Research ArticleDEVELOPMENTAL NEUROSCIENCE

Single-cell transcriptome analysis reveals cell lineage specification in temporal-spatial patterns in human cortical development

See allHide authors and affiliations

Science Advances  21 Aug 2020:
Vol. 6, no. 34, eaaz2978
DOI: 10.1126/sciadv.aaz2978

Abstract

Neurogenesis processes differ in different areas of the cortex in many species, including humans. Here, we performed single-cell transcriptome profiling of the four cortical lobes and pons during human embryonic and fetal development. We identified distinct subtypes of neural progenitor cells (NPCs) and their molecular signatures, including a group of previously unidentified transient NPCs. We specified the neurogenesis path and molecular regulations of the human deep-layer, upper-layer, and mature neurons. Neurons showed clear spatial and temporal distinctions, while glial cells of different origins showed development patterns similar to those of mice, and we captured the developmental trajectory of oligodendrocyte lineage cells until the human mid-fetal stage. Additionally, we verified region-specific characteristics of neurons in the cortex, including their distinct electrophysiological features. With systematic single-cell analysis, we decoded human neuronal development in temporal and spatial dimensions from GW7 to GW28, offering deeper insights into the molecular regulations underlying human neurogenesis and cortical development.

INTRODUCTION

The human brain contains billions of neurons that were originally generated from neuroepithelial cells (1). The cerebral cortex (CC) can be divided into the following lobes: the frontal lobe (FL), parietal lobe (PL), occipital lobe (OL), and temporal lobe (TL), with each showing specialized functions in sensory and motor control and having specific projections to different targets of the nervous system (2). The cellular landscape for molecular census at single-cell resolution has been fully delineated in the developing and adult mouse nervous system (36). The electrophysiological and morphologic characters of these systems have also been identified (7, 8). In humans, cellular anatomy has been partially revealed in recent years (912). Our group and others have either illustrated the developmental process of a specific brain region, for example, the human prefrontal cortex (10), or randomly collected samples just to learn the cellular types and developmental regulations (9). Comparisons between the human developmental prefrontal cortex and visual cortex (12) offered a brief view of distinct molecular regulations and developmental path between the caudal and rostral regions of the CC. This difference might lead to the disproportionate expansion of different cortical regions, which is also reported in other species (13). Nevertheless, we still lack a comprehensive view of neuronal development in the developing human cortex based on pairwise regional information within cortical comparisons and outside cortical controls, with sufficient sample size and data resolution to illustrate the diversity of cell states/populations during neurogenesis at the early fetal stage to neuronal maturation on an accurate temporal scale.

Neural development includes not only neuron differentiation and maturation but also glial cell development. The molecular architecture of glial cells in the mouse nervous system has been systematically investigated (3, 14, 15), while this process remains largely unexplored in humans. In this study, we used single-cell RNA sequencing (RNA-seq) to survey cells from different regions of developing human cortices together with the pons as a control to capture the global characteristics of the CC. In light of the comprehensive dataset, we specified the time-dependent neuronal events and enlarged the neural progenitor cell (NPC) pool with newly identified transient cells. We found that regional differences on the lateral side of the developing cortex appeared more conspicuous during the neuron maturation stage by exhibiting more intense electrophysiological properties in the FL. Moreover, we summarized the different regional molecular regulations within each cell type and emphasized the important role of long intergenic noncoding RNA (lincRNA) in regional and cell type maintenance.

RESULTS

Cell types revealed developmental patterns in human fetal brain development

We collected 14,239 individual cells from embryos ranging from gestational week 9 (GW9) to GW28 for single-cell RNA-seq analysis of the four lobes of the CC and pons when vigorous neurogenesis occurs in the human brain (Fig. 1A). Two sets of earlier embryonic brain samples (GW7 and GW8) without separation of the regions due to their tiny sizes were also collected. After stringent filtration, the transcriptome data of a total of 12,541 high-quality single-cell samples (on average 4410 genes and 127,324 transcripts detected in each individual cell; fig. S1A) were retained for further analysis. The number of unique molecular identifier (UMI) caused little clustering artifacts as cells of different clusters showed almost no difference on detected number of UMI (fig. S1B). There were minimal potential batch effects among the samples because cells were clustered by types instead of by individuals, and biological replicates at the same developmental stage showed high correlation coefficients for gene expression; in addition, housekeeping genes were consistently detected in every cell from all samples (fig. S1, C to E). Combined with previously published cortical single-cell RNA-seq data (11), we identified 33 clusters of cells using unsupervised clustering analysis (see more details in the Supplementary Materials). These clusters included four clusters of NPCs; three clusters of cortical excitatory neurons (CC-EX); two clusters of excitatory neuron-like cells in the pons (pons_neu) and Cajal-Retzius cells (CRs); two clusters of cortical interneurons (CC-IN); two clusters of astrocytes; four clusters of oligodendrocyte lineage cells; seven clusters of immune cells including microglias, macrophages, and T cells; and three clusters of endothelial cells and blood cells (Fig. 1, B and C, fig. S2A, and table S1). The actively dividing cells were mainly in the embryonic and early fetal stages, especially NPCs, consistent with the dividing cell ratios determined by immunofluorescence and verifying the accuracy of our data (fig. S2, B and C). NPCs enriched in embryonic to early fetal stages and CC-EX were generated throughout the GWs that we collected (Fig. 1C). As the first cohort of neurons, CR cells appeared in the GW9 cortex and were enriched in early GWs in our dataset (Fig. 1C) (16).

Fig. 1 Cell types in human embryonic and fetal brain development.

(A) Schematic diagram of sample collecting and data analyzing. We collected the pairwised regions including the FL, PL, OL, and TL of the CC and the pons from the same embryos across developmental stages from GW9 to GW26. The transcriptome data of single cells was collected and used to do clustering using Seurat. The purple dots in the t-distributed stochastic neighbor embedding (tSNE) map at the bottom show cells from embryonic samples before GW9, which were too tiny to accurately dissect the cortical regions. (B) Cell types identification on the tSNE map and their marker genes’ expressions. The early cluster included majority of the cells in GW7/8 cortices and GW8/9 pons. CC-EX, cortical excitatory neuron; CR, Cajal-Retzius cell; IN, inhibitory neuron; Pons-neu, projection neuron in pons; Oligo, oligodendrocyte; Astro, astrocyte; MG, microglia; MP, microphage; SMC, smooth muscle cell; VEC, vascular endothelial cell; VLMC, vascular leptomeningeal cell. PAX6 is a marker for NPC, NEUROD6 is a marker for cortical excitatory neuron, GAD1 is a marker for inhibitory neuron, AQP4 is a marker for astrocyte, OLIG1 is a marker for oligodendrocyte, and CX3CR1 is a marker for microglia. (C) Boxplot showing the expressions of the classic marker genes for each subtype of cells. The pie charts in the right showing the regional distribution and the stage distribution for each cell type. Specifically, embryonic stage indicates cells from GW7 to GW10, early fetal stage stands for GW11 to GW14, early mid-fetal stage refers to GW16 to GW22, and mid-fetal stage represents GW24 to GW28. The number of cells in each type is revealed by the size of the blue dots in the right.

