Forward programming of human pluripotent stem cells to generate glutamatergic and GABAergic neurons in a tri-culture model with astrocytes

Abstract

Background

Over the past decade, forward programming of human pluripotent stem cells (hPSCs) using various transcription factor (TF) combinations has been widely applied in neuroscience research. Ectopic NGN2 expression in hPSCs has been widely used for rapidly generating in vitro models of induced neurons (iNs) that are predominantly composed of excitatory glutamatergic neurons. Achieving a more balanced synaptic communication between excitatory and inhibitory neurons is essential for physiologically relevant in vitro models. Additionally, incorporating hPSC-derived astrocytes into models, rather than commonly used primary astrocytes, would more closely mimic in vivo disease phenotypes, especially for those associated with astrocyte dysfunction.

Methods

Inducible hPSC lines were generated by targeting the AAVS1 safe harbor site with TF transgene cassettes using CRISPR/Cas9. Forward programming was achieved through forced expression of NGN2 for glutamatergic neurons (iGlutNs), ASCL1/DLX2 for GABAergic neurons (iGABANs) and SOX9/NFIB for astrocytes (iAstros). Cell identity was validated by immunocytochemistry and bulk RNA sequencing. Functional properties were characterized on multielectrode arrays (MEAs).

Results

Bulk RNA sequencing confirmed lineage-specific differentiation while revealing distinct transcriptomic profiles between iAstros and human primary astrocytes. Functional assays demonstrated robust inhibitory control of network dynamics in co-culture with iGABANs on MEA, with enhanced responses to GABA_A receptor-targeting drugs including picrotoxin, bicuculline and clonazepam. Neurons co-cultured with iAstros showed reduced spontaneous activity compared to those cultured with primary astrocytes.

Conclusion

We successfully generated hPSC-derived excitatory and inhibitory neurons to establish an appropriate E/I balance in vitro, supported by primary astrocytes. Although astrocyte identity was confirmed in our hPSC-derived astrocytes, further optimization is required to achieve full functional maturation. This approach to developing an isogenic co-culture system derived from a single hPSC line may more faithfully replicate native neural network dynamics, offering a physiologically relevant platform for studying neurological disorders and screening therapeutic compounds.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13287-026-04917-6.

Background

Neurons and astrocytes are key components for maintaining the normal physiology of the central nervous system (CNS), with their interactions governing critical biological processes such as signaling modulation, metabolic support, microenvironment homeostasis and neuroprotection [1]. While mouse primary neural cell culture models have made significant contributions to neuroscience, they are unable to fully recapitulate the complexity of human brain cells [2]. The isolation of human primary neurons and astrocytes is challenging, costly, and has ethical implications, making human pluripotent stem cell (hPSC)-derived neuronal models a promising alternative source of brain cells [2].

Human somatic cells can be converted to a pluripotent state by introducing the reprogramming factors OCT4, SOX2, KLF4, and MYC [3]. Traditional strategies for the differentiation of hPSCs into neurons rely on the stepwise addition of growth factors and neural supplements, a process which can take months and results in heterogeneous populations of neurons and astrocytes with mixed states of maturity [4]. Recent advances in transcription factor (TF)-based forward programming have accelerated the process of disease modeling and drug screening. Combinations of different TFs can convert hPSCs into a diverse array of neuronal and glial lineages [4–6]. For instance, glutamatergic and GABAergic neurons, the two major neuronal populations in the brain, can be derived from hPSCs within several weeks via induced expression of NGN2 and ASCL1/DLX2, respectively [7, 8].

The healthy brain requires balanced excitation and inhibition (E/I balance) from synaptic transmission between glutamatergic and GABAergic neurons to fulfil its normal function [9, 10]. Impaired E/I balance is linked to many neurodevelopmental disorders including epilepsy, schizophrenia, autism spectrum disorder (ASD) and Alzheimer’s disease [11–13]. Additionally, astrocytes, being the most numerous glial cells in the mammalian brain, exert a variety of essential functions including metabolic and homeostatic support for neurons, neurotransmitter uptake and release, maintenance and regulation of synapses, modulation of synaptic transmission, and pain and inflammation processing [14–16]. Ectopic expression of SOX9/NFIB has been shown to drive the differentiation of hPSCs to mature astrocytes, which can support induced neurons (iNs) to develop functional networks in co-culture experiments [17]. Although lentiviral delivery systems are highly efficient DNA vehicles for stable transgene expression, their random integration into the cell genome can lead to potentially unpredictable risks for the host cells. The safety and reproducibility of forward programming can be enhanced by inserting an inducible TF-expression cassette into a safe harbor site such as the AAVS1 site [18, 19]. However, while the NGN2 induction method has been widely utilized, studies incorporating other types of neurons or glial cells remain limited [20].

In this study, we adopted a novel approach of targeting and modifying the AAVS1 site by CRISPR/Cas9 for the differentiation from hPSCs into three neural/glial lineages, including induced glutamatergic neurons (iGlutNs), GABAergic neurons (iGABANs) and astrocytes (iAstros). We validated successful cell fate determination after induced TF expression and identified distinct transcriptomic patterns between iAstros and primary astrocytes. Furthermore, we established tri-cultures with a predefined ratio of excitatory and inhibitory neurons, which exhibited robust GABAergic modulation of neuronal activity within the networks. Importantly, the application of compounds specific to GABA_A receptors elicited differential responses between networks with and without iGABANs, highlighting the functional integration of inhibitory neurons. In contrast, iAstros showed limited support for spontaneous neuronal activity compared to the primary astrocyte co-cultures, indicating that further optimization is required to enhance astrocyte functional maturation. Together, our findings establish a standardized method to incorporate inhibitory neurons into in vitro differentiated neural cultures to shape network dynamics, which is crucial for disease modeling of GABAergic dysfunction. This approach provides a scalable and translationally relevant platform for therapeutic screening and precision medicine applications in neurological disorders.

