Close this search box.

An epigenetic barrier sets the timing of human neuronal maturation – Nature

PSC lines and cell culture

Experiments with hPSCs and iPSCss was approved in compliance with the Tri-Institutional ESCRO at Memorial Sloan Kettering Cancer Center, Rockefeller University and Weill Cornell Medicine. hPSC lines WA09 (H9; 46XX) and WA01 (H1; 46XY) were from WiCell Stemcell Bank. The GPI::Cas9 line was derived from WA09 hPSCs. MSK-SRF001 iPSCs were from Memorial Sloan Kettering Cancer Center. hPSCs and iPSCs were authenticated by STR. hPSCs and iPSCs were maintained with Essential 8 medium (Life Technologies A1517001) in feeder-free conditions onto vitronectin-coated dishes (VTN-N, Thermo Fisher A14700). hPSCs and iPSCs were passaged as clumps every 4–5 days with EDTA (0.5 M EDTA/PBS) and routinely tested for mycoplasma contamination. The GPI::Cas9 knock-in hPSC line was generated using CRISPR–Cas9-mediated homologous recombination by transfecting H9 hPSCs with the Cas9-T2A-Puro targeting cassette downstream of the GPI gene (Supplementary Fig. 6b). Selected clones were validated by genomic PCR and Cas9 mRNA and protein expression by RT–qPCR and western blot, respectively and screened for Karyotype banding. CHD5-KO and JADE2-KO WA09 hPSC lines were generated by the SKI Stem Cell Research Core at Memorial Sloan Kettering Cancer Center (MSKCC) via CRISPR–Cas9 using the following gRNA targets: CHD5, CGTGGACTACCTGTTCTCGG; JADE2, CAGTTTGGAGCATCTTGATG. Mouse epiblast stem cells (EpiSCs) B6.129_4 were a gift from the Vierbuchen laboratory at Memorial Sloan Kettering Cancer Center and were maintained on mouse embryonic fibroblasts as previously described64. Rat primary astrocytes were purchased from Lonza (R-CXAS-520) and cultured according to manufacturer instructions.

Synchronized generation of hPSC-derived cortical neurons

hPSCs (passage 40–50) were differentiated toward cortical excitatory neurons using an optimized protocol based on dual SMAD inhibition and WNT inhibition as follows. hPSCss were dissociated at single cells using Accutase and plated at 300,000 cells per cm2 onto Matrigel-coated wells (354234, Corning) in Essential 8 medium supplemented with 10 μM Y-27632. On day 0–2, cells were fed daily by complete medium exchange with Essential 6 medium (E6, A1516401, Thermo Fisher Scientific) in the presence of 100 nM LDN193189 (72142, Stem Cell Technologies), 10 μM SB431542 (1614, Tocris) and 2 μM XAV939 (3748, Tocris) to induce anterior neuroectodermal patterning. On day 3–9 cells were fed daily with Essential 6 medium (E6, A1516401, Thermo Fisher Scientific) in the presence of 100 nM LDN193189 (72142, Stem Cell Technologies), 10 μM SB431542. On day 10–20 cells were fed daily with N2/B27 medium (1:1 NB:DMEM/F12 basal medium supplemented with 1× N2 and B27 minus vitamin A) to generate a neurogenic population of cortical NPCs. N2 and B27 supplements were from Thermo. At day 20, NPCs were either cryopreserved in STEM-CELLBANKER solution (Amsbio) or induced for synchronized neurogenesis as following: NPCs were dissociated at single cells following 45 min incubation with Accutase and seeded at 150,000 cells per cm2 onto poly-l-ornithine and laminin/ fibronectin-coated plates in NB/B27 medium (1× B27 minus vitamin A, 1% l-glutamine and 1% penicillin-streptomycin in Neurobasal medium) in the presence of 10 μM Notch pathway inhibitor DAPT for 10 days (until day 30). For long-term culture, neurons were maintained in NB/B27 supplemented with BDNF (450-10, PreproTech), GDNF (248-BD-025, R&D biosystems), cAMP (D0627, Sigma) and ascorbic acid (4034-100, Sigma). From day 20 onwards, cells were fed every 4–5 days by replacing 50% of the medium volume. For neurons-astrocytes co-cultures, rat primary astrocytes were plated onto poly-l-ornithine and laminin/fibronectin-coated plates in NB/B27 medium supplemented with BDNF, GDNF, cAMP and ascorbic acid and allowed to adhere for few days. hPSC-derived neurons at day 25 were dissociated using Accutase and seeded on top of rat astrocytes. Neurons-astrocytes co-cultures were maintained on NB/B27 medium supplemented with BDNF, GDNF, cAMP and ascorbic acid.

