Mice and in utero surgeries

All experiments were conducted according to institutional guidelines of the Max Planck Society and the regulations of the local government ethical committee (Beratende Ethikkommission nach §15 Tierschutzgesetz, Regierung von Oberbayern). All mouse colonies were maintained in accordance with protocols approved by the Bavarian government at the Max Planck Institute for Biological Intelligence or the Helmholtz Zentrum in Munich. Mice at the Max Planck Institute were group housed in isolated ventilated cages (room temperature 22 ± 1 °C, relative humidity 55% ± 5%) under a 12 h dark/light cycle with ad libitum access to food and water. C57BL/6NRj wild-type females (from inhouse breeding) were crossed to C57BL/6NRj wild-type or to CAS9-EGFP (B6.Gt(ROSA)26Sortm1.1(CAG-cas9*,-EGFP)Fezh/J, JAX, 026179) males10. Embryos were staged in days post coitus, with E0.5 defined as 12:00 of a day that a vaginal plug was detected after overnight mating. Timed pregnant mice were anesthetized with isoflurane (5% induction, 2.5% during the surgery) and treated with the analgesic Metamizol (WDT). A microsyringe pump (Nanoject III Programmable Nano-liter Injector, DRUM3-000-207) was used to inject ~700 nl of DNA plasmid solution made of 0.6 μg μl−1 of pEF1a-pBase (piggyBac transposase; a gift from R. Platt) and the sgRNA plasmid 0.7 μg μl−1, diluted in endo-free TE buffer and 0.002% Fast Green FCF (Sigma, F7252), into the lateral ventricle. pCAG-Cas9-EGFP (a gift from R. Platt) plasmid was added when wild-type males were used for plugs. For TrackerSeq experiments, a barcode library (final concentration of 0.4 μg μl−1) was added to the DNA plasmid solution. Embryos were then electroporated by holding the head between platinum-plated tweezer electrodes (5 mm in diameter, BTX, 45-0489) across the uterine wall, while five electric pulses (35 V, 50 ms at 1 Hz) were delivered with a square-wave electroporator (BTX, ECM830)59. We used these relatively large electrodes to target all areas of the GE (MGE, CGE and LGE)60. Pups were kept with their mothers. To assess cellular distribution after in utero electroporation, embryos were collected at E16.5 and E18.5. Dissected brains were fixed overnight in 4% paraformaldehyde (Electron Microscopy Sciences, 15710) and washed with PBS. The 50 μm tissue sections were prepared on a Leica VT1200S Vibratome and mounted on slides with ProLong Glass Antifade Mountant (P36980, ThermoFisher). All images were acquired using a STELLARIS 5 confocal microscope system (Leica). For immunohistochemistry, C57BL/6 wild-type brains were prepared from three E13.5 embryos, postfixed in 4% PFA solution for 2.5 h and subsequently washed with PBS.

TrackerSeq library preparation and validation

TrackerSeq is a piggyBac transposon-based61 lineage tracing tool that is compatible with the 10x Genomics Chromium platform8. It records clonal lineages of single cells through the integration of oligonucleotide sequences into the genome of mitotic progenitors. Each lineage barcode is a 37 bp long synthetic nucleotide that consists of short random nucleotides bridged by fixed nucleotides. We followed the protocols from ref. 8 to prepare TrackerSeq plasmids. Briefly, an oligo library was cloned downstream of the Read2 partial primer sequence in the purified donor plasmid via Gibson Assembly reactions (NEB, E2611S). Gibson assembly reactions were then pooled and desalted with 0.025 μm MCE membrane (Millipore, VSWP02500) for 40 min and concentrated using a SpeedVac. A total of 3 μl of the purified assembly was incubated with 50 μl of NEB 10-β-competent Escherichia coli cells (NEB, C3019H) for 30 min at 4 °C, then electroporated at 2.0 kV, 200 Ω, 25 μF (Bio-Rad, Gene Pulser Xcell Electroporation Systems). Electroporated E. coli were incubated for 90 min shaking at 37 °C and plated on prewarmed sucrose/ampicillin plates. The colonies were scraped off the plates 8 h later and the plasmids were grown in LB medium with ampicillin up to optical density 0.5. The plasmid library was purified using a column purification kit (Zymo, D4202). We first assessed the integrity of the TrackerSeq barcode library by sequencing it to a depth of ~42 million reads to test whether any barcode was over-represented. Around 3.6 million valid lineage barcodes which had a quality score of 30 or higher were extracted from the R2 FASTQ files using Bartender62. One-thousand barcodes were randomly sampled from the extracted lineage barcodes to assess hamming distance. To group similar barcodes into putative barcodes, Bartender assigns a UMI to each barcode read to handle polymerase chain reaction jackpotting errors and clusters them. The cluster distance was set to 3. A total of 2 × 105 clusters of barcodes were identified.

