Mosaic human cortical organoids model mTOR-related focal cortical dysplasia through DEPDC5 deletion

Abstract

Focal cortical dysplasia type II (FCDII), a major cause of paediatric drug-resistant focal epilepsy, results from brain somatic variants in mTOR pathway genes, including germline and somatic second-hit loss-of-function variants in the mTOR repressor DEPDC5.

Here, we present a proof-of-concept model of DEPDC5 two-hit inactivation mosaicism using patient-derived human cortical organoids (hCOs). Mosaic hCOs displayed increased mTOR activity that was rescued by the mTOR inhibitor rapamycin. Mosaic hCOs also exhibited dysmorphic-like neurons and enhanced neuronal excitability, recapitulating key FCDII pathology hallmarks. Single-cell transcriptomics across three developmental stages revealed aberrant differentiation trajectories leading to premature upper-layer neuron generation, upregulated Notch and Wnt signalling pathways in neural progenitors, and altered expression of synaptic- and epilepsy-associated genes in excitatory neurons. In addition, we identified cell-autonomous alterations in metabolism and translation in mosaic DEPDC5 two-hit hCOs.

This study provides novel insights into how DEPDC5 deficiency perturbs human corticogenesis, highlighting that mosaic biallelic inactivation of the gene is necessary for FCDII pathogenesis.

Maletic et al. used patient-derived mosaic cortical organoids to model focal cortical dysplasia caused by two-hit DEPDC5 mutations. They show that biallelic DEPDC5 loss disrupts corticogenesis and increases neuronal excitability, providing mechanistic insight into how these mutations lead to drug-resistant epilepsy.

Introduction

Focal cortical dysplasia type II (FCDII) is a cortical malformation that leads to a spectrum of neurodevelopmental disorders collectively known as mTORopathies. FCDII-associated epilepsy is typically resistant to antiseizure medications, necessitating neurosurgical resection of the epileptogenic zone for seizure control, and is the most prevalent cortical malformation in paediatric epilepsy surgeries.¹ FCDII can vary in extent, ranging from small, distinct cortical areas to an entire hemisphere, as seen in hemimegalencephaly (HME). The neuropathological hallmarks of FCDII and HME include cortical dyslamination and the presence of dysmorphic neurons (FCDIIa type), and in some cases, balloon cells (FCDIIb type).² Dysmorphic neurons, characterized by an enlarged soma and an accumulation of neurofilaments, are thought to play a critical role in contributing to epileptic discharge generation, due to their hyperexcitability and presence within the epileptogenic focus.³

Recent studies have revealed the contribution of somatic mosaic variants to various neurodevelopmental disorders.^4,5 Notably, pathogenic somatic variants in genes of the mTOR pathway have been increasingly recognized as a major cause of FCDII.⁶ In FCDII, postzygotic somatic variants are thought to arise during corticogenesis in dorsal pallium progenitors lining the ventricular zone,⁷ but recent single-cell genomic and transcriptomic analyses reveal that mutations are present across multiple brain cell types, suggesting they may originate at earlier developmental stages.⁸ The level of mosaicism (percentage of cells in the tissue that carry the variant), reflected by the variant allele frequency (VAF), can vary widely, being as low as ∼1% in FCDII and up to 30% in HME.⁹

Deep sequencing of surgical FCDII brain tissue has uncovered somatic gain-of-function (GoF) heterozygous variants in genes coding for PI3K (phosphoinositide 3-kinase)-mTOR pathway upstream activators such as PIK3CA, AKT3 and MTOR. Furthermore, two-hit loss-of-function (LoF) variants—germline and somatic—have been identified in genes encoding the PI3K-mTOR pathway upstream repressors, such as DEPDC5 or TSC1/TSC2.^10-14 All these variants cause hyperactivation of the mTOR pathway, which in turn may dysregulate cell growth, differentiation, metabolism and function.

DEPDC5 is a key component of the GATOR1 (Gap Activity TOward Rags 1) complex that represses mTORC1 signalling in response to amino acids (leucine and arginine specifically).¹⁵  DEPDC5 mutations are major contributors to both lesional and non-lesional focal epilepsies.^16-18 A recent study identified DEPDC5 as the most significant gene in non-acquired focal epilepsy among 9219 cases.¹⁹ Moreover, DEPDC5 variants account for 13% of FCDII or HME cases.¹⁶ Notably, DEPDC5 variants are predominantly associated with FCDIIa, rather than FCDIIb, and do not generate balloon cells.²⁰ Several studies support Knudson’s two-hit mechanism in FCDII pathogenesis—where a somatic second-hit occurs in addition to the germline DEPDC5 variant in brain tissue—to explain the incomplete penetrance of the phenotype in DEPDC5 families and the focal and mosaic nature of FCDII lesions.^11,12,21-25 We and others have shown that these somatic variants are enriched in dysmorphic neurons.^11,24 However, key questions remain regarding the relative contributions of heterozygous versus biallelic inactivation to epileptogenesis and FCDII pathogenesis.

In this study, we examined how mosaic DEPDC5 biallelic inactivation affects cortical development and contributes to epileptogenesis in FCDII. For this purpose, we generated patient-derived heterozygous (Het) and mosaic (Mos)—heterozygous/two-hit—human cortical organoids (hCOs) and employed longitudinal single-cell RNA sequencing (scRNA-seq) across three developmental stages and electrophysiological recording. Our findings reveal that Het and Mos hCOs display variable neurodevelopmental, transcriptomic and functional alterations. Notably, however, only Mos hCOs recapitulated hallmark features of FCDII, including cytomegalic mTOR-hyperactive neurons and neuronal network hyperactivity.

Materials and methods

Generation and characterization of human induced pluripotent stem cell lines

Patient-derived human induced pluripotent stem cells (hiPSCs) were generated from peripheral blood mononuclear cells of two male subjects previously reported²⁶ using viral-free episomal reprogramming (Phenocell). The study received ethical approval (Inserm No. C1456 GENEPI) with informed consent. The resulting hiPSC lines met standard reprogramming criteria, including negative serology and mycoplasma testing, pluripotency marker expression and normal karyotype (Supplementary Fig. 1A).

Gene editing

A DEPDC5^−/− line was generated by introducing the p.L1420Ffs*154 mutation into the second allele of patient DEPDC5^+/− hiPSCs using CRISPR-Cas9-mediated homologous repair (Supplementary Fig. 1B). A CAG-EGFP-pA construct was inserted at the hTIGRE locus by CRISPR-Cas9, with reagents delivered via electroporation (Supplementary Fig. 1C). CAG-EGFP-pA is a construct where the CAG promoter (CMV enhancer, β-actin promoter, β-globin elements) drives enhanced GFP (EGFP) expression, followed by a polyadenylation signal for proper mRNA processing. Two independent isogenic DEPDC5^−/− clones (DEPDC5^−/−1 and DEPDC5^−/−2) were validated based on EGFP expression and normal karyotype (Supplementary Fig. 1D) and were used interchangeably throughout the study. An isogenic control line was also generated by correcting the mutated allele with the corrected sequence. All hiPSC lines are deposited in the hPSCreg^® database (https://hpscreg.eu/).

Human induced pluripotent stem cell culture

Human iPSC lines were maintained in mTeSR^™1 medium on laminin-521- or Vitronectin XF^™-coated dishes at 37°C with 5% CO₂. Cells were passaged at ∼80% confluency using either Accutase or ReLeSR (Stem Cell Technologies) and were tested monthly for mycoplasma contamination.

