Structure‐Based Screening of Tetrazolylhydrazide Inhibitors versus KDM4 Histone Demethylases

Abstract Human histone demethylases are known to play an important role in the development of several tumor types. Consequently, they have emerged as important medical targets for the treatment of human cancer. Herein, structural studies on tetrazolylhydrazide inhibitors as a new scaffold for a certain class of histone demethylases, the JmjC proteins, are reported. A series of compounds are structurally described and their respective binding modes to the KDM4D protein, which serves as a high‐resolution model to represent the KDM4 subfamily in crystallographic studies, are examined. Similar to previously reported inhibitors, the compounds described herein are competitors for the natural KDM4 cofactor, 2‐oxoglutarate. The tetrazolylhydrazide scaffold fills an important gap in KDM4 inhibition and newly described, detailed interactions of inhibitor moieties pave the way to the development of compounds with high target‐binding affinity and increased membrane permeability, at the same time.


Introduction
Gene expression regulationt hrough epigenetic factorsi sa concept that is well established nowadays. It is defined as a pattern of changes in gene expression without altering the un-derlyingD NA sequence. [1] Living cells must be able to dynamically respond to changes in physiological and environmental stimuli and to modify their chromatin structure accordingly.A complexs et of regulatory events, fore xample, DNA modifica-tion by cytosine methylation or covalenth istonem odifications have been shown to alter the interactions between DNA and histones. Modifications promotee ither compaction or relaxation of the chromatin, and thus, play adirect role in controlling the transcription level of particularg enes. Epigenetic control occurs in both healthy and cancerous cells. [2] For instance, aberranth istonem odification exhibits ac lear relationship with disease and is often associated with cancer. [3] The ability to interfere with these processes is therefore of increasing interest for researchers from the academic sector andf rom pharmaceuticalcompanies.
One of the most studied histonem odificationsi st he addition of am ethyl group to lysine residues, as ar esulto ft he action of histonel ysine methyltransferase enzymes( KMTs). [4] The histonet ails, which are packed more loosely than the histone cores,have been identified as the main targets for modifications by KMTs. [5] Despite the subtle physicochemical effect of lysine methylation, it has at remendous regulatory impact on cellular functions. Combining the number of lysiner esidues in histones, which can be methylated, the variousd egrees of methylation (mono-, di-, and trimethylation), and the fact that methylation is cooperative with other modificationsr esults in a very complex picturef or gene expression control. [6] Until about 2004, histone lysine methylation was believed to be irreversible. Since then, two classes of reverse-action (eraser)e nzymes-histone lysine demethylases-have been discovered and studied in depth. Usually,t hese enzymes are selective for ap articular lysine residuei naparticular methylated state. The first class comprises the flavin-dependent lysine-specific demethylases( LSDs or KDM1s). The second class, whichh as more members than the first, comprises all oxygen-dependent demethylases, which use iron(II) and 2-oxoglutarate (2OG, also called a-ketoglutarate) as cofactors.T he enzymes, which Humanh istone demethylases are known to play an important role in the development of severalt umor types. Consequently, they have emergeda si mportant medical targets for the treatment of human cancer.H erein, structural studies on tetrazolylhydrazidei nhibitors as an ew scaffold for ac ertainc lass of histone demethylases, the JmjC proteins, are reported. As eries of compounds are structurally described and their respective binding modes to the KDM4D protein, which serves as ah ighresolution model to represent the KDM4 subfamilyi nc rystallo-graphic studies, are examined. Similar to previously reported inhibitors, the compounds described herein are competitors for the natural KDM4 cofactor,2 -oxoglutarate. The tetrazolylhydrazides caffold fills an important gap in KDM4 inhibition and newly described, detailed interactions of inhibitor moieties pave the way to the development of compounds with high target-bindinga ffinity and increased membranep ermeability, at the same time.
