Mechanistic and structural studies of KDM‐catalysed demethylation of histone 1 isotype 4 at lysine 26

N‐Methylation of lysyl residues is widely observed on histone proteins. Using isolated enzymes, we report mechanistic and structural studies on histone lysine demethylase (KDM)‐catalysed demethylation of N ε‐methylated lysine 26 on histone 1 isotype 4 (H1.4). The results reveal that methylated H1.4K26 is a substrate for all members of the KDM4 subfamily and that KDM4A‐catalysed demethylation of H1.4K26me3 peptide is similarly efficient to that of H3K9me3. Crystallographic studies of an H1.4K26me3:KDM4A complex reveal a conserved binding geometry to that of H3K9me3. In the light of the high activity of the KDM4s on this mark, our results suggest JmjC KDM‐catalysed demethylation of H1.4K26 may be as prevalent as demethylation on the H3 tail and warrants further investigation in cells.

Widespread N-methylation of both arginyl and lysyl residues occurs on the tails of core histones (histones H2A, H2B, H3 and H4), where it has roles in regulating gene expression [1]. Lysyl residues can undergo successive methylations on the N e -nitrogen to give three methylation states (mono-N e -methyllysine, di-N emethyllysine and tri-N e -methyllysine respectively), as catalysed by S-adenosylmethionine-dependent methyltransferases (KMTs) [2,3]. In many cases, KMTs catalyse methylation at multiple sites to give different methylation states, on both histones and non-histone proteins [1,4]. Enzyme-catalysed demethylation of methylated lysyl residues also occurs at many sites on histones H3 and H4 [1]. Thus, at least at certain sites, lysyl methylation is dynamically regulated by the interplay between the two enzyme-catalysed processes.
Demethylation is catalysed by two families of histone lysine demethylases (KDMs) [5]. Members of the larger family of KDMs, JmjC KDMs, catalyse methyl group oxidation coupled to oxygen-dependent oxidative decarboxylation of 2-oxoglutarate (2OG), forming succinate and carbon dioxide (Fig. 1A) [5]. The available evidence implies that the resultant hemiaminal is normally unstable, fragmenting to give the demethylated lysyl residue and formaldehyde [6]. Dysregulation of JmjC KDM family members is implicated in many diseases, including developmental disorders and many cancers. The JmjC KDMs are thus the focus of drug development programmes [7], although their physiological roles are, in many cases, incompletely defined. There is emerging evidence that certain KDMs have multiple substrates including non-histone proteins and potentially methylated arginyl residues [8,9], suggesting the possibility for in vivo pleiotropy in KDM catalysis.
Though less well studied, lysyl methylation has been observed on histones other than H3 and H4. Histone 1 isotype 4 (H1.4), as opposed to the core histones H2A, H2B, H3 and H4, is one of multiple 'linker' histones that bind to nucleosomal DNA, but which do not form part of the histone octamer [10]. Linker histones have been traditionally viewed as playing important structural roles in stabilising higher order chromatin structure [11]. More recent evidence suggests that histone 1 isotypes and their post-translational modifications also have additional functional roles within the cell, including in transcriptional regulation [11,12]. To date, a number of sites of histone H1.4 lysyl N e -methylation have been identified (Fig 1B), with H1.4K26 being the most abundantly methylated site on H1.4 [13]. The methylation status of H1.4K26 is reported to vary dynamically during the cell cycle [14], and to undergo KDM4-catalysed demethylation, as supported by in vitro studies on mouse KDM enzymes [15]. The methylation status of the H1.4K26 residue is commonly associated with repressed genes, and is known to regulate binding of the family of heterochromatin protein 1 (HP1) proteins and related chromodomains, such as CDYL2 [16,17]. Loss of H1.4K26 (by substitution with a histone H1.4 variant bearing an alanine residue at this position, H1.4K26A) results in a reduction of cell proliferation, deregulation of gene expression and a reduction of stabilisation of H1.4 in heterochromatic regions [14]. Methylation of H1.4K26 also enhances phosphorylation of H1.4K27 by the Aurora B kinase, which in turn hinders recruitment of HP1 to H1.4K26me2 [17,18]. H1.4K26 is also reported to be acetylated by an unknown acetyltransferase, and to be deacetylated by SIRT1 [19]. Thus, emerging, and incompletely characterised, results indicate that complex regulatory loops exist for H1.4, in addition to those established with the core histones.