Mouse epiblast stem cell differentiation

Mouse epiblast stem cells (mEpiSCs) B6.129_4 were differentiated as following: on day 0, mEpiSC colonies were lifted from feeders using 0.5 U µl−1 collagenase IV in HBSS + +, dissociated to single-cell solution using Accutase, then plated at 220,000 cells per cm2 on Matrigel-coated wells in mN2/B27 media64 supplemented with 10 µM Y-27632, 100 nM LDN193189, 10 µM SB431542 and 2 µM XAV939. Cells were fed daily with mN2/B27 supplemented with 2 µM XAV939 (day 1), 100 nM LDN193189 (day 1–5), 10 µM SB431542 (day 1–5). On day 6 NPCs were dissociated to single-cell suspension using Accutase and replated at 200,000 cells per cm2 onto poly-l-ornithine and laminin/fibronectin-coated plates in NB/B27 medium (10% Neurobasal, 90% Neurobasal A, 1× B27 minus vitamin A, 1% Glutamax, 0.5% penicillin-streptomycin, 0.1% BDNF, 0.1% cAMP, 0.1% ascorbic acid, 0.1% GDNF) supplemented with 10 µM Y-27632 (day 6) and 10 µM DAPT (day 6 and 8). Cells were fed every other day by replacing 50% of the medium volume.

Cerebral organoids differentiation

On day −1, WA09 (H9) hPSCs were dissociated with EDTA for 10 min at 37 °C and allowed to aggregate into spheroids of 10,000 cells each in V-bottom 96 well microplates (S-Bio) in E8 medium with ROCK inhibitor (Y-27632, 10 μM) and WNT inhibitor (XAV939, 5 μM, Tocris 3748). The next day (day 0), the medium was changed to E6 supplemented 100 nM LDN193189, 10 μM SB431542 and 5 μM XAV939. On day 5, medium was switched to E6 supplemented with 100 nM LDN193189, 10 μM SB431542. On day 8, medium was changed to N2/B27-based organoid medium as previously described65. From day 0 to day 14 medium was replaced every other day. On day 14, organoids were transferred to an orbital shaker on 10 cm dishes and half of the medium was changed on a Monday–Wednesday–Friday schedule. Treatment with 4 μM GSK343 or DMSO was performed transiently from day 17–25 or day 17–37 depending on the experiment as indicated in the figures.

EdU labelling and small molecule treatments

For birth-dating experiments of WA09 (H9) hPSC-derived cortical neurons, 3μM EdU (5-ethynyl-2′-deoxyuridine, A10044 Invitrogen) was added to the culture for 48 h in the following time windows: d18–19, d20–21, d22–23, d24–25, d26–27, d28–29. After treatment, EdU was washed out and neurons were fixed at day40 of differentiation and processed for immunostaining. Treatment of hPSC-derived cortical NPCs with small molecules inhibitors of chromatin regulator was performed from day 12 to day 20 of differentiation (Fig. 4b). Small molecules were washed out and withdrawn starting at day 20 before the induction of synchronized neurogenesis and neurons derived from all the treatments were maintained in the same conditions. Small molecules were dissolved in DMSO and added to the N2/B27 medium at 2 or 4 μM depending on the experiment. DMSO in control conditions was added at the corresponding dilution factor as for epigenetic inhibitors.

Treatment of mEpiSC-derived NPCs was performed as follows: For Ezh2i experiments, 0.04% DMSO or 4μM GSK343 was added to NPC medium on day 4 and 5. For Ezh2i+ experiments this treatment was extended with 0.02% DMSO or 2μM GSK343 being added to medium on day 6, 8 and 10. GSK-J4 was used at 1 μM and added to the medium on day 4 and 5.

The following small molecules targeting epigenetic factors were used in the study and purchased from MedChemExpress: GSK343 (HY-13500), UNC0638 (HY-15273), EPZ004777 (HY-15227), GSK2879552 (HY-18632), CPI-455 (HY-100421), A-196 (HY-100201), GSK-J4 (HY-15648F). A List of small molecules and relative molecular target is reported in Extended Data Fig. 3b.

Morphological reconstructions and quantification of synaptic puncta

For the morphological reconstruction of WA09 (H9) hPSC-derived neurons, NPCs were infected at day 20 with low-titre lentiviruses expressing dTomato reporter. Following induction of neurogenesis, the resulting neurons were fixed at day 25, 50, 75 and 100. The dTomato reporter signal was amplified by immunofluorescence staining and individual neurons were imaged at Zeiss AXIO Observer 7 epi-fluorescence microscope at 10× magnification. Neuronal morphology was reconstructed in Imaris v9.9.1 software using the filamentTracer function in autopath mode and using the nucleus (using DAPI channel) as starting point. Traces were eventually manually corrected for accuracy of cell processes detection. Neurite length and Sholl Analysis (every 10 μm radius) measurements were performed in the Imaris platform and extracted for quantifications and statistics. For staining with synaptic markers, cells were cultured on μ-plate 96 Well Black (Ibidi) and stained for SYN1 and PSD95 antibodies to visualize pre and post -synaptic puncta respectively and MAP2 to visualize neuronal dendrites. Confocal images were acquired using a 63× immersion objective at a Leica SP8WLL confocal laser-scanning microscope. Three fields of view for each sample from two independent differentiations (total of 6 fields of view per condition) were analysed as following. Single-plane confocal images were open in Fiji v2.9.0 and puncta were detected using the SynQuant plugin ( The z-score for particle detection was adjusted for accuracy of puncta detection. The other parameters were set as default value. Dendrite length was extracted from the reference MAP2 channel.