Immunostainings

Paraformaldehyde-fixed brains at E13.5 and E18.5 were incubated in 10%, 20% and 30% sucrose for 24 h each, embedded in Neg-50 Frozen Section Medium (Epredia, 22110617) and subsequently snap-frozen in isobutane at −70 °C. The 16 μm tissue sections were prepared on a Thermo Scientific CryoStar NX70 Cryostat and transferred to glass slides. Sections were incubated overnight with primary antibodies anti-MEIS2 (SCBT, sc-515470-AF594, 1:250), anti-LHX6 (SCBT, sc-271433-AF488, 1:50), anti-PROX1 (R&D Systems, AF2727, 1:250) and anti-CTIP2 (Abcam, ab18465, 1:500). Sections were then incubated with secondary antibodies at room temperature for 2 h at 1:500 dilution: anti-rabbit AF594 (Invitrogen, A21207); anti-rat AF488 (Invitrogen, A21208); and anti-goat AF488 (Invitrogen, A11055). Nuclei were counterstained with DAPI and slides mounted with Aqua-Poly/Mount (Polysciences, 18606). Fluorescence imaging was conducted on a LSM880 confocal microscope (Zeiss Microscopy) using Plan-Apochromat 20/0.8 M27 or C-Aprochromat 63×/1.2 W Korr M27 objectives.

Sample collection

Before preparing brain tissue for scRNA-seq, each brain was examined under a stereo microscope and only brains that met the following criteria were selected for scRNA-seq:

(1)

Dispersed tdTomato+ neurons throughout the neocortex. This indicates that we targeted MGE/CGE-derived INs which migrate long distances and disperse to different cortical brain regions.

(2)

Dense tdTomato+ neurons throughout the striatum. MSNs are known to originate from the LGE and account for ~90% of the neurons in the striatum.

(3)

tdTomato+ neurons in the olfactory bulb. GABAergic precursors are known to migrate from the LGE to the olfactory bulb.

We performed immunohistochemical labeling to validate that after in utero electroporation, individual brains express sgRNAs in cortical INs derived from the MGE (anti-SST) and CGE (anti-PROX1), as well as in striatal MSNs derived from the LGE (anti-CTIP2). We collected electroporated brains from mouse embryos (both sexes) at E16.5 in ice-cold Leibovitz L-15 Medium (ThermoFisher, 11415064) with 5% FBS or at P7–8 in ice-cold Hibernate-A Medium (ThermoFisher, A1247501) with 10% FBS and B-27 supplement (ThermoFisher, 17504044). Forebrain tissue was manually dissected. A papain dissociation system (Wortington, LK003150) was used according to the protocol described in ref. 40 on the gentleMACS Octo Dissociator (Miltenyi Biotec) to generate a cell suspension. To isolate positive cells, flow cytometry was performed using a BD FACSAria III Cell Sorter (BD FACSDiva Software, v.8.0.2) with a 100 μm nozzle. EGFP and tdTomato+ cells were collected in bulk to test sgRNA Meis2 knockout efficiency following the in vitro protocol (above; results in Supplementary Table 1) or for downstream processing on the 10x Genomics Chromium platform. After sorting in PBS (Lonza, 17-516) with 0.02% BSA (B9000, NEB), 5,000–16,000 individual cells per sample were loaded onto a 10X Genomics Chromium platform for gel beads-in-emulsion and complementary DNA generation, carrying cell- and transcript-specific barcodes using the Chromium Single Cell 3' Reagent Kit v.3.1 with Feature Barcoding technology (10X Genomics, PN-1000121) following the manufacturer’s protocol (document no. CG000205, 10X Genomics).

