By E. Braun, M. Danan-Gotthold, L. E. Borm, E. Vinsland, K. W. Lee, P. Lönnerberg, L. Hu, X. Li, X. He, Z. Andrusivova, J. Lundeberg, E. Arenas, R. A. Barker, E. Sundström, S. LinnarssonSourcebioRxiv
Fork
Clone
git clone https://openversions.com/git/index/biorxiv-10-1101-2022-10-24-513487v1.git
Cite
Version link
https://openversions.com/index/biorxiv-10-1101-2022-10-24-513487v1/commit/e29b257c2bb874a6e8f7f02c305912bd4f9bf03f
Version identifier (commit hash)
e29b257c2bb874a6e8f7f02c305912bd4f9bf03f

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

Latest version

Checks complete

Archive from bioRxiv Source: https://www.biorxiv.org/content/10.1101/2022.10.24.513487v1 License: https://creativecommons.org/licenses/by-nc-nd/4.0/ Manuscript unchanged.

Committed

Claims

Findings

Headline finding

Using single-cell RNA sequencing of 1,665,937 high-quality cells from 26 human brains at 5–14 post-conceptional weeks (p.c.w.) and spatial transcriptomics at 5 p.c.w., we identified 12 major cell classes and 616 distinct cell states that map to precise anatomical domains and reveal the emergence of cellular diversity during first-trimester human brain development.

Over 5–12 p.c.w., neurons and neuroblasts maintained relatively constant abundances, neuronal IPCs, immune, oligodendrocyte, and vascular lineages increased with age, and the major developmental change was a radial glia-to-glioblast switch occurring around 10.5, 9.5, and 7.5 p.c.w. in the forebrain, midbrain, and hindbrain, respectively.

In the developing human neocortical excitatory lineage (EMX1+), the vast majority of IPCs start the cell cycle as radial glia-like IPCs, make a fate decision between self-renewal and differentiation at late G1 or early S phase, and execute differentiation after mitosis before re-entering the cell cycle.

In the cortical excitatory lineage, most cycling radial glia are in G1 phase whereas most cycling IPCs are in S phase, and based on RNA velocity magnitude, late radial glia/glioblasts have a higher rate of gene expression change than IPCs, indicating that IPCs have a longer S phase and, at later time points, a longer cell cycle than radial glia/glioblasts.

Neuronal-like IPCs (expressing BCL11B, NEUROD6, and INA but not SOX2) have significantly higher total RNA molecule (UMI) counts throughout the cell cycle than radial glia-like IPCs (SOX2+).

Between early and late states of the cortical excitatory lineage, radial glia and glioblasts exhibit 728 differentially expressed genes—enriched for extracellular matrix, focal adhesion, and axon guidance in late radial glia—and show a significant overlap with mouse early-late radial glia genes, including genes previously defined as basal radial glia markers (TNC, HOPX, FAM107A).

Human forebrain GABAergic interneurons (DLX2+) develop along branched trajectories from radial glia/glioblasts and two types of IPCs without expressing the neuroblast marker NHLH1, with MGE-derived neurons migrating to the cortex by 6 p.c.w. and CGE-derived neurons by 9 p.c.w., while non-telencephalic (FOXG1-) DLX2+ interneurons in the thalamus and hypothalamus likely originate locally (including from FOXD1+ progenitors) rather than from the telencephalon, except for a potential MGE-to-thalamus migration of CRABP1+ interneurons at 14 p.c.w.

Putative human glioblasts (BCAN+ TNC+, 44 clusters) emerge from 6 p.c.w.—earliest in the hindbrain, followed by midbrain and forebrain—and differentiate into region-specific pre-astrocytes (AQP4+ GJA1+) in the diencephalon, midbrain, and hindbrain, as well as pre-OPCs (EGFR+ DLL3+).

Developing human OPCs (8 clusters, 6,190 cells) and COPs emerge as early as 5.5–6 p.c.w. in the hindbrain and ~8 p.c.w. in the midbrain and forebrain without producing mature oligodendrocytes by 14 p.c.w., exhibit distinct anteroposterior transcriptional identities (including FOXG1, EN1, HOXD3, PRRX1, AGMO, and ZNF703) already established in cycling OPCs, and show both conserved (32%, e.g., PDGFRA, ENPP6, MAG) and species-specific (e.g., human RIT2 and BMPER vs. mouse Cdo1 and Cbln2) differential gene expression compared to mouse.

