TSC tunes progenitor balance and upper-layer neuron generation in neocortex

Abstract

The appropriate generation of upper-layer neurons is necessary to create the circuits that underlie complex brain functions. Radial progenitors divide asymmetrically to generate neurogenic intermediate progenitors (IPs; also known as intermediate precursors), and the symmetric proliferation of IPs rapidly expands the cortical neuronal population. The dynamic maintenance of balanced diversity of cortical progenitors and the resultant generation, placement and connectivity of appropriate numbers of different classes of neurons serve to guide the formation of a properly wired cerebral cortex^1–12. However, the molecular logic that instructs progenitor balance remains unclear. Here we show that members of the tuberous sclerosis complex (TSC)—proteins that are major regulators of cellular metabolism—function to sculpt radial progenitor–intermediate progenitor balance, radial unit organization and the resultant generation of upper-layer neurons. Developmental deletion of TSC proteins alters the radial progenitor and IP balance and changes radial unit composition, leading to increased upper-layer neuron generation and aberrant cortical connectivity. Human-specific modulation of TSC protein expression through human-gained enhancers affects progenitor balance and generation of upper-layer neurons. Evolutionary downregulation of TSC protein expression may therefore provide an effective route to radial unit sculpting and the expanded generation of upper-layer neurons necessary for higher-order brain functions in humans.

The effective temporal maintenance of balance between radial progenitors and IPs underlies the generation of the correct number and types of neurons at the appropriate time in the developing cerebral cortex. As such, these progenitors form the foundation for the construction of the cerebral cortex. Radial glia (RG; also known as radial glial cells), through their function as a source of neurogenic or gliogenic progenitors and neuronal migratory guides, provide a template for the formation of the cerebral cortex^1–6. The migration of newborn neuronal clones along related RG basal processes leads to the formation of cortical radial units (or radial columns) and the laminar organization of neurons^1,2,5,6,12. During the neurogenic period, RG divide asymmetrically to generate cortical neurons and neurogenic IPs or outer radial glia (oRG)^1–8. The symmetric proliferation of IPs and oRG serves to rapidly expand the cortical neuronal population, particularly upper-layer neurons. During brain evolution, the expansion of upper-layer neurons in primates underlies enhanced cognitive flexibility and working memory in primates^5,6,9–11.

During corticogenesis, fine-tuning of the balance between radial progenitors and IPs and the resultant differences in the generation and guidance of neurons enables the emergence of functionally distinct cortical areas with characteristic patterns of neuronal density, diversity and connectivity^1–6. Abnormalities in progenitor balance and the ensuing changes in the generation, placement and connectivity of neurons underlie cortical malformations (for example, microcephaly, lissencephaly and cortical band heterotopia) and complex neurodevelopmental disorders such as epilepsy, schizophrenia and autism spectrum disorder (ASD)^1–6. Despite the significance of progenitor balance, the essential mechanisms that drive this fundamental feature of the developing cerebral cortex remain poorly understood.

The autosomal-dominant disorder tuberous sclerosis complex is caused by mutations in the tumour suppressor genes TSC1 or TSC2 and is characterized by altered cortical development and high incidences of developmental delay, ASD and epilepsy^13–17. Depletion of TSC1 or TSC2 in the developing mouse brain causes a range of changes in progenitor and neuronal differentiation, lamination defects, gliosis, myelination abnormalities and enlarged brain size^18–25. Loss-of-function mutations in TSC1 or TSC2 in humans lead to cortical tubers that contain dysmorphic neurons, subependymal nodules, subependymal giant cell astrocytomas and hemimegalencephaly, all of which indicate disrupted cortical development and connectivity^26–28. TSC1 and TSC2, together with TBC1D7, form a complex that integrates diverse signalling pathways of cell identity, growth and differentiation (for example, mTORC1 and mTORC2, AMPK, Notch and WNT)^{13–17,29–33}, and these pathways are substantially dysregulated in several human neurodevelopmental and neural circuit disorders^16,30,34. The TSC may therefore serve as an effective molecular platform to organize the mechanisms that govern progenitor balance and the consequent development and differentiation of diverse cortical neurons. We therefore examined the role of the entire TSC as an ensemble in progenitor balance and the resultant patterns of neuronal organization in the developing cerebral cortex.

Here we show that the TSC modulates radial progenitor and IP balance, radial unit composition and the resultant generation of upperlayer neurons necessary to sculpt cerebral cortical organization. Furthermore, human-specific downmodulation of TSC expression through human-gained enhancers (HGEs) affects progenitor balance and genesis of upper-layer neurons and may therefore contribute to the increase in cortical neuron numbers in humans.

TSC1/2 regulate cortical progenitor diversity and balance

To examine how radial progenitor and IP identities and balance are maintained as the cerebral cortex develops, we conditionally depleted both TSC1 and TSC2 (TSC1/2) in cortical progenitors using the Emx-Cre mouse line. Control (Emx^creTsc1^lox/+Tsc2^lox/+) and Tsc1/2 Emx-cKO (Emx^creTsc1^lox/loxTsc2^lox/lox) cortices from embryonic day 16 (E16) and postnatal day 0 (P0) mice were labelled with radial progenitor (that is, apical) and IP (that is, basal) markers and with different neuronal-layer-specific markers. Deletion of Tsc1/2 in cortical progenitors led to altered apical to basal progenitor balance and significant expansion of the cortex in Tsc1/2 Emx-cKO mice. The basal progenitor population was significantly increased in Tsc1/2 Emx-cKO mice, which in turn altered the ratio of TBR2⁺ basal progenitors to PAX6⁺ apical progenitors (Fig. 1a–c). Consistent with these changes, an increase in mitotic cells (stained using phospho-histone 3 (PH3)) was detected in the ventricular zone (VZ) and the subventricular zone (SVZ) of Tsc1/2 Emx-cKO mice (Extended Data Fig. 1a–c). Furthermore, an atypical population of PAX6⁺TBR2⁺ cells was present in Tsc1/2 Emx-cKO mice (Extended Data Fig. 1d–f). The average nucleus size of these cells was 53.3 ± 2.65 μm² (n = 30) compared with PAX6⁺ cells (41.34 ± 1.73 μm²; n = 30) and TBR2⁺ cells (35.64 ± 1.10 μm²; n = 30 nuclei; P < 0.001, unpaired two-sided t-test). No significant changes in caspase-3⁺ apoptotic cells were detected in the cortex of Tsc1/2 Emx-cKO mice (Extended Data Fig. 1g–i). GFAP immunolabelling revealed an increase in cortical astrocytes in Tsc1/2 Emx-cKO mice (Extended Data Fig. 1j–l), which suggests that there is aberrant astrogliosis in the neocortex of these mice. The disrupted cortical development in Tsc1/2 Emx-cKO mice was reflected in increased brain weight, cortical width and circumference in the mutants (Extended Data Fig. 2). As expected, deletion of Tsc1/2 in cortical progenitors led to increased mTORC1 signalling, as indicated by enhanced S6 phosphorylation (pS6) in progenitors and the cortex in Tsc1/2 Emx-cKO mice (Extended Data Fig. 3).

Fig. 1 |. Effect of Tsc1/2 deletion on RG/IP balance.

a,b, Expansion of TBR2⁺ IPs (red) in control cells (a) and Tsc1/2 cKO cells (b). c, Quantification of changes in the IP (TBR2⁺) to RG (PAX6⁺) ratio in Tsc1/2 Emx-cKO mice. Data are the mean ± s.e.m. (n = 4 mice per group). Unpaired two-sided t-test. ***P < 0.0001. Scale bar, 20 μm (a,b).

We next examined whether neuronal generation is altered as a result of altered apical to basal progenitor balance in the cortex of Tsc1/2 Emx-cKO mice. Upper-layer neurons were labelled with anti-CUX1 or anti-SATB2 antibodies, whereas deeper-layer neurons were labelled with anti-TBR1 antibodies. Tsc1/2 deletion led to a substantially expanded population of CUX1⁺ upper-layer neurons that were distributed ectopically in all layers of the cortex (Fig. 2a–c). Labelling with another upper-layer neuronal marker, SATB2, and with the pan-neuronal markers TUJ1 (an anti-β-tubulin 3 (TUBB3) antibody) and NeuN also indicated expanded numbers of cortical neurons in Tsc1/2 Emx-cKO mice (Fig. 2d–n). No significant changes in deeper-layer TBR1⁺ neuronal numbers were observed in Tsc1/2 Emx-cKO mice (Fig. 2o–q), although the width of the TBR1⁺ layer was wider (290.1 ± 8.1 μm for control compared with 362.3 ± 7.3 μm for Tsc1/2 Emx-cKO; unpaired two-sided t-test, P < 0.001; n = 7 brains per group). This result reflected the expanded cortical width and lamination defects observed in Tsc1/2 Emx-cKO mice.

Fig. 2 |. Tsc1/2 deletion promotes upper-layer neuronal generation.

a,b, Images of CUX1⁺ upper-layer neurons in control (a) and Tsc1/2 Emx-cKO (b) cells. c, Quantification of changes in CUX1⁺ neuronal density in Tsc1/2 Emx-cKO cells. d,e, Images of SATB2⁺ upper-layer neurons and cortical width in control (d) and Tsc1/2 Emx-cKO (e) mice. f–h, Images (f,g) and quantification (h) of SATB2⁺ neurons in control (f) and Tsc1/2 Emx-cKO (g) cells. i,j,l,m, Immunolabeling of neurons with anti-NeuN (i,j) or anti-TUJ1 antibodies (l,m) indicates an expansion of cortical neuronal numbers in Tsc1/2 Emx-cKO cells (j,m) compared with controls (i,l). k,n, Quantification of NeuN⁺ (k) and TUJ1⁺ (n) neurons. o–q, Images (o,p) and quantification (q) show no significant changes in TBR1⁺ deeper-layer neuronal density in Tsc1/2 cKO cells (p) compared with controls (o). Data are the mean ± s.e.m. (n = 4 mice per group). Unpaired two-sided t-test. **P < 0.01 and ***P < 0.001 (P = 0.0008 (c), P = 0.0032 (h), P = 0.0001 (k), P = 0.0017 (n), P = 0.9279 (q)). Scale bars, 50 μm (a,b,f,g,i,j,l,m,o,p) or 200 μm (d,e).