Human cortical organoid generation

Human induced PSCs were differentiated into hCOs following a directed protocol towards the dorsal forebrain cortical fate,²⁷ with adaptations: dimethyl sulfoxide (DMSO) omitted at Day −2, mTeSR1 medium used instead of Essential 8 and NeuroCult1 supplement (vitamin A-free) instead of B27 (Supplementary Fig. 1E). To initiate differentiation, hiPSCs were dissociated into single-cell suspensions using Accutase. Approximately 4 000 000 cells were seeded in 24-well AggreWell 800 plates in mTeSR1 medium supplemented with 10 µM Rock inhibitor at 5% CO₂ and 37°C for 24 h to form embryoid bodies. For mosaic organoids, EGFP+ DEPDC5^−/− and EGFP– DEPDC5^+/− cells were mixed at defined ratios prior to aggregation. Two levels of mosaicism were generated: Mos High, in which approximately 50% of the cells were DEPDC5^−/−, and Mos Low, in which 15%–20% of the cells were DEPDC5^−/−. Embryoid bodies were transferred to TeSR-E6 medium containing dorsomorphin (2.5 µM), SB-431542 (10 µM) and XAV-939 (2.5 µM). Neural induction used Neurobasal^™-A (Life Technologies) supplemented with FGF2/EGF (20 ng/ml, Days 6–24), then BDNF/NT-3 (20 ng/ml, Days 25–42). From Day 43, hCOs were maintained in Neurobasal-A with penicillin-streptomycin (100 U/ml). Medium was changed daily until Day 14, then every other day, with orbital shaking from Day 30.

For amino acid deprivation, 6-month hCOs were incubated for 4 h in leucine/arginine-deficient stable isotope labelling with amino acids in cell culture (SILAC) Neurobasal medium (Athena Enzyme Solutions) before immediate harvest.

For mTOR inhibition, hCOs received 20 nM rapamycin or vehicle (ethanol) in Neurobasal medium from Day 25 until analysis, as previously described.²⁸

Western blotting

Samples were lysed with protease and phosphatase inhibitors, centrifuged at 14 000 rpm and supernatants stored at −80°C. Protein concentrations were determined by the bicinchoninic acid (BCA) protein assay. Equal protein amounts were separated onto 4%–12% NuPAGE^™ gels, transferred to membranes and blocked in PBS with 0.1% Tween-20 and 3% bovine serum albumin (BSA), before overnight primary antibody incubation at 4°C, followed by horseradish peroxidase (HRP)-coupled secondary antibody. Protein bands were visualized with enhanced chemiluminescence (ECL) peroxidase substrates, quantified with FIJI software and normalized to β-actin. Four to five hCOs were pooled per genotype and per experiment. The number of technical replicates is specified in the figure legends for each result. Statistical analyses for imaging were performed with GraphPad Prism 9 software (v.9.4.1). Specific tests and sample sizes are detailed in the figure legends. The selection of hCOs for amino acid deprivation treatment was randomized. All samples were included in the analysis to ensure comprehensive and unbiased results. Significance thresholds: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Immunostaining

Human COs were fixed in 4% paraformaldehyde (PFA), cryopreserved in 30% sucrose, embedded in optimal cutting temperature compound (OCT) and frozen on dry ice; 20 µm sections were blocked (PBS, 0.4% Triton X-100, 10% goat serum), incubated with primary antibodies overnight, then with secondary antibodies and 4',6-diamidino-2-phenylindole (DAPI) for 2 h. Slides were mounted and imaged on Zeiss Apotome or Nikon confocal microscopes.

For 3D immunostaining, whole fixed hCOs underwent serial detergent permeabilization steps, blocking and prolonged antibody incubations with continuous shaking. Human COs were clarified with RapiClear before confocal imaging.

The number of hCOs and replicates are specified in the figure legends for each result. Statistical analyses for imaging were performed with GraphPad Prism 9 software (v. 9.4.1). Specific tests and sample sizes are detailed in the figure legends. The selection of hCOs for amino acid deprivation and rapamycin treatment was randomized. All samples were included in the analysis to ensure comprehensive and unbiased results. Significance thresholds: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Antibodies

Primary antibodies used included: β-Actin (Cell Signaling Technology, Cat. No. 4970; Sigma-Aldrich, Cat. No. A2066), DEPDC5 (Abcam, Cat. No. ab213181, ab185565), DLK1 (Abcam, Cat. No. ab21682), BLBP (Abcam, Cat. No. ab32423), GFP (Aves Labs, Cat. No. GFP-1020; Invitrogen, Cat. No. A-6455), NeuN (Millipore, Cat. No. MAB377, ABN78), phospho-S6 ribosomal protein (Ser240/244) (Cell Signaling Technology, Cat. No. 5364), SATB2 (Abcam, Cat. No. ab51502), neurofilament SMI311 (BioLegend, Cat. No. BLE837801), SOX2 (Thermo Fisher Scientific, Cat. No. 14-9811-82; Millipore, Cat. No. AB5603), S6 (Cell Signaling Technology, Cat. No. 2317S) and ZO-1 (Thermo Fisher Scientific, Cat. No. 33-9100; Cell Signaling Technology, Cat. No. 13663S). Secondary antibody: anti-rabbit immunoglobulin G HRP-linked (Cell Signaling Technology, Cat. No. 7074).

Multielectrode array recordings

Spontaneous extracellular recordings in 6-month whole hCOs were acquired using a 64-channel MEA system (MED64, Alpha MED Scientific). Human COs were plated on poly-L-ornithine and laminin-coated probes and recordings were performed in neurobasal medium at 37°C, 48 h post-medium change. Recordings started after a 2-min recovery period and were acquired over a time period of 3 min, as previously reported.^29,30 Extracellular field potentials were filtered (100 Hz–10 kHz band-pass filter) and sampled at 20 kHz. Spike detection was performed using MOBIUS software with a detection threshold set at 6× the standard deviation of noise followed by manual validation of events with a custom data visualization script (https://gitlab.com/icm-institute/dac/biostats/MEASpikeR).³¹ Electrodes with ≥5 spikes/min were considered active and included in the analysis. Firing rate was calculated as the total spike number per electrode over 3 min.

Statistical analyses were performed with GraphPad Prism 9 software (v.9.4.1) and data are shown as violin plots with the median and the 25th and the 75th percentiles. Statistical significance was analysed using the Kruskal–Wallis test followed by uncorrected Dunn’s multiple comparisons test. Symbols of statistical significance were used as follows: *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

Single-cell RNA sequencing

Human COs were prepared for scRNA-seq following a previously reported protocol.³² Five hCOs were harvested per library with the following batches per condition: control (CT) (1 month n = 1, 3 months n = 1, 6 months n = 2); Het (1 month n = 2, 3 months n = 1, 6 months n = 1); Mos (1 month n = 2, 3 months n= 1, 6 months n = 2). All experiments were performed using Mos High. Dissociated cells were resuspended in ice-cold solution (0.1% BSA and 25 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) in Dulbecco’s phosphate buffered saline (DPBS)), incubated with viability dye eFluor 660 (1:1000) for 30 min, filtered (50 µm) and sorted (BioRad S3e). Viability gating used eFluor660 dye, EGFP gating used the intrinsic EGFP signal. Viable cells (≥80%) were collected in BSA-coated tubes, counted and processed using the 10× Genomics Chromium Next GEM Single Cell 3′ Kit v. 3.1. Approximately 20 000 cells were loaded for gel bead-in emulsion (GEM) generation, barcoding and cDNA amplification to retrieve ∼10 000 cells per library. Paired-end sequencing was performed on an Illumina NovaSeq6000 targeting ∼50 000 read-pairs per cell.

