Volume 54, Issue 1, Pages 180-192 (April 2014) Hormone-Responsive Enhancer-Activity Maps Reveal Predictive Motifs, Indirect Repression, and Targeting of Closed Chromatin  Daria Shlyueva, Christoph Stelzer, Daniel Gerlach, J. Omar Yáñez-Cuna, Martina Rath, Łukasz M. Boryń, Cosmas D. Arnold, Alexander Stark  Molecular Cell  Volume 54, Issue 1, Pages 180-192 (April 2014) DOI: 10.1016/j.molcel.2014.02.026 Copyright © 2014 Elsevier Inc. Terms and Conditions

Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 1 Identification of Genome-wide Sets of Hormone-Responsive Enhancers (A) Schematic representation of the STARR-seq pipeline to screen for hormone-induced enhancers. (B) The three types of STARR-seq peaks considered for this study, the criteria for each group, and the number of peaks per group (“ratio” denotes the ratio of STARR-seq enrichments after versus before treatment). (C–E) UCSC genome browser (Kent et al., 2002) screenshots of STARR-seq and strand-specific RNA-seq tracks for the Eip75, Pvf2, and UGP gene loci (only representative isoforms are shown). Red shading highlights induced, blue repressed, and green constitutive peaks, respectively. Other known ecdysone target gene loci are shown in Figures S1C–S1J. Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 2 Validation of Hormone-Responsive Enhancers and Gene Induction via Multiple Enhancers (A) Validation of peaks from the induced, repressed, and constitutive classes by luciferase assays. Log2 fold induction of normalized luciferase signal is shown (see also Figure S1A). Wilcoxon rank-sum test: ∗p ≤ 0.0001, ∗∗p ≤ 10−13. (B) Venn diagram showing the number of genes near induced, repressed, and constitutive STARR-seq enhancers (±3 kb TSS). (C) The percent of assigned genes that are up- or downregulated at least 2-fold for each class of enhancers. See Figure S1B for a 4-fold cutoff and Figure S2 for a quantitative analysis. (D) Boxplot showing the gene expression changes (log2 scale) depending on the number of gained enhancers after treatment. Wilcoxon rank-sum test: p = 0.0015. (E and F) UCSC genome browser screenshots of STARR-seq and strand-specific RNA-seq tracks for E23 (E) and Alh (F), which are induced by multiple and a single enhancer, respectively. Red shading highlights induced peaks. See Figure S2D for an additional gene locus. Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 3 Differential Motif Content of Ecdysone-Induced Enhancers Allows the Identification of EcR Partner Motifs (A) Heatmap of predictive and differentially enriched motifs (p ≤ 0.01) when comparing induced enhancers versus negative control regions. (B) Logos for the EcR and srp motifs used in the analysis. (C) Percentage of correctly classified sites for each comparison when using an SVM based on motif counts and leave-one-out cross-validation. The black line shows the expected prediction accuracy for random classification (50%). Data are shown as mean ± SD of three replicates. (D) Heatmaps for predictive and differentially enriched motifs (p ≤ 0.01) when comparing induced enhancers versus negative controls selected to contain EcR motifs (note that the enrichment of the EcR motif above the genome merely reflects that the regions have been selected to contain such motifs). (E) Fold induction of normalized luciferase signal for induced enhancers near Eip75, sn, Dip-B, and Nhe-2, assessing the wild-type sequences and variants in which the EcR or srp or both motifs were mutated. Negative region: sequence from the exon of tj as in (Arnold et al., 2013). All DNA sequences are listed in Table S2. NS, not significant. Welch two-sample t test: ∗p ≤ 0.01; ∗∗p ≤ 10−5. Data are shown as mean ± SD of four experiments. See also Figure S3A. Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 4 Functionally Required EcR Partner Motifs Differ between Cell Types (A and B) UCSC genome browser screenshots of STARR-seq tracks in S2 cells and OSCs for the sky (A) and Hr46 (B) gene loci. Orange shading highlights OSC-specific induced peaks. (C) Percent of correctly classified sites when using an SVM based on motif counts and leave-one-out cross-validation (the black line shows the expected prediction accuracy for random classification [50%]). Data are shown as mean ± SD of three replicates. (D) Heatmap for predictive and differentially enriched motifs (p ≤ 0.01) when comparing OSC-induced enhancers versus negative controls (the srp motif is added to show that OSC-induced enhancers are depleted of it). (E) Fold induction of normalized luciferase signal for induced enhancers near sky and Eip75, assessing the wild-type sequences and variants in which the EcR, tj, or srp motifs are mutated. Negative region: sequence from the exon of tj as in Arnold et al. (2013). All DNA sequences are listed in Table S2. NS, not significant; NA, not applicable. Welch two-sample t test: ∗p ≤ 0.01; ∗∗p ≤ 10−4. Data are shown as mean ± SD of two experiments (two replicates each). See also Figure S3D. Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 5 Ecdysone-Mediated Repression Is Independent of EcR but Involves Eip74 Motifs (A) Percent of correctly classified sites when comparing induced and repressed enhancers using an SVM (50%, at random). Data are shown as mean ± SD of three replicates. (B) Heatmap for predictive and differentially enriched motifs (p ≤ 0.01) when comparing S2-induced versus repressed enhancers. (C) Fold repression of normalized luciferase signal for repressed intronic enhancer of mbc assessing the wild-type sequence and a variant in which the Eip74 motif was mutated. Negative region: sequence from the exon of tj as in Arnold et al. (2013). All DNA sequences are listed in Table S2. NS, not significant. Welch two-sample t test: ∗∗p ≤ 10−7. Data are shown as mean ± SD of four experiments. See also Figure S3E. (D–F) UCSC genome browser screenshots of STARR-seq and EcR ChIP-seq tracks for ecdysone-induced or repressed enhancers in the E23, btn, and mbc gene loci (only representative isoforms are shown). See main text for genome-wide numbers of EcR binding to induced and repressed enhancers. Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions

Figure 6 Ecdysone Signaling Can Target Inaccessible Chromatin (A–C) UCSC genome browser screenshots of STARR-seq, strand-specific RNA-seq, and DHS-seq tracks for the br, Sur-8, and Dip-B gene loci. Red and blue shading highlights STARR-seq enhancers that become open or closed upon treatment, respectively. (D) DHS-seq read density profiles ±5 kb around the summit of enhancers from three different classes before (black) and after (purple) treatment. (E) Boxplots showing the difference in chromatin accessibility upon ecdysone treatment at the summit positions (±5 bp) for the different enhancer classes (log2 scale). Wilcoxon rank-sum test: ∗p ≤ 0.001. Molecular Cell 2014 54, 180-192DOI: (10.1016/j.molcel.2014.02.026) Copyright © 2014 Elsevier Inc. Terms and Conditions