RESEARCH NEURODEVELOPMENT bined sample (Fig. 1A and fig. S1). At P0, spinal neurons have been generated and the Conserved genetic signatures parcellate cardinal basic functional features of adult spinal cir- cuitry have been established, but expression spinal neuron classes into local and persists for many developmental markers of neuronal subtype diversity (2, 9, 10, 18). Using projection subsets marker analyses, 6743 cells passed quality con- trol and were identified as neurons (Fig. 1A). Peter J. Osseward II1,2†, Neal D. Amin1‡, Jeffrey D. Moore3, Benjamin A. Temple1,2, Glutamatergic and GABAergic neurons largely Bianca K. Barriga1,4, Lukas C. Bachmann1, Fernando Beltran Jr.1, Miriam Gullo1, Robert C. Clark1, segregated in the uniform manifold approxi- Shawn P. Driscoll1, Samuel L. Pfaff1*, Marito Hayashi1†§ mationandprojection(UMAP)space,anda graph-based comparison using the highly vari- Motor and sensory functions of the spinal cord are mediated by populations of cardinal neurons arising from able genes conservatively identified 45 distinct separate progenitor lineages. However, each cardinal class is composed of multiple neuronal types with clusters(Fig.1,BandC).Asexpected,manyof distinct molecular, anatomical, and physiological features, and there is not a unifying logic that systematically these clusters were enriched for genes that have accounts for this diversity. We reasoned that the expansion of new neuronal types occurred in a stepwise been previously identified as marker genes for manner analogous to animal speciation, and we explored this by defining transcriptomic relationships using cardinal classes (Fig. 1D and table S1) (1, 2). a top-down approach. We uncovered orderly genetic tiers that sequentially divide groups of neurons by To explore the conserved relationships their motor-sensory, local-long range, and excitatory-inhibitory features. The genetic signatures defining that exist among lumbar spinal neurons, we neuronal projections were tied to neuronal birth date and conserved across cardinal classes. Thus, the developed a divisive clustering pipeline using intersection of cardinal class with projection markers provides a unifying taxonomic solution for the k-means algorithm to split spinal neurons Downloaded from systematically identifying distinct functional subsets. into pairwise groups iteratively on the basis of their overall transcriptional composition (Fig. 1, E and F). We focused on transcription factors rom a functional perspective, spinal neu- One perspective of neuronal diversity is that becausetheyregulateneuronalidentityand rons can be divided along several axes, it arose through evolution in order to expand function (2, 14, 15). We anticipated that this including motor-sensory, excitatory- neural functions (14–16). In this view, primi- approach would first reveal the glutamatergic http://science.sciencemag.org/ F inhibitory, and locally connected neurons tive neuron types served as the precursors and GABAergic neurotransmitter division in for intracord processing versus projection for more specialized subtypes, leading to the the cell population because it is such a promi- neurons for communication with the brain prediction that individualneuronalattributes nent phenotypic difference, and the two cell (1–6). Neuron heterogeneity is often charac- mayhaveemergedinastepwisefashionwith types were isolated from separate litters. How- terized on the basis of landmarks such as ordered hierarchical relationships (14, 16, 17). ever, the neuronal populations split along an their neurotransmitter type, connectivity, cyto- This analogy to animal speciation prompted us axis unrelated to neurotransmitter identity (Fig. architecture, morphology, physiology, develop- to investigate whether molecular and cellular 1, B and G) (11, 19). Differential gene expression mental origin, and genetic profile (7). In the correlates for this type of stepwise diversifica- of the first division indicated that transcription spinal cord, separate cardinal neuron classes tion could be detected among large heteroge- factor Ebf1 is highly expressed in a cell group arise from molecularly distinct progenitor neous populations of spinal neurons. Thus, we defined as group-E neurons, whereas cells arrayed along the dorsoventral axis of rather than focusing on fine-grained molecular transcription factor Hoxc10 is enriched in a on April 30, 2021 the neural tube (2, 8). Neurons within each differences that may or may not correspond different group we termed group-H neurons cardinal class share properties such as the to functional cellular features, we used a top- (Fig. 1G). Histological analysis of Ebf1+ and same neurotransmitter identity and have been down approach to identify the transcriptomic Hoxc10+ neurons identified a spatial divi- targeted in functional studies (1, 2). However, signatures linked to the emergence of differ- sion of these groups (Fig. 2, A and B, and fig. it is increasingly apparent that each cardinal ent neuronal types. At each branch point in S2). Ebf1+ neurons were located in