Establishment of versatile GAL4-dependent control of cell lineage-specific DamID
To establish DamID in WIDs, we employed a transgenic fly line carrying an inducible Dam or Dam-Pc fusion construct [5, 7]. Briefly, a full-length Hsp70 promoter is separated from the Dam or the Dam-Pc coding sequence by a cassette containing a transcriptional terminator flanked by FRT sites, which prevents transcription of Dam or Dam fusion proteins (Fig. 1a). Ubiquitous or cell-type-specific expression of a FLIP recombinase (FLP) mediates site-directed recombination of flanking FRT sites and removal of the terminator cassette, allowing expression of Dam or Dam fusion proteins [5, 7]. Indeed, only upon ubiquitous expression of a heat-shock-induced FLP, we observed the characteristic DNA smear formed by the methylation-dependent PCR products amplified from genomic DNA (gDNA) extracted from WIDs (Fig. 1b, Additional file 1: Fig. S1A). In addition, genotyping PCR confirmed the genomic elimination of the terminator cassette from the DamID constructs only after FLP induction (Additional file 1: Fig. S1B, B′). Combined, these observations indicate that the terminator cassette prevents transcription of Dam or Dam-Pc proteins in WIDs and that their expression can be efficiently induced by the presence of FLP.
We wanted to optimize this inducible DamID system for flexible cell-type-specific targeting by the rich repertoire of GAL4 driver lines available. We thus screened a number of UAS-FLP constructs from different sources for their ability to mediate efficient removal of the FRT-flanked transcriptional terminator cassette. Moreover, we specifically searched for a UAS-FLP line that did not show leaky expression in the absence of a GAL4 driver to prevent unspecific removal of the terminator cassette. Indeed, combining a UAS-FLP(JD2) transgene [21] with the inducible DamID system caused GAL4-independed removal of the terminator cassette (Additional file 1: Fig. S1B′). In contrast, a UAS-FLP(EXEL) transgene [22] did not induce removal of the terminator cassette in WIDs in the absence of a GAL4 driver (Additional file 1: Fig. S1B″). Only combining a DamID;UAS-FLP(EXEL) line with a rotund(rn)GAL4 driver caused partial removal of the terminator cassette in WIDs, consistent with the restricted expression of rnGAL4 in the central domain of the disc (Additional file 1: Fig. S1B″). This region was visualized using the G-trace system (Additional file 1: Fig. S1C) [23], which maps cell lineage history and real-time expression of a GAL4 driver of choice. To prove that Dam and Dam-Pc fusion proteins are really expressed in a cell-type-specific and GAL4/UAS-FLP(EXEL)-dependent manner, we sought to visualize expression of the Myc-tag encoded by both constructs [5, 7]. To this end, we induced removal of the terminator cassette by crossing a stable DamID;UAS-FLP(EXEL) line to a patched(ptc) GAL4 driver. ptcGAL4 is active in a row of cells anterior to the anterior–posterior compartment boundary in WIDs (Additional file 1: Fig. S1C′). However, most of the anterior compartment derives from cells that had expressed ptc earlier during development (Additional file 1: Fig. S1C′). Thus, the early removal of the terminator cassette during development under the control of ptcGAL4 is expected to cause expression of Myc-tagged Dam and Dam-Pc proteins in all cells of the anterior WID compartment. Notably, Dam and Dam-Pc proteins expressed under the control of the heat-shock promoter are present at undetectable levels if flies were kept at 21 °C. However, if boosted by a heat shock (see Experimental procedures), high expression of the Myc-tag could be detected specifically in the anterior compartment, if FLP expression was induced by ptcGAL4 (Fig. 1c, d). Importantly, Myc-tag expression was completely absent in the posterior compartment. Similarly, when DamID was induced using the posterior compartment driver engrailed(en)GAL4, boosted expression of the Myc-tag was exclusively detected in the posterior compartment (data not shown). These results indicate that UAS-FLP(EXEL) allows for the specific and flexible induction of cell-type-specific DamID in WIDs under the versatile control of cell-type-specific GAL4 drivers.
