Structural Basis of LSD1-CoREST Selectivity in Histone H3 Recognition*

Histone demethylase LSD1 regulates transcription by demethylating Lys4 of histone H3. The crystal structure of the enzyme in complex with CoREST and a substrate-like peptide inhibitor highlights an intricate network of interactions and a folded conformation of the bound peptide. The core of the peptide structure is formed by Arg2, Gln5, and Ser10, which are engaged in specific intramolecular H-bonds. Several charged side chains on the surface of the substrate-binding pocket establish electrostatic interactions with the peptide. The three-dimensional structure predicts that methylated Lys4 binds in a solvent inaccessible position in front of the flavin cofactor. This geometry is fully consistent with the demethylation reaction being catalyzed through a flavin-mediated oxidation of the substrate amino-methyl group. These features dictate the exquisite substrate specificity of LSD1 and provide a structural framework to explain the fine tuning of its catalytic activity and the active role of CoREST in substrate recognition.

Histone demethylase LSD1 regulates transcription by demethylating Lys 4 of histone H3. The crystal structure of the enzyme in complex with CoREST and a substrate-like peptide inhibitor highlights an intricate network of interactions and a folded conformation of the bound peptide. The core of the peptide structure is formed by Arg 2 , Gln 5 , and Ser 10 , which are engaged in specific intramolecular H-bonds. Several charged side chains on the surface of the substrate-binding pocket establish electrostatic interactions with the peptide. The three-dimensional structure predicts that methylated Lys 4 binds in a solvent inaccessible position in front of the flavin cofactor. This geometry is fully consistent with the demethylation reaction being catalyzed through a flavin-mediated oxidation of the substrate amino-methyl group. These features dictate the exquisite substrate specificity of LSD1 and provide a structural framework to explain the fine tuning of its catalytic activity and the active role of CoREST in substrate recognition.
Lysine methylation is among the most well characterized histone modifications, and its existence has been known since the early days of chromatin research (1,2). This type of epigenetic mark provides a huge potential for functional responses in that it can occur in different forms (mono-, di-, and tri-methylation) and on different histone sites, each having a specific physiological meaning. Histone methylation has been long thought to be a low turnover epigenetic mark, but the recent discovery of histone demethylases (3,4) has challenged this view by demonstrating that histone lysine methylation can be actively and dynamically regulated. Two classes of histone demethylases have been uncovered; the enzymes of the JmjC family use iron as cofactor, whereas lysine-specific demethylase 1 (LSD1) 4 employs FAD as the prosthetic group (5).
LSD1 catalyzes the oxidative demethylation of mono-and dimethyl Lys 4 of histone H3, generating hydrogen peroxide and formaldehyde (3,4). The enzyme is implicated as a key component of distinct co-activator and co-repressor complexes in a surprisingly wide range of cellular processes where it participates in the dynamic transition of transcriptional programs (6). Its catalytic activity is finely tuned by the epigenetic marks present on the H3 N-terminal tail (7) and by other protein partners, such as CoREST, that form a stable complex with the enzyme (8,9). The three-dimensional structure of LSD1 in its native state (10,11) and in complex with the LSD1-binding domain of CoREST (12) have revealed that the catalytic center is located in the core of the enzyme main body. A protruding tower domain consisting of two remarkably long helices forms the docking site for the co-repressor protein (Fig. 1A).
Here, we describe the structural analysis of LSD1-CoREST bound to a 21-amino acid H3 peptide in which pLys 4 ("p" is for peptide) is mutated to Met. The structural analysis illuminates the molecular properties that enable LSD1 to function as a key transcriptional regulator through both its histone demethylase activity and its highly specific binding to the H3 N-terminal tail.