Methods

Molecular cloning

The AAVS1-Ngn2 and AAVS1-Ascl1-P2A-Dlx2 (AAVS1-APD) donor plasmids were a gift from Dr Michael Peitz (University of Bonn) as described previously [18, 19, 21]. To generate the AAVS1-Sox9-P2A-Nfib (AAVS1-SPN) plasmid, the NGN2 coding sequence was removed from the AAVS1-Ngn2 plasmid via SalI and MluI digest and replaced with a Sox9-P2A-Nfib cassette, which was amplified from the TetO-Sox9-Puro and TetO-Nfib-Hygro plasmids (Addgene #117269 and #117271) [17]. The digested AAVS1 backbone and amplified insert were assembled using the NEBuilder HiFi DNA Assembly Cloning Kit (New England Biolabs).

hPSC transfection for safe harbor targeting

The human embryonic stem cell (hESC) line H9 (WAe009-A, WiCell Research Institute) was used for this study. When hPSC cultures reached 70–80% confluence, cells were dissociated into single cells with Accutase (Thermo Fisher Scientific). The cells were seeded in a 24-well plate coated with vitronectin (Thermo Fisher Scientific) at 1 × 10⁵ cells/well in E8 medium (Thermo Fisher Scientific) supplemented with 10 µM Y-27632 (Hello Bio). Following overnight incubation, the medium was replaced with fresh E8 medium added with 10 µM Y-27632. In tube A, 3 µl of 1 µM AAVS1 Alt-R sgRNA (Integrated DNA Technologies) was mixed with 3 µl of 1 µM Alt-R S.p. Cas9 Nuclease V3 (Integrated DNA Technologies) in 19 µl of Opti-MEM (Thermo Fisher Scientific). Tube A was incubated for 5 minutes at room temperature to allow the formation of ribonucleoprotein (RNP) complexes. After that, 500 ng of the donor plasmid and 100 ng of the pCE-mp53DD (Addgene #41856) were added to tube A. In tube B, 1 µl of Lipofectamine STEM reagent (Thermo Fisher Scientific) was added to 25 µl of Opti-MEM. The tube B solution was added to tube A and mixed well. The mixture was incubated for 10 min at room temperature and then added dropwise to the well. After 24 h of transfection, the cells were dissociated into single cells with Accutase and replated onto a 6-well plate. On day 3 post-transfection, cells were selected with 0.3 µg/ml puromycin for approximately 7 days. Stable colonies were manually picked for clonal expansion and genotyping. Similarly, CLYBL targeting was performed using the CLYBL sgRNA and pC13N-iCAG.copGFP (Addgene #66578). Guide sequences are provided in Supplemental Table S1.

Genotyping

Phire Tissue Direct PCR Master Mix (Thermo Fisher Scientific) was used for genotyping following the manufacturer’s instructions. Transgene insertions at the AAVS1 and CLYBL sites were validated using the primers listed in Supplemental Table S1.

Neuronal differentiation

Generation of iGlutNs and iGABANs followed a previously published protocol with slight modifications [21]. Briefly, on day − 1, 60–80% confluent NGN2- and ASCL1/DLX2-transgenic hPSCs were dissociated with Accutase and seeded onto vitronectin-coated 6-well plates at a density of 600,000 cells/well in E8 medium supplemented with 10 µM Y-27632. On day 0, the medium was replaced with neuronal induction medium (1% N2 in DMEM/F12 medium, from Thermo Fisher Scientific) supplemented with 2 µg/ml doxycycline (Dox). Dox remained in the medium until day 8 for iGlutNs and day 14 for iGABANs. On day 2, cells were dissociated with Accutase and plated on T75 flasks coated with 15 µg/ml mouse laminin (Sigma Aldrich) at a density of 5 × 10⁶ cells/flask for iGlutNs and 9 × 10⁶ cells/flask for iGABANs. At this point, cells were cultured in neuronal maturation medium (2% B27 and 1% GlutaMAX in Neurobasal medium, from Thermo Fisher Scientific) supplemented with 2 µg/ml Dox, 10 ng/ml BDNF (PeproTech) and 10 µM Y-27632. BDNF was subsequently maintained in the medium from now on. On days 3 and 5, medium was changed to Dox- and BDNF-supplemented medium containing 10 µM DAPT (Hello Bio). DAPT remained in the medium from day 3 to day 7. On day 7, neurons were dissociated with Accutase and plated in poly-D-lysine/laminin-treated cultureware for relevant assays. The proliferative non-neuronal cells were removed by addition of 2.5 µM Ara-C (Sigma Aldrich) on day 7 for 24–48 h. Half of the culture medium was replaced every 2–3 days.

Astrocyte differentiation

Generation of iAstros followed the Canals protocol with some modifications [17]. On day − 1, 60–80% confluent SOX9/NFIB-transgenic hPSCs were dissociated with Accutase and seeded onto vitronectin-coated 6-well plates at a density of 1 × 10⁵ cells/well in E8 medium supplemented with 10 µM Y-27632. On day 0, the medium was replaced with fresh E8 medium containing 2 µg/ml Dox. Dox was maintained in the medium until day 14. On days 1 and 2, cells were cultured in Expansion medium: DMEM/F12, 10% FBS, 1% N2 and 1% GlutaMAX, from Thermo Fisher Scientific. From day 3 to day 6, Expansion medium was gradually switched to FGF medium: Neurobasal, 1% B27, 1% NEAA, 1% GlutaMAX, 1% FBS, from Thermo Fisher Scientific; 8 ng/ml bFGF, 5 ng/ml CNTF and 10 ng/ml BMP4, from PeproTech. On day 7, cells were dissociated with Accutase and replated in vitronectin-coated flasks in FGF medium at a density of 5 × 10⁴ cells/cm² according to the desired output. On day 8, the culture medium was changed with fresh FGF medium. On day 10, half of the medium was replaced with Maturation medium: 1:1 DMEM/F12 and Neurobasal, 1% N2, 1% sodium pyruvate, 1% GlutaMAX, from Thermo Fisher Scientific; 5 µg/ml N-acetylcysteine, 500 µg/ml dbcAMP, from Sigma Aldrich; 5 ng/ml EGF-like growth factor (Thermo Fisher Scientific), 10 ng/ml CNTF, 10 ng/ml BMP4, from PeproTech. From here, the medium was half-changed with Maturation medium every 2–3 days.

Immunostaining

For immunocytochemistry, cells were plated on poly-D-lysine/laminin-coated 8-well chamber slides. At the indicated timepoints, the neuronal cultures were washed once with DPBS, fixed with 4% paraformaldehyde (PFA) for 10 min at room temperature and then washed three times with DPBS. Primary antibodies were incubated overnight at 4 °C in blocking solution containing 1% normal goat serum (NGS; Millipore) and 0.1% Triton X-100. Cells were washed three times with DPBS and incubated with secondary antibodies in blocking solution for 2 h at room temperature. Samples were washed three times with DPBS before being mounted using Prolong Gold Antifade Mountant with DAPI (Thermo Fisher Scientific) and covered with a coverslip. The primary and secondary antibodies used in this study are listed in Supplemental Table S2. Confocal images were acquired with a Nikon A1R HD25 Confocal Microscope and processed using an analysis computer equipped with NIS-Elements software and ImageJ.

Multielectrode array (MEA)

For MEA co-culture experiments, 24-well MEA plates (Multi Channel Systems) were treated with 100 µl of 100 µg/ml poly-D-lysine overnight at 4°C. The plates were rinsed three times with sterile water and then coated with 75 µl of 15 µg/ml laminin for 2 h at 37 °C or overnight at 4°C. Cryopreserved human primary astrocytes (hpAs, ScienCell) were recovered in HA astrocyte medium (ScienCell) before being co-cultured with iNs. At 7 days of in vitro differentiation (DIV), each well was seeded with 1 × 10⁵ iGlutNs and 5 × 10⁴ iAstros or hpAs in neuronal maturation medium supplemented with 2 µg/ml dox, 10 ng/ml BDNF and 10 µM Y-27632. For co-culture with inhibitory neurons, an extra 2.5 × 10⁴ iGABANs were added to each well. The next day Y-27632 was withdrawn from the culture medium. Proliferative cells were removed by adding 2.5 µΜ Ara-C for 48 h. The medium was changed every 2–3 days and on the day of MEA recording. Dox was withdrawn on day 14 for all cultures.