High expression levels of Dam are known to interfere with DamID specificity [24] and viability [5] (Additional file 1: Fig.S1D, E). Therefore, to understand whether expression of Dam by the low basal activity of the Hsp70 promoter at 21 °C is suitable for DamID profiling by maintaining wing disc cell viability, we monitored the occurrence of mitosis and apoptosis by immunodetection of phospho-H3S10 (pH3) and the activated effector caspase Dcp-1, respectively. No differences in mitotic or apoptotic activity between the anterior and posterior compartment could be observed when larvae were maintained at 21 °C and the terminator cassette was removed under the control of ptcGAL4/UAS-FLP(EXEL) (Fig. 1e–f″). Furthermore, immunodetection of developmental regulators such as Ptc itself (Fig. 1g–g″) or wingless (Wg) (Fig. 1h–h″) revealed appropriate patterning activity, and adult wings arising from these discs displayed only subtle alterations, such as extra vein tissues (Fig. 1i–i″). Combined these results suggest that inducible DamID profiling does not interfere with WID viability and developmental progression and thus presents an excellent option for cell-type-specific mapping of DNA binding sites in WIDs in vivo.
DamID and ChIP profiles of Polycomb-binding sites correlate
To provide a proof of principle that DamID sensitively detects differences in DNA binding activity in vivo, we wanted to compare Pc-binding profiles between wild-type (WT) and scrib1 tumourous wing discs (Fig. 2a, a′, Additional file 1: Fig. S2). We used scrib1 as a classic example of a polarity-deficient tumour suppressor gene [25] for which genetic interactions with and defects in Polycomb silencing have been reported [17].
We first induced ubiquitous expression of Dam and Dam-Pc in whole larvae using a FLP under the control of a heat-shock promoter (hsflp). We isolated and amplified methylated genomic DNA from WIDs of 10 WT or scrib1 third-instar larvae expressing either Dam alone or a Dam-Pc fusion protein (Fig. 2b) and generated NGS libraries using protocols devoid of additional PCR amplification steps to avoid PCR biases (see Experimental procedures).
The PCR-free NGS library preparation from 20 WIDs generated sequencing profiles with relatively low correlation coefficients across replicates (Additional file 1: Fig. S1F), likely due to high noise in profiles. However, assessment of multiple reproducibility parameters, such as correlation coefficients (Additional file 1: Fig. S1F), hierarchical clustering approaches using 94 DamID-Seq profiles (Additional file 1: Fig. S3A) and autocorrelation of neighbouring GATC sites at Lag 2 (Additional file 1: Fig. S3B) [26], revealed that technical replicates within genotypes are always more similar to each other than replicates across genotypes. Thus, PCR-free DamID-seq libraries can reproducibly reveal DNA binding profiles for small in vivo tissue samples.
While a subset of PcG target genes was previously reported to be upregulated in scrib1 WIDs [17], we found that total levels of H3K27 modifications were comparable between WT and scrib1 WIDs (Additional file 1: Fig. S4A). Our DamID-seq profiles confirmed that Pc-binding at individual sites (as defined by any genomic sequences flanked by Dam-targeted GATC motifs, also referred to as GATC fragments hereafter) was not globally altered in scrib1 (Fig. 2c). Indeed, when the genome-wide distribution of Pc-binding intensities at these sites was compared, the correlation between WT and scrib1 discs (Pearson’s correlation, r = 0.47, Fig. 2d) was only slightly lower than for biological replicates (Pearson’s correlation r = 0.51, Additional file 1: Fig.S1F). Importantly, broad binding of Pc to the Bithorax complex (BX-C) observed in Pc-DamID profiles could also be detected in Pc ChIP profiles from S2 cells, DmBG3 cells and whole embryo (Fig. 2e) [27, 28]. The Pearson’s correlation coefficients calculated for a comparison of the genome-wide Pc-binding intensities at individual GATC fragments in our Pc-DamID-seq and the corresponding GATC fragments in individual Pc ChIP-chip profiles ranged from 0.25 to 0.4 (Fig. 2e′). This finding is in agreement with previous comparisons of the two techniques [29,30,31] (for example Pearson’s correlation r = 0.37 in [30]). Our analysis thus indicates that DamID-seq is a suitable method to reveal DNA binding profiles of Polycomb in WID in vivo.
