Purification and characterization of human neural stem and progenitor cells

SUMMARY

The human brain undergoes rapid development at midgestation from a pool of neural stem and progenitor cells (NSPCs), which give rise to the neurons, oligodendrocytes, and astrocytes of the mature brain. Functional study of these cell types has been hampered by a lack of precise purification methods. We describe a method for prospectively isolating ten distinct NSPC types from the developing human brain using cell surface markers. CD24⁻THY1^−/lo cells were enriched for radial glia, which robustly engrafted and differentiated into all three neural lineages in the mouse brain. THY1^hi cells marked unipotent oligodendrocyte precursors committed to an oligodendroglial fate, and CD24⁺THY1^−/lo cells marked committed excitatory and inhibitory neuronal lineages. Notably, we identify and functionally characterize a transcriptomically-distinct THY1^hiEGFR^hiPDGFRA⁻ bipotent glial progenitor cell (GPC), which is lineage-restricted to astrocytes and oligodendrocytes, but not neurons. Our study provides a framework for the functional study of distinct cell types in human neurodevelopment.

In Brief

A framework for the prospective isolation of neural stem and progenitor cells from the developing human brain is provided to facilitate the functional study of distinct cell types in human neurodevelopment.

Graphical Abstract

INTRODUCTION

The human brain boasts an intricate architecture built from billions of cells of diverse identities. Yet, this extraordinary complexity arises during development from a relatively uniform neuroepithelium. In the developing cerebral cortex, radial glia (RG) serve as neural stem cells (NSCs) that self-renew and give rise to progressively more lineage-restricted progenitors, ultimately generating three major neural lineages: neurons, oligodendrocytes, and astrocytes^1–3. Recently, single-cell technologies have offered unprecedented spatial and temporal resolution on the transcriptomic diversity of neural stem and progenitor cells (NSPCs) throughout development^4–7, resulting in detailed atlases of human NSPCs. However, while transcriptomic signatures have great utility in identifying cell types and informing cell type properties, stem cells should ultimately be defined by their function, specifically with respect to self-renewal and differentiation potential.

The essence of stem cell biology thus lies in the ability to prospectively isolate pure populations of cells for functional characterization—warranting reliable tissue dissociation protocols and cell-surface markers for purification and assessment of their in vivo properties of self-renewal and differentiation. This is especially true in human tissues, for which researchers are not afforded the usual tools of genetic lineage tracing. Without such capabilities, questions regarding neural stem and progenitor mechanisms, that could include hierarchical organization of distinct intermediate progenitor states that are progressively restricted in their lineage output potential, have remained difficult to explore. Therefore, we sought to develop a method for purifying distinct NSPC subsets via fluorescence-activated cell sorting (FACS), allowing for prospective isolation based on the quantitative expression of over a dozen cell-surface markers.

We have previously shown that CD133⁺CD24^−/lo cells isolated from 16–20-gestational week-old (GW16–20) human brains using FACS are highly enriched in NSCs, with an appreciable fraction (1 in 23; range: 1 in 15 to 1 in 34) giving rise to clonally derived neurospheres in culture¹. The neurospheres, which consist of undifferentiated cells, expand for many passages, can differentiate into neurons and glia in vitro, and when transplanted into the brains of neonatal immunodeficient mice, engraft, migrate, proliferate, and differentiate in a site-appropriate manner^1,8. Cultured neurospheres are further capable of rescuing disease phenotypes in animal models of various central nervous system disorders^8–10, and have been extensively investigated for their clinical applications in regenerative medicine¹¹.

However, despite new NSC cell-surface markers that have since been identified in rodents^12,13, the precise identity and purity of cells sorted using combinations of these markers have not been rigorously characterized, and significant heterogeneity likely exists among the isolated cell populations. For example, radial glia exist in several distinct types: ventricular radial glia (vRG) which reside in the ventricular zone (VZ) and maintain both apical and basal processes, and outer radial glia (oRG) which reside in the outer subventricular zone (OSVZ) and maintain only their basal processes^14,15. However, a paucity of isolation methods for these rare, more nuanced cell types has made their functional characterization challenging.

Here, we combine high-dimensional flow cytometry with single cell transcriptomics to comprehensively purify and functionally characterize neural stem cells and distinct downstream progenitor subsets from the developing human brain during mid-gestation (GW17–19). We utilize index-sorting based on the combinatorial expression of cell-surface markers (immunophenotype) to directly link surface marker expression to the expressed transcriptomes of single cells, allowing for rigorous validation of sort purity. Our tissue processing and sorting strategy allows for the simultaneous purification of ten defined NSPC populations, including radial glia, neuron precursors, oligodendrocyte precursors, and astrocyte lineage cells, reflective of a NSC lineage hierarchy as described in other tissue stem cell systems¹⁶. We further demonstrate their functional properties in vitro, and in vivo. Within this framework, we also identify a bipotent glial progenitor cell (GPC) that gives rise exclusively to astrocytes and oligodendrocytes. Our findings aim to lay the groundwork for isolating NSPCs to establish the self-renewal and differentiation potential of these cells, and for future molecular and cellular studies on human neurodevelopment and the development of cell transplantation-based therapeutic regimens.

RESULTS

Immunophenotypic definition of neural stem and progenitor cell types

To purify cell-intrinsically distinct putative NSPC fractions, we mechanically and enzymatically dissociated GW17–19 human brain tissue into a single-cell suspension (Figure 1A). The cortex was dissected out prior to dissociation whenever possible. The cells were stained with a panel consisting of antibodies against PROM1 (CD133), CD24, THY1 (CD90), CXCR4, EGFR, PDGFRA, and non-neural lineages (CD45, PECAM1 [CD31], CD34, ENG [CD105], and GYPA [CD235a]). The stained cell suspension was analyzed using fluorescence-activated cell sorting (FACS) (Figure 1B, S1A). The observed immunophenotypic profile was extremely consistent across over a dozen samples of this gestational age group. Various populations that stratified based on cell-surface marker expression (immunophenotype) were index-sorted into 96-well plates for full-length single-cell RNA-sequencing (scRNA-seq) using Smart-seq2 or Smart-seq3^17,18. Index-sorting records the fluorescence intensities for each sorted single cell, allowing for direct mapping between surface marker profile and the expressed transcriptome.

Figure 1. Prospective isolation of NSPCs from the developing human brain.

(A) Tissue processing and experimental workflow for isolation and characterization of NSPCs via transcriptomic and functional methods.

(B) Gating scheme for isolation of distinct NSPC populations using FACS, based on varying cell-surface expression of THY1 (CD90), CD24, EGFR, PDGFRA, and CXCR4. Events are pre-gated on live single cells and negative for non-neural lineage markers PECAM1 (CD31), CD34, PTPRC (CD45), ENG (CD105), and GYPA (CD235a).

(C) Single cell RNA sequencing of index-sorted NSPCs using Smart-seq3. Plot showing PAGAembedded Leiden clusters with annotated cell type identities based on expressed transcripts of known genes.

(D) Index-sort data was used to map the sequenced single cells to their original immunophenotype with respect to cell-surface CD24 and THY1 expression levels.

Unsupervised clustering of single-cell transcriptomes identified various putative NSPC populations, which we annotated using input from a combination of functional assays and expressed mRNAs of known marker genes (Figure 1C, S1B). These transcriptomically-defined cell types include putative ventricular radial glia (vRG), outer radial glia (oRG), astrocytes (AC), pre-oligodendrocyte precursor cells (pre-OPC), OPCs, oligodendrocytes (OLs), intermediate progenitor cells (IPCs), and excitatory and inhibitory neuronal lineage cells (ExN, InN) (see details below).

Index-sort data allowed for each individual sequenced cell to be mapped back to its original immunophenotype. RNA and cell-surface protein expression did not always correlate, most notably with the glycosylphosphatidylinositol (GPI)-anchored surface molecules THY1 and CD24 (Figure S1C). Using the index-sort information, we found that the putative NSCs were CD24⁻THY1^−/lo, the putative OPCs were THY1^hi, and the putative neuronal lineage cells were CD24⁺THY1^−/lo (Figure 1D), with additional heterogeneity present in each compartment. We have previously reported that CD133⁺CD24⁻ cells are highly enriched for NSCs¹; indeed, our index-sort analysis confirms that this gate largely consists of vRG (Figure S1D, E). Our gating strategy resulted in remarkably pure populations of sorted cells as validated by scRNA-seq, allowing for prospective isolation of various putative NSPCs for downstream functional assays.

CD24⁻THY1^−/lo expression identifies phenotypic NSCs

Index-sort data showed that putative NSCs, broadly expressing SOX2, GFAP, and VIM transcripts, were enriched by the CD24⁻THY1^−/lo gate, with further immunophenotypic heterogeneity observed in cell-surface expression of EGFR and CXCR4 (Figure 2A). Within the CD24⁻THY1^−/lo gate, the EGFR^hi population was enriched for putative vRG (expressing CRYAB and FBXO32 transcripts), while the EGFR⁻ population was enriched for putative oRG (expressing HOPX and LIFR transcripts (Figure 2B), and LIFR protein (Figure S2A))¹⁵. Histological analysis confirmed that EGFR is expressed in the ventricular and subventricular zones (VZ/SVZ) but absent in the outer subventricular zone (OSVZ) (Figure S2B). In the VZ/SVZ, EGFR colocalized with GFAP and the vRG-specific marker CRYAB, supporting our finding that EGFR distinguishes vRG from oRG. We also observed a population of EGFR⁺CXCR4⁺ putative astrocyte (AC) lineage enriched for PAX3 and EDNRB transcripts. Intracellular staining followed by flow cytometry confirmed that, similar to their expression at the transcript level (Figure S1B), the CD24⁻THY1^−/lo cells also highly expressed the protein versions of known NSC marker genes, including SOX1, SOX2, GFAP, and PAX6 (Figure S2D). Notably, the expression of PAX3 transcript was restricted to the putative AC lineage (Figure S1B), suggesting that PAX3 may have a role in early lineage bifurcation between the NSC and AC lineages. This finding is in contrast with studies in mouse showing that Pax3 downregulation is associated with differentiation towards an astrocytic fate ¹⁹. It is possible that our AC population represents an early progenitor population where PAX3 induction is initially required before being downregulated with maturation. Alternatively, there could be considerable cellular heterogeneity within the AC lineage with respect to PAX3 expression in human prenatal brains.

Figure 2. CD24⁻THY1^−/lo expression identifies NSCs.

(A) Gating scheme for CD24⁻THY1^−/lo NSC compartment.

(B) Index-sort data was used to map the transcriptomically-identified vRG (red), oRG (orange), and AC (brown) clusters to their original immunophenotype.

(C) Plot showing the quantification of neurosphere initiation frequency of each CD24⁻THY1^−/lo subset based on in vitro limiting dilution assays. n=5–11 donors per population, mean ± standard deviation (S.D.)