Globally, neurons from the CC and pons were grouped into separate clusters. By contrast, macroglial cells from these two regions, including astrocytes and oligodendrocyte progenitor cells (OPCs), were more mixed (Fig. 1C). Nevertheless, we noticed that most of the macroglias that we captured were from the pons, especially astrocytes. To remove sampling bias that might mask regional differences for astrocytes, we also merged cortical astrocytes from our previous studies (10, 11). According to the t-distributed stochastic neighbor embedding (tSNE) map in Fig. 2A, we found that astrocytes were separated mainly between the cortex and the pons, indicating a pattern similar to that in mice (14). We further performed glial fibrillary acidic protein (GFAP) immunostaining in cortical regions and the pons at GW20 and confirmed the much higher abundance of astrocytes in the pons than in the cortex (Fig. 2B). Except for the differences in abundance, we also observed more astrocytes with complex processes in the pons, indicating earlier development and maturation of astrocytes in the pons than in the cortex at the same developmental stage (Fig. 2B).

Fig. 2 Glial cell types and development.

(A) tSNE plot showing the regional differences of astrocytes in CC and pons. The expression levels of typical markers AQP4 and GFAP are shown in the bottom. (B) Costaining of NEUROD2 and GFAP in the CC and pons at GW20. NEUROD2 positive cells represent for neurons and GFAP-positive cells represent for astrocytes. The pons contains much larger abundance of astrocytes. NEUROD2, neuronal differentiation 2. (C) Expression levels of oligo subtype markers are shown by boxplot. OLIG1 is a pan marker for all oligo subtypes, PDGFRA lacks expression in Oligo_4, CDK1 is specifically expressed by Oligo_1 cells, APOD lacks expression in Oligo_2, and MOG and PDGFA are specifically expressed by Oligo_4 cells. (D) Co-immunostaining of APOD and PDGFRA in the CC at GW17. The white arrowheads show the double positive cells. DAPI, 4′,6-diamidino-2-phenylindole; PDGFRA, platelet-derived growth factor receptor A. (E) Pseudotime map of the subtypes of oligodendrocytes and their progenitors. The regional identity for each cell is also shown on the bottom.

OPCs undergo a long developmental path to form functional oligodendrocytes, and the sequential transition subtypes during OPC development to mature oligodendrocytes in mice have been characterized in detail (17). Consistent with the molecular regulation of transition subtypes in mouse oligodendrocyte development, we showed that the oligo_1 cells were proliferating OPCs (PDGFRA+/CDK1+), and oligo_2 were OPCs exited cell cycle (PDGFRA+/CDK1). Oligo_3 cells were OPCs starting to express APOD (apolipoprotein D), which were generated throughout all the fetal stages we covered (Fig. 2, C and D), and oligo_4 was newly formed oligodendrocytes (NFOLs) expressing MOG, ITPR2, and PDGFA (Fig. 2C). These four oligo subclusters clearly revealed a stepwise developmental path from OPCs to NFOLs at fetal stages by the end of the second trimester (Fig. 2E and fig. S2, D and E). Similar to that in mice, OLIG1 was consistently expressed during oligodendrocyte lineage development (Fig. 2C) (18). PDGFRA was specifically expressed in OPCs, and myelination proteins such as MBP appeared when OPCs differentiated into oligodendrocytes (Fig. 2E, fig. S2E, and table S2). Zeisel et al. (14) reported a rather mixed model of OPCs from regionally patterned progenitors that shared a concordant differentiation process to mature oligodendrocytes in the mouse nervous system. We supposed that, comparing to the astrocytes and neurons, the early development of immature oligodendrocytes in different regions was more similar in humans because we observed relatively mixed cell distributions from the cortical regions and pons in all subculsters on the tSNE and pseudotime map (Fig. 2E and fig. S2, F and G).

NPC subtypes and their development in human cortical regions

NPCs mainly exist in the ventricular zone (VZ) and sub-VZ (SVZ) at embryonic and early fetal stages (19). According to the expression of the marker genes for these four subtypes of NPCs, we identified NPC_1 cells as ventricular radial glial cells (vRGs), NPC_2 cells as outer sub-RGs (oRGs), and NPC_4 cells as intermediate progenitor cells (IPCs) (Fig. 3A and fig. S3A) (20). In addition, we detected a novel NPC subcluster (NPC_3) that expressed NSC genes, such as HES1, VIM, and HMGA2, as well as neuronal genes, such as STMN2, DCX, and NEUROD1/2/6 (Fig. 3B and fig. S4B). NPC_3 cells also expressed IPC genes such as EOMES and PPP1R17 but at lower levels than those in IPCs (Fig. 3B). To further confirm our data, we integrated a previous dataset by Nowakowski et al. (12), and cells were clustered by similar cell types instead of batches (fig. S3, C and D), further supported the accuracy of sampling and analyses in our study. Nevertheless, the NPC_3 cells existed as a subcluster between NPCs and early excitatory neurons (Fig. 1B). We supposed that NPC_3 was a transient state of neural progenitors that maintained proliferation activity and further became deep-layer neurons (DLNs) when exiting the cell cycle (fig. S3, E and F). We costained MKI67, DCX (doublecortin), and neuronal differentiation 1 (NEUROD1) to locate NPC_3 cells. As expected, NPC_3 cells were abundant in the SVZ region at GW13 (Fig. 3C). We further searched for marker genes of this cell subcluster (Fig. 3D). ADAMTS3 and DOCK9 were specifically expressed in NPC_3 cells, and RNA in situ hybridization of these marker genes also showed clear enrichment in the SVZ region (Fig. 3E). The reported layer V gene NPR3 was also enriched in NPC_3 cells, and the WNT pathway receptor FZD8 was highly expressed in NPC_3 cells. Therefore, we further investigated the canonical WNT signaling pathway in these four NPC subtypes (Fig. 3F and fig. S3G). Both vRGs and NPC_3 cells expressed high levels of β-catenin and WNT signaling pathway target genes, indicating the activation of the WNT signaling pathway (Fig. 3G). The WNT–β-catenin pathway must be inhibited to facilitate NSC differentiation (21). However, we continued to observe high expression levels of WNT–β-catenin pathway genes in NPC_3 cells, although these cells already showed high expression levels of neuronal genes (Fig. 3B and fig. S3A). We deduced that when NSC cells differentiated to acquire some neuronal features, such as NPC_3, they continued to maintain high levels of WNT signaling activity and remained in the active cell cycle, which will permit them to generate many more neuronal progenies. This idea was compatible with observations in mice; specifically, overexpression of stabilized β-catenin repressed neuronal differentiation for embryonic day 10.5 (E10.5) NPCs, but inverse results were observed for E13.5 NPCs.

Fig. 3 NPC subtypes and their molecular regulations.

