Developmental Origins of Human Cortical Oligodendrocytes and Astrocytes

Human cortical radial glial cells are primary neural stem cells that give rise to cortical glutaminergic projection pyramidal neurons, glial cells (oligodendrocytes and astrocytes) and olfactory bulb GABAergic interneurons. One of prominent features of the human cortex is enriched with glial cells, but there are major gaps in understanding how these glial cells are generated. Herein, by integrating analysis of published human cortical single-cell RNA-Seq datasets with our immunohistochemistical analyses, we show that around gestational week 18, EGFR-expressing human cortical truncated radial glial cells (tRGs) give rise to basal multipotent intermediate progenitors (bMIPCs) that express EGFR, ASCL1, OLIG2 and OLIG1. These bMIPCs undergo several rounds of mitosis and generate cortical oligodendrocytes, astrocytes and olfactory bulb interneurons. We also characterized molecular features of the cortical tRG. Integration of our findings suggests a general picture of the lineage progression of cortical radial glial cells, a fundamental process of the developing human cerebral cortex. Supplementary Information The online version contains supplementary material available at 10.1007/s12264-021-00759-9.

During human cortical development, three types of RGCs have been identified. At the onset on PyN genesis around gestational week 7 (GW7), human cortical neuroepithelial cells transform into the ventricular radial glial cells (vRGs, also referred to as apical RGs). Like mouse cortical RGCs, human cortical vRGs have apical-basal polarity, contacting both the ventricle and the pia [4,[21][22][23][24][25][26]. Around GW15-GW16, vRGs begin to generate outer radial glia cells (oRGs) and truncated radial glial cells (tRGs); as a result the radial glia scaffold becomes discontinuous [27,28]. oRGs (also known as basal RGs) no longer contact the ventricle but they directly inherit the basal process from their vRG mother cells, thereby maintaining contact the pial surface. On the other hand, tRGs no longer contact the pia, and their basal processes mainly terminate in the outer SVZ (OSVZ) [27]; they inherit the apical domain from vRGs, which contact the lateral ventricle throughout the development. vRGs mainly generate deep layer PyNs, whereas oRGs mainly generate upper layer PyNs [21,26,27,29,30]. The progeny of tRGs is unknown.
A prominent feature of human brain is its relative enrichment with macroglial cells (oligodendrocytes and astrocytes), which make up at least 50% of the cells in the human cortex (approximately 20% more than in mouse cortex) [31][32][33][34]. However, the developmental origins of cortical glial cells and cortex-derived OBiNs in the human brain remain largely unknown. Here we explore whether recent findings from the mouse [6] shed light on this subject. We re-analyzed published single-cell RNA-Seq datasets, including 3,355 EGFR ? cells from human frontal cortex at GW21-GW26 [35] and 33,976 cells from human cortex at GW17-GW18 [36], in the context of our immunohistochemistry results, and provide evidence that human cortical EGFR ? tRGs generate bMIPCs that coexpress EGFR, ASCL1, OLIG2 and OLIG1. These MIPCs in turn give rise to cortical oligodendrocytes, astrocytes and OBiNs. Thus, we conclude that developmental origins of human cortical glial cells are similar to that in the mouse cortex. Further, we present a general model of RGC lineage progression and molecular identity of tRGs in the human cerebral cortex.

Human Tissue Collection
Three fetal brain samples (GW18, n = 1, male; GW23, n = 2, male and female) were collected with informed consent and in accordance with the Fudan University Shanghai Medical College guidelines, and the study design was approved by the institutional review board (ethics committee) of the Fudan University Shanghai Medical College (20110307-085 and 20120302-099). Fetal brains were obtained at autopsy within 3-5 h of spontaneous abortion, fixed in 4% PFA at 4°C for 3-7 days and then cryoprotected in 30% sucrose in 0.1 mol/L phosphate buffer for 72 h at 4°C. The brain tissue samples were frozen in in embedding medium (OCT; Sakura Finetek) on a dry ice and ethanol slush. All of these 3 fetal brain samples had been used in previous study [37,38].