Polycomb-binding is altered only at a subset of target sites in scrib
1 wing discs
To understand whether alterations in Pc-binding at specific target genes may contribute to tumour phenotypes in scrib1 disc, we performed a three-state hidden Markov model (HMM) analysis of Pc-binding at individual GATC fragments to define ‘depleted’, ‘intermediate’ and ‘enriched’ Pc-binding states and analysed transitions between these states when comparing scrib1 to WT WIDs (see Experimental procedures, Additional file 1: Fig. S4B, Additional files 2: SF2 and 3: SF3) [26, 32,33,34,35]. As expected, we obtained three possible clusters that described the changes between the two profiles, namely (1) ‘no change’, which defined GATC fragments that did not vary in their Pc-binding classification between WT and scrib1 profiles, irrespective of whether these sites were bound by Pc in WT and scrib1 WIDs or not; (2) ‘loss’ defined GATC fragments, which were bound by Pc in WT but not in scrib1 discs; and (3) ‘gain’ defined GATC fragments, which were not bound by Pc in WT but in scrib1 WID samples. This analysis revealed that about 11% of ‘intermediate’ and ‘enriched’ Pc-binding states present in WT were lost in scrib1 WIDs and about 18% of scrib1 ‘intermediate’ and ‘enriched’ Pc-binding states were arising de novo (Fig. 2f). This suggests that Pc-binding dynamics are altered in a loci-specific manner in scrib1 discs.
To learn more about the effects that gain and loss of Pc-binding may have on transcriptional activity of Pc target genes in scrib1 discs, we related DamID Pc-binding sites to previously published WT and scrib1 WID transcriptome dataset [17]. To this end, we extracted the presumptive regulatory region spanning across the transcriptional start site (TSS)(− 2.5 kb ~ + 1 kb) of all genes differentially expressed in scrib1 (Fig. 3a) and recovered all included GATC fragments, hereafter referred to as transcription-associated GATC fragments (taGATCf) (Additional file 1: Fig. S4C). We compared changes in Pc-binding (gain, loss or no change) at an individual taGATCf with changes in the transcription levels of the associated differentially expressed gene (Fig. 3a). When comparing WT and scrib1 WIDs, many transcriptional changes at differentially expressed genes whose presumptive regulatory region contained at least one Pc-bound taGATCf occurred in the absence of changes to Pc-binding (data not shown). In numerous instances, however, a gain or loss of Pc-binding at any one taGATCf was linked to a gain or loss in transcript levels of the associated gene (Fig. 3b, Additional file 1: Fig. S4D, Additional file 4: Table S1). Surprisingly, we found that, in some cases, gain in Pc-binding could occur in the context of upregulated transcription (group I) and loss of Pc-binding could occur when transcription was downregulated (group IV) (Fig. 3b). While this unexpected behaviour appears to contradict the established role of Pc as promoter of gene silencing, we speculate that, instead, additional regulatory inputs at these target sites dominate target gene expression or, alternatively, that the bulk of transcriptional changes and changes in Pc-binding states may arise in two different cell populations.
Polycomb-binding at differentiation and tumour-associated targets is altered in scrib
1 discs
Our approach indicated the presence of multiple genes associated with transcriptional upregulation upon loss of Pc-binding (group III) and with transcriptional repression upon gain of Pc-binding (group II) in scrib1 (Fig. 3b), which is consistent with the described function of Pc in gene silencing [11,12,13,14,15]. We thus focused our subsequent analysis on these genes.
Surprisingly, group II included genes implicated in axon guidance, for example dsx, Lrt, caps, PlexB, pdm3, Toll-7 and Fas3 (Fig. 3b), possibly reflecting a failure to develop wing and thorax sensory neurons. While all group II genes gained Pc-binding for at least one taGATCf in scrib1 discs, we wanted to provide additional evidence for a role of PcG in regulating their expression. An analysis of transcript levels in WIDs mutant for the PRC1 components Psc/Su(z)2 [17] revealed that specifically dsx, Toll-7 and the neuronal Notch target pnt were upregulated upon loss of repressive PcG complex function (Fig. 3c, Additional file 1: Fig. S4E). This suggests that at least a subset of group II genes are bona fide Pc target genes.
Strikingly, however, group III was comprised of many genes implicated in promoting tumorigenic transformation, but which had not yet been identified as Pc target genes. Foremost among them are Ets21C [36, 37], Atf3 [38] and Ilp8 [39, 40]. As reported previously, we also found the tumour-associated genes upd3 [16, 17], SOCS36E [16, 41, 42] and chinmo [43, 44] to be Pc target genes (Fig. 3b, c). Ets21C, Atf3, Ilp8 and upd3 are known JNK target genes [37, 45], whereas chinmo and SOCS36E are important effectors of JAK/STAT signalling [41, 43]. Importantly, Pc-binding at all but one gene can also be identified in Pc ChIP profiles from S2 cells (Fig. 3d). A recent study [42] suggests that a large number of PRC1 targets involved in proliferation and signalling, like SOCS36E, may only acquire PRC1-binding but not PRC2-dependent H3K27me3 modifications. We thus specifically asked whether H3K27me3 and Pc may be found at Ets21C, Atf3 and Ilp8 loci in WT WIDs. To do so, we compared our data with H3K27me3 and Pc ChIP-seq profiles published by Loubiere et al. [42] (Fig. 4b). Like chinmo [42], Ets21C and Atf3 carry both H3K27me3 and Pc signatures (Fig. 3d), suggesting that Ets21C and Atf3 may be canonical PcG target genes utilizing PRC2-dependent H3K27me3 modifications for transcriptional regulation. On the other hand, like SOCS36E [42], upd3 only acquires PRC1-binding but lacks H3K27me3 (Fig. 3d). Interestingly, neither H3K27me3 nor Pc signatures from previous studies mapped to Ilp8 (Fig. 3d).