tCROP-seq

To investigate the effects of TF perturbation on cellular fate decisions in a sparse population of precursors in the GE, we modified CROP-seq9, a method that enables CRISPR–Cas perturbation with scRNA-seq readout. Instead of lentiviral vectors, we applied a piggyBac transposon-based strategy (tCROP-seq) and in utero electroporation to deliver sgRNAs to cycling progenitors in the GE (Fig. 1b). The transposon system allows genes to be stably integrated into the genomes of electroporated cells and thus to be transmitted to their postmitotic daughter cells61. This increases the pool of perturbed cells and ensures that the perturbation occurs during a period covering the peak of neurogenesis8. We also added specific capture sequences to the sgRNA vectors which efficiently link sgRNAs to cell barcodes and enable sequencing of the protospacer from the transcriptome63. The tCROP-seq sgRNA vectors also encode tdTomato to enable the labeling and enrichment of perturbed neurons. The efficiency of sgRNA Meis2 to induce frame-shift mutations was validated in vitro and in vivo before the tCROP-seq experiments (Supplementary Table 1).

Preparation of tCROP-seq libraries

We used the Feature Barcode technology from 10X Genomics to prepare tCROP-seq libraries. The assay captures transcriptomes and guide RNAs from the same cell. We generated 3' gene expression and gRNA libraries according to the manufacturer’s manual (document no. CG000205) using the Chromium Library v.3.1 kit (PN-1000121), Feature Barcode Library Kit (PN-1000079) and Single Index Kit (PN-1000213) from 10X Genomics. The quantification of the libraries was performed with an Agilent BioAnalyzer.

Preparation of TrackerSeq NGS libraries

The TrackerSeq lineage libraries were amplified from 10X Genomics cDNA libraries with the Q5 polymerase (NEB, M094S) in a 50 μl reaction, using 10 μl of cDNA as template8. Specifically, each PCR contained: 25 μl of Q5 High-fidelity 2X Master Mix, 2.5 μl of 10 μM P7-indexed reverse primer, 2.5 μl of 10 μM i5-indexed forward primer, 10 μl of molecular grade H2O, 10 μl of cDNA (for primer sequences and indices, see Supplementary Table 4). Libraries were purified with a dual-sided selection using SPRIselect (Beckman Coulter, B23318) and quantified with an Agilent BioAnalyzer.

Sequencing and read mapping

We sequenced the transcriptome and CRISPR barcode libraries using an Illumina NextSeq 500 at the Next-Generation Sequencing Facility of the Max Planck Institute of Biochemistry or a NovaSeq at the Genomics Core Facility at the Helmholtz Center in Munich. Full details on each dataset are provided in Supplementary Table 2. The sequencing reads in FASTQ files were aligned to a reference transcriptome (mm10-2.1.0) and converted into UMI counts using the 10X Genomics Cell Ranger software (v.3.0.2 or 5.0.1).

tCROP-seq preprocessing

We loaded the UMI count data into R and processed it using the Seurat (v.4) package13. To recover the CRISPR gRNAs, we used Cell Ranger64, which produced a CSV file listing the cell barcodes and the sgRNA detected for each cell.

Processing embryonic tCROP-seq datasets

Electroporation of ventral progenitors using the 5 mm electrode targets additional progenitors located adjacent to the GE. These include progenitors of excitatory neurons located at the border between the pallium and the subpallium. Thus, our dataset consisted of: inhibitory, 16,098 neurons; excitatory, 10,010 neurons; glial, 5,915 cells; pericytes, 1,008 cells; fibroblasts, 537 cells; macrophages, 523 cells; and blood, 390 cells. We focused only on cells from inhibitory clusters where a gRNA could be recovered and excluded the others. We integrated inhibitory neurons with scRNA-seq datasets from wild-type mice8 to get a higher resolution of inhibitory cell states (Fig. 1) using the integration tool from Seurat13. We obtained cluster-specific marker genes by performing differential expression analysis (see below). Clusters were assigned to cell types on the basis of the expression of known marker genes, primarily using http://mousebrain.org/development/ (ref. 65) and https://DropViz.org (ref. 66).

Processing postnatal tCROP-seq datasets

To process the P7 datasets, we integrated Harmony (v.1.0)12 into our Seurat13 workflow for batch correction, using default settings (theta = 2, lambda = 1, sigma = 0.1). We used the first 30 Harmony embeddings for uniform manifold approximation and projection (UMAP) visualizations and clustering analysis. To group cells into clusters, we first constructed a shared-nearest neighbor graph from Harmony embeddings using the FindNeighbors() algorithm, then input the graph into the FindClusters() function in Seurat (dimensions = 30, res = 0.8). To test whether our postnatal dataset was subject to non-specific background expression, we applied DecontX67 using the default parameters. We retrieved the count matrix from our Seurat object, created an SCE object, ran DecontX and then added the corrected count matrix back to the Seurat object. The difference before and after correction was relatively small. Therefore, we decided to use the uncorrected counts for the subsequent analysis.