Illumina binary base calls (BCLs) were demultiplexed with cellranger mkfastq into FASTQ files. Alignment to hg38 customized with the EGFP gene sequence, and barcode and unique molecular identifier (UMI) counting for generation of feature/barcode count matrices were done with cellranger count. Filtered count matrices were used using Seurat (v.4.1.0).³³ Further filters were applied to keep: (i) genes expressed in ≥5 cells; (ii) cells expressing ≥200 genes; (iii) cells with <8% mitochondrial RNA; (iv) cells with <70% ribosomal RNA; and (v) cells with 500–20 000 UMIs. Additional filters included the removal of all cells found before the local minimum based on the function (f): nFeature -> density(nFeature). DoubletFinder³⁴ removed doublets (<10% per sample). Since we did not observe significant batch effects on our data, integration was unnecessary; libraries were normalized using NormalizeData with the LogNormalize method, and by regressing out mitochondrial percentage and UMI count. Variable features were identified using the FindVariableFeatures function. Data were further corrected for cell cycle, scaled with ScaleData and dimensional reduction used RunPCA and RunUMAP.

Cell type annotation

Clusters were annotated in CT hCO data using canonical cortical cell markers^35,36 (Supplementary Fig. 2A). To verify cluster annotation, label transfer using a reference atlas was performed using SingleR.³⁷ For label transfer, we used temporally matched reference data: 23 days to 1.5 months reference for 1-month hCOs, 2–4 months reference for 3-month hCOs and 4–6 months reference for 6-month hCOs. Het and Mos hCO clusters were subsequently annotated by label transfer using Seurat FindTransferAnchors followed by TransferData using CT hCOs data as reference (Supplementary Fig. 2B), and further refining the annotation based on canonical cell type marker expression.

RNA velocity and trajectory analysis

Each genotype (pooled time points) was analysed with scVelo’s dynamic model³⁸ to calculate RNA velocities, latent time and driver genes.

Differential gene expression analysis

Differential gene expression used Seurat FindMarkers (Wilcoxon Rank Sum test) with a minimum 1% cell percentage and a log fold change threshold at 0.1. Only genes with adjusted P-value <0.05 (Bonferroni correction) were retained.

Gene ontology analyses

Gene ontology (GO) biological process enrichment analysis was performed with ClusterProfiler’s enrichGO (v. 4.2.2),³⁹ with the following parameters: (i) P-value <0.01; (ii) q-value <0.1; (iii) gene size at 3–1000; and (iv) adjusted P-values calculated with the Benjamini and Hochberg correction. We set as background (universe) all genes expressed across hCOs genotypes and time points. We used synaptic gene ontology (SynGO) with the differentially expressed gene (DEG) lists obtained for each comparison as input.

Results

DEPDC5 mosaic hCOs display mTOR-hyperactive dysmorphic-like neurons

To investigate how DEPDC5 mutation alters early cortical development and model FCDII pathogenesis, we derived hiPSCs from two brothers: a patient with focal epilepsy and FCDII carrying a heterozygous LoF DEPDC5 variant (p.L1420Ffs*154), and his unaffected non-carrier sibling who serves as an age- and sex-matched control sharing 50% of the genetic background (Fig. 1A and Supplementary Fig. 1A). Using CRISPR/Cas9 genome editing, we engineered an isogenic EGFP-tagged DEPDC5 two-hit (homozygous) mutant line by introducing the same variant into the wild-type allele of the patient’s heterozygous hiPSCs (Supplementary Fig. 1B and C and see the ‘Materials and methods’ section). An isogenic control hiPSC line was also generated by CRISPR-Cas9 (DEPDC5^+/+) but was withdrawn from most experiments since its quality was questionable as we repeatedly observed a tendency to spontaneously differentiate. Therefore, we used three distinct hiPSC lines in this study: (i) a DEPDC5^+/− (EGFP−) line from the patient; (ii) a control DEPDC5^+/+ line from the asymptomatic brother; and (iii) an isogenic DEPDC5^−/− (EGFP+) line. To validate the efficiency of gene editing, we quantified the levels of DEPDC5 and EGFP proteins by western blotting. We confirmed the absence of the DEPDC5 protein in the DEPDC5^−/− line, and half the amount of DEPDC5 in DEPDC5^+/− compared with DEPDC5^+/+ (Supplementary Fig. 1D).

Figure 1.

DEPDC5 mosaic human cortical organoids display pS6+ dysmorphic-like neurons. (A) Left: Family pedigree showing inherited DEPDC5 loss-of-function germline variant. HPSCs were derived from the proband and unaffected sibling. Right: Schematic overview of the different hiPSC lines used to produce each organoid type. Bottom: Endogenous EGFP signal in clarified 1-month Mos hCOs. Scale bar = 300 µm. Created in BioRender. Noe, E. (2026) https://BioRender.com/4yr0tbk. (B) Schematic of mTOR pathway activation (measured by S6 phosphorylation) under amino acid deprivation (AA–) with and without functional DEPDC5. (C) Western blot showing DEPDC5, pS6 and actin protein levels in 6-month hCOs under basal (AA+) and amino acid-deprived (AA–) conditions. Quantification normalized to actin (five batches, five hCOs/genotype). Mean pS6 expression (au): CT AA+ = 1.11; CT AA− = 0.28; Het AA+ = 1; Het AA– = 0.12; Mos High AA+ = 1.27; Mos High AA− = 0.82. Two-way ANOVA with Šidák’s correction. *P < 0.05, **P < 0.01. (D) pS6 immunofluorescence quantification in 3-month hCOs treated with vehicle or rapamycin (n = 4–5 hCOs/condition, 1 section/hCO, 1 batch). Mean pS6 intensity (au): CT Vehicle = 7.82; CT Rapamycin = 5.09; Het Vehicle = 11.54; Het rapamycin = 5.38; Mos High Vehicle = 25.04; Mos High rapamycin = 11.81. Two-way ANOVA with Šidák’s correction. Scale bar = 1 mm. **P < 0.01, ****P < 0.0001. (E) pS6 levels in EGFP− and EGFP+ NeuN+ neurons in 6-month Mos Low hCOs under AA+ and AA− conditions. Arrows: EGFP+/pS6+ neurons. Scale bar = 20 µm. Number of cells analysed: EGFP+/AA+ = 33, EGFP+/AA− = 17, EGFP−/AA+ = 573, EGFP−/AA− = 535 from two batches (n = 14 AA+ and 11 AA− Mos Low hCOs). Mean pS6 intensity (au): EGFP+/AA+ = 8.4; EGFP+/AA− = 7.8; EGFP−/AA+ = 4.39; EGFP−/AA− = 3.73. Kruskal–Wallis test with Dunn’s correction. Median differences between groups were estimated using bootstrap resampling (10 000 iterations) and 95% bootstrap CIs: EGFP+/AA+ versus EGFP+/AA− = 89.05 (95% CI: −687.6 to 275.9); EGFP−/AA+ versus EGFP−/AA− = 37.23 (95% CI: 15.5 to 57.2); EGFP+/AA+ versus EGFP−/AA+ = 188.32 (95% CI: 43.1 to 316.2). ***P < 0.001, ****P < 0.0001. (F) Quantification of soma size and SMI311 intensity (normalized to soma area) in EGFP+ versus EGFP− NeuN+ neurons in 6-month Mos Low hCOs. Scale bar = 100 µm. Number of cells analysed from n = 4 hCOs: EGFP+ = 1437; EGFP− = 6045. Mean soma size (µm²): EGFP− = 78.59, EGFP+ = 99.24. Mean SMI311 intensity (au): EGFP− = 27.95, EGFP+ = 49.76. Mann–Whitney test. Median differences between groups were estimated using bootstrap resampling (10 000 iterations) and 95% bootstrap CIs: soma size (EGFP− versus EGFP+) = −15.21 (95% CI: −20.3 to −11.8); SMI311 intensity (EGFP− versus EGFP+) = −20.08 (95% CI: −21.4 to −18.7). ****P < 0.0001. au = arbitrary unit; CI = confidence interval; CT = control; EGFP = enhanced GFP; hCO = human cortical organoids; Het = heterozygous; hiPSC = human induced pluripotent stem cell; Mos = mosaic; ns = non-significant; pS6 = phosphorylated ribosomal protein S6.

