Molecular Mimicry and Ligand Recognition in Binding and Catalysis by the Histone Demethylase LSD1 – CoREST Complex

SUMMARY

Histone demethylases LSD1 and LSD2 (KDM1A/B) catalyze the oxidative demethylation of Lys4 of histone H3. We employed molecular dynamics simulations to probe the diffusion of the oxygen substrate. Oxygen can reach the catalytic centre independently from the presence of a bound histone peptide, implying that LSD1 can complete subsequent demethylation cycles without detaching from the nucleosomal particle. The simulations highlight the role of a strictly conserved active-site Lys residue providing general insight into the enzymatic mechanism of oxygen-reacting flavoenzymes. The crystal structure of LSD1-CoREST bound to a peptide of the transcription factor SNAIL1 unravels a fascinating example of molecular mimicry. The SNAIL1 N-terminal residues bind to the enzyme active-site cleft, effectively mimicking the H3 tail. This finding predicts that other members of the SNAIL/Scratch transcription factor family might associate to LSD1/2. The combination of selective histone-modifying activity with the distinct recognition mechanisms underlies the biological complexity of LSD1/2.

INTRODUCTION

Large chromatin complexes finely regulate eukaryotic gene expression and are selectively recruited to DNA sequences by specific transcription factors. The posttranslational modifications on the histone N-terminal tails protruding from the nucleosomal particle play fundamental roles in gene expression by dictating an epigenetic code that flags the activation or repression status of a gene (Jenuwein and Allis, 2001). These histone marks are recognized by transcription factors and are dynamically regulated by specific histone-modifying enzymes (Ruthenburg et al., 2007).

Methylation of histone Lys residues is catalyzed by histone methyl-transferases, a process thought to be irreversible for decades. This view was challenged by the discovery of the first histone lysine demethylase, Lysine-Specific Demethylase 1 (LSD1 or KDM1A, according to the newly adopted nomenclature) (Shi et al., 2004; Forneris et al., 2005). This enzyme acts on mono- and di-methylated Lys4 of histone H3 through a FAD-dependent oxidative process (Fig.1A). LSD1 is often associated to the histone deacetylases (HDAC) 1 and 2 and to the corepressor protein CoREST, which tightly binds to LSD1 enhancing both its stability and enzymatic activity. A number of studies published in the past two-three years have indicated that LSD1-CoREST interacts with various protein complexes involved in gene regulation and chromatin modification (see Forneris et al., 2008; Mosammaparast and Shi, 2010 and references therein). Especially relevant for our investigations is the finding that LSD1 is recruited to target gene promoters by interacting with the N-terminal SNAG domain of SNAIL1, a master regulator of the epithelial-mesenchymal transition, which is at the heart of many morphogenetic events including the establishment of tumor invasiveness (Lin et al., 2010) (Fig. 1B). A further addition to this biological complexity has been the discovery of LSD2 (or KDM1B), a mammalian homolog of LSD1 (Karytinos et al., 2009). LSD2 exhibits the same H3-Lys4 demethylase activity as LSD1 but it functions in distinct transcriptional complexes with specific biological functions (Ciccone et al., 2009; Fang et al., 2010; Yang et al., 2010).

Figure 1.

Histone demethylation by LSD1. (A) Scheme of the LSD1-catalyzed amine oxidation reaction. This enzyme acts on mono- and di-methylated Lys4 of histone H3 and is able to subsequently remove two methyl groups from a dimethylated substrate. The reactants involved in the simulated system are highlighted in red and correspond to the protein-bound FADH⁻ (Fig. 2), the histone H3 N-terminal peptide (Fig. 1B), and O₂. (B) Sequence alignment of the N-terminal residues of histone H3 and the N-terminal sequences of an evolutionary-related family of C₂H₂ zinc-finger transcription factors which includes SNAIL1 (Barrallo-Gimeno and Nieto, 2009). OVO-like1 is an epidermal proliferation/differentiation factor homologous to a protein originally identified in D. melanogaster ovary cells (Nair et al., 2006). Scratch proteins are expressed in the nervous system in both developing and adult cells (Marín and Nieto, 2006). Growth factor independence 1 (gfi1) is a gene repressor involved in hematopoiesis whose expression was already demonstrated to be regulated by LSD1-containing complexes (Saleque et al., 2007). Insulinoma-associated 1 (insm1) was originally isolated from neuroendocrine tumour cells and data suggest a role in differentiation of both neural and pancreatic precursors (Lukowski et al., 2006). Conserved residues are highlighted in green (identity) or magenta (similarity). Lys4 of LSD1 is indicated in blue.

