GLP1R agonists activate human POMC neurons

Drugs like Semaglutide (a.k.a. Ozempic/Wegovy) that activate the glucagon-like peptide-1 receptor (GLP1R) are a promising therapy for obesity and type 2 diabetes (T2D). Animal studies suggest that appetite-suppressing proopiomelanocortin (POMC) neurons in the arcuate nucleus of the hypothalamus are a likely target of these drugs, but the mechanisms by which they reduce food intake in humans are still unclear. We therefore generated POMC neurons from human pluripotent stem cells (hPSCs) to study their acute responses to GLP1R agonists by calcium imaging and electrophysiology. We found that hPSC-derived POMC neurons expressed GLP1R and many of them robustly responded to GLP1R agonists by membrane depolarization, increased action potential firing rate, and extracellular calcium influx that persisted long after agonist withdrawal and was likely mediated by L-type calcium channels. Prolonged administration of Semaglutide upregulated transcriptional pathways associated with cell survival in POMC neurons, and downregulated pathways associated with oxidative stress and neurodegeneration. These findings suggest that POMC neurons contribute to the long-term appetite-suppressive effects of GLP1R agonists in humans.


Introduction
Obesity affects over a billion people worldwide and significantly increases the risk of chronic diseases such as type 2 diabetes and cardiovascular disease (NCD Risk Factor Collaboration (NCD-RisC) 2024).It has a very strong genetic basis, and genetic variants associated with obesity predominantly act in the brain (Locke et al. 2015).Studies of individuals with severe, early-onset obesity revealed mutations in brain circuits that are essential for appetite regulation, such as the leptin-melanocortin system (van der Klaauw et al. 2019;S. Farooqi and O'Rahilly 2006;Loos and Yeo 2022).Specifically, populations of pro-opiomelanocortin (POMC) neurons in the arcuate nucleus of the hypothalamus increase their firing rate in response to the adipocyte-derived hormone leptin (Cowley et al. 2001;Qiu et al. 2018) and the predominantly gut-derived hormone glucagon-like peptide-1 (GLP-1) (Gabery et al. 2020;Secher et al. 2014;Dong et al. 2021;Péterfi et al. 2021).Activated POMC neurons then release more of the POMC-derived peptides α-melanocyte stimulating hormone (α-MSH) and β-MSH (Kirwan et al. 2018;Lee et al. 2006) that stimulate the melanocortin 4 receptor (MC4R) on neurons that suppress food intake and increase energy expenditure (Fenselau et al. 2017;Baldini and Phelan 2019;Cone 2005).Although other appetite-regulatory cell populations respond to GLP-1, including cells of area postrema (AP) (Kawatani, Yamada, and Kawatani 2018) and nucleus tractus solitarius (NTS) of the hindbrain (Hayes et al. 2011), the loss of either POMC or MC4R is sufficient to increase appetite and body weight in both humans (I. S. Farooqi et al. 2003;Yeo et al. 2003;Krude et al. 1998) and animals (Dittmann et al. 2024;Yaswen et al. 1999;Huszar et al. 1997), demonstrating that POMC neurons are essential for normal long-term body weight regulation.
The ability of GLP-1 to suppress appetite has recently been harnessed by modified peptide agonists of the GLP-1 receptor (GLP1R) that are revolutionizing the treatment of T2D and obesity (Meier 2012;Bessesen and Van Gaal 2018;Tan et al. 2022).GLP1R agonists delay gastric emptying and reduce food intake in humans to reduce body weight and improve metabolic health for long periods of time (Halawi et al. 2017).The efficacy of GLP1R agonist monotherapies such as Semaglutide (a.k.a.Ozempic, Wegovy) can be further improved by combining it with agonism of the gastric inhibitory polypeptide receptor (GIPR) and/or the glucagon receptor (GCGR) (Knerr et al. 2022;Coskun et al. 2022;Zhao et al. 2022), but some users experience side effects with these drugs, and their efficacy still falls short of what is seen with bariatric surgery.
To rationally design more effective treatments for obesity, it is important to understand the mechanisms by which existing treatments work in humans.GLP1R signaling has been well-studied in pancreatic β-cells (Shilleh et al. 2024), where its coupling to G αs stimulates adenylate cyclase to elevate the intracellular concentration of cyclic adenosine monophosphate, or [cAMP] i .This in turn activates protein kinase A (PKA) which increases the influx of extracellular Ca 2+ through L-type voltage-gated calcium channels to promote insulin secretion (Gomez, Pritchard, and Herbert 2002).GLP1R can also couple to β-arrestin upon ligand binding, which promotes its rapid internalization (Sonoda et al. 2008) and continued signaling from the endosome (Irannejad et al. 2013).In the nervous system, signaling pathways downstream of GLP1R are less well understood but likely also involve G αs coupling, since PKA becomes activated in NTS neurons (Hayes et al. 2011).Due to the inaccessibility of human neurons, endogenous GLP1R signaling pathways are unknown, and extrapolating insights from rodent studies is hampered by species-specific differences in appetite-regulatory cell types (Steuernagel et al. 2022;Tadross et al. 2023;Kirwan et al. 2018).
To address these limitations, we derived hypothalamic POMC neurons from human pluripotent stem cells (hPSCs), including a POMC-GFP knock-in reporter cell line that enabled their prospective identification in live cultures.We found that many human POMC neurons expressed GLP1R mRNA and were durably activated by GLP-1 and other GLP1R agonists including Exendin-4 derivatives, Semaglutide, and the dual GLP1R/GIPR agonist Tirzepatide (a.k.a.Mounjaro).Electrophysiological recordings and pharmacological studies suggested that POMC neurons were depolarized by GLP1R agonists for at least 20 minutes and fired substantially more action potentials via mechanisms that required functional L-type calcium channels.Prolonged stimulation with Semaglutide also induced significant transcriptional changes in human POMC neurons related to cell survival and neurodegenerative pathways.Together, these studies suggest that the activation of POMC neurons likely contributes to the appetite-suppressing effects of GLP1R agonist drugs in humans.