Immunohistochemistry
Immunohistochemistry was performed using standard protocols [6,37]. Immunohistochemical staining was performed on 60 lm thick fetal brain coronal sections which were mounted on glass slides. E17.5 and E18.5 mouse brains were cryosectioned into thickness of 12 lm. In the present study, all human brain sections were subjected to an antigen retrieval protocol (slides were briefly boiled in 10 mmol/L sodium citrate, pH 6.0). Sections were then permeabilized with 0.05% Triton X-100 for 1 h, followed by an incubation in blocking buffer (5% donkey serum and 0.05% Triton X-100 in TBS) for 2 h. The blocking buffer was removed, and the sections were incubated with primary antibodies (diluted in the blocking buffer) for 24-48 h at 4°C. The sections were washed in TBS, and incubated with secondary antibodies conjugated to Alexa Fluor488, Cy3 or Cy5 for 2 h at room temperature. All secondary antibodies were from Jackson ImmunoResearch. Finally, the sections were counterstained with DAPI for 2 mins before being mounted in the fluorescence mounting medium (DAKO S3023).

scRNA-Seq Datasets Analysis
Strict quality control of cells is required for scRNA-Seq analysis, which has been described in our previous study [6]. For analysis of human scRNA-Seq data from different stages, the pre-processing raw sequencing data from 3,355 human frontal cortical EGFR ? cells at GW21-26 (GSE144462) [35] were analyzed by Seurat 3.2.2.9007. Preprocessed data was normalized by log-normalization using a scale-factor of 10,000. The top 2,000 variable genes were identified by the variance stabilizing transformation method, and subsequently scaled and centered. We calculated cell cycle phase scores based on canonical markers [40], and regressed these scores during pre-processing.
Principal components analysis was performed for dimensional examination using the 'elbow' method. The first 20 dimensions showed the majority of data variability. Therefore, UMAP dimensional reduction was performed on the first 20 dimensions. For cluster annotation, the most comprehensive and reliable cell type markers were sought through extensive literature review [6]. Differentially expressed genes (DEGs) of compatible clusters were identified using following parameters: fold-change-threshold of log (0.25) and with min.pct = 0.25.

Image Acquisition and Cell Counting
Immunohistochemistry staining images were acquired using Olympus VS120 Automated Slide Scanner. Images for quantitative analyses were acquired using Olympus FV1000 confocal microscope system. Cell counting was performed on Z-stack confocal images. Briefly, the confocal images were imported into Photoshop CC (Adobe Systems Incorporated). We divided the human dorsal and dorsal-lateral neocortex into different number of bins with equal size (250 lm length and 125 lm width in 60 lm thick sections) spanning the cortical VZ, ISVZ (inner subventricular zone), IFL (inner fiber layer) [41], OSVZ [42], IZ (intermediate zone), SP (subplate), CP (cortical plate) and MZ (marginal zone). Cortical layers were delineated by DAPI staining. Labeled cells within each bin were manually counted by using a cell-counter and switching between the red, green, and blue channels in Photoshop. Three representative images of the dorsal or dorsal-lateral neocortex from randomly sampled sections were counted at each age (GW18 and GW23).

Statistics
Numbers of cells counted per bin in the human dorsal or dorsal lateral cortical VZ, ISVZ, IFL, OSVZ, IZ, SP and CP were shown as mean ± SEM. There were no statistical comparisons performed on cell counts in the present study.

