Distinct mechanisms decommission redundant enhancers to facilitate phenotypic evolution

Abstract

The evolutionary loss of morphological traits is often driven by changes in gene regulation. Many developmental genes are controlled by multiple redundant enhancers, raising the question of how robust regulatory systems can be dismantled to permit phenotypic transitions. Here, we show that the loss of larval trichomes in Drosophila sechellia resulted from the independent inactivation of four embryonic enhancers of the shavenbaby gene. Each enhancer was extinguished by a distinct mechanism: (i) a large deletion that removed essential sequences, (ii) the loss of activator sites and gain of repressor sites, (iii) the acquisition of a long-range silencer, and (iv) the unmasking of preexisting repression. Notably, three of these mechanisms relied on repression, pointing to repression as a rapid route for the evolutionary loss of robust regulatory elements. These results show that robustness in gene regulation does not prevent morphological change but instead provides multiple opportunities for mutations to reduce enhancer activity, giving selection many paths to reshape form.

Robust regulatory systems yield to evolutionary change via distinct mechanisms of enhancer elimination.

INTRODUCTION

The loss of morphological features has occurred frequently throughout the evolutionary history of life. This process, known as regressive evolution, usually occurs when certain traits become disadvantageous or neutral in a given environment (1, 2). Some of the best-known examples of adaptive regressive evolution include the loss of tails in primates (3), limbs in snakes (4, 5), eyes and pigmentation in cavefish (1, 6–8), pelvic spines in freshwater sticklebacks (9, 10), and spots in Drosophila wings (11, 12). Most of these evolutionary transitions involved sequence changes in cis-regulatory regions of key developmental genes, leading to reduction or loss of gene expression in a specific domain during development. For example, the repeated loss of the pelvic girdle in freshwater populations of sticklebacks resulted from deletions of a pelvic enhancer of the Pitx1 gene (9, 10). Similarly, hindlimb loss in snakes evolved through the accumulation of single-base mutations and small deletions in the ZRS enhancer of Sonic hedgehog (Shh), which affected transcription factor binding sites (4, 5). Loss of wing spots in Drosophila, on the other hand, partially evolved through the acquisition of a silencer element near the spot enhancer of yellow (11). In all these cases, the loss of morphological features resulted from a loss of function of a single developmental enhancer. These losses are thought to be facilitated by the modular nature of cis-regulatory regions (13), which allows a single enhancer of a given gene to evolve without affecting the function of other enhancers, thereby minimizing pleiotropic effects (14, 15).

However, the modularity paradigm has been challenged by recent experimental data [reviewed in Sabarís et al. (16), Kittelmann et al. (17), and McDonald and Reed (18)]. For example, many developmental genes have multiple enhancers with redundant functions (19–24). These so-called “shadow enhancers” have been demonstrated to ensure robust gene expression in the face of genetic and environmental variability (20–22). Another level of robustness may exist at the level of individual enhancers that contain multiple binding sites for the same transcriptional activator (25–28). How this inherent robustness can be overcome to evolve new phenotypes remains unclear, as there are almost no examples of the loss of multiple redundant enhancers underlying an evolutionary change.

One exception to this is the partial loss of function of the shavenbaby (svb) gene in Drosophila sechellia. The svb gene encodes a transcription factor that controls the formation of nonsensory, hairlike structures, named trichomes, on the cuticle of insects. Previous work done in Drosophila melanogaster revealed that the embryonic expression of svb is controlled by seven enhancers (designated 7, E6, E3, A, Z1.3, DG3, and DG2) that are scattered across a 90-kb region upstream of the core promoter (Fig. 1) (21, 29–31). These enhancers drive both unique and overlapping expression patterns, providing robustness to svb expression under environmental or genetic variation (21). In addition, it has been shown that some of these enhancers contain multiple binding sites for the same transcription factor (25, 26).

Fig. 1. Trichome patterns have evolved between Drosophila species by changes in the regulatory region of the svb gene.