LSD1 and LSD2 are distinguished from the histone demethylases of JmjC class that have been identified in the past years. The JmjC enzymes display wider substrate specificity by acting on mono-, di- and/or tri-methylated Lys residues and function through an iron-dependent catalytic mechanism that produces formaldehyde as side product (Horton et al., 2010; Tsukada et al., 2006). Conversely, LSD1/2 are flavoenzymes that use FAD to oxidatively demethylate their substrate. The reduced flavin, generated upon Lys demethylation, is re-oxidized by molecular oxygen (O₂) with the production of hydrogen peroxide in addition to formaldehyde (Figs. 1A and 2) (Forneris et al., 2005; Shi et al., 2004). This peculiar hydrogen-peroxide generating activity of LSD1/2 represents an important aspect because the reaction product and its reactive oxygen species are potentially dangerous in the context of the chromatin environment. An intriguing hypothesis is that they might have a signaling role in cellular processes, further expanding the biological roles and functions of LSD1/2 (Amente et al., 2010; Forneris et al., 2008; Winterbourn, 2008).

Figure 2.

Chemical representation of the reduced flavin adenine dinucleotide (FADH⁻) molecule. The C4a and C5a atoms relevant for the present study are labeled. The flavin is facing the viewer with its re-side (the si-side is on the opposite side). See also Figure S5.

Here, we study the mechanisms and processes of molecular recognition in LSD1 with a focus on the recognition of substrates and protein ligands. How does oxygen diffuse into the catalytic site? Do the very different enzyme substrates, a small molecule such as oxygen and a large protein complex such as the nucleosome, interfere with each other? What are the molecular mechanisms underlying the ability of LSD1/2 to interact and recruit many distinct protein partners and does binding to these partners affect substrate recognition (Forneris et al., 2008; Mosammaparast and Shi, 2010)? In order to advance our understanding about these points, we have investigated the crystal structure of LSD1-CoREST in complex with an N-terminal peptide of human SNAIL1 (Fig. 1B). Furthermore, we have probed the mechanisms of O₂ diffusion in LSD1-CoREST using molecular dynamics (MD) simulations. Our studies provide a molecular framework for the processivity of LSD1, which is able to catalyze the subsequent removal of two methyl groups from dimethylated H3-Lys4 residue (Fig. 1A). They also highlight the factors that determine peptide recognition and predict that a whole class of transcription factors is likely to employ a most unusual and fascinating “histone-mimicking” mechanism for binding to LSD1.

RESULTS

Structural analysis of SNAIL1 recognition by LSD1

We employed a combination of biochemical and structural experiments to investigate the interaction between LSD1 and the transcription factor SNAIL1, which was originally discovered by Lin et al (Lin et al., 2010). On the basis of a weak sequence similarity between SNAIL1 and the histone H3 N-terminal tail (in essence, a conserved pattern of positively charged residues; Fig. 1B), the authors of this study proposed that SNAIL1 could bind to LSD1 in the same site as the histone H3 substrate, i.e. in the catalytic site. This model for SNAIL1-LSD1 interactions implied that SNAIL1 should inhibit LSD1 enzymatic activity. Consistently, we found that a 20-amino-acid peptide corresponding to SNAIL1 N-terminal residues effectively inhibits LSD1-CoREST. In more detail, fitting of the enzyme initial velocities to the equation for competitive inhibition resulted in a K_i values of 0.21 ± 0.07 mM (using a mono-methylLys4 H3-peptide as substrate) and 0.22 ± 0.09 mM (using a di-methylated substrate), indicating a rather tight binding. Likewise, we found that also LSD2 binds the SNAIL1 peptide (albeit with lower affinity; K_i value of 2.22 ± 0.36 µM), which is in line with the previously observed similarities in the binding properties of LSD1 and LSD2 (Karytinos et al., 2009). Thus, the possibility exists that, in addition to LSD1, SNAIL1 might interact and represent a protein partner also for LSD2.