To identify the role of TBR2⁺ basal progenitors in the expansion of the upper layer, tamoxifen-mediated Tsc1/2 deletion was induced in IPs in Tbr2^cre-ERTsc1^lox/loxTsc2^lox/lox;Ai14 (Tsc1/2 Tbr2-cKO) mice. Significant expansion of upper-layer neurons was evident in Tsc1/2 Tbr2-cKO cortices (Fig. 3a–h). Consistent with the increased number of upper-layer neurons in Tsc1/2 Tbr2-cKO mice, the thickness of their main axonal projection tract (that is, the corpus callosum) was increased in the cortex of Tsc1/2 Tbr2-cKO mice (Fig. 3a,b and Extended Data Fig. 4). Underlying this upper-layer neuronal expansion was an increase in IPs in the cortex of Tsc1/2 Tbr2-cKO mice (Fig. 3i–l). Together, these observations suggest that the balance between radial progenitors and IPs is altered as a result of Tsc1/2 deletion, which in turn leads to an expanded upper-layer neuronal population and increased cortical size.

Fig. 3 |. Deletion of Tsc1/2 in IPs leads to expansion of upper-layer neurons.

a,b, Tsc1/2 deletion in TBR2⁺ IPs in Tsc1/2 Tbr2-cKO mice leads to the expansion of the cortical upper-layer, as indicated by the increased width of Ai14 (tdTomato⁺)-labelled upper layer in Tsc1/2 cKO mice (b) compared with control mice (a).Lines in a and b indicate Ai14⁺ layer width. c, Quantification of the increased width of Ai14-labelled upper layer in Tsc1/2 Tbr2-cKO mice (n = 5 mice per group). d,e, Ai14⁺ cells co-label with the upper-layer neuronal marker SATB2 in control (d) and Tsc1/2 Tbr2-cKO mice (e). f,g, Control (f) and Tsc1/2 Tbr2-cKO (g) cortical slices illustrate the expansion of Ai14⁺SATB2⁺ neurons (yellow) in knockout mice. h, Quantification of increased Ai4⁺SATB2⁺ cell density in Tsc1/2 Tbr2-cKO mice (n = 6 mice per group). i–l, Neuronal expansion is facilitated by an increase in Ai14⁺TBR2⁺ IP cell density, and the TBR2⁺/PAX6⁺ cell ratio is increased in Tsc1/2 Tbr2-cKO mice. i,j, The VZ in control (i) and Tsc1/2 Tbr2-cKO ( j) mice was co-labelled with anti-PAX6 and anti-TBR2 antibodies. Arrowheads (i,j) indicate a sample of Ai14⁺TBR2⁺ cells. k,l, Quantification of increased Ai14⁺TBR2⁺ cell density (k) and Ai14⁺TBR2⁺/Ai14⁺PAX6⁺ ratio (l) in Tsc1/2 Tbr2-cKO mice (n = 8 mice per group). Data are the mean ± s.e.m. Unpaired two-sided t-test; *P < 0.05, **P < 0.01 and ***P < 0.001 (P < 0.0001 (c), P < 0.0001 (h), P = 0.0012 (k), P = 0.018 (l)). Scale bars, 400 μm (a,b,d,e), 100 μm (f,g) or 50 μm (i,j). CC, corpus callosum.

Disrupted radial units in Tsc1/2 cKO mice

Changes in radial unit composition are thought to underlie neuronal expansion in the cerebral cortex^1,2,6. We therefore examined the effect of Tsc1/2 deletion on radial unit composition. We used the mosaic analysis with double markers (MADM) system to trace cell lineages^35,36 to evaluate changes in the cellular composition of individual radial units. We expressed Cre recombinase in isolated RG from the dorsal cortex of E14 MADM-11^GT/TG;Tsc1/2^lox/+ mice (MADM-control) and of MADM-11^GT/TG;Tsc1/2^lox/lox (MADM-Tsc1/2 cKO) mice. As we applied MADM as a lineage tracer without linking Tsc1/2-floxed alleles with the TG or GT MADM cassettes³⁵, all of the MADM-labelled (green, red or yellow) clones in MADM-control and MADM-Tsc1/2 cKO cortices are Tsc1/2^lox/+;cre or Tsc1/2^lox/lox;cre, respectively. This labelling of isolated radial units results from the deletion of Tsc1/2 in a single RG. At E16, we analysed MADM⁺, Tsc1/2-deleted radial units for the number and types of progenies (progenitors and neurons) (Fig. 4a–e). Compared with MADM-control, MADM-Tsc1/2 cKO radial units had 41% more progenitors and neurons per radial unit (Fig. 4e), which indicated the presence of altered radial unit composition and neuronal expansion in the cortex of MADM-Tsc1/2 cKO mice. Co-labelling of the radial units with anti-PAX6 and anti-TBR2 antibodies revealed that the ratio of TBR2⁺ basal progenitors to PAX6⁺ apical progenitors in radial units was increased in MADM-Tsc1/2 cKO mice (Fig. 4f–l). Furthermore, co-labelling the radial units with anti-SATB2 antibodies indicated that the number of SATB2⁺ upper-layer neurons in radial units also increased (Fig. 4m–q). Consistent with this result, in the cerebral cortex of P21 mice, an age at which neuronal placement is completed, MADM-Tsc1/2 cKO radial unit clones contained relatively more upper-layer neurons (Fig. 4r–v). The clones showed significantly wider distribution and altered, highly branched neuronal morphology than MADM-control clones (Fig. 4r–s,w). This finding indicates that Tsc1/2 deletion leads to disrupted laminar and columnar organization of expanded upper-layer cortical neurons.

Fig. 4 |. Effect of Tsc1/2 deletion on radial unit composition and upper-layer neuron generation.

a–d, MADM-based visualization of a single RG and all its progenies (that is, radial unit) in control (a,b) and Tsc1/2 cKO mice (c,d). e, Expansion of the number of cells per radial unit in Tsc1/2 cKO mice (n = 50 radial units) and controls ((n = 50 radial units). f–k, Co-labelling of control (f–h) and Tsc1/2 cKO (i–k) radial units with anti-PAX6 and anti-TBR2 antibodies. Yellow arrowheads indicate co-labelled PAX6⁺ (g,j) and TBR2⁺ (h,k) cells. l, Quantification of altered TBR2⁺ basal progenitor/PAX6⁺ apical progenitor ratio in Tsc1/2 cKO radial units (n = 11 radial units). m–p, Co-labelling of control (m,n) and Tsc1/2 cKO (o,p) radial units with anti-SATB2 antibodies. Red arrowheads indicate co-labelled SATB2⁺ (n,p) neurons. q, Quantification of increased SATB2⁺ neuronal numbers in Tsc1/2 cKO radial units (n = 6 radial units). GFP label of the merged panels (g,h,j,k,n,p) were superimposed to highlight co-labelled cells. Data are the mean ± s.e.m. Unpaired two-sided t-test; ***P < 0.001 (P < 0.0001 (e), P < 0.0001 (l), P = 0.001 (q)). r–w, Altered distribution and number of clonally related upper-layer neurons generated from Tsc1/2-deficient progenitors. r,s, Representative images of clonally related upper-layer neurons generated from a single radial progenitor in P21 MADM-control (r) and MADM-Tsc1/2 cKO (s) cerebral cortices in layers II–VI. t–w, Quantification of the total number of upper-layer neurons per radial unit (t), the total number of deep-layer neurons per radial unit (u), the upper-layer to deeper-layer neuronal ratio per radial unit (v) and the lateral distribution width of clonally related neurons (w). Data are the mean ± s.e.m. (n = 25 clones for control, n = 25 clones for MADM-Tsc1/2 cKO). Unpaired two-sided t-test; **P < 0.01 and ***P < 0.001 (P < 0.0001 (t), P = 0.4931 (u), P = 0.001 (v), P = 0.0003 (w)). Scale bars, 20 μm (a–d,f,i,m,o) or 50 μm (r,s).

To gain further insights into general changes in cortical neuronal morphology and connectivity resulting from expansion of the neuronal population in Tsc1/2 cKO mice, we analysed laminar organization and connectivity in MADM-11^GT/TGTsc1/2^lox/+Emx^cre (MADM-control^Emx) and MADM-11^GT/TGTsc1/2^lox/loxEmx^cre (MADM-Tsc1/2 Emx-cKO) cortices at age P5. Neurons displayed normal layer distribution in cortices of control mice, whereas in MADM-Tsc1/2 Emx-cKO mice, large, MADM-labelled ectopic collections of neurons were evident throughout the cortex (Extended Data Fig. 5a–d) and cortical layering was disrupted (Extended Data Fig. 5c,d). The axodendritic architecture of MADM-Tsc1/2 Emx-cKO upper-layer neurons showed excessive dendritic branching and aberrant axon projection patterns (Extended Data Fig. 5e–g). The increased number of upper-layer neurons in Tsc1/2 Emx-cKO mice led to increased thickness of the main axonal projection tract (corpus callosum) (Extended Data Fig. 5h–j). Collectively, these observations suggest that Tsc1/2 deletion leads to altered radial unit formation, and the resultant expansion of upper-layer neurons in turn change cortical circuitry.

Effect of inactivation of the entire TSC

TSC1 and TSC2 form a complex with TBC1D7 to mediate the functions of the TSC signalling complex^13,14,33. We therefore inactivated TBC1D7 in the Tsc1/2-null background to investigate the effect of the TSC1–TSC2–TBC1D7 complex as a whole on upper-layer neuronal development. Acute inactivation of TBC1D7 with short-hairpin RNA (shRNA) against Tbc1d7 in the cortex of Tsc1/2 Emx-cKO mice led to further disruption of neuronal lamination in these mice. Cells deficient in TSC1, TSC2 and TBC1D7 were ectopically placed throughout the cortex, away from their normal laminar locations (Extended Data Fig. 6a–h). A disrupted placement of neurons in the cortical plate (arrows in Extended Data Fig. 6e–h) and ectopic accumulation of neurons in the SVZ region were evident (asterisks in Extended Data Fig. 6e–h). Co-labelling of shRNA-expressing neurons (GFP⁺) with the upper-layer neuronal marker SATB2 indicated increased numbers of GFP⁺SATB2⁺ upper-layer neurons after Tbc1d7 knockdown in the cortex of Tsc1/2 cKO mice (Extended Data Fig. 6i–m). GFP⁺TBR1⁺ deep-layer neuronal numbers were not affected (Extended Data Fig. 6n–r). These results suggest that the deletion of all three members of the TSC leads to significantly disrupted generation, migration and distribution of upper-layer cortical neurons.

Single-cell RNA-seq of Tsc1/2 cKO cerebral cortex

To define how the TSC1–TSC2 complex regulates and restricts the cell-intrinsic identities and proliferative patterns of distinct cortical progenitors and the resultant RG/IP balance, we performed single-cell RNA sequencing (scRNA-seq) analysis of control and Tsc1/2-null cortices from P0 mice. We transcriptionally profiled around 75,000 cells across three biological replicates. We examined whether cell types, cell proportions and gene expression in a cell type are altered following Tsc1/2 deletion in the developing cerebral cortex. Analyses of scRNA-seq profiles in an unbiased manner led to the identification of 12 distinct groups of cells, and we characterized the identity of each of them using cell-type-specific gene expression annotations from previous studies^37,38 (Fig. 5a,b). Comparisons of cell-type composition in control and Tsc1/2-depleted cortices indicated a clear change in the representation of RG, IPs and upper-layer projection neuronal groups after Tsc1/2 deletion (Fig. 5c,d). Progenitor balance (that is, RG/IP ratio) shifted towards increased IPs in Tsc1/2 cKO mice. That is, the number of RGs was 7.44 ± 0.11 for control compared with 6.1 ± 0.11 for Tsc1/2 cKO (P < 0.001, unpaired two-sided t-test), and the number of IPs was 4.68 ± 0.02 for control compared with 6.04 ± 0.02 for Tsc1/2 cKO (P < 0.001, unpaired two-sided t-test) (Fig. 5c,d). Consistent with our in vivo observations, an increase in upper-layer neurons was evident in Tsc1/2 cKO mice (6.21 ± 0.04 for control compared with 8.60 ± 0.1 for Tsc1/2 cKO; P < 0.001, unpaired two-sided t-test) (Fig. 5c,d). Furthermore, expression of the oRG-like progenitor markers HOPX, HERC2 and ID4 (refs. 7,39) was significantly increased in Tsc1/2 cKO cortices (Fig. 5e).

Fig. 5 |. TSC1/2 mediate the RG/IP balance and lead to the expansion of upper-layer neurons.

a, Violin plots of number of unique molecular identifiers (UMIs), number of genes and percentage of mitochondrial genes per cell in control (n = 3) and Tsc1/2 cKO (cKO; n = 4) mice. b, t-distributed stochastic neighbour embedding (tSNE) visualization of scRNA-seq datasets assigned by cell type or genotype. OPC, oligodendrocyte precursor cell. c, Composition of each cell type by genotype. d, The RG/IP balance is significantly altered, and upper-layer (UL) neuronal percentage is increased in Tsc1/2 cKO mice. Unpaired two-tailed t-test; ***P < 0.001. False discovery rate (FDR): RG, FDR = 6.800281 × 10⁻⁶; IP, FDR = 3.663794 × 10⁻⁷; upper-layer neurons, FDR = 1.393733 × 10⁻⁶. Box plots represent first, second (median, centre line) and third quartiles, with whiskers drawn as 1.5× the interquartile range. e, Left, altered expression of progenitor markers (HOPX, HERC2 and ID4) that preferentially define human oRG-type progenitors in Tsc1/2 cKO cells. Right, images of the increased expression of HOPX in Tsc1/2 cKO progenitors in the VZ, and quantitative PCR with reverse transcription (RT–qPCR) confirmation of increased Herc2 and Id4 expression in Tsc1/2 cKO cortices (n = 8 mice). Data are the mean ± s.e.m. Unpaired two-sided t-test; **P < 0.01 and ***P < 0.001 (P = 0.0008 (middle), P = 0.0015 (right)). f, Changes in upper-layer neuronal lineage trajectory in Tsc1/2 cKO cells. Pseudotime trajectory of analysis was performed with Slingshot, filtering out genes expressed at low levels and using a two-dimensional embedding based on principal component (PC) analysis of z scores of gene expression patterns in RG (green), IP (blue) and upper-layer neurons (red) in control (darker hues) and Tsc1/2 cKO (lighter hues) cells. Three lineages were identified: lineages 1 and 2 both represent trajectories that originate with RG and terminate at the upper layer neurons (RG→IP→UL), and both have fewer immature cells (IPs) and more mature upper-layer cells in cKO relative to control mice. Lineage 3 describes a RG to IP differentiation trajectory that presumably will give rise to other projection neurons. Significant differences in both the abundance of cells along these trajectories and pseudotemporal gene expression were detected between control and cKO samples. g, The density of cells along pseudotime. Control and cKO cells are marked with different colours. Tsc1/2 cKO lineages have fewer immature cells (IPs) and more mature upper-layer cells than control samples. Two-sided Kolmogorov–Smirnov test; P ≤ 2.2 × 10⁻¹⁶ (lineage 1), P = 5.329 × 10⁻¹⁵ (lineage 2) and P = 5.526 × 10⁻¹¹ (lineage 3). h, Gene ontology terms of dysregulated biological pathways associated with differentially expressed genes in RG, IPs and upper-layer neurons in Tsc1/2 cKO cortices (Fisher’s one-tailed test; P < 0.05). Circle size indicates P values. Scale bar, 50 μm (e).

Lineage trajectory-based gene-expression analysis of scRNA-seq data indicated significant changes in the progression of RG to IP to upper-layer neuronal lineage in Tsc1/2 cKO cortex. Lineages 1 and 2 that terminate in the upper-layer neurons were enhanced in Tsc1/2 cKO cortex (Fig. 5f,g). The abundance of cells along these trajectories and pseudotemporal gene expression were altered in Tsc1/2 cKO cortex (P < 0.001, two-sided Kolmogorov–Smirnov test). These lineages had fewer progenitors and more upper-layer neurons in Tsc1/2 cKO cortex than in control cortex (Fig. 5g; P < 0.001, two-sided Kolmogorov–Smirnov test).

To investigate the underlying regulatory mechanisms of these cell-type changes in Tsc1/2 cKO mice, we analysed differentially expressed genes in distinct cell groups (for example, RG, IPs and upper-layer neurons) to identify dysregulated biological pathways in these Tsc1/2 cKO cells. In line with the phenotypic changes in neurogenesis, neuronal growth and connectivity observed in the cortex of Tsc1/2 cKO mice, Tsc1/2 deletion was associated with the dysregulation of pathways related to cell metabolism, brain development, neurogenesis, neuronal development, neuronal projection, axon development and synaptic contacts (Fig. 5h). Moreover, consistent with the known role of the TSC in ASD, distinct patterns of gene expression changes in 101 high-confidence ASD genes (identified using the SFARI Gene database)⁴⁰ were evident in progenitors and upper-layer neurons of Tsc1/2 cKO mice (Extended Data Fig. 7).

Together, these observations suggest that the TSC is a critical determinant of progenitor balance in the developing cerebral cortex, and this balance is necessary for the appropriate generation of upper-layer neurons. Loss of Tsc1/2 disrupts the RG/IP balance and results in the expansion of upper-layer neurons, thus contributing to the cortical organizational changes that underlie Tsc1/2-related neurodevelopmental disorders.

TSC1/2 regulate human-specific progenitor markers

Human oRG-type cells that enable the generation and expansion of upper-layer neuronal population are distinguished by the expression of markers such HOPX and ID4 (refs. 7,39). TSC1/2 deletion led to an increase in progenitors expressing these markers (Fig. 5e). Furthermore, increased S6 kinase expression is a characteristic of human oRG cells^41,42, and TSC1/2 deletion led to enhanced S6 expression and activation in cortical progenitors (Extended Data Fig. 3f–h).

TSC gene expression in mouse and human

Comparisons of TSC1, TSC2 and TBC1D7 expression patterns across different species showed strong conservation of the TSC genes, especially between human and mouse (Extended Data Fig. 8a). To delineate the potential function of TSC genes (TSC1, TSC2 and TBC1D7) in cerebral cortical development, we analysed their developmental expression in human and mouse brains, particularly in the germinal zone (VZ and SVZ) and the cortical plate of the developing cerebral cortex^37,43–47. TSC1, TSC2 and TBC1D7 expression levels were lower in human neural cells (that is, RG, IPs and neurons) than in mice, which suggests that TSC1, TSC2 and TBC1D7 expression is suppressed in the developing human brain (Extended Data Fig. 8b). Of these three TSC members, Tsc1 expression was lowest in the mouse brain, whereas TSC2 was lowest in the human brain (Extended Data Fig. 8b). The evolutionary suppression of TSC2 during human embryonic development suggests that it may have a role in expansion of the human brain.

Epigenetic regulation of TSC genes

Trimethylation of histone H3 at lysine 27 (H3K27me3) is associated with transcriptional silencing^48,49, whereas acetylation of histone H3 at lysine 27 (H3K27ac) in promoter regions is linked to active promoters and transcriptional activation⁵⁰. H3K27me3 and H3K27ac are mutually exclusive, and H3K27ac can antagonize H3K27me3-mediated gene silencing^51,52. To understand the potential role of H3K27me3 and H3K27ac in the regulation of TSC genes, we analysed these histone modifications around the transcription start site (TSS) of TSC genes in both human and mouse brain tissue samples^53–55. H3K27me3 peaks were present in the TSSs of TSC1, TSC2 and TBC1D7 in the fetal human brain, but not in the embryonic or adult mouse brain (Extended Data Fig. 8c). By contrast, strong H3K27ac peaks were present in the TSSs of Tsc1, Tsc2 and Tbc1d7 in the embryonic and adult mouse brain (Extended Data Fig. 8c). Similar results were observed in embryonic and adult human brain tissue, although the H3K27ac peaks in TSC2 in the adult human brain were weaker (Extended Data Fig. 8c). These results suggest that human-specific modulation of TSC genes by H3K27me3 and H3K27ac, specifically embryonic suppression of TSC genes, might facilitate neurodevelopmental changes specific to the human brain.

