Scalable hypothalamic neuron differentiation from human pluripotent stem cells suitable for modeling metabolic disorders Authors: Vukasin M. Jovanovic, Narisu Narisu, Lori L. Bonnycastle, David Castellano, Seungmi Ryu, Ravi Tharakan, Kendall T. Mesch, Qiang Chen, Fahrünisa Meryem Betül Erol, Hannah J. Glover, Tingfen Yan, Neelam Sinha, Chaitali Sen, Shu Yang, Dvir Blivis, Daniel F. Bennett, Giovanni Rosales-Soto, Jason Inman, Pinar Ormanoglu, Christopher A. LeClair, Natalie D. Shaw, Menghang Xia, Martin Schneider, Erick O. Hernandez-Ochoa, Michael R. Erdos, Anton Simeonov, Shuibing Chen, Francis S. Collins, Claudia A. Doege, Carlos A. Tristan Published: 2026 ABSTRACT SummaryThe hypothalamus, composed of multiple nuclei, is essential for maintaining the body’s homeostasis. Within the mediobasal hypothalamus, the arcuate nucleus (ARC) contains key neuronal populations, including appetite-suppressing pro-opiomelanocortin (POMC) neurons that regulate energy and glucose balance. Here, we present a chemically defined, scalable method for differentiating human pluripotent stem cells (hPSCs) into hypothalamic neurons enriched for POMC cells, compatible with robotic cell culture platforms for high-throughput use. Neuronal identity was validated by MERFISH single-cell transcriptomics, RNA-Seq, ATAC-Seq, and comparison to human hypothalamus. The method is robust across multiple hPSC lines, showing consistent induction of ventral diencephalon and hypothalamic markers. Derived neurons display metabolic disease-relevant features, including body mass index (BMI)-associated gene enrichment, and ATAC-Seq identifies potential candidate regulatory regions linked to hypothalamic development and metabolic traits. Functional assays reveal neuronal responses to insulin and the GLP-1 receptor agonist Exendin-4, and transcriptional responses to altered glucose conditions. This platform delivers a physiologically relevant model of human hypothalamic neurons that enables deeper mechanistic and therapeutic studies of metabolic disease. Introduction In the mid-1930s, Ernst and Berta Scharrer first proposed that the hypothalamus contains neurosecretory cells (Sarnat, 1983). Numerous physiological functions are regulated by the hypothalamus, such as the maintenance of body temperature, the regulation of hormone secretion, and the circadian rhythm-based sleep-wake cycle (Ranson, 1937). The arcuate nucleus (ARC) of the hypothalamus regulates energy balance, reproduction, and neuroendocrine control of growth hormone and prolactin release (Woods and D'Alessio, 2008; De Rosa et al., 2021). Thus, impaired function of ARC neurons contributes to metabolic disorders such as obesity and T2D (Jais and Bruning, 2022; McNay et al., 2012; Seeley and Tschop, 2006), as well as reproductive dysfunction (Silva et al., 2019; McIlwraith and Belsham, 2020). During early brain development, the ARC emerges from the ventral diencephalon and is marked by the expression of specific transcription factors, such as the homeodomain transcription factors retina and anterior neural fold homeobox (RAX) and NK2 homeobox 1 (NKX2.1) (Xie and Dorsky, 2017; Szabo et al., 2009; Shimogori et al., 2010; Blaess et al., 2014; Bedont et al., 2015; Yoo and Blackshaw, 2018; Diaz and Puelles, 2020; Placzek et al., 2025). Neurogenesis unfolds through the transcriptional activation of achaete-scute family bHLH transcription factor 1 (ASCL1), neuronal differentiation 1, neurogenin 3, distal-less homeobox 1/2/5 (DLX1,2,5), aristaless-related homeobox (ARX), orthopedia homeobox (OTP), and ISL LIM homeobox 1 (ISL1) as these progenitors differentiate and transition into neuroblasts (Xie and Dorsky, 2017; Lee et al., 2016, 2018; Kim et al., 2020; Nasif et al., 2015). Subsequently, distinct neuronal populations form within the ARC. Among them are the anorectic pro-opiomelanocortin (POMC) neurons (Hael et al., 2020; Rojo et al., 2024; Yu et al., 2024), which are essential for regulating feeding behavior and energy balance (Quarta et al., 2021). POMC neurons respond to postprandially secreted hormones such as insulin, elevated levels of circulating glucose, and the adipose secreted hormone leptin, all of which result in appetite suppression. In contrast, orexigenic agouti-related peptide (AgRP) neurons within the ARC, specified later in hypothalamic development (Zhou et al., 2022; Herb et al., 2023; Padilla et al., 2010; Yu et al., 2022), counteract activation of POMC neurons to promote appetite. Additional ARC neuronal populations include gonadotropin-releasing hormone (GnRH) neurons (Lehman et al., 2010), growth hormone releasing hormone (GHRH) also known as somatocrinin neurons, which stimulate growth hormone release from the pituitary (Balthasar et al., 2003), kisspeptin (KISS1) neurons that stimulate GnRH neurons (Plant, 2019), somatostatin (SST) neurons that regulate GHRH release (Mano-Otagiri et al., 2006) and tuberoinfundibular dopaminergic neurons that regulate prolactin release (Ben-Jonathan, 1985; Zhang and van den Pol, 2015), as well as several neuronal subtypes, with unique combinations of neuropeptides and transcription factors, whose functions have yet to be described (Zhou et al., 2022; Herb et al., 2023; Siletti et al., 2023; Tadross et al., 2025). Human hypothalamic tissue in living individuals is inaccessible for direct functional interrogation. Thus, the ability to efficiently and reliably produce large quantities of human pluripotent stem cell (hPSC)-derived hypothalamic neurons in vitro would provide a valuable resource for investigating the molecular causes of obesity, T2D, and hypogonadotropic hypogonadism, and rare neuroendocrine syndromes associated with the hypothalamus (Cassidy et al., 2012; Ansley et al., 2003). However, progress has been hindered by the limited scalability of previously published protocols for differentiating hPSCs into hypothalamic neurons (Rajamani et al., 2018; Merkle et al., 2015; Wang et al., 2015, 2021; Miwata et al., 2023; Yao et al., 2017; Pahl et al., 2021; Ogawa et al., 2018; Kasai et al., 2020; Ozaki et al., 2022; Taga et al., 2023). As a result, producing well-characterized hypothalamic neurons at scale in a reproducible and robust manner is necessary to enable the development of suitable models for translational applications. In this study, we report an efficient, scalable, robust, and controlled approach for the differentiation of hPSCs into hypothalamic neurons, featuring predominantly POMC neurons. These neurons were comprehensively validated on a cellular and molecular level. They display the expression of pertinent hypothalamic markers, as well as functional response to humoral and hormonal stimuli. The scalability of this approach is demonstrated by adapting the protocol to an established robotic cell culture platform (Tristan et al., 2021; Jovanovic et al., 2023). The robustness of this approach is shown by highly reproducible results across several distinct hPSC lines, and the translational applicability is documented by the generation of dose-response curves in targeted drug challenges. The data presented in this study will serve as a valuable reference resource, and the protocol will facilitate at-scale acquisition of previously limited patient-derived cell types for mechanistic studies of hypothalamus-associated disorders, as well as genomic and drug screening campaigns. Results A highly efficient and robust method for large-scale automated production of human hypothalamic neurons Human PSC lines were routinely cultured under feeder-free, chemically defined conditions (E8 medium) and vitronectin as a coating substrate. We utilized the CompacT SelecT robotic cell culture platform previously established at the National Center for Advancing Translational Sciences (NCATS) (Tristan et al., 2021). This platform now consists of three robotic cell culture systems that, when combined, allow for the simultaneous expansion of up to 270 hPSC lines in a fully automated manner, including media changes and cell passaging (Figure 1A) (Tristan et al., 2021). To ensure stress-free cell expansion, the hPSCs underwent EDTA-mediated passaging, followed by treatment with the CEPT small molecule cocktail for 24 h to enhance cell viability (Chen et al., 2021). As part of the pre-differentiation quality control pipeline, all hPSC lines underwent daily live cell imaging using an integrated imaging system (IncuCyte). This enabled daily monitoring of hPSC colony morphology and verification of the absence of spontaneous differentiation. Following routine robotic cell culture, hPSC lines were 1) karyotyped to confirm that chromosomal abnormalities did not develop during cell maintenance, 2) tested for mycoplasma contamination, and 3) differentiated into endoderm, mesoderm, and ectoderm to confirm their multi-lineage differentiation potential (Figure 1B).Figure 1Automated pipeline for routine hPSC culture, quality control, and differentiation into hypothalamic neurons(A) The CompacT SelecT automated cell culture system can accommodate up to 90 flasks and perform media changes, cell passaging, seeding, and live imaging. Our robotic platform, consisting of three CompacT SelecT systems, can manage up to 270 flasks simultaneously.(B) Quality control pipeline for hPSCs routinely cultured in CompacT SelecT, including live imaging, karyotype analysis, mycoplasma, and multilineage differentiation potential testing.(C) Schematic overview of the differentiation protocol with phase contrast images shows the appearance of cells differentiated using the robotic platform at days 0, 7, 15, 21, and 28, and Hypospheres at day 14.(D) Overview of the automated differentiation process using the CompacT SelecT robotic cell culture system with T175 flasks. Note that cells are cryopreservable at day 14. Starting with 3.5 million hPSCs per flask and processing 3 flasks during the sphere stage, the automated differentiation results in approximately 100 million hypothalamic neurons within 28 days. With the full integration of our robotic platform, this process can be scaled up by a factor of 90.(E) Hypothalamic neurons are available “off-the-shelf” and require thawing in Hypo-3 media with CEPT and 14 days of maturation before assaying.(F–G) Representative images of day 7 progenitors derived from LiPSC-GR 1.1 iPSC line immunostained for the ventral diencephalic markers NKX2.1 (F) and RAX (G).(H) High-content microscope whole-well image (24-well plate) composed of 89 high-magnification images shows robust POMC protein expression in day 28 cultures.(I) Representative images of day 28 hypothalamic neurons immunostained against POMC (red) and HOECHST (blue).(J) Bar graph displays unbiased quantification of POMC relative to Hoechst (nuclei) using Columbus software, across three independent hPSC lines (LiPSC-GR 1.1, NCRM5; iPSC, and WA09; hESC).(K) Western blot analysis of TUBB3, POMC, and GAPDH protein expression in samples derived from the time-course of hypothalamic neuron differentiation.Scale bars 50 μm (C and I), 100 μm (F and G) and 500 μm (H; entire well). Automated pipeline for routine hPSC culture, quality control, and differentiation into hypothalamic neurons (A) The CompacT SelecT automated cell culture system can accommodate up to 90 flasks and perform media changes, cell passaging, seeding, and live imaging. Our robotic platform, consisting of three CompacT SelecT systems, can manage up to 270 flasks simultaneously. (B) Quality control pipeline for hPSCs routinely cultured in CompacT SelecT, including live imaging, karyotype analysis, mycoplasma, and multilineage differentiation potential testing. (C) Schematic overview of the differentiation protocol with phase contrast images shows the appearance of cells differentiated using the robotic platform at days 0, 7, 15, 21, and 28, and Hypospheres at day 14. (D) Overview of the automated differentiation process using the CompacT SelecT robotic cell culture system with T175 flasks. Note that cells are cryopreservable at day 14. Starting with 3.5 million hPSCs per flask and processing 3 flasks during the sphere stage, the automated differentiation results in approximately 100 million hypothalamic neurons within 28 days. With the full integration of our robotic platform, this process can be scaled up by a factor of 90. (E) Hypothalamic neurons are available “off-the-shelf” and require thawing in Hypo-3 media with CEPT and 14 days of maturation before assaying. (F–G) Representative images of day 7 progenitors derived from LiPSC-GR 1.1 iPSC line immunostained for the ventral diencephalic markers NKX2.1 (F) and RAX (G). (H) High-content microscope whole-well image (24-well plate) composed of 89 high-magnification images shows robust POMC protein expression in day 28 cultures. (I) Representative images of day 28 hypothalamic neurons immunostained against POMC (red) and HOECHST (blue). (J) Bar graph displays unbiased quantification of POMC relative to Hoechst (nuclei) using Columbus software, across three independent hPSC lines (LiPSC-GR 1.1, NCRM5; iPSC, and WA09; hESC). (K) Western blot analysis of TUBB3, POMC, and GAPDH protein expression in samples derived from the time-course of hypothalamic neuron differentiation. Scale bars 50 μm (C and I), 100 μm (F and G) and 500 μm (H; entire well). To enable large-scale production of hypothalamic neurons from hPSCs, we developed a new hPSC differentiation protocol for the efficient, controlled, reproducible, and automated generation of human hypothalamic-like neurons (Figure 1C). In this protocol, on day −1, 20,000 hPSCs/cm2 were seeded on vitronectin in E8 medium with CEPT for 24 h. The following day (day 0), differentiation into ventral diencephalon progenitors was initiated. We previously reported that hPSCs exposed to recombinant protein modulators of JAK/STAT and NOTCH signaling in addition to BMP inhibition (LDN193189 + JAG1 + DLL1 + LIF + CNTF) directly differentiated into radial glial cells expressing fatty acid binding protein 7 (FABP7) and paired box 6 (PAX6), the latter being a marker of dorsal neural tube progenitors (Georgala et al., 2011; Jovanovic et al., 2023). Here, to identify culture conditions that promote the generation of ventral diencephalon progenitors, we assessed FABP7 and PAX6 expression on day 7 after treatment with TGF-β inhibitor A83-01, WNT inhibitor XAV939 and sonic hedgehog (SHH) activator SAG21K in four distinct conditions: LDN193189 + A83-01 (LA) (Figure S1A), LDN193189 + JAG1 + DLL1 + LIF + CNTF (A1) (Figure S1B), LA + XAV939 + SAG21K (Figure S1C), or A1 + XAV939 + SAG21K (Figure S1D). In agreement with directing the cells toward the ventral fate, we observed a decrease of PAX6 expression in both treatment groups containing XAV939 and SAG21K small molecules, while FABP7 expression was maintained, indicating that cells were directed toward a neural progenitor stage and ventralized under these culture conditions. To ascertain the robustness of this culture condition, these findings were further validated by western blot in two independent hPSC lines (WA09 human embryonic stem cells (hESCs) and LiPSC-GR 1.1 hiPSC (Figure S1E). As previously described (Jovanovic et al., 2023), A1 treatment induced strong expression of FABP7 protein in both cell lines (Figure S1E). Importantly, PAX6 mRNA expression after Hypo-1 medium treatment was also not observed in an orthogonal mRNA-seq time-course experiment, while RAX, an essential regulator of hypothalamus specification (Orquera et al., 2016), was strongly induced at day 7 (Figures S1F and S1G). Thus, Hypo-1 medium, with E6 medium as the base, was supplemented with BMP inhibitor LDN193189, TGF-β inhibitor A83-01, WNT inhibitor XAV939, and SHH activator SAG21K (see STAR Methods for concentrations and sources). DLL1, JAG1, LIF, and CNTF (see STAR Methods for concentrations and sources) were included to support neural induction and patterning into ventral diencephalon radial glia progenitors. Cells were kept in Hypo-1 medium, with daily medium changes, for 7 days to reach a fully confluent monolayer (Figure 1C). To further confirm ventral diencephalic identity, we performed RAX and NKX2.1 immunostaining on day 7 progenitors and found that nearly all cells across three independent lines expressed these markers (Figures 1F, 1G, S1H, and S1I). On day 7, ventral diencephalon progenitors were dissociated and transferred into AggreWell plates (see STAR Methods for details) for Hyposphere formation. After 24 h, Hypospheres were cultured in suspension in Hypo-2 medium composed of DMEM/F12 containing N2, B27 without vitamin A, BDNF, GDNF, ascorbic acid, and db-cAMP. Since WNT inhibition and Activin/Nodal signaling have been associated with hypothalamic development (Kapsimali et al., 2004; Mathieu et al., 2002), Hypo-2 medium was also supplemented with XAV939 and Activin A (see STAR Methods for exact concentration and sources). Hypospheres were maintained as free-floating spheres to maximize neurogenesis (Figures 1C and 1D) with daily Hypo-2 medium changes, until day 11 (Figure 1C). On day 11, Hypo-2 medium was replaced with Hypo-3 medium; a standard neuronal maturation medium comprised of DMEM/F12 with N2 and complete B27, supplemented with BDNF, GDNF, ascorbic acid, db-cAMP, with the addition of gamma-secretase inhibitor dibenzazepine (DBZ) (Lasagni et al., 2010). Spheres were dissociated on day 14, and cells were further matured in a monolayer on Geltrex-coated vessels in Hypo-3 medium for an additional two weeks, with medium changes every other day (Figure 1C). On day 28, approximately half of the cells expressed POMC (Figures 1H and 1I) as quantified using immunocytochemistry (ICC) across three independent lines (56.3% for LiPSC-GR 1.1, 48.1% for NCRM5, and 46.18% for WA09) (Figure 1J). To further validate neuronal and POMC neuropeptidergic identity of the derived cells, we performed time-course western blot experiments and detected the induction of POMC at day 14 that correlated with the onset of expression of beta III tubulin (TUBB3) (Figure 1K). The vast majority of cells expressed the neuronal marker microtubule-associated protein 2 (MAP2) and inhibitory marker gamma-aminobutyric acid (GABA) (Figure S1J). ISL1 is known to be expressed in the developing hypothalamus immediately before the onset of POMC expression, and from then on, ISL1 and POMC are co-expressed (Nasif et al., 2015). Accordingly, day 28 hypothalamic neurons derived from three independent hPSC lines, exhibited on average 80 percent of cells with ISL1 expression (78.11% for LiPSC-GR 1.1, 82.85% for NCRM5, and 80.87% for WA09; Figures S1K and S1L). The ventral hypothalamus contains different neuronal subpopulations (Campbell et al., 2017; Steuernagel et al., 2022; Romanov et al., 2017, 2020; Kim et al., 2020; Herb et al., 2023), including tuberoinfundibular dopaminergic neurons (Zhang and van den Pol, 2015), and a small population of SST-expressing neurons that play a role in regulating growth hormone release (Mano-Otagiri et al., 2006). High-content, whole-well imaging of neuronal cultures derived from three independent hPSC lines (Figure S1M) demonstrated the presence of tyrosine hydroxylase (TH)/TUBB3, and SST/TUBB3 co-expressing subpopulations (Figure S1M). Unbiased high-content imaging analysis allowed quantification of Hoechst+ nuclei and specific markers of interest (Figure S1N). This analysis demonstrated consistency in the type and number of additional neuronal subpopulations expressing specific markers across three independent hPSC lines, with dopaminergic TH+ neurons ranging from 12 to 15%, and SST+ neurons ranging from 3 to 6% when normalized to Hoechst+ nuclei (Figures S1O and S1P). Given the reproducibility and robustness of this differentiation protocol across multiple hPSC lines, we reasoned that fully automating the entire process on a non-stop robotic cell culture platform for scale-up and biomanufacturing could provide an unlimited source of hypothalamic neurons (Figure 1D). To this end, this protocol was automated using the CompacT SelecT platform, where 3.5 million hPSCs were differentiated per T175 flask, followed by daily automated Hypo-1 medium changes at exact 24 h intervals. At day 7, the cells were dissociated and manually transferred into AggreWell plates for Hyposphere formation, followed by transfer into T175 flasks and imported back into the automated system for maintenance with Hypo-2 medium changes at 24 h intervals. At day 14, hypothalamic neuroblasts can be dissociated and cryopreserved or undergo a 2-week maturation process in the Hypo-3 medium before assaying (Figures 1D and 1E). Upon dissociation, Hypospheres from each T175 flask yielded up to 35 million cryopreservable neuroblasts, available for “off-the-shelf” downstream applications such as high-throughput chemogenomic screens (Figure 1E). Since all experiments in this study were conducted using “off-the-shelf” hypothalamic neurons cryopreserved at day 14, we assessed post-thaw viability across three independent batches from two hPSC lines. The neuroblasts consistently exhibited post-thaw viability ranging from 60 to 80% (Figure S2A). In addition, to assess the consistency of neuronal quality across wells, we performed live imaging of neuroblast maturation in a 96-well plate over a 10-day period post-thawing and observed no “line-to-line” or “well-to-well” differences in morphology (Figures S2B–S2D). Furthermore, neurite outgrowth analysis demonstrated consistent results for both neurite length (Figure S2E) and branching points (Figure S2F) between neurons derived from two independent hPSC lines. Transcriptomic analysis confirms hypothalamic identity of differentiated neurons In vivo development of the human hypothalamus is guided by complex transcription factor programs that shape the molecular architecture of the region into specific nuclei populated by distinct neuronal populations (Xie and Dorsky, 2017). NKX2.1 and RAX-positive hypothalamic progenitors undergo basic-helix-loop-helix transcription factor ASCL1-mediated neurogenesis to transition into postmitotic ventral hypothalamus neuronal precursor subpopulations. Further specification of these postmitotic precursors into ventral hypothalamus neuronal subpopulations is driven by the combinatorial induction of transcription factors, including T-box transcription factor 3 (TBX3), ISL1, OTP, and DLX1/2/5 (Ma et al., 2021; Nasif et al., 2015; Hael et al., 2020; Rojo et al., 2024; Yu et al., 2024; Quarta et al., 2019). To capture the stepwise transcriptome changes throughout the differentiation of hPSCs into hypothalamic neurons, we profiled a total of 44 RNA samples isolated from three independent hPSC lines (LiPSC-GR 1.1, NCRM5, and WA09) at five stages of differentiation (days 0, 7, 14, 21, and 28). Principal component analysis (PCA) showed clustering of samples according to the time points profiled during differentiation (Figure 2A), demonstrating the reproducibility of our protocol and automated pipeline. We used the top upregulated genes at each time point to map the stepwise acquisition of hypothalamic identity throughout the in vitro differentiation process for all three hPSC lines tested. The hypothalamus specification of hPSCs was reflected by the early induction of RAX in day 7 progenitors, followed by an upregulation of ASCL1, DLX1/2 on day 14, and finally by day 28, strong expression of POMC (Figure 2B). To characterize the presence of major neuronal subpopulations based on markers for neurotransmitters, we analyzed the time-course expression of several markers, solute carrier family 32 member 1 (SLC32A1) for inhibitory GABA neurons, TH for dopamine neurons, histidine decarboxylase (HDC) for histamine-producing neurons, solute carrier family 17 member 6 (SLC17A6) for excitatory glutamatergic neurons, and solute carrier family 6 member 4 (SLC6A4) for serotonin neurons, as previously described (Herb et al., 2023; Tadross et al., 2025) (Figure 2C). This analysis demonstrated that transcripts of SLC32A1 and TH were most abundantly upregulated and detected by day 28, with some evidence of HDC , albeit at lower levels. Low upregulation of SLC17A6 was detected only in hypothalamic neurons derived from one out of three cell lines (WA09 hESC) (Figure 2C). This is consistent with our ICC findings (Figures 1I and S1M).Figure 2Time-course transcriptomic analysis of hPSCs differentiated into ventral diencephalon and hypothalamic neurons(A) PCA of a time-course mRNA-seq experiment including samples from three independent cell lines. Note the clustering of the samples according to the time-point of differentiation, suggesting the same developmental trajectory between samples from different cell lines.(B) The heatmap represents the top 20 gene markers upregulated (log2FC) at days 7, 14, and 21, and the top 60 gene markers upregulated at day 28, relative to day 0 (hPSC). Due to the overlap of top markers between the time points, the total number of genes represented on the heatmap is 71. Note the stepwise upregulation of ventral diencephalon markers RAX, ASCL1, DLX1, and ARX, as well as ARC nucleus markers ISL1, POMC, SST, and TH.(C) Time-course expression of SLC32A1, TH, HDC, SLC17A6, and SLC6A4 in three independent cell lines (LiPSC-GR 1.1, WA09, NCRM5).N = 3 independent hPSC differentiation experiments for all datasets presented (LiPSC-GR 1.1, WA09, NCRM5). Time-course transcriptomic analysis of hPSCs differentiated into ventral diencephalon and hypothalamic neurons (A) PCA of a time-course mRNA-seq experiment including samples from three independent cell lines. Note the clustering of the samples according to the time-point of differentiation, suggesting the same developmental trajectory between samples from different cell lines. (B) The heatmap represents the top 20 gene markers upregulated (log2FC) at days 7, 14, and 21, and the top 60 gene markers upregulated at day 28, relative to day 0 (hPSC). Due to the overlap of top markers between the time points, the total number of genes represented on the heatmap is 71. Note the stepwise upregulation of ventral diencephalon markers RAX, ASCL1, DLX1, and ARX, as well as ARC nucleus markers ISL1, POMC, SST, and TH. (C) Time-course expression of SLC32A1, TH, HDC, SLC17A6, and SLC6A4 in three independent cell lines (LiPSC-GR 1.1, WA09, NCRM5). N = 3 independent hPSC differentiation experiments for all datasets presented (LiPSC-GR 1.1, WA09, NCRM5). To further validate the hypothalamic identity of our neurons, we performed gene pathway enrichment analyses and compared them to publicly available datasets from previous studies. Indeed, when we used the top 100 genes upregulated at day 28 as compared to day 0 (log2FC), we found a strong association with human hypothalamus samples in the Human Gene Atlas database (Figure 3A). The hypothalamus was also a top feature in the GTEx database, while circadian entrainment was found in the KEGG database (Figures S3A and S3B).Figure 3Enrichment of BMI-associated genes in hPSC-derived hypothalamic neurons(A) Gene pathway enrichment analysis using the top 100 genes upregulated at day 28 vs. day 0 (log2FC) results in the hypothalamus as the main hit in the Human Gene Atlas human tissue database.(B) Principal component analysis (PCA) shows transcriptomic similarities between in vitro and in vivo hypothalamic neurons. PCA was performed using transcripts per million (TPM) data from in vitro neurons at sequential differentiation timepoints (days 0, 7, 14, 21, and 28), as well as from in vivo hypothalamic neurons at fetal (trimesters 1 and 2) and adult (29, 42, and 50 years old) stages. The analysis revealed distinct transcriptomic profiles between fetal and adult hypothalamic neurons. Notably, as in vitro neurons differentiate (from days 0 to 28), their transcriptomic profiles increasingly resemble those of fetal hypothalamic neurons.(C) Distribution of gene expression specificity index (ESI) at day 28. ESI values were calculated using normalized expression data of cells across various developmental stages. Genes located to the right of the vertical red line represent the top 10 percent of genes with the highest ESI at day 28.(D) Bar graph summarizes LDSC-derived enrichment scores for selected metabolic, neurological traits, as well as random trait (height), highlighting BMI as the only significantly enriched trait within the top 10th percentile of genes expressed in hypothalamic neurons, as identified by the ESI analysis in C.(E) Transcript per million (TPM) plots from mRNA-seq show that POMC and ACHE expression is markedly upregulated at day 28 of hypothalamic neuron differentiation compared to earlier stages. POMC is associated with both type 2 diabetes (T2D) and obesity in GWAS studies, whereas ACHE has been linked to obesity.(F) Bar graphs depict the RT-qPCR expression of POMC and ACHE mRNA at the midpoint and endpoint of cortical and hypothalamic neuron differentiation. For each gene and cell line, Welch’s one-way ANOVA revealed a significant effect (p < 0.05). Games-Howell post hoc testing was used to compare all timepoints, and only the comparisons for the final time point between the protocols (day 35 cortical vs. day 28 hypothalamic) are reported in the figure. Corresponding p values are POMCWA09p = 0.0138, POMCLiPSCp = 0.0484; ACHEWA09p = 0.0074, ACHELiPSCp = 0.0709. Fold change was represented vs. day 0 (hPSC stage) and normalized to ACTB.(G) TPM plots from mRNA-seq demonstrate significant upregulation of ACVR1C, GNG4, DGKG, and DGKB at day 28 of hypothalamic neuron differentiation. These genes have been identified as T2D-associated loci in GWAS analyses.