Enhancer-core-promoter specificity separates developmental and housekeeping gene regulation

Abstract

Gene transcription in animals involves the assembly of RNA polymerase II at core promoters and its cell-type-specific activation by enhancers that can be located more distally. However, how ubiquitous expression of housekeeping genes is achieved has been less clear. In particular, it is unknown whether ubiquitously active enhancers exist and how developmental and housekeeping gene regulation is separated. An attractive hypothesis is that different core promoters might exhibit an intrinsic specificity to certain enhancers. This is conceivable, as various core promoter sequence elements are differentially distributed between genes of different functions, including elements that are predominantly found at either developmentally regulated or at housekeeping genes. Here we show that thousands of enhancers in Drosophila melanogaster S2 and ovarian somatic cells (OSCs) exhibit a marked specificity to one of two core promoters--one derived from a ubiquitously expressed ribosomal protein gene and another from a developmentally regulated transcription factor--and confirm the existence of these two classes for five additional core promoters from genes with diverse functions. Housekeeping enhancers are active across the two cell types, while developmental enhancers exhibit strong cell-type specificity. Both enhancer classes differ in their genomic distribution, the functions of neighbouring genes, and the core promoter elements of these neighbouring genes. In addition, we identify two transcription factors--Dref and Trl--that bind and activate housekeeping versus developmental enhancers, respectively. Our results provide evidence for a sequence-encoded enhancer-core-promoter specificity that separates developmental and housekeeping gene regulatory programs for thousands of enhancers and their target genes across the entire genome.

Extended Data Figure 1. Setup of STARR-seq with different core promoters.

a, STARR-seq detects enhancers but no promoters (after ref. 12). (Left) STARR-seq couples the enhancer activities of candidate fragments to the candidates’ sequences in cis by placing the candidates to a position within the reporter transcript. Enhancer activities can therefore be assessed by the presence of candidates among cellular mRNAs, which allows the parallel assessment of millions of candidates, enabling genome-wide screens. Sequences that activate transcription from the intended core promoter of the STARR-seq vector lead to a full-length reporter transcript and can be detected by STARR-seq. Shown are the RT and nested PCR steps of the STARR-seq reporter RNA processing protocol that ensure this. (Right) In contrast, STARR-seq does not detect truncated transcripts that result if a candidate fragment functions as a promoter to initiate transcription. Thus, core promoter-containing (i.e. TSS-overlapping) sequences that are detected by STARR-seq exhibit enhancer activity as they can activate transcription from a remote position, in addition to their ability to serve as core promoter endogenously. b, Luciferase signals (Firefly/Renilla) assessing the intrinsic (or basal) activity of the core promoters used in this study in luciferase reporter setups that differ only in the respective core promoter sequences and do not contain any enhancer. The basal activities differ as expected, but do not differ consistently between housekeeping (RpS12, eEF1delta, NipB, x16) and developmental (DSCP, eve[long], eve and pnr) core promoters, nor between core promoters for which the STARR-seq screens appear most similar (e.g. RpS12 and eEF1delta; see Figure 3). Note that all luciferase assays and STARR-seq screens are corrected for differences in intrinsic activity. c, Reproducibility of hkCP and dCP STARR-seq in D. melanogaster S2 cells. The reproducibility of hkCP and dCP STARR-seq as assessed by the STARR-seq enrichments (replicate 1 vs. 2) at the summits of enhancer peaks called in the merged experiments (hkCP: 5,956; dCP: 5,408; scatter plots are enlarged versions of the insets in Figure 1d; PCC: Pearson correlation coefficient; enr. rep X: STARR-seq enrichment in replicate X). Note that the data for dCP are from ref. , but have been re-analyzed.

Extended Data Figure 2. Transcription initiates within the core promoter of the STARR-seq construct.