Immunocytochemistry and histology

Cultured cells were fixed with 4% PFA in PBS for 20 min at RT, washed three times with PBS, permeabilized for 30 min in 0.5% Triton X-100 in PBS and then blocked in a solution containing 5% Normal goat serum or Normal donkey serum, 2% BSA and 0.25% Triton X-100 for 1 h at room temperature. Primary antibodies were incubated overnight at 4 °C in the same blocking solution. EdU+ cells were detected using the Click-iT EdU Imaging kit (Molecular Probes) with Alexa Fluor 488 according to manufacturer’s instructions. Secondary antibodies conjugated to either Alexa 488, Alexa 555 or Alexa 647 (Thermo) were incubated for 45 min at 1:400 dilution in blocking solution. Cell nuclei were stained with 5 μM 4′-6-diamidino-2-phenylindole (DAPI) in PBS.

Organoids were fixed in 4% PFA overnight at 4 °C, washed 3 times with PBS and cryoprotected in 30% sucrose/PBS. Organoid tissue was sectioned at 30 μm on a cryostat (Leica 3050 S), mounted on microscope slides, allowed to dry at room temperature and stored at −80 °C. On the day of the staining, slides we defrosted for 20 min at room temperature. Sections were first permeabilized in 0.5% Triton X-100 in PBS, blocked for 1 h in 5% normal goat serum, 1% BSA, 0.25% triton in PBS and incubated in the same solution with primary antibodies overnight. The next day, sections were washed with PBS and incubated in secondary antibodies for 2.5 h at room temperature at 1:400 dilution. DAPI 5 μM stain was used to identify cell nuclei. Images were captured using a Leica SP8WLL confocal laser-scanning microscope.

The following primary antibodies and dilutions were used: rabbit anti-PAX6 1:300 (901301, Biolegend); rabbit anti-FOXG1 1:500 (M227, Clonetech); mouse anti-Nestin 1:400 (M015012, Neuromics); mouse anti-MAP2 1:200 (M1406, Sigma); chicken anti-MAP2 1:2000 (ab5392, Abcam); rabbit anti-class III β-tubulin (TUJI) 1:1,000 (MRB-435P, Covance); mouse anti-Ki67 1:200 (M7240, Dako); rabbit anti-Ki67 1:500 (RM-9106, Thermo Scientific); rabbit anti-TBR1 1:300 (ab183032, Abcam); rabbit anti-TBR1 1:500 (ab31940, Abcam); rat anti-CTIP2 1:500 (ab18465, Abcam); mouse anti-SATB2 1:1,000 (ab51502, Abcam); rabbit anti-synapsin I 1:1,000 (S193, Sigma); mouse anti-PSD95 1:500 (MA1-046, Thermo); mouse anti-neurofilament H 1:500 (non-phosphorylated) (SMI32, Enzo Life science); mouse anti c-FOS 1:500 (ab208942, Abcam); mouse anti-HLA Class I ABC 1:150 (ab70328, abcam); goat anti-RFP 1:1,000 (200-101-379, Rockland); rabbit anti-DsRed 1:750 (632496, Clontech); rabbit anti-H3K27me3 1:200 (9733, Cell Signaling Technologies); rabbit anti-GFAP 1:500 (Z033429-2, Dako); chicken anti-GFP 1:1,000 (ab13970, Abcam); rat anti-SOX2 1:200 (14-9811-82, Thermo); rabbit anti-AQP4 1:500 (HPA014784, SIGMA); sheep anti-EOMES 1:200 (AF6166, R&D). The primary antibodies including anti-GFAP antibody were validated for recognition of human antigens to confirm lack of human astrocytes in our synchronized cortical cultures.


smRNA-FISH was performed on WA09 (H9) hPSC-derived and mEpiSC-derived neurons using ViewRNA Cell Plus Assay Kit (Invitrogen) in RNAse-free conditions according to manufacturer’s instructions to simultaneously detect RNA targets by in situ hybridization and the neuronal marker MAP2 (Alexa Fluor 647) by immunolabelling. Neurons were plated on μ-plate 24 Well Black (Ibidi) plates, fixed and permeabilized for 15 min at room temperature with fixation/permeabilization solution and blocked for 20 min followed by incubation with primary and secondary antibody for 1 h at room temperature. Target probe hybridization with mouse or human -specific viewRNA Cell Plus probe sets was carried at 40 °C under gentle agitation for 2 h. Type 1 (Alexa Fluor 546) and type 4 (Alexa Fluor 488) probe sets were used to detect EZH2 and TBP RNA respectively, using the same fluorophore scheme for neurons derived from mEpiSCs and hPSCs. Pre amplification, amplification and fluorescence labelling steps were carried at 40 °C under gentle agitation for 1 h each. Washes were performed as indicated in the kit’s procedure. Samples were incubated with 5 μM DAPI to visualize cell nuclei and a coverslip was gently placed inside each well using ProLong Glass Antifade Mountant. z-stack images at 0.4 μm step and covering the entire cell volume were acquired using a Leica SP8WLL confocal laser-scanning microscope with a 63× immersion objective at 3× digital zoom. z-stacks were loaded and projected in Imaris v9.9.1 software for RNA puncta visualization and quantification within each single MAP2 positive neuron. Eight different fields of view (2–5 neurons per field) for each condition (mouse versus human) from two independent batches of differentiations (16 fields of view per condition) were obtained for downstream analysis. The nuclear volume for each neuron was reconstructed and calculated using the Surface function in Imaris Software.