(A) Lateral-view drawing of first instar larva of D. melanogaster. The rectangle marks the region shown in (B) to (D). (B to D) Dorsolateral cuticle of the fifth abdominal segment D. melanogaster (B), Drosophila simulans (C), and D. sechellia (D). The circles indicate three sensory bristles, which are found in all three species. Adapted from McGregor and colleagues (31). (E and F) Expression patterns of svb mRNA in stage 14 embryos of D. melanogaster (E) and D. sechellia (F) visualized by Hybridization Chain Reaction (HCR) RNA fluorescent in situ hybridization. (G) Schematic representation of the svb locus, indicating embryonic enhancers (colored boxes). The four enhancers that lost function in D. sechellia are marked with *. (H to O) Expression of D. melanogaster DG2::LacZ [(H) melDG2], D. sechellia DG2::LacZ [(I) secDG2], D. melanogaster Z1.3::LacZ [(J) melZ1.3], D. sechellia Z1.3::LacZ [(K) secZ1.3], D. melanogaster A::LacZ [(L) melA], D. sechellia A::LacZ [(M) secA], D. melanogaster E6::LacZ [(N) melE6], and D. sechellia E6::LacZ [(O) secE6] reporter constructs in D. melanogaster stage 15 embryos.

Despite the intrinsic robustness of the svb regulatory region (21, 25, 26, 32), trichome patterns have repeatedly evolved in larvae of the genus Drosophila through cis-changes in svb regulation during embryogenesis (33–37). For example, the dorsal and lateral cuticle of first instar larvae of D. melanogaster and Drosophila simulans exhibit broad patches of fine trichomes known as quaternary trichomes (Fig. 1, A to C). In contrast, their sister species D. sechellia has lost these trichomes, thus having a naked quaternary region (Fig. 1D). This morphological change is associated with reduced embryonic activity of four svb enhancers (Fig. 1, H to O), leading to a loss of svb expression in quaternary cells of the embryonic epidermis (Fig. 1, E and F) (21, 29, 31). We previously identified the genetic and molecular basis for the inactivation of the svb E6 enhancer in D. sechellia (Fig. 1, N and O) (25, 29). In D. melanogaster, E6 encodes multiple binding sites for the transcriptional activators Arrowhead and Pannier, which are required for robust svb expression. In D. sechellia, nucleotide substitutions in the E6 enhancer disrupted four Arrowhead binding sites and created a novel binding site for the transcriptional repressor Abrupt, resulting in complete loss of enhancer activity (25). Although we understand the inactivation of E6 in detail, the genetic mechanisms underlying the loss of function of the other three enhancers remain unknown. This leaves an open question as to whether evolution operated through similar or distinct mechanisms to decommission different enhancers in a complex regulatory region.

Here, we uncover the genetic basis for the loss of activity of three additional svb enhancers in D. sechellia, providing the first complete portrait of the genetic events underlying a case of regressive evolution. Each of the three svb enhancers lost activity through a different genetic mechanism: The Z1.3 enhancer acquired a 120-nucleotide deletion in its core sequence, the A enhancer evolved through the acquisition of a silencer element, and the DG2 enhancer lost an activator binding site, rendering it susceptible to preexisting repression. These findings demonstrate that independent elements within robust and redundant regulatory regions can evolve through different genetic routes to converge on the same functional outcome.

RESULTS

One large-effect deletion underlies the loss of the Z1.3 enhancer activity in D. sechellia

The embryonic expression of svb in D. melanogaster is driven by seven enhancers, four of which have lost activity in D. sechellia (Fig. 1, H to O). For example, the D. melanogaster Z1.3 enhancer (melZ1.3), a 1.3-kb sequence located 61 kb upstream of the svb promoter, drives patches of expression in the dorsal and lateral epidermis of stage 15 embryos (Fig. 1J). In contrast, the orthologous sequence from D. sechellia (secZ1.3) drives expression only in the most posterior abdominal segment (Fig. 1K).

To identify the genetic changes responsible for the reduced activity of secZ1.3, we first performed a multiple sequence alignment, comparing the secZ1.3 sequence to its orthologs from four related species that produce quaternary trichomes (fig. S1). This analysis revealed 16 single-nucleotide substitutions and two deletions of 120 and 35 nucleotides that are unique to the D. sechellia enhancer (Fig. 2A and fig. S1).

Fig. 2. One large-effect deletion and three small-effect nucleotide substitutions underly the loss of Z1.3 in D. sechellia.