To address these questions, we report studies investigating human JmjC KDM-catalysed demethylation of H1.4K26 methylation. Our findings reveal that demethylation of methylated lysyl residues in H1.4K26 peptide fragments is catalysed by all members of the human KDM4 subfamily of JmjC KDMs (KDM4A-E), supporting previously reported cell-based studies (predominantly with mouse enzymes) [15]. Unexpectedly, we also observed low level H1.4K26me2 demethylation activity by KDM7A and KDM7B (PHF8), in the context of fragment peptides, though at levels that may not be biologically relevant. While demethylation of H1.4K26me2/1-containing peptides is relatively inefficient for KDM7A and KDM7B, kinetic analyses indicate that H1.4K26me3 is a comparably efficient substrate for KDM4A as H3K9me3 (at least in the context of fragment peptides) and is a better substrate than H3K36me3. This assignment is supported by crystallographic studies indicating that the H1.4K26me3 peptide binds to the KDM4 active site in a similar manner to the H3K9me3 peptide. Overall, H1.4K26me3 appears to be a good substrate for KDM4A, and indeed other KDM4 enzymes, which is  supportive of its assignment as a KDM4 substrate in mouse and human cells [15].

Peptide synthesis
Peptides were produced as C-terminal amides using a Liberty Blue automated microwave peptide synthesiser (CEM Corporation, Matthews, NC, USA), using standard fluorenylmethyloxycarbonyl-mediated solid-phase chemistry and NovaPEG rink amide resin (Merck, Kenilworth, NJ, USA). Following synthesis, peptides were cleaved from the resin by incubation with trifluoroacetic acid/water/triisopropylsilane/dimethoxybenzene (92.5 : 2.5 : 2.5 : 2.5) for 3 h followed by precipitation with ice-cold diethyl ether. For kinetic experiments, lyophilised peptides were purified by reverse-phase high-performance liquid chromatography using a Vydac C18 column (Solvent A: 0.1% trifluoroacetic acid in H 2 O, Solvent B: 0.1% trifluoroacetic acid in acetonitrile). Sequences are given in Table S1.

Kinetic analyses
Kinetic parameters for KDM4 enzymes were determined by use of an FDH (formaldehyde dehydrogenase)/NAD + -coupled) assay for quantification of the formaldehyde reaction byproduct, as previously described [27]. Assays were carried out in 50 mM HEPES, pH 7.5, 0.01% Tween-20, with addition of Fe (II) ammonium sulfate (10 lM), sodium ascorbate (100 lM), 2oxoglutarate (200 lM), NAD + (500 lM), KDM4 enzymes (1 lM for specific activities, 500 nM for Michaelis-Menten kinetics) and FDH enzyme (0.001 UÁlL À1 , Sigma-Aldrich, St Louis, MO, USA) in a 30-lL volume in black 384-well plates. Specific activities were measured using a 100-lM peptide. For Michaelis-Menten experiments, peptide concentrations were varied. Reactions were monitored using a PHERAstar FS plate reader (BMG Labtech, Ortenberg, Germany) with 355 nm excitation and 460 nm emission. Kinetic parameters were calculated from the reaction rate during the initial linear phase of formaldehyde production, which were used to calculate specific activities or fitted to Michaelis-Menten equations using GRAPHPAD PRISM (GraphPad Software, La Jolla, CA, USA).

X-ray crystallography
Co-crystals of KDM4A  .Ni(II).H1.4(18-32)K26me3. NOG were obtained in sitting drops grown at 4°C with a ratio of 1 : 2 sample to well solution [0.1 M MIB buffer (malonic acid/imidazole/boric acid, pH 6.0, 25% w/v PEG 1500)] [26]. Drops contained 10 mgÁmL À1 KDM4A, 5 mM H1.4(18-32)K26me3, 5 mM NOG and 4 mM NiCl 2 . Crystals were cryoprotected with 25% glycerol then flash-frozen in liquid N 2 . Data were collected using a single crystal at 100 K at the Diamond I04-1 MX beam line and were processed with HKL2000 [31]. The structure was solved by molecular replacement using PHASER [32] (search model: PDB ID 2OX0) and was refined by alternative cycles of CNS [33] and PHENIX [34], with iterative rebuilding of the refined model using COOT [35]. All residues were in the allowed regions of the Ramachandran plot as calculated by PROCHECK [36]. Data collection and refinement statistics are given in Table S2. The crystal structure has been deposited under PDB accession code 6H8P.