In order to dissect the mechanism of SNAIL1 recognition by LSD1, we soaked LSD1-CoREST crystals in a solution containing the SNAIL1 peptide used for the inhibition studies (Fig. 1B) and we determined the crystal structure of the ternary complex at 3.0 Å resolution (Figs. 3A and 3B; Table 1). Inspection of the unbiased electron density map (Fig. 3B) allowed us to trace the polypeptide chain for the N-terminal nine residues of the SNAIL1 sequence whereas residues 10–20 of the peptide could not be identified in the electron density most likely because they lack an ordered conformation. Peptide binding does not induce any conformational change in the active site as compared to the ligand-free structure (root-mean-square deviation RMSD = 0.4 Å for the 799 equivalent C^α atoms of LSD1-CoREST). The peptide occupies the open cleft that has been shown to form the binding site for the H3 tail (Forneris et al., 2007). The N-terminus (residues 1–4) adopts a helical turn conformation, closely resembling that of the first residues of the H3 tail (see Figs. 3C and 3D). In particular, the N-terminal amino group of Pro1 and the side chains of Arg2 and Ser3 bind deeply into the cleft and establish several H-bonding interactions with the surrounding protein residues, in an arrangement almost identical to that exhibited by the Ala1-Arg2-Thr3 residues of the H3 peptide. Thus, Pro1 binds to the carbonyl of Ala539 at the C-terminus of α-helix 524–540 whereas Arg2 interacts with Asp553 and Asp556. This binding conformation positions Phe4 of SNAIL1 in front of the flavin to occupy a location corresponding to that of Lys4 of the H3 tail (Fig. 3D). Phe4 snugly fits in its binding niche by making edge-to-face interactions with the rings of the flavin cofactor and Tyr761. The conformation of the SNAIL1 peptide deviates from that of the H3 tail after residues 4–5 to compensate for the deletion of one residue compared to the histone sequence (see Figs. 1B, 3C and 3D). However, the comparative analysis of the H3 and SNAIL1 complexes clearly indicates that Arg7 of SNAIL1 occupies the same position and establishes similar interactions (with Cys360, Asp375, and Glu379) as Arg8 of H3. Likewise, Lys8 of SNAIL1 falls in the same solvent-exposed position observed for Lys9 of H3. Taken together, these findings highlight three key points: (i) the cavity of LSD1 specifically recognizes the N-terminal amino group of the peptide ligands, (ii) the conserved pattern of positively charged groups and small hydroxyl side chains shared by SNAIL1 and H3 N-terminal tails enable them to bind to LSD1 in a similar conformation, and (iii) SNAIL1 N-terminal residues act as a mimic of H3 being able to effectively bind to the enzyme active-site cleft.

Figure 3.

X-ray structure of LSD1-CoREST in complex with the SNAIL1 peptide. (A) Overall ribbon representation of the ternary complex of LSD1 (cyan), CoREST (blue) and the SNAIL1 peptide (orange). The FAD cofactor is in yellow sticks. (B) Fitting of the SNAIL1 peptide (carbons in orange) into the unbiased electron density map contoured at 1.2 σ calculated with weighted 2Fo-Fc coefficients. Color-coding and orientation are as in panel A. (C) Binding of the SNAIL1 peptide in the LSD1 active site. Color-coding and orientation are as in panel A with H-bonds shown as dashed lines. Red labels are used to highlight residues in direct contact with oxygen pathways in O₂-bound simulations. (D) Superposition between SNAIL1 (orange) and histone H3 (gray) peptides. Conserved peptide side chains (Fig. 1B) are drawn in stick representation. The FAD cofactor and the Lys661 side chain are shown as reference. The water bridging Lys661 and flavin N5 (as observed in the unbound enzyme structure that has been solved at higher resolution) are also displayed (Yang et al., 2006). See also Figure S1.

Table 1.

Data collection and refinement statistics for LSD1-CoREST- SNAIL1 peptide complex

Space group	I222
Unit cell axes (Å)	119.2, 181.5, 234.4
Resolution (Å)	3.0
Rsyma,b (%)	9.8 (63.7)
Completenessb (%)	99.7 (100.0)
Unique reflections	50,937
Redundancy	3.6 (3.7)
I/σb	9.5 (1.9)
N° of atoms protein/FAD/ligandc	6286/53/77
Average B value for ligand atoms (Å2)	75.2
Rcrystd (%)	21.2
Rfreed (%)	24.6
Rms bond length (Å)	0.014
Rms bond angles (°)	1.49

^aR_sym=∑|I_i−<I>|/∑I_i, where I_i is the intensity of i^th observation and <I> is the mean intensity of the reflection.

^bValues in parentheses are for reflections in the highest resolution shell.

^cThe final model consists of residues 171–836 of LSD1, a FAD molecule, residues 308–440 of CoREST, and residues 1–9 of the SNAIL1 peptide.

^dR_cryst=∑|F_obs−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 2V1D.

