Identification of M1BP, a novel zinc-finger protein that associates with Motif 1

Although Motif 1 has been subjected to extensive bioinformatic analyses (FitzGerald et al, 2006; Engstrom et al, 2007; Li et al, 2010; Ni et al, 2010), the identities of the proteins that bind this motif are not known. Hence, we carried out a series of biochemical experiments to track Motif 1 binding activity and identify the protein associated with Motif 1. Genomic footprinting with DNAse I detected the presence of protein in Drosophila embryos that is bound to Motif 1 residing in the core promoter region in either orientation (Figure 2A, ZAP3 and slmb) and upstream from the core promoter (Figure 2A, cib and smo). A gel-shift assay detected Motif 1 binding activity in nuclear extracts from Drosophila embryos (Figure 2B). This assay was used to track Motif 1 binding activity during its purification from embryo nuclear extracts (Supplementary Figure S1A). A protein with an apparent size of 55kDa that bound specifically to Motif 1 was purified by DNA affinity chromatography (Figure 2C), and protein–DNA crosslinking confirmed that a protein of this size interacted with Motif 1 (Supplementary Figure S1B). Mass spectrometry identified the protein as a novel zinc-finger protein encoded by the gene designated CG9797 at FlyBase (McQuilton et al, 2012). The DNA binding specificity of the protein and its zinc dependence were confirmed with recombinant protein isolated from E. coli (Supplementary Figure S1C). We named the protein M1BP (Motif 1 Binding Protein). It has a zinc-associated domain (ZAD) towards the N-terminus and five C2H2 zinc-fingers toward the C-terminus (Figure 2D). The zinc-fingers are likely to be involved in recognizing Motif 1. Zinc-associated domains have been proposed to be protein–protein interaction modules but their functions are unknown (Jauch et al, 2003).

Open in a separate window

Figure 2

Identification of M1BP. (A) DNAse I genomic footprinting analysis of ZAP3, slmb, cib, and smo in nuclei from Drosophila embryos. The vertical bar indicates the location of Motif 1. ‘+1’ and the arrows mark the start site and the direction of transcription. (B) Detection of Motif 1 binding activity in nuclear extract using a gel-shift assay. DNA containing the region from −311 to −274 of the smo promoter was used as wild-type (wt) probe and wild-type competitor, and DNA with three point mutations (shown in blue) in Motif 1 was used as mutant (mut) probe and mutant competitor. (C) Silver stained SDS–PAGE gel of eluates from DNA affinity columns containing either Motif 1 DNA (lane 1) or mutant Motif 1 DNA (lane 2). The arrow indicates M1BP. (D) Schematic of structural domains of M1BP predicted by SMART (Schultz et al, 1998). ZAD is a zinc-associated domain and Znf-C2H2 are zinc-fingers.

Motif 1 is a unique core promoter element that hardwires promoters for association of M1BP and Pol II

To investigate the relationships between M1BP, Motif 1, and paused Pol II, we raised antibody against recombinant M1BP (Supplementary Figure S2A) and performed ChIP-seq on Drosophila tissue culture cells. A total of 2000 M1BP peaks were detected on the genome (FDR <0.001), of which 1800 were within 300 base pairs of a transcription start site of 2187 genes (Figure 3A). M1BP correlates with the distribution of Motif 1 and is concentrated ~30 nucleotides upstream from the transcription start site (Figure 3B). This region is within the core promoter where the general transcription machinery assembles (Thomas and Chiang, 2006). The near one-to-one correspondence between Motif 1 and M1BP at promoters (Figure 3A) is unusual for Metazoa, since genome-wide studies indicate that the interactions of DNA binding proteins with their cognate sites are dependent on the chromatin environment and are constrained by DNA accessibility (Filion et al, 2010; Li et al, 2011). Hence, Motif 1 appears to hardwire promoters for association with M1BP.

