mTORC1 drives cell-autonomous astrocyte reactivity in tuberous sclerosis

Tuberous sclerosis complex (TSC) is a multisystem developmental disorder caused by mutations in TSC1 or TSC2⁴. TSC leads to hamartomas in several organs and a constellation of neurological and psychiatric conditions, including severe, early-onset epilepsy¹. Malformations of cortical development, known as cortical tubers, are a hallmark of TSC⁵. A subset of tubers and surrounding regions can become seizure foci, and surgical removal can be beneficial for individuals with intractable epilepsy⁶. The origin of tuber cells and the molecular changes that lead to their abnormal development are open questions. On a cellular level, tubers contain heterogeneous populations of dysmorphic neurons, dysplastic and gliotic astrocytes, activated microglia and giant or balloon cells within a background of normal-appearing cells^7,8,9. Although much research has focused on neuronal mechanisms in TSC^10,11,12,13, some studies have suggested that the primary pathological cells within tubers may be astrocytes³. Indeed, gene-expression analysis of tuber tissue has revealed increased expression of astrocyte-enriched genes and prominent neuroinflammation^14,15. However, because tuber resections are performed in individuals with intractable epilepsy, it has been unclear whether glial changes are a cause or consequence of continuing seizures. A further challenge is the lack of a clear cell type signature within tuber lesions, as giant cells can express progenitor cell markers as well as neuronal and glial proteins^8,16,17.

A prevailing hypothesis is that tubers arise by means of a ‘second-hit’ mechanism, whereby patients with a germline heterozygous mutation acquire a somatic mutation that inactivates the remaining functional TSC1 or TSC2 allele in a subset of neural progenitor cells². Loss of the TSC1–TSC2 protein complex leads to hyperactivation of mTOR complex 1 (mTORC1)¹⁸, a key signalling node that regulates cellular anabolic and catabolic processes¹⁹. Whereas second-hit progenitor cells can survive and proliferate, they are impaired in their differentiation and development, giving rise to focal regions of abnormal cells. This mechanism explains the localized, stochastic and variable nature of tubers, as well as the observation that only a subset of cells within tubers show high mTORC1 activity^7,20. Second-hit mutations are frequently observed in TSC-related hamartomas and have been identified in some resected tubers from patients^14,21,22,23. In addition, somatic mutations in mTOR pathway regulators are often found in focal cortical dysplasia type II, which has overlapping histological and clinical features with TSC^24,25. However, not all tubers have an identified second-hit mutation, and it has been suggested that tuber-like cells may arise from heterozygous cells in some contexts²⁶.

Here we use a human brain organoid model to recapitulate a second-hit mutation in TSC2 in neural progenitor cells. We demonstrate that TSC2 deficiency cell-autonomously induces the formation of reactive astrocytes showing distinct molecular and morphological alterations, including upregulation of genes associated with neurodegeneration. Cyclic immunostaining of tubers from patients confirms a distinct, mTORC1-hyperactive cell population expressing these same reactive astrocyte markers. Because glial alterations emerge early in developing organoids, our findings provide evidence that astrocyte dysfunction is an early driver of TSC pathophysiology, rather than solely a secondary consequence of chronic epilepsy.

To understand the developmental processes leading to malformations of cortical development in TSC, we performed single-cell RNA sequencing (scRNA-seq) of a genetically mosaic human brain organoid model that recapitulates a second-hit mutation²⁷. These cells have a loss-of-function mutation in one allele of TSC2 (exon 5 deletion), and a conditional allele that can be rendered non-functional by Cre recombinase (Extended Data Fig. 1a–d). Second-hit cells are identified by red fluorescence through a Cre-dependent tdTomato knock-in reporter (Fig. 1a and Extended Data Fig. 1c). We differentiated TSC2^{c/−;LSL-TdTom} human embryonic stem (hES) cells into brain organoids using a cortical differentiation protocol that recapitulates developmental transitions, including neural progenitor proliferation, neurogenesis and subsequent gliogenesis and cellular maturation²⁸. Cre recombinase was delivered by means of lentiviral transduction at day 8 to delete TSC2 in a subset of neural progenitor cells in the organoid (Fig. 1b). This timing was chosen to model a somatic mutation during early cortical progenitor expansion. We performed fluorescence-activated cell sorting (FACS) and scRNA-seq of TSC2^{c/−;LSL-TdTom} organoids at day 50 (neurogenesis), day 120 (early gliogenesis) and day 220 (cell maturation) (Fig. 1b). Organoids from three independent differentiations were collected at each time point, with each organoid contributing both heterozygous (TSC2^c/−; tdTomato-negative) and homozygous (TSC2^−/−; tdTomato-positive) cells.

After processing, 39,539 cells were mapped to a fetal brain tissue atlas²⁹ to determine their broad cell type identity (Fig. 1c). Cell type mapping was confirmed with feature plots of key neuronal genes including DCX, NEFL, NEUROD1 and STMN2 (Fig. 1d), glial genes such as VIM, S100B, GLUL, SLC1A3 and GFAP, and the proliferation marker MKI67 (Fig. 1e). Analysis of cell type proportions revealed a profound shift, whereby TSC2^−/− cells preferentially generated glioblasts compared with TSC2^c/− cells in the same organoid. The ‘glioblast’ identity was defined in first trimester brain tissue as cells expressing TNC and BCAN. Given that we included organoids at later developmental stages, glioblast-mapped cells in our data include putative glial intermediate progenitor cells (gIPCs; as defined in ref. ³⁰, EGFR⁺, ASCL1⁺), astrocyte-biased gIPCs (gIPC-A; SOX9⁺), oligodendrocyte-biased gIPCs (gIPC-O; OLIG1⁺, OLIG2⁺, DLL3⁺), oligodendrocyte progenitor cells (PDGFRA⁺) and differentiated astrocytes (S100B⁺, AQP4⁺, SLC1A3⁺). Therefore, we refer to these cells collectively as glial-lineage cells (GLCs). The GLC bias of TSC2^−/− cells was observable as early as day 50 and grew more prominent over time (Fig. 1f). This finding is consistent with reports that tuber tissue is often enriched for glial cells, including astrocytes^3,31.

To investigate gene-expression changes, we measured the expression distance between genotypes in each cluster at each time point (Fig. 1g) and found the greatest transcriptional changes in cluster 7, an astrocyte cluster defined by expression of AQP4, SLC1A3 and GLUL. Compared with TSC2^c/− cells, TSC2^−/− cluster 7 cells had striking upregulation of astrocyte genes including S100B, GFAP, AQP4, APOE, SPARCL1, CLU and PTGDS, together with reduced expression of oligodendrocyte-lineage genes (OLIG1, OLIG2 and PDGFRA) (Fig. 1h). Both the number of differentially expressed genes and the magnitude of the differential expression increased at later time points (Fig. 1h and Supplementary Table 1).

We noted that many of the differentially expressed genes in TSC2^−/− astrocytes have been associated with disease-associated ‘reactive’ states, such as SERPINA3, CD44, CCL2, CTSH, CRYAB and S100A11³². To investigate this further, we calculated a reactive astrocyte module score for each GLC-mapped cell based on the expression of genes that are upregulated when astrocytes become reactive³² (Supplementary Table 2). Compared with TSC2^c/− cells, TSC2^−/− GLCs had significantly higher expression of this module at all time points (Fig. 1i). As this analysis compares TSC2^c/− and TSC2^−/− cells from the same organoids, these differences cannot be explained by variation in culture conditions or signals from other cells. Therefore, cell-autonomous loss of TSC2 induces a reactive astrocyte transcriptional signature.

In addition to astrocytic genes, autophagy genes were upregulated in TSC2^−/− astrocytes, including SQSTM1, which encodes p62. A module score related to the initiation of autophagy³³ showed that TSC2^−/− cells had increased expression of this module (Fig. 1j and Supplementary Table 2). As loss of TSC1–TSC2 and activation of mTORC1 are typically associated with decreased functional autophagy¹⁹, this may reflect a compensatory transcriptional response. In addition to autophagy, several other biological processes were altered at the transcriptional level in TSC2^−/− astrocytes including upregulation of neuroinflammation and immune signalling, antioxidant response, extracellular matrix remodelling and lipid metabolism and transport (Extended Data Fig. 2 and Supplementary Table 3). We also observed transcriptional changes in TSC2^−/− neurons that were related to synaptic transmission and excitability, oxidative stress, upregulation of glycolysis and mitochondrial dysfunction (Extended Data Fig. 2 and Supplementary Table 3).

It has been suggested that tubers might arise from cells with heterozygous loss of TSC1 or TSC2²⁶; however, other studies have reported that heterozygous cells do not show the full characteristics of tuber cells^27,34,35. To address this, we generated a TSC2^{c/+;LSL-TdTom} hES cell line, which contains a conditional loss-of-function allele and tdTomato reporter but with a functional second allele. In this model, Cre recombinase generates a ‘single-hit’, resulting in loss of one copy of TSC2. TdTomato-positive cells in TSC2^{c/+;LSL-TdTom} organoids did not have dysmorphic morphology, a glial differentiation bias or increased phosphorylation of the mTORC1 pathway target S6 (p-S6) (Extended Data Fig. 3a,b). We performed scRNA-seq of TSC2^{c/+;LSL-TdTom} organoids at day 50 and 120 and observed modest changes in cell type proportions and gene expression (Extended Data Fig. 3c–h and Supplementary Tables 4 and 5). Although TSC2^+/− GLCs did show mild activation of the reactive astrocyte gene module (Extended Data Fig. 3i), this was less pronounced than in TSC2^−/− cells. These data show that biallelic loss of TSC2 is required to produce dysmorphic, tuber-like cells in our brain organoid model.