Conformational dynamics of free and substrate-bound LSD1-CoREST complex systems

We extended our studies on peptide recognition in LSD1-CoREST by analyzing conformational dynamics of the LSD1-CoREST system by MD simulations in solution. This study was based on two 50-ns-long MD trajectories using LSD1-CoREST crystallographic coordinates (Yang et al., 2006) in the ligand-free state (hereafter referred to as “unbound”) and in complex (“bound”) with the histone H3 peptide used in the crystallographic studies (Fig. 3) in which Lys4 is replaced by Met (Lys4Met mutation was previously shown to greatly increase binding affinity; Forneris et al., 2007). Figure 4 summarizes the analysis of the backbone C^α atom-positional RMSD values of LSD1-CoREST MD trajectory snapshots from the X-ray structures. After a first initial phase of about 4 ns – during which the most flexible loop regions relax in solution – RMSDs fluctuate stably. Overall, CoREST and the LSD1 tower domain is somewhat more flexible compared to LSD1, largely owing to relative motion between the CoREST SANT2 domain and the LSD1 tower domain (Fig. 3A). Differences in molecular fluctuations between LSD1 unbound and bound are moderate (Fig. S1) and changes upon substrate binding are limited to the side chain motion of residues in direct contact with the H3 peptide (not shown). These observations are further supported by analysis along the trajectories of the backbone C^α atoms of the scaled size-independent similarity parameter (Maiorov and Crippen, 1995) as described in Supplemental Information (Fig. S2). The ensemble averaged ρ_SC values are 0.2 and 0.3 for LSD1 and CoREST, respectively, with no differences between unbound and bound simulations.

Figure 4.

Backbone C^α atom-positional RMSD of LSD1-CoREST MD trajectory snapshots. Time series (left panels) and normalized probability distributions (right panels) of the RMSD values were calculated either for all C^α atoms of the complex (top panels) or for the individual LSD1 or CoREST subunits (middle and bottom panels) from both unbound and bound 50 ns simulations using the X-ray structures as reference (Yang et al., 2006; Forneris et al., 2007). Structures were superimposed using all backbone C^α atoms for which RMSD values are reported. Corresponding time series and distributions of the scaled size-independent similarity parameter (Maiorov and Crippen, 1995) are reported as Supplemental Information, Fig. S2.

This first set of simulations indicated that binding of the histone H3 N-terminal peptide has only local packing effects. No significant impact on the overall dynamics of the LSD1-CoREST complex as well as of LSD1 and CoREST individual partners was observed over the simulated timescales. The binding site for the histone tail on LSD1-CoREST is essentially pre-organized to host and stabilize the peptide in the observed folded conformation (Forneris et al., 2007). This is consistent with the X-ray data on the SNAIL1 complex. Also in that case, the bound peptide seems to adapt its conformation to the shape of the binding cleft whose H-bond acceptors function as anchoring elements for the positively charged groups of the peptide (Figs. 3C and 3D).

Oxygen diffusion into free and substrate-bound LSD1-CoREST complex

Having inspected the structure and dynamics of peptide binding, we investigated the diffusion of O₂ molecules into the active site of LSD1-CoREST to probe the mechanisms of oxygen migration and binding in this enzymatic system (Fig. 1A). The integration of biochemical and structural experiments with powerful MD simulations has already shown to be an effective strategy to investigate the relationship between enzyme dynamics and oxygen biocatalysis (Baron et al., 2009a; Baron et al., 2009b; Saam et al., 2007; Saam et al., 2010). Our computational approach intended to address a crucial question about LSD1 molecular function: does binding of the histone peptide in the active site cleft interfere with the diffusion of molecular oxygen into the active center? Two sets of five independent MD simulations were initialized based – respectively – on the X-ray structures of the unbound and histone peptide-bound LSD1-CoREST complexes and were analyzed in order to probe the effect of peptide binding on oxygen diffusion. The FAD was set to be in the reduced FAD form (FADH⁻; Fig. 2) as it is the case for the enzymatic state that undergoes re-oxidation following Lys demethylation (Fig. 1A). After equilibration, 100 O₂ molecules were added to the bulk water in each system according to a non-arbitrary procedure that avoids biasing their starting positions. We name these ten simulation runs “O₂-unbound” and “O₂-bound” (peptide-free and peptide-bound, respectively) and distinguish them by color-coding in Figures 5 and 6. These runs were propagated for different simulation periods, till any O₂ molecule was observed entering the LSD1 active site and arriving in close contact with the flavin, and were terminated when such O₂ molecule would leave the active site (Fig. 5). Paths carrying O₂ molecules inside the protein though not in proximity of the flavin (distance >7 Å) were observed as well, but were excluded from further analysis. No relevant differences were observed on the simulated timescales between conditions of O₂-saturation and absence of oxygen regarding LSD1-CoREST dynamics and deviation from the X-ray reference structures (Yang et al., 2006; Forneris et al., 2007) (see also Supplemental Information, Fig. S3). Out of a total of 500 O₂ trajectories per system, these simulations allowed collecting 10 complete spontaneous diffusion pathways that bring O₂ molecules at close distance from FADH⁻ starting from random configurations in the bulk solvent (Fig. 5). A total of about 83 and 50 ns MD sampling times was required for O₂-unbound and O₂-bound simulations, respectively, to observe a total of ten spontaneous diffusion routes. Successful paths typically cover overall distances of about 20–60 Å from the protein surface to the flavin and display a stepwise behavior from the time of entrance through the protein surface (Fig. 6). In fact, O₂ molecules may temporarily reside on the surface and/or generally visit several cavities along each of these paths. Similar diffusion routes were observed in the unbound and bound simulations (not shown), suggesting that H3 peptide binding has a minor effect on the migration of O₂ molecules in full agreement with the kinetic data (Forneris et al., 2006).