Human hypothalamic neurons express GLP1R
To facilitate the study of human POMC neurons, we differentiated them from a POMC-GFP knock-in cell line in which GFP is co-translationally expressed with the endogenous POMC gene (Chen, Yang, et al. 2023).This cell line allows live GFP+ POMC neurons to be prospectively identified and studied by fluorescent microscopy (Fig. 1A,B).To determine if these hPSC-derived neurons could be useful for the study of human GLP1R biology, we first analyzed single-cell transcriptomic data from five genetically distinct hPSC lines that were differentiated to hypothalamic cells (Chen, Yang, et al. 2023) (Fig. 1C).Whereas only a median of 1.1% of cells expressed GLP1R (UMI ≥ 1) across all clusters, GLP1R transcript was detectable at appreciable levels (>10% of cells) in cell clusters annotated as POMC neurons (2578/22892 cells, 11.3%), or OTP+ SST+ neurons (1424/12641 cells, 11.3%) (Fig. D,E).These findings suggest that POMC neurons are enriched in GLP1R expression.To further validate these results, we tested for GLP1R expression in the POMC-GFP cell line and found that 594/7634 (7.8%) of the GFP-expressing (UMI ≥ 1) cells in the POMC cell cluster also expressed GLP1R.
The observed GLP1R+ POMC neuron percentages are lower than the 37-100% of rodent POMC neurons reported to functionally respond to GLP1R agonists (Rønnekleiv et al. 2014;He et al. 2019;Péterfi et al. 2021;Secher et al. 2014).We reasoned that this discrepancy might be due to either the immaturity of hPSC-derived cells relative to their counterparts in the adult brain, or the inefficiency of transcript capture in single-cell data.Specifically, single-cell sequencing methods systematically underestimate the fraction of cells positive for genes expressed at low to moderate levels (Brennecke et al. 2013), such as G-protein coupled receptors like GLP1R.We therefore tested for Glp1r expression (UMI ≥ 1) in an integrated single-cell and single-nucleus RNAseq atlas of 384,925 mouse hypothalamic cells (Steuernagel et al. 2022).We found that mouse hypothalamic cell clusters annotated as Pomc neurons (C66-19) (Fig. 1F, arrow) or Sst neurons (C66-47) expressed Glp1r in 380/4655 (8.2%) and 479/5414 (8.8%) of cells, respectively, while the median percentage of Glp1r-expressing cells in other clusters was 1.1%.(Fig. 1G,H).These findings are similar to those seen in hPSC-derived hypothalamic cultures (Fig. 1E), suggesting that the modest fraction of GLP1R+ human and mouse POMC neurons we observed is likely attributable to inefficient mRNA capture in single-cell data rather than cellular immaturity.Together, these results demonstrate that hPSC-derived POMC neurons express GLP1R, and suggest that broad cell type-specific expression patterns of this gene are similar between the mouse and human hypothalamus.

GLP-1 activates human POMC neurons
The expression of GLP1R on human POMC neurons suggested that they might functionally respond to GLP-1 peptide and other GLP1R agonists.Since these agonists increase the electrical activity of rodent POMC neurons (Péterfi et al. 2021;Secher et al. 2014;Dong et al. 2021;Rønnekleiv et al. 2014;He et al. 2019;Singh et al. 2022), which raises their intracellular calcium concentration or [Ca 2+ ] i , we used calcium imaging to test for GLP1R agonist-induced calcium responses in human hypothalamic neurons derived from a POMC-GFP knock-in hPSC line.We loaded cultures of hypothalamic neurons derived from this cell line with a red calcium-sensitive dye (Fig. 2A), and imaged them on an upright microscope in a buffered physiological recording medium (HBSS) fed by a gravity perfusion system.All media and solutions contained synaptic blockers to isolate cell-autonomous responses from network activity.We selected cells for analysis if they exhibited stable baseline fluorescence and showed a clear increase in fluorescence (ΔF/F 0 ) in response to a depolarizing stimulus of 50 mM KCl (Fig. 2B-D), which we observed for the vast majority or recorded cells.These findings suggested that hPSC-derived hypothalamic neurons were electrophysiologically capable of responding to depolarising stimuli.
Next, we diluted GLP-1 peptide to a final concentration of 200 nM in recording media with synaptic blockers, since similar concentrations of GLP1R agonists (100 nM -1 µM) were previously used used in studies with rodent slice cultures (He et al. 2019;Péterfi et al. 2021;Secher et al. 2014;Gabery et al. 2020;Dong et al. 2021).After 10 minutes of recording to identify cells with stable baseline fluorescence, we added GLP-1 to cultures for two minutes followed by a wash-out period of 10-20 minutes and a terminal application of 50 mM KCl. Upon identifying hypothalamic neurons with stable baseline fluorescence and robust responses to KCl, we found that some such cells did not respond to GLP-1 (Fig. 2B), whereas others showed decreased (Fig. 2C) or increased (Fig. 2D) fluorescence intensity that persisted for tens of minutes, suggesting prolonged [Ca 2+ ] i changes indicative of inhibition or activation, respectively.
To categorize non-responsive, inhibited, and activated cells in an unbiased manner, we tested for significant differences between mean fluorescence intensities at 18 time points in a three-minute pre-stimulus window just before GLP-1 addition, and a three-minute post-stimulus window after GLP-1 wash-out, when responses in activated or inhibited cells had plateaued (Fig. 2E).We conservatively defined responsive cells as those having significant (P <0.01) changes in ΔF/F 0 , and either activated or inhibited depending on the direction of the response, whereas we defined cells whose response fell below this significance threshold non-responsive (Fig. 2F).Using these criteria, we found that the majority of GFP+ POMC neurons (6/10, 60%) were activated by 200 nM GLP-1.In contrast, few (25/154, 16%) GFP-neurons were activated and some (15/154, 10%) appeared to be inhibited (Fig. 2F).We then quantified the response magnitudes by calculating the post-stimulus area under the curve (AUC) normalized to the pre-stimulus AUC (Fig. 2G) calculated using the same time windows used for significance testing for both GFP-and POMC neurons (Fig. 2H).Since we observed both activated and inhibited populations among GFP-cells (Fig. 2F,H) but only activated POMC neurons, these results suggested cell type-specific responses.We therefore hypothesized that probing GLP1R signaling using biased agonists might shed light on the observed divergent responses.
Next, we hypothesized that cells should remain responsive to GLP1R ligands that do not effectively induce receptor internalization (e.g.Ex-F1), whereas cells should lose responsiveness to ligands that promote internalization (e.g.Ex-D3).We therefore exposed hypothalamic cultures to either 200 nM Ex-F1 or Ex-D3 for two minutes, washed ligands off for 10 minutes, and again exposed them to the same ligands for another two minutes and recorded cellular responses.We then compared ΔF/F 0 in three minute windows taken before and after the second ligand administration and found that 9/14 (64%) POMC neurons that had initially responded to Ex-F1 again responded to a second administration (Fig. 3A), whereas none of the seven (0%) POMC neurons exposed to Ex-D3 showed such a second response (Fig. 3D).These results suggest that either receptor internalization removes the pool of GLP1R necessary to generate responses detectable in our experimental system, or that these two ligands are engaging distinct intracellular signaling pathways.