(A) Schematic representation of the mutated melZ1.3 enhancer reporter constructs (magenta boxes). The D. sechellia Z1.3 enhancer is shown as a gray box. The positions of D. sechellia–specific substitutions and deletions are indicated with vertical and dashed red lines, respectively. (B to G) Expression of melZ1.3 (B) and secZ1.3 (C) reporter constructs and of melZ1.3 reporter constructs carrying the indicated D. sechellia–specific deletions and substitutions in stage 15 embryos. The embryo images in (B) and (C) are also shown in Fig. 1. (H) Box plots showing the integrated intensity of reporter activity in nuclei carrying the indicated constructs in the region outlined in (B) (n = 10 embryos for each genotype). In the plots, each point represents an individual embryo. Asterisks denote significant difference from melZ1.3 wild-type activities; *P < 0.05, **P < 0.01, and ***P < 0.001 (Kruskal-Wallis test with Hochberg-adjusted post hoc comparisons). WT, wild type; a.u., arbitrary units.

To test the effect of the D. sechellia–specific changes in Z1.3 on enhancer function, we introduced all of these changes into the melZ1.3 sequence while keeping the remaining sequence D. melanogaster derived, and assessed its activity using reporter gene assays (Fig. 2). In stage 15 embryos, the melZ1.3 construct carrying all the D. sechellia–specific changes (melZ1.3_mutALL) drove little to no expression, similar to the secZ1.3 enhancer (compare Fig. 2, C and G). Next, we grouped the D. sechellia–specific substitutions into six clusters and introduced each cluster of mutations separately into melZ1.3 (Fig. 2 and figs. S1 and S2). We also removed the nucleotides in melZ1.3 that correspond to the two D. sechellia–specific deletions. This analysis revealed that the 120-nucleotide deletion is the primary cause of the reduced embryonic activity of secZ1.3 (Fig. 2, E and H). This deletion removes sequences from Z0.3, a 300–base pair (bp) region within the melZ1.3 sequence that contains the information for driving its embryonic expression (30). In addition, three D. sechellia–specific nucleotide substitutions, also within the region that corresponds to Z0.3 (melZ1.3_mut2; Fig. 2, D and H), caused a slight but significant reduction in melZ1.3 activity.

Together, these results indicate that the loss of embryonic activity of Z1.3 in D. sechellia most likely resulted from a large-effect deletion, which removed essential activator binding sites. It is not possible to disentangle whether the substitutions that affect the activity of the enhancer occurred before or after the deletion. In any case, this mechanism is markedly different from that of the E6 enhancer, which evolved reduced function in the D. sechellia lineage through single-nucleotide substitutions that disrupted activator binding sites and created a site for a potent repressor.

Subregions within the D. sechellia A and DG2 enhancers retain activity

We next extended our analysis to the A and DG2 enhancers to determine which evolutionary paths underlie their reduced activity in D. sechellia. The D. melanogaster A (melA) enhancer drives expression in the cells that produce the thoracic tertiary trichomes and in lateral patches of the abdominal segments (Fig. 1L), whereas the D. melanogaster DG2 (melDG2) enhancer drives expression in dorsal and lateral patches (Fig. 1H). Unlike the E6 and Z1.3 enhancers, which have been dissected into small fragments that recapitulate the expression patterns of larger elements (29, 30), the A and DG2 enhancers reside in larger regions of ~5 kb. This makes it challenging to identify the specific genetic changes underlying their reduced activity in D. sechellia using an approach similar to that applied for E6 and Z1.3.

To overcome this challenge, we first dissected the melA and melDG2 enhancers into shorter fragments with the goal of identifying regions that recapitulate the spatial expression patterns of the larger fragments. To this end, we leveraged published single-cell assay for transposase-accessible chromatin with sequencing (scATAC-seq) data from D. melanogaster embryos (38) and chromatin immunoprecipitation sequencing (ChIP-seq) data for histone H3 lysine 27 acetylation (H3K27ac) from svb-expressing cells (39). We used the scATAC-seq dataset to map regions of accessible chromatin in embryonic epidermal cells (Fig. 3). These analyses revealed the regions of open chromatin coincide with previously dissected svb enhancers, including Z1.3 (30), E6 (29), and 7H (26) (Fig. 3). These enhancer regions are also flanked by H3K27ac, a histone mark associated with active enhancers (Fig. 3). For both A and DG2, we identified clear open-chromatin peaks within the full-length enhancer regions.

Fig. 3. Chromatin landscape at the svb locus predicts functional enhancer fragments.

Genome browser representations show pseudobulk ATAC-seq profiles from the embryonic epidermis (top) and H3K27ac profiles from svb-expressing nuclei (bottom) across the svb genomic region. ATAC-seq and H3K27ac peaks are highlighted with blue and purple boxes, respectively, above each profile. Gray shading marks the positions of the svb enhancers, also shown in the schematic at the top. The locations of the 7H enhancer and the newly identified enhancer fragments are shown below this schematic. chrX, chromosome X.