Figure 5.

Successful oxygen diffusion paths in the LSD1-CoREST complex. Time series of the O₂ – C4a distance are displayed with different colors (green, yellow, red, blue, and black) for independent MD simulations in O₂-saturated conditions (O₂-unbound and O₂-bound). The dashed horizontal line defines the flavin as a sphere of 7 Å radius centered on C4a. See Fig. 2 for the atom numbering of FADH⁻. See also Figure S3.

Figure 6.

O₂ spontaneous diffusion into the bound LSD1-CoREST complex. (A) Overall location, (B) top view, and (C) side view of the O2 diffusion pathways displayed with color-coding corresponding that of the right panel of Fig. 5. For graphical purposes the blue pathway is displayed in Supplemental Information, Fig. S4. The bound histone H3 tail (gray coil) and LSD1 (cyan coil) are also shown. (D) Side view of the time-dependent representations along the simulation time: O₂ molecules are colored from red (entrance into the protein matrix) to blue (exit from the active site). All displayed paths conduct O₂ molecules to the C4a-N5 locus of the FADH⁻ reduced flavin cofactor (yellow sticks). Residues Lys661 (orange sticks) and Tyr761 (purple sticks) are also highlighted.

Insights about the reaction of LSD1-CoREST with oxygen

Can we identify an orientation model for the approach of O₂ molecules to the flavin cofactor? Answering this question is of basic importance to understand the mechanism of the oxidative half-reaction that follows Lys demethylation (Fig. 1A). To understand preferential models for the approach of O₂ molecules to the reduced flavin cofactor, we analyzed the successful O₂ diffusion pathways employing the statistics obtained from all ten independent simulations. In eight of them (four O₂-unbound + four O₂-bound), O₂ diffusion pathways clearly converge above the re-face of FADH⁻ where Lys661 is located (Fig. 6 and Supplemental Information, Fig. S4). We find that oxygen molecules diffuse towards the flavin edge and, once within 5–7 Å distance from the flavin C4a-N5-C5a atoms (Fig. 2), they transiently displace the water that bridges the Lys661 side chain to flavin N5 (Fig. 3D). In this way, oxygen molecules can intercalate between the Lys661 side-chain amino group and bound H3 peptide to directly contact (~4–5 Å distances) either the edge or the re face of the cofactor, depending on the trajectory (Figs. 2 and 6). The statistical analysis of the O₂ - FADH⁻ encounter events further indicates a shift towards smaller values of the C4a - O₂ distances with respect to the corresponding values for C5a atom, implying that C4a is the site for preferential (but not exclusive) approach (Fig. S5).

The MD simulations highlight the role of Lys661 as the “entry residue” for oxygen into the active centre. This amino acid is engaged in a water-mediated interaction with the flavin ring (Fig. 3D). This peculiar Lys-water-flavin triad is a highly conserved feature, which characterizes the amine oxidase flavoenzymes that share the same folding topology as the catalytic domain of LSD1 (Binda et al., 2002). Lys661 is known to be essential for enzyme function in LSD1 and polyamine oxidase (Shi et al., 2004; Polticelli et al., 2005). Moreover, studies on sarcosine oxidase indicated that mutations of this Lys to neutral residues drastically reduce (up to 9000-fold) the reactivity of reduced flavin with oxygen (Jorns et al., 2010; Zhao et al., 2008). Taken together, these data suggest a twofold role of the conserved Lys residue in oxygen reaction: (i) a gating function to channel oxygen molecules to the catalytic site (Baron et al., 2009a), and (ii) a catalytic function in oxygen activation possibly through the electrostatic stabilization of the superoxide-flavin semiquinone pair, which is thought to transiently form during the electron-transfer process underlying flavin re-oxidation (Mattevi, 2006; Jorns et al., 2010).