A Multiwell-MEA-System plate reader (Multi Channel Systems) was used to record the electrical activity of neuronal cultures. Data extraction and offline analysis were conducted using Multiwell Analyzer software and GraphPad Prism. The sampling rate of the raw signal was 20 kHz, which was filtered with a high-pass filter (2nd-order Butterworth, 100 Hz cutoff frequency) and a low-pass filter (2nd-order Butterworth, 3500 Hz cutoff frequency). The threshold for spike detection was set at ± 4.5 standard deviations. All compound tests were performed after DIV 35 and all electrodes with spike rate < 1 Hz were excluded from the analysis.

For principal component analysis (PCA) plot, raw data were analyzed using the open source toolbox MEA-ToolBox [22]. Default values of the parameters were used for data filtration as recommended in their method section. Raster plots were generated via MATLAB code as described in Mossink et al. 2022 [10].

RNA sequencing (RNA-seq) analysis

Total RNA extraction was performed for iGlutNs, iGABANs and iAtros at DIV 28 using the Qiagen RNeasy Mini Kit according to the manufacturer’s protocol. Four biological replicates were prepared for each cell type. RNA quality and quantity were measured using QIAxpert Instrument System (Qiagen). RNA sequencing libraries were constructed by BGI Genomics (Hong Kong). Standard bulk RNA-seq analysis was conducted by Monash Genomics and Bioinformatics Platform. Differential gene expression (DGE) analysis was performed in Degust [23] and R program. Plots were generated using R packages in RStudio. In addition, the ClusterGVis package was employed to visualize clustering of the gene expression matrix [24]. Bulk RNA-seq data for H9 hESCs, hpAs (ScienCell) and human induced astrocytes (hiAs) were obtained from NCBI GEO Database (GSE84639), with supplemental information from corresponding published papers [25, 26].

Statistics

Statistics of the MEA data were performed in GraphPad Prism. Data are presented as mean (standard error of mean (SEM)), unless otherwise specified. A two-way ANOVA model was used if there were no missing values and a mixed-effects model was used if there were missing values. Multiple comparisons were corrected using Post hoc Tukey statistical hypothesis testing. A threshold of p < 0.05 was used to determine statistical significance under different experimental conditions.

Results

Generation of inducible NGN2, ASCL1/DLX2 and SOX9/NFIB transgenic hPSC lines

Previously, we generated NGN2 transgenic mouse ESCs (mESCs) and human induced PSCs (hiPSCs) via the piggyBac transposon and AAVS1 targeting systems, which effectively produced excitatory neurons [27–29]. Here, we utilized the H9 cell line to create stably inducible cell lines using the safe harbor targeting strategy for TF overexpression (Fig. 1A). During targeting experiments, cells were co-transfected with an episomal plasmid (pCE-mp53DD) which expresses a dominant-negative p53 variant known to prevent apoptosis induced by double-strand breaks (DSBs) and enhance homology-directed repair (HDR) efficiency [30]. Following puromycin selection, we confirmed transgene insertion at one or both AAVS1 alleles in all selected colonies. Multiple clones from each transfection were tested for Dox induction and those exhibiting homogeneous differentiation were expanded for further characterization.

Fig. 1.

Rapid generation of tri-lineages from hPSC lines. A Schematic depiction of the AAVS1 safe harbor-targeting for inducible expression of TFs. B Overview of the experimental setup for generating iNs and iAstros from transgenic hESCs. EM: Expansion Medium; MM: Maturation Medium. C Bright field images of neural and astrocyte induction at early timepoints showing rapid morphological changes. Scale bar = 100 μm. D Immunocytochemistry of iNs and iAstros stained with neuronal and astrocyte markers. iGlutNs co-cultured with hpAs (i-iii) were stained with MAP2, VGLUT1, GABA, Tuj1, NeuN and synaptophysin (SYP) at DIV 28. iGABANs co-cultured with hpAs (iv & v) were stained with VGAT, α-tubulin, GABA and Tuj1 at DIV 28. iAstros (vi) were stained with GFAP at DIV 17. Scale bar = 50 μm. E Immunocytochemistry for the trilineage coculture (E/I ratio = 80:20) at DIV 42. The iNs were stained with MAP2 and the primary astrocytes were stained with GFAP. Synapses were stained with markers of excitatory synapses, VGLUT2 (pre-synaptic) and PSD-95 (post-synaptic), and of inhibitory synapses, VGAT (pre-synaptic) and gephyrin (post-synaptic). Scale bar = 50 μm. Illustrative diagrams were created in BioRender

Forward programming to human glutamatergic neurons, GABAergic neurons and astrocytes

We established and optimized a protocol for rapidly and reproducibly generating hPSC-derived excitatory neurons (iGlutNs) and inhibitory neurons (iGABANs), which can be co-cultured with primary astrocytes or hPSC-derived astrocytes (iAstros) (Fig. 1Bi). The transient application of the Notch signaling blocker DAPT was found to promote neuronal differentiation from hPSCs [31]. Compared with iGlutN cultures, iGABAN cultures showed significant cell death between days 3 and 5 during Dox induction. By DIV 7, iGlutNs developed nascent neuronal networks whereas iGABANs only exhibited neural progenitor morphology with short projections (Fig. 1C). Extended culture periods improved the complexity of neurite arborization in both neuronal subtypes.

For astrocyte generation (Fig. 1Bii), SOX9/NFIB induction led to rapid cellular changes, with cells adopting elongated or polygonal fibroblast-like morphology (Fig. 1C). These cells showed high proliferation rates during the first two weeks of induction, with cell division slowing around day 14 in astrocyte maturation medium. However, we observed reduced cell viability under serum-free conditions, which was improved when the cells were cultured in commercial serum-containing astrocyte medium.

