Environmental and genetic risk factors of depression converge on neuronal dysfunction driven by changes in cholesterol homeostasis

Summary

Major depressive disorder (MDD) is a complex multifactorial neuropsychiatric disorder influenced by both genetic and environmental factors, but how these factors impact human neuronal function remains unclear. Using a highly defined human pluripotent stem cell (hPSC)-based prefrontal cortex (PFC) platform, we examined three high confidence environmental and genetic factors associated with depression: chronic exposure to high levels of cortisol or interferon alpha (IFN-a) and mutation in SIRT1. All three conditions induced overlapping phenotypes of neuronal dysfunction, characterized by dendritic atrophy, synaptic loss and neuronal hypoactivity across multiple cell lines. RNA sequencing uncovered converging alterations in neuronal cholesterol homeostasis. Depleting cholesterol in control neurons reproduced core depression-associated neuronal phenotypes, while cholesterol supplementation was sufficient to rescue these phenotypes in depression-associated conditions. These findings point to cholesterol imbalance as a common driver of neuronal dysfunction in MDD, linking diverse genetic and environmental risk factors through a shared cellular pathway.

Graphical Abstract

eTOC blurb

Oberst and Xu et al. identify cholesterol imbalance as a shared driver of neuronal dysfunction across genetic and environmental risk factors for depression. Their findings suggest that diverse causes of depression may converge on a common molecular pathway, offering insights into the cellular basis of this complex disorder.

Introduction

Major depressive disorder (MDD) is a complex multifactorial neuropsychiatric disorder involving both genetic and environmental factors. Major challenges in understanding its underlying pathophysiological mechanisms include the heterogeneity of clinical characteristics, suggesting the existence of different subtypes of depression with potentially distinct or only partially overlapping pathophysiology¹, limited reliability of the diagnosis², limited characterization of patient cohorts and varying definitions of depression used in genetic and other studies involving human subjects¹.

Despite these challenges, decades of research have revealed risk factors that are consistently associated with an increased incidence of depression. One of the most established risk factors for depression is chronic stress³. Increased levels of serum cortisol are consistently reported in severe depression with melancholic features⁴, a subtype of depression recognized in the DSM-5, and elevated morning cortisol has been reported to precede depression in adolescence⁵. Similarly, patients with Cushing’s disease, a rare endocrine disorder characterized by hypersecretion of cortisol, have a strongly elevated incidence of depression (>50%)^6–8, and patients undergoing corticosteroid treatment commonly develop depressive symptoms during chronic treatment^9,10, pointing to a causal role of chronically elevated cortisol levels in the development of depression in subsets of patients.

A second significant risk factor for depression is chronic inflammation. Meta-analyses have shown increased inflammatory markers in MDD^11–13 which decrease upon successful antidepressant treatment^14,15. MDD is also more common in patients who suffer from conditions associated with an activated immune system, such as systemic lupus erythematosus (SLE)¹⁶ and rheumatoid arthritis¹⁷. Notably, 25-50% of patients treated with the cytokine interferon alpha (IFN-a) develop depressive symptoms during treatment^18–22. Animal experiments have shown that depression-related behavior during IFN-a treatment is dependent on the CNS expression of the receptor for IFN-a, IFNAR1²³, suggesting a direct link between the action of altered levels of immune mediators, and IFN-a in particular, on brain cells and depression in subsets of patients.

In addition to these largely environmental factors, depression is also associated with genetic risk factors. The heritability of depression is estimated at 37%²⁴, and studies focusing on specific subsets of depression, such as early-onset or recurrent depression, have estimated higher genetic heritability^24–26. Genetic studies have identified an increasing number of gene variants that associate with MDD^27–33. A recent sparse whole-genome sequencing study used an extraordinarily well defined, clinically homogenous cohort of patients with severe and recurrent melancholic MDD³⁴, and identified SIRTUIN 1 (SIRT1), an NAD⁺ dependent deacetylase, as a locus contributing to MDD risk. Association of SIRT1 with MDD has been replicated in additional human studies^35–38, and work in rodent models has shown that pharmacological inhibition or genetic deletion of Sirt1 in the prefrontal cortex (PFC) or hippocampus is sufficient to induce depressive-like behavior^39–41.

While there is strong evidence that chronic exposures of cortisol^3–10,42 or IFN-a and mutations in SIRT1^34–38 can induce depressive symptoms in humans and animal models, little is known about potential shared disease-associated cellular or molecular phenotypes across these conditions. Human pluripotent stem cell (hPSC)-based approaches have greatly improved our ability to model neuropsychiatric disorders where it remains challenging to obtain relevant primary human tissue for functional studies and to address concerns about differences in the biology of human vs. rodent cells. In this study, we describe a combined pharmacological / genetic disease modeling approach in human PSC-derived PFC neurons, a cell type whose dysfunction has been strongly associated with depression^43,44. We have established MDD models based on chronic exposure to high levels of cortisol or IFN-a, compounds which trigger depressive symptoms in both humans and animals, and in parallel have investigated the functional consequences of loss of function of the MDD risk locus SIRT1. Our combined pharmacological / genetic disease modeling approach revealed convergent, MDD-associated cellular and molecular changes in disease-relevant human PFC neurons, characterized by neuronal atrophy and hypoactivity. We identified changes in neuronal cholesterol homeostasis as a driver of these phenotypes and further demonstrated that restoring cholesterol levels is sufficient to reverse depression-associated neuronal phenotypes.

Results

hPSC-based platform for disease modeling of depression reveals dendritic atrophy across three depression-associated conditions

Here, we developed a combined pharmacological / genetic disease modeling approach to investigate cellular and molecular phenotypes across different conditions that have been strongly associated with depression in humans and animals: chronic exposure to high levels of the glucocorticoid cortisol^3–10,42, chronic exposure to the cytokine interferon alpha (IFN-a) ^18–23, loss of function of the deacetylase SIRT1^34–41 (Fig. 1a). To define cellular phenotypes and their molecular underpinnings across these depression-associated conditions, we made use of a hPSC-based cellular platform that generates a pure population of human glutamatergic deep layer PFC neurons. We used dual-SMAD inhibition⁴⁵ in combination with transient WNT inhibition to induce a homogenous population of cortical progenitors marked by expression of the forebrain marker FOXG1 and the dorsal pallium marker PAX6 at day 20 of differentiation (Fig. 1b). We further patterned cortical progenitors into a PFC-like identity by timed exposure to FGF8⁴⁶, an organizer of anterior cortical development⁴⁷. By day 20, progenitors expressed high levels of the frontal cortex marker SP8 and low levels of the occipital cortex marker COUPTF1 (Fig. 1c). This expression pattern contrasts with the high expression of COUP-TF1 and the low expression of SP8 observed in a cortical differentiation protocol without FGF8 that gives rise to occipital cortex-like neurons^48,49 (Supplemental Fig. 1a). Post-mitotic identity was induced by replating at day 20 and transient Notch inhibition from day 20 to day 30, resulting in a nearly pure population of MAP2⁺ neurons expressing the forebrain marker FOXG1 and the cortical deep layer marker TBR1 by day 30 of differentiation (Fig. 1d). A similar differentiation efficacy was observed across different human embryonic stem cell (hESC) lines (WA09 and MEL-1) and in an induced pluripotent stem cell (iPSC) line (MSK-SRF001⁵⁰) used in this study (Fig. 1d; Supplemental Fig. 1b, c). The culture system remains largely neuronal throughout neurogenesis and maturation, as it is mostly devoid of glial cells, such as astrocytes (Supplemental Fig. 1d, e).

