A concerted metabolic shift in early forebrain alters the CSF proteome and depends on MYC downregulation for mitochondrial maturation

Neural tube closure (neurulation) is a fundamental initial step in brain development. Prior to the massive cellular rearrangements involved in neural tube closure (Massarwa and Niswander, 2013), the neural plate comprises multipotent neural stem cells, including forebrain neurectodermal precursor cells. These neurectodermal precursors become progressively lineage restricted, first as neuroepithelial cells and then radial glial cells, and eventually give rise to all neurons and glia in the forebrain (Bjornsson et al., 2015). The initial steps of brain development associated with neurulation are accompanied by parallel and massive changes in the greater embryonic environment. Prior to neurulation, chorioallantoic branching and placental development have not been completed, contributing to a hypoxic external environment that is now known to be necessary for proper CNS development (Clough and Whittingham, 1983; Scully et al., 2016). Concurrently, the fluid that is in direct contact with the developing CNS changes from amniotic fluid (AF) to cerebrospinal fluid (CSF). These early brain fluids vary in composition and instructive capacities (Chau et al., 2015; Lehtinen et al., 2011). Although advances in understanding the complex cellular, morphological and environmental processes surrounding neurulation have been made (Greene and Copp, 2014; Li et al., 2018; Massarwa and Niswander, 2013; Miyazawa et al., 2017; Wallingford et al., 2013; Wilde et al., 2014), still relatively little is known about the molecular processes instructing this foundational stage of brain formation.

The developmental time period surrounding neurulation is accompanied by changes in precursor cell transcriptomes and, in turn, organelle landscape (Chau et al., 2018). These changes include decreased MYC expression, mTOR pathway signaling and ribosome biogenesis by E10.5 in mouse (Chau et al., 2018). In addition to these concerted changes, this developmental time involves increased vascularization and oxygenation of the brain (Vasudevan et al., 2008; Walls et al., 2008). Transcriptional analyses also show that genes associated with mitochondrial function are enriched at E8.5 (Chau et al., 2018). Older studies using whole embryos identified changes in mitochondrial structure in cranial neural folds during the transition from gastrulation to neurulation. These changes include shifting from mitochondria with hallmarks of hypoxia, including swollen cristae and dense matrix (Morriss and New, 1979), which are associated with glycolysis, to mitochondria with flattened cristae and lucid matrices that are more characteristic of high levels of oxidative phosphorylation. More recently, metabolomic analyses point to changes in metabolic intermediates and ATP content in the whole embryo and along the neural tube (Miyazawa et al., 2017, 2018; Scully et al., 2016; Yamaguchi et al., 2017). Metabolic shifts can herald changes in cell fate, and mis-steps in metabolic pathways are implicated in a growing number of neurodevelopmental conditions (Fernandez et al., 2019; Salbaum and Kappen, 2010; Vallée and Vallée, 2018; Yamaguchi et al., 2017). Nevertheless, very little is known regarding the fundamental process of mitochondrial maturation and progenitor cell metabolism around the time of neural tube closure, when the mammalian forebrain first forms.

In light of these data, we have investigated metabolic changes during the early specification of the mammalian forebrain. Transmission electron microscopy (EM) and mitochondrial gene expression analyses revealed striking changes in mitochondrial morphology and activity. We then employed bioinformatics and molecular biology techniques to define a transcriptionally regulated metabolic transition in forebrain precursor cells from glycolysis at E8.5 to oxidative phosphorylation at E10.5. We probed intrinsic metabolic functional capability of E8.5 versus E10.5 and E12.5 neural progenitor cells, and uncovered large-scale metabolic shifts in glucose metabolism, indicating that the metabolism of E8.5 progenitors had the capacity to rely heavily on glycolysis, while E10.5 cells used more oxidative phosphorylation and E12.5 progenitors employed even more oxidative phosphorylation than either of the earlier stages. Finally, using transgenic mice to prolong MYC expression in neural progenitors beyond its natural downregulation before E10.5, we found that the massive mitochondrial morphology shift is dependent on MYC downregulation. Taken together; these data demonstrate that the forebrain neurectoderm undergoes metabolic switching during this early crucial stage of forebrain specification.

We previously observed major differences in transcriptional signatures of precursor cells from mouse forebrain at embryonic days E8.5 and E10.5 (before and after neural tube closure), including decreased MYC expression, mTOR pathway signaling and ribosome biogenesis at E10.5 (Chau et al., 2018). These changes also suggested an accompanying shift in the organelle landscape and nutrient sensing in E10.5 neuroepithelium (see figure 2 in Chau et al., 2018). Prompted by these findings and together with previous reports of mitochondrial and metabolite changes in the whole embryo during the transition from gastrulation to neurulation (from ∼E7-E9 (Miyazawa et al., 2017, 2018), we investigated structural changes in mitochondria before and after neurulation. We performed transmission electron microscopy (EM) on neuroepithelial cells at E8.5 and E10.5, and characterized mitochondrial morphology (Fig. 1A,B) based on their classic features (Suldina et al., 2018). We observed a shift in mitochondrial morphology from predominantly displaying hallmarks of hypoxia, glycolysis and immaturity, including condensed matrices with swollen cristae at E8.5, to more traditional mature mitochondria with lucid matrices and flat cristae at E10.5 (Fig. 1B). We also observed a trend in differential mitochondrial-associated membranes (MAMs) that was characterized by a shift toward fewer endoplasmic reticulum-mitochondrial contacts from E8.5 to E10.5 (Fig. 1A,C). This finding is consistent with reports of metabolic changes associated with an excess cellular requirement for glucose and hypoxia-induced mitochondrial fragmentation (Rieusset, 2018; Wu et al., 2016).

