Spatiotemporal development of sparse excitatory neuronal types within the deep mouse cortex Authors: Shalini Iyer, Mark S. Cembrowski Published: 2026 ABSTRACT SummaryRevealing the spatiotemporal organization of cell-type development is critical for understanding cortical patterning and function. To map the temporal evolution and spatial patterning of cell types in the developing mouse brain, here we employ single-cell spatial transcriptomics across four developmental time points and both sexes. We generate a dataset of >1.5 million brain cells across embryonic and postnatal time points and focus on deep cortical neurons of the subplate and claustrum, which have typically been thought to have similar developmental patterns due to shared molecular properties in adulthood. We find significant differences between the developing subplate and claustrum, including distinct origins, transcriptomic evolution, and spatiotemporal maturation, illustrating that these neurons likely play critical yet distinct roles in development. Our single-cell spatial transcriptomics dataset, encompassing a variety of cell types across time points and sexes, will facilitate future study of cell types in brain development. Introduction The developing brain comprises a diverse range of cell types that emerge in a tightly regulated spatiotemporal manner.1,2 The developmental processes underlying these cell types are coordinated by intricate transcriptional programs and signaling cues that guide cell fate and maturation.3 Understanding how distinct cell types arise during development is essential for delineating the mechanisms that underlie healthy brain development and for identifying signaling pathways that may be disrupted in neurodevelopmental disorders. Despite this importance, early embryonic developmental windows—when cell types typically emerge and begin migrating to their final destinations—remain underexplored. Such single-cell and spatial resolution is especially important for assessing sparse populations that have unique spatial localization and play critical roles in neurodevelopment, both of which are hallmarks of specific types of neurons within the deep cortex. One of the most prominent examples of this is the subplate (SP), a transient and heterogeneous neuronal population that plays an essential role in the migration of neuronal populations into the cortical plate.4 SP neurons are among the earliest born during neurogenesis, emerging around embryonic days 10.5–13.5 (E10.5–E13.5) and contributing to the formation of the cortical layer through preplate splitting.5,6 They facilitate neuronal migration and early network formation via transient synaptic connections,4 with a subset of SP neurons persisting into adulthood as layer 6b (L6b) neurons,7,8 shown to be involved in brain state modulation and arousal regulation.9,10,11 Spatially adjacent to the SP in the deep cortex is the claustral complex (CLA), an evolutionarily conserved structure that forms a thin sheet of neurons between the insular cortex and striatum.12 Similar to SP neurons, CLA neurons are born within a similar window within the cortical ventricular zone and are thought to migrate through the lateral cortical stream and reach their terminal destination around E15–E19.13,14 Although little is understood about the CLA function in development, in adulthood, it is involved in both shared and distinct functions relative to L6b, with both cell types implicated in sleep, consciousness, attention, salience and memory.12,15,16,17,18 The SP and CLA share spatial patterning, molecular properties, temporal progression, and behavioral contributions that suggest a potential developmental relationship.13,19,20,21 Previous studies have provided foundational insights into SP development and its potential link to the CLA, identifying key marker genes and time-dependent expression patterns from embryonic to postnatal stages.6,13,19,20,22,23,24 However, these studies largely relied on a set number of marker genes and lacked single-cell or spatial resolution at developmentally matched timepoints, making it difficult to resolve molecular heterogeneity or developmental progression within and between the SP and CLA. Thus, it is critical to elucidate the temporal patterns of these populations to resolve how the SP and CLA emerge within the developing cortex. Here, we performed single-cell spatial transcriptomic analysis of the developing mouse brain across E12.5, E15.5, E18.5, and postnatal day 1 (P1) time points for both male and female mice. In doing so, we spatially mapped >1.5 million cells from the brain across these four time points. We use this dataset to perform a detailed investigation of the spatiotemporal progression of cell types in the developing brain, including SP and CLA populations. Our work illustrates spatiotemporal divergence of the SP and CLA populations, suggesting that they likely govern distinct aspects of neurodevelopment. Our findings provide new insight into how these deep cortical structures transcriptomically emerge and diversify during neurodevelopment. To facilitate the exploration of the dataset, we have developed an interactive web portal, CellestialCortex (https://developingbrainatlas-cellestialcortex.com/), which will allow users to visualize and explore developing cortical cell types. Our dataset and web portal complement and extend recent efforts generating comprehensive atlases of developing cell types in mouse brains25,26,27,28 by providing an early-embryonic dataset at a single-cell resolution within a spatial context. Results Single-cell spatial transcriptomics captures major cell types within the developing brain To understand the transcriptomic profile of the developing mouse brain, we conducted single-cell spatial transcriptomics (scST) on whole brains across four developmental stages (E12.5, E15.5, E18.5, and P1) in both male and female mice (n = 16 sections from 16 total animals in a sex-balanced fashion: Figures 1A and S1; dataset quality: Figure S2; per-animal analysis: Figure S3). We used a panel of 347 genes, including 247 from the 10x Genomics mouse brain panel and 100 additional genes selected for enrichment in the developing cortex, SP, and CLA (Table S1).Figure 1Single-cell spatial transcriptomics reveals major cell types in the developing brain(A) Single-cell spatial transcriptomics experimental and analytical workflow.(B) Single-cell spatial transcriptomics images of whole brain sections across developmental stages, colored according to cluster identities (n = 4 biological replicates per developmental stage).(C) UMAP embedding of the complete dataset, colored according to cluster identities.(D) As in (C), but colored according to developmental stage.(E) Dotplot showing average gene expression (color) and percentage of expressing cells (dot size) for key marker genes across broad cell identities. Single-cell spatial transcriptomics reveals major cell types in the developing brain (A) Single-cell spatial transcriptomics experimental and analytical workflow. (B) Single-cell spatial transcriptomics images of whole brain sections across developmental stages, colored according to cluster identities (n = 4 biological replicates per developmental stage). (C) UMAP embedding of the complete dataset, colored according to cluster identities. (D) As in (C), but colored according to developmental stage. (E) Dotplot showing average gene expression (color) and percentage of expressing cells (dot size) for key marker genes across broad cell identities. We generated an scST atlas by merging and integrating all datasets, followed by dimensionality reduction and relatively coarse clustering to provide an overview of cortical development. This atlas, hereafter referred to as the “complete” atlas, consisted of >1.5 million cells. Cell types in this complete atlas were phenotyped based on spatial location, developmental stage, and canonical marker expression (Figures 1B, 1D, and 1E), with the majority of clusters capable of being associated with known cell types in this way (n = 12/17 clusters defined). To extract and analyze cortical cells from this complete atlas, cells were subset based upon cluster of origin being largely restricted to the cortex, yielding 