Radial glia require PDGFD/PDGFRß signaling in human but not mouse neocortex

RG are the physical substrate⁸ and progenitor population that underlie production of most cells in human neocortex². We sought to determine a general transcriptional “signature” of human neocortical RG (hRG) as a starting point for identifying genes that may regulate uniquely human aspects of cortical development. We and others have previously shown that gene coexpression analysis of heterogeneous tissue samples can deconvolve transcriptional signatures of distinct cell types without cell isolation or purification^{9, 10}. Because prenatal samples of human neocortex are scarce, we developed a novel strategy called Gene Coexpression Analysis of Serial Sections (GCASS) that exploits variation in cellular abundance across serial sections of a single tissue sample to reveal cell type-specific patterns of gene expression (Fig. 1a-c; Extended Data Fig. 1; see Supplementary Information for methods, rationale, and further discussion). We applied GCASS to 87 150μm sections of a single human cortical sample from gestational week 14.5 (GW14.5, corresponding to peak layer V neurogenesis¹¹; Supplementary Table 1) and identified 55 modules of coexpressed genes. Six modules overlapped significantly with a set of genes we determined were expressed significantly higher in FACS-sorted mouse RG (mRG) vs. intermediate progenitor cells (“FACS-mRG”: Extended Data Fig. 1; Supplementary Table 2), suggesting that they might represent transcriptional signatures of hRG (Fig. 1d). Analysis of laser-microdissected samples from three independent transcriptomic datasets^{12, 13} confirmed that genes in these modules are most highly expressed in the ventricular zone (VZ) and subventricular zone (SVZ) of developing human neocortex, where both ventricular (vRG) and outer subventricular (oRG) subtypes of RG reside⁴ (Extended Data Fig. 2).

To produce a consensus transcriptional signature for GW14.5 hRG, we first summarized each of these six modules by its first principal component/module eigengene^{14, 15} (ME) and calculated the WGCNA¹⁶ measure of intramodular gene connectivity, kME^{10, 14} (concept: Fig. 1c). kME quantifies the extent to which a gene conforms to the characteristic expression pattern of a module and can predict gene expression specificity for individual cell types¹⁰. kME values for the six modules were combined into a single measure (Z.hRG), with higher scores predicting greater expression specificity for hRG (Fig. 1e). Genes with high Z.hRG included markers of neocortical RG such as SLC1A3 (GLAST1), VIM, SOX2, NOTCH1, and PAX6 (Fig. 1e: blue lines). Genes with low Z.hRG included markers of committed neuronal lineages such as TBR1, FEZF2, RELN, and SATB2 (Fig. 1e: black lines). We performed in situ hybridization (ISH) and immunostaining on independent prenatal human neocortical samples for genes with high Z.hRG that have not, to the best of our knowledge, previously been implicated in RG biology (Fig. 1e: red lines; Extended Data Fig. 3). In all cases, expression of these genes was restricted to the VZ/SVZ (Fig. 1f; Extended Data Fig. 3). These results indicate that GCASS can discern a general transcriptional signature of hRG from a single, heterogeneous tissue sample without cell labeling, isolation, or purification. Moreover, because the sample derives from a single individual, this strategy implicitly controls for genotype and developmental stage and has broad implications for the molecular analysis of rare tissue samples.

To establish the robustness of the hRG transcriptional signature, we analyzed four additional prenatal human cortex gene expression datasets^{12, 13, 17} that were generated with diverse sampling strategies and technology platforms (Extended Data Table 1). In parallel, we also analyzed three embryonic mouse cortex gene expression datasets^{12, 18, 19} (largely E14-14.5, corresponding to peak layer V neurogenesis¹¹; Extended Data Table 1) to establish a robust mRG transcriptional signature. For each dataset, we constructed an unsupervised coexpression network, identified the module with the most significant overlap with the FACS-mRG gene set, and calculated kME values (RG.kME) for every gene with respect to this RG module (Supplementary Table 3). To facilitate comparisons of RG transcriptional signatures, we mapped all probe sets/transcripts to a common identifier (HomoloGene ID) and converted RG.kME values for each dataset into percentile ranks (RG.PR; workflow schematic: Extended Data Fig. 4). Genome-wide correlations of RG.PR among all datasets ranged from 0.36 – 0.95 (human) and 0.47 – 0.70 (mouse) (P < 2.2e-16; Extended Data Fig. 5), demonstrating the robustness of RG transcriptional signatures in both species.