In other systems, including in cancers that have altered oxygenation of tumor tissue, distinct mitochondrial morphologies are associated with either hypoxic environments (immature mitochondria) or normoxic environments (mature mitochondria) (García-Prat et al., 2017; Gasparre et al., 2011; Suldina et al., 2018; Vander Heiden et al., 2009). Hif1α is a master regulator of oxygen homeostasis that, in other systems, controls angiogenesis, erythropoiesis and glycolysis through transcriptional activation of target genes harboring an HRE (hypoxia response element) under hypoxic conditions (0.1-5.0% O₂). Accordingly, we observed increased Hif1α protein expression at E8.5, but no change in either the transcript levels or hydroxylation status (Fig. 1D,E), suggesting that Hif1α protein levels are stabilized at the translation or protein degradation levels, and not via differential transcription or post-translational modification. These elevated levels of Hif1α protein are in agreement with previous reports that the early embryo prior to neural tube closure and choriallantoic branching is hypoxic (Dunwoodie, 2009; Miyazawa et al., 2017). The vascular system is also maturing during this developmental stage, as periventricular vessels first reach the dorsal telencephalon at E10.5 (Vasudevan et al., 2008; Walls et al., 2008). Consistent with these biological processes in the developing embryo, the E10.5 forebrain had increased levels of mitochondrial components associated with oxidative phosphorylation (e.g. citrate synthase, CS) as well as markers of non-hypoxic mitochondria [e.g. mitochondrial dynamism protein mitofusin 2 (Mfn2)] (Fig. 1F). At the cellular level, each E8.5 neurectodermal cell had fewer CS⁺ mitochondria than did the cells of the E10.5 neuroepithelium (Fig. 1G,H). Together, these data demonstrate widespread morphological and molecular changes in mitochondria of precursor cells during critical and early stages of forebrain development.

Further investigation of changing metabolic components in forebrain precursors uncovered robust downregulation of the molecular machinery that governs glucose metabolism from E8.5 to E10.5. Glucose metabolism is a complex process, interconnected with multiple aspects of cellular function (Fig. 2A, reviewed by Khacho and Slack, 2017, 2018). It can therefore be modulated at multiple levels beginning from the initial process of glucose uptake by glucose transporters (e.g. Glut1 and Glut3) to the presence of glycolytic, gluconeogenic and oxidative phosphorylation machinery. Even if all the molecular machinery is present to perform oxidative phosphorylation and/or glycolysis, metabolism can quickly adapt to cellular needs with key switching components such as Ldh, Ucp2, Mtch1 and Atpif1, which alter the ratio of cellular materials that are shuttled through traditional oxidative phosphorylation versus other pathways, such as glycolysis and the pentose phosphate pathway (Cliff et al., 2017; Levine and Puzio-Kuter, 2010). Gene expression analysis from RNA-sequencing data in E8.5 neurectoderm and E10.5 neuroepithelium (Chau et al., 2018) revealed that members of multiple metabolic pathways were enriched at E8.5 and E10.5 at the mRNA level, including general glucose use, metabolic shifting, glycolysis, gluconeogenesis and the pentose phosphate pathway proteins (Fig. 2B, Fig. S1A). Key players in these processes were validated via qPCR, including glycolytic enzymes phosphoglycerate kinase 1 (Pgk1), lactate dehydrogenase a (Ldha) and aldolase a (Aldoa), and the glucose transporters Slc2a1 (Glut1) and Slc1a3 (Glut3) (Fig. 2C, Fig. S1B). Protein levels reflected a similar trend in increased glucose uptake and glycolysis components at E8.5, evidenced by high levels of Glut1 and Ldha, while increased oxidative phosphorylation was suggested at E10.5 by increased expression of ATP synthase/complex V (Fig. 2D,E). Together, these data demonstrate that forebrain precursor cells before and after neural tube closure house distinct metabolic machineries.

In addition to providing instructive signals for developing forebrain progenitors (Chau et al., 2015; Lehtinen et al., 2011), the early AF and CSF proteomes provide a biomarker signature of forebrain development by matching their fluid proteomes to the transcriptional changes of the forebrain, e.g. as was shown for changes associated with ribosome biogenesis (Chau et al., 2018). Consistent with these findings, nearly all metabolic proteins detected and enriched in the E8.5 AF proteome (Fig. 3A,B) matched the concurrent transcriptional changes in the E8.5 and E10.5 transcriptomes (Fig. 2B). Not only were metabolic proteins less abundant in E10.5 CSF, but many were no longer detected therein (Fig. 3A,B) (Chau et al., 2018, 2015). The E8.5 AF and E10.5 AF were even more divergent (Fig. 3C,D). Together, these data demonstrate that the changes in the AF and CSF proteomes during early forebrain development match the downregulation of metabolic machinery in the adjacent neuroepithelium (Fig. 3E,F). Collectively, these data suggest that early forebrain precursors harness different metabolic pathways for ATP synthesis and production of anabolic intermediates to meet changing cellular energy and substrate demands.