Given the activity observed with KDM4E and KDM7A, we decided to investigate the intra-subfamily conservation of this demethylation activity. The demethylation activity of the other subfamily members (KDM4A-D and KDM7B) was therefore screened. For all KDM4 subfamily members, we observed H1.4K26 demethylation activity similar to that already observed with KDM4E (Table 1, Figs S8-S11); however, no H1.4K26 demethylation activity was observed with KDM7B at this enzyme concentration (Fig. S12). Note that as with H3K9, demethylation of the monomethylated H1.4K26 peptide was only observed with KDM4E, under the stated conditions [24].   [29,44]. To test whether similar targeting may occur with H1.4, promoting activity at H1.4K26, an H1.4 peptide was synthesised containing di-N e -methyllysine at K26 and tri-N e -methyllysine at K21 (i.e. at the equivalent position in the sequence to K4 in the H3 peptide, Fig. 1B, Table S1). Monomethylation at H1.4K21 has previously been observed by proteomic mass spectrometry [45]. With the aim of enhancing activity, the enzyme concentration was increased from 1 lM to 5 lM, and time course analyses were conducted with both KDM7A and KDM7B, both with the new peptide (H1.4(18-32) K21me3K26me2) and the H1.4 peptides methylated only at K26 (Fig. 2B). At this increased enzyme concentration, clear evidence for demethylation of H1.4K26me2 was observed with KDM7B in addition to KDM7A (Fig. S12E), but no significant difference was observed between the peptide with and without methylation at H1.4K21 (Fig. 2B). Thus, it appears unlikely that N e -trimethylation at H1.4K21 promotes demethylation at H1.4K26 by KDM7B. These time courses, and similar ones with KDM7A, confirmed that the H1.4K26 peptides were significantly poorer substrates for the KDM7 subfamily than the H3K9 peptides (the weak demethylation efficiency observed for KDM7A/B precluded further kinetic analyses).
Studies then focused on determining the proficiency of H1.4K26 demethylation by the KDM4 enzymes.
Specific activities of each enzyme with H1.4K26me3 were compared to those with H3K9me3, and more detailed studies were carried out with a representative KDM4 enzyme, KDM4A. 1 H NMR time course analyses with KDM4A and the trimethylated H1.4K26me3 peptide confirmed time-dependent demethylation of the Kme3 residue (Figs 2C, S13). The 1 H NMR experiments also revealed concomitant turnover of 2OG into succinate (emergence of a singlet 1 H resonance at d H 2.28 ppm, Fig. S13) that is consistent with the proposed demethylation mechanism [5].
Kinetic analyses were then undertaken using a fluorescence-based formaldehyde dehydrogenase (FDH)coupled demethylation assay, which monitors the formation of NADH during FDH-catalysed oxidation of formaldehyde [27]. Initially, specific activities for each KDM4 enzyme with either H1.4K26me3 or H3K9me3 (each at 100 lM) were determined for comparison (Figs 2D, S14). These results revealed that in all cases demethylation of the two different marks was roughly comparable.
Michaelis-Menten kinetics were then conducted with KDM4A. Samples containing KDM4A, 2OG, ascorbate, H1.4K26me3 peptide, NAD + and FDH were analysed over time and kinetic parameters were determined by recording the initial reaction rates of NADH production (corresponding to enzymatic production of formaldehyde) at varying peptide concentration. Similar experiments were carried out with H3K9me3 and H3K36me3 substrate peptides and the obtained values compared ( Table 2). The experiments revealed that the H1.4K26me3 and H3K9me3 peptides are similarly efficient substrates of KDM4A, giving similar K M (32.3 AE 6.4 lM and 25.5 AE 4.9 lM respectively) and k cat (0.32 AE 0.049 s À1 and 0.31 AE 0.045 s À1 respectively) values. Demethylation of the H3K36me3 peptide was the least efficient, with higher K M (66.8 AE 5.4 lM) and lower k cat (0.12 AE 0.004 s À1 ) values than with the other two peptides.
