Experimental fine-mapping of QTLs from scRNA-seq data

One of the fundamental problems in translating genetic associations with disease into actionable insights is identifying the molecular mechanism by which they operate. What do we mean by “identifying a molecular mechanism”? Ideally, we would identify a cell type of action, a collection of regulatory elements, a target gene, and collection of upstream regulators (e.g., Smemo et al. 2013, Claussnitzer et al. 2015, Sekar et al. 2016).

Considerable progress has been made on identifying genomic annotations enriched for associations, which can potentially be used to identify putative cell types of action (e.g., Maurano et al. 2012, Gusev et al. 2015, Finucane et al. 2015, Finucane et al. 2016). Progress has also been made on statistically identifying putative causal variants directly from genotype and phenotype, possibly using enrichments as prior information (e.g., Hormozdiari et al. 2014, Pickrell 2014, Kichaev et al. 2014, Hormozdiari et al. 2014, Chung et al. 2014, Li and Kellis 2016). Recent attention has focused on the statistical problem of identifying the target gene from genotype, phenotype, and gene expression data (e.g., Gamazon et al. 2015, Gusev et al. 2016, Mancuso et al. 2017, Zhu et al. 2018).

Together, these statistical approaches can potentially be combined to identify a putative mechanism by which a disease-associated locus causes disease risk/onset. However, they are fundamentally limited by the availability of cell type–specific molecular data at the transcriptomic and epigenomic levels, which are required to derive relevant genomic annotations. These limitations have, in part, motivated recent efforts to generate relevant single cell genomic data (e.g., Regev et al. 2017).

In parallel, experimental approaches have been developed to identify functional regulatory elements using transfected reporter constructs (e.g., Melkinov et al. 2012, Arnold et al. 2013, Tewhey et al. 2016, Ernst et al. 2016, Wang et al. 2018), or more recently, combining CRISPR-perturbation with scRNA-seq phenotyping (e.g., Adamson et al. 2016, Dixit et al. 2016, Jaitin et al. 2016, Xie et al. 2017, Datlinger et al. 2017). These approaches have been limited by the cell type in which experiments are performed.

Here, we outline an alternative approach, which exploits the fact that naturally occuring stochastic variation in protein abundances of upstream regulators will lead to downstream variation in mRNA levels of target genes. This fact suggests that it could possible to identify upstream regulators, functional regulatory elements, and target genes directly from scRNA-seq data. Such an approach could avoid problems of collecting large numbers of donor individuals, and of picking the correct cell type/developmental time point a priori. Further, it could more directly answer the relevant questions for interpreting disease-associated loci. Longer term, this methodology could be extended to spatial transcriptomic measurements, allowing us to account for microenvironments and cell-cell interactions.