(D) Right: immunofluorescence (IF) images of CD24⁻THY1^−/lo subsets showing DCX (green) and GFAP (red) expression after 5 days of in vitro differentiation post-sort. Scale bar 50 μm. Left: bar-graph showing quantification of percent DCX⁺ and GFAP⁺ cells in images similar those shown in right panel. Error bars show standard deviation.

(E–G) Confocal IF images of mouse brains engrafted with CD24⁻THY1^−/lo human NSCs. 40 μm thick sections were stained with anti-human GFAP (E) or human cytoplasmic antigen-specific STEM121 antibody (F, G), in addition to species cross-reactive antibodies against SOX2, OLIG2, or NeuN and MAP2 (E, F, G, respectively). Imaged regions: medulla (Med), third ventricle (3V), hypothalamus (HY), olfactory bulb (OB), cerebellum (CB), corpus callosum (cc), optic chiasm (och), and subventricular zone (SVZ). Scale bar 25 μm.

(H) Visualization of CD24⁻THY1^−/loEGFR⁻ (top) and CD24⁻THY1^−/loEGFR^hi (middle) NSCs engrafted into the mouse brain. Each dot represents an engrafted GFAP⁺ (orange) or an OLIG2⁺ (green) human cell. Bottom: bar graph showing quantification of orange GFAP⁺ and green OLIG2⁺ cells per mm³ in brain sections similar to those shown in the upper two panels. n=4 quantified brains per population; mean ± S.D.

To assess whether our putative NSC subsets are functional, we next assessed their self-renewal and differentiation capabilities in vitro and in vivo. We have previously shown that CD133⁺CD24⁻ cells can self-renew and be maintained in the presence of LIF, FGF, and EGFR and expanded as neurospheres¹. Consistent with our previous results CD24⁻THY1^−/lo putative NSCs readily formed neurospheres when cultured. In addition, in vitro limiting dilution assays showed that within the CD24⁻THY1^−/lo gate, the EGFR^hi population possesses markedly higher neurosphere initiation frequency (1 in 4.5) compared to the EGFR⁻ population (1 in 104) (Figure 2C)—a difference that was not attributable to cell cycle (Figure S2C). To assess the functional behavior of individual CD24⁻THY1^−/lo putative NSCs in vitro, we generated clonal neurospheres from single sorted cells. We dissociated individual single-cell-derived clonal neurospheres and subjected them to differentiation conditions. Both EGFR⁻ and EGFR^hi NSC clones could be differentiated into DCX⁺, GFAP⁺, and rare O4⁺ cells, demonstrating their competence to give rise to neuron, astrocyte, and oligodendrocyte lineages (Figure S2E, F). In contrast, bulk-sorted EGFR⁺CXCR4⁺ cells gave rise almost exclusively to GFAP⁺ cells which also stained positive for AQP4, suggesting they are largely committed to an astrocytic fate (Figure 2D, S2G).

To functionally characterize our putative NSCs in vivo, we transplanted acutely isolated cells directly, without intervening culture, into the lateral ventricles of neonatal NOD-scid-IL2Rg^null (NSG) immunodeficient mice. After 6 months, we assessed the level of engraftment by immunofluorescent (IF) staining for human-specific antigens. Transplanted NSCs migrated and engrafted extensively throughout the brain and differentiated to give rise to all three major neural lineages: GFAP⁺ astrocytes/NSCs, OLIG2⁺ oligodendrocytes, and NeuN⁺ or MAP2⁺ neurons (Figure 2F–H). Remarkably, the CD24⁻THY1^−/loEGFR^hi NSCs resulted in significantly higher levels of engraftment compared to CD24⁻THY1^−/loEGFR⁻ NSCs, with 1.8-fold more OLIG2⁺ cells and 2.3-fold more GFAP⁺ cells observed across four replicates. This result is consistent with previous studies in the mouse demonstrating the role of EGFR in promoting proliferation and astrogliogenesis^20,21. Collectively, our in vitro and in vivo results confirm that our putative NSC fractions are indeed functional with respect to their self-renewal capability and differentiation potential. In addition, among the CD24⁻THY1^−/lo NSCs, EGFR^hi cells are likely enriched for relatively primitive multipotent NSCs compared to EGFR⁻ cells.

THY1^hi expression identifies oligodendrocyte lineages

Index-sort data combined with unsupervised clustering showed that the putative oligodendrocyte (OL) lineages, broadly expressing OLIG1, OLIG2, and SOX10 transcripts, was enriched by the THY1^hi gate. This gate highly enriched all OL lineages, suggesting THY1 to be a pan-oligodendrocyte lineage marker at least at this developmental stage. Among the THY1^hi cells, we observed additional heterogeneity with respect to cell-surface expression of EGFR and PDGFRA, resulting in four immunophenotypic populations: THY1^hiEGFR⁺PDGFRA⁻ (T^hiE⁺P⁻), THY1^hiEGFR⁺PDGFRA⁺ (T^hiE⁺P⁺), THY1^hiEGFR⁻PDGFRA⁺ (T^hiE⁻P⁺), and THY1^hiEGFR⁻PDGFRA⁻ (T^hiE⁻P⁻) (Figure 3A). Index-sort analysis showed that the T^hiE⁺P⁺ gate highly enriched for putative pre-OPCs (expressing EGFR, PDGFRA, and PCNA transcripts), the T^hiE⁻P⁺ gate for putative OPCs (expressing PDGFRA and PCDH15 transcripts), and the T^hiE⁻P⁻ gate for more mature oligodendrocytes (expressing MYRF and MBP transcripts) (Figure 3B). The T^hiE⁺P⁻ gate consisted of a heterogeneous population likely consisting of both putative OPCs and NSCs (Figure 3B, S3A), further discussed below. Consistent with our gene expression-based inference, pseudotime analysis also recapitulated the oligodendrocyte maturation trajectory (Figure S3B). Plotting the pseudotime by immunophenotype shows a gradient of maturation from the T^hiE⁺P⁺ to the T^hiE⁻P⁺, and then to the T^hiE⁻P⁻ gates (Figure S3C). Indeed, the T^hiE⁺P⁺ gate was enriched for actively-cycling cells (Figure S3D), suggesting that they are a transit-amplifying population that expands the OPC pool, likely corresponding to pre-OPCs²². Histological analysis confirmed the presence of both EGFR^–PDGFRA⁺ and EGFR⁺PDGFRA⁺ cells with OPC-like morphology in the fetal human cortex (Figure 3C, S3E). EGFR⁺PDGFRA⁺ cells had simpler morphology compared to EGFR^–PDGFRA⁺ cells (average 2.2 vs. 6.7 processes per cell, respectively), suggesting that the former represent a less mature cell type.

Figure 3. THY1^hi expression identifies OPCs.

(A) Gating scheme for THY1^hi OPC compartment.

(B) Index-sort data was used to map the transcriptomically-identified pre-OPC (light green), OPC (green), and OL (dark green) clusters to their original immunophenotype.

(C) (left) Quantification of processes on EGFR⁺PDGFRA⁺ and EGFR⁻PDGFRA⁺ cells in fetal human cerebral cortex (18 gestational weeks, GW18). (right) Confocal IF images of GW18 fetal human cerebral cortex stained for EGFR (green), PDGFRA (red), and DAPI (blue). Empty arrowheads to EGFR⁺PDGFRA⁺ cells; solid arrowheads point to EGFR⁻PDGFRA⁺ cells. Scale bar 50 μm.

(D) Plot showing the quantification of neurosphere initiation frequency of each THY1^hi subset based on limiting dilution assays. n=4–7 donors per population, mean ± S.D.

(E) IF images of the bulk-sorted THY1^hi subsets that were cultured in the absence of growth factors for 4 days, and subsequently stained with anti-O4 (green) antibody and DAPI (blue) marking the nuclei, along with quantification of percent O4⁺ cells in images similar to those in the left panel. Mean ± S.D. Scale bar 50 μm.

(F) Visualization of THY1^hi subsets engrafted into the mouse brain. Each dot represents an engrafted GFAP⁺ (orange) or OLIG2⁺ (green) human cell.

(G) Confocal IF images of mouse brains engrafted with THY1^hi human OPCs. 40 μm thick sections were stained with human cytoplasmic antigen specific STEM121 antibody (green) and OLIG2 (red). Imaged regions: corpus callosum (cc), cerebellum (CB), dentate gyrus (DG), and midbrain (MB). Scale bar 25 μm.

We next assessed the functional behavior of the putative OPC subsets in vitro. Limiting dilution assays showed that the T^hiE⁺P⁻ and the T^hiE⁺P⁺ cells had a high neurosphere initiation rate (1 in 9.8 and 1 in 5.6, respectively), in contrast to the T^hiE⁻P⁺ and the T^hiE⁻P⁻ cells (1 in 872 and 1 in 1634, respectively) (Figure 3D), further supporting the notion that EGFR expression marks a more proliferative pre-OPC subtype. Next, we sorted each of the four OPC populations and cultured them in the absence of growth factors to induce differentiation. Subsequent IF imaging of resulting cells post-differentiation revealed O4⁺ cells with distinctive oligodendroglial morphology (Figure 3E). When the T^hiE⁺P⁻, T^hiE⁺P⁺, and T^hiE⁻P⁺ populations were subjected to the in vitro differentiation, the resulting proportions of O4⁺ cells showed an increasing trend from 2.3% to 9.6%, and then to 83.5%, respectively. The T^hiE⁻P⁻ cells did not survive well in culture likely due to their more mature state—nevertheless, the survivors still gave rise to 61.6% O4⁺ cells in our culture conditions (Figure 3E). Strikingly, no other populations gave rise to any appreciable numbers of O4⁺ cells (Figure S3F). To determine whether the putative pre-OPCs and OPCs are lineally related or represent interconvertible states, we reanalyzed sorted T^hiE⁺P⁺ and T^hiE⁻P⁺ populations after 5 days of in vitro differentiation using flow cytometry. In samples from three separate donors, while the T^hiE⁺P⁺ putative pre-OPCs gave rise to both EGFR⁺ and EGFR⁻ cells, the T^hiE⁻P⁺ putative OPCs remained EGFR⁻, suggesting a hierarchical organization (Figure S3G). These results provide supportive evidence for the maturation trajectory predicted by the pseudotime analysis described above.

To assess their functional behavior in vivo, we then transplanted the THY1^hi OPC subsets into the lateral ventricles of neonatal NSG mice. The T^hiE⁺P⁺ and T^hiE⁻P⁺ subsets engrafted robustly in the brain (Figure 3F, G), and remarkably, gave rise exclusively to OLIG2⁺ cells, with no observed GFAP⁺ or NeuN/MAP2⁺ cells, suggesting that these cells are indeed committed to an oligodendroglial fate. The T^hiE⁻P⁻ cells did not engraft, possibly because they are too mature and post-mitotic. In summary, our collective in vitro and in vivo results confirm that the purified OPC subsets are engraftable and are indeed lineage committed to form only oligodendrocytes.

