Analysis of the PHD finger proteome via protein domain microarrays
To define the histone binding preferences of the PHD finger proteome, we expressed and purified 123 annotated human PHD-containing domains as GST-tagged recombinant fusions from E. coli. The recombinant proteins consisted of either PHD fingers in isolation, or as tandem domains if a given PHD finger was located adjacent to another reader domain (e.g., one or more PHD fingers, Tudor, chromo and/or bromodomains) (Additional file 1: Table S1). These GST fusions were printed in duplicate on nitrocellulose-coated microarray slides and probed with biotinylated peptides that represented the N-termini of H3, H4, H2A or H2B (Fig. 1a and Additional file 2: Figure S1). As the majority of PHD readers thus far characterized are H3K4me0/3 readers [6], we included additional peptides (H3K4 as either mono-, di-, or trimethylated) to further determine any H3K4 methyl preference (Additional file 2: Figure S2 and Fig. 1b). As a control, we also probed these microarrays with an α-Tubulin peptide (a.a. 30–50) that would not be predicted to interact with PHD fingers (Additional file 2: Figure S1). As in Fig. 1a, b, 31 of the 123 PHD-containing fusions showed positive binding to the H3 N-terminus, with the majority of these interactions showing preference for trimethylated H3K4. In contrast, the H2A, H2B, H4, and tubulin peptides showed little to no positive interactions, suggesting that the PHD finger family broadly prefers the histone H3 tail (Additional file 2: Figure S1). We note that the absence of binding in these experiments does not rule out the possibility of PHD-finger:histone PTM recognition under different hybridization conditions. We also cannot exclude the possibility that some PHD fingers might not be functionally active on the microarrays (perhaps due to misfolding or the lack of an important adjacent region).
Based on the above, we were able to classify the [PHD–H3 tail] interactions into three groups, namely PHD fingers that: (1) bound specifically with methylated H3K4; (2) interacted only with unmethylated H3K4; or (3) bound without preference to the H3K4 methylated state. Many of the PHD fingers found to only bind H3K4 methylation have previously been described and include the well characterized domains from the ING and PHF protein families [6, 24]. The PHD finger of MLL5, a member of the MLL/KMT2 family [25,26,27,28], showed strong preference for H3K4me2 and H3K4me3. This finding adds to the relatively small number of MLL5-histone PTM observations reported to date [25]. Of the PHD fingers that bound to H3K4 methylation specifically, we observed that H3K4me3 or H3K4me2 were largely recognized equivalently and these domains did not detect H3K4me1 to the same degree (Additional file 2: Figure S2)—a result in agreement with other reports showing H3K4me binding occurs largely on higher methylated states [6]. Again, as with the H3K4me3 interacting PHDs, our findings for proteins such as KDM5A [third PHD finger (PHD3)] and KDM5B [third PHD finger (PHD3)] are consistent with their current classification as H3K4me3 binders [10, 11]. In contrast to H3K4me2/3 binding, a smaller number of PHD fingers [e.g., PHD1 from KDM5A and KDM5B, PHF21A, AIRE (PP), and TRIM66 (PB)] showed preference for the unmethylated H3K4 state (Fig. 1a, c). Furthermore, three PHD fingers we tested showed no preference between the H3K4me0 and H3K4me3 peptides: PHRF1 (RP), CHD5 (PCC), and KDM5B (PHD3) (Fig. 1). Collectively, these experiments identified 31 PHD-containing reader domains that showed positive interaction with the H3 N-terminus. While a majority of these reader domains preferentially interacted with H3K4me3 (18 out of 31) or H3K4me0 (10 out of 31), three showed no preference for the state of modification at K4. Importantly, these analyses uncovered several reader:histone interactions for poorly characterized PHDs (i.e., TRIM66, PHRF1, and SP140L): such insight could provide new avenues of investigation to these disease-relevant proteins [29,30,31,32].