Semaglutide depolarises human POMC neurons
Persistent increases in [Ca 2+ ] i could be caused by the Ca 2+ release from internal stores such as the endoplasmic reticulum, or from neuronal depolarization and the entry of extracellular Ca 2+ through calcium and non-selective cation channels.Electrophysiological recordings from hypothalamic mouse brain slices demonstrated that GLP-1 or Liraglutide stimulation depolarized POMC neurons in a manner that was dependent on the presence of GLP1R and that persisted for at least tens of minutes after agonist removal (Secher et al. 2014;Dong et al. 2021).We therefore hypothesized that the prolonged nature of the calcium signal we observed was due to the elevated action potential firing rate of POMC neurons.
To test this hypothesis, we performed electrophysiology of hPSC-derived GFP+ POMC neurons using a perforated patch preparation under a gravity-driven perfusion system in a buffer of HBSS.All cells in which we injected a modest amount of current (5 pA, 1 second) were able to fire trains of action potentials (3/3, 100%) and most cells (5/7, 71%) showed spontaneous action potentials in the absence of current injection, suggesting that hPSC-derived POMC neurons were healthy and capable of stimulus-evoked electrical activity.Next, we selected cells for analysis that had a membrane potential (Vm) under baseline conditions of -80 ± 15 mV, and that gave stable recordings for at least 10 minutes before and 20 minutes after administration of 200 nM Semaglutide for 6 minutes in HBSS, followed by a wash-out period in HBSS.We found that while the baseline activity of POMC neurons was low to moderate (Fig. 5A, see also E), their membrane potential and rate of action potential firing increased upon administration of Semaglutide (Fig. 5B), and persisted at elevated levels for at least 20 minutes after washout (Fig. 5C).We observe similar results from all cells tested (3/3, 100%) including both a significant (P<0.05)depolarization of their membrane potential by an average of 22.4 ± 3.8 mV (Fig. 5D), and a consistently increased rate of action potentials fired per minute (Fig. 5E), though firing activity varied from cell to cell.Together, these results suggest that the persistent elevation of [Ca 2+ ] i we observed in POMC neurons upon Semaglutide is likely caused by their increased electrical activity and the accompanying influx of extracellular calcium.

Neuronal responses to Semaglutide require L-type calcium channels
The persistent electrical activation of human POMC neurons in response to Semaglutide might increase [Ca 2+ ] i via voltage-gated calcium channels as described in pancreatic β-cells, where L-type calcium channels are engaged upon GLP1R agonist binding (Gomez, Pritchard, and Herbert 2002).We therefore examined scRNAseq data of hPSC-derived POMC and non-POMC neurons, and found that most calcium channels we tested for were expressed at some level, and POMC neurons were significantly enriched in CACNA1D relative to non-POMC neurons (Fig. 6A).To identify which calcium channels might contribute to the prolonged increase in [Ca 2+ ] i we observed, we treated cultures with 200 nM Semaglutide for two minutes to identify activated cells, waited for responses to plateau, and then added a cocktail of voltage-gated calcium channel (VGCC) blockers.These VGCC blockers included Benidipine (blocks L-type, N-type, and T-type channels), TTA-P2 (blocks T-type channels) and ω-Agatoxin (blocks P/Q-type channels) (Zamponi et al. 2015) (Fig. 6B).We found that upon adding these channel blockers, calcium indicator dye fluorescence dropped precipitously, sometimes below pre-stimulation baseline levels (Fig. 6C).These findings suggest that Semaglutide-induced increases in [Ca 2+ ] i require VGCCs, rather than being due to Ca 2+ release from internal stores.
To identify which channel blockers drove this effect, we repeated these studies with one candidate VGCC blocker at a time.We saw no significant drop in fluorescence intensity in Semaglutide-activated cells upon administration of either ω-Agatoxin (8 cells, Fig. 6D,E) or TTA-P2 (21 cells, Fig. 6F,G), but observed a significant (P<0.01) drop in fluorescence intensity in response to the L-type-specific blocker Benidipine (11 cells, Fig. 6H,I).To replicate these findings, we tested the effects of a second L-type calcium channel blocker, Nifedipine, on Semaglutide-activated cells and again observed a significant (P<0.0001) drop in fluorescence intensity (34 cells, Fig. 6J,K).Together, these findings suggest that L-type voltage-gated calcium channels are required for the sustained increase in [Ca 2+ ] i observed in human POMC neurons.

