Structural insights into human KAP1 PHD finger–bromodomain and its role in gene silencing

Structure of the PHD finger–bromodomain of KAP1

We determined the three-dimensional solution structure of the tandem PHD finger–bromodomain, encompassing residues 624–812 of human KAP1 (Fig. 1a) with a total of 3,726 NOE-derived distance and torsion-angle restraints obtained from heteronuclear multidimensional NMR³⁷. An ensemble of the 20 lowest-energy NMR structures is depicted in Figure 1b. Excluding the linker residues 674–695, the structures of the PHD finger (residues 623–673) and the bromodomain (residues 696–801) are well defined. The linker showed high mobility in solution as evidenced by reduced or negative backbone ¹H-¹⁵N heteronuclear NOEs compared to the structural regions of the protein (Fig. 1c).

The PHD finger–bromodomain forms an integrated structural unit, despite the flexible linker. The two individual modules have conserved structural folds, as seen in the corresponding domain family^8,32. The bromodomain adopts a left-handed four-helical bundle (helices α_Z, α_A, α_B and α_C), whereas the Cys4-His-Cys3 PHD finger contains an antiparallel β-sheet of two β-turn-β units that coordinate two zinc atoms in a cross-brace topology (Fig. 1d and Supplementary Fig. 1 online). In the tandem motif, the two domains are brought together to anchor around the first helix, α_Z, of the bromodomain (Fig. 1d). Unlike an amphipathic α_Z in most bromodomains, α_Z in the atypical KAP1 bromodomain consists of a contiguous stretch of amino acids with large hydrophobic and aromatic side chains in its C-terminal half (that is, Val707–Phe712; Supplementary Fig. 1). The hydrophobic C-terminal segment of α_Z in KAP1 is buried in the center of the tandem-domain structure. Packed on one side of α_Z are residues from the other three helices of the bromodomain bundle, and on the other side are Val630, Phe647, Leu653 and Leu668 of the PHD finger on a twisted antiparallel β-sheet (Fig. 1d,e). Thus, although both hydrophobic and electrostatic interactions mediate domain-domain association, α_Z serves as the hydrophobic core of the unified tandem-domain structure.

Structural integration of the PHD finger–bromodomain

The integrated structure of the KAP1 PHD finger–bromodomain is unique and different from the structure of the PHD finger–bromodomain of human BPTF¹⁴, or those of the tandem bromodomains of human transcriptional protein TAFII250 (ref. 9) and yeast remodeling protein Rsc4 (ref. 22). As seen by structural comparisons, the PHD finger and bromodomain in BPTF do not contact each other, but they are connected by a rigid three-turn α-helix that spans about 18 Å (Fig. 2a, middle). In TAFII250, the corresponding α_Z in the second bromodomain is a typical amphipathic helix packed to the hydrophobic core of the helical bundle on one side and exposed to the solvent on the other (Fig. 2a, right). The two bromodomains are connected by a short loop followed by a two-turn α-helix, and they interact between α_B of the first bromodomain and α_A of the second. The domain-domain association is reinforced by interactions between the ZA loops of the two domains. In BPTF or TAFII250, the acetyllysine binding pocket located between the ZA and BC loops is accessible to the solvent, which indicates that the bromodomain and the PHD finger have independent functions in protein-protein interactions even in the tandem motif^9,14. However, these structural features contrast sharply with the distinct integrated nature of the structure of the KAP1 tandem PHD finger–bromodomain (Fig. 2a, left).

To characterize the structural integration of the PHD finger–bromodomain, we engineered an enterokinase cleavage site into the interdomain linker by mutating the sequence LKEEDGSLSL (residues 675–684) to LKDDDDKLSL. The two-dimensional ¹H-¹⁵N HSQC (heteronuclear single quantum correlation) spectrum of the mutated KAP1 superimposes nearly perfectly to that of the wild type, except for the resonances of a few residues at the cleavage site (Fig. 2b, left). Although yielding only fragments corresponding to the PHD finger and the bromodomain (Supplementary Fig. 2a,b online), enterokinase treatment of mutated KAP1 resulted in both loss and perturbations of many protein peaks in the NMR spectra (Fig. 2b, middle). Notably, the remaining NMR signals match those of the PHD finger alone (Fig. 2b, right), with the exception of a few signals of different terminal residues in the corresponding protein constructs. Thus, the lost signals are those of the bromodomain and could possibly be due to nonspecific enterokinase cleavage at several sites in the protein or aggregation of the cleaved protein that forms a species of large molecular weight. The former is not a likely cause, as indicated by our SDS-PAGE analysis of the protein upon enterokinase treatment (Supplementary Fig. 2b); if it were, the cleaved KAP1 bromodomain fragments would yield strong NMR signals in the HSQC spectrum. Given the hydrophobic nature of α_Z, the latter is more likely. Indeed, when expressed alone in Escherichia coli, the bromodomain was unstable in solution (data not shown). On the other hand, these results show that the stable PHD-finger fold is established via coordination of two zinc atoms. This is supported by the NMR observation that only the residues located at the domain interface show major chemical shift differences between the PHD finger alone and the tandem-domain motif (Fig. 2c and Supplementary Fig. 2c). Taken together, these results argue that the tandem PHD finger–bromo-domain of KAP1 is one integrated structural unit and probably functions cooperatively.