belong to the second class, contain the Jumonji domains; hence,t hey are also called JmjC KDMs. [7] The discoveryo f these enzymes established an entirely new concept, with respect to the dynamic reversible regulation of histonem ethylation by KMTs and KDMs. In humans, the JmjC histone lysine demethylases encompass distinct families,e ach of which consist of several multidomain members. There are at least six KDM families recognized in the literature( KDM2-7) and still new subfamilies (KDM8 and KDM9) are being reported.T he catalytic function of these enzymes is provided by the JmjC domain, whereas the JmjN domain closely interacts with JmjC and provides the three-dimensional scaffold.A ll members are classified based on the presenceo fc onserved structural elements, in addition to the catalytic JmjC domain, including PHD, Tudor,C XXC, FBOX, ARID, LRR, and JmjN domains. [7,8] The five members of the KDM4 subfamily( KDM4A-E) catalyze the removal of the methylg roup from tri-andd imethylated lysine residue 9i nh istone3 (H3K9me3/me2). At the same time, KDM4A-C also processt he tri-and dimethylated lysine residue 36 (H3K36me3/me2).I ncreasing evidence has been accumulated that KDMs are also linked with various disease states. [9] For instance, KDM4Bw as found to be involved in breast, colon, and gastric cancer, [10] whereas KDM4A, KDM4C, and KDM4D are overexpressed in prostate cancer,d uring which they act as androgen receptor coactivators. [11] Consequently,t hese enzymes have been identified as potential targets for intervention by drugs.Atpresent, four compounds targeting LSD1 are in clinical trials, whereas for the JmjC KDM family of enzymes no such developmentsh ave been reported. [12] In spite of this, al arge number of small-molecule inhibitors of JmjC KDMs have been established. For the KDM4 subfamily alone, about 94 structures with bound ligands [13] have been deposited in the Protein Data Bank (PDB). [14] Most of the inhibitors are 2OG competitors that coordinate the Fe 2 + ion in the catalytic center and sometimes penetrate into the histonepeptide binding pocket.M any of these compounds mimic the natural cofactor and, similar to 2OG, possess ac arboxylic acid moiety.T he carboxylic group is ac ommon building block in endogenous substances, but it is also part of many pharmacophores. [15] Its successisdue to its acidic character and potential to engage in relativelys trong electrostatici nteractions and hydrogen bonds. On the other hand, the carboxylic moiety reduces the ability of ad rug to passively move across biological membranes, and thus, limits cell permeability.T wo different strategies are suited to overcoming this problem associated with drugs and drug candidates:1 )bioisosteric replacement or 2) prodrug strategies,s uch as use of carboxylic acid esters. Consequently,i nadrug development process of carboxylic acid leads against JmjC KDMs,s everal ester compounds have been investigated. [16] Because this well-established strategy potentially yields prodrugs that are transformed into the active form differently in individual patients,t he replacement of the carboxylic acid by groups with similar volume, shape, and physicochemical properties is an attractive alternative. From this perspective,i ti sq uite surprising that structurald ata for KDM4 proteins and compoundsc ontaining tetrazolem oieties have not been reported, although the tetrazole moiety is part of many biologicallya ctive compounds, including drugs. [17] Althought he electrostatic potentials of carboxyl and tetrazole groups are similar,t etrazoles could improve the cellular permeability due to their more lipophilicc haracter. Tetrazoles are low-toxic, metabolically stable bioisosteres of carboxylic acids, and thus, promising building blocks for analogues of the natural KDM4 cofactor,2 OG. [18] However,n one of the KDM4-ligand complexes reported so far in the PDB contain at etrazole ring as af unctional group.
In 2015, Rüger et al. identifiedt etrazolylhydrazidesa ss elective fragment-like inhibitors of the KDM4 subfamilyo fp roteins. [19] Arrays of such molecules have since been synthesized and analyzed mainly by molecular modeling approaches. [20] The most promising compounds suggested am ode of binding in the KDM4 active site. The assumption was that the tetrazole group acted through binding to the 2OG site as ac ompetitor for the substrate. However,d irect support for this hypothesis by experimental structurala nalysis has so far been lacking.
Herein, we presentt he first successful attemptt oo btain high-resolution X-ray structures of both hydrazide and tetrazole moieties bound to the enzyme KDM4D. These small molecules constitute promising frameworks for early drug discovery stages because their initial selectivity has been reported. [19] The structures presented herein show the specific bindingo ft etrazolylhydrazide ligands [19,20] (Figure 1) to the metal ion in the active site, serving as2 OG competitors. As eries of molecules with hydrazide and tetrazole ring modifications show different modes of binding in the active centero ft he enzyme. Mapping of the active site allowed us to localizet he residues that were able to adjust their position upon ligand binding. Our attempt to describe compounds that exclusively inhibit this relatively new target KDM4 hasr esulted in structural data that show the mode of binding of new functionalg roups that have not been describedp reviously:t etrazole and hydrazide. Uncovering and describing these details is as tep forwardi nt he development of drug-like molecules in the fight against cancer.

