Nucleosome binding by the lysine specific demethylase 1 (LSD1) enzyme enables histone H3 demethylation

Abstract

The essential human enzyme lysine specific demethylase 1 (LSD1) silences genes by demethylating mono- and dimethylated lysine 4 in histone H3 (H3K4me1/2). Studies of the minimal requirements for LSD1 activity are complicated by the heterogeneity of histone modification states in cells. We overcame this challenge by generating homogeneous mononucleosome substrates containing semisynthetic H3K4me2. Biophysical and biochemical assays with full-length LSD1 revealed its ability to bind and demethylate nucleosomes. Consistent with a requirement for nucleosome binding prior to demethylation, a competing nucleosome binding peptide from the high mobility group protein effectively inhibited LSD1 activity. Thus, our studies provide the first glimpse of nucleosome demethylation by LSD1 in the absence of other scaffolding proteins.

Graphical Abstract

Histone post-translational modifications (PTMs) or marks include the acetylation and methylation of lysine side-chains that play critical roles in regulating gene transcription.¹ Lysine 4 in histone H3 (H3K4) is methylated by the Set1/MLL family of lysine methyl-transferases.² Extensive histone di- and trimethylation (H3K4me2 and H3K4m3) are associated with transcriptionally active genes, while monomethylation (H3K4me1) marks active and primed enhancer elements.³ The lysine specific demethylase protein 1 (LSD1 or KDM1A) is a flavin adenine dinucleotide (FAD)-dependent histone mark eraser repressing gene transcription by demethylating H3K4me1/2.⁴ LSD1 is an essential enzyme in mammals and regulates the epithelial-mesenchymal transition, cell survival and proliferation.⁵ The misregulation of LSD1 is associated with aggressive cancers of the prostate and breast, non-small cell lung cancer and neuroblastoma.⁶ Therefore, identifying the minimal requirements for its function on chromatin is an essential step in developing therapeutics targeting LSD1.

LSD1 is a 95 kDa multi-domain enzyme consisting of an unstructured N-terminal region followed by a Swi3p, Rsc8, and Moira (SWIRM) domain and the catalytic amine oxidase domain (AOD) (Figure 1A).⁷ An antiparallel coiled-coil of two long α-helices forms the tower domain that is inserted into the AOD and serves as the binding site for the co-repressor of RE1 silencing transcription factor protein (CoREST).⁸ LSD1 uses flavin adenine dinucleotide (FAD) and molecular oxygen to catalyze the stepwise oxidation of dimethylated lysine to monomethylated and then unmethylated lysine, forming formaldehyde and hydrogen peroxide (H₂O₂) as by-products (Figure 1B).^{4, 9} LSD1 isolated from human cells is associated with several gene-repressive complexes, including the CoREST complex.¹⁰ These complexes also contain histone deacetylases (HDACs), which preclude definitive conclusions regarding the minimal requirements for LSD1 function.¹¹ Additionally, PTMs such as acetylation and phosphorylation at proximal residues in the H3 tail directly inhibit LSD1 activity, complicating mechanistic conclusions from assays with heterogeneously modified cellular chromatin.^{12, 13} Therefore, we undertook studies of the minimal requirements for LSD1 function with semisynthetic nucleosome substrates uniformly modified with H3K4me2.¹⁴

Figure 1.

Structure and mechanism of lysine demethylation by LSD1. (A) Structure of the LSD1(171–836) truncant showing tower, SWIRM and amine oxidase domains. PDB code 2IW5. (B) Chemical mechanism of lysine demethylation and H₂O₂ production by LSD1.

We began our studies by purifying the full-length LSD1 enzyme (Figure S1). In order to identify potential differences between full-length LSD1 and previously characterized LSD1 truncants, we undertook demethylation assays with an H3(1–21)K4me2 peptide substrate (Supplementary Information and Figure S2). The kinetics of demethylation were measured by a continuous horseradish peroxidase-coupled assay (Figure S3). In this platform, the k_cat of 8.9 ± 0.5 min⁻¹ and K_M of 15.1 ± 2.4 μM for full-length LSD1 were similar to reported values for an N-terminally truncated LSD1(151–852) enzyme (Table 1).¹⁵

Table 1.

