Heterogeneous populations of hypothalamic neurons orchestrate energy balance via the release of specific signatures of neuropeptides. However, how specific intracellular machinery controls peptidergic identities and function of individual hypothalamic neurons remains largely unknown. The transcription factor T-box 3 (Tbx3) is expressed in hypothalamic neurons sensing and governing energy status, whereas human TBX3 haploinsufficiency has been linked with obesity. Here, we demonstrate that loss of Tbx3 function in hypothalamic neurons causes weight gain and other metabolic disturbances by disrupting both the peptidergic identity and plasticity of Pomc/Cart and Agrp/Npy neurons. These alterations are observed after loss of Tbx3 in both immature hypothalamic neurons and terminally differentiated mouse neurons. We further establish the importance of Tbx3 for body weight regulation in Drosophila melanogaster and show that TBX3 is implicated in the differentiation of human embryonic stem cells into hypothalamic Pomc neurons. Our data indicate that Tbx3 directs the terminal specification of neurons as functional components of the melanocortin system and is required for maintaining their peptidergic identity. In summary, we report the discovery of a key mechanistic process underlying the functional heterogeneity of hypothalamic neurons governing body weight and systemic metabolism.
Energy-sensing neuronal populations of the hypothalamic arcuate nucleus (ARC), including proopiomelanocortin (Pomc)- and agouti-related protein (Agrp)-expressing neurons, release specific neuropeptides that control energy homeostasis by modulating appetite and energy expenditure. Dysregulated activity of these neurons, which constitute key components of the melanocortin system1, is causally linked with energy imbalance and obesity2,3,4. Considering the constantly changing input into these neurons throughout development and adult life, an intricate intracellular regulatory network must be in place to accommodate plasticity adjustments (as an adequate response to energy state) as well as maintenance of cell identity. Whether extrinsic signals can induce in vivo reprogramming of neuropeptidergic identity has not been resolved, partly because of the limited knowledge of the intracellular factors involved.
To identify genes implicated in the maintenance of ARC neuronal identity and energy-sensing function, we took advantage of cell-specific transcriptomic approaches that allow profiling of subpopulations of hypothalamic neurons under basal and metabolically stimulated conditions. We cross-referenced publicly available analysed datasets from phosphorylated ribosome profiling5, translating ribosome affinity purification (TRAP)-based sequencing of leptin-receptor-expressing neurons6, and single-cell sequencing7. We determined that the transcription factor termed Tbx3 is expressed with unique abundance in hypothalamic neuronal populations critically involved in energy balance regulation, including ghrelin- and leptin-responsive cells5,6, and that its expression is regulated by scheduled feeding5.
Although Tbx3 is known to influence proliferation8, fate commitment and differentiation9,10,11 of several non-neuronal cell types, its functional role during neuronal development or in post-mitotic neurons located in the CNS is currently uncharted. Intriguingly, this factor appears to be selectively expressed in the ARC in the adult murine hypothalamus12. Moreover, TBX3 mutations in humans have been described to cause ulnar–mammary syndrome (UMS), exhibiting hallmark symptoms theoretically consistent with ARC neuron dysfunction, including impaired puberty, deficiency in growth hormone production and obesity13,14.
Thus, we hypothesized that Tbx3 in ARC neurons may control neuronal identity and consequently be of critical relevance for systemic energy homeostasis. To test this hypothesis, we explored the functional role of Tbx3 in both mouse and human hypothalamic neurons and investigated whether loss of neuronal Tbx3 affects systemic energy homeostasis in mice and in Drosophila melanogaster.
We report that Tbx3 directs postnatal fate and is critical for defining the peptidergic identity of both immature and terminally differentiated mouse melanocortin neurons, a biological process essential for the regulation of energy balance.
Tbx3 expression profile in the CNS and pituitary
To characterize Tbx3 expression in the central nervous system (CNS), we generated a targeted knock-in mouse model in which the Venus reporter protein is expressed under the control of the Tbx3 locus (Tbx3-Venus mice) (Supplementary Fig. 1). Two areas of the brain displayed a detectable Venus signal: the ARC (Fig. 1a) and the nucleus of the solitary tract (NTS; Supplementary Fig. 1), both of which are important in the regulation of systemic metabolism15,16. This hypothalamic expression pattern was confirmed via qRT–PCR (Supplementary Fig. 1) and anti-Tbx3 immunohistochemistry (Supplementary Fig. 1), using an antibody validated in house with Tbx3-deficient embryos (Supplementary Fig. 1). All Venus-positive cells in the ARC and NTS of Tbx3-Venus mice coexpressed Tbx3, as assessed by immunohistochemistry (Supplementary Fig. 1), and the model was further validated via Southern blot analysis (Supplementary Fig. 1), thus underlining the quality of the newly developed transgenic model.
To further address the cell-specific expression profile of Tbx3, we performed bioinformatic-based reanalysis of a publicly available single-cell RNA sequencing (RNA-seq) dataset from the ARC7. Our analysis demonstrated overlap of Tbx3 with neurons expressing Pomc, Agrp, kisspeptin (Kiss) and somatostatin (Fig. 1b and Supplementary Fig. 1), in addition to overlapping with the transcriptional profile of tanycytes, the ‘gateway’ cells to the metabolic hypothalamus17 (Supplementary Fig. 1).
Neuroanatomical analysis in Tbx3-Venus mice demonstrated Tbx3 (Venus) expression in almost all ARC Pomc neurons (Fig. 1c) and NTS Pomc neurons (Supplementary Fig. 1), with a comparable pattern of expression from embryonic (embryonic day (E) 18.5) to postnatal life (postnatal day (P) 0, P4 and adults) (Fig. 1d), thus indicating that Tbx3 expression in Pomc neurons is switched on embryonically and maintained throughout adult life. As suggested by the analysis of the single-cell RNA-seq data from the cells from the arc-median eminence, a considerable fraction of Tbx3-positive cells do not express Pomc. Tbx3 transcripts have been observed within the pituitary gland18. We found that Tbx3 (Venus) expression was restricted to the posterior pituitary and that no signal was observed in Pomc-expressing cells of the anterior pituitary (Supplementary Fig. 1). Moreover, no signs of Tbx3 (Venus) expression were detected in glial fibrillary acidic protein (GFAP)-positive astrocytes (Supplementary Fig. 1) or in microglia (Iba1-positive glial cells) (Supplementary Fig. 1), whereas a substantial number of Tbx3 (Venus)-positive cells coexpressed the tanycyte and reactive astrocyte marker vimentin (Supplementary Fig. 1), in agreement with results from single-cell sequencing analysis (Supplementary Fig. 1).
Thus, within the CNS, Tbx3 is expressed in both neuronal and non-neuronal cells known to affect energy homeostasis.
Loss of Tbx3 in hypothalamic neurons promotes obesity
ARC neurons detect changes in energy status, via both direct and indirect sensing of circulating nutrients and hormones, and accordingly modulate their activity to maintain energy balance16. Overnight fasting significantly decreased hypothalamic Tbx3 mRNA levels in the ARC in C57BL/6J mice, whereas refeeding partially restored Tbx3 expression (Fig. 1e). This finding suggests that changes in hypothalamic Tbx3 levels are likely to be involved in the control of systemic metabolism. To test this notion, we used a viral-based approach to selectively ablate Tbx3 via Cre-LoxP recombination (adeno-associated virus (AAV)-Cre) from the mediobasal hypothalamus (MBH) of 12-week-old Tbx3loxP/loxP littermate mice (Supplementary Fig. 2).
AAV-Cre-treated mice developed pronounced obesity over the course of 7 weeks, with elevated cumulative food intake and higher fat mass relative to control mice (AAV-green fluorescent protein (GFP)-treated Tbx3loxP/loxP mice), whereas no difference was observed in lean mass (Fig. 1f–i). Indirect calorimetry did not reveal changes in hourly uncorrected energy expenditure (Fig. 1j), nor in the relationship between total uncorrected energy expenditure and body weight, as demonstrated by analysis of covariance (ANCOVA)19 (Fig. 1k). Although the average respiratory exchange ratio (RER) was not altered in AAV-Cre-treated mice, these mice displayed metabolic inflexibility relative to controls, as indicated by a flat RER with minimal diurnal fluctuations (Fig. 1l,m).