448,261 cells (Figure 2A). This cortical subset was subjected to dimensionality reduction and high-resolution clustering to resolve finer-scale heterogeneity within developing cortical cell types. The resulting clusters captured distinct cortical cell populations (Figure 2B; uniform manifold approximation and projection [UMAP] expansion: Figure S4), which were subsequently grouped into coarse categories to provide an overview of major cortical cell types (Figure 2C). UMAP visualization revealed a developmental patterning encompassing transient to mature states: E12.5–E15.5 clusters localized to proliferative and transient regions, while E18.5–P1 clusters transitioned to more defined, non-proliferative populations (Figures 2B–2E). Cortical identities were further validated by canonical markers (Figure 2E), including radial glial cells (RGCs), intermediate progenitor cells (IPCs), excitatory cells, inhibitory cells, superficial-layer cells, deep-layer cells, and neurons comprising the SP and CLA. No significant sex-specific differences were found in the dataset (Figure S5). Ultimately, this scST atlas identifies major known cell populations at a single-cell resolution, from progenitors to terminally differentiated cell types, both across entire brain sections as well as within the cortex.Figure 2Cortical cell types across development(A) Single-cell spatial transcriptomic profiles of cortical cell-type clusters across developmental stages, with colors representing cluster identities (n = 4 biological replicates per developmental stage). Scale bars: 500 μm.(B) UMAP embedding of cortical dataset, colored by cluster identity.(C) As in (B), but colored by broad cell type.(D) As in (B), but colored by age.(E) Average gene expression (color) and percentage of expressing cells (dot size) for key marker genes across broad cell identities. Cortical cell types across development (A) Single-cell spatial transcriptomic profiles of cortical cell-type clusters across developmental stages, with colors representing cluster identities (n = 4 biological replicates per developmental stage). Scale bars: 500 μm. (B) UMAP embedding of cortical dataset, colored by cluster identity. (C) As in (B), but colored by broad cell type. (D) As in (B), but colored by age. (E) Average gene expression (color) and percentage of expressing cells (dot size) for key marker genes across broad cell identities. The SP and claustrum are developmentally distinct The SP and CLA are thought to share developmental origins and a similar developmental progression based on shared marker gene expression during postnatal development, including Nr4a2.13,19 However, there is a lack of insight and comparisons of their marker gene expression at time-matched points during early development, which we sought to directly compare in our scST dataset. To do so, we identified the SP and CLA clusters based on spatial location, with both cell populations becoming first anatomically identifiable at E15.5 (Figures 3A and 3B). Both SP and CLA expressed markers known to be associated with the claustral and cortical SP, such as Nxph4, Kcnab1, Ccn2, Cplx3, and Tle4. Interestingly, these genes show biased expression toward the SP, whereas the CLA exhibited significantly higher expression of other known CLA-specific markers including, Nr2f2 and Cux2 (Figure 3C). Thus, this analysis illustrated divergence in marker-gene expression, including genes that are known to be associated with both SP and CLA populations.Figure 3Subplate and claustrum are developmentally distinct(A) Spatial location of the subplate and claustrum across developmental stages (n = 4 biological replicates per developmental stage). Light gray dots represent Neurod6 transcripts. Scale bars: 500 μm.(B) Subplate and claustrum clusters highlighted within the cortical UMAP embedding.(C) Boxplots of pseudobulked marker genes of the subplate and claustrum. Boxplots show median and interquartile range (IQR), with whiskers extending 1.5× the IQR.(D) Gene expression of markers of cortical deep layers (left) and superficial layers (right).(E) Volcano plot of differential gene expression of subplate and claustrum (significant genes in red, obeying log2FC > 2 and pADJ < 0.001).(F) UMAP embedding of the subplate, claustrum, and transient cell types (left), and a representative section (P1). Scale bars: 500 μm and inset: 100 μm.∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (C). Subplate and claustrum are developmentally distinct (A) Spatial location of the subplate and claustrum across developmental stages (n = 4 biological replicates per developmental stage). Light gray dots represent Neurod6 transcripts. Scale bars: 500 μm. (B) Subplate and claustrum clusters highlighted within the cortical UMAP embedding. (C) Boxplots of pseudobulked marker genes of the subplate and claustrum. Boxplots show median and interquartile range (IQR), with whiskers extending 1.5× the IQR. (D) Gene expression of markers of cortical deep layers (left) and superficial layers (right). (E) Volcano plot of differential gene expression of subplate and claustrum (significant genes in red, obeying log2FC > 2 and pADJ < 0.001). (F) UMAP embedding of the subplate, claustrum, and transient cell types (left), and a representative section (P1). Scale bars: 500 μm and inset: 100 μm. ∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (C). Within the cortical UMAP, we observed a cell-type-specific divergence toward superficial- and deep-layer patterning, with example genes providing expression patterns for anatomical and molecular context for interpreting SP and CLA cells within broader cortical maturation (Figure 3D). Within this divergence, the SP cluster localized to the endpoint of deep-layer cortical maturation, consistent with its molecular similarity to deep-layer excitatory cells. In contrast, the CLA cluster localized to the endpoint of superficial-layer cortical maturation, suggesting a molecular profile that is more aligned with superficial-layer cell types. Reinforcing this, we identified that the deep-layer marker Foxp2 expression was significantly greater in the SP compared to the superficial layer and CLA, while the superficial-layer marker Satb2 was significantly lower in the SP compared to the superficial layer (Figures S6A and S6B). In contrast, the superficial-layer marker Cux2 was expressed significantly greater in the CLA compared to the deep layer and SP. Canonical SP and CLA markers were specific to their respective cell types across developmental stages, while deep- and superficial-layer markers showed preferential expression in SP and CLA populations within each developmental stage, respectively (Figure S6C). To explicitly analyze the differences between these CLA and SP clusters, we computed differentially expressed genes (DEGs) on pseudobulked samples. A substantial number of genes from the total panel were differentially expressed within the SP and CLA (19 and 23 genes obeyed adjusted p value (pADJ) ≤ 0.001 and absolute log fold change (FC) > 2, respectively; Figure 3E). Within the SP, known markers were significantly expressed, such as Nxph4, Kcnab1, Cplx3, and Tle4. Other newly discovered SP markers included those involved in neuronal differentiation and development (Zfmp2, Pcsk5, and Foxp2), extracellular matrix (ECM) and cell adhesion (Tmem132d, Col19a1, and Fmod), and synaptic function (Syt6). Within the CLA, genes involved in ECM and cell adhesion (Col16a1 and Arhgap6) and synaptic function (e.g., Slc44a5, Cbln1, Unc13c, and Syt17) were significantly expressed in addition to neurotrophic growth and guidance factors (Bdnf and Nrp2). Ultimately, our results demonstrate that the SP and CLA have unique developmental progression, which is embodied in their unique transcriptomic profiles and spatial location. To further validate that the SP and CLA represent molecularly distinct populations, we subset transient populations (RGCs, IPCs, and intermediate zone [IZ]) together with the SP and CLA clusters and performed dimensionality reduction and clustering. At a coarse resolution, this analysis transcriptomically and spatially resolved the cortical SP and the CLA. The CLA consisted of the principal CLA (CLp) and claustral SP (CLSp), with the