scRNA-Seq Analysis Reveals Human Cortical tRG Lineage at GW21-GW26
Our recent study has demonstrated that the presence of EGFR ? progenitors in the E16.5 mouse cortex is a strong signal for the onset of cortical gliogenesis (and OBiN genesis) [6]. To investigate whether a similar process may be occurring in humans, we studied the molecular identity of human cortical EGFR ? cells using published scRNA-Seq data. First, we analyzed data from 3,355 EGFR-  immunopanned cells obtained from GW21-GW26 human frontal cortex (Fig. 1A) [35]. 13 clusters were recognized (C0-C12) using t-distributed stochastic neighbor embedding (tSNE) and Seurat clustering [43] (Fig. 1B). We identified the identity of the cell types in these clusters based on gene expression. To our surprise, we found that cluster 6 contained cortical tRGs (see below). Furthermore, we obtained evidence that these tRGs (cluster 6) may be lineally related to the PyN-IPC clusters (C4, C2 and C0, 1,531 cells, 46% of EGFR ? cells) and bMIPC clusters (C10, C3 and C1) (Fig. 1B), as they all expressed EGFR and ASCL1 (Fig. 1G). Other major cell types were APCs, OPCs and OBiN-IPCs. Consistent with previous analysis, immature ependymal cells (EP), pericytes and endothelial (Endo) cells also expressed EGFR (Fig. 1B), but the authors did not recognize EGFR expression in the cortical tRG [35]. Below, we focus on the molecular features of the tRG, PyN-IPC, bMIPC, APC, OPC and OBiN-IPC cell types.
Our genetic fate mapping study demonstrated that mouse cortical Egfr ? Ascl1 ? Olig2 ? Olig1 ? bMIPC population give rise to most of the OPCs and APCs in the mouse cortex and a subpopulation of OBiN-IPCs [6]. Thus, our reanalysis of scRNA-Seq from human cortical EGFR ? cells provides evidence for that mouse and human share the same developmental origins of cortical oligodendrocytes and astrocytes, and cortex-derived OBiNs.
Here, we propose that vRGs might directly give rise to oRGs and tRGs simultaneously at the end of their final mitosis because vRGs, tRGs and oRGs expressed many shared glial genes, such as HOPX, FAM107A, LIFR, ALDOC, AQP4, BCAN, GFAP and TNC (Fig. 2B, C). Perhaps because tRGs contact the lateral ventricle, they may gradually acquire their unique properties, including expression of EGFR, ASCL1, ANXA1, CRYAB and CXCL12 (Fig. 2D); these genes were not expressed in vRGs or oRGs. On the other hand, oRGs, contacting the pial surface, expressed ACSBG1 and FBN2, which were not Fig. 2 vRGs, tRGs and oRGs coexist in the human cortex at GW17-GW18. A scRNA-Seq analysis of 33,976 cortical cells from the midgestation human brain (GW17-GW18) [36]. B, C Expression patterns of PDGFD (vRG and tRG marker), HOPX, FAM107A, LIFR, ALDOC, AQP4, BCAN, GFAP and TNC indicated the coexistence of cortical vRGs, tRGs and oRGs. D EGFR, ASCL1, ANXA1, CRYAB and CXCL12 were expressed in tRGs but not in oRGs or vRGs. Note that EGFR was weakly expressed in a subpopulation of PyN-IPCs, and ASCL1 was expressed (indicated by arrows) in a subpopulation of PyN-IPCs, maturing PyNs, and MGE-and CGE-derived cortical interneurons, and MGE-derived OPCs. E ACSBG1 and FBN2 were expressed in oRGs but not in tRGs. F Wide expression of PyN-IPC marker genes (PAX6, NEUROG2 and EOMES) indicating activated PyN genesis. G Except MGE-derived OPCs, cortexderived OPC or APC clusters were not identified, suggesting that gliogenesis had not commenced yet in the human cortex at GW17-GW18. expressed in tRGs (Fig. 2E). In the cortex at GW17-GW18, IPCs mainly consisted of PyN-IPCs that expressed PAX6, NEUROG2 and EOMES (Fig. 2D, F), suggesting that PyN genesis is the primary cellular product. Indeed, except for the medial ganglionic eminence (MGE)-derived OPCs (Fig. 2G), we did not observe cortex-derived OPC or APC clusters, suggesting that gliogenesis had not yet commenced at GW17-GW18. Notably, while EGFR was expressed in tRGs and a subpopulation of PyN-IPCs (progeny of tRG), EGFR expression was absent in the OPCs and cortical interneurons that were derived from the MGE and caudal ganglionic eminence (CGE) (Fig. 2D). This demonstrated that EGFR ? cells in the cortex were derived from the cortex itself. ASCL1, in contrast, was continuously expressed in a subpopulation of immature PyNs, MGE-and CGE-derived cortical interneurons, and MGE-derived OPCs in the cortex (arrows in Fig. 2D). Consistent with previous report [36], we did not find IPC clusters in the human cortex for generating cortical interneurons, further providing evidence that most if not all human cortical interneurons are derived from the ventral telencephalon [29,37].
Taken together, by a re-analysis of published scRNA-Seq datasets from human fetal brains at GW17-GW18 and GW21-GW26, we have found that: 1) EGFR is expressed in tRGs but not in vRGs or oRGs; 2) EGFR ? tRGs appear to generate EGFR ? PyN-IPCs and EGFR ? bMIPCs; these bMIPCs then give rise to cortical OPCs, APCs, and OBiN-IPCs; 3) HOPX, FAM107A, TNC and LIFR are expressed in tRGs, in addition to vRGs and oRGs; 4) tRGs express numerous hallmarks of cells in the astrocyte lineage.