Guided by these data, we cloned a 1.2-kb fragment from the A enhancer that corresponds to the open-chromatin peak and named it melA1.2 (Fig. 4A). This fragment drives higher levels of expression than the full-length melA enhancer, with expanded activity in the abdominal tertiary trichome cells and broader expression in the lateral patches of the abdominal segments (Fig. 4J). Unexpectedly, the orthologous sequence from D. sechellia, secA1.2, drove a comparable spatial pattern and similar expression levels to melA1.2 (Fig. 4, K and L). These findings suggest that secA1.2 retains conserved activator binding sites and that repressive elements located outside this region are responsible for the reduced activity of the full-length secA enhancer.

Fig. 4. The D. sechellia DG2B fragment drives reduced activity, while the D. sechellia A1.2 region encodes conserved activator binding sites.

(A) Top: Structure of the svb locus, indicating the position of embryonic enhancers. Bottom: Schematics of the DG2 and A enhancer fragments tested for epidermal enhancer activity in reporter gene assays. Gray fragment did not drive expression. (B to E) Expression of melDG2 (B), secDG2 (C), melDG2B (D), and secDG2B (E) reporter constructs in stage 15 embryos. The embryo images in (B) and (C) are also shown in Fig. 1. (F) Box plots showing the integrated intensity of reporter activity in nuclei carrying the indicated constructs in the region outlined in (B) (n = 10 embryos for each genotype). In the plots, each point represents an individual embryo. Asterisks denote significant difference between species; ****P < 0.00001 (Student’s t test). (G to L) Expression of melA (G), secA (H), melA1.2 (J), and secA1.2 (K) reporter constructs in stage 15 embryos, juxtaposed to box plots [(I) and (L)] showing the integrated intensity of reporter activity in nuclei carrying the indicated constructs as in the region outlined in (G) and (J) (n = 10 embryos for each genotype). Each point in the plot represents an individual embryo. Asterisks denote significant difference between species; ***P < 0.001 and n.s. (not significant) (Student’s t test). The embryo images in (G) and (H) are also shown in Fig. 1.

To further dissect the melDG2 enhancer, we combined a similar bioinformatic approach with a systematic functional dissection. We first generated a series of overlapping 1-kb fragments spanning the 5.2-kb melDG2 region (Fig. 4A and fig. S3). Only one of these fragments, named melDG2B, drove expression within the svb expression domain that resembled the pattern driven by the full-length enhancer, although with some ectopic expression (Fig. 4D and fig. S3). All other fragments tested did not drive detectable expression in the embryo (fig. S3). As expected, melDG2B resides in a region of open chromatin that is flanked by H3K27ac in the D. melanogaster embryonic epidermis (Fig. 3). Next, we examined expression driven by the orthologous DG2B sequences from D. simulans and D. sechellia. The D. simulans fragment (simDG2B) drove expression similar to melDG2B (fig. S4). The D. sechellia orthologous fragment, secDG2B, drove low levels of expression in a similar spatial pattern (Fig. 4, E and F, and fig. S4). This observation suggests that, as with E6 and A, secDG2B retains conserved activator inputs, while the full-length secDG2 enhancer likely contains repressor binding sites that suppress its activity.

The D. sechellia A enhancer acquired long-range repression upstream of secA1.2

To identify the source of repression within the D. sechellia A enhancer, we tested larger enhancer fragments from both D. melanogaster and D. sechellia that included the A1.2 element and compared their activity (fig. S5). All tested fragments, regardless of species, drove lower expression levels than the A1.2 constructs. This reduction may reflect conserved repressive inputs.

In all fragments that included sequences upstream of A1.2, the D. sechellia ortholog consistently drove significantly lower expression than the D. melanogaster ortholog. Moreover, the magnitude and significance of this difference increased with the length of the upstream region, with the most pronounced difference observed between the full-length melA and secA enhancers (fig. S5). These findings suggest that repressive activity in D. sechellia is distributed across a broad genomic region upstream of A1.2, consistent with the presence of a long-range silencer element that evolved in the secA enhancer. The secA sequence contains an 833-bp insertion upstream of the A3.6 fragment (fig. S5), which may harbor some of this repressive activity.