Structural features of the bromodomain and the PHD finger of KAP1

The bromodomain functions as an acetylly-sine binding domain in the epigenetic control of gene transcription¹⁰. The acetyllysine of a bromodomain target protein is bound in a pocket between the ZA and BC loops, wherein the acetyl group is typically recognized through hydrogen-bonding interactions of the amide nitrogen and carbonyl oxygen of the ace tyl group with three highly conserved residues in the domain family corresponding to Tyr760 in the ZA loop and Tyr802 and Asn803 in the BC loop of the p300/CBP-associated factor (PCAF) bromodomain (Fig. 3a). Notably, although the four-helical bundle fold is conserved, the ZA and BC loops—except for the three conserved acetyllysine binding residues—vary in length and amino acid composition, which contributes to distinct ligand binding specificity by different bromodomains. However, in the KAP1 bromodomain, these three conserved residues are substituted with glutamine, leucine and threo-nine, respectively, which have completely different functionality (Fig. 3a). Moreover, because of the direct packing of the PHD finger, the ZA loop is positioned in a different orientation with respect to the BC loop compared to that of the PCAF bromodomain. Consequently, the characteristic pocket between the ZA and BC loop in the canonical bromodomain of PCAF, CBP or BPTF is distorted in the atypical bromodomain of KAP1 (Fig. 3b and Supplementary Fig. 3a,b online). These distinct structural features, however, are consistent with our observation that the tandem PHD finger–bromodomain of KAP1 does not bind to lysine-acetylated peptides derived from histones H3 or H4, as shown by NMR binding.

The PHD finger in KAP1 is also functionally different from that of BPTF. The latter recognizes methylated lysine 4 of histone H3 (H3K4) by a cage of four aromatic side chains located on the surface of the β-sheet of the PHD finger¹⁴ (Fig. 3c, right). This aromatic cage is absent in the corresponding site in KAP1; instead, a four-aromatic-aminoacid cluster in KAP1 lies on the opposite side of the β-sheet packing against α_Z of the bromodomain (Fig. 3c, left and middle). However, the KAP1 PHD finger, either alone or in tandem with the bromodomain, does not bind to lysine-methylated or nonmodified peptides derived from methylation sites in histones, such as H3K4, H3K9 and H3K27, as shown in the NMR binding study (data not shown). Moreover, in a dot blot assay we also ruled out possible KAP1 PHD-finger binding to phosphoinositides, binding of which has been reported for inhibitor of growth family, member 2 (ING2) and other PHD fingers³⁸. Collectively, these results show that the PHD finger and the bromodomain of KAP1 do not have typical activities of methyllysine and acetyllysine binding, respectively.

SUMOylation and binding of the KAP1 tandem domain to UBC9

We recently reported that the co-repressive activity of KAP1 is dependent on lysine SUMOylation, which is unique within the TIF1 protein family (Supplementary Fig. 4a,b online), and that the PHD finger of KAP1 functions as a SUMO E3 ligase for KAP1 SUMOylation³³. Of six SUMOylation sites mapped in KAP1, four reside within the PHD finger–bromodomain unit (Fig. 4) and two in a region N-terminal to the PHD finger³³. Using an in vitro SUMOylation assay and lysine-to-arginine mutations of the PHD finger–bromodomain, we showed that two major SUMOylation sites are located at Lys779 in the BC loop and Lys804 in the loop following α_C of the bromodomain (Supplementary Fig. 4c, lanes 6 and 7), and two minor sites reside at Lys676 in the interdomain linker and Lys750 in the AB loop of the bromodomain (Supplementary Fig. 4c, lanes 4 and 5).