We then performed immunofluorescence analysis to verify the neural/glial lineage subtypes after induced expression of TFs. In co-cultures with primary astrocytes at DIV 28, iNs expressed neuronal markers including MAP2, NeuN, Tuj1, Synaptophysin and α-tubulin (Fig. 1Di-v). In particular, we detected the expression of glutamatergic neuronal marker VGLUT1 in iGlutNs (Fig. 1Di) and GABAergic neuronal markers GABA and VGAT in iGABANs (Fig. 1Div & v). Notably, we also identified a low percentage of GABA-positive neurons in iGlutN cultures (Fig. 1Dii), consistent with previous reports [19, 21]. By DIV 17, iAstros presented moderate signals of the typical astrocyte marker GFAP (Fig. 1Dvi). To confirm whether cells from different lineages formed synapses, we performed immunofluorescence on co-cultures comprising iGlutNs, iGABANs and primary astrocytes at DIV 42. The cells exhibited mature markers of excitatory synapses, VGLUT2 (pre-synaptic) and PSD-95 (post-synaptic), and of inhibitory synapses, VGAT (pre-synaptic) and gephyrin (post-synaptic) (Fig. 1Ei & ii). Moreover, MAP2 and GFAP overlapping regions showed increased VGLUT2 and VGAT puncta (Fig. 1Eiii & iv), indicating that synaptic terminals are enriched near sites of close neuron-astrocyte apposition. Overall, these results suggest that the transgenic hPSC lines can rapidly differentiate into specific neural/glial lineages iGlutNs, iGABANs and iAstros. Furthermore, when combined in a trilineage culture with primary astrocytes, iNs form mature synaptic connections.

Bulk RNA-seq analysis reveals distinct signatures of the three neural/glial cell types

To determine the transcriptional profiles of the cell lines, we performed bulk RNA-seq on iGlutN, iGABAN and iAstro cultures at DIV 28 (n = 4 biological replicates). Transcriptome data of the three differentiated cell types were compared with the original pluripotent hESCs population. PCA demonstrated that replicates within each group were tightly clustered and clearly separated from undifferentiated H9 cells by principal component 1 (PC1: 65.28% variation) (Fig. 2A). The iAstros group was distinct from iGlutNs and iGABANs, which were closely clustered along principal component 2 (PC2: 28.45% variation), indicating highly similar transcriptomes between the two neuronal subtypes. Following that, we conducted PCA without including the hESCs group to focus on the differences between the three neural/glial lineages (Fig. 2B). The results revealed that iAstros were distinctly separated from iGlutNs and iGABANs by PC1 (83.47% variation). Furthermore, the two neuronal subtypes iGlutNs and iGABANs were separated from each other along PC2 which accounted for only 14.31% of the variation. In summary, these results indicate that the transcriptomic differences between iGlutNs and iGABANs were relatively subtle, with most of the variation attributed to the iAstros.

Fig. 2.

Transcriptomic profiling of the three neural/glial lineages generated by directed differentiation. A PCA of bulk RNA-seq data comparing the three lineages and hESCs. Comparisons between induced cells and hESCs control explained the majority of the variance. B PCA of the three lineages excluding the hESCs control. A comparison between iNs and iAstros explained the majority of the variance. C Dot plot of the expression of interneuron subtype genes in iGABANs samples. Gene counts are shown in counts per million (CPM). D Supervised clustering and heatmap analysis of the neural/glial lineages and hESCs. Typical marker genes of each cell type were selected for clustering the samples. The expression data are transformed into Z-scores. E Volcano plot comparing iGlutNs and iGABANs. Thresholds are set as |log2(Fold Change)| > 1 and Adjusted P-Value < 0.05. Upregulated, not significant (NS) and downregulated genes are marked in red, gray and blue, respectively. Glutamatergic and GABAergic markers are highlighted in green. F GO terms related to synaptic transmission enriched in DEGs between iGlutNs and iGABANs. The -log10(padjust) values displayed in the color bar are for visualization purposes only. Statistical significance should be interpreted using absolute values. G Venn diagram showing epilepsy-related genes covered in the three lineages. CPM > 1 is considered to be expressed in the samples

We next conducted DGE analysis against hESCs to identify lineage-specific markers in the samples. A heatmap of characteristic markers for each cell type confirmed distinct signatures of glutamatergic neurons, GABAergic neurons and astrocytes (Fig. 2D). Moreover, a comparison between iGlutNs and iGABANs revealed over 1000 differentially expressed genes (DEGs). As shown in the volcano plot (Fig. 2E), the glutamatergic neuronal markers SLC17A7 (VGLUT1) and SLC17A6 (VGLUT2) were enriched in iGlutNs, whereas the GABAergic markers SLC32A1 (VGAT), GAD1 and GAD2 were upregulated in iGABANs. Notably, we also observed moderate VGAT and GAD1 expression in iGlutNs, which is consistent with our immunostaining results showing a small proportion of GABA-expressing neurons.

Expression patterns of neuropeptides and calcium-binding proteins serve as hallmarks to distinguish the subtypes of mature cortical interneurons, including parvalbumin (PV), somatostatin (SST), neuropeptide Y (NPY), calretinin (CR), calbindin (CB), reelin (RELN), neuronal nitric oxide synthetase (nNOS) and vasoactive intestinal peptide (VIP) [8, 32, 33]. We detected minimal expression of these markers in iGABANs (Fig. 2C), suggesting incomplete maturity of iGABANs into subtypes at the DIV 28 time point. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment were performed to uncover the biological process (BP), cellular component (CC), molecular function (MF) categories and KEGG pathways associated with the mRNAs expressed by the three hPSC-derived neural/glial lineages. The analysis revealed shared neuronal terms between iGlutNs and iGABANs, including synaptic organization, modulation of chemical synaptic transmission, synaptic and postsynaptic membrane, neuronal cell body, monoatomic ion channel activity, gated channel activity and calcium signalling pathway (Supplemental Figures S1). iAstros were enriched in key functional categories including extracellular structure/matrix organization, connective tissue development, basement membrane and extracellular matrix structural constituent (Supplemental Figure S1). KEGG pathways analysis further highlighted iAstros involvement in neuroactive ligand-receptor interaction and ECM-receptor interaction, underscoring their essential roles in support, regulation and protection within nervous tissues. A direct comparison between neuronal subtypes revealed enrichment of excitatory and inhibitory synaptic terms in iGlutNs and iGABANs, respectively (Fig. 2F). These findings reflected the physiologically relevant expression characteristics of the three neural/glial lineages generated via directed differentiation of hPSCs.

We then sought to evaluate the relevance of these lineages for disease phenotyping in epilepsy. We compared the expression of genes related to epilepsy with a recently updated table of 977 epilepsy-associated genes [34], A total of 919 genes (94%) were detected across the three lineages (CPM > 1), with 830 genes expressed in all three lineages (Fig. 2G). This supports the potential utility of our tri-culture system for epilepsy modeling.

iAstros exhibit distinct transcriptional profiles from human primary astrocytes

To evaluate how closely iAstros resemble human primary astrocytes (hpAs), we compared RNA-seq publicly available data for hpAs (ScienCell) and human induced astrocytes (hiAs) generated by another group, who used ectopic NGN2 expression alongside small molecule patterning to induce neural progenitor stage, followed by maturation in astrocytes media [26]. PCA clearly separated all astrocyte groups from hPSCs along PC1 (47.32%), whereas iAstros were separated from closely clustered hpAs and hiAs by PC2 at a percentage of variance similar to PC1 (Fig. 3A). Excluding hPSCs, PCA showed that the most of the variance (PC1 82.27%) was due to differences between iAstros and hpAs/hiAs, whereas hpAs and hiAs showed modest separation along PC2 (15.04%) (Fig. 3B).