To explore the global extent of gene expression conservation between hRG and mRG, we calculated the mean RG specificity for each gene across the five human ( $\overline{\text{hRG}.\text{PR}}$), three mouse ( $\overline{\text{mRG}.\text{PR}}$), and all eight ( $\overline{\text{RG}.\text{PR}}$) datasets (Supplementary Table 3). For genes present in at least one dataset in each species (n = 15,576), the correlation between $\overline{\text{hRG}.\text{PR}}$ and $\overline{\text{mRG}.\text{PR}}$ was 0.48 (P < 2.2e-16; Fig. 2a), indicating broad conservation of transcriptional programs that are active during cortical development in humans and mice. Genes with $\overline{\text{hRG}.\text{PR}}$ and $\overline{\text{mRG}.\text{PR}}$ > 80 (Fig. 2a, +RG box) or < 20 (Fig. 2a, -RG box) were significantly enriched with sets of genes expressed by RG or neurons, respectively (Supplementary Table 4). Furthermore, the most conserved set of core RG genes ( $\overline{\text{hRG}.\text{PR}}$ and $\overline{\text{mRG}.\text{PR}}$ > 95: Fig. 2a, green box) included the canonical RG markers VIM and PAX6, and SOX family members implicated in nervous system development such as Sox2, SOX3, SOX9, and SOX21 (Fig. 2b). Also present among this core set were elements of the Notch (NOTCH2 and HES1) and Wnt-βcatenin (SFRP1, FZD8, and LRP4) signaling pathways, which critically regulate neurogenesis². In addition, we observed a cohesive subgroup of genes involved in cell cycle regulation, including CKS2, RACGAP1, MELK, CCNB1, CCNA2, ASPM, and MKI67. Finally, several genes in this group have not, to the best of our knowledge, previously been implicated in RG biology, including PSAT1, DDAH1, AIF1L, PSRC1, and ACSS1. Together, these results provide a framework for identifying conserved and distinct aspects of gene expression in hRG and mRG.

To identify evolutionary changes in RG gene expression that might underlie neurodevelopmental differences between human and mouse, we assessed homologous genes based on predicted differences in RG expression specificity between the species (differential RG specificity: $\text{DS}=\overline{\text{hRG}.\text{PR}}-\overline{\text{mRG}.\text{PR}}$). Because such differences could emerge from gene expression changes in RG or neighboring cell types (Fig. 2c), we also compared the relative expression levels of homologous genes between the species (differential expression: $\text{DE}=\overline{\mathrm{h}.\text{Expr}.\text{PR}}-\overline{\mathrm{m}.\text{Expr}.\text{PR}}$). Genome-wide analysis of DS/DE revealed four distinct quadrants of genes that differed substantially between human and mouse (Fig. 2d). Consistent with the proposed model (Fig. 2c), these quadrants were enriched with sets of genes expressed in distinct patterns in developing human and mouse cortex (Supplementary Table 4).

We reasoned that quadrant 1 might contain genes with “Boolean” expression differences (i.e. ON in hRG and OFF in mRG) required for human but not mouse neocortical development. We applied stringent criteria to identify and validate 18 candidate genes in this quadrant with strong and consistent evidence of expression in hRG and no evidence of expression in mRG: ABHD3, ASAP3, BMP7, C5, C8orf4, FAM107A, FOXN4, ITGA2, LRIG3, LRRC17, PAM, PDGFD, PDLIM3, RFTN2, SLC2A10, SP110, STOX1, and ZC3HAV1 (Fig. 2d, cyan circles; Extended Data Fig. 6). Because secreted growth factors could alter the size of developing neocortex by influencing proliferation, we focused on PDGFD (Fig. 2d', green square), which has not, to the best our knowledge, previously been implicated in cortical development of any species. Compared to genes with the highest $\overline{\text{RG}.\text{PR}}$ (conserved in both species), PDGFD expression was highly correlated in human, but not mouse (Fig. 3a-b). ISH confirmed that PDGFD was expressed by RG throughout the VZ of GW14.5 human neocortex (Fig. 3c). In contrast, Pdgfd was not detected in RG (or any cell type) in E15.5 mouse neocortex (Fig. 3d). These expression differences were consistent for multiple ages in human (GW14.5-GW18.2) and mouse (E14.5-E17.5) prenatal cortex (Extended Data Fig. 7).

The effects of PDGFD are specifically mediated by the PDGFRß receptor, which upon phosphorylation can trigger signaling pathways that promote cell proliferation^{6, 7, 20}. Although PDGFRβ did not meet the same stringent criteria as PDGFD, its location on the DS/DE plot was proximal to PDGFD (Fig. 2d', red square), also suggesting species differences in PDGFRβ expression. Compared to genes with the highest $\overline{\text{RG}.\text{PR}}$, PDGFRβ expression was moderately correlated in human, but not mouse (Fig. 3e-f). Immunostaining for PDGFRß in GW14.5 human brain revealed strong expression throughout the telencephalic germinal zones (VZ/SVZ) and in vascular pericytes, with highest levels in dorsolateral cortical progenitors and the lateral ganglionic eminence (LGE) (Fig. 3g). In contrast, immunostaining for PDGFRß in E15.5 mouse brain revealed expression in LGE progenitors but no evidence of expression in cortical progenitors (Fig. 3h). However, we did observe very low levels of Pdgfrβ transcript in the VZ of lateral mouse cortex²¹ (Extended Data Fig. 8), leaving open the possibility of modest, region-specific function. Collectively, these results indicate that expression patterns of PDGFD and PDGFRß in developing neocortex have diverged considerably during human and mouse evolution, despite retaining amino-acid sequences that are ∼85% identical between the species.