EXPERIMENTAL PROCEDURES
Protein Purification-Escherichia coli cells overproducing the GST-fused LSD1-binding domain of CoREST (residues 305-482) were harvested 8 h after the addition of 0.25 mM isopropyl ␤-D-thiogalactopyranoside. Cells (10 g) were disrupted with an Emulsiflex C3 cell disruptor and the extract loaded onto a Glutathione-Sepharose 4B column (GE Healthcare) followed by extensive washing with 50 mM Tris/ HCl, pH 7.2, 300 mM NaCl, 1 mM dithiothreitol, and 1 mM EDTA. GST-tag cleavage was performed by incubating the column-bound protein with 80 units of Prescission protease (GE Healthcare) overnight at 4°C. Next, the eluted protein was gel-filtered using a Superdex 200 column equilibrated in 5% (w/v) glycerol, 25 mM potassium phosphate, pH 7.2. Human recombinant LSD1 was expressed and purified following published protocols (7). The protein lacks the N-terminal 122 amino acids (with reference to the sequence deposited in the NCBI data base under accession code NP_055828). Purified CoREST was mixed with LSD1 in a 1.5:1 molar ratio for 1 h and gel-filtered on a Superdex 200 column equilibrated in 5% (w/v) glycerol, 25 mM potassium phosphate, pH 7.2. The elution peak had the retention volume expected for the CoREST-LSD1 complex, the formation of which was further verified by SDS-PAGE analysis.
Enzymatic Assays-Enzymatic activities were measured under aerobic conditions by using a peroxidase-coupled assay on a Cary 100 UV-visible spectrophotometer (7). Peptides were * This work was supported in part by the Italian Ministry of Science (PRIN06 and FIRB programs) and the Italian Association for Cancer Research (AIRC). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. purchased from Thermo Electron Corp. Their purity was Ͼ90% as checked by analytical high-pressure liquid chromatography and mass spectrometry. Inhibition was tested in the presence of varied concentrations (2-100 M) of mono-methylated H3-K4 substrate and pLys4Met peptide (Table 1).
Crystallization and Structure Determination-The LSD1-CoREST complex was crystallized by the hanging-drop vapor diffusion method at 20°C by mixing equal volumes of 100 M LSD1-CoREST solution in 5% (w/v) glycerol, 25 mM potassium phosphate, pH 7.2, with reservoir solutions containing 1.2-1.3 M sodium/potassium tartrate and 100 mM N-(2-acetamido)-2-iminodiacetic acid, pH 6.5. Crystals were isomorphous to those obtained by Yang et al. (12); their cell parameters and space group are reported in Table 2. Crystals were soaked in a solution consisting of 1.6 M sodium/potassium tartrate, 100 mM N-(2-acetamido)-2-iminodiacetic acid, pH 6.5, 10% glycerol, and 2 mM pLys4Met peptide (measured final pH of the soaking solution is 7.0) for 3 h and then flash-cooled in a stream of gaseous nitrogen at 100 K. X-ray diffraction data were collected at the X06SA beam-line of the Swiss Light Source (Villigen, Switzerland). Data processing and scaling (Table 2) were carried out using MOSFLM (13) and programs of the CCP4 package (14). The structure of the LSD1-CoREST complex (PDB entry code 2IW5) (12) after removal of all water atoms was used as the initial model for refinement. Unbiased 2F o Ϫ F c and F o Ϫ F c maps were used to manually build the protein-bound peptide inhibitor (Fig. 1B); its conformation was clearly defined by the electron density map (Fig. 1B). Crystallographic refinements were performed with Refmac5 (15), and manual rebuilding was done with Coot (16). Refinement statistics are listed in Table 2. Pictures were produced with PyMol (www.pymol.org), Bobscript (17), and Raster3d (18).  a Apparent steady-state kinetic parameters were determined as described (7) by using a 21-amino acid H3 peptide monomethylated at Lys 4 . b The inhibition constant for the pLys4Met peptide was measured following previously described protocols (20). c Data taken from Ref. 7.
where I(h) i is the scaled observed intensity of the ith symmetry-related observation for reflection h, and ͗I(h)͘ is the average intensity. b Values in parentheses are for reflections in the highest resolution shell. c The final model consists of residues 171-836 of LSD1, a FAD molecule, residues 308 -440 of CoREST, and residues 1-16 of pLys4Met peptide. d R cryst ϭ ⌺͉F obs Ϫ F calc ͉/⌺͉F obs ͉, where F obs and F calc are the observed and calculated structure factor amplitudes, respectively. The set of reflections used for R free calculations and excluded from refinement was extracted from the structure factor file relative to the PDB entry code 2IW5 (12). e r.m.s., root mean square.

