All validated silencers act as transcriptional enhancers in other cellular contexts 

Get Complete Project Material File(s) Now! »

Screening a library of elements for silencer activity in whole Drosophila embryos

We adapted our previously published enhancer-FACS-Seq technology, for highly parallel screening of elements for enhancer activity in Drosophila embryos (Gisselbrecht et al., 2013), into ‘silencer-FACS-Seq’ (sFS) technology, which enriches for elements that tissue-specifically silence reporter gene expression (see Methods). We generated a reporter vector, pSFSdist, which drives GFP expression under the control of a strong, ubiquitous enhancer and an element from a library of candidate silencers, in a genomically integrated context (Fig. 1a,b, Supplementary Fig. 5). The reporter construct is integrated in a single location in the haploid genome via phiC31 integrase (Gisselbrecht et al., 2013). Flies carrying single insertions from the reporter library are crossed to a strain in which expression of the exogenous marker protein CD2 is driven in a tissue or cell type of interest, and the resulting informative embryos dissociated to produce a single cell suspension. By sorting CD2+ cells in which GFP expression is reduced from the level driven by the strong ubiquitous enhancer in the absence of silencing activity, we enrich for cells containing silencers active in the cell type of interest, which can be recovered and identified by high-throughput sequencing.

Selection of elements to test for silencer activity in Drosophila embryos

We designed a first library of 576 genomic elements (Supplementary Table 1), chosen to represent a variety of chromatin states or enhancer activities, to test for silencer activity. Since the general features of silencers are unknown, we pursued a few different strategies to identify elements that might correspond to repressive elements. The availability of genome-wide chromatin immunoprecipitation data for well-characterized transcriptional corepressors (Celniker, 2009) provided one source of candidate silencers to test. Next, we reasoned that some CRMs might function as enhancers in one context and as silencers in other contexts, as two such bifunctional CRMs had been identified previously in Drosophila (Jiang et al., 1993; Stathopoulos and Levine, 2005). Therefore, since we were designing a library to screen for silencing activity in the mesoderm, we selected CRMs from the REDfly and CAD2 databases (Gallo et al., 2011; Bonn et al., 2012) that exhibited no or highly restricted mesodermal expression at embryonic stage 11. We furthermore filtered out elements associated with genes that show widespread mesodermal expression at this stage. Another potential source of such bifunctional elements that we selected was genomic regions associated with markers of both active and repressed chromatin structure in whole-mesoderm or whole-embryo experiments (Bonn et al., 2012; Rosenbloom et al., 2015; Thomas et al., 2011). All sequences identified from genome-wide ChIP methods were associated with nearby genes (see Methods) and filtered for absence of widespread mesodermal expression. Finally, we included three positive control sequences previously shown to have mesodermal silencing activity, and two types of negative controls: broadly active mesodermal enhancers, and length-matched regions of E. coli genomic sequence. Using sFS, we screened our library of genomic elements for silencer activity in embryonic mesoderm. Testing of this library yielded a readily detectable population of mesodermal cells in which green fluorescent protein (GFP) expression is reduced (Fig. 1b,c). Of the 576 sequence elements chosen for inclusion in this library, we detected 372 in cells derived from transgenic flies. We found 79 elements significantly enriched (see Methods) in the reduced-GFP cell population in either of two biological replicate screens (Fig. 2a, Supplementary Fig. 2, Supplementary Table 3).

Promoter competition in sFS-positive elements

The most enriched feature among elements that were identified as ‘hits’ in our silencer screen is overlap with regions surrounding the transcriptional start sites of genes, which likely reveals the presence of promoter competition. Promoter competition previously has been observed to restrict the enhancer-driven activity of reporter gene promoters (Ohtsuki et al., 1998). Consistent with this interpretation, three of the sequences that caused reduced GFP expression were broadly active mesodermal enhancers that we included in our library as negative controls; all three of these sequences contain promoters. Overall, the set of 41 ‘hits’ that overlapped promoter regions was significantly enriched for mapped instances of the TATA box (Zhu et al., 2009). While these are technical positives in our silencer screen, since our goal was to analyze CRMs that silence gene expression by other means, we omitted any identified ‘hits’ that overlapped promoter regions from further analysis, resulting in a filtered set of 262 detected library elements, of which 38 are genomic regions showing silencer activity in our assay. In the second step of our silencer identification strategy, we then generated pure reporter lines and individually tested each of the 38 non-promoter sFS ‘hits’ for silencer activity by fluorescence-activated cell sorting (FACS) (see Methods). This resulted in a final, high-confidence set of 15 validated, mesodermal silencers (Fig. 2b,c, supplementary figure 5), which we considered a rather low validation rate.

Characterization of silencing activity in the context of distinct enhancers

To confirm that the observed silencing activity does not represent an artifact of the FACS-based assay, we tested the ability of one of the newly discovered silencer elements to suppress activation by two different mesodermally restricted enhancers and visualized reporter gene expression in the resulting embryos. Both constructs showed silencing by the tested element relative to a negative control sequence (Fig. 2d–g). Interestingly, we observed different patterns of silencing in the context of these two mesodermal enhancers, which are active at different times in development; silencing activity was much weaker in the posterior germband at embryonic stage 12 (Fig. 2f,g), a pattern not observed in the earlier embryo (Fig. 2d,e). Thus, the repressive activity of silencers may exhibit complex spatiotemporal regulation similar to that of many enhancers.

