Single cell transcriptomics reveals spatial and temporal dynamics of gene expression in the developing mouse spinal cord

Key results

1. ShownThe atlas reproduces known spinal cord composition and growth. Progenitor fractions shrink over time, and domain dynamics match an independent quantification. The data also extend these dynamics past e11.5, showing dorsal progenitors depleting faster than ventral ones by e13.5.

   41,025 cells sequenced; 38976 kept after QC (cells with >6% mitochondrial UMIs or <500 genes removed); 33754 (87%) assigned to tissues; about 1% mixed neural-mesoderm transcriptomes; progenitors fell from >65% of cells at e9.5 to <15% at e13.5; FP, p3, pMN, pI and pD trajectories closely matched Kicheva et al (2014). · Assignment of transcriptomes to cell identities; Developmental dynamics of cell types; Quality filtering

2. ShownTesting every combination of dorsoventral domains for differential expression recovers known neurotransmitter and transcription factor patterns and predicts new ones. These include an adhesion-gene split between inhibitory and excitatory neurons and Cldn3 expression in motor neurons that is limited to early stages.

   8192 progenitor models gave 102 pattern categories and 4096 neuronal models gave 147 (P < 10^-9, log2-fc > 2); 9 adhesion genes, with Nrxn3 in inhibitory classes and Cbln1 in excitatory classes plus MNs; immunofluorescence showed Cldn3 in MNs at e10.5 but not at e11.5/e12.5, and Pou3f1 in dorsal dI1-dI3 neurons at e11.5. · Prediction of novel gene expression patterns; Combinatorial testing for differential…

3. ShownGene-module-based hierarchical subclustering of each neuron class recovers known subtypes and predicts a new Nfia/b/x and Neurod2/6 V3 subtype. Staining links Pou2f2, Nr2f1 and Nkx6.2 to specific subsets of V3 neurons, motor neurons and V2a neurons.

   59 subpopulations, of which two (MN.5, V2a.5) were judged misassigned; at least 4 V3 clades and four major V2a subtypes; Foxp1+ MNs split into cervical LMC (MN.6) and thoracic PGC (MN.7); staining at e11.5 and e12.5 confirmed Pou2f2 in V3 neurons, Nr2f1 in Foxp1+ MNs, and Nkx6.2 in V2a cells lacking Shox2 and Nfib. · Clustering of neurons predicts novel neuronal…; Subclustering of neuronal populations

4. SuggestedThe same transcription factor modules subdivide several ventral neuron classes and switch on in the same temporal order in each. This suggests that birth time adds a second axis of neuronal subtype diversity alongside dorsoventral patterning.

   Pou2f2 modules (usually with Zfhx2/3/4) and Nfib modules (often with Nfia, Nfix, Neurod2, Neurod6, Tcf4) recur in V0-V3 and MNs. Fractions of Onecut2+ and Pou2f2+ neurons peak before e11.5, while Nfib and Neurod2 come on at e12.5 or e13.5. Staining showed Pou2f2 in neurons at e10.5; Nfib was absent from neurons at e11.5 but present at e13.5. · Modular and temporal specification of neuronal…; Temporal specification of neuronal subtype…; Figure legends

5. ShownPseudotime ordering along the first principal component, rebuilt separately for each domain, predicts that domain-specific transcription factors rise briefly as neurogenesis begins, while broadly expressed progenitor factors do not. Immunostaining confirmed this.

   PCA on 100 genes (114 from 4 modules minus 14 with dorsoventral bias); spline fits for 36 regionally patterned genes. Olig3 peaked with Neurog1/Neurog2 in dp2/dp3, Dbx1 rose in Neurog1+ p0 cells, and Nkx2.2 rose in Olig3+ V3 neurons at e11.5. Msx1, Pax3, Pax6 and Sox2 showed no such rise. · Reconstruction of the gene expression dynamics…; Dynamics of gene expression during neurogenesis; Neuronal trajectories and Pseudotime…; Figure legends