Identification of HGEs of TSC genes

Single-cell transcriptomic and epigenomic studies have identified genetic changes that influence developmental gene regulatory mechanisms among different species. HGEs are active regions or sequences in the human genome that are conserved in many species but show significant human-specific features, including gained enhancer activity⁵⁶. HGEs have epigenetic changes that may modify gene regulation in a human-specific manner^57–59. As deletion of Tsc1/2 in neural progenitors led to the expansion of the mouse cortex and the TSC is epigenetically regulated and expressed at low levels in the human cortex, we hypothesize that HGEs might be involved in the regulation of the TSC in human neural progenitors to mediate the expansion of upper-layer neurons in the human cerebral cortex.

HGEs can be defined by open chromatin, low levels of DNA methylation and physical interactions with gene promoters⁵⁸. Previous studies have identified potential HGEs of TSC genes in human brain cells^56,60–63. To profile accessible chromatin that may regulate the expression of the TSC, we analysed assay for transposase-accessible chromatin with sequencing (ATAC–seq) data of developing human brain neural progenitor cells (NPCs) and neurons^61,63. Looping of the accessible chromatin enables distal regulatory elements, including HGEs, to interact with gene promoters⁶⁰. To identify the presence of HGEs that interact with TSC promoters, we first analysed the correlation between potential enhancer and promoter ATAC–seq signals from human neurons and neural progenitors⁶³. An ATAC–seq peak near the loci of TBC1D7 (TBC1D7-P1) overlapped with a HGE detected in a comparative analysis of human, rhesus monkey and mouse brains designed to identify HGEs⁵⁶ (Extended Data Fig. 9a). TBC1D7-P1 is located upstream of the promoter and showed significantly higher chromatin accessibility in neural progenitors than neurons. As described above, expression of TBC1D7 is relatively suppressed in human neural progenitors (Extended Data Fig. 8b). The higher signal of TBC1D7-P1 in neural progenitors suggests that this HGE may be a repressive cis-regulatory element of TBC1D7.

To further identify the interaction between TSC promoters and their HGEs, we analysed a publicly available Hi-C seq dataset^57,60. Two HGEs near TSC2 (TSC2 HGE1 and TSC2 HGE2) showed physical interactions with the promoter in both cortical plate and germinal zone cells (Fig. 6a). The Hi-C peaks of the two HGEs also overlapped with single-cell ATAC–seq peaks of the developing human brain⁶¹, a result that suggests that they may have a regulatory function. As TSC2 expression is repressed in the human brain, this evidence suggests that TSC2 HGE1 and TSC2 HGE2 may have a role in negatively regulating human TSC2. We did not find any evidence of a HGE interacting with TSC1 (Extended Data Fig. 9b).

Fig. 6 |. Identification and analysis of the effect of HGEs on TSC2 expression and function.

a, Hi-C data of TSC2 from cortical plate (CP) and germinal zone (GZ) cells showing two HGEs (data from refs. 56,57) (blue boxes; TSC2 HGE1 and TSC2 HGE2). Single-cell ATAC–seq (scATAC) peaks representative of open chromatin were also found in these regions in human cortical progenitors and neurons (data from ref. 61). DL, deep layer; EN, excitatory neurons. b, CRISPRi targeting of HGEs of TSC2 disrupts TSC2 expression. RT–qPCR analysis of TSC2 expression in primary human NPCs after lentiviral transduction with dCas9-KRAB-gRNAs targeted to TSC2 HGEs. Two TSC2 HGE sgRNAs targeted to HGE1 (HGE1-sgRNA1 and HGE-sgRNA2) led to upregulation of TSC2 by 1.39-fold and 1.28-fold relative to the control sgRNA (Cntrl sgRNA1). Data are the mean ± s.e.m. (n = 12 independent experiments). Unpaired two-sided t-test; **P < 0.01. c–f, HGE1 disruption did not affect the expression of TSC1 (c), TBC1D7 (d) or the nearby genes RAB26 (e) or SNORD60 (f). sgRNAs targeted to HGE2 (HGE2-sgRNA1 and HGE2-sgRNA2) did not show significant changes in TSC2 or other gene levels. All data were normalized to GAPDH expression. Data are the mean ± s.e.m. (n = 3 independent experiments for each genotype). Unpaired two-sided t-test; P > 0.05. g–j, TSC2 HGE1 inhibition disrupts the RG/IP balance and neurogenesis. g, PAX6 and TBR2 immunostaining of 9-day post-transduction primary human NPCs. Anti-PAX6 and anti-TBR2 label RGs and IPs, respectively. GFP label of the merged panels were superimposed to highlight co-labelled cells (arrowheads). h, Quantification of changes in the ratio of TBR2⁺GFP⁺/PAX6⁺GFP⁺ cells. Data are the mean ± s.e.m. (n = 14 independent experiments). Unpaired two-sided t-test; ***P < 0.001. i, TUJ1 immunostaining of post-transduction primary human NPCs. TUJ1 antibodies label neurons. j, Changes in the percentage of TUJ1⁺GFP⁺ cells indicating decreased neurogenesis in TSC2-HGE1-inhibited progenitors. Data are the mean ± s.e.m. (Cntrl sgRNA1, n = 7 independent experiments; HGE1-sgRNA1, n = 6 independent experiments; and HGE1-sgRNA2, n = 7 independent experiments). Unpaired two-sided t-test; **P < 0.01 and ***P < 0.001. k,l, Altered pS6 expression in TSC2-HGE1-inhibited primary human NPCs. k, pS6 (red) immunostaining of post-transduction primary human NPCs (GFP⁺) indicating decreased pS6 in TSC2-HGE1-inhibited cells (arrowheads). l, Quantification of changes in pS6 levels. The fluorescence intensity of pS6 was normalized to GFP. Data are the mean ± s.e.m. (Cntrl-sgRNA1, n = 31 cells; HGE1-sgRNA1, n = 24 cells; and HGE1-sgRNA2, n = 18 cells). Unpaired two-sided t-test; *P < 0.05 and ***P < 0.001. m–s, TSC2 HGE inhibition disrupts the RG/IP balance and neurogenesis in cerebral organoids. m, Cerebral organoids (30 days old) were treated with TSC2 HGE sgRNAs and analysed at day 60. PAX6 and TBR2 immunolabelling of the ventricular niche (top, m) indicates decreased ratio of TBR2⁺GFP⁺/PAX6⁺GFP⁺ cells after TSC2 HGE1 inhibition (n). Data are the mean ± s.e.m. (n = 11 organoids per group). Unpaired two-sided t-test; ***P < 0.001. HOPX⁺ oRG cells were also decreased (middle, m,o). Data are the mean ± s.e.m. (Cntrl-sgRNA1, n = 10 organoids; HGE1-sgRNA1, n = 10 organoids; and HGE1-sgRNA2, n = 12 organoids). Unpaired two-sided t-test; ***P < 0.001. Quantification of TUJ1⁺GFP⁺ cells indicate decreased neurogenesis in TSC2-HGE1-inhibited organoids (m (bottom), p). Data are the mean ± s.e.m. (n = 15 organoids per group). Unpaired two-sided t-test; ***P < 0.001. At day 90, quantification of SATB2⁺GFP⁺ cells indicate decreased upper-layer neurons in TSC2-HGE1-inhibited organoids (top, q,r). Data are the mean ± s.e.m. (n = 15 organoids per group). Unpaired two-sided t-test; **P < 0.01 and ***P < 0.001. TBR1⁺GFP⁺ upper-layer neurons were not affected (bottom, q,s). Data are the mean ± s.e.m. (n = 15 organoids per group). Unpaired two-sided t-test; P > 0.05. Arrows (insets of areas outlined by boxes in m and q) indicate GFP co-labelled TBR2⁺ progenitors, HOPX⁺ oRGs, TUJ1⁺, SATB2⁺ or TBR1⁺ neurons. t–u, TSC2 HGE inhibition disrupts the RG/IP balance in human fetal cortices. t, Human fetal cortical slices at GW12–GW14 were electroporated with control or TSC2 HGE1 sgRNAs. After 7 days, slices were immunolabelled with anti-GFP, anti-TBR2 and anti-PCNA antibodies. PCNA⁺TBR2⁻ cells are radial progenitors, whereas TBR2⁺ cells are IPs. u, HGE1 disruption led to a significant decrease (*P = 0.0376; one sample t-test) in the IP/RG ratio. The IP/RG ratio of control sgRNA1 is set to 1. Number of fetal cortices: Cntrl-sgRNA = 3; HGE1-sgRNA1 = 3. Scale bar, 50 μm (g,i,m,q,t (left)), 20 μm (k) or 10 μm (t, right).

These observations suggest that human-specific regulation of TSC genes, particularly TSC2 and TBC1D7, occurs during embryonic brain development. Human-specific changes in the gene regulatory mechanisms of the TSC modulated through HGEs may contribute to the evolution of human brain development, specifically cortical neuronal expansion.