Fig. 3.

Molecular signatures of hpAs, hiAs and iAstros. A PCA of bulk RNA-seq data comparing the three astrocyte groups and hPSCs. PC1 and PC2 shared nearly equal percentages of variance in the comparisons. B PCA of the three astrocyte groups excluding the hPSCs. Comparisons between hpAs/hiAs and iAstros explained the majority of the variance. C Volcano plot of DGE analysis between hpAs and iAstros. Thresholds are set as |log2(Fold Change)| > 2 and Adjusted P-Value < 0.05. Upregulated, not significant (NS) and downregulated genes are marked in red, gray and blue, respectively. Astrocyte markers are highlighted in green. D Dot plots showing expression levels of astrocyte genes in the three types of astrocytes. Gene counts are shown in CPM. E Supervised clustering, heatmap and GO enrichment analysis of the three types of astrocytes, graphed using the ClusterGVis package

Subsequent DGE analysis revealed upregulation of astrocyte genes S100B, ITGA6 and ID3 in iAstros compared to hpAs (Fig. 3C). As displayed in Fig. 3D, we found divergent expression patterns of canonical astroglial markers across the three groups. CLU and GFAP showed robust expression in hpAs, with minimal expression in both hiAs and iAstros. CD44 and SLC1A3 also presented highest expression in hpAs, with notably reduced levels in hiAs and iAstros. Conversely, iAstros showed higher levels of markers such as S100B, ITGA6 and ID3, compared with hpAs and hiAs.

We next sought to characterize a total of 4351 DEGs using supervised clustering analysis. Heatmap and GO enrichment identified four clusters of DEGs, with the transcriptional differences mostly related to iAstros (Fig. 3E). hiAs and hpAs showed similar patterns in Clusters 2 and 4, which were enriched in immune response and axonogenesis, while iAstros showed a closer expression pattern to hpAs in Cluster 3 associated with synaptic organization. Furthermore, Cluster 1, involved in cell adhesion and synapse organization, was uniquely upregulated in iAstros. Overall, iAstros exhibited a distinct transcriptional profile from hpAs and hiAs, whereas hiAs more closely approximated the molecular signatures of hpAs.

iGABANs modulate network activity in MEA recordings

Following generation of various neural/glial cell types, we tested their functional properties. MEAs are a non-invasive system for sensitive and repeated measurements of network properties of in vitro cell cultures [35, 36]. We first compared two conditions: iGlutNs alone (E/I ratio = 100:0) versus iGlutNs co-cultured with iGABANs (E/I ratio = 80:20), both in the presence of primary astrocytes (hpAs). The E/I 80:20 approximates the physiological proportions of inhibitory interneurons (20–30%) in the human cerebral cortex [37–39]. MEA recordings from DIV 14–35 demonstrated progressive network maturation in both E/I ratios. Functional maturity of the two conditions was assessed in three aspects: spikes, bursts and network bursts. As shown in Fig. 4A, only sparse, random spiking activity was observed in both compositions at DIV 14. By DIV 21, the cultures exhibited local bursting activity, followed by synchronized network bursts appearing around DIV 28 (Fig. 4B). Both ratios showed increased spike rate and bursting activity over time, indicating the development of interconnected neuronal networks. They also exhibited a similar distribution of spike amplitudes, with increased spike counts over time (Figure S2). There were, however, significant differences in firing patterns between E/I 100:0 and E/I 80:20, with E/I 80:20 cultures showing lower spike counts than E/I 100:0 at all timepoints. To further explore network activity, we focused on two key parameters: network burst duration (NBD) and percentage of spikes in network bursts (PSNB) at DIV 28, 32 and 35 (Fig. 4C & D). Both culture conditions exhibited elevated values of these two parameters, which suggested increased synchronicity during maturation as more spikes became involved in network activity. Notably, E/I 80:20 resulted in consistently lower spike rate, NBD and PSNB, even though their mean network burst rate (NBR) reached comparable levels (~ 0.18 Hz) to those observed in E/I 100:0 at DIV 35 (Fig. 4B). These results demonstrate the impact of iGABANs in modulating network behavior, with the most consistent and marked effect being to reduced burst duration.

Fig. 4.

Functional characterization of E/I 100:0, E/I 80:20 and iNs co-cultured with iAstros on MEA. Comparisons of neuronal activity between E/I 100:0 and E/I 80:20 including A Spike rate, B Network burst rate, C Network burst duration, D Percentage of spikes in network burst (n = number of electrodes (432) for spike rate and n = number of wells (36) for network parameters). E Representative raster plots showing firing patterns of E/I 100:0 and E/I 80:20 at DIV 28 and DIV 35. Scale bar = 10 s. F PCA of the two E/I ratios at DIV 28 and DIV 35 using 18 parameters extracted from MEAToolBox. PCA loadings are provided in Supplemental Table S3. G Oscillation parameters at different timepoints indicated by CV of ISI and CV of NBIBI. H Synchronicity parameter at different timepoints indicated by mean ISI distance. I Representative raster plots showing firing patterns of iGlutNs co-cultured with iAstros that were differentiated in Canals medium (iAstros_C) and HA astrocyte medium (iAstros_S). J Comparison of spike rate at weekly timepoints between iGlutNs co-cultured with hpAs, iAstros_C and iAstros_S (n = number of electrodes (432, 288 and 288)). K Network comparison of DIV 35 network burst rate of iGlutNs with the three astrocyte groups (n = number of wells (36, 20 and 18). Wells with no network activity were excluded from the analysis. L Representative images showing iAstros with GFP reporter. Images i & ii show iAstros at DIV 26. Images iii & iv show iNs co-cultured with iAstros or hpAs on MEA at DIV 28. Scale bar = 100 μm. MEA data are presented as mean ± standard error of mean (SEM). ^nsp > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, and ****p < 0.0001. For A-D, G, H and J, mixed model two-way ANOVA was performed between timepoints and groups. Multiple comparisons were corrected using Post hoc Tukey statistical hypothesis testing. For K, Kruskal–Wallis non-parametric test was performed, with correction using the Dunn’s method

To better illustrate the functional features of the two E/I ratios, PCA was conducted at DIV 28 and DIV 35 using 18 parameters exported from the open source toolbox for MEA data analysis [22]. PCA revealed distinct neuronal activity patterns between the two conditions (Fig. 4F). The PC1, accounting for 57% and 51% of the variance at DIV 28 and DIV 35 respectively, clearly separated the two culture compositions. We observed that E/I 100:0 tended toward higher positions along PC1 for spike rate and burst-related parameters. Conversely, co-cultures with iGABANs demonstrated stronger correlations with interspike intervals (ISI) and network burst interburst intervals (NBIBI). However, these distinctions between the two groups became less pronounced by DIV 35 as both were more scattered on both PCs. In addition, the E/I 80:20 co-cultures, located mainly toward the negative PC1 axis, showed a broader distribution along PC2, indicating more complex network dynamics in the presence of iGABANs.

