Microglia regulate GABAergic neurogenesis in prenatal human brain through IGF1

Human tissue samples

De-identified human specimens were collected from the Autopsy Service in the Department of Pathology at the University of California San Francisco (UCSF) (Supplementary Table 2) with previous patient consent in strict observance of the legal and institutional ethical regulations. Autopsy consents and all protocols for human prenatal brain tissue procurement were approved by the Human Gamete, Embryo and Stem Cell Research Committee (Institutional Review Board GESCR no. 10-02693) at UCSF. All specimens received diagnostic evaluations by a board-certified neuropathologist as control samples and were free of brain-related diseases. The diagnostic panel included assessments of neural progenitor and immune cells using IHC to ensure that all control cases were not affected by any inflammatory diseases. Tissues used for snRNA-seq were snap-frozen, either on a cold plate placed on a slab of dry ice or in isopentane on dry ice. Tissues later used for IHC were cut coronally into 1-mm tissue blocks, fixed with 4% paraformaldehyde (PFA) for 2 days, cryoprotected in a 15–30% sucrose gradient, embedded in optimal cutting temperature (OCT; SciGen; 4586) compound, sectioned at 30 μm using a Leica cryostat and mounted onto glass slides.

Authentication of cell lines used

All hiPSC and hESC lines used in this study were karyotyped and regularly tested for Mycoplasma. The eWT-1323-4 hiPSC line⁸⁴ (female; Research Resource Identifier (RRID): CVCL_0G84) was obtained from the Conklin Laboratory (UCSF). WA09/H9 (female; RRID: CVCL_9773; National Institutes of Health (NIH) registration number: NIHhESC-10_0062) and WA01/H1 (male; RRID: CVCL_9771; NIH registration number: NIHhESC-10-0043) were obtained from the WiCell Research Institute. NKX2.1-GFP hESC line⁵⁶ (female) was obtained from Murdoch Children’s Research Institute and Monash University.

Mice

All mice were handled in accordance with the guidelines of the Institutional Animal Care and Use Committee of UCSF. Minimal sample sizes were chosen on the basis of standards commonly used in the field and previous experience with similar experiments. All animals of the same genotype and sex were randomly selected for breeding and/or experimentation in this study. Wild-type C57/B6 mice were purchased from Taconic Biosciences and bred in the laboratory. Igf1^f/f mice (strain number 012663) and Cx3cr1^CreERt/+ mice (strain number 020940) were purchased from The Jackson Laboratory. For timed pregnancy, males and females were paired, and females were observed daily for the presence of a copulation plug. The noon of the day when a plug was observed was noted as embryonic day 0.5. For Igf1 cKO experiments, 100 mg kg⁻¹ of tamoxifen in corn oil was injected intraperitoneally into pregnant dams on embryonic days 11.5 and 12.5. Igf1^f/f; Cx3cr1^CreERt/+ fetuses were used as Igf1 cKO mice, and their littermates Igf1^+/+; Cx3cr1^CreERt/+ and Igf1^f/f; Cx3cr1^+/+ fetuses were used as controls. During all subsequent experimental procedures, including sample collection, processing, imaging and quantification, the experimenter was blinded to the genotype, sex and age of the mice. Both males and females were included in the mouse experiments. In the EdU labelling experiment, a single dose of EdU (10 mg kg⁻¹; provided in the Click-iT EdU Alexa Fluor 647 Imaging Kit from Invitrogen; C10340) was injected intraperitoneally into pregnant mice at embryonic day 14.5. At embryonic days 14.5 and 16.5, the pregnant dams were killed, and the fetal brains were collected, fixed in 4% PFA at 4 °C overnight, cryopreserved in 30% sucrose at 4 °C overnight, embedded in OCT and cryosectioned at 20 μm (EdU labelling and IGF1 staining experiments) or 40 μm (microglia staining experiments) using a Leica cryostat. Additionally, wild-type P5 pups were transcardially perfused with 4% PFA, and their brains were extracted and post-fixed in 4% PFA overnight, cryoprotected in 30% sucrose overnight, embedded in OCT and cryosectioned at 20 μm.