KDM4D crystals and KDM4Ds tructure
All structural work reported herein wasp erformedo nt he catalytic core of KDM4D, which comprised amino acid residues 1-342. This construct crystallizes in af ormt hat renders the active site of the protein fully accessible, with no obstructions from neighboring protein molecules in the crystal lattice. In all KDM4 subfamily members,t he 2OG binding site is literally the same and engages the same residues.T he tetragonal crystal form features one molecule in the asymmetric unit and permits the diffusion of ligand molecules through the solvent channels to the substrate binding pocket of the protein.T his crystal form of KDM4Dh as been observed previously (PDB ID: 3DXU, unpublished), and based on our own observations, it turned out to be superior to others for ligand soaking experiments due to its diffraction properties, mechanical stability, and tolerance for highD MSOc oncentrations. In all obtained structures, the part of the amino acid sequence that was visible in the electron density included residues 11-342. Overall, the structure of KDM4Di sg lobular and may be divided into the JmjN and JmjC domains. The N-terminal JmjN domain comprises residues 18-60 andt he catalytic JmjC domain comprises residues 146-312 (Figure2). These domainsa re highly similar to the homologous domainso fo ther KDM4 proteins. As in other KDM4 subfamily members, ac onserved zinc-binding site composed by the residues His244, Cys238, Cys310, and Cys312 was observed. This binding site is of mainly structural importance.T he Zn 2 + ion is present with 100 %o ccupancy and has its origin in the Escherichia coli host organism because no Zn 2 + wasa dded to any of the purification or crystallization buffers. The active-site metal center contains Ni 2 + as the metal ion. Although in the natural form of the enzymet his site is occupied by an Fe 2 + ion, it is widely accepted in crystallographic studies to replace rather oxygen-sensitive Fe 2 + with Ni 2 + or Co 2 + .A ll of the ligands (1-7;F igure 1) reported in this study occupy the native cofactor binding site, which is in close vicinity to the metal binding site and the site binding the methylated histone lysine.B ased on available statistics( Ta ble 2) and the quality of experimental data, the structures reported herein are of sufficiently high quality to ascertain the binding of the soaked-in ligands. The entirec atalytic core and the binding of the cofactor 2OG and at rimethylatedp eptidet hat mimics the histone3 tail (H3K9me3) has been thoroughly described by )-cofactor2 OG and the "incoming" trimethylated lysine (Kme3, part of the histonel ike peptide)-can be seen in the active-site pocket. Superpositiono fligand 1 with the 2OG-bound structure showsh igh structural similarities between bioisosteres. Substrate binding site residues with semitransparent secondarystructure elements can be visible. The cofactor and trimethylated lysiner esidue are given in ball-and-stick representation in magenta, whereas the tetrazolehydrazideliganda nd binding residues are in yellow.B ottom:Surface representation of KDM4D with the ligand in the binding pocket and the histone-like peptide bound on the surfacei ss uperimposed in magenta as stick representation.  Krishnan and Trievel. [13m] Cofactor2 OG chelates the active-site Ni 2 + ion by using both the C2 keto group and C1 carboxylate group. In addition, the octahedral coordinations phereo ft he metal centerc ontainsG ln194, which binds opposite to the C2 keto group;H is192, which binds opposite to the C1 carboxylate group;H is280; and aw ater molecule. The other end of cofactor 2OG is held in place by Asn202, Lys210, and Tyr136 ( Figure 2). The active-site residues Tyr181, Glu194, and Gly174 are in close vicinity aroundt he trimethylated lysine of the histone. [13m] The peptidic ligand was not used in our experiments; thus, it is not observed in the structuresr eported herein, but superimposed in Figure 2f or visualization of the histoneb inding site in KDM4 proteins.

