Robust target gene discovery through transcriptome perturbations and genome-wide enhancer predictions in Drosophila uncovers a regulatory basis for sensory specification

Abstract

A comprehensive systems-level understanding of developmental programs requires the mapping of the underlying gene regulatory networks. While significant progress has been made in mapping a few such networks, almost all gene regulatory networks underlying cell-fate specification remain unknown and their discovery is significantly hampered by the paucity of generalized, in vivo validated tools of target gene and functional enhancer discovery. We combined genetic transcriptome perturbations and comprehensive computational analyses to identify a large cohort of target genes of the proneural and tumor suppressor factor Atonal, which specifies the switch from undifferentiated pluripotent cells to R8 photoreceptor neurons during larval development. Extensive in vivo validations of the predicted targets for the proneural factor Atonal demonstrate a 50% success rate of bona fide targets. Furthermore we show that these enhancers are functionally conserved by cloning orthologous enhancers from Drosophila ananassae and D. virilis in D. melanogaster. Finally, to investigate cis-regulatory cross-talk between Ato and other retinal differentiation transcription factors (TFs), we performed motif analyses and independent target predictions for Eyeless, Senseless, Suppressor of Hairless, Rough, and Glass. Our analyses show that cisTargetX identifies the correct motif from a set of coexpressed genes and accurately predicts target genes of individual TFs. The validated set of novel Ato targets exhibit functional enrichment of signaling molecules and a subset is predicted to be coregulated by other TFs within the retinal gene regulatory network.

Figure 1. cisTargetX predictions of Dorsal target genes.

(A) Example of scoring for homotypic clusters of binding sites with Cluster-Buster, using the Dorsal PWM and 1,980 other PWMs. (B) The scoring is applied to all 5-kb upstream sequences and introns (k = 1 to k = 93,330) of 13,667 genes across 12 Drosophila species (i = 1 to i = 12). (C) For each region k of species i, the highest score is retained and used to rank all regions in each species independently. The ranks R _k,i are integrated across the species into one ranking R _k. The highest ranking region for each gene is retained to yield a final ranking of all D. melanogaster genes. (D) Using a set of candidate coexpressed genes, a receiver operating characteristic (ROC) curve is drawn using the dl-PWM based ranking on the x-axis and the recovery of the candidate genes in the y-axis. In this example, a set of 80 genes expressed downstream of dl is used . The blue curve (using 12 species) shows significant enrichment of direct Dl targets within the set of 80 candidates. The optimal cut-off at position 220 yields a subset of 13 direct target predictions. (E) Histogram of AUC for all 1,981 PWMs tested, with the best performing PWMs being Dl PWMs, illustrating the use of motif discovery using ROC curves. (F) Predicted target regions are cloned in an enhancer reporter vector.

Figure 2. cisTargetX predictions of Ato target genes.

(A) Histogram of AUC values for all ROC curves generated from the rankings using 1,981 PWMs and 204 upregulated genes under Ato GOF conditions. The six best PWMs are all E-box motifs. (B) ROC curve for the best performing motif, RACASCTGY (blue) and the seventh best motif (M00234 for Su(H)). The arrow indicates the optimal cut-off for RACASCTGat position 674, yielding 36 direct Ato target predictions.

Figure 3. In vivo GFP reporter activities of predicted Ato target enhancers.

Enhancer GFP-reporter assays for six positive enhancers. (A,B) Wild-type enhancer activity in wild-type eye-antennal discs showing GFP (A), and GFP plus Ato and Sens protein (B). The arrow and line indicate the initiation of GFP expression. GFP maturation causes a slight delay in GFP appearance posterior to Ato, as observed for positive controls (Figure S4). (C) Activity of the same enhancers with mutated Ato binding sites. (D) Response of the same wild-type enhancers to ectopic expression of Ato along the anterior-posterior boundary in wing imaginal discs of dppGAL4-UASAto animals. (E) Schematic of the enhancer sequences indicating the predicted Ato E-boxes in blue (see Methods). These E-boxes were mutated from CANNTG to CGNNCG.

Figure 4. Validation of orthologous Ato target enhancers.

(A) GFP reporter activity in the eye-antennal disc produced by D. yakuba and D. virilis orthologous sequences of the nmo and Dscam Ato target enhancers. Orthologous sequences were selected from the UCSC Genome Browser Multiz multiple alignments across 12 Drosophila genomes. (B) Screenshots from the UCSC Genome Browser showing the nmo enhancer with predicted E-boxes (blue track) and sequence constraint across the 12 Drosophila genomes as PhastCons scores (black track) and as aligned Nets, where the majority of the D. melanogaster sequence is conserved with D. ananassae and D. virilis, including the predicted E-box cluster. (C) Similar screenshot for the Dscam enhancer, showing a fragment of the D. melanogaster that is absent in the D. virilis orthologous sequence (red box), which contains the predicted E-box cluster.

Figure 5. Ato target enhancer activity in other SOPs.

(A) Reporter GFP activity for two examples of Ato target enhancers across different imaginal discs. Ato-overlapping activity is found in the photoreceptors, for nmo and Spn, in the antennal SOPs (Spn), in the leg chordotonal SOPs (nmo and Spn), in the wing chordotonal SOPs (nmo and Spn), and in other SOPs specified by another proneural factor Scute (Spn). Ato- and Scute-dependent activity is shown by ectopic expression of Atonal and Scute (nmo and Spn). (B) Unique combinations of signaling molecules are activated by Atonal in each sensory organ. The binary active/inactive summaries shown as green and red boxes are derived from GFP-reporter assays for all 20 Ato target enhancers (Figures S8 and S9). Atonal target genes analyzed are signaling molecules or TFs.

Figure 6. Motif analysis across Ato target enhancers.

Motif over-representation analysis using the Clover algorithm finds E-boxes (red) and Su(H) motifs (blue) as highly over-represented (p<0.001) across the Ato target enhancers. The ey motif (green) is not statistically over-represented but was found by independent ey target discovery with cisTargetX (the Eyeless site predictions are generated by Cluster-Buster).

Figure 7. Target gene predictions for Atonal and associated retinal TFs.

Predicted target genes for Ey and Su(H) and validated target genes for Ato, showing two coregulated targets of Ato and Ey (Fas2 and CG30492) and extensive coregulation between Ato and Su(H). Full arrows represent validated target genes, dashed arrows represent predicted target genes by cisTargetX. Genes in bold face are previously known target genes. ⁰, previously known target genes that are not detected in this study; ^*, Su(H)-predicted target genes by cisTargetX that are also Ato targets yet without predicted binding sites in the Ato target enhancer; ^$, genes not predicted as Su(H) by cisTargetX yet the Ato target enhancers contain predicted Su(H) binding sites. The full GRN can be found in Figure S11 and Table S10.