Abstract
Convergent phenotypic evolution is often caused by recurrent changes at particular nodes in the underlying gene regulatory networks (GRNs). The genes at such evolutionary ‘hotspots’ are thought to maximally affect the phenotype with minimal pleiotropic consequences. This has led to the suggestion that if a GRN is understood in sufficient detail, the path of evolution may be predictable. The repeated loss of larval trichomes among Drosophila species is caused by the loss of shavenbaby (svb) expression. svb is also required for development of leg trichomes, but the evolutionary gain of trichomes in the ‘naked valley’ on T2 femurs in Drosophila melanogaster is caused by the loss of microRNA-92a (miR-92a) expression rather than changes in svb. We compared the architectures of the larval and leg trichome GRNs to investigate why the genetic basis of trichome pattern evolution differs in these developmental contexts. We found key differences between these two networks in both the genes employed, and in the regulation and function of common genes. These differences in the GRNs reveal why mutations in svb are unlikely to contribute to leg trichome evolution and how instead miR-92a represents the key evolutionary switch in this context. Our work shows that differences in the components and wiring of GRNs in different developmental contexts, as well as whether a morphological feature is lost versus gained, influence the nodes at which a GRN evolves to cause morphological change. Therefore our findings have important implications for understanding the pathways and predictability of evolution.
Significance Statement A major goal of biology is to identify the genetic cause of organismal diversity. Convergent evolution of traits is often caused by changes in the same genes - evolutionary ‘hotspots’. shavenbaby is a ‘hotspot’ for larval trichome loss in Drosophila, however microRNA-92a underlies the gain of leg trichomes. To understand this difference in the genetics of phenotypic evolution, we compared the underlying gene regulatory networks (GRNs). We found that differences in GRN architecture in different developmental contexts, and whether a trait is lost or gained, influence the pathway of evolution. Therefore hotspots in one context may not readily evolve in a different context. This has important implications for understanding the genetic basis of phenotypic change and the predictability of evolution.
Introduction
A major challenge in biology is to understand the relationship between genotype and phenotype, and how genetic changes modify development to generate phenotypic diversification. The genetic basis of many phenotypic differences within and among species have been identified (e.g. 1, 2-15), and these findings support the generally accepted hypothesis that morphological evolution is predominantly caused by mutations affecting cis-regulatory modules of developmental genes (16). Moreover, it has been found that changes in the same genes commonly underlie the convergent evolution of traits (reviewed in 17). This suggests that there are evolutionary ‘hotspots’ in GRNs: changes at particular nodes are repeatedly used during evolution because of the role and position of the gene in the GRN, and the limited pleiotropic effect of the change (18-21).
The regulation of trichome patterning is an excellent system for studying the genetic basis of evolutionary morphological change (22). Trichomes are actin protrusions from epidermal cells that are overlaid by cuticle and form short, non-sensory, hair-like structures. They can be found on various parts of insect bodies during different life stages, and are thought to be involved in, for example, thermo-regulation, aerodynamics, oxygen retention in semi-aquatic insects, grooming, and larval locomotion (23-27) (Fig. 1).
The GRN underlying trichome formation on the larval cuticle of Drosophila species has been characterised in great detail (reviewed in 21, 22, 28) (Fig. 1). Several upstream transcription factors, signalling pathways, and tarsal-less (tal)-mediated post-translational proteolytic processing lead to the activation of the key regulatory transcription factor Svb, which, with SoxNeuro (SoxN), activates a battery of downstream effector genes (6, 29-35). These downstream factors modulate cell shape changes, actin polymerisation, or cuticle segregation, which underlie the actual formation of trichomes (30, 35). Importantly, ectopic activation of svb during embryogenesis is sufficient to drive trichome development on otherwise naked larval cuticle, and loss of svb function leads to a loss of larval trichomes (36).