We next asked whether loss of function of Tbx3 selectively in either Agrp or Pomc neurons would recapitulate the obesity-prone phenotype observed in the MBH loss-of-function model. No difference in body weight, food intake, glucose tolerance, fat or lean mass was observed in littermate mice bearing a conditional deletion of Tbx3 in Agrp-expressing neurons (Agrp-Cre;Tbx3loxP/loxP) relative to controls (Fig. 2a–e). The quality of this previously validated20 transgenic model was confirmed by the presence of reduced Tbx3 mRNA levels in ARC homogenates (Supplementary Fig. 2), together with a specific decrease in Tbx3 expression within Npy-positive neurons (Supplementary Fig. 2).
In contrast, mice bearing Tbx3 deletion in Pomc-expressing cells21(Pomc-Cre;Tbx3loxP/loxP) displayed body weight higher than that of control littermates, independently from changes in food intake (Fig. 2f,g). They also had glucose intolerance (Fig. 2h) and increased fat and lean mass (Fig. 2I,j). Indirect calorimetry demonstrated similar hourly energy expenditure, in spite of higher body weight (Fig. 2k), and further revealed lower energy expenditure with respect to body weight in Pomc-Cre;Tbx3loxP/loxP mice than controls (Fig. 2l), thus suggesting that lower systemic energy dissipation may contribute to their obese phenotype. These mice also displayed a higher average RER (Fig. 2m,n), thereby indicating that lower lipid utilization might favour the increased adiposity of these mice. A significant decrease in Tbx3 mRNA levels was observed in the hypothalamus in Pomc-Cre;Tbx3loxP/loxP mice, whereas no changes in Tbx3 mRNA levels were detected in extra-hypothalamic sites expressing Pomc, including the pituitary and adrenals (Supplementary Fig. 2). This transgenic model was further validated via costaining between Tbx3 and a Cre-dependent membrane GFP reporter, an analysis that revealed blunted Tbx3 immunoreactivity in Cre-positive neurons of Pomc-Cre;Tbx3loxP/loxP mice relative to controls (Supplementary Fig. 2). Thus, the metabolic alterations observed in this model are attributable to the specific deletion of Tbx3 in Pomc neurons located in the CNS. Collectively, these data demonstrate that ablation of Tbx3 in ARC neurons has profound functional consequences on energy balance and that most of these metabolic alterations can be reproduced after specific deletion of this gene in Pomc-positive neurons located in the brain.
Loss of Tbx3 impairs the postnatal melanocortin system
Although Tbx3 is known to control the cell cycle and programming of highly proliferative stem cells and cancer cells9,10,11,22, its functional role in neurons has remained unexplored. To investigate possible biological mechanisms underlying the metabolic phenotypes observed, we performed Tbx3-focused RNA sequencing and proteomic analyses in hypothalamic tissue as well as in primary hypothalamic cultures. The effect of Tbx3 deletion on transcription in hypothalamic neurons was assessed by using primary neurons isolated from Tbx3loxP/loxP mice and infected with adenoviral (Ad) particles carrying the coding sequence for Cre recombinase (Ad-Cre) or GFP (Ad-GFP) as a control (Supplementary Fig. 3), an approach that effectively allows knockdown of Tbx3 (Supplementary Fig. 3) in the absence of cell toxicity (Supplementary Fig. 3). Because we had found the most important in vivo metabolic effects with Tbx3 deletion uniquely within Pomc-expressing cells, we performed RNA sequencing of the wild-type (WT) and Tbx3-knockout (Tbx3-KO) primary hypothalamic cultures and identified genes that were both differentially expressed in this in vitro model and known to be expressed in Pomc neurons. This analysis highlighted 449 transcripts that were differentially expressed (243 downregulated and 206 upregulated). Unbiased pathway analysis revealed that Tbx3 deletion significantly downregulated the expression of genes controlling cellular proliferation, differentiation and determination of cellular fate (Supplementary Fig. 3). In turn, several genes linked with intracellular metabolic pathways were upregulated, albeit in a less significant way (Supplementary Fig. 3). To complement this unbiased approach, in silico analysis of the genomic loci coding for Pomc, Cart and Agrp for potential Tbx3-binding sites (T-box-binding motifs) was performed23. Potential Tbx3-binding sites were found in all three genes, thus suggesting that Tbx3 altered their transcription directly (Supplementary Fig. 3). To further explore the molecular machinery linked with Tbx3 in hypothalamic neurons, we performed immunoprecipitation of Tbx3 from adult C57BL/6J mouse hypothalami, then used mass spectrometry to identify Tbx3-interacting proteins. We identified 142 proteins that were significantly enriched by Tbx3 precipitation (Supplementary Fig. 3 and Supplementary Table 2), including previously known Tbx3 interactors such as Kif21 (ref. 24), AES25 and Tollip26. Pathway analysis of these interacting proteins highlighted their roles in several processes, notably including inter- and intracellular signalling and neuronal development (Supplementary Fig. 3). These genomic and proteomic data led us to test the hypothesis that a lack of Tbx3 in the ARC might interfere with the cellular fate and differentiation stage of these neurons and therefore affect their peptidergic profile in addition to potentially affecting neuropeptide generation via direct transcriptional actions.
Accordingly, we measured Pomc and Agrp mRNA expression in WT and Tbx3-KO primary hypothalamic neurons through qRT–PCR. Both transcripts were significantly downregulated after Ad-Cre-mediated Tbx3 deletion (Supplementary Fig. 4). These changes were reproducible in vivo, because we found significantly lower expression levels of Pomc and Agrp mRNA in the ARC in Pomc-Cre;Tbx3loxP/loxP mice than in control littermates (Fig. 3a). No changes in Kiss or growth-hormone-releasing hormone (Ghrh) mRNA levels were observed in these animals, whereas the mRNA levels of tyrosine hydroxylase were elevated, and there was a trend toward elevated levels of somatostatin (Fig. 3a). To explore whether these changes in the peptidergic expression profile were caused by neurodevelopmental alterations, Pomc-Cre;Tbx3loxP/loxP mice were crossed with Pomc-GFP reporter animals to precisely quantify Pomc-expressing cells during both embryonic and postnatal life, when ARC-Pomc neurons are generated and acquire their terminal peptidergic identity27,28. No difference in Pomc neuronal cell number was detected in this model at E14.5, E15.5, or E18.5, thus implying normal neuronal generation in utero (Supplementary Fig. 4). No change in Pomc counts was observed at P0, whereas a substantial decrease in the number of Pomc-positive neurons was found at P4, and this relative decrement remained at P14 and 12 weeks (adult) (Fig. 3b,c). Despite progressive loss of Pomc expression at P2 and P4, no significant apoptotic activity was observed in this region (Supplementary Fig. 4), nor did we detect any proliferation leading to new Pomc-positive neurons between P0 and P3, as assessed with BrdU (Supplementary Fig. 4), thus confirming that most Pomc neurons are generated during embryonic life28 and suggesting that neurogenesis and/or cellular turnover do not contribute to the Tbx3-mediated control of Pomc expression observed during neonatal life. Furthermore, no compensatory change was observed in the Pomc-processing enzymes of Pomc-Cre;Tbx3loxP/loxP mice (Supplementary Fig. 4). Collectively, these data demonstrate that constitutive loss of Tbx3 in Pomc-expressing neurons undermines the melanocortin system, probably by interfering with the proper terminal differentiation of this neuronal population during postnatal life and possibly via direct transcriptional actions.
Loss of Tbx3 alters the peptidergic profile of Agrp neurons
In agreement with the results from the mRNA analysis documenting decreased ARC Agrp mRNA (Fig. 3a), Pomc-Cre;Tbx3loxP/loxP mice displayed a diminished number of neuropeptide Y (Npy)-expressing neurons (co-expressed in most Agrp neurons29) in the ARC (Fig. 3d,e). This finding was also reflected by reduced Npy projection density in the paraventricular nucleus of the hypothalamus (PVN) (Fig. 3d–f), as demonstrated by crossing Pomc-Cre;Tbx3loxP/loxP mice with Npy-GFP reporter mice. Because a substantial fraction of Agrp and Npy neurons are derived from Pomc-expressing cells27, Cre-mediated ablation of Tbx3 in these cells may interfere with Agrp and Npy expression in this animal model, thus suggesting that Tbx3’s action in Agrp-expressing neurons may be similarly implicated in controlling the peptidergic profile of this specific neuronal subpopulation. To test this hypothesis, we crossed Agrp-Cre;Tbx3loxP/loxP mice with Npy-GFP reporter mice and quantified the number of Npy-positive neurons and their neuronal projections. Significantly fewer Npy-positive neurons in the ARC (Fig. 3g,h) and less Npy immunoreactivity in the PVN (Fig. 3g–i) were observed in Agrp-Cre;Tbx3loxP/loxP;Npy-GFP mice than in littermate controls, as well as a significant decrease in ARC Agrp mRNA levels (Supplementary Fig. 4). Thus, Tbx3 action in hypothalamic ARC neurons controls the peptidergic expression profiles of different neuronal subpopulations.
Tbx3 is critical for the differentiation of Pomc neurons
To further delineate the process underlying Tbx3-mediated control of neuropeptide expression, we used a cell lineage approach and crossed Pomc-Cre;Tbx3loxP/loxP mice with ROSAmT/mG reporter mice to genetically and permanently label cells undergoing Cre-mediated recombination (via the Pomc-Cre driver) as well as their neuronal projections. We then quantified Pomc expression and assessed its colocalization with GFP, which was indicative of Cre-mediated recombination. The P4 Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG pups had a significantly smaller number of Pomc-positive cells than controls (Fig. 4a,b; raw counts available in Supplementary Table 3), thus reproducing our previously obtained results (Fig. 3b,c). However, no change was observed in the number of neurons or in neuronal-fibre density by analysing Cre-recombined (GFP-expressing) cells (Fig. 4a–c). These data are in agreement with the absence of apoptotic events at P2–4 (Supplementary Fig. 4) and demonstrate that loss of Tbx3 function in Pomc-expressing cells does not affect cellular survival or neuronal architecture during embryonic or early postnatal development. Instead, most Cre-recombined neurons in Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG mice lacked Pomc immunoreactivity (Fig. 4a,b, arrows), thus suggesting that Tbx3 ablation in Pomc-positive cellular populations disrupts their normal peptidergic identity. Such an alteration in Pomc neuronal identity in Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG was also observed in adult animals (Fig. 4d,e, arrows; raw counts available in Supplementary Table 3). Similarly, Cre-recombined cells in Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG had lower expression of Cart than controls, thus indicating that the peptidergic alterations in this model are not limited to Pomc (Supplementary Fig. 5; raw counts available in Supplementary Table 3). A slight decrease in the number of Cre-recombined cells and in neuronal fibre density in the ARC and PVN was observed in adult Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG mice compared with controls (Fig. 4d–i). We hypothesize that this finding is indicative of cellular loss in Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG mice during adulthood, because this phenomenon occurred only after the peptidergic identity impairment observed at P4. We speculate that Tbx3 deletion in hypothalamic Pomc neurons may impair neuronal maturation during postnatal life, which might in turn provoke cell death in a subpopulation of neurons during the transition into adult life. However, these results could also be linked with decreased postnatal neurogenesis and/or impaired neuronal turnover of Pomc-positive cells in Pomc-Cre;Tbx3loxP/loxP;ROSAmT/mG mice, perhaps linked with the condition of obesity observed in these animals. The concept of postnatal hypothalamic neurogenesis, however, remains controversial30. These data collectively demonstrate that Tbx3 has a fundamental role in maintaining the identity of ARC Pomc-expressing cells, a process that underlies changes in the neuropeptidergic profile of these neurons and consequently in systemic energy homeostasis.
Tbx3 controls the identity and plasticity of mature Pomc neurons
Because ARC Tbx3 levels are modulated by nutritional status in mice (Fig. 1e), we asked whether Tbx3 in hypothalamic Pomc neurons might be implicated in the previously observed plastic ability of these cells to adjust Pomc expression and release in response to changes in nutritional status31. Pomc-positive cells and Pomc immunoreactivity were measured in adult Pomc-Cre;Tbx3loxP/loxP;Pomc-GFP and control animals in the ad libitum–fed condition and after exposure to a fasting–refeeding paradigm. In controls, fasting reduced Pomc-positive cell counts (Fig. 4j,k) and Pomc immunoreactivity (Supplementary Fig. 5) relative to what occurred in ad libitum–fed mice, whereas refeeding normalized Pomc expression, as previously reported31. In contrast, changes in nutritional status did not alter Pomc expression in adult Pomc-Cre;Tbx3loxP/loxP;Pomc-GFP mice (Fig. 4j,k and Supplementary Fig. 5), thus implicating Tbx3 in fine-tuning Pomc expression in response to energy needs. These data also suggest that Tbx3 is likely to control the peptidergic profile of fully differentiated hypothalamic neurons in adult mice. To assess this possibility, we quantified Pomc-positive neurons in our adult-onset model of viral-mediated hypothalamic Tbx3 deletion. A prominent decrease in ARC Pomc-positive cells was observed (Fig. 4l,m), with no changes in apoptotic events (Supplementary Fig. 5), thus implying that loss of Tbx3 in fully mature and specified neurons alters their peptidergic identity. To uncover whether such an alteration might underlie hyperphagia, and therefore the obese phenotype observed in AAV-Cre-treated mice, we challenged these animals with intracerebroventricular (ICV) injections of the biologically active Pomc-derived peptide alpha-melanocyte-stimulating hormone (α-MSH) at a subeffective dose. ICV injection of this dose of α-MSH had a slight, non-significant hypophagic effect in control (AAV-GFP) mice. In contrast, this approach significantly normalised food intake in AAV-Cre-treated animals to the level of control AAV-GFP mice (Fig. 4n). Together, these results indicate that Tbx3 knockdown in fully differentiated ARC neurons impairs their peptidergic expression profile under non-stimulated conditions and undermines the ability of Pomc neurons to adjust Pomc expression and release in response to changes in nutritional status. These alterations in turn provoke dysregulated central melanocortin tone, a blunted neuronal response to the organism’s nutritional status, and ultimately obesity.
Tbx3 functions are conserved in Drosophila and human neurons
The T-box family of transcription factors is remarkably conserved among species32. In Drosophila melanogaster, a Tbx3 homologue protein is encoded by the gene omb (or bifid). Omb is expressed in the CNS in adult flies, as assessed by double immunohistochemistry between omb and the synaptic marker bruchpilot (labelled by the Nc82 antibody) (Fig. 5a and Supplementary Fig. 6). To address whether neuronal Tbx3 action on energy homeostasis might be conserved in Drosophila, we generated flies bearing an inducible nervous-system-specific omb-knockdown system (Fig. 5b). Relative to the restults for controls (RNAi off), knockdown of omb (RNAi on) induced a significantly higher body-fat content (Fig. 5c). These results were reproduced in a second transgenic Drosophila model by using a different omb RNAi targeted sequence (Supplementary Fig. 6).
To determine whether Tbx3 loss-of-function phenotypes could be recapitulated in a relevant human neurocellular model system, we investigated the role of TBX3 in the control of differentiation and the peptidergic profile of human hypothalamic neurons. H9 human embryonic stem cells (hESC; WA09; WiCell) were differentiated into ARC-like neurons over the course of 27 d (Fig. 5d), as previously described33,34,35. In this in vitro human hypothalamic neuronal model, NKX2.1 expression was observed by day 12 of differentiation, corresponding to the hypothalamic progenitor stage (Fig. 5e). Low-level expression of class III β-tubulin (TUBB3), a neuronal differentiation marker, occurred by day 12 and reached a maximum at day 27 (Fig. 5f). Expression of POMC and its processing enzyme proprotein convertase subtilisin/kexin type 1 (PCSK1) was detected after neuronal maturation at day 27 (Fig. 5g,h). TBX3 expression was observed in this model at day 12 of differentiation, corresponding to the NKX2.1-positive hypothalamic progenitor stage, and TBX3 levels remained stable in differentiated ARC-like neurons, as obtained on day 27 (Fig. 5i).
To assess the effect of TBX3 deletion on human hypothalamic neuronal differentiation, we generated two independent TBX3-KO hESC lines by using CRISPR–Cas9 (Supplementary Fig. 6). Despite efficient TBX3 ablation (Supplementary Fig. 6), no change in the hypothalamic progenitor marker NKX2.1 was observed at day 12 in either TBX3-KO line (Supplementary Fig. 6), thus suggesting normal differentiation into hypothalamic progenitors. At day 27, NKX2.1 as well as TUBB3, the marker for neuronal differentiation, were greatly diminished in TBX3-KO cells compared with WT cells (Supplementary Fig. 6 and Fig. 5j, respectively), a result indicative of an impaired neuronal maturation state in the TBX3-KO condition in this in vitro human neurocellular model system. In silico analysis of the genomic loci of genes encoding human POMC, CART and AGRP for potential Tbx3-binding sites (T-box-binding motifs) revealed, as in mice, Tbx3-binding sites in all three genomic loci (Supplementary Fig. 6). However, because the strong decrease in TUBB3 in the absence of TBX3 indicated that some hypothalamic differentiation programmes were halted, further analysis of expression levels for neuropeptides such as POMC was precluded.