The hiPSC lines were differentiated into dorsally patterned cortical organoids following an established protocol.²⁷ We generated three types of hCOs: control (CT, DEPDC5^+/+ cells), heterozygous (Het, DEPDC5^+/− EGFP− cells) and mosaic (Mos), the latter containing both DEPDC5^+/− EGFP− and DEPDC5^−/− EGFP+ cells, with either low (Mos Low, 15%–25% EGFP+ cells at the time of embryoid body assembly) or high (Mos High, ∼50% EGFP+ cells) levels of mosaicism (Fig. 1A, Supplementary Fig. 1E and Supplementary material ‘Results and methods’ section). For simplicity, these three conditions (CT, Het and Mos) are referred to as ‘genotypes’ throughout, although Mos hCOs contain cells of two distinct genotypes. Mos High hCOs were used for most experiments in the study, unless stated otherwise. Clarification and imaging of 1-month Mos hCOs, as well as flow cytometry analysis, confirmed the presence of EGFP+ cells in the expected proportions, with no evidence of positive or negative selection for the EGFP+ population (Fig. 1A and Supplementary Fig. 1F). By Day 14, Mos hCOs were slightly larger than CT, while isogenic control (Iso, DEPDC5^+/+ cells) and Het remained similar in size (Supplementary Fig. 1G). However, Iso hCOs degraded rapidly, with their numbers decreasing significantly. This, along with the Iso hiPSC line’s tendency to differentiate spontaneously, indicated poor quality of the cell line, leading to its exclusion from downstream experiments. All other hCO genotypes (CT, Het and Mos) maintained structural integrity and characteristic neural rosette formation throughout 6 months of differentiation, indicating that DEPDC5 deficiency does not significantly impair cortical organoid development.

DEPDC5, within the GATOR1 complex, is a key repressor of mTOR activity in response to amino acid availability, especially leucine and arginine. In the absence of DEPDC5, mTOR signalling is predicted to remain constitutively active during amino acid deprivation (Fig. 1B). To assess mTORC1 activity in mature 6-month hCOs, we stimulated the GATOR1 branch of the pathway by using a modified neurobasal medium deprived of leucine and arginine. Western blot analysis showed reduced phosphorylated S6 (pS6), a downstream effector used as an mTORC1 activity readout, in both CT and Het hCOs under amino acid deprivation (AA–). In contrast, Mos hCOs maintained pS6 levels comparable to baseline (AA+), indicating that GATOR1 failed to inhibit mTORC1 activity (Fig. 1C and Supplementary Fig. 1H). Phosphorylated S6 immunofluorescence analysis confirmed constitutive mTOR activation in Mos and, to a lesser extent, Het hCOs (Supplementary Fig. 1I). Administration of the mTORC1 inhibitor rapamycin significantly reduced pS6 levels in both Het and Mos hCOs, confirming that the constant pS6 levels resulted from impaired GATOR1-mediated mTORC1 inhibition (Fig. 1D).

To determine the cell-autonomous effects of DEPDC5 loss, we analysed pS6 immunofluorescence in NeuN-positive neurons from Low Mos hCOs, where sparse EGFP labelling enables clear distinction between DEPDC5^+/− (EGFP−) and DEPDC5^−/− (EGFP+) cells. EGFP+ neurons exhibited significantly higher basal pS6 levels compared with neighbouring EGFP− neurons within the same Mos hCO and maintained these levels during amino acid deprivation, indicating constitutive mTORC1 activation. In contrast, EGFP− neurons showed decreased pS6 levels under amino acid-deprived conditions (Fig. 1E).

We searched for the presence of FCDII hallmark dysmorphic neurons by performing co-immunostaining of EGFP, NeuN and SMI311, a canonical neurofilament marker of dysmorphic neurons, in mature 6-month hCOs. EGFP+ (DEPDC5^−/−) neurons in Mos hCOs exhibited enlarged soma size compared with EGFP− (DEPDC5^+/−) neurons (26% increase; mean EGFP− = 78.59 μm² ± 51.67, mean EGFP+ = 99.24 μm² ± 64.47) and SMI311 accumulation (Fig. 1F). Notably, no balloon-like cells were detected by vimentin staining, consistent with DEPDC5-related FCDII pathology, which typically presents with dysmorphic neurons but lacks balloon cells.^9,20

These findings indicate that biallelic DEPDC5 inactivation in hCOs leads to constitutive mTOR pathway activation, and is necessary to reproduce dysmorphic-like neurons, the main FCDII hallmark.

DEPDC5 loss-of-function leads to premature generation of upper-layer neurons

To investigate how DEPDC5 deficiency affects dorsal forebrain development and cell fate acquisition, we performed scRNA-seq of CT, Het and Mos hCOs at three key developmental stages: (i) progenitor expansion and early neurogenesis (1 month); (ii) early-to-mid corticogenesis, when diverse cortical cell types start emerging (3 months); and (iii) later corticogenesis, with mature neuron specification (6 months). Following quality filtering, we analysed a total of 73 685 cells for all three stages (28 637 CT; 14 972 Het; 30 076 Mos). Cell clusters were annotated in CT hCOs at each time point based on the expression of canonical brain cell type marker genes (Supplementary Fig. 2A). Subsequently, we annotated Het and Mos hCOs using a semi-unsupervised approach that consisted of label transfer from CT, followed by verification of marker gene expression (Supplementary Fig. 2B). In CT hCOs, we identified the expected cell types at each time point compared with similar studies.^27,36,40 At 1 month, CT hCOs were made primarily of progenitor cells, including FOXG1-negative progenitors and FOXG1-positive apical radial glia (aRG) expressing HES1. Furthermore, at 1 month, we detected FOXG1-negative neurons expressing STMN2, GRIA2 and SLC17A6. In smaller proportions, we observed unspecified FOXG1-positive neurons (expressing STMN2 and GRIA2), preplate cells (expressing EMX2, FOXP2 and LHX9), and Cajal-Retzius-like cells (expressing LHX5 and RELN). At 3 months, we observed the appearance of EOMES-expressing intermediate progenitors (IP) and a proportion of lower-layer excitatory neurons (LL.Exc.N) expressing NEUROD6, FEZF2 and PDE1A. By 6 months, in addition to aRG and IP, we detected outer radial glia (oRG) expressing TNC, HOPX, FAM107A and MOXD1. Concurrently, upper-layer excitatory neurons (UL.Exc.N) emerged, expressing BHLHE22, PLXNA4, CUX2 and SATB2. At this time point, we also detected interneuron progenitors (IN.progenitors) expressing BIRC5, GAD1/2, DLX1/2/5 and SP8, and inhibitory neurons (Inh.N) expressing GAD1/2, DLX1/2/5, SP8, SCGN, PROX1, CALB2 and SLC32A1 (Fig. 2A and Supplementary Fig. 2A).

Figure 2.

DEPDC5 deficiency alters differentiation trajectories. (A) Single-cell transcriptomic analysis of cortical organoids. Top: UMAP visualization of cell clusters from CT, Het and Mos hCOs at 1, 3 and 6 months. Bottom: Quantification of cell type proportions across developmental time points. (B) SATB2 and EGFP immunofluorescence in 3-month whole hCOs. SATB2+ particle density quantified relative to section area (μm²). Scale bar = 500 µm. n = 4 hCOs/genotype (one section/hCO, one batch). Mean SATB2+ staining (au): CT = 6.08, Het = 14.28, Mos High = 6.26. One-way ANOVA with Tukey’s correction. *P < 0.05. (C) RNA velocity analysis across development. Top: UMAP visualization colour-coded by scVelo latent time. Middle: Cell type distribution along latent time for each genotype. Bottom: Heat maps showing expression dynamics of top 300 likelihood-ranked genes resolved along latent time. Coloured stars indicate UL.Exc.N markers; other cell type marker genes are also annotated. CT = control; EGFP = enhanced GFP; hCO = human cortical organoids; Het = heterozygous; Mos = mosaic; ns = non-significant; UL.Exc.N = upper-layer excitatory neuron; UMAP = Uniform Manifold Approximation and Projection.