Open in a separate window

Figure 3

Motif 1 is a unique core promoter element that hardwires promoters for association of M1BP and paused Pol II. (A) Correlations of Motif 1 and M1BP with active (top, Pol II promoters) or inactive (bottom, non-Pol II promoters) promoters. Active promoters were defined as described in Figure 1. The area of each circle and the overlapping regions are proportional to the number of promoters associated with the indicated feature. (B) Composite plots of the distribution of Motif 1 and M1BP. A total of 12124 promoters without a neighbouring TSS within 500bp were used. (C) Distribution of M1BP and Motif 1 on genes that alter expression upon depletion of M1BP. P-values were calculated by Fisher’s exact test. (D, E) Changes in M1BP and Pol II occupancy on different promoters resulting from M1BP depletion. Cells were treated with RNAi, and M1BP and Pol II occupancies were determined by ChIP. CG3625, CG4609, and CG8224 contain Motif 1 at their promoters, while Actin5C does not. LacZ RNAi served as a negative control for RNAi treatment. The intergenic region, which is ~63kb downstream of hsp70Bc, has no Motif 1 within 2kb and no annotated genes within 5kb. Error bars represent the SEM of biological triplicates.

Since M1BP associates almost exclusively with Pol II-bound promoters (Figure 3A, P<1.66E−321, Fisher’s exact test), we investigated if the association of Pol II with these genes was dependent on M1BP. M1BP was depleted from Drosophila cells with RNAi and changes in transcript levels were analysed with microarrays. Western blot analysis showed that M1BP RNAi treatment for 5 days selectively depleted M1BP (Supplementary Figure S2B). This RNAi treatment did not alter the rate of cell proliferation, whereas longer treatment caused the cells to arrest (Supplementary Figure S2C). Microarray analysis identified 1152 genes that changed expression by at least 1.5-fold (5% FDR in four RNAi experiments; Figure 3C). M1BP resides at a significant portion (66%, P<1 × E−100) of the genes that were downregulated by depletion of M1BP, indicating that the loss of M1BP was directly affecting expression of these genes. In contrast, no correlation (P>0.57) was observed between the genes that were upregulated by the depletion of M1BP and the presence of M1BP, indicating that loss of M1BP was probably indirectly affecting these genes. This is not unexpected since many M1BP target genes are involved in transcription regulation, including genes encoding transcriptional repressors such as E(z) and Su(var)3-3 (Muller et al, 2002; Rudolph et al, 2007) (Supplementary Table S1). The number of M1BP-associated genes that were downregulated by M1BP RNAi was probably underestimated due to the limited sensitivity of the microarrays. Indeed, quantitative PCR detected modest but reproducible decreases in expression of M1BP-associated genes that fell below the significance threshold of the microarrays (Supplementary Figure S2D).

To determine if M1BP directly affects Pol II at promoters, we monitored the effects of depleting M1BP on the association of M1BP and the Pol II subunit Rpb3 with several promoters. ChIP analysis revealed that depletion of M1BP with RNAi decreased both M1BP and Pol II at M1BP-associated promoters but had no effect on the M1BP-independent Actin 5C promoter (Figures 3D and E). Collectively, our results indicate that M1BP functions as a transcriptional activator that is involved in the recruitment of Pol II at promoters.

M1BP genes exhibit lower transcriptional plasticity than GAF genes

A striking outcome of our genomic analysis of M1BP is that it rarely cohabits promoters with GAF, although both factors are highly enriched at promoters with Pol II (Figure 4A). To begin to understand the biological significance of this separation, we examined the expression levels of these two groups of genes. The median expression level of M1BP genes was approximately two-fold higher than that of GAF genes (Figure 4B). More strikingly, the dynamic range of expression for M1BP genes was much narrower than GAF genes (F-test: 7.25E−11). Hence, mechanisms of regulation linked to GAF exhibit much greater transcriptional plasticity than those linked to M1BP.

