Investigating D-Lysine Stereochemistry for Epigenetic Methylation, Demethylation and Recognition

Abstract

Histone lysine methylation is regulated by N^ε-methyltransferases, demethylases, and N^ε-methyl lysine binding proteins. Thermodynamic, catalytic and computational studies were carried out to investigate the interaction of three epigenetic protein classes with synthetic histone substrates containing l- and d-lysine residues. The results reveal that out of the three classes, N^ε-methyl lysine binding proteins are superior in accepting lysines with the d-configuration and identify key electrostatic interactions involved.

Histone tails are subject to a plethora of posttranslational modifications (PTMs); acetylation, methylation, phosphorylation¹ are established with many more recently discovered.^2–4 Histone lysine methylation is linked to both gene activation and repression, depending on the methylation state and modification site. Methylation is catalysed by S-adenosylmethionine (SAM) dependent histone lysine methyltransferases (HKMT) that install one (Kme1), two (Kme2), or three (Kme3) methyl groups on the lysine N^ε-amino group (Fig. 1).⁵ N^ε-Methyl group removal is catalysed by flavin-dependent lysine specific demethylases (KDM1 in humans) that accept only Kme1 or Kme2 modifications, or by the Fe(II) and 2-oxoglutarate (2OG) dependent JumonjiC (JmjC) demethylases (KDM2-7) that catalyse demethylation of all three types of N^ε-methylated lysines (Fig. 1).⁶ Methylated lysines are recognised by different classes of reader proteins, including plant homeodomain (PHD) zinc fingers, tandem tudor domains (TTD), chromodomains (CD) and malignant brain tumor (MBT) domains (Fig. 1).⁷ In general, hydrogen bonds and electrostatic interactions appear relatively more important in binding of Kme1 and Kme2 to the reader proteins, whereas Kme3 groups bind to aromatic cage-containing reader proteins via cation–π interactions^8,9 involving displacement of water molecules.⁹

Fig. 1.

A) N^ε-Methylation, demethylation and binding of histones. B) View from a SETD7 structure (magenta) complexed with H3K4me1 (yellow) and S-adenosylhomocysteine (cyan) (PDB: 1O9S). C) View from a KDM4A_JmjC (green) structure complexed with H3K9me3 (yellow) and 2OG (cyan) (PDB: 2OQ6). D) View on KDM5A_PHD3 (blue) complexed with histone mimic peptide H3K4me3 (yellow) (PDB: 2KGI).

Structural and mechanistic work implies that three classes of histone N^ε-methyl lysine interacting proteins accept the N^ε-(methyl) l-lysine as their natural substrate/ligand; however, their selectivity with respect to lysine C^α stereochemistry has not been investigated. This is of interest because protein residue epimerisations and methylation patterns are implicated in ageing¹⁰/disease⁶ and because some 2OG oxygenases, including some JmjC KDMs and hydroxylases, have a broad substrate selectivity.¹¹ Along with rearrangements, including Asn and Gln residues, d-amino residues have been observed in ageing/diseased cells¹² and it has been proposed that d-lysine residues occur in tumour cells.¹³

We hypothesised that there might be a selectivity between the three classes of epigenetic proteins for the acceptance of (methylated) d-lysine residues on histones. Here, we report studies comparing the selectivity of histone lysine methyltransferases, histone lysine demethylases and epigenetic readers for (methylated) l- and d-lysine residues.

We first used isothermal titration calorimetry (ITC) to investigate the binding of representative PHD zinc fingers and TTDs recognising the H3K4me3 mark, i.e: i) the PHD zinc fingers of KDM5A_PHD3, TAF3_PHD, and BPTF_PHD and ii) the TTDs of SGF29_TTD and KDM4A_TTD.⁷ Thermodynamic parameters for the associations between reader proteins and 10-mer histone peptides that bear l-Kme3 and d-Kme3 were measured. In all cases, the readers form stronger complexes with the l-Kme3, compared to the d-Kme3-histone sequences (Table 1, Fig. S1 in ESI). d-H3K4me3 bound to KDM5A_PHD3, TAF3_PHD and BPTF_PHD with an ~10-fold lower affinity than l-H3K4me3, whereas the association of d-H3K4me3 to SGF29_TTD and KDM4A_TTD decreased more substantially (~30–35 fold, compared to l-H3K4me3). The results reveal that decreased affinity for d-H3K4me3 relative to l-H3K4me3 derives from less favourable enthalpy of binding (ΔH°); values of ΔΔH° are in the range of the 2.1 kcal mol^-1 (for TAF3_PHD) to 10.5 kcal mol^-1 (for KDM4A_TTD) (Table 1). The entropy of binding (–TΔS°) is relatively more favourable for d-Kme3-containing histone peptides, but this does not compensate for the decreased enthalpy values (Table 1). We have reported that binding of 10-mer H3G4 (where l-Kme3 is substituted by glycine) results in a large decrease (>500-fold) in binding strength to the same readers.⁹ The observation that replacement of l-Kme3 by d-Kme3 causes a smaller reduction in affinity (8–36-fold) and enthalpy suggests that the d-Kme3 side chain is involved in energetically favourable interactions, possibly within the aromatic cage; it is also possible that binding of d-Kme3 leads to less favourable interactions of neighbouring amino acids in the histone peptide with the reader proteins.