Het and Mos hCOs exhibited distinct cell type compositions compared with CT hCOs. This was observed as early as 1 month and persisted across all developmental stages. At 1 month, all clusters in Het and Mos hCOs expressed FOXG1 confirming telencephalic fate commitment, with Het already displaying a cluster expressing oRG markers. By 3 months, Mos and Het hCOs showed higher proportions of UL.Exc.N (18% and 60%, respectively), which were almost absent (<1%) in CT hCOs (Fig. 2A). This was confirmed by SATB2 immunostaining, a marker of upper-layer neurons (Fig. 2B). At 6 months, Het and Mos hCOs still exhibited higher proportions of UL.Exc.N compared with CT and a near absence of IN.progenitors and Inh.N (Fig. 2A).

To further explore differences in cell type differentiation across the three genotypes, we performed RNA velocity analysis using scVelo. This analysis revealed distinct differentiation trajectories along latent time, reflected by genotype-specific transcriptional cascades defined by the expression dynamics of putative driver genes, including several neuronal subtype marker genes. In Het and Mos hCOs, driver gene expression patterns showed a more abrupt and possibly premature transition from progenitors to differentiated cell types, possibly underlying the earlier production of UL.Exc.Ns. Consistent with this, Het and Mos hCOs displayed a higher representation of UL.Exc.N marker genes among the top drivers. Whereas in CT hCOs only the UL.Exc.N marker PLXNA4 was identified as a trajectory driver, Het additionally displayed PLXNA4, and Mos uniquely featured EPHA4, CUX2 and SATB2 (the last two in Mos only; Fig. 2C), suggesting a preferential and premature expression of the UL.Exc.N transcriptional programme. Thus, our data show that DEPDC5 loss causes aberrant cell type differentiation trajectories irrespective of whether a mosaic second hit is also present.

Dysregulation of Notch and Wnt pathways in neural progenitors

To investigate whether altered cell type proportions in Het and Mos hCOs originate from progenitor structural defects, we analysed the number and structure of neural rosettes—radially organized neuroepithelial structures that recapitulate the developing cortical ventricular zone. Quantification of neural rosettes in 1-month hCOs, identified by immunostaining for the progenitor marker SOX2, revealed significantly reduced rosette density in both Het (144 ± 72 rosettes/mm²) and Mos (118 ± 76 rosettes/mm²) compared with CT hCOs (350 ± 105 rosettes/mm²) (Fig. 3A). This reduction persisted in 3-month Mos hCOs (CT: 87 ± 53, Het: 48 ± 39, Mos: 28 ± 43 rosettes/mm²) and was rescued by rapamycin treatment (Supplementary Fig. 3A and B). Notably, rapamycin also increased rosette density in CT hCOs at 3 months (Supplementary Fig. 3B), indicating that mTOR activity regulates neural progenitor self-renewal. Mos High hCOs showed a tendency towards lower density of rosettes compared with Mos Low hCOs at 1 month (Supplementary Fig. 3C), suggesting that the proportion of DEPDC5^−/− cells, and therefore the dosage of DEPDC5, plays a role in rosette formation and/or maintenance. Co-immunostaining for SOX2 and the tight-junction marker ZO1, which labels the ventricular lining, revealed no significant structural abnormalities in Het and Mos rosettes (Fig. 3B). Furthermore, rosette ventricular size was comparable across all genotypes (Supplementary Fig. 3D and E). These data indicate that DEPDC5 LoF and mTOR hyperactivity do not cause gross progenitor structural defects, and that the lower density of rosettes observed in Het and Mos may be a consequence rather than a cause of the accelerated differentiation revealed by scRNA-seq.

Figure 3.

DEPDC5 deficiency alters developmental signalling pathways. (A) SOX2 and EGFP immunofluorescence in 1-month whole hCOs (confocal images). Scale bar = 500 µm. Rosette density quantified as the number of rosettes per mm². n = 3 CT, 10 Het, 8 Mos hCOs (three batches). Mean number of rosettes per mm²: CT = 349.7; Het = 144.3; Mos High = 118.4. One-way ANOVA with Tukey’s correction. **P < 0.01. (B) Confocal z-stack analysis of neural rosettes immunostained for SOX2, ZO-1 and EGFP at 1 month. Top: ZO-1 ring plane; Bottom: ZO-1 surface plane. Scale bar = 50 µm. (C) Differential expression of cell type marker genes in apical radial glia (aRG) at 1 month. Differentially expressed genes (Wilcoxon Rank Sum test with Bonferroni correction) between Het versus CT, Mos versus CT and Mos versus Het are reported in separate dot plots. For each dot plot, the three conditions are reported, with the dismissed condition shown in grey. CT = control; EGFP = enhanced GFP; hCO = human cortical organoids; Het = heterozygous; Mos = mosaic; ns = non-significant

We performed differential gene expression analysis comparing Het and Mos aRG with CT at 1 month. DEGs included several markers of both immature and mature cortical cell types (Fig. 3C and Supplementary Table 1), suggesting that cell type specification changes may start at the progenitor stage. Notably, both Het and Mos hCOs showed dysregulation of neuronal marker genes, and downregulation of aRG markers (PAX6, EMX1 and HES1) as well as FEZF2, a key determinant of lower-layer subcortical projection neuron specification. DEGs between Mos and Het included progenitor markers such as BIRC5, MOXD1, SOX5 and FABP7 (Fig. 3C).

We focused on DEGs with an absolute average log₂ fold change (avg_Log2FC) > 0.5 and an adjusted P-value < 0.01 (Fig. 4A). Among the top upregulated genes in both Het and Mos, we identified DLK1, HES5, FABP7 and CXXC4, which are associated with Notch and Wnt signalling pathways (Fig. 4A and B). Only HES5 and FABP7 were differentially expressed between Mos and Het aRG; however, the avg_Log2FC were lower than 0.5 (0.22 and 0.15, respectively; Fig. 4B). Immunostaining against DLK1 and FABP7 confirmed their increased expression in 1-month Mos hCOs, but not Het (Fig. 4C). Comparison of EGFP+ (DEPDC5^−/−) and EGFP− (DEPDC5^+/−) populations in Mos showed no significant differential expression of DLK1, HES5, FABP7 or CXXC4. When comparing the two populations with CT and Het hCOs, only FABP7 was differentially expressed in the EGFP+ population but not in the EGFP− population compared with Het. The other three genes showed the same pattern in the two Mos populations in comparison to CT and Het (Supplementary Fig. 3F and Supplementary Table 1). Notch and Wnt signalling pathways regulate cortical progenitor pool size, self-renewal versus differentiation balance and neuronal laminar fate specification.^41,42 DLK1 promotes cell cycle exit in both mouse and human embryonic stem cell (ESC)-derived neural progenitors.⁴³ HES5 is a key effector of Notch signalling that regulates the transition timing of neurogenesis, with Hes5 overexpression in mouse cortical progenitors accelerating the transition from deep to superficial layer neurogenesis.⁴⁴ Thus, our data indicate that DEPDC5 loss in hCOs causes alterations in the cortical progenitor transcriptional programme that may underlie cell type specification differences observed later on in more mature hCOs. Upregulation of key Notch and Wnt signalling genes in particular may explain the premature differentiation of cortical progenitors and production of upper-layer excitatory neurons at a stage when only lower-layer neurons are generated in CT hCOs.

Figure 4.

