2020 Virtual AIChE Annual Meeting

(505c) Single-Cell Transcriptional Dynamics of Retinal Ganglion Cell Diversification

Authors

Shekhar, K., UC Berkeley
Vision begins in the retina, a neural tissue that lines the back of the eye. The retinal circuitry that processes visual information is composed of six major neuronal classes: rods, cones, amacrine cells, horizontal cells, bipolar cells, and retinal ganglion cells (RGCs), each of which contains numerous types with distinct morphological, physiological and molecular properties. Despite accounting for less than 2% of retinal cells, the RGC class is highly diverse (i.e. contains numerous types), and its axons project the retina’s output to the rest of the brain. Recent advances in single-cell RNA sequencing (scRNA-seq) have revealed the existence of 46 distinct RGC types in mice. Here, we interrogate this remarkable diversity during retinal development by decoding the logic of RGC specification and identifying the key molecular players that orchestrate it.

Using scRNA-seq, we profiled mouse RGCs at 5 stages of development: embryonic days 13, 14 (during and slightly after the period of peak genesis), 16 (as RGC axons begin to reach targets) along with postnatal day 0 (as dendrite elaboration begins) and 5 (shortly after they begin to receive synapses). These data were combined with a recently published transcriptomic atlas of adult mouse RGCs [3] to complete a six-time point developmental dataset that enabled the investigation of 75,115 RGCs. As the destructive nature of scRNA-seq precludes direct capture of the lineage dynamics of developing cells, we employ a computational approach termed RNA velocity [1-2] to infer transcriptional dynamics. RNA velocity leverages the relative abundance of exonic and intronic reads within the framework of a dynamical model to infer a cell’s future state.

A key assumption of the original RNA velocity framework was that the underlying dynamical process is at steady state, which is violated during early development. To apply RNA velocity to RGC diversification, we use scVelo [2], which applies an expectation-maximization framework to extend RNA velocity estimation to transient processes. We estimate an RNA velocity vector for each immature RGC in our dataset, and use it to resolve developmental asynchrony within each time point and connect RGC states between consecutive time points. We also compare velocity-based estimates of temporal couplings between cells to those based on Optimal Transport [4], an alternative method to infer developmental relationships in time series scRNA-seq data. We then use the velocity estimates to identify genes that contribute most to driving velocity at each time point and evaluate their biological significance through gene ontology enrichment analyses. Assessing how these driver genes change over time will enable understanding of the dynamic processes underlying RGC diversification and maturation. Comparing the temporal variation of these driver genes to that of differentially expressed, transcriptomic cluster-defining genes will reveal the interplay between absolute gene expression and gene expression dynamics. Taken together, these efforts will further our understanding of the transcriptional logic of RGC diversification and identify novel candidate genes for experimental investigation.

[1] La Manno et. al., RNA velocity of single cells, Nature (2018)

[2] Bergen et. al., Generalizing RNA velocity to transient cell states through dynamical modeling, bioRxiv (2019)

[3] Tran et. al., Single-Cell Profiles of Retinal Ganglion Cells Differing in Resilience to Injury Reveal Neuroprotective Genes, Neuron (2019)

[4] Schiebinger et. al., Optimal-Transport Analysis of Single-Cell Gene Expression Identifies Developmental Trajectories in Reprogramming, Cell (2019).