latter being continuous with the cortical SP (Figure 3F). This indicates that the SP and CLA remain transcriptomically and spatially distinct from E15.5 onwards; however, no distinct SP or CLA cell types could be resolved within the transient cell types. scRNA-seq analysis recapitulates SP and claustral divergence in development In order to validate the developmental patterning and transcriptomic profile of the SP and CLA identified in our scST analysis, we analyzed a previously published single-cell RNA sequencing (scRNA-seq) dataset1 that spans embryonic (E12.5, E15.5, and E18.5) and early postnatal (P1) stages (Figure 4A). This dataset was generated from single-cell suspensions prepared from the microdissected somatosensory cortex at E15.5, E18.5, and P1, with the medio-lateral region of the somatosensory cortex microdissected at E12.5,1 with one biological replicate per time point. High-quality cells defined by low mitochondrial gene expression were selected from each stage, and stage data were merged and integrated using reciprocal principal component analysis (RPCA; see STAR Methods), resulting in 36,686 cells being retained for analysis. We performed dimensionality reduction and clustering algorithms, which produced 52 initial clusters that were subsequently consolidated into broad cortical cell-type categories based on developmental stage and canonical marker gene expression (Figures 4B–4D; comparison to scST data: Figure S7).Figure 4Subplate and claustrum cells have a unique transcriptomic profile across development(A) Workflow of scRNA-seq analysis.(B) UMAP embedding of scRNA-seq dataset, colored by broad cell type.(C) As in (B), but colored by stage.(D) Barplot of SP and CLA cell types plotted by stage.(E) Average gene expression (color) and percentage (size) of marker genes used to classify broad cell types.(F) Expression of superficial and deep layer markers.(G) Boxplots of raw gene expression of SP and CLA marker genes. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR.(H) UMAP embedding and representative section depicting cell-type predictions on scST cortical dataset. Gray dots represent Neurod6 transcripts. Scale bars: 500 μm.(I) Upset plot of total DEGs for SP and CLA cell types. Subplate and claustrum cells have a unique transcriptomic profile across development (A) Workflow of scRNA-seq analysis. (B) UMAP embedding of scRNA-seq dataset, colored by broad cell type. (C) As in (B), but colored by stage. (D) Barplot of SP and CLA cell types plotted by stage. (E) Average gene expression (color) and percentage (size) of marker genes used to classify broad cell types. (F) Expression of superficial and deep layer markers. (G) Boxplots of raw gene expression of SP and CLA marker genes. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR. (H) UMAP embedding and representative section depicting cell-type predictions on scST cortical dataset. Gray dots represent Neurod6 transcripts. Scale bars: 500 μm. (I) Upset plot of total DEGs for SP and CLA cell types. This analysis provided a complementary approach to identify the major cortical cell types that emerged across development (Figures 4B–4E). The developmental progression began with RGCs and IPCs at early stages (E12.5–E15.5). RGCs were defined by Nes, Hes5, Pax6, and Sox2, while IPCs were defined by Eomes. Both populations also expressed proliferative cell-cycle markers such as Mki67 and Ccnd2. These early transient cell populations gave rise to divergent deep- and superficial-layer excitatory patterning, spanning later stages at E15.5–P1, consistent with our scST findings (cf. Figure 3D). Consistent with our scST findings, at the endpoints of the deep and superficial-layer developmental progression, the SP cluster localized to the terminus of the deep-layer population (characterized by markers including Bcl11b, Pcp4, Fezf2, and Foxp2) and the CLA cluster localized to the terminus of the superficial-layer population (characterized by Satb2, Pou3f2, Cux1, and Cux2) (Figures 4B and 4F). As expected, there was good general concordance between this scRNA-seq dataset and our scST dataset (FCs between datasets: R value = 0.52 and p value = 2.2e−16, Figure S7; note that because the scRNA-seq dataset contained only one replicate/time point, Wilcoxon rank-sum testing was required for scRNA-seq differential expression analysis). Altogether, these results confirm that the SP and CLA cell types diverge during cortical development in a superficial versus deep patterning. Further analysis of these cell types suggested three transcriptionally distinct SP cell types and a singular CLA cell type. The SP comprised of (1) an SP-IPC cell type (E12.5–E15.5) marked by Nxph4 and the IPC marker Eomes; (2) a putative SP cell type (E12.5–E18.5) expressing Nxph4, Tle4, and Zfpm2, and a lack of RGC/IPC marker gene expression; and (3) a terminal SP cell type (E15.5–P1) expressing Nxph4, Kcnab1, Tle4, Zfpm2, Nr4a2. The CLA cell type (E18.5–P1) was defined by Nr4a2, Gnb4, Rgs6, and Lxn expression, similar to our scST results, with no evidence of putative or intermediate CLA subtypes (Figures 4B, 4E, and 4G). To further validate that the CLA was captured within the scRNA-seq dataset, we performed single-cell resolution analysis (SingleR) using the annotated scRNA-seq dataset with our cortical scST dataset (cf. Figure 2). The scRNA-seq identified the CLA within the UMAP embedding and spatially in our cortical scST dataset (Figure 4H). Additionally, no evidence of sex differences was observed within the SP and CLA (Figure S8). Examining DEGs across SP and CLA cell types, we identified 60 DEGs unique to the SP-IPC, 111 to the putative SP, 317 to the SP, and 139 to the CLA. Shared markers included 112 genes between the SP and CLA, 32 between the putative SP and SP-IPC, and 37 between the SP and putative SP (Figure 4I). The SP-IPC expressed markers related to early developmental processes (e.g., Epha3, Nrp1, Ddr1, Acvr2b, and Unc5d) and transcription factors (e.g., Eomes, Nhlh1, and Ebf2), while putative SP expressed markers related to neuronal differentiation (e.g., Nefl, Nefm, and Draxin) and synaptic development (e.g., Ctnna2, Grik2, and Gabra2). The SP expressed markers associated with synaptic development and maturation (e.g., Sv2a, Snap91, Kcnab1, and Scn2a1) and ECM/cell-surface modification (e.g., Fut9, Hs3st4, Timp2, and Col19a1). The CLA also expressed markers related to synaptic transmission (e.g., Gria4, Cplx1, and Grm5), as well as neuronal signaling/cell adhesion (e.g., Rgs6, Gnb4, Cdh12, and Cdh8) (Table S2). Collectively, this validates the terminally differentiated SP and CLA populations identified in our scST analysis and suggests previously unresolved SP subtype heterogeneity within progenitor populations. The putative SP exhibits a unique transcriptomic profile and may arise from a distinct RGC population Interestingly, we observed a lack of E12.5 cells within the SP cluster in our scST analysis (Figures 2B–2D) and in our scRNA analysis (Figures 4B–4D). Additionally, while our scST results resolved the SP as a single population, the scRNA-seq analysis suggested putative SP subtypes, potentially indicative of additional transcriptomic heterogeneity not captured at coarse resolution. Together, these results motivated a detailed analysis of progenitor populations with our scST dataset to understand SP maturation. To do this, we subset transient cell types (RGCs, IPCs, and IZ cells) from the cortical dataset and performed dimensionality reduction and clustering, which we then grouped into broad identities based on marker gene expression, spatial location, and developmental stage (Figures 5A–5D). Within our progenitor subset, we identified three unique clusters expressing known SP markers: a cortical RGC population (Nr4a2), a putative SP/Hippocampal (HPC) RGC subtype (Nxph4), and a putative gliogenic RGC subtype (Ccn2) (Figures 5A, 5B, and 5D–5F). The gliogenic-like RGC subtype exhibited a transcriptional profile distinct from SP-associated populations (including the glial markers Aqp4, Gfap, and Olig1) and was not considered further.Figure 5Putative subplate cells emerge from a unique RGC population(A) UMAP embedding of ventricular zone subset, colored by broad cell identity (n = 4 biological replicates per developmental stage).