Table 1.

Thermodynamic parameters for binding of 10-mer histone peptides l-H3K4me3 and d-H3K4me3 (ART(l-Kme3/d-Kme3)QTARKS) to reader proteins. (Measured by ITC ± Standard Deviation (3–5 repeats).)

	l-H3K4me3				d-H3K4me3
	Kd (μM)	ΔG° (kcal mol-1)	ΔH° (kcal mol-1)	−TΔS° (kcal mol-1)	Kd (μM)	ΔG° (kcal mol-1)	ΔH° (kcal mol-1)	−TΔS° (kcal mol-1)
KDM5APHD3	0.11	−9.5 ± 0.1	−11.1 ± 0.1	1.6 ± 0.1	1.2	−8.1 ± 0.1	−7.8 ± 0.1	−0.3 ± 0.1
TAF3PHD	0.082	−9.7 ± 0.1	−10.1 ± 0.1	0.4 ± 0.2	0.73	−8.4 ± 0.1	−8.0 ± 0.1	−0.4 ± 0.2
BPTFPHD	0.44	−8.7 ± 0.1	−13.2 ± 0.1	4.5 ± 0.1	3.8	−7.4 ± 0.1	−7.7 ± 0.1	0.3 ± 0.1
SGF29TTD	1.7	−7.9 ± 0.1	−7.7 ± 0.1	−0.2 ± 0.1	50	−5.9 ± 0.1	−2.3 ± 0.1	−3.6 ± 0.1
KDM4ATTD	1.1	−8.1 ± 0.1	−13.1 ± 0.2	5.0 ± 0.2	39	−6.0 ± 0.2	−2.6 ± 0.1	−3.4 ± 0.2

Temporal atomistic molecular dynamic (MD) simulations were then used to investigate the relative flexibility of the readers and their ability to accommodate d-Kme3. MD approaches have proven to be valuable for elucidating properties concerning binding of histone PTM residues by epigenetic proteins.^14,15 Previously, MD simulations (10 ns with the Amberff99sb force field) of H2AK5ac, H4K12ac, and H3K14ac binding to the BRPF1 bromodomain were used to understand selectivity.¹⁴ Binding modes to the multi-domain JARID demethylase protein, KDM5C, have also been studied by MD (10 ns with Amberff14sb).¹⁵

We simulated epigenetic reader proteins with the l-Kme3 and modified d-Kme3 residues. Starting structures were taken from representative crystal structures where the l-Kme3 stereocenter was manually inverted to generate the d-Kme3 complex, with priority given to replicating the position of l-Kme3 in the original PDB structure (see ESI). All simulations used AMBER12.¹⁶ Systems were solvated in a 10 Å truncated octahedral box of TIP3P¹⁷ water, neutralised explicitly with either sodium or chloride ions, and simulated for a total of 10 ns. The binding poses of the l- and d-Kme3 residues relative to the surrounding reader aromatic cage following minimization and equilibration are shown in Fig. S2. Qualitative analysis of snapshots taken after 5 ns and 10 ns of MD simulation show the behaviour of the complexes over time (Fig. S3-7). Orientations of the two Kme3 stereoisomers within the aromatic pocket of the same reader are similar at 0 ns (Fig. S2A-D), with the exception of SGF29_TTD (Fig. S2E). KDM5A_PHD3 bound to both d- and l-Kme3 exhibit the most similar pose regarding placement of the W18-W28 aromatic cage and modified residues (Fig. S2A). In the case of SGF29_TTD, d-Kme3 more fully occupies the aromatic Y238-Y245-F264 cage than does l-Kme3; however, l-Kme3 adopts a more similar orientation to d-Kme3 by 5 ns that is observed throughout the simulation (Fig. 2SE). For the two H3-PHD reader complexes, KDM5A_PHD3 (Fig. S3) and TAF3_PHD (Fig. S4), major differences in the H3 chain backbone geometry were observed throughout the simulation when comparing the l- and d- systems (Table S1). The reorientation of H3 backbone for d-H3K4me3 bound to TAF3_PHD and KDM5A_PHD3 is stabilized by favourable cation-π interactions with the W868-W891 and W18-W28 cages, respectively. Movement of the H3 backbone is minimal for the other 3 reader complexes (Fig. S5-7). We analysed the electrostatic energies (E_ele) between N^ε of l-Kme3 and d-Kme3 and the π-system of reader protein aromatic cage residues during each simulation. Electrostatically-dominated cation-π interactions are present for both stereoisomers. Those involving l-Kme3 are slightly more favourable than d-Kme3 for all reader proteins except SGF29_TTD (Table S2). This can be explained by the non-optimal starting geometry (Fig. S2E), wherein l-Kme3 is pointing away from the Y238-Y245-F264 aromatic cage of SGF29_TTD.