As metabolic processes can be regulated at multiple levels, including transcription, translation, post-translational modification and based on external cues such as substrate availability, we directly investigated whether the capacity for metabolic activity differs between E8.5 and E10.5 neuroepithelial cells. When glycolysis-derived lactate is exported from the cell, protons are also exported; therefore, extracellular acidification rates (ECARs) indicate glycolysis (Fig. 4A, Fig. S1C). Using the Agilent XF Seahorse technology, we simultaneously measured ECAR and mitochondrial oxygen consumption rate (OCR) to generate a live-cell bioenergetic profile. Unlike endpoint assays such as lactate detection, this test allowed us to detect changes in metabolic function in real time, providing a window on the earliest events of metabolic switching. The ECAR was then used to calculate glycolytic parameters in primary cultures of E8.5 and E10.5 forebrain neuroepithelium (Fig. 4A, Fig. S1D). After microdissection of the E8.5 and E10.5 neuroepithelium, followed by dissociation and plating in glucose-free media (Fig. 4B), we measured glucose use with the Seahorse XF Glycolysis Stress Test. This test is designed to measure the capacity of the glycolytic pathway in cell populations by driving them toward glycolysis and measuring their increase in glycolytic activity to meet bioenergetic demand (Fig. 4A). The greater ECAR increase after glucose addition indicated that E8.5 neuroepithelial progenitors were more dependent on glucose than were E10.5 neuroepithelial progenitors (Fig. 4C-E, Fig. S1C,D). After blocking mitochondrial ATP synthase and, therefore, also the electron transport chain, with oligomycin, E8.5 progenitors still maintained higher ECAR, indicating a larger capacity to buffer against compromised mitochondrial function via glycolysis use, i.e. glycolytic capacity (Fig. 4C-E, Fig. S1D). The baseline OCR was higher in E8.5 progenitors than in E10.5 progenitors, suggesting the possibility of an early intracellular energy source (Fig. S1C); however, both E8.5 and E10.5 progenitors showed a boost in the OCR after glucose addition and a decrease in OCR after oligomycin, indicating that both cell types have functional mitochondria that employ oxidative phosphorylation (Fig. S1C). It should be noted that, owing to glycolytic rates and cellular availably, E8.5 and E10.5 cells were plated at different densities for comparison. Cellular density, even when normalized by cell number leads to somewhat different rates (Fig. S1D). The results remain consistent with those presented, and as these data obtained from cultured cells show a snapshot in time, it may be that the native in vivo differences are even greater than these in vitro analyses are able to show.

To investigate the proportion of acidification of each cell type due to glycolysis, rather than to other processes (including increased glutamate levels in excitatory neuron progenitors), we employed the Seahorse XF Glycolytic Rate Assay. This assay takes into account the ECAR, OCR and buffer capacity of the media, and uses a standard contribution rate of CO₂ to acidification that is determined experimentally in multiple cell lines and primary cell cultures, including neural progenitors. Therefore, the Glycolytic Rate Assay allowed calculation of the proton efflux rate (PER) in the presence of glucose followed by inhibition of the electron transport chain with rotenone and antimycin A (Rot/AA), which inhibit NADH ubiquitone reductase and cytochrome c reductase, respectively. This was followed by inhibition of all glycolysis with the glucose analogue 2-deoxy-d-glucose (2-DG) (Fig. 4F). We determined that the proportion of proton efflux due to glycolysis was highest in E8.5 progenitors, reduced in E10.5 progenitors and lowest in E12.5 progenitors (Fig. 4F,G), the developmental stage when changes in mitochondrial function have been shown to induce the maturation factor NRF2 (Khacho et al., 2016; Khacho and Slack, 2017). Additionally, the ratio of mitochondrial oxygen consumption to glycolytic proton efflux drastically increased as the neuroepithelium matured from E8 to E12.5 (Fig. 4H). Taken together, these data functionally demonstrate that E8.5 forebrain precursors have more glycolytic capacity and are more dependent on extracellular glucose for their metabolism than E10.5 forebrain neuroepithelial progenitors. Conversely, E10.5 and E12.5 progenitors are more heavily reliant on mitochondrial oxidative phosphorylation processes for their metabolism. These data suggest key roles for glucose availability and regulation during this early crucial stage of forebrain development.

Developmental downregulation of MYC is crucial for controlling a shift in translation machinery and ribosome biogenesis during neural tube closure between E8.5 and E10.5 in mice (Chau et al., 2018). Because MYC targets are also involved in many mitochondrial and hypoxia-related processes (Doe et al., 2012; Goetzman and Prochownik, 2018; Morrish and Hockenbery, 2014), we investigated mitochondria after experimentally extending the time and level of MYC expression in neuroepithelial progenitors by expressing a human MYC transgene in neuroepithelial progenitors by crossing StopFLMYC mice (Calado et al., 2012) with Foxg1-Cre mice (Hébert and McConnell, 2000). First, we performed transmission EM on neuroepithelial cells at E12.5 and characterized mitochondrial morphology (Fig. 5A,B, as in Fig. 1A,B) based on their classic features (Suldina et al., 2018). In contrast to traditional, more mature mitochondria with lucid matrices and thin cristae, E12.5 forebrain progenitors with prolonged MYC expression retained the striking mitochondrial morphology of immature glycolytic condensed matrices with swollen cristae (Fig. 5A,B), approximating E8.5 forebrain precursors (Fig. 1A,B). However, this morphological change was not reflected in MAMs (n=4 mice, percentage of mitochondria with EM contact±s.e.m.: wild type, 20.2%±0.8; MYC-OE, 21.9%±1.6; P=0.63, t-test) or expression of canonical mitochondrial dynamics pathway-associated proteins such as OPA1 or PGC1α (Fig. S2C). While MYC expression regulated a robust change in mitochondrial morphology, it was insufficient to drive a preference for glycolytic metabolism over oxidative phosphorylation, as demonstrated by unchanged expression of glycolytic genes (Fig. 5C, Fig. S2A), although some genes in other metabolic pathways did change (Fig. 5D, Fig. S2B). Functionally, there was no change in the PER balance, OCR or ECAR in progenitors under the conditions tested (Fig. 5E, Fig. S2D). This observation could potentially result from combinatorial effects due to the potentially limited accessibility of MYC to chromatin following neurulation and/or the oxygenated/normoxic environment of the E12.5 embryo. Taken together, our findings support a model in which MYC expression in neuroepithelial progenitor cells regulates mitochondrial maturation in the early developing forebrain.

