Competing programs shape cortical sensorimotor–association axis development

Association and sensorimotor gene module curation

Independently generated spatiotemporal human brain exon microarray data and RNA-seq datasets were obtained from ref. ²⁶ (accessed through https://hbatlas.org/) and ref. ²⁸ (accessed through http://development.psychencode.org/#) and the developing macaque brain RNA-seq data across brain regions were obtained from ref. ²⁹ (accessed through http://www.evolution.psychencode.org/#). The estimated correspondence between macaque developmental timepoints and equivalent human developmental periods has been reported previously²⁹. We split association areas representative of pericentral programs into two groups: Af, which included the OFC, MFC, DFC/dlPFC and VFC; and At, which included the posterior STC and ITC. Sensorimotor groups representative of central programs contained M1C, S1C, A1C and V1C. Areas dissected during fetal development are depicted in Fig. 1a. When defining our GMs, we used human samples from period 3 to period 7 and macaque samples from predicted period 5 to predicted period 7 with at least two biological replicates for each predicted period (the earliest macaque sample corresponded to approximately human period 4 and there were no biological replicates). For each of the three sequencing datasets (human microarray²⁶, human RNA-seq ²⁸ and macaque RNA-seq ²⁹), we conducted differential expression analysis using all samples across ten regions for each period. For RNA-seq differential expression analysis, we retained genes with sufficiently large counts using the filterByExpr function from the edgeR⁸⁰ package and conducted trimmed mean of M values normalization using the normalizeCounts function from the tweeDEseq⁸¹ package. We then applied RNentropy⁸² to identify genes differentially expressed among the above ten neocortical regions in each developmental period. The resulting DEGs from RNentropy were further selected using criteria adapted from a previous study⁴¹ in each period. In brief, a gene to be considered as an upregulated DEG in a certain A/S subgroup, (1) there is at least one region in that subgroup where the gene is significantly upregulated; (2) the gene is not upregulated in any region of the opposing (S/A) groups; (3) the gene is underexpressed in at least 30% of the areas in the opposing (S/A) group; and (4) the gene is not in a module gene list of any of the opposing (S/A) subgroups. For exon microarray data in each period, to conduct an equivalent analysis to the RNA-seq data, we used limma⁸³ to compare each cortical region within a subgroup independently as treatment to all other cortical regions of the opposing group as replicates of control for each period. DEGs were selected with FDR < 0.05 and absolute foldchange (FC) > 1 (the same FC threshold used in RNentropy)⁸⁴. We then defined the overexpressed DEGs in a certain S/A subgroup in a certain period following the same criteria as we described for the RNA-seq data. Lastly, we merged the genes across all prenatal periods of each gene module within each dataset independently as the final module gene list respectively. To identify shared genes, we found the overlap of each concatenated module gene lists among three datasets.

To show the module gene expression in heat maps, we averaged the gene expression across genes in each module for each sample and took the median of the averaged expression values among samples from a given region and period. To generate continuous expression trajectories for each cortical region across developmental periods (log₂[time (days)]), we applied locally weighted scatterplot smoothing (LOESS). For each region, the median of the sample-level mean log₂-transformed expression values were modelled as a function of developmental period using the loess() function in R (stats package). The fitted LOESS models were then used to predict smoothed expression values at 1,000 equally spaced points spanning the observed developmental time range for each region.

Aggregated module PCA and Euclidean distance plots

PCA for plots in Figs. 1d and 2a and Extended Data Figs. 1d,h and 9c was performed using the prcomp function in R, using the expression matrix of either the combined shared sensorimotor and association GMs or combined SEMA7A and PLXNC1 expression. Ellipses for each region were centred on the mean of the points within the region, with their axes sized according to the s.d. values on each component. Euclidean distances between different peripheral and central groups were calculated based on the full module gene list for each developmental period. Smooth average lines were generated with a 95% confidence interval.

Gene set enrichment analysis for GMs

Gene set enrichment analysis (GSEA) of each GM was conducted using ClusterProfiler⁸⁵ based on reference databases including GO, Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome and WikiPathways.

Human diffusion MRI analysis

Diffusion MRI (dMRI) data were acquired from ex vivo human fetal brains at PCW13, PCW15 and PCW17 (refs. ^86,87,88). Cortical masks were generated using ITK-Snap (v.4.0.2)⁸⁹. Whole-brain tractography was performed using Diffusion Toolkit with a deterministic tracking algorithm, reconstructing fibre pathways based on dMRI data⁹⁰. To isolate corticocortical connections, only streamlines with both end points located within the manually delineated cerebral wall mask were retained. For region of interest (ROI)-based analysis, tracts were further filtered to include only those with at least one end point in a given cortical ROI. Surface ROIs were converted to volumetric space and slightly dilated to compensate for partial volume effects and minor misalignments between the segmentation and underlying diffusion data. Streamlines were visually inspected and cleaned to remove short and anatomically incorrect fibres. Tractograms were visualized in DSI Studio with brains aligned to a consistent orientation for cross-subject comparison⁹¹. To complement this, whole-brain corticocortical tracts were filtered by length using the anterior–posterior extent (L) of each brain as a reference. Thresholds of 0.25L and 0.50L were applied, in addition to unfiltered tracts (0L), to highlight the developmental progression of long-range association pathways. Tractograms were colour coded by principal diffusion orientation (red, left–right; green, anterior–posterior; blue, superior–interior) and visualized using TrackVis.

Individual gene module PCA

The PCA scores from this analysis is presented Fig. 1g and Extended Data Fig. 9a,b. For each of the two spatiotemporally distinct, molecularly defined modules (sensorimotor GM and association GM), we derived the first PC of gene expression using PCA separately. We then quantified the relationship between these two GM components, separately for six developmental periods (developmental periods 3–7, and adult). This enabled us to explore the dominant axis of variation specific to each module at each period.

For periods 3–7, we used the expression data from ref. ²⁶. We first computed the mean gene expression over replicate samples for each unique period, region and gene. Gene expression values were then normalized column-wise using min–max normalization to ensure comparability across genes. Then using the shared GM gene list (Supplementary Table 5), we extracted an 11-region × 43-gene matrix and an 11-region × 171-gene matrix for the sensorimotor GM and association GM, respectively.

For adult, we used the Allen Human Brain Atlas (AHBA) dataset. AHBA gene expression data were obtained using the abagen toolbox with the default parameters^31,92,93. The AHBA is an open-access database containing microarray gene expression data collected from six human post-mortem brains. The samples were assigned to brain regions in the Schaefer 400 region, 17 network atlas⁹⁴. Five regions lacked reliable gene expression so were not included in subsequent analysis (region ID, label: 59, lh_17Networks_LH_SomMotB_Cent_5; 173, lh_17Networks_LH_DefaultB_IPL_1; 252, rh_17Networks_RH_SomMotB_S2_5; 299, rh_17Networks_RH_SalVentAttnA_FrMed_2; 303, rh_17Networks_RH_SalVentAttnA_FrMed_4). The resulting gene expression matrix used was 395 regions by 15,632 genes. In this matrix, each cell contained the normalized gene expression level for a given region. Using the shared sensorimotor GM and association GM gene list described above, we then extracted a 395-region × 42-gene matrix and a 395-region × 161-gene matrix for the sensorimotor GM and association GM, respectively. Note that there was no gene expression data available for 11 GM genes in the AHBA dataset (BOC, DCHS2, DLX5, DSC2, GCNT2, HCRTR2, NEUROG2, STC1, STK32B, UBASH3B and WBSCR17).