Regions of dorso-lateral larval trichomes have been independently lost at least four times among Drosophila species (37, 38). In all cases, this phenotypic change is caused by changes in svb enhancers, resulting in a loss of svb expression (13-15, 37-39). The modular enhancers of svb are thought to allow the accumulation of mutations that facilitate the loss of certain larval trichomes without deleterious pleiotropic consequences. Note, however, that evolved enhancers underlying differences in larval trichomes also drive expression in other tissues (40). It is thought that changes in larval trichome patterns cannot be achieved by mutations in genes upstream of svb because of deleterious pleiotropic effects, while changes in individual svb target genes would only affect trichome morphology rather than their presence or absence (19-21, 30, 35). Given the position and function of svb in the larval trichome GRN, these data suggest that svb is a hotspot for the evolution of trichome patterns more generally because it is also required for the formation of trichomes in adult epidermis and can induce ectopic trichomes on wings when over expressed (36, 40). Therefore, one could predict that changes in adult trichome patterns are similarly achieved through changes in svb enhancers (20, 21).
The trichome pattern on femurs of second legs varies within and between Drosophila species (1, 41) (Fig. 1). In D. melanogaster, an area of trichome-free cuticle or ‘naked valley’ varies in size among strains from small to larger naked valleys. Other species of the D. melanogaster species subgroup only exhibit larger naked valleys (1, 41). Therefore trichomes have been gained at the expense of naked cuticle in some strains of D. melanogaster. Differences in naked valley size between species have been associated with differences in the expression of Ultrabithorax (Ubx), which represses the formation of leg trichomes (41). However, smaller naked valley size in populations of D. melanogaster is caused by a reduction of m/R-92a expression, which represses trichome formation by repressing the svb target gene shavenoid (sha) (1, 42). Therefore, while svb is thought to be a hotspot for the evolutionary loss of patches of larval trichomes, it does not appear to underlie the evolutionary the gain of leg trichomes in D. melanogaster.
Differences in GRN architecture among developmental contexts may affect which nodes can evolve to facilitate phenotypic change in different tissues or developmental stages. In addition, an evolutionary gain or loss of a phenotype may also result from changes at different nodes in the underlying GRN, i.e. alteration of a particular gene may allow the loss of a trait but changes in the same gene may not necessarily result in the gain of the same trait. Therefore, a better understanding of the genetic basis of phenotypic change and evaluation of the predictability of evolution requires characterising GRN architecture in different developmental contexts and studying how the loss versus the gain of a trait is achieved.
Here we report our comparison of the GRN underlying trichome development in legs versus embryos. Our results show that differences in GRN composition and architecture in these two developmental contexts mean that it is likely svb is unable to act as a switch for the gain of leg trichomes because it is already expressed throughout the legs in both naked and trichome-producing cells. Instead, regulation of sha by miR-92a appears to act as the switch between naked and trichome-producing cells in the leg. This shows that the architecture of a GRN in different developmental contexts can affect the pathway used by evolution to generate phenotypic change.
Materials and Methods
Fly strains, husbandry and crosses
Fly strains used in this study are listed in Table S1. Flies were reared on standard food at 25 °C if not otherwise indicated.
Replacement of the P{lacW}I(3)S011041 element, which is inserted 5’ of the tal gene, by a P{GaWB} transposable element was done by mobilization in omb-Gal4; +/CyO Δ2−3; l(3)S011041/TM3Sb flies as described in Galindo, Pueyo, Fouix, Bishop and Couso (32). Replacements were screened by following UAS-GFP expression in the progeny. The P{GaWB} element is inserted in the same nucleotide position as P{lacW}S011041. Clonal analysis of tal S18.1 and svbR9 alleles were performed as described in Pueyo and Couso (43).
A transgenic line that contains the cis-regulatory region of svb upstream of a GFP reporter (svbBAC-GFP) (40) was used to monitor svb expression. Legs of pupae were dissected 24 h after puparium formation (hAPF), fixed and stained following the protocol of Halachmi et al. (2012) (44), using a chicken anti-GFP as primary antibody (Aves Labs, 1:250) and an anti-chicken as secondary (AlexaFluor 488, 1:400). Images were obtained on a confocal microscope with a 60X objective. SUM projections of the z-stacks were generated after background subtraction. A filter median implemented in ImageJ software (http://rsb.info.nih.gov/ij/) was applied. The proximal femur image was reconstructed from two SUM projections using Adobe Photoshop.