(A) tSNE map of the four NPC subtypes. (B) Dotplot showing the marker genes’ expressions of vRG (HMGA2), oRG (HOPX), IPC (EOMES and PPP1R17), and neurons (STMN2 and NEUROD6). (C) Costaining of NEUROD1, MKI67, and DCX in the GW13 CC indicating the existence of NPC_3 cells. The white arrowheads show the triple positive cells and they are mainly located in the SVZ region. (D) Boxplot showing the marker genes for oRGs (LINC00043) and the NPC_3 subcluster of cells. TPM, transcripts per million mappped reads. (E) In situ hybridization of the NPC_3 marker genes. The staining of cresyl violet indicates cell density distributions in each section. (F) Boxplot showing the expressions of WNT–β-catenin signaling pathway genes in the NPC subtypes. The NPC_3 cells are active in the signaling pathway. ***P < 0.001 and ****P < 0.0001. n.s., no significance. (G) Costaining of CCND1, MKI67, and DCX indicating the activation of WNT–β-catenin signaling pathway in the NPC_3 cells. The white arrowheads indicate the positive cells.

lincRNAs are often species- and tissue-specific, and of which some have been proved to play a role in the evolutionary expansion of the human neocortex. They are usually lowly detected in heterogeneous bulk tissues but can regulate essential functions in a cell type–specific manner (22). In our dataset, we were excited to notice the lincRNA LINC00943 was specifically expressed by oRG cells (Fig. 3D). The specific expression of LINC00943 was further validated by RNA in situ hybridization (Fig. 4A). To test whether LINC00943 could affect the oRG identity, we overexpressed the gene in the mouse embryonic CC at E13.5 (fig. S4A). We first validated the overexpression through reverse transcription quantitative polymerase chain reaction (RT-qPCR) and collected the green fluorescent protein (GFP)–positive cells at E16.5 (fig. S4B). Cells overexpressing LINC00943 (GFP+) are enriched in the germinal structures compared to control cells (only overexpressing GFP) in the intermediate zone (Fig. 4B). In addition, the GFP+/Hopx+ cells are much more abundant in the vascular zone and subvascular zone of the CC when overexpressing LINC00943, indicating more oRG cells (Fig. 4C). We validated more oRG genes including Fabp7 and Tnc using RT-qPCR of the GFP+ cells, and they all showed elevated expression levels in the LINC00943 overexpressed cells (fig. S4B). We also checked the IPCs (Tbr2+ cells), and they were not affected by the overexpression of LINC00943 (fig. S4, B and C) nor did other genes, such as the global NPC marker genes Sox2 and Pax6 and the neuron markers Dcx and Neurod2. The inhibitory neuron gene Sst did not show differences between the LINC00943 overexpressed cells and control cells either (fig. S4B). The ratios of dividing cells were not changed by overexpressing LINC00943 (fig. S4C).

Fig. 4 LncRNA regulation in oRGs and cortical NPC development.

(A) RNA in situ hybridization of the oRG gene LINC00943 in both GW15 and GW17 CC indicates its enrichment in the SVZ. (B) Immunostaining of Hopx and Tbr2 in the E16.5 mouse CC transfected with GFP control (left) or GFP-LINC00943 (right). (C) Calculation of Hopx+ cell ratio and Tbr2+ cell ratio in the VZ and SVZ cells [n = 3 embryos, including slides shown in (B)] transfected with the pCAG plasmid, respectively. *P < 0.05 and **P < 0.01. (D) Pseudotime map of the NPC cells reflects the developing relationships of the NPC subclusters. The corresponding time is also shown on the bottom. The blue and green arrows indicate the two developmental trajectories of NPCs. The dashed circle indicates the determined start of the NPC development because they include the earliest cells as vRGs. (E) Pseudoage evaluation of NPCs in the four lobes of the CC reveals relatively late maturation between rostral and caudal regions.

Then, we analyzed the developmental trajectory of NPCs, and as expected, the percentage of NPCs in the cortex decreased over time (fig. S5A). VRGs were the dominant NPC subtype before GW11, and NPC_3 cells were mainly enriched from GW10 to GW14, following the time window of vRG cells. The generation of oRGs began to accelerate from GW11, and IPCs were generated throughout the developmental stages that we analyzed. We wonder whether the NPC_3 cells might be derived from vRGs according to their emergence time window and their expression of vRG markers. We further performed pseudotime analysis on these four NPC subtypes (Fig. 4D). In accordance with previous studies, the pseudotime result showed that the vRGs could develop into oRGs and IPCs (1, 23) (branch 1). Moreover, we observed that the vRGs could also generate NPC_3 cells (branch 2). The expression of the oRG marker gene HOPX gradually increased during lineage development in branch 1. The vRG marker gene HMGA2 and the neuronal gene NEUROD6 were up-regulated in branch 2 (fig. S5, B to D).

We further compared the pseudoage of NPCs in each lobe. In agree with previous study, the rostral region (FL) showed earlier maturation of NPCs than the caudal region (OL). In addition, we observed synchronized development of NPCs in the TL and PL, keeping the same pace with those in FL in the early fetal stage (Fig. 4E).

Excitatory neurons developed in a nonsynchronized manner across the cortex

We obtained 4347 CC-EX peaking from GW9 to GW26, showing temporally regulated characteristics in clustering, consistent with that observed in human prefrontal cortex development (Fig. 5A and fig. S6, A and B). In addition to the temporal patterns, we noticed that the excitatory neuron subtypes showed distinct structural distributions. The EX_1 subcluster of excitatory neurons was mainly generated before GW16 and widely expressed deep-layer genes such as FEZF2 and its downstream gene BCL11B (Fig. 5A and fig. S6C) (24). Thus, we defined this subcluster of neurons as DLNS. On the contrary, the EX_2 subcluster of excitatory neurons was mainly generated from GW16 to GW21, and these cells highly expressed some upper-layer markers such as POU3F2 and CUX2 (25), as well as the noncoding RNA LINC00158 (Fig. 5A and fig. S6C). So we defined this subcluster of neurons as upper-layer neurons (ULNs). According to Gene Ontology (GO) enrichment analysis with subtype-specific genes, the cells were enriched for neuronal functional terms such as ion transport, neurotransmitter transport, and chemical synaptic transmission (fig. S6D). Thus, we defined the EX_3 as a mature neuron (MN) subtype. More layer-specific genes were detected in DLN and ULN cells, respectively (fig. S6E), with some of them validated in the GW26 cortices (fig. S6F). These layer-specific genes were conserved in humans and in mice (fig. S6, G and H). Thus, we figured out the time windows for the generation of DLNs and ULNs in humans. The former were generated first, predominantly before GW16, followed by the ULNs that largely accumulated after GW16 (Fig. 5B) (26). CTIP2-positive (also known as BCL11B) neurons were nearly present in all four lobes of the GW13 cortical plate, while CTIP2-negative but SATB2-positive neurons were much more abundant in the GW20 cortical plate (Fig. 5, C to E, and fig. S6, I and J). We observed more SATB2+ neurons in the cortical plate of GW13 TL according to the immunostaining results (Fig. 5E and fig. S6, I and J). We further calculated the SATB2+ cell ratios in DLNs across the four cortical regions in our single-cell dataset, and, indeed, we found higher ratio of SATB2+ cells in TL (fig. S6K). The concordance of the validation experiment and sequencing result that there were more SATB2+ neurons in TL during DLN generation time might indicate different molecular regulations for neurogenesis in different regions of the CC (12, 27).

Fig. 5 Excitatory neuron development in the CC.

(A) tSNE map of the subtypes of CC-EX and the dotplot on the right showing the expressions of deep-layer gene BCL11B and upper-layer gene SATB2. (B) Time course of generation for DLN, ULN, and mature neuron in the human fetal CC and important molecular regulations. (C to E) Cell ratios of each excitatory neuron subtypes along GWs in each lobe, respectively. The cell ratios in GW24 PL are not shown because we failed to collect the cells in the first time. (D) Statistics analysis of cell ratios in each region using one-way analysis of variance (ANOVA). The GWs are divided into three stages: GW9 to GW14, GW16 to GW21, and GW24 to GW26 based on the time window for neuron subtypes. Each dot represents the ratio of corresponding subtype of cells in all neurons of an embryonic sample. *P < 0.05, **P < 0.01, and ****P < 0.0001. n.s., no significance. (E) Immunostaining of deep-layer marker CTIP2 and upper-layer marker SATB2 in each lobe of GW13 (stage GW9 to GW14) and GW20 (stage GW16 to GW21) are shown in the bottom. The GW13 cortical plate is dominant by CTIP2+ cells, while in GW20 there are much more SATB2+ cells.