Together, these data reveal that TBX3 is essential for the maturation of hypothalamic progenitors into ARC-like POMC-expressing neurons. Furthermore, our data suggest that Tbx3 has a conserved role in the regulation of energy homeostasis in invertebrates and mammals, including humans, although the molecular and cellular underpinnings might differ across different species.
The heterogeneity of hypothalamic ARC neurons allows for rapid and precise physiological adaptation to changes in body energy status and is thus highly relevant for adequate maintenance of energy homeostasis. Although several transcriptional nodes are known to establish hypothalamic neuronal identity by controlling early neurogenesis and cellular fate during embryonic life28,36,37,38, the molecular programme driving the terminal specification and identity maintenance of ARC neurons during postnatal life remains incompletely understood; some advances have identified Islet-1 (refs. 39,40), Bsx41 and microRNAs42 as crucial regulators.
In the present experiments, we demonstrate that the transcription factor Tbx3 is required for terminal specification of hypothalamic ARC melanocortin neurons during neonatal development and is also required for the normal maintenance and plasticity of their peptidergic programme throughout adulthood.
Our work highlights a previously uncharacterized role of Tbx3 in the regulation of energy metabolism. The brain expression profile and the functional data presented reveal that Tbx3 action in hypothalamic neurons contributes to the CNS-mediated control of systemic metabolism. Loss of Tbx3 in Pomc-expressing neurons during development causes glucose intolerance and obesity secondary to decreased energy expenditure and lipid utilization in adult mice.
These metabolic alterations are accompanied by a massive decrease in the number of Pomc-expressing neurons during postnatal life, independently of changes in cell number, which probably underlies the observed obesity phenotype. In agreement, neonatal Pomc neuronal ablation promotes similar metabolic alterations43. Intriguingly, constitutive loss of Tbx3 specifically in Agrp/Npy-co-expressing neurons does not translate into phenotypic metabolic changes, although there is a significant decrease in Agrp and Npy expression. Such a lack of metabolic alterations in this model is probably the result of compensatory developmental mechanisms masking the ability of Agrp and Npy to modulate systemic metabolism44, a phenomenon previously observed after neonatal Agrp/Npy neuronal ablation45,46. Thus, Tbx3 affects systemic energy homeostasis by controlling the peptidergic identity profile of different populations that directly modulate the activity of the melanocortin system in ARC neurons during neonatal life, when maturation of the melanocortin system occurs28,47.
Importantly, Tbx3 deletion in fully mature adult hypothalamic ARC neurons selectively decreases the number of Pomc-expressing neurons, a phenotype mimicking the observations in mice with Pomc-promoter-driven deletion of Tbx3 from the genome at mid-term developmental stages. This translates into dysregulated central melanocortin tone that is in turn linked to hyperphagia, alterations in systemic lipid oxidation capacity and obesity. All of these findings are in agreement with the physiological role of Pomc neurons and the central melanocortin system during adulthood48,49. Thus, Tbx3 not only is required for establishing Pomc identity during neonatal life but also is likely to play a key role in maintaining the peptidergic identity and functional activity of fully differentiated ARC neurons.
The cellular and metabolic effects provoked by Tbx3 ablation in hypothalamic ARC neurons are independent of neuronal survival and/or turnover, as demonstrated by our cell-lineage tracing approach. Instead, Tbx3 seems to direct intracellular programmes controlling the neuronal differentiation state, in agreement with previous studies linking Tbx3 intracellular activity with differentiation and cell fate commitment in non-neuronal cells9,10,11. Whether Tbx3 loss of function in immature and/or fully differentiated hypothalamic neurons may induce cellular reprogramming and a peptidergic identity switch is a compelling hypothesis requiring further scrutiny, but it is supported by evidence of neuronal developmental plasticity within the mammalian CNS50,51,52. In this context, our in silico–based prediction of Tbx3-binding sites suggests that the observed changes in the peptidergic identity profiles might also be explained by direct transcriptional effects in Pomc, Cart and Agrp genomic loci. However, a more comprehensive and unbiased analysis, such as by chromatin immunoprecipitation followed by high-throughput sequencing, will be required to directly test this hypothesis. Similarly, a detailed characterization of the molecular machinery controlled by Tbx3 in hypothalamic neurons will be necessary to elucidate the main intracellular mechanisms underlying the metabolic effects observed. Our profiling of genes and proteins linked with Tbx3 does not allow them to be causally linked with the metabolic changes observed, but this initial effort may spur future research addressing the role of such Tbx3-linked machinery in the context of obesity. It will also be of paramount importance to determine whether Tbx3 influences neuropeptidergic profiles and systemic metabolism via interactions with known metabolic signals implicated in neuronal specification, such as neurogenin 3, Mash1, OTP or Islet-1 (refs. 37,38,39).
Our observations in Drosophila melanogaster suggest that the link between neuronal Tbx3 action and systemic energy homeostasis is probably evolutionarily conserved; however, our data do not enable understanding of the cellular and molecular mechanisms underlying the obese-like phenotype observed in flies or whether these mechanisms are conserved across different species. Because Drosophila does not express Pomc, Agrp or any homologue peptide, another neuronal population might link Tbx3 action with adiposity regulation in this species. Intriguingly, our data show that TBX3 is essential for the maturation of hypothalamic progenitors into ARC-like human neurons. Because human subjects with TBX3 mutations display pathological conditions consistent with ARC neuronal dysfunction (obesity, impaired GHRH release and alterations in reproductive capacity13,14), we speculate that mutations affecting TBX3 in humans might undermine ARC neuronal differentiation status and/or peptidergic profiles, changes that ultimately affect body weight regulation, reproduction and growth. Thus, our findings might have implications for human pathophysiology.
Neurons sensitive to the orexigenic hormone ghrelin express Tbx3, such that ghrelin may also directly regulate Tbx3 expression5. Moreover, we uncovered a clear link among nutritional status, Tbx3 action and neuropeptide expression in hypothalamic Pomc neurons. Whether hormonal factors and nutritional status in turn alter the peptidergic identity of hypothalamic neurons in physiological or pathophysiological conditions via modulation of Tbx3 remains a critical question. A detailed characterization of the role of Tbx3 in the context of nutritional and hormone-based regulation of hypothalamic neuronal activity might help in deciphering the main environmental factors controlling peptidergic identity development, maintenance and potential plasticity in mammalian CNS neurons.
We uncovered a molecular switch implicated in the terminal differentiation of body-weight-regulating ARC neurons into specific peptidergic subtypes, unravelling one of the mechanisms responsible for the neuronal heterogeneity of hypothalamic ARC neurons. Our findings represent another step toward the identification the key molecular machinery controlling the functional identity of hypothalamic neurons, particularly during postnatal life, and may consequently facilitate understanding of the fundamental neuronal mechanisms implicated in the pathogenesis of obesity and its associated metabolic perturbations.
Ethical compliance statement
All animal experiments were approved and conducted under the guidelines of Helmholtz Zentrum Munich and of the Faculty Animal Committee at the University of Santiago de Compostela.