Identification of a bipotent glial progenitor

Within our scRNA-seq data, we identified a cluster of cells that expressed transcripts for genes characteristic of both astrocyte (GFAP, SLC1A3, HEPACAM, AQP4) as well as oligodendrocyte (OLIG1, OLIG2) lineages (Figure 4A, S4A), leading us to hypothesize it as a putative bipotent glial progenitor cell (GPC). Pseudotime trajectory analysis showed that these putative GPCs maintain some oRG markers (HOPX, TNC) but also turn on other unique markers (ETV4, ADM, METTL7B) not found in other neural cell types (Figure 4B, S4B). Through confocal imaging of fetal brain tissue, we were able to identify cells that stained positive for both GFAP and OLIG2 in the cortex (Figure 4C). Further immunostaining confirmed that GFAP⁺OLIG2⁺ cells also stain positive for HOPX and ETV4 (Figure 4D), consistent with our transcriptomic findings. Unlike HOPX, which is also expressed in oRG, ETV4 staining was specific to GFAP⁺OLIG2⁺ cells, validating it as a cell type-specific transcription factor for the putative GPCs. These GFAP⁺OLIG2⁺ cells were preferentially located in the OSVZ, with lesser numbers found in the VZ/SVZ and the subplate, and none found in the cortical plate (Figure 4E). This anatomical bias as well as their residual expression of oRG markers suggest that GPCs may preferentially arise from oRG, though it does not exclude the possibility they may also arise directly from vRG. Indeed, in vitro differentiation of CD24⁻THY1^−/lo radial glia gives rise to some GFAP⁺OLIG2⁺ cells (Figure S4C), supporting their hierarchical relationship.

Figure 4. Identification of a bipotent glial progenitor.

(A) mRNA expression matrix showing astrocyte and oligodendrocyte marker genes in transcriptomic cell clusters.

(B) PAGA pseudotime analysis of expressed transcripts along the maturation trajectory from ventricular and outer radial glia (vRG, oRG) to glial progenitor cells (GPC) to oligodendrocyte precursor cells (OPC) to oligodendrocytes (OL).

(C) Confocal immunofluorescence (IF) images of fetal human brain sections (18 gestational weeks; GW18) from the cortex. 14 μm thick sections were stained with antibodies against GFAP (green) and OLIG2 (red). Scale bar 50 μm.

(D) Confocal IF images of GW18 fetal human cerebral cortex, stained with antibodies against GFAP (green), OLIG2 (red), ETV4 (cyan, top), and HOPX (cyan, bottom). Scale bar 10 μm.

(E) Anatomical distribution of GFAP⁺OLIG2⁺ cells in GW18 fetal human cortex across the ventricular/subventricular zone (VZ/SVZ), outer subventricular zone (OSVZ), intermediate zone/subplate (IZ/SP), and cortical plate (CP).

(F) Index-sort data was used to map the transcriptomically-identified glial progenitors to their original immunophenotype. Glial progenitors were found to be enriched in the THY1^hiEGFR^hiPDGFRA⁻ gate.

(G) Experimental strategy for clonal differentiation assay.

(H) IF images of cells after differentiation stained with DAPI (blue) and antibodies against GFAP (red), O4 (green), and DCX (cyan). Clonal neurospheres were derived from single THY1^hiEGFR^hiPDGFRA⁻ cells, the cells from clonal neurospheres were dissociated then subjected to differentiation conditions. Scale bar 50 μm.

(I) Quantification of lineage output of clonal neurospheres derived from THY1^hiEGFR^hiPDGFRA⁻, THY1^hiEGFR^midPDGFRA⁻, THY1^hiEGFR⁺PDGFRA⁺, or CD24−THY1^−/lo cells. Each column represents a distinct clonal neurosphere. Differentiated cells were classified based on their expression of GFAP, DCX, or OLIG2.

(J) Confocal IF images of mouse brains engrafted with THY1^hiEGFR^hiPDGFRA⁻ putative glial progenitors. 40 μm thick sections were stained with antibodies against human GFAP (left, green), SOX2 (left, red), human cytoplasmic antigen (right, green) and OLIG2 (right, red). Imaged regions: medulla (Med) and midbrain (MB). Scale bar 50 μm.

(K) Visualization of THY1^hiEGFR^hiPDGFRA⁻ cells engrafted into the mouse brain. Each dot represents an engrafted GFAP⁺ (orange) or OLIG2⁺ (green) human cell.

Even with this transcriptomic and histologic data, functional characterization of our putative GPC would only be possible if they can be prospectively isolated for downstream experiments. Fortunately, index-sort analysis showed that this cell type was enriched within the THY1^hiEGFR^hiPDGFRA⁻ gate (Figure 4E), thus allowing for their prospective isolation. To assess the developmental potential of this cell type, we index-sorted single THY1^hiEGFR^hiPDGFRA⁻ cells and grew them into clonal neurospheres. We dissociated the clonal neurospheres and subjected the cells to differentiation in vitro by cytokine withdrawal (Figure 4F). Strikingly, nine out of ten clonal neurospheres derived from THY1^hiEGFR^hiPDGFRA⁻ cells gave rise to both OLIG2⁺ oligodendroglial and GFAP⁺ astrocytic cells, but no DCX⁺ neuronal cells after differentiation (Figure 4G, H), suggesting that at a clonal level a significant majority of the THY1^hiEGFR^hiPDGFRA⁻ cells are restricted to oligodendroglial and astrocytic fates. One of the ten neurospheres gave rise exclusively to OLIG2⁺ cells, indicating some unipotent pre-OPCs copurified within this gate (Figure 3B). In contrast, clonal neurospheres derived from THY1^hiEGFR⁺PDGFRA⁺ (T^hiE⁺P⁺) cells—which we have identified above as pre-OPCs and OPCs—gave rise exclusively to OLIG2⁺ cells. Clonal neurospheres derived from the CD24⁻THY1^−/lo population enriched in radial glia gave rise to all three neural lineages, including DCX⁺, GFAP⁺, and rare OLIG2⁺ cells (Figure 4H, S2E). Clonal neurospheres derived from THY1^hiEGFR^midPDGFRA⁻ cells gave rise to a mix of unipotent, bipotent, and tripotent colonies, consistent with the index sort analysis showing that this gate consists of heterogeneous mix of pre-OPCs, glial progenitors, and radial glia (Figure 3B, S3A, 4C). Though these clonal differentiation assays require a period of in vitro outgrowth which may affect the character of the cells, the clear contrast in potency and lineage output between different NSPC types appears to be conserved.

To assess the in vivo behavior of the putative glial progenitors, we transplanted acutelyisolated THY1^hiEGFR^hiPDGFRA⁻ cells into the lateral ventricles of neonatal NSG mice. Six months post-transplant, we observed engraftment of both GFAP⁺ and OLIG2⁺ human cells (Figure 4I, J). None of the engrafted human cells stained positive for either DCX or MAP2. Though it is not technically feasible to transplant single human cells, these transplant assays are still consistent with the identity of a bipotent GPC. The relative proportion of oligodendrocytes versus astrocytes was significantly higher in the vivo setting (67%) than in vitro (6.3%), suggesting that oligodendroglial differentiation of GPCs is directed by signals present in the neonatal mouse brain. Indeed, differentiation of GPC-derived clonal neurospheres in OPC induction media (containing PDGF, IGF, and NT3)²³ resulted in 21.2% OLIG2⁺ cells compared to the 6.3% obtained from cytokine withdrawal (Figure S4D). Taken together, the transcriptomic identification, prospective isolation, clonal in vitro differentiation assays, and in vivo engraftment results offer strong evidence for the existence of a functionally bipotent glial progenitor cell in the fetal human brain.

CD24⁺THY1^−/lo expression identifies neuronal precursors

Index-sort analysis showed that neuron precursors, broadly expressing DCX, SOX4, and SOX11 transcripts, were highly enriched within the CD24⁺THY1^−/lo gate, with further immunophenotypic heterogeneity observed with respect to EGFR and CXCR4 expression (Figure 5A). Within the CD24⁺THY1^−/lo gate, the EGFR⁺CXCR4⁻ gate enriched for early glutamatergic excitatory neuron (ExN) lineage cells, the EGFR⁻CXCR4⁻ gate enriched for later ExN, while the EGFR⁻CXCR4⁺ gate enriched for GABAergic inhibitory neuron lineage cells (InN) (Figure 5B). These markers are consistent with known biology, as interneurons are known to migrate towards gradients of CXCL12 via interactions with its receptor CXCR4²⁴. In particular, during early fetal brain development, the ganglionic eminences give rise to InN precursors that undergo a long-range tangential migration to reach the cortex²⁵. Pseudotime analysis confirmed mRNA expression changes along the neuron maturation trajectory, with intermediate progenitor genes (EOMES) preceding early (NEUROD2) and late (SATB2) neuron markers (Figure 5C).

Figure 5. CD24⁺THY1^−/lo expression identifies neuron precursors.

(A) Gating scheme for CD24⁺THY1^−/lo neuron compartment.

(B) Index-sort data was used to map the transcriptomically-identified early ExN (light blue), late ExN (dark blue), and InN (purple) clusters to their original immunophenotype.

(C) PAGA pseudotime analysis of expressed transcripts along the neuronal maturation trajectory.

(D) Top: IF images of cultured CD24⁺THY1^−/loCXCR4⁻ cells stained with anti-DCX (red) and anti-MAP2 (green) antibodies. Bottom: bar-graph showing the quantification of percent DCX⁺, MAP2⁺, GFAP⁺, and OLIG2⁺ cells in images similar to those shown in the top panel. Scale bar 50 μm.

(E) IF images of cultured CD24⁻THY1^−/loCXCR4⁻ cells stained with anti-SYN1 (red) and anti-MAP2 (green) antibodies. Scale bar 50 μm.

To assess their functional behavior in vitro, we then sorted CD24⁺THY1^−/loCXCR4⁻ cells for culture, whereupon they differentiated into a homogeneous population of DCX⁺MAP2⁺ cells with neuronal morphology (Figure 5D). Synapsin-1 (SYN1) puncta were observed between the neurons, suggesting the formation of synaptic connections. No GFAP⁺ or OLIG2⁺ cells were observed, suggesting that these cells are indeed committed to a neuronal fate. CD24⁺THY1^−/loCXCR4⁺ InN could not be successfully cultured. We attempted to transplant both ExN and InN into mice, but observed no engraftment for either, even when transplanting up to 2 million cells. This result likely reflects the cells being too mature to robustly engraft. This finding may have implications for neuronal transplantation studies for applications in regenerative medicine.