Electrophysiological recording

For electrophysiological recordings, neurons were plated in 35 mm dishes. Whole-cell patch clamp recordings during the maturation time course were performed at day 25, 50, 75 and 100 of differentiation as previously described22. In brief, neurons were visualized using a Zeiss microscope (Axioscope) fitted with 4× objective and 40× water-immersion objectives. Recordings were performed at 23–24 °C and neurons were perfused with freshly prepared artificial cerebral-spinal fluid (aCSF) extracellular solution saturated with 95% O2, 5% CO2 that contained (in mM): 126 NaCl, 26 NaHCO3, 3.6 KCl, 1.2 NaH2PO4, 1.5 MgCl2, 2.5 CaCl2, and 10 glucose. Pipette solution for recordings in current clamp configuration contained (in mM): 136 KCl, 5 NaCl, 5 HEPES, 0.5 EGTA, 3 Mg-ATP, 0.2 Na-GTP, and 10 Na2-phosphocreatine, pH adjusted to 7.3 with KOH, with an osmolarity of ~290 mOsm. For mEPSCs, the pipette solution contained (in mM): 140 CsCl, 10 NaCl, 10 HEPES, 0.5 EGTA, 3 Mg-ATP, 0.2 Na-GTP, and 10 Na2-phosphocreatine, pH adjusted to 7.3 with CsOH. 20 μM (−)-bicuculline methochloride (Tocris), 1 μM strychnine HCl (Sigma), and 0.5 μM tetrodotoxin (TTX) (Alomone Labs) were added to aCSF for mEPSC recordings to block GABAA receptors, glycine receptors, and voltage-gated Na+ channels, respectively. Input resistance was measured from a voltage response elicited by intracellular injection of a current pulse ( − 100 pA, 200 ms). Membrane voltage was low-pass filtered at 5 kHz and digitized at 10 kHz using a Multiclamp 700B amplifier connected to a DigiData 1322 A interface (Axon Instruments) using Clampex 10.2 software (Molecular Devices). Liquid junction potentials were calculated and corrected off-line. Action potentials were generated in current clamp from currents injected in 10 pA intervals from 0 to 250 pA. Recordings were analysed for: resting membrane potential, input resistance, rheobase, threshold, as well as action potential amplitude, overshoot, duration, half-width, rise and decay. Neurons were held at −80 mV and continuous recordings of mEPSCs were made using Axoscope software (Molecular Devices). Data processing and analysis were performed using MiniAnalysis (Synaptosoft) version 6 and Clampfit 10.2 (Molecular Devices). Events were detected by setting the threshold value, followed by visual confirmation of mEPSC detection. Whole-cell patch clamp recordings in neurons derived from DMSO and EZH2i conditions (pipettes 3–6 MΩ) were performed in aCSF containing (in mM): 125 NaCl, 2.5 KCl, 1.2 NaH2PO4, 1 MgSO4, 2 CaCl2, 25 NaHCO3 and 10 d-glucose. pH and osmolarity were adjusted to 7.4 and 300–310 mOsm, respectively. For firing recordings, pipettes were filled with a solution containing (in mM): 130 potassium gluconate, 4 KCl, 0.3 EGTA, 10 Na2-phosphocreatine, 10 HEPES, 4 Mg2-ATP, 0.3 Na2-GTP and 13 biocytin. pH and osmolarity were adjusted to 7.3 (with KOH) and 285–290 mOsm respectively. For mEPSCs recordings the ACSF was supplemented with 1 µM TTX and 100 µM 4-AP and pipettes were filled with a caesium-based solution that contained (in mM): 120 CsMeSO4, 8 NaCl, 10 HEPES, 0.3 EGTA, 10 TEA-Cl, 2 Mg2-ATP, 0.3 Na2-GTP, 13.4 biocytin and 3 QX-314-Cl. pH: 7.3 (adjusted with CsOH) and 290–295 mOsm. Recordings were acquired with a computer-controlled Multiclamp 700B amplifier and a Digidata 1550B (Molecular Devices, California) at a sampling rate of 10 kHz and low-pass filtered at 1 kHz. pClamp 10 software suite (Molecular Devices) was used for data acquisition (Clampex 10.6) and data analysis (Clampfit 10.6). The quantification of the amplitude and inter-event interval of mEPSCs shown in the cumulative probability plots in Fig. 4j was performed taking all the events together. To isolate the NMDA component from mEPSCs recorded at +40 mV, we measured current amplitude 20 ms after the mEPSC onset, where AMPA receptors are desensitized (depicted by the dotted line in Extended Data Fig. 5f)66,67,68. For the calculation of the NMDA/AMPA ratio, the amplitude of the NMDA component was then divided by the amplitude of the peak of the AMPA currents recorded at – 70 mV. Statistical analysis and plots were done in Prism 9 (GraphPad, California). Evoked action potential and traces shown in DMSO versus EZH2i groups in Fig. 4g were elicited with 20 pA injected current.