Immunohistochemical Identification of tRGs and oRGs in the Human Developing Cerebral Cortex
To validate the molecular signatures of tRG and oRG identified by the scRNA-Seq analysis, we examined expression of cell type specific marker proteins in fixed cortical sections at GW18 (n = 1) and GW23 (n = 2) (Fig. 1A). HOPX and GFAP were strongly expressed in somas and/or basal processes of oRGs in the GW18 cortex (Fig. 3A, D, E). At GW23, we observed similar expression patterns of these proteins in the cortex (Fig. S3A-D). On the other hand, double-or triple-immunofluorescence analysis of GW18 cortical sections showed that EGFR, CRYAB, HOPX and GFAP were expressed in tRGs ( Fig. 3A-C, E). Consistent with previous observations [27], CRYAB ? tRG basal processes terminated in the cortical OSVZ, and thus did not contact the pia (Fig. 3C, Fig. S3A).
CRYAB is a marker for tRGs at GW18 as it is mainly expressed in tRGs. However, at GW23, we observed CRYAB ? astrocyte lineage cells that also expressed HOPX and GFAP in the cortical IFL and MZ (Fig. S3B, C) suggesting that CRYAB is more widely expressed in cortical cells at this later fetal stage (Fig. S2C). Taken together, HOPX is expressed in tRGs, in addition to expression in vRGs and oRGs, consistent with our scRNA-Seq analysis (Fig. S1C). This is also consistent with the expression pattern of HOPX in the macaque monkey cortical VZ and OSVZ at E70 and E125 [52] and in the ferret cortical VZ and OSVZ at E36 [53].

A Proposed Model of the Generation of tRGs and oRGs from vRGs
It has been proposed that human cortical oRGs are generated from vRGs by a process that resembles epithelial-mesenchymal transitions, because oRGs express genes that promote extracellular matrix production, such as TNC, ITGB5, PTN and PTPRZ1 [27,44]. However, these genes were also expressed in tRGs; furthermore, TNC, PTN and PTPRZ1 were also expressed in vRGs (Fig. 2C, Fig. S1) [36,44]. Thus, our scRNA-Seq analysis demonstrated that oRGs expressed most of the genes that were also expressed in tRGs (Fig. 1, Figs. S1, S2).
HOPX expression marked multiple progenitor subtypes at GW18. Immunohistochemistry of fixed cortical sections showed HOPX expression in tRGs in the VZ, oRGs in the OSVZ, and in cells in the ISVZ and IFL with long basal processes that appear to be oRGs migrating into the OSVZ (Fig. 4A, B). Previous study using time-lapse imaging have observed that oRGs and tRGs emerge as the daughter cells of horizontally dividing vRGs [54]. Thus, based on HOPX immunostaining results (Fig. 4A, B), combined with scRNA-Seq data (Fig. 1, 2; Figs. S1, S2) and time-lapse imaging analysis [54], we propose that the final mitosis of a HOPX ? vRG concomitantly generates a HOPX ? oRG and a HOPX ? tRG (Fig. 4C). oRGs inherit the long basal fiber of vRGs while tRGs inherit the apical domain of vRGs. Like cortical vRGs, both oRGs and tRGs can self-renew (Fig. 4C).
By GW23, the density of EGFR ? cells and HOPX ? cells were greatly increased in the cortex compared with GW18 ( Figs. S4A and 5A). Again, tRGs expressed EGFR and oRGs did not (Fig. S4B). EGFR ? HOPX ? APCs were widely distributed from the cortical ISVZ to SP, but EGFR was not expressed in most of the HOPX ? cells in the cortical CP (Fig. S4B), indicating that EGFR expression is downregulated in HOPX ? immature astrocytes. GFAP and HOPX double immunostaining revealed that nearly all HOPX ? GW23 cortical cells were GFAP ? tRGs, oRGs and astrocyte lineage cells (Fig. 6A-C). The high density of To further confirm that HOPX expression marks astrocyte lineage cells, we performed HOPX, SOX10 and OLIG2 triple immunostaining on GW18 cortical sections (Fig. S5A). All SOX10 ? cells (OPCs) in the cortex expressed OLIG2, but none of them expressed HOPX (Fig. S5B-F), confirming that HOPX is not expressed in OPCs. On the other hand, At GW18, most HOPX ? APCs expressed OLIG2 (Fig. S5B-F), consistent with the scRNA-Seq data (Fig. 1G). The ratio of OPCs (SOX10 ?-OLIG2 ? cells) to APCs (HOPX ? OLIG2 ? cells) was 2:1 (Fig. S5G), demonstrating that there were more OPCs than APCs in the GW18 cortex. We propose that in the GW18 cortex, most OPCs were derived from the ventral telencephalon [55], whereas most APCs were derived from tRGs, because both APCs and tRGs expressed EGFR and HOPX, and because only a small number of cortical bMIPCs were observed at this stage (Figs. 7, 8, 9). Thus, the earliest macroglial cells in the human cortex are MGEderived OPCs, followed by cortical tRG-derived EGFR ?-HOPX ? APCs.
We also observed that there were more GFAP ? HOPX ? APCs in the ventral cortex than in the dorsal cortex at GW18 (Fig. S6A-C). Thus, in both mouse and human brains, PyNs and glial cells are first generated in the ventral cortex, followed by dorsal and medial cortex [6], suggesting this is a common rule in mammalian brain. Taken together, our immunohistochemistry results demonstrated that in addition to its expression in vRGs, tRGs and oRGs, HOPX marks astrocyte, but not oligodendrocyte, lineage cells in the human cortex.
Some EGFR ? EOMES ? cells had an apical process extended to the ventricular surface (Fig. 7C, G), exhibiting morphologies of cortical SNPs. At GW18 and GW23, EGFR ? EOMES ? PyN-IPCs were mainly in the cortical ISVZ, whereas EGFR ? OLIG2 ? bMIPCs were mainly in the IFL (Fig. 7D, E). At GW18, only a few EGFR ?-OLIG2 ? bMIPCs were observed in the cortical IFL (Fig. 7C, D), suggesting that at this stage tRGs mainly produce PyN-IPCs. Thus, there were two germinal zones (niches) for PyN genesis in the GW18 cortex: one in the VZ and the other in the OSVZ. By GW23, there was a high density of EGFR ? OLIG2 ? bMIPCs in the IFL (Fig. 7E, G), a high density of astrocyte lineage cells (Fig. 6) and a high density of OPCs (see below, Fig. S10) in the cortex, indicating a high level of gliogenesis. Thus, at GW23 there were two germinal zones: one was the VZ for tRG PyN genesis and gliogenesis, and the other was the OSVZ mainly for PyN genesis from oRGs (see below, Fig. S12, S13). We propose that tRGs and oRGs both generate upper layer PyNs (Fig. 7H, see discussion).