Figure 1: hPSC-based platform for disease modeling of depression reveals dendritic atrophy across three depression-associated conditions.

a, Schematic of depression-associated experimental models used in this study. b, Differentiation protocol to derive glutamatergic deep layer PFC-like neurons from hPSCs. c, Patterning into PFC progenitors at day 20. Left: Progenitors express high levels of the dorsal pallium marker PAX6 and the forebrain marker FOXG1. Right: Progenitors express low levels of the caudal cortical marker COUP-TF1 and high levels of the frontal cortical marker SP8. d, Differentiation into PFC neurons at day 30. Neurons express high levels of the neuronal marker MAP2, the forebrain marker FOXG1 and the cortical deep layer marker TBR1. e, Day 50 neurons express high levels of glucocorticoid receptor, GR, and IFN-a receptor, IFNAR1. f, Acute 4h treatment with cortisol or IFN-a induced the expression of known cortisol- or IFN-a induced genes, assessed by qPCR. Unpaired t-test. g, Summary of the experimental approach to assess dendritic morphology. h, j, Cortisol and IFN-a treatment induce dendritic atrophy in PFC neurons. Left, top: staining with the dendritic marker MAP2 and GFP to reveal transfected neurons. Bottom: representative reconstructions of neuronal morphology. i, k, Left: quantification of total dendritic length. One-way Anova with post-hoc Dunnett’s multiple comparison test. Right: quantification of dendritic complexity, measured by the number of dendritic branches as a function of the distance from the soma. Two-way Anova with post-hoc Dunnett’s multiple comparison test. GR, glucocorticoid receptor. PFC, prefrontal cortex. dSMADi, dual SMAD inhibition. WNTi, WNT inhibition. NOTCHi, Notch inhibition. Data are represented as mean ± SEM.

To develop pharmacological models of depression using chronic exposures of cortisol or IFN-a, we first examined the expression of the glucocorticoid receptor (GR) and the receptor for IFN-a IFNAR1 in hPSC-derived PFC neurons at day 50-51. Bulk RNA-sequencing (RNA-seq) showed high expression of NR3C1 (encoding GR) and IFNAR1 (Supplemental Fig. 1f), and immunofluorescence revealed that almost all hPSC-derived PFC neurons expressed these receptors (Fig. 1e). In addition, we observed low expression of the mineralocorticoid receptor (encoded by the gene NR3C2), which is known to be expressed in PFC neurons⁵¹ (Supplemental Fig. 1g, h). We next tested the functional activation of GR and IFNAR1 by a 4h pulse application of cortisol or IFN-a followed by qPCR to examine the expression of known cortisol or IFN-a induced genes (Fig. 1f). These experiments revealed upregulation of the known cortisol responsive genes FKBP5, ZBTB16 and TSC22D3 upon exposure to 1500 nM cortisol, while not affecting expression of the neuronal genes TUJ1 and TBR1. This cortisol concentration was chosen based on serum cortisol levels reported in Cushing’s disease patients⁵², more than 50% of whom develop clinically significant depression⁸, and represents some of the higher reported values. Similarly, 1000 U/ml IFN-a treatment, chosen based on previous studies^53,54, induced upregulation of the known IFN-a stimulated genes IRF9, ISG15 and STAT1, while left levels of TUJ1 and TBR1 unaffected (Fig. 1f). These findings show that hPSC-derived PFC neurons express the machinery necessary to respond to cortisol and IFN-a and respond transcriptionally to acute exposure to these compounds.

In addition to pharmacological conditions, our approach also includes studying the functional impact of loss of function of SIRT1, which has been associated with depression in humans and animals^34–41. We generated KO lines targeting SIRT1 in hPSCs using CRISPR-Cas9 gene editing and confirmed successful gene editing by Sanger sequencing (Supplemental Fig. 1i). Targeted hPSCs were karyotypically normal (Supplemental Fig. 1j). In addition, we confirmed full KO of SIRT1 at the pluripotency stage by Western blot (Supplemental Fig. 1k) and at the neuronal stage (day 64) by immunofluorescence (Supplemental Fig. 1l). SIRT1 KO hPSCs efficiently differentiated into PFC neurons and showed similar expression of FOXG1, MAP2 and TBR1 as wildtype neurons (Supplemental Fig. 1m).

Depressive symptoms in patients undergoing treatment with IFN-a are first reported two weeks after treatment initiation^20,55 and a similar time line of symptom onset is seen in patients that undergo corticosteroid treatment and develop psychiatric symptoms⁵⁶. Thus, to investigate potential depression-associated molecular or cellular phenotypes in hPSC-derived PFC neurons, we treated neurons with cortisol or IFN-a for 14 days to mimic the timeline observed in patients. We chose day 50 as the starting point for the treatments, as neurons are relatively mature at this stage, express high levels of synaptic proteins (Supplemental Fig. 2a, b) and have the relevant machinery to respond to cortisol or IFN-a treatment (Fig. 1e, f). Importantly, two-week treatment with either compound from day 50 to day 64 did not affect cell viability (Supplemental Fig. 2c).

Depression and chronic stress are associated with structural and functional changes in the PFC, including atrophy of dendrites⁵⁷, a phenotype also observed in cultured mouse Sirt1 KO neurons⁴⁰. We thus first assessed dendritic morphology of hPSC-derived neurons across depression-associated conditions. Day 48 neurons were transduced with a low titer lentiviral vector expressing an eGFP reporter and neuronal morphology of individual infected neurons was reconstructed at day 64 (Fig. 1g). Two-week treatment with cortisol or IFN-a from day 50 to day 64 reduced total dendritic length and complexity across PFC neurons differentiated from three different hPSC lines (Fig. 1h, i; Supplemental Fig. 2d, e). Similarly, we also observed reduction in total dendritic length and complexity in SIRT1 KO neurons compared to wildtype neurons (Fig. 1j, k). Interestingly, occipital cortex like neurons, generated by removing FGF8 during differentiation⁴⁸, did not exhibit consistent defects in dendritic morphology (Supplemental Fig. 2f–h), suggesting that the treatment-induced dendritic atrophy is more specific to PFC neurons. These data indicate that our hPSC-based disease modeling approach can capture clinically relevant features of depression pathology across all three paradigms.

Reduced synaptic density and alteration in functional excitatory synapses across depression-associated conditions

Depression is associated with hypoactivity in prefrontal cortical areas, while antidepressant treatment is associated with an increase in PFC activity^58,59. Studies in humans have reported synaptic loss in the PFC of depressed subjects^60,61 and rodent studies have revealed that fast-acting antidepressants rapidly reverse synaptic deficits in chronically stressed animals^62,63, suggesting a close link between changes in PFC synaptic transmission and modulation of depressive symptoms.

To investigate potential changes in synapses and neuronal activity across our pharmacological and genetic depression models, we next examined synaptic availability and function in depression-associated conditions (two-week treatment with with cortisol or IFN-a and SIRT1 KO) at day 64 (Fig. 2a). We first morphologically examined synaptic density by immunofluorescence using the pre-synaptic marker Synapsin 1 (SYN1) and the post-synaptic marker PSD95 together with the dendritic marker MAP2. Across all depression-associated conditions, we observed reduced density of SYN1 and PSD95 positive puncta normalized to dendritic length, as well as reduced apposition of both markers (Fig. 2b, c), pointing to loss of synaptic connections in PFC neurons under these conditions. The synaptic defect appears to be less prominent in occipital neurons, as we did not observe a consistent reduction in synaptic density across the three hPSC lines we tested (Supplemental Fig. 2i–k). Interestingly, fluoxetine, a selective serotonin reuptake inhibitor (SSRI), did not rescue the synaptic density defect across depression-associated conditions (Supplemental Fig. 2l), likely because our PFC-like neurons express very low level of fluoxetine target, serotonin transporter (SLC6A4, TPM < 1). On the other hand, ketamine, a rapid-acting antidepressant, only mildly alleviated this defect in the cortisol-treated neurons (Supplemental Fig. 2m). The lack of a rescue effect may be explained by ketamine’s proposed mechanism of disinhibiting glutamatergic neurons via NMDA receptor binding on inhibitory neurons⁶⁴, which are absent in our system.

Figure 2: Reduced synaptic density and alteration in functional excitatory synapses across depression-associated conditions.

a, Left: summary of the experimental approach of synaptic analysis. Right: representative image of staining with the postsynaptic marker PSD95, the presynaptic marker SYN1 and the dendritic marker MAP2 at day 64. b, The density of SYN1⁺, PSD95⁺ and SYN1⁺ PSD95⁺ (“Co”, double-positive) puncta is decreased in cortisol and IFN-a-treated neurons. One-way Anova with post-hoc Dunnett’s multiple comparison test. c, The density of SYN1⁺, PSD95⁺ and SYN1⁺ PSD95⁺ (“Co”, double-positive) puncta is decreased in SIRT1 KO neurons. Unpaired t-test. d, Representative traces from mEPSC recordings across conditions. e, f, Left: the amplitude of mEPSCs is reduced across depression-associated conditions compared to their respective control. Right: the frequency of mEPSCs is reduced in IFN-a treated and SIRT1 KO neurons compared to their respective control. e: two-way Anova with post-hoc Bonferroni multiple comparison correction. f: unpaired t-test. mEPSCs, miniature excitatory postsynaptic currents. SYN1, Synapsin 1. Data are represented as mean ± SEM.