laminae I class is itself composed of heterogeneous popu- our established hierarchy, we defined the to III of the superficial dorsal horn, corresponding lations of neuron types, impeding a detailed molecular, developmental, cytoarchitectural, to the site where many exteroceptive sensory understanding how spinal circuits function neurotransmitter, and connectivity properties inputs terminate. By contrast, Hoxc10+ neurons (9–13). We sought to determine whether there of the neurons. Our analysis uncovered molec- were predominantly located in laminae IV to was a coherent logic for spinal neuron diver- ular markers for local and projection neurons X, areas involved in proprioception and motor sification in mice that could be used to sys- regardless of cardinal class or neurotransmitter. control (3, 4). Although group-E and group-H tematically describe the heterogeneity within By combining markers for cardinal classes neurons broadly differ in motor-sensory func- all the cardinal classes. with those for conserved genetic signatures tions associated with their positions, both types for projection status, we established a simple arecomposedofamixtureofglutamatergic combinatorial matrix that systematically iden- and GABAergic neurons (Fig. 1, B and G, and 1Gene Expression Laboratory, Salk Institute for Biological tified discrete subsets of spinal neurons. fig. S3) (3, 4). Studies, La Jolla, CA 92037, USA. 2Neurosciences Graduate Program, University of California, San Diego, 9500 Gilman Drive, Group-E and -H neurons divide the spinal Group-N and -Z neurons are arrayed along the 3 La Jolla, CA 92037, USA. Howard Hughes Medical Institute, motor-sensory circuitry mediolateral axis of the spinal cord Department of Molecular and Cellular Biology, Center for Brain Science, Harvard University, Cambridge, MA, USA. 4Biological To enrich for neurons while minimizing tech- Using the hierarchical-divisive strategy outlined Sciences Graduate Program, University of California, San Diego, nical and biological variability, we genetically above (Fig. 1F), we further split the group-E 9500 Gilman Drive, La Jolla, CA 92037, USA. †These authors contributed equally to this work. labeled glutamatergic and g-aminobutyric neurons into two subgroups and found that ‡Present address: Department of Psychiatry and Behavioral acid (GABA)–ergic neurons in postnatal day this division identifies glutamatergic-related Sciences, Stanford University, Stanford, CA 94305, USA. §Present 0 (P0) mice, sorted lumbar spinal neurons and GABAergic-related genes and their cor- address: Howard Hughes Medical Institute, Department of Cell 2 4 Biology, Harvard Medical School, Boston, MA 02115, USA. separately, pooled them, and performed single- responding neurons (Fig. 1, B and H) ( , ). *Corresponding author. Email: [email protected] cell RNA-sequencing (scRNA-seq) on the com- Division of the group-H population produced Osseward et al., Science 372, 385–393 (2021) 23 April 2021 1of8 RESEARCH | RESEARCH ARTICLE A postnatal day 0 B Glutamatergic GABAergic 24 C 12 D Glutamatergic Vglut2:Cre; Vgat:Cre; 21 tdTomato tdTomato 40 dI1 18 02 dI3 17 31 28 01 29 16 dI5/LB 010203 09 12 16 19 21 24 28 3229 40 32 V2a 35 39 MN 15 31 18 27 09 03 38 35 15 19 V3 17 37 25 23 pool 38 GABAergic 22 06 scRNA-seq 11 04 03 dI4 0405 10 200733 34 36 41 100% 43 13 44 dILA 08 302614 42 41 34 10 05 dI6 22 50% 36 20 07 33 V1 11 13 23 37 0% V2b 11 23 UMAP2 08 14 30 26 45 neuron UMAP1 non-neuron E F Group-E Group-H functions? spinal neurons enriched enriched k-means 2 Group-E 300 Tshz2 cluster 1 Ebf1 Ebf2 Fig. 1G Cck Hoxc10 ... Npy UMAP 2 Nefl cluster 4 Frzb Nap1l5 200 UMAP 1 k-means 2 k-means 2 Zfhx3 Elavl2 Fig. 1H Fig. 1I Hoxd10 Nova1 Downloaded from functional 100 Cplx1 feature 1? Group-H −log10(p−value) k-means 2 k-means 2 Group-E functional Fig. 1J Fig. 1K feature 2? 0 −2 0 2 log2 Fold Change Group-Z Group-N Group-E (Glut) Group-E (GABA) enriched enriched HIenriched enriched http://science.sciencemag.org/ 300 Nfix Group-E (Glut) 300 Nfib Npy Zfhx3 Tcf4 Cck Zic1 Group-Z Group-N 200 Zic2 200 Sst Hoxb9 Lhx1 Prox1 Gad1 Nefl Sncg Tlx3 Pax2 100 100 Neurod2 −log10(p−value) Slc6a1 −log10(p−value) Zfhx4 Slc17a6 Otp Esrrg Foxp2 0 0 −2 −1 0 1 2 Group-E (GABA) −2 0 2 log2 Fold Change log2 Fold Change Group-N (Glut) Group-N (GABA) Group-Z (GABA) Group-Z (Glut) J enriched enriched K enriched enriched on April 30, 2021 Tfap2b 200 Slc17a6 150 Group-N Group-Z 150 (Glut) (Glut) Slc6a5 Pou4f1 Slc32a1 100 100 Npy Gad1 Slc17a6 Gad2 −log10(p−value) Slc6a5 50 −log10(p−value) 50 En1 Group-N Lhx1 Group-Z Lhx1 (GABA) Gad1 (GABA) Pax2 0 0 −2 0 2 −2 −10 1 2 log2 Fold Change log2 Fold Change Fig. 1. scRNA-seq identifies relationships among groups of spinal neurons. populations can be used to define branch points. The functional differences at the (A) Lumbar segments from P0 Vglut2:Cre; Ai14 and Vgat:Cre; Ai14 neonates (two branch points can be used to identify increasingly more specific neuronal females and one male from each litter) were microdissected. tdTomato+ cells were attributes linked to function.