Long-term treatment with Semaglutide alters gene expression in POMC neurons
The sustained activation we observed in human POMC neurons resembled that seen in mouse POMC neurons in brain slices in response to GLP1R agonists (He et al. 2019;Péterfi et al. 2021;Secher et al. 2014;Gabery et al. 2020;Dong et al. 2021) that, on the time frame of hours to days, suppresses appetite due to increased MSH release (Baldini and Phelan 2019).We therefore wondered what transcriptional changes might accompany the sustained activation of human POMC neurons, and treated replicate cultures of hypothalamic cells differentiated from POMC-GFP reporter hPSCs with vehicle or Semaglutide for 20 hours.We then dissociated six replicate cultures per treatment group, purified POMC neurons by FACS, and analyzed their transcriptomes by bulk RNA sequencing.After adjusting for multiple comparisons, we found that Semaglutide induced significant (FDR<0.05)changes in gene expression, including 393 significantly down-regulated and 257 significantly up-regulated genes (Fig. 7A), of which 23 and 107 genes showed a |log 2 (fold-change)|>0.5, respectively.To gain insight into the pathways regulated by Semaglutide treatment in POMC neurons we performed gene ontology analysis of differentially expressed genes against the Kyoto encyclopedia of genes and genomes (KEGG) pathway database (Kanehisa et al. 2023).We observed a significant upregulation of pathways associated with cell survival and cancer (Fig. B), and a significant downregulation of pathways associated with oxidative stress and neurodegeneration, among others (Fig. C).
Upregulated genes included members of multiple signaling pathways including FGF (FGF8, FGF17, FGFR1, FGFRL1), WNT (FZD7, WNT5A) and Hedgehog (GLI2, GLI3), whereas downregulated genes included multiple members of a key mitochondrial ATP synthase (ATP5F1E, ATP5F1B, ATP5F1A, ATP5MF, ATP5PF), and vacuolar H+ pumps important for the acidification of intracellular organelles including synaptic vesicles, and receptor-mediated endocytosis (ATP6V0B, ATP6V0C, ATP6V0E2, ATP6V1B2, ATP6V1G1, ATP6V1G2, ATP6AP1).We also observed the differential expression of genes that may help sustain higher levels of neuronal activity, including increased expression of the voltage-gated L-type calcium channel CACNA1D and decreased expression of the inward rectifying potassium channel KCNJ12.Notably, CACNA1D is enriched in hPSC-derived POMC neurons and is a target of the L-type blockers we applied (Fig. 6C, H-K), Together, these pathways suggest that prolonged exposure to Semaglutide alters POMC neuron cell state, which may be relevant for their responses to long-term treatment with anti-obesity drugs targeting GLP1R.

Discussion
The remarkable success of GLP1R agonist drugs in humans follows a long history of research in animal models, where cell populations in the hindbrain and hypothalamus that mediate its appetite-regulatory effects have been identified (Müller et al. 2019).While histological and single-cell studies confirm that GLP1R is expressed in the human hypothalamus (Tadross et al. 2023), the accessibility of these cells has previously precluded their functional analysis.Here, we demonstrate that many human POMC neurons and some other hypothalamic neurons functionally respond to GLP1R agonists, but that the nature of these responses varies by both cell type and ligand.We identify a likely mediator of the persistent electrical excitability of POMC neurons, and describe the transcriptional targets of prolonged GLP1R activation in these cells.Below, we discuss the implications of these findings, remaining questions, and the strengths and limitations of our study.
We used single-cell sequencing to reveal cell populations that expressed GLP1R mRNA in the hPSC-derived cultures and mouse brain (Steuernagel et al. 2022).We found that the human POMC cell cluster had one of the highest percentages of GLP1R expressing cells in our dataset (11.3%), resembling results from mouse Pomc clusters (8.2%).Due to the relatively low sequencing coverage characteristic of single-cell sequencing and modest expression levels of GPCRs, these values likely underestimate the true percentage of POMC neurons that express functional levels of GLP1R protein.Indeed, we observed that the majority of POMC-GFP hPSC-derived POMC neurons responded to GLP1R agonists by calcium imaging.We considered cells to be POMC neurons if they had clearly detectable levels of GFP fluorescence, since GFP and POMC expression are tightly correlated at the protein level (Chen, Yang, et al. 2023).However, the identity of GFP-cells that responded to GLP1R agonists is unknown, other than that they were likely neuronal given their morphology and responsiveness to KCl.We were surprised to find that a small but highly reproducible fraction of these GFP-neurons appeared to reduce [Ca 2+ ] i upon administration of GLP1R agonists, as suggested by the sustained decrease in calcium indicator dye fluorescence.These findings suggest that signaling pathways downstream of GLP1R vary in a cell type-specific manner.
Electrophysiological analysis and pharmacological blockade of candidate channels suggested that the prolonged increases (at least 20 minutes) in action potential firing rate and increased [Ca 2+ ] i observed upon brief administration of these agonists was likely mediated by the increased electrical activity of POMC neurons and the associated influx of extracellular calcium through L-type calcium channels.L-type calcium channels can be indirectly modulated by GPCRs (Dolphin 2018) via the cAMP-dependent kinase PKA, which phosphorylates C-terminal serine residues (Mitterdorfer et al. 1996) to increase the likelihood of channel opening in some contexts, and accelerate recovery period (Cachelin et al. 1983).We speculate that the L-type channel CACNA1D (Ca V 1.3) may play a key role in the responses we observed as it is expressed at appreciable levels in hPSC-derived POMC neurons (~28 CPM based on bulk RNAseq data) and is significantly enriched relative to non-POMC neurons (Fig. 6A).Moreover, this channel is inhibited by Binidipine and Nifedipine, and is significantly upregulated transcriptionally in response to prolonged exposure to Semaglutide.
We also found that 20-hour Semaglutide treatment was associated with the significant downregulation of pathways associated with neurodegeneration.While these findings require validation and mechanistic follow-up, it has not escaped our attention that GLP1R agonists show neuroprotective and cognitive benefits to mice (During et al. 2003;Li et al. 2009;Panagaki, Gengler, and Hölscher 2018), and humans (Wu et al. 2018;Vadini et al. 2020), and are in numerous clinical trials for Alzheimer's Disease (Michael Gejl et al. 2017;M. Gejl et al. 2016;Egefjord et al. 2012), Parkinson's Disease (Aviles-Olmos et al. 2013, 2014;Hogg et al. 2022) and other neurological conditions (Mitchell et al. 2023;Glotfelty et al. 2019).The mechanisms by which GLP1R agonists exert their neuroprotective effects remain largely obscure (Reich and Hölscher 2022;Kopp et al. 2022), and our findings suggest that they could involve direct effects on neurons.
Finally, we note that the release of MSH by POMC neurons plays a dose-dependent and central role in body weight homeostasis (Biebermann et al. 2006;Baldini and Phelan 2019).Therefore, although other brain regions also likely contribute to the weight loss effects of GLP1R agonist drugs, the increased activity of POMC neurons upon acute administration of GLP1R agonists, together with the transcriptional changes following longer-term exposure, suggests that human POMC neurons are likely one of their mechanistic targets.