We further searched for different molecular regulations for excitatory neurons and their progenitors across the four lobes, which could probably promote different regional functions of the cortex (Fig. 6A and see more details in the Supplementary Materials). We were excited to find that NR2F1 and NR2F2 (also known as COUP-TF1 and COUP-TF2) were more widely expressed in the TL during NPC differentiation into excitatory neurons (Fig. 6, A to C, and fig. S7A). Essentially, more than 90% of the NR2F2-positive cells were in a quiescent state or at the postmitotic stage (Fig. 6D). The FL and TL also showed differential expression of the lincRNA LINC00643, with the former exhibiting enrichment in DLNs (Fig. 6A and fig. S7A).

Fig. 6 Regional differences in the development of CC.

(A) Boxplot showing regional different regulated genes in each neuronal subtype. NR2F2 is highlighted in red as it is a TL-specific gene in all cell subtypes. (B) Immunostaining of NR2F2 and MKI67 in the four lobes of the GW13 cortex. (C and D) Barplot showing the ratios for NR2F2+ cells (C) and the MKI67+ cell ratios of NR2F2+ cells (D) calculated in each lobe [n = 3 embryos, including slides shown in (B)]. (E) GO-enriched terms of increased expressed genes in GW24 (GW25 for PL) in excitatory neurons of each lobe. (F) Visualization of the morphology of neurons in FL and PL after whole-cell patch-clamp recording at GW24. Scale bars, 20 μm. (G to I) Representative electrophysiological properties of FL (red) and PL (yellow) neurons at GW24. (G) Phase plot of evoked action potential (insert). (H) Voltage responses elicited by current injections of −20, 0, and 40 pA. (I) Current responses evoked by a series of voltage steps. The magnification of Na+ currents indicates by the green bar. (J) Quantification of Na+ currents of FL and PL neurons at GW24. Twenty-three FL neurons from six independent replicates and 17 PL neurons from five independent replicates were recorded. **P < 0.01 (P = 0.00525, 0.00559, and 0.00846 for voltage step from −60 to −40 mV, −60 to −30 mV, and −60 to −20 mV respectively). Data are shown as the means ± SEM. (K) Resting membrane potential of FL and PL neurons. There are 23 FL neurons from six independent replicates and 17 PL neurons from five independent replicates. Data are shown as the means ± SEM. (L) Representative trace of spontaneous excitatory postsynaptic currents (sEPSCs) of FL neuron. The magnification of the arrow marked sEPSC event is showed as below.

MNs appeared as a subcluster detected from GW24 and later stages (Fig. 5, B and C, and fig. S6B). To determine whether the neuronal developmental process along the fetal stages was uniform across the four cortical lobes, we searched for differentially expressed genes (DEGs) between excitatory neurons in adjacent GWs and then performed GO enrichment analysis on each DEG set (see more details in the Supplementary Materials). We extracted neuronal maturation-related GO terms that were enriched in all cortical regions, including axon development, neuron projection development, and synaptic transmission. The average expression levels of genes belonging to these three GO terms and a series of typical genes such as CAMK2B, UNC13A, GRIN1, and NRXN2 were calculated in each lobe during the developmental period (fig. S7, B and C). In accordance with the temporal appearance of MNs, the expression levels of these sets of genes were markedly elevated from the GW24/GW25 stage onward (Fig. 6E). However, we inferred that the maturation degrees for different cortical regions were unsynchronized, as the enrichment in the FL was the most drastic whereas that in the PL was less drastic (Fig. 6E). The expression of ion channel genes also exhibited the highest RNA levels in the FL (fig. S7E). To confirm this hypothesis, we examined the electrophysiological properties of neurons in the FL and PL at GW24 (fig. S7, F and G). Neurons in the FL exhibited more complex morphology, while their counterparts in the PL showed bare processes (Fig. 6F). Action potentials (APs) with mature kinetics could be evoked by current stimuli in FL neurons (Fig. 6, G to I). Although the resting membrane potential (RMP) of neurons in FL was roughly equivalent to that in PL, an inward Na+ current with a significantly larger amplitude was observed in FL neurons (Fig. 6, J and K). Spontaneous excitatory postsynaptic currents (sEPSCs), which represented the abilities of synaptic transmission, were frequently encountered in FL neurons (Fig. 6L). These evidences demonstrated that neurons in the FL reach a more physiologically mature state than do those in the PL at GW24.

Comparison between human and mouse in neuronal development

A recent work elucidated a well-conserved cellular architecture between human and mouse cell types with extensive differences in gene expression in the adult CC (28). We wonder whether this conservation existed from the developmental stages. Thus, we made a comparison of the NPCs and excitatory neurons in the human developing CC to those in the mouse (29). The temporal development of mouse cortical cells showed conserved pattern to those in human, exhibiting a neuronal differentiation tendency (fig. S8A). It is worth emphasizing that the NPC_3, which identified as a novel subcluster of transient cells we captured, also turned to be human specific. As expected, a total of 12,786 merged cells (6396 human cells and 6390 mouse cells) could be divided into four clusters with biological inferences. As shown in fig. S8b, the newly defined radial glia cells contained human vRG and oRG cells together with the majority of mouse E11.5 cells and mouse RG cells from E13.5 to E17.5, highly expressed HES1 (fig. S8C). We further checked the defined clusters with the human-mouse conserved cell type marker genes including EOMES as intermediate progenitor marker, FEZF2 as DLN marker, and SATB2 as ULN marker (fig. S8C). Then, we searched for human specific genes in each cell type (table S1). The endothelial signaling gene, APOLD1, showed human specific expression, especially in the intermediate progenitors (fig. S8d). We supposed that this gene might be an important evolution-related gene in neural development, as it was also reported to show different expression between human and chimpanzee in the cerebral organoids (30).

Cortex showed tremendous differences to pons in neural developmental course

Compared to excitatory neurons in the cortex, excitatory neurons in the pons showed marked molecular distinctions (Fig. 7A and fig. S9A). As expected, patterning genes such as FOXG1 and EMX2 (2, 31) were expressed in cortical neurons, whereas HOX family genes were enriched in the pons only (32). ZIC family genes, which reportedly regulate the dorsal regions of the midbrain and hindbrain (33, 34) were also clearly detected in the pons. The neurogenesis genes NEUROG2 and NEUROD1 were enriched in cortical neurons (fig. S9, A and B), indicating a difference in molecular regulation for neuron generation in pons. Similar to those in the cerebellum and medulla, projection neurons in the pons also highly expressed PAX6 (35), demonstrating a different role of PAX6 in neuron development in the hindbrain. More region-specific genes for the cortex and pons were revealed (fig. S9, A and B, and table S1). LIM homeobox gene 1 (LHX1) showed preferential expression in the pons, whereas LHX2 in the same gene family was predominantly expressed in the cortex (Fig. 7, B and C). The regional differences included not only the protein coding genes, such as transcription factors, but also noncoding RNAs, such as LOC400043, indicating markedly distinct regulations of projection neurons between the cortex and pons.

Fig. 7 Cortex showed tremendous differences to pons in neural developmental course.