bMIPCs Are Mainly Distributed in the Human Cortical IFL at GW18-GW23
To further confirm bMIPCs were mainly distributed in the human cortical IFL, we examined cell types in the cortical ISVZ. Immunostaining showed that there were many EOMES ? IPCs in the cortical ISVZ (Fig. 8A), but there were many more cortical interneurons, forming a migratory stream, in the cortical ISVZ at GW18 (Fig. 8B). These cortical interneurons expressed GABA, NR2F2 (COUP-TFII) and SP8 (Fig. 8B) suggesting that they are mainly CGE-derived cortical interneurons [29,37]. Again, very few EGFR ? ASCL1 ? OLIG2 ? bMIPCs were in the ISVZ; they were mainly in the IFL (Fig. 8C, D). In the mouse cortex, chemokine CXCL12 is mainly expressed in PyN-IPCs [56,57], whereas in the human cortex, our scRNA-Seq analysis showed that CXCL12 is mainly expressed in tRGs (Fig. S2C). CXCR4, a CXCL12 receptor, is expressed in nearly all human cortical interneurons [36]. Because CXCL12 is mainly expressed by tRGs in the human cortical VZ, but not by oRGs and their progeny in the OSVZ, cortical ISVZ becomes the main corridor for tangentially migrating cortical interneurons. This may be a key reason for the distribution of bMIPCs mainly in the cortical IFL at GW18-GW23 (Fig. 8D).