To investigate functional correlates, we performed patch-clamp whole cell recordings at days 64-70 (Supplemental Fig. 3a). We found no changes in membrane capacitance or input resistance between depression-associated conditions and controls, suggesting a similar maturation stage of neurons across conditions (Supplemental Fig. 3b). However, in line with our morphological findings, we observed alterations in functional excitatory synapses, characterized by reduction in the amplitude of miniature excitatory postsynaptic currents (mEPSCs) across all conditions compared to control (Fig. 2d, e), suggesting alterations at the level of the postsynapse. The reduction in mEPSC amplitude may also reflect changes in presynaptic glutamate packaging or release, including altered function of vesicular glutamate transporters, which could reduce the glutamate content per vesicle. Alternatively, altered presynaptic lipid composition—particularly cholesterol depletion—could affect vesicle fusion dynamics and release efficiency, as cholesterol-rich domains are known to influence SNARE complex function and neurotransmitter release⁶⁵. In addition, we observed a reduction in mEPSC frequency in IFN-a treated and SIRT1 KO neurons compared to control (Fig. 2d–f), which could suggest decreased release probability or a decrease in the total number of synapses. However, since synaptic loss is consistent across conditions (Fig. 2b, c), the reduced mEPSC frequency in IFN-a treated and SIRT1 KO neurons likely indicates additional alterations in the presynaptic compartment. While we observed a significant reduction in synaptic puncta density and in mEPSC amplitude in the cortisol treatment group, mEPSC frequency remained unchanged. One possible explanation is that the remaining synapses may compensate by maintaining or increasing their release probability, thereby preserving overall mEPSC frequency despite a reduction in synapse number. This has been observed in other contexts where homeostatic mechanisms preserve synaptic output in the face of structural deficits⁶⁶. Additionally, it is possible that a subset of functionally silent synapses was preferentially lost, while active synapses remained, thus preserving spontaneous synaptic transmission as measured by mEPSCs⁶⁷. Taken together, these experiments show largely overlapping morphological as well as functional synaptic deficits across conditions.

Calcium imaging reveals neuronal hypoactivity and reduced response to glutamate in depression-associated conditions

Since the reduced synaptic density and reduced mEPSC amplitude across all conditions pointed to changes in neuronal activity and deficits in the postsynaptic compartment, we next sought to investigate spontaneous neuronal activity and glutamate receptor dependent activity by performing calcium-imaging experiments (Fig. 3a). We first imaged spontaneous Ca²⁺ transients at day 64 using a calcium sensitive dye. In line with the patch-clamp recordings, we observed a small but consistent reduction in the frequency of spontaneous Ca²⁺transients across all depression-associated conditions compared to control (Fig. 3b, c), suggesting hypoactivity of neurons across depression-associated conditions. Reduced frequency of spontaneous Ca²⁺ transients upon chronic treatment with cortisol or IFN-a was consistently observed in PFC neurons differentiated from two different hPSC lines (Fig. 3b; Supplemental Fig. 3c). We next examined the response of neurons to stimulation with potassium chloride (KCL), which depolarizes neurons by activating voltage sensitive calcium channels. Bath application of KCL during the recording resulted in strong increase in calcium signal in both control and depression-associated conditions and we detected no statistically significant difference in the magnitude of KCL-induced calcium signal intensity increase between control and conditions (Supplemental Fig. 3d, e). These data suggest that there are no major alterations in voltage sensitive calcium channels and in membrane integrity across conditions.

Figure 3: Calcium imaging reveals neuronal hypoactivity and reduced response to glutamate in depression-associated conditions.

a, Summary of the experimental approach to assess neuronal activity using calcium-imaging experiments. b, c, Reduced number of spontaneous calcium transients in depression-associated conditions at day 64. b: one-way Anova with post-hoc Dunnett’s multiple comparison test. c: unpaired t-test. d, Pre-treatment with the AMPA antagonist NBQX and/or the NMDA antagonist MK-801 decreases glutamate-induced calcium signal intensity increase in control neurons. One-way Anova with post-hoc Dunnett’s multiple comparison test. e, f, Reduced increase in calcium signal intensity upon addition of glutamate in neurons from depression-associated conditions. e: one-way Anova with post-hoc Dunnett’s multiple comparison test. f: unpaired t-test. Images in d-f are shown using the ”Fire” LUT from ImageJ. Data are represented as mean ± SEM.

Since our patch-clamp recordings revealed a reduction in mEPSC amplitude across all depression-associated conditions suggesting changes at the level of the postsynaptic compartment, we hypothesized that changes in the availability of glutamate receptors could underlie the observed neuronal hypoactivity. To test this hypothesis, we next directly investigated functional glutamate receptor availability by stimulation with glutamate during the recording. Treatment of control neurons with 100 μM glutamate resulted in a strong increase in calcium signal (Fig. 3d). To ascertain that increase in calcium signal upon glutamate stimulation is dependent on glutamate receptors, we pre-treated cultures with 2,3-dioxo-6-nitro-7-sulfamoyl-benzo[f]quinoxaline (NBQX) which is a competitive antagonist of AMPA and kainite glutamate receptors, and/or Dizocilpine (MK-801) which is an antagonist of the NMDA glutamate receptor, respectively. Pre-treatment with NBQX or MK-801 decreased glutamate-induced calcium signal by approximately half fold (NBQX: 42% ± 5.9% of control; MK-801: 60% ± 9.2% of control), and combined pre-treatment with both compounds almost completely abolished response to glutamate (10% ± 1.3% of control) (Fig. 3d). In comparison to control neurons, neurons across depression-associated conditions showed markedly reduced calcium signal intensity upon addition of glutamate across three different hPSC lines (Fig. 3e; Supplemental Fig. 3f, g). Interestingly, at 1 and 5 days of cortisol or IFN-a treatment, we did not observe differences in spontaneous neuronal activity or response to glutamate stimulation (Supplemental Fig. 3h–k), suggesting that only chronic exposures of cortisol or IFN-a, but not acute treatment, lead to neuronal hypoactivity. Together, these experiments reveal marked functional changes in PFC neurons across all depression-conditions characterized by reduced spontaneous activity and a diminished response to glutamate.

RNA-seq reveals dysregulation of cholesterol homeostasis in depression-associated conditions

Our disease modeling approach revealed largely overlapping cellular disease phenotypes across all depression-associated conditions, marked by neuronal atrophy, hypoactivity and reduced response to glutamate stimulation. To gain insight into molecular changes underlying the observed phenotypes, and to investigate if different conditions act through common molecular pathways to induce these phenotypes, we next performed bulk RNA-sequencing at day 51, day 56 and day 64 (corresponding to 1 day, 5 days and 14 days of treatment) for the pharmacological conditions, and at day 64 for SIRT1 KO neurons (Fig. 4a, Supplemental Fig. 4a, b).

Figure 4: RNA-seq reveals dysregulation of cholesterol homeostasis in depression-associated conditions.

a, Summary of the experimental approach of the RNA-sequencing experiments. b, Expression of selected cortisol-regulated genes in cortisol-treated neurons versus control. TPM, transcript per million. c, Expression of selected interferon stimulated genes in IFN-a-treated neurons versus control. TPM, transcript per million. d-f, Gene Set Enrichment Analysis (GSEA) was performed using pairwise comparisons of control vs. condition. Waterfall plot of gene sets depleted in cortisol-treated (d), IFN-a-treated (e) and SIRT1 KO (f) neurons, compared to the control, are shown. Selected pathways are highlighted in color. g, GSEA plots of representative pathways related to cholesterol homeostasis that are depleted in cortisol-treated, IFN-a-treated and SIRT1 KO neurons. h, Reduced neuronal cholesterol content in depression-associated conditions. Left: staining with the neuronal marker MAP2 and Filipin III, a fluorescent compound that stains membrane-associated free cholesterol. Right: quantification of Filipin III staining intensity. One-way Anova with post-hoc Dunnett’s multiple comparison test. Data are represented as mean ± SEM.