RESULTS AND DISCUSSION
Our structural analysis was based on human LSD1-CoREST crystals, which are isomorphous to those reported by Yang et al. (12) although they were grown under slightly different conditions (see "Experimental Procedures"). The structure of the ternary complex was obtained using a peptide bearing Met in place of Lys in position 4. With respect to the wild-type peptide (K i ϭ 1.8 M) (7), such a mutation led to a 30-fold increase in binding affinity (K i ϭ 0.05 M), making the pLys4Met substrate analogue a strong inhibitor and an ideal candidate for structural studies ( Table 1). The electron density map allowed us to unambiguously position the 16 N-terminal amino acids of the ligand (Fig. 1B). Peptide binding does not induce any conformational change, as indicated by the root-mean-square deviation of 0.4 Å from ligand-free LSD1-CoREST (799 equivalent C␣ atoms). The peptide binds in a funnel-shaped pocket (Figs. 1A and 2A) adopting a folded conformation in which three structural elements can be identified: a helical turn (residues 1-5) located in front of the flavin, a sharp bent (residues 6 -9), and a more extended stretch (residues 10 -16) that remains Nitrogens are blue, oxygens red, sulfurs yellow, and carbons green. The positions of negatively charged residues lining the peptide-binding site are labeled. The C␣ trace of CoREST residues 308 -314 is shown in red, highlighting their proximity to the LSD1 372-395 ␣-helix that is integral part of the peptide-binding site. The orientation is the same as in Fig. 1A. B, three-dimensional view of the peptide-binding mode. Nitrogens are blue, oxygens red, and sulfurs yellow. Carbons of peptide and protein residues are in green and gray, respectively. The flavin cofactor is yellow. With respect to Fig. 1A, the structure is rotated by ϳ180°about the vertical axis in the plane of the drawing. C, schematic representation of the peptide-protein interactions. D, model of dimethyl Lys 4 peptide substrate bound in the active site. Orientation and atom colors are the same as described in B. The modeling was carried out assuming that the C␤-C␥-C␦ atoms of dimethyl Lys 4 adopt the same conformation of the C␤-C␥-S␦ atoms of pMet 4 in the crystal structure. In this way, the predicted position of the N-bound CH 3 group of dimethyl Lys 4 falls exactly in front of the N-5 atom of the flavin. This type of substrate-binding geometry is similar to that found in other flavin-dependent oxidases and is fully consistent with an oxidative attack of the flavin on the N-CH 3 group of the substrate leading to the formation of an NHϭCH 2 imine that is then hydrolyzed to generate the demethylated Lys 4 and formaldehyde.
partially solvent-exposed on the rim of the binding pocket ( Fig.  2A). All amino acids fall in the allowed regions of the Ramachandran plot and therefore do not appear to adopt strained conformations. The core of the peptide structure is formed by pArg 2 , pGln 5 , and pSer 10 , which are engaged in several intramolecular interactions (Fig. 2, B and C). The side chain of pArg 2 is crucial in that it establishes H-bonds with the carbonyl oxygens of pGly 12 and pGly 13 and the side chain of pSer 10 . Moreover, all peptide residues define an intricate network of very specific intermolecular contacts with LSD1, often involving electrostatic interactions (Fig. 2). Among these, the N-terminal amino group of pAla 1 favorably interacts with the C terminus of helix 524 -540 (Figs. 1A and 2C), whereas the pArg 2 and pArg 8 form salt bridges with the carboxylate groups of, Asp 375 , Asp 553 , Asp 556 , and Glu 379 . The side chains of pLys 9 and pLys 14 are partially disordered, pointing toward the solvent ( Fig. 2A). However, they both can establish favorable electrostatic interactions with a cluster of negatively charged residues on the LSD1 surface that involve Asp 553 , Asp 556 , Asp 557 , and Glu 559 (Fig. 2A).