Overall, the combined MD simulations and ITC binding studies indicate the ability of reader proteins to efficiently accommodate d-H3K4me3 residues. Electrostatic energy calculations between N^ε of Kme3 and surrounding aromatic residues suggest the presence of favourable stabilizing cation-π interactions for both stereoisomers. Examination of the H3 orientation for d-Kme3 complexed with readers KDM5A_PHD3 and TAF3_PHD emphasize the flexibility of the histone backbone to potentially prioritize the cation-π interactions. The observed reduction in binding enthalpy for d-Kme3 is presumably due to weaker cation–π interactions, weaker H-bonding, and/or the release of smaller number of water molecules in the aromatic cage. The more favourable entropy of binding for d-Kme3 may arise from a higher conformational degree of freedom of d-Kme3 side chain (than l-Kme3 side chain) in the complex.

We then explored whether histone lysine methyltransferases catalyse methylation of d-lysine residues. Four representative human methyltransferases were chosen to investigate d/l-stereoselectivity: SETD7 that monomethylates H3K4, SETD8 that monomethylates H4K20, and G9a and GLP that di- and trimethylate H3K9.⁵ MALDI-TOF MS analyses of SETD7-catalysed methylation of H3K4 manifested quantitative production of monomethylated H3K4 (i.e. l-Lys at position 4) under standard assay conditions. In contrast, d-H3K4 (i.e. d-Lys at position 4) was not observed to be a substrate under the same conditions (2 μM SETD7, 100 μM histone peptide, 200 μM SAM, pH 8.0, 37 °C, 1 hour) (Fig. 2A, S8-9). Increased amounts of SETD7 (10 μM) and SAM (1 mM), and prolonged incubation (3 hours) did not yield any monomethylated product d-H3K4me1. Similarly, SETD8 was observed to efficiently catalyse monomethylation of H4K20, but no methylation of d-H4K20 was detected under standard conditions (Fig. 2B), or with prolonged incubation with additional SETD8/SAM (Fig. S10-11). MALDI-TOF MS assay of G9a/GLP-catalysed methylation of 15-mer H3K9 peptide revealed efficient formation of H3K9me3 (Figs. 2C,2D and S12,13). Under the same conditions, we only observed trace evidence for monomethylation of d-H3K9 with G9a/GLP. A significantly larger amount (20-35%) of the d-H3K9me1 product was observed with increased G9a/GLP (10 μM) and SAM (1 mM) after 1 hour (37 °C); all 3 methylated products were observed after 6 hours (Fig. S14-15). Competitive experiments between 14-mer l-H3K9 and 15-mer d-H3K9 indicated that d-H3K9 does not inhibit G9a-catalysed methylation of H3K9 within detection limits (Fig. S16). Collectively, the results imply the tested methyltransferases are specific for the l- over the d-lysine stereoisomers. Notably, d-lysine is poorly accepted by some methyltransferases (G9a and GLP). Structural analyses of histone lysine methyltransferases with substrates reveal binding in a narrow apolar tunnel;¹⁸ it is likely that the positioning of D-lysine in this tunnel is non-optimal, thus slowing catalysis (Fig. 1B).

Fig. 2.

MALDI-TOF MS assays for SAM-mediated methylation of l-lysine (top panel) and d-lysine (bottom panel) containing histone peptides by histone lysine methyltransferases. A) l-H3K4 and d-H3K4 with SETD7. B) l-H4K20 and d-H4K20 with SETD8. C) l-H3K9 and d-H3K9 with G9a. D) l-H3K9 and d-H3K9 with GLP. (black = starting peptide, red = after HKMT-catalysed reaction).