Methods

Human embryonic and fetal brain tissues (5–14 p.c.w.) were dissociated using Worthington's Papain Dissociation System and sequenced using 10X Genomics Chromium Single Cell 3' Reagent Kits v2 and v3 on an Illumina NovaSeq 6000.

Source excerpts

“All brain regions were dissociated separately using the Worthington’s Papain Dissociation System (Worthington) (Protocols.io, https://dx.doi.org/10.17504/protocols.io.xmbfk2n).” · Show in paper

“Single cells were captured using the droplet-based single-cell RNA sequencing platform Chromium (10X Genomics). Roughly half of the sampling was done using the Chromium Single Cell 3' Reagent Kits Version 2 and the other half with Version 3 (see detailed sampling in table S1).” · Show in paper

Spatial gene expression at 5 p.c.w. was profiled using Enhanced Electric fluorescent in-situ hybridization (EEL FISH) targeting 440 genes and 10X Genomics Visium Spatial Gene Expression, with single-cell clusters mapped to spatial bins using FISHscale and Bonefight.

Source excerpts

“440 genes were selected to cover major cell types in the human developmental brain using single-cell RNA-seq data and literature (table S3). EEL FISH was performed as previously described (8) with a minor change in the digestion step, that was performed two times 10 minutes with a concentration of 1 U/ml of proteinase K.” · Show in paper

“Mean expression profiles of single cell RNA-seq clusters of cells from 5 week old samples were manually selected and mapped to the hexagonally binned spatial data using FISHscale and Bonefight (8, 10).” · Show in paper

Sequencing reads were demultiplexed with cellranger mkfastq v4.0.0, quantified with STARSolo v2.7.10a against a custom-filtered GRCh38.p13 Gencode V35 reference, and quality-controlled using DoubletFinder and a 2D Gaussian maximum likelihood estimate on log(total UMIs) versus unspliced UMI fraction.

Source excerpts

“Sequencing runs were demultiplexed with cellranger mkfastq version 4.0.0 (10x Genomics) and filtered through the index-hopping-filter tool version 1.1.0 (10x Genomics). Unique molecular identifier (UMI) counts were determined using STARSolo version 2.7.10a” · Show in paper

“First, we used the DoubletFinder (73) algorithm to mark putative doublets. Next, we used the relationship between total UMI count and the fraction of unspliced UMIs (fig. S2A-D) to classify droplets” · Show in paper

Global clustering and annotation were performed with the cytograph-shoji pipeline using Harmony batch correction, Leiden clustering, support-vector classification for orphan cells, and trinarization scores, while cortical excitatory lineage integration used Scanorama and scVI.

Source excerpts

“We used the latest version of our in-house cytograph pipeline, which implements standard analysis steps in a modular and scalable fashion. The version used here is called cytograph-shoji and uses a custom tensor database called shoji.” · Show in paper

“We selected 1000 highly variable genes for the Scanorama analysis and 700 highly variable genes for the scVI analysis using scanpy.pp.highly_variable_genes based on 'seurat_v3' flavor (79) and performed dimensionality reduction and batch correction for each p.c.w. using Scanorama (14) and for all time points using the scVI model (15) as implemented in scvi-tools (version 0.15.5) (80).” · Show in paper

RNA velocity (scVelo), DeepCycle cell-cycle phase inference, Milo differential abundance testing (milopy), a sklearn RandomForestClassifier for region provenance, and diffxpy differential gene expression testing were applied for downstream trajectory and comparative analyses.

Source excerpts

“To compute RNA velocity we used scvelo package (version 0.2.4) (81) and run stochastic mode analysis on each batch (cortical subregion or donor as needed) that had more than 1000 cells in every p.c.w.” · Show in paper

“Differences in cell abundances associated with post conceptional age was tested using the Milo framework (26), with the Python implementation milopy (https://github.com/emdann/milopy).” · Show in paper

Cell-class proportions across developmental ages were normalized to expected tissue volume (excluding ventricles) based on morphometric estimates of forebrain, midbrain, and hindbrain size.

Source excerpts

“We therefore normalized our sampling to the expected tissue volume (excluding ventricles) based on morphometric estimates of forebrain, midbrain and hindbrain size by age (83, 84).” · Show in paper