Sequence conservation and combinatorial complexity of Drosophila neural precursor cell enhancers

Neural precursor cell enhancers share highly conserved core sequence elements

Most c DTs discovered in this analysis represent elements that are shared pairwise, i.e., by only two of the NB enhancers examined (see the website for a list of cDTs that are shared by only two of the enhancers examined). The fact that the majority of c DTs are shared two ways, with only a small subset of sequences being shared three or more ways, suggests that the cis-regulation of early neural precursor genes is carried out by a large number of factors acting combinatorially and/or that many of the identified c DTs may in fact represent interlocking sites for multiple factors, and the exact orientation and spacing of these sites may differ among enhancers.

Neural specific cDTs that contain bHLH TF DNA-binding sites

During Drosophila neurogenesis, bHLH proteins function as proneural TFs to initiate neurogenesis in both the central and peripheral nervous system. TFs encoded by the achaete-scute complex function in both systems, while the related Atonal bHLH protein functions exclusively in the PNS [31]. Different proneural bHLH TFs, acting together with the ubiquitous dimerization partner Daughterless, bind to distinct E-boxes that contain different core sequences [32]. In addition to the core recognition sequence, flanking bases are important to the DNA binding specificity of bHLH factors [33].

Our analysis also reveals that not only are the core bases of E-boxes shared between similarly regulated enhancers, but bases flanking the E-box were also found to be highly conserved and are also frequently shared by these enhancers. Among the E-boxes found in CSBs of NB enhancers (many are illustrated in Table 2) aaCAGC TG (core bases of E-box are bold, flanking bases lower case) is repeated three times in nerfin-1 and once in scrt; gCACT TG is repeated three times in scrt; CAGC TGCA is repeated twice in wor, and CAGC TGctg is repeated twice in scrt (see Fig 1). In the dpn CNS NB enhancer, the E-box CAGCTG is found twice, separated by a single base (CAGC TGaCAGC TG). None of these sequences were present in mesodermal enhancers examined, but each is found in PNS enhancers; CAGC TGCA is repeated multiple times among PNS enhancers. Among the conserved PNS enhancer E-boxes (CAAA TGca, gcCAAA TG, cacCAAA TGg, CACA TGttg, gCACG TGtgc, ttgCACG TG, agCACG TGcc, aCAGA TG, ggCAGA TGt, CAGC TGccg, CAGC TGcaattt, gCAGG TGta and cCAGG TGa) each, including flanking bases, is found in two or three PNS enhancers, and these are distributed among all 13 enhancers. Of these, only agCACG TGcc, CAGC TGccg, cCAGG TGa were found once in our sample of neuroblast enhancers and none were found in our sample of mesodermal enhancers. The sequence aaCAAG TG is found in 4 E(spl) complex enhancers, those for E(spl)m8, mγ, HLHmδ and m6, and the sequence aCAGC TGc is found twice in E(spl)m8 and once in m4 and m6; neither sequence was found in our mesodermal enhancers. Therefore, although a given hexameric sequence may often be shared by all three types of enhancers, NB, PNS and E(spl), when flanking bases are taken into account there appears to be enhancer type-specific enrichment for different E-boxes.

Neural specific cDTs that contain Antennapedia class homeodomain DNA-binding sites

Antennapedia class homeodomain proteins play essential roles in multiple aspects of neural development including cell proliferation and cell identity [35]. The segmental identity of Drosophila NBs is conferred by input from TFs encoded by homeotic loci of the Antennapedia and bithorax complexes [36–38]. For example, ectopic expression of abd-A, which specifies the NB6-4a lineage, down-regulates levels of the G1 cyclin, CycE [38]. Loss of Polycomb group factors has been shown to lead to aberrant derepression of posterior Hox gene expression in postembryonic NBs, which causes NB death and termination of proliferation in the mutant clones [39].

We have examined the enhancer-type specificity of sequences flanking the Antennapedia class core DNA-binding sequence, ATTA [40]. Nearly 25% of the NB and PNS CSBs examined in this study contain this core recognition sequence. ATTA-containing sites were found multiple times in selected NB and PNS enhancers (Figure 1). The cis-Decoder analysis identified 18 different neural specific ATTA containing c DTs that were exclusively shared by two or more PNS enhancers or CNS enhancers and 10 were found to be shared between PNS and CNS. The most common c DT, ATTA gca, was shared by two CNS and two PNS enhancers (Figure 1; consensus homeodomain-binding sites are bold, flanking sequence lower case). In addition, 6 homeodomain-binding site c DTs were found twice in wor CSBs, aATTA ccg, tttgaATTA, aatcaATTA, ATTAAT ctt and aaacaaATTA g, but not in other CNS or PNS enhancer CSBs. In some cases these c DTs were found repeated in given enhancer CSBs. Only one of these c DTs aligned with CSBs of enhancers of the E(spl) complex. Given that 2/3 of the occurrences of HOX sites in these promoters can be accounted for by c DTs whose flanking sequences are shared between enhancers, it is unlikely that the appearance of these shared sequences occurs by chance.