We tested the requirement of PDGFD/PDGFRß signaling for hRG proliferation in GW17.5 human neocortical slice cultures, screening four chemical inhibitors of PDGFRß signaling (Sutent, Tivozanib, Imatinib, and CP673451). Three out of four PDGFRß inhibitors reduced the percentage of SOX2+ progenitors (RG) that incorporated BrdU over two days in slice culture (Extended Data Fig. 9). For replication we focused on CP673451, which exhibits the greatest selectivity for PDGFRβ over other receptors²² and caused the greatest reduction in SOX2+BrdU+ cells among tested inhibitors (Extended Data Fig. 9). PDGFRß inhibition by CP673451 in GW17.5 human neocortical slice cultures reduced the number of RG and intermediate progenitors (IP) that incorporated BrdU by >50%, affecting progenitors in the VZ and SVZ (Fig. 4a-b). The percentage of progenitor cells co-staining with cleaved caspase-3, an apoptosis marker, was slightly elevated by CP673451, but sufficiently low to attribute reduced BrdU incorporation to cell cycle dysregulation rather than cell death (Fig. 4b). Furthermore, CP673451 treatment of E13.5 mouse cortical slice cultures did not decrease BrdU incorporation or the cycling proportion (Ki67+) of RG or IP populations over multiple time points (Fig. 4b, Extended Data Fig. 9). These results indicate that PDGFRß signaling is required for hRG but not mRG to progress through the cell cycle and expand at a normal rate.

We next investigated whether introducing PDGFD into embryonic mouse cortex, where it is normally absent, could promote mRG proliferation. We injected recombinant-PDGF-DD protein into the lateral ventricles of E13.5 mouse embryos, bypassing the need for PDGFD to be generated and dimerized in vivo, and analyzed the number and spatial distribution of SOX2+ progenitors at E15.5 in lateral cortex. Relative to vehicle, PDGF-DD increased the proportion of RG (SOX2+/DAPI+) in lateral cortex by ∼10% (Fig. 4c-d). Furthermore, PDGF-DD induced a modest subventricular shift in the distribution of mRG in the developing cortical wall (Fig. 4d). These effects were not observed in dorsomedial cortex, where PDGFRβ was not detected in RG (Fig. 4c-d, Extended Data Fig. 8).

To test if mRG can respond to PDGFRβ activation, we ectopically expressed two forms of constitutively active PDGFRβ (PDGFRβ:D850V [homologous to human D849V²³] or the TEL-PDGFRβ²⁴ fusion protein) in mRG by in utero electroporation at E13.5. By E15.5, expression of PDGFRβ:D850V approximately doubled the proportion of SOX2+ or Ki67+ progenitors among electroporated (GFP+) cells and dramatically dispersed SOX2+ progenitors in the basal direction (Fig 4e-f; Extended Data Fig. 9). Similar but less dramatic effects were observed following TEL-PDGFRβ electroporation (Extended Data Fig. 9). These results indicate that PDGFRβ signaling in mRG functions analogously to its known role as an oncogenic pathway that promotes proliferation and epithelial-mesenchymal transition²⁰. We therefore propose that physiological levels of PDGFD/PDGFRβ signaling in hRG may contribute to the proliferation and subventricular dispersion of neural progenitors that characterizes OSVZ formation^{4, 5}.

Mouse studies have demonstrated that the size and shape of cerebral cortex depend on precise regulation of molecular pathways controlling RG proliferation and differentiation^{1, 2}. While many of these pathways are likely to be conserved in humans, mouse studies alone cannot reveal uniquely human aspects of cortical development^25-30. By analyzing human tissue as a starting point, we found that PDGFD/PDGFRß signaling is required for normal RG proliferation in developing human but not mouse neocortex, and is sufficient to promote some “humanizing” characteristics in mouse. Our analysis has also identified other genes that likely contribute to differences between human and mouse cortical development. BMP7, another secreted growth factor, was predicted and validated to be expressed by hRG but not mRG (Fig. 2d, Extended Data Fig. 6), raising the intriguing possibility that local production of growth factors by hRG may be necessary to support the expanded germinal region and progenitor heterogeneity of developing human neocortex. We anticipate that the analytical and experimental strategies described here will help determine the extent to which these and other pathways are shared among primates or uniquely required for human cortical development.