Calcium imaging and analysis

hPSC-derived cortical neurons were infected with lentiviruses encoding GCaMP6m and cultured on μ-plate 96 Well Black (Ibidi). In rat astrocytes co-culture experiments, hPSC-derived neurons were infected with GCaMP6m lentiviruses four days before dissociation and prior to seeding onto rat primary astrocytes. For each batch of experiments, the infection and measurement of Ca2+ spikes in neurons under control or genetic/pharmacological perturbation has been done in parallel on the same day to account for the variability in the absolute expression of GCaMP6m due to lentiviral delivery. Ca2+ imaging was performed as previously described69. In brief, on the day of the imaging, cells were gently washed twice in modified Tyrode solution (25 mM HEPES (Invitrogen), 140 mM NaCl, 5 mM KCl, 1 mM MgCl2, 10 mM glucose, 2 mM CaCl2, 10 μM glycine, 0.1% BSA pH 7.4, pre-warmed to 37 °C) and equilibrated in imaging buffer for 1-2 min (25 mM HEPES, 140 mM NaCl, 8 mM KCl, 1 mM MgCl2, 10 mM glucose, 4 mM CaCl2, 10 μM glycine, 0.1% BSA pH 7.4, pre-warmed to 37 °C). GCaMP6m fluorescence was recorded on Celldiscover7 (ZEISS) inverted epi-fluorescence microscope with the 488 nm filter under environmental control (37 °C; 95% O2, 5% CO2) using ZEN Blue 3.1 software at the Bio-Imaging Resource Center (BIRC) at Rockefeller University. Neuronal cultures were imaged for ~3 min at a frame rate of 4–6 frames per second (600–800 frames per time lapse) using a 10× or 20× objective.

hPSC-derived cortical brain organoids were infected with lentiviruses encoding GCaMP6m at day 45 of differentiation and cultured in BrainPhys Imaging Optimized Medium (Stem Cell Technologies) for a week before the imaging. On the day of the imaging, DMSO control and organoids transiently treated with GSK343 were equilibrated in imaging buffer for 30 min and transferred into imaging cuvettes. GCaMP6m fluorescence on intact organoids was recorded by light-sheet microscopy on TruLive3D Imager (Bruker) under environmental control (37 °C; 95% O2 – 5% CO2). Multiple fields of view from 3–4 organoids per condition from 2 independent batches each were imaged for ~2–4 min at a frame rate of 5–10 frames per second at 31.3× effective magnification.

Analysis was performed as previously described69. In brief, the live-imaging image stack was converted to TIFF format and loaded into optimized scripts in MATLAB (Mathworks) R2020b and R2021b. Region of Interest (ROI) were placed on the neuron somas to calculate the raw GCaMP6m intensity of each neuron over time. The signal intensity of each raw trace was normalized to baseline fluorescence levels (ΔF/F0) for spike detection. Single-neuron amplitude was calculated from the normalized GCaMp6m intensity for all the detected spikes in each trace (mean ΔF/F0 of detected spikes for each neuron). Single-neuron frequency was calculated as the number of detected spikes in each trace per minute of recording. Network activity was assessed by calculating the synchronous firing rate, defined as the number of detected synchronous Ca2+ spikes from all ROI in one FOV per minute of recording. In Figs. 1k and 4k, coloured lines depict the normalized (ΔF/F0) GCaMP6m signal traces of individual neurons in 1 field of view during 1 min of imaging; the black line is the averaged normalized GCaMP6m signal among neurons in 1 field of view. Images in Figs. 1j Fig. 4m were displayed as royal lookup table in FIJI. Supplementary Videos 16 show 20 frames per second, Supplementary Videos 7 and 8 show 100 frames per second.

Image analysis and quantification

Microscopy images were visualized with Adobe Photoshop 2022, Fiji 2.9.0 or Imaris software version 9.9.1. Morphological reconstruction of neurons was performed using Imaris software version 9.9.1. Ca2+ imaging analysis was performed using MATLAB software. Quantification of immunofluorescence images was performed in FIJI (ImageJ) version 2.9.0 or using the Operetta high content imaging system coupled with Harmony software version 4.1 (PerkinElmer).

Protein extraction and western blot

Cells were collected and lysed in RIPA buffer (Sigma) with 1:100 Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientific) and then sonicated for 3× 30sec at 4 °C. Protein lysates were centrifugated for 15 min at more than 15,000 rpm at 4 °C and supernatant was collected and quantified by Precision Red Advanced Protein Assay (Cytoskeleton). 5–10 μg of protein were boiled in NuPAGE LDS sample buffer (Invitrogen) at 95 °C for 5 min and separated using NuPAGE 4–12% Bis-Tris Protein Gel (Invitrogen) in NuPAGE MES SDS Running Buffer (Invitrogen). Proteins were electrophoretically transferred to nitrocellulose membranes (Thermo Fisher Scientific) with NuPAGE Transfer Buffer (Invitrogen). Blots were blocked for 60 min at room temperature in TBS-T + 5% nonfat milk (Cell Signaling) and incubated overnight in the same solution with the respective primary antibodies at 4 °C. The following primary antibodies were used: mouse anti-neurofilament H 1:500 (non-phosphorylated) (SMI32; Enzo Life science); mouse anti-syntaxin 1 A 1:500 (110 111; SYSY); mouse anti-actin 1:500 (MAB1501; Millipore); mouse anti-Cas9 1:500 (14697; Cell Signaling Technology); rabbit anti-CHD3 1:1,000 (ab109195, Abcam); rabbit anti-KDM5B 1:1,000 (ab181089, abcam). The following secondary antibodies were incubated for 1 h at room temperature at 1:1,000 dilution: anti-mouse IgG HRP-linked (7076; Cell Signaling Technology) and anti-rabbit IgG HRP-linked (7074; Cell Signaling Technology). Blots were revealed using SuperSignalTM West Femto Chemiluminescent Substrate (Thermo Fischer Scientific) at ChemiDoc XRS+ system (Bio-Rad). Chemiluminescence was imaged and analysed using Image lab software version 6.1.0 (Bio-Rad). Controls samples were run within each gel and the signal intensity of protein bands of interest was normalized to the intensity of the actin band (loading control) for each sample on the same blot. Uncropped and unprocessed images are shown in Supplementary Figure 1. One sample t-test on Fig. 3d was performed by comparing the mean of logFC for each genetic perturbation with the hypothetical mean logFC = 0 (null hypothesis of no changes). Two-tailed ratio-paired t-test in Fig. 4c was calculated on normalized marker/actin expression in manipulations versus DMSO.

