Evolutionary mirages: selection on binding site composition creates the illusion of conserved grammars in Drosophila enhancers

Abstract

The clustering of transcription factor binding sites in developmental enhancers and the apparent preferential conservation of clustered sites have been widely interpreted as proof that spatially constrained physical interactions between transcription factors are required for regulatory function. However, we show here that selection on the composition of enhancers alone, and not their internal structure, leads to the accumulation of clustered sites with evolutionary dynamics that suggest they are preferentially conserved. We simulated the evolution of idealized enhancers from Drosophila melanogaster constrained to contain only a minimum number of binding sites for one or more factors. Under this constraint, mutations that destroy an existing binding site are tolerated only if a compensating site has emerged elsewhere in the enhancer. Overlapping sites, such as those frequently observed for the activator Bicoid and repressor Krüppel, had significantly longer evolutionary half-lives than isolated sites for the same factors. This leads to a substantially higher density of overlapping sites than expected by chance and the appearance that such sites are preferentially conserved. Because D. melanogaster (like many other species) has a bias for deletions over insertions, sites tended to become closer together over time, leading to an overall clustering of sites in the absence of any selection for clustered sites. Since this effect is strongest for the oldest sites, clustered sites also incorrectly appear to be preferentially conserved. Following speciation, sites tend to be closer together in all descendent species than in their common ancestors, violating the common assumption that shared features of species' genomes reflect their ancestral state. Finally, we show that selection on binding site composition alone recapitulates the observed number of overlapping and closely neighboring sites in real D. melanogaster enhancers. Thus, this study calls into question the common practice of inferring "cis-regulatory grammars" from the organization and evolutionary dynamics of developmental enhancers.

Figure 1. Simulation of enhancers under a compositional constraint.

(A) Starting state for a simulation of an enhancer constrained to have five sites for each of two different transcription factors (red triangles and blue circles). (B) A deletion (red bar) eliminates a site, bringing the total number for that factor to four and leading to the rejection of the mutation. (C) A mutation creates a new site (bringing the total to six) and is accepted. The subsequent deletion of an original site (red bar) does not reduce the total below five and is accepted, leading to a binding site turnover event. (D) Sample run of a simulation of an enhancer required to have five sites each for the D. melanogaster transcription factors Bicoid and Krüppel over 1,500 mutation-selection rounds. The course of the simulation proceeds from top to bottom, with all Bicoid sites in the enhancer shown in red and Krüppel sites in blue. Overlapped BCD/KR sites are darker and purple.

Figure 2. Turnover rates vary widely in proportion to information content of transcription factor binding site model.

The log of the half-life of different artificial and real binding sites against their specificity. Synthetic binding sites are plotted in gray, while sites derived from Drosophila transcription factors are highlighted: Krüppel (blue circle), Bicoid (red triangle), Giant (green diamond), and Hunchback (cyan hexagon). Specificity is defined as the difference in the information between the binding site and a random sequence of the same length.

Figure 3. Overlapping binding sites are enriched and appear preferentially conserved in simulated sequences.

(A) The post-simulation (S) probability of observing a Krüppel site conditioned on seeing a Bicoid site (blue) and a Bicoid site conditioned on seeing a Krüppel site (red) is always significantly higher than the expected probability (E) in random DNA for binding matrices derived from in vitro footprinting experiments. (B) Overlapping sites (solid line) are more likely than isolated sites (dashed line) to persist in simulations at a wide range of mutational distances.

Figure 4. A deletion bias leads to clustering of sites and the apparent conservation of clustered sites.

(A) The distribution of spacer lengths between binding sites during simulations in which 0% (black), 20% (light green), and 40% (dark green) of mutation events are indels with a 3∶2 deletion∶insertion bias. (B) The percent probability that a deletion event affecting a given binding site is accepted by our selective process for adjacent sites (Adj; sites that are touching) or far sites (Far; those with a spacer of at least twenty bases to the nearest neighboring site). (C) The distribution of the average age of binding sites as a function of their distance to their nearest neighbor shows that clustered sites appear more conserved than isolated sites, even though no such selection was applied in the simulations.

Figure 5. A deletion bias creates the appearance of conserved site spacing.

(A,B) Following an initial starting condition where two binding sites are 100 base pairs apart, the evolution of their spacing is simulated where either (A) there is no bias towards deletions or (B) the distribution of indels approximates that found in Drosophila. The probability of observing the sites separated by a given distance after a given number of substitutions is shown on a scale of deep blue (zero) to deep red (≥2%). Without a deletion bias, site spacing rapidly becomes unpredictable. However, the deletion bias, on average, ratchets sites together over time, correlating any two pairs of sites' evolution. (C–E) After starting 30 (C), 50 (D), or 100 (E) base pairs apart at a speciation event, orthologous pairs of sites are subjected to a simple test of spacing conservation. If both pairs of sites are separated by a distance of 30 base pairs or less after diverging by a certain number of substitutions, their close spacing is considered ‘conserved.’ We plot the chance that, given that none of the sites themselves have degraded, this apparent conservation could be created by a neutral model. This neutral model may have a balance of insertions and deletions (blue) or a deletion bias approximating Drosophila's (green). When no deletion bias is present, the chance that apparently conserved spacing is explained by neutral forces decreases over time, allowing better discrimination of ‘true’ conservation via negative selection. Drosophila's neutral mutation pattern not only reverses this trend (C), but also induces a substantial fraction of originally distantly spaced sites to appear to have a conserved close spacing (D, E).

Figure 6. Simulations of a well-characterized D. melanogaster enhancer.

One thousand simulations of the eve stripe 2 enhancer (see Methods) resulted in variable numbers of overlapping Bicoid and Krüppel sites (A, grey histogram) and sites within 10 basepairs of each other (B, grey histogram). The number of overlapping Bicoid/Krüppel site pairs, and closely spaced sites in the real eve stripe 2 enhancer are shown in red. That the real numbers are comfortably within the range produced by these simulations demonstrates that the higher-order structure in real D. melanogaster enhancers could plausibly have arisen solely from deletion biased mutation and selection to maintain binding site composition.