In summary, the appearance of Hox sites in the context of conserved sequences shared by functionally related enhancers suggests that the specificity of consensus homeodomain-binding sites is conferred by adjacent bases, either through recognition of adjacent bases by the TF itself or in conjunction with one or more co-factors.

Neural specific cDTs that contain Pbx/Extradenticle sites

Examination of the c DTs from Drosophila NB and PNS enhancers revealed that many contained the core Pbx/Extradenticle docking site ATGA [41, 42]. In Drosophila, Extradenticle has been shown to have Hox-dependent and independent functions [43]. Studies have also shown that Pbx factors provide DNA-binding specificity for homeodomain TFs, facilitating specification of distinct structures along the body axis [43]. In the CNS enhancers of Drosophila, most predicted Pbx/Extradenticle sites are not, however, found adjacent to Hox sites.

Our analysis revealed that 8 of the Pbx motifs were shared between CNS and PNS enhancer types, and 16 were shared between similarly expressed enhancers (Figure 2), thus indicating that there appears to be some degree of specificity to Pbx site function when flanking bases are taken into account. Three of the Pbx binding-site containing elements also exhibit ATTA Hox sites: 1) the dodecamer GATGATTAAT CT (Pbx site is ATGA, Hox sites in bold) shared by the PNS enhancers edl and amos (references in Table 1), contains a homeodomain ATTA site that overlaps the Pbx site by a single base, and 2) the smaller heptamer ATGATTA, shared by pfe and ato, likewise contains a homeodomain ATTA site (bold) that overlaps ATGA Pbx site by a single base. Adjacent Hox and Pbx sites have been documented to facilitate synergy between the two factors [44]. Taken together our findings suggest that, as with homeodomain-binding sites, the conserved bases flanking putative Pbx sites are functionally important. These flanking bases are likely to confer different DNA-binding affinities for Pbx factors or are required for binding of other TFs.

Neural specific cDTs that contain Suppressor of Hairless binding sites

Also indicating a degree of biological specificity of enhancer types is the distribution of Suppressor of Hairless Su(H) binding sites among neural enhancers. Su(H) is the Notch pathway effector TF of Drosophila [45]. The members of the E(spl) complex, both the multiple basic helix-loop-helix (bHLH) repressor genes and the Bearded family members, have been shown to be Su(H) dependent [23, 26]. The consensus in vitro DNA binding site for Su(H) is RTGRGAR (where R = A or G) [25]. Notch signaling via Su(H) occurs through conserved single or paired sites [46] and the presence of conserved sites for other transcription regulators associated with CSBs containing Su(H) binding sites has been documented [47].

Within the CSBs of the six NB enhancers examined, only two, dpn and wor, contained conserved putative Su(H)-binding sites; two dpn sites matched one of the Su(H) consensus sites (GTGGGAA) and two wor sites match the sequence ATGGGAA. Only one of the two dpn sites contained flanking bases conforming to the widely distributed CGTGGGAA site of E(spl) Su(H) binding sites and none of the NB enhancers contained paired Su(H) sites typical of the E(spl) enhancers [25, 46]. Of the 13 PNS cis-regulatory regions examined, only four enhancers contained putative Su(H)-binding sites [sna and ato (ATGGGAA), brd (GTGGGAG)] and dpn (GTGGGAA). dpn also contained a pair of sites that conforms to the SPS configuration frequently found in Su(H) enhancers (CSB sequence: AATGTGAGAA AAAAACTTTCTCAC GATCACCTT, Su(H) sites in bold, Pbx site is ATCA). The lack of Su(H) sites in PNS enhancers was noted by Reeves and Posakony [23], who suggested that these enhancers are directly regulated by the proneural proteins but not activated in response to Notch-mediated lateral inhibitory signaling. Among the conserved sequences of E(spl) gene enhancers there is an average of 3.4 consensus Su(H) binding sites per enhancer, with most enhancers containing both types of sites, i.e., those with either A or G in the central position (data not shown).