The D. sechellia DG2 enhancer lost two activator binding sites

We next sought to identify the nucleotide changes responsible for the reduced function of the D. sechellia DG2B enhancer. First, we compared the DG2B sequence between D. sechellia and five related species, as was done for the Z1.3 enhancer (fig. S6). This analysis revealed 13 single-nucleotide substitutions, a three-nucleotide deletion, and two single-nucleotide deletions unique to the D. sechellia sequence. To test the effect of these substitutions and deletions on enhancer activity, we introduced all of them into the simDG2B sequence and tested its function using quantitative reporter gene assay (Fig. 5). We chose to manipulate the D. simulans sequence because it is the species most closely related to D. sechellia. The resulting construct, designated simDG2B_mutALL, drove expression indistinguishable in pattern and levels from the secDG2B enhancer (Fig. 5, E and F).

Fig. 5. The D. sechellia DG2 enhancer has lost two activator binding sites.

(A) Schematic representation of the mutated simDG2B enhancer reporter constructs (yellow boxes). The D. sechellia DG2B enhancer is shown as a gray box. Positions of D. sechellia–specific substitutions and deletions are indicated by vertical and zigzagged red lines, respectively. (B to E) Expression of simDG2B (B) and secDG2B (C) reporter constructs and of simDG2B reporter constructs carrying the indicated D. sechellia–specific deletions and substitutions [(D) and (E)] in stage 15 embryos. (F) Box plots showing the integrated intensity of reporter activity in nuclei carrying the indicated constructs in the region outlined in B (n = 10 embryos for each genotype). In the plots, each point represents an individual embryo. Asterisks denote significant difference from simDG2B wild-type activities; **P < 0.01 and ***P < 0.001 (Kruskal-Wallis test with Hochberg-adjusted post hoc comparisons). (G) Sequence alignment of the nucleotides surrounding the D. sechellia–specific substitutions in mut7 from D. simulans and D. sechellia, and the locations of the polyadenine tracks introduced to disrupt these sites (underscored). (H) Box plots showing the integrated intensity of reporter activity in nuclei carrying the indicated constructs as in (F). Asterisks denote significant difference from wild-type activities; *P < 0.05, **P < 0.01, and n.s. (Kruskal-Wallis test with Hochberg-adjusted post hoc comparisons).

To determine which of the D. sechellia–specific nucleotide changes contributed to the reduced expression, we grouped them into 10 clusters and introduced each cluster separately into the simDG2B sequence (Fig. 5A and figs. S6 and S7). Notably, only one construct, named simDG2B_mut7, drove reduced expression levels, which are similar to secDG2B and simDG2B_mutALL (Fig. 5, C to F).

The simDG2B_mut7 construct includes two D. sechellia–specific substitutions (Fig. 5G). We next asked which of these substitutions affect enhancer activity and whether their effect results from the loss of activator binding sites or the gain of repressor binding sites. To investigate this, we conducted an adenine scan. We separately replaced the sequence surrounding each of the two sites with polyadenylate [poly(A)] stretches in both the simDG2B and secDG2B enhancers (Fig. 5G and fig. S8). The rationale for this experiment is that if a D. sechellia mutation caused the loss of an activator site, replacing the corresponding site in simDG2B with a poly(A) stretch should reduce its activity without affecting secDG2B. Conversely, if the mutation introduced a repressor site, then replacing it with a poly(A) stretch should increase secDG2B activity while having no effect on simDG2B.

When we replaced each of the two candidate sites in simDG2B with poly(A) stretches, both resulting constructs showed reduced enhancer activity (Fig. 5H and fig. S8). However, only simDG2B_polyA1 drove significantly lower expression levels compared to simDG2B, with levels comparable to those driven by simDG2B_mut7. In contrast, introducing the same poly(A) substitutions into secDG2B did not significantly affect its activity (Fig. 5H and fig. S8). These results suggest that the D. sechellia–specific substitutions primarily reduce DG2B activity by disrupting activator binding sites.

Loss of activator sites unmasks conserved repression in the DG2 enhancer of D. Sechellia

Our finding that the secDG2B fragment can drive residual expression, while the full secDG2 enhancer does not drive any detectable expression, suggests that the secDG2 enhancer contains repressive sequences outside of secDG2B. To identify the source of this repression, we tested larger fragments surrounding DG2B (Fig. 6). A fragment that included upstream sequences of secDG2B, named secDG2AB, drove similar expression to secDG2B (Fig. 6, G and L). The orthologous sequence from D. melanogaster, melDG2AB, drove slightly lower expression compared to melDG2B (Fig. 6, F and L), suggesting that this upstream region contains a repressor binding site specific to the D. melanogaster sequence.