For each period (3–7, adult), we then decomposed the region × gene matrix using PCA to identify the first principal component (PC1) separately for each GM. Each component is represented by a set of PC scores, one per brain region, defining the dominate axes of variation in gene expression, and a set of PC loadings that capture how strongly expression of particular genes contribute to a component. Note that principal component scores and loadings were sign-aligned for interpretability. Specifically, if the correlation between module PC1 scores and the average expression of the module was negative, we multiplied both the PC1 scores and loadings by −1. This step ensured that positive module PC1s consistently reflected relative gene enrichment within each module.

To evaluate the relationship between the modules, we used Spearman’s rank correlation between the sensorimotor GM PC 1 and the association GM PC 1. For the fetal data (periods 3–7), at each period, we evaluated statistical significance using a null model where we randomly shuffled gene expression values across the 11 regions for sensorimotor GM, computed the correlation between the permuted sensorimotor module and the observed association GM 10,000 times, and then compared to the observed correlation between the sensorimotor GM and association GM. For the AHBA dataset, we evaluated statistical significance using a spin-permutated Vasa null model (10,000 repetitions)^95,96.

Cortical surface renderings of this analysis (Fig. 1f), and subsequent plots using cortical surface renderings (Fig. 2c,e and Extended Data Figs. 6a and 9b) were generated using Surfplot (https://github.com/danjgale/surfplot/tree/main)^97,98. For the adult data, we used the fsLR 32k template, which is a standard human cortical surface-based template⁹⁹. For the fetal data, we used surfaces derived from dMRI data described above^86,87,88.

Correlation of SEMA7A and PLXNC1 expression in fetal and adult brain regions

We quantified the relationship between SEMA7A (sensorimotor-enriched gene) and PLXNC1 (association-enriched gene). For periods 3–7, we extracted region-wise normalized gene expression data for SEMA7A and PLXNC1 for each of the 11 cortical regions using ref. ²⁶. For adult³¹, we extracted expression data for the marker genes for 395 cortical regions. For each period, we then conducted separate PCAs for each gene and used Spearman’s rank correlation and tested for statistical significance between the PC1 for each of the two marker genes.

Disease gene enrichment analysis

We conducted a hypergeometric test to examine whether the indicated modules were enriched for genes associated with various diseases. We analysed gene lists associated with neurodegenerative, neurodevelopmental or psychiatric diseases, including Alzheimer’s disease, ADHD, anorexia nervosa, autism spectrum disorder, bipolar disorder, developmental delay, major depressive disorder, neuroticism and Parkinson’s disease, as well as intelligence quotient as described previously¹⁰⁰. We also tested whether GMs were enriched for diseases that are probably not associated with central nervous system development using genes identified in various genomic analysis and genome-wide association studies for coronary artery disease¹⁰¹, Crohn’s disease¹⁰², lupus¹⁰³, metabolic syndrome¹⁰⁴ and type 2 diabetes¹⁰⁵.

Post-mortem human and macaque brain tissue

Post-mortem human brain samples were collected at PCW17, PCW20 and PCW22. Parental or next of kin consent and approval by institutional review boards was obtained before tissue was collected. Rhesus macaque brain samples were collected post-mortem at PCD165. Whole slabs or whole hemispheres were post-fixed in 4% paraformaldehyde (PFA) for 48 h and then cryoprotected in an ascending gradient of sucrose 10%, 20%, 30% for 1 week at each timepoint.

Ethics statement regarding use of post-mortem human tissue

Tissue was handled in accordance with ethical guidelines and regulations for the research use of human brain tissue set forth by the NIH and the World Medical Association Declaration of Helsinki (https://www.wma.net/policies-post/wma-declaration-of-helsinki/).

All experiments using macaques were carried out in accordance with protocols approved by Yale University’s Committee on Animal Research and NIH guidelines.

Human tissue immunohistochemistry

Fixed frozen sections were equilibrated to room temperature and washed with 1× PBS for 10 min. The sections were additionally fixed with 1.6% PFA for 10 min at room temperature. To improve tissue adherence, the slides were then baked at 60 °C for 30 min and cooled to room temperature. The samples were then incubated in acetone for 10 min at room temperature. After washing three times with PBS, the slides were incubated with a sodium citrate buffer (10 mM citric acid monohydrate, 0.05% Tween-20, pH 6.0) and brought to a boil in a microwave followed by cooling to room temperature. After washing, the slides were transferred to a slide holder and incubated in autofluorescence quenching buffer (2.25% H₂O₂ and 10 mM NaOH in PBS) for 90 min at 4 °C. A high intensity broad-spectrum LED light source was placed over the slide holder during this time for additional photobleaching. After washing in PBS, the slides were blocked in blocking buffer containing 5% donkey serum, 1% BSA diluted in staining buffer (2.5 mM EDTA pH 8.0, 0.5× PBS, 0.25% BSA, 0.01% NaN₃, 0.122 M Na₂HPO₄, 0.078 M NaH₂PO₄ in double-distilled H₂O) for 45 min at room temperature. Next, primary antibodies diluted in blocking buffer was added to the slides and incubated overnight at 4 °C. The slides were then washed with staining buffer and post-fixed with 1.6% PFA at room temperature for 10 min. After washing with 1× PBS, the slides were incubated in ice-cold methanol at 4 °C for 5 min. The slides were washed with PBST (0.1% Tween-20 in PBS) and secondary antibodies (Jackson Labs) were added (1:500, diluted in 5% donkey serum, 1% BSA in PBST) and incubated at room temperature for 2 h. Next, DAPI nuclear stain was applied to the slides for 10 min at room temperature. After two washes with PBST and then PBS, the sections were mounted with Fluoromount-G (SouthernBiotech 0100-01). The following antibodies were used: hPLXNC1 (1:250, MAB 544232, R&D Systems), hSEMA7A (1:250, AF2068, R&D Systems), SLC17A6 (1:400, Abcam, ab305254).

Whole-mount fluorescence in situ hybridization

Whole-mount fluorescence in situ hybridization was performed according to the Tris-buffer-mediated retention of in situ hybridization chain reaction (isHCR) signal in cleared organs (TRISCO) method¹⁰⁶. Before dissection, P0 and P3 mice were perfused with 1× PBS. Dissected brains were fixed in 4% PFA in 1× PBS overnight at 4 °C with gentle agitation. Fixed brains were dehydrated through a graded methanol series in 1× PBS (20%, 40%, 60%, 80%, 100% and 100%), each for 1 h at room temperature with rotation. After dehydration, brains were transferred to fresh 100% methanol and incubated overnight at room temperature with rotation. Dehydrated brains were stored at −20 °C until use. The brains were transferred to 100% dichloromethane (DCM) and rotated overnight at room temperature. Subsequently, the brains were transferred back to 100% methanol and immersed twice for 1 h at room temperature with rotation. Brains were then treated in 5% H₂O₂ in methanol overnight at 4 °C with rotation. After two washes in 100% methanol for 1 h each at room temperature with rotation, brains were immersed and equilibrated in 50% formamide wash buffer (50% formamide, 5× saline sodium citrate (SSC), 9 mM citric acid (pH 6.0), 0.1% Tween-20, 50 μg ml⁻¹ heparin and 0.3% poly(vinylsulfonic acid, sodium salt) solution (PVSA) in H₂O) at room temperature. Subsequently, brains were transferred to 50% formamide hybridization buffer (50% formamide, 5× SSC, 9 mM citric acid (pH 6.0), 1× Denhardt’s solution (Thermo Fisher Scientific), 10% Tween-20, 50 μg ml⁻¹ heparin, 10% dextran sulfate and 0.3% PVSA in H₂O) and incubated overnight at 37 °C with rotation. The next day, the brains were transferred to 400 μl of formamide hybridization buffer containing 0.8 μl of DiYO-1 and 8 μl each of isHCR initiator probes targeting individual transcripts (Molecular Instruments). Brain samples were incubated for 3 days at 37 °C with gentle agitation (days 6–8). Target regions of each mRNA were as follows: Plxnc1, 1,177–2,135 base of RefSeq NM_018797.2; Cyp26b1, 460–1308 base of RefSeq NM_175475.3; Sema7a, whole sequence of GenBank BC057875. After hybridization, the brains were washed three times in 50% formamide wash buffer at 37 °C, followed by three times in 5× SSCT buffer (5× SSC, 0.1% Tween-20 and 0.3% PVSA in H₂O) at room temperature. Each wash lasted 1 h with rotation. Brains were equilibrated in amplification buffer (5× SSC, 0.1% Tween-20, 10% dextran sulfate, 0.3% PVSA in H₂O) overnight at 4 °C with rotation. Subsequently, the brains were transferred directly to 400 μl of amplification buffer containing 8 μl each of amplifier corresponding to the respective isHCR probe and incubated for 3 days at 4 °C with gentle inversion shaking. Hairpin formation of amplifiers was performed according to the manufacturer’s instructions. The emission wavelengths of the fluorophores conjugated to the amplifiers corresponding to each target mRNA were as follows: Plxnc1, 800 nm; Cyp26b1, 647 nm; Sema7a, 546 nm. After fluorescence labelling, the brains were washed twice in 5× SSCT buffer, followed by two washes in 0.1 M Tris-HCl (pH 7.0). Each wash was performed for 1 h at room temperature with rotation. The brains were transferred to fresh 0.1 M Tris-HCl (pH 7.0) and washed overnight at room temperature with rotation. The washed brains were treated twice with 100% methanol for 1 h at room temperature with rotation, followed by incubation in fresh 100% methanol overnight at room temperature with rotation. For tissue clearing, brains were transferred to 66% DCM/33% methanol solution and incubated for 3 h at room temperature with rotation, followed by two 15 min incubations in 100% DCM at room temperature with rotation. The brains were then treated with dibenzyl ether for 1 h, followed by incubation in fresh dibenzyl ether overnight at room temperature without shaking (day 15). The next day, cleared brains were equilibrated in ethyl cinnamate overnight for imaging. All rotations were performed at 20 rpm.