We offer three insights with respect to Su(H) binding sites. First, although in vitro DNA-binding studies suggest there is a flexibility in the Su(H) binding site, like the bHLH E-box, comparative analysis shows that within any one the Su(H) sites there is no sequence flexibility. Except for the pair of Su(H) sites in the dpn PNS enhancer, none of the CNS or PNS sites contained a central A; less that a quarter of the E(spl) sites consisted of a central A, and all these were conserved across all species examined. In light of the high conservation in these regions the invariant core and flanking sequences are important for the unique Su(H) function at any particular site.

A second finding was the extensive conservation of bases flanking the consensus Su(H) sequence in the E(spl) complex genes (data not shown). For example, the c DT GTGGGAA ACACACGAC [Su(H) site bold] was present in HLHm3 and HLHm5 enhancer CSBs, and ACCGTGGGAA AC was conserved in HLHm3 and HLHmβ enhancers. The conservation of bases flanking the consensus Su(H) binding site suggests that the Su(H) site may be flanked by additional binding sites for co-operative or competitive factors, or else, that Su(H) contacts additional bases besides the consensus heptamer.

A third observation is that in most cases Su(H) binding sites are imbedded in larger CSBs, suggesting that CSB function is regulated by the integrated function of multiple TFs. For example the dpn NB enhancer Su(H) site is imbedded in a CSB of 24 bases, and the atonal PNS enhancer Su(H) site is imbedded in a CSB of 45 bases. In the E(spl) complex, CSB #6 of HLHmγ, consisting of 30 bases and CSB#13 of m8, consisting of 31 bases (each contains a GTGGGAA Su(H) site, a CACGAG element, conforming to a Hairy N-box consensus CACNAG [48, 49], and an AGGA Tramtrack (Ttk) DNA-binding core recognition sequence [50], but the order and context of these three sites is different for each enhancer). Although Su(H) binding sites were present in only a minority of NB and PNS enhancers, the conservation of core bases, as well as the complexity of their flanking conserved sequences points to a diversity of Su(H) function and interaction with other factors.

Neural specific cDTs that contain core DNA-binding sites for other known TFs

Two of these elements, one exclusively present in NB enhancers (CAGGA TA) and a second exclusively present in PNS enhancers (GTAGGA), contained consensus core AGGA DNA-binding sites for Ttk [50], a BTB domain TF that has been shown to regulate pair rule genes during segmentation and to repress neural cell fates [51–53]. Another site (CACCCCA), shared by both NB and PNS enhancers, conforms to the consensus binding site of IA-1 (ACCCCA), the vertebrate homolog of nerfin-1 [54]. Most of the cDTs of Table 2 do not contain sequences corresponding to consensus binding-sites of known regulators of NB expression. The fact that they are represented multiple times in NB CSB sequences suggests that they contain binding sites for unknown regulators of neurogenesis in Drosophila.

Neural-enriched cDTs

Neural enriched c DTs that are shared between multiple NB enhancers and also exhibit a low frequency in the sample of mesodermal enhancers examined in this study serve as a resource for understanding enhancer elements that may not have an exclusive neural function [see Additional file 1]. Notable here is the presence of CAGCTG bHLH DNA binding sites (all with flanking A, CC and TC) and Antennapedia class homeobox (Hox) core DNA binding site ATTA [40], as well as additional Ttk and Pbx/Extradenticle sites. Present in this list are portions of sequences conforming to Su(H) binding sites described above. Of particular interest in this table are sequences that are also enriched in the PNS (p); these sites may bind factors that play similar developmental roles in different tissues. For example, the presumptive Ttk site, AAAGGA (core sequence in bold) is highly enriched in segmental enhancers. Thus, some of these sites can be identified as targets of known TFs, but the identity of most are as yet unknown. These elements shared by multiple enhancers may be useful in identifying other enhancers driving expression in NBs.

cis-Decoder analysis reveals a complex sub-structure of enhancer CSBs

EvoPrint analysis revealed that all of the enhancer regions examined in this study contained multiple CSBs that were greater that 15 to 20 bases in length. The occurrence of overlapping DNA-binding sites for different TFs is currently the best explanation for the maintenance of intact CSB sequences across ~160 millions of years of collective species divergence. Our analysis has revealed that the sequence context, order and orientation of shared c DTs can differ between co-regulating enhancers.