Variations in the genetic background of human pluripotent stem (hPS) cell lines can affect how mutations influence neural differentiation and development³⁶. The WIBR3 TSC2^{c/−;LSL-TdTom} line is a female hES cell line. To test the robustness of our findings, we obtained a male induced pluripotent stem (hiPS) cell line derived from BJ fibroblasts, which we engineered to carry the conditional TSC2 knockout machinery (Extended Data Fig. 1a–d). These stem cells were differentiated into cortical organoids in five batches, exposed to Cre at day 8 and subject to the scRNA-seq workflow on day 140 (Fig. 2a–d). This dataset was analysed separately to preserve the internally controlled experimental design. We noted that organoids derived from BJ hiPS cells showed stronger mapping to the dorsal pallium compared with WIBR3 hES cell-derived organoids, with neuronal clusters corresponding to populations of cortical excitatory and inhibitory neurons (Extended Data Fig. 4a,b,e,f).

In BJ TSC2^{c/−;LSL-TdTom} organoids, we again observed a pronounced GLC differentiation bias, the generation of cell-autonomously reactive astrocytes and activation of the autophagy induction module (Fig. 2e–i, Supplementary Tables 6 and 7 and Extended Data Fig. 2c,d). To test whether TSC2^−/− GLCs represented a distinct cell type or state, we performed a combined analysis of GLCs from the WIBR3 and BJ TSC2^{c/−;LSL-TdTom} lines (Fig. 2j). Reclustering of GLCs from both lines revealed that TSC2^−/− cells were more highly represented in cluster 0, which was defined by astrocyte reactivity genes such as S100A10, S100B and CRYAB (Fig. 2k,l). TSC2^c/− cells were more represented in cluster 5, which emerged at later stages and expressed oligodendrocyte-lineage genes such as PDGFRA, OLIG1 and OLIG2 (Fig. 2k,l). This shift demonstrates that TSC2^−/− cells preferentially differentiate into astrocytes and not other glial fates such as oligodendrocyte progenitor cells. A combined analysis of the day 120–140 cells from both lines showed clear upregulation of reactivity-associated genes in TSC2^−/− astrocytes, including GFAP, VIM, SERPINA3, LGALS3, CLU, APOE, B2M and CXCL12 (Fig. 2m and Supplementary Table 8). In addition, we observed high expression of several genes associated with extracellular matrix synthesis and remodelling, including COL3A1, MMP14, DCN, CD44 and TIMP2³⁷ (Fig. 2m, Extended Data Fig. 2b,d and Supplementary Table 8). Upregulation of these genes indicates that TSC2^−/− astrocytes may induce remodelling of the surrounding brain tissue, potentially contributing to the reported stiff texture of cortical tubers⁵.

Previous studies have suggested that tuber cells may be in a progenitor-like or immature state due to their expression of markers such as nestin and vimentin^16,17. However, these proteins are also induced in mature astrocytes when they enter an activated state³². Organoid-derived astrocytes recapitulate maturity-related changes in astrocyte gene expression across development³⁸. To assess how loss of TSC2 affected cell maturity, we calculated immature and mature astrocyte gene module scores for astrocytes from both stem cell lines (Supplementary Table 2). We found that TSC2^−/− GLCs had higher maturity scores and lower immaturity scores than corresponding TSC2^c/− GLCs at each time point (Extended Data Fig. 4c,d,g,h).

In summary, our scRNA-seq analysis across several genetic backgrounds demonstrates that TSC2^−/− cells produce disproportionately more astrocytes than TSC2^c/− cells in the same organoid, and that these astrocytes express higher levels of genes associated with reactivity. These changes are cell autonomous and triggered by loss of TSC2 alone, in the absence of signals from microglia or other immune cells.

Given the pronounced transcriptional alterations we observed in TSC2^−/− astrocytes, we performed a series of experiments to characterize the properties of these cells, assessing their protein expression, morphology and cytokine secretion profile. To generate large populations of wild-type and TSC2 mutant astrocytes for comparison, we engineered isogenic TSC2^+/+ and TSC2^−/− hiPS cells on two different genetic backgrounds (8858 and 8119 hiPS cells²⁸, Fig. 3a). TSC2^−/− organoids showed mTORC1 activation, evidenced by increased phosphorylation of S6 and 4EBP1 (Fig. 3b,c and Extended Data Fig. 5a,b). We confirmed that TSC2^−/− organoids had a significantly higher astrocyte to neuron ratio at day 120–130 across both hiPS cell lines on the basis of immunostaining for the astrocyte protein S100β and the neuronal protein HuC/D (Fig. 3d–f). We performed bulk RNA-seq of these organoids at around day 200 and observed a marked increase in GFAP, CLU, CRYAB, APOE and SQSTM1 expression in TSC2^−/− organoids, together with strongly decreased expression of neuronal mRNAs including MAP2, RBFOX3, NEUROD1 and STMN2 (Extended Data Fig. 5c–p).

As TSC can be caused by mutations in either TSC1 or TSC2⁴, we generated cortical organoids from TSC1^−/− hES cells and isogenic TSC1^+/+ controls in the WIBR3 background²⁷ (Extended Data Fig. 6a). We confirmed that these organoids had activated mTORC1 signalling (Extended Data Fig. 6b,c). Consistent with the TSC2^−/− organoids, TSC1^−/− organoids had an increased astrocyte to neuron ratio at days 50 and 70 (Extended Data Fig. 6d–f). The fact that S100β-expressing cells were already observable at days 50–70 (Extended Data Fig. 6d,e) suggests that the increased abundance of astrocytes at later stages is a result of precocious differentiation, occurring during a period when organoids are typically undergoing neurogenesis.

To assess the protein expression profile of astrocytes selectively and at greater scale, we performed immunopanning from day 210–350 TSC2^+/+ and TSC2^−/− organoids with the astrocyte-specific cell surface protein HepaCAM⁴⁷ (Fig. 4a). This enabled the purification of a highly enriched astrocyte population (Fig. 4b). Among these cells, we observed a population of multinucleated cells (up to 17 nuclei per cell) that were readily apparent in TSC2^−/− astrocyte cultures but only rarely seen in TSC2^+/+ cultures (Fig. 4c). This observation is consistent with the fact that giant cells in tubers are often multinucleated⁵. Western blotting of roughly day 325 astrocytes immediately after immunopanning revealed increased expression of GFAP, clusterin and Kir4.1 (Fig. 4d,e). We again observed decreased expression of EAAT1 (Fig. 4d,e), confirming that decreased glutamate transport by TSC2^−/− astrocytes could be a pathophysiological mechanism contributing to neuronal dysfunction within tubers.

To quantify astrocyte size, we immunopanned astrocytes from 240-day-old organoids and cultured them for 7 days. Whereas two-dimensional (2D) culture substantially changes astrocyte morphology compared with the native three-dimensional (3D) organoid environment (Fig. 3n versus Fig. 4f), we again observed that TSC2^−/− astrocytes in culture were highly enlarged and dysmorphic compared with TSC2^+/+ astrocytes (Fig. 4f); with significantly larger cell bodies (Fig. 4g,h). Immunopanned TSC2^−/− astrocytes from 240–350-day-old organoids had greater expression of reactive astrocyte proteins, including S100β, GFAP, CRYAB and vimentin at the level of individual cells (Fig. 4i–n and Extended Data Fig. 7i–t). In addition, we found increased expression of Alzheimer’s disease-associated proteins clusterin and apolipoprotein E (APOE) (Fig. 4o–r and Extended Data Fig. 7r,s). Immunopanned TSC2^−/− astrocytes did not show increased proliferation, as the percentage of Ki67⁺ cells was not different from controls (Extended Data Fig. 7g,h). Together, these data show that, in addition to a reactive transcriptomic signature, TSC2^−/− astrocytes show morphological and protein changes consistent with a reactive state.

The data above show that activation of mTORC1 signalling is sufficient to alter the molecular and morphological profile of astrocytes. To test whether this is reversible with mTOR blockade, we immunopanned astrocytes from day 315–355 TSC2^+/+ and TSC2^−/− organoids and treated them with the mTOR inhibitors rapamycin or Torin. Rapamycin is an inhibitor of mTORC1, although chronic treatment also suppresses mTORC2 signalling¹³. Torin is a selective, ATP-competitive inhibitor that potently suppresses both mTORC1 and mTORC2⁴⁸. After exposure to mTOR inhibitors or vehicle for 7 days, TSC2^+/+ and TSC2^−/− astrocytes were processed using 4i, a multiplexed cyclic immunostaining approach, to probe the same cells with many rounds of immunostaining (Extended Data Figs. 8 and 9a,b). The resulting images were computationally aligned to identify the same cells across each cycle.

To examine protein expression trends across cells in an unbiased manner, we developed an ‘image UMAP’ approach (Extended Data Figs. 8 and 9c). In conjunction with per-cell intensity measurements, this approach integrates broader changes in cellular protein expression and morphological profiles. The image uniform manifold approximation and projection (UMAP) shows that TSC2^+/+ and TSC2^−/− astrocytes are well separated in UMAP space (Extended Data Fig. 9c). Treatment of TSC2^−/− astrocytes with mTOR inhibitors did not fully restore the wild-type state but rather shifted them to an intermediate state, with rapamycin and Torin having largely similar effects (Extended Data Fig. 9c). In terms of specific proteins, the mTOR inhibitors reduced p-S6 in TSC2^−/− astrocytes as well as clusterin, APOE and p62 expression compared with vehicle (Extended Data Fig. 9d,e). The mTOR inhibitors also drove morphological changes in TSC2^−/− astrocytes that increased the measured signal of structural proteins such as GFAP and vimentin (Extended Data Fig. 9d,e). These results indicate that under these conditions and time frame, mTOR inhibition partially rescues the observed reactivity phenotype of TSC2^−/− astrocytes but does not fully restore a homeostatic state.

The analyses above demonstrate that loss of TSC2 in cortical organoids profoundly alters astrocyte properties. To test whether these changes also occur in tubers from patients with TSC, we performed immunostaining on two tuber samples from individuals with TSC who had undergone resection surgery for seizure control (Supplementary Table 9). Compared with neighbouring cells with low p-S6 levels, cells with high p-S6 in tubers had increased expression of several top differentially expressed proteins identified in organoids: APOE (Fig. 5a,c), clusterin (Fig. 5b,d), CRYAB (Fig. 5e,g) and p62 (Fig. 5f,h).