All experiments were conducted on male mice. The mice were fed a standard chow diet and group housed under a 12 h:12 h light:dark cycle at 22 °C and given free access to food and water unless indicated otherwise. C57BL/6J mice were provided by Jackson Laboratories. Tbx3loxP/loxP mice were generated previously53 and back-crossed on a C57BL/6J background for five generations. Pomc-Cre mice (Jax mice stock 5965 (ref. 21)) and Agrp-Cre mice (Jax mice stock 012899 (ref. 20)) were mated with Tbx3loxP/loxP mice to generate Pomc- and Agrp-specific Tbx3-knockout mice (Pomc-Cre;Tbx3loxP/loxP or Agrp-Cre;Tbx3loxP/loxP). Pomc-Cre;Tbx3loxP/loxP and control (Pomc-Cre) mice were crossed with a ROSAmT/mG reporter line (Jax mice stock 007576 (ref. 54)) so that neurons expressing Pomc were permanently marked. Pomc-Cre;Tbx3loxP/loxP, Agrp-Cre;Tbx3loxP/loxP or control mice (Tbx3loxP/loxP) were crossed with mice selectively expressing GFP in Npy-expressing neurons (Npy-GFP, Jax mice stock 006417 (ref. 55) or in Pomc-expressing neurons (Pomc-GFP Jax mice stock 009593 (ref. 56)).The Tbx3-Cre-Venus mouse line was created by using CRISPR–Cas9 technology. The coding sequences for 2A peptide bridges, Cre recombinase, Venus fluorescent protein and bovine growth hormone polyadenylation signal were cloned into a targeting vector between 5′ and 3′ homology arms flanking the stop codon of the Tbx3 locus. Homologous recombination was confirmed by PCR and Southern blot analysis (using the DIG system from Roche). For the studies involving embryos, the breeders were mated 1 h before the dark phase and checked for a vaginal plug the next day. The day of conception (sperm-positive vaginal smear) was designated as E0. The day of birth was considered P0. Additional information can be found in the Nature Research Reporting Summary.
To measure food consumption, we housed mice at two or three per cage. Body composition (fat and lean mass) was measured with quantitative nuclear magnetic resonance technology (EchoMRI). Energy expenditure and the respiratory exchange ratio were assessed with a combined indirect calorimetry system (TSE PhenoMaster, TSE Systems). O2 consumption and CO2 production were measured every 10 min for a total of up to 120 h (after a minimum of 48 h of adaptation). Energy expenditure (EE, kcal/h) values were correlated to the body weight of the animals recorded at the end of the measurement with analysis of covariance (ANCOVA)19. For the analysis of glucose tolerance, mice were injected intraperitoneally with 1.75 g glucose per kg of body weight (Agrp-Cre;Tbx3loxP/loxP mice) or 1.5 g glucose per kg of body weight (Pomc-Cre;Tbx3loxP/loxP mice). 20% (w/v) d-glucose (Sigma-Aldrich) in 0.9% (w/v) saline was used. Tail-blood glucose concentrations (mg/dl) were measured with a handheld glucometer (TheraSense Freestyle).
Viral-mediated deletion of Tbx3
To ablate Tbx3 in the MBH, recombinant adeno-associated viruses (AAV) carrying the Cre recombinase and the haemagglutinin (HA)-tag (AAV-Cre) or control viruses carrying Renilla GFP (AAV-GFP) were generated as previously described57 and injected bilaterally (0.5 µl per side; 1.0 × 1011 viral genomes ml–1) into the MBH in Tbx3loxP/loxP mice (12 weeks old), with a motorized stereotaxic system from Neurostar. The nuclear localization signal (nls) of the simian virus 40 large T antigen and the Cre-recombinase coding region was fused downstream of the HA tag, in an rAAV plasmid backbone containing the 1.1-kb CMV immediate early enhancer/chicken β-actin hybrid promoter (CBA), the woodchuck post-transcriptional regulatory element (WPRE) and the bovine growth hormone poly(A) (bGH) to obtain rAAV-CBA-WPRE-bGH carrying Cre-recombinase (AAV-Cre). The rAAV-CBA-WPRE-bGH backbone carrying the Renilla GFP cDNA (Stratagene) was used as negative control. rAAV chimeric vectors (virions containing a 1:1 ratio of AAV1 and AAV2 capsid proteins with AAV2 ITRs) were generated by transfection of HEK293 cells with the AAV cis plasmid, the AAV1 and AAV2 helper plasmids, and the adenovirus helper plasmid through standard PEI transfection methods. At 60 h after transfection, cells were harvested, and the vector was purified through an OPTIPREP density gradient (Sigma). Genomic titres were determined with an ABI 7700 real-time PCR cycler (Applied Biosystems) with primers designed for WPRE. Virus was injected bilaterally (0.5 µl per side; 1.0 × 1011 viral genomes ml–1) into the MBH in Tbx3loxP/loxP mice (12 weeks old), with a motorized stereotaxic system from Neurostar. Stereotaxic coordinates were −1.6 mm posterior and ±0.25 mm lateral to the bregma and −5.8 mm ventral from the dura. During the same procedure, a stainless-steel cannula (Bilaney Consultants) was implanted into the lateral cerebral ventricle. Stereotaxic coordinates for ICV injections were −0.8 mm posterior, −1.4 mm lateral from the bregma and −2.0 mm ventral from the dura. Surgeries were performed with a mixture of ketamine and xylazine (100 mg per kg and 7 mg per kg, respectively) as anaesthetic agents and Metamizol (200 mg per kg, subcutaneous), then Meloxicam (1 mg per kg, on three consecutive days, subcutaneous) for postoperative analgesia. For ICV studies, mice were infused with 1 µl of either vehicle (aCSF; Tocris Bioscience) or α-MSH (1 nmol, R&D systems, Tocris) 2 h before the onset of the dark cycle, and food intake followed immediately for 24 h.
Bromodeoxyuridine (BrdU, 50 mg per kg) in ~50 µl of sterile saline was injected daily at postnatal days 0, 1, 2 and 3 in the dorsal neck fold of pups. The pups were euthanized at P7, and the brains were processed for immunohistochemistry.
Adult mice were transcardially perfused with PBS, then with 4% neutral buffered paraformaldehyde (PFA) (Fisher Scientific). Brains from embryos and pups were isolated from non-PFA-perfused animals. After dissection, brains were post-fixed for 24 h with 4% PFA, equilibrated in 30% sucrose for 24 h and sectioned on a cryostat (Leica Biosystems) at 25 μm. Brain sections were incubated with the following primary antibodies: rabbit anti-Pomc precursor (Phoenix Pharmaceuticals, H-029-30), goat anti-Agrp (R&D systems, AF634), chicken anti-GFP (Acris, AP31791PU-N), goat anti-GFP (Abcam, ab6673), rabbit anti-Npy (Abcam, ab30914), rabbit anti-Cart (Phoenix Pharmaceuticals), mouse anti-BrdU (Sigma), goat anti-Tbx3 (A20, Santa Cruz Biotechnology), rabbit anti-Tbx3 (A303–098A, Bethyl Laboratories), rabbit anti-cleaved caspase-3 (5A1E, Cell Signaling), goat anti-Iba1 (Abcam ab107519), rabbit anti-GFAP (Dako, Z0334), chicken anti-vimentin (Sigma, Abcam ab24525) and rabbit anti-HA tag (C29F4, Cell Signaling). Primary antibodies were incubated at a concentration of 1:500 overnight at 4 °C in 0.1 M Tris-buffered saline (TBS) containing gelatine (0.25%) and Triton X-100 (0.5%). Sections were washed with 0.1 M TBS and incubated for 1 h at room temperature with 0.1 M TBS containing gelatine (0.25%) and Triton X-100 (0.5%), using the following secondary antibodies (1:1,000) from Jackson ImmunoResearch Laboratories: goat anti-rabbit (Alexa 647), goat anti-chicken (Alexa 488), donkey anti-goat (Alexa 488) and donkey anti-mouse (Alexa 488).
Images were obtained with a BZ-9000 microscope (Keyence) or a Leica SP5 confocal microscope, and automated analysis was performed in Fiji 1.0 (ImageJ) when technically feasible. Manual counts were performed blinded. When anatomically possible, neuronal cell counts were performed on several sections spanning the medial arcuate nucleus and averaged.
Gene expression analysis by qRT–PCR
Dissected tissues were immediately frozen on dry ice, and RNA was extracted with RNeasy Mini Kits (Qiagen). Whole hypothalamus was isolated and immediately frozen on dry ice. To obtain RNA from ARC micropunches, freshly dissected whole brains were immersed in RNAlater (AM7021, Thermo Fisher) for a minimum of 24 h at 4 °C. The RNAlater-immersed brains were subsequently cut coronally in 280-μm slices with a vibratome, and the ARC was dissected from each slice with a scalpel, as visually aided by binoculars. RNA was extracted with RNeasy Mini Kits (Qiagen). cDNA was generated with a reverse-transcription QuantiTech reverse transcription kit (Qiagen). Quantitative real-time RT–PCR (qRT–PCR) was performed with a ViiA 7 Real-Time PCR System (Applied Biosystems) with the following TaqMan probes (Thermo Fisher): Hprt (Mm01545399_m1), Ppib (Mm00478295_m1), Npy (Mm03048253_m1), Pomc (Mm00435874_m1), Agrp (Mm00475829_g1), Kisspeptin (Mm03058560_m1), Somatostatin (Mm0043667_m1), Tyrosine hydroxylase (Mm00447557_m1), Ghrh (Mm00439100_m1), Tbx3 (Mm01195726_m1), Pcsk1 (Mm00479023_m1), Pcsk2 (Mm00500981_m1), Pam (Mm01293044_m1) and Cpe (Mm00516341_m1). Target gene expression was normalized to expression of the reference genes Hprt or Ppib. Calculations were performed with a comparative method (2−ΔΔCT).