(H) Bar graphs show the RT-qPCR expression of ACVR1C, GNG4, DGKG, and DGKB at the midpoint and endpoint of cortical and hypothalamic neuron differentiation. One-way Welch’s ANOVA was significant for all four genes across groups in both lines (p < 0.05). Games-Howell post hoc p values for final timepoints between the protocols (cortical day 35 vs. hypothalamic day 28) are listed for each gene and cell line as represented in the figure. ACVR1CWA09p = 0.0085, ACVR1CLiPSCp = 0.0165; GNG4WA09p = 0.0249, GNG4LiPSCp = 0.0934; DGKGWA09p = 0.0074, DGKGLiPSCp = 0.0885; DGKBWA09p = 0.0041, DGKBLiPSCp = 0.0011. Fold change was represented vs. day 0 (hPSC stage) and normalized to ACTB.N = 3 independent hPSC differentiation experiments for all RNA-seq datasets presented (LiPSC-GR 1.1, WA09, NCRM5). N = 2 independent hPSC lines (WA09, LiPSC-GR 1.1) for all RT-qPCR datasets presented. Enrichment of BMI-associated genes in hPSC-derived hypothalamic neurons (A) Gene pathway enrichment analysis using the top 100 genes upregulated at day 28 vs. day 0 (log2FC) results in the hypothalamus as the main hit in the Human Gene Atlas human tissue database. (B) Principal component analysis (PCA) shows transcriptomic similarities between in vitro and in vivo hypothalamic neurons. PCA was performed using transcripts per million (TPM) data from in vitro neurons at sequential differentiation timepoints (days 0, 7, 14, 21, and 28), as well as from in vivo hypothalamic neurons at fetal (trimesters 1 and 2) and adult (29, 42, and 50 years old) stages. The analysis revealed distinct transcriptomic profiles between fetal and adult hypothalamic neurons. Notably, as in vitro neurons differentiate (from days 0 to 28), their transcriptomic profiles increasingly resemble those of fetal hypothalamic neurons. (C) Distribution of gene expression specificity index (ESI) at day 28. ESI values were calculated using normalized expression data of cells across various developmental stages. Genes located to the right of the vertical red line represent the top 10 percent of genes with the highest ESI at day 28. (D) Bar graph summarizes LDSC-derived enrichment scores for selected metabolic, neurological traits, as well as random trait (height), highlighting BMI as the only significantly enriched trait within the top 10th percentile of genes expressed in hypothalamic neurons, as identified by the ESI analysis in C. (E) Transcript per million (TPM) plots from mRNA-seq show that POMC and ACHE expression is markedly upregulated at day 28 of hypothalamic neuron differentiation compared to earlier stages. POMC is associated with both type 2 diabetes (T2D) and obesity in GWAS studies, whereas ACHE has been linked to obesity. (F) Bar graphs depict the RT-qPCR expression of POMC and ACHE mRNA at the midpoint and endpoint of cortical and hypothalamic neuron differentiation. For each gene and cell line, Welch’s one-way ANOVA revealed a significant effect (p < 0.05). Games-Howell post hoc testing was used to compare all timepoints, and only the comparisons for the final time point between the protocols (day 35 cortical vs. day 28 hypothalamic) are reported in the figure. Corresponding p values are POMCWA09p = 0.0138, POMCLiPSCp = 0.0484; ACHEWA09p = 0.0074, ACHELiPSCp = 0.0709. Fold change was represented vs. day 0 (hPSC stage) and normalized to ACTB. (G) TPM plots from mRNA-seq demonstrate significant upregulation of ACVR1C, GNG4, DGKG, and DGKB at day 28 of hypothalamic neuron differentiation. These genes have been identified as T2D-associated loci in GWAS analyses. (H) Bar graphs show the RT-qPCR expression of ACVR1C, GNG4, DGKG, and DGKB at the midpoint and endpoint of cortical and hypothalamic neuron differentiation. One-way Welch’s ANOVA was significant for all four genes across groups in both lines (p < 0.05). Games-Howell post hoc p values for final timepoints between the protocols (cortical day 35 vs. hypothalamic day 28) are listed for each gene and cell line as represented in the figure. ACVR1CWA09p = 0.0085, ACVR1CLiPSCp = 0.0165; GNG4WA09p = 0.0249, GNG4LiPSCp = 0.0934; DGKGWA09p = 0.0074, DGKGLiPSCp = 0.0885; DGKBWA09p = 0.0041, DGKBLiPSCp = 0.0011. Fold change was represented vs. day 0 (hPSC stage) and normalized to ACTB. N = 3 independent hPSC differentiation experiments for all RNA-seq datasets presented (LiPSC-GR 1.1, WA09, NCRM5). N = 2 independent hPSC lines (WA09, LiPSC-GR 1.1) for all RT-qPCR datasets presented. To better understand how our cells differ from those derived using a previously published hPSC-differentiation method (Rajamani et al., 2018), we performed a PCA analysis and demonstrated that our hPSC-derived hypothalamic neurons clustered separately from those derived in a previous study and contained higher levels of canonical ARC marker expression, such as POMC and TBX3 (Figures S3C and S3D) (Rajamani et al., 2018). As demonstrated by the time-course transcriptome profiling, at the final time point of differentiation, our cultures have acquired hypothalamic identity featuring high POMC expression (Figures S4A and S4C). However, AgRP mRNA (Figure S4C), a marker of the appetite-promoting ARC neuron population (van den Top et al., 2004), could not be reliably detected. Notably, this was not due to delayed progenitor differentiation, as markers of proliferative neural progenitors such as NES, SOX2, FABP7, and MKI67 were downregulated by days 21 and 28 (Figure S4B). We detected transcripts of previously described marker genes for the dorso-medial (GPR50) and ventro-medial (NPTX2) hypothalamus (Herb et al., 2023) (Figure S4D), which are two nuclei in close proximity to the ARC nucleus (Figure S4A). There was little to no evidence of marker genes representing other hypothalamic nuclei (Figures S4E and S4F). To effectively use this in vitro model in functional studies, it is essential to further understand the heterogeneity of the derived cultures and whether the genes of interest are expressed in a sufficiently large number of neurons. To investigate this, we performed a multiplexed error-robust fluorescence in situ hybridization (MERFISH) assay using a panel of 110 probes targeting genes expressed in human hypothalamic neurons (Zhou et al., 2022; Herb et al., 2023) in 62,294 iPSC-derived hypothalamic neurons (Figures 4A, 4B, and S7; Table S4). UMAP clustering analysis revealed 14 distinct clusters (Figure 4C), with POMC strongly expressed in the largest (red) cluster, while also showing a scattered expression pattern in other clusters (Figures 4C–4E). PR/SET domain 12 (PRDM12), a marker co-expressed in POMC neurons (Hael et al., 2020; Yu et al., 2022; Tadross et al., 2025), exhibited a similar UMAP distribution pattern (Figures 4D and 4E). Overall, POMC transcripts were found in 69.72% the single cells profiled on day 28, while PRDM12 was found in 51.31% (Figure 4F). We also quantified POMC/PRDM12 co-expression and found that 47.06% of cells in the culture expressed transcripts for both markers. Other neuronal subpopulations included 3 clusters with varying degrees of OTP and SST expression, 2 clusters of GHRH-expressing neurons, and separate clusters expressing SRY-box transcription factor 14 (SOX14), nescient helix-loop-helix 2 (NHLH2), FEZ family zinc finger 1 (FEZF1), and H6 family homeobox 2 (HMX2) (Figures 4D and 4E). In summary, our transcriptomic analysis and validation methods confirm that the differentiation protocol robustly and reproducibly generates hypothalamic neurons, predominantly of POMC identity, with minimal expression of marker genes associated with other hypothalamic nuclei.Figure 4MERFISH reveals the molecularly diverse neuronal subtypes comprising the iPSC-hypothalamic neuronal culture(A) High-resolution image of neurons segregated based on DAPI/PolyT staining, displaying different transcript profiles.(B) Spatial map of clusters in the culture of iPSC-derived hypothalamic neurons (LiPSC-GR 1.1).(C) Uniform manifold approximation and projection (UMAP) plots show the distribution of clusters and cluster annotations.(D) Violin plots depict the top markers across the annotated clusters. Note the abundance of POMC expression.(E) UMAP feature plots depicting the expression of top marker genes across annotated clusters.(F) Radar plot depicts the percentage of cells expressing POMC, PRDM12, OTP, SST, SOX14, NHLH2, FEZF1, GHRH, and HMX2 mRNA transcripts.N = 62,294 cells used in the analysis (after QC) for MERFISH datasets. MERFISH reveals the molecularly diverse neuronal subtypes comprising the iPSC-hypothalamic neuronal culture (A) High-resolution image of neurons segregated based on DAPI/PolyT staining, displaying different transcript profiles. (B) Spatial map of clusters in the culture of iPSC-derived hypothalamic neurons (LiPSC-GR 1.1). (C) Uniform manifold approximation and projection (UMAP) plots show the distribution of clusters and cluster annotations. (D) Violin plots depict the top markers across the annotated clusters. Note the abundance of POMC expression. (E) UMAP feature plots depicting the expression of top marker genes across annotated clusters. (F) Radar plot depicts the percentage of cells expressing POMC, PRDM12, OTP, SST, SOX14, NHLH2, FEZF1, GHRH, and HMX2 mRNA transcripts. N = 62,294 cells used in the analysis (after QC) for MERFISH datasets. We also compared day 0 with each of the other profiled stages (days 7, 14, 21, and 28) to identify up-regulated genes relative to day 0. Using a stringent threshold of FDR<10−6 and fold change >4, we found over 6000 upregulated genes by days 21 and 28 (Figure S5A). Top upregulated pathways at day 7 included TGF-β and Hedgehog signaling pathways, corresponding to ventral diencephalon patterning, while synaptic vesicle cycle, circadian entrainment, and GnRH secretion were upregulated at later time points (days 14, 21, and 28), suggesting hypothalamus neuronal specification (Figure S5B). Downregulated pathways at later time points correlated with cells exiting the cell cycle (Figure S5C). A stepwise comparison between day 7 and day 14 provided interesting insights, outlining the onset of hypothalamic neurogenesis in the upregulated pathways and the downregulation of neural progenitor and patterning-related pathways (TGF-β, Hedgehog, MAPK, Hippo, Rap1, and Ras signaling) (Figures S4F and S5D). Altogether, these findings suggest that hPSC-derived neurons presented here exhibit dynamic transcriptome changes that align with prenatal hypothalamus development. Enrichment of disease-relevant genes in hPSC-derived hypothalamic neurons Genome-wide association studies (GWASs) have identified hundreds of genomic loci associated with risk for obesity and T2D. Thus, we asked whether any of the associated genes are expressed in our hPSC-derived hypothalamic neurons. To this end, we determined the expression specificity index (ESI), which measures the likelihood that a gene is specifically expressed at the given differentiation time point (Timshel et al., 2020). We identified 1216 genes within the highest (top 10%) ESI at day 28 (Figure 3C; on the right of the dashed line; Table S2). To perform an unbiased enrichment analysis for the traits body mass index (BMI) (Locke et al., 2015), T2D (Suzuki et al., 2024), Alzheimer’s disease (AD) (Bellenguez et al., 2022), Parkinson’s disease (PD) (Leonard and Global Parkinson’s Genetics Program, 2025), amyotrophic lateral sclerosis (ALS) (van Rheenen et al., 2021), and height (Yengo et al., 2022), we conducted linkage disequilibrium score regression (LDSC) analysis on Hapmap 3 SNPs (Table S3) within the 1,215 genes highly expressed in hPSC-derived hypothalamic neurons at day 28 and observed significant enrichment for BMI (Figure 3D). Next, we compared the set of 1,216 ESI-defined genes with a curated subset of 257 T2D-associated genes obtained from the T2D Knowledge Portal (Costanzo et al., 2023) as well as 200 obesity-associated genes compiled from prior studies (Goodarzi, 2018; De Rosa et al., 2021; Loos and Yeo, 2022; Hagg et al., 2015; Hoggart et al., 2014; Locke et al., 2015; Scuteri et al., 2007; Speliotes et al., 2010; Willer et al., 2009), with the aim of identifying genes expressed in hypothalamic neurons. We identified a total of 11 obesity-associated genes showing expression in hypothalamic neurons, including acetylcholinesterase (ACHE), erb-b2 receptor tyrosine kinase 4 (ERBB4), glutamate ionotropic receptor delta type subunit 1 (GRID1), potassium two pore domain channel subfamily K member 3 (KCNK3), potassium two pore domain channel subfamily K member 9 (KCNK9), kinase suppressor of ras 2 (KSR2), neurexin 3 (NRXN3), neurotrophic receptor tyrosine kinase 2 (NTRK2), plexin A4 (PLXNA4), POMC and syntaxin 1B (STX1B) (Figure S6). Additionally, we identified 13 T2D-associated genes that were expressed in hypothalamic neurons namely activin A receptor type 1C (ACVR1C), adenylate cyclase 5 (ADCY5), diacylglycerol kinase beta (DGKB), diacylglycerol kinase gamma (DGKG), glucagon-like peptide 1 receptor (GLP1R), G protein subunit gamma 4 (GNG4), POMC, RAS guanyl releasing protein 1 (RASGRP1), sodium voltage-gated channel alpha subunit 3 (SCN3A), hyperpolarization activated cyclic nucleotide gated potassium channel 1 (HCN1), inositol-trisphosphate 3-kinase A (ITPKA), KSR2, and NRXN3 (Figure S6). We validated the expression of these genes in hypothalamic neurons and compared them to hPSC-derived cortical neurons. Alongside POMC, ACHE was selectively expressed in hypothalamic neurons (Figures 3E and 3F). ACHE may contribute to BMI-related phenotypes by modulating the cholinergic regulation of melanocortin-dependent satiety signaling and hypothalamic energy-balance circuits (Jeong et al., 2017). T2D-associated genes, DGKB and DGKG, which encode diacylglycerol kinases involved in the diacylglycerol-phosphatidic acid (DAG-PA) signaling axis, were also expressed at higher levels in hypothalamic neurons (Figures 3G and 3H). GNG4, a G-protein γ-subunit involved in G-protein coupled receptor (GPCR) signaling that converges on the same lipid-signaling pathway, and ACVR1C, known to be involved in controlling energy balance in the hypothalamus, were also displaying statistically stronger expression in hypothalamic neurons (Figures 3G and 3H). These findings suggest that the coordinated regulation of DAG-mediated signal transduction may be important for metabolic regulation in the hypothalamus and warrants further investigation. In conclusion, hPSC-derived hypothalamic neurons generated using this protocol are enriched for GWAS genes associated with BMI phenotypes and display the expression of several genes linked to T2D and obesity when compared to hPSC-derived cortical neurons, making them suitable for functional studies and drug target discovery. The abundance of relevant gene transcripts and canonical hypothalamus markers highlights their relevance and potential utility for modeling a spectrum of rare and common metabolic disorders. Integrated analysis of chromatin accessibility and transcriptomic changes over the course of hypothalamic neuron differentiation To provide insight into the epigenetic modulation occurring over the course of human hypothalamic neuron differentiation, we conducted ATAC-Seq. We profiled a total of 43 samples, aligning with the time-course sample profile used in our transcriptomic analysis (Figures 2 and 3). PCA of ATAC-Seq samples demonstrated clustering of Day 0, Day 7, and Day 14 samples according to differentiation time points, while samples from Day 21 and Day 28 displayed intermixing (Figure 5A). To better understand how the chromatin landscape changes throughout differentiation and to compare it to transcriptomic changes, we compared the numbers of differentially regulated genes between adjacent differentiation time points (Figure 5B) and observed a similar trend in ATAC peaks (Figure 5C). Volcano plot analysis of differential ATAC-Seq peaks between the differentiation starting point (day 0; hPSC) and endpoint (day 28; hypothalamic neurons) demonstrated significant changes in the chromatin landscape (Figure 5D). As expected, these ATAC-Seq peaks were highly enriched around the transcription start sites (TSSs) of genes. Among the top differential peaks with increased accessibility in hypothalamic neurons were genes associated with obesity and T2D (Figure 5D).Figure 5Epigenetic dynamics of human hypothalamus development revealed through ATAC-Seq profiling(A) PC1 and PC2 of hPSC-derived samples from different timepoints of differentiation based on ATAC-Seq data.