Fig. 6. The D. sechellia DG2 enhancer evolved through minimizing enhancer robustness to conserved repression.

(A) Schematics of the DG2 enhancer fragments tested for epidermal enhancer activity in reporter gene assays. Gray fragment did not drive expression. (B to K) Expression of the indicated DG2 reporter constructs from D. melanogaster [(B), (D), (F), (H), and (J)] and D. sechellia [(C), (E), (G), (I), and (K)] in stage 15 embryos. The embryo images in (B) and (C) are also shown in Fig. 1. The embryo images in (D) and (E) are also shown in Fig. 4. (L) Box plots showing the integrated intensity of reporter activity in nuclei carrying the indicated constructs in the region outlined in (D) (n = 10 embryos for each genotype). In the plot, each point represents an individual embryo. Asterisks denote significant difference from wild-type activities; *P < 0.05, **P < 0.01, ***P < 0.001, and n.s. (Kruskal-Wallis test with Hochberg-adjusted post hoc comparisons).

In contrast, the fragment secDG2BC, which included downstream sequences of secDG2B, showed markedly reduced activity, indicating the presence of repressive elements in this region (Fig. 6, I and L). Unexpectedly, the orthologous sequence from D. melanogaster, melDG2BC, also drove lower expression compared to melDG2B, suggesting that repression within the DG2C fragment is conserved across species (Fig. 6, H and L). To account for the possibility that this reduction resulted from increased distance between the enhancer and the minimal promoter in the reporter construct, we inverted the enhancer orientation, so that the DG2B fragments remained adjacent to the promoter (fig. S9). Inverting melDG2B (melDG2B-flipped) and secDG2B (secDG2B-flipped) both led to reduced expression, suggesting that their activating elements are located near the end adjacent to the DG2C fragment and that distance from the promoter does play a role (fig. S9, C, F, H, and K). Nonetheless, adding the DG2C fragment upstream of DG2B-flipped from either species (melDG2CB and secDG2CB) led to a further reduction in expression (fig. S9, E, F, J, and K). These results support the conclusion that the DG2C fragment contains conserved repressive sequences, although increased distance from the promoter may also contribute to the observed reduction in activity.

To further dissect the repressive activity within secDG2C, we divided the region into six ~100-bp fragments and tested each one individually by placing it upstream of the secDG2B-flipped enhancer (fig. S10). Four fragments (secDG2C1, secDG2C2, secDG2C3, and secDG2C4) reduced expression compared to secDG2B-flipped (fig. S10Q), but only secDG2C2 caused a significant decrease (fig. S10, K and Q). In contrast, none of the corresponding melDG2C fragments significantly lowered expression (fig. S10P), suggesting that the repressive effect in D. melanogaster cannot be attributed to a short, isolated sequence.

These results indicate that in both species, the DG2C region contains repressive elements that reduce DG2B activity. We propose that the secDG2 enhancer was inactivated through the loss of activator binding sites in secDG2B, thereby unmasking a conserved repressive activity in the DG2C region.

DISCUSSION

Genes controlled by redundant regulatory elements raise a central paradox: How can robust expression systems buffered by enhancer redundancy be dismantled during evolution? Using the Drosophila svb gene as a model, we demonstrate that the evolutionary loss of quaternary trichomes in D. sechellia larvae occurred through the independent inactivation of four embryonic enhancers (Fig. 7). Each enhancer lost its function through a different molecular mechanism: single-nucleotide substitutions that removed activator binding sites and created a repressor binding site in E6, acquisition of long-range repression in A, loss of activator binding that unmasked conserved repression in DG2, and a large-effect deletion in the core of Z1.3. Notably, three of these mechanisms involved repression, highlighting repressive interactions as an underappreciated driver of regulatory change and suggesting that they may provide a particularly rapid path to phenotypic evolution.

Fig. 7. The svb enhancers evolved reduced functionality in D. sechellia through diverse mechanisms.

Schematics illustrate predicted transcription factor occupancy at the svb DG2 (yellow), Z1.3 (purple), A (green), and E6 (cyan) enhancers in D. melanogaster (top) and D. sechellia (bottom). Circles represent putative activator binding sites; octagons represent putative repressor binding sites.