Cleared brains were imaged using a light-sheet microscope, UltraMicroscope Blaze (Miltenyi Biotec) equipped with a ×4 objective lens with a ×1 magnifier. The brain samples were mounted onto the stage in a cuvette filled with ethyl cinnamate. Images were acquired in light-speed mode with 2.5 μm step intervals. The light sheet was set to 5 μm thickness and 60% width with bidirectional illumination. Imaging conditions for each excitation were as follows: 785 nm, 90% laser transmission and 500 ms exposure; 640 nm, 90% laser transmission and 40 ms exposure; 561 nm, 90% laser transmission and 20 ms exposure; 488 nm, 5% laser transmission and 10 ms exposure. Excitation at 488 nm was used to detect nuclear signals stained with DiYO-1 for orientation and focus adjustment. Brain samples at PCD15.5, P0 and P3 that exceeded a single field of view were captured with tiled imaging.

All analyses of light-sheet microscopy images were performed using Imaris software (v.10.2.0, Oxford Instruments) unless otherwise noted. Acquired TIFF images were converted to IMS format for 3D rendering, and tiled images were stitched using Imaris Stitcher (v.10.2.0). Hemisphere and cortical regions were manually cropped using the Surface function. Image processing, including contrast adjustment, was applied manually across the entire dataset. All 3D-rendered images were visualized in the perspective view; note that scale bars are not provided, as scale information is not accurately preserved in this view.

Retrograde tracing in the short-tailed opossum

Opossum surgeries were carried out in sterile conditions following all United States Department of Agriculture requirements (USDA) for animals under their oversight, and all procedures were approved by the Yale Institutional Animal Care and Use Committee. Opossums were anaesthetized using institutional protocols. Body temperature, heart rate, oxygen saturation and respiration were closely monitored during the procedure. Anaesthetized opossums were placed onto a small animal stereotaxic instrument (Kopf, model 940), with a rat head-holder attachment (model 929-B rat gas anaesthesia head holder with model 955 ear bars). The skull was exposed, and craniotomies were made using a dental drill above the prospective dorsolateral frontal cortical area and somatosensory area. In the dorsolateral frontal area, a bevelled glass needle was inserted 1 mm below the pial surface of the brain. To label all cortical layers, three injections of retrograde AAV carrying pCAG-tdTomato (Addgene, 59462-AAVrg) were performed (50 nl each) at 1 mm, 0.6 mm and 0.3 mm below the surface of the brain. The needle was left at 3 mm for 5 min before retracting. To label the somatosensory area, three injections of retrograde AAV carrying pCAG-eGFP (37825-AAVrg) were performed (75 nl each) at 1 mm, 0.6 mm and 0.3 mm below the surface of the brain. The needle was left at 3 mm for 5 min before retracting. Then, 2 weeks later, the opossums were euthanized and subjected to transcardial perfusion with about 200 ml of 1× DPBS followed by 4% PFA in 1× DPBS. The brains were post-fixed in 4% PFA overnight at 4 °C and then switched to 30% sucrose 1× DPBS and left at 4 °C until equilibrated (brains sank). Brains were sectioned at 60 µm using the Leica Cryostat and staining and imaging was performed as described in the ‘Immunohistochemistry using mouse and opossum tissue’ section.

Birthdating of thalamic nuclei and cortical areas in the mouse brain

All data on the order of neurogenesis of mouse cortical and thalamic neurons was extrapolated from the collection of EdU birthdating experiments accessible online (https://neurobirth.org)⁵⁰. We downloaded the processed quantification of EdU-positive cells within our cortical and thalamic regions of interest. Specifically, the following cortical areas were selected: the basolateral amygdala (BLA) was chosen as representing the major nucleus of the amygdaloid complex; the hippocampal CA1 (CA1) for the hippocampus; the infralimbic area (ILA) for the mPFC; the primary motor cortex (MOp); primary somatosensory cortex (SSp); primary auditory cortex (AUDp); primary visual cortex (VISp); ventral auditory area (AUDv); temporal association area (TEa); and posterior parietal association area (PTLp). These areas were considered the closest correlates of the following human cortical areas analysed in this work: AMY, HIP, MFC, M1C, S1C, A1C, V1C, STC, ITC and IPC, respectively.

From the same dataset, the following thalamic nuclei projecting to the selected cortical areas were considered: ventral medial geniculate nucleus (MGv), dorsal lateral geniculate (LGd), lateral posterior (LP), ventral postero medial (VPM), ventral anterior lateral (VAL), lateral dorsal (LD), Reuniens (RE), anterior ventral (AV), mediodorsal, medial subdivision (MDm), centromedial (CM), paraventricular (PVT), paratenial (PT) and rhomboid (RH).

To determine the temporal order of genesis of neocortical areas, we compared the peak of neurogenesis of layer 6a, as it is one of the earliest generated in the mammalian brain^107,108 and data were available for all areas of our interest. To directly compare the time of neurogenesis of neocortical areas with the genesis of hippocampus, which is not organized into six layers, we consider the peak of neuronal production of the earliest and latest generated laminae of CA1, namely stratum orienses (CA1so) and stratum lacunosum molecolare (CA1slm). To birthdate the amygdala, we selected the peak of neurogenesis of ventral, posterior and anterior subdivisions of the BLA (BLAv, BLAa and BLAp, respectively).