NSPC surface markers are conserved across diverse brain regions

Though our studies have been focused on the cortex broadly, we also sought to determine how well our purification scheme applies to specific cortical regions and subcortical structures. From intact fetal human brain specimens, we dissected out 13 distinct regions: the frontal lobe (FL), motor cortex (precentral gyrus, PrG), somatosensory cortex (postcentral gyrus, PoG), parietal lobe (PL), subventricular zone (SVZ), frontal/temporal lobe border (FTL), temporal lobe (TL), hippocampus (Hip), parietal/occipital lobe border (POL), occipital lobe (OL), thalamus (Th), brainstem (BS), and cerebellum (CB) (Figure 6A). Each region was separately digested, stained, and index-sorted for Smart-seq3 as above.

Figure 6. NSPC surface markers are conserved across diverse brain regions.

(A) Dissected regions of fetal human brain.

(B) Sorting strategy for neural stem and progenitor cell types. Each gate has been assigned a letter from a-i. Stacked bar graphs show the composition of transcriptomically-defined cell types purified from each gate, for each brain region. (a) CD24⁻THY1^−/lo, (b) CD24⁺THY1^−/lo, (c) THY1^hi, (d) CD24⁺THY1^−/loCXCR4⁻, (e) CD24⁺THY1^−/loCXCR4⁺, (f) THY1^hiEGFR^hiPDGFRA⁻, (g) THY1^hiEGFR⁺PDGFRA⁺, (h) THY1^hiEGFR⁻PDGFRA⁺, (i) THY1^hiEGFR⁻PDGFRA⁻.

Single cell RNA-seq of the dissected brain regions recovered all lineages described above (Figure S5A). As expected, ventricular and outer radial glia as well as cortical neuron lineages were represented in all cortical regions, but not the subcortical structures (i.e. thalamus, brainstem, and cerebellum) (Figure S5B). Glial progenitor cells, astrocytes, and oligodendrocyte lineages were represented in all brain regions. We also observe a cluster of RELN-expressing neurons found only in the subcortical structures. With this anatomically annotated dataset, we were also able to confirm that even in segmented brain samples (where dissecting out the cortex prior to dissociation was not possible), the vast majority of sequenced cells (86–99%) mapped to a cortical identity (Table S1), and of the cells that did not map to a cortical identity, nearly all were accounted for by subcortical astrocytes (Figure 6Ba). Limiting dilution assays from each brain region showed that CD24⁻THY1^−/loEGFR^hi cells gave rise to neurospheres at a high frequency in all cortical regions (1 in 5.8 to 1 in 13.8), less so in the brain stem (1 in 45) and cerebellum (1 in 48), and rarely in the thalamus (1 in 1509) (Figure S5C).

We then performed index-sort analysis to determine how well our previously described purification scheme applies to each brain region. Consistent with our earlier results, the CD24⁻THY1^−/lo gate highly enriched for radial glia and astrocytes in all cortical regions; in the subcortical structures, where radial glia are not present, CD24⁻THY1^−/lo still enriched for astrocytes (Figure 6B, a). The CD24⁺THY1^−/lo gate was highly specific for neuronal lineages in all brain regions (Figure 6B, b). CD24⁺THY1^−/loCXCR4⁻ enriched for excitatory neurons in all brain regions. CD24⁺THY1^−/loCXCR4⁺ enriched for interneurons in all cortical regions as well as the thalamus, and, to a lesser extent, the brainstem and cerebellum (Figure 6B, d–e). The THY1^hi gate was highly specific for glial progenitors and oligodendrocyte lineages, regardless of brain region (Figure 6B, c). The subpopulations within the THY1^hi too remained consistent across all brain regions: EGFR^hiPDGFRA⁻ enriched glial progenitors, EGFR⁺PDGFRA⁺ enriched pre-OPCs, EGFR⁻PDGFRA⁺ enriched OPCs, and EGFR⁻PDGFRA⁻ enriched mature oligodendrocytes (Figure 6B, f–i). Overall, we have shown that our surface markers for NSPCs in the fetal human brain are remarkably conserved even across diverse brain regions, allowing for its broad application to any brain structure at this gestational age of development.

We then used our gating scheme to quantify the frequency of each immunophenotypic population as a percentage of live lineage (blood and vessel) negative cells in all 13 dissected brain regions (Figure S6). The frequency of THY1^hi cells was consistently 0.5–1% in cortical regions, and enriched to 4.0% in the SVZ, possibly reflecting active oligodendrogenesis in the SVZ and OSVZ (Figure S6B, C). In the thalamus and brainstem, THY^hi cells made up over 40% of cells, reflecting the higher extent of myelination in more caudal brain structures at this developmental stage. Our method thus allows for rapid profiling of fetal human brain tissue and their relative cellular makeup via flow cytometry.

Profiling of the NSPC surfaceome reveals new cell type-specific markers

In addition to our base gating scheme, we measured the expression of 352 additional surface markers using flow cytometry, allowing for cell type-specific profiling of surface antigen expression (Figure S7A, B, Table S2). We identified several markers specific to CD24⁻THY1^−/lo radial glia (CD49b [ITGA2], CD49d [ITGA4], CD142 [F3], PSMA [FOLH1]), as well as those specific to THY1^hi oligodendrocyte lineages (CD111 [NECTIN1], CD146 [MCAM], CD81, CD73 [NT5E], CD172a/b [SIRPA/SIRPB], CD202b [TEK]). No single marker was specifically enriched in CD24⁺THY1^−/lo neurons, emphasizing the importance of using combinatorial marker expression in isolating pure cell types. Within the CD24⁻THY1^−/lo gate, we identify several markers enriched in EGFR⁺ vRG (CD193 [CCR3], CCR10, CD164, CD165, CD51 [ITGAV], CD58), as well as a few enriched in EGFR⁻ oRG (CD325 [CDH2], CD200).

By integrating our transcriptome and surfaceome data, we were able to correlate the percent of cells expressing RNA versus the corresponding surface protein for each marker for each of the identified cell type (Figure S7C, D). Though some markers showed good correlation between the percent of cells expressing RNA and surface protein (CD56 [NCAM1], CD147 [BSG], β2-microglobulin [B2M], CD29 [ITGB1], CD325 [CDH2], CD164), other markers showed high RNA expression with little surface protein expression (GPR56 [ADGRG1], CD46, Notch2, CD220 [INSR]), or high surface protein expression with little RNA expression (CCR10, CD57 [B3GAT1], CD276, CD298 [ATP1B3]). Thus, though the transcriptomic differences between NSPC types have been previously studied, characterization of surface antigen expression remains essential due to their utility in live sorting, as well as the often-poor correlation between RNA and protein (Figure S1C). Our broad profiling of the NSPC surfaceome thus facilitates future studies on functional heterogeneity within specific cell types, potential molecular mechanisms of lineage restriction, fate specification, migration, and homing in the fetal human brain.

DISCUSSION

Our data demonstrate that NSPCs can be prospectively isolated from the developing human brain based on the expression of defined surface markers. We identify three major neural compartments: CD24⁻THY1^−/lo radial glia and astrocytes, THY1^hi oligodendrocyte lineages, and CD24⁺THY1^−/lo neuron precursors, with further immunophenotypic heterogeneity present within each main gate (Figure 7). These markers are conserved across diverse brain regions. Whereas NSCs maintain multilineage potential, OPCs, astrocytes, and neuron precursors are heavily skewed if not outright committed to their specific lineage. We additionally identify a transcriptomically- and functionally-distinct glial progenitor cell (GPC), which we found to be lineage-restricted to oligodendroglial and astrocytic fates.

Figure 7. Prospective isolation strategy for human fetal NSPCs.

Graphical summary of NSPC types and their corresponding surface marker expression identified in this study.

Index sorting as a method for developing prospective isolation strategies

Index-sorting is a powerful tool that bridges the technologies of scRNA-seq (transcriptome) and flow cytometry (immunophenotype), allowing for each sequenced single cell to be mapped back to its original cell-surface profile. We have utilized this method here to rigorously quantify the purity of transcriptomically-defined cell types with respect to their isolation strategy based on combinations of cell-surface markers. With a suitable panel of specific antibodies against cell-surface markers, scRNA-seq with index-sorting is a generalizable method for developing and validating sort strategies in other tissues and cell types of interest. While scRNA-seq, index-sorting, and pseudotime analyses are information-rich methods, they can be thought of as highly sophisticated molecular morphologies. Similar to other morphology-based interpretations, these methods are also insufficient to define lineage relationships. Here we demonstrate that transplantation into the lateral ventricles of neonatal immunodeficient mice can reveal at least some site appropriate activities, perhaps allowing direct testing of lineage differentiation potential.

Previous characterization of NSPCs from the developing human brain has often been done without the ability for prospective isolation. Pollen et al. sorted single cells from the germinal zone for culture, and distinguished oRG from IPCs based on their morphology and division kinetics using timelapse imaging¹⁵. Though these experiments yielded valuable insights on their functional differences, the ability to distinguish these cell types prospectively would be desirable. Immunopanning has been another popular method to purify cell types from the brain, a technique in which specific antibodies are adsorbed to the surface of a dish. Subsequently, cell suspensions are serially incubated in said dishes to enrich or deplete cell types of interest²⁶. While this method has been valuable in advancing our understanding of neural cell biology, it has been confounded by nonspecific adsorption of cells to surfaces, and therefore is limited in its ability to precisely and quantitatively isolate pure cell populations based on the expression, especially non-binary expression gradients, of multiple cell-surface markers. For example, Huang et al. isolated a putative “pre-OPC” population via immunopanning with negative selection for PDGFRA followed by positive selection for EGFR²². However, based on our data, such a protocol using FACS would in fact yield ~99% vRG cells (CD24⁻THY1^−/loEGFR⁺), early neuron precursors (CD24⁺THY1^−/loEGFR⁺), and GPCs (THY1^hiEGFR^hiPDGFRA⁻), with only <1% of cells being true pre-OPCs arising from the heterogeneous THY1^hiEGFR⁺PDGFRA⁻ population described earlier (Figure 3B). This highlights the importance of using combinatorial markers, both positive and negative, for isolating truly pure functional NSPC populations.

In vivo transplantation remains the gold standard in stem cell biology to interrogate a cell’s developmental potential. Previous transplantation studies of prenatal human NSCs have generally relied on cultured cell lines^1,8–10,27. In contrast, in this study we orthotopically transplanted acutely purified NSPCs from the developing human brain into mice without any intervening culture. Transplanted CD24⁻THY1^−/lo cells engrafted and gave rise to astrocytes, oligodendrocytes, and neurons, consistent with their identity as multipotent NSCs. Remarkably, we found that THY1^hi cells can also robustly engraft in the mouse brain, and moreover, gave rise exclusively to oligodendrocyte lineages. THY1 is classically thought of as a neuron marker, and its promoter is often used in genetic studies in mice to drive neuron-specific expression. Our findings, however, demonstrate that THY1 in human fetal brains is in fact a marker of oligodendrocyte lineages, at least during the window of human brain development (GW17–19) under investigation in this study. Of course, most studies using THY1 as a neuron marker are in mice, and in addition, it is possible that our tissue processing depletes high THY1 expressing neurons. Nevertheless, our results indicate more nuanced stage-specific variability of marker expression across species and developmental stages.