To gain a more comprehensive understanding of protein expression changes in tuber cells, we again used 4i to probe individual sections of tubers with more than 20 antibodies (Fig. 5i,j) and computationally aligned the cells across cycles. In our antibody panel, we included top differentially expressed candidates, general neuron and glia markers, and proteins identified in tubers from a prior study²⁶. Cyclic staining of ten tubers from eight different patients with TSC (Supplementary Table 9) revealed that cells with high p-S6 were more likely to be glial cells and overexpress candidate differentially expressed proteins compared with low p-S6 cells in the same tuber. Specifically, high p-S6 cells were more likely to be positive for the reactive astrocyte proteins S100β, nestin and vimentin, but were not more likely to be positive for the neuronal markers NeuN or HuC/D (Fig. 5k–o). Accordingly, high p-S6 cells had higher expression levels of APOE, clusterin, CRYAB and p62 (Fig. 5p–s). Although these broad trends were found across tubers, individual tuber cells showed varied protein expression profiles, even between adjacent cells (Fig. 5t,u and Extended Data Fig. 10a–e). Notably, in certain samples, we observed that cells with high p-S6 were associated with regions of lower TSC2 signal intensity (Extended Data Fig. 10f,g), suggesting cell-autonomous loss of TSC2 protein.

The analyses above were based on comparisons of high versus low p-S6 cells within the same tuber. To examine protein expression trends across tubers in an unbiased manner, we generated an image UMAP from 117,901 cells spanning all tubers (Fig. 5v). Whereas different tubers generally separated in UMAP space, unbiased clustering identified a unified outlying cluster of 385 cells from 9 of the 10 tubers (Fig. 5w). When mapped by p-S6 intensity, this small cluster contained the cells with the highest p-S6 levels (Fig. 5x). Differential intensity analysis between the cells in the outlying cluster versus all other cells showed that their expression profile aligned with the reactive astrocyte profile identified in TSC2^−/− cells in vitro. These tuber cells had higher expression of p62, vimentin, CRYAB, S100β, nestin, APOE, GFAP, PTGDS and clusterin (Fig. 5y). Furthermore, cells in the high p-S6 cluster had reduced expression of neuronal proteins including HuC/D and MAP2 (Fig. 5y).

In summary, we find an abundance of reactive astrocyte markers and a paucity of neuronal markers in the hyperactive mTORC1 cells in tubers from patients with TSC. Our image UMAP approach identifies a small, unique population of abnormal cells that have hyperactive mTORC1 signalling and expression profiles consistent with reactive astrocytes. Taken together with our data in organoids, this suggests a model whereby a small population of cortical progenitor cells with activated mTORC1 signalling are biased to become astrocytes, which are dysmorphic and express proteins indicative of a reactive state.

In this study, we show that TSC2^−/− cells in human brain organoids preferentially generate astrocytes with several features of disease-associated reactivity. These findings are mirrored in surgically resected tubers from patients with TSC, in which cells with high mTORC1 signalling show protein expression profiles matching those in our in vitro models. These results demonstrate that TSC1–TSC2 dysfunction triggers the production of autonomously reactive astrocytes with neuroinflammatory potential. Because these cells emerge early in cortical development, they are poised to be primary drivers of neuropathophysiology in TSC.

Glial pathology has been consistently reported in malformations of cortical development, including tubers^3,49, but it has been unclear whether this was a consequence or a cause of chronic seizures and epilepsy. Our results in brain organoids show that activation of mTORC1 through biallelic loss of TSC2 is sufficient to induce premature gliogenesis and reactivity, in conditions in which surrounding cells are not reactive. Notably, our organoid models lack microglia, vascular cells and peripheral immune cells, all of which can drive astrogliosis in response to injury or disease⁵⁰. Although their absence may influence astrocyte development, it also demonstrates that the TSC2^−/− phenotype is cell autonomous and does not depend on signals from these populations. Our findings are congruent with recent studies showing that pro-gliogenic factors converge on mTORC1 signalling during human astrocyte development⁵¹ and that cytokine-induced astrocyte reactivity triggers mTORC1 activation and mTOR-dependent remodelling of the endolysosomal system in diseased astrocytes⁵². Likewise, several studies in mouse models have identified glial abnormalities and astrocyte dysfunction resulting directly from TSC1–TSC2 loss in these cells⁴⁹. Together with our work, these studies provide a compelling case for the idea that mTORC1 activation is both pro-gliogenic and promotes the transition to a reactive state.

It is important to note that the highly mTORC1-activated cells we identify in tuber tissue represent less than 1% of cells in the tuber. Despite their low abundance, there are several mechanisms by which TSC2^−/− reactive astrocytes could influence a broad population of cells. Reactive astrocytes could drive neuronal hyperexcitability through failure of homeostatic functions, such as glutamate uptake³¹. Indeed, we observed a striking downregulation of the glutamate transporter EAAT1 in TSC2^−/− organoids and acutely isolated astrocytes. Astrocytes can also gain neurotoxic functions, such as the secretion of inflammatory cytokines⁵³, as observed in TSC2^−/− organoids. In other contexts, such as acquired epilepsy, the presence of reactive astrocytes alone is sufficient to drive epileptogenesis⁵⁴. In addition, whereas this work highlights astrocytes, we also observed transcriptomic changes in neurons, and considerable previous work has shown that TSC1–TSC2 loss affects neuronal development and properties^{11,13,35,55,56,57}. The combination of dysfunctional neurons within a glial environment that fails to constrain their activity may help explain how a small population of mTORC1-activated cells generates focal regions capable of driving seizures while influencing cortical networks more broadly.

Finally, our work hints at common underlying factors connecting developmental disorders, such as TSC, and neurodegenerative diseases, such as Alzheimer’s disease. Genetic variants of many of the differentially expressed genes identified in this study, such as CLU and APOE, have been associated with increased Alzheimer’s disease risk^58,59. Moreover, mTORC1 dysregulation has been reported in Alzheimer’s disease². Alterations in SQSTM1 expression and autophagy have also been implicated broadly in neurodegenerative disorders⁶⁰. There is a growing body of work demonstrating the role of reactive glia, both astrocytes and microglia, in driving pathophysiology in these diseases⁶¹. Activation of mTORC1 signalling, either cell autonomously due to loss of TSC1–TSC2 complex function, or by some other means, may drive astrocytes to an aberrant state that contributes to several neurological disorders. Future study of the mechanisms by which mTORC1 regulates astrocyte reactivity, as well as how this state may be reversed, will lead to deeper insights into the role of glia in establishing and maintaining brain function, with important therapeutic potential.

hPS cell culture

Derivation, maintenance and differentiation of human pluripotent stem cell lines was approved by the University of California, Berkeley Stem Cell Research Oversight Committee (protocol no. 2014-10-029). WIBR3 hES cells (National Institutes of Health (NIH) stem cell registry 0079) were initially obtained from R. Jaenisch’s laboratory⁶². WIBR3 hES cell lines were cultured according to published protocols^63,64. hES cells were maintained on a layer of inactivated mouse embryonic fibroblasts (CD-1 strain, Charles River) in hPS cell medium, consisting of DMEM/F12 supplemented with 20% KnockOut Serum Replacement (Thermo Fisher, 10828010), 2 mM l-glutamine (Thermo Fisher, A2916801), 1% non-essential amino acids (Thermo Fisher 11140050), 0.1 mM 2-mercaptoethanol (Sigma, M6250) and 4 ng ml⁻¹ fibroblast growth factor (FGF)-Basic (AA 1-155) recombinant human protein (Thermo Fisher, PHG0261). Cultures were passaged every 7 days with collagenase type IV (1.5 mg ml⁻¹; Thermo Fisher, 17104019) and gravitational sedimentation by washing 3 times in wash media composed of DMEM/F12 supplemented with 5% fetal bovine serum (Thermo Fisher, A5670801) and 1,000 U ml⁻¹ penicillin–streptomycin (Thermo Fisher 15070063).

The BJ⁶⁵, 8858 and 8119 hiPS cell lines were maintained in feeder-free conditions. BJ hiPS cells were generated by D. Hockemeyer and the 8858 and 8119 hiPS cells were obtained from S. Pasca at Stanford University. iPS cells were cultured on tissue-culture-treated six-well plates (Corning, 3516) coated with vitronectin (Gibco, A14700) and maintained in E8 media (Gibco, A1517001). Cultures were passaged using 7-min of room temperature incubation with EDTA (Thermo Fisher, 15575020).

All cell lines were tested regularly for Mycoplasma contamination. On-target gene editing was confirmed by PCR (Extended Data Fig. 1b). Pluripotency status was confirmed by immunostaining with OCT4 and NANOG (Extended Data Fig. 1d). Genomic integrity of the stem cell lines was verified after gene editing using array comparative genomic hybridization (Cell Line Genetics) (Supplementary Table 11).

Gene editing of hPS cells to generate TSC2 loss of-function was performed using CRISPR–Cas9 editing and validated in our laboratory as previously reported in ref. ²⁷. In brief, constitutive TSC2 exon 5 deletion mutants were generated by electroporating cells with two px330 plasmids⁶⁶ containing single-guide RNAs targeting the genomic regions of interest, as well as a GFP-encoding plasmid. After recovery, GFP-positive cells were selected using FACS, and single-cell-derived hPS cell colonies were manually picked, replated and expanded.

WIBR3 TSC2^{c/−;LSL-TdTom} hES cells, TSC2^{c/+;LSL-TdTom} hES cells and BJ TSC2^{c/−;LSL-TdTom} hiPS cells were generated using CRISPR–Cas9 gene editing as previously described in ref. ²⁷. A single-guide RNA containing a TSC2 exon 5 cassette flanked by loxP sites was cloned into px330, electroporated into TSC2^+/− or TSC2^+/+ hES cells, and colonies underwent puromycin selection. The puromycin resistance cassette was removed, and the Ai9 tdTomato Cre reporter cassette was added to the AAVS1 safe harbour locus⁶⁷ using the same genome editing methods described above.

All unique biological materials in this paper (for example, gene-edited human stem cell lines) will be provided to qualified users on request and on completion of the relevant material transfer agreements.