DISCUSSION

LSD1 was originally identified as H3-Ly4 demethylase and this is the only activity that, so-far, has been demonstrated in vitro (Forneris et al., 2008; van Essen et al., 2010). The three-dimensional structure of the enzyme in complex with an H3 N-terminal peptide (Forneris et al., 2007) displayed a substrate binding mode which is fully consistent with this specificity. However, several in vivo evidences indicated that LSD1 might also act on H3-Lys9 as well as on a non-histone substrate such as p53 and that a switch in substrate specificity might be promoted by the association with other proteins (Nair et al., 2010; Huang et al. 2007). Thus, probing the binding properties and conformational dynamics is especially relevant to fully understand the biology of LSD1. The discovery of the tight association between LSD1 and N-terminal residues of the SNAIL1 transcription factor was particularly insightful because of the peculiar mode of association proposed to underlie this protein complex (Lin et al., 2010). Furthermore, we could see that this interaction occurs also in vitro using recombinant proteins, paving the way to our molecular dynamics and structural investigations. These studies demonstrate that the active site of LSD1-CoREST complex can bind the N-terminal sequences of other proteins, such as SNAIL1, that are only distantly related to H3. In more detail, they identify the elements that determine peptide recognition: (i) the conformation of the bound peptide whose N-terminal helical-turn binds deeply into the LSD1 active site, (ii) the conserved pattern of positively charged and small polar side chains which are all involved in specific H-bond interactions, (iii) the peptide N-terminus which is specifically recognized by the enzyme, and (iv) the high degree of pre-organization of the binding cleft as indicated by the lack of detectable conformational changes in LSD1-CoREST crystal structure upon peptide binding as well as be the overall stability of the active-site conformation in all MD simulations (Figs. 1B, 3C, and 3D). The emerging general conclusion is that the conformation of LSD1 associated to CoREST is tailored for binding the N-terminal residues of a peptide ligand with the side chain of its fourth amino acid pointing towards and in direct contact with the flavin ring of the cofactor. It remains to be seen if association of LSD1 with other protein partners can induce conformational changes that switch the binding specificity of the enzyme.

The comparative analysis of the SNAIL1 and H3 peptide binding reveals a further intriguing feature: the folded conformation adopted by the peptides shields the flavin from the solvent by forming a sort of plug that occludes the active site cleft (Fig.3) (Forneris et al., 2007; Yang et al., 2007). This observation raises the question of the possible interference exerted by the bound peptide on the diffusion and reaction of molecular oxygen, the electron-acceptor substrate required to regenerate the oxidized state of the flavin (Fig. 1A). To investigate this point, we used MD to simulate oxygen diffusion. The key observation from these studies is that oxygen reaches the active site following multiple routes that converge to a residue which acts as entry point for admission of oxygen into the catalytic site (Lys661; see Figs. 3 and 6). This concept of “multiple pathways to an entry point” is emerging as the common feature highlighted by computational and biophysical studies on several oxygen-reacting enzymes and finds support in mutagenesis data (Baron et al., 2009a; Jorns et al., 2010). Most importantly, in the case of LSD1, this model for diffusion implies that O₂ can reach the flavin cofactor even in the presence of the bound H3 peptide as clearly confirmed by the MD simulations of oxygen trajectories in the peptide-bound enzyme (Figs. 5–7). This feature becomes very insightful when analyzed in light of the kinetic properties of the LSD1-CoREST reaction, showing that the reactivity of oxygen with the reduced flavin is virtually unaffected by peptide binding and that the Km values for the mono- and di-methylated substrates are very similar (Shi etal., 2004; Forneris et al., 2005; Forneris et al., 2006). Taken together, these kinetic and computational data strongly suggest that that LSD1 can function processively – i.e. the possibility for oxygen to bind and react while a peptide is bound to the active site, enabling the enzyme to complete subsequent demethylation cycles without detaching from the nucleosomal particle.