Two examples are given here of the complex contextual appearance of c DTs that appear frequently in CNS and PNS enhancers (Figure 3). Each of the eight CSBs shown was nearly fully 'covered' by c DTs of the NB library (data not shown), suggesting that each contains multiple overlapping binding sites for a number of TFs. First, examination of the distribution of c DT GCTGCA reveals that it overlaps, by one and two bases, adjacent but different consensus bHLH sites in scrt CSB^#32, while in scrt CSB^#23 it overlaps a third consensus bHLH sequence by two bases. In the PNS enhancer char, in CSB^#17, GCTGCA overlaps a bHLH site, but in a different configuration (overlapping four bases) than found in the two CNS enhancers illustrated in Figure 3A. In amos CSB^#26, GCTGCA appears adjacent to a HOX site and does not overlap a bHLH site. Second, examination of the distribution of the c DT GGCACG reveals that it overlaps different consensus bHLH sites in scrt CSB^#32 and wor CSB^#106, overlapping the bHLH site in the former by one base and in the latter by four bases. GGCACG overlaps a CAGCTG bHLH-binding site in rho CSB^#18, but in a different configuration than the overlap with CAGCTG in the wor CSB. In the PNS enhancer scrt, GGCACG in CSB^#5 overlaps a Hairy site N-box (consensus CACNAG) [48, 49]. N-boxes were most common in E(spl) CSBs, but were also present in NB and PNS enhancer CSBs. In these two examples, and others we have examined, there is no consistent spatial constraints to the association of known TF-binding sites (i.e., bHLH-binding E-box sites) with novel c DTs; a picture that emerges is one of combinatorial complexity, in which known or novel c DTs are associated with each other in different contexts on different CSBs.

The use of cis-Decoder, FlyEnhancer and EvoPrinter to identify novel enhancers

We have used the information derived from cis-Decoder analysis of neural precursor cell enhancers to search for other genomic sequences with similar cis-regulatory properties. Having identified c DTs found multiple times among NB enhancers, we used the genomic search tool FlyEnhancer [55] to identify Drosophila melanogaster genomic sequences that contained clusters of the following c DTs (number in parenthesis is the total number of each c DT in our sample of six NB enhancers): GGCACG (6), GGAATC (4), TGACAG (6), TGGGGT (4), CAGCTG (14), TGATTT (9) CAAGTG (7), CATATTT (5), TGATCC (7) and CTAAGC (6). As a lower limit, a minimum of three CAGCTG bHLH sites was set for this search, because of the prevalence of this site in nerfin-1 and deadpan NB enhancers. Each sequence detected by this search was subjected to EvoPrinter analysis to determine the extent of its sequence conservation. Among the c DT clusters identified, our search identified a 5' region adjacent to the nervy gene ([] that contained three conserved CAGCTG sites as well five other sites identical to TGACAG, GGAATC, TGGGGT, GGCACG and CATATTT (see below). nervy, originally identified as a target of homeotic gene regulation, is expressed in a subset of early CNS NBs, as well as in PNS SOP cells [56]. Later studies have implicated nervy, along with cyclic adenosine monophosphate (cAMP)-dependent protein kinase (PKA) in antagonizing Sema-1a-PlexA-mediated axonal repulsion [57], and nervy has been shown to promote mechanosensory organ development by enhancing Notch signaling [58].

EvoPrinter analysis revealed that the cluster of neural precursor cell enhancer c DTs positioned 90 bp upstream from the nervy transcribed sequence contains highly conserved sequences (Figure 4A; chr2R:20,162,556-20,163,290). This region contains 10 CSBs that include six conserved E-boxes, three of which conform to the CAGCTG sequence that was prominent in nerfin-1 and deadpan promoters. To determine if this region functions as a neural precursor cell enhancer, we generated transformant lines containing the nervy CSB cluster linked to a minimal promoter/GFP reporter transgene (see methods section). Our analysis of the reporter expression driven by the nervy upstream fragment revealed a pattern indistinguishable from early nervy mRNA expression [56] (Figure 5). Specifically, we detected expression in a large subset of early delaminating NBs and in SOPs and secondary precursor cells of the PNS. Significantly, the nervy enhancer, unlike nerfin-1 and deadpan NB enhancers, activates reporter expression in then PNS and not just in early NBs.

A new c DT-library was generated combining the nervy enhancer CSBs and the NB and PNS enhancer CSBs used to generate the libraries described above. The new c DTs, along with the previously defined c DTs were aligned back to nervy CSBs (Figure 4b). Most c DTs were found only once in previously examined NB or PNS CSBs, but 21 cDTs appeared in our original analysis, described above, that did not include the nervy enhancer. The addition of this new enhancer to our analysis resulted in the discovery of a significant number of c DTs that had not been found previously. Three c DTs that were identified in the previous analysis, tCAGC TGc, cagCAGC TG and aaCAGC TG, contain bHLH DNA-binding sites (central bases of E-box in bold, flanking sequence are lower case). Aligning c DTs that are specific to the CNS or PNS may indicate sequences required to specifically drive expression in either the CNS or PNS.