Ligand bindingexamined by crystalstructure analysis
Compounds 1-7,which all contain at etrazole group ( Figure 1), were individually soaked into KDM4Dc rystals. This resulted in as eries of seven crystal structures of KDM4D ligand complexes ( Figure 3). The full pictureo fs patial positioning and detailed web of interactions of protein residues with compounds are discussed in the following subsections. All structures are of high quality,a se videnced by their resolution and refinement statistics( Ta bles 1a nd 2). Although some ligands exhibit less than 100 %o ccupancy,w hich means that they are only bound to af raction of the protein molecules, their clear appearance in the differencee lectron density map allows their unambiguous placement in the structure. All compounds in this series, except for compounds 4 and 5,a re composed of two building blocks intendeda si nteraction motifs:t he tetrazole ring and the hydrazide group. Ligands mainly differ in the alternations and modifications incorporated between them. The functional groups of the compounds were designed with binding to the KDM4 proteins through these two functional groups in mind. In addition to compounds 1-5,w hich exhibit simple binding motifs and some modifications, two more complicated tetrazolylbenzohydrazide compounds (compounds 6 and 7)w ere investigated. In these compounds, the tetrazoler ing is attached in meta and para positions to the benzene ring.

Structure of KDM4D with compound1
The lead structure, compound 1,r epresents the molecule with the shortestd istance between the two interaction motifs:o nly one methylene group separates the tetrazole group and the hydrazide( n = 1). In the structure of compound 1 bound to KDM4D, ac lear positive difference electron density for the ligand could be observed and the ligand could be modeled with 77 %o ccupancy ( Figure 3A). Its hydrazideg roup chelates to the active site Ni 2 + ,w hich is as urrogate of iron, as discussed above. The mode of chelation is very similar to that observed in the bindingo ft he natural2 OG cofactor (Figure 2). [13m] The aromatic tetrazole ring is sandwiched between the two active-sitea romatic residues Phe189 andT yr181. The distances between the tetrazole and the centers of the aromatic rings are 4.2 and 4.9 for Phe189 and Tyr181, respectively; thus indicating an aromatic stacking interaction. The distance of 2.6 between the hydroxy of Tyr136a nd the second nitrogen atom from the tetrazole ring indicates strong hydrogen bonding. All of the other nitrogen atoms in the tetrazole ring are connected to the protein residues through clearly defined water molecules.

Structure of KDM4D with compound 2
Compound 2 is characterizedb yt wo methyleneg roups between the functional groups (n = 2). In the complex structure with KDM4D, this leads to as hift of the tetrazole ring relative to its position in the complex with compound 1,w hereas the hydrazide group remains unchanged ( Figure 3B). Theo bserved electron density for the side chain of Tyr136 reveals that its position has changed. In its major conformation( 64 %o ccupancy), the side chain has movedb y3 .6 relative to its position in the structure with shorter compound 1.T he first nitrogen atom of the tetrazole ring interacts with Tyr136-OH at ad istance of 2.7 and the third nitrogen atom with Lys210-NZ ata distance of 2.8 .T he aromatics tacking interactions are still present, with distances between the centers of the aromatic rings of 4.3 and 4.6 for Phe189 and Tyr181, respectively.

Structure of KDM4D with compound 3
In the case of an even longer ligand with n = 3 ( 3), the hydrazide group remains in the same position as those reported for both n = 1a nd 2. The tetrazole ring still interacts with Lys210-NZ, but it is shiftedb ya bout 1.8 away from the positiono bserved in the case of compound 1.I ti sn ow located at the position of Tyr136, and thus, forces the tyrosine aromatic ring further away,s ot hat the residue-ligand interaction is no longer observed ( Figure 3C).

Structure of KDM4D with compound4
Compound 4 is characterized by the substitution of the hydrazide with an ester group. It could be modeled with clear electron density in the binding groove. Its tetrazole group is coordinated by two residues,L ys245 and Tyr136. Lys245-NZ, the electron density of which usually identifies two distinct conformations, is clearly defined and pointingi nward into the active site ( Figure 3D). The two remaining nitrogen atoms of the tetrazole are hydrogen bondedt ow ater molecules and through one water molecule to the active-site metal ion. The orientation of the ligand is flipped by 1808 around the tetrazole ring relative to the orientation observed for compounds 1 and 2. The position of the tetrazole ring is 1.4 away from the tetrazole in the case of ligand 1.T he ester tail of the ligand spans the bindingc avity at its entrance, close to the surface of the protein, forming hydrophobic interactions with the protein.

Structure of KDM4D with compound5
Compound 5 is av ariant of compound 1 with am ethyl substituenta tp osition 1o ft he heterocyclic ring. The hydrazide group of the ligand is placed in the same positiona st hat in all ChemMedChem 2019ChemMedChem , 14,1828ChemMedChem -1839 www.chemmedchem.org 2019 The Authors. Publishedb yWiley-VCH Verlag GmbH &Co. KGaA, Weinheim structures described so far and it chelates the active-site metal ion. However, the whole decorated tetrazoleg roup is turned by 858,a nd its carbon atom points toward Tyr181, whereas the ring is placed between Asn202a nd Lys210 ( Figure 3E). The third nitrogen atom of the ring binds to aw ater molecule through ahydrogen bond.

