SUMMARY Patterns of gene expression can be used to characterize and classify neuronal types. It is challenging, however, to generate taxonomies that fulfill the essential criteria of being comprehensive, harmonizing with conventional classification schemes, and lacking superfluous subdivisions of genuine types. To address these challenges, we used massively parallel single-cell RNA profiling and optimized computational methods on a heterogeneous class of neurons, mouse retinal bipolar cells (BCs). From a population of ~25,000 BCs we derived a molecular classification that identified 15 types including all types observed previously, and two novel types, one of which has a non-canonical morphology and position. We validated the classification scheme and identified dozens of novel markers using methods that match molecular expression to cell morphology. This work provides a systematic methodology for achieving comprehensive molecular classification of neurons, identifies novel neuronal types, and uncovers transcriptional differences that distinguish types within a class.

Neuronal types in the central nervous system differ dramatically in their resilience to injury or other insults. Here we studied the selective resilience of mouse retinal ganglion cells (RGCs) following optic nerve crush (ONC), which severs their axons and leads to death of $80% of RGCs within 2 weeks. To identify expression programs associated with differential resilience, we first used single-cell RNA-seq (scRNA-seq) to generate a comprehensive molecular atlas of 46 RGC types in adult retina. We then tracked their survival after ONC; characterized transcriptomic, physiological, and morphological changes that preceded degeneration; and identified genes selectively expressed by each type. Finally, using loss-and gain-of-function assays in vivo, we showed that manipulating some of these genes improved neuronal survival and axon regeneration following ONC. This study provides a systematic framework for parsing type-specific responses to injury and demonstrates that differential gene expression can be used to reveal molecular targets for intervention.

Amacrine cells (ACs) are a diverse class of interneurons that modulate input from photoreceptors to retinal ganglion cells (RGCs), rendering each RGC type selectively sensitive to particular visual features, which are then relayed to the brain. While many AC types have been identified morphologically and physiologically, they have not been comprehensively classified or molecularly characterized. We used high-throughput single-cell RNA sequencing to profile .32,000 ACs from mice of both sexes and applied computational methods to identify 63 AC types. We identified molecular markers for each type and used them to characterize the morphology of multiple types. We show that they include nearly all previously known AC types as well as many that had not been described. Consistent with previous studies, most of the AC types expressed markers for the canonical inhibitory neurotransmitters GABA or glycine, but several expressed neither or both. In addition, many expressed one or more neuropeptides, and two expressed glutamatergic markers. We also explored transcriptomic relationships among AC types and identified transcription factors expressed by individual or multiple closely related types. Noteworthy among these were Meis2 and Tcf4, expressed by most GABAergic and most glycinergic types, respectively. Together, these results provide a foundation for developmental and functional studies of ACs, as well as means for genetically accessing them. Along with previous molecular, physiological, and morphologic analyses, they establish the existence of at least 130 neuronal types and nearly 140 cell types in the mouse retina.

Summary The sizes of different neuronal populations within the central nervous system are precisely controlled, but whether neuronal number is coordinated between cell types is unknown. We examined the covariance structure of twelve different retinal cell types across thirty genetically distinct lines of mice, finding minimal co-variation when comparing synaptically-connected or developmentally-related cell types. Variation mapped to one or more genomic loci for each cell type, but rarely were these shared, indicating minimal genetic co-regulation of final number. Multiple genes, therefore, participate in the specification of the size of every population of retinal neuron, yet genetic variants work largely independent of one another during development to modulate those numbers, yielding substantial variability in the convergence ratios between pre- and post-synaptic populations. Density-dependent cellular interactions in the outer plexiform layer overcome this variability to ensure the formation of neuronal circuits that maintain constant retinal coverage and complete afferent sampling.

The network of starburst amacrine cells plays a fundamental role in the neural circuitry underlying directional selectivity within the retina. Individual sectors of the starburst dendritic field are directionally selective by virtue of a mutually inhibitory relationship between starburst amacrine cells with overlapping dendrites. These features of the starburst amacrine cell network suggest that starburst cells regulate their dendritic overlap to ensure a uniform coverage of the retinal surface. The present study has compared the dendritic morphology of starburst amacrine cells in two different strains of mice that differ in starburst amacrine cell number. The A/J (A) strain contains about one-quarter fewer starburst amacrine cells than does the C57BL/6J (B6) strain, although the mosaics of starburst amacrine cells in both strains are comparably patterned. Dendritic field size, however, does not compensate for the difference in density, the A strain having a slightly smaller dendritic field relative to the B6 strain, yielding a significantly larger dendritic coverage factor for individual cells in the B6 strain. The area of the distal (output) annulus of the dendritic field occupies a comparable proportion of the overall field area in the two strains, but overlapping annuli establish a finer meshwork of co-fasciculating processes in the B6 strain. These results would suggest that the architecture of the dendritic network, rather than the overall size of the dendritic field, is dependent on the density of starburst amacrine cells.