RNA isolation and RT–qPCR

Samples were collected in Trizol. Total RNA from hPSC-derived samples was isolated by chloroform phase separation using Phase Lock Gel-Heavy tubes, precipitated with ethanol, and purified using RNeasy Mini Kit (Qiagen) with on-column DNA digestion step. RNA from mouse cells was isolated using Direct-zol microprep kit (Zymo research, R2060). cDNA was generated using the iScript Reverse Transcription Supermix (Bio-Rad) for RT–qPCR and quantitative PCR (qPCR) reactions were performed using SsoFast EvaGreen Supermix (Bio-Rad) according to the manufacturer’s instructions in 96 or 384-well qPCR plates using CFX96 and CFX384 Real-Time PCR Detection systems (Bio-Rad) using 5–10 ng cDNA / reaction. Primers were from Quantitect Primer assays (QUIAGEN) except for the ones in Supplementary Table 4. Results were normalized to the housekeeping genes GAPDH or TBP.

DNA constructs and lentivirus production

A Cas9-T2A-PuroR cassette flanked by 5′ and 3′ homology arms for the GPI locus was generated by NEBuilder HiFi DNA Assembly Cloning Kit of PCR-amplified fragments according to manufacturer’s instruction. EF1alpha-GCaMP6m lentiviral vector was generated by PCR amplification of GCaMP6m from pGP-CMV-GCaMP6m (Addgene 40754) using with Q5 High Fidelity master mix (NEB) and subcloned into pWPXLd (Addgene 12258) into BamH1 and EcoRI restriction site using standard cloning methods. For the simultaneous expression of gene-specific gRNA under transcriptional control of U6 promoter and dTomato fluorescent reporter driven by EF1alpha promoter, the SGL40.EFs.dTomato vector (Addgene 89398) was modified by inserting a P2A-Basticidin cassette downstream of dTomato sequence to generate the SGL40.EFs.dTomato-Blast backbone. gRNA sequences specific to each gene were designed using SYNTEGO CRISPR design tool ( and validated using CRISPOR tool70 ( DNA oligos (IDT) were annealed and subcloned into BsmBI restriction sites of SGL40.EFs.dTomato-Blast lentiviral backbone by standard cloning methods. Lentiviruses were produced by transfection of HEK293T cells (ATCC) using the Xtreme Gene 9 DNA transfection reagent (Sigma) with the respective lentiviral vectors along with the packaging vectors psPAX2 (Addgene, 12260) and pMD2.G (Addgene, 12259). Arrayed CRISPR gRNA lentiviral libraries were produced simultaneously. Viruses were collected 48 h post transfection, filtered with 0.22-μm filters and stored in aliquots at −80 °C. The sequence of each gRNA used is reported in Supplementary Table 5.