Open in a separate window

Figure 4

Divergent transcriptional plasticity of M1BP, GAF, and TATA box genes. (A) Venn diagram showing the negative correlation between M1BP and GAF on active promoters (P<1.16E−34, Fisher’s exact test). (B) Box plot of expression profiles showing that GAF genes have a broader expression range than M1BP genes (P=7.25E−11, F-test). Published expression data were used (Gilchrist et al, 2008). Boxes depict 25th through 75th percentiles, and whiskers show 10th through 90th percentiles. (C, D) Comparison of the variation in gene expression levels in different tissues (C) and at different developmental stages (D) for genes associated with M1BP, GAF, or the TATA box. Microarray (Chintapalli et al, 2007) or RNA-seq (Graveley et al, 2010) data provided measures of expression. To focus on factors recruited directly by the DNA sequence, genes having both the binding motif and ChIP-seq or ChIP-chip peaks were selected as M1BP or GAF genes. Cumulative plots for all genes provide a reference. SD: standard deviation.

Transcriptional plasticity can also be interrogated on a single gene basis by analysing the expression variation displayed by each gene (Figures 4C and D). We determined the relative standard deviation in the expression levels of each gene among different tissues or at different developmental stages, and generated a cumulative plot for M1BP and GAF genes. We also included an analysis of TATA box genes because the TATA box was underrepresented among genes with paused Pol II. The cumulative plots show that M1BP genes exhibit the least plasticity among tissues and developmental stages while TATA box genes exhibit the greatest plasticity. GAF genes exhibit less expression variation than TATA box genes but more than M1BP genes. The difference between any two groups of genes is statistically significant (Mann–Whitney test, for any pair, P<0.0001), implying divergent regulatory mechanisms are employed at these three groups of genes.

TATA box and GAF genes tend to use distinct developmental transcription factors while M1BP genes function independently of these regulators

The difference in the transcriptional plasticity of M1BP, GAF, and the TATA box genes raised the possibility that these groups of genes are regulated by different sets of sequence-specific factors. To test that, we queried a list of 39 sequence-specific transcription factors that associate with experimentally defined DNA sequences (Kulakovskiy et al, 2009), and determined the distribution of these binding sites among M1BP, GAF, and TATA box promoters (Figure 5). GAF and TATA box promoters each exhibit correlations with largely distinct transcription factors. This suggests that gene-specific transcription factors are required for the regulation of these promoters, but that these two classes of promoters are regulated by different factors. In contrast, Motif 1 shows neutral or negative correlations with the majority of the binding sites for the 39 factors. Hence, M1BP might govern transcription of its target genes in a ‘stand-alone’ fashion that has less dependence on other sequence-specific factors than GAF and TATA box genes.

Open in a separate window

Figure 5

GAF, M1BP, and TATA box promoters employ distinct transcriptional regulators. Heat map showing the correlation of transcription factor binding sites with M1BP, GAF, and TATA box genes. Genes having both the binding motif and ChIP-seq or ChIP-chip peaks were selected as M1BP or GAF genes. Each row corresponds to 1 of 39 transcription factors with experimentally determined binding sites (Kulakovskiy et al, 2009). The frequency of a certain binding site occurring at promoters associated with M1BP, GAF, or the TATA box was calculated, and the significance of correlation was expressed as a P-value (Fisher’s exact test). Hierarchical clustering analysis was performed to group factors with similar correlation patterns.

Different gene ontologies are linked to M1BP, GAF, and the TATA box

To understand the biological rationale for employing different transcriptional controls, we identified the types of genes in each group. Gene ontology analysis showed that M1BP genes are enriched for ones that function in basic cellular processes, such as the cell cycle, metabolism, and the cytoskeleton (Figure 6A). Hence, these genes are likely to be active in most cells. In contrast, GAF genes are mostly involved in development and morphogenesis, and would require greater tissue and developmental-specific patterns of expression (Figure 6B). This observation is consistent with the proposed function of paused Pol II in synchronous induction of developmental genes (Levine, 2011). The TATA box is enriched at genes involved in responding to external stimuli or functioning in particular tissues (Figure 6C), such that TATA box genes show the most cell type-specific expression profiles. As a group, these genes may require fuller, cell type-specific repression to avoid being miss-expressed in particular cells.