To characterize the functional role of the PHD finger–bromodomain in KAP1 SUMOylation, we assessed the tandem-domain interactions with the SUMO E2 ligase UBC9 using NMR. As shown in the ¹H-¹⁵N HSQC spectra of the PHD finger–bromodomain, a substantial number of KAP1 backbone amide resonances underwent line broadening upon addition of UBC9 fused to glutathione S-transferase (GST-UBC9) but not GST alone, indicative of direct protein-protein interactions (Supplementary Fig. 5a online). Notably, the residues that showed major chemical shift perturbations are localized at the tandem-domain interface (Fig. 4a), consistent with the NMR-mapped location of UBC9 binding sites on the PHD finger alone³³ (Supplementary Fig. 5b,c). Although UBC9 recognizes the SUMOylation consensus sequence φKxE/D^39,40, we did not observe major perturbations of residues at any of the four SUMOylation sites. Localization of UBC9 binding at the domain-domain interface suggests that tandem-domain association may be coupled to UBC9 binding and facilitated by a flexible interdomain linker. The observation of different KAP1 residues perturbed upon UBC9 binding of the PHD finger alone or the tandem PHD finger–bromodomain (Supplementary Fig. 5b,c) indicates that the two domains would become rearranged to accommodate UBC9–PHD finger–bromodomain association. Viewed from the top of the tandem-domain structure, the four SUMOylation sites (Fig. 4b, magenta spheres) are spatially located on the periphery of the globular structural fold, making the PHD finger–bound E2 conjugating enzyme UBC9 accessible to all four sites. Taken together, these results confirm a coherent nature of the PHD finger–bromodomain association in the tandem unit and their interactions with UBC9.

Functional role of the structural integrity of the tandem domain

To assess the functional importance of its structural integration, we performed in vitro SUMOylation of the tandem PHD finger–bromodomain of KAP1. As shown in Figure 4c, either the PHD finger or the bromodomain alone was a poor substrate for SUMOylation (lanes 2 and 3), whereas the tandem-domain motif showed at least ten-fold stronger SUMOylation (lane 4). The increased level of SUMOylation cannot be ascribed to a possible difference in protein stability, as the PHD finger alone is as stable as the tandem domain. As such, these results strongly suggest that the two domains cooperate. The marked increase in SUMOylation is due to the stimulatory effect of the PHD finger on the bromodomain as the major SUMO conjugation sites, Lys779 and Lys804, are located in the bromodomain (Fig. 4c). Disruption of the PHD finger fold by C651A mutation due to the ablation of zinc coordination resulted in a complete loss of the PHD finger–mediated stimulatory effect on bromodomain SUMOylation (lane 6), which is consistent with the inability of the PHD C651A mutant to bind to UBC9 (ref. 33).

To further characterize the functional role of the interdomain association, we made alanine mutations at three leucine residues, Leu653, Leu668 and Leu709, located at the interdomain interface, and tested the effects of these mutations on in vitro KAP1 SUMOylation (Fig. 4c) and KAP1 transcriptional repression in a transient gene reporter assay (Fig. 4d). Although these conservative amino acid substitutions should not cause major structural changes, alanine mutation of Leu653 in the PHD finger or Leu709 in the bromodomain showed detrimental effects on KAP1 SUMOylation to a level similar to that of the C651A mutation (Fig. 4c, lanes 7 and 9 versus lane 6), or about 65% and 40% of the C651A mutation effects on in vivo transcription-repression activity, respectively (Fig. 4d, lanes 5 and 7 versus lane 4). It is important to note that, when compared to the mutation effects on in vitro SUMOylation of the PHD finger–bromodomain, the less profound reduction observed in in vivo repression activity in these measurements is probably due to the use of full-length KAP1, which contains two additional SUMOylation sites located N-terminal to the PHD finger, and the fact that SUMOylation of these sites occurs in the absence of or upon structural perturbation of the PHD-bromodomain module³³. Indeed, only when all six SUMOylation sites in KAP1 were mutated (the K6R mutant) did we observe a nearly complete loss of repression activity (Fig. 4d, lane 3), indicating that SUMO modification has a major role in KAP1 repression function.

On the basis of the NMR structure and the UBC9 binding data as described above, we argue that the tandem-domain interactions and their UBC9 binding are coupled and that amino acid mutations at the interdomain interface can affect PHD finger–bromodomain association and/or their binding to UBC9. Thus, the effects of single, double or triple leucine-to-alanine mutations on KAP1 SUMOylation probably result from a net balance of these two related factors. The effects of single leucine-to-alanine mutations can indicate their relative functional importance in KAP1 SUMOylation through contributions to domain-domain interactions and UBC9 binding (such as Leu653 and Leu709 versus Leu668), whereas a double or triple mutant does not necessarily produce an additive effect when compared to the corresponding single or double mutant (Fig. 4c). Along the same line of reasoning, we predicted and confirmed that deletion of the linker would have a major impact on SUMOylation (Fig. 4c, lane 4 versus 14). As the linker contains only one minor SUMOylation site at Lys676 (Supplementary Fig. 4c, lane 4), a nearly complete loss of SUMOylation of the linker deletion mutant does not result from elimination of a SUMOylation site, but rather from a major structural perturbation of the tandem-domain motif (data not shown). Taken together, these results argue that the tandem PHD finger–bromodomain is an integrated structural and functional unit that is important for KAP1 SUMOylation and co-repression activity.