Limitations of study
Though we took care to ensure the reproducibility and relevance of our work, we acknowledge several limitations.First, much of our work was based on a single cell line (HUES9-POMC-GFP) analyzed at time points when cells were sufficiently mature to fire spontaneous action potentials and show clear calcium responses.Although POMC neurons from this cell line closely resemble POMC neurons derived from diverse genetic backgrounds at the transcriptional level (Chen, Yang, et al. 2023), it is possible that different results could have been obtained on different genetic backgrounds, maturation time points, or in different culture conditions.Furthermore, while we reduced the likelihood of recorded activity being driven by network activity with the use of a synaptic blocker cocktail in all experiments, we cannot formally exclude the possibility that cells communicated with each other via other mechanisms, such as gap junctions or paracrine signals.
We tested the hypothesis that L-type calcium channels are required for sustained GLP1R agonist-mediated activation of POMC neurons and found evidence supporting this hypothesis.However, other channels may also contribute.For example, in mouse hypothalamic hypocretin (HCRT) neurons, Exendin-4 induced an extracellular sodium-dependent and intracellular GDP-dependent depolarization and inward current in the absence of extracellular Ca 2+ consistent with G-protein dependent depolarization and the involvement of a non-selective cation channel (Acuna-Goycolea and van den Pol 2004).We found that brief administration of GLP1R agonists typically resulted in prolonged POMC neuron activation over a period of up to 20 minutes.In future studies, it would be interesting to determine the full duration of this increased activity, the kinetics by which [Ca 2+ ] i returns to baseline, and compare these kinetics to any changes in intracellular cAMP concentrations to study responses more proximal to putative GLP1R-G αs coupling.
Much attention has been paid to the heterogeneity of POMC neurons based on the genes they express, their anatomical location and connectivity, or by their functional responses to factors like leptin and GLP-1 as determined by electrophysiological recording or calcium imaging (Quarta et al. 2021).There is little agreement about the identity of these POMC sub-populations, or how identified mouse sub-populations might map onto the human counterparts.Specifically, differences seen between mouse and human datasets could be attributable to true species-specific differences, the maturation state of adult mouse versus hPSC-derived human cells, or technical factors such as sequencing depth or library quality.In order to draw robust conclusions about functionally relevant human POMC neuron heterogeneity, further functional and molecular characterization that is beyond the scope of the current study is necessary.This includes cAMP imaging and characterization of POMC neuron responses to a broad panel of candidate ligands, followed by confirmation in primary human brain tissue or in snRNAseq data.Finally, we did not consider non-hypothalamic cell populations in this study, and it would be of great interest to study other GLP1R-responsive human neuron types, such as NTS neurons.

Analysis of single-cell sequencing data
Processed single-cell RNA sequencing data obtained from five distinct hPSC lines differentiated to hypothalamic neurons (Chen, Yang, et al. 2023) as well as published data from HypoMap mouse single cell hypothalamus atlas (Steuernagel et al. 2022) were analyzed for the expression of GLP1R and other genes of interest using expression cutoffs of ≥1 UMI.

Preparation of drugs and chemicals
Peptides and small molecule drugs used in this study were reconstituted according to manufacturer instructions, aliquoted into single-use vials, and stored at -70°C or -20°C until the day of each experiment.