Kinetic constants of LSD1 with histone and mononucleosome substrates.

Substrate K4me2 in	LSD1	kcat min−1	KM μM	kcat/KM min−1μM−1
H3 (1–21)	151–852	6.0 ± 0.1	4.5 ± 0.3	1.3 ± 0.1a
H3 (1–21)	1–852	8.9 ± 0.5	15.1 ± 2.4	0.6 ± 0.1b
H3	1–852	10.9 ± 1.2	3.2 ± 0.8	3.4 ± 0.9b
H3 (1–21)	1–852 + CoREST	5.3 ± 0.5	15.4 ± 2.9	0.3 ± 0.1b
H3	1–852 + CoREST	3.8 ± 0.2	1.9 ± 0.3	2.0 ± 0.3b
MN	1–852	0.11 ± 0.01c	8.2 ± 2.0d	0.013 ± 0.003
MN (Kc4me2)	171–852 + CoREST (286–482)	0.32 ± 0.02e	0.39 ± 0.06e	0.82 ± 0.14

^aData for His₆-LSD1(151–852) from McCafferty and coworkers.¹⁵

^bMeasured by a continuous horseradish peroxidase-coupled assay.

^cMaximum rate of catalysis under single-turnover conditions measured by quantifying western blots with an anti-H3K4me2 antibody.

^dApparent K_M under single-turnover conditions measured by quantifying western blots with an anti-H3K4me2 antibody.

^eData from Tan and coworkers measured by a fluorescence-based continuous assay.¹⁶ MN= mononucleosome. K_c= thialysine analog of lysine.

Next, we synthesized the full-length dimethylated histone, H3K4me2 (Figure S4), and assayed it with LSD1. Consistent with a recent report of enhanced binding of full-length H3 by LSD1,¹⁵ we observed a 5-fold decrease in K_M for H3K4me2 with no significant changes in the k_cat of demethylation (Table 1). Given their strong association in cells, we asked if CoREST binding affects LSD1 kinetics. Surprisingly, the inclusion of a 1:1 LSD1-CoREST complex (Figure S5) in either peptide or histone demethylation assays only had small detrimental effects on the specificity constant (k_cat/K_M) for each substrate (Table 1), indicating that CoREST does not allosterically activate LSD1. CoREST also does not change the distributive mechanism of demethylation by LSD1, as seen from the similar build-up of partially demethylated intermediates in demethylation assays (Figure S6).

As CoREST did not enhance LSD1 activity nor change its distributive mechanism of demethylation, we wondered if it is absolutely essential for nucleosome demethylation by LSD1. To answer this question, we assembled nucleosomes with human histones and the 147 bp Widom 601 double-stranded DNA sequence (Figure S7).¹⁷ Two sets of nucleosomes containing either wild-type H3 or H3K4me2 were generated and subjected to electrophoretic mobility shift assays (EMSA) with LSD1 (Figure 2). Interestingly, while an N-terminally glutathione S-transferase (GST)-tagged LSD1Δ(1–171) truncant did not bind nucleosomes (Figure S8), we observed concentration dependent gel-shifts of both methylated and unmethylated nucleosomes with full-length LSD1 that showed similar mid-points with K_1/2 ~4 μM (Figure 2A and S9). Consistent with specific binding rather than non-specific aggregation, the mid-point of gel-shifts in EMSA decreased to K_1/2 ~1 μM when LSD1 bound 12-mer nucleosome arrays at identical histone concentrations (Figure 2B and S10).

Figure 2.