Human pluripotent stem-cell-derived organoids

The hPSC-derived organoids were generated largely following a previously established protocol^37,38. In brief, 1323-4 hiPSCs or WA01/H1 and WA09/H9 hESCs were expanded in StemFlex Basal Medium (Gibco; A3349401). After reaching 80% coverage, hPSCs cultured on Matrigel were dissociated into clumps using ReLeSR (STEMCELL Technologies; 100-0483) and equally distributed into a V-bottom 96-well ultra-low-attachment PrimeSurface plate (S-BIO; MS-9096VZ). The rho kinase inhibitor Y-27632 (10 μM) was added during the first 24 h of neural induction to promote survival. Between days 0 and 5, organoids were cultured in neural induction medium (Dulbecco’s modified Eagle medium/F-12, 20% knockout serum, 1% non-essential amino acids, 0.5% GlutaMAX, 0.1 mM β-mercaptoethanol and 1% penicillin–streptomycin) supplemented with the SMAD inhibitors SB431542 (10 μM) and dorsomorphin (5 μM). Between days 6 and 24, organoids were cultured in neural differentiation medium (Neurobasal-A medium, 2% B27 supplement, 1% GlutaMAX and 1% penicillin–streptomycin) supplemented with human recombinant EGF (20 ng ml⁻¹) and human recombinant FGF2 (20 ng ml⁻¹). Between days 25 and 43, organoids were maintained in neural differentiation medium supplemented with human recombinant brain-derived neurotrophic factor (20 ng ml⁻¹) and human recombinant neurotrophin 3 (20 ng ml⁻¹). For MGEOs, the media were also supplemented with 5 μM wnt inhibitor IWP-2 on days 4–23, 100 nM smoothened agonist on days 12–23, 100 nM retinoic acid on days 12–15 and 100 nM allopregnanolone on days 16–23 for ventral forebrain patterning. Cortical organoids were not supplemented with IWP-2, smoothened agonist, retinoic acid and allopregnanolone. Each organoid was then moved to six-well plates for long-term culture after week 5. All media and supplements used for organoid cultures were the same as those in a previously published protocol^37,38.

Induced microglia

Induced microglial cells were generated from WA01/H1 or WA09/H9 hESC cells using STEMdiff kits, according to the manufacturer’s protocols. In brief, hESCs were differentiated into CD43-expressing haematopoietic progenitor cells for 12 days using a STEMdiff Hematopoietic Kit (STEMCELL; 05310). Haematopoietic progenitor cells were differentiated for 24 days using the STEMdiff Microglia Differentiation Kit (STEMCELL; 100-0019) and matured for an extra 4 days using the STEMdiff Microglia Maturation Kit (STEMCELL; 100-0020) before being added to the organoid cultures for co-culture.

iMG–organoid engraftment and co-culture

Mature iMG were immediately added to 4-week-old MGE organoids in 96-well ultra-low attachment PrimeSurface plates at 80–100 × 10³ microglia per organoid. Trophic factors (100 ng ml⁻¹ of IL-34 (PeproTech; 200-34), 25 ng ml⁻¹ of CSF1 (PeproTech; 300-25) and 50 ng ml⁻¹ TGFβ1 (PeproTech; 100-21)) were added to the culture medium to support microglial survival. One wpt, co-cultured organoid–microglia (neuroimmune organoids) were transferred to a six-well plate and placed on an orbital shaker. The co-cultures were then maintained following the usual organoid maintenance protocol, with the addition of trophic factors.

Pharmacological manipulation of organoids

Six-week-old organoids were treated with PBS, 100 ng ml⁻¹ of recombinant human IGF1 (Abcam; ab269169), 1 μg ml⁻¹ of IGF1-neutralizing antibodies (Abcam; ab9572), 1 μM GSK4529 (GSK1904529A; Selleckchem; S1093) or 1 μΜ picropodophyllin (Selleckchem; S7668) for 48 h. Then 10 μΜ BrdU (Abcam; ab142567) was added during the last 4 h to label proliferating cells. Organoids were collected immediately after the treatment for IHC analysis. For the DAPT treatment experiment, PBS or 10 μM DAPT (Abcam; ab120633) was applied to MGEOs transplanted with iMG from 10 to 14 dpt. Organoids were then collected at 14 dpt for IHC analysis.

Immunohistochemistry

We followed the IHC protocol, as previously reported^14,32. Human tissue samples were fixed and cryosectioned, as described above. Mouse samples were prepared, as described above in ‘Mice’. Organoids were fixed in 4% PFA for 30–45 min at room temperature and cryopreserved in 30% sucrose in PBS overnight. The organoids were then embedded in OCT and cryosectioned at 14 μm using a Leica cryostat.