Despite these different behaviours with respect to H3K27me3 modifications, Ets21C, Atf3, Ilp8, SOCS36E, upd3 and chinmo are all upregulated upon loss of repressive PRC1 complex function in Psc/Su(z)2 mutant WIDs, demonstrating a role for Pc in silencing these tissue-stress-responsive genes in wild-type WIDs (Additional file 1: Fig. S3D). Thus, we identify at least three tumour-associated genes as novel bona fide Pc target genes and imply that the tumour-suppressive function of PcG proteins [16] integrates with regulation by the two important tumour-promoting pathways JNK and JAK/STAT.
Modulation of Polycomb-binding and target gene expression is associated with enrichment of specific regulatory elements
A question we wanted to address is how epigenetic mechanisms may intersect with changes in signalling environment of cells, and more specifically, how Pc-binding may be affected by cross-talk with transcription factors that act as effectors of signalling cascades activated during tumorigenesis. Thus, to advance our insight into how gain or loss of Pc-binding in scrib1 WIDs may be regulated, we analysed GATC fragments classified by the three-state HMM analysis to be ‘enriched’ in Pc-binding, for predicted transcription factor binding motifs or modENCODE-identified chromatin domains [27, 46] using i-cisTarget [47] (see Experimental procedures). In parallel, we performed an i-cisTarget on GATC fragments classified as gain or loss of ‘enriched’ Pc-binding states in scrib1 WIDs (Fig. 4a). As expected, Pc-bound GATC fragments in WT were enriched for PRC1 and PRC2 binding, as well as H3K27me3 and H3K9me3 modifications (Fig. 4a′, a″). In contrast, regulatory regions exhibiting dynamic Pc-binding transitions in scrib1 displayed high NES scores for RNA-mediated silencing machineries (Piwi, Ago2), transcriptional activation by histone acetylation (Nejire/CBP) or recruitment of RNAPol II (Fig. 4a′), all of which may cooperate with CTCF (Fig. 4a′) in insulator-dependent transcriptional regulation and spatial organization of chromatin [48,49,50,51,52]. Interestingly, histone modifications previously observed to occur at genes that are expressed, but importantly, at intermediate levels [53], were also detected at dynamic Pc-binding sites (Fig. 4a″). This suggests that Pc target genes, which experience altered Pc-binding in scrib1, may be subject to transcriptional modulation rather than absolute repression by Pc.
Next, we wondered whether tumour-associated transcripts upregulated upon loss of Pc-binding in scrib1 (group III, Fig. 3b) were characterized by a specific signature of regulatory elements. We thus repeated an i-cisTarget analysis for the presumptive regulatory region spanning the transcriptional start site (TSS)(− 2.5 kb ~ + 1 kb) of genes belonging to group III (Fig. 4b). Strikingly, AP-1 (Jra/Kay), Atf3, Cnc and Lola-binding motifs enriched in group III loci (Fig. 4b′, Additional file 5: Table S2) and align with the stress-dependent activation of chinmo, Atf3, Ets21C, Ilp8, upd3 and SOCS36E associated with high JNK and JAK/STAT activity during wound healing, regeneration and tumorigenesis [38, 44, 54,55,56,57].
We repeated an i-cisTarget analysis for group II genes, whose transcripts were downregulated upon gain of Pc-binding in scrib1 (Fig. 4b) to ask how Polycomb may be recruited to these sites. In agreement with the observation that group II genes were enriched for axon guidance targets, we found that transcription factors specifically expressed in neurons, such as Jumu and CG12299, were enriched in regulatory regions of group II (Fig. 4b′, Additional file 5: Table S2). Importantly, however, wing patterning regulators, such as the transcription factor Rn and the Dpp/TGF-β signalling effectors Med and Mad, were also enriched, confirming that wing differentiation is affected in a Polycomb-dependent manner in scrib1 WID (Fig. 4b′) [17]. These data, however, may indicate that transcriptional downregulation of genetic circuits involved in neuronal and wing disc patterning promotes binding of Pc to these target genes.