The three-dimensional structure of the ternary complex explains several biological properties of LSD1. The peptide residue in position 4 binds right in front of the cofactor, with the side chain pointing toward the flavin ring of FAD (Fig. 2B). This residue is made solvent-inaccessible by the peptide C-terminal portion that effectively seals the active center. A modeling analysis (Fig. 2D) in which pMet 4 is replaced by dimethyl Lys shows that the dimethyl group can be easily accommodated in the space between the peptide and the flavin. The predicted position of the pLys 4 -N-CH 3 group (the site of oxidative demethylation) is at about 3 Å distance from the reactive N-5 atom of the flavin, and the geometry of the flavin-methylamino group interactions is identical to those of the flavodehydrogenases-oxidases (19). In particular, the methylamino group of the substrate is fixed in position through the interaction with an aromatic side chain (Tyr 761 ; Fig. 2, B and D) as typically found in other flavin-dependent amine oxidases. Thus, the structure is fully compatible with the proposed oxidative mechanism for demethylation that involves flavin-mediated oxidation of the N-CH 3 group of methyl-Lys 4 .
An astonishing feature is the intricacy of the interactions between LSD1 and the histone peptide, which is consistent with the exquisite specificity of the enzyme (Fig. 2). LSD1 is not active toward peptides shorter than 16 amino acids, and optimal activity is detected with peptides comprising the N-terminal 21 amino acids of H3 (7). Consistently, the structure shows that the first 16 amino acids of the H3 peptide all specifically interact with the protein. After pPro 16 , the electron density is too poorly defined to allow model building (Fig. 1B). We speculate that the flexibility of amino acids 17-21 may be required to allow the tail to protrude from the nucleosomal particle to reach the binding pocket. The complexity of the peptide-enzyme interactions rationalizes the ability of LSD1 to recognize epigenetic modifications on H3. pSer 10 is solvent-inaccessible, being part of the intramolecular interactions that stabilize the peptide conformation. Phosphorylation of this residue is predicted to disrupt this binding geometry in keeping with the fact that pSer 10 -phosphorylated peptides are unable to bind to LSD1 (7). Likewise, both pArg 2 and pArg 8 side chains establish several inter-and intramolecular H-bonds, explaining the reduced binding affinity caused by arginine methylation (20). pLys 9 and pLys 14 are solvent-accessible and close to a cluster of negatively charged residues on the LSD1 surface (Fig. 2a). Indeed, methylation of pLys 9 (that preserves the side chain positive charge) does not affect the normal kinetic parameters and, therefore, is not expected to perturb the peptide binding (7). Conversely, lysine acetylation with consequent removal of the positive charge is predicted to diminish the favorable interaction with the surrounding negatively charged protein environment, accounting for the 6-fold increase in K m caused by pLys 9 acetylation and the lack of LSD1 activity on hyperacetylated peptides and nucleosomes (9,20). Taken together, these structural data explain the unique combination of specificity and affinity of LSD1 for deacetylated/dephosphorylated H3 that enables the enzyme to demethylate Lys 4 and to have non-enzymatic functions in the recruitment of other chromatin-modifying proteins and protein complexes (6,21,22).
The co-repressor protein CoREST is necessary to make LSD1 able to act on intact nucleosomal particles (8). Moreover, CoR-EST-bound LSD1 exhibits a 2-fold increase in the rate of catalysis measured with peptide substrate ( Table 1). The three-dimensional structure offers a clue to understanding these functional properties, indicating that the co-repressor indirectly takes part in substrate binding (Fig. 2, A and C). In particular, residues 14 -16 of the histone peptide are in contact with helix 372-395 (Figs. 1A and 2A) of LSD1, which in turn extensively interacts with CoREST. The shortest distance between a CoREST residue and the peptide is 7.2 Å (Pro 311 -pGly 13 ). These structural features combined with the functional data suggest that CoREST may have a dual role in optimizing LSD1 activity; it contributes to the overall architecture of the peptide-binding site and is very likely to interact physically with the nucleosomal particles, thus facilitating their recognition by LSD1.