Crystallographic analyses were undertaken to investigate the binding mode of trimethylated H1.4K26 in  [46]. Density was observed for H1.4 residues 24-29. Refinement revealed the H1.4K26me3 peptide bound in a cleft along the surface of KDM4A with the methylated K26 residue protruding towards the active site-bound metal centre (Figs 3A, S1), positioning the carbon of the methyl group~4 A from the active site metal, apparently productively poised for enzymatic hydroxylation and subsequent demethylation.
Comparison of the H1.4K26me3 complex structure with those for a KDM4A:Ni(II):NOG:H3K9me3 peptide complex (PDB: 2OQ6 [26], Fig. 3B) and a KDM4A:Ni(II):NOG:H3K36me3 peptide complex (PDB: 2P5B [47], Fig. 3C) reveals that all three peptides bind with the side chain of the trimethyllysine residue in the same orientation. All three peptides have the same N-to C-directionality through the active site. The two residues N-terminal to the trimethyllysine residue (A24, R25 in the H1.4 peptide, A7, R8 in the H3K9 peptide and G34, V35 in the H3K36 peptide) adopt a similar conformation/orientation in all three structures, with backbone hydrogen bonds between KDM4A E169 and the peptide residues at the À1 and À2 positions (Fig. 3B/C). However, differences arise in the conformations of the C-terminal regions of the peptides. The H3K9me3 and H1.4K26me3 peptides manifest similar conformations/orientations for the three residues C-terminal to the target trimethyllysine residue, sharing conserved interactions with KDM4A K241, D135 and N86 (black dashes, Fig. 3B). By contrast with the H3K9me n and H1.4K26me3 substrate structures, the peptide backbone of H3K36 adopts a more 'bent' conformation, likely enforced by the Fig. 3. Binding mode of H1.4K26me3 to KDM4A. The figure shows views from (A) an X-ray crystal structure of KDM4A bound to nickel (green, substitute for iron), NOG (pink, a 2OG mimetic inhibitor) and an H1.4(18-32)K26me3 peptide (yellow), alone, or overlaid with: (B) a structure of KDM4A bound to H3(7-14)K9me3 peptide (cyan, PDB ID 2OQ6) [26], and (C) a structure of KDM4A bound to an H3K36me3 peptide (violet, PDB ID 2P5B) [47]. Residues from KDM4A in the H1.4K26 structure are shown as pale grey sticks, those from KDM4A in the H3K9 structure are shown as pale green sticks and those from KDM4A in the H3K36 structure are shown as pale pink (A-C). Some of the key interactions between KDM4A and the H1.4K26me3 peptide are shown by black dashes. Interactions between the H1.4K26me3 peptide and KDM4A conserved in the H3K9me3 peptide structure, but not in the H3K36me3 peptide structure are marked as pink dashes. Interactions only observed in the H3K36me3 peptide structure are shown as green dashes. (D) Overlay of the peptide backbones (and Kme3 side chain) of each peptide. The amide flip between the H3K36 peptide, and the H3K9 and H1.4K26 peptides, is highlighted with a dashed black box. presence of a proline (H3P38) in its sequence. Thus, in the K36 peptide, the backbone carbonyl of the H3K36 to K37 amide bond is 'flipped'~180°compared to the analogous amide in the other two substrates (Fig. 3D). Unlike the H3K9 and H1.4K26 peptides (pink dashes, Fig. 3C), the H3K36 peptide displays no interactions with either KDM4A K241 or D135, nor with the sidechain of N86. Instead, the backbone -NH of the H3K36-K37 amide link interacts with the phenol of Y175 (green dashes, Fig. 3C).
Overall, the crystallographic analyses reveal the H1.4K26me3 substrate adopts a binding mode more similar to that of H3K9me3 than H3K36me3. This common binding geometry is supportive of their similar demethylation efficiencies.