(B) As in (A), but colored by cell-type clusters.(C) As in (A), but colored by developmental stage.(D) Average raw gene expression (color) and percentage of expressing cells (dot size) for key marker genes across broad cell identities.(E) Spatial location of transient subtypes expressing known subplate markers. Light gray dots represent Neurod6 transcripts. Scale bars: 500 μm.(F) Boxplots of pseudobulked marker genes. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR.(G) As in (A), but colored by cell-type predictions assigned by SingleR via scRNA-seq reference dataset.(H) As in (E), but colored by predicted cell types, with representative sections at E12.5 and E15.5. Light gray dots represent Neurod6 transcripts (scale bars: 500 μm).∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (F). Putative subplate cells emerge from a unique RGC population (A) UMAP embedding of ventricular zone subset, colored by broad cell identity (n = 4 biological replicates per developmental stage). (B) As in (A), but colored by cell-type clusters. (C) As in (A), but colored by developmental stage. (D) Average raw gene expression (color) and percentage of expressing cells (dot size) for key marker genes across broad cell identities. (E) Spatial location of transient subtypes expressing known subplate markers. Light gray dots represent Neurod6 transcripts. Scale bars: 500 μm. (F) Boxplots of pseudobulked marker genes. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR. (G) As in (A), but colored by cell-type predictions assigned by SingleR via scRNA-seq reference dataset. (H) As in (E), but colored by predicted cell types, with representative sections at E12.5 and E15.5. Light gray dots represent Neurod6 transcripts (scale bars: 500 μm). ∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (F). The cortical RGC population, located in the ventricular region at E12.5, had significantly higher expression of an SP marker (Nr4a2), an RGC marker (Hes5), an IPC marker (Eomes), and a proliferative marker (Mki67), consistent with an SP-derived RGC identity (Figures 5A, 5B, 5F, and 5G). The putative SP/HPC RGC subtype, located most superficially at E12.5, had significantly higher expression of an SP marker (Nxph4), IPC marker (Eomes), and non-proliferative marker (Rbfox3) (Figures 5E and 5F), spatially and transcriptomically consistent with the putative SP. To further validate the localization of Nxph4 within the putative SP cluster, we confirmed that Nxph4 was enriched within putative SP cells and largely absent from adjacent progenitor populations (Figure S9). Notably, the spatial location of this cluster shifted between E12.5 and E15.5, transitioning from the cortical ventricular region to the hippocampal ventricular region by E15.5 (Figure 5E). This spatial shift suggests that the E12.5 putative SP may share a developmental relationship with hippocampal progenitor populations. To validate this identity, we performed single-cell resolution analysis (SingleR) using the annotated scRNA-seq dataset (cf. Figure 4). The IPC-SP and putative SP identified in the scRNA-seq dataset spatially aligned with the putative SP identified in the scST data (Figures 5E, 5G, and 5H), confirming that the E12.5 putative SP shared a molecular profile with hippocampal RGC/IPC populations (Figures 5G and 5H). Together, these results suggest that SP cells may arise from a unique RGC subtype, and that the putative SP shares molecular similarity with RGC/IPC populations. The SP has unique subtypes that emerge across development Given that the scRNA-seq analysis suggested the existence of SP subtypes, we next sought to examine the possibility of spatially and transcriptomically distinct SP subtypes within our scST dataset. To do this, we extracted the SP cluster and performed dimensionality reduction followed by coarse-resolution clustering, which resolved four SP clusters (Figures 6A–6C), hereafter defined as subtypes I (immature), II (immature), III (transitional), and IV (mature). This analysis also revealed two additional subtypes (V and VI), which had high expression of genes typically associated with inhibitory neurons (Gabra2, Calb2, Gad1, and Gad2; Figure S10), and thus were not analyzed further. Further validating this subtype identity, there was minimal contamination of inhibitory and glial-related marker genes across SP subtypes (Figure S11).Figure 6Subplate subtypes across development(A) Atlas schematics of the subplate.(B) Spatial location of subplate subtypes across developmental stages (n = 4 biological replicates per developmental stage). Light gray dots represent Neurod6 transcripts (scale bars: 500 μm, inset: 100 μm).(C) UMAP embedding of the SP subset, colored by subtype.(D) As in (C), but colored by developmental stage.(E) Barplot of subtypes plotted by age.(F) Boxplots of pseudobulked SP marker genes. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR.(G) As in (F), but colored by age.(H) Upset plot depicting intersecting DEGs between subtypes and total DEGs per subtype.(I) Volcano plot of differential gene expression between E15.5 and P1 (significant genes in red, obeying log2FC > 2 and pADJ < 0.001).∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (F and G). Subplate subtypes across development (A) Atlas schematics of the subplate. (B) Spatial location of subplate subtypes across developmental stages (n = 4 biological replicates per developmental stage). Light gray dots represent Neurod6 transcripts (scale bars: 500 μm, inset: 100 μm). (C) UMAP embedding of the SP subset, colored by subtype. (D) As in (C), but colored by developmental stage. (E) Barplot of subtypes plotted by age. (F) Boxplots of pseudobulked SP marker genes. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR. (G) As in (F), but colored by age. (H) Upset plot depicting intersecting DEGs between subtypes and total DEGs per subtype. (I) Volcano plot of differential gene expression between E15.5 and P1 (significant genes in red, obeying log2FC > 2 and pADJ < 0.001). ∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (F and G). Subtypes I and II had the highest proportion of E15.5-derived cells, and thus were classified as immature subtypes. Subtype III had a mix of cells from all stages; as such, it was classified as a transitional subtype. Lastly, subtype IV had the highest proportion of cells derived from E18.5–P1; as such, it was classified as a mature subtype (Figures 6D and 6E). This suggests there are unique subtypes emerging throughout the developmental stages. We next characterized each subtype by canonical SP marker expression. Subtype IV (mature) had significantly higher expression of some canonical SP markers, including Nxph4, Nr4a2, Kcnab1, and Ccn2. Consistent with this, visualization of canonical SP markers at a single-cell resolution demonstrated enrichment of these SP marker genes within the broader SP and mature SP cells, with minimal expression in adjacent deep-layer populations (Figure S12). In contrast, subtype I (immature) had significantly lower Nr4a2 expression, while subtype II (immature) had significantly lower expression of Kcnab1 and, although not significant, reduced expression of Nxph4 (Figure 6F). Furthermore, Nr4a2, Kcnab1, Cplx3, and Ccn2 had significantly greater expression within P1 cells compared to earlier developmental stages (Figure 6G). To further understand the differences between these subtypes and their developmental progression, we performed differential gene expression analysis. Each subtype could be associated with its own unique suite of DEGs (35, 24, 4, and 14, respectively; Figure 6H and Table S3). From this, we postulated that DEGs could also be directly obtained from developmental age and computed a similar analysis between E15.5 and P1. At E15.5, enriched genes included those involved in transcriptional regulation and neuronal identity (Nhlh2, Zfp366, Prdm8, and Lhx9). By contrast, at P1, genes involved in synaptic function (e.g., Cplx3, Syt6, Necab2, and Nrgn) and ECM/adhesion/guidance (Ccn2, Cd44, and Sdk2) were significantly expressed (Figure 6I). Together, these results suggest that the SP consists of unique subtypes that dynamically change throughout development, consistent with the SP having specialized roles at specific time points. Developing claustral cells have unique transcriptomic profiles consistent with emergence from the lateral cortical stream We next sought to examine the transcriptomic heterogeneity within the CLA across development. To accomplish this, we generated a subset of the lateral cortical stream (LCS), a region associated with migrating cell types and CLA (Figure 7A), and applied dimensionality reduction and clustering at a coarse resolution to identify distinct subtypes (Figures 7B and 7C). Notably, the CLA clusters were predominantly composed of later developmental stage cells (namely, E15.5, E18.5, and P1) and intriguingly recapitulated known distinct subdomains of the CLA. Specifically, the CLA consists of the CLp, comprised of a compact sheet of cells adjacent to the insular cortex, as well as the CLSp that lies medial to the CLp and is continuous with the cortical SP, with both of these domains identified in our scST clusters (Figures 7B and 7C).Figure 7Claustral subtypes emerging from the lateral cortical stream(A) Atlas schematics of the CLA complex and LCS.