Open in a separate window

Figure 6

(A) M1BP, (B) GAF, and (C) TATA box genes have different functions. Gene ontology terms were queried from GO_FAT categories and then subjected to functional annotation clustering analysis to group GO terms having similar biological meanings. The collapsed GO terms and the group enrichment scores (cutoff at 5) are shown.

Evidence for chromatin-dependent and -independent mechanisms of pausing

Our analysis of the nucleosome organization on M1BP and GAF genes suggests that promoter proximal pausing on M1BP genes is strongly linked to the +1 nucleosome, while pausing on GAF genes is largely independent of this nucleosome. Additional support for this conclusion is provided by our finding that the occupancy of the +1 nucleosome is higher for the efficiently paused M1BP genes than for the less efficiently paused M1BP genes. Since the leading edge of Pol II is located 15 to 20bp downstream from the front of the transcription bubble, Pol II on M1BP-associated genes appears to penetrate the upstream edge of the first nucleosome. This region is known to cause Pol II to pause and is due to contacts between histones and the DNA (Bondarenko et al, 2006; Hall et al, 2009; Bintu et al, 2012).

In contrast to M1BP genes, the distance between the paused Pol II and the +1 nucleosome on GAF genes spanned from 100 to 250bp upstream from the distal border of the first nucleosome, with over half of the Pol II not being near the nucleosome. Moreover, while the density of paused Pol II on GAF genes is higher than M1BP genes, the nucleosome occupancy is substantially lower. Thus, much of the Pol II that pauses on GAF genes does not involve a collision between Pol II and the nucleosome. Recent results show that nucleosomes in the vicinity of GAF-associated regions are highly dynamic and this could be the basis for the low nucleosome occupancy (Deal et al, 2010). Since GAF recruits NELF to promoter DNA (Li et al, 2013), this recruitment could facilitate binding of NELF to the elongation complex, thus assuring efficient pausing without a nucleosome collision.

Previous biochemical studies fully support the notion of chromatin-dependent and -independent promoter proximal pausing mechanisms. Reconstitution of a stably paused Pol II in the promoter proximal region of human hsp70 in nuclear extracts required the template to also be reconstituted into chromatin (Brown et al, 1996). In contrast, promoter proximally paused Pol II was reconstituted on the Drosophila hsp70 promoter with naked DNA templates (Benjamin and Gilmour, 1998).

Several studies have made conclusions about the relationship between chromatin structure and pausing, but none have taken into consideration the striking difference in the chromatin structures on M1BP and GAF genes. Since both groups of genes account for a substantial portion of paused genes, they significantly impact bioinformatic analyses. One study observed that as the pausing efficiency increased, the nucleosome occupancy decreased (Gilchrist et al, 2010). This observation is readily explained because GAF-associated genes dominate the group with the highest level of paused Pol II (Supplementary Figure S7A and B). Another study noted that RNA polymerase II pauses at H2Av-depleted genes (Weber et al, 2010). This too can be explained by the paucity of nucleosomes residing on the highly paused, GAF-associated genes (Supplementary Figure S7C). In the future, it will be important to separate M1BP and GAF genes, since the two classes of genes constitute significant portions of Pol II-associated promoters, yet are divergent in chromatin structure, in efficiency of pausing, and in the transcription factors that regulate their expression. Our results indicate that M1BP and GAF genes maintain their unique chromatin structures irrespective of the level of paused Pol II, suggesting that these unique features are likely dictated by M1BP and GAF rather than the paused Pol II (Supplementary Figure S7C).