The binding mode exhibited by the SNAIL1 peptide has also far-reaching implications for the functions of LSD1-CoREST and LSD2 in chromatin biology. SNAIL1 is part of a large family of transcription factors with key roles in development and oncogenesis (the so-called SNAIL/Scratch superfamily; Fig. 1B) (Barrallo-Gimeno and Nieto, 2009; Lin et al., 2010; Marín and Nieto, 2006). Alignment of the N-terminal sequences of their common SNAG domain reveals that the residues that are key for binding to LSD1, as gathered from the SNAIL1-LSD1-CoREST crystal structure, are conserved among these SNAIL1-related proteins. This feature predicts that other (if not all) transcription factors of the SNAIL/Scratch family are likely to associate to LSD1 (and possibly LSD2) following the same molecular mechanism highlighted by SNAIL1. Consistently, a member of this transcription factor family (gfi; Fig. 1B) has been already shown to exert its regulatory roles in hematopoiesis partly through the recruitment of LSD1 to specific target DNA sequences (Saleque et al., 2007). An intriguing feature of this mode of function – mimicking the histone peptide – is that it makes these proteins inherently able to inhibit LSD1/2. The competitive inhibition constant K_i measured for the SNAIL1 peptide (~0.2 µM) indicates a relatively tight binding when compared to the Km values of 3–4 µM the methylated H3 peptides (Forneris et al., 2006). Clearly, the N-terminal residues of SNAIL1 must be released from the LSD1 active site to make the enzyme catalytically active. Consistently, Lin et al. have shown that, in vivo, nucleosomes can indeed displace SNAIL1 so that, after being targeted to the SNAIL1-binding DNA sequences, LSD1 can exert its demethylase activity (Lin et al., 2010). An obvious fascinating question for future studies will be to evaluate the potential regulatory role played by SNAIL1 and related transcription factors as endogenous inhibitors of LSD1 and LSD2.

The mix of flexibility and processivity highlighted by our studies provides a molecular rationale for the dual (catalysis versus binding) activity of LSD1-CoREST which can act as a histone-modifying enzyme as well as multi-docking element for a number of functionally and structurally distinct protein partners (Forneris et al., 2008; Mosammaparast and Shi, 2010; Shi et al., 2004). The (demethylated) nucleosomal particle(s) and transcription factors such as SNAIL1 have their primary anchoring site in the active site cleft whereas other proteins (such as HDACs) will associate to other regions of the LSD1-CoREST complex (possibly including the tower domain). The challenge for the future will be to translate these findings to the complexity of the crowded chromatin environment.

EXPERIMENTAL PROCEDURES

Biochemical and crystallographic studies

All chemicals were purchased from Sigma-Aldrich unless specified. E. coli over-expression, purification and crystallization of the human LSD1-CoREST complex were carried out following previously published procedures (Forneris et al., 2007). A His-tagged recombinant form of LSD1 comprising residues 171–836 was co-purified with a GST-tagged CoREST protein (residues 308–440) by tandem-affinity chromatography followed by gel filtration on a Superdex200 column (GE Healthcare). Recombinant mouse LSD2 was purified as described (Karytinos et al., 2009). The enzymatic activity of purified and homogeneous samples LSD1-CoREST and LSD2 were measured on a 21-residue H3 peptides mono- and di-methylated at Lys4 (Thermo Electron Corporation) by the peroxidase-coupled assay at 25 °C using a Cary 100 UV/Vis spectrophotometer (Varian Inc.) (Forneris et al., 2005). Briefly, reactions were started by adding 2 µl of protein solution (40 µM protein in 50 mM potassium phosphate buffer pH 7.5 and 5% (w/v) glycerol) to reaction mixtures (150 µl) consisting of 50 mM Hepes/NaOH buffer pH 7.5, 0.1 mM 4-aminoantipyrine, 1 mM 3,5-dichloro-2-hydroxybenzenesulfonic acid, 0.35 µM horseradish peroxidase, and variable concentrations (2–100 µM) of mono- or di-methylated H3-K4 peptides. The 20-residue SNAIL1 peptide (Fig. 1B) (Thermo Electron Corporation) was tested in inhibition studies by enzymatic assays in the presence of varied concentrations (2–100 µM) of H3 substrates. The best fit was obtained with the equation describing a competitive inhibition.