(B) Differential genes shared across comparisons of adjacent time points, taking into account directionality of expression changes. Note, by day 7 and day 14, the largest number of genes are up/down regulated.(C) Similar to Figure 4B, differential peaks are shared across comparisons of adjacent developmental time points.(D) Volcano plot shows differential ATAC-Seq peak accessibility between day 0 and day 28. T2D- and obesity-associated genes are highlighted among the peaks exhibiting significant changes in accessibility.(E) UCSC Genome Browser Plot shows time-course ATAC-Seq and mRNA-seq data for the POMC gene and its upstream flanking region. This plot visualizes the accessibility of two phylogenetically conserved POMC gene enhancer regions over the time course of differentiation.(F) ACHE association with a potential regulator element in a likely promoter.(G) NR5A1 association with an intronic ATAC-peak, likely an enhancer. NR5A1 is an important transcription factor in hypothalamus development.N = 3 independent hPSC differentiation experiments for all ATAC-seq datasets presented (LiPSC-GR 1.1, WA09, NCRM5). Epigenetic dynamics of human hypothalamus development revealed through ATAC-Seq profiling (A) PC1 and PC2 of hPSC-derived samples from different timepoints of differentiation based on ATAC-Seq data. (B) Differential genes shared across comparisons of adjacent time points, taking into account directionality of expression changes. Note, by day 7 and day 14, the largest number of genes are up/down regulated. (C) Similar to Figure 4B, differential peaks are shared across comparisons of adjacent developmental time points. (D) Volcano plot shows differential ATAC-Seq peak accessibility between day 0 and day 28. T2D- and obesity-associated genes are highlighted among the peaks exhibiting significant changes in accessibility. (E) UCSC Genome Browser Plot shows time-course ATAC-Seq and mRNA-seq data for the POMC gene and its upstream flanking region. This plot visualizes the accessibility of two phylogenetically conserved POMC gene enhancer regions over the time course of differentiation. (F) ACHE association with a potential regulator element in a likely promoter. (G) NR5A1 association with an intronic ATAC-peak, likely an enhancer. NR5A1 is an important transcription factor in hypothalamus development. N = 3 independent hPSC differentiation experiments for all ATAC-seq datasets presented (LiPSC-GR 1.1, WA09, NCRM5). Next, we sought to identify associations between differential gene regulation and ATAC-peaks detected at intronic or promoter regions, thereby identifying potential regulatory regions or enhancers for the specific gene, as previously described (Xue et al., 2023). We fit a general linear model to assess the association considering the hPSC lines and differentiation time point. We detected over sixty thousand accessible chromatin regions associated with gene expression across the differentiation time course (FDR <0.05; Table S5). These include many genes that were previously associated with metabolic disorders and hypothalamus development. To validate this extensive dataset, we sought proof of concept by identifying two well-described and evolutionarily conserved POMC gene enhancers (de Souza et al., 2005). Indeed, as previously described, we observed ATAC peaks of 600 base pairs (bp) and 150 bp located approximately 10–12 kb upstream of the POMC transcriptional units, with the highest accessibility at the onset of POMC mRNA expression at day 14 (Figure 5E). In a different example, a 2878 bp peak around the promoter of ACHE positively correlated with ACHE gene expression (Figure 5F). Likewise, a 1314 bp peak in the last intron of the nuclear receptor subfamily 5 group A member 1 (NR5A1) gene positively correlated with the time-course expression of this key regulator of medio-basal hypothalamus development (Dellovade et al., 2000) (Figure 5G). In conclusion, our ATAC-Seq analysis revealed significant changes in the chromatin landscape during the differentiation of hPSCs into hypothalamic neurons, aligning with transcriptomic changes and identifying potential regulatory regions that merit further investigation. Functional characterization of human hypothalamic neurons Next, we sought to demonstrate the translational potential of hypothalamic neurons derived from hPSCs in this study. To test this, we investigated the capacity of hPSC-derived hypothalamic neurons to respond to humoral and hormonal stimuli. Signals of nutrient-state, such as insulin and glucose, as well as leptin and glucagon-like peptide 1 (GLP1), are important regulators of appetite and satiety (Jais and Bruning, 2022; Secher et al., 2014). The ARC is well-positioned to react quickly to circulating levels of hormones and nutrients because of its proximity to the portal blood capillary network and the median eminence. For example, POMC neurons of the ARC nucleus project into the median eminence outside the blood-brain barrier, where they can detect circulating levels of insulin and glucose for appetite regulation (Horvath, 2005; Ibrahim et al., 2003). Following a meal, circulating insulin levels increase and exert depolarization and activation of anorexigenic POMC neurons, partly mediated by the activation of TRPC-5 channels (Qiu et al., 2014). Thus, we assessed their molecular response to challenges with insulin. Complete insulin withdrawal results in massive cell death of neurons (Rhee et al., 2013). Therefore, we developed an assay to assess the functional response of the differentiated hypothalamic neurons to insulin in modified culture conditions. First, we evaluated their basal spontaneous action potentials using multi-electrode arrays (MEAs) following maintenance in Hypo-3 medium with reduced insulin for 96 h (Figure 6A). After 96 h in reduced insulin conditions, the hypothalamic neurons were then treated with increasing doses of human recombinant insulin (0, 0.01, 0.1, and 1 mg/mL) for 2 h, and their spontaneous activity was re-evaluated using MEAs. A significant dose-dependent functional response to insulin was observed via increased mean firing rates of hypothalamic neurons (Figure 6B). To validate these findings across biological replicates, we treated hypothalamic neurons derived from three independent cell lines with low (0.01 mg/mL) and high (1 mg/mL) insulin for 2 h. We observed a significant treatment effect in response to increasing insulin concentrations, as measured by changes in mean firing rate (Figure 6C). While a tendency toward increased burst frequency was observed, this did not reach statistical significance (Figure 6D). Of note, we also measured spontaneous action potential firing in hypothalamic neurons derived from two independent lines on days 28 and 35. The number of spikes (Figure S2G) and mean firing rate (Figure S2H) remained consistent during this time period, indicating that all functional assays can be reliably performed within this 7-day window.Figure 6Action potential firing changes of hypothalamic neurons in response to insulin(A) Schematic overview of MEA assay design for capturing insulin responses.(B) Mean firing rate of hypothalamic neuron cultures captured after 2 h treatment with increasing doses of human recombinant insulin. Data were normalized to baseline (0 h, no treatment). Data were analyzed using one-way ANOVA across four treatment groups (DMSO, 0.001, 0.01, and 1 mg/mL insulin; n = 16 total observations), revealing a significant treatment effect (F(3,12) = 4.10, p = 0.0323). Tukey’s multiple comparisons test identified a significant difference between DMSO and 1 mg/mL insulin (p = 0.0197); all other comparisons were not significant. Data are presented as mean ± SEM.(C and D) Differential response to low and high insulin concentrations validated across neurons derived from three independent hPSC lines (LiPSC-GR1.1, WA09, NCRM-5) as measured by changes in mean firing rate (C) and burst frequency(D) after 2 h treatment. For mean firing rate (C), two-way ANOVA showed a significant treatment effect (F(1,17) = 9.44, p = 0.0069), with no cell line or interaction effects. Šídák post hoc identified a significant difference in WA09 only (p = 0.0196). Data are presented as mean ± SEM; (E) Representative microscopic images of WA09 tdTomato::POMC-derived hypothalamic neuron pierced by an electrode for selective patch-clamp electrophysiology recordings in POMC neurons.(F) Spontaneous action potential (AP) firing recorded from a holding potential of −50 mV before (baseline, gray) and after the local perfusion of BrainPhys media (Ctrl, red), followed by BrainPhys containing 0.1 mg/mL insulin (blue).(G) Quantification of action potentials before and after insulin treatment. Note the significant increase in firing rate of insulin-treated cells. One-way repeated measures ANOVA (F(2,6) = 6.77, p = 0.029, n = 4).(H) Representative traces of a WA09 tdTomato::POMC-derived hypothalamic neuron with increased AP firing after 100 nM Leptin (green).(I) Quantification of AP firing in WA09 tdTomato::POMC expressing hypothalamic neurons before (baseline, gray) and after 100 nM Leptin (green). Vehicle addition alone did not produce consistent changes in neuronal firing rates in control experiments performed under identical recording conditions (Figure 6F). Action potential firing changes of hypothalamic neurons in response to insulin (A) Schematic overview of MEA assay design for capturing insulin responses. (B) Mean firing rate of hypothalamic neuron cultures captured after 2 h treatment with increasing doses of human recombinant insulin. Data were normalized to baseline (0 h, no treatment). Data were analyzed using one-way ANOVA across four treatment groups (DMSO, 0.001, 0.01, and 1 mg/mL insulin; n = 16 total observations), revealing a significant treatment effect (F(3,12) = 4.10, p = 0.0323). Tukey’s multiple comparisons test identified a significant difference between DMSO and 1 mg/mL insulin (p = 0.0197); all other comparisons were not significant. Data are presented as mean ± SEM. (C and D) Differential response to low and high insulin concentrations validated across neurons derived from three independent hPSC lines (LiPSC-GR1.1, WA09, NCRM-5) as measured by changes in mean firing rate (C) and burst frequency (D) after 2 h treatment. For mean firing rate (C), two-way ANOVA showed a significant treatment effect (F(1,17) = 9.44, p = 0.0069), with no cell line or interaction effects. Šídák post hoc identified a significant difference in WA09 only (p = 0.0196). Data are presented as mean ± SEM; (E) Representative microscopic images of WA09 tdTomato::POMC-derived hypothalamic neuron pierced by an electrode for selective patch-clamp electrophysiology recordings in POMC neurons. (F) Spontaneous action potential (AP) firing recorded from a holding potential of −50 mV before (baseline, gray) and after the local perfusion of BrainPhys media (Ctrl, red), followed by BrainPhys containing 0.1 mg/mL insulin (blue). (G) Quantification of action potentials before and after insulin treatment. Note the significant increase in firing rate of insulin-treated cells. One-way repeated measures ANOVA (F(2,6) = 6.77, p = 0.029, n = 4). (H) Representative traces of a WA09 tdTomato::POMC-derived hypothalamic neuron with increased AP firing after 100 nM Leptin (green). (I) Quantification of AP firing in WA09 tdTomato::POMC expressing hypothalamic neurons before (baseline, gray) and after 100 nM Leptin (green). Vehicle addition alone did not produce consistent changes in neuronal firing rates in control experiments performed under identical recording conditions (Figure 6F). To selectively study POMC neuron responses, we established a new, and now publicly available WA09 tdTomato::POMC reporter line and used it to generate hypothalamic neurons (Jovanovic et al., 2026). To delve deeper into the electrical properties of POMC neurons, we conducted patch-clamp electrophysiological recordings of four tdTomato::POMC+ neurons. Upon challenge with insulin, there was a noticeable and significant increase in the firing rate of tdTomato::POMC+ neurons, while there was no change in the firing rate in response to sham control (Figures 6E–6G). The observed tonic spontaneous firing pattern and insulin-evoked responses are consistent with previous studies using primary POMC neurons derived from mouse hypothalamus (Na et al., 2022), further supporting that these measurements largely reflect POMC neuron activity. Leptin is produced by the adipose tissue. Upon secretion into the bloodstream, it acts primarily on the neurons of the hypothalamus, regulating feeding. We exposed our hypothalamic neuron cultures to leptin. Patch-clamp electrophysiological recordings showed that out of five tdTomato::POMC+ neurons, two exhibited increased firing rates, while another two exhibited decreased firing rates, and one neuron showed no detectable response (Figures 6H and 6I). This variability aligns with previous reports describing heterogeneous responses to leptin among POMC neurons (Biglari et al., 2021; Hill et al., 2008; Sun et al., 2016; Na et al., 2022; Al-Qassab et al., 2009; Roberts et al., 2019; Plum et al., 2006; Cowley et al., 2001; Campbell et al., 2017; Williams et al., 2010; Gao et al., 2017). Next, we examined how the hypothalamic neurons respond at the molecular level to metabolic challenges with glucose, insulin, and incretin mimetic drugs (GLP1R agonists). We quantified POMC and insulin receptor (IR; α and β subunits) protein levels following a 4-day (96 h) treatment under different media formulations: (1) low glucose with insulin, (2) low glucose with reduced insulin, (3) high glucose with insulin, (4) high glucose with reduced insulin, and (5) regular Hypo-3 medium as a control. Under high-glucose conditions, POMC protein levels were elevated compared to low-glucose conditions (Figures 7A and 7B). A statistically significant increase was observed specifically between the low-glucose with insulin (−g/+i) and high-glucose with reduced insulin (+g/−i) conditions (p = 0.0159, unpaired Student’s t test; Figure 7B). GLP1R expressed in the hypothalamus has been implicated as a mediator of weight loss via effects associated with antidiabetic drugs acting as incretin mimetics (Secher et al., 2014). Building on our ESI analysis, which indicated that GLP1R is one of the genes expressed in our hypothalamic neurons on day 28 (Figure 3F), and MERFISH single-cell analysis, which suggested 12.77% of cells carry the GLP1R mRNA transcript (Figures S7H and S7I). We first validated that GLP1R expression is induced in day 28 hypothalamic neurons as compared to day 0, hPSC stage (Figure 7C). Next, we sought to measure GLP1R transcript expression using RT-qPCR under different culture conditions: low glucose (1 g/L), medium glucose (4 g/L), or high glucose (9 g/L) supplemented Hypo-3 medium for 96 h (Figure 7D). Interestingly, we observed the highest expression of GLP1R under medium glucose, although this was not significantly different from the low glucose condition. The high glucose condition displayed a statistically significant reduction in GLP1R expression (unpaired Student’s t test; high vs. medium glucose p = 0.0042; low vs. high glucose p = 0.045; Figure 7D). After we confirmed which culture conditions support the expression of GLP1R, we investigated the pharmacological activation of GLP1R and its resultant phosphorylation of cAMP response element binding (CREB) protein using a fluorescence resonance energy transfer (FRET) assay, which is amenable to high-throughput experiments. Hypothalamic neurons were challenged with the FDA-approved drug for glycemic management, Exendin-4, for periods of 5 or 60 min. Forskolin, a strong adenylate cyclase activator, was included as a positive control as GLP1R is a GPCR and GPCR activation is followed by adenylate cyclase stimulation prior to CREB phosphorylation (Jewell et al., 2019). Forskolin treatment resulted in the significant phosphorylation of CREB after 60 min (Figure 7E, unpaired t test, p = 0.0132), while there was an increase but with no statistical significance at 5 min. Similarly, treatment with Exendin-4 led to detectable and statistically significant phosphorylation of CREB at the 60 min time point (Figure 7F, unpaired t test, p = 0.0052).Figure 7Pulse assay experiments to measure glucose, insulin, incretin, and kisspeptin mimetic drug responses(A) Western blot analysis of triplicate samples of hypothalamic neurons at day 28 (LiPSC-GR 1.1) treated with a combination of low (−) or high (+) glucose, with B27 with insulin (+) or with B27 without insulin (−) over a 96 h period. POMC upregulation is observed in high glucose samples.(B) Bar graph representations of peak area normalized to beta-actin (ACTB) and no-treatment control show the strongest expression of POMC in high glucose-treated neurons. (C) RT-qPCR analysis of GLP1R transcript expression in iPSC and iPSC-derived hypothalamic neurons. Statistical significance between groups was assessed using a two-tailed unpaired Student’s t test (p = 0.012). Data are presented as mean ± SD.(D) RT-qPCR analysis of hypothalamic neurons treated with Hypo-3 medium with low (1 g/L), medium (4 g/L), and high glucose (9 g/L). Unpaired two-tailed Student’s t test high vs. medium glucose ∗∗p = 0.0042, medium vs. low glucose p = 0.0753, low vs. high glucose ∗p = 0.045 (E) Forskolin stimulation increases CREB phosphorylation in hypothalamic neurons. FRET-based pCREB/CREB measurements show a significant rise at 60 min (unpaired t test, p = 0.0132) with a non-significant increase at 5 min.(F) Exendin-4 activation of GLP1R leads to time-dependent CREB phosphorylation. A significant increase in pCREB is detected at 60 min following Exendin-4 treatment (unpaired t test, p = 0.0052).(G) KISS1R expression was significantly upregulated in day 28 hypothalamic neurons as compared to day 0 hPSC. Data points represent three replicate samples for the LiPSC-GR1.1 and NCRM-5 lines.(H) Immunostaining for GnRH at day 28 following 24-h stimulation with 5 nM kisspeptin-10. Scale bars, 100 μm.(I–J) Mid-throughput (96-well plate format) RT-qPCR assay for GNRH1 and FOS mRNA expression in response to increasing doses of Kisspeptin-10 (1 nM, 2.5 nM, and 5 nM) or no treatment control. Note the peak increase in GNRH1 expression at 1 h and its downregulation at later time points captured. Target gene expression level was normalized to that of the housekeeping gene GAPDH. Three replicate wells of neurons derived from LiPSC-GR 1.1 were used per treatment group. Pulse assay experiments to measure glucose, insulin, incretin, and kisspeptin mimetic drug responses (A) Western blot analysis of triplicate samples of hypothalamic neurons at day 28 (LiPSC-GR 1.1) treated with a combination of low (−) or high (+) glucose, with B27 with insulin (+) or with B27 without insulin (−) over a 96 h period. POMC upregulation is observed in high glucose samples. (B) Bar graph representations of peak area normalized to beta-actin (ACTB) and no-treatment control show the strongest expression of POMC in high glucose-treated neurons. (C) RT-qPCR analysis of GLP1R transcript expression in iPSC and iPSC-derived hypothalamic neurons. Statistical significance between groups was assessed using a two-tailed unpaired Student’s t test (p = 0.012). Data are presented as mean ± SD. (D) RT-qPCR analysis of hypothalamic neurons treated with Hypo-3 medium with low (1 g/L), medium (4 g/L), and high glucose (9 g/L). Unpaired two-tailed Student’s t test high vs. medium glucose ∗∗p = 0.0042, medium vs. low glucose p = 0.0753, low vs. high glucose ∗p = 0.045 (E) Forskolin stimulation increases CREB phosphorylation in hypothalamic neurons. FRET-based pCREB/CREB measurements show a significant rise at 60 min (unpaired t test, p = 0.0132) with a non-significant increase at 5 min. (F) Exendin-4 activation of GLP1R leads to time-dependent CREB phosphorylation. A significant increase in pCREB is detected at 60 min following Exendin-4 treatment (unpaired t test, p = 0.0052). (G) KISS1R expression was significantly upregulated in day 28 hypothalamic neurons as compared to day 0 hPSC. Data points represent three replicate samples for the LiPSC-GR1.1 and NCRM-5 lines. (H) Immunostaining for GnRH at day 28 following 24-h stimulation with 5 nM kisspeptin-10. Scale bars, 100 μm. (I–J) Mid-throughput (96-well plate format) RT-qPCR assay for GNRH1 and FOS mRNA expression in response to increasing doses of Kisspeptin-10 (1 nM, 2.5 nM, and 5 nM) or no treatment control. Note the peak increase in GNRH1 expression at 1 h and its downregulation at later time points captured. Target gene expression level was normalized to that of the housekeeping gene GAPDH. Three replicate wells of neurons derived from LiPSC-GR 1.1 were used per treatment group. As noted above, differential gene expression analysis suggested the presence of a GnRH neuron population at later stages of differentiation (Figure S5B). GnRH neurons are a poorly understood subgroup present in the hypothalamus that exhibit pulsatile GnRH secretion in response to the neuropeptide kisspeptin to promote the release of gonadotropin hormones by the pituitary. In all mammals, this mechanism is essential for both fertility and the onset of puberty (Herbison, 2018; Plant et al., 1978). To validate the presence of a potentially functional GnRH neuron population, we conducted RT-qPCR to validate KISS1R expression in hypothalamic neurons as KISS1R is required for the neuronal response to kisspeptin. Utilizing two independent iPSC lines, we observed higher levels of KISS1R mRNA in hypothalamic neurons compared to iPSCs (Figure 7G). Next, we performed ICC for GnRH to visualize GnRH-expressing neurons in the cultures following 24 h treatment with Kisspeptin-10 (Figure 7H). We also quantified GNRH1 mRNA levels at multiple time points (0.5, 1, 2, 4, and 6 h) during a dose-response experiment using increasing concentrations (0, 1, 2.5, and 5 nM) of the KISS1R agonist Kisspeptin-10. GNRH1 mRNA levels in hypothalamic neurons increased with increasing dosages of Kisspeptin-10 (Figure 7I). Expression of GNRH1 peaked 1 h after treatment and returned to baseline 4 h later (Figure 7I). We also measured FOS expression, which is a downstream effector of KISS1R and is strongly induced in GnRH neurons upon kisspeptin administration (Irwig et al., 2004). Indeed, FOS was significantly upregulated 1 h after kisspeptin treatment, with its expression correlating to the kisspeptin-10 concentration and GNRH1 levels (Figure 7J). This transient upregulation of GNRH1 indicates the neurons’ ability to dynamically respond to Kisspeptin-10 pulses and lays the groundwork for the use of this hPSC-derived model in reproductive endocrinology research. Discussion Robotic cell culture platform to biomanufacture human hypothalamic neurons at industrial scale Drug screening and large genomic phenotyping campaigns in pre-clinical settings require scalable and robust cell manufacturing strategies to ensure effective functional differentiation from numerous patient-derived hPSC lines. In this study, we leveraged an established robotic platform (Tristan et al., 2021) to automate our novel method for hPSC differentiation into hypothalamic neurons under chemically defined conditions. This innovative approach enables the rapid, reproducible, and standardized production of billions of hypothalamic neurons, with the capacity to simultaneously process up to 270 different patient- and disease-specific cell lines. A fully automated 7-day monolayer culture in T175 flasks, designed for direct patterning into RAX+/NKX2.1+ hypothalamic progenitors, is followed by an automated 7-day spheroid stage. By day 14 of differentiation, the protocol can yield approximately 35–50 million cryopreservable neuroblasts per T175 flask. Our modular design includes defined checkpoints at days 7, 14, and 28, each corresponding to key developmental and practical milestones, such as ventral patterning during early neural induction (day 7), cryopreservation, and large-scale “off-the-shelf” availability of neuroblasts (day 14), and the generation of ready-to-use hypothalamic neurons with high reproducibility across independent patient-derived hPSC lines suitable for high-throughput assays (day 28). Additionally, the ability to handle up to 270 patient-specific cell lines in parallel opens new avenues for personalized medicine, enabling the identification of patient-specific drug responses and genetic susceptibility. To validate our approach, we conducted extensive molecular and cellular characterizations, including transcriptomic and epigenomic time-course profiling. Notably, we implemented miniaturized drug response assays that allow for direct plating into 24-, 96-, and 384-well plate formats. These assays benchmarked the performance of our industrial-scale derived hypothalamic neurons, ensuring their functional equivalence to bona fide human hypothalamic neurons. The high level of fidelity and reproducibility of our automated platform makes it a powerful tool for large-scale phenotypic screening and the development of new therapeutic strategies. This technology holds promise for accelerating the pace of discovery in obesity, T2D, and reproductive disease modeling, drug development, and regenerative medicine. hPSC-derived hypothalamic neurons are functional The combination of dual SMAD and WNT inhibition, along with SHH activation, has been central to establishing hypothalamus identity in most previously published hPSC differentiation protocols. However, our protocol is distinct in that it introduces early and simultaneous dual SMAD and WNT inhibition, combined with SHH activation, starting directly from the hPSC stage. This approach differs from previous reports, which typically involved either dual SMAD inhibition combined with SHH activation followed by Notch inhibition, or dual SMAD and WNT inhibition followed by SHH activation (Merkle et al., 2015; Huang et al., 2021). Our protocol produces a robust population of hypothalamic neurons, predominantly POMC-positive, alongside cells expressing dopaminergic, GHRH, GLP1R, and GnRH/KISS1R neuron markers that are suitable for functional studies of hormonal modulation, including responses to insulin, Exendin-4, and kisspeptin, as well as transcriptional responses to glucose shifts. Within four weeks of in vitro differentiation, our hPSC-derived neurons reach a developmental stage that, based on transcriptomic correlation analysis, shows greater similarity to the human prenatal hypothalamus than to other brain regions. This finding is consistent with other hPSC differentiation protocols that generate neurons, all of them with characteristics of a prenatal stage in brain development (Maroof et al., 2013; Nicholas et al., 2013). In line with recent single-cell profiling of the human prenatal hypothalamus (Zhou et al., 2022; Herb et al., 2023) reporting the detection of the appetite-promoting AgRP/NPY neurons at later developmental stages, we did not observe AgRP/NPY neurons in our cultures. Potential promoter regulatory elements and gene enhancer regions identified in this study can serve as a foundation for further functional validation studies. Subsequent studies can also explore approaches to induce epigenetic modifications to further enhance the maturity of the derived cells (Ciceri et al., 2024; Hergenreder et al., 2024). Hypothalamic dopamine neurons, expressing TH and GABA, tonically inhibit prolactin release from the pituitary and can be modulated by endogenous opioids met-enkephalin and dynorphin (Ben-Jonathan, 1985; Zhang and van den Pol, 2015; Kawano and Daikoku, 1987). These neurons comprised 10–15% of the total neurons generated using our approach and provide additional value to the established in vitro model of hypothalamic neurocircuitry. The GHRH neuronal subpopulation generated with this protocol can be used for modeling different growth hormone deficiency syndromes caused by disorders affecting the hypothalamus. However, further studies are needed to establish functional assays targeting the dopaminergic and GHRH subpopulations generated with this approach. GnRH neurons originate in the olfactory placode and migrate to the hypothalamus during embryonic development, where their activity is largely suppressed until puberty, when pulsatile GnRH secretion drives sexual maturation (Wray, 2010). Reports have also suggested that hypothalamic neural progenitor cells have the capacity to directly generate GnRH neurons (Markakis et al., 2004). hPSC-derived hypothalamic neuron cultures show functional KISS1R expression, which can be stimulated to elicit GNRH1 expression. This establishes evidence for the presence of GnRH neurons that merits further investigation. Resource for modeling hypothalamic pathophysiology in metabolic disorders Understanding the molecular and functional responses of these hPSC-derived neurons to metabolic signals could provide valuable insights into the pathophysiology of metabolic disorders associated with the hypothalamus. The expression of obesity and T2D-associated genes (Costanzo et al., 2023; Goodarzi, 2018) in these hPSC-derived hypothalamic neurons further supports the use of this resource in future studies modeling metabolic disorders. Insight into the mechanisms underlying susceptibility to obesity and insulin resistance is crucial in combating these prevalent metabolic diseases. Previous studies have found that a selective decrease in ARC nucleus insulin receptor levels can cause weight gain, increased adiposity, and rapid onset of hepatic insulin resistance (Obici et al., 2001, 2002). The dynamic modulation of the POMC, IR alpha, and IR beta protein levels by glucose and insulin, as demonstrated here, supports this notion and the utility of the derived neurons in high-throughput assays and phenotypic screens. The multi-omics datasets presented here should serve as a rich data mining resource and help provide insight into the development of the ventral hypothalamus. Finally, the combination of this robust differentiation method with a fully automated robotic platform will allow for the generation of hundreds of millions of human hypothalamic neurons from various human cell lines carrying obesity and T2D-associated variants, creating opportunities for cross-institutional collaborations for the core efforts of the Hypothalamus Consortium, to address metabolic disorders such as obesity and T2D. A path toward translational research in neuroendocrinology The hypothalamus is a central research focus of neuroendocrinologists studying metabolic disorders (Palka et al., 1971; Deng et al., 2020; Zhu et al., 2020) and reproductive dysfunction (True et al., 2017; Silva et al., 2019; Brawer, 1971; Schipper et al., 1981). Recent advancements in iPSC technology have provided a new avenue for establishing human in vitro models of neuropeptidergic hypothalamic neurons (Rajamani et al., 2018; Merkle et al., 2015; Wang et al., 2015; Huang et al., 2021). While these models have offered valuable insights (Sobreira et al., 2021; Razolli et al., 2020; Kirwan et al., 2018; Naule et al., 2023; Lechner et al., 2023; Wang et al., 2017, 2019, 2021; Quarta et al., 2019; Lasconi et al., 2021; Hammond et al., 2021; Pahl et al., 2021), they have lacked the scalability required for successful translation to genome-wide genetic and large library drug screens. Primary human hypothalamic neurons are not available to support therapeutic drug discovery, and the use of ex vivo human hypothalamus tissue has been limited to single-cell profiling experiments to map hypothalamic cell subpopulations (Herb et al., 2023; Zhou et al., 2022; Tadross et al., 2025). The human PSC-differentiated hypothalamic neuron model presented in this study consistently achieves a gene expression signature resembling that of the hypothalamic neurons across multiple independent hPSC lines. This suggests high reproducibility of the method, which is crucial for its application in translational research, particularly when utilized across various patient-derived iPSCs. Moreover, the strongest rationale for using an iPSC-based model lies in its presumed physiological relevance (Vincent et al., 2015). In this regard, the hypothalamic neurons presented here demonstrate responses to humoral, pharmacological, and hormonal stimuli involved in the regulation of appetite and satiety as well as the initiation and maintenance of reproductive competence through the control of gonadotropin release. Given the scalability and reproducibility of the approach, these hPSC-derived neurons will be useful for a variety of additional phenotypic or drug response screens. The comprehensive characterization of the derived hypothalamic neurons reported in this study provides compelling evidence that supports their physiological relevance to metabolic disorders such as obesity and T2D, as well as endocrine reproductive disorders such as hypogonadotropic hypogonadism. Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Vukasin Jovanovic (vukasin.jovanovic@nih.gov). Materials availability The tdTomato::POMC WA09 reporter line will be shared with a completed Materials Transfer Agreement (MTA). This study did not generate new unique reagents. Data and code availability Bulk RNA-Seq and ATAC-Seq FASTQ files have been deposited into the Sequence Read Archive (SRA) under SRA BioProject: PRJNA1085406. Acknowledgments The contribution of the NIH authors was made as part of their official duties as NIH federal employees, is in compliance with agency policy requirements, and is considered a Work of the United States Government. However, the findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH of the U.S. Department of Health and Human Services. The authors would like to thank Matthew Hall, Ann Knebel, Fiona Daly, Yeliz Gedik, Glib Lirazan, Kathryn Garish, Hannah Baskir, Christopher Cherry, and Charles “Pepper” Bonney for their support throughout this work, and Ilyas Singeç for his oversight of conceptualization and development of the protocol. We are grateful to Allan Hoofring and Erina He from the NIH Medical Arts Design Section for the art designs. We also gratefully acknowledge funding from the Regenerative Medicine Program (RMP) of the 10.13039/100000002NIH Common Fund and by the intramural research program of the National Center for Advancing Translational Sciences (10.13039/100006108NCATS) grants TR000244 and TR000410, Intramural 10.13039/100000002NIH HHS/United States grant awarded to the National Human Genome Research Institute (NHGRI) ZIA-HG000024, and the 10.13039/100023343InterAccess #NTR 12003 from the 10.13039/100000066National Institute of Environmental Health Sciences (NIEHS)/Division of Translational Toxicology (DTT) to the 10.13039/100006108NCATS, NIH. C.A.D. is supported by the Frontiers program of the 10.13039/100020059Russell Berrie Foundation, New York Obesity Research Center, P30 DK026687-43, and DK52431-25. E.H.-O. has contributed to this work while employed at the School of Medicine, 10.13039/100008510University of Maryland. The funders had no role in study design, data collection, and analysis; the decision to publish; or the preparation of the manuscript. Figure 6A was generated using Biorender.com. Author contributions V.M.J., C.A.T., C.A.D., F.S.C., and S.C. conceived the study and experiments. V.M.J., K.T.M., C.S., P.O., S.Y., S.R., E.O.H., D.F.B., G.R., and D.B. performed all experiments. T.R., N.N., Q.C., F.M.B.E., T.Y., N.S., H.J.G., and J.I. performed the bioinformatics analyses. V.M.J., C.A.L., N.D.S, M.X., M.S., E.O.H., M.R.E., L.L.B., S.C., N.D.S., D.C., A.S., C.A.D., F.S.C., and C.A.T. contributed to data analysis and discussions. V.M.J. wrote the manuscript with contributions from C.A.T., C.A.D., F.S.C., L.L.B., S.C., H.J.G., Q.C., and N.N. Declaration of interests The authors declare no competing interests. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesPOMCThermoFisherCat#MA5-38604; RRID:AB_2898516CALB (d28k)SWANTCat#CB300PUR; RRID:AB_3542811IR-betaCell SignalingCat#23413; RRID:AB_2924796PAX6BD BiosciencesCat#561462; RRID:AB_10715442PAX6BioLegendCat#901301; RRID:AB_2565003ISL1AbcamCat#ab86472; RRID:AB_1951287GABAThermoFisherCat#PA5-32241; RRID:AB_2549714IR-alphaCell SignalingCat#74118; RRID:AB_2799850Beta ActinMillipore SigmaCat#SAB1305554CREBCell SignalingCat#9104; RRID:AB_490881pCREBCell SignalingCat#9198; RRID:AB_2561044GAD65/67ThermoFisherCat#PA1-84572; RRID:AB_2263114Tyrosine hydroxylase (TH)Peel FreezCat#P40101; RRID:AB_2313713Somatostatin (SST)Santa CruzCat#sc-55565; RRID:AB_831726TUBB3 (TUJ1)BioLegendCat#801201; RRID:AB_2313773FABP7EMD MilliporeCat#ABN14; RRID:AB_11210770GAPDHSanta CruzCat#sc-25778; RRID:AB_10167668Donkey anti-rabbit IgG 488ThermoFisherCat#A21206; RRID:AB_2535792Donkey anti-mouse IgG 568ThermoFisherCat#A10037; RRID:AB_11180865Anti-Rabbit Secondary HRP AntibodyProtein SimpleCat#042–206; RRID:AB_2860577Anti-Mouse Secondary HRP AntibodyProtein SimpleCat#042–205; RRID:AB_2860576Chemicals, peptides, and recombinant proteinsE6 mediumThermoFisherCat#A1516401E8 mediumThermoFisherCat#A1517001BrainPhys mediaStemCell TechnologiesCat#05790LDN189193TocrisCat#6053XAV939TocrisCat#3748SAG21KTocrisCat#5282Ascorbic AcidTocrisCat#4055db-cAMPTocrisCat#1141DBZTocrisCat#4489Chroman-1MedChemCat#HY-15392EmricasanSelleckChemCat#S7775Trans-ISRIBTocrisCat#5284Polyamines supplementMilliporeCat#P8483Jagged1R&D systemsCat#1277-JGActivin AR&D systemsCat#338-ACLIFR&D systemsCat#7734-LFCNTFR&D systemsCat#257-NTDLL1R&D systemsCat#1818-DLBDNFR&D systemsCat#11166-BDGDNFR&D systemsCat#212-GDStemPro AccutaseThermoFisherCat#A1110501DMEM/F12ThermoFisherCat#11320033Anti-adherence solutionStemCell TechnologiesCat#07010Neuron freezing mediaCell ApplicationsCat#042-50N2ThermoFisherCat#A1370701B27 w/o vit. AThermoFisherCat#12587010B27ThermoFisherCat#17504044VitronectinThermoFisherCat#A14700GeltrexThermoFisherCat#A1413302MetforminTocrisCat#2864ForskolinTocrisCat#1099Exendin-4TocrisCat#1933Kisspeptin-10TocrisCat#2570Critical commercial assaysTaqMan gene expression assay KISS1RThermoFisherHs00261399_m1TaqMan gene expression assay GNRH1ThermoFisherHs00171272_m1TaqMan gene expression assay GAPDHThermoFisher4310884ERNeasy plus mini kitQiagenCat#74136Illumina TruSeq stranded mRNA kitIlluminaCat#20020492BCA protein assay kitThermoFisherCat#23227SingleShot cell lysis kitBio-RadCat#1725080Deposited dataBulk RNA-Seq and ATAC-Seq FASTQ filesThis paperSRA BioProject: PRJNA1085406Experimental models: Cell linesWA09WiCellRRID:CVCL_9773LiPSC-GR1.1NIH Common FundRRID:CVCL_RL65NCRM5NIH Common FundRRID:CVCL_1E75WA09 POMC-tdTomato (homozygousclone 1G)NIH/NCATSWAe009-A-3ISoftware and algorithmsfastqc (version 0.11.9)https://biodockerfiles.github.io/fastqc-0-11-9multiqc (version 1.20)https://pypi.org/project/multiqctrim-galore (version 0.6.7)https://github.com/FelixKrueger/TrimGalore/releasescutadapt (version 4.0)https://cutadapt.readthedocs.io/en/v4.0Python (version 3.8.13)https://www.python.org/downloads/release/python-3813Salmon aligner (version v1.10.0)https://github.com/COMBINE-lab/salmon/releasesEnsembl (hg38, Homo_sapiens.GRCh38.cdna.all.fa, release 109)https://useast.ensembl.org/info/website/archives/index.html?R (release 4.2.2)https://cran.r-project.org/bin/windows/base/oldfactoextra (version 1.0.7)https://CRAN.R-project.org/package=factoextratximeta (version 1.15.3)https://bioconductor.org/packages/release/bioc/html/tximeta.htmlDESeq2 (version 1.39.5)https://bioconductor.org/packages/release/bioc/html/DESeq2.htmlapeglm R package (version 1.21.0)https://bioconductor.org/packages/release/bioc/html/apeglm.htmlggplot (release 3.4.1)https://github.com/tidyverse/ggplot2/releasesRUVseq RUVg (version 1.36.0)https://www.bioconductor.org/packages//release/bioc/vignettes/RUVSeq/inst/doc/RUVSeq.htmlOrigin-Pro softwareOriginLabhttps://www.originlab.comPatchmasterHEKA Instrumentshttps://www.heka.comFitmasterHEKA Instrumentshttps://www.heka.comGraphPad PrismGraphPadhttps://www.graphpad.com REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesPOMCThermoFisherCat#MA5-38604; RRID:AB_2898516CALB (d28k)SWANTCat#CB300PUR; RRID:AB_3542811IR-betaCell SignalingCat#23413; RRID:AB_2924796PAX6BD BiosciencesCat#561462; RRID:AB_10715442PAX6BioLegendCat#901301; RRID:AB_2565003ISL1AbcamCat#ab86472; RRID:AB_1951287GABAThermoFisherCat#PA5-32241; RRID:AB_2549714IR-alphaCell SignalingCat#74118; RRID:AB_2799850Beta ActinMillipore SigmaCat#SAB1305554CREBCell SignalingCat#9104; RRID:AB_490881pCREBCell SignalingCat#9198; RRID:AB_2561044GAD65/67ThermoFisherCat#PA1-84572; RRID:AB_2263114Tyrosine hydroxylase (TH)Peel FreezCat#P40101; RRID:AB_2313713Somatostatin (SST)Santa CruzCat#sc-55565; RRID:AB_831726TUBB3 (TUJ1)BioLegendCat#801201; RRID:AB_2313773FABP7EMD MilliporeCat#ABN14; RRID:AB_11210770GAPDHSanta CruzCat#sc-25778; RRID:AB_10167668Donkey anti-rabbit IgG 488ThermoFisherCat#A21206; RRID:AB_2535792Donkey anti-mouse IgG 568ThermoFisherCat#A10037; RRID:AB_11180865Anti-Rabbit Secondary HRP AntibodyProtein SimpleCat#042–206; RRID:AB_2860577Anti-Mouse Secondary HRP AntibodyProtein SimpleCat#042–205; RRID:AB_2860576Chemicals, peptides, and recombinant proteinsE6 mediumThermoFisherCat#A1516401E8 mediumThermoFisherCat#A1517001BrainPhys mediaStemCell TechnologiesCat#05790LDN189193TocrisCat#6053XAV939TocrisCat#3748SAG21KTocrisCat#5282Ascorbic AcidTocrisCat#4055db-cAMPTocrisCat#1141DBZTocrisCat#4489Chroman-1MedChemCat#HY-15392EmricasanSelleckChemCat#S7775Trans-ISRIBTocrisCat#5284Polyamines supplementMilliporeCat#P8483Jagged1R&D systemsCat#1277-JGActivin AR&D systemsCat#338-ACLIFR&D systemsCat#7734-LFCNTFR&D systemsCat#257-NTDLL1R&D systemsCat#1818-DLBDNFR&D systemsCat#11166-BDGDNFR&D systemsCat#212-GDStemPro AccutaseThermoFisherCat#A1110501DMEM/F12ThermoFisherCat#11320033Anti-adherence solutionStemCell TechnologiesCat#07010Neuron freezing mediaCell ApplicationsCat#042-50N2ThermoFisherCat#A1370701B27 w/o vit. AThermoFisherCat#12587010B27ThermoFisherCat#17504044VitronectinThermoFisherCat#A14700GeltrexThermoFisherCat#A1413302MetforminTocrisCat#2864ForskolinTocrisCat#1099Exendin-4TocrisCat#1933Kisspeptin-10TocrisCat#2570Critical commercial assaysTaqMan gene expression assay KISS1RThermoFisherHs00261399_m1TaqMan gene expression assay GNRH1ThermoFisherHs00171272_m1TaqMan gene expression assay GAPDHThermoFisher4310884ERNeasy plus mini kitQiagenCat#74136Illumina TruSeq stranded mRNA kitIlluminaCat#20020492BCA protein assay kitThermoFisherCat#23227SingleShot cell lysis kitBio-RadCat#1725080Deposited dataBulk RNA-Seq and ATAC-Seq FASTQ filesThis paperSRA BioProject: PRJNA1085406Experimental models: Cell linesWA09WiCellRRID:CVCL_9773LiPSC-GR1.1NIH Common FundRRID:CVCL_RL65NCRM5NIH Common FundRRID:CVCL_1E75WA09 POMC-tdTomato (homozygousclone 1G)NIH/NCATSWAe009-A-3ISoftware and algorithmsfastqc (version 0.11.9)https://biodockerfiles.github.io/fastqc-0-11-9multiqc (version 1.20)https://pypi.org/project/multiqctrim-galore (version 0.6.7)https://github.com/FelixKrueger/TrimGalore/releasescutadapt (version 4.0)https://cutadapt.readthedocs.io/en/v4.0Python (version 3.8.13)https://www.python.org/downloads/release/python-3813Salmon aligner (version v1.10.0)https://github.com/COMBINE-lab/salmon/releasesEnsembl (hg38, Homo_sapiens.GRCh38.cdna.all.fa, release 109)https://useast.ensembl.org/info/website/archives/index.html?R (release 4.2.2)https://cran.r-project.org/bin/windows/base/oldfactoextra (version 1.0.7)https://CRAN.R-project.org/package=factoextratximeta (version 1.15.3)https://bioconductor.org/packages/release/bioc/html/tximeta.htmlDESeq2 (version 1.39.5)https://bioconductor.org/packages/release/bioc/html/DESeq2.htmlapeglm R package (version 1.21.0)https://bioconductor.org/packages/release/bioc/html/apeglm.htmlggplot (release 3.4.1)https://github.com/tidyverse/ggplot2/releasesRUVseq RUVg (version 1.36.0)https://www.bioconductor.org/packages//release/bioc/vignettes/RUVSeq/inst/doc/RUVSeq.htmlOrigin-Pro softwareOriginLabhttps://www.originlab.comPatchmasterHEKA Instrumentshttps://www.heka.comFitmasterHEKA Instrumentshttps://www.heka.comGraphPad PrismGraphPadhttps://www.graphpad.com Experimental model and study participant details Cell lines Human pluripotent stem cell lines used in this study included LiPSC-GR1.1 (male; RRID:CVCL_RL65) and NCRM5 (male; RRID:CVCL_1E75), obtained from the NIH Common Fund, and WA09 (female; RRID:CVCL_9773), obtained from WiCell. A tdTomatoPOMC reporter line was generated in-house from the WA09 line as described previously (Jovanovic et al., 2026) (see key resources table for additional details). All experiments involving human pluripotent stem cell lines were conducted in accordance with NIH guidelines and institutional oversight policies. No human participants or live animals were involved in this study. Cells were maintained in E8 medium on vitronectin (VN)-coated surfaces and passaged using CEPT for a minimum of seven passages prior to initiating differentiation in coated microplates or T175 flasks. Method details Scalable differentiation of hPSCs into hypothalamic neurons The hPSCs utilized in this study underwent mycoplasma testing prior to differentiation and karyotype analysis to exclude common cell culture contamination and chromosomal abnormalities. Each cell line was maintained for a series of 7 passages under feeder-free conditions (using E8 Media, ThermoFisher, A1517001) on vitronectin (ThermoFisher, A14700) as a coating substrate. CEPT was included at each passage for 24 h to support viability. On Day (−1), 20,000 hPSCs/cm2 were seeded on vitronectin-coated vessels in E8 supplemented with CEPT (e.g., 180,000 cells per well of a 6-well plate or 3.5 million cells per T175 flask). On day 0, the media was replaced with Hypo-1 medium. From Day 0–6, full media changes were performed daily, without passaging. Hypo-1 medium composition.•E6 medium (ThermoFisher, A1516401)•100 nM LDN193189 (Tocris, 6053)•2 μM A83-01 (Tocris, 2939)•4 μM XAV939 (Tocris, 3748)•0.5 μM SAG21K (Tocris, 5282)•Jagged 1 (1277-JG), LIF (7734-LF), CNTF (257-NT), Delta-like protein 1 (DLL1) (R&D systems), all at 10 ng/mL. •E6 medium (ThermoFisher, A1516401) E6 medium (ThermoFisher, A1516401) •100 nM LDN193189 (Tocris, 6053) 100 nM LDN193189 (Tocris, 6053) •2 μM A83-01 (Tocris, 2939) 2 μM A83-01 (Tocris, 2939) •4 μM XAV939 (Tocris, 3748) 4 μM XAV939 (Tocris, 3748) •0.5 μM SAG21K (Tocris, 5282) 0.5 μM SAG21K (Tocris, 5282) •Jagged 1 (1277-JG), LIF (7734-LF), CNTF (257-NT), Delta-like protein 1 (DLL1) (R&D systems), all at 10 ng/mL. Jagged 1 (1277-JG), LIF (7734-LF), CNTF (257-NT), Delta-like protein 1 (DLL1) (R&D systems), all at 10 ng/mL. On day 7, cells were dissociated with StemPro Accutase (ThermoFisher, A1110501) for a 5 min at 37°C (1 mL per well of a 6-well plate; 10 mL for T175 flask), and a cell suspension of 2.5 million cells per mL was prepared in Hypo-2 medium with CEPT. Hypo-2 medium composition.•DMEM/F12 (ThermoFisher, 11320033) + N2 (ThermoFisher, A1370701) + B27 w/o vit. A (ThermoFisher, 12587010)•4 μM XAV939 (Tocris, 3748)•25 ng/mL Activin A (R&D systems, 338-AC)•20 ng/mL BDNF (R&D systems, 11166-BD)•20 ng/mL GDNF (R&D systems, 212-GD)•200 μM Ascorbic Acid (Tocris, 4055)•50 μM db-cAMP (Tocris, 1141) •DMEM/F12 (ThermoFisher, 11320033) + N2 (ThermoFisher, A1370701) + B27 w/o vit. A (ThermoFisher, 12587010) DMEM/F12 (ThermoFisher, 11320033) + N2 (ThermoFisher, A1370701) + B27 w/o vit. A (ThermoFisher, 12587010) •4 μM XAV939 (Tocris, 3748) 4 μM XAV939 (Tocris, 3748) •25 ng/mL Activin A (R&D systems, 338-AC) 25 ng/mL Activin A (R&D systems, 338-AC) •20 ng/mL BDNF (R&D systems, 11166-BD) 20 ng/mL BDNF (R&D systems, 11166-BD) •20 ng/mL GDNF (R&D systems, 212-GD) 20 ng/mL GDNF (R&D systems, 212-GD) •200 μM Ascorbic Acid (Tocris, 4055) 200 μM Ascorbic Acid (Tocris, 4055) •50 μM db-cAMP (Tocris, 1141) 50 μM db-cAMP (Tocris, 1141) For Hyposphere formation, 3 mL of prepared cell suspension was added per well of an AggreWell 800 6-well plate (StemCell Technologies) to achieve 5000 cells per microwell, and plates were maintained in a cell culture incubator overnight. The next day, spheres were carefully transferred using a 1 mL wide bore pipet into 40 mL of Hypo-2 medium and placed in T175 flasks pre-rinsed with an anti-adherence solution (StemCell Technologies, 07010). One full AggreWell plate was transferred into each T175 flask. Spheres were maintained in Hypo-2 medium until day 11, with daily medium changes. On day 11, the medium was replaced with Hypo-3 medium. Hypo-3 medium composition.•DMEM/F12 + N2 + B27 complete (ThermoFisher, 17504044)•20 ng/mL BDNF (R&D systems, 11166-BD)•20 ng/mL GDNF (R&D systems, 212-GD)•200 μM Ascorbic Acid (Tocris, 4055)•50 μM db-cAMP (Tocris, 1141)•1 μM DBZ (Tocris, 4489) •DMEM/F12 + N2 + B27 complete (ThermoFisher, 17504044) DMEM/F12 + N2 + B27 complete (ThermoFisher, 17504044) •20 ng/mL BDNF (R&D systems, 11166-BD) 20 ng/mL BDNF (R&D systems, 11166-BD) •20 ng/mL GDNF (R&D systems, 212-GD) 20 ng/mL GDNF (R&D systems, 212-GD) •200 μM Ascorbic Acid (Tocris, 4055) 200 μM Ascorbic Acid (Tocris, 4055) •50 μM db-cAMP (Tocris, 1141) 50 μM db-cAMP (Tocris, 1141) •1 μM DBZ (Tocris, 4489) 1 μM DBZ (Tocris, 4489) On day 14, spheres were dissociated using the neural tissue enzyme dissociation kit and platform (Miltenyi Biotec). At this point, cells were cryopreserved or plated for final maturation. Cryopreservation medium: Neuron freezing medium (Cell Applications, 042–50) with CEPT. For final maturation, 100,000 dissociated cells were seeded per cm2 on Geltrex (ThermoFisher, A1569601) coated vessels in Hypo-3 medium with CEPT. Every other day, half-media changes were performed until day 28. On day 28, cells can be used for functional assays described in this study. All experiments described in this manuscript were conducted using hypothalamic neurons matured to days 28–35, which had been previously cryopreserved at day 14. Automated cell culture Scalable robotic cell culture and differentiation were carried out using the CompacT SelecT platform as previously described (Tristan et al., 2021) and following the steps described above. For sphere formation, T175 flasks were pretreated with anti-adherence solution (STEMCELL Technologies). Detailed information on immunocytochemistry and high-content imaging, live-cell neurite analysis, RNA-Seq, ATAC-Seq, MERFISH, electrophysiology, neuronal marker quantification, MEA recordings, metabolic and incretin treatments, Western blotting, RT-PCR, RNA-seq alignment and QC, differential gene expression, gene set enrichment, expression specificity analysis, LDSC, RUV-seq comparisons, ATAC-Seq processing, differential chromatin accessibility, and RNA–chromatin integration is provided in theSTAR Methods. Immunocytochemistry and high content imaging analysis hPSC-derived hypothalamic neurons at different time points were cultured as described above on glass-bottom multiwell plates (Cellvis, P24-1.5H-N). The cultures were fixed with 4% formaldehyde (Thermo-Fisher, 28908) in PBS for 15 min. Then the cultures were incubated with PBS +4% donkey serum +0.1% Triton X- for 1 h at room temperature, followed by incubation with primary antibodies overnight at 4°C. Secondary antibodies were incubated at room temperature for 1 h. Refer to Table S1 for complete list of primary and secondary antibodies. The cultures were stained with 4 μM Hoechst 33342 (VWR, PI62249) in PBS for 10 min before imaging on the Opera Phenix high-content microscope (Revvity) or Zeiss LSM 710 confocal microscope. For GnRH neurons, prior to fixation cells were treated with 5 nM Kisspeptin-10 (TOCRIS, catalog No. 2570). Live-cell imaging and neurite outgrowth analysis Neurite outgrowth was quantified using the NeuroTrack module of the IncuCyte Live-Cell Analysis System. Live-imaging of neuronal maturation was performed over a defined time course of 10 days post-thawing, and neurite length and branching were automatically analyzed using the NeuroTrack software algorithms. RNA seq iPSCs from two different healthy individuals (LiPSC-GR 1.1 and NCRM-5) and hESC (WA09) as well as derived neural progenitors and hypothalamic neurons at different differentiation time points were lysed using RLT plus buffer (Qiagen, 1053393) supplemented with 2-mercaptoethanol (Millipore-Sigma, 63689) directly in wells and RNA was extracted and purified using RNeasy Plus Mini Kit (Qiagen, 74136) according to the manufacturer’s instruction. QIAcube automated workstation was used for the extraction (Qiagen). Genomic DNA was eliminated by both the gDNA eliminator column and on-column incubation with DNase I (Qiagen, 79256). Sequencing libraries were constructed and sequenced at the Azenta sequencing facility using Illumina TruSeq Stranded mRNA kit. ATAC-Seq ATAC-Seq was performed on 50,000 to 100,000 cells according to the manufacturer’s instructions (Active Motif, 53150). The cells were pelleted and washed with 100 μL of ice-cold PBS. They were then lysed with 100 μL of ATAC Lysis Buffer on ice for 3 min and centrifuged at 500 g for 10 min at 4°C. The nuclei were incubated in 50 μL of Tagmentation Master Mix at 37°C for 30 min in a thermomixer set to 800 RPM. During this process, the Tn5 Transposase probed accessible DNA regions in the nuclei by inserting sequencing adapters into open regions. The tagged DNA was then purified and PCR-amplified. Prior sequencing, the libraries were quantified by qPCR using the KAPA library quantification kit for Illumina platforms (Roche, KR0405) using QuantStudio 12K Flex Real-Time PCR System (Thermo Fisher Scientific). All samples were normalized according to concentration and pooled. Libraries were loaded and sequenced using the NovaSeq 6000 System (Illumina). MERFISH Probe design and sample preparation A total of 110 genes previously associated to ARC nucleus (see Table S3) were identified and evaluated using the MERSCOPE Gene Panel Design Portal available at Vizgen (portal.vizgen.com) to ensure that each gene was suitable in length for probe binding and that abundance was below the recommended threshold to avoid optical crowding during imaging. Human iPSC derived neurons were grown on MERSCOPE Slides (Vizgen, 20400001), fixed with 4% Paraformaldehyde-fixed (PFA). Samples were then washed three times with 5 mL 1× phosphate buffered saline (1× PBS) and incubated with 70% ethanol at 4°C. Probe hybridization Samples were briefly washed with Sample Prep Wash Buffer (Vizgen, 20300001) and then incubated in Formamide Wash Buffer (Vizgen, 20300002) for 30 min at 37°C followed by probe hybridization with 50 μL of a custom designed MERSCOPE Gene Panel Mix at 37°C for 18–36 h. Imaging Following incubation, the samples were washed twice with 5 mL Formamide Wash Buffer (Vizgen, 20300002) at 47°C for 30 min, followed by two washes with Sample Prep Wash Buffer (Vizgen, 20300001), the samples were then stained with DAPI and Poly T Reagent (Vizgen, 20300021) for 15 min at room temperature, washed for 10 min with 5 mL of Formamide Wash Buffer (Vizgen, 20300002), then washed briefly with 5 mL of Sample Prep Wash Buffer (Vizgen, 20300001). The samples were imaged on the MERSCOPE according to the MERSCOPE Instrument User Guide https://vizgen.com/resources/merscope-instrument/. Seven 1.5 μm-thick z planes were imaged for each field of view at 60× magnification. Image analysis Vizgen’s Merlin analysis pipeline (Emanuel et al. 2020) was used to analyze the raw image files from the MERSCOPE experiment to align image stacks from the different MERFISH rounds, filter out background noise, and enhance RNA spot detection. Individual RNA molecule barcodes were then decoded using a pixel-based algorithm and an adaptive barcoding scheme that corrects misidentified barcodes not matching the provided codebook. Next, cells were segmented using information from DAPI and PolyT stain with the Cellpose algorithm. The decoded RNA molecules were then grouped into individual cells, generating single-cell RNA count matrices. Data processing We used Seurat v5.1 (Hao et al., 2024) to process and analyze MERSCOPE data. Seurat object was created with cell-by-gene.csv and cell-metadata.csv, which were generated by MERSCOPE. cell-by-gene.csv includes cell by gene count matrix, and cell-metadata.csv includes information of cell volume and cell centroids, and total gene counts. Total gene counts and cell volume were used in QC procedure. Cells that are too small or too large in volume, and cells with too low gene counts were excluded in analysis. Thresholds of volume and total gene counts were determined according to the corresponding density plots. Cell volume and total gene counts were transferred with log2(x+1) before density plots were generated. In the density plots, we identified several local minima and determined the best thresholds for keeping cells for analysis. Data analysis Before analysis, we normalized gene counts by cell volume. In detail, to account for variation in cell volume due to different cutting angles and areas, transcript counts were normalized to the median cell volume, which helps to adjust the tendency of larger volume cells to have more transcripts detected. SCT transformation method in Seurat was used to globally normalize the feature expression measurements for each cell by the total expression. Clusters were identified using Seurat function FindClusters with resolution = 0.5 using data normalized and scaled with SCT. UMAP was created with Seurat function RunUMAP. Clusters including immature cells were identified and excluded from the further analysis and display. We generated bar plot to explore overall expression of several candidate genes. Bar plot represents percentage of cells with non-zero count of corresponding gene vs. total number of cells. Patch-clamp electrophysiology POMC::tdTomato (WA09)-expressing hypothalamic neurons were plated on 12 mm glass coverslips (Carolina Biological Supply) coated with Poly-ʟ-Ornithine (Sigma Aldrich) and iMatrix-511 SILK (Amsbio). D28-D35 tdTomato-expressing hypothalamic POMC neurons were visually identified by fluorescence for patch-clamp electrophysiology recordings. The extracellular media contained BrainPhys media (STEMCELL Technologies). The intracellular solution contained (in mM): 135 K-gluconate, 5 KCl, 5 Mg-ATP, 0.5 Na2GTP, 5 HEPES, 2 MgCl2, 5 EGTA, 0.5 CaCl2, pH 7.4. Recordings were made using an Axopatch 200B (Molecular Devices) amplifier at a sampling rate of 10 kHz and a 2 kHz Bessel low-pass filter. The data was digitized using a Digidata 1550B with pClamp11 (Molecular Devices). Whole-cell configuration was obtained in voltage clamp mode and switched to current clamp mode after waiting 3 min for the membrane potential to stabilize. Spontaneous action potentials were recorded at a – 50 mV holding potential before and after local perfusion of BrainPhys media, followed by BrainPhys containing 0.1 mg/mL insulin (I9278., Sigma Aldrich) or 100 nM leptin (398-LP, R&D systems). Local perfusion was achieved using a 4 multi-barrel manifold perfusion pencil (Automate Scientific) with a 360 μm tip positioned approximately 100–200 μm away from the recorded cell. Action potentials within a 60 s window were detected using a template-based event detection algorithm (Clampfit). Neuronal marker quantification Imaging was performed on hypothalamic neurons derived from 3 independent lines and 3 replicate wells of the 24 well plate for each of the marker sets POMC, SST and TH, using high-throughput confocal microscope Opera Phenix (Revvity). Image sets were imported into Columbus (PerkinElmer) and analyzed via a standardized pipeline built from modular “building blocks” such as Find Nuclei and Find Cytoplasm/Cell. Following segmentation, automated batch analysis was performed and the number of marker-positive cells and was quantified and expressed as a percentage of the total number of nuclei within each cell line. Multi-electrode array (MEA) Neuronal activity was analyzed using the Maestro Pro multi-electrode array system (Axion Biosystems) according to the manufacturer’s protocol. Briefly, day 14 hypothalamic neurons were thawed in Hypo-3 medium with CEPT and plated at a density of 80,000 cells/well on PLO/Laminin coated 48-well or 96-well MEA plates. Twenty-four hours post-plating the medium was replaced by Hypo-3 medium and full media change was performed every 2–3 days. On day 28, to reduce insulin media was changed to Hypo-3 with B27 without insulin, and hypothalamic neuron activity was recorded on day 32, 18 days post plating. Human recombinant insulin concentration used were as follows: 0.01 mg/mL (low insulin), 0.1 mg/mL and 1 mg/mL (high insulin). Differential glucose, insulin, and incretin mimetic treatments Hypo-3 media was modified as follows to create four media conditions: low glucose [1 g/L] and B27 with insulin, low glucose [1 g/L] with B27 without insulin, high glucose [9 g/L] and B27 with insulin, high glucose [9 g/L] with B27 without insulin. Matured hypothalamic cultures were treated with one of the four modified Hypo-3 medium conditions for four days, with 3/4 media change on day 29 and day 30. On day 32, cells were gently rinsed with PBS (−/−) two-three times and lysed with RIPA buffer (ThermoFisher Scientific, 89900) supplemented with Halt Protease Inhibitor Cocktail (ThermoFisher Scientific, 78440) and collected. Lysed cells were sonicated, and debris-free lysis was collected for BCA assay and Western Blot Analysis. Incretin mimetic drug treatment was performed by supplementing unmodified Hypo-3 medium with Exendin-4 at 10 nM (Tocris, 1933) and Forskolin (Tocris, 1099) at 10 μM. We quantified both total CREB and phosphorylated CREB (Ser133) in monolayer-cultured hypothalamic neurons using the HTRF Total CREB Detection Kit (Revvity, 63ADK052PEG) in conjunction with the corresponding phospho-CREB (S133) HTRF kit. Neurons were plated in 96-well culture plates, treated with drug or vehicle for the indicated duration, and then lysed directly in-plate according to kit instructions. Lysates were transferred into low-volume white 384-well detection plates, and 16 μL of sample was incubated with donor- and acceptor-conjugated anti-CREB antibodies. The no-wash HTRF readout was acquired using a PheraStar plate reader (excitation and emission filter set according to kit manual). The ratio of phospho-CREB signal to total CREB signal was calculated for each well and used as a normalized readout of CREB activation in response to treatment. Western Blot The Jess automated Western blotting systems (ProteinSimple) were utilized for quantitative analysis of protein expression, as previously described (Tristan et al., 2021). All Western blot data are presented by lanes in virtual blot-like images. In brief, cells were harvested by scraping, pelleted, washed with PBS, flash-frozen using dry ice, and stored at −20°C until processed. Cell pellets were then resuspended in RIPA buffer (ThermoFisher Scientific, 89900) supplemented with Halt Protease and Phosphatase Inhibitor Cocktail (ThermoFisher Scientific, 78440) and lysed by sonication. Lysates were subsequently cleared of debris by centrifugation at 14,000g for 15 min and quantified using the BCA protein assay kit (Thermo Fisher Scientific). Protein quantification was performed using the 12–230 kDa plates (ProteinSimple, PS-MK15) in a JESS Capillary Western Blot analyzer following the manufacturer’s recommendations. Protein quantification was conducted using the Compass software. For a full list of primary and secondary antibodies used, please refer to Table S1. Real-time reverse transcription polymerase chain reaction (RT-PCR) RNA was extracted using the SingleShot Cell Lysis Kit (Bio-Rad Laboratories, 1725080) per the manufacturer’s instructions. Briefly, a 96-well plate was loaded with 100k cells per cm2 of iPSC-derived human hypothalamic neurons (LiPSC-GR 1.1) in complete Hypo-3 medium. After each treatment, cells were lysed for 10 min at room temperature in 40 μL of lysis buffer containing DNase and proteinase, followed by 5 min at 37°C and 5 min at 75°C. Each sample underwent RT-PCR in triplicate using Reliance One-Step Multiplex Supermix (Bio-Rad Laboratories, 12010220) and specific probe sets for each gene. The total volume of each reaction was 10 μL, and the final concentration of each gene’s primer is 1×. On a QuantStudio 3 Real-Time PCR System, the real-time PCR reactions were performed according to the manufacturer’s instructions (1 cycle of 50°C 10 min and 95°C 10 min for reverse transcription and enzyme activation, followed by 40 cycles of 95°C 15 s and 60°C 30 s for annealing and extension). Primer probe sets KISS1R (Hs00261399_m1), GNRH1 (Hs00171272_m1), and GAPDH (4310884E) were all purchased from ThermoFisher Scientific. Using the comparative threshold cycle (Ct) method, the target gene’s expression level was normalized to that of the housekeeping gene GAPDH. Kisspeptin-10 was purchased from TOCRIS (TOCRIS, 2570). For GLP1R RT-PCR experiments, differentiated neurons (LiPSC-GR 1.1) at day 32 were cultured for 96 h in one of the three different glucose conditions (low glucose 1 g/L, medium 4 g/L and high 9 g/L), followed by RNA isolation using the RNeasy Plus 20 mini kit (Qiagen, 74136) in the QIAcube equipment. cDNA was synthesized using the high-capacity RNA to cDNA kit (Applied Biosystems, 4387406). PrimeTime Gene Expression Master Mix (IDT, 1055771) was used to perform the qRT-PCR in the QuantStudio 12K Flex Real-Time PCR system. GLP1R (Hs.PT.58.39163702.g) and ACTB primers (Hs.PT.39a.22214847) were purchased from IDT. RNA-seq read alignment, quantification, and quality control We processed RNA-Seq reads as described in Xue et al. (2023). Briefly, we aligned the sequencing reads to the GRCh38 genome assembly using STAR (v2.73a) (Dobin et al., 2013) with default parameters and quantified expression levels of Gencode v19 genes (Ensembl release 103) using QoRTs (v1.3.6) (Hartley and Mullikin, 2015) per library/replicate. We generated a raw read count matrix of gene by library and a normalized mRNA expression matrix of transcripts per million (TPM). On average, we generated 60,111,933 (49,044,888 - 70,752,140) paired-end reads per library, of which 89.95% were uniquely aligned to the genome. Out of the aligned reads, 48.33% were unambiguously assigned to unique genes, indicating high quality data. We normalized the gene by library raw count matrix to TPM considering transcript length and library size information. We excluded genes with no reads in at least one library and retained only protein-coding genes. We scaled the TPM data using the “scale” function of R package (version 3.5.1) and performed a Principal Component Analysis (PCA) using the PCA function of R and used ggplot2 (version 3.5.1) for generating figures. We did not find any outlier replicate per iPSC line at any time point during differentiation. We used a similar normalization approach to compare the transcriptome of iPSC derived cells with data obtained from previously published studies. Differential gene expression analysis We performed differential gene expression analysis as described in Xue et al. (2023). We assessed gene expression changes in the iPSC lines at each differentiation stage (D7, 14, 21, or 28) relative to the initial iPSC stage (D0), as well as that of adjacent differentiation stages (D0 vs. D7, D7 vs. 14, D14 vs. D21 and D21 vs. D28) using R package DESeq2 (v1.42.1) (Love et al., 2014). We tested genes with read Counts Per Million (CPM) > 0.5 in at least 50% libraries for each comparison. We normalized the raw read counts using the median-of-ratios method, which accounts for differences in library size and RNA composition between samples. A generalized linear model was fit to the count data and the Wald test was used to identify differentially expressed genes. p-values were adjusted for multiple testing using the Benjamini-Hochberg (BH) procedure to control the false discovery rate (FDR). Genes with an FDR <5% and a fold change (FC) > 1.5 were considered differentially expressed. Gene set enrichment analysis To assess the impact of changes in gene expression through differentiation, we selected the top differentially expressed genes (FDR <10−6 and FC ≥4) and evaluated the representation of these genes in previously known biological pathways of KEGG (Liberzon et al., 2011) using the “enrichKEGG” function of the clusterProfiler (v4.11.0) R package (Yu et al., 2012). p-values were adjusted for multiple testing using the Benjamini-Hochberg (BH) procedure to control the FDR, and gene sets with FDR <5% were considered to be over-represented in the differential genes. Gene expression specificity index We normalized the raw read counts by library size and transcript length for each gene and generated a TPM matrix of differentiation time point by gene. We applied CELLEX v1.2.2 (Timshel et al., 2020) on the TPM matrix and calculated expression specificity index for each gene at a given time point. Linkage Disequilibrium Score Regression We conducted LDSC analysis (Bulik-Sullivan et al., 2015; Finucane et al., 2015) to estimate the heritability of differentially expressed genes in different traits and diseases. LDSC is a statistical method widely applied in genetic study to estimate heritability and genetic correlations based on genome-wide association study (GWAS) summary statistics. It helps us in understanding if differentially expressed genes significantly contribute to the phenotypic variation of complex traits and diseases by analyzing the relationship between linkage disequilibrium (LD) of reference genetic data (e.g., 1000 genomes) and association statistics without requiring individual-level genotype data used in GWAS. We first estimated LD scores for SNPs within 1,215 differential genes (GENCODE GRCh37 release 48 annotations) using the 1000 Genomes EUR Phase3 as reference data. SNPs mapped to HapMap3 and GWAS summary statistics were used for LDSC analysis. Heritability was estimated as enrichment of LD scores of study genes compared to pre-estimated LD scores using 1000 genome as baseline. Numbers of SNPs used in enrichment analysis for each GWAS were included in Table S3. Gene set enrichment and RUV-Seq analysis We performed enrichment of the top 100 differentially expressed genes by fold-change in the KEGG 2021 Human, GTEx Tissues (v8 2023), and Human Gene Atlas databases using Enrichr (Kuleshov et al., 2016). We generated heatmaps using the pheatmap package in R (version 1.0.12). For comparison to the Rajamani et al. study (Rajamani et al., 2018), we obtained sequencing data of the control iPSC-derived hypothalamic neurons or adult hypothalamic tissue from SRA (Leinonen et al., 2011) Bioproject PRJNA376498. We transformed the read counts using upper-quartile normalization (Bullard et al., 2010). As no spike-in controls were available, we performed a first-pass analysis using edgeR with a negative binomial GLM, and used all genes but the top 5000 by edgeR p-value as empirical control genes to normalize all samples together using RUVseq RUVg (version 1.36.0). For comparison to ARC nucleus neurons, a set of marker genes was derived from the Zhou et al. publication (Zhou et al., 2022) by compiling a list of top 10 genes for each neuronal subtype of the ARC nucleus (Table S4 in Zhou et al.), and normalized count values, using the offsets derived from RUVg were used. ATAC-seq read alignment, quantification, and quality control We processed ATAC-Seq reads and quantified sequencing reads using the approach used in Xue et al. (2023). Briefly, we removed duplicate reads and retained 97,076,117 uniquely aligned primary reads per library/replicate on average (minimum of 41,330,939) for downstream analyses. Using the filtered reads, we called ATAC-peaks as described in Rai et al. (2020), removing candidate peaks that overlap with ENCODE blacklists (Consortium, 2012) and controlling for an FDR of 5%. For each iPSC line at a differentiation time point (D0, 7, 14, 21, and 28), we merged peaks across replicates and retained peaks present in ≥2 replicates. Next, we created a master set of peaks by merging peaks across all libraries, generating a total 395,573 peaks. Finally, we used this master set of peaks to quantify the number of reads mapping to the peaks within each library and generated an accessible chromatin region count matrix. We used MultiQC (Ewels et al., 2016) to generate and aggregate QC metrics across libraries and determined all libraries are of good quality (both “per_base_sequence_quality_scores” and “per_sequence_quality_scores” meeting thresholds of “pass”). We applied ataqv (v1.2.1) (Orchard et al., 2020) to generate QC metrics, such as the number of high-quality autosomal alignments (HQAA) and transcription start site (TSS) enrichment. We identified no outlier library. Differential chromatin accessibility analysis We performed differential chromatin accessibility analysis using the ATAC-Seq read count matrix corresponding to the comparisons in differential gene expression analysis using the approach used in Xue et al. (2023). We considered peaks with FDR <5% and an FC > 1.5 to be differential open chromatin regions. Association of gene expression and chromatin accessibility We annotated any ATAC-Seq peak with a gene where the ATAC-Seq peak overlaps 50kb flanking regions of the TSS or gene body (gencode v43, https://www.gencodegenes.org/human/). To infer association of likely cis-regulatory elements and nearby gene expression, we fit a general linear regression to model the inverse-normalized TPM for a gene as the dependent variable, the inverse-normalized ATAC-Seq peak read counts as the independent variable taking into account the iPSC line and differentiation time point (0, 7, 14, 21 and 28 days) as covariables (“lm” function from the R package stats (v4.3.2)). Since RNA-Seq and ATAC-Seq libraries of each iPSC line were not simultaneously generated, we summed up raw counts of three replicates for each gene/peak of a cell line, calculated CPM prior to the inverse normalization. We used the Benjamini-Hochberg procedure to control for the number of tests and considered tests with FDR<5% and FC > 1.5 to be associated. Quantification and statistical analysis Data are presented as mean ± standard deviation (SD), unless otherwise indicated in the figure legend. Comparisons between two groups were performed using an unpaired, two-tailed Student’s t test. For comparisons among multiple groups with one variable one-way ANOVA followed by appropriate post hoc tests was used, while two variables have been assessed using two-way ANOVA with appropriate post-hoc test. Statistical significance was defined as p < 0.05 (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). Statistical analyses were conducted using GraphPad Prism version 11 (GraphPad Software, Boston, MA, USA). Experiments included at least three independent biological replicates, and sample sizes are specified in the corresponding figure legends. Additional resources This study did not generate additional resources.