RNA-seq
Pupae were collected within 1 hAPF and allowed to develop for another 20 to 28 h at 25 °C. Second legs were dissected in PBS from approximately 80 pupae per replicate and kept in RNAlater. RNA was isolated using phenol-chloroform extraction. This was done in three replicates for two different strains (e4,wo1,ro1 and OregonR). Library preparation and sequencing (75 bp paired end) were carried out by Edinburgh Genomics. Reads were aligned to D. melanogaster genome version 6.12 (45) using TopHat (46). Transcriptomes were assembled using Cufflinks and analysed using Cuffdiff (47) (Supplementary Files 1-7). Genes expressed below 1 FPKM were considered not expressed. The raw reads will be deposited in the Gene Expression Omnibus.
ATAC-seq
Pupae were reared and dissected as described above. Dissected legs were kept in ice cold PBS. Leg cells were lysed in 50 μl Lysis Buffer (10 mM Tris-HCl, pH = 7.5; 10 mM NaCl; 3 mM MgCl2; 0.1 % IGEPAL). Nuclei were collected by centrifugation at 500 g for 5 min. Approximately 60,000 nuclei were resuspended in 50 μl Tagmentation Mix [25 μl Buffer (20 mM Tris-CH3COO−, pH = 7.6; 10 mM MgCl2; 20 % Dimethylformamide); 2.5 μl Tn5 Transposase; 22.5 μl H2O] and incubated at 37 %C for 30 min. After addition of 3 μl 2 M NaAC, pH = 5.2 DNA was purified using a QIAGEN MinElute Kit. PCR amplification for library preparation was done for 15 cycles with NEBNext High Fidelity Kit; primers were used according to (48). This procedure was repeated for three replicates in each of two strains (e4,wo1,ro1 and OregonR). Paired end 50 bp sequencing was carried out by the Transcriptome and Genome Analysis Laboratory Göttingen, Germany. Reads were end-to-end aligned to D. melanogaster genome version 6.12 (FlyBase) (45) using bowtie2 (49). After filtering of low quality reads and removal of duplicates using SAMtools (50, 51), reads were re-centered according to Buenrostro, Giresi, Zaba, Chang and Greenleaf (48). Peaks were called with MACS2 (52) and visualisation was done using Sushi (53) (Supplementary Files 8, 9.)
Results
The composition of the leg trichome GRN differs from the larval trichome GRN
To better characterise the GRN underlying leg trichome development we first carried out RNA-Seq of T2 pupal legs between 20 and 28 hAPF, which is the window when leg trichomes are specified (41) (Supplementary File 1-6). We found that key genes known to be involved in larval trichome formation are expressed in legs. These include Ubx, SoxN, tal, svb, and sha, as well as key components of the Delta-Notch, Wnt and EGF signalling pathways (Table S2). However, we did not detect expression of Dichaete, arrowhead or abrupt, which are also known to regulate svb expression during larval trichome development (33, 39) (Table S2). Furthermore, we did not detect expression of 24 of the 163 known targets of svb in embryos (30, 35) in our dataset (Table S2). In addition, 10 out of the 43 genes thought to be involved in larval trichome formation independently of svb (34, 35) are not expressed in legs (Table S2). These changes in both Svb targets and other trichome effector genes possibly reflect differences in trichome morphology between larvae and legs (see 54). It also suggests that other factors, in addition to Svb, are required to activate these genes specifically during larval trichome development that are not used during leg trichome development. Alternatively, the Svb-dependent cis-regulatory elements of some of these genes may not be accessible during leg trichome formation. Overall, our RNA-Seq data exemplify differences in both upstream and downstream components of the leg trichome GRN when comparing it to the embryonic GRN that specifies larval trichomes.
We next compared our leg RNA-Seq data to published RNA-Seq datasets for embryos 12-14 and 14-16 h after egg lay (55). Svb activation during these developmental windows is critical for larval trichome formation (31). We identified a set of 105 genes expressed in our leg RNA-Seq data that showed little to no expression in embryos 12-16 hours AEL (Table S3). 94 of these ‘leg-specific’ genes are protein-coding while the other eleven produce non-coding RNAs. Gene ontology (GO) analysis of the protein-coding genes showed that nine are associated with chitin and cuticle development and hence may play a role in trichome formation, and a further five genes encode potential transcription factors (Table S3). Therefore, these genes represent candidates for the development of leg trichomes that are not used during larval trichome production.