Logistic regression model to predict IN and PN genes

We used a recently published scRNA-seq dataset from ref. 8 to explore genes that are predictive for IN or PN fate. Raw counts for samples from GE-specific microdissections collected from wild-type mice at E13.5 and E15.5 were processed using Seurat (v.4.1.0)13. After integration across batches, counts were normalized and scaled. Cluster annotations from ref. 8 were summarized into four broad cell classes: mitotic, trunk, IN and PN. For performing logistic regression, we subsetted cells from IN and PN cell classes. Logistic regression was performed using the 3,000 most variable genes. To account for balanced design, cells were subsampled to have a equal number of cells in both classes. A logistic regression model was trained on the scaled expression matrix of the corresponding cells and genes, where two-thirds of cells were used for training and the other third for validation. This was implemented using the cv.glmnet(family = ”binomial”) function from the R package glmnet68. The model achieved 99.15% accuracy on the held-out validation set. For each gene, the model predicts a coefficient which reflects whether high expression of the gene is predictive of a cell being an IN (coefficient ∈ [0,0.5]) or a PN (coefficient ∈ [0.5,1]).

Comparing cell-type composition between perturbations

We compared the perturbation effect on cell-type composition using the method described by ref. 40. A script of the analysis is deposited on GitHub (https://github.com/mayer-lab/Dvoretskova-et-al). Compositional change was investigated using the CellComp_Poisson R function from ref. 40. It performs Poisson regression analysis to identify genes that are differentially expressed across different cell types, perturbations and batches. First, the function performs data cleaning by creating a metadata data frame and filtering out cells with low counts. It then fits a Poisson regression model for each combination of cell type and perturbation and extracts the coefficients for the perturbation variable. These coefficients are then used to calculate P values and adjusted P values for each gene.

Differential gene expression analysis

We used the Libra package (v.1.0) to perform differential gene expression analysis69. We ran the run_DE function on Seurat objects using the following parameters: de_family = pseudobulk, de_family = pseudobulk, de_method = edgeR, de_type = LRT. We obtained DEGs of PNs or INs by using the run_DE function on cells grouped into classes (mitotic, PNs and IN). We filtered for statistically significant genes (false discovery rate (FDR)-adjusted P value threshold = 0.05). Genes were considered differentially expressed if log2(fold change) <−0.3 and >0.3 for embryonic and log2(fold change) <−1 and >1 for postnatal datasets.

We also used the R packagage Libra to calculate the DEGs for each cluster (i_Calb2/Nxph1, i_Cck/Reln, i_Ebf1/Zfp503, i_Foxp1/Isl1, i_Foxp1/Six3, i_Isl1/Bcl11b, i_Lhx6/Npy, i_Meis2/Bcl11b, i_Nfib/Tcf4, i_Nr2f2/Nnat, i_Tiam2/Zfp704 and i_Tshz1/Pbx1). The result of the DEG analysis is in the Source Data of Fig. 3. We applied thresholds (adjusted P ≤ 0.05 and log2(fold change) <−1.0 and >1.0) to select the genes for intersection with the ChIP–seq data. For the Venn diagram, we combined DEGs from all subtypes and split them into upregulated or downregulated genes.

TrackerSeq (lineage tracing) barcode processing and analysis

For a subset of datasets (ED210204, ED210215, ED211111 and ED211124), we included TrackerSeq lineage barcodes to perform a clonal analysis. We followed the protocol outlined in ref. 8 to process the TrackerSeq barcodes to obtain cloneIDs for each corresponding cell barcode. The resulting cloneIDs were added to the Seurat object metadata. To quantify clonal relationships between cell classes, the inhibitory clusters were first merged into cell classes (Fig. 2c) on the basis of whether they were annotated as mitotic (Ube2c and Top2a) or as INs and PNs (Gad2). The UpsetR library was used to count the number of clones shared between the neuronal classes, as well as the proportion of clonal relationships in gMeis2 and gLacZ datasets. The set size is the number of cells in the class. The Upset bar plot shows the calculated proportion of each type of clonal distribution category within the perturbation. The calculated percentage stemmed from dividing the number of clones in a given category (for example, clones containing only mitotic cells and IN) by the total count of clones spread across all clonal distribution categories.