Organoid differentiation

Cortical organoid generation was performed as described previously in ref. ⁶⁸. For feeder-based cultures, hES cells were isolated and removed from mouse embryonic fibroblasts using accutase (Stemcell Technologies, 07920) for 20 min. The cell suspension was collected and strained through a 40-μm strainer in hES cell media. This cell suspension was spun down for 5 min at 1,000 rpm. The supernatant was removed and resuspended in 5 ml of hES cell media without FGF2, supplemented with 10 μM Y-27632 dihydrochloride (Selleckchem, S1049). Cells were counted and resuspended to a concentration of 2.7 × 10⁶ cells per ml and 6 ml of this suspension was deposited into 1 well of a 6-well Aggrewell 800 plate (Stemcell Technologies, 34825). After the aggregation of single cells into embryoid bodies overnight, the embryoid bodies were removed and put into 10-cm ultra-low attachment dishes (Corning, 4615). On days 1–5, embryoid bodies were cultured in DMEM/F12 supplemented with 20% KnockOut Serum Replacement, 2 mM l-glutamine, 1% non-essential amino acids, 0.1 mM 2-mercaptoethanol, 1,000 U ml⁻¹ penicillin–streptomycin, supplemented with 10 μM dorsomorphin (Abcam, ab146597) and 10 μM SB-431542 (R&D Systems, 1614/10).

For feeder-free cultures, hiPS cells were cultured to high density (80–90% confluency). Then 24 h before aggregation, hiPS cells were pretreated with 1% dimethylsulfoxide (DMSO) (Sigma, D2438) in E8 media. For aggregation, hiPS cells were isolated using 7 min of incubation with accutase and 3 million cells in 2 ml of E8 were transferred into 1 well of a 24-well Aggrewell 800 plate (Stemcell Technologies, 34815). The following day, aggregates were dislodged and transferred into 10-cm ultra-low attachment dishes, with aggregates from 1 well being transferred into 2 10-cm dishes. From days 1 to 5, organoids were cultured in E6 media (Thermo Fisher, A1516401), supplemented with 2.5 μM dorsomorphin and 10 μM SB-431542.

After day 5, feeder-based and feeder-free cultures followed the same protocol. On day 6, organoids were cultured in neural induction media, consisting of Neurobasal-A (Thermo Fisher, 10888022), B-27 Supplement minus vitamin A (Thermo Fisher, 12587010), 50 U ml⁻¹ penicillin–streptomycin and 1× GlutaMAX (Thermo Fisher, 35050-061). During this time, neural induction media was supplemented with 20 ng ml⁻¹ FGF2 (R&D, 233-FB) and 20 ng ml⁻¹ epidermal growth factor (R&D, 236-EG). A full media change was performed every day from days 6 to 15 and then every other day until day 25. From days 25 to 43, the organoids were grown in neural induction media supplemented with 20 ng ml⁻¹ brain-derived neurotrophic factor (BDNF) (Peprotech, 450-02) and 20 ng ml⁻¹ NT-3 (Peprotech 450-03), with media changes every 4 days. From day 43 onward, organoids were maintained in neural induction media without BDNF or NT-3, with media changes every 4 days until collection.

To generate mosaic organoids (TSC2^{c/−;LSL-TdTom} and TSC2^{c/+;LSL-TdTom}), organoids were transduced on day 8 postdifferentiation from hPS cells with UBC-Cre-RFP lentivirus (Kerafast, FCT224) by adding 5 μl of 1.0 × 10⁸ virus to each 10-cm dish containing roughly 20 organoids, with a media change after 24 h.

Organoid dissociation for FACS

Dissociation of organoids for FACS followed a protocol for dissociation of mouse cortex for primary neuronal culture⁶⁹. First, dissociation media was made consisting of calcium and magnesium free Hanks buffered saline solution (Invitrogen, 14185-052), 1 mM sodium pyruvate (Life Technologies, 11360070), 0.1% d-glucose (Sigma, G8769) and 10 mM pH 7.3 HEPES (Invitrogen, 15630-080). Next, the dissociation solution was made consisting of 5-ml dissociation media, 256 μl of Papain Solution (Worthington, LS003126), 0.067 mM 2-mercaptoethanol (Gibco, 21985-023), 1.1 mM EDTA (Thermo Fisher, 15575020) and 5.5 mM l-cysteine (Sigma, 168149). This solution was warmed at 37 °C for 15 min and then filter sterilized through a 0.22-μm filter. Organoids were transferred into dissociation media and incubated at 37 °C for 40 min. During this time, trypsin inhibitor solution was made, consisting of 10 mg of Trypsin Inhibitor (Sigma) in 10 ml dissociation media, prewarmed at 37 °C for more than 15 min and then filter sterilized. After incubation, the intact organoid was washed twice with trypsin inhibitor, then incubated in trypsin inhibitor for 4 min at 37 °C. During this time, the sorting buffer of 1× Dulbecco’s PBS with calcium and magnesium (Thermo Fisher, 14040117) with 10 μM Y-27632 was made and placed on ice. After 4 min at 37 °C, the trypsin inhibitor was removed from the tube with the organoid and 2 ml of sorting buffer was added. The organoid was then mechanically dissociated by triturating 5–10 times through a 5-ml serological pipette within this solution. The dissociated cell solution was then taken up into the serological pipette and passed through a 70-μm cell strainer into a 50-ml conical tube. This passed-through solution was then placed into a polypropylene FACS tube on ice.

FACS and scRNA-seq

Dissociated cells were sorted on a BD Aria Fusion cell sorter with a 70-μm nozzle. When sorting for fluorophores, a negative control of a dissociated non-fluorophore labelled organoid was sorted first to ensure proper gating. After sorting, the cells were centrifuged at 300g for 5 min at 4 °C and then counted on a haemocytometer. Cells were then processed through the 10x Genomics 3′ single-cell sequencing pipeline for v2, v3 or v3.1 according to the manufacturer’s protocol. Complementary DNA from the 10x protocol was assessed for quality at the UC Berkeley Functional Genomics Laboratory using an Agilent 2100 Bioanalyzer, and libraries were prepared using the 10x Genomics protocol. Sequencing was performed at the UC Berkeley Genomics Sequencing Laboratory (QB3 Genomics, UC Berkeley, RRID SCR_022170) or at the Chan Zuckerberg Biohub San Francisco Genomics Platform (Supplementary Table 12).

Processing and analysis of single-cell sequencing data

FASTQ files were aligned to a modified version of the human genome (GRCh38) using Cell Ranger v.6.1.2 (10x Genomics). Cell Ranger gene-expression matrix outputs were then loaded into Seurat v.5.1 (ref. ⁷⁰). Data from each individual sample was turned into a Seurat object and metadata regarding time point, genotype and batch was added to each object. Each object was subset, extracting cells in which more than 500 RNA features were expressed and less than 20% of the genes expressed were mitochondrial. To integrate datasets within stem cell lines, each Seurat object was normalized using the SCTransform pipeline⁷¹. Seurat objects were then integrated by first finding the integration anchors using the oldest control samples as the reference. For WIBR3 TSC2^{c/−;LSL-TdTom} hES cells, the day 220 TSC2^c/− cells were used as a reference. For WIBR3 TSC2^{c/+;LSL-TdTom} hES cells, the day 120 TSC2^c/+ cells were used as a reference and for BJ TSC2^{c/−;LSL-TdTom} hiPS cells, the day 140 TSC2^c/− cells were used as a reference. Other parameters were left at their defaults and then the anchor set was integrated. Principal component analysis was then run on the postintegration cells, followed by UMAP dimensionality reduction using the first 20 principal components. Shared nearest neighbours for each cell and cluster were identified.

The organoid datasets generated in this paper were projected onto a primary fetal tissue dataset²⁹ by creating Seurat objects from the publicly available gene-expression matrices, finding the integration anchors between the organoids and the primary tissue data, and then integrating the two datasets into a single Seurat object. Downstream analysis of the integrated dataset was then performed as above.

Differential expression was performed using MAST through Seurat’s FindMarkers function, with batch included as a latent variable⁷². Clusters with fewer than 3 cells in either condition were excluded from analysis. Cell type proportions were plotted using the DittoSeq v.1.18.0R package⁷³, and volcano plots were plotted using the EnhancedVolcano v.1.20.0R package⁷⁴. The genes used to calculate module scores are listed in Supplementary Table 2. Expression distance analysis was performed using the Cacoa v.0.4.0R package⁷⁵. Pathway changes were analysed using the Single Cell Pathway Analysis v.1.6.1R package⁷⁶. Spatial similarity maps were generated using the VoxHunt v.1.0.1R package⁷⁷.

Immunostaining of organoid sections

Organoids were fixed with 4% paraformaldehyde (PFA) (Electron Microscopy Sciences, 15710) for 2 h at 4 °C. After fixation, organoids were transferred to a 30% sucrose solution and allowed to settle at 4 °C overnight or at room temperature for 1 h. For cryosectioning, organoids were embedded in Tissue-Tek optimal cutting temperature (OCT) compound (Fisher Healthcare, 4585), frozen in an ethanol and dry ice bath, and sectioned on a cryostat (Leica, CM3050S) into 18-μm sections. Sections were washed once with 1× PBS and blocked in Block-Aid (Thermo Fisher, B10710) with 0.3% Triton X-100 (Sigma, X100) for 1 h at room temperature. Sections were incubated overnight at 4 °C in primary antibodies in Block-Aid. The following day, sections were washed three times with PBS, incubated in secondary antibody (1:500 in Block-Aid) and Hoechst stain (1:1,000, Thermo Fisher, H1399) for 1 h at room temperature and washed again 3 times with 1× PBS. Slides were coverslipped with ProLong Glass Antifade Mountant (Thermo Fisher, P36980) and allowed to cure before imaging. Antibody vendors, catalogue numbers and dilutions are listed in Supplementary Table 13.