Regulation of svb during leg trichome patterning
Given the important role of svb in trichome development and patterning, we investigated the regulatory sequences for this gene used in T2 legs. To do this we carried out ATAC-Seq (48, 56) on chromatin from T2 legs during the window of 20 to 28 hAPF when leg trichomes are specified.
Embryonic expression of svb underlying larval trichomes is regulated by several enhancers spanning a region of approximately 90 kb upstream of the transcription start site of this gene (15, 57) (Fig. 2). Several of these larval enhancers also drive reporter gene expression during pupal development (40). We observed that the embryonic enhancers DG3, E and 7 contained regions of open chromatin according to our T2 leg ATAC-Seq data. However, we found additional accessible chromatin regions that do not overlap with known embryonic svb enhancers (Fig. 2).
Deletion of a region including the embryonic enhancers DG2 and DG3 [Df(X) svb108] (Fig. 2) results in a reduction in the number of dorso-lateral larval trichomes when in a sensitized genetic background or at extreme temperatures (57). Moreover, Preger-Ben Noon and colleagues (2017) (40) recently showed that this deletion, as well as a larger deletion that also removes embryonic enhancer A ([Df(X) svb106], see Fig. 2), results in the loss of trichomes on abdominal segment A5, specifically in males. We found several peaks of open chromatin in the regions covered by these two deficiencies in our second leg ATAC-seq dataset (Fig. 2) and therefore tested the effect of Df(X) svb106 on leg trichome development. We found that deletion of this region and consequently enhancers DG2, DG3, Z and A did not affect the size of the naked valley or the density of trichomes on the femur or other leg segments of flies raised at 17°C, 25°C, or 29°C (compared to the parental lines) (Fig. S1). This suggests that while this region may contribute to svb expression in legs, its removal does not perturb the robustness of leg trichome patterning.
Next, to try to identify enhancer(s) responsible for leg expression, we employed all available GAL4 reporter lines for cis-regulatory regions of svb (Table S1) that overlap with regions of open chromatin downstream of the above deficiencies (Fig. 2). All 10 regions that overlap with open chromatin are able to drive GFP expression to some extent in second legs between 20 and 28 hAPF, as well as in other pupal tissues (Fig. S2). While some of the regions only produce expression in a handful of epidermal cells or particular regions of the T2 legs, none are specific to the presumptive naked valley, and VT057066, VT057077, VT057081, and VT057083 appear to drive variable levels of GFP expression throughout the leg (Fig. S2). Note that the two regions overlapping with larval enhancers E and 7 (VT057062 and VT057075, respectively) only drive weak expression in a few cells in the tibia and tarsus (Fig. S2).
To further test whether the expression of any of these regions is consistent with a role in trichome formation, we used them to drive expression of the trichome repressor miR-92a and the trichome activator sha-ΔUTR (see 1). Intriguingly, driving miR-92a under control of one of the fragments (VT057077) caused the repression of trichomes on all legs (Fig. 2, Fig. S3) as well as on wings and halteres (Fig. S3). However, expressing miR-92a under control of VT057062 or VT057075 had no noticeable effect. UAS-miR-92a under control of some of the other fragments (VT057053, VT057056) only led to repression of trichomes in small patches along the legs consistent with the GFP expression pattern (Fig. S2, Fig. S3).
Driving sha-ΔUTR with VT057077 is sufficient to induce trichome formation in the naked valley (Fig. 2) and on the posterior T3 femur (Fig. S3). Driving sha-ΔUTR under control of any of the other nine regions did not produce any ectopic trichomes in the naked valley on T2 or on any other legs. These results indicate that VT057077 is the only tested enhancer that bears sufficient regulatory information to be involved in trichome development throughout the second leg. Its ability to drive in the whole of the second leg, i.e. in regions which normally produce trichomes as well as in naked areas also suggests that svb is in fact expressed throughout the T2 leg, including the naked valley.