Based on our finding that GATC fragments gaining Pc-binding in scrib1 were enriched for CTCF (Fig. 4a′), we asked whether insulator elements locate to group II genes. Strikingly, 71% of group II genes contained Flybase-mapped class I and II insulator elements within their gene body. In contrast, insulator features mapped to only 19% of group III genes. This suggests that insulator-dependent modulation of Pc function or Pc-dependent modulation of insulator function may have important consequences for Pc-targeted gene expression in scrib1.
Polycomb-binding transitions fail in scrib
1 imaginal discs development
Previous studies indicate that abnormal differentiation in scrib1 discs may be linked to deregulation of Pc function [17]. To better characterize the differentiation state of scrib1 discs, we asked whether scrib1 Pc-DamID profiles correlated better with developmentally younger than with older WIDs, indicative of a failure to acquire PcG-regulated wing fates during development. We thus compared Pc-DamID profiles from WT and scrib1 late third-instar WIDs to Pc-DamID profiles from young WT WIDs isolated 2 days earlier in development (120 h AEL at 21 °C, early third instar) (Additional file 6: SF4). Strikingly, Pc-DamID profiles of scrib1 WIDs correlated more strongly with young WIDs than with older WIDs (Fig. 5a). Importantly, while the percentage of Pc-‘enriched’ GATC fragments gained in scrib1 and older WT WIDs stayed relatively constant if compared to young WIDs, the percentage of Pc-‘enriched’ GATC fragments that was lost was strongly reduced in scrib1 (Fig. 5b). Furthermore, target sites that normally gained Pc-binding during development failed to gain Pc-binding in scrib1 WIDs (Figs. 2f, 5c). Combined, this suggests that early Pc-bound sites stay bound as scrib1 discs progress through development and that sites which should gain Pc-binding in older scrib1 discs fail to do so. These results imply that a failure to execute Pc-dependent fate specification may contribute to the lack of wing disc differentiation in scrib1 discs.
A subsequent i-cisTarget analysis of young WID profiles revealed that Pc-‘enriched’ GATC fragments in young WIDs displayed PRC1 and PRC2-binding, confirming that they are canonical Pc target sites (Fig. 5d). GATC fragments that specifically lost ‘enriched’ Pc-binding in late development scored high for binding sites of the wing differentiation regulators nubbin (Nub) and scalloped (Sd) (Fig. 5d), reflecting the expansion of the central wing domain. GATC fragments that gained Pc-binding in late development were enriched in binding sites for Atf3 and Adf1 (Fig. 5d). Adf1 was recently identified to be critical for recruitment and tethering of Pc to target sites [58]. The enrichment of Atf3 motifs may suggest that Atf3 target genes are increasingly silenced as wing discs development progresses, which has indeed been observed for Atf3 expression [59]. This may also have important implications for the reduction in regenerative capacity previously attributed to Pc silencing of critical tissue-stress-responsive enhancers in late WIDs [60].
However, GATC fragments with dynamic Pc transitions during development were also enriched for CTCF and Su(Hw) insulator components, as well as for the histone demethylase Lsd1. Combined, these invoke earlier observations of insulator signatures at dynamic Pc-targeted sites (Fig. 4a′) and imply that Pc-binding dynamics at insulator elements, which are critical for organization of chromatin in the nucleus [48,49,50,51,52], are crucial to Pc function during differentiation. Intriguingly, a detailed analysis of our DamID profiles revealed that the Pc-bound GATC fragment sizes recovered from earlier developmental stages were larger than those recovered from late imaginal discs (Additional file 1: Fig. S5). Moreover, in scrib1 datasets, GATC fragment sizes occupied an intermediate distribution (Additional file 1: Fig. S5). The size range differences cannot be recapitulated by Dam profiles alone (data not shown). It may suggest that Pc-binding to genome regions characterized by different GATC motif frequencies is developmentally regulated and may reflect different distributions at promoters, introns or intergenic regions. However, it may also suggest a link between changes to Pc-binding and chromatin accessibility, where chromatin compaction during development may reduce the likelihood of distant GATC motifs to be methylated by Pc-Dam fusion proteins.