Confocal imaging and image analysis

Images for all experiments, with the exception of the rapamycin and Torin experiments in Extended Data Fig. 9, were acquired using an Olympus Fluoview FV3000 confocal microscope, using Olympus FV31S 2.3.2.169. For organoid sections and 2D cultures, tile scans were collected using a ×10 or ×20 objective and stitched using ImageJ’s ‘Grid/Collection Stitching’ plugin using a PyImageJ v.1.8.0 wrapper. Individual cells were segmented using Stardist v.0.9.1 (ref. ⁷⁸) on the nuclei channel, and debris was excluded using an area filter. For statistical comparisons, sample sizes were determined based on pilot experiments and prior publications in the field. Experimenters were not blind to genotype during experiments, however analysis pipelines were automated and the same settings were applied to all conditions. Randomization is not relevant to this study. For in vitro cultures, all comparisons were made within isogenic controls, so group membership was determined by genotype, rather than allocation. For human tuber data, comparisons were made between cell populations in the tuber.

For intensity-based measurements, the regionprops_table() function from scikit-image v.0.25.2 was used with the Stardist labels and the corresponding intensity image. For certain protein targets that were expressed in the cell body but did not include the nucleus, the labels were expanded by 5 pixels to include the soma.

To determine whether a cell was positive or negative for a particular marker, a binarization approach was used. The intensity image was postprocessed to remove noise or excessive background, and thresholded according to a mean filter. Stardist labels were then overlaid on the resulting binary image, and labels that contained more than 95% positive pixels were considered positive. Each analysis was validated by visualizing positive and negative cells using Napari v.0.6.6 (ref. ⁷⁹) to confirm that cells were correctly classified.

In the whole-organoid p-S6 analysis (Fig. 3c), a hybrid approach was used in which intensities from expanded labels were extracted only from cells that contained more than 50% positive pixels in the binarization approach. This approach restricted the analysis only to cells above a minimum detectable level of p-S6, as labelled nuclei with no detectable p-S6 were probably unviable.

In the analysis of standard immunostaining in human tuber samples, only selected regions of tubers were imaged. Cells with fluorescence intensities greater than one standard deviation above the mean in a given image were considered to be high p-S6 cells, with all other cells considered to be low p-S6 cells. In the quantification of cyclic immunofluorescence, the full tuber section was imaged, and the high and low p-S6 thresholds were chosen manually for each tuber due to the variability across samples.

Imaging astrocyte morphology in whole organoids

To visualize astrocyte morphology in whole organoids, organoids were exposed to pAAV.GfaABC1D.PI.Lck-GFP.SV40 virus (Addgene, 105598)⁸⁰ at low dilutions (TSC2^−/−: 1:100,000; TSC2^+/+: 1:10,000) at day 220 or 224. After 14 days, intact organoids were fixed with 4% PFA for 2 h at 4 °C and immunostained as described above (section ‘Immunostaining of organoid sections’) with the only difference being 48 h of primary antibody incubation at 4 °C. Whole-mount stained organoids were transferred to 1× PBS in a glass-bottom 96-well plate (Cellvis P96-1.5H-N) and imaged using an Olympus Fluoview FV3000 microscope.

After whole-mount imaging, tissue clearing was performed by incubating the organoids in 50% CUBIC R+ (TCI America T3741-25ML) for 24 h, then transferring the organoids to 100% CUBIC R+ for at least 48 h. Organoids were then re-imaged as described above.

Western blotting

Cortical organoids were harvested in lysis buffer containing 1% SDS in 1× PBS with Halt phosphatase inhibitor cocktail (Thermo Fisher, PI78420) and Complete mini EDTA-free protease inhibitor cocktail (Roche, 4693159001). Immunopanned astrocytes were harvested immediately after immunopanning with 100 μl of lysis buffer (lysis buffer: 10 mM Na-PPi (Sigma, 221368), 10 mM Na-beta-glycerophosphate (Sigma, G5422), 40 mM HEPES, 4 mM EDTA (Sigma, E5134), 1% Triton X-100 (Sigma, T8787), phosphatase inhibitor and protease inhibitor in 1× PBS adjusting pH to 7.4). Total protein was determined by bicinchoninic acid assay (Thermo Fisher, PI23227) and 4 μg (whole-organoid samples) or 8–10 μg (immunopanned astrocyte samples) of protein in 1× Laemmli sample buffer (Bio-Rad, 161-0747) were loaded onto 4–15% Criterion TGX gels (Bio-Rad, 5671084). Proteins were transferred overnight at low voltage to polyvinyl difluoride membranes (Bio-Rad, 1620177), blocked in 5% milk in 1× Tris-buffered saline with Tween (TBS-Tween) for 1 h at room temperature and incubated with primary antibodies diluted in 5% milk in 1× TBS-Tween overnight at 4 °C. The following day, membranes were washed 3 × 10 min in 1× TBS-Tween and incubated with HRP-conjugated secondary antibodies (1:5,000) for 1 h at room temperature, washed 6 times for 10 min in 1× TBS-Tween, incubated with chemiluminescence substrate (Revvity Health Sciences, NEL105001EA) and developed on GE Amersham Hyperfilm ECL (VWR, 95017-661) or imaged using a ChemiDoc (Bio-Rad). Membranes were stripped by two 6-min incubations in stripping buffer (6 M guanidine hydrochloride (Fisher Scientific, ICN10190505) with 1:150 β-mercaptoethanol) with shaking followed by 4 2-min washes in 1× TBS with 0.05% NP-40 to reblot on subsequent days.

Bands were quantified by densitometry using ImageJ v.1.52p software (NIH). For all experiments, phospho-proteins were normalized to their respective total proteins. β-actin was used as a loading control for the whole-organoid samples. Antibody vendors, catalogue numbers and dilutions are listed in Supplementary Table 13. All antibodies were used in accordance with manufacturer guidelines and were validated by the manufacturer for use in human samples for the specific assays used in this study.

Immunopanning

Immunopanning⁴⁷ was performed on non-treated six-well plates (Corning, 3736). The day before the experiment, each well was incubated overnight at 4 °C with 1:400 goat anti-mouse IgG + IgM antibody (Jackson ImmunoResearch, 115-005-044) in 50 mM Tris-HCl pH 9.5 (Thermo Fisher, J62084.K2). Plates for plating cells were prepared by coating Corning BioCoat Poly-d-Lysine 24-well plates (Corning, 356414) with laminin (Sigma, 11243217001) diluted 1:20 in 1× PBS at 37 °C overnight.

On the day of immunopanning, panning plates were rinsed 3 times with 1× PBS, then incubated at room temperature with 1:1,000 anti-HepaCAM antibody (R&D systems, MAB4108, resuspended at 500 μg ml⁻¹) in 1 ml of 1× PBS. Plates were incubated until use (roughly 2 h). Between 6 and 15 organoids (immunofluorescence experiments) or 2–4 organoids (western blotting experiments) were dissociated with the same dissociation media as for the FACS protocol listed above. Organoids were triturated and resuspended in 1 ml of room temperature 0.2% BSA (Sigma, A9418) in 1× PBS with 10 μM Y-27632. After dissociation, cells were filtered with a 70-µm pore size strainer (Greiner, 542170) to eliminate the clumps. The anti-HepaCAM plates were rinsed 4 times with 1× PBS and the cell suspension was added to the plates and incubated at room temperature for 20 min. After incubation, non-bound cells were washed off carefully three times with the BSA/PBS/Y27 solution.

For the drug treatment and immunofluorescence experiments, the anti-HepaCAM plates with bound astrocytes were treated with 1 ml of accutase and incubated at 37 °C for 7 min to release the cells. The accutase was then inactivated with 1 ml of neural induction media (section ‘Organoid differentiation’) supplemented with BDNF and NT-3. Cells were dislodged using trituration and counted using a haemocytometer. The laminin was removed from the plating plates, the purified astrocyte suspension was added to the plate without rinsing, and cells were returned to the incubator for 1 h. After 1 h, a full media change was performed with neural induction media supplemented with BDNF and NT-3. Astrocytes were cultured in neural induction media for 7 days, fixed with 4% PFA for 15 min at room temperature and immunofluorescence was performed as described above.

For acute western blotting experiments, instead of adding accutase, the immunopanned astrocytes were lysed directly from the anti-HepaCAM plates with lysis buffer and western blotting was carried out as described above.

Rapamycin and Torin experiments

For mTOR inhibitor experiments, astrocytes were immunopanned from day 315–355 organoids. After 24 h of recovery, cells were treated with DMSO at a 1:1,000 dilution, rapamycin (Cayman Chemicals, 13346) at a concentration of 50 nM, or Torin-1 (Tocris, 4247) at a concentration of 100 nM, with half media changes every other day. Cells were fixed with 4% PFA after 7 days of drug treatment.

For 4i cyclic staining^81,82, antibody staining was performed as described above, except that the sample well was filled with imaging buffer (freshly prepared 0.7 M N-acetyl-cysteine (Sigma, A7250) in 0.2 M phosphate buffer at pH 7.4). Samples were imaged using an Opera Phenix Plus High-Content Screening System (Revvity) using Revvity Harmony v.5.2, with a 20× water immersion objective. Antibody elution was performed immediately after imaging. Before elution, TCEP-HCl (Sigma, C4706) was added to a stock solution (0.5 M glycine (Fisher, BP381-1), 3 M urea (Fisher, U15) and 3 M guanidine hydrochloride (MP Biomedicals, 101905), stored at 4 °C) to a final concentration of 0.07 M (20 mg ml⁻¹). To elute antibodies, samples were rinsed with 1× PBS, incubated for 5 min with elution buffer and rinsed with water. This process was repeated for a total of three washes. After elution, samples were washed with 1× PBS, stained with Hoechst and selected regions were re-imaged to ensure that elution was successful. Antibodies and cycles were excluded from the analysis if residual signal was still present.