The mounted human slides were defrosted overnight at 4 °C and then dried at 37 °C for 3 h. The mounted organoids and mouse slides were dried directly at 37 °C for 30 min. Antigen retrieval was performed for 5–12 min at 95–99 °C using antigen retrieval buffer (BD Pharmingen; 550524). After antigen retrieval, tissue slices were washed with 1× PBS plus 0.1% or 0.3% Triton X-100 and then blocked in blocking buffer (5–10% serum, 1% bovine serum albumin (BSA) and 0.1% Triton X-100 in PBS, or 1% BSA in 0.3% Triton X-100 in PBS) for 1–1.5 h at room temperature before proceeding to incubation with primary antibodies (Supplementary Table 3) overnight at 4 °C. After washing, sections were incubated with species-specific secondary antibodies conjugated to Alexa Fluor dyes (1:500; Invitrogen) for 1.5–2 h at room temperature. For human and embryonic mouse slides, TrueBlack Lipofuscin Autofluorescence Quencher (1:20 in 70% alcohol; Biotium; 23007) was applied for 3–5 min to block autofluorescence. For EdU staining, the EdU working solution was applied to embryonic mouse brain slices after secondary antibody application following the manufacturer’s instructions. Nuclei were counterstained with DAPI (1:1,000 from 1 mg ml⁻¹ of stock; Invitrogen; 2031179) for 5 min. Images were captured using a Leica STELLARIS 8 confocal microscope. For organoid experiments, three slices of each organoid were imaged, quantified using ImageJ (1.54) and averaged for the final statistical analysis.

Three-dimensional reconstruction and image analysis

Three-dimensional reconstructions were generated using the Imaris software (Oxford Instruments). For distance analysis, microglia were reconstructed using surface modules, whereas Ki-67⁺ or DAPI⁺ cells were reconstructed with spot modules. The distance from the centre of each cell (spot) to the nearest microglia (surface) was determined using Imaris. The distance distributions of the Ki-67⁺ and Ki-67⁻ cells to the nearest microglia were calculated accordingly.

Single-nucleus preparation

Single-nucleus suspensions were prepared from postmortem human samples. About 50 mg of sectioned freshly frozen human brain tissue was homogenized in lysis buffer (0.32 M sucrose, 5 mM CaCl₂, 3 mM MgAc₂, 0.1 mM EDTA, 10 mM Tris-HCl, 1 mM dithiothreitol and 0.1% Triton X-100 in diethyl pyrocarbonate-treated water) plus 0.4 U μl⁻¹ of RNase inhibitor (Takara; catalogue no. 2313A) on ice. Then, the homogenate was loaded into a 30-ml-thick polycarbonate ultracentrifuge tube (Beckman Coulter; catalogue number 355631), and 9 ml of sucrose cushion solution (1.8 M sucrose, 3 mM MgAc₂, 1 mM dithiothreitol and 10 mM Tris-HCl in diethyl pyrocarbonate-treated water) was added to the bottom of the tube. The tubes with tissue homogenate and sucrose cushions were then ultracentrifuged at 107,000g for 2.5 h at 4 °C. The pellet was recovered in 250-μl ice-cold PBS for 20 min, resuspended in nuclei sorting buffer (PBS, 1% BSA, 0.5 mM EDTA and 0.1 U μl⁻¹ of RNase inhibitor) and filtered through a 40-μm cell strainer to obtain single-nucleus suspensions for FACS/fluorescence-activated nucleus sorting.

Single-cell preparation

Single-cell suspensions of 1323-4 hiPSC-derived organoids were prepared using neural tissue dissociation kits (P) (Miltenyi Biotec; 130-092-628) following the manufacturer’s instructions. In brief, 12–16 organoids per experimental condition were processed through a gentle two-step enzymatic dissociation procedure, as instructed. Five mg ml⁻¹ of Actinomycin D (Sigma-Aldrich; A1410), 10 mg ml⁻¹ of anisomycin (Sigma-Aldrich; A9789) and 10 mM triptolide (Sigma-Aldrich; T3652) were added before tissue digestion to inhibit the cellular transcriptome. Following digestion, organoids were mechanically triturated using fire-polished glass pipettes, filtered through a 40-μm cell strainer test tube (Corning; 352235), pelleted at 300g for 5 min and washed twice with Dulbecco’s phosphate-buffered saline (DPBS) before proceeding to 10× genomics scRNA library preparation. For samples that needed FACS, the single-cell pellet was resuspended in cell sorting buffer (DPBS, 1% BSA and 0.1 U μl⁻¹ of RNase inhibitor).