Dysregulation of Notch and Wnt pathway genes in DEPDC5 mutant human cortical organoids. (A) Up- and downregulated genes (Wilcoxon Rank Sum test with Bonferroni correction) in Het and Mos 1-month aRG compared with CT. Significance threshold was established as average log₂ fold change > 0.5, adjusted P < 0.01 (based on Bonferroni correction using all genes in the dataset). Sig = significant. (B) Expression levels of Notch (DLK1, HES5, FABP7) and Wnt (CXXC4) pathway genes in 1-month aRG across genotypes. Significance values were obtained with a Wilcoxon Rank Sum test with Bonferroni correction. (C) DLK1 and FABP7 immunofluorescence intensity quantified relative to SOX2+ rosette area (confocal images). Scale bar = 30 µm. n = 5 CT, 4 Het, 4 Mos High hCOs (one batch). DLK1: Kruskal–Wallis with uncorrected Dunn’s correction. Mean DLK1 intensity per rosette (au): CT = 28.71; Het = 29.9; Mos High = 34.22. **P < 0.01. FABP7: one-way ANOVA with Tukey’s correction. Mean FABP7 intensity per rosette (au): CT = 56.92; Het = 66.64; Mos High = 70.81. **P < 0.01. aRG = apical radial glia; au = arbitrary unit; CT = control; Het = heterozygous; Mos = mosaic; NS/ns = non-significant; pS6 = phosphorylated ribosomal protein S6.

DEPDC5 Mos two-hit hCOs exhibit altered synaptic gene expression and increased spontaneous neuronal spike activity

We then performed DEG analysis at more mature stages of development focusing on neuronal cells to identify dysregulated processes linked to FCDII pathophysiology. Given that our hCO model primarily recapitulates dorsal telencephalic development and exhibits limited interneuron specification, we restricted our analysis to early-born LL.Exc.Ns and late-born UL.Exc.N at time points corresponding to their peak abundance in CT hCOs, meaning 3 and 6 months, respectively (Supplementary Tables 2 and 3). LL.Exc.N exhibited a higher ratio of DEGs over expressed genes compared with UL.Exc.N in both Het and Mos organoids, suggesting increased vulnerability of early-born neurons to DEPDC5 loss (Fig. 5A). GO enrichment analysis of DEGs for each excitatory neuron subtype revealed distinct enrichment patterns between genotypes. In Het LL.Exc.N, RNA processing and translation were predominantly affected (among the top 10 most enriched terms) while Mos LL.Exc.N showed enrichment of terms related to neuron projection and neuronal morphogenesis (Fig. 5B and Supplementary Table 4). RNA processing terms were also enriched when comparing Het to Mos LL.Exc.N (Supplementary Fig. 4A and Supplementary Table 4), indicating a distinguishing feature between Het and Mos. Both Het and Mos UL.Exc.N also displayed altered expression of genes involved in neuron projection and neuronal morphogenesis compared with CT (Fig. 5B and Supplementary Table 5) but also when comparing Het and Mos (Supplementary Fig. 4A and Supplementary Table 5). Genes associated with neuronal migration were specifically dysregulated in Mos UL.Exc.N (Fig. 5B). These transcriptional alterations align with key histopathological features of FCDII, particularly defects in neuronal morphogenesis. The dysregulation of RNA metabolism and protein synthesis machinery in DEPDC5 mosaic hCOs likely reflects downstream effects or compensatory mechanisms of activation of mTORC1 signalling.

Figure 5.

DEPDC5 deficiency alters neuronal gene expression. (A) Proportion of differentially expressed genes (DEGs) in lower-layer excitatory neurons (LL.Exc.Ns, 3 months) and upper-layer excitatory neurons (UL.Exc.Ns, 6 months). (B) Top 10 enriched gene ontology (GO) biological process (hypergeometric distribution with Benjamini and Hochberg correction) terms enriched in DEGs from LL.Exc.Ns and UL.Exc.Ns comparing Het and Mos hCOs to CT. (C) Proportion of synaptic genes (SynGO annotated) among DEGs in LL.Exc.Ns (3 months) and UL.Exc.Ns (6 months) across genotype comparisons. CT = control; hCOs = human cortical organoids; Het = heterozygous; Mos = mosaic; SynGo = synaptic gene ontology.

Given that DEPDC5 is a major epilepsy-causing gene, we next investigated potential molecular mechanisms underlying neuronal hyperexcitability and seizures in FCDII. We examined differential expression of synaptic, epilepsy-related and ion channel genes. We interrogated the SynGO database,⁴⁵ which catalogues 1112 annotated synaptic genes (full SynGO analysis results are available in Supplementary Tables 6 and 7). We found that ∼20% of DEGs in both UL.Exc.Ns and LL.Exc.Ns from Mos hCOs were SynGO annotated. While Het UL.Exc.N showed comparable synaptic gene enrichment, Het LL.Exc.N exhibited only 10% overlap with SynGO genes (Fig. 5C). Mos LL.Exc.Ns showed the highest −log₁₀(Q-values) in SynGO enrichment analysis compared with Mos UL.Exc.Ns, Het LL.Exc.Ns and Het UL.Exc.Ns (Fig. 6A and Supplementary Fig. 4B), with significant enrichment in genes controlling synapse organization, synaptic vesicle cycle and regulation of postsynaptic membrane neurotransmitter receptor levels (Fig. 6A).

Figure 6.

Dysregulation of epilepsy-associated genes and increased neuronal activity in DEPDC5 mutant human cortical organoids. (A) SynGO enrichment analysis (one-sided Fisher exact test with multiple testing correction using false discovery rate) of DEGs between Mos and CT LL.Exc.Ns at 3 months. Terms with −log₁₀ Q-value >6 are shown. (B) Epilepsy-associated DEGs in Het and Mos LL.Exc.Ns at 3 months. Differentially expressed genes (Wilcoxon Rank Sum test with Bonferroni correction) between Het versus CT, Mos versus CT and Mos versus Het are reported in separate dot plots. For each dot plot, the three conditions are reported, with the dismissed condition shown in grey. (C) Ion-channel DEGs in Het and Mos LL.Exc.Ns and UL.Exc.Ns. (D) Spontaneous neuronal activity of whole hCOs measured by multielectrode array (MEA). Left: Representative raster plots showing spike activity of the 64 channels over 3 min and superimposed spikes from 32 electrodes. Middle: Quantification of firing rate (FR) of active electrodes. Violin plots show median (middle line). Number of hCOs displaying active electrodes out of the total number of recorded hCOs (≥3 batches): CT = 7/23, Het = 3/14, Mos High = 4/14. Kruskal–Wallis test with uncorrected Dunn’s correction. Mean number of active electrodes per hCO: CT = 2.14, Het = 1.67, Mos High = 7.75. *P < 0.05. Right: Quantification of active electrodes per hCOs (only hCOs with at least one active electrode are included). Total number of active electrodes from active organoids (with at least one active electrode): CT = 15 electrodes from 7 hCOs, Het = 5 electrodes from 3 hCOs, Mos High = 31 electrodes from 4 hCOs. Kruskal–Wallis test with uncorrected Dunn’s correction. Mean firing rate: CT = 0.26, Het = 0.35, Mos High = 0.39. *P < 0.05. CT = control; DEGs = differentially expressed genes; hCOs = human cortical organoids; Het = heterozygous; LL.Exc.N = lower-layer excitatory neurons; Mos = mosaic; ns = non-significant; SynGo = synaptic gene ontology.