Identification of a glial progenitor cell in human cortical development

The existence of a bipotent glial progenitor cell (GPC), giving rise exclusively to oligodendrocytes and astrocytes, has previously been speculated. In our study we identify, prospectively isolate, and functionally characterize the GPC within the developing human cortex. In mice, the existence of GFAP⁺OLIG2⁺ cells in the neonatal subventricular zone has been observed²⁸, and histological studies on fetal macaque brains suggest that EGFR⁺OLIG2⁺ radial glia become glial-restricted after E92²⁹. Lineage tracing experiments have shown that OLIG2expressing lineages can give rise to cells expressing markers for either astrocytes or oligodendrocytes³⁰. The strongest functional evidence for a bipotent glial progenitor comes from studies in the rat optic nerve, where it has been shown that bulk or single-cell cultures of A2B5⁺ cells can give rise to either GFAP⁺ astrocytes or GC⁺ oligodendrocytes^31,32. A2B5⁺ cells from cryopreserved human fetal brain cells have also been shown to be expandable in vitro to generate cells that give rise to oligodendrocytes and astrocytes, but not neurons³³. However, because these studies lack clonal analysis, they cannot definitively distinguish between the possibilities that there is (1) a truly bipotent glial progenitor cell type, or (2) a heterogeneous mix of separate oligodendrocyte- and astrocyte-committed cells. Indeed, our data show that while GPCs are A2B5⁺, radial glia, astrocytes, and OPCs are also A2B5⁺ (Figure S4E). Recent droplet-based scRNA-seq studies of the fetal human brain have described transcriptomic cell clusters expressing both astrocyte and oligodendrocyte markers^7,34,35. However, cell types should not be defined solely by their transcriptomic signature, but rather by their function³⁶. The power to prospectively isolate a cell type is critical for functional studies, especially in human tissues where we lack the usual tools of genetic lineage tracing. Only by developing a method to purify the glial progenitor cell (GPC) were we able to functionally demonstrate its bipotent fate, both in vitro, and in vivo. We identify ETV4 (ETS Variant Transcription Factor 4) as a GPC-specific transcription factor. ETV4 has been associated with the progression of various cancers, including glioblastoma, where its overexpression is a poor prognostic marker³⁷. Its role in normal neurodevelopment, however, has not been previously documented. Our initial work provides both a purification strategy and genetic handle on the glial progenitor cell, facilitating further investigation on its role in human brain development.

The results presented here are strong proof of concept that distinct cell types from the developing brain can be prospectively isolated based on the expression of multiple surface markers. The combined results from scRNA-seq, in vitro differentiation, and in vivo transplantation offers the most rigorous interrogation of NSPCs’ transcriptomic identities and functional behavior. The modular nature of our antibody panel allows for the easy expansion or simplification of the panel based on the cell-type of interest, thus providing a valuable tool for future investigation of both molecular and functional NSPC heterogeneity.

Limitations of the study

Our study describes prospective isolation strategies for ten distinct neural cell types. However, further heterogeneity undoubtedly exists within each population, as evidenced by our cell surface marker screen. These additional markers may further stratify our defined cell types into transcriptomically- and functionally-distinct subtypes or substates, all of which may be the focus of future studies. Additionally, it should be noted that surface marker-based cell type identities are often specific to the tissue under investigation both spatially and temporally—in this case, the fetal human brain at mid-gestation. While our gating strategy may certainly guide the development of prospective isolation strategies in other related systems (e.g. brain organoids, postnatal brain tissue, disease states), it should not be blindly applied without prior validation in that new system.

Finally, as noted earlier^1,8 as well as in this study, it is remarkable that human fetal NSPCs transplanted into the brains of newborn NSG mice successfully engraft, migrate, self-renew, and differentiate, despite an estimated 90 million years since a last common ancestor. We speculate that for such functions to be preserved over this interval, several independent modes of selection acting on NSCs are likely operating, so that single mutations or other genetic events do not result in animals with altered properties of stem cells and their immediate progeny. Nevertheless, while this study has allowed many of the human fetal stem cells and primitive downstream progenitors to operate as if they were mouse cells in a mouse brain, strict adherence to this hypothesis beyond the general functions of human fetal NSPCs will need to be demonstrated by experiment.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Rahul Sinha (sinhar@stanford.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• Single cell RNA-seq data have been deposited at GEO and are publicly available as of the date of publication. Accession numbers are listed in the key resources table. Microscopy data reported in this paper will be shared by the lead contact upon request.

• All original code has been deposited at GitHub and is publicly available as of the date of publication. DOIs are listed in the key resources table.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse anti-human CD24-APC, Clone 32D12	Miltenyi	Cat#130–095-954
Mouse anti-human CD90-FITC, Clone 5E10	Biolegend	Cat#328108
Mouse anti-human CXCR4-APC/Cy7, Clone 12G5	Biolegend	Cat#306528
Mouse anti-human EGFR-PE-Cy7, Clone AY13	Biolegend	Cat#352910
Mouse anti-human PDGFRA-biotin, Clone 16A1	Biolegend	Cat#323504
Mouse anti-human CD31-PE, Clone WM59	BD Bioscienc es	Cat#563652
Mouse anti-human CD34-PE, Clone 581	BD Bioscienc es	Cat#562383
Mouse anti-human CD45-PE, Clone HI30	Biolegend	Cat#304010
Mouse anti-human CD105-PE, Clone 266	BD Bioscienc es	Cat#562380
Mouse anti-human CD235a-PE, Clone HIR2	Biolegend	Cat#306606
Mouse anti-A2B5-APC, Clone 105HB29	Miltenyi	Cat#130–098-039
Mouse anti-human LIFR-PE, Clone 12D3	BD Bioscienc es	Cat#566384
Goat anti-human DCX, Clone C-18	Santa Cruz Biotechnol ogy	Cat#sc-8066
Mouse anti-human GFAP, Clone STEM123	Takara	Cat#Y40420
Chicken polyclonal anti-MAP2	Abcam	Cat#ab5392
Rabbit polyclonal anti-Synapsin 1	ThermoFisher Scientific	Cat#A-6442
Mouse anti-O4, Clone O4	R&D Systems	Cat#MAB1326
Mouse anti-human cytoplasmic protein, Clone STEM121	Takara	Cat#Y40410
Rabbit polyclonal anti-SOX2	Abcam	Cat#ab97959
Rabbit anti-OLIG2, Clone EPR2673	Abcam	Cat#ab109186
Rabbit polyclonal anti-NeuN	Abcam	Cat#ab104225
Rabbit anti-EGFR, Clone EP38Y	Abcam	Cat#ab52894
Goat polyclonal anti-human PDGFRA	R&D Systems	Cat#AF-307
Mouse anti-Alpha B Crystallin antibody, Clone 1B6.1–3G4	Abcam	Cat#ab13496
Rabbit polyclonal anti-human HOPX	Sigma	Cat#HPA030180
Rabbit polyclonal anti-ETV4	ThermoFisher Scientific	Cat#PA5–76825
BUV395 Streptavidin	BD Bioscienc es	Cat#564176
Chemicals, Peptides, and Recombinant Proteins
Liberase	Roche	Cat#5401119001
DNase I	Worthingt on	Cat#LS002007
Accutase	Innovative Cell Technologies	Cat#AT104
Polyvinyl alcohol	Sigma	Cat#P8136
Poly-L-ornithine hydrobromide	ThermoFisher Scientific	Cat#P3655
Laminin	ThermoFisher Scientific	Cat#23017015
X-VIVO 15 media	Lonza	Cat#04–744Q
DMEM/F12	ThermoFisher Scientific	Cat#11320082
N-2	ThermoFisher Scientific	Cat#17502048
Heparin	STEMCELL Technologies	Cat#07980
N-acetylcysteine	VWR	Cat#E-3710
Fibroblast growth factor 2	Shenando ah Biotechnology	Cat#100–146
Epidermal growth factor	Shenando ah Biotechnology	Cat#100–26
Leukemia inhibitory factor	Sigma	Cat#LIF1010
Platelet-derived growth factor AA	Shenando ah	Cat#100–16-10ug
	Biotechnol ogy
Insulin-like growth factor 1	PeproTech	Cat#100–11
Neurotrophin-3	PeproTech	Cat#450–03
B27	ThermoFi sher Scientific	Cat#17504044
Paraformaldehyde	Electron Microscopy Sciences	Cat#15710
Normal goat serum	ThermoFi sher Scientific	Cat#50062Z
Triton-X	ThermoFi sher Scientific	Cat#85111
RNase inhibitor	Clontech	Cat#2313B
UltraPure water	ThermoFi sher Scientific	Cat#10977015
Tris-HCl pH 8.5	Teknova	Cat#T5085
Dithiothreitol	Promega	Cat#P1171
GTP	ThermoFi sher Scientific	Cat#R0461
Maxima H-minus reverse transcriptase	ThermoFi sher	Cat#EP0753
KAPA HiFi HotStart ReadyMix	Kapa Biosystems	Cat#KK2602
AMPure XP beads	Beckman Coulter	Cat#A63882
Fast Green dye	Sigma	Cat#F7252
Propidium iodide	Sigma	Cat#P4170
ProLong Gold Antifade Mountant	ThermoFisher Scientific	Cat#P36934
Chlorinated water	Innovive	Cat#M-WB-300C
Teklad global 18% protein rodent diet	Envigo	Cat#2918
Teklad global 19% protein extruded rodent diet	Envigo	Cat#2919
Critical Commercial Assays
LEGENDScreen™ Human PE Kit	Biolegend	Cat#700007
Deposited Data
scRNA-seq data	This paper	PRJNA798712
Original code	This paper	https://github.com/transcriptomics/NSPC
Experimental Models: Organisms/Strains
Mouse: NSG (NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ)	The Jackson Laboratory	JAX: 005557
Human: human fetal samples	Advanced Bioscience Resources	N/A
Oligonucleotides
Oligo-dT30VN (5’-AAGCAGTGGTATCAACGCAGAGTACT30VN-3’)	IDT	N/A
Template-switching oligonucleotide (TSO) (5’-AAGCAGTGGTATCAACGCAGAGTA CATrGrG+G-3’)	IDT	N/A
ISPCR primers (5’-AAGCAGTGGTATCAACGCAGAGT-3’)	IDT	N/A
ERCC (External RNA Controls Consortium) ExFold RNA Spike-In Mixes	Invitrogen	Cat#4456739
dNTP	Invitrogen	Cat#10297–018
Software and Algorithms
ImageJ	NIH	http://wsr.imagej.net/distros/
GraphPad Prism 9.0	GraphPad Software	http://www.graphpad.com/scientific-software/prism
Fragment Analyzer Controller	Agilent	https://www.agilent.com/en/product/automatedelectrophoresis/fragment-analyzer-systems/fragmentanalyzer-systems-software/fragment-analyzersoftware-1149185
Prosize	Agilent	https://www.agilent.com/en/product/automatedelectrophoresis/fragment-analyzer-systems/fragmentanalyzer-systems-software/fragment-analyzersoftware-1149185
BD FACSDiva	BD Biosciences	https://www.bdbiosciences.com/en-us/products/software/instrument-software/bd-facsdivasoftware
FlowJo	FLOWJO LLC	https://www.flowjo.com/
Mantis	Formulatrix	https://formulatrix.com/life-science-automationblog/liquid-handling-software-update-mantis-47/
Mosquito	SPT Labtech	https://www.sptlabtech.com/products/mosquito
bcl2fastq2 2.19	Illumina	https://support.illumina.com/sequencing/sequencing_software/bcl2fastq-conversion-software.html
Skewer 0.2.2	Jiang et al., 201439	https://sourceforge.net/projects/skewer
STAR 2.6	Dobin et al., 201340	https://github.com/alexdobin/STAR
RSEM 1.3.3	Li and Dewey, 201141	https://deweylab.github.io/RSEM/
Scanpy 1.8.2	Wolf et al., 201842	https://scanpy.readthedocs.io/en/stable/
Extreme Limiting Dilution Analysis	Hu & Smyth, 200938	https://bioinf.wehi.edu.au/software/elda/