Nucleosome binding and demethylation by LSD1. (A) Electrophoretic mobility shift assay (EMSA) of methylated mononucleosomes (MNs) with increasing amounts of LSD1. (B) EMSA of 12-mer methylated nucleosome arrays with increasing amounts of LSD1. Asterisks indicate sub-saturated LSD1-bound species. (C) Representative microscale thermophoresis profiles of LSD1 binding to mononucleosomes at the indicated concentrations. (D) Quantitation of LSD1-mononucleosome binding affinity. n=3, error bars show standard deviation. (E) Time-course of LSD1-catalyzed demethylation of H3K4me2 in mononucleosomes followed with an H3K4me2-specific antibody. Residual H3K4me2 in each lane was quantitated relative to histone H4 in each lane that was detected with an H4-specific antibody. (F) Quantitation of residual dimethylated H3K4me2 from western blots. n ≥3, error bars show standard deviation. (G) Kinetic parameters for LSD1 function determined from quantifying single-turnover demethylation assays. n ≥3, error bars show standard deviation.

In order to quantitate the strength of nucleosome binding by LSD1, we employed a recently developed MicroScale Thermophoresis (MST) approach.¹⁸ The thermophoretic movement of nucleosomes varies with their size, charge and hydration shell. Upon LSD1 binding to a nucleosome one or more of these properties change, leading to a change in diffusion behavior of the nucleosome that can be used to calculate the equilibrium dissociation constant, K_d. Using unmethylated mononucleosomes prepared with 601 DNA labeled at its 5’-end with the fluorescent dye Cy5 and by varying LSD1 concentrations from 9 nM to 145 μM in the MST capillary (Figure 2C), we measured a K_d of 6.1 ± 1.7 μM for LSD1 binding to the mononucleosome (Figure 2D).

Having established that full-length LSD1 binds nucleosomes, we next tested its ability to demethylate H3K4me2 in the nucleosomal context. Single-turnover assays were conducted with 50 nM mononucleosomes containing two copies of H3K4me2 and varying LSD1 concentrations from 2 to 20 μM in HEPES buffer containing 50 mM KCl (Figure 2E,F and S11). The observed rate of demethylation was plotted against LSD1 concentrations and the graph fit to the equation k_obs=k_max[LSD1]/(K’_M+[LSD1]) using non- linear regression analysis (Figure 2G).^{19, 20} This revealed an apparent Michaelis constant (K’_M) of 8.24 ± 2.01 μM and the maximum rate of catalysis (k_max) to be 0.11 ± 0.01 min⁻¹. Thus, although LSD1 binds nucleosomes and histone tail peptides with similar affinities, the rate of nucleosomal demethylation is ~50-fold slower than peptide demethylation. The addition of an equimolar amount of full-length CoREST increased the extent of demethylation by 5 μM LSD1 about two-fold in HEPES buffer containing 125 mM KCl (Figure S12).

In the course of nucleosome demethylation assays, we observed that free DNA inhibits LSD1 function (Figure S13). Hence, we asked if LSD1 binds free DNA, which may compete with its nucleosome binding and thereby limit H3 demethylation. Interestingly, full-length LSD1 bound both the 601 DNA sequence and the Mouse Mammary Tumor Virus Long Terminal Repeat A region (MMTV-A) DNA sequence with similar affinities in EMSA assays (Figure 3A).²¹ Although LSD1 has a SWIRM domain, this was suggested to participate in RNA binding and to not bind DNA.²² Therefore, we sought to identify the region of LSD1 responsible for DNA binding. The degree of DNA-binding by full-length LSD1 and its truncated forms was quantified in fluorescence polarization assays with 1 nM of 5’-AlexaFluor 488-labeled 601 DNA (Figure 3B). The addition of increasing amounts of LSD1 to the fluorescent DNA led to an increase in anisotropy, and the observed binding curves were fit by non-linear regression using fluorescent DNA concentration as the constraint. This revealed a K_d of 127 ± 20 nM for full-length LSD1 binding to 601 DNA. The disordered N-terminus of LSD1, LSD1(1–170), alone or in combination with the SWIRM domain, LSD1(1–285), bound DNA ~6-fold weaker than LSD1(1–852). The LSD1(286–852) construct, which contains both the amine oxidase and tower domains of LSD1, bound DNA with a K_d of 185 ± 25 nM, very similar to full-length LSD1 (Figure 3B). Our observations explain why LSD1(171–852) was observed to photo-crosslink with nucleosomal DNA,¹⁶ and agree with a recent crystal structure that observed the direct interaction between LSD1(171–852) and nucleosomal DNA in the context of a truncated LSD1-CoREST complex.²³ The fact that multiple regions of LSD1 bind DNA suggests the interaction is electrostatic in nature.

Figure 3.

Nucleosome binding by LSD1 underlies histone demethylation. (A) Electrophoretic mobility shift assay of 147 bp Widom 601 DNA (top) and 147 bp MMTV-A DNA (bottom) with the indicated concentrations of LSD1. An asterisk indicates sub-saturated LSD1-bound DNA. (B) Variation in fluorescence polarization of 1 nM AlexaFluor 488-labeled Widom 601 DNA as a function of varying concentrations of full-length LSD1 or truncants thereof. n ≥3, error bars show standard deviation. (C) The effect of increasing amounts of nucleosome binding domain (NBD) 29-mer peptide on mononucleosome demethylation by LSD1. Signal intensities in western blots, using an H3K4me2-specific antibody, were normalized to the amount of histone H4 detected with an H4-specific antibody at each time point. n ≥3, error bars show standard deviation. * p<0.05, ** p>0.05.

Finally, we focused on the mechanistic requirement for nucleosome-binding toward histone demethylation. We asked if inhibiting the LSD1-DNA and/or LSD1-octamer surface interactions may inhibit H3K4me2 demethylation. In thinking of proteins that bind nucleosomal DNA and the octamer surface that may compete with LSD1, we considered the high mobility group nucleosome-binding family of proteins (HMGNs).²⁴ These multi-domain proteins contain an N-terminal bipartite nuclear localization sequence interrupted by a nucleosome binding domain (NBD) and a C-terminal chromatin-unfolding domain (Figure S14). The NBD of HMGN1 is a 29-mer peptide that was previously shown to bind the octamer surface and nucleosomal DNA.^{25, 26} We tested LSD1 activity on methylated nucleosomes in the presence of the HMGN1 NBD peptide (Figure S15). Increasing amounts of NBD incrementally inhibited nucleosome demethylation by LSD1, suggesting that nucleosome binding is indeed essential for H3K4me2 demethylation by LSD1 (Figure 3C). Importantly, the highest concentration of NBD tested did not significantly inhibit demethylation of the H3(1–21)K4me2 peptide (Figure S16), indicating that inhibition arises solely from competing with LSD1 for nucleosome binding.

In conclusion, our biophysical and biochemical studies with purified full-length LSD1 and homogenously methylated designer nucleosomes have revealed the previously unknown ability of LSD1 to demethylate non-acetylated nucleosomes in the absence of other DNA-binding proteins. Furthermore, nucleosome demethylation may involve sequence-independent electrostatic DNA-binding by the tower and/or amine-oxidase domains of LSD1. Finally, the effective inhibition of demethylation by an excess of nucleosome-binding peptide from the high mobility group protein suggests that nucleosome-binding by LSD1 is a pre-requisite for H3K4me2 demethylation.

ABBREVIATIONS

KDM1A

lysine demethylase 1A

LSD1

lysine specific demethylase 1

FAD

flavin adenine dinucleotide

SWIRM

Swi3p, Rsc8, and Moira domain

AOD

amine oxidase domain

CoREST

Co-repressor of RE1-silencing transcription factor

HDACs

histone deacetylases, MN, mononucleosome

DNA

deoxyribonucleic acid

GST

glutathione S-transferase

MMTV-A

Mammary Tumor Virus Long Terminal Repeat A region

EMSA

electrophoretic mobility shift assay

MST

microscale thermophoresis

HMGNs

high mobility group nucleosome-binding family of proteins

NBD

nucleosome-binding domain