We next interrogated DEGs for a curated list of 143 established epilepsy-associated genes.⁴⁶ Several of these genes overlapped with DEGs in both excitatory neuron subtypes, with LL.Exc.N showing the highest number of matching genes compared with UL.Exc.N. Nearly half of these epilepsy-associated genes were also differentially expressed between Het and Mos (Fig. 6B and Supplementary Fig. 4C). Among the LL.Exc.Ns DEGs, we identified a subset of mTOR- and FCDII-related genes. Notably, AKT3 and PIK3CA were upregulated only in Mos, with AKT3 showing a significant increase in Mos versus Het (Fig. 6B). In addition, we identified 13 and 6 DEGs matching ion channel genes in LL.Exc.Ns and UL.Exc.Ns, respectively (Fig. 6C). Among these, three voltage-gated ion channel genes previously implicated in epilepsy were identified: KCNB1 (potassium voltage-gated channel subfamily B member 1) was specifically upregulated in Mos LL.Exc.N compared with CT, CACNA1A (calcium voltage-gated channel subunit α1A) was downregulated in Het LL.Exc.N compared with CT and Mos, and KCNQ3 (potassium voltage-gated channel subfamily Q member 3) showed increased expression in Mos compared with both CT and Het (Fig. 6C). To further assess cell-autonomous effects, we examined ion channel gene expression within DEPDC5^+/− (EGFP−) and DEPDC5^−/− (EGFP+) populations of Mos hCOs compared with CT and Het. This analysis revealed a more pronounced dysregulation in the EGFP+ population (Supplementary Fig. 5A), supporting a direct role of biallelic DEPDC5 loss in modulating ion channel expression. Together, these findings indicate that Het and Mos excitatory neurons exhibit distinct degrees of dysregulation in genes and pathways linked to neuronal excitability, potentially contributing to the epileptogenic phenotype observed in DEPDC5-related FCDII.

To examine whether the differential expression of epilepsy-associated genes translates into functional alterations in neuronal networks, we recorded spontaneous neuronal spike activity in 6-month whole hCOs (n = 23 CT, n = 14 HET, n = 14 Mos) using extracellular recording with a 64-channel multielectrode array (MEA). Our analysis focused on two key activity parameters: (i) the number of active electrodes per hCO; and (ii) the firing rate (FR) of active electrodes. Both CT and Het hCOs displayed similarly low numbers (<5) of active electrodes (CT: 2.14 ± 0.71 across 7/23 hCOs; Het: 1.67 ± 0.67 across 3/14 hCOs). In contrast, Mos hCOs exhibited a significantly higher number of active electrodes compared with CT (7.75 ± 3.15 across 4/14 hCOs, P = 0.049). Among active electrodes, the FR was comparable between CT (CT: 0.26 ± 0.07 Hz, n = 15 active electrodes from 7 hCOs) and Het (Het: 0.35 ± 0.08 Hz, n = 5 active electrodes from 3 hCOs), whereas Mos exhibited a significantly higher FR (Mos: 0.39 ± 0.05; n = 31 active electrodes from 4 hCOs, P = 0.025) (Fig. 6D). We detected no significant differences in the number of active electrodes nor the FR between Het and Mos hCOs, which may indicate a susceptibility to neuronal network excitability due to heterozygosity.

Together, these results indicate that despite transcriptional alterations in Het hCOs—including changes in synaptic and epilepsy-associated genes—the type and/or magnitude of such changes are insufficient to increase network excitability. In contrast, biallelic DEPDC5 inactivation in Mos hCOs results in significant neuronal hyperactivity, consistent with the presence of dysregulated synaptic and ion channel gene expression identified at the transcriptomic level.

DEPDC5 two-hit cells in mosaic hCOs exhibit metabolic and translational dysregulations

To examine cell-autonomous effects of biallelic DEPDC5 loss, we compared DEPDC5^−/− (EGFP+) and DEPDC5^+/− (EGFP−) cell populations within mosaic hCOs at both early and mature developmental stages. We first assessed whether EGFP+ cells participated in rosette formation to the same extent as EGFP− cells. Analysis of neural rosette composition in 1-month organoid sections revealed three distinct patterns: nearly entirely EGFP− rosettes, mosaic (EGFP− and EGFP+) rosettes and nearly entirely EGFP+ rosettes (Fig. 7A). Quantification of the EGFP+ area within SOX2 rosettes demonstrated proportions consistent with the expected mosaicism rates of ∼50% and ∼20% in High Mos and Low Mos hCOs, respectively (Supplementary Fig. 5B). These findings indicate that DEPDC5^−/− cells integrate into rosettes without bias, suggesting that biallelic DEPDC5 loss does not significantly alter early neuroepithelial organization.

Figure 7.

Comparison of DEPDC5^+/− and DEPDC5^−/− cell populations in mosaic cortical organoids. (A) SOX2 and EGFP immunofluorescence in 1-month Mos hCOs showing three rosette types: entirely DEPDC5^+/− EGFP− rosettes (top left), mosaic (EGFP− and DEPDC5^−/− EGFP+) rosettes (top right) and nearly entirely EGFP+ rosettes (bottom left and right). (B) Single-cell transcriptomic analysis of EGFP+ and EGFP− populations in Mos hCOs. UMAP visualizations showing conserved cell clusters between EGFP+ and EGFP− populations. Right: Cell type composition at each time point. EGFP+ defined as cells with EGFP counts >0. (C) RNA velocity analysis comparing EGFP+ and EGFP− populations across all time points. Top: Cell type evolution along latent time. Bottom: Heat maps representing gene expression dynamics resolved along latent time show the cascade of transcription for the top 300 likelihood-ranked genes. Coloured stars indicate cell type-specific markers. (D) Top 10 most-significant GO biological process terms (hypergeometric distribution with Benjamini and Hochberg correction) enriched in DEGs between EGFP+ and EGFP− aRG at 1 month. (E) Differential gene expression analysis between EGFP+ and EGFP− populations (Wilcoxon Rank Sum test with Bonferroni correction). Left: Proportion of DEGs in LL.Exc.Ns (3 months) and UL.Exc.Ns (6 months). Right: Proportion of SynGO annotated synaptic genes among DEGs. (F) Top 10 most-significant GO biological process terms among DEGs between EGFP+ and EGFP− LL.Exc.Ns. (G) SynGO enrichment analysis (one-sided Fisher exact test with multiple testing correction using false discovery rate) of DEGs between EGFP+ and EGFP− LL.Exc.Ns (−log₁₀(Q-value) > 10 shown). (H) Epilepsy-associated DEGs between EGFP+ and EGFP− LL.Exc.Ns and UL.Exc.Ns. Ion channel genes are indicated in red. DEGs = differentially expressed genes; EGFP = enhanced GFP; GO = gene ontology; Het = heterozygous; LL.Exc.Ns = lower-layer excitatory neurons; Mos = mosaic; SynGo = synaptic gene ontology; UL.Exc.Ns = upper-layer excitatory neurons; UMAP = Uniform Manifold Approximation and Projection.

To assess whether developmental alterations were differentially affecting DEPDC5^+/− and DEPDC5^−/− cells, we analysed EGFP+ and EGFP− cell populations within Mos hCO using scRNA-seq data. Cell type proportions showed only minor differences between EGFP+ and EGFP− populations across all time points (1, 3 and 6 months; Fig. 7B). scVelo trajectory analysis confirmed globally similar cell type differentiation dynamics between both populations (Fig. 7C). Consistent with the similarity observed between Het and Mos hCOs, these results indicate that altered cell differentiation affects equally heterozygous and two-hit cells also within Mos hCOs (Fig. 2A).