M1BP and GAF mediate divergent regulatory mechanisms that can be linked to distinct transcriptional dynamics

We posit that the divergent regulatory mechanisms involving M1BP and GAF help to define the transcriptional dynamics of their target genes. The striking absence of nucleosomes at the promoter of M1BP genes could play a key role in maintaining the steady levels of expression of these genes throughout development because the promoter region remains accessible to the transcriptional machinery. As a group, M1BP genes exhibit a significantly narrower range of expression levels than GAF genes. The efficiency of promoter proximal pausing could provide the basis for this difference. Both types of promoters associate with NELF, indicating that pausing occurs. However, the efficiency of pausing on GAF genes exceeds that of M1BP genes. Reactivation of the paused Pol II at GAF genes is likely to involve additional sequence-specific transcription factors as is the case for the heat shock genes (Lis, 1998). Several DNA binding factors in mammals have been reported to recruit P-TEFb, the kinase that promotes productive elongation (Peterlin and Price, 2006). A similar mechanism could be used by those factors that we identified with binding sites enriched at GAF genes. This two-step process allows GAF genes to be poised for induction until the second transcription factor is activated by internal or external cues. In contrast, pausing on M1BP genes is less efficient and due to a transient impediment by the first nucleosome. One function of this pause could be to allow the elongation complex to transition to a state that is able to efficiently transcribe through nucleosomes that are present at significantly higher occupancy throughout the body of M1BP genes than GAF genes.

Gel-shift assay

Drosophila nuclear extracts were made as previously described (Biggin and Tjian, 1988), except that extracts were dialysed to a conductivity equal to that of 0.15M HEMG (25mM HEPES (pH 7.6), 0.1mM EDTA, 12.5mM MgCl₂, 10% glycerol, and 150mM KCl). Double-strand DNA probes for gel-shift assays were generated by ³²P-labelling one oligonucleotide with T4 polynucleotide kinase (New England Biolabs), and annealing three-fold molar excess of a complementary oligonucleotide in 10mM Tris-HCl (pH 7.5), 50mM NaCl, and 5mM EDTA. To prepare DNA competitors, an equal molar amount of each oligonucleotide was annealed.

A total of 50fmol of radiolabelled probe (0.5−1 × 10⁵c.p.m., Supplementary Table 3) was incubated with protein samples in 20μl of binding buffer (10mM HEPES, pH 7.6, 75mM KCl, 1mM EDTA, 5% glycerol, 1mM DTT, 250μg/ml BSA, and 25μg/ml HaeIII-cut E. coli DNA) at room temperature for 15min. If present, unlabelled competitors were included in 100-fold molar excess at the beginning of the binding reaction. Binding reactions contained 15μg embryo nuclear extract, 6μl of eluates from DNA affinity purification, or 250ng of recombinant M1BP. Some binding reactions contained 10μM ZnCl₂. Protein–DNA complexes were analysed on 6% polyacrylamide gels in 0.5 × TBE buffer.

Purification of M1BP from embryo nuclear extracts

Drosophila embryo nuclear extracts were fractionated on heparin-agarose and then Sephacryl S-300 (Austin and Biggin, 1996). Sephacryl S-300 column fractions were assayed by gel-shift, and fractions containing Motif 1 binding activity were pooled for DNA affinity chromatography. DNA affinity columns were prepared by coupling DNA to CNBr-activated sepharose (GE Healthcare) (Kadonaga and Tjian, 1986). Complementary oligonucleotides of the sequence 5′-GATCCAGTGTGACCG-3′ and 5′-GATCCGGTCACACTG-3′ were used for the Motif 1 affinity resin, while oligonucleotides of the sequence 5′-GATCCACTCTGACGG-3′ and 5′-GATCCCGTCAGAGTG-3′ were used for the control resin. DNA affinity chromatography was done as previously described (Kadonaga and Tjian, 1986), except that 15μg/ml of HaeIII-cut E. coli DNA was used as competitor.