(B) Spatial location of claustral subtypes across developmental stages (n = 4 biological replicates per developmental stage).(C) UMAP embedding of CLA complex and LCS subsets, colored by subtype.(D) As in (C), but colored by developmental stage.(E) Subtype proportions plotted by age. Light gray dots represent Neurod6 transcripts (scale bars: 500 μm, inset: 100 μm).(F) Boxplots of pseudobulked expression for CLA and CLA-subplate-specific marker gene expression. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR.(G) Upset plot depicting intersecting DEGs between subtypes and total DEGs per subtype.(H) Volcano plot of differential gene expression between E15.5 and P1 (significant genes in red, obeying log2FC > 2 and pADJ < 0.001).∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (F). Claustral subtypes emerging from the lateral cortical stream (A) Atlas schematics of the CLA complex and LCS. (B) Spatial location of claustral subtypes across developmental stages (n = 4 biological replicates per developmental stage). (C) UMAP embedding of CLA complex and LCS subsets, colored by subtype. (D) As in (C), but colored by developmental stage. (E) Subtype proportions plotted by age. Light gray dots represent Neurod6 transcripts (scale bars: 500 μm, inset: 100 μm). (F) Boxplots of pseudobulked expression for CLA and CLA-subplate-specific marker gene expression. Boxplots show median and interquartile range, with whiskers extending 1.5× the IQR. (G) Upset plot depicting intersecting DEGs between subtypes and total DEGs per subtype. (H) Volcano plot of differential gene expression between E15.5 and P1 (significant genes in red, obeying log2FC > 2 and pADJ < 0.001). ∗pADJ < 0.05, ∗∗pADJ < 0.01, ∗∗∗pADJ < 0.001 by Kruskal-Wallis test with Dunn’s post hoc and Benjamini-Hochberg correction (F). At E12.5, defined CLA cells were not identified. However, we identified two LCS subtypes: LCS subtype I and LCS subtype II/reservoir (Figures 7B and 7C). LCS subtype I was spatially continuous with the ventricular zone and transitioned into subtype II, forming the LCS reservoir. Interestingly, LCS subtypes I and II had cells that expressed a canonical claustral marker (Nr2f2) (Figure 7F). This suggests that at E12.5, the CLA cells are still migrating to their terminal location, which contrasts with the putative SP, which at this stage had reached their terminal location (cf. Figures 5A, 5B, and 5E). To further assess whether Nr2f2-positive cells within the E12.5 LCS represented migrating CLA cells in the context of a heterogeneous migratory stream, we examined the single-cell localization of Nr2f2 together with inhibitory and progenitor marker genes (Figure S13). We found that inhibitory transcripts were localized primarily to cells outside the LCS, with minimal contamination by progenitor marker genes within the LCS itself. Moreover, the E12.5 LCS contained both Nr2f2-positive cells, consistent with putative migrating CLA cells, and Nr2f2-negative cells, likely reflecting other cell types within this heterogeneous migratory stream. By E15.5, Nr2f2 expression was observed within the CLA and was reduced in the LCS, supporting Nr2f2-positive cells in the E12.5 LCS, representing migrating CLA cells that subsequently reach the developing CLA. The CLA structure first emerged at E15.5 and was defined by the expression of Nr4a2 and Nr2f2. At a coarse resolution, we identified three CLA subtypes that each had unique spatial locations and gene expression profiles, hereafter referred to as CLA subtype I, CLA subtype II, and the CLSp (Figures 7B and 7F). CLA subtype I was located dorsally relative to CLA subtype II and included cells within the insular cortex at E18.5 and P1. Both subtypes expressed significantly higher levels of the CLA canonical marker Nr4a2. There was no significant difference in the expression of superficial layer markers (Cux1 and Cux2) between subtypes. We also identified the CLSp, which, although continuous with the cortical SP, clustered with the CLA in the broader cortical dataset (Figures 2A–2C). The CLSp had significantly greater expression of the canonical SP markers (Nxph4, Kcnab1, and Ccn2). Overall, these findings suggest that CLA cells may begin migrating within the LCS as early as E12.5, before emerging as a distinct structure around E15.5 and continuing to develop thereafter. To understand the difference between these subtypes, we performed differential gene expression analysis (Figure 7G and Table S4). DEGs were enriched across LCS subtypes I and II, CLA subtypes I and II, and the CLSp (62, 19, 29, 25, and 63 genes, respectively). Furthermore, we identified shared DEGs between LCS subtypes I and II, CLA subtypes I and II, and LCS subtype I and CLA subtype I (45, 8, and 6, respectively). Finally, as in our SP analysis, we identified genes that were differentially expressed between E15.5 and P1. At E15.5, enriched genes included those involved in growth and progenitor identity (Igf2 and Eomes), while at P1, genes involved in synaptic function and plasticity (e.g., Pcp4, Scg2, Cpne6, Fos, and Drd1) were significantly expressed (Figure 7H). Collectively, these findings highlight that the LCS and CLA subtypes are defined by distinct transcriptomic profiles, with stage-specific differences, consistent with developmental maturation. Discussion Distinct developmental patterns of the SP and claustrum despite shared select features A variety of spatial, maturational, and behavioral features are shared between the SP and CLA. Spatially, the SP and CLA occupy adjacent regions in the deep cortex, with the SP forming the deepest cortical layer and the CLA located ventrolaterally beneath the insular cortex. Developmentally, both populations are among the earliest-born pallial neurons (E10.5–E12.5) and have been associated with similar marker genes.19 Specifically, during embryonic development, both the SP and CLA share expression of markers of the deep layer (Satb2) and superficial layer neuronal fate (Cux2, Pou3f3, Pou3f2, and Lmo4)19. In maturity, the SP-derived L6b and CLA in the adult both exhibit expression of Nr4a2,13,19 and other SP-specific markers (Cplx3, Tle4, and Ctgf) form a continuous band of expression from L6b into the deepest layer of the CLA.13,16,19,29 Furthermore, both structures make contributions to similar functional and behavioral properties, including integration of diverse inputs and modulation of the sleep-wake cycle.9,10,11,12,15,16,17,18,30 This overall body of work highlights similarities between the SP and CLA and is suggestive of a shared origin and developmental progression. In contrast to this, in both our scST and scRNA-seq analyses, we identified that the SP and CLA diverge during postnatal development. In particular, the mature SP and CLA, respectively, resemble the molecular profile of cortical deep layers (including expression of Bcl11b, Pccp4, Foxp2, and Fezf2) and cortical superficial layers (including expression of Satb2, Pou3f2, Cux1, and Cux2) (Figures 3D and 4F). Notably, we identified the expression of Nr2f2 to be significantly greater within the