Calcium imaging
Calcium Imaging experiments were performed as previously described (Chen, Mazzaferro, et al. 2023).Briefly, hypothalamic progenitors differentiated from the HUES9-POMC-GFP cell line were replated onto 35 mm x 10 mm polystyrene imaging dishes (Corning) coated with 1 µg/cm 2 of iMatrix-511.All analyses were performed on neurons 40-50 days post-differentiation and transitioned from SynaptoJuice medium to BrainPhys medium 24-48 hours prior to recording.On the day of recording, neuronal cultures were loaded with Cal-590 AM fluorescent calcium indicator dye (Stratech) following manufacturers' instructions.An extracellular bath solution consisting of Hanks' Balanced Salt Solution without Phenol Red (HBSS; ThermoFisher Scientific) and synaptic blockers were used for all aspects of calcium imaging, including dye loading, perfusion, and dilution of GLP1R agonists and other experimental substances.Synaptic blockers were added to final concentrations of 100 µM DL-AP5, 50 µM Picrotoxin, 30 µM CNQX, and 20 µM Strychnine to inhibit NDMA, GABA, AMPA/Kainate, and glycine receptors, respectively.Immediately before each experiment, GLP1R agonists were diluted from stock solutions to final concentrations of 200 nM in unless otherwise indicated.
Imaging dishes were placed on an Olympus BX51WI Fixed Stage Upright Microscope (RRID: SCR_023069) and imaged using a 16-bit high-speed ORCA Flash4.0LT plus digital sCMOS camera (RRID: SCR_021971).Neurons were identified by Cal-590 AM fluorescence using an excitation wavelength of 540 nm and an emission wavelength of 590 nm.Images were taken using a 20x objective and acquired at a frequency of 0.1 Hz (100 ms exposure/frame) using a CoolLED pE-300 white illumination system (RRID: SCR_021073) and HCImage (RRID: SCR_015041) software for acquisition.Neurons were perfused continuously at room temperature (3 mL/min) using a gravity-driven perfusion system for the entire length of the experiment.For recordings, cells were perfused with extracellular bath solution for 10 minutes to record baseline fluorescence, and then perfused with extracellular bath solution containing the experimental compound for two minutes, followed by a second washout period (10-20 minutes) and stimulation with 50 mM KCl for two minutes followed by a washout period to identify responsive neurons.Videos of fluorescence intensity in the green (GFP) and red (Cal-590 AM) channels were recorded at an acquisition rate of 6 frames per minute.

Imaging and figure preparation
Images of hypothalamic cultures were collected on the calcium imaging microscope as described above, and separate cultures were plated onto PhenoPlateTM 96-well microplates (Perkin Elmer), fixed in 4% paraformaldehyde (Thermo Fisher Scientific) for 10 minutes, rinsed three times with Tris-buffered saline containing 0.1% Triton X-100 (TBS-T; Sigma-Aldrich), and incubated with primary antibodies including anti-αMSH (A1H5, detects epitopes in the αMSH region of POMC and developed by Professor Ann White; 1:5000) diluted in TBS-T containing 1% normal donkey serum (NDS; Stratech) overnight at 4 °C.Cells were washed three times with TBS-T and incubated with secondary antibodies diluted in TBS-T with 1% NDS for 2 hours at room temperature on an orbital shaker.Cells were washed three more times with TBS-T, and 300 μM of 4′,6-diamidino-2-phenylindole (DAPI; Thermo Fisher Scientific) was used to stain cell nuclei.
Image acquisition was performed using the Perkin Elmer Opera Phenix Plus High-Content Screening System with a 20x or 40x water objective.Some figure elements were prepared with the help of BioRender, and some photomicrographs were background-subtracted, adjusted for brightness and contrast, and false-colored to enable merging of channels and aid in visualization.Figures were prepared in Adobe Illustrator or Affinity Designer.

Quantification of responses from calcium imaging
To quantify changes in fluorescence intensity in GFP+ and GFP-neurons, regions of interest were manually drawn around candidate cells and raw fluorescence values from the 590 nm channel were exported to Microsoft Excel (RRID: SCR_016137) for each fluorescence time course.Only cells displaying a stable baseline prior to experimental compound administration and also a robust response to 50 mM KCl were considered for analysis.The change in fluorescence intensity as a function of time was expressed as (F − F 0 )/F 0 or ΔF/F 0 , where F was the measured fluorescence intensity and F 0 was the mean fluorescence intensity recorded during the 10 minute baseline period.Cellular responses to experimental compounds (e.g.GLP1R agonists) were quantified by comparing the change in the area under the curve (AUC) of normalized fluorescence (ΔF/F 0 ) intensity in a three minute time window before and after addition of experimental compounds.The post-stimulus window started two minutes after compound administration to account for a slow rise in fluorescence intensity followed by a plateau period.We considered the raw ΔF/F 0 values and did not normalize to KCl response amplitude.

Electrophysiology
Whole-cell recordings were obtained using the perforated Patch-Clamp technique as previously described (Goldspink et al. 2020).Briefly, microelectrodes were pulled from borosilicate glass (1.5 mm OD, 1.16 mm ID; Harvard Apparatus), and the tips were coated with refined yellow beeswax.Microelectrodes were fire-polished using a microforge (Narishige) and had resistances of 2-3 MΩ when filled with internal pipette solution.The internal pipette solution contained: 76 mM K 2 SO 4 , 10 mM NaCl, 10 mM KCl, 10 mM HEPES, 55 mM sucrose, and 1 mM MgCl 2 ; adjusted to pH 7.2 with KOH.Amphotericin B (10 μg/ml) dissolved in DMSO was added to the pipette solution on the day of recording to perforate cell membranes.A silver/AgCl wire connected to the bath solution via a 0.15 M NaCl agar bridge was used as a ground.
Once cells had been patched in this configuration (Lippiat 2008), recordings were acquired using an Axopatch 200B amplifier connected through a Digidata 1440A A/D converter and pCLAMP software (Molecular Devices).Spontaneous action potential firing was recorded in current-clamp mode without injecting current (I = 0).Neurons were perfused continuously at room temperature with HBSS solution (3 mL/min) in the presence of synaptic blockers using a gravity-driven perfusion system.For each experiment, we established a baseline recording period under perfusion, followed by perfusion with HBSS containing 200 nM Semaglutide.Membrane potential and the spontaneous action potential firing were recorded for 10-30 minutes before and after applying Semaglutide for six minutes.For each cell, mean membrane potential and action potential frequency was measured during 180 second-long time windows taken immediately before Semaglutide application, and 60-120 seconds after Semaglutide application, at which point Semaglutide induced depolarization had plateaued.