Chromatin immunoprecipitation (ChIP)

ChIP assays were performed as previously described (Petesch and Lis, 2008), with the following modifications. S2R+ (DRSC) cells were grown in Schneider’s media with 10% FBS to a density of 5 million cells per ml before crosslinking. After crosslinking, extracts were sonicated at 4°C on the high-sonication setting of the Bioruptor (Diagenode) with three 15min intervals of 30s bursts, followed by 1min of inactivity to achieve an average fragment size of 150bp. For standard ChIP assays, 4μl of an antiserum (M1BP, pre-immune or Rpb3) was used per immunoprecipitation of 2.5 million cells. Quantitative PCR was used to determine percent input for each primer pair (list in Supplementary Table S5). For ChIP-seq analysis, sonicated DNA from 20 million cells was immunoprecipitated by either M1BP antisera or pre-immune sera. DNA precipitated with either antibody in two biologically independent experiments was combined and subjected to library preparation for SOLiD sequencing.

SOLiD sequencing

The DNA libraries were prepared as previously described (Motallebipour et al, 2009), with some modifications. Only one size selection was performed after ligation of barcoded adapters and PCR amplification. DNA was purified from an agarose gel with the MinElute Gel Extraction Kit (Qiagen). DNA samples were sequenced with 50bp read lengths on an ABI SOLiD Genome Sequencer by the Penn State Genomics Core Facility. SOLiD sequencing reads for ChIP with M1BP and pre-immune antisera were mapped to the Drosophila genome with the NGS toolbox on Galaxy (Blankenberg et al, 2010; Goecks et al, 2010). Mapping with Bowtie used default settings, except only uniquely aligned reads were kept (Langmead, 2010). A total of 6583283 reads from two M1BP ChIP experiments and 4244033 reads from the pre-immune control were mapped to unique sites on the genome. To define the centre of a peak, sequencing reads on both strands were shifted by 50bp, which was determined with tools on GeneTrack (Albert et al, 2008). M1BP peak calling was done with the cis-genome software package (Ji et al, 2008) using the two-sample model.

M1BP RNAi

RNAi experiments with S2R+ cells were done as previously described (Clemens et al, 2000), with the following modifications. dsRNA (30μg) was applied to 1 million cells in 1.5ml of FBS-free Schneider’s media. The cells were incubated for 1h at 24°C followed by addition of 1.5ml of Schneider’s media containing 20% FBS. The cells were incubated for an additional 5 days before expression analysis or ChIP assays. LacZ RNAi served as control. The primers for generating the templates for in vitro synthesis of RNAi are listed in Supplementary Table S4.

RNA expression analysis

Total RNA was isolated from M1BP or lacZ RNAi-treated S2R+ cells using the RNeasy Mini Kit (Qiagen). For transcriptome analysis, RNA samples from four individual RNAi experiments were analysed on Roche NimbleGen 4 × 72K multiplex arrays by the Penn State Genomics Core Facility. The ArrayStar 3 software package was used for microarray data analysis. To validate the microarray data, total RNAs were converted into cDNA libraries with SuperScript III Reverse Transcriptase (Invitrogen) in the presence of both oligo-dT and random hexamers. Quantitative PCR was used to examine the mRNA changes on individual genes with gene-specific primers (list in Supplementary Table S5).

For expression and purification of recombinant M1BP, in vitro protein–DNA crosslinking analysis, DNA motif analysis, and other bioinformatic analyses, see Supplementary Methods.

We thank Saurabh Sinha and Majid Kazemian at University of Illinois for helpful discussion on DNA motif analysis. We thank Debashis Ghosh for helpful discussion on normal mixture modelling of the size of nucleosomes. SOLiD sequencing and expression microarray experiments were done by the Penn State Genomics Core Facility. We also thank Frank Pugh, Tracy Nixon, and members of the Gilmour Lab for helpful insights on the manuscript. This research was supported by NIH grant GM47477.

Author contribution: JL designed and executed experiments, performed bioinformatic analyses, interpreted results, and wrote the manuscript. DSG designed experiments, interpreted results, and wrote the manuscript.