FACS and fluorescence-activated nucleus sorting

Single-nucleus suspensions from fresh-frozen human samples were stained with antibodies of PU.1 (Cell Signaling Technology; 81886S; 1:100) and OLIG2 (Abcam; ab225100; 1:2,500) overnight at 4 °C. PU.1 and OLIG2 antibodies were conjugated with fluorescence upon purchase. DAPI (1:1,000) was added for 5 min on the second day. Single-cell suspensions from organoids were stained with DAPI (1:1,000) for 5 min in cell sorting buffer (DPBS; 1% BSA and 0.1 U μl⁻¹ of RNase inhibitor). The single-nucleus/cell suspension was then centrifuged at 300g for 5 min, resuspended in nucleus/cell sorting buffer and filtered through a 40-μm cell strainer for final analysis and sorting using a FACSAria II Cell Sorter (BD Biosciences). Target cells were collected in nucleus/cell sorting buffer for future sequencing library preparation.

Single-cell and single-nucleus RNA library preparation

Nuclei and cells were counted using a haemocytometer and resuspended to a final concentration of 300–1,000 cells/nuclei per microlitre in PBS. Single-nucleus/cell RNA-seq libraries were prepared using 10× Genomics Chromium Next GEM Single Cell v.3.1 kit according to the manufacturer’s instructions, targeting for 5,000 nuclei/cells per sample. Single-cell/nucleus libraries were then sequenced on the NovaSeq 6000 machine, with a sequencing depth of 50,000 reads per cell.

Single-cell and single-nucleus RNA-seq data analysis

Sequencing results were then aligned to the GRCh38 genome (gex-GRCh38-2020-A) using Cell Ranger v.6.1.2 (10× Genomics). Then ‘–include-introns’ was used to include premature messenger RNA in single-nucleus samples. Gene counts then underwent a doublet removal step using DoubletFinder v.2.0.3 (https://www.cell.com/cell-systems/fulltext/S2405-4712(19)30073-0).

The output (count matrix) was used as the main input file for all downstream analyses using Seurat v.5.1.0. For human snRNA-seq, nuclei with UMIs of less than 1,000, gene features of less than 1,000 or more than 100,000 or percentage of mitochondrial genes of more than 3% were filtered out. For organoid scRNA-seq, cells with UMIs of less than 800 or more than 50,000, gene features of less than 500 or more than 10,000 or percentage of mitochondrial genes less than 2% or more than 25% were filtered out. For FACS-isolated iMG scRNA-seq, cells with UMIs of less than 1,000 or more than 80,000, gene features of less than 1,000 or more than 20,000 or mitochondrial genes less than 20% were filtered out. MALAT1, mitochondrial genes (MT-), ribosomal protein-encoding genes (RPS- and RPL-) and haemoglobin genes (HB-) were excluded from further analysis. Standard data normalization, variable feature identification, linear transformations, dimensional reduction, UMAP embedding and unsupervised clustering were conducted using the standard Seurat pipeline³⁵. Cell-type cluster identification was performed on the basis of the expression of known marker genes, as shown in Extended Data Figs. 2, 7 and 10. For scRNA-seq of GFP-labelled iMG, iMG were purified in silico using canonical microglia/macrophage markers, including AIF1, CX3CR1, C3, PTPRC, ITGAM and CD68.

We analysed cell–cell interaction using CellChat v.2 (ref. ³⁶). For development-based analysis, independent CellChat files were generated from ‘embryonic’ and ‘perinatal’ Seurat objects, and a comparison analysis was conducted between them. A heat map was created using GraphPad Prism 9 according to the interacting probability of significant ligand–receptor interactions involved in microglial regulation of interneurons (CIN).

DEG analysis was conducted on the basis of the Seurat-default non-parametric Wilcoxon rank-sum test. Pathways with enriched DEGs were generated using Enrichr (https://maayanlab.cloud/Enrichr/#) on the basis of the Reactome Pathway Database, Kyoto Encyclopedia of Genes and Genomes, GEO and Gene Ontology database. The full names of the pathways shown in Fig. 4 are as follows: IGF1R 46: IGF1R drug inhibition 46 (kinase perturbations from GEO down; GSE14024); IGF1R 52: IGF1R knockdown 52 (kinase perturbations from GEO down; GSE16684); mitotic sister: mitotic sister chromatid segregation (GO:0000070); aerobic electron: aerobic electron transport chain; respiratory: respiratory electron transport, ATP synthesis by chemiosmotic coupling, heat production by uncoupling proteins (R-HSA-163200).