All validated silencers act as transcriptional enhancers in other cellular contexts

We analyzed these 15 validated mesodermal silencers to determine which genomic features that we explicitly sampled in the design of our element library were predictive of silencer activity. Despite the inclusion in our library of ChIP peaks for two well-known transcriptional corepressors (Groucho, CtBP) (Mani-Telang and Arnosti, 2007; Orian et al., 2007) and for the repressive chromatin mark trimethylated Lys27 of histone H3 (H3K27me3) (Kharchenko et al., 2011), the only screened element type significantly enriched among active mesodermal silencers was non-mesodermal enhancers (Fig. 3a; P = 0.0147, Fisher’s exact test). In fact, all but two of the 15 high-confidence mesodermal silencers were previously reported to have enhancer activity. Testing of the remaining two silencers for enhancer activity revealed that they both also function as non-mesodermal enhancers in the embryo (Fig. 3b,c). Thus, our results suggest that most if not all mesodermal silencers are also enhancers in other cellular contexts.

TF compositional complexity and chromatin features of active silencers

Combining both libraries, we found out that the validated silencers were still somewhat enriched for overlap with HOT regions (as previously defined), with a complexity score in validated silencers very close to the previously set threshold of 8 (see methods), with a score of ~7.75 (AUROC = 0.732, P = 2.815e-05) compared to ~4.09 in sFS negative elements. We therefore ran a motif finding on these new elements (see Methods) as we previously did on our first set of validated elements, but no combination of motif was significantly detected. As this thesis is being written, an analysis (see Methods) on all validated silencers pooled together is running, and therefore could not be included here.
Regarding histone marks, the mild enrichment of the H3K27me3 mark that we observed previously appears to hold up. We found that whole embryo H3K27me3 ChIP signal is mildly enriched (area under the receiver operating characteristic curve [AUROC] ~ 0.61, P < 0.04, two-tailed Wilcoxon test (see Methods)) among validated mesodermal silencers, yet not significantly after correction for multiple hypothesis testing (adjusted P ~ 0.16) (see supplementary figure 10). Looking at other additional marks beyond H3K27me3, analysis and biclustering (see mMethods and Fig. 6) of whole-embryo and sorted-mesoderm ChIP-chip and ChIP-seq data for histone modifications (see Methods) revealed that our validated silencers were significantly depleted for the H3K36me1 mark, which is commonly associated with elongating Polymerase II (Ernst and Kellis, 2010), and that a subset of our silencers showed statistically significant enrichment for moderate levels of repressive histone marks (see Fig. 6).
Once again, these results suggest that H3K27me3 may be a potential feature of active, tissue-specific silencers, but it is not sufficient to distinguish silencers from other genomic elements. Therefore, it is obvious to say that further studies, including larger sets of active silencers, are needed to discover significant silencer-associated chromatin marks which together might enable accurate prediction of silencers and offer mechanistic insights on silencer activity.