Structure of KDM4D with compound 6
Compound 6 has at etrazole ring in the para position of the benzene ring. The benzene ring is placed 4.4 away from Phe189 and 5.4 away from Tyr181. The second nitrogen atom of the tetrazole ring is hydrogen bondedt oA sn284-N and is oriented parallel to Tyr136( Figure 3F). Compared with all compounds analyzed previously,t he hydrazide group is rotated by 908.T he nitrogen atom adjacent to the carbonyl group occupies the position of aw ater molecule, which is bound to the metal ion in previous cases,w hile the water molecule is now modeledi napositiono ccupied by the nitrogen atom of the hydrazide group in other cases.

Structure of KDM4D with compound 7
Compound 7 carriest he tetrazole ring in the meta positiono f the benzene ring. The hydrazide group binds in the same way as that observed in all other structures ( Figure 3G). The benzene ring is positioned between both functional groups and is engaged in stacking interactions with Tyr161 and Phe189 at distances of 4.3 and 5.1 ,r espectively.T he first tetrazole nitro-gen atom hydrogen bonds with Lys210 (2.9 )a nd the second nitrogen atom with Asn284( 3.1 ). The position of the ring is only 0.8 further away compared with the most similar ligand, compound 3.A lso, the aliphatic chain that links the tetrazole and hydrazideg roups in compound 3 ligand is found at the same position as that of the benzene group of this ligand.

Isothermal titration calorimetry (ITC)
The affinity and energetics of the binding of tetrazole ligands to KDM4Dw ere studied by meanso fI TC. In all cases, the ligand was titrated into the solution of protein in the calorimeter cell. Theb inding of compound 1 to KDM4D is exothermic and characterized by ad issociation constant, K d ,o f2 .3 mm.A l-  (27 166) 247 119( 39 529) 165 316 (26 447) 202 280 (32 624) 127 491 (20 523)1 95540 (31550)1 74 441 (28 048 thought he interaction is favored by both enthalpic (DH = À2.6 kcal mol À1 )a nd entropic (ÀTDS = À4.9 kcal mol À1 )c ontributions,t he entropic part dominates and has an early twofold higher impact on the Gibbs free energy of binding (DG = À7.5 kcal mol À1 ;T able 3). Compound 2 displayed ar ather large differential powers ignal, but this signal could not be attributed to binding. It most likely originated from side effects of the system,s uch as ligand dilution.N ob inding isothermsc ould be observedf or compounds 3-6.T he binding of compound 7 to KDM4D was found to be endothermic. With K d = 41 mm,i ti s weaker, relative to the binding of compound 1.This interaction is characterizedb ya nu nfavorable positive binding enthalpy (DH = 4.0 kcal mol À1 ), which is overcompensated for by the rather stronge ntropic contribution (ÀTDS = À9.9 kcal mol À1 ).

In vitro KDM4A inhibition assays
Ligands 1-5 were previously tested by us in af ragment-based screening against the closely related isozymeK DM4A. In brief, ligand 1 was the most potent compound, with an IC 50 value of 46.64 AE 0.94 mm in af ormaldehyde dehydrogenase (FDH)-coupled assay.I nt he antibody-based LANCE Ultra assay,t he potency was as high as 2.38 AE 0.37 mm.T he two control compounds, 4 and 5,i nw hicho ne structuralm otif required for potent binding (hydrazide or tetrazole) is replaced or masked, exhibit no inhibition. [19] The two derivatives 6 and 7,w ith an aromatic spacer,w ere characterized as noninhibiting. Apparent IC 50 values from both in vitro assays are reported in Table 4. The apparent potency is much higher in the FDH-coupled assay than that in the LANCE Ultra assay,w hich is in contrast to previous observations for this and other structural classes of inhibitors. [21] These resultsi ndicatea nassay artifact in the FDH assay.I ndeed,acounterscreening assay revealed that compounds 6 and 7 also negativelyi nfluencedt he conversion of formaldehyde by FDH in aKDM4A-free setup, potentially by directly inhibiting FDH, by reactingw ith released formaldehyde, or by quenching the fluorescences ignal (see Figure S1 in the Supporting Information). As such, the resultsf rom the orthogonal LANCE Ultra assay appear to be more reliable;t his means that compounds 6 and 7 are, at best, very weak KDM inhibitors. This is in agreement with the much less favorable binding properties observed in the ITC measurements (Table 3).