Semaglutide simulation and FACS purification of POMC neurons
To test for the transcriptional consequences of longer-term exposure to GLP1R agonists, HUES9-POMC-GFP hypothalamic cultures at 40 ± 2 days post-differentiation were cultured in N2B27 medium for 72 hours prior to treatment with vehicle (DMSO) or Semaglutide (2 µM, MedChem Express) in N2B27 medium for 20 hours.Cultures were then digested with TrypLE Express with 200 U/mL of papain and dissociated in wash media (N2B27) supplemented with 2 mg/mL DNAse I, 10 µg/mL Rock inhibitor, and 40 µg/mL actinomycin D. The cell suspension was centrifuged at 300 g for 5 min and washed with wash medium and centrifuged again.The resulting cell pellets were resuspended with 500 µL of sorting buffer (DPBS -/-with 1 mM EDTA, 25 mM HEPES, and 0.2% BSA) and transferred to DNA Lo-bind tubes for FACS sorting.Immediately prior to sorting, cells were filtered through a 40 µm Flowmi cell strainer and loaded onto the Aria Fusion Spyro set at 4°C and sorted using a 100 µm sort nozzle based on DAPI (0.1 µg/mL) for live/dead cells and GFP fluorescence.Sorted live cells were collected directly into RLT lysis buffer and RNA was immediately extracted using the RNeasy Plus Micro kit (Qiagen).
Bulk RNA sequencing and analysis RNA from sorted cells were analyzed on the Agilent 2100 Bioanalyser system using the Eukaryote Total RNA Pico kit (Agilent Technologies) and the mean RIN number was 9.5.Libraries of cDNA were then generated using the SMARTer Stranded Total RNA-Seq Kit v3 -Pico Input Mammalian (Takara).The resulting cDNA libraries were then quantified by the High Sensitivity DNA assay, pooled, and sequenced to a targeted depth of 50M reads/sample on Illumina flow cells to generate 100bp paired-end reads.
The resulting RNA sequencing libraries were pre-processed with TrimGalore (v0.6.7)(Krueger et al. 2021) to remove adapters and filter of low-quality reads before aligning them to the human (GRCh38, Ensembl release 98) reference genome using using STAR (v2.7.9a) (Dobin et al. 2013).Read counts per gene were then quantified using featureCounts (v2.0.0) (Liao, Smyth, and Shi 2014) against the human (GRCh38, gencode v32) annotation.Qualimap (v2.2.2d) (Okonechnikov, Conesa, and García-Alcalde 2015) was used to generate QC metrics of aligned reads for evaluating sample library quality, together with QC metrics produced by STAR and featureCounts.Size factor normalization and removal of lowly expressed genes was performed using edgeR.Further quality control analysis of the samples were performed using principal component analysis and expression correlation analysis to identify any outlier samples.For differential expression analysis, estimation of both negative binomial dispersions and quasi-likelihood dispersion were first performed using the edgeR package, followed by a quasi-likelihood F-test to identify genes which are differentially expressed between different treatment groups, with significance threshold set to FDR < 0.05.Finally, gene set enrichment analysis of the differentially expressed genes were performed using the g:Profiler R package (Raudvere et al. 2019), with custom background set to expressed gene list, analytical-adjusted p-value threshold set to 0.05 and data sources set to KEGG.

Statistical analysis
Graphing and statistical analysis were performed using GraphPad Prism (version 7.02; RRID: SCR_002798).For all one-way analyses of variance (ANOVA), data were first analyzed for normality using the D'agostino Pearson test.For all instances in which multiple paired t-tests were used, data were first analyzed for normality using the D'agostino Pearson test and results were presented as volcano plots with a P value < 0.01 adjusted by Holm-Sidak method.All the test data were executed with Welch's correction, which does not assume equal SD.For frequency analysis, data were organized in contingency tables and differences were tested by chi square test.Results were presented as mean with 95% CI for all calcium imaging experiments and mean ± standard error of the mean (SEM) for all other experiments.After adjusting for multiple comparisons, P values < 0.05 were considered as statistically significant.

Data, code, and materials availability
Upon publication in a peer-reviewed journal, sequencing data will be made available on ENA, and code used for analysis and generating figures will be made available in GitHub.Reagents will be made available upon request from the lead author (F.T.M.).

Figure 1 .
Figure 1.Expression of GLP1R mRNA in human POMC neurons.A,B) Hypothalamic cells differentiated from human pluripotent stem cells (hPSCs) contain a significant number of POMC neurons as visualized by immunostaining for aMSH (A) and in live cultures derived from a POMC-GFP reporter cell line (B).Scale bars represent 25 and 12 µm in A and B, respectively.C) UMAP of single-cell RNAseq data from five genetically distinct hPSC lines differentiated into hypothalamic neurons, separated by cells annotated as being in POMC (dark blue) or non-POMC (light blue) clusters.D,E) Expression of GLP1R mRNA in hPSC-derived hypothalamic neurons (UMI ≥ 1), color-coded by expression level (D) and quantified for POMC or SST cell clusters, and median expression across all cell clusters (E).F) UMAP with 66 annotated clusters from mouse single-cell and single-nucleus RNAseq data, with the arrowhead indicating the Pomc-expressing cell cluster.G,H) Expression of Glp1r mRNA in mouse hypothalamic neurons (UMI ≥ 1), color-coded by expression level (G) or quantified across Pomc or Sst clusters, or the median across all cell clusters.