Further characterization of H3-reading PHD fingers by peptide microarrays
To more comprehensively define the histone interactions of the 31 PHD readers identified from the domain microarray analyses, we probed each on an alternate microarray platform containing a library of 293 synthetic histone peptides with single or combinatorial PTMs [33] (Additional file 2: Figure S4 and Additional file 3: Table S2). All screening results can be found in Additional file 3: Table S2, but for brevity, findings pertaining to peptides that contain K4 and K9 modifications as well as neighboring phosphorylation sites that impinge on the observed binding by reader domains are displayed in the form of a normalized heatmap (Fig. 2). In general, the 31 PHD fingers were confirmed to associate with the H3 tail with the same H3K4 methyl preferences as in the domain microarrays (Fig. 2; Additional file 3: Table S2). Notably, the MLL5 PHD finger displays a strong preference for H3K4me3 over the un-, mono-, or di-methylated H3K4 peptides (Fig. 2), and further, over all other histone peptides on the array (Additional file 3: Table S2), consistent with results from the domain array (Fig. 1). Since CHD4, a protein annotated to recognize H3K9me3 [13, 14], was a positive binder in this assay, we compared its binding to H3K9me3 or H3K4 methyl peptides along with their unmodified counterparts at each position (K4me0/K9me0). The CHD4 (PPCC) fusion bound H3 N-terminal peptides more strongly when H3K4 was unmodified and dually acetylated at K9 and K18 versus when H3K4 is methylated in an identically acetylated context (Fig. 2); additionally, there was no difference in binding to the H3K4me0 peptide versus the H3K9me3 peptide. Interestingly, there also seems to be increased binding with CHD4 (PPCC) to the H3 K9ac peptide, potentially due to the “surface effect” (described in detail below). In addition, we confirmed the newly identified interactions observed with the domain microarrays for PHRF1 and TRIM66 (Fig. 2).
While findings between the domain microarrays and peptide microarrays largely agreed, there were some interesting differences. For example, PHRF1 (RP) showed no preference for the H3K4 methyl state on the domain array but strong preference for H3K4me0 on peptide microarray. Furthermore, KDM5B (PHD3), is reported to bind H3K4me3 [11], and showed such a preference on peptide microarrays but not on domain microarrays (Figs. 1 and 2). It should be noted that the comparison made here is between the H3K4me3 + K9ac + K18ac and the H3K4me0 + K9ac + K18ac peptides. Due to the limited binding, if any, observed by the non-acetylated versions of these peptides, it is difficult to assess the binding preference displayed by KDM5B (PHD3) with this comparison. Of note, certain PHD readers [i.e., DIDO1 and DPF2 (PPC2W2)] also showed some interaction with a number of H4 N-terminal peptides (Additional file 3: Table S2), consistent with published reports [33, 34].
During the course of this study, we observed that domain binding to H3 peptides tended to be enhanced when neighboring lysine residues were additionally acetylated (e.g., [K9ac + K18ac] for H3K4me0 or H3K4me readers) (Fig. 2). While at first approximation it might appear that these readers have an enhanced affinity for poly-acetylated states that neighbor H3K4, we note that solution-based peptide pulldown or AlphaScreen (see below) assays with several of these readers (i.e., KDM7A that binds H3K4me3 and KDM5B (PHD1) that reads H3K4me0) did not support this idea (Additional file 2: Figure S5 and Fig. 3i). We surmise that the enhanced binding caused by poly-acetylation is a property of the charged surface of the streptavidin-coated glass slides: when modified with bulky and neutral acetyl groups the highly charged histone N-terminal tail peptides become more accessible to reader domains.