Birthdating of thalamic nuclei in the rhesus macaque brain

To determine the relative birth order of thalamic nuclei, we performed a meta-analysis of previously published works. In cases in which we could not locate published data for a given area, we obtained images using archival brain sections from tritiated thymidine experiments and quantified tritiated thymidine-positive cells in the specified areas. The specimens that we used to generate new birth dating relationships are part of the collection previously generated by Pasko Rakic and previously used for various studies of neurogenesis in the rhesus macaque (Macaca mulatta)^{13,108,109,110,111,112}. A detailed description of the methods used for the original experiments can be found in the original works. In brief, in the studies cited, timed-pregnant macaques received an intravenous injection of 10 mCi per kg of thymidine-methyl-³H (also known as tritiated thymidine, ³H-TdR). The offspring were euthanized postnatally on the specified day, and autoradiography was used to reveal the ³H-TdR-positive signal on Nissl-stained (toluidine blue) brain sections. We integrated the results from previous works that used this material with our cell counting analysis on the archival samples to determine the peak time of neurogenesis in all thalamic regions of interest. Archival material is available and can be requested from the MacBrain-Resource (https://medicine.yale.edu/neuroscience/macbrain/). Metadata on the five samples that were further analysed in the current work for integration of the datasets are included in Supplementary Table 16.

The relative order of generation of the following areas was examined: MFC (BA 24c); DFC (BA 46); OFC (BA 11); A1C (BA 41, 42); M1C (BA 4); S1C (BA 1–3); STC (caudal superior temporal gyrus, STGc); IPC (BA 7a); ITC (BA 36r); V1C (BA17). VFC (BA 44, 45) was not included, as more rostral sections containing this area were not available for any of the selected cases. The timing and patterns of neurogenesis in MFC, DFC, OFC and V1C have been described in more detail previously¹¹³ (see figure 2 of ref. ¹¹³). Sections containing these regions were scanned on an Aperio ScanScope HR CS2scanner at ×20 magnification. The resulting images were manually analysed for quality of the tissue, staining and imaging. Annotation of the cortical areas of interest was done separately for each individual case based on the anatomical regions defined in the macaque brain atlas reported previously¹¹⁴. For each area of interest, we selected two serial coronal sections for quantification and generated a region of interest 800 µm wide spanning the layer 6 and white matter border to the layer 1 and 2 border. Cells positive for tritiated thymidine were identified by the presence of at least 3–5 silver grains and counted manually using FIJI. Both heavily and lightly labelled cells were included in the positive count. By PCD110, cortical neurogenesis is completed and most ³H-TdR-positive cells observed at this age are glial cells, which were excluded from the analysis. These were clearly distinguished from neurons by their smaller size, and they are often found physically adjacent to neuronal bodies.

Birthdating of thalamic nuclei in macaque was determined by performing a meta-analysis of previous studies. The work in ref. ⁵¹ using similar archival material described above detailed the birth order of the following nuclei used in our timeline: anterior thalamic nuclear group (Ant); infralaminar nuclei (IL), which includes the central medial (CM), central lateral (CL), and parafascicular (Pf) nuclei; mediodorsal nucleus (MD); midline thalamic nuclei (Mid), including reuniens (Re), rhomboid (Rh), paratenial (Pt) and paraventricular (PV) nuclei; ventroanterior nucleus (VA); ventrolateral nucleus (VL); ventroposterior nucleus (VP). These data were integrated with previous birthdating studies of primate thalamic nuclei that included pulvinar (PUL) subdivisions including lateral, medial and inferior pulvinar (Pl, Pm, Pi, respectively) and medial geniculate nucleus (MGN)¹¹⁵. Finally, neurogenesis of macaque lateral geniculate nucleus was based on previously reported findings¹¹⁶.

Not all cortical or thalamic areas have exact homologous counterparts between rodents and primates. For example, the DFC/dlPFC was sampled only for the macaque as no equivalent exists in the mouse. We detailed the correspondence between mouse and macaque thalamic nuclei and cortical areas in Supplementary Table 16.

Connectivity between thalamic nuclei and cortical targets

To generate a graphical representation of the anatomical connectivity strength between various thalamic nuclei and cortical areas (Extended Data Fig. 12), we extracted connectivity measures between areas from previously published anterograde viral tracing experiments. The high-resolution mouse anatomical connectome was derived from a voxel-scale model of Allen Mouse Brain Connectivity Atlas (http://connectivity.brain-map.org/)^117,118. The connectome data were derived from imaging enhanced green fluorescent protein (eGFP)-labelled axonal projections from 428 viral microinjection experiments in WT C57BL/6J mice. We extracted the ipsilateral directed normalized connectivity from thalamic nuclei to regions in the mouse cortex (Supplementary Table 16).

In Extended Data Fig. 12, the thalamocortical connections are represented by lines, and the thickness and opacity is derived from connectivity strength. The line thickness was derived by converting the thresholded weighted connection values into thicknesses ranging from 0.25 pt to 1 pt, with the more heavily weighted values being represented by lines of higher thickness. Moreover, the line opacity was calculated by assigning opacity values based on the thickness of the lines, with the thickest lines having the highest opacity. The criteria used for opacity assignment were as follows: lines of thickness between 0.25 pt and 0.5 pt were assigned opacity of 60%; thicknesses between 0.5 pt and 0.75 pt were assigned opacity of 80%; thicknesses between 0.75 pt and 1 pt were assigned opacity of 100%. Grey lines indicating connections between FO nuclei and association areas as well as connections between HO nuclei and sensorimotor areas were assigned line opacity of 50%. Connections derived by adjusting mouse connectivity data to reflect accurate macaque connectivity were assigned a line thickness of 0.5 pt and an opacity of 70%. We removed connections from pulvinar to V1C, S1C and A1C as the corresponding region in mouse is highly divergent compared with macaque. We also removed MGN to ITC, IPC and V1C connections, as there are extensive primate data charting MGN connectivity, and the coarse nature of tracing experiments in mice probably contributed to false positives. The conclusions are all based on birth order of thalamic nuclei, while the connectivity between brain regions provides a visual aid and does not alter our conclusions.

Analysis of human cerebral organoid data

Human cerebral cortex organoid (hCO) data were collected from the Broad Institute’s Single Cell Portal (https://singlecell.broadinstitute.org/single_cell/study/SCP1756)⁵³. The excitatory neurons without GAD1 expression were extracted from each age. Module gene lists of neocortex (Ncx), HIP, striatum (Str) and thalamus (Thal) were defined previously²⁵. The AMY module gene list (containing TFAP2D, TBL1X and NR2F2) was identified on the basis of in-house comparison of the human exon microarray data²⁶ and RNA-seq data²⁸ used in this study using in situ hybridization validation from the Allen Developing Mouse Brain Atlas¹¹⁹. The RA synthesis (RA_syn) module included ALDH1A1, ALDH1A2 and ALDH1A3. The module score was calculated using AddModuleScore function from Seurat (v.5)¹²⁰. A given regional (Ncx, HIP or AMY) cell group was defined with positive module score only of the given region but not the other two regions. To compare the pseudo-bulk gene expression across three brain regions, each regional cell group was randomly split into three batches, which were then aggregated as the pseudo-bulk replicates for that region using AggregateExpression function from Seurat v.5. Each module expression value was calculated using the median of all genes from the corresponding module for each pseudo-bulk replicate. Regional expression difference of selected genes and modules was tested using t_test function from R with the FDR threshold set as 0.05.

Mice used in this study

All studies using mice (Mus musculus) and macaques (M. mulatta) were performed in accordance with protocols approved by Yale University’s Institutional Animal Care and Use Committee and National Institutes of Health (NIH) guidelines. The mice were housed and kept under consistent environmental conditions at 25 °C, 56% relative humidity under a 12 h–12 h light–dark cycle. Food and water were consumed ad libitum. Experimental cohorts comprised of both sexes. The following mouse lines were used: Celsr3-flox⁷¹, C57BL/6, P7; Dlx5/6-cre¹²¹, FVB/N, P7; Emx1-cre¹²², 129S2/SvPas, P7 and P37; Fezf2-flox¹²³, C57BL/6J, P7 and P37; Gbx2-flox¹²⁴, 129S6/SvEvTac, P7 and P37; Neurod6-cre¹²⁵, C57BL/6J, P2, P7 and P37; Pax6-flox¹²⁶, C57BL/6, P7 and P37; Plxnc1-flox, C57BL/6J, P2, P37; Olig3-cre¹²⁷, C57BL/6J, P7 and P37; Rarb-null⁴¹, B6SJLF1/J, P0; Rxrg-null⁴¹, B6SJLF1/J, P0; Satb2-flox¹²⁸, C57BL/6 P7 and P37; Sema7a-flox, C57BL/6JGpt, P37; Sema7a-null³³, 29S6/SvEvTac, P2; Zbtb18-flox⁶⁶, C57BL/6, P7 and P37.