Elucidating the metabolic contributions at the inception of the mammalian forebrain is paramount to understanding neural development. Our study reveals concerted changes in metabolic machinery during early specification of the mammalian forebrain. This work (1) uncovers striking changes in the mitochondrial landscape between neuroectodermal progenitors at E8.5 and neuroepithelial progenitors at E10.5; (2) shows that enhanced glycolysis and anabolic pathways serve as transcriptional signatures of early forebrain progenitor cells at E8.5; (3) reveals that the changing proteomes of the AF and CSF provide a biomarker signature that matches the concurrent normal metabolic changes in the adjacent brain; (4) functionally demonstrates that E8.5 progenitors are more glycolytic than their E10.5 progeny; and (5) demonstrates prolonged MYC expression as sufficient to maintain mitochondria in a state with swollen cristae and condensed matrices, one part of the metabolic machinery present in proliferative, glycolytic or hypoxic cells. We conclude that neural stem cells dynamically regulate their metabolism to meet their changing energy and anabolic needs.

Changes in organelle content and maturity are a common theme in fate specification. We have previously demonstrated downregulation of ribosome biogenesis that corresponds with accompanying changes in protein synthesis in neural progenitors during the same timeframe that we investigate here (E8.5-E10.5; Chau et al., 2018). Additionally, tightly regulated proteostatic mechanisms are required for maintaining neural stem cell pluripotency (García-Prat et al., 2017; Vilchez et al., 2012, 2014; You et al., 2015), and during human neuronal differentiation, widespread remodeling of translational machinery occurs via differential mTOR signaling and differential translation of 3′ UTRs (Blair et al., 2017). Although the inputs to proteostasis are diverse and involve multiple organelles and molecular machines, the endoplasmic reticulum (ER) plays a key role in proteostasis via the ER unfolded protein response and through its roles regulating intracellular calcium (García-Prat et al., 2017; Lee et al., 2017) and metabolism (García-Prat et al., 2017). Both the ER unfolded protein response and calcium regulation are achieved through coordination between the ER and the mitochondria. Here, we directly implicate mitochondria and metabolism as changing concurrently with linked protostatic shifts in the forebrain.

At later stages of corticogenesis, mitochondrial dynamics regulate stem cell fate decisions, mediated by reactive oxygen species (ROS) that directly stimulate NRF2 signaling, which suppresses self-renewal and promotes differentiation (Khacho et al., 2016; Khacho and Slack, 2017). Our new data fit into this context, showing that this process is not stagnate but, rather, is initiated prior to cortical progenitor specification, and continues into later stages of brain development and maturation. For example, cortical neuronal differentiation in vitro and postnatally in vivo results in changes in mitochondrial mass, increased oxidative phosphorylation and decreased glycolysis dependent on PI3K-Akt-mTOR signaling (Agostini et al., 2016); in mature neurons, mitochondrial dynamics in presynaptic boutons are crucial for maintaining neuronal health and avoiding excitatory neurotoxicity via mitochondrial regulation of intracellular calcium clearance to control neurotransmitter release (Kwon et al., 2016). Thus, close coordination between proteostasis, intracellular calcium and metabolism govern neural cell fate and long-term neuronal health. Our data bridge the gap between stem cell and cortical progenitor by demonstrating a transition in mitochondrial morphology and gene expression as the neurectoderm develops and becomes the more lineage-restricted neuroepithelium.

The overall embryonic environment changes dramatically from E8.5 to E10.5, with the pre-neurulation embryo being hypoxic compared with the post-neurulation embryo (Clough and Whittingham, 1983; Scully et al., 2016). Studies have shown that this hypoxia is necessary for proper development of neural folds and neural crest (Morriss and New, 1979; Scully et al., 2016). Hypoxia can also alter mitochondrial dynamics by inducing mitochondrial fission and mitophagy, which is dependent on mitochondrial-ER interactions (Wu et al., 2016). More generally, in mature cells that normally rely on oxidative phosphorylation, hypoxia alone or in concert with gene expression changes can induce metabolic switching, which is associated with transferring metabolism to glycolysis and away from oxidative phosphorylation (Cliff et al., 2017; Goetzman and Prochownik, 2018; Morrish and Hockenbery, 2014). Recently, additional metabolic developmental studies in the context of neural tube closure have corroborated older studies (Morriss and New, 1979) showing that glucose metabolism along the neural tube is sensitive to extracellular oxygen levels (Miyazawa et al., 2018), and is also at least partially under the genetic control of Lin28a (Miyazawa et al., 2017; Yamaguchi et al., 2017). Our original observation of changed organelle landscape (Chau et al., 2018), followed by unbiased transcriptome analyses performed in microdissected developing forebrain provide a ‘parts list’ of metabolic machinery that, when paired with Seahorse data, demonstrate a functional shift in glucose use in the developing forebrain. Collectively, this active area of research is beginning to disentangle the extent to which early metabolism is flexible in response to specific external signals compared with simply executing tightly regimented genetically programmed events.