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

All mouse studies were performed strictly in adherence to the protocols and procedures approved by the Administrative Panel on Laboratory Animal Care (APLAC) at Stanford University and the NIH Guidelines for the Care and Use of Laboratory Animals. All mice were housed at Stanford’s mouse facility under supervision of the Stanford Veterinary Service Center (VSC). All mice were monitored on a regular basis to ensure good health until the termination of the study.

Immunodeficient NSG (NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ, JAX: 005557) mice were used as recipients for transplantation of human NSPCs. Mice were maintained in a 12:12 light/dark cycle at 18–23°C and 40–60% humidity, given bedding (FiberCore), and provided food and water ad libitum. Feed consisted of chlorinated water (Innovive, cat. #M-WB-300C) and an irradiated 19% protein / 9% fat rodent diet for pregnant/nursing dams (Envigo, cat. #2919) or a 18% protein / 6% fat rodent diet after weaning (Envigo, cat. #2918). All mice were transplanted between postnatal day 1–2 with no restriction on sex. Litters were weaned at 21 days, separated by sex, and housed at a maximum of 5 mice per cage. Transplanted mice were sacrificed at 6 months post-injection.

Human samples

Human fetal brain samples were obtained from Advanced Bioscience Resources (ABR, Alameda, CA) and shipped overnight in BioWhittaker RPMI-1640 media supplemented with L-glutamine. Samples ranged in age from 17 to 19 weeks of gestation (GW17–19) with no restrictions on sex or race. Information regarding sex, health/immune status, and previous procedures or drug exposures was not available for fetal samples. All fetal samples were processed within 6–8 hours after delivery and used immediately for live cell sorting.

No Institutional Review Board (IRB) approval was required for procurement of deidentified samples from ABR as per Stanford University guidelines. All subsequent in vitro and in vivo experiments that utilized cells from fetal brain tissues thus obtained were performed strictly as per pre-approved guidelines set by Stem Cell Research Oversight (SCRO) (SCRO protocol #735 and APLAC protocol #26209) at Stanford university.

Cell Lines

To generate neurospheres, acutely isolated NSPCs were cultured at a density of 10⁵ cells/mL in fetal growth media (FGM) consisting of X-VIVO 15 media (Lonza, cat. #04–744Q) supplemented with N-2 (ThermoFisher Scientific, cat. #17502048), heparin (STEMCELL Technologies, cat. #07980), N-acetylcysteine (VWR, cat. #E-3710), 20 ng/mL fibroblast growth factor 2 (FGF2) (Shenandoah Biotechnology, cat. #100–146), 20 ng/mL epidermal growth factor (EGF) (Shenandoah Biotechnology, cat. #100–26), and 10 ng/mL leukemia inhibitory factor (LIF) (Sigma, cat. #LIF1010). No established cell lines were used in this study.

METHOD DETAILS

Human fetal brain dissociation

Deidentified prenatal human brain samples were obtained from Advanced Bioscience Resources (Newark, California) and shipped overnight. The age range for samples was 17 to 19 gestational weeks with no restrictions on race or sex. Prenatal brain sample procurement and handling was done in accordance with guidelines set by the Stanford Institutional Review Board (IRB) and the Stanford Stem Cell Research Oversight (SCRO) Panel. Intact samples were dissected into distinct anatomical regions by licensed neuropathologists. Samples were gently chopped with a razor blade and resuspended in digestion buffer A, which consists of Hank’s balanced salt solution (HBSS) (Thermo, cat. 24020117) with 10 μg/mL Liberase (Roche, cat. #5401119001) and 200 μg/mL DNase I (Worthington, cat. #LS002007). The sample was then incubated twice at 37°C under constant agitation for 40 minutes. Afterwards, cells were spun down and resuspended in digestion buffer B, consisting of Accutase (Innovative Cell Technologies, cat. #AT104) supplemented with 200 μg/mL DNase I, then incubated at 37°C for 15 minutes under constant agitation. Red blood cell removal was performed using a density gradient by layering the cell suspension on top of Histopaque (Sigma, cat. #10771) at a 2:1 ratio, then centrifuging at 400×g for 30 minutes at 25°C with low acceleration and no brakes. The buffy coat at the fluid interface was collected, washed, and counted before staining with fluorochrome-conjugated antibodies. All washes were performed in HBSS supplemented with 0.1% polyvinyl alcohol (PVA) (Sigma, cat. #P8136).

Fluorescence-activated cell sorting (FACS)

Brain tissue was dissociated into a single cell suspension as described above in the “Human fetal brain dissociation” section. The cell suspension was stained with fluorochrome-conjugated antibodies against CD133 (clone 13H2), CD24 (clone 32D12, Miltenyi, cat. #130–095-954), CD90 (clone 5E10, Biolegend, cat. #328108), CXCR4 (clone 12G5, Biolegend, cat. #306528), EGFR (clone AY13, Biolegend, cat. #352910), PDGFRA (clone 16A1, Biolegend, cat. #323504), CD31 (clone WM59, BD Biosciences, cat. #563652), CD34 (clone 581, BD Biosciences, cat. #562383), CD45 (clone HI30, Biolegend, cat. #304010), CD105 (clone 266, BD Biosciences, cat. #562380), and CD235a (clone HIR2, Biolegend, cat. #306606), all at a dilution of 1:50. Propidium iodide (PI) (Sigma, cat. P4170) was added as a viability marker immediately prior to analysis (1 μg/mL). Flow cytometry was performed on a FACS Aria II (BD Biosciences). Debris was excluded using forward and side scatter area, and doublets were excluded using a stringent two-step gating based on forward and side scatter height versus width. Gating schemes were guided by fluorescence-minus-one (FMO) controls. The highest purity setting (“single cell”) was used to collect cells for scRNA-seq and limiting dilution neurosphere initiation assays. The “4-way purity” setting was used for bulk sorting of cells for in vivo transplant and in vitro differentiation assays.

Surfaceome profiling of human fetal brain cells

Brain tissue was dissociated into a single cell suspension and prepared for flow cytometry as described above in the “Human fetal brain dissociation.” Cells were stained with the antibody panel described above in “Fluorescence activated cell sorting (FACS),” omitting the antibody against CD133 to free up the PE channel. Cells were screened for 352 additional surface markers using the LEGENDScreen™ Human PE Kit (Biolegend, cat. #700007) per manufacturer protocol. Cells were analyzed on a BD FACSymphony A5 in 96-well plate format. An average of 130,000 events were recorded for each marker. The percentage of cells staining positive for each marker was determined for the corresponding parent gate using the appropriate isotype control.

For the comparison between antigen and RNA expression, each surface marker on the LEGENDScreen panel was matched to its corresponding gene. Glycan antigens were matched to the gene responsible for their synthesis (e.g. FUT4 for CD15, B3GAT1 for CD57). The percentage of antigen-positive cells within each immunophenotypic gate was correlated with the percentage of RNA-expressing cells within the corresponding transcriptomic cluster for that gate.

Single cell RNA sequencing (scRNA-seq)

Single cell capture and RNA extraction

Brain tissue was dissociated into a single cell suspension and prepared for FACS as described above in the “Human fetal brain dissociation” and “Fluorescence activated cell sorting (FACS)” sections above. Single cells were index-sorted into 96-well plates containing 2 μL lysis buffer (1 U/μL RNase inhibitor (Clontech, cat. #2313B), 0.1% Triton (ThermoFisher Scientific, cat. #85111), 2.5 mM dNTP (Thermo Fisher Scientific, cat. #10297018), 2.5 μM oligo dT₃₀VN (Integrated DNA Technologies, 5′-AAGCAGTGGTATCAACGCAGAGTACT₃₀VN-3’), and 1:600,000 ERCC (external RNA controls consortium) RNA spike-in mix at 1:600,000 (Thermo Fisher Scientific, cat. #4456739) in UltraPure water (ThermoFisher Scientific, cat. #10977015)¹⁷. Immediately after sorting, plates were centrifuged at 3000×g for 30 seconds at 4°C, snap frozen on dry ice, and stored at −80°C.