5’ Rapid Amplification of cDNA Ends (5’ RACE) demonstrates that transcription initiates at the TCT and Initiator motifs within the hkCP and dCP, respectively. a, Setup of the 5’RACE experiment including the STARR-seq plasmid, used here with two defined enhancers, the STARR-seq transcript and location of all primers used to specifically amplify 5’-capped STARR-seq transcripts. b, 5’ RACE nested PCR products separated on a 1% agarose gel. c, Screenshot of Sanger sequencing results (chromatogram and called bases) compared with the template sequence. Annotations are shown in green, in the following order: 5’ RACE adapter, hkCP with TCT motif (only the part downstream of the transcription start is annotated, as the 5’ part is not present in the sequenced cDNA), spliced intron, GFP; the sequencing primer is shown in red (top). Also shown is a version that displays the template and Sanger sequencing results for the core promoter region only (zoom in). d, same as in (c) but for the dCP for which transcription initiates within the Initiator (Inr) motif.

Extended Data Figure 3. Specificity of hkCP and dCP enhancers towards the hkCP and dCP assessed by luciferase assays.

a, Luciferase reporter setup with the hkCP or dCP (see also Figure 1e). b, Luciferase signals of 24 hkCP-specific enhancers tested in a hkCP- (purple bars) as well as in a dCP-containing (brown bars) luciferase reporter. 21 out of 24 hkCP enhancers showed luciferase activity (>1.5 fold over negative, P<0.05 via one-sided unpaired Student’s t-test, n=3) with the hkCP, while only 1 out of 24 showed activity with the dCP (error bars are s.d. of three biological replicates, ‘x’ flags candidates that are not active with the correct core promoter, and ‘+’ flags candidates for which the activity with the wrong core promoter is above the threshold (note that the activity with the correct core promoter is still higher in all 3 cases). c, as in (b) but testing dCP-specific enhancers. 10 out of 12 are positive with the dCP while only 2 out of 12 are positive with the hkCP. d, as in (b and c) but testing shared enhancers that were found by STARR-seq with hkCP and dCP; 6 out of 7 are active with both core promoters. See Supplementary Table 17 for the genomic coordinates of the enhancers and the primers used to amplify them.

Extended Data Figure 4. hkCP and dCP STARR-seq signal in S2 cells around different core promoter types.

Average hkCP (top) and dCP (bottom) S2 STARR-seq enrichment in 40kb intervals around transcription starting sites (TSS) that contain different combinations of known core promoter motifs. Shown are (left to right) TATA-Initiator (179 TSSs), Initiator (that do not contain either TATA or DPE; 1901), Initiator-DPE (100), TCT (303) and Motif1-Motif6 (266). According to their motif contents, the first 3 are developmental-type core promoters and the last 2 are housekeeping-type core promoters. Indeed, only the housekeeping-type core promoters show a strong enrichment of hkCP S2 STARR-seq signals at the TSS, which is not seen for the dCP STARR-seq signal (due to enhancer–core promoter specificity) nor for the developmental-type core promoters (due to the dCP enhancers location at more distal sites).

Extended Data Figure 5. TSS-overlapping hkCP enhancers function independent of their orientation.

Luciferase signals for all 17 TSS-overlapping hkCP enhancers (i.e. containing 1 TSS or 2 divergent TSSs; see Supplementary Table 17) from Extended Data Figure 3 cloned in the second orientation with respect to the TSS of the luciferase gene (lower bar plot; the upper bar plot corresponds to the initial orientation as in Extended Data Figure 3 and is shown for comparison). In both orientations, 15 out of 17 enhancers showed activity towards the hkCP (details as in Extended Data Figure 3). These results together with the findings in Extended Data Figure 3 challenge the widespread notion that TSS-proximal sequences are promoters and even the concept of promoters more generally: sequences that autonomously activate gene expression – and are therefore often termed promoters – might in fact be the combination of a core promoter and a proximal enhancer. The TSS-proximal location of many housekeeping enhancers might be evolutionarily more ancient, consistent with regulatory mechanisms in simple eukaryotes such as yeast. In contrast, enhancers of genes with more complex regulation are typically located more distally, potentially simply because the several different cell type-specific enhancers of these genes would not all fit to positions near TSSs. Consistently, such genes frequently have larger intergenic and intragenic regions known to accommodate enhancers with diverse activity patterns.