Discussion
The aim of this work wast oi nvestigate the mode of binding of the histone demethylaseK DM4Di nc omplex with tetrazolylhydrazide compounds. Structural data describe,i nd etail, the interactions of this molecular scaffold with the KDM4 cofactor bindings ite and might be very helpful in the furtherd evelopment of inhibitors of the KDM4 subfamily.I no ur opinion, the most desired compounds should exhibit high affinity to make them potent binders and, at the same time, they should possess physicochemical properties that would allow them to penetrate cells across membranes. Onlyt he combination of the two properties in one compound would identify ac ompound as one displaying high anti-KDM4 activity in vivo. To the best of our knowledge,t here is al imited number of KDM4 binding compounds, to date, that feature both properties [13b] and tetrazolylhydrazides might fill this gap.

Role of the carboxyl group in KDM4 inhibitors
Early developeda nalogues of 2OG usually contain either one or two carboxyl groups,f or example, N-oxalylglycine (NOG; PDB ID:4 D6S, unpublished) or 2,4-pyridinedicarboxylic acid (2,4-PDCA;P DB ID:4 D6Q, unpublished;F igures 4A,C ). In all cases, one of the carboxylic groups is positioned in the same way as that in the naturally occurring cofactor,2 OG. Other scaffolds, which contain one carboxylic group, are, for instance, 2,2-bipyridines [13d] (PDB ID:3 PDQ), 8-hydroxyquinoline (8HQ, PDB ID:4 BIS), and derivatives of 8HQ [13i] (8-hydroxy-3-(piperazin-1-yl)quinoline-5-carboxylic acid; PDB ID:3 RVH;F igure 4D-F). Several attempts were made to enhance the cell permeability of such KDM4 bindingm olecules. For instance, ester modifications were introduced to produce ap rodrug, for example, in the case of MethylStat, which is apan-KDM inhibitor. [22] However,i th as been reported repeatedly that the carboxyl group was needed to maintain the potency of the inhibitor molecules. This can easily be explained by examining the interactions of the carboxyl groups with the enzyme, whicha re highlighted in several crystal structures of KDM4 complexes with ligands ( Figure 4). Another strategy to eliminate carboxylate [a] K d , DH,a nd N (stoichiometry) are directlym easured parameters, which are determined from the fit of binding data.T he errori st he fitting error, so it showsh ow good the fit (and corresponding bindinge quation) describesthe experimental data. N < 1means that the real concentration of binding protein is lower than the total protein concentrationi nt he sample (as determined by UV spectroscopy). The Gibbs free energy and entropicc ontribution to binding are estimated from the following equations: DG = RTln(K d )a nd DG = DHÀTDS. groups,d ifferent from as imple ester modification, is their replacementb yi sosteric groups. One such group is the tetrazole moiety.G iven that, it seems surprising that no such structures of complexes have been reported for KDM4 proteins, to date, althought etrazolylhydrazide compounds in KDM4 inhibition were first reported in 2015 by Rüger et al. [19]

Role of the hydrazide group in KDM4 inhibitors
Hydrazide-containing ligands display ab identate coordination of the metal ion. This is clearly evidenced by the structures of KDM4D in complex with compounds 1 to 3 and 5 to 7.S ubstitution of the hydrazide group in compound 4 by as imple ethyl ester was not tolerated and resulted in ad ifferent mode of binding of compound 4 to KDM4D with respectt ot he aliphatic part of the molecule. It seems as if the unmodifiedh ydrazidei sn ecessary for bindingt ot he active-site metal ion, but also for interacting with neighboring active-site residues, Ser200 and Asn202. Amonga ll reported structures,t od ate, the nitrogen atom coordinating the metal in the same position as that in the hydrazide ligands described herein is observed only in daminozide, which is ac ompound described as ap otential JmjC oxygenase inhibitor (PDB ID:4 AI9). [13p] Closer inspection of the hydrogen-bonding interactions clearly shows that the dimethylamino group in daminozide prevents binding to Ser200a nd Asn202 ( Figure 4B). This interaction is observed in our compounds that exhibit low micromolar affinity:c ompounds 1 and 7.T he binding potencyo ft he hydrazide is also shown in the case of compound 6.T he benzene ring is incorporated between the two main functional groups of the ligand;thus rigidifyingi t. Steric clashes prevent it from binding in the same bidentate manner as that reported for other tetrazolylhydrazide ligandsa nd for 2OG. Nevertheless, it is still chelating the metal ion by replacing the water molecule for ah ydraziden itrogen atom. Consequently,t he hydrazide appearst o be av ery powerful motif for chelatingt he metal ion in the active site of KDM4 enzymes.