Human Cortical oRGs Do not Generate OPCs
Two recent studies suggested that cortical oRGs were an additional source of OPCs in the developing human [49] and macaque monkey [52] cortex. The authors found some HOPX ? cells expressing EGFR in the cortical OSVZ, therefore suggesting that oRGs generated cortical Pre-OPCs and OPCs [49,52]. In the human cerebral cortex, Huang et al., also found a small number of HOPX ? cells expressing PDGFRA, further proposed that oRGs  of EGFR ? OLIG2 ? bMIPCs with a large soma in the cortical IFL were shown. Note that EOMES ? cells did not express OLIG2. Also note that EOMES was expressed in some EGFR ? SNP like cells (had an apical process extended to the ventricular surface, arrowheads in C and G). H Schematic of human cortical tRGs and their direct progeny. Initially, EGFR ? ASCL1 ? tRGs produce EGFR ? ASCL1 ?-EOMES ? PyN-IPCs, but later they produce bMIPCs that express higher levels of EGFR, ASCL1 and OLIG2. We propose that EGFR ? PyN-IPCs differentiate into upper layer PyNs. generated Pre-OPCs [49]. However, our results do not support these conclusions. We have identified that HOPX ?-EGFR ? cells were cortical tRGs and APCs, but not oRGs (Fig. 5, 6 and Fig. S4).
Next, we examined the expression of PDGFRA in the human cortex at GW18 and GW23. The vast majority of PDGFRA ? cells were OPCs in the cortex based on their co-expression with OLIG2 and SOX10 (Fig. S9A-F,  Fig. S10). In addition, we found that about 16% of PDGFRA ? cells between the GW18 cortical ISVZ and SP expressed HOPX (Fig. S9E, G); they were cortical APCs.
We also found that about 16% of PDGFRA ? cells in the GW18 cortical IFL and OSVZ expressed neither HOPX nor SOX10 (Fig. S9E, G); they were bMIPCs. These observations were consistent with the scRNA-Seq analysis results (Fig. 1H). Notably, in the GW23 human fetal brain, a large number of OPCs (Fig. S10) and APCs (Fig. 6) were already generated in the cortex, but oRGs were still making PyN-IPCs (see below, Fig. S12, S13), further suggesting that oRGs do not produce OPCs.
Thus, based on the expression patterns of EGFR, ASCL1, PDGFRA, OLIG2, HOPX and EOMES in the cortex, as well as the scRNA-Seq analysis results, we concluded that EGFR ? tRG-derived bMIPCs are the major source of cortical OPCs (and APCs) in the human brain. There are a small number of OPCs in the human cortex that are derived from the MGE, but we did not find any evidences supporting the claim that oRGs are an additional source of cortical OPCs.

PyN Genesis Continues at GW23
There were two germinal zones, VZ and OSVZ, in the GW18 and GW23 cortex. HOPX and EOMES double immunostaining revealed a large number of EOMES ? cells in both the cortical ISVZ and OSVZ at GW18; some of them expressed ASCL1 (Fig. S11A-D) [28,37]. In the GW23 cortex, a large number of EOMES ? cells in the OSVZ and ISVZ were observed (Fig. S12A-C), further confirming that the GW23 cortical OSVZ is still in the peak of PyN genesis, whereas cortical VZ is making both PyNs and glia. A recent study reported that the number of EOMES ? cells was reduced in the GW19-GW20 cortex and proposed that the production of PyNs ceases in the human fetal cortex after GW20 [63]. However, as indicated Fig. 9 bMIPCs generate cortical OPCs and APCs. A-F EGFR, ASCL1 and OLIG2 triple immunostained GW18 and GW23 human cortical sections. tRGs that expressed EGFR and ASCL1 in the cortical VZ were shown (arrows in C, F). bMIPCs that expressed EGFR, ASCL1 and OLIG2 in the cortical IFL were shown (arrows in B, E). In the cortical IZ, OPCs (arrowheads) expressed ASCL1 and OLIG2, but downregulated EGFR expression, and APCs (arrows) expressed EGFR and OLIG2, but downregulated ASCL1 expression (A, D). above, we observed a large number of EOMES ? cells in the GW23 cortex (Fig. S12); all of these EOMES ? cells expressed PAX6 (Fig. S13) and none of them expressed OLIG2 (Fig. 7G), suggesting that human cortical PyN genesis indeed occurs at GW23. Our observation was again supported by the scRNA-Seq data: among 3,355 tRGderived EGFR-immunopanned cells obtained from GW21-GW26 human frontal cortex, there were 1,531 cells belonged to PyN-IPCs (46% of EGFR ? cells) (Fig. 1A) [35]. Thus, during GW21-GW26, human cortical tRGs and oRGs are still making PyN-IPCs.