Bulk RNA-seq

For bulk RNA-seq, each organoid was transferred to a 1.5-ml tube, residual media was removed and samples were snap-frozen in liquid nitrogen. RNA was extracted using the RNeasy Mini kit (Qiagen, 74104) according to the manufacturer’s instructions. Library preparation and sequencing was performed by the QB3-Berkeley Genomics core laboratories. Total RNA quality as well as poly-dT enriched mRNA quality were assessed on an Agilent 2100 Bioanalyzer. Libraries were prepared using the KAPA mRNA Hyper Prep kit (Roche, KK858). Truncated universal stub adapters were ligated to complementary DNA fragments, which were then extended through nine cycles of PCR using unique dual indexing primers into full length Illumina adapters. Library quality was checked on an AATI Fragment Analyzer. Library molarity was measured by use of quantitative PCR with the KAPA Library Quantification Kit (Roche, KK4824) on a Bio-Rad CFX Connect thermal cycler. Libraries were then pooled by molarity and sequenced on an Illumina NovaSeq X with the 25B flowcell for 2 × 150 cycles, targeting at least 25 M reads per sample. Fastq files were generated and demultiplexed using Illumina BCL Convert v.4 and default settings. Transcript alignment was performed using Kallisto v.0.48.0 (ref. ⁸³), and statistical analysis was performed using DESeq2 v.1.46.0 (ref. ⁸⁴).

Cytokine release assay

To assess cytokine release from organoids, three organoids per hiPS cell line and genotype were transferred to a 1.5-ml tube with 500 µl of Neurobasal/B27 culture medium. Organoids were incubated in standard incubator conditions (37 °C at 5% CO₂) for 48 h. The media was removed to a fresh tube and centrifuged at 300g for 5 min and the supernatant was stored at −80 °C. Cytokine concentrations were measured using the R&D Systems Proteome Profiler Human XL Cytokine Array Kit (R&D ARY022B) according to the manufacturer’s protocol, using 300 µl of culture supernatant as input. Membranes were developed on GE Amersham Hyperfilm ECL (VWR, 95017-661).

Immunostaining of human cortical tuber sections

Surgically resected cortical tuber tissue was collected under Stanford University Institutional Review Board protocol IRB-12625 (‘The Neuroscience Brain Bank: Collection of Neurosurgical Tissue for Research’). Patients were recruited on the basis of clinical criteria, including a diagnosis of TSC and medically intractable seizures. Informed consent was obtained from the caregivers and legal guardians of all participants. After surgical resection, samples were placed into Eppendorf tubes and frozen and stored at −80 °C. Portions of samples were cut and embedded in OCT. OCT sample blocks were cryosectioned (Leica, CM3050S) to create 18-μm sections.

For standard immunohistochemistry, cryosectioned samples were fixed with 4% PFA in 1× PBS for 10 min and then washed 3 times in 1× PBS. Sections were blocked in buffer containing 10% normal donkey serum (Jackson ImmunoResearch, 017-000-121), and 0.3% Triton X-100 in 1× PBS for 1 h at room temperature. Sections were then incubated overnight at 4 °C with primary antibodies in antibody dilution buffer (10% normal donkey serum in 1× PBS). The following day, sections were washed 3 times with 1× PBS, incubated in secondary antibody (1:500 in antibody dilution buffer) and Hoechst stain (1:1,000) for 1 h at room temperature and washed 3 times with 1× PBS. Slides were coverslipped with ProLong Glass Antifade Mountant and allowed to set for at least 1 day before imaging. Antibody vendors, catalogue numbers and dilutions are listed in Supplementary Table 13.

For cyclic imaging, a well was created by cutting a rectangular shape from a sheet of cured polydimethylsiloxane (Dow, Sylgard 184) and pressed to the glass slide with cryosectioned samples. Antibody staining was performed as described above, except that the sample well was filled with an imaging buffer (freshly prepared 0.7 M N-acetyl-cysteine (Sigma, A7250) in 0.2 M phosphate buffer at pH 7.4) instead of mounting media^81,82, and imaging of the full tuber section was performed through the bottom glass slide. Confocal images were acquired using a ×10 objective and 4,096 × 4,096 pixel resolution on an Olympus Fluoview FV3000 microscope. Antibody elution was performed immediately after imaging. Before elution, TCEP-HCl (Sigma, C4706) was added to a stock solution (0.5 M glycine (Fisher, BP381-1), 3 M urea (Fisher, U15) and 3 M guanidine hydrochloride (MP Biomedicals, 101905), stored at 4 °C) to a final concentration of 0.07 M (20 mg ml⁻¹). To elute antibodies, samples were rinsed with 1× PBS, incubated for 5 min with elution buffer and rinsed with water. This process was repeated for a total of three washes. After elution, samples were washed with 1× PBS, stained with Hoechst and selected regions were re-imaged to ensure that elution was successful (Extended Data Fig. 8b). Antibodies and cycles were excluded from analysis if residual signal was excessive. The blocking step for the next round of antibody staining was started immediately afterwards. Eluted samples were stored in the dark at 4 °C in 1× PBS for up to 48 h between cycles, and stained samples containing fluorescent antibodies were stored in the dark at 4 °C in imaging buffer for up to 24 h.

Cyclic staining analysis

Images from each cycle were stitched and aligned on the Hoechst channel using a developmental branch of the Alignment by Simultaneous Harmonization of Layer/Adjacency Registration (ASHLAR) software package containing a rotation correction^85,86. Standard intensity-based image analysis was performed as described above, using manually chosen p-S6 intensity thresholds.

The image UMAP process is described in Extended Data Fig. 8a. Beginning with the ASHLAR-aligned image channels, Stardist was used to segment nuclei for each cycle individually. Because ASHLAR alignment results in the same coordinate system for all channels, each nucleus label in the first channel was matched to the nearest label in each subsequent channel to identify the same cell across channels. Poor matches were filtered by a defined maximum allowable distance between any pair of label centroids across channels. After identifying confidently aligned cells, intensity data for each cell were extracted. All pixels within a 30 pixels (9.3 μm for tubers) or 50 pixels (14.8 μm for immunopanned astrocytes) radius circle centred on each cell’s label centroid across all cycles were extracted. These data were then reshaped into a one-dimensional vector, and data from all cells were stacked to generate a 2D matrix of (cell index) × (pixel intensity). The UMAP-learn package was used to perform dimensionality reduction on this matrix, and HDBSCAN was used on the UMAP embeddings to determine clustering. For differential intensity measurements, intensity data, cluster assignments and UMAP embeddings were used to generate a Seurat object, and differential intensity was calculated using the Wilcoxon rank-sum test with the Bonferroni correction.

For the mTOR inhibitor experiments in Extended Data Fig. 9, images were acquired using an Opera Phenix Plus High-Content Screening System (Revvity) using Revvity Harmony v.5.2. The image UMAP was computed on a subset of the full dataset in which the numbers of cells from each genotype and treatment combination were equalized. The full dataset was then projected into the embedding derived from the balanced subset.

Reporting summary

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

Code capsules demonstrating the image UMAP and image binarization pipelines on sample data are available at Code Ocean (binary quantification https://doi.org/10.24433/CO.4440783.v1; image UMAP https://doi.org/10.24433/CO.2878862.v1).

1. Crino, P. B., Nathanson, K. L. & Henske, E. P. The tuberous sclerosis complex. N. Engl. J. Med. 355, 1345–1356 https://doi.org/10.1056/NEJMra055323 (2006).

   Article  PubMed  Google Scholar 

2. Crino, P. B. The mTOR signalling cascade: paving new roads to cure neurological disease. Nat. Rev. Neurol. 12, 379–392 https://doi.org/10.1038/nrneurol.2016.81 (2016).

   Article  PubMed  Google Scholar 

3. Sosunov, A. A. et al. Tuberous sclerosis: a primary pathology of astrocytes? Epilepsia 49, 53–62 https://doi.org/10.1111/j.1528-1167.2008.01493.x (2008).

   Article  PubMed  Google Scholar 

4. Salussolia, C. L., Klonowska, K., Kwiatkowski, D. J. & Sahin, M. Genetic etiologies, diagnosis, and treatment of tuberous sclerosis complex. Annu. Rev. Genomics Hum. Genet. 20, 217–240 https://doi.org/10.1146/annurev-genom-083118-015354 (2019).

   Article  PubMed  Google Scholar 

5. Mizuguchi, M. & Takashima, S. Neuropathology of tuberous sclerosis. Brain Dev. 23, 508–515 https://doi.org/10.1016/s0387-7604(01)00304-7 (2001).

   Article  PubMed  Google Scholar 

6. Mohamed, A. R. et al. Intrinsic epileptogenicity of cortical tubers revealed by intracranial EEG monitoring. Neurology 79, 2249–2257 https://doi.org/10.1212/WNL.0b013e3182768923 (2012).

   Article  PubMed  Google Scholar 

7. Talos, D. M. et al. Cell-specific alterations of glutamate receptor expression in tuberous sclerosis complex cortical tubers. Ann. Neurol. 63, 454–465 https://doi.org/10.1002/ana.21342 (2008).

   Article  PubMed  PubMed Central  Google Scholar 

8. Zhang, J. et al. Iconography of abnormal non-neuronal cells in pediatric focal cortical dysplasia type IIb and tuberous sclerosis complex. Front. Cell. Neurosci. 18, 1486315 https://doi.org/10.3389/fncel.2024.1486315 (2024).

   Article  PubMed  Google Scholar 

9. Muhlebner, A. et al. Novel histopathological patterns in cortical tubers of epilepsy surgery patients with tuberous sclerosis complex. PLoS ONE 11, e0157396 https://doi.org/10.1371/journal.pone.0157396 (2016).

   Article  PubMed  PubMed Central  Google Scholar 

10. Sundberg, M. & Sahin, M. Modeling neurodevelopmental deficits in tuberous sclerosis complex with stem cell derived neural precursors and neurons. Adv. Neurobiol. 25, 1–31 https://doi.org/10.1007/978-3-030-45493-7_1 (2020).

    Article  PubMed  Google Scholar 

11. Meikle, L. et al. A mouse model of tuberous sclerosis: neuronal loss of Tsc1 causes dysplastic and ectopic neurons, reduced myelination, seizure activity, and limited survival. J. Neurosci. 27, 5546–5558 https://doi.org/10.1523/JNEUROSCI.5540-06.2007 (2007).