HGEs modulate TSC expression and upper-layer neuron generation

CRISPRi-mediated disruption of two TSC2 HGEs was used to test their effects on TSC2 expression in multiple unique primary human NPC donor lines^64,65. Repression of HGE1 activity resulted in around 28–39% increased TSC2 expression, whereas HGE2 disruption did not induce any changes in TSC2 expression (Fig. 6b). Furthermore, HGE1 disruption did not affect the expression of TSC1, TBC1D7 or nearby genes such as RAB26 and SNORD60 (Fig. 6c–f). This result validates the influence of the HGE1 regulatory element on TSC2, which was predicted through its chromatin interaction. Disruption of the TBC1D7 HGE did not lead to any changes in TBC1D7 expression (Extended Data Fig. 6c). Control, non-targeting guide RNAs (gRNAs) did not lead to any changes in TSC2 or TBC1D7 expression (Fig. 6b and Extended Data Fig. 9c). These observations suggest that the TSC2 HGE1 may regulate TSC2 expression in NPCs of the human fetal cortex and therefore may influence neuronal generation from these progenitors.

To test the effect of HGE1 on human cortical neuronal generation, we examined whether HGE1 induces a change in RG/IP balance and neuronal generation. HGE1 disruption in primary human NPCs led to a significant decrease in the generation of IPs, the IP/RG ratio (Fig. 6g,h) and, consequently, about a 30% decrease in the generation of TUJ1⁺ cortical neurons (Fig. 6i,j). Consistent with the increased expression of TSC2 following HGE1 inhibition, pS6 expression levels in HGE1 gRNA-expressing cells were significantly decreased (Fig. 6k,l).

To extend these observations, we treated 30-day-old cortical organoids⁶⁶ with control or HGE1 gRNA lentivirus and analysed IP/RG balance and neuron generation after 30 days (Fig. 6m). A significant decrease in IP/RG ratio, oRGs and the generation of TUJ1⁺ cortical neurons (about −33%) was observed (Fig. 6n–p). Analyses of 90-day-old organoids indicated SATB2⁺ upper-layer neuronal generation was decreased, but not TBR1⁺ deep-layer neurons (Fig. 6q–s). Last, we electroporated gestation week 12.5 (GW12.5) to GW14.5 human embryonic cortical slices⁶⁷ with control or HGE1 gRNA. After 7 days, slices were immunolabelled with anti-GFP, anti-TBR2 and anti-PCNA antibodies (Fig. 6t). PCNA⁺TBR2⁻ cells are presumably radial progenitors, whereas TBR2⁺ cells are IPs. HGE1 disruption led to a significant decrease in the IP/RG ratio (Fig. 6u). Together, these results suggest that evolutionary activation of human-specific TSC2 enhancers may serve to downmodulate TSC2 levels in the TSC and therefore enable enhanced neuronal generation in human brains.

Discussion

The dynamic balance between radial progenitors and IPs during development underlies the orderly formation of the cerebral cortex. During the neurogenic period, asymmetric proliferation of RG generates cortical neurons and neurogenic IPs. Rapid symmetric division of IPs then leads to the generation of most of the upper-layer neurons^1–6. However, the mechanisms that regulate progenitor balance and radial unit composition during corticogenesis and the significance of these mechanisms for cortical neuronal generation, organization and neurodevelopmental disorders are unclear. Here we defined a TSC-related mechanism that underlies the radial progenitor/IP balance in the developing cerebral cortex and revealed how the resultant generation of upper-layer neurons sculpts cerebral cortical organization in mice and humans.

TSC downmodulation facilitates enhanced upper-layer neuronal generation by altering the radial progenitor to IP balance and radial unit composition. scRNA-seq analyses indicated that the proportion and types of RG, IPs and upper-layer neuron cell types change in Tsc1/2 cKO mice in a manner indicative of altered RG/IP balance and enhanced generation of upper-layer neurons. Cells that express oRG markers such as HOPX and ID4 were increased in Tsc1/2 cKO mice. Consistent with this finding, MADM-based labelling of radial units (that is, RG and all their progeny) in Tsc1/2 cKO samples revealed quantitative changes in radial progenitor differentiation and upper-layer neuronal output. Furthermore, a HGE in TSC2 serves as a negative regulator of TSC2 expression and therefore enhances IP production of upper-layer neurons in humans. Although we did not identify any HGEs in TSC1, its evolutionary modulation by other genetic or epigenomic mechanisms remains to be characterized⁶⁸. Together, these observations support the hypothesis that TSC signalling modulates the RG/IP balance necessary for upper-layer neuron generation. Moreover, evolutionary downmodulation of the TSC signalling pathway leads to the enhanced generation of upper-layer neurons in humans. The evolutionary effects of the TSC in generating more neurons may not be limited to upper-layer neurons. Recent studies using patient-derived human brain organoid models showed that TSC2 mutations lead to overproliferation of a specific progenitor, caudal late intermediate progenitor (CLIP), thus resulting in the generation of excess numbers of interneurons, brain tumours and cortical malformations⁶⁹.

Developmental deletion of individual TSC genes (TSC1, TSC2 or TBC1D7) disrupts cortical development in distinctly different ways^18–25. Emx-Cre-mediated deletion of TSC1 led to an increase in cortical progenitors but a decrease in upper-layer and lower-layer neurons (CUX1⁺, BRN1⁺, FOXP2⁺ and TBR1⁺)^20,21. By contrast, hGFP-Cre-mediated deletion of TSC2 resulted in increased numbers of progenitors, fewer layer VI FOXP2⁺ neurons and no changes in CUX1⁺ upper-layer neuronal numbers¹⁸. Deletion of TSC1/2 through hGFP-Cre also resulted in a similar phenotype²⁴. Germline deletion of TBC1D7 leads to increased thickness of cortical layers IV/V (CTIP2⁺) and layer VI (TBR1⁺)³³. In contrast to the mouse models, homozygous deletion of TSC1 or TSC2 in human embryonic stem cells did not affect the generation or proliferation of PAX6⁺ forebrain progenitors. Spheroids derived from TSC1^−/− and TSC2^−/− human embryonic stem cells exhibit reduced expression of neuronal markers (MAP2 and NeuN) and increased expression of glial markers (GFAP and S100B)^70,71. Analyses of cortical tubers from patients with tuberous sclerosis complex revealed that dysmorphic neurons in cortical tubers express CUX2 (layers II/III), SATB2 (layers II–V), RORB (layer IV) and ER81 (layers V–VI) throughout all layers of the cortex. Giant cells of cortical tubers express mainly SATB2, but also RORβ or ER81 (ref. 28). Vimentin and brain lipid binding protein (BLBP), markers for RG progenitors, are highly expressed in tubers²³. Together, these studies underscore the significance of the TSC pathway for the appropriate development of the cerebral cortex. Furthermore, they indicate that the developmental stage and cell type in which inactivation of the specific TSC gene occurs affects the outcome in a manner specific to the TSC gene member and species. Expanding on and in contrast to these observations, our results revealed how the TSC as a collective regulates the balance of progenitors, changes the radial unit composition and upper-layer neuron generation to shape the organization and connectivity of the cerebral cortex. Although our Tsc1/2 cKO mouse models do not recapitulate the exact TSC genetics or pathology of humans, they provide valuable insights into essential cellular developmental programs that underlie normal cerebral cortical development and evolution.

Cross-species comparison of cell-type proportions has indicated that the upper-layer, intratelencephalic neuronal population substantially expanded in humans⁹. Consistent with an evolutionary role for the TSC in primate and human brain expansion, human oRGs, which enable rapid neuronal generation, are distinguished by high levels of pS6 (refs. 41,72). Multi-omic profiling of human malformations of cortical development (MCD) indicates strong association of MCDs with mTOR deregulation, which is linked with the disrupted function of excitatory neurogenic pools (dividing RG, IPs and their excitatory neuronal descendants) in MCD⁷³. Greater activation of mTOR signalling pathways is associated with an increased risk of schizophrenia, epilepsy and ASD, as well as cortical malformations (for example, cortical dysplasia and hemimegalencephaly)^{26,27,34,74,75}. Moreover, genomic landscape analyses of TSC-associated cortical tubers indicated increased expression of neurogenic differentiation markers such as NeuroD6 (ref. 30) and overrepresentation of gene ontology modules such as neurogenesis, nervous system development, dendrite development and synaptic signalling⁷⁶. These observations raise the possibility that the downmodulation of TSC activity and the resultant changes in the tempo of mTOR-mediated cellular metabolism during evolution may prolong the neuronal development program and neurogenic period^10,77–80. Consequently, the evolution of a larger, more cellularly complex cerebral cortex is enabled. Disruption of this mechanism in humans may trigger cortical malformations and associated neurodevelopmental disorders.

Collectively, our results revealed how appropriate RG/IP balance, radial unit formation and upper-layer neuron generation are organized through the TSC signalling pathway to promote normal cortical development. Moreover, we demonstrated how changes in this process can trigger brain malformations and neural circuit changes relevant to neurodevelopmental diseases such as tuberous sclerosis complex, epilepsy and ASD. Further deconstruction of the evolutionary mechanisms that underlie TSC expression levels and their impact on progenitor balance, diversity, radial unit composition and the resultant upper-layer neuronal organization and connectivity presents a challenge and opportunity. Such studies will provide deeper insights into the salient mechanisms necessary for normal human brain development and help to uncover the origins of cortical malformations in neurodevelopmental disorders triggered by progenitor imbalance.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-025-09810-5.

Methods

Mice

This study is based on data from mice at various developmental stages (E16, E18, P0, P5, P7 and P21) and includes both male and female mice. Mice were in a mixed 129 and C57/Bl6 background and cared for according to the University of North Carolina’s Institutional Animal Care and Use Committee guidelines. The light–dark cycle in the vivarium is 7–7 h, with 18–23 °C ambient temperature and 40–60% relative humidity. Animals were housed in groups of three adults per cage. Tsc1/2 were conditionally inactivated in cortical progenitors by crossing Tsc1^lox/loxTsc2^lox/lox mice with Tsc1^lox/+Tsc2^lox/+cre+ mice^81,82. The Emx1-Cre line was used to inactivate Tsc1/2 in cortical radial progenitors⁸³. The Tbr2-Cre^ER line was used for tamoxifen-induced inactivation in intermediate progenitors⁸⁴, and the Ai14 line was used as a reporter⁸⁵. To generate MADM-11^GT/TGTsc1^lox/loxTsc2^lox/lox mice, MADM-11^TG/TG Tsc1^lox/+Tsc2^lox/+ mice were crossed with MADM-11^GT/GTTsc1^lox/loxTsc2^lox/lox mice⁸⁶.