Oscillation, quantified by coefficient of variation (CV) of ISI or NBIBI, reflects the temporal coordination and regularity of neuronal firing patterns across the network [40, 41]. Both E/I ratios showed CV of ISI close to 1.0 at DIV 21, with a significant increase over time (Fig. 4G). The CV of ISI increased the most in E/I 100:0, significantly surpassing that in E/I 80:20 at both DIV 28 and DIV35, when network bursts were formed. CV of NBIBI in E/I 80:20 was initially higher, 0.51 at DIV 28, before significantly decreasing to 0.36 at DIV 35, although remaining significantly higher than E/I 100:0, which remained stable (0.27 to 0.25). The increase in CV of ISI in both groups indicates the emergence of alternating periods of high and low activity in the neuronal cultures, which is a hallmark of functional neuronal networks. Furthermore, the higher CV of NBIBI in E/I 80:20 at DIV 28 reflected greater variability in network burst timing during the initial emergence of synchronized activity, suggesting the early influence of GABAergic signaling. Its reduction by DIV 35 indicated more regular network bursts with the establishment of a proper E/I balance. Additionally, we quantified the synchronicity index using the mean ISI distance (Fig. 4H) in MEA-ToolBox [22] and observed a gradual decrease in both cultures, indicating that the spike trains became more synchronized as the network matured. Similar to a previous report [10], these data suggest a hyperpolarizing shift of the GABA reversal potential shaping network activity patterns over time.

iAstros show limited support to neurons at the network level

Since in vitro neuronal models often require astrocytes to promote synaptic formation, rodent primary astrocytes have been routinely used for co-culture with hPSC-derived neurons [7, 42, 43]. Given the differences in genetic background between rats and humans, we used human primary astrocytes (hpAs) in the experiments above, but there is also a need to explore the use of hPSC-derived astrocytes as an alternative source [44]. We assessed whether iAstros can support the formation of functional synapses on MEA as a key aspect of co-culture experiments. We compared neuronal activity of iGlutNs co-cultured with iAstros differentiated in either Canals medium or HA astrocyte medium. A representative raster plot of the network bursts generated by iGlutNs co-cultured with iAstros at DIV 35 is shown in Fig. 4I. Notably, iGlutNs exhibited markedly longer synchronous bursts in the presence of iAstros cultured in Canals medium than in HA astrocyte medium. As shown in Fig. 4J, the spike rate of iGlutNs cultured with iAstros in either media, was considerably lower than that in co-cultures with hpAs at all timepoints. Although burst firing behavior was detected in iGlutNs with iAstros around DIV 30, NBR was significantly lower than in co-cultures with hpAs (Fig. 4K). Consequently, iNs co-cultured with iAstros showed reduced network maturity on MEA, compared with those with primary astrocytes.

While neuronal medium was able to support hpAs in co-culture experiment, co-culturing iAstros in this medium may lead to reduced astrocyte viability. To better track iAstros in co-cultures with iNs, we targeted our SOX9/NFIB transgenic cell line at another safe harbor site CLYBL with a GFP reporter cassette using CRISPR/Cas9. The dual-targeted cell line showed bright GFP expression in the differentiated state (Fig. 4L). iAstros were induced for 21 days in HA astrocyte medium before being combined with DIV 7 iGlutNs. When plated on a transparent 24-well MEA plate, even distributions of iAstros and iGlutNs were observed across the wells after seeding. However, by DIV 28 co-cultures with iAstros became highly clustered, while iGlutNs formed evenly distributed networks with hpAs (Fig. 4L). These findings indicate that, while neuronal medium did not significantly reduce astrocyte viability, the clustering of iGlutNs with iAstros and potentially reduced synaptic competence may contribute to the lower NBR observed. Thus, while iAstros support synaptogenesis and network formation, there is room for further optimization to improve their functionality.

Co-culture with iGABANs enhances the response to compounds targeting GABA_A receptors

We next investigated how GABAergic neurotransmission controls network activity dynamics by using drugs that specifically target GABA_A receptors on co-cultures with primary astrocytes (hpAs) after DIV 35. The MEA drug testing protocol is outlined in Fig. 5A, with a 5-minute baseline recording before incubation with the drug for 15 min, followed by a 5-minute “Drug” recording. For screening potential antiseizure medications (ASMs), the pro-convulsant 4-aminopyridine (4-AP) will be added to induce hypersynchronous epileptiform-like activity, and drug effects will be assessed during a 5-minute “Epileptiform” recording. We tested two antagonists of GABA_A receptors, picrotoxin (PTX) and bicuculline (BIC), in E/I 100:0 and E/I 80:20 cultures. Following acute treatment with 3 µM BIC or 1 µM PTX [45, 46], both E/I conditions showed an increase in neuronal activity, with varying degrees of response as seen in raster plots (Fig. 5B). Both E/I compositions exhibited significantly higher spike rates in the presence of BIC or PTX, compared to the vehicle control, but the increase was markedly greater in the presence of iGABANs (Fig. 5C). Spike amplitude distribution analysis also demonstrated increased spike counts after PTX or BIC treatment (Figure S3). In terms of bursting activity, both E/I networks exhibited an increased burst rate and NBR after PTX or BIC, however, the increase in NBR was not statistically significant between the two network compositions (Figure S4A&B). Notably, NBD was significantly and markedly increased in E/I 80:20 treated with both compounds, whereas E/I 100:0 exhibited only a slight increase with PTX condition but not with BIC. Overall, co-cultures with iGABANs showed a greater response to these two compounds compared to that of iGlutNs alone, reflecting a greater role for GABA_A receptors in regulating network activity in these cultures.

Fig. 5.