READ  Hardware Design of a Tethered Robot Deployment and Recovery System 

Downstream analysis of the validated silencers

1. Enrichment of input data types. Each tested element belonged to one of nine categories, as described above (« Design of the candidate silencer library »). We compared the prevalence of each category among high-confidence validated silencers to its prevalence among non-TSS-overlapping windows confidently detected in either or both of the two experimental repetitions. Statistical significance of enrichment or depletion was calculated using the fisher.test function in R.
2. Enrichment of histone marks and TF ChIP signal. Mesoderm-specific histone modification ChIP-seq datasets were downloaded from the European Nucleotide Archive (Bonn et al., 2012) or from GEO (Gaertner et al., 2012). Reads mapping to each tested library element (i.e., each non-TSS-overlapping element confidently detected in either or both of the two experiments, each comprising three biological replicates) were counted and, where available, normalized by dividing by total H3 ChIP read count. Whole embryo histone modification ChIP-seq and ChIP-chip datasets were downloaded from modENCODE as bedfiles. ChIP-chip or ChIP-seq data for individual TFs, coactivators, and corepressors were assembled from modENCODE and other sources (see Supplementary Table 9 and 10). Mean signal over all tested library elements was calculated using bedtools. Enrichment or depletion was measured by calculating the area under the receiver-operator characteristic curve (AUROC), considering high-confidence validated silencers to be « true positives, » using the auROC function of the limma package in R. Statistical significance was assessed using the wilcox.test function. Where independent replicates were available, p-values were calculated separately and combined using Fisher’s method (Mosteller and Fisher, 1948). P-values were corrected for multiple hypothesis testing using the p.adjust function in R with the « fdr » method. TF complexity scores and HOT regions were downloaded from (Roy et al., 2010). Using the previously defined TF complexity score of 8.0 as the cutoff to define a HOT region, tested elements that overlap HOT regions were defined and the enrichment of high-confidence validated silencers in this population was calculated with the fisher.test function in R. To generate the heatmap, all histone mark data were Z-transformed, subtracting from each element in each column (i.e. each histone mark dataset) the mean of that column over all library elements and dividing that deviation by the standard deviation over the column; the < 0.5% of all Z-scores over 5 were truncated to 5. Truncated Z-scores were biclustered, using 1-Pearson’s R as a distance metric and Ward’s minimum variance method for clustering.
3. Motif enrichment. We curated a list of 93 repressive TF binding site motifs (see Supplementary Table 4). Gene lists were downloaded from FlyBase (download date: February 3, 2015) with the Molecular Function Gene Ontology term GO:0043565 (sequence-specific DNA binding) and either the Biological Process term GO:0000122 (negative regulation of transcription from RNA polymerase II promoter) or the Biological Process term GO:0045892 (negative regulation of transcription, DNA-templated). These were combined and intersected with the list of Drosophila TFs with experimentally determined DNA binding site motifs from CisBP (Weirauch et al., 2014), UniPROBE (Hume et al., 2015), and FlyFactorSurvey (Zhu et al., 2011). For TFs with multiple similar PWMs available, a single representative (learned from ChIP data, where available) was chosen; where a single TF (or its isoforms or heterodimers) gave two unalignable motifs, both were included. We then used the Lever algorithm (Warner et al., 2008) to search for combinations of 1, 2, or 3 motifs enriched among high-confidence validated silencers relative to matched random genomic background sets, as previously described (Gisselbrecht et al., 2013). We consider a motif or motif combination significantly enriched if it targets ≥50% of the foreground sequences and has AUROC ≥ 0.65 and FDR ≤ 0.1; all such combinations are shown in Fig. 4, for the validated elements from the first library we tested.
4. Testing effects of binding site mutations on silencer/enhancer activity. High-quality instances of the dve motif were chosen for site-directed mutagenesis in pDONR clones of example high-confidence validated silencers. Mutations were designed to avoid altering or introducing overlapping binding sites for known or suspected regulators; introduced mutations are shown in Supplementary Fig. 4. To assess effects on silencer activity, sequence-validated mutant silencers were LR-cloned into pSFSdist and introduced into flies; silencer activity of the resulting constructs was tested in parallel with their wild type counterparts as described above. To assess effects on enhancer activity, mutant silencers were LR-cloned into pEFS and introduced into flies. Wild type versions of the same elements were LR-cloned into pWattB-nlacZ (Busser et al., 2012) and introduced into flies; crossing of the resulting CRM:LacZ and CRMmut:GFP lines together produces embryos in which the wild type- and mutant-driven expression patterns can be compared directly, which were fixed, stained, and imaged as above.

Table of contents :

Acknowledgements
Introduction
Regulation of Transcription
Transcription factors and cis regulatory modules
Enhancers
Silencers
Predictions of CRMs
Predictions using motifs and conservation
Predictions from ChIP-seq
Histone marks and chromatin accessibility
Spatial proximity between genomic regions
Experimental identification of enhancers
Current challenges
Goals of the dissertation
References
Chapter 1: bifunctionality of CRMs
Author contributions
Abstract
Acknowledgments
Introduction
Initial library and first experiment
Screening a library of elements for silencer activity in whole Drosophila embryos .
Selection of elements to test for silencer activity in Drosophila embryos
Promoter competition in sFS-positive elements
Characterization of silencing activity in the context of distinct enhancers
All validated silencers act as transcriptional enhancers in other cellular contexts
Transcription factor compositional complexity at silencers
Chromatin features of active silencers
Second library and ongoing analyses
Second library
Results and validations
TF compositional complexity and chromatin features of active silencers
Current and future experiments and analyses
Genome editing: CRM knock-out
Hi-C: mesoderm specific interactions
Spatiotemporal activity of silencers
Discussion
Methods
Generation of reporter vector pSFSdist
Design of the candidate silencer libraries
Performing silencer-FACS-Seq experiments
Statistical analysis of sFS sequencing reads.
Validation of sFS results
Assessing CRM bifunctionality
Downstream analysis of the validated silencers
Cell sorting and fixation with formaldehyde
Guide-RNA and primer design
Cas9 and gRNA preparation and microinjection
In situ hybridization: probes primer design
References
Supplementary figures
Supplementary References:
Chapter 2: rare cell purification
Introduction
Cell panning approach
Preliminary results
pCD8 vector
Cell panning
Future directions
Methods
Generation of vector pCD8
Creation of dpp_VRR:CD8 vector and fly lines
Cell purification protocol – a work in progress
Supplementary figure
References
Conclusion and Future directions
Summary
Limitations
Future directions
Concluding remarks
References
Annexes and supplementary tables
Annex 1: full sequence of the pCD8 plasmid
Annex 2: Protocol for positive panning from Drosophila embryos
Annex 3 : gRNA list for bifunctional element knockout

GET THE COMPLETE PROJECT

Related Posts