Immunohistochemistry

Immunohistochemical labelling of embryonic and postnatal brain sections or isolated cortical cells was performed as previously described¹². The following primary antibodies were used: cleaved-caspase 3 (1:400, rabbit, 9661, Cell Signaling); CTIP2 (1:500, rat, ab18465, Abcam); CUX1 (1:500, rabbit, sc-13024, Santa Cruz Biotechnology); GFAP (1:250, rabbit, Z0334, DAKO); GFP (1:1,000, chicken, ab13970, Abcam; 1:1,000, chicken, GFP-1020, Aves); HOPX (1:500, rabbit, 11419-1-AP, Proteintech); NeuN (1:500, mouse, MAB377, Chemicon); PAX6 (1:500, rabbit, 901301, BioLegend; 1:500, mouse, MA1-109, Invitrogen); PCNA (1:200, mouse, CBL407, Millipore); phospho-histone H3 (1:500, rabbit, 06570MI, Millipore); phospho-S6 ribosomal protein (Ser240/244) (D68F8)XP (1:1,000, rabbit, 5364s, Cell Signaling Technology); phospho-vimentin (1:1,000, mouse, ab22651, Abcam); POU3F2 (1:500, mouse, sc-393324, Santa Cruz Biotechnology); PTPRZ1 (1:500, rabbit, HPA015103, Sigma-Aldrich); RC2 (5 μg ml⁻¹, mouse IgM, AB 531887, Developmental Studies Hybridoma Bank, University of Iowa); RFP (1:1000, rabbit, 600-401-379, Rockland); SATB2 (1:500, mouse, ab51502, Abcam); TBR1 (1:500, rabbit, ab31940, Abcam); TBR2 (1:500, rat, 14-4875, eBioscience; 1:250, sheep, AF6166, Fischer Scientific; 1:250, rabbit, ab23345, Abcam); and TUJ1 (anti-TUBB3) (1:1,000, mouse, 1409, Stemcell Technologies; rabbit, 18020, BioLegend). Appropriate Cy2, Cy3 or Alexa dye-conjugated secondary antibodies (Jackson ImmunoResearch, Molecular Probes) were used to detect primary antibody binding. DAPI (Sigma-Aldrich, D9542) was used as a nuclear counterstain. Images were obtained using a Zeiss LSM780 confocal microscope. Quantification of cells labelled with PH3, caspase, PAX6, TBR2, TUJ1, NeuN or SATB2 in the cerebral cortex was performed as previously described¹². Cell counts are presented as cells per unit area (10,000 or 100,000 μm²).

In utero electroporation

In utero electroporation of shRNAs was performed as previously described¹². Plasmid DNA (0.5 μl of 2 μg μl⁻¹) was injected into the lateral ventricle of mice at E14.5 and electroporated into the VZ (five 50-ms pulses of 40 V at 950-ms intervals). Cortices were removed from treated animals for analysis at E18. The neuronal position in the cortex was measured as previously described¹².

To label single radial units, lateral ventricles of E14 embryos were electroporated with low-titre pcDNA-Cre (0.5 μl of 1 μg μl⁻¹ DNA) to express Cre recombinase in isolated RG¹². Cortices were collected at E16 or P21 for immunolabelling and analyses of isolated radial units.

Generation and characterization of Tbc1d7-specific shRNA

Tbc1d7 unique target sequence oligonucleotides (shRNA-1: TAGCGGTAGAAATACTATTAA; shRNA-2: ACAAGTACAGGGACGCTTTAC) and a nontargeting scramble shRNA sequence oligonucleotide (GCGCGATAGCGCTAATAATTT) were subcloned into pU6-shRNA-CAG-GFP vectors (Addgene, 10878 and 11150), which contain chicken β-actin promoter-driven eGFP and a U6 promoter for shRNA transcription. To test the efficiency of shRNA-mediated Tbc1d7 knockdown, targeting constructs were transfected into N2A cells using Lipofectamine 3000 (ThermoFisher) per the manufacturer’s instructions. N2A cells were authenticated (STR profiling; mycoplasma negative) by the American Type Culture Collection (ATCC). High transfection efficiency (>80% of total cells on the dish) was confirmed by eGFP expression. After 48 h, total RNA was prepared using a RNeasy Plus Mini kit (Qiagen), and the knockdown efficiency of shRNAs (fold change compared with control; shRNA-1 = −97 ± 0.05%; shRNA-2 = −96 ± 0.8%) was measured by qPCR. Gapdh expression was measured as a control.

The following qPCR primers were used: Tbc1d7-Forward: GGGTTTCGTGGTGTCGAAGA and Tbc1d7- Reverse: GTGTCATGGTGCGGAGGTAA; and Gapdh-Forward: CATCACTGCCACCCAGAAGACTG and Gapdh-Reverse: ATGCCAGTGAGCTTCCCGTTCAG.

scRNA-seq

scRNA-seq was performed as previously described^37,38. In brief, P0 cerebral cortical neurons from control and Tsc1/2 cKO mice were dissociated and stored in 80% methanol at −80 °C. Cells were then lysed in TRI-reagent and RNA was extracted using a Direct-zol kit (Zymo Research, R2051) with on-column DNase treatment, following the manufacturer’s instructions. Strand-specific total RNA-seq libraries from ribosomal-RNA-depleted RNA were prepared using a 10xG 3Prime v3 single-cell RNA kit (10x Genomics) according to the manufacturer’s protocol. Libraries were sequenced single end for a minimum of 76 bp for each index to a depth of approximately 40 million host genome reads on an Illumina NextSeq 500 instrument at the Yale Center for Genomic Analysis.

Raw and processed scRNA-seq data, cell metadata and cluster annotations have been deposited into the Gene Expression Omnibus (GEO) under accession GSE281619.

scRNA-seq data analysis

FASTQ files were processed using alevin with GENCODE vM25 transcript annotations to generate cell × gene matrices⁸⁷. Data were then imported into R (v.3.6) and Seurat objects were created for each individual sample before being merged⁸⁸. Cells were filtered such that there were at least 1,000 UMIs and 500 genes detected, and fewer than 10% of the transcripts were mitochondrially contributed. Data were then scaled and normalized using scTransform and integrated using Seurat⁸⁹. Clusters were identified using Louvain–Jaccard clustering with multilevel refinement (resolution = 1.5) performed on the top 100 principal components. Marker genes of each cluster were identified using Presto (https://github.com/immunogenomics/presto), and differential expression analysis was performed using distinct with log₂ counts per million in R (v.4.1)⁹⁰. Differential abundance testing comparing cellular proportions for each genotype was performed using the propeller test in the speckle package⁹¹.

We determined whether the loss of TSC1 and TSC2 significantly affected neurodevelopmental disease-relevant genes by intersecting our data with gene ontology databases using g:Profiler, as well as with the Simons Foundation ASD gene database and other large-scale genome-wide association study databases^40,53,58.

Inference of developmental trajectories

Pseudotime trajectory analysis was performed using Slingshot (v.2.16.0)⁹², filtering out genes expressed at low levels and using two-dimensional embedding based on the principal component analysis of z scores of gene expression patterns in RG, IPs and upper-layer PN (projection neuron) cells in control and Tsc1/2 cKO samples. The statistical comparison of cell density along pseudotime between cKO and control of each trajectory identified was conducted using a two-sided Kolmogorov–Smirnov test.

Analysis of the conservation and expression TSC genes

Ensembl (https://uswest.ensembl.org/index.html) was used to identify conserved regions of TSC1, TSC2 and TBC1D7 across different species. Ensembl gene trees were generated to illustrate the evolutionary history of these genes and to predict orthologous relationships between them. Expression of TSC genes during cerebral cortical development was examined using RNA-seq data from mouse and human embryonic and adult brains^37,43–47.

Analysis of epigenetic modifications of TSC genes

To investigate H3K27me3 and H3K27ac ChIP–seq signals and peaks around the TSSs of TSC genes in humans and mice, publicly available ChIP–seq data (GEO identifiers: GSE90694, GSE62193, GSM818043, GSM1013151, GSE96300, GSE101250, GSE96421, GSE96264, GSE96178, GSE96414, GSE96463, GSE95935, GSM896165, GSM818033, GSE95944, GSE96183, GSE96292, GSE177130, GSE178047 and GSE176723) from embryonic and postnatal mouse forebrain, neurons and NPCs of mouse, adult and fetal human brain tissue, germinal matrix of the human fetal brain, human cortical organoids, and in vitro neurosphere cultures derived from the ganglionic eminence of human were used^53–55.

Identification of HGEs of TSC genes

HGEs were acquired from a previous study⁵⁶ and were identified on the basis of an increase in H3K4me2 and H3K27ac signals compared with those in rhesus and mouse. To establish the regulatory role of potential HGEs to the promoters of TSC genes, the correlation between promoters and other ATAC–seq peaks within 200 kb of the TSS for each TSC gene was calculated in both neurons and neuronal progenitors⁶³. HGEs were selected if they overlapped with ATAC–seq peaks and significantly correlated with the promoters of TSC genes or if HGEs overlapped with ATAC–seq peaks present in single-cell chromatin-accessibility data from developing human cortical tissue⁶¹. Hi-C data from the cortical plate and germinal zone of the developing human cortex (post-conception week 15 (PCW15) to PCW17) were used to confirm the interaction between the promoters of the TSC genes and the potential HGEs^{57,60,63–65}.

Human NPC culture

Fetal tissue (GW15–GW18) collection and neurosphere collection were done as previously described^12,63,65,93 and cells were derived from frozen primary human NPC stocks. Cells from both male and female donors were used. Cells were kept in proliferation medium (Neurobasal A (Invitrogen) supplemented with 10% BIT (Stem Cell Technologies), antibiotic–antimycotic (Thermo Fisher Scientific), GlutaMAX (Gibco) and heparin (1 μg ml⁻¹; Sigma)) with freshly added EGF, FGF2 and PDGF (each at 20 ng ml⁻¹; Invitrogen) and LIF (2 ng ml⁻¹; Invitrogen) and passaged when confluent on Matrigel-coated plates. Before transduction, NPCs were plated onto polyornithine-coated and laminin-coated German glass coverslips at around 80,000 cells per cm².