Extended Data Figure 6. hkCP and dCP enhancers in S2 cells are associated with genes of different functions and core promoter elements.

a, Gene ontology (GO) analysis of genes next to hkCP and dCP-specific enhancers in S2 cells using different enhancer-to-gene assignment strategies (top-left: ‘closest TSS’ as in Figure 2, top-right: ‘1kb TSS’, bottom-left: ‘gene loci’; see Methods for details). Shown are 20 non-redundant GO categories selected from the 100 most significantly enriched categories associated with each enhancer class (see Supplementary Tables 11-13 for all categories). b, Enrichment of core promoter elements at genes next to hkCP and dCP-specific enhancers in S2 cells. Similar analysis as in Figure 2e, however using different enhancer-to-gene assignment strategies (see Methods for details). Consistent with Figure 2e, core promoters of genes assigned to hkCP-specific enhancers are enriched in Motifs 1, 5, 6, 7 and DRE, while core promoters of genes assigned to dCP-specific enhancers are enriched for TATA box, Initiator, MTE and DPE motifs, irrespective of the assignment strategy.

Extended Data Figure 7. Housekeeping and developmental core promoters differ characteristically in their global enhancer preferences.

As Figure 3b yet including biological replicates with an independently cloned BAC library covering around 5MB of genomic sequence (BAC) and assessing the Pearson correlation coefficient at each position along these regions (GW: genome-wide screens as in Figure 3b). The similarity observed for the TATA box and DPE containing core promoters (hsp70, pnr, and DSCP [dCP]) suggest that differences related to these core promoter elements might be more subtle or related to alternative mechanisms, including the potential preferences of more proximal or distal enhancers or Polymerase RNA polymerase II pausing and the dynamics versus stochasticity of initiation and elongation^,,.

Extended Data Figure 8. Difference between hkCP and dCP enhancers in OSCs (I).

a-b, Different enhancers activate transcription from hkCP and dCP in OSCs. As Figures 1c and d but for OSCs rather than S2 cells. c, Genomic distribution of hkCP and dCP enhancers in OSCs. As Figure 2a but for OSCs rather than S2 cells. d, hkCP and dCP STARR-seq signal in OSCs around different core promoter types. As Extendend Data Figure 4 but for OSCs rather than S2 cells.

Extended Data Figure 9. Difference between hkCP and dCP enhancers in OSCs (II).

a, Gene ontology (GO) analysis of genes next to hkCP and dCP-specific enhancers in OSCs. As Extended Data Figure 6a but for OSCs rather than S2 cells (see Supplementary Tables 15-17 for all categories). b, Enrichment of core promoter elements at genes next to hkCP and dCP-specific enhancers in OSCs. As Figure 2e and Extended Data Figure 6b but for OSCs rather than S2 cells (NS: non-significant [hypergeometric P>0.05]). c, Heat maps of hkCP (top) and dCP (bottom) STARR-seq enrichments in S2 cells and OSCs. Heat maps on the left and right are centered on the summits of core-promoter type-specific enhancers in S2 and OSCs, respectively.

Extended Data Figure 10. The activity of hkCP and dCP enhancers are dependent on DRE and Trl/GAGA motifs, respectively.

a, Differential motif enrichment in distally located hkCP and dCP-specific enhancers (as Fig. 5a but assessing enrichments of the same motif PWMs exclusively at distal enhancers >500bp away from the closest TSSs). Key motifs including DRE and Trl/GAGA are also differentially enriched in distal hkCP and dCP-specific enhancers (NS: non-significant [FDR-corrected hypergeometric P>0.01]; S2 cells: hkCP n=790, dCP n=3013; OSCs: hkCP n=556, dCP n=2555). b, Distal hkCP and dCP-specific enhancers are differentially bound by DREF and Trl/GAF, respectively. ChIP enrichments of DREF (left) and Trl/GAF (right) at S2 hkCP and dCP-specific enhancers that are distal (>500bp) from the closest TSSs. Equivalent to Figure 5b, but considering exclusively TSS-distal enhancers to exclude potentially confounding effects for TSS-proximal enhancers for which it is not possible to discern whether binding occurs due to the enhancer sequence or core promoter function. The differential binding between DREF and Trl/GAF to hkCP and dCP-specific enhancers respectively is also found in Kc cells, in which the DREF ChIP-seq experiment had been performed (data not shown). c, Addition of DRE motifs to dCP enhancers increases their activity towards hkCP. Relative luciferase activity values (Firefly/Renilla [FF/RL]) for 11 dCP enhancers without DRE motifs (WT, light purple) and with 3 DRE motifs flanking the enhancers on each side (+DRE, dark purple). Asterisks (*) indicate statistical significance (P<0.05 via one-sided unpaired Student’s t-test); error bars denote the s.d. of three biological replicates.

Figure 1. Distinct sets of enhancers activate transcription from the hkCP and dCP in S2 cells.

a, STARR-seq setup using the hkCP housekeeping (RpS12; purple) and dCP developmental core promoter (DSCP11; brown) b, Genome browser screenshot depicting STARR-seq tracks for both core promoters. c, Overlap of hkCP and dCP enhancers. d, hkCP versus dCP STARR-seq enrichments at enhancers (insets show replicates for hkCP and dCP; dCP data from ref. 12). e, hkCP, dCP, or shared enhancers that activate luciferase (>1.5-fold & P<0.05 [one-sided t-test]; n=3; Extended Data Figs 3 and 5) from hkCP (purple) or dCP (brown; numbers: positive/tested).

Figure 2. hkCP and dCP enhancers differ in genomic distribution and flanking genes.

a, Genomic distribution of hkCP and dCP enhancers (CDS: coding sequence; UTR: untranslated region). b-c, hkCP enhancers function distally in luciferase assays independent of their genomic positions (b) and orientation towards the luciferase TSS (c; orientation 1 from (b); Extended Data Figs 3 and 5). d-e, GO (5 of the top 100 terms shown per column; Supplementary Table 11) and gene expression (terms curated from BDGP and FlyAtlas) analyses (d) and enrichment of core-promoter elements at TSSs (e) for genes next to hkCP and dCP enhancers.

Figure 3. Housekeeping and developmental core promoters differ characteristically in their enhancer preferences.

a, Different housekeeping (top 4) and developmental-like (bottom 3) core promoters and their motif content (schematic). b, Bi-clustered heatmap depicting pairwise similarities of STARR-seq signals ([PCCs] at peak summits). PCCs and dendrogram (top) show the separation between housekeeping and regulated core promoters. c, Genome browser screenshot depicting STARR-seq tracks for all 7 core promoters.

Figure 4. hkCP enhancers are shared across cell types.

a, Genome browser screenshot showing tracks for hkCP (top) and dCP STARR-seq (bottom) in S2 cells and OSCs. b, Overlap of hkCP (top) and dCP (bottom) enhancers between S2 cells and OSCs. c-d, hkCP (c) and dCP (d) STARR-seq enrichments in S2 cells versus OSCs at hkCP- or dCP-specific enhancers (insets show replicates; dCP data from ref. ).

Figure 5. hkCP and dCP enhancers depend on DREF and Trl/GAF, respectively.

a-b, Motif enrichment (a) and ChIP-signals for DREF and Trl/GAF (b) in hkCP and dCP enhancers (NS: not significant [FDR-corr. hypergeometric P>0.01]; boxes: median and interquartile range; whiskers: 5^th and 95^th percentiles; two-sided Wilcoxon-rank-sum P-values). c, Luciferase assays (LAs) for 4 wildtype and DRE-motif-mutant hkCP enhancers (numbers: mutated motifs; error-bars: s.d. [n=3]; * P<0.005 [one-sided t-test]). d, LAs for 2 dCP enhancers (-) and Trl/GAGA→DRE-mutant variants (+) with hkCP (top) and dCP (bottom; details as in c). e, LAs for 6 DRE motifs with hkCP and dCP (details as in c). f, Model: housekeeping genes contain Motifs 1,5,6,7 and/or TCT and are activated by TSS-proximal hkCP enhancers via DREF. Regulated genes contain TATA-box, Initiator, MTE and/or DPE and are activated by distal dCP enhancers via Trl/GAF.