Primary mouse hypothalamic cell cultures
Hypothalami were extracted from Tbx3loxP/loxP mouse foetuses on E14 in ice-cold calcium- and magnesium-free HBSS (Life Technologies), digested for 10 min at 37 °C with 0.05% trypsin (Life Technologies), washed three times with serum-free MEM supplemented with l-glutamine (2 mM) and glucose (25 mM) and dispersed in the same medium. Cells were plated on 12-well plates coated with poly-l-lysine (Sigma-Aldrich) at a density of 1.5 × 106 per well in MEM supplemented with heat-inactivated 10% horse serum and 10% foetal bovine serum, 2 mM l-glutamine and glucose (25 mM) without antibiotics. On day 4, half the medium was replaced with fresh culture medium lacking foetal bovine serum and containing 10 μM of the mitotic inhibitor AraC (cytosine-1-β-d-arabinofuranoside, Sigma-Aldrich) to inhibit non-neuronal cell proliferation. On day 6, neurons were infected with a recombinant adenovirus carrying the coding sequence for the recombinase Cre (Ad5-CMV-Cre-eGFP, named Ad-Cre) to delete the loxP-flanked portion of the Tbx3 gene, or with a control virus (Ad5-CMV-eGFP, named Ad-GFP) from Vector Development Laboratory. On day 7, after 12 h of incubation, virus-containing medium was removed and replaced with fresh growth medium. Neurons were further incubated for 48 h to ensure efficient recombination before performing experiments. Cell cytotoxicity was assessed with a Pierce LDF Cytotoxicity Assay Kit (88953, Thermo Fisher).
For ChIP experiments followed by mass spectrometry (ChIP–MS), hypothalamic samples from 34 individual mice were pooled, and five hypothalami at a time were homogenized in 9 ml of 1% formaldehyde in PBS for 10 min. After quenching for 5 min with 125 mM glycine, samples were washed twice with PBS. Pellets were resuspended in 1 ml of lysis buffer (0.3% SDS, 1.7% Triton, 5 mM EDTA, pH 8, 50 mM Tris, pH 8, and 100 mM NaCl), and the chromatin was sonicated to an average size of 200 bp. After incubation with either an antibody to Tbx3 (A303–098A, Bethyl Laboratories) or an IgG antibody (rabbit IgG 2729 S, Cell Signaling Technology), antibody–bait complexes were bound by Protein G–coupled agarose beads (Cell Signaling Technology) and washed three times with wash buffer A (50 mM HEPES, pH 7.5, 140 mM NaCl and 1% Triton), once with wash buffer B (50 mM HEPES pH 7.5, 500 mM NaCl and 1% Triton) and twice with TBS. Beads were incubated for 30 min with elution buffer 1 (2 M urea, 50 mM Tris-HCl, pH 7.5, 2 mM DTT and 20 µg ml–1 trypsin) followed by a second elution with elution buffer 2 (2 M urea, 50 mM Tris-HCl, pH 7.5 and 10 mM chloroacetamide) for 5 min. Both eluates were combined and further incubated overnight at room temperature. Tryptic-peptide mixtures were acidified with 1% TFA and desalted with Stage Tips containing three layers of C18 reverse-phase material and analysed by mass spectrometry. Peptides were separated on 50‐cm columns packed in house with ReproSil‐Pur C18‐AQ 1.9 μm resin (Dr Maisch). Liquid chromatography was performed on an EASY‐nLC 1000 ultra‐high‐pressure system coupled through a nanoelectrospray source to a Q-Exactive HF mass spectrometer (all from Thermo Fisher). Peptides were loaded in buffer A (0.1% formic acid) and separated by application of a non-linear gradient of 5–32% buffer B (0.1% formic acid, 80% acetonitrile) at a flow rate of 300 nl min–1 over 100 min. Data acquisition switched between a full scan and 15 data‐dependent MS/MS scans. Full scans were acquired with target values of 3 × 106 charges in the 300–1,650 m/z range. The resolution for full-scan MS spectra was set to 60,000 with a maximum injection time of 20 ms. The 15 most abundant ions were sequentially isolated with an ion target value of 1 × 105 and an isolation window of 1.4 m/z. Fragmentation of precursor ions was performed by higher energy C-trap dissociation with a normalized collision energy of 27 eV. Resolution for HCD spectra was set to 15,000 with a maximum ion-injection time of 60 ms. Multiple sequencing of peptides was minimized by excluding the selected peptide candidates for 25 s. Raw mass spectrometry data were analysed with MaxQuant (version 126.96.36.199)58 and Perseus (version 188.8.131.52) software packages. Peak lists were searched against the mouse UniProt FASTA database (2015_08 release) combined with 262 common contaminants by the integrated Andromeda search engine59. The false discovery rate was set to 1% for both peptides (minimum length of seven amino acids) and proteins. ‘Match between runs’ (MBR) with a maximum time difference of 0.7 min was enabled. For a gain in peptide identification, MS spectra were matched to a library of Tbx3 ChIP MS data derived from murine neuronal progenitor cells. Relative protein amounts were determined with the MaxLFQ algorithm60, with a minimum ratio count of two. Missing values were imputed from a normal distribution, by applying a width of 0.2 and a downshift of 1.8 standard deviations. Significant outliers were defined by permutation-controlled Student’s t test (FDR < 0.05, s0 = 1) comparing triplicate ChIP–MS samples for each antibody. Additional information is in the Nature Research Reporting Summary.
RNA-seq was performed in primary neurons isolated from Tbx3loxP/loxP mice and treated with Ad-Cre or Ad-GFP viruses. Sequencing was performed in three independent neuronal isolations totalling 9 Ad-GFP-treated and 11 Ad-Cre-treated independent samples. Before library preparation, RNA integrity was determined with an Agilent 2100 Bioanalyzer and an RNA 6000 Nano Kit. All samples had RNA integrity number (RIN) values >7. One microgram of total RNA per sample was used for library preparation. Library construction was performed as described in the low-throughput protocol of the TruSeq RNA Sample Prep Guide (Illumina) in an automated manner, by using the Bravo Automated Liquid Handling Platform (Agilent). cDNA libraries were assessed for quality and quantity with a Lab Chip GX (Perkin Elmer) and the Quant-iTPicoGreendsDNA Assay Kit (Life Technologies). cDNA libraries were multiplexed and sequenced as 100-bp paired-end runs on an Illumina HiSeq2500 platform. Approximately 8 Gb of sequence per sample were obtained. The GEM mapper61 (v 1.7.1) with modified parameter settings (mismatches = 0.04, min-decoded-strata = 2) was used for split-read alignment against the mouse genome assembly mm9 (NCBI37) and UCSC knownGene annotation. Duplicate reads were removed. To quantify the number of reads mapping to annotated genes, we used HTseq-count62 (v0.6.0). We normalized read counts to correct for possibly varying sequencing depths across samples using the R/Bioconducter package DESeq2 (ref. 63) and excluded genes with low expression levels (mean read count <25) from the analysis. We combined RNA-seq data of the three independent neuronal isolations. Because the independence of the three neuronal isolations might have introduced batch effects, we applied surrogate variable analysis implemented in the R package sva64 to remove them. Gene expression levels between the two virus treatments were compared with DESeq2. We chose 0.001 to be the P-value cutoff after FDR correction (Benjamini–Hochberg). To obtain genes selectively expressed in Pomc neurons, we used the single-cell sequencing data set previously published7 and selected the n14 (Pomc/Ttr), n15 (Pomc/Anxa2) and n21 (Pomc/Glipr1) neuronal clusters as gene expression references. We chose all genes that had a normalized expression value above a noise level of 4.5. Additionally, we required the selected genes to be expressed in at least 10% of the 1,191 samples in our Pomc-neuron reference. We intersected the genes differentially expressed in our Tbx3 Ad-Cre65 to test these genes against GO biological process terms66. After the overrepresentation test, we excluded GO terms whose gene list overlapped the list of another term completely. All calculations were performed in R (v3.4.3).
The Drosophila melanogaster neuronal GeneSwitch Gal4 driver line (Elav-Gal4GS) was obtained from the Bloomington Drosophila Stock Center (BDSC43642). GeneSwitch drivers can be activated by progesterone steroids67. The RNAi transgenic lines for the Tbx3-homologue gene omb (line 1, UAS-ombRNAi-C4, line 2, UAS-ombRNAi-C1) were as described in ref. 68. Elav-Gal4GS virgin females and 15 omb RNAi transgenic males were crossed in big-fly food vials for 24 h, and approximately 600 F1 embryos were seeded and kept at 25 °C for growth on standard cornmeal medium (12 h:12 h light:dark cycle; 60–70% humidity). After eclosion (24 h), 50 young adult male and virgin females were fed on small fresh drug-food vials (mifepristone, 200 μM) and control food vials (ethanol, same volume as the mifepristone dissolved volume), respectively, for 6 d. At least eight technical replicates (five flies each) were collected for body-fat content measurement (TAG value normalized to protein value), on the basis of a coupled-colorimetric assay (triglyceride, Pointe Scientific (T7532))69 and bicinchoninic acid assay (protein, Pierce, Thermo; 23225)70,71; five adult male flies per technical replicate, 600 μl homogenization buffer (0.05% Tween-20 in water) and 5-mm metal beads (Qiagen 69989) were homogenized in 1.2-ml collecting tubes (Qiagen 19560; caps, 19566) with a tissue lyser II (Qiagen, 85300) and immediately incubated at 70 °C (water bath) for 5 min. The fly homogenates were spun down at 5,000 r.p.m. for 3 min, and 2 × 50 μl supernatant for each replicate, TAG standards solutions (Biomol, Cay-10010509; 0, 5.5, 11, 22, 33 and 44 μg in 50 μl homogenization buffer), BSA (bovine serum albumin) and protein-standard samples (0, 25, 125, 250, 500 and 750 mg ml–1) were measured at 500 nm (for TAG) and 570 nm (for protein). The assay kit for the colorimetric assay was from Pointe Scientific (T7532). Immunostainings were carried out in 5-d-old adult male flies (omb-Gal4>UAS-GFP transgenic line72) through a method reported previously73. Brains were dissected in cold PBS and fixed in 4% PFA in PBS at room temperature (RT) for 30 min. Brain tissues were incubated with 0.25% Triton X-100 in PBS (0.25% PBST) at RT for 25 min and blocked with 1% BSA & 3% normal goat serum (NGS) in 0.25% PBST for 1 h at RT with mild rotation. The following primary antibodies were used: mouse anti-Bruchpilot (nc82, 1:50) (nc82, deposited in the DSHB by Buchner, E. (DSHB Hybridoma Product nc82)), chicken anti-GFP (1:1000) (Acris, AP31791PU-N) and rabbit anti-Omb serum (1:1,000)74. The following secondary antibodies (Jackson ImmunoResearch Laboratories) were used: donkey anti-mouse (Alexa 568), goat anti-rabbit (Alexa 647) and goat anti-chicken (Alexa 488). Secondary antibodies were incubated at RT for 2 h. After 5 × 10 min washing with 0.25% PBST and 1× overnight washing with PBS, tissues were mounted on gelatine-coated glass slides and coverslipped for image analysis. Images were obtained with Leica SP8 confocal system (×20 air objective) and processed with Fiji 1.0 (ImageJ).
Human embryonic stem cells
The human H9 ESC line was purchased from WiCell. Cells were maintained in a humidified incubator at 37 °C on irradiated murine embryonic fibroblasts (MEFs; CF-1 MEF 4 M IRR; GLOBALSTEM) in DMEM KO medium (10829018; Thermo Fisher) supplemented with 15% KnockOut Serum Replacement (10828028; Thermo Fisher), 0.1 mM MEM non-essential amino acids (11140050; Thermo Fisher), 2 mM GlutaMAX (35050061; Thermo Fisher), 0.06 mM 2-mercaptoethanol (21985023; Thermo Fisher), FGF-basic (AA 1–155), (20 ng ml–1 medium; PHG0263; Thermo Fisher), and 10 µM Rock inhibitor (S1049; Selleckchem). Cells were passaged with Accutase (00–4555–56; Thermo Fisher). For CRISPR–Cas9-mediated deletion of Tbx3, pCas9_GFP was obtained from Addgene (Kiran Musunuru; 44719). As previously published, the GFP was replaced by a truncated CD4 gene from the GeneArt CRISPR Nuclease OFP Vector (Thermo Fisher) by GenScript through CloneEZ seamless cloning technology, thus resulting in vector pCas9_CD4 (ref. 75). The full vector sequence of pCas9_CD4 is given in Supplementary Table 4. The guide RNA sequence 5′-TCATGGCGAAGTCCGGCGCC-3′ was obtained by using Optimized CRISPR Design (MIT; http://crispr.mit.edu/). Cloning of the gRNA into pGS-U6-gRNA was performed by GenScript. 800,000 human ESCs were collected and mixed in nucleofection buffer (Human Stem Cell Nucleofector Kit 2; VPH-5022) with gRNA and pCas9_CD4 plasmids (2.5 μg each). Nucleofection was performed in an Amaxa Nucleofector II (Programme A-023) with a Human Stem Cell Nucleofector Kit 2 according to the manufacturer’s instructions. Cells were plated on MEFs for 2 d for recovery, and transfected cells were purified through positive selection of CD4-expressing cells by using human CD4 MicroBeads (130–045–101; MS Column, 130–042–20; MACS Miltenyi Biotec) and replated at clonal density in 10 cm2 tissue culture plates on MEFs. After 7–12 d, ESC colonies were picked into 96-well plates and, 4–5 d later were split 1:2 (one well for genomic DNA extraction followed by sequence analysis as described below, and one well for amplification of clones and further analysis and freezing, if indicated). For genomic-DNA extraction and PCR analysis, genomic DNA was extracted with HotShot buffer according to a published protocol76. The DNA region of interest was PCR-amplified with the following primers: 5′-GAGAGCGCCGCCGCGCCGT-3′ and 5′-GCTGCGGACTTGTCCCCGGCTGGA-3′76. Sequences were generated by Sanger sequencing (Macrogen). Sequence analysis was performed to identify clones carrying mutations resulting in TBX3 knockout. Positive clones were amplified, and genomic DNA was extracted with a Gentra Puregene Core Kit A (Qiagen). Topo TA Cloning Kit for sequencing (K457501; Thermo Fisher) was used to determine the zygosity of TBX3 knockout with the following primers: 5′- CACCTTGGGGTCGTCCTCCA-3′ and 5′- CGCAAGGCACAAGGACGGTCA-3′. G-band karyotyping analysis was done by Cell Line Genetics. Chromosome analysis was performed on 20 cells per cell line.