Discussion
Our study has 4 main findings: (1) The molecular identity of the human cortical tRG based on re-analysis of scRNA-Seq datasets; (2) A model of human cortical RGC lineage progression (Fig. 11); (3) Cortical EGFR ? tRGs first generate EGFR ? EOMES ? upper layer PyN-IPCs, and then generate EGFR ? ASCL1 ? OLIG1/2 ? bMIPCs which in turn give rise to most of the cortical OPCs and APCs, and cortex-derived OBiN-IPCs (Fig. 11). (4) The developmental origins of cortical glial cells (oligodendrocytes and astrocytes) and cortex-derived OBiNs in the human brain is evolutionary conserved based on this study and our previous one in mouse cortex using genetic lineage tracing (Fig. 11) [6].
In the human developing cortex, three types of cortical RGCs have been identified: vRGs, tRGs and oRGs [27,28,44]. At GW8-GW15, vRGs mainly generate deep layer PyNs. Around GW15-GW16, human cortical vRGs start to differentiate into oRGs and tRGs. We propose that the commencement of EGFR ? tRGs is a strong signal indicating onset of human cortical gliogenesis. Initially, these EGFR ? tRGs produce EGFR ? EOMES ? PyN-IPCs, and then they produce bMIPCs. Thus, onset of gliogenesis in the human cortex is also a sequence of steps, but one that takes place over a much longer period than in the mouse cortex. This one reason that we interpret that 46% of EGFR ? cells in the GW21-GW26 cortex (scRNA-Seq data) correspond to PyN-IPCs (Fig. 1B). EGFR and EOMES double immunostaining also revealed that most EOMES ? PyN-IPCs in the GW18 and GW23 cortical ISVZ expressed EGFR (Fig. 7A-D). It is well established that deep layer PyNs are the first PyNs to be generated [2,[72][73][74][75]. We thus propose that tRG-derived EGFR ?-EOMES ? PyN-IPCs become upper layer PyNs. Thus, a series of steps are required to generate tRG-derived PyN-IPCs, bMIPCs, OPCs, APCs and OBiN-IPCs.