Discussion
Though lysyl methylations have been reported on linker histones [10,45], the processes that both regulate these methylation levels and contribute to their functional roles are largely undefined. Demethylation of methylated lysine residues at K26 of linker H1.4 has been reported to be catalysed by the KDM4 subfamily of JmjC KDMs in cells [15,43]. In previous studies on H1.4, no activity was observed for the KDM3 or KDM6 subfamilies; however, the KDM7 subfamily, which, like the KDM4s, also act at H3K9 had not been tested for activity with H1.4 [29,48,49]. Our results demonstrate that at least in the context of histone fragment peptides, recombinantly produced members of both the human KDM4 and KDM7 subfamilies of JmjC KDMs are capable of catalysing demethylation of methylated lysine residues at K26 of linker H1.4. Demethylation of dimethylated H1.4K26 by KDM7A and KDM7B appears considerably less efficient than the well-characterised H3K9me2 substrates for these enzymes, possibly precluding any biological relevance [29,42,44]. However, kinetic analyses with KDM4A reveal that demethylation of the H1.4K26me3 peptide and the well-characterised KDM4 H3K9me3 substrate are comparably efficient, whereas the H3K36me3 substrate is less efficiently demethylated than either H3K9me3 or H1.4K26me3 under the conditions tested [23,24,50,51].
Crystallographic analyses of H1.4K26me3 bound to KDM4A imply very similar binding modes for the H1.4K26me3 and an H3K9me3 peptides [26,47], consistent with the kinetic studies which imply that they are substrates of approximately equal efficiency. Importantly, the H3K36me3 peptide, which at least in vitro is a less efficient substrate for KDM4A-C (and not a substrate at all for KDM4D/E) than either H3K9me3 or H1.4K26me3, binds differently at the KDM4A active site [23,47,50]. One clear difference is in the binding mode of the amide bond on the C-terminal side of the N-methylated substrate lysine for H3K9/H1.4K26 versus H3K36 (Fig. 3). Although how this observation relates to the different binding/catalytic efficiencies of the different peptide sequences is difficult to dissect, it suggests that subtle differences in binding conformation may have substantial effects on catalytic efficiency, an observation relevant to functional assignment work on JmjC KDMs [50,52]. Similar subtleties have been observed for chromatin reader domains; in some cases they bind similarly to the closely related H3K9 and H1.4K26 sequences (e.g. HP1) [17], whilst in others binding is only observed to one or the other (e.g. ankyrin repeats of G9A) [43].
The combined biochemical and structural studies thus identify methylated H1.4K26 peptides as substrates for JmjC KDMs with broadly comparable demethylation efficiencies to those reported for KDM4 substrates on core histones [23,24,50,51]. Given this potential for activity at H1.4, it is of interest to further investigate the consequence of methylation at this position in vivo, in particular in the context of ongoing medicinal chemistry studies targeting the KDM4 subfamily as a treatment for disease [7,53]. The methylation status of H1.4K26 is reported to regulate binding of HP1 and other chromodomain proteins [16,17], which are associated with formation of heterochromatin. Thus, the reversal of this methylation by KDM4 enzymes in cells may well have direct effects on chromatin compaction and gene expression. It remains unclear whether a complex and dynamic 'histone code' akin to that proposed for the H3 N-terminal tail, exists for H1. The dynamic modulation of PTMs would be a likely requirement for this. Regulation of PTMs on H1.4 has already been demonstrated to play important roles in gene regulation, as in the case of acetylation of H1.4K34, which may help recruit the general transcription factor IID through its bromodomain, and which increases the dynamic exchange of H1 [54], or citrullination of H1.2R54, which results in histone eviction and chromatin decondensation [55]. The ready demethylation of H1.4K26 as catalysed by the KDM4 demethylase family argues that this might also be the case for H1.4K26 methylation, at least for H1.4K26me3 [12]. Overall, we hope that our findings will stimulate further cellular work to characterise the full scope of JmjC KDM substrate selectivity and will be informative for ongoing studies with JmjC KDM inhibitors.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.      Fig. S11. KDM4D catalyses lysine demethylation at H1.4K26. Fig. S12. PHF8/KDM7B only catalyses lysine demethylation at H1.4K26 at high concentration. Fig. S13. Analysis of KDM4A demethylation by 1 H NMR. Fig. S14. Specific activity determination for KDM4 enzymes. Table S1. Peptide sequences used in this study. Table S2. Crystallographic data processing and refinement statistics.