Using DESEQ2, we identified a total of 96 differentially expressed genes (DEGs) upon treatment with cortisol across all time points (padj < 0.1), with the largest number of DEGs at 1 day and 5 days of treatment (Supplemental Table 1). Upregulated genes included known cortisol-induced genes, including FKBP5, a co-chaperone of HSP90 which regulates sensitivity of the glucocorticoid receptor, the Cannabinoid Receptor 1 (CNR1) and lipoprotein lipase (LPL), both of which have been shown to be upregulated in response to glucocorticoids in adipose tissue^68,69 (Fig. 4b). Interestingly, expression of LPL strongly increased over treatment duration. For IFN-a treatment, we identified a total of 300 DEGs across all time points (padj <0.1), with the largest number of DEGs at day 5 of treatment (Supplemental Table 1). IFN-a induced a well-described interferon-stimulated gene (ISG) response, including the increased expression of IFI6, IFIT1 and B2M, (Fig. 4c, Supplemental Fig. 4c). In addition, several members of the PARP protein family were also strongly upregulated, including PARP9, PARP12 and PARP14, which play important roles in lipid metabolism⁷⁰ and inhibition of which has been associated with antidepressant effects⁷¹ (Supplemental Fig. 4c). In comparison to the pharmacological conditions, SIRT1 KO neurons showed a much larger number of DEGs at day 64. We detected a total of 2068 DEGs (Supplemental Table 1), including genes that are known to be regulated by SIRT1, such as Monoamine oxidase A (MAOA)⁷², a mitochondrial enzyme involved in the catabolism of catecholamine neurotransmitters, and Lipin-3 (LPN), which is involved in the regulation of lipid storage, lipoprotein synthesis and inflammation. Across all conditions, many DEGs have previously been associated with depression, including the cannabinoid receptor CNR1 known to be upregulated specifically in the PFC in depression⁷³ (upregulated in cortisol treated neurons), and ISG15⁷⁴ (upregulated in IFN-a treated neurons) (Supplemental Fig. 4d). Among the top 250 genes enriched in each depression-associated condition, no gene was shared across all three conditions, and only a small number of genes were shared in two conditions (Supplemental Fig. 4e). These comprised 11 genes shared between cortisol and IFN-a treated neurons (including upregulation of the cytochrome CYP1B1 which is involved in cholesterol metabolism, and nicotinamide N-methyltransferase (NNMT) which is involved in the modulation of cholesterol, glucose and triglyceride metabolism⁷⁵); and 1 gene shared between SIRT1 KO and IFN-a treated neurons (downregulation of RTL1 which is involved in the regulation of neuronal excitability and has been associated with anxiety and depressive-like behavior in mice⁷⁶). The relatively small overlap of individual DEGs across conditions, and the lack of shared DEGs across all conditions, suggested that the overlap in cellular phenotypes is likely not mediated by changes in the expression of a common set of genes.

To investigate if depression-associated conditions converged on the observed neuronal phenotypes by affecting the expression of different genes within common molecular pathways, we next performed Gene Set Enrichment Analysis (GSEA)^77,78. We first examined gene sets that were enriched across depression-associated conditions at day 64 (Supplemental Fig. 5a–c). Consistent with DEGs, we observed enrichment of gene sets related to glucocorticoid receptor signaling in cortisol-treated neurons, gene sets related to antiviral signaling in IFN-a-treated neurons, and gene sets belonged to the categories “Immune” and “Metabolism, Energy” in SIRT1 KO neurons. We next examined gene sets that were depleted within and across different conditions (Fig. 4d–f). In cortisol-treated neurons, only a small number of gene sets were depleted, mainly in the “Epigenetic” category, along with “Regulation of Clathrin Dependent Endocytosis” and “NR1H3 NR1H2 Regulate Gene Expression Linked to Cholesterol Transport and Efflux”. Interestingly, liver X receptors (LXRs) (encoded by NR1H3 and NR1H2) are activated by oxysterols, a metabolite of cholesterol, and act as transcription factors that regulate cholesterol homeostasis. In IFN-a treated neurons, most depleted gene sets belonged to the category “Excitability, Synapse” and cholesterol homeostasis, including “Regulation of Cholesterol Efflux” and “Cholesterol Binding”. In SIRT1 KO neurons, most depleted gene sets belonged to the categories “Excitability, Synapse” and “Lipid, Hormone”, including “Cholesterol Biosynthesis” and “Cholesterol Metabolism with Bloch and Kandutschrussell Pathways”. Notably, across all depression-associated conditions, many of the top depleted gene sets were related to cholesterol homeostasis (Fig. 4g). Importantly, we did not observe depletion of cholesterol related gene sets at day 5 of cortisol or IFN-a treatment and found instead enrichment of cholesterol biosynthesis related gene sets at day 1 of treatment (Supplemental Fig. 5d, e), consistent with our finding that only chronic two-week treatment, but not 1- or 5-day treatments, induces functional defects in the PFC neurons (Supplemental Fig. 3h–k). This is in line with human studies that depressive symptoms in patients occur after chronic treatment with corticosteroids or IFN-a and not after acute exposure^20,55,56 and parallels previous work in rodent models where depressive symptoms and associated neuronal hypoactivity in response to corticosterone occur only after longer treatment^79,80. While we did not observe significantly increased neuronal activity or response to glutamate stimulation at 1 day of treatment in our culture system, it is possible that neuronal activity is transiently increased at later days of treatment.

Together, dysregulation of gene sets involved in cholesterol homeostasis across all depression-associated conditions at day 64, and lack of depletion of such gene sets at days 1 and 5 of treatment with cortisol or IFN-a when depression-associated neuronal phenotypes are not yet discernible in our pharmacological models, pointed to a potential role of cholesterol homeostasis in driving overlapping neuronal phenotypes across conditions. To corroborate the observed transcriptomic changes related to cholesterol homeostasis at day 64, we directly examined cholesterol availability in depression-associated versus control neurons. Staining with Filipin III, a fluorescent compound that stains membrane-associated free cholesterol, revealed reduced cholesterol content across conditions compared to control, as measured by fluorescence intensity (Fig. 4h). This effect was consistent across several different hPSC lines (Supplemental Fig. 6a). Measurement of total cholesterol using an Amplex Red assay also revealed reduced cholesterol content in cultures from depression-associated conditions (Supplemental Fig. 6b), further confirming changes in cholesterol homeostasis across conditions.

Reducing cholesterol induces core depression-associated neuronal phenotypes and cholesterol supplementation reverses these deficits

Cholesterol has been shown to influence neuronal transmission by affecting endocytosis, exocytosis and membrane lateral diffusion of neurotransmitter receptors, ion channels and growth factor receptors. Alterations in brain cholesterol metabolism and availability have been associated with changes in neuronal excitability^81,82 and have been linked to neuropsychiatric disorders including depression^83–85. Thus, to investigate if reduced levels of cholesterol could underlie the shared neuronal phenotypes observed across our pharmacological and genetic depression models, we sought to functionally manipulate cholesterol content in hPSC-derived PFC neurons.

We first tested the implications of acute cholesterol depletion from neuronal membranes using methyl-β-cyclodextrin (MβCD). MβCD binds plasma membrane cholesterol and makes it water-soluble, thus reducing plasma membrane cholesterol content. 15 minutes treatment with increasing concentrations of MβCD resulted in a dose-dependent reduction of neuronal cholesterol, as assessed by measuring fluorescence intensity of Filipin III staining (Supplemental Fig. 6c). To test the effect of acute cholesterol depletion on glutamate-dependent neuronal activity, one of the central phenotypes in our disease models, we pretreated cultures with MβCD before glutamate stimulation in calcium imaging experiments. Pretreatment of cultures with 1 mM MβCD significantly reduced calcium signal intensity upon stimulation with glutamate (Supplemental Fig. 6d) without affecting cell viability (Supplemental Fig. 6f), revealing that depletion of cholesterol from membranes affects glutamate receptor-dependent transmission in hPSC-derived PFC neurons.