Our work reveals that enhancer inactivation can occur through different molecular routes involving repression. We previously showed that the svb E6 enhancer lost function in D. sechellia through the combined effect of multiple single-nucleotide substitutions that eliminated several Arrowhead binding sites and created a novel binding site for the potent repressor Abrupt (25). This combination enabled complete inactivation of a previously robust enhancer. Here, we show that the svb A and the DG2 enhancers were also inactivated through repression. The repeated evolution of repression across independent enhancers, all contributing to reduced svb activity in the same phenotypic direction, suggests that the naked cuticle of D. sechellia was shaped by natural selection rather than by genetic drift (17, 40).

The A enhancer in D. sechellia was inactivated through the evolution of a long-range silencer that suppresses its activity. Silencers have previously been implicated in additional cases of morphological evolution in Drosophila. For example, gain and loss of silencer elements in the regulatory region of the ebony gene contributed to evolutionary changes in male abdominal pigmentation (41, 42). Similarly, the emergence of a silencer in the yellow gene regulatory region led to the secondary loss of wing spots in D. melanogaster (11). Our findings at the svb locus add a previously unknown example and suggest that modulation of silencer activities may be a common route for morphological change. Despite their importance, silencers remain an understudied class of regulatory elements, and their mechanisms of action are poorly understood (43). It remains unclear whether they act locally by interfering with enhancer activity or more broadly by inhibiting transcription at promoters. Addressing these questions will require genomic approaches such as ATAC-seq and chromatin conformation assays, which can provide insight into chromatin accessibility and long-range genomic interactions in an evolutionary context.

The DG2 enhancer reveals a previously unknown mode of inactivation: loss of resilience to conserved repression. While the secDG2B fragment retains residual activity, the full-length secDG2 is completely inactive. Dissection of adjacent sequences showed that the DG2C region carries repressive properties conserved between species. In D. melanogaster, this repression appears to fine-tune the expression domain, but in D. sechellia, the loss of activator input in DG2B renders the enhancer vulnerable to this preexisting repression, resulting in complete silencing. To our knowledge, this is the first described case in which enhancer inactivation occurred through loss of activator binding that allows conserved repression to dominate.

The Z1.3 enhancer is unique among the svb enhancers in that it was inactivated through a 120-bp deletion within its core sequence. Deletions of enhancer sequences have been documented in several cases of morphological evolution (9, 44–46), including the repeated loss of pelvic spines in freshwater stickleback populations (9). However, this type of mutation is thought to be less likely in enhancers that are pleiotropic, as deletions risk disrupting multiple regulatory functions. We previously showed that all embryonic svb enhancers can drive gene expression in multiple tissues and at different developmental stages (30). Unlike other svb enhancers, where the same transcription factor binding sites are reused across developmental contexts, the Z1.3 enhancer exhibits a physical separation between the sequences required for embryonic and pupal expression (30). Notably, the 120-bp deletion in D. sechellia removes only the sequences necessary for embryonic activity while leaving the adjacent pupal regulatory region intact. As a result, the pupal activity of Z1.3 is preserved in D. sechellia (30). This organization of regulatory information thus permitted the selective loss of embryonic function through deletion, without compromising the enhancer’s roles in other developmental contexts.

While we have focused here on sequence-level changes within enhancers, the broader regulatory landscape of the svb locus also warrants attention. Given the extensive cis-regulatory inactivation observed in D. sechellia, an important open question is whether chromatin architecture has also evolved in D. sechellia. For example, how have the sequence changes in D. sechellia affected chromatin accessibility? On one hand, the acquisition of a silencer, as seen in secA, could increase chromatin accessibility, as silencers often exhibit open chromatin that enables repressor binding (11). On the other hand, potent repressors such as Abrupt, which represses the activity of secE6, might function by promoting chromatin compaction, leading to reduced accessibility. These possibilities could be tested using comparative, cell-type–specific ATAC-seq, which may also help pinpoint the location and extent of repressive activity in the D. sechellia A enhancer.

Another timely question concerns the three-dimensional (3D) organization of the svb locus in D. sechellia. Do physical interactions between enhancers and the svb promoter persist in D. sechellia despite the loss of enhancer activity? Or has the loss of activator input and the acquisition of repression reshaped the 3D architecture of this locus? We have recently found that in D. melanogaster, the svb enhancers form a 3D hub that brings multiple enhancers into contact with the svb promoter. This hub persists even in the absence of transcription factor binding to individual enhancers. This observation raises the possibility that the genetic changes affecting enhancer function in D. sechellia may not disrupt the higher-order chromatin structure of the locus. If so, then this would suggest that 3D enhancer-promoter organization is robust to genetic perturbations of this scale. Testing this hypothesis will require cell-type–specific chromosome conformation approaches that can directly compare the spatial organization of the svb locus between D. melanogaster and D. sechellia embryos.