Reverse transcription and pre-amplification

Reverse transcription (RT) and cDNA pre-amplification was performed using the Smart-seq3 protocol with minor modifications¹⁸. In brief, lysis plates were thawed on ice and incubated at 72°C for 3 minutes then immediately snap chilled on ice to anneal the oligo dT₃₀VN primer. For reverse transcription, 3 μL of RT mix (25 mM Tris-HCl pH 8.5 (Teknova, cat. #T5085), 0.5 U/μL RNase inhibitor (Clontech, cat. #2313B), 8 mM dithiothreitol (DTT) (Promega, cat. #P1171), 30 mM NaCl (Thermo Fisher Scientific, cat. #AM9760G), 2.5 mM MgCl₂ (Thermo Fisher Scientific, cat. #AM9530G), 1 mM GTP (ThermoFisher Scientific, cat. #R0461), 2 μM TSO (Integrated DNA Technologies, 5′-AAGCAGTGGTATCAACGCAGAGTGAATrGrGrG-3′), 5% polyethylene glycol (Sigma, cat. #P1458), and 2 U/μL Maxima H-minus reverse transcriptase (ThermoFisher Scientific, cat. #EP0753) in UltraPure water) was added to each well using a Mantis liquid handler (Formulatrix), and incubated in a C1000 Touch Thermal Cycler (BioRad) at 42°C for 90 min, and then 70°C for 15 min to terminate the reaction. Afterwards for preamplification, 7.5 uL of PCR mix (1.67X KAPA HiFi HotStart ReadyMix (Kapa Biosystems, cat. #KK2602) and 0.17 μM IS PCR primer (Integrated DNA Technologies, 5′-AAGCAGTGGTATCAACGCAGAGT-3′) in UltraPure water) was added to each well and cycled using the following program: (1) 98°C for 3 min, (2) denaturing at 98°C for 20 sec, (3) annealing at 67°C for 15 sec, (4) elongation at 72°C for 6 minutes, (5) repeat from step 2 24 times, and (6) 72°C for 5 minutes. Preamplified cDNA was then purified using 0.65–0.75X volume of calibrated AMPure XP beads (Beckman Coulter, cat. #A63882) to remove residual reaction components and oligos smaller than 400 base pairs, and eluted in 12.5 μL UltraPure water.

Quality control

From the purified cDNA, 1 μL was taken for quality control for each well. cDNA concentration and size distribution for each well was determined on a capillary electrophoresis-based Fragment Analyzer (Advanced Analytical). Wells with a concentration less than 1.7 ng/μL were excluded; this cutoff was determined by measuring the concentration blank wells with ERCC but no sorted cell. The wells within the 96 well plates with cDNA concentration above the cutoff value were then consolidated and reformatted to a new 384 well plate using the Mosquito X1 liquid handler (SPT Labtech), such that in the destination 384 well plate each well’s concentration was also normalized to a desirable concentration range of 1.7–4.0 ng/μL by diluting with UltraPure water.

Library preparation and sequencing

Normalized cDNA was used to prepare Illumina sequencing libraries. Tagmentation was performed by combining 0.4 μL cDNA with 1.2 μL homebrew Tn5 mix consisting of1 ng/μL Tn5 enzyme, 16 mM Tris-HCl pH 7.6, 16 mM MgCl₂ (Thermo Fisher Scientific, cat. #AM9530G) and 8% dimethylformamide (DMF) (Thermo Fisher Scientific, cat. #AC327171000) in UltraPure water. The reaction was stopped by adding 0.4 μL neutralization buffer (0.1% SDS). Indexing PCR reactions were performed by adding 0.4 μL of 5 μM i5 indexing primer, 0.4 μL of 5 μM i7 indexing primer (Integrated DNA Technologies, custom made 7680-plex unique dual index-primer set), and 1.2 μL KAPA HiFi HotStart ReadyMix (Kapa Biosystems, cat. #KK2602). PCR amplification was performed on a C1000 Touch Thermal Cycler (BioRad) using the following program: (1) 72°C for 3 min, (2) 95°C for 30 sec, (3) denaturing at 98°C for 10 sec, (4) annealing at 67°C for 30 sec, (5) elongation at 72°C for 60 sec, (6) repeat from step 3 10 times. For each 384 well plate, 1 μL was taken from each well for pooling, followed by purification using 0.8X volume of AMPure XP beads (Beckman Coulter, cat. #A63882). The 384-cell library pool from each plate was analyzed for concentration and size distribution using Fragment Analyzer (Advanced Analytical). Twenty 384-cell library pools were then normalized for concentration, further pooled to get a 7680-plex library pool, which was purified once more and concentrated using 0.8X AMPure beads. The final 7680-plex library pool was then sequenced on a NovaSeq 6000 S4 flow cell (Illumina) to obtain ~1–2 million 2×150 base-pair paired-end reads per cell.

In vivo transplantation of NSPCs

NSPCs were purified using FACS and resuspended in 2 μL HBSS with Fast Green dye (Sigma, cat. #F7252) for better visualization of the injected site. Neonatal mice (postnatal day 1–2) were randomized, anesthetized using hypothermia, and placed on a stereotaxic device (Harvard Apparatus) fitted with a mouse neonate adaptor (Cunningham). Light illumination was used to identify the sinus above Lambda as the reference point. Burr holes were made in the skull cartilage at the injection site using a 30-gauge needle. The cell suspension was injected using a Hamilton syringe with a 33-gauge needle into the lateral ventricles using the Micro4 microsyringe pump controller (World Prevision Instruments) at a rate of 1 μL/min, with 1 μL injected per side. Successful injection into the ventricles was confirmed visually with light illumination. The following coordinates were used for transplant; anteroposterior from midline (A), lateral from midline (L), ventral from surface of brain (V). Lateral ventricles (A, L, V) = (0.8, ±1.5, 2.0) mm with reference to lambda. All in vivo experiments described in this section that utilized cells from fetal brain tissues were performed strictly as per pre-approved guidelines set by Stem Cell Research Oversight (SCRO) at Stanford university (SCRO protocol #735; also see above).

Limiting dilution neurosphere initiation assay

Limiting dilution assays were conducted to quantify the frequency of cells in a population that initiate neurospheres. Known numbers of primary NSPCs were sorted at single cell purity into 96 well plates containing 100 μL of FGM (see above). Half of the media was replaced with fresh media every week. Plates were scored at 4 weeks, with wells containing at least one neurosphere being considered positive. Linear regression analysis of the proportion of positive wells at each cell concentration was used to determine the neurosphere initiation frequency³⁸.

In vitro differentiation of cells

For differentiation assays, primary NSPC populations were bulk-sorted on the 4-way purity setting, and directly plated into 96 well plates coated for 12 hours with poly-L-ornithine hydrobromide (Millipore Sigma, cat. #P3655) and laminin mouse protein (ThermoFisher Scientific, cat. #23017015) in differentiation media consisting of DMEM/F12 (ThermoFisher Scientific, cat. #11320082), heparin, N-acetylcysteine, N-2, and B27 (ThermoFisher Scientific, cat. #17504044), but no other growth factors or cytokines. For OPC induction media²³, the following growth factors were added to the differentiation media at 10 ng/mL: PDGF-AA (Shenandoah Biotech, cat. #100–16-10ug), IGF1 (PeproTech, cat. #100–11), NT3 (PeproTech, cat. #450–03). For clonal differentiation assays, we index-sorted single cells into 96 well plates containing FGM, and cultured them for 4 weeks to obtain clonally-derived neurospheres. Single cell-derived individual neurospheres were then dissociated, and either reanalyzed via flow cytometry or plated onto polyornithine/laminin-coated wells in differentiation media. After 4 days of in vitro culture, cells were fixed using 4% paraformaldehyde (PFA), and then stained with antibodies for immunofluorescence (IF). Clonal differentiation results were replicated on at least 3 donors.

Histology

Mice were sacrificed 6 months post-transplant and perfused using phosphate buffered saline (PBS) supplemented with 20 mM EDTA. The brain was dissected out and fixed in freshly prepared 4% paraformaldehyde (PFA) solution in PBS (Electron Microscopy Sciences, cat. #15710) for 16 hours and then transferred to 30% sucrose solution for cryoprotection. Afterwards, 40 μm sagittal sections of the brains were sliced on a Leica SM2010 R Sliding Microtome. Tissue slices were kept as floating sections in PBS at 4°C until ready for staining.

For fetal human brain sections, cortical tissues were fixed in 4% PFA for 16 hours, transferred to 30% sucrose solution for cryoprotection, then embedded in optimal cutting temperature (OCT) compound (Tissue-Tek^®) and frozen at −20°C. Afterwards, 14 μm coronal sections were sliced on a Microm HM550 cryostat (ThermoFisher Scientific) and mounted on slides.

Immunofluorescence

For immunofluorescence (IF) of cultured samples, cells were fixed using 4% PFA for 10 minutes at 4°C, then washed 3 times with PBS. Samples were permeabilized and blocked in PBS supplemented with 0.3% Triton and 3% normal goat serum (ThermoFisher Scientific, cat. #50062Z) for 1 hour at room temperature. Primary antibodies were diluted in antibody diluent (PBS supplemented with 0.3% Triton and 1% normal goat serum) and incubated with samples for 2 hours at room temperature. Primary antibodies used include: DCX (1:500, Santa Cruz Biotechnology, cat. #sc-8066), human GFAP (1:1000, Takara, clone STEM123, cat. #Y40420), MAP2 (1:5000, Abcam, cat. #ab5392), SYN1 (1:1000, ThermoFisher Scientific, cat. #A-6442), O4 (1:500, R&D Systems, cat. #MAB1326). Samples were washed 3 times using PBS, then incubated with the appropriate secondary antibodies diluted in antibody diluent for 1 hour at room temperature (ThermoFisher Scientific). Samples were incubated in DAPI (1 μg/mL in PBS) and washed 3 times prior to imaging on a Leica DMi8 inverted microscope. For O4 staining, primary antibody was localized to the cells prior to fixation.

IF of fixed mouse brain sections was carried out in a similar manner. Samples were blocked and permeabilized, then incubated with primary antibodies diluted in antibody diluent overnight at 4°C with agitation. Primary antibodies used include: human cytoplasmic antigen (1:1000, Takara, clone STEM121, cat. #Y40410), human GFAP (1:1000, Takara, clone STEM123, cat. #Y40420), SOX2 (1:1000, Abcam, cat. #ab97959), OLIG2 (1:500, Abcam, cat. #ab109186), MAP2 (1:5000, Abcam, cat. #ab5392), NeuN (1:1000, Abcam, cat. #ab104225). Following incubation with secondary antibodies and DAPI, sections were floated onto slides and mounted using ProLong Gold Antifade Mountant (ThermoFisher Scientific, cat. #P36934). Imaging was done on a Leica Stellaris 8 confocal microscope.

IF of fixed human fetal brain sections was carried out similarly, except in a slide mounted format. Following sectioning and mounting onto slides, Optimal Cutting Temperature OCT compound (Tissue-Tek^®) was washed off with 3 changes of PBS. Tissue sections were outlined with a PAP pen (Vector Laboratories, cat. #H-4000), then stained as detailed above. Primary antibodies used include: human GFAP (1:1000, Takara, clone STEM123, cat. #Y40420), OLIG2 (1:500, Abcam, cat. #ab109186; 1:200, R&D Systems, cat. #AF2418), EGFR (1:500, Abcam, cat. #ab52894), PDGFRA (1:500, R&D Systems, cat. #AF-307), CRYAB (1:300, Abcam, cat. #ab13496), HOPX (1:1000, Sigma, cat. #HPA030180), ETV4 (ThermoFisher Scientific, 1:200, cat. #PA5–76825).

Quantification of IF images was performed manually using ImageJ, during which researchers were blinded to experimental group.