Tetrazole ring as ac arboxyl group isostere
The second carboxyl group of the cofactor,w hich is distal to the chelated metal ion, was replaced by at etrazole ring. It was anticipated that the tetrazoler ing or its anionic counterpart, deprotonated tetrazolate, would occupy as imilarp osition to that of the carboxylate group of 2OG and would thus maintain the interactions that were reported to have ag reat impact on binding affinity ( Figure 4). Also, apart from the binding affinity, the hope was that, by introducingt he tetrazole group, other properties would be introduced that would turn out to be advantageousf or further developmenti nto drugs against KDM4 enzymes.S ubstitution of the carboxylic acid with the tetrazole moiety should make the compounds more lipophilic, and thus, improvec ell permeability.I ti sc lear that this notionr emains speculative, until cell penetration experiments have been performed. However,i th as been shown previously that this small, nitrogen-richf unctional heterocycle can serve simultaneously as an aromatic platform and as af unctionali nteraction motif. [20] Tetrazoles were reported as KDM4 bindersb yR üger et al. in 2015. [19] Amongo ther reported KDM4 inhibitors, the most similar moiety to tetrazole is at riazole, which has been incorporated into ap yridinec arboxylate scaffold. Many such derivatives have been synthesized and tested. [13f] However, their spatial position and orientation in the active site of the KDM4 enzymes is very different from that observed in the ligands reported herein. In contrast, tetrazole rings of ligands presented herein are never engaged in direct metal chelation, although an indirect, water-mediatedm etal interaction could be observed for compound 4.I nm osto ft he structures reported herein, the tetrazole ring is sandwiched between two aromatic residues in the actives ite, Phe189 and Tyr181. This position must be considered favorable based on the inspection of all of the structures. Ap roof of concept comesf rom the analysis of the structure of KDM4D in complex with compound 4. Compound 4 does not have ah ydrazide group chelating the metal ion, and thus, does not bind the metal directly.T his lack of metal chelation explains why compound 4 cannotc ompete efficiently with the co-substrate 2OG andd oes not inhibit KDM4. Nevertheless, its tetrazole ring is retained in the same positionv ery close to those observed in the other tetrazole compounds 1 to 4 and 6 to 7.I tm ust therefore be concluded that this positioni sf avorable for an aromatic platform that is able to engage in stacking interactions. The presence of two aromatic systems, for example, at etrazole and ab enzene group, for example, in compound 7 seems to increaset he entropic contribution of the binding in KDM4D,b ut not in KDM4A.A sam atter of fact, it is much higher than that in the case of ligand 1.T he position of the aromatic system, as reported herein, has not been observed previously,p robablyb ecause most of the previously reported KDM4 inhibitors possess an aromatic ring as part of the metal chelating motif.

