Gene module reconstruction elucidates cellular differentiation processes and the regulatory logic of specialized secretion

During differentiation, cells become structurally and functionally specialized, but comprehensive views of the underlying remodeling processes are elusive. Here, we leverage scRNA-seq developmental trajectories to reconstruct differentiation, using two secretory tissues as a model system – the zebrafish notochord and hatching gland. First, we present an approach (MIMIR) to integrate expression and functional similarities for gene module identification, revealing dozens of gene modules representing known and newly associated differentiation processes and their temporal ordering. Second, we focused on the unfolded protein response (UPR) transducer module to study how general versus cell-type specific secretory functions are regulated. By profiling loss- and gain-of-function embryos, we found that the UPR transcription factors creb3l1, creb3l2, and xbp1 are master regulators of a general secretion program. creb3l1/creb3l2 additionally activate an extracellular matrix secretion program, while xbp1 partners with bhlha15 to activate a gland-specific secretion program. Our study offers an integrated approach for functional gene module identification and illustrates how transcription factors confer general and specialized cellular functions. Overall design: scRNA-seq experiments were conducted to study the regulatory roles of the unfolded protein response (UPR) pathway transcription factors (TFs) during zebrafish embryogenesis. We focused on the four UPR TFs (xbp1, atf6, creb3l1, and creb3l2) that are expressed during the early differentiation of the notochord and/or the hatching gland. We hypothesized that the UPR TFs regulate the secretory machinery genes in these two cell types. To identify the endogenous target genes of the UPR TFs, we performed scRNA-seq of UPR TF loss-of-function embryos at 12 hours post fertilization (hpf), when UPR TFs and secretory pathway genes are normally highly expressed in the notochord and hatching gland. Loss-of-function embryos were generated by CRISPR-mediated mutagenesis. Crispants were created by injecting Cas9 protein and guide RNAs (gRNAs) into 1-cell stage embryos. We additionally raised a stable mutant line for creb3l1. Co-injecting Cas9 with gRNA targeting multiple genes in wild-type or the stable mutant embryos allowed testing for redundant functions by generating double, triple, and quadruple loss-of-function embryos. Tyrosinase (tyr) crispants, which do not exhibit developmental defects, were generated as the control. Using this loss-of-function dataset, we defined target genes for UPR TFs as the genes whose expression levels changed in corresponding crispants and/or mutants. Next, we performed scRNA-seq on UPR TF gain-of-function embryos to test the sufficiency of the UPR TFs in regulating their endogenous targets. To generate gain-of-function embryos, we globally mis-expressed each UPR TF by injecting mRNAs encoding their activated forms into 1-cell stage embryos. We then performed scRNA-seq on the resulting embryos at 8hpf, when 1) a sizable population of transcriptionally distinct notochord and hatching gland cells are detectable, and 2) the endogenous UPR TFs and secretory pathway genes are not yet highly expressed (except for xbp1). Since xbp1 can be activated by UPR TFs including creb3l1, creb3l2, and atf6, and atf6 can be activated by xbp1, we mis-expressed the UPR TFs in xbp1 or xbp1+atf6 double crispants to reduce potential indirect effects. Single-cell transcriptomes were obtained for mCherry control, the activated forms of xbp1 (xbp1-s), creb3l1 (creb3l1-N), creb3l2 (creb3l2-N), and atf6 (atf6-N), as well as the non-activated, full-length creb3l2 (creb3l2-full). All scRNA-seq was performed using the 10x genomics platform (V3 and V3.1).