As an embryo develops, it needs to generate a myriad of cell types, all in specific locations relative to each other and with correct abundances. Retina, for example, contains more than a hundred neuronal subtypes, organized into three layers of cell bodies and two layers of neuropil. These cell types have different features and frequencies and have to be tiled across the retina in a specific way to support the overall function of the tissue. This seemingly impossible task is accomplished during development with incredible robustness. All the instructions for making retina, as well as all the other tissues in the body, has to be encoded in the genome. So we hypothesize that the blueprints for making tissues like retina are compressible and involve relatively simple principles. Our goal is to identify these principles and utilize them to develop new therapeutic strategies for neurodegenerative disorders.
Our approach involves imaging based genetic barcoding for tracing the lineage and molecular history of individual cells, spatial transcriptomics for mapping cell states, computational modeling for interpreting the results, and synthetic biology for developing molecular tools to manipulate cell fate decisions.
Question: Are there distinct classes of progenitor cells in the retina?
Tools: Spatial Transcriptomics, Lineage Tracing
All neurons and glia of the retina are made by retinal progenitor cells (RPCs). These cells are multipotent. They can give rise to virtually any combination of retinal cell types. The clones that individual progenitors make are highly variable, both in the number of cells they contain and their identity. Do RPCs know what cell types their are going to make in advance? or are differences in the fate outcome of RPCs due to extrinsic and stochastic factors? The answer to this question has profound implications for our ability to make retinal cells for transplantation or research purposes.
Single cell RNA sequencing (scRNAseq) has shown significant heterogeneity in gene expression of RPCs. But distinct clusters are not identified among RPCs at any given developmental stage based on scRNAseq data alone. We hypothesize that to classify RPCs, their gene expression state has to considered in conjunction with their spatial context. Informed with the existing scRNAseq data, we use spatial transcriptomics to systematically identify RPC subtypes. In the process, we also develop innovative new approaches to scale up in situ gene expression profiling and integrate data across multiple modalities. Using our image readable barcode libraries, we can also trace the lineage of many progenitors in the same sample and classify RPCs based on their diverse fates.
Question: How signals that cells receive early on relate to their eventual fate?
Tool: Molecular recording
Cells use signaling pathways to orchestrate their behavior during development. A surprisingly limited number of key signaling pathways are used to elicit numerous cellular responses. Cells use intensity and order of these signals to interpret their meaning and make appropriate decisions.
We use genome engineering and imaging based barcoding to link the intensity and order of signals that a cell receives to the phenotype of its progeny. If mutations in an array of barcodes are made at a rate proportional to the intensity of a signal, a permanent record of that signal will be made in the genome of the cells. Mutation rate in our genetic recording system depends on the expression level of the CRISPR base editor and the guide RNA. By coupling the expression of base editor or gRNA to the signal of interest we can develop a molecular recording system and explore how early events in the progenitors are linked to cellular diversification. The same scheme can also be used to study how aberrant signaling leads to cancer and developmental disorders.
Question: To what extent the lineage of a cell affect its fate?
Tool: Lineage recording
Neuroblasts in Drosophila and C. elegans produce neurons and glia in a stereotypical sequence instructed by cell fate regulators. Neurons and glia of the mouse retina are also born in an evolutionarily conserved order, although with significant overlap between the cell types. It is tempting to think that mammalian retinal progenitors also transition through a reproducible sequence of competence states and the overlap is only caused by asynchrony between lineages. However, this process in vertebrates may be regulated differently and allow for more stochastic cell fate choices.
Knowing the set of cell types that each progenitor makes is not enough to resolve this type of issues. We need to know the sequence of cell fate decisions along individual lineage trees. In systems where time lapse microscopy is not feasible, either due to inaccessibility of the tissue or the time scale of development, reconstruction of lineage trees poses a significant challenge. Lineage recording, in which mutations accumulate in heritable barcodes as cells grow, offers a promising solution to this problem. We have developed a synthetic system in which a CRISPR base editor can gradually make stochastic mutations in an array of compact image readable barcodes. This approach allows reconstruction of lineage relationships between cells. Analysis of the reconstructed trees will reveal the extent to which programmed lineages versus stochastic processes operate in cell fate decisions.
We are generously supported by: