Entangled and non-modular enhancer sequences producing independent spatial activities

Abstract

The modularity of transcriptional enhancers is central to our understanding of morphological evolution, allowing specific changes to a gene expression pattern component, without affecting others. Enhancer modularity refers to physically separated stretches of regulatory sequence producing discrete spatiotemporal transcriptional activity. This concept stems from assays that test the sufficiency of a DNA segment to drive spatial reporter expression resembling that of the corresponding gene. Focusing on spatial patterns, it overlooks quantitative aspects of gene expression, underestimating the regulatory sequence actually required to reach full endogenous expression levels. Here, we show that five regulatory activities of the gene yellow in Drosophila, classically described as modular, result from extensively overlapping sequences, with broadly distributed regulatory information. Nevertheless, the independent regulatory activities of these entangled enhancers appear to be nucleated by specific segments that we called enhancer cores. Our work calls for a reappraisal of enhancer definition and properties, as well as of the consequences on regulatory evolution.

Regulatory mapping at a developmental gene reveals an unexpected architecture of entangled, rather than modular, enhancers.

INTRODUCTION

The embryonic blueprint of gene expression prefiguring morphology results from the activity of separate cis-regulatory elements (1–5). It was hypothesized (6, 7) and then demonstrated (8–11) [for review, see (12)] that mutations in enhancers are a primary driver for the evolution of discrete morphological traits, such as gains and losses of limbs in vertebrates or decorative elements in insects. Owing to their reduced pleiotropy, mutations affecting single enhancers facilitate changes in specific aspects of morphology with no deleterious effect on other traits (12, 13). Hence, a modular representation of enhancers bears strongly on our understanding of regulatory evolution.

The original notion of enhancer modularity (14, 15) was quickly generalized in transgenic animals and plants with reporter assays testing the capacity of arbitrarily chosen DNA segments to recapitulate elements of the gene’s spatial expression (16). This notion is based on the sufficiency of DNA segments to drive specific expression patterns, leaving out completely how much transcript these sequences yield compared to the endogenous levels of transcript. Yet, the amount of gene transcript is key to normal development (17, 18). Moreover, a minimal enhancer producing a correct spatiotemporal pattern may be insufficient for a robust expression under unfavorable environmental or genetic conditions (19–21). It has been repeatedly shown that mutations occurring outside minimal enhancers can contribute to phenotypic evolution (21–23). Last, several papers have recently challenged the lack of pleiotropy of enhancers (24–26).

The modular definition of enhancers devolved from functional assays has been reinforced by genomic approaches. Discrete regions of open chromatin or peaks of specific epigenetic marks for active enhancers have been used to map enhancers genome-wide (27–32). Yet, the extent of sequence that these peaks span is influenced by the choice of cutoff values or other methodological biases. For instance, the processing of datasets assessing genome-wide chromatin accessibility starts with the binning of raw data into putative regulatory elements (33), leading to a circular argument. As a result, of these different approaches, transgenic assays, or genomic methods, enhancers are deemed to span 100 to 1000 base pairs, although the sequence necessary for a full regulatory activity, pattern and levels, may be larger (26).

Additional considerations undermine the notion of enhancer modularity or their compactness. While short regulatory segments usually drive spatially restricted expression, they are often found to have broad activity, targeting multiple tissues (2, 24, 25, 34–36). By contrast, single enhancers are sometimes insufficient to produce robust gene expression in a given tissue: The additive activity of redundant or shadow enhancers was shown to enable robust gene expression (19, 20, 37). Other sequences, referred to as facilitator elements, do not drive activity on their own but enhance the function of neighboring enhancers (38). They are part of so-called super-enhancers, which are themselves described as large clusters of enhancers sufficient to assign cell identity, a concept that was criticized for its lack of functional support (39). These different concepts assume some level of sequence modularity, but a systematic analysis of intervening regions between these regulatory elements has been lacking.

Does the picture of discrete and modular enhancers hold when one considers the quantitative dimension of transcription? Using a quantitative reporter assay, we addressed this question by precisely mapping regulatory information at the yellow (y) locus in Drosophila, a typical locus with several independent enhancers. Their activities, classically mapped with reporter assays, prefigure a gray background or a black spot on the wing, stripes, a longitudinal band, a sexually dimorphic block of pigmentation on the abdomen, or black sensory bristles over the body, respectively (9, 40–42). A recent systematic dissection of y regulatory regions in Drosophila melanogaster (34), however, suggested more distributed activities than the textbook picture of separate and independent enhancers would predict. Along the same lines, focusing on the wing-spotted species Drosophila biarmipes, we showed that the wing blade and spot enhancers, respectively prefiguring the background gray pigmentation of the wing and a dark distal wing spot, were actually broader, extensively overlapping, and shared regulatory information (43). The sequence overlap of these two enhancers may, however, simply result from a recent cooption process, and the regulatory activities may resolve with enough evolutionary time.

To investigate the relationship between distinct regulatory activities, we compared the wing enhancers to an ancient regulatory sequence, the body enhancer. The body enhancer, found in various Drosophila species (42, 44), is active in the head, thoracic, and abdominal epidermis during metamorphosis (45, 46), defining a complex spatial pattern. We mapped regulatory information of the body enhancer and compared it to our previous map of the wing enhancers (43). The body enhancer spans the entire sequence of the two wing enhancers, refuting their modularity. The sequences of these activities are entangled with, however, different distributions of regulatory information.

RESULTS

The y body enhancer encompasses the entire y 5′ region

y is expressed in body pupal epidermis (42, 47) under the control of regulatory sequences mapping 5′ of y transcription start site (TSS) (41, 45, 46). The dissection of these sequences from D. melanogaster with transgenic reporter assays identified a 1-kb segment, the body enhancer, sufficient to recapitulate y spatial expression (42). Whether this segment is sufficient to drive endogenous transcription levels is, however, unknown. We therefore first mapped the entire regulatory sequence necessary and sufficient to drive y transcription in the fly body, relying on a quantitative reporter assay. For the sake of comparison with our previous mapping, we used D. biarmipes sequences, and, for simplicity, we focused on abdominal expression. Starting from a 5.4-kb segment upstream of y TSS, we tested the regulatory activity of segments with deletions from the 5′ end (D series), as well as progressive sequence randomization from the 3′ end in male pupae [E series; Fig. 1, A and C, and (43)]. To map the boundaries of regulatory activities in the abdomen, we calculated how much reporter expression was lost or gained for each construct compared to the activity of the largest construct of the corresponding series (Fig. 1B; see Materials and Methods). For instance, removing a distal segment (line D1), about 4 kb upstream of the original body enhancer (42), already had a significant effect on the abdominal expression (Fig. 1B). Generally, we found that all constructs differed in activity from the largest construct of each series (table S1). We concluded that the regulatory information determining abdominal expression spans up to 5.4 kb and overlaps largely with the previously mapped regulatory activities of the wing (43), challenging the notion of short and modular enhancers. Average phenotypes (Fig. 1C) revealed changes not only in the levels of regulatory activity among constructs but also in spatial distribution. We therefore sought to understand how the distributed regulatory information upstream of y TSS was organized to control different spatial pattern elements.

Fig. 1. Mapping regulatory activities upstream of the yellow gene.

(A) Reporter construct series used to evaluate the regulatory content of regions located 5′ of y. The first series, D (purple), is progressively trimmed from the 5′ end, while the proximal end of constructs of the second series, E (ocher), is increasingly randomized, without changing the distance to the TSS. Note that construct D2 appears in the E series, too, as the reference construct from which all E constructs were derived. (B) Relative regulatory information (fluorescence levels) for each reporter line compared to the reference line of the series. bp, base pairs. (C) Average reporter expression (average phenotype) in the abdominal epidermis for each line. ø denotes a line with an empty reporter vector. In all figures, only male abdomens were examined.

PCA highlights the composite nature of body activities

To this end, we relied on a principal components analysis (PCA) of phenotypic variation across all constructs (Fig. 2, A and B, and fig. S1), as it gave us access to the fine relationship between regulatory content and enhancer activity. The first component [principal component 1 (PC1), 72% of the total variation; Fig. 2, A and B], mostly captured broad uniform expression changes across the abdomen. The second component (PC2, 18% of the total variation) captured changes in two distinct spatial activities simultaneously, in the upper four segments and the last two segments, respectively corresponding to broad uniform expression (34) and the sexually dimorphic male expression (Fig. 2, A and B). The third component showed an identified artifact caused by gaps between the last two abdominal segments, due to ventral bending (fig. S1). We did not consider PC3 for further analysis. The fourth component (PC4, 1% of the total variation) shows changes in the basal part of each segment, identifying the banded pattern of the fly abdomen (Fig. 2A). In summary, PC1, PC2, and PC4 appear to jointly capture variation in the three main pattern components of the global body activity, although these PCs are not necessarily independent.

Fig. 2. The body enhancer pattern results from three distinct activities.

(A) Phenotypic directions captured by three principal components (PCs) in the space of variation for all reporter lines shown in Fig. 1. These directions correspond to broad uniform expression (PC1), a combination of broad uniform expression and sexually dimorphic male expression (PC2), and stripes (PC4). PC3 identifies an artifact caused by stretching between consecutive segments and is shown in fig. S1. (B) Principal components analysis (PCA) of phenotypic space showing the first two PCs. Each dot represents a single abdomen. All samples from the same construct are connected by lines to their average. (C) Subtraction of average phenotypes of selected lines identifies specific pattern elements: D3-D4, stripes; E1-E2, broad uniform expression; and D2-E0, sexually dimorphic element. (D) Constructs with randomized sequence segments according to (C). (E) Resulting average phenotypes of lines with randomized segments with an annotation of pattern elements.

stripes, broad, and dimorphic, three independent enhancers defining yellow abdominal expression

To test the possible regulatory independence of these activities, we sought to identify sequence segments that affect single spatial pattern components. To this end, we first calculated differences between average phenotypes of consecutive lines (fig. S2). We observed that the average phenotype differences between certain lines corresponded precisely to pure individual activities (Fig. 2C): D3-D4 reflects the stripes component, E1-E2 reflects the broad uniform component, and D2-E0 reflects the dimorphic component, suggesting that the corresponding sequence segments may affect the respective activities independently. To directly test these independent regulatory contributions, we examined reporter constructs with randomized sequence in place of these segments (Fig. 2D): Δ stripes (segment D3-D4 randomized in the context of D0), Δ broad (segment E1-E2 randomized in the context of D0), and Δ dimorphic (segment D2-E0 randomized in the context of D2 = construct E0 of Fig. 1). Compared to the activity of the largest construct D0, which contains all three pattern components—stripes, broad expression, and dimorphic expression, we observed that only a single component was missing in each randomized line (Fig. 2E): Δ stripes had no visible stripes, Δ broad missed broad expression in the upper segments, and Δ dimorphic was devoid of strong expression in the posterior segments. These results showed that the three body activities are, to some extent, functionally independent of each other. Further, they suggested that each of these segments could be an enhancer core, meaning a stretch of sequence necessary to seed the spatial regulatory activity but insufficient to account for the endogenous levels of expression (43). To evaluate the extent of independence of the three activities—broad, dimorphic, and stripes—and to lastly examine the distribution of regulatory information upstream of the y TSS, we went on to decompose the expression patterns as a sum of these three activities in the PC space. The decomposition is achieved through a simple operation, a change of basis, whereby the PC space is re-projected in a new coordinate system made of three vectors corresponding to the pure activities. The stripes activity is defined by the vector (D0 to Δ stripes) in the original PC space, the broad activity by the vector (D0 to Δ broad), and the dimorphic activity by the vector (D2 to Δ dimorphic). The resulting mixed components of the re-projected space show changes only in stripes, broad, or dimorphic pattern elements, respectively (Fig. 3).

Fig. 3. Regulatory architecture upstream of yellow.

The abdomen and wing images with a colormap correspond to the mixed components, representing the variation of different regulatory activities in the PC space, after a change of basis. While PCs of the original PCA captured multiple activities, the mixed components resulting from a change of basis capture variation in single activities (i.e., phenotypic directions). The contribution of each sequence segment along y 5′ region to each mixed PC is a map of the regulatory information for each enhancer. While all activities span most of the 5.4-kb region, we note that enhancer cores (pink shading) identified in Fig. 2 and (43) are not overlapping.

Entangled wing and body enhancers upstream of yellow

In this re-projected space, we could then measure the independent contribution of each fragment in the dissection series to each pattern component and compare these distributions of regulatory information to those of the wing enhancers that we calculated in (43) (Fig. 3). The results show that, in spite of their extensive overlap, the distribution of regulatory information along the sequence is very different among activities. Together, our results outline an alternative architecture of y regulatory regions, whereby enhancer cores necessary to initiate a regulatory activity are surrounded by regulatory sequences contributing to the full endogenous levels of transcription. It is at odds with the current thread-like image of modular regulatory elements at y or, generally, at developmental loci. This could be a peculiarity of the y locus. Yet, the methodological biases of ascertainment caused by classical reporter assays and genomic data processing (33) speak for an entangled architecture of regulatory regions in general. This is only comforted by circumstantial evidence that regulatory regions have been underestimated (21–23, 26). This emerging possibility raises a fundamental question.

If enhancers are not modular stretches of sequence, then how do they independently produce regulatory activities in distinct subsets of cells? At the y locus, each enhancer responds to a subset of transcription factors (TFs) governing its activity, but TFs present in the wing may also be present in the abdomen and could activate wing enhancers in abdominal cells. We compared the TFs expressed in pupal wings (48, 49) to those expressed in abdominal epidermis at the same stage (using a previously unpublished microarray dataset, see Materials and Methods). Of 302 wing TFs and 240 abdominal TFs, we found that 78 are expressed in both tissues during pupal life (table S2), potentially exposing their respective enhancers to bleed-through activities.

DISCUSSION

Our reappraisal of yellow regulatory architecture with a quantitative method challenges the textbook picture of enhancers as small and discrete boxes surrounding the coding sequence of a gene (50). The difference arises, we reckon, from taking expression levels into consideration. We would therefore not be surprised if other loci with complex expression also had large and overlapping enhancers. The notion of distributed regulatory information has previously been proposed for the y locus (34) and may generally relate to the concepts of redundant enhancers (51) as well as super-enhancers (38, 39). It is conceivable that intervening regions between shadow enhancers contribute to increase regulatory activity but do not have the ability to drive expression alone, reminiscent of facilitator elements (38).

We mapped regulatory information at the y locus in a transgenic and heterologous context. This could, in principle, bias our assessment. We have shown, however, that the same spot enhancer reporter construct drove similar expression in D. melanogaster and in D. biarmipes transgenic flies (48). The only significant difference was the shape of the spot pattern, extending proximally toward the wing hinge in D. melanogaster, but confined distally in D. biarmipes. This difference, we found, was due to a change in the spatial distribution of the activator Distal-less (48). Apart from this, we have also established that this region and other Drosophila enhancers contain the information determining their own accessibility (43, 52). Hence, studying enhancers in a different genomic context than their endogenous locus should not bias how their regulatory content is read by TFs.

Entangled regulatory information at the y locus resolves into functionally independent enhancers. We submit that the regulatory independence may reside in the function of enhancer cores and hypothesize that each core subordinates all flanking regulatory information through an unknown mechanism to selectively produce the full activity of a given enhancer. This could involve selective accessibility or selective use of TF binding sites, or both, as we have shown in the case of the spot enhancer (43). With the advent of methodologies describing the dynamics of the three-dimensional (3D) genome, the representation of gene regulation is changing from selective and transient looping of a remote enhancer onto the core promoter region to more stable multi-enhancer hubs gathered around the TSS of a gene (53, 54). It is conceivable that a complex architecture of entangled enhancers forms a relatively stable hub around a core promoter, where core enhancer segments control the selective involvement of each activity in the respective cell types.

Last, the notion of enhancer modularity is the cornerstone of Evolutionary Developmental Biology, predicting that regulatory changes in developmental genes drive morphological diversification because these changes circumvent pleiotropy and foster evolvability (12). If enhancers share extensive portions of sequence, then how are their independent changes possible? One possibility is that binding sites for TFs involved in distinct activities are intermingled rather than shared. This would still afford changes in selected activities with minimal effects on other enhancers. On the other hand, if the 3D folding of a regulatory region with entangled enhancers is relevant for its function, then the evolvability of regulatory sequences may be more limited than previously assumed.

MATERIALS AND METHODS

Fly husbandry

Our D. melanogaster stocks were maintained on standard cornmeal medium at 25°C with a 12-hour:12-hour day:night light cycle.

Transgenesis

All reporter constructs were injected as in (48). We used ɸC31-mediated transgenesis (55) and integrated all constructs at the genomic attP site VK00016 on chromosome 2 (56). The enhancer sequence of all transgenic stocks was genotyped before imaging. Of note, all reporter constructs contain regulatory regions from D. biarmipes and are tested in D. melanogaster transgenic flies.

Molecular biology

Lines from the D and E series and the randomized sequence for the Δ stripes and Δ broad were taken from (43). Fragments of y 5′ sequences for lines D8, Δ stripes, and Δ broad were amplified with Phusion polymerase (New England Biolabs) and cloned into our transformation vector pRedSA (43) digested with Eco RI and Bam HI using In-Fusion HD Cloning Kits (Takara, catalog no. 121416). Randomized sequences were amplified from the fragment used in (43). Primers are listed in table S3. The sequences are provided in table S4.

Imaging

Sample preparation

All transgenic abdomens imaged in this study were from male flies heterozygous for the reporter construct, obtained from a cross between homozygous reporter construct flies and a marker line [;en-Gal4, UAS-GFP/CyO; pnr-Gal4/TM6B, (57), FlyBaseID FBrf0098595]. White pupae were left to age 90 hours at 25°C. Pupal case was removed, and pupae were mounted in halocarbon oil (Sigma-Aldrich, CAS no. 9002-83-9) on a microscope slide with cover slips and appropriate spacers and immediately imaged.

Microscopy

All abdomen images were acquired as 12-bit images on a Leica SP5-2 confocal microscope using an HC PL APO CS 10×/0.40 IMM lens with a HyD SP Hybrid-Detector. Each image comprises two fluorescent channels (DsRed to report y regulatory activities and en-Gal4, pnr-Gal4, UAS-GFP as positional landmark for image registration). Image stacks were acquired in 4.99-μm steps.

Image quantification and analysis

We used a comprehensive pipeline for the registration, segmentation, and analysis of multichannel 3D fluorescence microscopy images. This pipeline, now supporting images with three color channels, facilitates the analysis of gene expression patterns in the epithelial cells of the Drosophila abdomen by producing a registered 2D map of expression on the abdomen surface starting from a 3D image stack (fig. S3A). The pipeline begins with an automated preprocessing step (fig. S3B), where segmentation of the image stack is performed using an automatically determined threshold. This threshold is set to ensure that the segmented volume covers a specified range of the entire image volume. After segmentation, the object is refined using morphological transformations and mesh fitting to fill any holes. Following preprocessing, the registration step involves rendering the segmented volume in an interactive interface. Corresponding points on the source and reference objects are selected, and the image stack is rigidly rotated and rescaled to minimize the distances between these points (fig. S3C). In the projection phase, the surface of the registered objects is projected into 2D images using a modified sinusoidal projection. The object is sliced, and, for each slice, the profile of the bright object is interpolated with a spline curve. Image brightness is then read out along the curve by taking the local maxima along the local normal direction. The resulting 1D brightness profiles from each slice form the rows of the 2D projected image, which are aligned at a predefined meridian plane (fig. S3D). The final step involves labeling and elastic warping. A graphical interface built with the PYSimpleGUI library enables manual labeling and registration of the 2D images. User-selected points are used to elastically warp the images onto a reference model using thin-plate spline registration (fig. S3E). This methodology ensures precise registration and analysis of multichannel 3D images, thereby facilitating the detailed study of gene expression patterns in the Drosophila abdomen. The Python-based code, together with an example of Jupyter Notebooks, installation instructions, and a minimal test dataset, is available at https://github.com/UniBonn-GompelLab/3D-DrosophilaRegistration.

Relative regulatory information and statistical analysis

We showed how much regulatory information (fluorescence levels) is contained in each line and compared it to the construct containing the largest regulatory fragment. First, to measure the amount of fluorescence, we calculate a squared Euclidean distance from each sample to the average phenotype of the line with an empty vector (ø). Practically, we subtracted from each projected abdomen image the average image of the empty line (ø) and then squared and sum all the values. Next, we calculated an average value of these distances for each line. Last, to show the relative amount of regulatory information, we divided all the line averages by the average of the biggest construct in the series (D0 line for the D series and D2 line for the E series). The statistical significance of differences among lines was assessed with Kolmogorov-Smirnov tests (table S1).

Density of regulatory information

We calculated the PCA from the matrix of dimensions (n_individuals × n_pixels on the abdomen), regrouping intensities of all pixels for every individual. Mixed components representing the stripes, broad, and dimorphic activities were obtained by a change of basis of the original PC space. We calculated vectors defining a new coordinate system as differences between the control lines and randomized lines without enchancer cores: D0 and Δ stripes, D0 and Δ broad, and D2 and Δ dimorphic. The amount of regulatory information brought by a segment of DNA for a given activity was calculated as the absolute value of the difference between the phenotypic distances of two consecutive fragments, relative to the phenotypic distance from the empty line (ø) to the largest construct in the series, using independent measurements for the broad, stripes, dimorphic, wing blade, and spot activities. To represent regulatory information, we used the absolute value of the change in the measure of activity, resulting in a similar depiction of repression and activation.

Abdominal transcriptome

RNA isolation

RNA isolation was performed on wild type flies using standard TRIzol (Invitrogen) protocol from abdominal dorsal epithelium of pupae 72 hours after puparium formation. The resulting RNA was further purified using “RNeasy Mini Protocol” for RNA cleanup (QIAGEN). RNA concentration was measured by spectrophotometer, and RNA quantity, quality, and size distribution were checked on a Bioanalyzer (Agilent Technologies).

Microarray experiment

In microarray experiments, each tissue was analyzed in three replicates consisting of 25 individuals each. RNA amplification was done according to Affymetrix “One cycle cDNA synthesis” and “Synthesis of biotin-labeled cRNA” protocols. Microarray hybridizations were conducted by Affymetrix facility.

Supplementary Materials

The PDF file includes:

Figs. S1 to S3

Table S4

Legends for tables S1 to S3

References

Other Supplementary Material for this manuscript includes the following:

Tables S1 to S3