In the disease setting, cancer cells usually remain glycolytic, even in the presence of a high oxygen concentration (sometimes termed ‘oxidative glycolysis’). Key to these findings, the products of glycolysis itself, such as lactate, can induce a hypoxic response independently of the canonical Hif1α-dependent pathways using NDRG3 stabilization and Raf/Erk signaling (Lee et al., 2015). Thus, our data showing that the metabolic state of the developing forebrain is reflected in the associated AF/CSF raises the hypotheses that these surrounding fluids may provide biomarkers of normal developmental signatures or could even act as synchronizing feedback to the entire CNS about the metabolic state of the developing embryo.

From cancer research, it is known that transformed cells can alter their metabolism and mitochondrial morphology drastically to adapt to high proliferation rates and changing anabolic and energetic demands (Vander Heiden et al., 2009). Classically, Otto Warburg observed in 1924 that cancer cells metabolize glucose via glycolysis, in a manner that is distinct from that of cells in differentiated tissues (Warburg et al., 1927). Subsequent studies show that far from reflecting a mitochondrial defect, this metabolic switch adapts to proliferation by supporting the higher demand for equivalents of carbon and NADH than for ATP, in order to provide the raw components needed for nucleotide and amino acid biosynthesis (Fantin et al., 2006; Vander Heiden et al., 2009). Pluripotent stem cells (induced and embryonic) also rely heavily on glycolysis (Cliff et al., 2017; Elstrom et al., 2004; Kondoh et al., 2007) and display similar mitochondrial morphologies to those that we observed in neurectoderm at E8.5 (Suldina et al., 2018). Comparing similarities between rapidly proliferating cancer cells and developmentally regulated proliferative progenitors reveals the common theme that cellular metabolism, through glycolysis, is adaptive for the demands of proliferative cells. Some caveats do exist from using the ECAR, PER and OCR readouts as a proxy for full metabolic analysis, including the possibility of NADH contribution from the cytosol that results in O₂ consumption without equivalent acidification, CO₂ consumption from carboxylation reactions and lack of direct measurements of reactive species that may alter ECAR and OCR. However, the real-time readout enables an overview of progenitor metabolic capacity and glucose dependence. Here, we demonstrate that the shift from glycolysis to oxidative phosphorylation is a tightly regulated transition between the E8.5 neurectodermal progenitors versus the more fate-restricted E10.5 neuroepithelial progenitors, and continues to be regulated as differentiation progresses to E12.5.

MYC is well-known to be associated with regulating the pathways necessary for maintaining proliferative progenitors (Kerosuo and Bronner, 2016). While upstream controls, including nutrient-sensing pathways (mTOR/PI3K, LKB1; Agostini et al., 2016; Buzzai et al., 2005; Hardie, 2007), have been shown to contribute to the glycolytic switching, MYC is also poised to be one of the master regulators of this process. Transcriptional targets of MYC include metabolite transporters and enzymes required for glucose metabolism, glutaminolysis and biosynthesis, and inhibiting these downstream activities suppresses growth of MYC-driven tumors (Altman et al., 2015). Importantly, MYC targets include a large number of genes associated with mitochondrial morphology and function (Morrish and Hockenbery, 2014). In differentiating stem cells, MYC pathways are crucial for developmental processes in the untransformed setting to maintain glycolysis in undifferentiated cells; as MYCN and MYC are downregulated in endoderm and mesoderm, metabolic switching shifts more committed cells lineages to oxidative phosphorylation (Cliff et al., 2017; Goetzman and Prochownik, 2018). In fact, overexpression of MYCN in endoderm and mesodermal lineages is sufficient to re-induce glycolytic metabolism (Cliff et al., 2017). Strikingly, this metabolic switching does not occur as stem cells are allocated into the early ectodermal lineage (Cliff et al., 2017). Because MYC is endogenously low in the E10.5 neuroepithelium, our MYC-overexpression data support the hypothesis that ectodermal lineages maintain high levels of glycolysis for longer, and also that MYC downregulation and metabolic shifting occur later as the neuroectoderm commits to neuroepithelial lineage between E8.5 and E10.5.

Although we report here that MYC must be downregulated for appropriate development of mitochondrial morphology in E12.5 neuroepithelium, the ex vivo glycolysis assay and the RNA-Seq of the Nestin-driven MYC overexpression system show that MYC may not be the sole regulator of glycolysis. These observations must be interpreted in the context of the development of the entire embryo, and it may be that MYC does not have access to its targets due to changes in chromatin landscape or other molecular checks and balances that are poorly understood that may prevent comprehensive evaluation of mechanisms at play via these assays. However, both hypoxia and MYC can stabilize Hif1α and promote glycolysis and PI3K-AKT, which are also involved in glycolytic switching. In addition to influencing metabolism, MYC overexpression can induce over-proliferation phenotypes and tumors in brain epithelium (Shannon et al., 2018). Although the pre-neurulation embryo is hypoxic compared with the post-neurulation embryo (Clough and Whittingham, 1983; Scully et al., 2016), glycolytic by-products can also induce a hypoxic response independently of Hif1α (Lee et al., 2015). Because the time period of MYC downregulation corresponds with the time period of shift from hypoxia to normoxia in the embryos, it is likely that while hypoxia itself can induce glycolytic responses, the concurrent high MYC levels act as an only partially redundant or reinforcing pathway to maintain the crucial hypoxic and glycolytic state of these early progenitors (Cliff et al., 2017). In conclusion, while MYC itself stabilizes the hypoxic response in some systems (Doe et al., 2012; Goetzman and Prochownik, 2018), in the developing neuroepithelium, MYC is sufficient to dramatically regulate mitochondrial morphology, but more than MYC alone is required to maintain glycolytic metabolism.

The metabolic capacity of neural progenitors is of particular importance to the natural fluid environment of the developing brain. The ex vivo Seahorse analyses were performed in HEPES-buffered pH 7.4 minimal media to allow the calculations and to investigate the intrinsic potential of the progenitors. However, AF and CSF have buffering capacities that are distinct from each other and from the media. While the in vivo buffering capacity of CSF (pH 7.33) is better than that of blood due to high bicarbonate levels (Husted and Reed, 1977; Lee et al., 1969), early CSF is likely not buffered with bicarbonate, as the choroid plexus (ChP), which expresses Cl⁻/HCO₃⁻ transporters, has not yet developed and matured (Liddelow et al., 2012). However, the embryonic CSF has a higher protein concentration than adult CSF, which can provide additional buffering capacity. AF, on the other hand, has virtually no buffering capacity in chick at the analogous pre-neurulation stages in development (Randles and Romanoff, 1954). Taken together with our results, these data suggest that the acidic metabolic byproducts of glycolysis, particularly an early bolus of glucose, could severely alter the pH of the fluid environment of the developing brain. Such unbuffered acidosis can impact progenitor proliferation, survival, epigenetics, physiology, motility and even metabolic reprogramming (White et al., 2017). All of these pH-dependent effects in combination with intrinsic roles of altered metabolism might render the neurectoderm particularly sensitive to glucose availability.

Mis-steps in development increase vulnerability to neurological disease. Our data open up a pathway for understanding how early forebrain metabolism critically contributes to more mature brain health. One relevant example is that of maternal diabetes, which has been shown to be a risk factor for neural tube closure defects (Kucera, 1971; Martínez-Frías, 1994; Martínez-Frías et al., 1998). While glucose availability is not a usual limiting factor for glycolytic switching (Vander Heiden et al., 2009), high maternal glucose levels are associated with coordinated gene expression changes in the developing embryo (Pavlinkova et al., 2009; Salbaum and Kappen, 2010). Strikingly, the gene expression changes that result from high maternal blood glucose include enrichment of motifs for transcription factors that regulate response to oxidative stress or hypoxia in the E10.5 embryo (Salbaum and Kappen, 2010). These associations, coupled with our data supporting a massive change in glucose metabolism and potential for acidosis at times adjacent to neural tube closure, suggest a relationship between natural and diseased processes, whereby metabolic misregulation increases vulnerability for neurodevelopmental disease and, therefore, indicate a potential window of intervention.

Taken together, our findings indicate large coordinated changes in metabolic machinery in forebrain precursors during early brain development concurrent with neural tube closure, MYC downregulation and the embryonic exit from hypoxia. Such a tight regulation of glycolytic metabolism at this time informs the relationship between neural precursor fate restriction and glucose use, with the potential to guide future studies linking maternal metabolic dysregulation (maternal diabetes) and neural tube closure defects.

Timed pregnant CD1 dams (Mus musculus) were obtained from Charles River Laboratories. StopFLMYC (Calado et al., 2012) mice (JAX: 020458) were maintained in a C57BL/6J genetic background and crossed with Foxg1-cre line (JAX: 004337) to generate MYC-OE mice, in which the human MYC transgene is selectively expressed in neural progenitor cells. Owing to phenotypes associated with Foxg1-cre^+/− mice (Hébert and McConnell, 2000), Foxg1-cre^+/−; StopFLMYC^+/− served as MYC-OE mice and Foxg1-cre^+/−; StopFLMYC^+/+ were used as controls. All analyses were carried out using male and female mice. All animal experimentation was carried out under protocols approved by the IACUC of Boston Children's Hospital.

All analyses were performed using genes with FPKM>1, which we considered as the threshold of expression, using previously published data (Chau et al., 2018). Functional annotation clustering was performed using DAVID v6.7 (david.ncifcrf.gov/home.jsp; Huang et al., 2009). Gene set enrichment analysis was performed using GSEA v3 (Subramanian et al., 2005); gene sets were obtained from the Broad Institute Molecular Signatures Database (software.broadinstitute.org/gsea/msigdb).

Samples were fixed in 4% paraformaldehyde (PFA). For cryosectioning, samples were incubated in the following series of solutions: 10% sucrose, 20% sucrose, 30% sucrose, a 1:1 mixture of 30% sucrose and OCT (overnight), and OCT (1 h). Samples were frozen in OCT.