Protein expression and purification

The tandem PHD finger and bromodomain of human KAP1 (residues 624–812) was expressed in E. coli BL21(DE3) cells in the pET15b vector (Novagen). We prepared isotope-labeled proteins from cells grown on a minimal medium containing ¹⁵NH₄Cl, with or without ¹³C₆-glucose, in H₂O or 75% ²H₂O. The protein was purified by nickel-IDA affinity chromatography followed by thrombin cleavage to remove an N-terminal polyhistidine tag. GST-fused bromodomain–PHD finger of KAP1 were expressed in E. coli in the pGEX4T3 vector (Pharmacia), and purified with glutathione-Sepharose beads. NMR spectra of the purified KAP1 proteins were acquired to ensure proper protein folding.

In vitro SUMOylation

In vitro SUMOylation reactions were carried out at 30 °C for 3 h. Briefly, 2 μg of GST-fused proteins of KAP1 were loaded onto glutathione beads and incubated in the presence of 500 ng of UBC9, 700 ng of human E1 (BostonBiochem) and 2 μg of ³²P-labeled SUMO1 in 20 μl reaction mix with buffer containing ATP (LAE Biotech International). After incubation, beads were washed five times with 1 ml of BLB500 buffer, proteins were eluted in 2× Laemmli buffer, resolved by SDS-PAGE and visualized by Coomassie blue staining and subsequent autoradiography. ³²P-labeled SUMO1 was prepared from GST–PKA-SUMO1 as described⁴⁵. We performed in vitro SUMOylation reactions for the tandem PHD finger–bromodomain motifs from three other TIF1 family members and the tandem bromodomain–PHD finger from p300/CBP.

Gene repression assay

U2OS cells were cultured in DMEM supplemented with 10% (v/v) FBS. Transfections were performed on six-well plates in triplicate using FuGene6 (Roche), according to the manufacturer’s instructions. Cells were lysed 36 h after transfection in Tween 20 buffer⁴⁶. Luciferase Assay System (Promega) and β-galactosidase Assay Reagent (Pierce) were used to measure respective enzymatic activities and subsequent normalization. Nuclei were pelleted, lysed in Triton X-100 buffer⁴⁶ and used for detection of transfected proteins by western blotting. Lentiviral vector–based KAP1 gene knockdown was performed as described previously⁴⁷.

NMR spectroscopy

NMR samples contained KAP1 PHD finger–bromodomain of ~0.5 mM in 100 mM phosphate buffer of pH 6.5 containing 200 mM sodium chloride in H₂O/²H₂O (9/1) or ²H₂O. All NMR spectra were acquired at 30 °C on a Bruker 500 MHz, 600 MHz, 800 MHz or 900 MHz NMR spectrometer. The ¹H, ¹³C and ¹⁵N resonances of the protein were assigned by triple-resonance NMR spectra collected with a uniformly ¹³C- and ¹⁵N-labeled and 75% deuterated protein³⁷. The distance restraints were obtained from three-dimensional ¹³C- or ¹⁵N-NOESY spectra. Slowly exchanging amides, identified in two-dimensional ¹⁵N-HSQC spectra recorded after a H₂O buffer was changed to a ²H₂O buffer, were used with structures calculated with only NOE distance restraints to generate hydrogen-bond restraints for final structure calculations.

Structure calculations

Structures of the KAP1 PHD finger–bromodomain were calculated using a distance geometry-simulated annealing protocol using the X-PLOR program⁴⁸. Initial protein structure calculations were performed with manually assigned NOE-derived distance restraints. Hydrogen-bond distance restraints, generated from the H/D exchange data, were added at a later stage of structure calculations for residues with characteristic NOE patterns. ARIA-assigned⁴⁹ distance restraints agree with structures calculated using only the manually determined NOE distance restraints. For the final 20 lowest-energy NMR structures, no distance or torsional angle restraint was violated by more than 0.5 Å or 5°, respectively (Table 1). Ramachandran plot analysis of the final structures, excluding the linker and the ZA loop, with Procheck-NMR⁵⁰ showed that 77.4 ± 2.9%, 18.3 ± 2.6%, 3.0 ± 1.5% and 1.4 ± 1.2% of the non-glycine and non-proline residues were in the most favorable, additionally allowed, generously allowed and disallowed regions, respectively. The corresponding values for the residues in the secondary-structure regions of the PHD finger–bromodomain were 88.8 ± 2.7%, 10.7 ± 2.5%, 0.4 ± 0.7% and 0.2 ± 0.5%, respectively.