We did not formally calculate sample sizes; we estimated the number of animals needed on the basis of established practices in the field and previous studies published utilizing these experimental approaches (see the Reporting Summary for additional details). For all experiments, the number of animals or biological replicates (n) is indicated either in the figure legend or in the associated Supplementary Table. Groups of animals included specific genotypes and therefore randomization was not applicable here. We did not separate animals on the basis of sex.

Experimental designs required comparing specific genotypes that must be identified before surgery, tissue collection and experimentation. Before surgery or tissue collection, the animals were given identification numbers that did not contain genotype information; however, visual differences between mutant and control mice precluded true blinding. Standardized automated cell counting algorithms, image analysis and imaging pipelines were used to in all cases to quantify experimental data to avoid experimenter bias, and regions of interest were identified using standardized anatomical landmarks.

Generation of mouse Plxnc1-flox and Sema7a-flox alleles

Mice with a conditional Plxnc1 floxed allele were generated using CRISPR–Cas9-mediated gene editing according to previously described methods^129,130. A floxed allele of exon 1 was created by using the CRISPick tool (https://portals.broadinstitute.org/gppx/crispick/public) to screen potential Cas9 target guide (protospacer) sequences approximately 1 kb upstream of the Plexin C1 5′ UTR and in intron 1 on the reverse strand of chromosome 10. sgRNAs incorporating these protospacers were transcribed in vitro using the MEGAShortscript kit (Invitrogen), purified using the MEGAclear kit (Invitrogen) and eluted using microinjection buffer. The 156–157 bp repair template oligonucleotides (ssODN) containing loxP sites was synthesized by IDT Technologies. The floxed allele was created in two steps: targeting the 5′ loxP site, followed by a generation of breeding and subsequently targeting the 3′ loxP site. sgRNA–Cas9 RNP and the corresponding template oligo were electroporated into C57BL/6J (JAX) zygotes¹³⁰, after which the embryos were transferred to the oviducts of pseudopregnant CD-1 foster female mice according to standard methods¹³¹. Founder animals were identified by PCR and sequencing of the loxP target sites. These animals were then mated to C57BL/6J mice to confirm correct targeting and germline transmission of the cKO allele.

Upstream (5′) guide protospacer + PAM sequence: ATTGAGAACTTGCTCCGTTCTGG. Intron 1 (3′) guide protospacer + PAM sequence: CAGCCTGAACCAGCGCGAGGGGG. 5′ loxP template oligo: AGTTCTGCTCACTATCTAACCCTAGTGACAAAAGCTAACATTTATTGAGAACTTGCTCCGATAACTTCGTATAATGTATGCTATACGAAGTTATGAATTCTGGGCATAGTTCTAAATGCACCATGTAGATTTTGATTCACTTGATGCTTGGGACAAA. 3′ loxP template oligo: TGGTAAATAAATGCTATGGTGCAAAAGCCGCTCGCAGACCGCCCAGCCTGAACCAGCGCGATAACTTCGTATAGCATACATTATACGAAGTTATAAAGGGGGCGGGGCTGTGGCTAGCCAAGGGTGGCTGTGGTCTCAACCTGGTGGAGAGAACCC.

Plxnc1 mice were genotyped with the following primers: CAGGTGGCACTGGACTAGC and TCTCCCTTGGCAACGGAGTC, resulting in a 335 bp product for WT and 373 bp product for the Plxnc1-flox allele.

Mice with a conditional Sema7a floxed allele were generated using CRISPR–Cas9-mediated gene editing by GemPharmatech. In brief, CRISPR–Cas9 and guide RNAs were microinjected into fertilized eggs of C57BL/6JGpt mice. loxP sites were inserted between exons 1 and 2 (gRNA1: GCTTTGAAGCTCCCCGGTTA) and exons 8 and 9 (gRNA2: TCAGCCTGTGACTGTCGGGG). Fertilized eggs were transplanted to obtain positive F₀ mice, which were confirmed by PCR and sequencing. A stable F₁ generation was obtained by mating positive F₀ mice to C57BL/6JGpt mice.

Sema7a mice were genotyped with the following primers: AGCTGCTCAAACACTGCCTGATAA and CCTAAAACCACAACCTTGGGTTCTG, resulting in a 313 bp WT product and 416 bp product for the Sema7a-flox allele.

Immunohistochemistry using mouse and opossum tissue

For mouse brain staining, we performed transcardial perfusion with 10 ml 1× DPBS, followed by 10 ml 1× DPBS, 4% PFA and post-fixed in 4% PFA, 1× DPBS at 4 °C overnight. Post-fixed brains were then transferred to 30% sucrose 1× DPBS at 4 °C until the brains sank in the solution, indicating that they were equilibrated and ready for the next steps. The equilibrated brains were immersed and frozen in optimal cutting temperature (OCT) compound (Thermo Fisher Scientific, 23-730-571) and sectioned at 60 µm using the Leica Cryostat (CM3050S). The sections were washed three times with 1× DPBS at room temperature for 5 min each to remove the OCT and were permeabilized by incubating in 1× DPBS 0.6% Triton X-100 for 1 h. Antigen blocking was performed by incubating the sections with blocking buffer (1× DPBS, 5% normal donkey serum, 0.3% Triton X-100) at room temperature for 1 h. Primary antibodies were incubated with the sections at 4 °C overnight, followed by three 5 min washes with DPBS, 0.3% Triton X-100. Individual secondary antibodies (Jackson laboratories) all at a dilution of 1:1,000 with a 1:10,000 dilution of 4′,6-diamidino-2-phenylindole (DAPI) to label cell nuclei were incubated with sections for 2 h at room temperature, followed by three 10 min washes with DPBS, 0.3% Triton X-100. The sections were mounted onto Superfrost Plus Microscope Slides (Fisherbrand, 22-037-246) and sealed with Fluoromount-G mounting medium (Invitrogen, 00-4958-02). Antibody dilutions used for immunostaining were as follows: 5-HT (1:500, Immunostar, 20080), ALDH1A3 (1:250, Proteintech, 25167-1-AP), BCL11A (1:1,000, Abcam, ab19487), BHLHE22 (1:500, Sigma-Aldrich, HPA064872), GFP (1:2,000, Aves Labs, NC9510598), MEIS2 (1:1,000, Santa Cruz Biotechnology, sc-81986), PLXNC1 (1:250, R&D Systems, AF5375), NRP2 (1:250, R&D Systems, AF567), NTNG1 (1:100, R&D Systems, AF1166), SEMA7A (1:250, R&D Systems, AF1835), SLC17A6 (1:500, Synaptic Systems, 135404), TD-TOMATO (1:500, Scigen, AB8181) and TLE4 (1:100, Santa Cruz Biotechnology, sc-365406).

SEMA7A and PLXNC1 antibodies were raised in sheep and goat, respectively, which precludes them from individually being recognized by secondary antibodies due to the similarity of goat and sheep IgGs. Thus, to co-stain for SEMA7A and PLXNC1, we conjugated these antibodies to Alexa fluorophores. Using antibody conjugation kits, SEMA7A was conjugated to Alexa 647 (Invitrogen A2186) and PLXNC1 was conjugated to Alexa 568 (Invitrogen A2184). The resulting primary conjugated antibodies were incubated with no changes to the protocol described above, but the secondary antibody step was omitted.