Quantitative assessment of poorly defined PHD readers by the AlphaScreen dCypher assay
We next employed a highly sensitive proximity-based AlphaScreen histone peptide assay (dCypher®) to provide a third and orthogonal approach to analyzing the histone binding preferences for a subset of the 31 PHD proteins with respect to various histone tail PTMs. In this assay, biotinylated peptides are bound to streptavidin “donor” beads and the GST-tagged reader domains bound to Glutathione “acceptor” beads. The donor beads are excited by 680 nm light, releasing a singlet oxygen which causes light emission (520–570 nm) in proximal acceptor beads (within 200 nm); emission intensity is then correlated to binding strength [35]. For further examination with this more sensitive approach we chose the PHD fingers with positive binding data from the domain and peptide microarrays that were less characterized in the literature [i.e., MLL5, PHRF1 (RP), and TRIM66 (PB)], or those that displayed weak interactions on the domain and/or peptide microarrays [i.e., CHD4 (PPCC) and CHD5 (PPCC)]. Additionally, we examined several well characterized PHD–PTM interactors [DIDO1, KDM7A, and DPF2 (PPC2W2)] for positive controls and to provide a benchmark. Initial binding assays were conducted for each fusion protein using three peptides [H3 (1–20) with K4me0, H3K4me3 or H3K9me3] to determine the optimal reader domain concentration for full peptide library studies (Fig. 3a–h; Additional file 4: Table S3 and Additional file 5: Table S4). This is an important first step as signal often declined after query protein saturation (the ‘hook point’, caused by excess free query competing with bead bound).
Once the optimal protein concentration ranges for each of the eight readers were determined, we conducted the full dCypher peptide screen (293 histone peptides) (Fig. 3i; Additional file 4: Table S3). In agreement with our previous findings, the dCypher peptide assay demonstrated KDM7A to be a reader of H3K4me3. Furthermore, TRIM66 (PB) showed a preference for H3K4me0 and me1, consistent with findings from the peptide microarrays. For CHD4 (PPCC), the dCypher approach showed a clearer specificity for the H3K4me0 peptide over the methylated species in comparison to the peptide microarray results (Fig. 3i versus Fig. 2). In the case of CHD5 (PPCC), the peptide microarray indicated this protein to be insensitive to the methylation status at H3K4 (Fig. 2), but the dCypher assay identifies a preference for H3K4me0/1 (Fig. 3i), consistent with the domain microarray (Fig. 1a, c).
Consistent with the results from the domain and peptide microarrays, dCypher assays confirmed that the PHD finger of DIDO1 and MLL5 recognized the higher methyl states of K4 (me3/2), but also identified interaction of these domains with the H3K4me1 peptide. Interestingly, the four H3K4me0 readers analyzed—CHD4 (PPCC), DPF2 (PPC2W2), TRIM66 (PB), CHD5 (PPCC)—also showed the ability to bind to the peptides containing H3K9me3; this may be due to H3K4me0 in the H3K9me3 peptide. However, CHD4 (PPCC) and TRIM66 (PB) showed stronger interaction with H3K9me3 compared with the unmodified peptide over a range of protein concentration (Fig. 3d, f). We note that while the initial protein concentration optimizations in Fig. 3a–h were performed over a range of protein concentrations, the full peptide screen (Additional file 5: Table S4; summarized in panel Fig. 3i) was performed at a single protein concentration. When presented with the [H3K9me3 + S10p] peptide, four out of five of the H3K4me0 readers lose binding capacity, suggesting that these readers are sensitive towards the bulky negative phosphate group at S10; this phenomenon is also observed with the H3S10p peptide alone (Additional file 4: Table S3). To our knowledge, this would be the first report of a H3 tail binder outside the H3K9 position to be impacted by S10 phosphorylation, suggesting the phospho-methyl switch may function more broadly than previously thought. Intriguingly, PHRF1 (RP) binding specificity at 15 nM showed more limited interactions to H3K4me0 and H3K9me3 peptides (Fig. 3i), which will be discussed further below. Finally we note that the shift for poly-acetyl peptides seen in the peptide microarrays (reflecting a possible “surface effect”; Fig. 2) is not observed in the dCypher screen (Fig. 3i) which more closely resembles the peptide pulldown assays (Additional file 2: Figure S5).