Differentiation of human ESCs into arcuate-like neurons
Human ESCs differentiated into hypothalamic arcuate-like neurons were derived from human ESCs through a previously published protocol33,34,35. H9 cells were plated on dishes coated with Matrigel (08–774–552; Thermo Fisher) dishes at a density of 100,000 cells per cm2) in human ESC medium, as described above, supplemented with bFGF and Rock inhibitor. Cell density was observed after 24 h. If the cells were not yet at 100% confluency, medium was aspirated and replaced with ESC medium with bFGF and Rock inhibitor for another 24 h. After cells reached 100% confluency, differentiation was initiated. 10 μM SB 431542 (S1067; Selleckchem) and 2.5 μM LDN 193189 (S2618; Selleckchem) were used from day 1 to day 8 to inhibit TGFβ and BMP signalling to promote neuronal differentiation from human ES cells77. 100 ng ml–1 SHH (248-BD; (R&D Systems) and 2 μM purmorphamine (PM; S3042; Selleckchem) were added from days 1–8 to induce ventral brain development and NKX2.1 expression. Cells were cultured on days 1–4 in ESC medium, from days 5–8, the medium was switched stepwise from ESC medium to N2 medium (3:1, 1:1, 1:3). N2 medium (500 ml) consisted of 485 ml DMEM/F12 (11322; Thermo Fisher) supplemented with 5 ml MEM non-essential amino acids (11140050; Thermo Fisher), 5 ml of a 16 % glucose solution and 5 ml N2 (1370701; Thermo Fisher). Ascorbic acid (A0278; Sigma-Aldrich) was added just before use at a final concentration of 200 nM. From Day 9 onward, cells were cultured in N2-B27 medium consisting of 475 ml DMEM/F12 (11322; Thermo Fisher) supplemented with 5 ml MEM non-essential amino acids (11140050; Thermo Fisher), 5 ml of a 16 % glucose solution, 5 ml N2 (1370701; Thermo Fisher) and 10 ml B27 (12587010; Thermo Fisher). Ascorbic acid (A0278; Sigma-Aldrich) was added just before use at a final concentration of 200 nM. Inhibition of Notch signalling by 10 μM DAPT (S2215; Selleckchem) was performed from days 9 to 12. Nkx2.1 + progenitors were collected and re-plated on extracellular matrix (poly-l-ornithine (A-004-C; Millipore) and laminin (23017015; Thermo Fisher)) to enhance the attachment and differentiation of neuron progenitors. The Notch inhibitor DAPT was used to inhibit the proliferation of progenitor cells and promote further neuronal differentiation78,79. The neurotrophic factor BDNF (20 ng ml–1; 450–02; PeproTech) was introduced after DAPT treatment to improve the survival, differentiation and maturation of these neurons. For RT–PCR analyses, Cells at day 0, day 12 and day 27 of differentiation were homogenized in Trizol reagent (15596026; Thermo Fisher), and total RNA was extracted with an RNeasy Plus Micro Kit (74034; Qiagen) with on-column DNase I (79254; Qiagen) treatment to remove genomic DNA contamination and stored at −80 °C until further processing. A total of 500 ng of total RNA was used for reverse transcription with a Transcriptor First Strand cDNA Synthesis Kit (04897030001; Roche Diagnostic) by using a mixture of anchored oligo(dT)18 and random-hexamer primers according to the manufacturer’s instructions. Quantitative PCR was performed with a Light-Cycler 480 (Roche Diagnostics) with SYBR Green in a total volume of 10 μl with 1 μl of template, 1 μl of forward and reverse primers (10 μM) and 5 μl of SYBR Green I Master-Mix (04707516001; Roche Diagnostic). Reactions included an initial cycle at 95 °C for 10 min, followed by 40 cycles of denaturation at 95 °C for 10 s, annealing at 60 °C for 5 s and extension at 72 °C for 15 s. Crossing points were determined by Light-Cycler 480 software, by using the second-derivative maximum technique. Relative expression data were calculated with the delta-delta Ct method, with normalization of the raw data to expression of TBP. Quantitative PCR was performed to determine the mRNA levels of TBX3, POMC, TUBB3, PCSK1, NKX2.1. Primer sequences are shown in Supplementary Table 1.
Western blot analysis
Human ESC (H9) and hypothalamic arcuate-like neurons (day 27 of differentiation) were washed with DPBS and lysed in RIPA Lysis and Extraction Buffer (Thermo Fisher) with protease and phosphatase inhibitors (78442; Thermo Fisher), incubated at 4 °C for 15 min and then centrifuged at 12,000 r.p.m. for 15 min at 4 °C. Fifteen micrograms of total protein from each extract was loaded on a 4–12% gradient Bis-Tris gel (NP0335BOX; Thermo Fisher) and transferred onto nitrocellulose membrane with an iBlot 2 Dry Blotting System (Thermo Fisher). The membrane was blocked for 1 h at room temperature with SuperBlock T20 (TBS) Blocking Buffer (37536; Thermo Fisher) and then incubated with primary antibody to TBX3 (1:100; ab99302; Abcam) overnight at 4 °C, washed three times with TBS with 0.1% Tween-20 (1706531; Bio-Rad) and incubated with secondary antibody anti-rabbit HRP (1:10,000; 7074 S; Cell Signaling) for 1 h at room temperature. Specific bands were then detected through electrochemiluminescence analysis with SuperSigna West Pico PLUS Chemiluminescent Substrate (34577; Thermo Fisher). An antibody to beta-actin (1:1,000; ab8226; Abcam), with anti-mouse HRP (1:10,000; 7076 S; Cell Signaling) as a secondary antibody, was used as a loading control. Validation of the goat anti-Tbx3 antibody (A20, Santa Cruz Biotechnology) was performed by using Tbx3-deficient embryos (E13.5) kindly provided by A. Kispert80. Proteins were extracted with RIPA buffer containing protease- and phosphatase-inhibitor cocktails (Thermo Fisher) 1 mM phenyl-methane-sulfonyl fluoride (PMSF) and 1 mM sodium butyrate (Sigma-Aldrich). Proteins were transferred on nitrocellulose membranes by using a Trans Blot Turbo transfer apparatus (Bio-Rad), and stained with primary antibody goat anti-Tbx3 (1:500) and a secondary antibody anti-goat HRP (1:1,000). Detection was carried out on a LiCor Odyssey instrument (software Image studio 2.0), by using electrochemiluminescence (Amersham). Additional information is available in the Nature Research Reporting Summary.
Tbx3-focused single-cell RNA-sequencing analysis
Data for the scRNA-seq analysis were obtained from GEO accession codes GSE90806 and GSE93374 (ref. 7). The data matrix comprised 21,086 cells and 22,802 genes generated from the arcuate-median eminence (Arc-ME) of the mouse hypothalamus by Campbell et al.7. We used Seurat software81 to perform clustering analysis. We identified the 2,250 most variable genes across the entire dataset, controlling for the known relationship between mean expression and variance. After scaling and centring the data along each variable gene, we performed principal component analysis and identified 25 significant prinicipal components for downstream analysis that were used to identify 20 clusters. Similarly to those identified by Campbell et al.7, a total of 13,079 neurons and 8,007 non-neuronal cells were identified in our study. We further used the neuronal identities assigned by the authors for clustering the neurons into their respective neuronal clusters. For differential expression between cell type clusters, we used the negative binomial test, a likelihood-ratio test assuming an underlying negative binomial distribution for UMI-based datasets.
Statistical analyses were conducted in GraphPad Prism (version 5.0a). For each experiment, slides were numerically coded to obscure the treatment group. Statistical significance was determined with unpaired two-tailed Student’s t test, one-way ANOVA or two-way ANOVA followed by an appropriate post hoc test, as indicated in figure legends, and linear regression when appropriate. P ≤ 0.05 was considered statistically significant. Additional information is provided in the Nature Research Reporting Summary.
Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.
The authors declare that all data supporting the findings of this study are available within the paper and its supplementary information files. The RNA-seq database generated in our paper has been made publicly available through Gene Expression Omnibus (GEO accession number GSE119883).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank A. Kispert and M.-O. Trowe (Institut für Molekularbiologie, Medizinische Hochschule Hannover, Hannover, Germany) for kindly providing Tbx3-deficient embryos, J. Friedman (The Rockefeller University, Howard Hughes Medical Institute, New York, NY, USA) for scientific guidance and for graciously providing access to data shown in ref. 5, M. Guzmán (Complutense University, Madrid, Spain) for assistance with the generation of AAV-GFP and AAV-Cre viral particles, the Bloomington Drosophila Stock Center (BDSC) (NIH P40OD018537) for fly stocks, and C. Layritz, H. Hoffmann, N. Wiegert and C. L. Holleman for technical assistance and assistance with animal studies. A.F. is supported by a postoctoral fellowship from the Canadian Institutes of Health Research (Funding reference no. 152588). V.V.T. is supported by NIH-NIDDK grant 5K23DK110539 and in part by the Baylor-Hopkins Center for Mendelian Genomics through NHGRI grant 5U54HG006542. C.A.D. is supported by funding from the NIH (R01 DK52431, R01 DK110113 and P30 DK26687) and Columbia Stem Cell Initiative Seed Fund Program. We thank the Fondation Recherche Medicale (ARF20140129235, L.B.). This work was strongly supported by the Helmholtz Alliance ICEMED & the Helmholtz Initiative on Personalized Medicine iMed by Helmholtz Association. This work was supported in part by the Helmholtz cross-program topic ‘Metabolic Dysfunction’, the European Research Council ERC (AdG HypoFlam no. 695054) and in part by funding to M.H.T., Y.L., B.L. and V.K. from the Alexander von Humboldt Foundation.