CLA, indicating Nr2f2 may represent a selective marker of CLA identity (Figure 3C). We also identified a range of markers that were significantly different between the SP (e.g., Zfpm2, Col19a1, and Foxp2) and CLA (e.g., Slc44a4, Cbln1, and Angpt1) (Figure 3E). This suggests that although the SP and CLA do, indeed, share expression of certain markers, they have unique molecular profiles that contribute to a divergence in cortical development. Interestingly, a recent macaque CLA atlas reported that CLA glutamatergic neurons resemble deep-layer insular neurons,31 whereas the mouse CLA in our study showed greater similarity to superficial layer cells. Although species-specific molecular differences may contribute to this discrepancy, an alternative interpretation is that the upper-layer-like neurons are conserved in both mouse and macaque CLA, but constitute a larger proportion of the CLA population in the mouse. Several aspects of our results are consistent with these differences being driven by distinct developmental progressions. First, during the birth window of these regions, we identified that the putative SP resembles the molecular profile of a proliferative IPC cell type, which may reflect a transitional population differentiating toward an SP-like fate (Figures 5A, 5E, and 5F). Notably, we did not identify the putative CLA at this time point (Figure 5A). Second, at E12.5 we identified a unique RGC population consistent with SP derivation (Figures 5A, 5E, and 5F); conversely, Nr2f2-expressing CLA-associated cells were identified in the LCS, indicating that these cells are still migrating to their terminal location during this time window (Figures 7A, 7E, and 7F). Third, the transcriptomic profile of SP and CLA subtypes evolved on distinct temporal scales: the SP exhibited heterogeneity across development (Figures 6A–6C, 6E, and 6F), whereas the CLA maintained stable subtypes across development with unique molecular properties (Figures 7A–7C, 7E, and 7F). These findings of distinct developmental progressions and mature identities of the SP and CLA are consistent with some observed distinct spatiotemporal dynamics of these cells during development. Notably, the CLA follows a radial/reverse migration pattern and forms a distinct structure around E14.5–15.5, while the SP migrates radially and can be observed at earlier time points,13,14,16 suggesting that these regions may have diverging developmental patterns. Implications of developmental heterogeneity in the SP and claustrum Although temporal shifts in SP-specific gene expression have been previously described,6,22 the degree of transcriptomic and spatial heterogeneity during development remains unresolved. Here, we aimed to understand the transcriptomic and spatial heterogeneity of the SP and resolved four SP subtypes with distinct molecular signatures across development. SP subtype I (immature; E15.5–E18.5 enriched) expressed genes enriched for ECM organization and axonal guidance (Reln, Sema3e, Rspo2, and Col19a1), as well as depletion of progenitor-specific markers (Hes5, Sox11, and Meis2), indicative of an early differentiating SP population establishing extracellular scaffolds and nascent synaptic connections. SP subtype II (immature; E15.5–E18.5 enriched) exhibited upregulation of genes linked to synaptogenesis (Gsg1l and Syt6), deep-layer identity (Foxp2), and downregulation of developmentally early markers (Reln, Rspo2, and Sema3e), reflecting an immature state. Subtype III (transitional; E15.5–P1 enriched) expressed genes involved in early developmental signaling (e.g., Arhgap6 and Dkk3). Subtype IV (mature; P1 enriched) expresses genes involved in guidance refinement and neuromodulatory signaling (Tacr1 and Nrp2), while downregulating guidance and laminar fate specification genes (Cdh6, Nell1, Rorb, and Pou3f1) (Figure 6I and Table S3). Ultimately, these findings highlight temporal and transcriptomic factors that drive SP heterogeneity, and suggest unique functional roles across narrow windows of development. Prior anatomical and developmental studies indicate that the CLA consists of spatially unique cell populations with specialized molecular signatures during development.13,16,18,19,21,24,29 Recently, two CLA subtypes were identified within the adult CLA, with CLA cells also present in the dorsal endopiriform nucleus (EpD), a region that is positioned ventrally to the CLp and has typically been differentiated from the CLA by having sparse expression of CLA-specific markers.13,21,24 Additionally, complementary work has identified CLA-associated dispersed cells called Arimatsu cells that reside within the insular cortex and express Nr4a2 and latexin,32,33,34 which may be established through reverse migration14 and represent early counterparts of shell projection neurons described in adult CLA.35,36 Within our study, we sought to take a data-driven approach to understand the transcriptomic heterogeneity and spatial patterning of neurons comprising the CLA. In doing so, we resolved two distinct CLA subtypes: CLA subtype I expressed markers associated with cell adhesion/axonal guidance (Cdh6, Rspo2, Rtn4r, and Tacr1) and lower expression of markers of progenitor/proliferative genes (Pax6, Mki67, and Ccnd2), suggesting a migratory population (Figure 7H and Table S4). This subtype also occupied a dorsal domain that includes a dispersed population extending into the insular cortex, corresponding to the Arimatsu cells described previously32,33,34 (Figure 7E). CLA subtype II exhibited upregulation of synaptic regulators (Cbln1, Unc13c, Cpne6, Nrgn, Rasgrf2, and Scg2) and neurotrophic factors (Bdnf, Htr1f, and Rbp4), which may reflect neurons forming mature synaptic connections (Figure 7H and Table S4). This subtype occupied a ventral domain encompassing the region corresponding to the EpD (Figure 7E). Notably, both subtypes I and II expressed CLA-specific markers, suggesting that the CLp and EpD form a molecularly continuous CLA. Together, these findings highlight the fundamentally different organizational properties governing the SP and CLA during development: the SP exhibits multiple subtypes with unique transcriptomic signatures with a relative lack of spatial organization, whereas the CLA exhibits distinct spatial domains. Expanding beyond SP and claustrum to developing cell types Recent transcriptomic studies in both the mouse25,26,27,28 and human25,37,38,39,40 have begun to create developmental cell atlases of the brain, capturing the molecular transitions and spatial organizations across embryonic and early postnatal stages. These large-scale datasets provide invaluable resources for understanding neuronal and non-neuronal developmental progression and spatial patterning. One study profiled ∼1.6 million nuclei from 101 embryos using combinatorial indexing small nuclear RNA sequencing (snRNA-seq) at a single time point, providing broad coverage but limited per-embryo depth and no spatial information.28 Another effort generated ∼1.2 million single-cell transcriptomes spanning prenatal, postnatal, and adult stages, but lacked spatial resolution.27 Our work here, capturing ∼1.5 million spatially resolved cells from the brain, with over 300 marker genes across four key developmental time points, complements and extends this work. This dataset yields a substantially greater number of cells per animal and enables direct mapping of transcriptomic identity to precise anatomical locations. Additionally, it provides an immense opportunity to understand the spatiotemporal progression of developing cell types, ranging from progenitors to terminally differentiated cell types, at a single-cell resolution. We further developed an interactive web portal (https://developingbrainatlas-cellestialcortex.com/) that allows users to explore cortical cell types and gene expression across developmental timepoints. Together, these features make our dataset one of the most comprehensive resources to date for defining the origins, maturation, and spatial organization of developing cell types in the brain. Limitations of the study There are several limitations to our study that should be noted. First, our scST dataset used a panel of 347 genes with a variety of genes targeted to SP and