Fate Specification of Cortical OPCs, APCs and OBiN-IPCs in the Human Cortex
Human cortical bMIPCs expressed higher levels of EGFR, ASCL1, OLIG1/2, and weak PDGFRA (Fig. 1F-H). The molecular features of bMIPCs closely resemble OPCs, therefore, many previous studies called bMIPCs as Pre-OPCs or Pri-OPCs [35,[46][47][48][49]. These studies, in conjunction with ours, provides strong evidence that as EGFR expression wanes cortical bMIPCs are biased toward OPC fates. Currently, we do not know why EGFR expression must be downregulated in the OPCs, but we hypothesize that it may interfere with PDGFR signaling in promoting OPCs proliferation and survival.
When bMIPCs give rise to APCs in the cortex, APCs maintain EGFR and OLIG1/2 expression, but downregulate ASCL1 expression (Fig. 1G). This process is prolonged in the human cortex than that in the mouse cortex, partially because the upregulation of ID3 expression in human astrocyte lineage cells is very late. Indeed, we only observed very few ID3 ? cortical APCs in the human brain at GW21-GW26 (Fig. S2B), but with the upregulation of ID1, ID2, ID4 and HES1, downregulation of ASCL1 and OLIG1/2 expression [6,76,77], and further downregulation of EGFR expression [78], human cortical APCs gradually differentiate into immature and mature astrocytes. It is also worth noting that downregulation of ASCL1 (protein and mRNA) expression in PyN-IPCs, immature PyNs and cortical interneurons in the human cortex takes a much longer time than that in the mouse cortex (Figs. 1G,  2D). Cortical astrocytes have two origins: bMIPCs are the major source of cortical astrocytes, and translocating cortical RGCs give rise to a smaller population of cortical astrocytes, as observed in the developing macaque monkey and ferret cortex [4,79,80]. We proposed that a subset of all three types of human cortical RGCs: vRGs, tRGs and oRGs, may directly transform to APCs in the human cortex. Currently, we do not know many differences between these two groups of astrocytes, but we do know transforming cortical RGC-derived APCs appear earlier in the cortex than bMIPC-derived APCs (Fig. 5A-E).
In the mouse cortex, we have identified that Egfr ? Id3 ? cells are earliest APCs, which then upregulate Aldh1l1 expression in nearly all astrocytes [6]. Aldh1l1 is expressed in mouse cortical RGCs in late embryogenesis, but downregulated in bMIPCs and re-upregulated in APCs and strongly expressed in immature and mature astrocytes [6,81,82]. In the developing human cortex, however, both ID3 and ALDH1L1 expression are delayed in astrocyte lineage cells (Fig. S2B, F). Instead, the expression of HOPX and GFAP in human cortical APCs is much earlier than that in the mouse cortex (Figs. 5, 6). Indeed, mouse GFAP protein is not widely expressed in the mouse cortical RGCs until P0, but 2.2-kb human-GFAP promoter driving expression of EGFP can be detected in the mouse cortical RGCs as early as E13.0 [83]. Thus, HOPX and GFAP are reliable markers to characterize astrocyte lineage cells in the prenatal and postnatal human cortex.
In the postnatal human cortical parenchyma, OPCs and APCs continue to divide [92]. Neural stem cells (type B1 neural stem cells, like tRGs) in the cortical SVZ may predominantly generate OBiN-IPCs [93] (Fig. 11), but this process only occurs during the first year of life [94]. In the adult human SVZ, newly born OBiNs are rare [95,96], suggesting that neurogenesis is not present in the adult human brain [97]. Fig. 11 Summary of cortical RGC lineage progression and origins of cortical oligodendrocytes, astrocytes and cortex-derived OBiNs in the human brain. Neuroepithelial cells (NEs) convert into vRGs as the developing human cortex thickens around GW7. During GW8-GW15, vRGs gradually begin to express HOPX, undergo asymmetric cell division to self-renew, and to produce short neural precursors (SNPs). Cortical SNPs, represent transitional cell form between vRGs and PyN-IPCs, differentiate into PyN-IPCs, which exclusively generate deep layer PyNs. Around GW15-GW16, human cortical HOPX ? vRGs horizontally divide into a HOPX ? oRG and a HOPX ? tRG. oRGs inherit long basal fibers of vRGs while tRGs inherit apical domains of vRGs. Both oRGs and tRGs can self-renew. HOPX ? oRGs generate upper layer PyNs. tRGs gradually express EGFR and ASCL1 and initially produce EGFR ? EOMES ? ASCL1 ? PyN-IPCs, which also differentiate into upper layer PyNs. Some tRGs may directly transform into HOPX ? EGFR ? APCs; some convert into ependymal cells (EPs). Around GW23, a large number of tRGs generate bMIPCs that express EGFR, ASCL1 and OLIG1/2. bMIPCs undergo several rounds of mitosis to generate OPCs, APCs and OBiN-IPCs in the human cortex. These lineage-restricted IPCs then divide symmetrically to generate cortical oligodendrocytes, astrocytes and OBiNs. In the neonatal human brain, cortical tRGs (B1 neural stem cells) may predominantly generate OBiN-IPCs, but this process only occurs during the first year of life. In the adult human SVZ, very few newly born OBiNs are detected.
Why Does Our Human Brain Need a Cortical OSVZ?
The findings described in this and other studies [24,30,98] raise several interesting theoretical issues. For example, why does the human brain need a cortical OSVZ? One likely reason is that with the evolutionary expansion of our neocortex it needs more PyNs. Indeed, it was reported that human cortical neurogenesis extends into the third trimester [99] (also see Fig. S12, S13). Another critical reason may be that the expanded neocortex needs more glial cells, as early as possible, because they have multiple functions in the developing cortex [100,101].
The human neocortex has two germinal zones. oRGs in the OSVZ mainly produce cortical upper layer PyNs, from GW16 to the third trimester. In contrast, the main function of tRGs in the cortical VZ is to generate upper layer PyNs first, and then generate cortical macroglial cells (and OBiNs). Thus, the human neocortex can simultaneously generate PyN-IPCs in the OSVZ and glial-IPCs (bMIPCs) in the ISVZ, so there is no way one can interfere with the other. According to this model, the specification of cortical RGCs (tRGs) may take place only in proximity to morphogens (e.g., SHH and FGF) in the cortical VZ that are implicated in glial and OBiN cell type specification [62,67,68,[102][103][104][105].
Taken together, these findings extend our understanding of the origins, lineage relationships, timing of differentiation, and molecular properties of cortical oligodendrocytes, astrocytes, and cortex-derived cortex-derived OBiNs in the human fetal brain. This also provides important insights into the origins of malignant glioma cells.
Conflict of interest All authors claim that there are no conflicts of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons. org/licenses/by/4.0/.