RNA-seq sample processing and analysis

Total RNA was extracted as described above. Sample for RNA-seq during chronological maturation at hPSC, NPC, d25, d50, d75 and d100 timepoints were submitted for TruSeq stranded ribo-depleted paired-end total RNA-seq at 40–50 million reads at the Epigenomic Core at Well Cornell Medical College (WCMC). Samples for RNA-seq studies on neurons upon perturbation with epigenetic inhibitors were submitted for paired-end poly-A enriched RNA-seq at 20–30 million reads to the MSKCC Integrated Genomic Core. Quality control of sequenced reads was performed by FastQC. Adapter-trimmed reads were mapped to the hg19 human genome using versions 2.5.0 or 2.7.10b of STAR71. The htseq-count function of the HTSeq Python package version 0.7.172 was used to count uniquely aligned reads at all exons of a gene. For the chronological maturation studies, the count values were transformed to RPKM to make them comparable across replicates. A threshold of 1 RPKM was used to consider a gene to be present in a sample and genes that were present in at least one sample were used for subsequent analyses. Differential gene expression across timepoints or treatments with epigenetic inhibitors was computed using versions 1.16 or 1.22.2 of DESeq2 respectively73. Variance stabilizing transformation of RNA-seq counts was used for the PCA plots and for heat maps of gene expression. For downstream analysis of trends of gene expression, transcripts were first grouped into ‘monotonically upregulated’ and ‘monotonically downregulated’ based on the characteristics of their expression from d25 to d100 and further categorized in strict: all the transitions satisfy the statistical significance criteria and relaxed: d25 versus d100 transition satisfy the significance criteria and intermediate transitions may not. For all comparisons a significance threshold of false discovery rate (FDR) ≤ 5% was used. Monotonically upregulated (strict): (d50 versus d25: FDR ≤ 5%) and (d100 versus d25: FDR ≤ 5%) and (d100 versus d50: FDR ≤ 5%) and (d50 versus d25:logFC > 0) and (d75 versus d50: logFC > 0) and (d100 versus d25 logFC > d50 versus d25 logFC). Monotonically downregulated (strict): (d50 versus d25:FDR ≤ 5%) and (d100 versus d25: FDR ≤ 5%) and (d100 versus d50: FDR ≤ 5%) and (d50 versus d25:logFC <0) and (d75 versus d50: logFC <0) and (d100 versus d25 logFC <d50 versus d25 logFC). Monotonically upregulated (relaxed): (d100 versus d25: FDR ≤ 5%) and (d50 versus d25:logFC > 0) and ((d100 versus d25:logFC >= d50 versus d25: logFC) OR (d75 versus d50: logFC > 0)). Monotonically downregulated (relaxed): (d100 versus d25: FDR ≤ 5%) and (d50 versus d25:logFC <0) and ((d100 versus d25:logFC <= d50 versus d25: logFC) OR (d75 versus d50: logFC <0)). GSEA74 was performed on d50 versus d25 and d100 versus d50 pairwise comparisons to test enrichment in KEGG pathways or gene sets from MSigDB using the following parameters: FDR ≤ 5%, minimum gene-set size=15, maximum gene-set size=500, number of permutations = 1000. GO term analysis was performed using v6.8 DAVID75 ( Venn diagrams were generated using Biovenn76.

The score for maturation in neurons upon epigenetic inhibition and control conditions (Extended Data Fig. 7b,c). was computed based on the geometric distribution of samples in the three-dimensional coordinate system defined by PCA1, 2 and 3. For each condition (treatment and day of differentiation), the coordinates defining the position of the samples in the 3D PCA space were determined based on the average across replicates. The DMSO d25 coordinates were set as the origin. The vectors defining maturation trajectories for each treatment and timepoint were then measured as the connecting segments between sample coordinates. The vector linking DMSO d25 and DMSO d50 conditions was used to define the chronological maturation trajectory and set as a reference (control vector) to calculate a similarity score for each treatment at any given timepoint. To account for vector magnitude and directionality, the dot product metric treatment vector · control vector was used to calculate the scores. Gene expression correlation heat maps in Extended Data Fig 7d were created from either all genes or maturation genes only by computing Pearson correlation and then running agglomerative hierarchical clustering using complete linkage. k-Means clustering in Extended Data Fig 7e was performed on z-score converted normalized counts and run using the kmeans function in R with nstart = 25 and k = 2:10, stopping when clusters became redundant (k = 4).

ATAC-seq sample processing and analysis

ATAC-seq libraries were prepared at the Epigenetic Innovation Lab at MSKCC starting from ~50,000 live adherent cells plated on 96-wells. Size-selected libraries were submitted to the MSKCC Genomic core for paired-end sequencing at 40–60 million reads. Quality control of sequenced reads was performed by FastQC (version 0.11.3) and adapter filtration was performed by Trimmomatic version 0.36. The filtered reads were aligned to the hg19 reference genome. Macs2 (version 2.1.0)77 was used for removing duplicate reads and calling peaks. Differentially accessible peaks in the atlas were called by DESeq2 version 1.1673. To define dynamic trends of chromatin accessibility during neuronal maturation as shown in Fig. 3g, agglomerative hierarchical clustering using Ward’s linkage method was done on the union of differentially accessible peaks in pairwise comparisons between d25, d50, d75 and d100 samples. HOMER (version 4.6)78 was used to investigate the motif enrichment in pairwise comparisons and unbiasedly clustered groups of peaks. Motif enrichment was also assessed by Kolmogorov–Smirnov and hypergeometric tests as previously described79. ATAC-seq peaks in the atlas were associated with transcription factor motifs in the updated CIS-BP database80,81 using FIMO82 of MEME suite version 4.1183. Hypergeometric test was used to compare the proportion of peaks containing a transcription factor motif in each group (foreground ratio) with that in the entire atlas (background ratio). Odds ratio represents the normalized enrichment of peaks associated with transcription factor motifs in the group compared to the background (foreground ratio/background ratio). Odds ratio ≥ 1.2 and transcription factor expression from parallel RNA-seq studies (reaching ≥ 1 RPKM) in neurons at any timepoint (d25, d50, d75, d100) was used to filter enriched transcription factor motif.

CUT&RUN sample processing and analysis

CUT&RUN was performed from 50,000 cells per condition as previously described84 using the following antibodies at 1:100 dilution: rabbit anti-H3K4me3 (aab8580, abcam); rabbit anti-H3K9me3 (ab8898, abcam); rabbit anti-H3K27me3 (9733, Cell Signaling Technologies); rabbit anti-H3K27ac (309034, Active Motif), normal rabbit IgG (2729, Cell Signaling Technologies). In brief, cells were collected and bound to concanavalin A-coated magnetic beads after an 8 min incubation at room temperature on a rotator. Cell membranes were permeabilized with digitonin and the different antibodies were incubated overnight at 4 °C on a rotator. Beads were washed and incubated with pA-MN. Ca2+-induced digestion occurred on ice for 30 min and stopped by chelation. DNA was finally isolated using an extraction method with phenol and chloroform as previously described84. Library preparation and sequencing was performed at the MSKCC Integrated Genomic Core.