Some studies have reported that MβCD might affect the distribution of different membrane components and not just affect cholesterol levels. Thus, to target cholesterol more specifically and to lower cholesterol levels in a more physiological manner, we next sought to treat neurons with cholesterol lowering agents. We chose two compounds that target two different key enzymes involved in cholesterol metabolism: zaragozic acid, which is a competitive inhibitor of squalene synthase, and mevastatin, which is a competitive inhibitor of the rate limiting enzyme in cholesterol synthesis, HMG CoA reductase (Fig. 5a). As previously reported in other in vitro studies⁸⁶, treatment of neurons with either compound for 5 days from day 59 to day 64 (Fig. 5b) resulted in a reduction of total cholesterol content as assessed by Filipin III staining (Fig. 5c) and by measuring total cholesterol using an Amplex Red assay (Supplemental Fig. 6e), with the magnitude of the effect comparable to the decrease in cholesterol levels observed across our depression models (Fig. 4h).

Figure 5: Pharmacological manipulation of cholesterol levels induces neuronal hypoactivity and mimics core depression-associated phenotypes.

a, Schematic showing cholesterol biosynthesis pathway and mode of action of mevastatin and zaragozic acid. b, Summary of the experimental approach to manipulate cholesterol levels using mevastatin and zaragozic acid. c, Decreased cholesterol content in neurons treated with mevastatin or zaragozic acid. Left: staining with the neuronal marker MAP2 and Filipin III, a fluorescent compound that stains membrane-associated free cholesterol. Right: quantification of Filipin III staining intensity. d, The density of SYN1⁺, PSD95⁺ and SYN1⁺ PSD95⁺ (double-positive) puncta is decreased in neurons treated with mevastatin or zaragozic acid. e, Reduced number of spontaneous calcium transients in neurons treated with mevastatin or zaragozic acid. f, Reduced increase in calcium signal intensity upon addition of glutamate in neurons treated with mevastatin or zaragozic acid. Images are shown using the ”Fire” LUT from ImageJ. c-f: One-way Anova with post-hoc Dunnett’s multiple comparison test. Data are represented as mean ± SEM.

We then sought to investigate potential alterations in synapses, as reduced synaptic density is a central finding in our depression-associated conditions, and cholesterol is known to play an important role in synaptic development and function. Staining with Syn1 and PSD95 revealed significant reduction in the density of synaptic puncta in neurons that were treated with mevastatin or zaragozic acid compared to controls (Fig. 5d). We next examined functional correlates by performing calcium-imaging experiments. Treated neurons showed a significantly reduced frequency of spontaneous calcium transients (Fig. 5e), mimicking the reduction in spontaneous activity observed across our depression-associated conditions. Similarly, stimulation with glutamate resulted in a blunted calcium signal intensity increase in neurons treated with cholesterol lowering agents compared to control neurons (Fig. 5f). Taken together, these findings show that pharmacological reduction of neuronal cholesterol content mimics the neuronal phenotypes described in our pharmacological and genetic depression models, suggesting that alterations in cholesterol content are sufficient to induce these phenotypes.

We next sought to investigate whether increasing cholesterol levels could rescue depression phenotypes across conditions. To this end, we exposed neurons to MβCD-cholesterol^87,88 (Fig. 6a). As previously reported^87,88, this treatment significantly increased cholesterol levels across conditions and resulted in total cholesterol levels similar to that of the control (Fig. 6a) without affecting neuronal viability (Supplemental Fig. 6f). We next analyzed core depression-associated phenotypes, including synaptic density and response to glutamate stimulation across conditions upon cholesterol rescue. Strikingly, increasing cholesterol levels was sufficient to rescue these phenotypes across all three depression-associated conditions (Fig. 6b–d). Taken together, these findings show that core depression-associated phenotypes of hPSC-derived PFC neurons can be rescued by increasing cholesterol levels and establish a central role of cholesterol metabolism in associated neuronal dysfunction.

Figure 6: Normalization of membrane cholesterol levels rescues core depression-associated phenotypes.

a, Summary of the experimental approach (left) and increased membrane cholesterol levels upon treatment with cholesterol-MβCD complex (right). b, Increased membrane cholesterol levels rescue synaptic phenotypes across depression-associated conditions. c, Increased membrane cholesterol levels rescue response to glutamate across depression-associated conditions. d, Schematic summarizing the rescue effect of increasing membrane cholesterol levels. One-way Anova with post-hoc Dunnett’s multiple comparison test. Data are represented as mean ± SEM.

Discussion

The core depression-associated phenotypes identified in this study, reduced synaptic density, neuronal hypoactivity and reduced response to glutamate stimulation, are in line with and expand on human studies that have reported hypoactivity in prefrontal cortical areas in depression^58,59. Our findings that reduced cholesterol levels act as a driver of depression-associated neuronal phenotypes are supported by previous studies that have reported reduced cholesterol levels in the PFC of depressed subjects and in animal models of depression^83–85, and are consistent with the known role of cholesterol in the regulation of neuronal function. As a structural component of neuronal plasma membranes and the major lipid component of lipid rafts, cholesterol plays an important role in the modulation of synaptic transmission by affecting pre- and postsynaptic function^81,89,90, and has been reported to directly regulate TrkB signaling in a concentration-dependent manner⁹¹. In this study, we show that restoring cholesterol levels was sufficient to rescue depression-associated neuronal phenotypes, underscoring the importance of the overall balance in the cholesterol synthesis in neuronal function. Previous studies have shown that treatment of primary neurons with statins decreases synaptic density and impairs synaptic vesicle release⁹². While statins lower cholesterol by inhibiting HMGCR and are widely used for cardiovascular disease, their effects on depression are mixed. Some studies link statin use to increased risk for depression^93,94, others describe a protective effect^95–97, or report no association^98–101. These discrepancies may stem from several factors, including potential bias in the form of more health consciousness in patients receiving statin treatment, indirect effects of statins on quality of life due to decreased risk of cardiovascular events and pleiotropic effects of statins on other cellular functions. Importantly, our study does not address if statin use increases risk of depression in humans and should not be interpreted in this context.

How does chronic exposure to high levels of cortisol or IFN-a and loss of function of SIRT1 induce changes in neuronal cholesterol homeostasis? IFN-induced hypocholesteremia is well-described in humans and linked to downregulation of gene expression involved in cholesterol biosynthesis and reduced cellular cholesterol content in immune cells as part of the antiviral response^{102,103,104,105,106}. Our finding in hPSC-derived PFC neurons show that chronic IFN-a treatment alters expression of genes involved in cholesterol efflux and homeostasis, leading to a reduction in total cholesterol content, in line with findings in other cell types. An elevated cortisol level has been shown to induce a reduction in cholesterol levels in the medial prefrontal cortex of rodents⁸⁵, and exogenous cholesterol was reported to have a protective effect on neuron morphology¹⁰⁷. We found that in hPSC-derived PFC neurons, cortisol regulates the expression of genes involved in cholesterol synthesis and homeostasis in a treatment duration-dependent manner: acute treatment transiently upregulates genes involved in cholesterol synthesis, while chronic two-week treatment alters expression of PCSK9, PPARG and liver X receptors (LXRs), contributing to dysregulation of cholesterol homeostasis. Finally, previous studies have shown a central role for SIRT1 in the regulation of cholesterol homeostasis in multiple cell types^108,109, potentially through epigenetic regulation of genes involved in cholesterol homeostasis¹¹⁰. In particular, Sirt1 inhibition decreases the expression of genes involved in cholesterol synthesis and reduces total cholesterol levels in rat primary cortical cultures¹⁰⁹. Consistent with this finding, we find that SIRT1 loss-of-function in hPSC-derived PFC neurons decreased expression of genes involved in cholesterol biosynthesis and increased expression of genes involved in its catabolism. In addition, we find dysregulated expression of genes involved in cholesterol transport, such as STARD3, increased expression of which has been shown to repartition cholesterol to endosomes at the expense of the plasma membrane¹¹¹. Taken together, our findings support the notion that chronic cortisol, IFN-a and loss of SIRT1 act via changes in the expression of distinct sets of genes, but involved in various aspects of cholesterol homeostasis, converging on a reduction in neuronal cholesterol availability.