We investigated potential demethylation of methylated d-lysine residues by histone lysine demethylases, using 6 human histone lysine demethylases, i.e. catalytic domains of KDM1A (that demethylate H3K4me2), KDM5B_JmjC and KDM5C_JmjC (that demethylate H3K4me3), and KDM4A_JmjC, KDM4D_JmjC and KDM4E_JmjC (that demethylate H3K9me3).⁶ LC-MS analyses confirmed activity with the l-lysine H3K4me2/H3K4me3/H3K9me3 peptides. Thus, under standard conditions (200 nM enzyme, 5 or 10 μM substrate), we observed KDM1A-catalysed demethylation of H3K4me2 to H3K4, KDM5B_JmjC/KDM5C_JmjC-catalysed demethylation of H3K4me3 to H3K4me2 and H3K4me1, and KDM4A_JmjC/KDM4D_JmjC/KDM4E_JmjC-catalysed demethylation of H3K9me3 to H3K9me2 and H3K9me1 (Fig. 3, S17-23). Under the same conditions, the results of analogous d-lysine substrates showed only traces (at most) of possible demethylation (Fig. 3, S17-23), implying that histone lysine demethylases require the l-stereochemistry for efficient catalysis. Use of higher enzyme concentrations (600 nM or 1 μM) did not enhance demethylation with methylated d-lysine, yielding only traces of demethylated products.

Fig. 3.

Deconvoluted LC-MS data showing demethylation of methylated l-lysine (top panel) and methylated d-lysine (bottom panel) containing histone peptides in the presence of histone lysine demethylases, Fe(II), 2OG and ascorbate. A) l-H3K4me2 and D-H3K4me2 with KDM1A. B) l-H3K9me3 and d-H3K9me3 with KDM4A_JmjC. C) l-H3K9me3 and d-H3K9me3 with KDM4D_JmjC. D) l-H3K4me3 and d-H3K4me3 with KDM5B_JmjC. (black = starting peptide, red = after KDM-catalysed reaction). Conditions for JmjC-KDM assay: 10 μM 2OG, 10 μM Fe(II), 100 μM Asc.

To test for potential stimulation of 2OG turnover by d-H3K9me3 we employed ¹H NMR, monitoring signals of H3K9me1/2/3, 2OG and succinate catalysed by KDM4E_JmjC. The results confirmed that KDM4E_JmjC catalyses l-K9me3 demethylation (Fig. S24,26).¹⁹ Analysis of the d-K9me3 potential substrate shows substantial lower formation of Kme1/2 under the same conditions, compared to l-K9me3 (Fig. S25,27). Use of an increased enzyme did not increase formation of d-Kme1/2 for d-K9me3. (Note, traces (≤5%, determined by ¹H NMR (Fig. S27)) of demethylation for the d-Kme3 may be due to the presence of very low levels of the l-epimers produced during synthesis.) Initial rates of uncoupled 2OG turnover (Fig. S26-28) do not indicate substantial differences between d-K9me3 and l-K9me3, suggesting that d-K9me3 does not stimulate significant 2OG turnover. Additional experiments to examine the influence of the d-K9me3 on l-K9me3 demethylation in a competition assays showed only a small decrease in succinate formation, suggesting only very weak binding/inhibition by the d-K9me3 peptide (Fig. S29,30).

Our observations that KDMs do not (or extremely poorly) accept methylated d-lysine as substrates are consistent with their crystal structures complexed with histone peptides (Fig. 1C). Proximate positioning of the quaternary ammonium group of l-Kme3 to the active site iron is essential for efficient demethylation;²⁰ d-Kme3 presumably orients the N^ε-methyl group away from the iron, consequently leading to a non-productive binding mode. It is notable that human trimethyllysine hydroxylase, the first enzyme in the carnitine biosynthesis pathway, catalyses C-3 hydroxylation of the natural free l-N^ε-trimethyllysine (obtained by the proteolytic degradation of l-N^ε-trimethyllysine containing histones), but not d-N^ε-trimethyllysine.²¹

Overall, the results reveal that histone lysine methyltransferases, histone lysine demethylases and epigenetic readers efficiently modify or bind the (methyl) l-lysine, but manifest different levels of acceptance of (methylated) d-lysine as substrates/ligands. Our results imply that the protein-histone interactions, especially for the KDMs, critically determine the enzyme activity, whereas the associations between reader proteins and histones appear to be less sensitive to changes in lysine C^α stereochemistry. Notably, we did observe methyltransferase activity with some of the HKMTs with the d-configured substrates. Coupled with the lack of KDM activity and the observations of binding of d-residues by some reader domains, this raises the possibility that aberrant PTM patterns may occur should d-residues be produced in diseased or aged cells.

Supplementary Material

Electronic Supplementary Information (ESI) available: Supporting figures, enzyme assays, peptide synthesis, protein production, synthesis, NMR data. See DOI: 10.1039/x0xx00000x