a 2 nd order polynomial. Profiles were then normalized between zero and one.
The oscillatory period was defined as the average time between two peaks in HES7-Achilles profiles.

Cell Cycle Length Measurements
To generate cultures with sparsely labeled cells, we mixed HES7-Achilles; AAVS1-CAG-H2B-mCherry human iPSCs or pMsgn1-Venus mouse ESCs with their parental line (NCRM1 or E14, respectively) at a ratio of 1:100 during the initial seeding. Cultures were then differentiated normally and subjected to timelapse imaging. Individual reporter cells were tracked manually on Fiji [9] and the time of cell division was recorded. The cell cycle length was defined as the time elapsed between the time a cell first divides and the time one of its daughter cells divides again.
To measure cell cycle length in primary mouse PSM explants, lentiviral infection was used to sparsely label cells with an SV40-mCherry reporter [5]. The plasmid E[beta]C (Addgene cat. no. 24312) was used. Lentivirus was produced in 293T cells, which were transfected using the CaCl2 method with the packaging plasmids psPAX2 (Addgene cat. no. 12260) and pVSVG (gift from M. Wernig lab). Supernatant was collected, filtered using a 0.45μM filter and concentrated by centrifugating 4 volumes of supernatant on 1 volume of TNE buffer (50mM Tris pH7.2, 100mM NaCl, 0.5mM EDTA, 15% sucrose) at 7197 relative centrifugal field (rcf) for 4 hours at 4°C. Explants were infected for ~4 hours and further incubated overnight before imaging. SV40-mCherry+ cells were manually tracked, and the cell cycle length was calculated as described above.

Flow Cytometry and Cell Sorting
To determine the fraction of PSM cells expressing MSGN1-Venus, cultures were dissociated with Accutase and analyzed by flow cytometry using an S3 cell sorter (Biorad). Undifferentiated ESCs or iPS cells, which do not express the fluorescent protein, were used as a negative controls for gating purposes. Samples were analyzed in biological triplicates. Results are presented as the percentage of Venus-positive cells among singlets. The same gating strategy was used to sort MSGN1-Venus+ for subsequent experiments. All other flow cytometry analyses were performed on a 5-laser Fortessa analyzer (BD). Automatic compensation was set up whenever more than one stain or fluorescent protein was used at a time. Flow cytometry data was analyzed in FlowJo. The mean fluorescence intensity (MFI) for 10,000 cells is presented. In the case of mouse vs. human comparisons, only MSGN1-Venus+ cells were considered in the analysis and MFI was normalized to cell mass.

Mitochondrial content and ΔΨm measurement
PSM cells were washed in PBS and incubated in DMEM supplemented with 25nM Mitotracker Deep Red FM (Invitrogen cat. no. M22426), 25nM Mitotracker Green (Invitrogen cat. no. M7514), and/or 20nM TMRM (Invitrogen, T668) for 30 minutes as previously described [11]. The cells were then dissociated with TrypLE, resuspended in PBS-1%FBS and analyzed by flow cytometry. As a control, 1uM FCCP was used to depolarize the inner mitochondrial membrane.

Seahorse Assays
PSM cells were dissociated on day 2 of differentiation and reseeded onto fibronectin-coated Seahorse plates (Agilent cat. no. 101085-004) at a density of 7 x 10 5 cells per cm 2 in Seahorse XF DMEM (Agilent cat. no. 103575-100) supplemented with 10mM glucose (Agilent cat. no. 103577-100), 1mM pyruvate (Agilent cat. no. 103578-100) and 2mM glutamine (Agilent cat. no. 103579-100). For mouse vs. human comparisons, MSGN1-Venus+ cells were pre-sorted. Cells were allowed to attach at room temperature for 20 minutes and then transferred to a 37°C incubator without CO2 for 1 hour. The Seahorse cartridge was hydrated and calibrated as per the manufacturer instructions. For the Mitochondrial Stress Test (Agilent cat. no. 103015-100), we used oligomycin at 1μM, FCCP at 1μM, rotenone at 0.5μM and antimycin A at 0.5μM. No glucose controls were used to calculate the CO2 contribution factor for glycolytic proton efflux rate determination. All samples were run in six to ten replicates in either a Seahorse XF96 or Stable isotope quantification: All metabolites except fructose 2,6-bisphosphate (FBP) and 3phosphoglycerate (3PG) were measured using the following mobile phase gradient: 0 min, 80% B; 5 min, 80% B; 30 min, 20% B; 31 min, 80% B; 42 min, 80% B. The mass spectrometer was operated in selected ion monitoring mode with an m/z window width of 9.0 centered 1.003355times half the number of carbon atoms in the target metabolite. The resolution was set at 70,000 and AGC target was 1 × 105 ions. Peak areas were corrected for quadrupole bias as previously described [12]. Mass isotope distributions for FBP and 3PG were calculated from full scan chromatograms as described below. Raw mass isotopomer distributions were corrected for natural isotope abundance using a custom R package (mzrtools, https://github.com/oldhamlab/mzrtools) employing the method of Fernandez, et al. [13].

Cell Volume Measurements
Cells were dissociated in TrypLE, washed and resuspended in PBS. Volume was measured on a Moxi Go II Coulter-principle cell sizer and flow cytometer. When mouse vs. human PSM cells were compared, only MSGN1-Venus+ cells were considered. Data was analyzed in FlowJo.

Mass and Density Measurements -Suspended Microchannel Resonator
Mouse and human PSM cells were dissociated in TrypLE and MSGN1-Venus+ cells were sorted. Cells were then counted and resuspended in DMEM/F12, 1% ITS, 5% FBS with 6 μM Chir 99021, 20ng/ml bFGF and 0.5 μM LDN193189 at a concentration of 3x10 5 cells/ml. The cells were kept on ice and their total mass and density were measured using the suspended microchannel resonator (SMR) according a previously developed fluid-switching method [14]. The SMR is a vibrating cantilever with a fluidic channel inside. In the absence of cells, the vibration frequency of the cantilever is proportional to the density of the fluid flowing through the cantilever. As a cell is flown through the cantilever, the vibration frequency of the cantilever changes proportionally to the buoyant mass of the cell. Following the measurements of media density and cell's buoyant mass in normal media, the cell is immersed in culture media that has been made denser by the addition of 35% OptiPrep (Sigma-Aldrich cat. no. D1556-250ML). The cell is then flow back through the cantilever in the high-density media to obtain a second set of buoyant mass and media density measurements for the cell. The total mass and density of a cell is calculated by comparing these two sets of measurements according to the equation = ( − ), where is buoyant mass, is volume, is density of the cell, and is density of the fluid. After each cell is measured, the cell is flushed out of the SMR and fresh media and cells are flushed in to the SMR. All measurements were carried out at +4°C. Calibration of the SMR frequency response to a cell was done using NIST traceable 10.12 μm diameter polystyrene beads (Thermo Scientific, Duke Standard beads, cat. no. 4210A), and the calibration of SMR baseline frequency to fluid density was done using NaCl solutions of known density [15].

Small Molecule Inhibitor Treatments
Human PSM cells were differentiated using the serum-free protocol [2] and treated chronically with the relevant inhibitors or supplements as indicated on Extended Data Table 1 starting on day 2 of differentiation. Timelapse imaging started approximately 2 hours after the inhibitor addition. In the case of aphidicolin, cells were pre-treated for 24 hours before imaging. All other assays (Seahorse, proteasome activity, NAD+/NADH, Peredox fluorescence, etc.) were performed after 16-24 hours of treatment to observe chronic effects.