To assess clonal coupling, we used a method from ref. 70. The method computes an observed/expected ratio of shared barcodes for each pair of cell states. A barcode is considered shared if it appears in at least one cell from both states. From the observed shared barcode matrix Oij, it derives an expected shared barcode matrix Eij under the assumption of no lineage couplings, as follows:

$$_=\frac\nolimits__\times \mathop\nolimits__}\nolimits__}$$

These matrices were recomputed 1,000 times, each time using a random 25% sample of clones. The lineage coupling scores shown in Extended Data Fig. 2g represent the median Oij/Eij from these 1,000 randomized trials. To assess significance, we calculated empirical P values for each pair of cell states. An observed/expected ratio of 1 indicates lineage coupling that is in line with random expectations, a ratio of <1 or >1 indicates lower or higher lineage coupling, respectively. Empirical P values were calculated by counting the number of random shuffles, where the simulated observed/expected ratio was higher than 1 for negatively coupled pairs or lower than 1 for positively coupled pairs of cell states. Empirical P values were subsequently corrected for multiple testing using FDR correction.

Hotspot gene module analysis

Hotspot (v.0.91) is a tool for identifying co-expressing gene modules in a single-cell dataset41. It computes gene modules by evaluating the pairwise correlation of genes with high local autocorrelation, then clusters the results into a gene–gene affinity matrix. To identify the inhibitory-specific modules in the postnatal dataset, we first separated the Gad2-expressing inhibitory neuron population from the rest of the P7 dataset. We ran the depth-adjusted negative binomial model on the entire count matrix and Harmony (v.1.0) corrected principal components. We computed a k-nearest-neighbors graph with 30 neighbors, 9,154 non-varying genes were subsequently detected and removed. Autocorrelations between each gene were calculated and the top 500 significant (FDR ≤ 0.05) genes were used to evaluate pairwise gene associations (local correlations). After pairwise local correlations were calculated, we grouped genes into modules. Modules were created through agglomerative clustering, where the minimum number of genes per module was set to 30. Eight modules were identified and 103 genes were not assigned to a module. Summary per-cell module scores is calculated using the calculate_module_scores() function as described by ref. 41. As described by ref. 40, linear regression was used to test the relationship between perturbation and Hotspot module gene scores. We fitted a linear regression model that accounted for the batch and number of genes and extracted the effect sizes to estimate how the module scores in the perturbed cells deviated from gLacZ control cells40. For the three TFs, the perturbations had significant effects across different modules.

GO term analysis

Gene ontology (GO) term analysis was done using the package enrichR (v.3.0)71. The DEGs and module genes of each module were queried against the following databases: GO_Molecular_Function_2018, GO_Cellular_Component_2018 and GO_Biological_Process_2018. Only GO terms that were significant (adjusted P ≤ 0.05) were kept.

Luciferase assay

CREs were amplified from mouse genomic DNA with the Q5 polymerase (NEB, M0491) using primers listed in Supplementary Table 5 and cloned into pGL4.24[luc2P/minP] (Promega, E8421) with the NEBuilder HiFi DNA Assembly kit (NEB, E2621). The enhancer hs1080 had to be cloned in reverse-complement. Mouse Meis2 isoform D (4) (the tag was removed) and Lhx6 variant 1 (C-DYK) expression vectors were purchased from Genscript. Dlx5 and Pbx1 coding sequences were amplified from mouse cDNA and cloned into pcDNA3.1 (Genscript). The Meis2 vector was mutated with the NEBuilder HiFi DNA Assembly kit (NEB, E2621) to harbor the human mutation p.(Arg333Lys), c.998G>A (Meis2*333)7. A short version of the enhD1 luciferase vector was mutated using gBlock (IDT) and the NEBuilder HiFi DNA Assembly kit. Luciferase reporter vectors were cotransfected with pNL1.1.PGK[Nluc/PGK] (Promega, N1441) and different combinations of pcDNA3, pcDNA3-Dlx5, pcDNA3-PBX1, pcDNA3-Meis2 and pcDNA3-Lhx6. Neuro2a cells were seeded in 24-well plates at 80,000 cells per well and were transfected on the next day with TransIT-LT1 Transfection Reagent (Mirus, MIR 2300), using 150 ng of luciferase reporter, 10 ng of Nluc/PGK and 350 ng of total of pcDNA3.1 plasmids per well (150 ng per TFs vector). pcDNA stands for a control plasmid (pcDNA3.1) which does not contain a protein coding sequence. The pcDNA was used to balance the DNA load during transfections. Cells were harvested 24 h after transfection and luciferases activity was measured using the Nano-Glo Dual-Luciferase Reporter Assay System (Promega, N1630) on a Berthold Multimode reader Tristar2S. A Nanoluc reporter was used for normalization. Statistical tests were performed using the GraphPad Prism software (v.10.0.2). Two-way analysis of variance (ANOVA) followed by Tukey’s honestly significant difference (HSD) test were used to determine the statistical significance between various conditions. Data distribution was assumed to be normal but this was not formally tested. All results for statistical analysis are listed in Source Data files.