Figure 2 .
Figure 2. Responses of human hypothalamic neurons to GLP1.A) Photomicrographs from a POMC-GFP reporter cell line showing (from left to right) neuronal morphology in brightfield, baseline fluorescence from the Cal-590 AM calcium sensitive dye, endogenous GFP fluorescence from the reporter, and a merge of these images.B-D) Representative traces from single neurons in response to administration of 200 nM GLP-1 for two minutes showing no response (B), decreased fluorescence (C), a prolonged increase in fluorescence (D), where the Y axis represents ΔF/F 0 .E) Representative trace showing how GLP1R responsive cells were classified by calculating whether values for ΔF/F 0 taken in a baseline period were significantly different from those taken after GLP1R agonist administration.F) Plot of the significance of response probability (Y axis) versus the magnitude of response (ΔF/F 0 ), where GFP+ POMC neurons are plotted in green, and a cutoff of P<0.01 was used to assign cells into inhibited (blue), activated (red), and non-responsive (gray) categories.Significance was calculated by multiple paired t-test from 18 pre-stimulus and 18 post-stimulus fluorescence intensity values.G) Representative trace showing how the magnitude of response to GLP1R was calculated by taking the area under the curve (AUC) in three-minute time intervals before and after its administration.H) Summary of the activated and inhibited cell responses for GFP-and POMC neurons.Non-responsive neurons are not shown.

Figure 3 .
Figure 3. Neuronal responses to GLP1R 'biased' agonists.A) Representative calcium imaging trace of a cell responding to a two-minute administration of Ex-F1, and then responding to a second administration given 10 minutes later.B) Distribution of the significance and magnitude of response to Ex-F1 across all analyzed cells.C) Response magnitude to Ex-F1 in activated and inhibited GFP-and POMC neurons.D-F) As in A-C above but for Ex-D3, for which calcium responses to a second administration after a 10 minute delay were not observed.Significance calculated by Welch's t test .

Figure 4 .
Figure 4. Effects of anti-obesity drugs on human hypothalamic neurons.A-C) Representative calcium imaging trace of a cell responding to 200 nM GLP-1 peptide (A), the classification of neuronal responses into activated, inhibited and non-responsive as described in Figure 2, and the fraction of GFP-and POMC neurons in each of these categories (B), comparison of responses between GFP-and POMC neurons (C).D-I) Panels as in A-C, but for 200 nM Semaglutide (D-F), or 200 nM Tirzepatide (G-I).Significance calculated by chi-squared test in B, E, H, and Welch's t test in C, F, I, where *p<0.05,**p<0.001.

Figure 5 .
Figure 5. Electrophysiological analysis of POMC neuron responses to Semaglutide.A-C) Representative trace of a hPSC-derived POMC neuron under perforated patch preparation that is relatively quiescent under baseline conditions (A, see also E) but is depolarized upon administration of 200 nM Semaglutide leading to an increased rate action potential firing (B) that persists for up to at least 20 minutes after Semaglutide withdrawal (C).D) POMC neuron Vm shifts to a significantly (P<0.05)depolarised state after addition of 200 nM Semaglutide (Post-Sema).N=3 cells.Significance calculated by paired t test.E) All tested POMC neurons show an increase in average action potential (AP) firing rate per minute in response to 200 nM Semaglutide (Post-Sema).

Figure 6 .
Figure 6.GLP1R agonist-induced calcium responses require voltage-gated calcium channels.A)Normalized expression levels (low to high = yellow to dark green) of candidate genes from scRNAseq data of hPSC-derived hypothalamic POMC neurons (top row), or non-POMC neurons (bottom row), where dot size indicates the fraction of cells in each population in which the candidate gene was detected (UMI ≥ 1).POMC, GLP1R, and CACNA1D are significantly enriched in POMC neurons.B) Tableof voltage-gated calcium channel (VGCC) gene names, channel type, and the identity of drugs that selectively inhibit them.C) In a Semaglutide-activated cell, a cocktail of VGCC blockers abolishes agonist-induced fluorescence of the calcium indicator.D,E) Administration of the P/Q-type VGCC blocker ω-Agatoxin does not significantly decrease Semaglutide-induced calcium indicator fluorescence, as seen in a representative trace (D) and in a summary of all Semaglutide-activated cells (E).F,G) Panels as in D,E but with the T-type VGCC blocker TTA-P2.H,I) Panels as in D-G, but with the L-type VGCC blocker Benidipine, which abolished Semaglutide-induced calcium dye fluorescence, as observed with the drug cocktail in (C).J,K) Panels as in D-I, showing that the results from Benidipine (H,I) are replicated with Nifedipine, a second L-type calcium channel blocker.*p<0.05,**p<0.01,***p<0.001****p<0.0001by one-way ANOVA with repeated measures in E, G, I, K.

Figure 7 .
Figure 7. Transcriptional effects of GLP1R activation in POMC neurons.A) Volcano plot of significantly differentially expressed genes between vehicle-and Semaglutide-treated POMC neurons, with a cutoff of FDR<0.05,B,C) KEGG enrichment of analysis of significantly (FDR <0.05) upregulated (C), or downregulated (D) genes.