Crystals of LSD1-CoREST were grown as previously published (Forneris et al., 2007) and were soaked in a cryoprotectant solution containing 0.3 mM SNAIL1 peptide for 3 h and then flash-cooled in a stream of gaseous nitrogen at 100 K. Crystals were obtained by both co-crystallization and soaking and many of them had to be screened at beam-lines of SOLEIL, ESRF, SLS synchrotrons due to poorly reproducible diffraction. The best data set was measured on a crystal obtained by soaking. Data processing and scaling (Table 1) were carried out using MOSFLM (Leslie, 1999) and programs of the CCP4 package (1994). The structure of the LSD1-CoREST complex (Yang et al., 2006) (PDB entry 2IW5) was used as initial model for refinement after removal of all water atoms. Unbiased 2Fo-Fc and Fo-Fc maps were used to manually build the protein-bound peptide inhibitor (Fig. 3B). Crystallographic refinement (Table 1) was performed with Refmac5 (Murshudov et al., 1997) and manual re-building was done with Coot (Emsley and Cowtan, 2004). Pictures were produced with PyMol (www.pymol.org). Crystallographic coordinates have been deposited with the Protein Data Bank (2y48).

Molecular Model and Computational Procedure

Conformational dynamics in solution was investigated for two (peptide-bound and peptide-free) LSD1-CoREST complex systems non-covalently bound to FADH⁻. The spontaneous, explicit diffusion of dioxygen in these two systems was separately addressed based on ten independent MD simulations in O₂-saturated conditions (O₂-unbound and O₂-bound), a successful approach for oxygen-consuming enzymes (Baron et al., 2009a; Baron et al., 2009b). Note that in the present study O₂ molecules have physical masses, thus MD timescales are realistic. Initial configurations for the LSD1-CoREST-FADH⁻ complexes, the ions, and the crystallographic water sites were taken from PDB 2IW5 (Yang et al., 2006); H3 peptide coordinates from PDB 2V1D (Forneris et al., 2007). Initial configurations were solvated in (pre-equilibrated) boxes large enough to avoid interactions between mirror images under rectangular periodic boundary conditions along the entire MD trajectory. For O₂-unbound and O₂-bound simulations, 100 O₂-molecules were substituted (after 5 ns equilibration) to randomly chosen water molecules (enforcing minimum reciprocal distances and minimum distances from any protein atom of 1.0 nm). All systems were neutralized with (unbound: 2; bound: 6) Cl⁻ ions and contained (unbound: 60106; bound: 60093; O₂-unbound: 60006; O₂-bound: 59993) water molecules.

All trajectories in explicit water were generated and analyzed in double precision using the GROMACS 4.0.4 software (Hess et al., 2008). Force field parameters and charges were set from the 53A6 GROMOS force field (Oostenbrink et al., 2004) to reproduce the experimental condition of apparent neutral pH. GROMOS compatible SPC water model (Berendsen, 1981) and ion parameters (Åqvist, 1990) were employed. Newton's equations of motion were integrated using the leap-frog algorithm (Ryckaert et al., 1977) with a 2 fs time step. The P-LINCS numerical algorithm (Hess et al., 2008) was applied to constrain all bond lengths, excluding water molecules which were kept rigid using the SETTLE analytical algorithm (Miyamoto and Kollman, 1992). All simulations were carried out in the N,p,T canonical ensemble (300 K; 1 Atm) by separately coupling the temperature of LSD1, COREST, FADH⁻, solvent (and O₂) degrees of freedom to a heat bath using the Nosé-Hoover thermostat (Hoover, 1989; Nosé, 1984) and the Parrinello-Rahman pressure coupling (Parrinello and Rahman, 1981). Full treatment of electrostatic interactions was achieved using a smooth Particle Mesh Ewald approximation (Essmann et al., 1995). Non-bonded interactions were shifted (Baron et al., 2007) to zero at a distance of 1.4 nm, re-calculated every time-step in the range 0.0 – 0.8 nm and every five time-steps in the range 0.8 – 1.4 nm, using a twin-range cutoff scheme (van Gunsteren and Berendsen, 1990). Trajectory snapshots were extracted every 10 ps for analysis along 50 ns of unbound or bound simulations or periods of varying length of ten O₂-unbound and O₂-bound simulations (totaling about 83 and 50 ns sampling, respectively). O₂-diffusion pathways to the FADH⁻ cofactor were analyzed as in Ref. (Baron et al., 2009b) with the VMD software (Humphrey et al., 1996). Additional computational details are reported as Supplemental Information.

Abbreviations

LSD1 (KDM1A)

Lysine-Specific Demethylase 1

LSD2 (KDM1B)

Lysine-Specific Demethylase 2

HDAC

histone deacetylase

MD

molecular dynamics

RMSD

root-mean-square deviation

FADH⁻

reduced FAD

O₂-unbound and O₂-bound

MD simulations in O₂-saturated conditions of peptide-free and peptide-bound LSD1-CoREST.