    Article  PubMed  PubMed Central  Google Scholar 

12. Feliciano, D. M. et al. A circuitry and biochemical basis for tuberous sclerosis symptoms: from epilepsy to neurocognitive deficits. Int. J. Dev. Neurosci. 31, 667–678 https://doi.org/10.1016/j.ijdevneu.2013.02.008 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

13. Karalis, V., Caval-Holme, F. & Bateup, H. S. Raptor downregulation rescues neuronal phenotypes in mouse models of tuberous sclerosis complex. Nat. Commun. 13, 4665 https://doi.org/10.1038/s41467-022-31961-6 (2022).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

14. Martin, K. R. et al. The genomic landscape of tuberous sclerosis complex. Nat. Commun. 8, 15816 https://doi.org/10.1038/ncomms15816 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

15. Boer, K. et al. Gene expression analysis of tuberous sclerosis complex cortical tubers reveals increased expression of adhesion and inflammatory factors. Brain Pathol. 20, 704–719 https://doi.org/10.1111/j.1750-3639.2009.00341.x (2009).

    Article  PubMed  PubMed Central  Google Scholar 

16. Crino, P. B., Trojanowski, J. Q., Dichter, M. A. & Eberwine, J. Embryonic neuronal markers in tuberous sclerosis: single-cell molecular pathology. Proc. Natl Acad. Sci. USA 93, 14152–14157 https://doi.org/10.1073/pnas.93.24.14152 (1996).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

17. Hirose, T. et al. Tuber and subependymal giant cell astrocytoma associated with tuberous sclerosis: an immunohistochemical, ultrastructural, and immunoelectron and microscopic study. Acta Neuropathol. 90, 387–399 https://doi.org/10.1007/BF00315012 (1995).

    Article  PubMed  Google Scholar 

18. Tee, A. R. et al. Tuberous sclerosis complex-1 and -2 gene products function together to inhibit mammalian target of rapamycin (mTOR)-mediated downstream signaling. Proc. Natl Acad. Sci. USA 99, 13571–13576 https://doi.org/10.1073/pnas.202476899 (2002).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

19. Saxton, R. A. & Sabatini, D. M. mTOR signaling in growth, metabolism, and disease. Cell 168, 960–976 https://doi.org/10.1016/j.cell.2017.02.004 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

20. Marcotte, L., Aronica, E., Baybis, M. & Crino, P. B. Cytoarchitectural alterations are widespread in cerebral cortex in tuberous sclerosis complex. Acta Neuropathol. 123, 685–693 https://doi.org/10.1007/s00401-012-0950-3 (2012).

    Article  PubMed  Google Scholar 

21. Henske, E. P. et al. Allelic loss is frequent in tuberous sclerosis kidney lesions but rare in brain lesions. Am. J. Hum. Genet. 59, 400–406 (1996).

    PubMed  PubMed Central  Google Scholar 

22. Crino, P. B., Aronica, E., Baltuch, G. & Nathanson, K. L. Biallelic TSC gene inactivation in tuberous sclerosis complex. Neurology 74, 1716–1723 https://doi.org/10.1212/WNL.0b013e3181e04325 (2010).

    Article  PubMed  PubMed Central  Google Scholar 

23. Qin, W. et al. Analysis of TSC cortical tubers by deep sequencing of TSC1, TSC2 and KRAS demonstrates that small second-hit mutations in these genes are rare events. Brain Pathol. 20, 1096–1105 https://doi.org/10.1111/j.1750-3639.2010.00416.x (2010).

    Article  PubMed  PubMed Central  Google Scholar 

24. Sim, N. S. et al. Precise detection of low-level somatic mutation in resected epilepsy brain tissue. Acta Neuropathol. 138, 901–912 https://doi.org/10.1007/s00401-019-02052-6 (2019).

    Article  PubMed  Google Scholar 

25. Baldassari, S. et al. Dissecting the genetic basis of focal cortical dysplasia: a large cohort study. Acta Neuropathol. 138, 885–900 https://doi.org/10.1007/s00401-019-02061-5 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

26. Eichmüller, O. L. et al. Amplification of human interneuron progenitors promotes brain tumors and neurological defects. Science 375, eabf5546 https://doi.org/10.1126/science.abf5546 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

27. Blair, J. D., Hockemeyer, D. & Bateup, H. S. Genetically engineered human cortical spheroid models of tuberous sclerosis. Nat. Med. 24, 1568–1578 https://doi.org/10.1038/s41591-018-0139-y (2018).

    Article  PubMed  PubMed Central  Google Scholar 

28. Paşca, A. M. et al. Functional cortical neurons and astrocytes from human pluripotent stem cells in 3D culture. Nat. Methods 12, 671–678 https://doi.org/10.1038/nmeth.3415 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

29. Braun, E. et al. Comprehensive cell atlas of the first-trimester developing human brain. Science 382, eadf1226 https://doi.org/10.1126/science.adf1226 (2023).

    Article  PubMed  Google Scholar 

30. Ramos, S. I. et al. An atlas of late prenatal human neurodevelopment resolved by single-nucleus transcriptomics. Nat. Commun. 13, 7671 https://doi.org/10.1038/s41467-022-34975-2 (2022).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

31. Zimmer, T. S. et al. Tuberous sclerosis complex as disease model for investigating mtor-related gliopathy during epileptogenesis. Front. Neurol. 11, 1028 https://doi.org/10.3389/fneur.2020.01028 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

32. Escartin, C. et al. Reactive astrocyte nomenclature, definitions, and future directions. Nat. Neurosci. 24, 312–325 https://doi.org/10.1038/s41593-020-00783-4 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

33. Bordi, M. et al. A gene toolbox for monitoring autophagy transcription. Cell Death Dis. 12, 1044 https://doi.org/10.1038/s41419-021-04121-9 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

34. Costa, V. et al. mTORC1 inhibition corrects neurodevelopmental and synaptic alterations in a human stem cell model of tuberous sclerosis. Cell Rep. 15, 86–95 https://doi.org/10.1016/j.celrep.2016.02.090 (2016).

    Article  PubMed  Google Scholar 

35. Winden, K. D. et al. Biallelic mutations in TSC2 lead to abnormalities associated with cortical tubers in human iPSC-derived neurons. J. Neurosci. 39, 9294–9305 https://doi.org/10.1523/JNEUROSCI.0642-19.2019 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

36. Pașca, S. P. et al. A framework for neural organoids, assembloids and transplantation studies. Nature 639, 315–320 https://doi.org/10.1038/s41586-024-08487-6 (2025).

37. Naba, A. et al. The matrisome: in silico definition and in vivo characterization by proteomics of normal and tumor extracellular matrices. Mol. Cell. Proteom. 11, M111.014647 https://doi.org/10.1074/mcp.M111.014647 (2011).

38. Sloan, S. A. et al. Human astrocyte maturation captured in 3D cerebral cortical spheroids derived from pluripotent stem cells. Neuron 95, 779–790.e6 https://doi.org/10.1016/j.neuron.2017.07.035 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

39. Giaume, C., Naus, C. C., Sáez, J. C. & Leybaert, L. Glial connexins and pannexins in the healthy and diseased brain. Physiol. Rev. 101, 93–145 https://doi.org/10.1152/physrev.00043.2018 (2020).

40. Olsen, M. L. & Sontheimer, H. Functional implications for Kir4.1 channels in glial biology: from K⁺ buffering to cell differentiation. J. Neurochem. 107, 589–601 https://doi.org/10.1111/j.1471-4159.2008.05615.x (2008).

    Article  PubMed  PubMed Central  Google Scholar 

41. Pajarillo, E., Rizor, A., Lee, J., Aschner, M. & Lee, E. The role of astrocytic glutamate transporters GLT-1 and GLAST in neurological disorders: potential targets for neurotherapeutics. Neuropharmacology 161, 107559 https://doi.org/10.1016/j.neuropharm.2019.03.002 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

42. Rothstein, J. D. et al. Knockout of glutamate transporters reveals a major role for astroglial transport in excitotoxicity and clearance of glutamate. Neuron 16, 675–686 https://doi.org/10.1016/s0896-6273(00)80086-0 (1996).

    Article  PubMed  Google Scholar 

43. Lee, H.-G., Lee, J.-H., Flausino, L. E. & Quintana, F. J. Neuroinflammation: an astrocyte perspective. Sci. Transl. Med. 15, eadi7828 https://doi.org/10.1126/scitranslmed.adi7828 (2023).

    Article  PubMed  Google Scholar 

44. Mireuta, M., Darnel, A. & Pollak, M. IGFBP-2 expression in MCF-7 cells is regulated by the PI3K/AKT/mTOR pathway through Sp1-induced increase in transcription. Growth Factors 28, 243–255 https://doi.org/10.3109/08977191003745472 (2010).

    Article  PubMed  Google Scholar 

45. Schiweck, J., Eickholt, B. J. & Murk, K. Important shapeshifter: mechanisms allowing astrocytes to respond to the changing nervous system during development, injury and disease. Front. Cell. Neurosci. 12, 261 https://doi.org/10.3389/fncel.2018.00261 (2018).

46. Oberheim, N. A. et al. Uniquely hominid features of adult human astrocytes. J. Neurosci. 29, 3276–3287 https://doi.org/10.1523/JNEUROSCI.4707-08.2009 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

47. Zhang, Y. et al. Purification and characterization of progenitor and mature human astrocytes reveals transcriptional and functional differences with mouse. Neuron 89, 37–53 https://doi.org/10.1016/j.neuron.2015.11.013 (2016).

    Article  PubMed  Google Scholar 

48. Thoreen, C. C. et al. An ATP-competitive mammalian target of rapamycin inhibitor reveals rapamycin-resistant functions of mTORC1. J. Biol. Chem. 284, 8023–8032 https://doi.org/10.1074/jbc.M900301200 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

49. Wong, M. & Crino, P. B. Tuberous sclerosis and epilepsy: role of astrocytes. Glia 60, 1244–1250 https://doi.org/10.1002/glia.22326 (2012).