For some experiments, the 5-HT signal was visualized using 3,3′-diaminobenzidine (DAB) staining. Primary and secondary antibody incubations were performed with no changes. A secondary horseradish peroxidase (HRP)-conjugated donkey anti-rabbit IgG (1:2,000, Jackson Immuno, 711-035-152) antibody was used. For colour development, we used the ImmPACT DAB kit (Vector Laboratories, SK-4105). One drop of ImmPACT DAB reagent was added to 1 ml ImmPACT DAB diluent and mixed by inversion. The sections were then incubated in the mixture on a shaker. Once sufficient colour development was observed by eye, the reaction was stopped by washing with 1× PBS. Control and experimental groups were incubated at the same time (that is, one batch) and stopped at identical times.

Unless otherwise stated in the following sections, all microscopy images were obtained using a ×10 objective on the Olympus VS-200 Slide Scanner.

Quantification of immunofluorescence data

P7 brain section images obtained using a ×10 objective on the Olympus VS-200 Slide Scanner were exported from QuPath, with a downsample factor of 8 and saved as TIF files. The resulting images were opened in FIJI and the segmented line tool with a width of 35 px was used to select the quantified region indicated. The straighten tool was then used to generate a rectangular image of the resulting selected area which was subsequently split into 200 equally sized bins and the fluorescence intensity of a given channel was measured in each bin. The values were normalized to the average fluorescence value in the control.

For retrograde traced brains, GFP⁺ cells were quantified using FIJI. In brief, the cortex was highlighted using the segmented line tool followed by the straighten function. The resulting rectangular image was split into 15 equal sized bins, and GFP⁺ cells were counted in each bin. For dual tracer injection experiments, bleed through was observed at the injection sites. We therefore manually segmented specific cortical and subcortical areas of interest in QuPath using defined areal landmarks using the DAPI channel as guidance. Cells were quantified using positive cell detection in QuPath. For dual injections, statistical analyses were performed using linear mixed-effects models implemented in the R package lme4 using the function lmer, with group (control versus dKO) and region as fixed effects, and animal as a random effect to account for repeated measurements across sections. Post hoc comparisons were conducted using estimated marginal means (emmeans) from R package emmeans.

All other statistical analyses related to immunofluorescence data were performed using Prism 9.

Preparation of flat mouse cortices

To generate sections from flattened cortices, we performed transcardial perfusion with 10 ml 1× DPBS to remove blood from the tissue. The brains were quickly dropped into ice cold 1× DPBS and cut in half in the sagittal plane. The subcortical structures were carefully removed, resulting in the cortex and olfactory bulb. The cortices were then placed between glass microscope slides with 1 mm spacers. These were then submerged in ice cold 4% PFA, 1× DPBS at 4 °C overnight. The next day, the microscope slides were carefully removed and the resulting flattened cortices were then transferred to 30% sucrose 1× DPBS at 4 °C until they sank in the solution, indicating that they were equilibrated. The flattened cortices were then section at 60 µm using the Leica Cryostat (CM3050S).

Bulk RNA-seq data analysis

Fastq files from bulk RNA-seq experiments using dissected neocortices from Satb2^flox/floxEmx1-cre (Satb2 cKO) and control mice at P0 were downloaded from the GEO (GSE68912)⁶⁴. Fastq files of bulk RNA-seq data of Fezf2 homozygous whole-body KO and WT mice at P0 were downloaded from the GEO (GSE160202)⁶⁵. The RNA-seq data of the neocortical tissue of the Zbtb18^flox/floxNeurod6-cre (Zbtb18 cKO) and WT mice at P0 were obtained from a previous study⁶⁶. FastQC¹³² was performed before and after trimming adapters using Trimmomatic¹³³. The alignment was performed using STAR¹³⁴ against the mouse primary genome assembly GRCm39¹³⁵. Reads with low quality or those that were improperly mapped were filtered out using SAMtools¹³⁶. Duplicated reads were removed using Picard¹³⁷. HTSeq¹³⁸ was then applied to annotate features using GENCODE M33 mouse gene annotation¹³⁵ and we then calculated the raw read counts mapped to each gene across samples. The differential expression analysis was then conducted using DESeq2 (v.1.44.0)¹³⁹. Differential expression of transcripts was detected using a false discovery rate (FDR) < 0.05 and |log₂[FC]| ≥ 0.6.

The count matrix and sample metadata of Rarb/Rxrg-dKO bulk RNA-seq data were obtained from a previous study⁴¹ (GEO: GSE142851). Differential expression analysis was performed using the DESeq2 package (v.1.44.0) with the default parameters in R. Two-sided Wald tests were applied to identify DEGs between Rarb/Rxrg-dKO and WT mice for each region of mPFC, OFC and MOp. P values were adjusted using the Benjamini–Hochberg procedure. DEGs were defined as those meeting a threshold of adjusted P < 0.05 and |log₂[FC]| ≥ 0.6.

Analysis of Visium data

The Visium spatial transcriptomics data from P7 control versus Celsr3-cKO mice generated previously⁷⁶ can be found in the Center for Brain Science (CBS) repository system (https://doi.org/10.60178/cbs.20230816-001).

Differential expression analysis was performed using a two-level framework to capture both sample-level and spot-level effects. First, spatial transcriptomic counts were aggregated at the sample ID level using Seurat’s AggregateExpression function, grouping by sample ID and genotype to generate pseudobulk profiles. Differential expression analysis was then conducted using DESeq2 and genotype as the primary factor. Genes with FDR < 0.05 were considered to be significant. Second, to account for within-sample spatial variability, spot-level expression was modelled using limma-voom (R package limma v.3.58.1) with dream (R package variancePartition v.1.32.5), incorporating genotype as a fixed effect and sample ID as a random effect to account for repeated measures across spots. For both approaches, log₂-transformed fold changes and adjusted P values were extracted and directly compared for selected genes of interest.

In utero electroporation

Plasmid mixtures were prepared containing 1 µg μl⁻¹ pCAG-GFP (Addgene, 11150), and 0.1% Fast Green (Sigma-Aldrich) was added to visualize the plasmid mixture during injections. In utero electroporation was performed on timed-pregnant PCD14.5 dams. Timed-pregnant dams were anaesthetized using institutional protocols. The uterine horns were exposed. Using a bevelled pulled glass needle, about 0.5 µl of the plasmid mixture was injected into the lateral ventricle of each embryo. After visually confirming that the lateral ventricles were sufficiently filled with the plasmid mixture, a handheld platinum coated electrode (Tweezertrode, 5 mm, 45-0489, Harvard Apparatus) attached to an electroporator (BTC, ECM830) was used to deliver five 35 V electrical pulses at 50 ms each with a 950 ms interval between pulses to each embryo. The positive end of the electrode was positioned to direct the DNA towards the frontal-medial region of the ventricle to label mPFC and the centre of the right or left hemisphere of the dorsal pallium to target SSp. The pups were then delivered without intervention at the expected time.