Across metazoa, many developmental traits are governed by regulatory architectures that buffer gene expression and ensure phenotypic reproducibility (20–24). This study provides a detailed genetic account of how such robust systems, once thought to buffer against evolutionary change, can themselves serve as the substrate for morphological divergence. Our results show that evolution can exploit diverse mutational paths to silence gene expression, including deleting enhancers, weakening their activity, or repressing them through existing or novel mechanisms. The inactivation of different svb enhancers through distinct molecular mechanisms indicates that similar morphological outcomes can be achieved through multiple, enhancer-specific regulatory routes. These findings underscore the flexibility of cis-regulatory evolution and reveal how robust developmental systems remain evolvable.

MATERIALS AND METHODS

Transgenic constructs and Fly strains

Enhancer fragments were amplified by polymerase chain reaction (PCR) from genomic DNA of the corresponding species (see table S1 for details). For mutagenesis experiments, the Z1.3 mutated fragments were synthesized by GenScript, and the DG2B mutated fragments were generated by site-directed mutagenesis using PCR, except for the DG2BmutAll fragment, which was synthesized by IDT. The Poly(A) DG2B fragments were synthesized by Twist Bioscience. All enhancer fragments were cloned into the placZattB reporter plasmid using Gibson Assembly (see table S2 for cloning details). Plasmids were integrated into the attP2 or attP40 landing sites by Rainbow Transgenic Flies.

Embryo staining and image analysis

Stage 15 embryos were collected, fixed, and stained using standard protocols with mouse anti–β-galactosidase (1:500; Promega) and Alexa Fluor 488 goat anti-mouse (1:500; Invitrogen) antibodies. Embryos carrying reporter constructs were imaged on a ZEISS LSM 900 Confocal Microscope. Image analysis and fluorescent intensity quantification were performed using Fiji (https://imagej.net/) (47). Briefly, confocal stacks were converted to maximum intensity projections, and background fluorescence was subtracted using a rolling-ball radius of 50 pixels. Nuclei in abdominal segments A2 to A5 were segmented using the “Analyze Particles” tool, and the mean fluorescence intensity of each nucleus was measured. Expression was quantified in two ways: at the single-nucleus level, by pooling all nuclear intensities across embryos and visualizing their distribution with violin plots; and at the embryo level, by summing the mean intensities of all segmented nuclei within each embryo to yield an Integrated Intensity value.

Statistical analysis

All statistical analyses and data visualizations were performed using R (version 4.3.1, 2023) (table S3). To assess the distribution of each dataset, we first tested for normality using the Shapiro-Wilk test and for homogeneity of variances using Levene’s test. For normally distributed data with equal variances, we applied Student’s t test when comparing two groups (e.g., the A enhancer fragments) and one-way analysis of variance (ANOVA) when comparing more than two groups. In cases where ANOVA indicated a significant difference (P < 0.05), we performed Tukey’s post hoc test to identify which specific groups are different while controlling for multiple comparisons. If variances were unequal, then we used Welch’s t test or Welch’s ANOVA with P value adjustment Hochberg, as appropriate. For nonnormally distributed data, we used the Kruskal-Wallis test for comparisons among more than two groups, followed by the Hochberg procedure to adjust P values.

Genomic data analysis

Single-cell ATAC-seq FASTQ files from Drosophila embryos (Gene Expression Omnibus: GSE101581) (38) were processed as pseudobulk profiles using a previously described pipeline (39). Reads from epidermal cells (stages 14 and 15), as annotated by the Descartes portal (38), were pooled before peak calling. Peaks were identified using MACS2 (v2.2.9.1) (48) with parameters adjusted to account for Tn5 insertion bias.

Similarly, H3K27ac ChIP-seq data from sorted E10::GFP-positive epidermal nuclei (39) were processed using bbduk.sh for adapter trimming, aligned with BWA-MEM2, and peak calling was performed with MACS2 (v2.2.9.1) using default parameters to identify sharp enrichment peaks characteristic of H3K27ac enrichment at active enhancers and promoters. Input H3 DNA was used as a control for background subtraction.

Supplementary Materials

The PDF file includes:

Figs. S1 to S10

Legends for data S1 to S3

Other Supplementary Material for this manuscript includes the following:

Data S1 to S3