Generation of EvoPrints and CSB-libraries

EvoPrinter analysis was performed as described [10, 61]. This analysis used EvoPrinterHD (please see Availability & requirements for more information) a second-generation EvoPrinter program that uses an enhanced-BLAT algorithm for increased resolution of conserved sequences [61]. Detailed instructions are provided at the EvoPrinter web site.

When possible, all twelve Drosophila species were used for the EvoPrint analysis, while species that exhibited sequencing gaps were excluded. CSBs within enhancers were curated from either an EvoPrint, which reveals bases conserved in all species, or a relaxed print (also known as an EvoDifference profile) that identifies base pairs that are conserved in all but one of the species. The collective evolutionary divergence for all of the EvoPrints was greater than 140 My and in most cases, when all twelve species were included in the analysis, EvoPrints represented over ~200 My of additive divergence. With the exception of two NB enhancers, scrt and wor, the size of each curated sequence was less than 1800 bases (Table 1). CSBs of 6 bp or longer were extracted from the EvoPrints using EvoPrint parser to generate CSB libraries. The number of CSBs in each enhancer, enhancer length, and relation of the enhancer with respect to the transcriptional start site is shown in Table 1. Lists of CSBs for each library are given at the cis-Decoder web site (please see Availability & requirements for more information).

Generation of cis-Decoder Tag libraries

In order to focus the analysis on neural-specific and neural-enriched c DTs, those cDTs that were found at high frequency in non-neural (mesodermal) enhancers were placed in a shared/common c DT-library. To identify neural specific c DT elements, the frequency of c DTs was scored against an out-group of mesodermal CSBs [11], and subsequently the common elements were removed. Prior to removal of mesodermal c DTs, the number of NB c DTs was 856, whereas after removal of shared c DTs, the number dropped to 272, indicating that the majority of c DTs shared by NB enhancers were also present in mesodermal enhancers.

Three c DT-libraries were generated by alignment of NB, PNS and E(spl) CSBs and are provided at the cis-Decoder web site (please see Availability & requirements for more information). The number of c DTs in each library was 272, 333 and 226 respectively. Of the 272 NB c DTs, less than half (120) aligned exclusively with NB CSBs, and did not align with PNS or E(spl) CSB sequences. Only 21% of the NB c DTs corresponded to PNS tags – in other words only 21% of the NB tags aligned two times or more with PNS CSBs.

Cytoscape analysis

We have adapted the biomolecular interaction network software Cytoscape [62] in order to display shared c DTs from different enhancer CSBs. The following data structure was used: node1 xx node2, where node1 is the name of an enhancer, xx refers to any designator and node2 is the c DT sequence. This data structure facilitates the display of enhancer identity and shared sequence elements in an interactive pattern. Cytoscape analysis requires elimination of the reverse complements of c DTs in order to avoid duplicate representation. To eliminate duplicate reverse-complement c DTs, we used the program c DT-Uncomplementer (please see Availability & requirements for more information). After removing duplicates, c DT-cataloger was used to name each node according to the enhancer aligning with that c DT.

Identification of novel neural precursor cell enhancers

To identify novel enhancers that direct gene expression in neural precursor cells, we curated c DTs that were shared by multiple identified NB enhancers and submitted them to the web-based genomic search tool FlyEnhancer [55], to discover other genomic regions with similar densities of c DTs. Candidate sequences that contained densities of c DTs alignments were subject to EvoPrinterHD analysis to determine the extent of conservation. Candidate enhancer regions were selected for enhancer/reporter studies.

Generation and analysis of nervy enhancer/reporter transformant lines

Genomic DNA containing the putative nervy enhancer (734 bp) was amplified by PCR using standard methods. Primers for the nervy upstream region including BglII and Nhe1 sites (bold) were respectively AGATCT CTAAAGCCCTCGATGTGCCC (5') and GCTAGC TCCGACCAGTCGTAAGTGGCG (3'). Fragments were gel purified and cloned into the pCRII-TOPO double promoter vector. Sequencing verified the fidelity of the PCR and cloning. After cutting with Bgl and Nhe1, gel purification was performed and fragments were cloned into pH-Stinger [63]. Details of our procedure are available upon request. The generation of transformant lines and embryo immunohistochemistry were carried out as described previously [64].