(A) tSNE map of the excitatory neurons from the CC and the pons. Neurons from both regions are clearly separated on the map, indicating huge differences in gene regulation. (B) Boxplot showing the different regulated genes between the cortical excitatory neuron subtypes and those in the pons. LOC400043 and LHX1 are specifically detected in neurons of the pons, while LHX2 shows specific expression in CC. (C) In situ hybridization in pons and cortex of GW10 embryo to validate the genes in (B). (D) tSNE map of the two subculsters of excitatory neurons in the pons and the histogram shows the number of cells in each subcluster at different stages. (E) Cell cycles distributions of neurons from different GWs in the pons. The cells of both G1-S and G2-M score less than 1 (the dash line) are recognized in quiescent state. Very few cells are in dividing activation state. (F) Histogram showing the GAD1+ cell ratios in early GWs in the CC and the pons, respectively, based on the single-cell sequencing data. (G) In situ hybridization of GAD1 in the GW10 CC and pons. The pons contains large abundant of GAD1+ cells, while there is little signal in the CC.

We further analyzed projection neurons in the pons and found that the temporal pattern of neuronal development was not as prominent as that in the cortex (Fig. 7D). The pons_neu1 subcluster contained neurons from GW9 to GW25, and pons_neu2 cells were rarely detected until GW21. Few cells were in active cell cycles (Fig. 7E). GO analysis of the pons_neu2-specific genes showed enrichment of biological terms like trans-synaptic signaling and synapse organization (fig. S9C). Thus, compared with the neuronal maturation from GW24 in the CC, the maturation of neurons in pons occurred earlier.

Inhibitor neurons constituted approximately 20% of total cortical neurons, and they mainly migrated from the ganglionic eminence (36). Inhibitory neurons were rarely detected in the CC at early stages (fig. S9D). In contrast, inhibitory neurons were already abundant at early embryonic stages in pons (Fig. 7, F and G) and might be involved in determining the distinct role of the pons in the central nervous system. The increase of inhibitory neurons in the cortex tended to start at later stages in the FL and TL (fig. S9D), consistent with our previous observations. This observation might be the result of inhibitory neuron migration during fetal development.

DISCUSSION

In the present study, we systematically analyzed the neuronal developmental process in four representative cortical lobes and the pons from the early to middle stages of human gestation, covering more than 5 months of critical developmental stages. Considering the limited sample size in each region at a specific time point, we merged all the data from each stage/region as there was negligible batch effects. After that, we accurately classify the identities of the cells and made further analysis within general cell types to obtain robust results. Meanwhile, the sequencing depth and gene detection sensitivity of our data are much higher than most of the reported datasets. Thus, we could obtain more information with valid supporting data. For example, we found more NPC subtype DEGs in table S2.

We summarized the characteristics of the whole developing CC using a control from other part of the brain. Regional comparisons of the whole transcriptome revealed neurons from the CC and the subcortical region, the pons, showing marked distinctions in both molecular regulations and developmental patterns. By comparing the CC to the pons, we determined the regulatory genes for neurons in the human CC, such as MEF2C, EMX2, and TBR1. The pons, which is evolutionarily more ancient than the CC, exhibited earlier development and maturation. Inhibitory neurons are already abundant in the pons at embryonic stages, while most of interneurons tangentially migrate to the cortex beginning at the early mid-fetal stage.

The systematization of astrocytes and oligodendrocytes in the mouse nervous system has been studied in detail (14). However, this information remains to be elucidated in humans. We summarized the cell subtype transitions of oligodendrocyte lineage cells until the mid-fetal stage in humans, including data from our previous studies, to show a relatively mixed model of regional OPC development than astrocyte and neuron. While regional DEGs could still be identified in the immature oligo cells between CC and pons, these genes showed high consistency to those identified in neurons between these two regions, indicating more regional characters than cell lineage associated regulations. In contrast, astrocytes show a regionalized pattern on the molecular level, and accordingly, the morphology of astrocytes in the pons exhibited much more complex processes. These developmental patterns tend to be conserved in humans and mice. According to the observation that MNs in the pons appeared earlier (before GW22) than did those in the cortex (after GW24), we proposed more delayed development of neurons in the cortex than in the subcortical regions.

The birth date of a neuron is highly in accordance with its final neocortical laminar location, and the same type of laminar neurons shares common projection targets. We crystallized the temporal windows for DLN, ULN, and neuronal maturation in the human CC. In our previous study (10), the electrophysiological activation of neurons in the prefrontal cortex occurred at some time point between GW23 and GW26. Here, we specified the accurate stage for neuron maturation at GW24 in the FL. Moreover, regional asynchrony in neuronal maturation was observed and validated by analyzing the electrophysiological features of neurons in the FL and PL, which revealed earlier maturation of the rostral regions than of the caudal regions. In addition to revealing spatial differences in neuronal maturation, the pseudoage analysis also exhibited earlier maturation of NPCs in the rostral region (the FL) than in the caudal region (the OL), similar to the result in the previous study. The asynchronous cell development in different regions of the cortex could be essential for proper neural network construction, which might offer us clues on neural developmental diseases.

Neurogenesis is a conserved process of neuron generation that occurs in an inside-to-outside manner in vertebrates. In our study, we summarized the molecular characteristics of these cells, and a series of lincRNAs for cell type and regional specification were emphasized. A previous study investigated the lincRNA population in bulk samples of human fetal brain tissue and integrated a group of single-cell RNA-seq data based on main cell types. As lincRNAs showed much lower expression levels than did protein-coding mRNAs, the best way to explore them is using hypersensitive single-cell data. Here, we revealed different lincRNA expressions not only in different subpopulations of NPCs and neurons but also in different brain regions, i.e., the CC and pons, indicating lincRNA functions in both cell type specification and regionalization. We and others (22) validated the important role of some specific lincRNAs in vivo and in vitro. Nevertheless, how these lincRNAs participate in preserving cell types and specific regions remains unknown. One mechanism of lincRNA function in the neural system is to sequester miRNAs (37), while more lincRNAs are supposed to bind proteins of transcription factors or epigenetic modifiers (38). Nevertheless, the mechanisms underlying how most lincRNAs function remain largely unexplored.

We identified a novel subcluster of NPCs that was enriched in a narrow time window from GW11 to GW14. This narrow window might be why a previous study hardly captured these cells. These newly identified transient cells exhibited both vRG and neuron characteristics and might be human specific to facilitate the rapid expansion of the human CC at the early gestation stage. These cells exhibited a unique regulatory pattern of the canonical WNT signaling pathway, which supports both active proliferation and quick differentiation. Moreover, these transient cells highly express deep-layer feature genes, with further pseudotime analysis reflecting restricted fate to DLNs. There was also a chance that these novel NPCs might be neurogenic progenitor cells or newly generated differentiating neurons that have not yet degraded progenitor cell mRNAs. If so, the phenomenon that these transient cells were only captured in the early mid-stage during the time window for DLN genesis, although large abundant of neurons were obtained after this stage in our dataset may indicate quite different molecular regulations for generation of DLN and ULN. As we analyzed many more cells with higher sensitivity single-cell RNA-seq method in cerebral cortices during human fetal development than any previous study, we also summarized 73 novel marker genes for vRG, oRG, and IPC in our study (table S1).