Sequencing reads were trimmed and filtered for quality and adapter content using version 0.4.5 of TrimGalore ( and running version 1.15 of cutadapt and version 0.11.5 of FastQC. Reads were aligned to human assembly hg19 with version of bowtie2 ( and MarkDuplicates of Picard Tools version 2.16.0 was used for deduplication. Enriched regions were discovered using MACS2 with a p-value setting of 0.001 and a matched IgG or ‘no antibody” as the control. The BEDTools suite version 2.29.2 ( was used to create normalized read density profiles. A global peak atlas was created by first removing blacklisted regions ( then merging all peaks within 500 bp and counting reads with version 1.6.1 of featureCounts ( Reads were normalized by sequencing depth (to 10 million mapped fragments) and DESeq2 (v1.22.2) was used to calculate differential enrichment for all pairwise contrasts. Clustering was performed on the superset of differential peaks using k-means clustering by increasing k until redundant clusters arose. Gene annotations were created by assigning all intragenic peaks to that gene, and otherwise using linear genomic distance to transcription start site. The annotations in each cluster were used to intersect with the RNA-seq time series by plotting the average expression z-score of all peak-associated genes which are differentially expressed across any stage. Motif signatures and enriched pathways were obtained using Homer v4.11 ( Tracks of CUT&RUN peaks were visualized in Integrative Genomics Viewer version 2.8.9 (IGV, Broad Institute).

scRNA-seq sample processing and analysis

Neuronal cultures at day 27 of differentiation were washed three times in PBS, incubated with Accutase supplemented with Neuron Isolation Enzyme for Pierce (Thermo 88285) solution at 1:50 at 37 °C for 45–60 min and gently dissociated to single-cell suspensions via pipetting. After washing in PBS, single-cell suspensions were diluted to 1,000 cells per μl in 1× PBS with 0.04% BSA and 0.2 U μl−1 Ribolock RNAse inhibitor (Thermo EO0382) for sequencing. scRNA-seq was performed at the MSKCC Integrated Genomic Core for a target recovery of 10,000 cells per sample using 10X Genomics Chromium Single Cell 3′ Kit, version 3 according to the manufacturer’s protocol. Libraries were sequenced on an Illumina NovaSeq. The CellRanger pipeline (Version 6.1.2) was used to demultiplex and align reads to the GRCh38 reference genome to generate a cell-by-gene count matrix. Data analysis was performed with R v4.1 using Seurat v4.2.085. Cells expressing between 200 and 5,000 genes and less than 10% counts in mitochondrial genes were kept for analysis. Gene counts were normalized by total counts per cell and ScaleData was used to regress out cell cycle gene expression variance as determined by the CellCycleScoring function. PCA was performed on scaled data for the top 2,000 highly variable genes and a JackStraw significance test and ElbowPlot were used to determine the number of principal components for use in downstream analysis. A uniform manifold approximation and projection (UMAP) on the top 12 principal components was used for dimensional reduction and data visualization. FindNeighbors on the top 12 principal components and FindClusters with a resolution of 0.3 were used to identify clusters. Published scRNA-seq datasets for hPSC cortical differentiation were from Yao et al.86 (PMID: 28094016) and Volpato et al.87 (PMID: 30245212). To compare our dataset to those generated by Yao et al.86 and Volpato et al.87, Seurat’s anchor-based integration approach85 was used using FindIntegrationAnchors with 5,000 features. Single-cell hierarchical clustering and plotting for Extended Data Fig. 1h was performed with HGC88 using the Louvain algorithm. Single-cell RNA-seq analysis for mouse cortical development in Fig. 3f,g were from the published dataset from Di Bella et al.41 Data were processed using the same pipeline as in the original publication and developmental trajectories were inferred using v1.1.1.URD algorithm89.

Statistics and reproducibility

Sample sizes were estimated based on previous publications in the field. Investigators were not blinded to experimental conditions. However, for knockout and small molecule treatment studies, samples were de-identified respect to the molecular target. Transcriptional and genomic studies were performed with the same bioinformatic pipeline between conditions. Statistics and data representation were performed in PRISM (GraphPad) version 8,9 or 10, excel and R software version 3.5.2 or 4.1. Statistical tests used for each analysis are indicated in the figures’ legend. Data are represented as arithmetical mean ± s.e.m. unless otherwise indicated.

Independent replication from representative micrographs were as following. Fig. 1b, 6 experiments; Fig. 1j, 3 experiments; Fig. 1n, 2 experiments, Fig. 2d, 2 experiments; Fig. 3c, 2 experiments for each genetic perturbation; Fig. 4m, 4 experiments; Supplementary Fig. 2a, 4 experiments; Supplementary Fig. 2f, 3 experiments; Supplementary Fig. 6e, 1 experiment; Extended Data Figs. 6a, 2 experiments; Extended Data Figs. 6c, 2 experiments; Supplementary Fig. 8e, 2 experiments for d12 and d16.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Latest Intelligence