Chromatin immunoprecipitation

Mice were handled in accordance with the CNIC Ethics Committee, Spanish laws and the EU Directive 2010/63/EU for the use of animals in research. GEs and part of the underlying striatum of 70 wild-type C57BL/6 embryos at E14.5 were microdissected and immediately fixed in 1% formaldehyde for 5 min. Tissue preparation, immunoprecipitation and sequencing on an Illumina HiSeq2500 were performed as previously described50. Immunoprecipitation was carried out using a combination of two anti-MEIS, one recognizing MEIS1A and MEIS2A, the other recognizing all MEIS2 isoforms72.

ChIP–seq data analysis

Single-end reads of 61 bp length were trimmed using Cutadapt (v.1.16) and mapped to GRCm38 using Bowtie2 (v.2.3.0)73 followed by duplicate removal with Picard (v.2.15) and peak calling with MACS2 (v.2.1.2)74 using a cutoff of q ≤ 0.01. TSS definitions were adapted from the eukaryotic promoter database (mmEPDnew v.003)75. We determined the distance of each peak to the nearest TSS using the R package Plyranges (v.1.180). Using custom R scripts, peaks were assigned to the TSS of a gene when overlapping an ~5 kb region around a TSS, defined as promoter region. Overlap with developmental enhancers20 was determined in the same way. Similarly, we determined overlap of MEIS2-binding sites with DLX5-binding sites at E13.5 from ref. 4 and LHX6-binding sites at E13.5 from ref. 35. Enrichment of enhancer-overlapping peaks among shared MEIS2/DLX5 peaks, compared to MEIS2- and DLX5-exclusive peaks, was determined using Pearson’s Chi-squared test of the R stats package (v.4.0.2). Genomic tracks and VISTA enhancers26 were visualized using the Integrated Genomics Viewer (v.2.4.1)76.

Motif identification and enrichment of known motifs were carried out by HOMER (v.4.10.4)77 using default settings. Motif enrichment within enhancer- and promoter-overlapping peaks was likewise performed with HOMER. We used SpaMo (v.5.4.1)78 to determine motif spacing between MEIS2 and DLX5-binding motifs in common MEIS2/DLX5-binding sites, within 100 bp upstream and downstream of MEIS2 peak summits.

Data used in this study

GSE167047 (snATAC-seq of E12.5 MGE and LGE; ref. 27), GSE85705 (LHX6 ChIP–seq GE E13.5; ref. 35), GSE124936 (DLX1, DLX2 and DLX5 ChIP–seq GE E13.5; ref. 4) and GSE188528 (scRNA-seq of LGE, MGE, CGE E13.5; ref. 8) were downloaded from https://www.ncbi.nlm.nih.gov/geo. Coordinates of developmental enhancers and interacting genes were taken from ENCODE20. VISTA enhancer images were downloaded from the VISTA Enhancer Browser (https://enhancer.lbl.gov)26.

Statistics and reproducibility

Data distribution was assumed to be normal but this was not formally tested. The exact values of n indicating the total number of animals per group are reported in each figure caption or in the Source Data files. Analyses were carried out using Prism v.10.0.2, R v.3.6 and R v.4.1. The sample size was chosen empirically or based on preliminary data to provide a sufficient level of statistical power for detecting indicated biological effects. No statistical methods were used to predetermine sample sizes but our sample sizes are similar to those reported in previous publications8. No data were excluded from the analyses. The experiments were not randomized.

Reporting summary

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