Enhanced inhibitory modulation in response to GABA_A receptor antagonists and PAM in E/I 80:20 networks. A Schematic diagram (created in BioRender) of drug testing protocol on the MEA platform, with drugs added at the arrowheads, and left on, and MEA recordings made for the period of the colored boxes. Firing patterns of E/I 100:0 and E/I 80:20 networks after application of BIC or PTX including B Representative raster plots, C Normalized spike rate (n = number of electrodes), and D Normalized network burst duration (n = number of wells (12 for all except 11 for E/I 80:20 PTX)). Firing patterns of the two E/I ratios after CZP treatment including E Representative raster plots, F Normalized spike rate (n = number of electrodes), G Normalized network burst rate (n = number of wells (9)) and H Normalized network burst duration (n = number of wells (9)). Firing patterns of the two E/I ratios after addition of 4-AP following CZP treatment including I Representative raster plots, J Normalized spike rate (n = number of electrodes) and K Normalized network burst rate (n = number of wells (9)). Electrodes with spike rate < 1 Hz or wells with no network burst were excluded from the analysis. Data are presented as min to max in box plots. ^nsp > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, and **** p < 0.0001. Mixed model two-way ANOVA was performed between timepoints and groups. Multiple comparisons were corrected using Post hoc Tukey statistical hypothesis testing

We also tested the GABA_A receptor agonist and positive allosteric modulator (PAM) clonazepam (CZP), which is a member of the commonly used ASMs benzodiazepines (BZDs). CZP is known to exert its anticonvulsant and anxiolytic effects in a highly potent and long-acting manner [47, 48]. Three concentrations (0.1, 1 and 10 µM) were tested on the cultures to assess concentration-dependent effects. CZP treatment significantly reduced the spike rate in both E/I compositions, with a slightly greater reduction at 0.1 µM CZP in E/I 80:20, relative to E/I 100:0 (Fig. 5F). CZP also significantly shortened NBD in both E/I ratios at all concentrations tested (Fig. 5H). It was for burst rate and NBR, however, that the impact of CZP addition was notably greater in E/I 80:20 networks, reducing both measures further than in E/I 100:0 networks (Fig. 5G and Figure S4C). Higher doses of CZP did not result in a greater reduction in network activity, suggesting near-saturation of positive modulatory effects on GABA_A receptors at lower concentrations. We further tested the effect of prior CZP treatment on hypersynchronous epileptiform-like activity induced by addition of 4-AP, as shown in Fig. 5A. CZP treatment reduced neuronal activity in a dose-dependent manner, with greater effects in the E/I 80:20 (Fig. 5J & K and Figure S4D). Together, these results confirmed that iGABANs exerted robust GABAergic synaptic inputs in co-cultured neuronal networks.

Discussion

In this study, we developed a tri-culture model composed of glutamatergic neurons, GABAergic neurons and astrocytes to model human brain electrophysiology, although the induced astrocytes require further maturation to fully support functional network activity. While NGN2-induced neurons have been widely used [49], it is not routine to incorporate inhibitory neurons into these culture networks, which is essential for establishing an appropriate physiological E/I balance. RNA-seq analysis revealed differential transcriptomic profiles of our NGN2-induced neurons and ASCL1/DLX2-induced neurons, characterizing them as glutamatergic neurons and GABAergic neurons. Furthermore, GO term and KEGG pathway analysis highlighted enrichment in synapse- and neural signaling-related terms in both neuronal populations. It was notable, however, that interneuron subgroup markers were largely absent in iGABANs, indicating that their maturity had not extended to subtype differentiation. Previous reports indicate that most subtype markers were detected in ASCL1/DLX2-induced neurons cultured on mouse glia at 5 weeks post-induction [8], and that PV, CB, CR and SST expression could be detected in iGABANs cultured on mouse astrocytes at later timepoints [19]. Thus, the immaturity of iGABAN monocultures may result from the absence of astrocyte support or the shorter culture period. Future studies could explore whether iNs reach full subtype specification when co-cultured with primary astrocytes over extended periods.

We then assessed the functional properties of iNs on MEA, demonstrating that the E/I 80:20 cultures showed a significant reduction in spike rate and NBD compared to the E/I 100:0 cultures, confirming the modulation of network activity through GABAergic inputs from iGABANs. Furthermore, administration of GABA_A receptor antagonists BIC and PTX resulted in more evident elevation of spike and bursting activity in co-cultures with iGABANs than iGlutNs alone. Similarly, treatment with the commonly used ASM, CZP, resulted in enhanced inhibition in E/I 80:20 networks, particularly on NBR, but also on spike rate at low CZP doses. The enhanced inhibitory effect of CZP in E/I 80:20 networks on both spike rate and NBR was more prominent in the hyperactive state induced by 4-AP. BZDs, such as CZP, bind to their unique allosteric site on GABA_A receptors, which enhances GABAergic signaling by enhancing GABA-activated channel opening and chloride influx, leading to neuronal hyperpolarization [48]. Since BZDs alone cannot activate the receptors without GABA binding, it is likely that greater GABAergic synaptic activity and higher GABA levels in E/I 80:20 cultures are responsible for the heightened effects of CZP in these conditions. These results highlight the essential roles of inhibitory neurotransmission in the presence of iGABANs.

Interestingly, our study shows that the differences in firing patterns between E/I 100:0 and E/I 80:20 networks were evident as early as DIV 28, in contrast to a previous report where significant differences between E/I 100:0 and E/I 65:35 were not observed until DIV 42 [10]. This report only observed an effect of PTX on spike rate and NBD of E/I 65:35 networks by DIV 49 but not at DIV 35. Since they generated iGABANs using a different protocol involving expression of ASCL1 alone with addition of forskolin (FSK), it is possible that these neurons require an extended maturation period to enable inhibitory control of network activity. Our results suggest that iGABANs significantly affect network dynamics earlier than previously reported. Contradictory results have been observed in NGN2-induced neurons regarding response to GABA_A receptor antagonists [10, 50–52]. We demonstrated that iGlutNs alone displayed modulation of network activity by PTX, BIC and CZP, albeit less robustly than co-cultures with iGABANs. In support of this finding, we confirmed the presence of GABA-positive neurons in iGlutNs by immunofluorescence and RNA-seq. Others have reported that GABA-expressing neurons accounted for approximately 1–2% of NGN2-induced neurons [19, 21]. Although NGN2-induced neurons alone could respond to compounds targeting GABA_A receptors, the low percentage of GABA-positive neurons likely prevents them from fully mimicking the network dynamics within the human brain.

While ASMs exert their function by reducing hyperactivity or hypersynchrony in pathological cortical networks, they are typically categorized on the basis of their molecular mechanisms, including sodium channel blockers and GABA receptor PAMs [53, 54]. Therefore, inclusion of GABAergic neurons in hPSC-derived neuronal models is particularly critical for pharmacological screening to identify compounds targeting GABAergic signaling. PV-positive and SST-positive interneurons are crucial regulators of E/I balance [55, 56], but the ASCL1/DLX2 induction method yielded a low frequency of these interneuron subclasses. An alternative ASCL1/FSK differentiation protocol enriches SST-expressing neurons to 30%, but PV-positive interneurons remain largely absent in the resulting population [10]. Thus, it may be necessary to refine directed differentiation methods in order to generate more defined neuronal subpopulations from hPSCs for better disease modelling.