To test this further we examined the expression of svb transcripts in pupal T2 legs using in situ hybridization. However, this method produced variable results among legs and it was difficult to distinguish between signal and background in the femur (not shown). Therefore we examined the expression of a nuclear GFP inserted into a BAC containing the entire svb cis-regulatory region, which was previously shown to reliably capture the expression of this gene (40). We detected GFP throughout T2 legs at 24 hours after puparium formation including in the proximal region of the posterior femur (Fig. S4). This indicates that svb is expressed in naked valley cells that do not produce trichomes as well as more distal cells consistent with expression driven by enhancer VT057077.
miR-92a is sufficient to repress leg trichomes and acts downstream of Ubx
The above results suggest that components of the GRN for trichome production, including svb, are expressed in naked valley cells of the posterior T2 femur but are unable to produce trichomes. This differs to the situation in naked embryonic/larval cells and therefore might explain the differences in the genetic basis of trichome pattern evolution between these contexts. To test this further we examined the ability of genes to activate or repress leg trichomes.
It was previously shown that mutants of miR-92a have small naked valleys (58), which is consistent with the evolution of this locus underlying natural variation in naked valley size (1). We confirmed these findings using a double mutant for miR-92a and its paralogue miR-92b (59), which exhibits an even smaller naked valley (Fig. 3). We examined the morphology of the trichomes gained from the loss of miR-92a compared to the trichomes found more distally. We found that the trichomes gained were indistinguishable from the other leg trichomes (Fig. S5). This suggests that all of the genes required to generate leg trichomes are already transcribed in naked valley cells, but that miR-92a is sufficient to block their translation. Indeed, the extra trichomes that develop in the naked valley in the absence of miR-92a are dependent on svb, i.e. in a svb mutant background no trichomes are gained after a loss of miR-92 (Fig. 3). Furthermore, these results also show that trichome repression by Ubx in the naked valley (41) requires miR-92a and that the former must play a role upstream of the latter. Thus while Ubx is part of the GRN for the development of trichome patterns in larvae and legs it plays opposite roles in these two contexts: in embryos Ubx activates svb to generate larval trichomes, while in legs it represses trichomes in a miR-92a-dependent mechanism (41, 60, 61) (Fig. 1).
Svb and Sha differ in their capacities to induce trichomes in larvae and legs
It was previously shown that miR-92a inhibits leg trichome formation by repressing translation of the svb target sha (1). However sha mutants are still able to develop trichomes in larvae, albeit with abnormal morphology (30). These data suggest that there are differences in the functionality of svb and sha in larvae versus leg trichome formation and therefore we next verified and tested the capacity of svb and sha to produce larval and leg trichomes.
As previously shown (30), ectopic expression of svb is sufficient to induce trichome formation on normally naked larval cuticle (Fig. 4). However, we found that ectopic expression of sha in the same cells does not lead to the production of trichomes (Fig. 4). svb is also required for posterior leg trichome production (40; Figs 3, S6), but over expression of svb in the naked valley does not produce ectopic trichomes (Fig. 4). Over expression of sha on the other hand is sufficient to induce trichome development in the naked valley (1) (Fig. 4). These results show that svb and sha differ in their capacities to generate trichomes in larvae versus legs.
Svb acts as a transcriptional repressor and requires cleavage by the proteasome to become a transcriptional activator. This cleavage is induced by small proteins encoded by the tal locus (6, 31, 32). We therefore tested if svb is unable to promote trichome development in the naked valley because it is not activated in these cells. We found that expressing the constitutively active form ovoB, or tal, in naked leg cells is sufficient to induce trichome formation (Fig. 4), which is consistent with loss of tal in clones of leg cells resulting in the loss of trichomes (Fig. S6). Furthermore, it appears that tal, like svb, is expressed throughout the leg (Fig. S6). It follows that svb and tal are expressed in naked cells but are unable to induce trichome formation under normal conditions because of repression of sha by miR-92a. Over expression of tal on the other hand must be able to produce enough active Svb to result in an increase of sha transcription to overwhelm miR-92a repression.