QUANTIFICATION AND STATISTICAL ANALYSIS

Single cell RNA-seq

Read mapping

Sequences were demultiplexed using bcl2fastq version 2.19.0.316. 3’ adapter sequences were removed from reads using skewer v0.2.2³⁹, and aligned to the hg38 genome (Gencode version GRCh38.p13) with STAR aligner version 2.6.1d using 2-pass mapping⁴⁰. Briefly, as a first pass, reads for every cell were aligned using STAR genome index generated using the Gencode transcript annotation for the human genome (version 34). Mapped splice junctions for each cell from the first-pass mapping were extracted, aggregated together and a new STAR index was created where any newly discovered splice junctions were included in addition to the existing Gencode annotation during genome index generation. The new STAR index with all known and newly identified splice-junctions was then used for second pass read mapping. Parameters used for STAR mapping were adapted from the ENCODE long-mRNA-pipeline (https://github.com/ENCODE-DCC/long-read-rna-pipeline) recommendations, also detailed in the STAR manual. In addition to the ENCODE recommended options we also used the “--quantMode TranscriptomeSAM” option during second-pass mapping to generate a bam file containing a catalog of all reads mapped to the transcriptome. This bam file was used as input to calculate expression levels of either genes (sum total of expression levels of all known transcript variants) or individual transcripts using RSEM version 1.3.3 with settings “--single-cell-prior”, which accounts for the sparse nature of mRNA detection usually prevalent in scRNA-seq⁴¹.

Data preprocessing

Gene count tables were combined with metadata using the Scanpy python package v.1.8.2⁴². We filtered out genes expressed in fewer than 3 cells, as well as cells with fewer than 500 detected genes or 5000 read counts. The data were normalized using size factor normalization so that every cell has 10,000 read counts, log transformed, and scaled to a maximum value of 10. Highly-variable genes were computed using default parameters. We then performed principle component analysis, computed the neighborhood graph, and clustered the data using the Leiden method⁴³. PAGA was used to visualize data, as well as reconstruct gene expression changes along maturation trajectories⁴⁴. A total of 9,454 cells across four donors were included in the final analysis, with an average gene count of 3962. Step-by-step instructions to reproduce preprocessing and analysis of data are available on GitHub.

k-nearest neighbor classifier

To determine the anatomical origin of cells sequenced from segmented samples (where dissection of the cortex prior to dissociation was not possible), we implemented a k-nearest neighbor classifier, using our intact sample for ground truth anatomical labels. Each cell of unknown anatomical origin was classified to a cortical or subcortical identity based on a plurality vote of its k most similar neighbors of known anatomical origin (k = 5).

Supplementary Material

1

Figure S1. Index sorting and scRNA-seq of NSPCs, related to Figure 1

(A) Pre-gating scheme for live, lineage negative single cells. Debris was removed based on forward and side scatter area, followed by two-step doublet discrimination using forward and side scatter height and width. Dead cells were removed based on staining for propidium iodide, and non-neural lineages were removed based on staining for fluorophore-conjugated antibodies against PECAM1 (CD31), CD34, PTPRC (CD45), ENG (CD105), or GYPA (CD235a).

(B) Dot plot showing marker gene expression for transcriptomically-defined clusters from single cell RNA sequencing.

(C) Scatter plots showing correspondence between RNA and protein expression in single cells. Cells are colored by their transcriptomically-defined clusters.

(D) Index-sort data was used to map the sequenced single cells to their original immunophenotype with respect to CD133 and CD24 cell-surface expression.

(E) Violin plot showing CD133 cell-surface protein expression as measured by flow cytometry in each transcriptomically-defined cluster as shown in Figure 1C.

2

Figure S2. Characterization of CD24⁻THY1^−/lo radial glia and astrocytes, related to Figure 2

(A) Flow cytometric analysis of cell-surface EGFR and LIFR expression within CD24⁻THY1^−/lo (orange) gate.

(B) Confocal immunofluorescence (IF) imaging of fetal human cerebral cortex (18 gestational weeks) stained for DAPI (blue), GFAP (red), EGFR (green), and CRYAB (magenta). The boxed region shown at higher resolution on the right. Labelled regions: ventricular zone (VZ), subventricular zone (SVZ), outer subventricular zone (OSVZ), cortical plate (CP). Scale bar 500 μm.

(C) Computationally predicted cell cycle status in CD24⁻THY1^−/lo subsets.

(D) Intracellular staining for GFAP, SOX1, SOX2, and PAX6 in CD24⁺THY1^−/lo (blue), THY1^hi (green), and CD24⁻THY1^−/lo (orange) gates, besides corresponding isotype controls (grey).

(E) Experimental strategy for in vitro clonal neurosphere differentiation assays. Single CD24⁻THY1^−/lo cells that were either EGFR⁻ or EGFR⁺ were sorted into a well and cultured for 4 weeks to generate clonal neurospheres, which were then dissociated and subjected to differentiation.

(F) IF images of clonally derived cells stained with either (left) anti-O4 (green), or (right) anti-DCX (green) and anti-GFAP (red) antibodies. Scale bar 50 μm.

(G) IF images of sorted CD24⁻THY1^−/loEGFR⁺CXCR4⁺ cells subject to 5 days of in vitro differentiation and then stained for GFAP (red) and AQP4 (green). Scale bar 50 μm.

3

Figure S3. Characterization of THY1^hi oligodendrocyte precursors, related to Figure 3

(A) Index-sort analysis of vRG and oRG cells found within the heterogeneous THY1^hiEGFR⁺PDGFRA⁻ gate.

(B) PAGA pseudotime analysis showing the oligodendrocyte maturation trajectory.

(C) PAGA pseudotime distribution within each THY1^hi subset.

(D) Computationally predicted cell cycle status in THY1^hi subsets.

(E) Confocal immunofluorescence (IF) images of fetal human cerebral cortex (18 gestational weeks) stained for EGFR (green), PDGFRA (red), and DAPI (blue). Scale bar 50 μm.

(F) Quantification of O4⁺ cells among bulk sorted CD24⁻THY1^−/lo and CD24⁺THY1^−/lo populations cultured for 5 days.

(G) Flow cytometric analysis of sorted THY1^hiEGFR⁺PDGFRA⁺ and THY1^hiEGFR⁺PDGFRA⁻ cells that were cultured in vitro for 5 days. Each row represents a separate donor.

4

Figure S4. Characterization of the glial progenitor cell, related to Figure 4

(A) mRNA expression matrix showing glial progenitor cell (GPC) marker genes and their expression in other cell types.

(B) Immunofluorescent (IF) imaging of sorted CD24⁻THY1^−/lo cells after being subjected to 5 days of in vitro differentiation. Cells were stained for DAPI (blue), GFAP (red), and OLIG2 (green). Scale bar 50 μm.

(C) Plot showing UMAP representation of single cell RNA sequencing of fetal NSPCs, with GPC cluster highlighted (left), and expression plots for ETV4, ADM, and METTL7B (right).

(D) Quantification of lineage output of clonal neurospheres derived from THY1^hiEGFR^hiPDGFRA⁻ cells, differentiated via either cytokine withdrawal or in OPC induction media. Each column represents a distinct clonal neurosphere. Differentiated cells were classified based on their expression of GFAP, DCX, or OLIG2.

(E) Flow cytometric analysis of A2B5 expression on NSPC subsets.

5

Figure S5. Characterization of NSPCs across brain regions, related to Figure 6.

(A) Single cell RNA sequencing of index-sorted NSPCs from dissected brain regions using Smart-seq3. Plot showing UMAP-embedded Leiden clusters with annotated cell type identities based on expressed transcripts of known genes.

(B) Stacked bar graph showing distribution of each transcriptomic cell type across brain regions.

(C) Plot showing the quantification of neurosphere initiation frequency of CD24⁻THY1^−/loEGFR⁻ (orange) and CD24⁻THY1^−/loEGFR^hi (red) cells from each brain region based on limiting dilution assays. Numbers underneath denote the reciprocal of the neurosphere initiation frequency. Error bars represent 95% confidence interval.

6

Figure S6. Frequencies of immunophenotypic populations by brain region, related to Figure 6.

(A) Dissected regions of fetal human brain.

(B) Representative flow cytometry plots showing the expression of THY1 and CD24 in cells derived from the subventricular zone or the thalamus. Events are pre-gated on live, lineage (blood and vessel) negative single cells.

(C) Frequency of THY1^hi, CD24⁻THY1^−/lo, and CD24⁺THY1^−/lo cells among live, lineage negative single cells.

(D) Frequency of EGFR⁻, EGFR^mid, EGFR^hi, and EGFR⁺CXCR4⁺ cells among CD24⁻THY1^−/lo cells.

(E) Frequency of EGFR^hiPDGFRA⁻, EGFR⁺PDGFRA⁺, EGFR⁻PDGFRA⁺, and EGFR⁻PDGFRA⁻ cells among THY1^hi cells.

(F) Frequency of CXCR4⁻EGFR⁻, CXCR4⁻EGFR⁺, and CXCR4⁺ cells among CD24⁺THY1^−/lo cells.

Numbers shown below each bar denotes percentage of parent population.

7

Figure S7. Surfaceomic profiling of NSPC types, related to Figure 7.

(A) Surface marker expression within each immunophenotypic population, presented as the percent of cells within the parent gate staining positive for that marker. Each bar graph represents one marker, with each bar representing an immunophenotypic population as denoted in the legend.

(B) Heat map showing the percentage of cells within each immunophenotypic population (rows) that stained positive for each surface marker (columns).

(C) Scatter plots showing correlation between RNA and surface antigen expression, plotted for each cell type. Each dot represents a marker, with the y-axis denoting the percentage of cells within the immunophenotypic gate staining positive for surface antigen, and the x-axis denoting the percentage of cells within the transcriptomic cluster expressing transcript for the marker.

(D) Scatter plot of markers, arranged on the x-axis by the percentage of cells staining positive for surface antigen (averaged across all 10 populations), and on the y-axis by the absolute difference between RNA and protein percent positivity, summed across all 10 populations.

8

Table S1. Classification of cells to brain regions, related to Figure 6. k-nearest neighbor (k-NN) classifier results, showing percentage of cells from each transcriptomic cluster classified to either a cortical or subcortical identity, for each of the 3 segmented samples in our dataset.

9

Table S2. Surface antigen profiling of NSPC populations, related to Figure 7. Percentage of cells within each immunophenotypic gate (columns) staining positive for each surface antigen (rows). Percent positivity was gated based on the corresponding isotype control of each antibody.

Highlights.

• Purification, functional validation of 10 NSPC types in human neurodevelopment.

• Index sorting maps transcriptome onto immunophenotype to derive gating strategy.

• Identification of transcriptomically, functionally defined glial progenitor cell (GPC).

• GPCs likely arise from oRG, give rise to oligodendrocytes/astrocytes but not neurons.