The cross species comparisons have been carried out in the adult cell types, which emphasized the necessities of directly studying human brain. However, we still not clear how conserved of cellular architecture betwesen human and mouse neural development in CC. We compared the human data here with previous mouse data generated using Drop-seq (29) and observed global cell type conservation between human and mouse in neuronal development in developing CC. Meanwhile, we found that the novel NPC state we captured in GW11 to GW14 turned out to be human specific. We found hundreds of DEGs between human and mouse in each specific cell type. However, affected by large batch effects between these two datasets (the median gene number and UMI number for mouse cells were only 1297 and 2064), many of these genes are not faithful and need more supporting data and proofs.

Unexpectedly, neuronal developmental terms regarding neuronal maturation were enriched not only at GW24/25 but also at GW14 in some regions (Fig. 6E and fig. S7, B and D). Because the electrophysiological properties implicated immature neurons before GW24 (Zhong et al. showed no Aps or potassium and sodium channel currents for neurons in the pre-frontal cortex at GW230) (10), the increased expression levels of neuron maturation genes in GW14 cortical regions such as the TL might be associated with inhibitory migration to the CC. We observed a marked inhibitory neuron ratio increase in the cortical regions, especially in the TL at the early mid-fetal stage (fig. S8D). Thus, the development of axons and transmitting chemicals, for example, PLXNA4, would preserve local cues for inhibitory neuron migration (39). However, this assumption needs to be further explored. In addition, the higher ratio of SATB2+ neurons in TL at this stage (Fig. 5D and fig. S6, I and J) might also associate with the migration of inhibitory neurons or just a coincidence.

The sequence for layer formation has been confirmed in numerous studies, and the temporal windows for laminar generation in mice have been well traced (40). Nevertheless, the corresponding timing in humans has not yet been reported. We summarized the specific time points for laminar neurogenesis that could help to better align neural development in mice to that in humans. Furthermore, we accurately defined the spatial-temporal neuronal maturation in GW24 FL. The priority of regional maturation might probably influence region-specific neuronal projections and interactions with subcortical regions, ensuring an integrated neural network. Together, our data provide a comprehensive resource for neuronal development in the human embryonic and fetal CC, including pairwise regional information, which may have potentially important indications for decoding the highly complex yet highly robust developmental trajectory of the human brain.

MATERIALS AND METHODS

Single-cell sample preparation

The human embryonic and fetal sample collection and single-cell transcriptome study were carried out under the approval of the Reproductive Study Ethics Committee of Peking University Third Hospital (2012SZ-013 and 2017SZ-043). All these experiments were done following the International Review Board and Institutional Animal Care and Use Committee guidelines. Fetal brain samples were collected after donors signed an informed consent document, which is compatible with International Society for Stem Cell Research guidelines.

Aborted embryos and fetuses from the Third Hospital of Peking University were dissected for intact brains. Then, we mapped the shape and anatomical regions to The Human Brain During the first Trimester and The Human Brain During the Second Trimester by S.A., Bayer and J. Altman for dissection of the four cortical regions and pons. The age of a brain tissue (GW) was measured in weeks from the first day of the donor’s last menstrual cycle to the day when it was aborted. The most central part of each region were cut from cortical plate to the ventricle and then digested in collagenase IV (2 mg/ml) (Gibco, 17104-019), papain (1 mg/ml; Sigma-Aldrich, P4762), and deoxyribonuclease I (10 U/μl; New England Biolabs, M0303L) with pipetting and then vortexed on a thermocycler at 37°C for 10 to 15 min. Next, the suspension was filtered using a 40-μm strainer (BD Biosciences, 352340) to remove undigested cell clumps. The flow through of cells was centrifuged at 500g for 3 min to obtain the cell pellet. The suspension was carefully removed, and we used hibernate E medium (Thermo Fisher Scientific, A1247601) to resuspend the cells, keeping it on ice before transferring single cells into lysis buffer by mouth pipet.

Single-cell RNA-seq library preparation and sequencing

We did barcoded single-cell transcriptome library preparation based on the STRT-method and in accordance with our previous studies (10, 11). Specifically, four cycles of preamplification with 98°C for 20 s, 65°C for 30 s, and 72°C for 5 min were carried out before 15 cycles of complementary DNA (cDNA) amplification with 98°C for 20 s, 67°C for 15 s, and 72°C for 5 min. After amplification, we pooled each 96 cells together (the cell barcode sequences shown in table S2), purifying with DNA clean and concentration kit (Zymo Research, D5044). The full-length cDNAs were purified twice more with 0.8 volumes of Ampure XP beads (Beckman, A63882). Then, we took about 30 ng of cDNAs for a second round of PCR for four cycles with 98°C for 20 s, 67°C for 15 s, and 72°C for 5 min. We used biotin modified illumine read2 primers (see detailed sequences in table S2) and the IS PCR primer of Smart-seq2 for the second round of PCR. Then, one round of purification using 0.8 volumes of Ampure XP beads was carried out before shearing the DNA into fragments between 300 and 500 base pair (bp) by Covaris M220 machine. We enriched the 3′ end of mRNAs using streptavidin-modified beads (Thermo Fisher Scientific, 65002) and did further library construction using the KAPA Hyper Prep Kits for Illumina (KAPA, KK8506). Each cell was sequenced for 0.5-giga bases of 150-bp paired-end reads on an Illumina platform with Hiseq4000.

Processing and quantification of single-cell RNA-seq data

Cells were distinguished according to the cell barcode info (without any mismatch) in Read2 (also see table S2), and for each cell, with the corresponding read identifier in Line1, Read1 was selected. To facilitate later UMI quantification, we attached the UMI info in Read2 to the read identifier of Read1. Subsequently, the template switch oligo (TSO) sequence, poly(A) tail sequence and adaptors were trimmed, and reads with too many (more than 50%) low-quality bases or a high ratio (>10%) of “N” or those shorter than 37 bp were discarded. Next, clean data were aligned to hg19 genome sequences using TopHat software. Then, uniquely mapped reads were counted by the HTSeq package in which reads with the same UMI info were denoted as “1.” Last, for each given individual cell, cell-gene matrices with UMI count values were generated.

To obtain high-quality data, we discarded cells with less than 1500 expressed genes, fewer than 10,000 transcripts, or a mapping ratio lower than 20%. For all 14,239 sequenced cells, 12,541 (88.08%) cells passed the quality control. A total of 952 cells from our previous study (including cells from the SF, FP, LO, and MT regions) were merged. In total, 13,493 cells were used for downstream analysis.

Clustering

Seurat v2.3.0 (https://satijalab.org/seurat/) was used to analyze the expression matrices (13,493 cells × 24,153 genes), and two-round clustering was performed. On the first level, genes were expressed in at least three cells, while cells with a minimum of 1500 expressed genes were retained. Data were normalized to a scale factor of 100,000, and variable genes were found by the “FindVariableGenes” function with the parameter “x.low.cutoff = 1, y.cutoff = 1.” Sequentially, cell cycle effects were regressed out, and 33 PCs were used to find clusters. With a “resolution” of 1.8 upon running “FindClusters,” we distinguished major cell types in the tSNE map according to known markers, including neuron, astrocyte, oligodendrocyte, NSC, IPC, immune cells, smooth muscle cell, vascular endothelial cell, vascular leptomeningeal cells, and blood. Here, some very early samples (≤9 W) did not cluster well according to cell type, possibly because early embryos developed too fast over time and cell type features were not mature at that time. On the second level, we successively focused on neuron, astrocyte, oligodendrocyte, NSC, IPC, and immune cells, which were distinguished on the first level. By following a similar pipeline, subclusters were identified. We finalized the resolution parameter on the FindClusters function once the cluster numbers did not increase when we increased the resolution. Then, we checked the DEGs between each of the clusters. To avoid overfragmented clustering, we merged clusters with less than 10 DEGs based on the cutoff value “P_value < 0.01, avg_LogFC >1.5” into one cluster. In this way, there were at least 10 DEGs between any two clusters.