CLA neurons. This approach offers high-sensitivity single-cell resolution with preserved spatial context, but as this panel does not provide whole-transcriptome breadth, it may introduce potential bias toward specific cell types. Second, our sex-specific analyses were limited by sample size, with two males and two females included per developmental stage. While we did not detect significant sex-specific differences, additional animals in future studies will help to fully elucidate and validate potential sex-dependent transcriptomic differences across development. Third, our cell-type maturation comparisons were generally age matched, but heterochronicity in development may account for additional differential expression observed between populations. Fourth, while our work suggests potential functions of subtypes based on marker gene expression, additional studies are needed to elucidate functional activity and how this relates to the transcriptomic and spatial properties of SP and CLA subtypes. To help guide this future work, the scST data generated through our study provide a range of targetable molecules in the SP and CLA that can be used to guide further studies linking molecular and cellular properties to higher-order circuit and functional properties. Resource availability Lead contact Additional information and requests for resources and/or reagents should be directed to and will be fulfilled by the lead contact, Mark S. Cembrowski (mark.cembrowski@ubc.ca). Materials availability This study did not generate new reagents. Data and code availability •Single-cell RNA sequencing data are available from Bella et al.1 (GEO: GSE153164).•Single-cell spatial transcriptomics data have been deposited on GEO and are available (GEO: GSE310599).•All original code has been deposited and made available.•Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. •Single-cell RNA sequencing data are available from Bella et al.1 (GEO: GSE153164). Single-cell RNA sequencing data are available from Bella et al.1 (GEO: GSE153164). •Single-cell spatial transcriptomics data have been deposited on GEO and are available (GEO: GSE310599). Single-cell spatial transcriptomics data have been deposited on GEO and are available (GEO: GSE310599). •All original code has been deposited and made available. All original code has been deposited and made available. •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments M.S.C. is supported by the Natural Sciences and Engineering Research Council of Canada (RGPIN-2024-03916), the Canadian Foundation for Innovation (John R. Evans Leaders Fund 38369), and the 10.13039/501100000096Scottish Rite Charitable Foundation of Canada (research grant no. 22101). S.I. is supported by the Canada Graduate Research Scholarship—Doctoral from the Natural Sciences and Engineering Research Council. This research was supported, in part, through computational resources and services provided by Advanced Research Computing at the 10.13039/501100005247University of British Columbia, as well as Jeff LeDue and the NeuroImaging and NeuroComputation Core at the 10.13039/501100025249Djavad Mowafaghian Centre for Brain Health (RRID: SCR_019086). We thank Dr. Jessica Rosin for insight and guidance regarding embryonic tissue dissections. We would also like to thank Ingrid Barta at 10.13039/501100005247UBC’s Diagnostic and Research Histology Laboratory for support in preparing and sectioning tissue for single-cell spatial transcriptomics. We thank members of the Cembrowski lab for helpful discussions, especially Dr. Larissa Kraus for providing guidance with experimental protocols and data analysis. Author contributions S.I. collected and prepared tissue for single-cell spatial transcriptomic experiments. S.I. also conducted all the data analysis, including single-cell RNA sequencing and single-cell spatial transcriptomics, and wrote the manuscript. M.S.C. supervised the project and wrote the manuscript. Declaration of interests The authors declare no competing interests. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the authors used ChatGPT to review grammar and spelling. After using this tool, the authors reviewed and edited the content as needed. No generative AI tools have been used to produce any new content. The authors take full responsibility for the content of the publication. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERDeposited datascRNA-seq dataBella et al.1GEO: GSE153164Single-cell spatial transcriptomics dataThis paperGEO: GSE310599Experimental models: Organisms/strainsTimed Pregnant C57BL/6 miceJackson LaboratoryRRID:IMSR_JAX:000664Software and algorithmsRR Development Core Team41RRID: SCR_001905SeuratSajita et al.42RRID: SCR_016341CodeThis paperhttps://doi.org/10.6084/m9.figshare.33233709 REAGENT or RESOURCESOURCEIDENTIFIERDeposited datascRNA-seq dataBella et al.1GEO: GSE153164Single-cell spatial transcriptomics dataThis paperGEO: GSE310599Experimental models: Organisms/strainsTimed Pregnant C57BL/6 miceJackson LaboratoryRRID:IMSR_JAX:000664Software and algorithmsRR Development Core Team41RRID: SCR_001905SeuratSajita et al.42RRID: SCR_016341CodeThis paperhttps://doi.org/10.6084/m9.figshare.33233709 Experimental model and study participant details Animals Experiments used time-pregnant wild-type C57BL/6 mice, with animals were maintained on a 12-h reverse light-dark cycle, with ad libitum access to food and water. All experimental protocols were approved by the University of British Columbia Animal Care and Use Committee (Animal Protocol #: A22-0255). Male and female C57BL/6 mice were paired together for breeding and females were checked for the presence of a vaginal plug each day. When a plug was observed, it was marked as embryonic day 0.5 (E0.5) and the females were housed individually for the remainder of their pregnancy. Method details Tissue collection and preparation for single-cell spatial transcriptomics Embryos were collected at embryonic day 12.5 (E12.5), E15.5 and E18.5 and postnatal samples were collected on postnatal day 1 (P1), where birth was marked as P0. In all cases, the pregnant female was deeply anesthetized, followed by an abdominal incision to remove the uterine horns, which were placed in ice-cold 1× PBS. The embryos were separated, decapitated and brains dissected using the Leica M165 FC stereo microscope. Pup brains were dissected in a similar manner. Tails were collected for sex genotyping, which was performed by The Biomedical Research Center Core Facility at the University of British Columbia. Following dissection, samples were drop-fixed in 10% Neutral Buffered Formalin at 4°C on a shaker for 24 h, followed by three 5-min 1× PBS washes and dehydration in 70%, 50% and 30% ethanol for 1 h at room temperature. Samples were stored in 70% ethanol at 4°C until paraffin embedding and sectioning, which was performed by the Diagnostic and Research Histology Laboratory at the University of British Columbia. Single-cell spatial transcriptomics data acquisition Samples were prepared according to Xenium In Situ FFPE Tissues – Tissue Preparation Guide (10× Genomics) by the Diagnostic and Research Histology Laboratory at the University of British Columbia. A total of 16 animals were used, with 2 males and females per developmental stage. Tissue was sectioned at 10 μm thickness on a 12 × 24 mm capture frame Xenium slide. Subsequently, slides were stored at room temperature in desiccant and used within five days. For use, slides were deparaffinized, assembled into the Xenium cassettes and decrosslinked according to the Xenium In Situ FFPE Tissues – Deparaffinization and Decrosslinking protocol (10× Genomics). Next, the Xenium In Situ Gene Expression with Cell Segmentation User Guide (10× Genomics) was followed. Probe hybridization was performed at 50°C for 16–24 h using the pre-designed 247-probe Xenium Mouse Brain Gene Expression panel (10× Genomics, cat#: 1000462) combined with a custom 100-probe add-on panel (10× Genomics, cat#: XER9KU; Data Table S1) targeting cortical developmental and subplate marker genes. Next, slides were washed to remove unhybridized probes, followed by a 2-h ligation procedure at 37°C to anneal the probes to the mRNA targets. Subsequently, the probes were amplified