    Article  PubMed  Google Scholar 

50. Liddelow, S. A. et al. Neurotoxic reactive astrocytes are induced by activated microglia. Nature 541, 481–487 https://doi.org/10.1038/nature21029 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

51. Voss, A. J. et al. Identification of ligand–receptor pairs that drive human astrocyte development. Nat. Neurosci. 26, 1339–1351 https://doi.org/10.1038/s41593-023-01375-8 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

52. Leng, K. et al. mTOR activation induces endolysosomal remodeling and nonclassical secretion of IL-32 via exosomes in inflammatory reactive astrocytes. J. Neuroinflammation 21, 198 https://doi.org/10.1186/s12974-024-03165-w (2024).

    Article  PubMed  PubMed Central  Google Scholar 

53. Ravizza, T. et al. mTOR and neuroinflammation in epilepsy: implications for disease progression and treatment. Nat. Rev. Neurosci. 25, 334–350 https://doi.org/10.1038/s41583-024-00805-1 (2024).

    Article  PubMed  Google Scholar 

54. Vezzani, A. et al. Astrocytes in the initiation and progression of epilepsy. Nat. Rev. Neurol. 18, 707–722 https://doi.org/10.1038/s41582-022-00727-5 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

55. Mills, J. D. et al. Coding and small non-coding transcriptional landscape of tuberous sclerosis complex cortical tubers: implications for pathophysiology and treatment. Sci. Rep. 7, 8089 https://doi.org/10.1038/s41598-017-06145-8 (2017).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

56. Bateup, H. S. et al. Excitatory/inhibitory synaptic imbalance leads to hippocampal hyperexcitability in mouse models of tuberous sclerosis. Neuron 78, 510–522 https://doi.org/10.1016/j.neuron.2013.03.017 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

57. Casingal, C. R. et al. TSC tunes progenitor balance and upper-layer neuron generation in neocortex. Nature 650, 417–427 https://doi.org/10.1038/s41586-025-09810-5 (2026).

    Article  ADS  PubMed  Google Scholar 

58. Foster, E. M., Dangla-Valls, A., Lovestone, S., Ribe, E. M. & Buckley, N. J. Clusterin in Alzheimer’s disease: mechanisms, genetics, and lessons from other pathologies. Front. Neurosci. 13, 164 https://doi.org/10.3389/fnins.2019.00164 (2019).

    Article  PubMed  PubMed Central  Google Scholar 

59. Raulin, A.-C. et al. ApoE in Alzheimer’s disease: pathophysiology and therapeutic strategies. Mol. Neurodegener. 17, 72 https://doi.org/10.1186/s13024-022-00574-4 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

60. Nixon, R. A. The role of autophagy in neurodegenerative disease. Nat. Med. 19, 983–997 https://doi.org/10.1038/nm.3232 (2013).

    Article  PubMed  Google Scholar 

61. Gleichman, A. J. & Carmichael, S. T. Glia in neurodegeneration: drivers of disease or along for the ride? Neurobiol. Dis. 142, 104957 https://doi.org/10.1016/j.nbd.2020.104957 (2020).

    Article  PubMed  Google Scholar 

62. Lengner, C. J. et al. Derivation of pre-X inactivation human embryonic stem cells under physiological oxygen concentrations. Cell 141, 872–883 https://doi.org/10.1016/j.cell.2010.04.010 (2010).

    Article  PubMed  Google Scholar 

63. Blair, J. D., Bateup, H. S. & Hockemeyer, D. F. Establishment of genome-edited human pluripotent stem cell lines: from targeting to isolation. J. Vis. Exp. https://doi.org/10.3791/53583 (2016).

64. Li, H. et al. Highly efficient generation of isogenic pluripotent stem cell models using prime editing. eLife 11, e79208 https://doi.org/10.7554/eLife.79208 (2022).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

65. Bodnar, A. G. et al. Extension of life-span by introduction of telomerase into normal human cells. Science 279, 349–352 https://doi.org/10.1126/science.279.5349.349 (1998).

    Article  ADS  PubMed  Google Scholar 

66. Cong, L. et al. Multiplex genome engineering using CRISPR/Cas systems. Science 339, 819–823 https://doi.org/10.1126/science.1231143 (2013).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

67. Hockemeyer, D. et al. Efficient targeting of expressed and silent genes in human ESCs and iPSCs using zinc-finger nucleases. Nat. Biotechnol. 27, 851–857 https://doi.org/10.1038/nbt.1562 (2009).

    Article  PubMed  PubMed Central  Google Scholar 

68. Yoon, S. J. et al. Reliability of human cortical organoid generation. Nat. Methods 16, 75–78 https://doi.org/10.1038/s41592-018-0255-0 (2019).

    Article  PubMed  Google Scholar 

69. Beaudoin, G. M. 3rd et al. Culturing pyramidal neurons from the early postnatal mouse hippocampus and cortex. Nat. Protoc. 7, 1741–1754 https://doi.org/10.1038/nprot.2012.099 (2012).

    Article  PubMed  Google Scholar 

70. Hao, Y. et al. Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat. Biotechnol. 42, 293–304 https://doi.org/10.1038/s41587-023-01767-y (2024).

    Article  ADS  PubMed  Google Scholar 

71. Choudhary, S. & Satija, R. Comparison and evaluation of statistical error models for scRNA-seq. Genome Biol. 23, 27 https://doi.org/10.1186/s13059-021-02584-9 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

72. Zimmerman, K. D., Espeland, M. A. & Langefeld, C. D. A practical solution to pseudoreplication bias in single-cell studies. Nat. Commun. 12, 738 https://doi.org/10.1038/s41467-021-21038-1 (2021).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

73. Bunis, D. G., Andrews, J., Fragiadakis, G. K., Burt, T. D. & Sirota, M. dittoSeq: universal user-friendly single-cell and bulk RNA sequencing visualization toolkit. Bioinformatics 36, 5535–5536 https://doi.org/10.1093/bioinformatics/btaa1011 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

74. Blighe, K., Rana, S. & Lewis, M. EnhancedVolcano. GitHub https://github.com/kevinblighe/EnhancedVolcano (2021).

75. Petukhov, V. et al. Case-control analysis of single-cell RNA-seq studies. Preprint at bioRxiv https://doi.org/10.1101/2022.03.15.484475 (2022).

76. Bibby, J. A. et al. Systematic single-cell pathway analysis to characterize early T cell activation. Cell Rep. 41, 111697 https://doi.org/10.1016/j.celrep.2022.111697 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

77. Fleck, J. S. et al. Resolving organoid brain region identities by mapping single-cell genomic data to reference atlases. Cell Stem Cell 28, 1177–1180 https://doi.org/10.1016/j.stem.2021.03.015 (2021).

    Article  PubMed  Google Scholar 

78. Weigert, M. & Schmidt, U. Nuclei instance segmentation and classification in histopathology images with Stardist. In IEEE International Symposium on Biomedical Imaging Challenges (ISBIC) https://doi.org/10.1109/ISBIC56247.2022.9854534 (IEEE, 2022).

79. Sofroniew, N. et al. napari: a multi-dimensional image viewer for Python, v0.6.6. Zenodo https://doi.org/10.5281/zenodo.17367124 (2025).

80. Shigetomi, E. et al. Imaging calcium microdomains within entire astrocyte territories and endfeet with GCaMPs expressed using adeno-associated viruses. J. Gen. Physiol. 141, 633–647 https://doi.org/10.1085/jgp.201210949 (2013).

    Article  PubMed  PubMed Central  Google Scholar 

81. Cole, J. D. et al. Characterization of the neurogenic niche in the aging dentate gyrus using iterative immunofluorescence imaging. eLife 11, e68000 https://doi.org/10.7554/eLife.68000 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

82. Gut, G., Herrmann, M. D. & Pelkmans, L. Multiplexed protein maps link subcellular organization to cellular states. Science 361, eaar7042 https://doi.org/10.1126/science.aar7042 (2018).

    Article  PubMed  Google Scholar 

83. Bray, N. L., Pimentel, H., Melsted, P. & Pachter, L. Near-optimal probabilistic RNA-seq quantification. Nat. Biotechnol. 34, 525–527 https://doi.org/10.1038/nbt.3519 (2016).

    Article  PubMed  Google Scholar 

84. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 https://doi.org/10.1186/s13059-014-0550-8 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

85. Muhlich, J. L. ASHLAR: alignment by simultaneous harmonization of layer/adjacency registration. GitHub https://github.com/jmuhlich/ashlar/tree/rotation-correction (2022).

86. Muhlich, J. L. et al. Stitching and registering highly multiplexed whole-slide images of tissues and tumors using ASHLAR. Bioinformatics 38, 4613–4621 https://doi.org/10.1093/bioinformatics/btac544 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

Download references

Author notes

1. These authors contributed equally: Thomas L. Li, John D. Blair

Authors and Affiliations

1. Department of Molecular and Cell Biology, University of California, Berkeley, Berkeley, CA, USA

   Thomas L. Li, John D. Blair, Taesun Yoo, Dirk Hockemeyer & Helen S. Bateup

2. Department of Neuroscience, University of California, Berkeley, Berkeley, CA, USA

   Thomas L. Li, Taesun Yoo & Helen S. Bateup

3. Department of Neurosurgery, Lucile Packard Children’s Hospital and Stanford University Medical Center, Stanford, CA, USA

   Gerald A. Grant

4. Innovative Genomics Institute, University of California, Berkeley, Berkeley, CA, USA

   Dirk Hockemeyer

5. Department of Neurology and Neurological Sciences, Stanford University, Stanford, CA, USA

   Brenda E. Porter

Contributions

T.L.L. and J.D.B. generated the CRISPR-edited cell lines and brain organoids, performed scRNA-seq experiments and performed immunostaining and imaging of human tuber sections. T.L.L. performed immunostaining and imaging of organoid sections and immunopanned cells, bulk RNA-seq, drug treatment experiments and conducted code-based analyses. T.Y. performed immunopanning, immunostaining and western blotting experiments. G.A.G. performed the surgical resection of tuber samples, which were collected and shared by B.E.P. D.H. and H.S.B. obtained funding and supervised the project. Article drafts were written by J.D.B., T.L.L. and H.S.B. All authors contributed to review and editing.

Corresponding author

Correspondence to Helen S. Bateup.

Competing interests

The authors declare no competing interests.

Peer review information

Nature thanks Peter Crino, Steven Sloan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions