Introduction

In eukaryotic cells, DNA is packaged within the complex and organized structure of chromatin. The basic unit of chromatin is the nucleosome, which is a compact core of four histones (H2A, H2B, H3 and H4, each one present in two copies forming an octamer) surrounded by a 147 bp stretch of DNA [1]. The histone N-terminal tails bear a number of sites for post-translational modifications that have a direct role in modulating gene expression. The term ‘epigenetics’ refers to the mechanisms that regulate gene transcription through the read-out of these covalent modifications on the histone tail. This epigenetic control is achieved either by simply modulating the accessibility of DNA or by recognition of histone post-translational modifications through specific protein domains. In particular, the lysine residues of histone tails are subject to both acetylation and methylation, and the meaning of such epigenetic marks depends on the site where they occur. Histone lysine acetylation is often associated with gene activation. In some cases, this condition results from the neutralizing effect of acetyl groups on the lysine positive charge, which weakens the electrostatic interaction between the histone tail and the DNA phosphate moiety, with a resulting loosening of the nucleosome compactness [2]. In addition, lysine acetylation at particular sites on the histone tail is specifically recognized by chromatin-interacting proteins such as the bromodomains [3]. Histone lysine methylation may signal either the activation or repression of gene expression depending on the site of methylation [4]. These histone post-translational modifications represent platforms for the binding of protein modules that recruit or instruct effector proteins regulating transcriptional activity [5].

Although histone lysine methylation has been known for decades, it was not until 2004 that the first human histone demethylase lysine-specific demethylase 1 (LSD1) was identified [6,7]. This led to the discovery of other human histone demethylases, the so-called JmjC domain-containing proteins, which revolutionized the concept of histone methylation as a dynamically regulated process, rather than a permanent epigenetic mark [8]. Moreover, orthologs of LSD1 in Caenorhabditis elegans, Arabidopsis thaliana and Drosophila melanogaster were identified and extensively investigated [9–11]. The ever-growing number of chromatin-remodeling enzymes led scientists in the field to create an organized nomenclature [12]. According to these guidelines, histone lysine demethylases are named K-demethylases and the LSD1-like enzymes are named K-demethylase 1 proteins. This minireview focuses on LSD1 without involving its orthologs or JmjC demethylases; the acronym LSD1 is adopted for consistency with previous publications on the same subject.

LSD1 specifically removes one or two methyl groups from Lys4 of histone H3 via a FAD-dependent reaction (Fig. 1). LSD1 activity can be measured in vitro on free histones and on a peptide of at least 21 amino acids corresponding to the N-terminal tail of histone H3 [6]. LSD1 is not active on trimethylated Lys4, which is consistent with the flavin-catalyzed amine oxidation reaction that requires a lone pair of electrons on the lysine amino group. It has been shown that the presence of a second post-translational modification on the same histone H3 peptide dramatically reduces LSD1 activity on methylated Lys4 [13]. This demonstrates that LSD1 is capable of reading the histone code and suggests a timing scheme for gene repression in which this enzyme removes the last activation mark [14,15]. The crystal structure of the protein in complex with a histone H3 peptide (Fig. 2) provides an explanation for this highly specific recognition mechanism, which is accomplished through an intricate network of electrostatic interactions between a long stretch of the histone tail and the active site residues [16,17]. This structural feature also accounts for the inhibitory effect of an unmodified histone peptide on LSD1 activity and suggests that the affinity of this enzyme for the histone tail may have a role in docking LSD1-associated co-factors to the nucleosome.

LSD1 was originally identified as part of the multiprotein repressor complex coordinated by the repressor element 1-silencing transcription (REST, also known as the neuron-restrictive silencing factor) factor in the regulation of neuronal genes [18]. Methylation on Lys4 of histone H3 is a gene-activation mark and LSD1 activity contributes to transcriptional repression of neuronal genes by removing this post-translational modification. LSD1 interacts with CoREST, a co-repressor protein that binds REST and recruits other histone-modifying enzymes such as histone deacetylases 1/2. The function of the LSD1–CoREST–histone deacetylase subcomplex in transcriptional repression events is not limited to REST-regulated neuronal genes, but can be extended to other contexts such as hematopoietic differentiation [19] and the telomerase reverse transcriptase genes [20]. Moreover, a very recent study showed that, in C. elegans, patterns of dimethylated Lys4 of histone H3 serve as epigenetic memory to maintain the transcriptional program during cell division, and LSD1 was shown to be essential in resetting this memory in the germline in order to ensure pluripotency [21]. In parallel, a role for LSD1 in gene activation was originally discovered in the regulation of androgen receptor target genes [22] and was also identified in other hormone-dependent transcriptional programs [23]. LSD1 is suggested to promote these gene activation events by acting on methylated Lys9 of histone H3, an epigenetic mark associated with gene repression. However, biochemical and structural studies on the recombinant enzyme showed that LSD1 is not active on H3 peptides methylated at Lys9 [6,7] and it features a highly specific substrate-recognition mechanism [14,16]. Unquestionably, the involvement of LSD1 in an ever-growing number of transcriptional protein complexes emphasizes the central role of this flavoenzyme in the chromatin environment. LSD1 activity has also been linked to tumorigenesis events [24], which highlights this protein as a potential drug target.

The aim of this minireview is to dissect LSD1 domain organization in light of its role in transcriptional regulation. The minireview focuses on the amine oxidase moiety of this protein which is responsible for the catalytic activity on methylated Lys4 of histone H3 and will involve other flavin-dependent amine oxidases that are evolutionary related to LSD1.

LSD1 is a flavin-dependent amine oxidase

Amine oxidations are widespread reactions in nature. Flavin-dependent amine oxidases are enzymes that catalyze oxidative cleavage of the C–N bond by two-electron reduction of the FAD coenzyme, which produces an imine intermediate that is then hydrolyzed nonenzymatically [25] (Fig. 1). Reduced FAD can be reoxidized by molecular oxygen, which generates hydrogen peroxide and makes the enzyme available for a new catalytic cycle [26]. The overall globular structure of these enzymes was defined as an amine oxidase domain (AOD; 3, 4) whose genes are present from eubacteria to eukaryotes [27]. The structural paradigm of these proteins is the Rossmann-fold topology [28], a dinucleotide-binding motif that is shared by other (nonamine oxidizing) enzymes. Typically, the AOD structure consists of two subdomains, a FAD-binding moiety characterized by the Rossmann-fold topology and a substrate-binding domain.

Evolution has ‘modeled’ the structure of these AOD enzymes to fit different amine substrates and cellular landscapes. LSD1 was the subject of a dali search [29] to investigate structural affinities with other proteins and the outcome of this analysis is summarized in Table 1. The highest score is found with polyamine oxidases (PAOs), such as maize PAO [30] (Fig. 3A) and yeast FMS1 [31], which are characterized by the typical AOD two-domain fold. Fungal monoamine oxidase MAO N [32] is a soluble enzyme supposed to be located in the peroxisome and has an overall globular structure similar to that of PAOs (Fig. 3A). During evolution, its mammalian homologs, the monoamine oxidases MAO A and MAO B, acquired an additional C-terminal segment which folds into an α helix that anchors these enzymes to the outer mitochondrial membrane (Fig. 3A) [33,34]. The rationale for such cellular localization relative to the role of MAO A and B in neurotransmitter metabolism remains unknown. A fascinating aspect of LSD1 is that this enzyme has an even higher level of complexity because its AOD sequence comprises an insertion of ∼100 amino acids folded into a helix–turn–helix motif (Tower domain) [35] and preceded by a segment of ∼200 residues that includes a SWIRM domain and a putative unstructured region (3, 4). The Tower domain protrudes from the globular AOD structure and provides the interface for CoREST binding (Fig. 2). Interestingly, the Tower domain originates on the surface of the AOD structure that corresponds (with respect to the position of the FAD coenzyme) to the area where the transmembrane helix branches off in MAO A and B (Fig. 3A). The SWIRM domain is a motif commonly found in chromatin enzymes and is responsible for protein–protein and DNA–protein interactions [36]. The name of this domain derives from its original identification in the proteins SWI3, Rsc8 and Moira, which are ATP-dependent chromatin remodeling enzymes. The N-terminal part of LSD1 is predicted to be a disordered region and is discussed in the next section.

The active site of all these amine oxidases is normally located in the proximity of the flavin ring of the FAD cofactor and extends along the substrate-binding domain. Despite the similar overall folding topology in the AOD enzymes, there is significant variability in the size and shape of their substrate-binding sites, which have been evolutionarily adapted to the enzyme function and type of substrate. PAO is outstanding in this regard because it is endowed with a long tunnel suited to the aliphatic chain of polyamines (Fig. 3) [30] and its yeast homolog FMS1 displays similar features [31]. In l-amino acid oxidase a 25 Å-long funnel provides access to the active site [37], whereas in MAO A, MAO B and MAO N the aromatic substrates bind in a hydrophobic cavity that, in MAO B (Fig. 3A), has a bipartite nature and is shaped by a loop that provides access to the enzyme active site [38]. In LSD1, the substrate-binding site is represented by a large funnel that originates from the flavin and opens wide towards the outside to accommodate the histone substrate (Fig. 3A) [35,39]. In contrast to other histone-modifying enzymes, the mechanism of histone tail recognition by LSD1 is complex and highly specific. LSD1 enzymatic activity can be assayed using peptides corresponding to the 21 N-terminal amino acids of histone H3 [7]. The structure of the enzyme in complex with a histone peptide revealed that the histone tail adopts a folded conformation when bound to the enzyme and creeps deep into the funnel cavity, establishing a network of interactions with the active site residues [16]. These specific interactions act together to fix the histone tail in the correct register, which positions Lys4 (the site of the demethylation reaction) in front of the flavin cofactor. The co-repressor CoREST, which stabilizes LSD1 and enhances its enzymatic activity on histone peptides, is essential for the in vitro demethylation of nucleosomal particles by LSD1 [40,41]. As shown in Fig. 2, CoREST embraces the LSD1 Tower domain and binds in the proximity of the funnel opening where the histone peptide C-terminus is located (i.e. where the globular part of the entire histone is likely to lie).

Despite differences in the shape of the substrate-binding site, the details of the active sites of flavin amine oxidases are strikingly conserved (Fig. 3B). In all enzymes listed in Table 1 (except for glycine oxidase), there is a lysine residue on top of the flavin, bridged to the N5 atom of the coenzyme via a water molecule [25], which is proposed to have a role in the enzymatic activity. Indeed, mutagenesis experiments have shown that in LSD1, replacement of Lys661 with Ala produced an inactive enzyme [35]. Very recently, it was shown that mutation of this residue in monomeric sarcosine oxidase dramatically reduced the enzymatic turnover rate and oxygen reactivity, providing definitive evidence for this lysine as the site of oxygen activation [42]. Another conserved feature in the structure of the AOD active site is the so-called ‘aromatic cage’ [25]. The substrate binds in front of the FAD cofactor (either the re or the si face depending on the enzyme), which has alongside one or two aromatic amino acids that form a sort of cage together with the flavin ring. In PAO and in MAO A, MAO B, MAO N the cage is composed of a Tyr–Phe and Tyr–Tyr couple, respectively. Mutagenesis experiments in MAO B demonstrated that these aromatic residues may have a steric role in substrate binding and in increasing the nucleophilicity of the substrate amine moiety [43]. In LSD1 one of these aromatic residues is conserved, with the other being replaced by a Thr residue (Fig. 3B). Although the role of Tyr761 in the histone demethylation reaction has not been clarified, it might be involved in recognition of the methylated Lys4 amino group.

Multidomain organization of LSD1 and its enzymatic activity in the chromatin context

Genes coding for LSD1/K-demethylase 1 proteins exist only in eukaryotes and have undergone distinct evolutionary development in plants and animals [27]. In contrast to AOD enzymes, which are amine-metabolizing catalysts generally functioning by themselves, LSD1 oxidizes the C–N bond of methylated histone lysines within the frame of the chromatin matrix. Table 2 compares the enzymatic activities of those flavin amine oxidases for which a thorough biochemical characterization has been reported. The reactivity with oxygen is high, consistent with the fact that, in the case of oxidases, reoxidation of FAD is not the rate-limiting step, whereas the enzymatic activity responsible for substrate oxidation is relatively low for LSD1 compared with that of the chemically similar amine oxidations of other AOD proteins. However, it is noteworthy that the rates of the histone demethylation reaction carried out by enzymes of the JmjC class are even lower [44]. The catalytic activity of these histone-modifying enzymes needs to be evaluated taking into account that they participate in finely tuned cellular processes in which the cascade of events is regulated by a number of protein factors. In this regard, we observed that the presence of the co-repressor CoREST increases in vitro LSD1 activity by approximately two-fold [16] and it has been reported that CoREST is essential for LSD1-mediated demethylation in intact nucleosomes [40,41]. So far, LSD1 is unique among the AOD proteins in catalyzing an amine oxidation reaction in the chromatin environment and its demethylase activity has to be pondered in the context of multiprotein transcriptional complexes.

LSD1 is a multidomain protein and its involvement in many diverse gene-expression programs [45] is strictly related to its domain organization (Fig. 4). In addition to the SWIRM domain, whose role in the interaction with chromatin proteins is well known, the N-terminal part of LSD1 is of interest. This 200-residue segment is predicted to be disordered and, indeed, all structural studies to date have focused on the SWIRM–AOD portion of the enzyme. However, in recent years, the idea has emerged that, contrary to the traditional view that protein function is necessarily determined by a stable 3D structure, disordered regions may represent flexible modules that fold upon binding to their biological target [46]. In order to investigate the putative disordered 200-residue segment of LSD1 by bioinformatics tools, we used the ELM database [47] which contains a collection of linear motifs corresponding to consensus sequences of known functional meaning. These short sequences may provide sites for post-translational modifications or docking modules that are recognized by specific, well-characterized protein domains. In this way, it is possible to identify potential functional sites in a putative disordered region and to formulate hypotheses regarding its biological role, which may suggest experimental investigations. It is important to note that the updated version of ELM produces results that are ‘filtered’ by additional information such as cell compartment, phylogeny and structural data, and this provides more reliable outcomes from a biological viewpoint. Figure 4 shows an overview of the results of ELM analysis on LSD1, which was restricted to the human cell nucleus compartment. The first two lines indicate that the globular and structured regions of LSD1 are located in the SWIRM–AOD part of the protein, as predicted by other tools and confirmed by the crystal structure [35]. Several linear motifs are identified as putative functional sites in the N-terminal low-complexity region of LSD1, others between the SWIRM and the AOD and two additional hits are found at the C-terminus (whose final 20 residues are not visible in the crystal structure). Some of these motifs are supposed to undergo Ser/Thr phosphorylation by well-characterized kinases (e.g. cyclin-dependent kinases), whereas others are predicted to be sites of other post-translational modifications such as sumoylation and ubiquitination, which are recognized by specific domains in protein–protein interaction mechanisms. These predictions need to be validated using experimental studies, but they suggest that this N-terminal putative disordered domain may have a role in providing the enzyme with the necessary flexibility to target different chromatin proteins and to adapt LSD1 enzymatic activity to distinct gene transcription events.

Conclusions

The globular domain responsible for the flavin-dependent amine oxidase activity is a highly conserved module used in nature to oxidize amine substrates of different types and participating in distinct biological processes. LSD1 is a remarkable example because it exerts its oxidase activity on a complex substrate such as the histone tail in the eukaryotic cell nucleus, and because it is one of the first examples of the insertion of a protein–protein interaction module (the Tower domain) into a highly conserved functional domain. Evolution has provided LSD1 with the necessary complexity to catalyze a basic chemical reaction in the context of gene transcription regulation. It will be a challenge for future studies to understand how LSD1 fulfils its multiple roles in these complexes mechanisms.