Retrograde tracing in mice

For P30 injections, animals were anaesthetized using institutional protocols. Mice were held in place using a Kopf Stereotaxic instrument (Model 940). The skull was exposed and the bregma point was located. The tip of a bevelled glass needle attached to a microinjection unit attached to the stereotaxic instrument was positioned directly atop bregma. This position was considered the zero-point. The mPFC was targeted by performing a craniotomy above the planned injection site. The injections were performed at the following position relative to bregma: 2.1 mm anterior, 0.3 mm lateral and 2.4 mm ventral. Due to the heterogeneity of mutant animals, the mPFC was targeted based on anatomical landmarks, and only matching injections determined after processing the brains were used for analysis. After the needle was placed in the brain, 50 nl virus was injected into the mPFC, and the needle was retracted after 5 min. The SSp was targeted by positioning the needle 1.46 mm posterior and 3 mm lateral to bregma. To obtain the maximum precision in the depth of the injection for SSp, a craniotomy was performed to expose the cortex, and the bevelled glass needle was moved into the relative posterior and lateral position and placed on the cortical surface, this represented the zero point in the dorsal-ventral axis. The needle was then moved to −0.9 mm relative to the cortical surface and 20 nl of virus was injected, the needle was then positioned at −0.6 and −0.3 mm for two additional 20 nl injections to cover the entire depth of the cortex. At the final injection point, the needle was left for 5 min before retracting. Targeted regions were injected with 50 nl retrograde AAV carrying pCAG-eGFP or retrograde AAV carrying pCAG-tdTomato (Addgene, 59462-AAVrg). For all experiments, incubation was for 1 week.

After incubation, the brains were collected for histology and stained for the indicated antigens. Brain collection and staining was performed as described in the ‘Immunohistochemistry using mouse tissue’ section. Three representative sections at each anterior–posterior position were imaged using the extended focus imaging (EFI) software associated with the Olympus VS200 slide scanner. EFI results in images spanning entire sections with all cells in focus, essentially doing the job of a confocal microscope in a sensible timeframe, facilitating cell counting. All retrograde traced sections for all experiments included in the P30 injections were imaged using the identical parameters, EFI with the following exposure times, DAPI 26 ms, GFP 56.5 ms, TdT 29.0 ms.

For P3 injections, the mice were anaesthetized on ice for several minutes. During surgery, the pups were maintained under hypothermia with ice and placed on a stereotaxic apparatus. A purified AAV solution (AAV2-retro-CAG-tdTomato; 50 nl per site; injection rate: 50 nl min⁻¹) was injected into the right mPFC. The injection site was located 1.0 mm anterior and 0.3 mm lateral to bregma, and 0.5 mm ventral from the surface. Experiments were carried out with KDS Legato 130 (KD Scientific) and stereotaxic frame (Muromachi Kikai). The brains were collected at P7. The fluorescence images were acquired using fluorescence microscope (BZ-X810, KEYENCE) with ×4 and ×10 objectives.

Fixed-tissue DiI tracing

Brains were quickly dissected at P7 and dropped into ice-cold 4% PFA, 1× DPBS and left to fix overnight at 4 °C. The next day, the brains were cut in half coronally, exposing the thalamus. DiI crystals were then carefully placed just below the surface, across the exposed thalamic region. The brains were then embedded in 4% low-melt agarose (invitrogen) and incubated at 37 °C for 1 week. The brains were then sectioned at a thickness of 60 µm using a Leica Vibratome and placed in a solution of DAPI in 1× PBS (1:10,000) for 10 min at room temperature before switching back to 1× PBS. Sections were placed on glass slides and sealed with Fluoromount-G, and all slides were imaged the same day that they were sealed.

Human telencephalic organoid culture

Human induced pluripotent stem cell lines (line HSB311, 36, derived from skin fibroblasts of a female donor, PCW8) were authenticated by morphology or genotyping and confirmed to be free of mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza). For maintenance of pluripotency, cells were dissociated into single cells using Accutase (Thermo Fisher Scientific, 00-4555-56) and plated at a density of 1 × 10⁵ cells per cm² on Matrigel-coated (BD) six-well plates (Falcon) in mTeSR1 (StemCell Technologies, 85850) supplemented with 5 μM Y27632 and a ROCK inhibitor (Sigma-Aldrich, SCM075). ROCK inhibitor was removed after 24 h, and cells were cultured for an additional 4 days before passaging.

Telencephalic organoids were generated using a directed differentiation protocol as previously described⁴¹. Cells were dissociated into single cells with Accutase (Thermo Fisher Scientific, 00-4555-56) and resuspended in neural induction medium containing 100 nM LDN193189 (StemCell Technologies, 72147), 10 μM SB431542 (Selleck Chemicals, S1067) and 2 μM XAV939 (Sigma-Aldrich, X3004-5MG) to achieve dual SMAD and WNT inhibition. Cells were plated at 10,000 cells per well in a 96-well V-bottom ultra-low-attachment plate (Sumitomo Bakelite). To promote cell survival and aggregation, 10 μM Y-27632 (Sigma-Aldrich, SCM075) was added for the first 24 h. After 10 days in stationary culture, the organoids were transferred to six-well ultra-low-attachment plates (Millipore Sigma) and maintained on an orbital shaker at 90 rpm to improve nutrient and gas exchange. Starting on day 18, the organoids were cultured with maturation medium supplemented with 1× CD lipid concentrate (Thermo Fisher Scientific, 11905031), 5 μg ml⁻¹ heparin (StemCell Technologies, 07980), 20 ng ml⁻¹ BDNF (Abcam, 9794), 20 ng ml⁻¹ GDNF (R&D Systems, 212-GD), 200 μM cAMP (Sigma-Aldrich, 20–198) and 200 μM ascorbic acid (Sigma-Aldrich, A92902) to support neuronal maturation. On day 148, all-trans RA (Sigma-Aldrich, R2625) or the pan-RA receptor antagonist (RA inhibitor) AGN193109 (Tocris, 5758) was added to the medium for 48 h before sample collection.

For the preparation of organoids for staining, telencephalic organoids were fixed in 4% PFA at 4 °C and cryoprotected in 20% sucrose, and embedded in OCT compound (Thermo Fisher Scientific, 23-730-572). Embedded samples were sectioned at 10 µm using the Cryostat (Leica, CM3050S). The organoid sections were washed three times in PBS for 5 min, then incubated in a blocking solution containing 0.5% (v/v) Triton X-100 and 10% (v/v) donkey serum (Jackson ImmunoResearch Laboratories, 017-000-121) in PBS for 2 h. Primary antibodies were added in 10% (v/v) donkey serum and applied at 4 °C overnight. The sections were washed three times in PBS for 5 min and incubated with fluorescent secondary antibodies for 2 h at room temperature in 10% (v/v) donkey serum. After a final PBS wash (3 times for 5 min), the sections were coverslipped with Vectashield mounting medium (Vector Laboratories, H-1000). The primary antibodies used targeted hMEIS2 (1:200, Abcam, ab244267), hPLXNC1 (1:250, MAB3887, R&D Systems) and hSEMA7A (1:250, AF2068, R&D Systems). Images were acquired using the LSM800 confocal microscope (Zeiss) and were processed using Zeiss ZEN and ImageJ software. z-stack images were analysed using Volocity (v.6.3.1) and Spotfire (v.11.2.0).

In situ hybridization in brain sections and whole-mount preparations

Whole-mount and section in situ hybridization using antisense digoxigenin-labelled RNA probes was performed as described previously⁴¹, except that 5% dextran sulfate was added to the hybridization buffer for whole-mount in situ hybridization. Brains from the embryonic mouse, Rhesus macaque (M. mulatta) and human were fixed in 4% PFA at 4 °C and sectioned at 20 μm and slides were stored in −80 °C until use.

The following commercially available probes (Horizon Discovery) were used: mouse Cyp26b1 (MMM1013-202798233), Plxnd1 (MMM1013-202765536), Sema3e (MMM1013-202769580) and Sema7a (MMM11013-202769010). The Plxnc1 probe was made from the following primers, forward (fwd): CAGCCAATCAAACCTTGAGCAC, reverse (rev): GTTGTTGAATAGAGGCCCAGTGAC. The template for mouse Ntf3 was a gift from J. L. R. Rubenstein. The templates for mouse Sema5b and Cdh8 probes were gifts from E. A. Grove.