Discussion
The GRNs for larval and leg trichome patterning differ in architecture and evolution
The causative genes and even nucleotide changes that underlie the evolution of an increasing number and range of phenotypic traits have been identified (17). An important theme that has emerged from these studies is that the convergent evolution of traits is often explained by changes in the same genes - so called evolutionary ‘hotspots’ (17, 62). This suggests that architecture of GRNs may influence or bias the genetic changes that underlie phenotypic changes (18, 19, 21). However, relatively little is known about the genetic basis of changes in traits in different developmental contexts and when features are gained versus lost (18).
It was shown previously that changes in svb underlie the convergent evolution of the loss of larval trichomes, while the gain of leg trichomes in D. melanogaster is instead explained by evolution of miR-92a (1, 13-15, 37, 38). We investigated this further by comparing the GRNs involved in both developmental contexts and examining the regulation and function of key genes.
Our results show that there are differences between the GRNs underlying the formation of larval and leg trichomes in terms of the components and the wiring used. These changes are found both in upstream genes of the GRN that help to determine where trichomes are made and in downstream genes whose products are directly involved in trichome formation (Fig. 1). The latter may also determine the differences in the fine-scale morphology of these structures on larval and leg cuticle (Fig. 1) (30).
Furthermore, while the key evolutionary switch in embryos, the gene svb, is also necessary for leg trichome production, this gene is not sufficient to produce leg trichomes in the naked proximal region of the T2 femur. This is because the leg trichome GRN employs miR-92a, which inhibits trichome production by blocking the translation of the svb target gene sha. In the legs of D. melanogaster, miR-92a acts as the evolutionary switch for trichome production, and the size of the naked valley depends on the expression of this gene (Fig. 5) (1).
Interestingly, we observed that the ectopic trichomes produced by expression of sha-ΔUTR in the naked valley are significantly shorter than those on the rest of the leg (Fig. S5). This suggests that although sha is able to induce trichome formation in these cells, other genes are also required for their normal morphology. Another svb-target gene, CG14395 (35), is also a strongly predicted target of miR-92a: its 3’UTR contains two conserved complete 8-mers corresponding to the binding site for this microRNA. We found that CG14395 is also expressed in our leg RNA-Seq data (Table S1). Therefore it is possible that miR-92a represses CG14395 and potentially other target genes in addition to sha to block trichome formation.
Other genetic bases for the evolution of leg trichome patterns?
In contrast to larvae, it is unlikely that mutations in svb can lead to evolutionary changes in legs to gain trichomes and decrease the size of the naked valley. This is because this gene (and all the other genes necessary for trichome production) is already transcribed in naked cells. In addition, a single svb enhancer is able to drive expression throughout the legs including the naked valley. Although other enhancer regions of this gene are able to drive some expression in patches of leg cells, none of these is naked valley-specific. This suggests that evolutionary changes to svb enhancers would be unlikely to only affect the naked valley. It remains possible that binding sites could evolve in this enhancer to specifically increase the Svb concentration in naked valley cells. This could overcome miR-92a-mediated repression of trichomes similar to experiments where tal and ovoB are over expressed in these cells, or when sponges are used to phenocopy the loss microRNAs (63). However, this does not seem to have been the preferred evolutionary route in D. melanogaster (1) (Fig. 5).
Our study also corroborates that Ubx represses leg trichomes (41) whereas it promotes larval trichome development through activation of svb (61). Moreover, our results indicate that Ubx acts upstream of miR-92a in legs because it is unable to repress leg trichomes in the absence of this microRNA. It is possible that Ubx even directly activates miR-92a since ChIP-chip data indicate that there are Ubx binding sites within the jigr1/miR-92a locus (64). Intriguingly, there is no naked valley in D. virilis and Ubx does not appear to be expressed in the second legs of this species during trichome development (41) (Fig. 5). However naked valleys are evident in other species in the virilis and montana groups and it would be interesting to determine if these differences were caused by changes in Ubx, miR-92a or even other loci (Fig. 5).