To identify potential transcriptomic differences between DEPDC5^+/− and DEPDC5^−/− cell populations, we conducted DGE and GO analyses across key developmental stages and cell types (Supplementary Tables 1–7) as before, focusing on aRG at 1 month, LL.Exc.Ns at 3 months and UL.Exc.Ns at 6 months. aRG showed alterations of metabolic pathways, particularly ATP metabolism, oxidative phosphorylation and cellular respiration (Fig. 7D). UL.Exc.Ns showed a higher percentage of DEGs compared with LL.Exc.Ns (Fig. 7E), with top GO terms such as translation, metabolic and neuron projection development in both excitatory neuron subtypes (Fig. 7F). A substantial proportion of DEGs in both LL.Exc.N (28%) and UL.Exc.N (22%) mapped to SynGO (Fig. 7E). Among these SynGO-annotated genes, 45% in LL.Exc.N and 19% in UL.Exc.N overlapped with those identified in the Mos versus CT hCO comparison. Consistent with GO enrichment analysis, SynGO enrichment analysis revealed predominant dysregulation of metabolic processes, synaptic translation as well as synapse organization in both LL.Exc.N and UL.Exc.N (Fig. 7G and Supplementary Fig. 5C). Among the DEGs in the two excitatory neuron subtypes, we identified several epilepsy-associated genes, including LL.Exc.Ns (71%) and UL.Exc.Ns (21%) which were also found in the Mos versus CT comparison (Fig. 7H). Notable dysregulated ion channel genes, previously linked to epilepsy, included KCNQ3 (upregulated in DEPDC5^−/− UL.Exc.Ns), CACNA1A (upregulated in DEPDC5^−/− in LL.Exc.Ns) and SCN2A (downregulated in DEPDC5^−/− in UL.Exc.Ns). These transcriptional changes affecting DEPDC5^−/− compared with DEPDC5^+/− excitatory neurons may exacerbate the Mos phenotype and contribute to the epileptogenic nature of FCDII lesions harbouring biallelic DEPDC5 loss.

Discussion

In this study, we leveraged hCOs to model both heterozygous (Het) and biallelic LoF DEPDC5 mutations in a mosaic context (Mos), providing insights into the role of DEPDC5 and its regulation of mTOR activity in neurodevelopment. Through longitudinal data across three developmental time points, we identified early neurodevelopmental phenotypes and altered differentiation trajectories in both Het and Mos hCOs including differential neural rosette densities and premature generation of upper-layer neurons, which were associated with dysregulation of Notch and Wnt signalling pathways. These shared neurodevelopmental phenotypes between Het and Mos, together with the finding that DEPDC5^+/− and DEPDC5^−/− cells in Mos follow similar differentiation trajectories, indicate that DEPDC5 haploinsufficiency alters neurodevelopment similarly to biallelic loss, suggesting a scenario where global mTOR activity levels influence timing of fate specification in the developing cerebral cortex. Coherent with this finding, a recent study showed that deletion of tuberous sclerosis complex (TSC) proteins, which also function as mTOR repressors, leads to increased upper-layer neuron generation in the mouse neocortex.⁴⁷

In spite of the shared neurodevelopmental phenotype between Het and Mos hCOs, our findings provide experimental support for the Knudson’s two-hit model in FCDII surgical tissues where a second-hit somatic mutation leading to biallelic DEPDC5 inactivation is required for the FCDII phenotype including mTOR-hyperactive dysmorphic-like neurons and spontaneous neuronal network hyperactivity, which were absent in Het hCO (Table 1). Notably, DEPDC5 mosaic organoids did not develop balloon cells, consistent with clinical observations that DEPDC5 mutations predominantly cause FCDIIa, characterized by dysmorphic neurons without balloon cells.^9,20 The necessity of a second hit for the development of pS6-positive dysmorphic neurons was also observed in another mTORopathy cortical organoid model of TSC with TSC1/TSC2 mutations.²⁸ Our mosaic organoid model enabled us to directly compare DEPDC5^+/− and DEPDC5^−/− cells within Mos hCOs, identifying alterations in metabolism (including cell respiration processes) and translation, which likely drive the formation of cytomegalic dysmorphic neurons, although a clear mechanistic link needs providing.

Table 1.

Highlighted phenotypes observed in heterozygous and mosaic (Mos) human cortical organoids compared with controls, and in DEPDC5^+/− and DEPDC5^−/− cells in Mos

	Phenotypes
Genotypes	Upregulation of UL.Exc.N	Dysregulated epilepsy genes	Increased mTOR activity	Dysmorphic neurons	Increased network activity
Het DEPDC5+/−	Yes	Yes	No	No	No
Mos DEPDC5+/−, DEPDC5−/−	Yes	Yes	Yes	Yes	Yes
DEPDC5 +/− cells	Yes	Yes	No	No	N/A
DEPDC5 −/− cells	Yes	Yes	Yes	Yes	N/A

Het = heterozygous; Mos = mosaic; N/A = does not apply; UL.Exc.N = upper-layer excitatory neurons.

Patients with germline heterozygous DEPDC5 mutations and non-lesional epilepsy rarely undergo brain surgery, limiting opportunities for histopathological and genetic analyses of brain tissue. Thus, it remains unclear whether single-hit heterozygous DEPDC5 LoF alone is sufficient to generate seizures. Our electrophysiological findings align with mouse models, where homozygous deletion causes spontaneous seizures while heterozygous deletion reduces seizure threshold.^48-52 Consistent with this observation, the increased firing rate in Mos hCOs suggests that Biallelic DEPDC5 inactivation can confer seizure susceptibility.²⁸ We identified transcriptional alterations in excitatory neuron subtypes shared between Het and Mos, as well as specific to each genotype, including altered expression of synaptic and ion channel genes, with KCNQ3, SCN2A and CACNA1A specifically in double-hit cells. Although our data do not allow us to clearly identify which gene(s) may be responsible for hyperactivity, we point towards several epilepsy-associated and ion channel genes that are more severely affected in Mos.

Our study has some limitations. First, our findings are based on a single DEPDC5 patient-derived line, necessitating validation in additional DEPDC5 two-hit iPSC lines to establish phenotypic reproducibility. We note, however, that most pathogenic DEPDC5 variants reported to date are LoF (nonsense, frameshift or splice-site) mutations that undergo degradation via nonsense-mediated decay, leading to haploinsufficiency.^53,54 This reduces the relevance of comparing different patient variants, since they converge on the same pathogenic mechanism. Second, the use of homozygous DEPDC5 knockout hiPSCs may not accurately model the temporal dynamics of somatic mutation acquisition in patients, where second-hit variants are likely to arise at later developmental stages. Such early depletion could trigger compensatory mechanisms that may attenuate the observed phenotypes. Third, we were unable to generate a reliable isogenic control; instead, we used an age- and sex-matched control line from the unaffected sibling. Importantly, some of our key findings rely on comparisons between DEPDC5^+/− and DEPDC5^−/− cells within the same mosaic organoids, which mitigates concerns about the absence of an external isogenic control. Future studies using inducible DEPDC5 knockout systems and additional patient-derived lines will be essential to extend and validate these proof-of-concept findings.

Despite these limitations, our study sheds light on the molecular mechanisms underlying DEPDC5-related FCDII, demonstrating the relevance of mosaic hCOs in modelling focal cortical malformations. We provide here the bases for future studies aiming at identifying potential therapeutic targets among the identified dysregulated pathways. These advances will pave the way for more personalized approaches in understanding and treating DEPDC5-related cortical malformations and epilepsies.

Data availability

Single-cell RNA sequencing data will be deposited at the European Genome-phenome Archive (EGA).

Funding

This work was supported by the European Research Council (ERC-CoG) (No. 682345 to S.Bau.), the programme ‘Investissements d'avenir’ (ANR-10-IAIHU-06 to S.Bau.), the Fondation pour la Recherche Médicale (ECO20160736027 to S. Bau), Agence nationale pour la recherche and the European commission (ERA-NET NEURON JTC 2021) (ANR-21-NEU2-0002-01 to S.Bau). S.Biz. was supported by the Horizon2020 Research and Innovation Program Marie Skłodowska-Curie Actions (MSCA) Individual Fellowship (grant agreement no. 101026484—CODICES), and is now supported by a European Research Council starting grant (Grant Agreement Project No. 101115984—LINMOS). T.R. was supported by the Fondation pour la Recherche Médicale (FDT201904008269) and the Ligue Française Contre l’Epilepsie.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.