Integrate analysis with previous datasets

Expression matrix and meta file of Nowakowski et al. (12) were downloaded from https://cells.ucsc.edu/?ds=cortex-dev. After creating its seurat object, “FindIntegrationAnchors” and “IntegrateData” functions in R (v3.5.1) package Seurat(v3.0.0) were used to perform the integration with our dataset. The single-cell RNA-seq mouse data (29) were download according to the Gene Expression Omnibus (GEO) number of GSE107122, and then similar pipeline was done for the integration in R (v3.5.1) package Seurat (v3.1.0).

For the astrocytes analysis, we fetched expression matrixes of astrocytes from our previous datasets including Zhong et al. (10) and Fan et al. (11). In addition, cells belong to insular gyrus and medulla regions in Fan et al., (11), Cell research data were discarded. Seurat package was performed for further clustering of all the astrocytes from different studies.

Detection of DEGs

To determine spatial features for each cell type, we calculated DEGs between the four lobes using the “FindMarkers” function in the Seurat package. As shown in Fig. 5E and fig. S7D, within the excitatory neuron in the cortex, we also calculated DEGs of adjacent weeks in each lobe using the FindMarkers function in the Seurat package with the cutoff value of “P_value < 0.01, |avg_logFC| > 0.25”. GO enrichment analysis was performed on the basis of these DEGs by Metascape (http://metascape.org). Similarly, to compare differences between subclusters, such as DLN and ULN, we used the FindMarkers function with specified “ident.1=, ident.2=.” All the statistical tests for different gene expressions were carried out using Wilcox.

Pseudotime analysis by monocle

The R package Monocle was used to estimate the lineage trajectory. We selected the “differentialGeneTest” module to perform differential expression analysis to obtain the ordering genes with parameter as “fullModelFormulaStr = ~subtype”, and then genes with q values less than 0.001 were retained for downstream analysis. Subtype here was the clustering result conducted by Seurat before. Followed by running the “setOrderingFilter” function, the “reduceDimension” module was used to reduce the dimensionality of the data with the parameter of “method = DDRTree.” When running “orderCells,” “root_state” can be changed according to the known biology background. Then, “plot_cell_trajectory,” “plot_genes_in_pseudotime,” and “plot_genes_branched_heatmap” functions were used for further visualization.

Evaluation of neuronal development related terms along GWs in the CC

We extracted the GO terms enriched during week by week comparisons of gene regulations for excitatory neurons in each lobe of the CC regarding biological processes such as axon development (GO: 0050770, GO: 0007411, GO: 0016358, GO: 0048813, and GO: 0050772), projection (GO: 0010975, GO: 0106027, GO: 1902284, GO: 0048812, GO: 0031344, and GO: 0010976), and synaptic transmission (GO: 0007268, GO: 0050807, GO: 0051924, GO: 0006816, GO: 0050804, GO: 0035249, and GO: 0050806). Genes located in these GO terms and expressed in our data were used for further analysis. Then, for each lobe, the averaged expression level of each biology process during each week (GW10 to GW26) was calculated and exhibited by heatmaps drawn with the “heatmap.2” function in the R package “gplots.”

Cell cycle analysis

The “CellCycleScoring” function in Seurat was used to obtain S and G2-M scores with input from cell cycle markers (41), and each cell could be predicted to either G2-M, S, or G1 phase. In fig. S2B, we assigned this info to the tSNE map. Using the same list of cell cycle markers, the average expression levels of G1-S genes and G2-M genes sets were calculated as the G1-S score and G2-M score for cells in fig. S3E and Fig. 7E, respectively. The “geom:point” function in the R package “ggplot2” was used for visualization. Cells with a G1-S score and G2-M score more than 1 were defined in an active dividing state.

Assignment of pseudoage score for NPCs

To select age correlated genes, we first identified variable genes using function FindVariableGenes with default parameters and constructed a vector contained age information of each cell. We then correlated expression value of each variable gene with the age vector. Genes with correlation of more than 0.2 were selected for subsequent analysis. Next, we performed principal components (PCs) analysis with selected age correlated genes using function RunPCA and determined statistically significant PCs using function JackStraw in R. Then, we computed the correlation between the age vector and these significant PCs and then selected the PC resulting in the highest correlation coefficient. Pseudoage score of cell was determined by the averaged expressions of the chosen PC genes.

Immunostaining

Tissue samples were fixed overnight in 4% paraformaldehyde (PFA), dehydrated in 30% sucrose in phosphate-buffered saline (PBS), embedded and frozen at −80°C in O.C.T. (optimal cutting temperature) compound, and sectioned with Leica CM3050S. Cryosections were subjected to antigen retrieval, pretreated (0.3% Triton X-100 in PBS), incubated for a blocking solution (10% normal donkey serum, 0.1% Triton X-100, and 0.2% gelatin in PBS), followed by incubation with the primary antibodies overnight at 4°C. The primary antibodies used were rabbit anti-NEUROD2 (1:300; Abcam, ab104430), mouse anti-GFAP (1:500; Cell Signaling Technology, 3670s), rabbit anti-APOD (1:300; Abcam, ab196569), goat anti–platelet-derived growth factor receptor A (PDGFRA) (1:300; R&D, AF-307-NA), mouse anti-NEUROD1 (1:200; Abcam, ab60704), mouse anti-SATB2 (1:200; Abcam, ab51502), rat anti-CTIP2 (1:200; Abcam, ab18465), mouse anti-Ki67 (1:200; BD, 550609), rabbit anti-Ki67 (1:200; Millipore, AB9260), goat anti-DCX (1:300; Santa Cruz Biotechnology, sc-8066), mouse anti-CTNNB1 (1:200; Abcam, ab6301), rabbit anti-NR2F2 (1:200; Abcam, ab41859). Secondary antibodies used were Alexa Fluor 488 (1:500), Alexa Fluor 594 (1:500), or Alexa Fluor 647 (1:500) conjugated to donkey anti-mouse, anti-rat, anti-rabbit, or anti-goat (Invitrogen). Cell nuclei were stained using 4′,6-diamidino-2-phenylindole (Life Technologies). Immunoflourence images were acquired with Olympus laser confocal microscope and analyzed with FV10-ASW viewer (Olympus), ImageJ (National Institutes of Health), and Photoshop (Adobe). Data are presented as means ± SEM.

RNA in situ hybridization

The in situ hybridization protocol has been described previously (42). For in situ hybridization, all operations were carried out under ribonuclease-free conditions. Digoxigenin (DIG)–labeled antisense riboprobes were synthesized by in vitro transcription using cDNA templates. Fetal brain sections with a thickness of 40 μm were hybridized with complementary RNA probes with a final concentration of 300 ng/ml overnight at 64°C in hybridization solution [50% formamide, 10% dextran sulfate, 0.2% tRNA (Invitrogen), and 1× Denhardts solution from a 50× stock; Sigma-Aldrich], 1× salt solution [containing 0.2 M NaCl, 0.01 M tris, 5 mM NaH2PO4, 5 mM Na2HPO4, and 5 mM EDTA (pH 7.5)]. In situ product was visualized using anti-DIG alkaline phosphatase–conjugated secondary antibody (Roche) with color development using BCIP-NBT (bromochloroindolyl phosphate–nitro blue tetrazolium) solution (Roche). Images were acquired by Leica SCN400 (Leica Microsystems).