CRISPR–Cas9-mediated transcriptional inhibition of HGEs and lentivirus preparation

gRNAs targeting the HGEs of TSC2 and TBC1D7 were designed using GT-Scan⁹⁴. Four gRNAs of TSC2 HGE and six gRNAs of TBC1D7 HGE were initially selected. Two nontargeting gRNAs were used as controls (Supplementary Table 1). The gRNAs were cloned into pLV-hU6-UbC-dCas9-KRAB-T2a-GFP (https://www.addgene.org/71237/) with Bsmb1 restriction sites. Lentiviruses for each plasmid were generated by co-transfecting the lentiviral constructs with psPax2 packaging plasmid (https://www.addgene.org/12260/) and pMD2.G envelope plasmid (https://www.addgene.org/12259/) in HEK293T cells using FuGENE HD transfection reagent (Promega). HEK293T cells were authenticated (STR profiling; mycoplasma negative) by the ATCC. At 12–15 h after transfection, the lentivirus packaging medium was removed and replaced with proliferation medium. The viral supernatant was collected after 24 h and 48 h after transfection, centrifuged at 1,250 rpm for 5 min at room temperature, filtered (0.45 μm, Olympus plastics) and then frozen at −80 °C until use. Viral titres (Supplementary Table 2) were calculated using a qPCR Lentivirus Titer kit (Applied Biological Materials).

Primary human NPC lentivirus infection and analysis of TSC2 and TBC1D7 expression

NPCs were plated in 24-well plates at a density of 50,000 cells per well. The next day, the cells were infected with pLV-hU6-UbC-dCas9-KRAB-T2a-GFP gRNA viruses (Supplementary Table 2) with a multiplicity of infection of 8, and the medium was replaced with fresh proliferation medium after 12–16 h. Cells were cultured for an additional 7 days before RNA extraction. Total RNA was isolated with a RNeasy Mini kit (Qiagen), and 200 ng of the RNA was reverse transcribed using a HiScript III 1st Strand cDNA Synthesis kit (Vazyme). Gene expression patterns were then measured by RT–qPCR using PowerUP SYBR Green master mix (Applied Biosystems) and a QuantStudio 3 Real-Time PCR system (Thermo Fischer). Gene expression values for each of the HGE gRNAs were compared with the nontargeting gRNA controls, with GAPDH as the normalizer. gRNAs with the highest impact on TSC gene expression were used for further analyses. Primers for qPCR are as follows: GAPDH-F (5′-CATCACTGCCACCCAGAAGACTG-3′), GAPDH-R (5′-ATGCCAGTGAGCTTCCCGTTCAG-3′), TSC1-F (5′-TGTTGTGATCGAGTGTGCCA-3′), TSC1-R (5′-AACAACATCAGCCGAGACGT-3′), TSC2-F (5′-TGGTCCGCAGCAGGATACAG-3′), TSC2-R (5′-ACAGGAGACCTCTTCGGGAC-3′), TBC1D7-F (5′-GTTGCCAAAAGCGTTTGAACA-3′), TBC1D7-R (5′-CATCCCGCAAAGCACCTCTT-3′), SNORD60-F (5′-TTGCTTTGACTTCTGACACCTCGTA-3′), SNORD60-R (5′-GTCTTGCTAAATAATCAGACTGCACG-3′), RAB26-F (5′-GTCTGCTGGTGCGATTCAAG-3′), RAB26-R (5′-GCATGGGT AACACTGCGGA-3′), HERC2-F (5′-CACCTGTGTATAGAGCCAAGTCA-3′), HERC2-R (5′-TTCAACCTCAAGGCTGAGAGT-3′), ID4-F (5′-GGACAGCTCCAGTAACTGCAC-3′), and ID4-R (5′-CTGCATCTTTGATGTGAGGCG-3′).

Cerebral organoid assay

A feeder-free cortical organoid differentiation protocol known to produce reproducible cell-type proportions in and across induced pluripotent stem (iPS) cell lines was used to generate cortical organoids⁶⁶. Thawed iPS cells (PGP-1) were passaged twice, seeded at 9,000 cells per well into a 96-well ultralow attachment plate, and compounds for Nodal–activin and WNT inhibition were applied to direct cells towards a dorsal telencephalic fate^95,96. After 18 days of differentiation, organoids were transferred to a 90 mm ultralow attachment dish on an orbital shaker to allow further growth and diffusion of the medium. After 30 days, organoids were transduced with pLV-hU6-UbC-dCas9-KRAB-T2a-GFP gRNA viruses (Supplementary Table 2) with a multiplicity of infection of 100. Twenty fours later, medium (cortical differentiation medium (CDM2); DMEM/F12, GlutaMAX, N2, CDLC, Fungizone and penicillin–streptomycin) was replenished with fresh medium. All organoids were collected for radial progenitor, IP and neuronal labelling and analyses at day 60 or 90. In brief, organoids were fixed in 4% paraformaldehyde, sunk in 30% sucrose, embedded in OCT, sectioned at 18 μm thickness and immunolabelled with anti-PAX6, anti-TBR2, anti-HOPX, anti-TUJ1, anti-SATB2 and anti-TBR1 antibodies as described above.

Human fetal brain tissue assay

Human fetal brain tissues (GW12–GW14) were obtained from the Helsinki University Hospital, with approval by the ethics committee of the Hospital District of Helsinki and Uusimaa (HUS/1170/2021). After dissecting the parietal cortex, neocortical tissues were electroporated with a mixture of 4 μg μl⁻¹ control sgRNA or HGE1-sgRNA1, 0.1% Fast Green (Sigma) and 5% glycerol in PBS as previously described⁶⁷. The electroporated tissues were cultured for 7 days in a humidified atmosphere of 5% CO₂, 60% O₂ and 35% N₂ (ref. 67). Fetal human brain tissues were then fixed with 4% paraformaldehyde overnight at 4 °C, washed with PBS followed by 30% sucrose in PBS, embedded in OCT compound (Sakura Finetek) and cut into 40-μm-thick sections. Sections were immunolabelled with anti-GFP, anti-PCNA and anti-TBR2 antibodies as previously described⁶⁷.

Images were acquired using LSM980 (Zeiss) with a ×40 objective. Tiling of Images was done using ZEN (v.2.3) software (Zeiss). All cell counting was performed in a blinded manner with ImageJ2 (v.2.9.0). GFP⁺ cells with a PNCA signal but not a TBR2 signal were categorized as RG, whereas GFP⁺ cells with a TBR2 signal were categorized as IPs. The IP/RG ratio of the control group was set to 1. One sample t-test was performed using GraphPad Prism 10.6.0 (GraphPad Software).

Quantification and statistical analysis

General statistical analysis.

Excel 2023 (v.16.77), GraphPad Prism 10.6.0 and R (v.4.1) language were used for data analyses. Significance was determined using unpaired two-tailed t-tests for comparisons between two groups and using analysis of variance with Tukey–Kramer’s test for comparisons among multiple groups. All experiments were independently repeated (biological repeats) at least three or more times. No a priori power analyses were performed. All data are expressed as the mean ± s.e.m. Statistical details, including P values, are indicated in the main text or figure legends.

Quantification of MADM-labelled radial units and dendritic branches.

For the quantification of MADM-labelled clones at E16, isolated RG clones located in the dorsal region of the cortex were confocally imaged (serial z axis scans) and analysed. The number of radial progenitors, IPs and neurons aligned along a single mother RG were counted. To quantify MADM-labelled clones at P21, isolated RG were electroporated at E14 with Cre vectors to enable single radial unit labelling. Next, 5-μm interval z scans of tissue sections, spanning 100 μm thickness, were used to reconstruct individual neuronal clones. Upper-layer and deeper-layer neurons per clone and clonal distribution width were quantified. The dendritic branching index was measured as the percentage of area occupied by the MADM⁺ dendritic branches per 450 μm² cortical area.

Quantification of the final placement of neurons.

To measure the final placement of neurons, the cerebral cortical wall was divided into ten equal bins from the ventricular surface to the pial surface. The number of GFP⁺ neurons in each bin was counted and presented as the percentage of the total number¹².

Quantification of transduced primary human NPCs.

Transduced control and HGE sgRNA-expressing cells (GFP⁺) were labelled with anti-PAX6, anti-TBR2, anti-TUJ1 or anti-SATB2 antibodies, and changes in cell-type proportions per unit area (50 mm²) were quantified as previously described¹².

Ethics statement

Human fetal brain tissue studies were approved by the ethics committee of the Hospital district of Helsinki and Uusimaa (HUS/1170/2021).

Reporting summary

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

Extended Data

Extended Data Fig. 1 |. Altered proliferative niche and astrogliosis in TSC1/2 cKO.

(A,B) Increased density of PH3⁺ cells in the ventricular zone of TSC1/2 cKO. (C) Quantification of changes in PH3⁺ cells (n = 6 mice/group). (D-E) An atypical collection of Pax6⁺, Tbr2⁺ double-positive cells in TSC1/2 cKO VZ. (F) Quantification of changes in Pax6⁺, Tbr2⁺ cells (n = 3 mice/group). (G-H) No changes in caspase-3⁺ apoptotic cells are evident in the TSC1/2 cKO cortex. (I) Quantification of caspase-3⁺ cells (n = 6 mice/group). (J,K) Increased density of GFAP⁺ astrocytes in TSC1/2 cKO cortex. (L) Quantification of changes in GFAP⁺ cells (n = 5 mice/group). Data shown are mean ± SEM. Unpaired two-sided t-test; **p < 0.01; ***p < 0.001 [p = 0.0026 (C), 0.0087 (F), 0.804 (I), 0.0001 (L)]. VZ, ventricular zone. Scale bar: A-B (50 μm), D-E (10 μm); G-H (50 μm); J-K (20 μm).