Evolutionary hotspots and developmental context
To the best of our knowledge, our study is the first to directly compare the GRNs underlying formation of similar structures that have evolved in different developmental contexts. Our results show that the GRNs for trichome production in larval versus leg contexts retain a core set of genes but also exhibit differences in the components used and in their wiring. These differences likely reflect changes that accumulate in GRNs during processes such as co-option (65) and developmental systems drift (66-68), although it remains possible that the changes have been selected for unknown reasons.
Importantly, we show that the differences in these GRNs may help to explain why they have evolved at different nodes to lead to the gain or loss of trichomes. This supports the suggestion that GRN architecture can influence the pathway of evolution and lead to hotspots for the convergent evolution of traits (17-19, 21). Indeed, such hotspots can also underlie phenotypic changes in different developmental contexts. For example, yellow underlies differences in abdominal pigmentation and wing spot pigmentation among Drosophila species (10, 11, 69, 70). However, we demonstrate that it cannot be assumed that evolutionary hotspots in one development context represent the nodes of evolution in a different context as a consequence of differences in GRN architecture.
Our findings also highlight that the genes that underlie the loss of features might not have the capacity to lead to the gain of the same feature. Therefore, while evolution may be predictable in particular contexts, it is very important to consider developmental context and whether a trait is lost versus gained. Indeed even when we map the genetic basis of phenotypic change to the causative genes it is important to understand the changes in the context of the wider GRN to fully appreciate how the developmental program functions and evolves. Since evolution is thought to favour changes with low pleiotropy (19, 71-74,) the effects of genetic changes underlying phenotypic change should be tested more widely during development. Such an approach recently revealed that svb enhancers underlying differences in larval trichomes are actually also used in other contexts (40). Interestingly, miR-92a is employed in several roles, including self-renewal of neuroblasts (59), germline specification (58), and circadian rhythms (75). It remains to be seen if the changes in this microRNA underlying naked valley differences also have pleiotropic consequences, and therefore if natural variation in naked valley size is actually a pleiotropic outcome of selection on another aspect of miR-92a function.
Supplementary material
Supplementary File 1. FPKM values (with high and low confidence values) after transcriptome assembly with cufflinks for Oregon R replicate 1.
Supplementary File 2. FPKM values (with high and low confidence values) after transcriptome assembly with cufflinks for Oregon R replicate 2.
Supplementary File 3. FPKM values (with high and low confidence values) after transcriptome assembly with cufflinks for Oregon R replicate 3.
Supplementary File 4. FPKM values (with high and low confidence values) after transcriptome assembly with cufflinks for e,wo,ro replicate 1.
Supplementary File 5. FPKM values (with high and low confidence values) after transcriptome assembly with cufflinks for e,wo,ro replicate 2.
Supplementary File 6. FPKM values (with high and low confidence values) after transcriptome assembly with cufflinks for e,wo,ro replicate 3.
Supplementary File 7. FPKM values (with high and low confidence values) for both Oregon R and e,wo,ro after comparison with cuffdiff.
Supplementary File 8. Oregon R svb locus ATAC-seq peaks (called with MACS2) with information about position, summit position, height, -log10 (p and q values), and enrichment.
Supplementary File 9. e,wo,ro svb locus ATAC-seq peaks (called with MACS2) with information about position, summit position, height, -log10 (p and q values), and enrichment.
Acknowledgements
We thank Georgina Haines-Woodhouse for technical assistance, Daniel Leite for help with bioinformatics, and members of the McGregor lab and Maike Kittelmann for comments and suggestions throughout the project. This work was funded by DFG Research Fellowships to SK (Ki 1831/1-1) and FAF (FR 3929/1-1), a BBSRC DTP studentship to ADB, grants from Ministerio de Economía y Competitividad (BFU2016-74961-P) and the Andalusian Government (BIO-396) to JLGS, and an Austrian Science Fund (FWF) Fellowship to APM (M1059-B09). RNA library preparation and sequencing were carried out by Edinburgh Genomics, The University of Edinburgh. Edinburgh Genomics is partly supported through core grants from NERC (R8/H10/56), MRC (MR/K001744/1) and BBSRC (BB/J004243/1). We thank Francois Payre (University of Toulouse), David Stern (Janelia Farm), Fen-Biao Gao (UMass Med School) and the Vienna Drosophila RNAi Center for providing fly stocks.