Although NGN2 induction has been extensively used for neuroscience research, its simplicity in cell composition raises concerns. Brain organoids, which offer more structural and functional complexity, may better recapitulate the human brain [57, 58]. Nevertheless, generation of 3D organoids involves multiple steps, making it labor-intensive and time-consuming. For instance, it requires over 250 days for cerebral organoids to reach later stages of development [59]. In contrast, 2D neuronal monolayers produced by forward programming remain advantageous in terms of homogeneity, scalability and reproducibility, which are essential for high-throughput pharmacological screens. The complexity of monolayer neural cultures can be further enhanced by incorporating other TF-induced neural/glial lineages to study interactions between various cell types, such as microglia, oligodendrocytes and dopaminergic neurons [60–62]. Recently, a novel technology involving ectopic TF expression in combination with signaling modulators was shown to generate region-specific neuronal subtypes from hiPSCs, known as patterned iNs [49, 63]. This combinatorial patterning approach may promote the creation of more precise brain region-specific disease models, improving our understanding of neurological disorders. In addition, safe harbor targeting strategies overcome the drawbacks associated with lentiviral-based techniques, allowing precise insertion of inducible transgene cassettes and creation of desired in vitro models. By combining CRISPR interference/activation platforms with forward programming, multimodal genetic screens have been performed in hiPSC-derived neurons and microglia to uncover modifiers of disease states [64, 65]. Hence, the ability to rapidly and efficiently engineer cell fates expedites research progress to develop relevant disease models for high-throughput screening.

Astrocytes play crucial roles in supporting neuronal function within the CNS. While significant progress has been made in neuronal differentiation, deriving functionally mature astrocytes remains challenging. In this study, we explored the generation of induced astrocytes through SOX9/NFIB induction to establish all hPSC-derived tri-cultures. However, RNA-seq analysis revealed distinct transcriptomic differences between iAstros and commercial primary astrocytes (hpAs), particularly in the expression of canonical astrocyte markers like GFAP. GFAP expression was found to be higher in hpAs than in freshly extracted fetal or adult astrocytes [66]. The low GFAP levels in iAstros may represent a more quiescent phenotype since GFAP expression is typically upregulated during astrocyte activation [67, 68]. GFAP levels can also be increased in the presence of serum in the culture medium [69], which might contribute to the high GFAP expression in hpAs cultured in HA astrocyte medium (2% FBS). Single-cell RNA-seq analysis of SOX9/NFIB-induced astrocytes could provide deeper insights into the astrocytic transcriptome and differentiation trajectory [70].

When co-cultured with iNs, we observed clumping of neurons and a lower level of neuronal activity recorded on MEA. Notably, the combination of NGN2 induction with patterning factors produced hiAs with molecular signatures more similar to those of hpAs [26]. These hiAs recapitulated key functions of hpAs including response to pro-inflammatory stimuli, calcium oscillation dynamics and synaptic formation. Evaluating iNs co-cultured with hiAs on MEA could provide insights into their ability to mirror the network activity observed with hpAs. Moreover, SOX9 induction alone has also been reported to generate functional astrocytes with better performance than primary rat neuronal cultures on MEA [66]. This study demonstrated that SOX9-induced astrocytes presented functional features similar to those of hpAs except that their activity on MEA was not directly compared. In another study, four different TF combinations that define astrocyte identity were systematically analyzed to explore their relevance to primary astrocytes [71]. SOX9/NFIB-induced astrocytes displayed transcriptomic profiles distinct from the other three combinations. Adding NFIA to the SOX9/NFIB protocol produced iAs that more closely resembled the maturity of brain astrocytes. When co-cultured with iNs, these iAs exhibited spike and burst firing patterns comparable to those in co-cultures with rat astrocytes at DIV 42 on MEA. Thus, future studies will be necessary to explore various astrocyte generation protocols for establishing reliable hPSC-derived co-cultures without requiring the isolation of primary tissues. This is particularly important for modeling astrocyte-related diseases such as Alexander disease, Alzheimer’s disease and Parkinson’s disease.

Conclusions

In summary, we have explored the possibility of developing an all-hPSC-derived tri-culture system for in vitro disease modeling and drug discovery, while highlighting the need for further optimization of astrocyte generation. Our work emphasizes the importance of incorporating inhibitory neurons into neural networks to achieve an appropriate E/I balance. We showed that this balance is critical for more disease-relevant phenotyping and drug screening, particularly when targeting GABAergic signaling. Our induced astrocytes support synaptogenesis and network formation but future efforts should focus on further refining astrocyte differentiation protocols to enhance the capabilities of such next-generation in vitro disease models.

Abbreviations

TF

Transcription factor

hPSCs

Human pluripotent stem cells

iNs

Induced neurons

iGlutNs

Induced glutamatergic neurons

iGABANs

Induced GABAergic neurons

iAstros

Induced astrocytes

MEA

Multielectrode array

CNS

Central nervous system

E/I

Excitation and inhibition

ASD

Autism spectrum disorder

hESCs

Human embryonic stem cells

RNP

Ribonucleoprotein

Dox

Doxycycline

PFA

Paraformaldehyde

hpAs

Human primary astrocytes

DIV

Days of in vitro differentiation

PCA

Principal component analysis

DGE

Differential gene expression

hiAs

Human induced astrocytes

SEM

Standard error of mean

mESCs

Mouse embryonic stem cells

hiPSCs

Human induced pluripotent stem cells

DSBs

Double-strand breaks

HDR

Homology-directed repair

PC1

Principal component 1

PC2

Principal component 2

DEGs

Differentially expressed genes

PV

Parvalbumin

SST

Somatostatin

NPY

Neuropeptide Y

CR

Calretinin

CB

Calbindin

RELN

Reelin

nNOS

Neuronal nitric oxide synthetase

VIP

Vasoactive intestinal peptide

CPM

Counts per million

GO

Gene ontology

KEGG

Kyoto encyclopedia of genes and genomes

BP

Biological process

CC

Cellular component

MF

Molecular function

NBD

Network burst duration

PSNB

Percentage of spikes in network bursts

NBR

Network burst rate

ISI

Interspike intervals

NBIBI

Network burst interburst intervals

CV

Coefficient of variation

PTX

Picrotoxin

BIC

Bicuculline

PAM

Positive allosteric modulator

CZP

Clonazepam

ASMs

Antiseizure medications

BZDs

Benzodiazepines

4-AP

4-Aminopyridine

FSK

Forskolin

Author contributions

JG, BR, BC and HS conceived the project and designed the study. BR, TJO, PK, BC and HS supervised the project. JG, BR, BC, and HS conducted the experiments. JG, BR, ZL, BC and HS analyzed the data. JG wrote the manuscript. BR, ZL, TJO, PK, BC and HS edited the manuscript.

Funding

This work is supported by Swinburne Tuition Fee Waiver and a Medical Research Future Fund Stem Cell Therapies Mission grant (MRF2015957).

Data availability

The RNA-seq dataset supporting the conclusions of this article can be found under Supplementary Material. Any other datasets supporting the conclusions of this article are available from the corresponding authors upon reasonable request.

Declarations

Conflict of interest

The authors declare that they have no competing interests.

Ethical approval and consent to participate

For the human ESC line, WiCell Research Institute, Inc. has confirmed that there was initial ethical approval for collection of human cells, and that the donors had signed informed consent (https://hpscreg.eu/cell-line/WAe009-A).

Consent for publication

Not applicable.