Cryosections were blocked and permeabilized (0.3% Triton X-100 in PBS, 5% serum), incubated with primary antibodies overnight and secondary antibodies for 2 h. Sections were counterstained with Hoechst 33342 (Invitrogen, H3570; 1:10,000) and mounted using Fluoromount-G (SouthernBiotech). The following primary antibodies were used: rabbit anti-CS (Cell Signaling, 14309; 1:1000), mouse anti-Glut1 (Abcam, ab40084; 1:1000). Secondary antibodies were selected from the Alexa series (Invitrogen, 1:500).

Tissues were homogenized in RIPA buffer supplemented with protease and phosphatase inhibitors. Protein concentration was determined using a BCA assay (Thermo Fisher Scientific, 23227). Samples were denatured in 2% SDS by heating at 37°C for 5 min. Equal amounts of proteins were loaded and separated by electrophoresis in a 4-15% gradient polyacrylamide gel (BioRad, 1653320), transferred to a polyvinylidene difluoride (PVDF) membrane (250 mA, 1.5 h, on ice), blocked in filtered 5% BSA or milk in TBST, incubated with primary antibodies overnight at 4°C followed by HRP-conjugated secondary antibodies (1:5000) for 1 h and visualized with ECL substrate. For hydroxylated protein analysis, the hydroxyl proteins were probed first, and then blots were stripped (Thermo Fisher Scientific, 21059) and reprobed for total proteins. The following primary antibodies were used: rabbit anti-CS (Cell Signaling, 14309; 1:1000), mouse anti-Glut1 (Abcam, ab40084; 1:1000), rabbit anti-Hif1α (Cell Signaling, 3716S; 1:1000), rabbit anti-hydroxy Hif1α (Pro564) (Cell Signaling, 3434S; 1:1000), rabbit anti-mitofusin2 (Cell Signaling, 9482S; 1:1000), rabbit anti-vinculin (Cell Signaling, 13901; 1:1000), rabbit anti-Ldha (Cell Signaling, 2012S; 1:1000), mouse anti-PGC-1α (4C1.3 EMD Millipore, ST1202; 1:1000), mouse anti-OPA-1 (BD Biosciences, 612606; 1:1000) and total OXPHOS Rodent WB Antibody Cocktail (Abcam, ab110413; 1:250).

All tissue processing, sectioning and imaging was carried out at the Conventional Electron Microscopy Facility at Harvard Medical School. Forebrain tissues were fixed in 2.5% glutaraldehyde/2% paraformaldehyde in 0.1 M sodium cacodylate buffer (pH 7.4). They were then washed in 0.1 M cacodylate buffer and postfixed with 1% osmiumtetroxide (OsO₄)/1.5% potassiumferrocyanide (KFeCN₆) for 1 h, washed in water three times and incubated in 1% aqueous uranyl acetate for 1 h. This was followed by two washes in water and subsequent dehydration in grades of alcohol (10 min each of 50%, 70% and 90%, and 2×10 min in 100%). Samples were then incubated in propyleneoxide for 1 h and infiltrated overnight in a 1:1 mixture of propyleneoxide and TAAB Epon (Marivac Canada). The following day, the samples were embedded in TAAB Epon and polymerized at 60°C for 48 h. Ultrathin sections (about 80 nm) were cut using a Reichert Ultracut-S microtome, and picked up onto copper grids stained with lead citrate. Sections were examined using a JEOL 1200EX transmission electron microscope or a TecnaiG² Spirit BioTWIN. Images were recorded using an AMT 2k CCD camera.

Mitochondrial quantification was performed by hand using the parameters outlined by Suldina et al. (2018) by an author blinded to all conditions (age and embryo). Any mitochondria with unclear or out-of-plane cristae, were labeled ‘blurry’ and excluded. For each condition, analyses were performed across multiple individual embryos (n=3 for each age; n=4 for each genetic condition in MYC-OE experiments). From each embryo, 15-20 fields of view were imaged at 10,000× and each mitochondrion was analyzed individually for morphology metrics. Each different field of view represented a unique cell or cells, and fields of view were chosen within the most apical layer of cells. The analyses between E8.5 and E10.5 included a total of 946 mitochondria; the analysis of control and MYC-OE included a total of 2196 mitochondria.

Mitochondrial quantification was performed using Imaris (Bitplane). For one image per individual (n=3 at each age), CS⁺ mitochondrial number was calculated within a 63× field of view. The region of interest was created by hand tracing a 3D surface containing only mesenchyme, only neuroectoderm or only neuroepithelium, depending on the age. Mitochondria were then counted using the Imaris software with the ‘spots’ tool, with an estimated xy diameter of 0.8µm and an estimated z diameter of 1.99 µm, and with automatic background subtraction ‘on’ and quality above the automatic threshold. Nuclei were counted using the ‘spots’ tool, with an estimated xy diameter of 3.5 µm and an estimated z diameter of 4.5 µm, and with automatic background subtraction ‘on’ and quality above the automatic threshold. The number of mitochondria per cell was calculated by dividing the total number of mitochondria by the total number of nuclei.