Plasmids and in-utero electroporation

Constructs used were pCAG-GFP (control) or pCAG-GFP-LINC00943 (GFP-LINC00943). For in utero electroporation, timed pregnant CD-1 mice at E13.5 was deeply anesthetized, the uterine horns were exposed, and ~1 μl of plasmid (1 to 2 μg μl−1) mixed with fast green (0.1 mg ml−1; Sigma-Aldrich) was manually microinjected through the uterus into the brain lateral ventricle using a beveled and calibrated glass micropipette (Drummond Scientific). For electroporation, five 50-ms pulses of 50 mV with a 950-ms interval were delivered across the uterus with two 9-mm electrode paddles positioned on either side of the head (BTX, ECM 830). After the procedure, the uterus was placed back in the abdominal cavity, and the wound was surgically sutured.

qPCR validation of oRG genes

GFP+ cells were isolated by fluorescence-activated cell sorting, and we extracted the total RNA using the RNeasy Mini Kit (QIAGEN, 74106). Because of the low yield of total RNA for directly RT-qPCR, we firstly converted the RNAs into cDNAs and amplified them by 10 cycles of PCR using Smart-seq2 (43). Then, the cDNA products were purified with 0.8 volume of AMPure XP (Beckman, A63882), and further, we did real-time PCR analysis with specific gene primers including LINC00943 (forward, 5′-CGATGAACCACCCATGGCCT-3′; reverse, 5′-ACTTCCAGGAATGGAAGCCACA-3′), Hopx (forward, 5′-TGCCCTTGGCCATCACCTTC-3′; reverse, 5′-CCCACGTTCTCATTCAACCACCA-3′), Tnc (forward, 5′-TGGGAATGGGAGAGGGGCAA-3′; reverse,5′-AACATCGAGGGTGGGGGTGG-3′), Fabp7 (forward, 5′-TGTGCAGAAGTGGGATGGCA-3′; reverse, 5′-TGGCTAACTCTGGGACTCCAGG-3′), Dcx (forward, 5′-GGACTGGAATGTTTGGCAAGGC-3′; reverse, 5′-ACAACACCTCACACACATGGGG-3′), Neurod2 (forward, 5′-TGGAGGTTCCCCTCGCAAGA-3′; reverse, 5′-TCGGGTGTCCGACAGGAGTT-3′), Pax6 (forward, 5′-CAGCATGCAGGGCAGGAGTG-3′; reverse, 5′- GTGGCCAGGACCCCAGAGTT-3′), and Sst (forward, 5′-CTGCCAACTCGAACCCAGCA-3′; reverse, 5′-TCAAGTTGAGCATCGGGGGC-3′).

Detection of neuronal electrophysiological properties

Neocortical tissues were obtained from GW24 embryos and sectioned at 500 μm in oxygenated (95% O2 and 5% CO2) ice-cold sucrose-based artificial cerebrospinal fluid (ACSF) (234 mM sucrose, 2.5 mM KCl, 26 mM NaHCO3, 1.25 mM NaH2PO4, 11 mM d-glucose, 0.5 mM CaCl2, and 10 mM MgSO4) with a vibratome (VT1200s, Leica). Coronal slices were incubated in an oxygenated ACSF (126 mM NaCl, 3 mM KCl, 26 mM NaHCO3, 1.2 mM NaH2PO4, 10 mM d-glucose, 2.4 mM CaCl2, and 1.3 mM MgCl2) at 34°C for 30 min and then at room temperature for about 60 min before use. After the recovery period, an individual slice was transferred to the recording chamber and continuously superfused with oxygenated ACSF at a rate of 3 to 5 ml/min at room temperature. Whole-cell patch clamp recording was performed on cells, which were randomly selected, with a clear apical process and located in the deep layer of cortical plate (800 to 1000 μm below the pial surface, as showed in fig. S7F). Patch pipettes had a 7 to 11 megohm resistance when filled with intracellular solution [130 mM potassium gluconate, 16 mM KCl, 2 mM MgCl2, 10 mM Hepes, 0.2 mM EGTA, 4 mM Na2–adenosine 5′-triphosphate, 0.4 mM Na3–guanosine 5′-triphosphate, 0.1% Lucifer yellow, and 0.5% neurobiotin (pH = 7.25) adjusted with KOH]. Currents with a series of amplitudes from −100 to 280 pA in increments of 20 pA were injected to evoke action potentials of the recording cells in current-clamp mode. Whole-cell currents were recorded using a basal holding potential of −60 mV, followed by stimulating pulses from −80 to 60 mV with a step size of 10 mV in voltage-clamp mode. sEPSCs were acquired when the membrane potential of the target cell was holding at −70 mV. A 40× Olympus water-immersion objective lens, a microscope (Olympus, BX51WI) conFig.d for DGC (Dodt Gradient Contrast), and a camera (Andor iXon3) were used to monitor the tissue during the recording. Stimulus delivery and data acquisition were conducted with a MultiClamp 700B amplifier and a Digidata 1440A (Molecular Devices). After the electrophysiological experiments, the slices were fixed and stained with fluorescein streptavidin (1:500; Vector Laboratories, SA-5001) or Texas Red streptavidin (1:500; Vector Laboratories, SA-5006) to visualize the morphology of the recording cells.

SUPPLEMENTARY MATERIALS

Supplementary material for this article is available at http://advances.sciencemag.org/cgi/content/full/6/34/eaaz2978/DC1

https://creativecommons.org/licenses/by-nc/4.0/

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.

REFERENCES AND NOTES

Acknowledgments: We thank X. Wu and Y. Cui for helping with the single-cell collecting and S. Tan for dealing with the donor patient issues. We thank the National Center for Protein Sciences at Peking University for assistance with the LINC00943 overexpression experiment in mouse embryos. Funding: These studies were supported by grants from the National Natural Science Foundation of China (31625018, 31230047, 81521002, and 91732301), the National Basic Research Program of China (2019YFA0110101, 2017YFA0102601, 2017YFA0103303, and 2018YFA0107601). X.F. was supported by the grant from China Postdoctoral Science Foundation (2018 M630031) and the Beijing Brain Initiative of Beijing Municipal Science & Technology Commission (Z181100001518004). X.F. was a Bayer Postdoc of the Bayer-Peking University Center for Translational Research and awarded by Boehringer Ingelheim Postdoc Fellowship. This work was also supported by the Beijing Advanced Innovation Center for Genomics at Peking University. Author contributions: F.T., J.Q., and X.W. designed the project. X.F. carried out most of the sequencing experiments and data preprocessing. Y.F. performed the bioinformatics analysis. X.Z. and X.F. performed the immunostaining and in situ hybridization. L.S. and R.C. collected and analyzed the electrophysiological data. M.W. did the pseudoage analysis. Y.F., X.F., and L.S. interpreted the results and wrote the manuscript. Q.W., J.D., and L.W. helped interpret the data results. M.Y. and J.Y. contributed to the brain sample collection. F.T. and X.W. led the experimental design and revised the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors. The sequencing data have been deposited into GEO with the accession number GSE120046 (https://tanglab.shinyapps.io/Human_Fetal_Brain_Cell/). The plasmid can be provided by X.W. with pending scientific review and a completed material transfer agreement. Requests for the plasmid should be submitted to X.W.
View Abstract

Stay Connected to Science Advances

Navigate This Article