enzymatically using rolling circle amplification PCR at 37°C for 2 h. Following amplification incubation, the tissue was blocked and stained for 16–24 h at 4°C. Next, a stain enhancer was applied, followed by an autofluorescence quencher and a DAPI nuclear stain. The slide cassettes were finally loaded onto the Xenium Analyzer Instrument for fluorescent probe hybridization, imaging and decoding. The Xenium analyzer provides high spatial precision in transcript localization, with XY localization below 30 nm and Z localization below 150 nm, with images acquired at a resolution of ∼0.2 μm/pixel. Single-cell spatial transcriptomics data analysis Data were analyzed using the Seurat package (v5) in R (v4.4.2) with in-house code analogous to previously published workflows.43,44 Default parameters were used for the following code, unless otherwise specified. Sixteen tissue sections were merged, and cells with zero counts (nCount_Xenium = 0) were removed. The merged dataset was normalized using SCTransform(), dimensionally reduced by RunPCA(), integrated using IntegrateLayers(method = HarmonyIntegration, orig.reduction = “pca”, new.reduction = “harmony”, normalization.method = “SCT”), UMAP embedded in two dimensions with RunUMAP(dims = 1:40), and clustered using a shared nearest-neighbor graph via FindNeighbors(dims = 1:40) and FindClusters(resolution = 0.1). To isolate the whole brain from the E12.5 whole-embryo samples, regions were cropped with Crop() using the corresponding x and y coordinates for each E12.5 sample. This complete dataset, spanning all four developmental stages, was then re-processed. Specifically, the dataset was split using SplitObject(split.by = “animal”). Each sample was normalized using SCTransform(), followed by variable features detection within each sample using SelectIntegrationFeatures(). The dataset was subsequently processed using ScaleData(), reduced by RunPCA(features = selected variable features), IntegrateLayers(), RunUMAP(dims = 1:30), and clustered FindNeighbors(dims = 1:30) and FindClusters(resolution = 0.2). The same workflow of variable feature selection, integration, dimensionality reduction, and clustering were applied to cortical (dims = 1:15; resolution = 0.8), transient (dims = 1:20; resolution = 1.0), subplate (dims = 1:10; resolution = 0.2), and claustrum (dims = 1:20; resolution = 0.4) subsets. For these analyses, only the number of UMAP dimensions and the clustering resolution were adjusted, as indicated. For differential gene expression analysis, pseudobulked expression profiles were generated using AggregrateExpression(). RNA counts were aggregated across cells within each sample (per orig.ident). Samples were grouped by cell subtype and stage, by stage or by sex. For pseudobulk comparisons, differential expression was assessed using FindMarkers(test.use = DESeq2). Stage-specific differences were tested between E15.5 and P1 and sex differences were tested between male and female samples for datasets. Significant genes were defined as those with adjusted pADJ ≤ 0.001 and absolute log FC > 2. In parallel, cluster-wide differential expression across all subtypes within the selected dataset was tested using FindAllMarkers(test.use = DESeq2, only.pos = FALSE, log FC threshold = 0.5). Significant genes were retained if they met pADJ < 0.05 and absolute log FC > 1. Gene expression was visualized by plotting normalized counts using FeaturePlot(), DotPlot(), ImageFeaturePlot(). For dotplots, the dot size corresponds to the percentage of cells in the group with non-zero expression of the specific gene, while the dot color represents average expression level. The average expression corresponds to the average z-scored expression (relative to the gene’s mean across all cells), where positive values indicate higher-than normal average expression, while negative values indicate lower-than-average expression. For boxplot visualizations pseudobulked data was plotted, where the boxes represent the interquartile range, the center line denoting the median, and whiskers extending up to 1.5× the interquartile range. In order to perform unbiased cell type annotation to validate cell types in our integrated scST dataset, we used SingleR(v2.8),45 a computational tool that leverages the reference transcriptomic dataset (scRNAseq) and predicts the cell of origin for each cell independent. The CellestialCortex (https://developingbrainatlas-cellestialcortex.com/) web portal was developed using in-house scripts using an RShiny environment in R. The shinyapp uses functions from the Seurat package: DimPlot(), FeaturePlot(), ImageDimPlot(), ImageFeaturePlot(). Single-cell sequencing data analysis Single-cell sequencing analysis was performed on mouse embryonic cortical data across four developmental stages, E12.5, E15.5, E18.5 and P1, using a publicly available dataset.1 Computational analysis was performed in R using Seurat v4. Data from all developmental stages was loaded in using Read10X_h5(use.names = TRUE, unique.features = TRUE) and Seurat objects were created using CreateSeuratObject(min.cells = 3, min.features = 200). For quality control, percent.mt was filtered and minimum cut-offs were identified for each developmental dataset. For each dataset, raw counts were normalized with NormalizeData(), and variable genes were identified with FindVariableFeatures (selection.method = “vst”, nfeatures = 2000). Shared integration features were selected using SelectIntegrationFeatures(). Each dataset was scaled ScaleData() and reduced by PCA using RunPCA(). Integration anchors were then identified using FindIntegrationAnchors(reduction.method = “rpca”). Datasets were combined with IntegrateData() into a single integrated object, with the integrated dataset was scaled using ScaleData() and reduced using RunPCA(). UMAP embedding was generated using RunUMAP(dims = 1:30), a nearest-neighbor graph was generated using FindNeighbors(dims = 1:30), and clustering was performed FindClusters(resolution = 1.0). To assess potential sex-related differences within the dataset, cells were classified as Xist+ if they exhibited Xist expression greater than 0, and Xist- otherwise. Cluster-specific marker genes were identified using FindAllMarkers(only.pos = TRUE, min.pct = 0.25, logfc.threshold = 0.25). By default, this function performs a Wilcoxon rank-sum test to compare expression levels of each gene between the cluster of interest and all other cells. The resulting p-values were adjusted for multiple testing using the Bonferroni correction (pADJ). Further, we filtered for upregulated genes with a minimum absolute log FC of 0.25, pct.1 of 0.25 and pct.2 of 0.3. Following, we ranked within-cluster expression and significant marker genes were defined as genes with an adjusted p-value <0.05 and an absolute average log2 fold change >1, which retained 5126 unique genes from a total of 18,274 genes. Gene expression was visualized by plotting raw RNA counts using FeaturePlots() or boxplots. The number and relationships of DEGs were visualized using an upset plot. Quantification and statistical analysis The following statistical criteria were applied, unless otherwise specified. Box-and-whisker plots show the distribution of gene expression within specific cell types, with the midline representing the median, the boxes representing the first and third quartiles, the whiskers extending 1.5× the interquartile range, with any points beyond this range considered outliers and not visualized. For differential gene expression analysis of scRNA-seq data, we used a Wilcoxon rank-sum test to compare gene expression between the cluster of interest and all other cells. Resulting p-values were adjusted for multiple testing using the Bonferroni correction. For pseudobulked DEG analysis on single-cell spatial transcriptomics data, we applied DESeq2,46 with pADJ obtained using the Bonferroni correction. For statistical comparison of expression levels of selected genes across cell types or subtypes, we performed the Kruskal–Wallis rank-sum test followed by Dunn’s post hoc test with Benjamini–Hochberg correction. Statistical significance for adjusted pADJ-values is specified as such: ns: pADJ > 0.05; ∗: pADJ < 0.05, ∗∗: pADJ < 0.01, ∗∗∗: pADJ < 0.001.