Principal component analysis

The published sequencing datasets for comparison were collected from eleven previous papers^{39,41,42,43,44,45,46,47,48,49,50}. The specific papers and corresponding NIH GEO datasets used were as follows: GSE89189 (ref. ³⁹); (GSE123021, GSE123022, GSE123024 and GSE123025) (ref. ⁴¹); GSE121654 (ref. ⁴²); GSE141862 (ref. ⁴³); (GSE133345 and GSE137010) (ref. ⁴⁴); GSE180945 (ref. ⁴⁵); GSE178317 (ref. ⁴⁶); (GSE139549 and GSE139550) (ref. ⁴⁷); GSE85839 (ref. ⁴⁸); GSE97744 (ref. ⁴⁹); GSE99074 (ref. ⁵⁰). Each dataset was collected, filtered and grouped by appropriate characteristics, including species, real/derived, bulk/single cell, age and protocol details. To facilitate comparison, the groups within each set were pooled into single representations.

Once the data were collected and preprocessed, the pooled samples were processed using Scanpy v.1.10.3 (https://github.com/scverse/scanpy). Specifically, the cells were normalized by total counts over all genes using scanpy.pp.normalize_total. They were then logarithmized using scanpy.pp.log1p. For use in downstream PCA, highly variable genes were calculated using scanpy.pp.highly_variable_genes. The number of top genes was configured to 2,000. In the final steps, PCA was performed using scanpy.tl.pca (default of 50 components), and a scatter plot using the coordinates of PCA 1 and 2 was plotted for each cell representation using scanpy.pl.pca.

CRISPR–Cas9 gene editing

A WA09/H9 stem cell line with an IGF1 loss-of-function mutation (IGF1 knockout) was generated using CRISPR–Cas9-based non-homology end joining, largely following the protocol of the Alt-R CRISPR–Cas9 System from Integrated DNA Technologies (IDT). The guide RNA (5′TCGTGGATGAGTGCTGCTTC3′) was selected from the predesigned Alt-R CRISPR–Cas9 guide RNA (IDT). Equal amounts of CRISPR RNA and ATTO 550 labelled tracrRNA (IDT; 1075927) were mixed to a final concentration of 100 μΜ, heated to 95 °C for 5 min and then cooled to room temperature for annealing followed by the formation of the ribonucleoprotein complex with Alt-R S.p. HiFi Cas9 Nuclease V3 (IDT; 1081061) at room temperature for 20 min. The ribonucleoprotein complex was delivered to single-stem-cell suspensions using the Neon Electroporation System (1,400 V; 20 ms; one pulse) according to the manufacturer’s instructions. After electroporation, ATTO 550+ cells were selected by FACS after 3 days of culture and sparsely seeded to form a single-cell colony. A loss-of-function mutation cell line was selected by Sanger sequencing with out-of-frame mutations at the target site, followed by exclusion of any mutations at the top 5 potential off-target sites. Further Sanger sequencing confirmation, reverse transcription–quantitative polymerase chain reaction and IHC were performed to confirm IGF1 knockout.

Time-lapse imaging of microglia–MGE progenitor interactions

To visualize the interactions between engrafted microglia and MGE progenitors in MGEOs, we used MGE organoids generated from NKX2.1-GFP cells and iMG derived from tdTomato-labelled WA09/H9 cells using lentiviral transduction (SignaGen Laboratories; SL100289). Live imaging was performed 2–3 weeks after iMG transplantation. For imaging, engrafted organoids were transferred to a flat glass-bottom six-well plate, with one organoid per well with 500 μl of culture medium. Time-lapse imaging was conducted using a Leica SP8 confocal microscope at 37 °C and 5% CO₂. Z-stacks were captured every 5 min over a 12-h period and processed using maximum intensity projections to visualize dynamic cellular interactions.

Data analysis, statistics and presentation

For all quantifications, images were acquired and quantified blindly to genotype or treatment. Statistical analyses were performed using GraphPad Prism (v.10.1.0), as shown in each figure legend.

Reporting summary

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