Previous studies using iPSC lines derived from MDD patients identified changes in electrophysiological properties¹¹² and neuronal morphology^113,114 that may contribute to disease pathology and antidepressant resistance. One study reported altered lower membrane capacitance in MDD patient-derived neurons, which was not significantly different between our control and depression-associated conditions, potentially due to different maturation stages (21 days in their study vs. 64 days in our study). As discussed in the limitations below, systematic, larger scale studies of well characterized MDD patient-derived cells and matched controls are needed to confirm and expand on these findings.

Limitations of the Study

While using a homogenous 2D culture system enabled precise dissection of pharmacological and genetic manipulations in a defined population of human cortical neurons, it does not capture of the complex interplay with other cell types, such as astrocytes and microglia. Future studies using more complex 3D culture systems as well as in vivo studies will be important to understand cellular interactions. Conceptually, our study focused on melancholic depression, but given the genetic and clinical heterogeneity of MDD, future studies using large sets of patient specific iPSC lines will be essential to assess whether similar phenotypic and molecular convergence occurs for other environmental or genetic risk factors contributing to distinct depression subtypes. Finally, many of the neuronal phenotypes described in this study are not exclusive to MDD and have also been described in schizophrenia¹¹⁵, Alzheimer’s disease¹¹⁶, Multiple Sclerosis¹¹⁷, and Huntington’s disease^118,119. Additional studies using patient-specific iPSC-derived neurons from various disease conditions will be needed to parse out fine-grained correlations between cellular phenotypes and specific sets of clinical symptoms across diagnoses.

Resource Availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Lorenz Studer (studerl@mskcc.org).

Materials availability

Cell lines generated in this study are available from the Lead Contact without restriction.

STAR Methods

Experimental model and study participant details

hPSC culture

All work using hPSCs was approved by and conducted in compliance with the Tri-Institutional Stem Cell Initiative Embryonic Stem Cell Research Oversight Committee (Tri-SCI ESCRO) at Memorial Sloan Kettering Cancer Center, Rockefeller University and Weill Cornell Medicine. hPSC line WA09 (H9; 46XX) was from WiCell Stemcell Bank. hPSCs were maintained in Essential 8 medium on vitronectin-coated plates at 37°C at 5% CO₂ with routine testing for mycoplasma. hPSCs were passaged as clumps every 4-5 days using 0.5 M EDTA in DPBS.

Method details

PFC differentiation

hPSC-derived glutamatergic prefrontal cortex (PFC) neurons were generated using an optimized protocol based on dual-SMAD and WNT inhibition^48,49 and timed exposure to FGF8⁴⁶. Briefly, hPSCs were dissociated into single cells using Accutase and plated at 300,000 cells / cm² onto matrigel-coated plates in E8 medium supplemented with 10 μM Y-27632 (day -1). On day 0-3, cells were fed daily with E6 medium supplemented with 10 μM SB431542, 100 nM LDN193189 and 2 μM XAV939 to induce anterior neuroectodermal patterning. XAV939 was removed from the medium from day 4 onward. From day 9-20, cells were cultured in 1:1 Neurobasal and DMEM/F12 medium supplemented with 1X N2, 1X B27 without Vitamin A and Penicillin Streptomycin. To achieve PFC patterning, 50 ng / ml FGF8 was added to the medium. On day 20, cells were dissociated using Accutase for 30 min and replated at a density of 220,000 cells / cm² in plates coated with poly-L-ornithine, laminin and fibronectin, or cryopreserved in STEM-CELLBANKER solution at approximately 10 million cells / vial. After replating, cells were cultured in low-glucose neuronal culture medium, containing 1:9 Neurobasal and Neurobasal-A medium supplemented with 1X B27 without Vitamin A, 1X GlutaMAX Supplement, Pen/Strep, 10 ng/ml BDNF, 10 ng/ml GDNF, 100 μM Cyclic Monophosphate Sodium Salt and 100 μM Ascorbic Acid. During replating, 10 μM Y-27632 was added to increase cell viability. During the first 10 days after replating, medium was supplemented with 10 μM DAPT. Half of the medium was replaced every 4-5 days.

OCC differentiation

PSC-derived occipital cortex like neurons were generated as previously described^48,49. The same differentiation protocol as provided above for PFC neurons was used, except that FGF8 was omitted from the differentiation medium.

Drug treatments

Treatment with cortisol (Hydrocortisone) or purified recombinant human IFN-a was started at day 50 and continued until day 64, unless otherwise specified. A concentration of 1500 nM cortisol or 1000 Units / ml IFN-a in neuronal culture medium was used unless otherwise specified. 75% of the medium was changed every 4 days during treatment. Treatment with Methyl-β-cyclodextrin was started on day 64. For immunocytochemistry, cells were treated with different concentrations of Methyl-β-cyclodextrin as indicated in the figure for 15 minutes before fixation. For calcium imaging experiments, neurons were pre-treated with 1 mM Methyl-β-cyclodextrin for 5 minutes before the recording. Treatment with mevastatin or zaragozic acid was started at day 59 and continued until day 64. A concentration of 10 μM mevastatin and 100 μM zaragozic acid in neuronal culture medium was used. 75% of the medium was changed every 4 days during treatment. Treatment with MbCD-cholesterol complex was started on day 64 and continued for 24 hours before analysis. Cells were treated 50 μM MbCD-cholesterol complex unless otherwise stated. Treatment with 1 μM ketamine was started on day 64 and continued for 24 hours before analysis. Treatment with 5 μM fluoxetine was started on day 64 and continued for 14 days before analysis.

Morphological reconstructions

hPSCs derived neurons were infected with a low titer lentivirus expressing an eGFP reporter on day 48 and fixed at day 64. The eGFP reporter signal was amplified by immunofluorescence staining using an anti-GFP antibody and individual GFP labeled neurons were imaged using a fluorescence microscope. Cells were co-stained with MAP2 to allow dendritic tracing. Neuronal morphology was reconstructed using the filament tracing function in Imaris software. Measurement of total neurite length and neuronal complexity using Sholl analysis (sampled every 20 μm) were performed in the Imaris platform and extracted for further analysis.

Immunofluorescence

Cultured cells were fixed with 4% PFA in PBS for 15 min at RT, permeabilized and blocked for 20 min in 0.5% Triton X-100 in PBS with 3% BSA. Primary antibodies were incubated overnight at 4°C in the same blocking solution. Primary antibodies used in this study are listed in STAR Methods. Cells were then washed 3x with PBST (PBS with 0.1% Tween-20). Secondary antibodies conjugated to either Alexa 488, Alexa 555 or Alexa 647 were incubated for 45 min and cell nuclei were stained with 5 μM 4′-6-diamidino-2-phenylindole (DAPI) in PBS. For Filipin III staining, DAPI was omitted and instead 2 μg / ml Filipin III was added during incubation with secondary antibodies. After incubation, cells were washed twice with PBST and once with PBS. Microscopy was performed using a standard inverted epifluorescence microscope (Olympus IX71) or a Nikon A1R HD25 confocal laser scanning microscope.

Image analysis

Quantification of immunofluorescence images was performed manually using FIJI¹²⁰. Tracing of dendrites for normalization of synaptic density was performed using the NeuronJ plugin in FIJI.

RNA extraction and qRT-PCR

Cells were collected in Trizol and total RNA was extracted using the RNeasy Mini Kit with an on-column DNA digestion step. cDNA was generated using the iScript Reverse Transcription Supermix for RT-qPCR and qPCR reactions were performed using SsoFast EvaGreen^®Supermix. Results were normalized to the housekeeping gene GAPDH. Primers used in this study are listed in STAR Methods.

Electrophysiological recordings

Neurons were plated in 35 mm dishes coated with PO/FN/LAM as described in the cell culture section, and whole-cell patch clamp recordings were performed on days 64-71 of differentiation. Neurons were kept at 28 – 30° C and perfused with freshly made, CO₂/O₂-bubbled artificial cerebrospinal fluid (aCSF) containing (in mM): 125 NaCl, 2.5 KCl, 1.2 NaH2PO4, 1 MgSO4, 2 CaCl2, 25 NaHCO3 and 10 D-glucose (pH 7.4 and 300-310 mOsmol / kg). Neurons were recorded under tetrodotoxin at 1 μM and 4-Aminopyridine at 100 μM. Pipettes of 3-7 MΩ resistance were filled with an intracellular solution containing (in mM): 120 CsMeSO4, 8 NaCl, 0.3 EGTA, 10 Tetraethylammonium-Cl, 10 HEPES, 2 Mg2-ATP, 0.3 Na2-GTP, 3 QX-314-Cl and 13 biocytin (pH 7.3 and 285–290 mOsmol / kg). Recordings were performed in voltage-clamp at −70 mV on a computer-controlled amplifier (MultiClamp 700B Axon Instruments, Foster City, CA) and acquired with an Axoscope 1550B (Axon Instruments) at a sampling rate of 10 kHz and low-pass filtered at 1 kH.

Calcium imaging and analysis

Neurons were incubated in loading solution containing 5 μM Calbryte 590 AM (Cal-590) calcium indicator diluted in HEPES-Buffered Hanks Balanced Salt Solution with 0.1% Pluronic F-127 at 37 °C for 20 minutes. Following incubation, neurons were washed twice with modified Tyrode’s solution containing 25 mM HEPES, 140 mM NaCl, 5 mM KCl, 1 mM MgCl₂, 10 mM glucose, 2 mM CaCl₂, 10 μM glycine (pH 7.2–7.4) which was pre-warmed to 37 °C. After 3-5 min equilibration, calcium imaging was performed using a Nikon A1R HD25 confocal laser scanning microscope equipped with a 20x long working distance objective and a Tokai Hit stagetop incubator for temperature and CO₂ control. Frames were captured for 5 minutes (spontaneous activity recordings) or 3 minutes (response to different compounds) at 1 frame/second. Data analysis, including automatic ROI selection, spike detection, and calculation of signal intensity over time (F/F0) was performed using the CALIMA open-source software¹²¹. For spontaneous calcium transient experiments, the number of spikes per second is shown. For KCL and glutamate stimulation experiments, a baseline of 60 seconds was recorded before addition of 50 mM KCL or 100 μM glutamate. KCL or glutamate remained in the imaging solution throughout the recording. Signal intensities recorded upon addition of KCL or glutamate were normalized to the average F/F0 of the baseline recording. For statistical analysis, averaged normalized signal intensities (F/F0 value normalized to 60 sec baseline) of frames 70-180 were included. Frames 60-70 were not included in analysis to avoid artifacts resulting from opening of the stagetop incubator during addition of KCL or glutamate. Images are depicted using the “Fire” Look Up Table (LUT) from ImageJ.

For glutamate receptor antagonist experiments, neurons were pre-treated with 10 μM MK-801 and/or 25 μM NBQX during the first minute of the recording before addition of glutamate. MK-801 and/or NBQX remained in the imaging solution throughout the recording.

For acute depletion of cholesterol, neurons were pre-treated with 1 mM Methyl-β-cyclodextrin for 5 minutes before the recording. Methyl-β-cyclodextrin remained in the imaging solution throughout the recording.

Protein extraction and Western blots

Cells were harvested in ice cold PBS and lysed in RIPA buffer (Sigma) with 1:100 HaltTM Protease and Phosphatase Inhibitor Cocktail on ice and then sonicated for 5 x 30 s at 4°C. Protein lysates were centrifuged for 15 min at > 15000 rpm at 4°C and the supernatant was collected and quantified by Precision Red Advanced Protein Assay. 10 μg of protein were boiled in NuPAGE LDS sample buffer at 95°C for 5 min and separated using a NuPAGE 4%–12% Bis-Tris Protein Gel in NuPAGE MES SDS Running Buffer. Proteins were electrophoretically transferred to nitrocellulose membranes with NuPAGE Transfer Buffer. Blots were blocked for 60 min at RT in TBS-T + 5% nonfat milk and incubated overnight in the same solution with the respective primary antibodies at 4°C. Antibodies used in this study are listed in STAR Methods. Blots were revealed using SuperSignalTM West Femto Chemiluminescent Substrate. Chemiluminescence was imaged and analyzed using Image lab software (Biorad) and FIJI.

Presto Blue Assay

Presto Blue viability reagent was diluted in neuronal culture medium at 1:10 and 85 μl of diluted Presto Blue solution was added to each well containing neurons and incubated at 37 °C with 5% CO2 for 2 hours. 75 μl was transferred to a new plate for absorbance readings at 570 and 600 nm. Results were calculated by subtracting 591 600nm absorbance values from 570 nm values, followed by subtracting background [600-570nm] readings (read from blank wells) from experimental samples.

Bulk RNA-sequencing and analysis

Flash frozen cell pellets were sent to the Integrated Genomics Operation (IGO) at MSKCC for RNA extraction and quality control, followed by library preparation and paired-end poly-A enriched sequencing (30-40 million reads).

Homopolymers, adapters and distribution of base quality of raw sequences from each sample were investigated using FastQC. Reads were aligned to the hg19 Human reference genome build using STAR 2.5.1b¹²². Read counts were tabulated using featureCounts (version 1.4.3), and multi-mapping reads were discarded according to suggested featureCounts practice. Genes without any expression counts in any sample were discarded. Differential gene expression across time-points was performed using DESeq2¹²³ R package that normalizes gene count data to transcription per million (TPM).

Gene Set Enrichment Analysis (GSEA) was performed using pairwise comparisons of control vs. condition to test for enrichment or depletion in Reactome, KEGG, Wikipathways and GO pathways using the following parameters: minimum gene-set size=15, maximum gene-set size=500, number of permutations = 1000. The top 200 enriched/depleted pathways were plotted.

Lipid extraction and cholesterol measurement

Cell pellets were collected in PBS and lipids were extracted using a lipid extraction kit according to the manufacturer’s instructions. Lipid were resuspended in 50 μl suspension buffer provided in the kit and sonicated for 10 minutes.

Cholesterol was measured using the Amplex Red Cholesterol Assay Kit according to the manufacturer’s instructions. Fluorescence was measured using a microplate reader with excitation at 540 nM and emission detection at 590 nM. Measurements were corrected for background fluorescence by subtracting the measurement values derived from two averaged no-cholesterol controls. Values were normalized to the experimental control.

Quantification and statistical analysis

Statistics were performed in PRISM (GraphPad) and R software. At least three independent biological replicates were used for each experiment, derived from at least three independent differentiations. Specific details on N and p-values are given in Supplemental Table 2. Data are represented as arithmetical means +/− standard error of the mean (SEM) unless otherwise indicated. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Supplementary Material

2

Table S1. RNA-seq data of differentially expressed genes (DEGs) in depression-associated conditions, related to Figure 4.

3

Table S2. Details on experimental N and p-values, related to STAR method ‘Quantification and statistical analysis’.