Forebrain epithelium at E8.5, E10.5 and E12.5 was dissected as described previously (Chau et al., 2015). Each sample comprised forebrain epithelial tissues pooled across one litter. Briefly, embryos were removed from the uteri in HBSS, and forebrain tissues were collected and digested in 2.5% pancreatin/0.25% trypsin for 30 min (E8.5), 40 min (E10.5) or 20 min (E12.5). The digest was stopped with 10% horse serum and 0.01% DNase in HBSS. Neuroectodermal/neuroepithelial cells were separated from adjacent tissue using tungsten needles and dissociated in 1% trypsin and 0.01% DNase in Ca²⁺/Mg-free HBSS. Cells were washed in 10% FBS, 0.01% DNase in 1× DMEM, counted, then transferred to Seahorse XF Base Medium (Agilent, 102353-100) supplemented with 1 mM L-glutamine at pH 7.4

For mRNA expression analyses, whole brains at E8.5 and E10.5 were dissected and pooled across one litter for each experiment. RNA was isolated using the RecoverAll Total Nucleic Acid Isolation Kit (Ambion) following the manufacturer's specifications. Extracted RNA was quantified spectrophotometrically and 100 ng was reverse-transcribed into cDNA using the ImProm-II Reverse Transcription System (Promega) following manufacturer's specifications. RT-qPCRs were performed in duplicate using Taqman Gene Expression Assays and Taqman Gene Expression Master Mix (Applied Biosystems) with eukaryotic 18S as an internal control. Cycling was executed using the StepOnePlus Real-Time PCR System (Invitrogen) and analysis of relative gene expression was performed using the 2−ΔΔCT method (Livak and Schmittgen, 2001). Technical replicates were averaged for their cycling thresholds and further calculations were performed with those means.

E8.5, E10.5 and E12.5 cells were plated on Seahorse XFe96 cell culture microplates (Agilent, 101085-004) coated with Cell TAK (Corning) in Seahorse XF Base Medium (Agilent, 102353-100) supplemented with 1 mM L-glutamine and 1 mM pyruvate at pH7.4, centrifuged at 500 g for 1 min without break and incubated for 1 h at 37°C in a non-CO₂ incubator. Extracellular acidification rates (ECAR) and oxygen consumption rates (OCR) were measured via the glycolytic stress test assay (Agilent, 103020-100) using a Seahorse XFe96Analyzer (Agilent) following the manufacturer's protocols. Data were processed using Wave software (Agilent) and standardized to cell number. The glycolytic rate measurements were calculated as the difference in ECAR measurements before and after glucose injection, and glycolytic capacity measurements were calculated as the ECAR measurements following oligomycin treatment subtracted from the initial baseline measurements, as described by the manufacturer's protocol (Agilent, 103020-100). OCR rates were calculated as the difference in OCR measurements following glucose injection and the OCR measurements after oligomycin injection. The combined analysis includes three distinct experiments. The individual analyses were performed with pooled litters at each age and distributed into 6-12 wells depending on cell availability. Because of the detection limits of the system and the vastly different metabolic signatures, E8.5 progenitors were plated at 25,000 cells per well and E10.5 and E12.5 progenitors were plated at 50,000 cells per well. Data were normalized by cell number (per 1000 cells). The low levels of E10.5 glycolysis precluded extensive analysis of 25,000 cells/well and the low yield from E8.5 progenitors precluded sufficient technical replicates at 50,000 cells/well. However, cellular density, even when normalized by cell number leads to different rates (Fig. S1). While the results remain consistent with those presented, it may be that the native in vivo differences are likely even greater than these in vitro analyses are able to show.

For proton efflux rate (PER) calculations, E10.5 and E12.5 cells were plated at 50,000 cells per well and E8.5 cells were plated at 25,000 cells per well on Seahorse XFe96 cell culture microplates (Agilent, 101085-004) coated with Cell TAK (Corning), in Seahorse XF Base Medium pH=7.4 (Agilent, 102353-100) supplemented with 2 mM L-glutamine, 0.18% glucose and 1 mM pyruvate, centrifuged at 500 g for 1 min without break, and incubated for 1 h at 37°C in a non-CO₂ incubator. PER was measured via the glycolytic rate assay (Agilent, 103344-100) using a Seahorse XFe96Analyzer (Agilent) following the manufacturer's protocols. Data were processed using Wave software (Agilent) using the CO₂ contribution factor (CCF) of 0.61, as calculated for XFe96 by Agilent for over 20 cell lines, including neuronal progenitors, and standardized to cell number (per 1000 cells). When only one experiment is represented, each data point represents the average of multiple wells for one condition (with error bars representing s.e.m.); when multiple experiments are represented, each data point is the average of an entire experiment.

Biological replicates (n) were defined as samples from distinct individuals, analyzed either in the same experiment or within multiple experiments. Some samples were pooled across multiple litters so as to reduce inter-litter variability. Samples were allocated into groups based solely on the gestational age without respect to sex (both males and females were included). Statistical analyses were performed using Prism 7 or R. Outliers were excluded using ROUT method (Q=1%). Appropriate statistical tests were selected based on the distribution of data, homogeneity of variances and sample size. F tests or Bartlett's tests were used to assess homogeneity of variances between datasets. Parametric tests (t-test, ANOVA) were used only if data were normally distributed and variances were approximately equal. Otherwise, nonparametric alternatives were chosen. Data are presented as mean±s.d. If multiple measurements were taken from a single individual, data are presented as mean±s.e.m.. Please refer to figure legends for statistical tests used and sample size. P<0.05 were considered significant (*P≤0.05, **P≤0.01, ***P≤0.001, ****P≤0.0001).

We thank members of the Lehtinen and Fleming labs for helpful discussions, and M. Ericsson for experimental advice and assistance.