Extended Data Fig. 2 |. Changes in brain weight and cortical size in TSC1/2 cKO.

(A,B) Increased brain size and weight (C) in TSC1/2 cKO (n = 5 mice/group). (D-F) Increased width (yellow bar) of cerebral cortical wall in TSC1/2 cKO (n = 8 mice/group). (G) Change in cortical size is also reflected in increased cortical circumference [somatosensory cortex level] in TSC1/2 cKO (n = 7 mice/group). Data shown are mean ± SEM. Unpaired two-sided t-test; *p < 0.05, **p < 0.01, and ***p < 0.001 [ p = 0.0147 (C), 0.0036 (F), 0.0009 (G)]. Scale bar: A-B (1 mm); D-E (500 μm).

Extended Data Fig. 3 |. Effect of TSC1/2 deletion on pS6 expression.

(A,B) Increased pS6 expression in TSC1/2 cKO cortex. (C) Quantification of changes in pS6 expression in TSC1/2 cKO cortex (n = 6 mice/group). (D) Immunoblot analysis and quantification (E) of increased pS6 expression in TSC1/2 cKO cortex (n = 4 mice/group). (F,G) Increased pS6 expression in TSC1/2 cKO RGCs (arrowhead). (H) Quantification of changes in pS6 expression in TSC1/2 cKO RGC soma (n = 25 soma). RGCs were co-labeled with RC2 antibodies. CP, cortical plate; IZ, intermediate zone; VZ, ventricular zone. Data shown are mean ± SEM. Unpaired two-sided t-test; ***p < 0.001 [ p = <0.0001 (C), 0.0001 (E) , <0.0001 (H)]. Scale bar: A-B (100 μm); F-G (50 μm). Sample RGC soma are outlined (dotted lines; F-G). A.U., arbitrary unit of pS6 intensity normalized to background. For immunoblot source data, see Supplementary Fig. 1.

Extended Data Fig. 4 |. Increased width of the corpus callosum following deletion of TSC1/2 in intermediate progenitors.

(A,D) Consistent with the increase in upper layer neurons (tdTomato⁺) in TSC1/2^Tbr2 cKO, the width of the corpus callosum (arrow), containing the axonal projections of these neurons, is increased. (E) Quantification of increased width of tdTom labeled corpus callosum (CC). Data shown are mean ± SEM (n = 10 mice/group). Unpaired two-sided f-test; ***p < 0.0001. Scale bar: A-D (100 μm).

Extended Data Fig. 5 |. Altered cortical size, laminar organization, neuronal morphology, neuronal projections, and connectivity in MADM labeled Tsc1/2 Emx-cKO cortex.

(A,B) Expansion of cortical size in MADM labeled Tsc1/2Emx-cKO brains. (C,D) TSC1/2 deletion leads to disrupted laminar organization and neuronal misplacement (circle; D) in MADM labeled Tsc1/2 Emx-cKO brains. (E,F) Changes in the morphology and increased branching of MADM labeled upper layer neurons (compare areas indicated by asterisk [E, F]). (G) Quantification of increased neurite branching in Tsc1/2 Emx-cKO cortex (n = 12 mice/group). (H,I) Consistent with the expanded upper layer neurons, the thickness of corpus callosum (CC) containing axons of upper layer neurons traversing to the contralateral cortex is increased in Tsc1/2 Emx-cKO. Bar (H, I) indicates callosal width. (J) Quantification of increased callosal width in Tsc1/2 Emx-cKO. Data shown are mean ± SEM (n = 5 mice/group). Unpaired two-sided t-test; **p < 0.01 and ***p < 0.001 [ p = 0.0002 (G), 0.0022 (J)]. Scale bar: A-B (200 μm); C-D (50 μm); E-F (20 μm); H-I (100 μm).

Extended Data Fig. 6 |. Effect of inactivation of TSC1, 2, and TBC1D7.

Embryonic cortices were electroporated with control and TBC1D7 shRNAs at E14.5 and analyzed at E18.5. (A-H) TBC1D7 knockdown (GFP⁺) further disrupts cortical malformation in TSC1/2 cKO. Compared to control (A-B), expanded upper layers and upper layer neuronal misplacement are evident in TSC1/2 cKO (C-D; arrow [A vs. C]). Inactivation of TBC1D7 in TSC1/2 cKO background (E-H) further disrupts cortical plate organization (compare areas indicated by arrows; C vs. E, G) and leads to ectopic collection of neurons near the ventricular zone (compare areas indicated by asterisks; C vs. E, G). (B, D, F, H) Quantification of GFP⁺ cell position in the developing cortical wall. Data shown are mean ± SEM (n [mice]: control = 5, TSC1/2^cKO = 6, TSC1/2^cKO+Tbc1d7shRNA1 = 9, and TSC1/2^cKO+ Tbc1d7shRNA2 = 7). Two-way ANOVA; Control [B] vs. TSC1/2^cKO [D], TSC1/2^cKO [D] vs. TSC1/2^cKO+Tbc1d7 shRNA1 [F] or TSC1/2^cKO+Tbc1d7 shRNA2 [H], *P<0.0001. (I-M) Co-labeling of GFP⁺ neurons in the cortical plate with upper layer neuronal marker Satb2, indicates increased GFP⁺, Satb2⁺ upper layer neurons in TSC1, 2, and TBC1D7 deficient cortices (K, L, M; yellow arrowheads) as compared to control (I, M; white arrowheads) or TSC1/2 cKO cortices (J, M; purple arrowheads). Data shown are mean ± SEM (n [mice]: Control = 5, TSC1/2^cKO = 5, TSC1/2^cKO+Tbc1d7shRNA1 = 7, and TSC1/2^cKO+Tbc1d7shRNA2 = 7). Unpaired two-sided t-test; **p < 0.01 and ***p < 0.001 ([M: Control vs. TSC1/2^cKO, p = 0.0003; TSC1/2^cKO vs. TSC1/2^cKO+Tbc1d7 shRNA1, p = 0.0017, or TSC1/2^cKO+ Tbc1d7 shRNA2, p = 0.0024]) (N-R) Co-labeling of GFP⁺ neurons with deep layer neuronal marker Tbr1, indicates no changes in GFP⁺, Tbr1⁺ neurons. Data shown are mean ± SEM (n [mice]: Control=5, TSC1/2^cKO = 5, TSC1/2^cKO+Tbc1d7shRNA1 = 5, and TSC1/2^cKO+Tbc1d7shRNA2 = 5). Unpaired two-sided t-test; **p < 0.01 and ***p <0.001 ([R: Control vs. TSC1/2^cKO, p = 0.9385; TSC1/2^cKO vs. TSC1/2^cKO+ Tbc1d7 shRNA1, p = 0.7682, or TSC1/2^cKO+Tbc1d7 shRNA2, p = 0.7790]). CP, cortical plate; IZ, intermediate zone; SVZ, subventricular zone; VZ, ventricular zone. Scale bar: A, C, E, G (50μm); I-L (20μm); N-Q (40 μm).

Extended Data Fig. 7 |. Changes in the average expression of the ASD genes in RG, IP, and UL cells of control and TSC1/2 cKO cortices.

(A) Log2-fold changes in high confidence ASD gene expression levels in cKO cells compared to controls are illustrated (P < 0.05, two-tailed hierarchical permutation test). High confidence ASD gene list is from SFARI ASD gene database (https://gene.sfari.org/). SFARI genes with no significant expression changes in any of the 3 cell classes are not included. RGC, radial glial cells; IP, intermediate precursors; UL, upper layer neurons. Circle size indicates the value of log fold change. Red-blue gradient indicates the direction of change.

Extended Data Fig. 8 |. Expression profile and epigenomic regulation of TSC genes.

(A) Sequence conservation analysis shows that the TSC complex genes are highly conserved between mice and humans. (B) Expression profile of the TS complex genes in human and mouse neural cell types⁴⁴. Data shown are mean ± SEM (n = 10 [human], 22 [mouse] brain cell clusters). Unpaired two-sided t-test; **p < 0.01 and ***p < 0.001 [ p = 0.0083 (TSC1), <0.0001 (TSC2)]. (C) Histone modifications around the transcription start site of TS complex genes. H3k27me3 peaks (orange box) are present in TS complex genes only in the human embryonic brain. In contrast, H3k27ac peaks (teal box) are present in Tscl, Tsc2, and Tbc1d7of the mouse embryonic and adult brain. They are also present in TSC1, TSC2, and TBC1D7of the human embryonic and adult brain tissue. RG, radial progenitors; IP, intermediate progenitors.

Extended Data Fig. 9 |. Identification of candidate gene regulatory elements in TBC1D7 and TSC1.

(A) ATAC-seq peaks showing a candidate HGE (teal box; TBC1D7-P1) in the TBC1D7 promoter⁵⁶. The black lines indicate the significantly correlated peak pairs across a population of human neural progenitors and their correlation values (r). (B) HGEs were not detected near TSC1 in ATAC-seq or Hi-C data from neurons, progenitors, or germinal zone (GZ). (C) qPCR analysis of relative TBC1D7 expression in hNPCs after lentiviral transduction with dCas9-KRAB-gRNAs targeted to the candidate TBC1D7 human-gained enhancer. No changes in TBC1D7 expression were detected. Data shown are mean ± SEM (n = 9 unique donors). Unpaired two-sided t-test p >0.05 (0.6363 [Cntrl sg vs sg1]; 0.6559 [Cntrl sg vs sg2]; 0.0702 [Cntrl sg vs sg3]; 0.3921 [Cntrl sg vs sg4]; 0.3130 [Cntrl sg vs sg5]; 0.4876 [Cntrl sg vs sg6]).

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1038/s41586-025-09810-5.

Data availability

Raw and processed scRNA-seq data, cell metadata and cluster annotations have been deposited into the GEO under accession GSE281619. The SFARI Gene database (https://gene.sfari.org/) was used identify human genes associated with ASD. Source data are provided with this paper.