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
mouse anti-COUP-TF1	Abcam	Cat#ab41858; RRID: AB_742210
rabbit anti-FOXG1	Clontech	Cat#M227
chicken anti-GFP	Abcam	Cat#ab13970; RRID: AB_300798
rabbit anti-GR	Cell Signaling	Cat#3660S; RRID: AB_11179215
rabbit anti-IFNAR1	Abcam	Cat#ab244357
chicken anti-MAP2	Abcam	Cat#ab5392; RRID: AB_2138153
mouse anti-MAP2	Sigma	Cat#M1406; RRID: AB_477171
rabbit anti-MR	Abcam	Cat#ab2774; RRID: AB_303287
rabbit anti-PAX6	Biolegend	Cat#901301; RRID: AB_2565003
mouse anti-PSD95	Thermo Fisher Scientific	Cat#MA1-045; RRID: AB_325399
mouse anti-SIRT1	Abcam	Cat#ab110304; RRID: AB_10864359
goat anti-SP8	Santa Cruz Biotechnology	Cat#sc-104661; RRID: AB_2194626
rabbit anti-Synapsin I	Sigma	Cat#S193; RRID: AB_261457
rabbit anti-TBR1	Abcam	Cat#ab183032; RRID: AB_2936859
chicken anti-TUJ1	Abcam	Cat#ab52623; RRID: AB_869991
Donkey anti-mouse IgG H&L (Alexa Fluor®488)	Invitrogen	Cat#R37114; RRID: AB_2556542
Donkey anti-rabbit IgG H&L (Alexa Fluor®488)	Invitrogen	Cat#A21206; RRID: AB_2535792
Donkey anti-chicken IgG H&L (Alexa Fluor®488)	Invitrogen	Cat#A78948; RRID: AB_2921070
Donkey anti-mouse IgG H&L (Alexa Fluor®555)	Invitrogen	Cat#A31570; RRID: AB_2536180
Donkey anti-rabbit IgG H&L (Alexa Fluor®555)	Invitrogen	Cat#A31572; RRID: AB_162543
Donkey anti-goat IgG H&L (Alexa Fluor®555)	Invitrogen	Cat#A-21432; RRID: AB_141788
Donkey anti-chicken IgG H&L (Alexa Fluor®647)	Jackson ImmunoResearch	Cat#703-605-155; RRID: AB_2340379
mouse anti-actin	Millipore	Cat#MAB1501; RRID: AB_2223041
mouse anti-SIRT1	Millipore	Cat#04-1557; RRID: AB_1977495
Anti-mouse IgG HRP-linked secondary antibody	Cell Signaling Technology	Cat#7076; RRID: AB_330924
Chemicals, peptides, and recombinant proteins
E8 medium	Thermo Fisher	A1517001
Vitronectin	Thermo Fisher Scientific	A14700
DPBS	Thermo Fisher Scientific	14190-136
Accutase	Innovative Cell Technologies	AT104-500
Matrigel	Corning	354234
Y-27632	R&D Systems	1254
E6 medium	Thermo Fisher	A1516401
SB431542	Tocris	1614
LDN193189	Stem Cell Technologies	72149
XAV939	Tocris	3748
Neurobasal	Thermo Fisher	21103049
DMEM/F12	Thermo Fisher	11330-032
N2	Thermo Fisher	17502048
B27 w/o Vitamin A	Thermo Fisher	12587010
Penicillin Streptomycin	Thermo Fisher Scientific	15140-122
FGF8	R&D Systems	423-F8
Poly-L-ornithine	Sigma Aldrich	P3655
Laminin	R&D Systems	3400-010-1
Fibronectin	Thermo Fisher	356008
STEM-CELLBANKER	Amsbio	11890
Neurobasal-A medium	Thermo Fisher	A247750
GlutaMAX Supplement	Thermo Fisher	35050061
BDNF	R&D Systems	248-BD
GDNF	Peprotech	450-10
Cyclic Monophosphate Sodium Salt	Sigma-Aldrich	D0627
Ascorbic Acid	Sigma-Aldrich	4034-100
DAPT	Tocris	2634
Hydrocortisone	Tocris	4093
IFN-a	Sigma-Aldrich	IF007
Methyl-□-cyclodextrin	Sigma-Aldrich	M7439
Mevastatin	Sigma-Aldrich	M2537
Zaragozic acid	Sigma-Aldrich	Z2626
MbCD-cholesterol	Sigma-Aldrich	C4951
Ketamine	Tocris	6094
Fluoxetine	Sigma Aldrich	F132
4’,6-diamidino-2-phenylindole (DAPI)	Sigma	D9542
Filipin III	Sigma-Aldrich	SAE0087
Tetrodotoxin	Tocris	1078
4-Aminopyridine	Tocris	0940
Calbryte 590 AM	Thermo Fisher Scientific	NC1792904
HEPES-Buffered Hanks Balanced Salt Solution with 0.1% Pluronic F-127	Sigma	P2443
MK-801	Sigma-Aldrich	M107
NBQX	Tocris	1044
Halt Protease Inhibitor	Thermo Fisher	87786
Halt Phosphatase inhibitor	Thermo Fisher	78420
Precision Red Advanced Protein Assay	Cytoskeleton, Inc	ADV02-A
NuPAGE™ LDS Sample Buffer	Invitrogen	NP0007
NuPAGE MES SDS Running Buffer	Invitrogen	NP0002
Nitrocellulose Membranes	Thermo Scientific	88018
NuPAGE Transfer Buffer	Invitrogen	NP00061
Nonfat Dry Milk	Cell Signaling	9999S
Critical commercial assays
RNeasy Mini Kit	Qiagen	74104
iScript™ Reverse Transcription Supermix for RT-qPCR	Biorad	1708841
SsoFast™ EvaGreen® Supermix	Biorad	1725201
NuPAGE 4%–12% Bis-Tris Protein Gel	Invitrogen	NP0335Box
SuperSignal™ West Femto Maximum Sensitivity Substrate	Thermo Fisher	34094
Presto Blue viability reagent	Thermo Fisher Scientific	A13261
Lipid extraction kit	Abcam	ab211044
Amplex Red Cholesterol Assay Kit	Thermo Fisher Scientific	A12216
Deposited data
RNA sequencing dataset: “Disease modeling in human pluripotent stem cell-derived prefrontal cortex neurons”	This study	NCBI database: PRJNA949591
Experimental models: Cell lines
H9 (WA-09) hESC line	WiCell Research Institute	NIHhESC-10-0062
H9 SIRT1 KO hESC line	This study	N/A
MEL-1 hESC line	University of Queensland	RRID:CVCL_B863
MSK-SRF001	MSKCC Stem Cell Core	N/A
731 iPSC line (parental line: primary fibroblasts)	pFIB: Coriell Institute	GM00731
Oligonucleotides
GAPDH F: TCAAGAAGGTGGTGAAGCAG	This study	N/A
GAPDH R: CGCTGTTGAAGTCAGAGGAG	This study	N/A
FKBP5 F: GCGAAGGAGAAGACCACGACAT	This study	N/A
FKBP5 R: TAGGCTTCCCTGCCTCTCCAAA	This study	N/A
ZBTB16 F: GAGCTTCCTGATAACGAGGCTG	This study	N/A
ZBTB16 R: AGCCGCAAACTATCCAGGAACC	This study	N/A
TSC22D3 F: ATCTGCAACCGCAACATCGACC	This study	N/A
TSC22D3 R: GCATACATCAGATGATTCTTCACC	This study	N/A
TUJ1 F: GGCCTTTGGACATCTCTTCA	This study	N/A
TUJ1 R: ATACTCCTCACGCACCTTGC	This study	N/A
TBR1 F: GGGCTCACTGGATGCGCCAAG	This study	N/A
TBR1 R: TCCGTGCCGTCCTCGTTCACT	This study	N/A
IRF9 F: CCACCGAAGTTCCAGGTAACAC	This study	N/A
IRF9 R: AGTCTGCTCCAGCAAGTATCGG	This study	N/A
ISG15 F: CTCTGAGCATCCTGGTGAGGAA	This study	N/A
ISG15 R: AAGGTCAGCCAGAACAGGTCGT	This study	N/A
STAT1 F: ATGGCAGTCTGGCGGCTGAATT	This study	N/A
STAT1 R: CCAAACCAGGCTGGCACAATTG	This study	N/A
Software and algorithms
Imaris	Oxford Instruments	https://imaris.oxinst.com/
FIJI	Schindelin et al.120	https://fiji.sc/; RRID: SCR002285
CALIMA open-source software	Radstake et al.121	https://github.com/EALRaaijmakers/CALIMA-2.0
Image lab software	Biorad	http://www.bio-rad.com/en-us/sku/1709690-image-lab-software; RRID: SCR_014210
STAR 2.5.1b	Dobin et al.122	https://github.com/alexdobin/STAR?tab=readme-ov-file
DESeq2	Love et al.123	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
GSEA	Subramanian et al.77 and Mootha et al.78	https://www.gsea-msigdb.org/gsea/index.jsp
R	N/A	https://cran.r-project.org/; RRID: SCR_001905
Prism 9	GraphPad	www.graphpad.com; RRID: SCR_002798

Highlights.

• Pharmacological and genetic modeling of depression using hPSC-derived neurons

• Depression-like neurons show morphological defects and reduced activity

• Cholesterol imbalance is a shared mechanism across depression-associated conditions

• Cholesterol supplementation rescues depression-like neuronal phenotypes

Data and code availability

Sequencing data in this study have been deposited in the National Center for Biotechnology Information (NCBI) database under accession code PRJNA949591 and are publicly available as of the date of publication.

This paper does not report original code.

Any additional information required to reanalyze the data reported in this work paper is available from the Lead Contact upon request.