To detect Rhesus macaque PLXNC1 and SEMA7A, DNA fragments containing a single exon of each gene (1,057 bp, 1,043 bp, 718 bp, respectively) were PCR-amplified from the genomic DNA. DNA fragments were subcloned into the pCRII vector using the TA-cloning kit (Invitrogen, K202020). All plasmids containing cDNAs and genomic DNA fragments were used as templates to generate probes after linearization by restriction enzymes. Primer sets used for PCR amplification were as follows: PLXNC1 fwd, ATGGAGGTCTCCCGGAGGA; rev, GGCTCTCGGCCGTCTTGAAG. SEMA7A fwd, ACAAGGCCCCACTGCAGAA; rev, TTCCCAGCCCCTCCCTTTC. The digoxigenin-labelled probes were synthesized from the linearized templates either by SP6 (NEB, M0207S), T7 (NEB, M0251S) and T3 (NEB, M0378S) RNA polymerases, respectively, and RNA labelling mix (Roche, 11277073910) according to the manufacturer’s instructions. Images of brain sections were taken using the VS200 microscope (Olympus Microscopy).

Whole-mount in situ hybridization preparations were imaged and colour balance was manually adjusted to normalize the background hue across images without altering signal intensity.

LacZ signal development in whole-mount preparations

Brains were dissected from PCD13.5 and PCD16.5 RARE-lacZ mouse embryos and drop-fixed in 4% PFA for 90 min at 4 °C, followed by rinsing with 1× PBS. The enzyme activity of β-galactosidase was visualized using Rad-Gal (Sigma-Aldrich, RES1364C) as a substrate for chromogenic reaction.

RNAscope analysis on chicken, opossum and mouse brain tissue

Brains from the chicken, opossum (Monodelphis domestica) and mouse were fixed by quickly dissecting and immersing in ice cold 4% PFA, 1× DPBS overnight, followed by equilibration in 30% sucrose, 1× DPBS. The brains were then sectioned at 20 µm using the Leica Cryostat and placed onto microscope slides and allowed to dry overnight at room temperature to enhance tissue adhesion. We used the following probes: chicken Sema7a (1276971, ACD) and Plxnc1 (1276981, ACD); opossum Sema7a (1570351, ACD) and Plxnc1 (1570361, ACD); mouse Sema7a (437261, ACD), Plxnc1 (495481, ACD), Zbtb20 (837641, ACD) and Tfap2d (551631, ACD). RNAscope was performed according to the RNAscope Multiplex FL v2 protocol and kit (323270) from ACD (Advanced Cell Diagnostics Bio). In brief, the slides were incubated at 60 °C for 30 min, then further fixed in 4% PFA, 1× DPBS on ice for 30 min and then dehydrated with graded ethanol immersions (50%, 70%, 100%). To inactivate endogenous peroxidases, the slides were incubated with hydrogen peroxide for 10 min and washed with distilled water. Protease treatment was performed with protease III for 10 min. Target retrieval was performed by immersing the slides in boiling RNAscope target retrieval buffer for 5 min. The samples were then washed with RNAscope wash buffer and incubated with probes for 2 h at 40 °C. After probe incubation, the samples were incubated sequentially with RNAscope Multiplex FL v1 Amp1, Amp2 and Amp3 for 30 min each. The slides were then incubated with HRP C1, and then subjected to tyramide signal amplification. This was repeated for each additional channel. All slides were imaged using a ×10 objective on the VS200 Slide Scanner (Olympus).

Statistics and reproducibility

All quantified experimental manipulations include the exact number of biological replicates, statistical methods and exact P values in the figure legends or accompanying Supplementary Tables. Some data shown in the figures are representative rather than quantified, including RNA-expression patterns, or derive from scarce tissue sources, particularly fetal human and macaque tissue. For representative micrographs of mouse, opossum and chicken tissue, experiments were performed two or more times. For fetal macaque and human micrographs, each experiment was performed using one biological sample. The scarcity of fetal human and macaque tissue precludes robust replication across samples and is an inherent limitation of studies using these tissues. Human fetal DWI analyses also used one brain per age, limiting quantitative analyses.

Organotypic slice co-culture

GFP⁺ regions were labelled by in utero electroporation as described in the ‘In utero electroporation’ section. The brains of P2 control (non-electroporated) and GFP⁺ mice were extracted and immediately transferred to ice-cold (<4 °C) carbogen-saturated (95% O₂:5% CO₂) artificial cerebrospinal fluid (aCSF; 92 mM NMDG, 20 mM HEPES, 5.5 mM glucose, 30 mM NaHCO₃, 5 mM sodium l-ascorbate, 2.5 mM KCl, 1.25 mM NaH₂PO₄, 2 mM thiourea, 3 mM sodium pyruvate, 5.5 mM urea, 10 mM MgSO₄, 0.5 mM CaCl₂·2H₂O). Coronal 200-µm-thick slices were obtained using a vibratome (Leica, VT1200S) in ice-cold aCSF continuously aerated by carbogen (95% O₂:5% CO₂). ROIs were dissected from coronal brain slices using scissors with tiny blades (FST, 15000-03). Dissected brain areas were transferred to 50 μg ml⁻¹ rat-tail-collagen-I-coated (Corning, 354236) culture inserts (Falcon, 353090) in a six-well plate containing 1.5 ml of slice culture medium (Neurobasal medium, Gibco, 21103049), supplemented with 1× B-27 supplement (Gibco, 17504044), 1% (v/v) GlutaMAX (Gibco, 35050061), 5 µg ml⁻¹ human insulin (Sigma-Aldrich, I9278), 20% (v/v) horse serum (Gibco, 26050088) and 1% (v/v) penicillin–streptomycin (Gibco, 15070063). The orientation of two dissected slices were arranged such that the ventral sides were in contact, while the dorsal sides (upper-layers) were at opposing ends. Organotypic brain slices were cultured at 37 °C for 2 days.

For immunofluorescence staining, co-cultured organotypic slices were fixed at DIV2 using 4% PFA for 20 min at room temperature. The slices were washed in PBS (three times for 10 min), and incubated in blocking solution containing 5% (v/v) normal donkey serum (Jackson ImmunoResearch Laboratories, 017-000-021) and 0.3% (v/v) Triton X-100 in PBS at 4 °C overnight. The slices were incubated with primary antibodies diluted in blocking solution overnight at 4 °C and washed with PBST (0.1% (v/v) Tween-20 in PBS; three times for 10 min). The slices were incubated overnight in the appropriate fluorescent secondary antibodies (all raised in donkey, Jackson ImmunoResearch Laboratories) diluted at 1:1,000 in blocking solution at 4 °C, and washed with PBST (three times for 10 min). The slices were then counterstained with DAPI for 10 min at room temperature, and finally washed with PBST (three times for 10 min). 2-by-2 tile-scan images of eight serial optical sections at 5.8-μm intervals over a total depth of 40.8 μm were acquired on the LSM880 confocal microscope (Zeiss). Orthogonal projection images were used for the quantification of axonal innervation. The mean fluorescence intensity of the electroporated and non-electroporated regions was measured in Fiji.

Manuscript preparation

All text was written by humans. The manuscript was written by J.T. and N. Sestan with edits and suggestions from all authors. In editing the manuscript text, ChatGPT (https://chatgpt.com) and Grammarly (https://www.grammarly.com) were used for proofreading and style suggestions.

Adobe Illustrator 2025 and 2026, and Adobe Photoshop 2025 were used to assemble all figures. Adobe Illustrator 2025 and 2026 was used to draw cartoons in Figs. 1h, 2i,k and 4b and Extended Data Figs. 1b,f and 14b. The mid-fetal human brain cartoons in Fig. 1a and Extended Data Figs. 1a, 8a and 9d are modified versions of a cartoon originally published by our group previously⁴¹.

Reporting summary

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