Role of the active-siteresidues of KDM4D
The obtained high-resolutions tructures allow an investigation of the conformational changes of the amino acid residues in the active site upon ligand binding. Amino acidsL ys210, Tyr136, Ser200, and Asn202 are all engaged in binding of the natural cofactor,2 OG. The positions and conformations of these residues are more or less the same in the cofactorbound and -free states of the enzyme. This suggestst hat the active site is in ar eady state to accommodate the cofactor in a lock-and-key binding mode. The same position of theser esidues is also observed upon binding of compound 4 in the distal area of the active-site cleft. In most instances, Asn202 points away from the metal center,e xcept in the structure with compound 6,i nw hich the metal is chelated differently.A superposition of KDM4D structures in complexw ith tetrazolylhydrazides and aK DM structure in the presence of 2OG leads one to assume that binding of the tetrazolylhydrazide to the protein induces ar estructuring of the active-site residues. Asn202i snow in an "inward"p osition if the hydrazide group chelates the metal ion in the same bidentate manner as that in the natural cofactor binding,t hanks to its internal nitrogen atom. This suggests an induced-fit modelf or tetrazolylhydrazide bindingt oK DM4D. Interestingly,s ome of the compounds describedh erein inducet he movement of Tyr136, whiche nlarges the bindingc left of the natural cofactor( Figure 5);t hus making room for even more ligand atoms. It could be clearly observedi nt he complex structure with compound 7 that the tetrazolering pushes Tyr136 away from Lys210 from ad istance of 3.4 to 9.7 .A tt he same time, the tetrazole establishes a new interaction with Asn284. This residuei su sually buried. It is ab it surprising, however,b ecause similari nteractions have been reportedt wicep reviously. [13g] They were observed in a pyrazolopyrimidine ligand structure (PDB ID:5 KR7) [13g] and in its unmethylatedf orm (PDB ID:5 FJK). The ligand architecture that triggersm ovement of Tyr136 was reported to be essential for maintaining ah igh binding constant.H owever, the loss in the enthalpy contribution of ligand 7 compared with 1 is significant and might be due to losing the strong hydrogen bond with the hydroxy group of Tyr136. The mapping of flexible residues in the active side of KDM4Dmight be helpful in aprocess of rational design of higher affinity compounds.

Conclusion
By describing as eries of tetrazolylhydrazide compounds and by examining their respective binding modes to the target protein KDM4D,w ehave unveiled severalu nique interactions that were not observed previously and could potentially be usefuli nf urther inhibitor optimization. The tetrazole ring appears to be as uitable replacement for carboxylg roups. Because of its uniquef eatures, it can establish both aromatic and polar interactions. The hydrazide moiety seems to be av ery powerful motif for metal chelation in KDM4 proteins. In contrast to other previously described compounds that target KDM4 proteins,t he new compounds discussed herein might display more desirable physicochemical properties. The new scaffold fills an important gap in KDM4 inhibition and opens up an ew path to developing compounds with high affinity and possibly increased permeability, at the same time. The tetrazolylhydrazides caffold also presents severalp ossibilities for ligand extension and branching. This would allow the creation of al igand that could extend into other binding sites, for instance,t he histone bindings ite, which has not yet been explored in the presents eries of compounds.

Ligands ynthesis
The synthesis of compounds 1-5 was described previously by Rüger et al. [19]  Protein expression and purification cDNA encoding the catalytic domain of human KDM4D, comprising residues 1-342, and KDM4A, comprising residues 1-359, was purchased from Source Bioscience (Nottingham, UK). It was cloned into the pQTEV expression vector,w hich encoded an N-terminal hexahistidine tag. The proteins were purified by means of affinity chromatography.T he binding buffer consisted of 50 mm HEPES, pH 7.5;5 00 mm NaCl;2 0mm imidazole;a nd 1mm tris(2-carboxyethyl)phosphine (TCEP). Protein was recovered from resins by using elution buffer consisting of 50 mm HEPES, pH 7.5;5 00 mm NaCl; 300 mm imidazole;a nd 1mm TCEP.T he His-tagged protein was then processed by TEV protease to remove the N-terminal affinity tag prior to further purification. In the final stage of purification, the protein sample was subjected to size-exclusion chromatography in the following buffer:1 0mm HEPES, pH 7.5;3 00 mm NaCl; 5% (w/v)g lycerol;a nd 0.5 mm TCEP. Data collection, processing, structure determination, and refinement

Protein crystallization
For data collection, crystals were presoaked in well solution supplemented with 10 mm NiCl 2 for 10 min. Subsequently,t hey were transferred to the ligand soaking solution, consisting of 100-150 mm ligand in well solution (10-15 %f inal DMSO concentration), and incubated from 30 min to 24 h. The crystals were then cryoprotected by quickly immersing them in well solution complemented with 20 %( v/v)e thylene glycol and then flash-cooled in liquid nitrogen. [23] XRD data were collected on beamlines BL14.1 and BL14.3 at the BESSY II electron storage ring operated by the Helmholtz-Zentrum Berlin [24] by using aP ILATUS 6M and aR ayonix MX225 charge-coupled device (CCD) detector,r espectively.T he data were integrated and scaled by using XDSAPP. [25] All relevant data collection and processing statistics are given in Ta ble 1. The structures were solved by simple molecular replacement by using ap ublished KDM4 structure (PDB ID: 4 HON [13m] )a sas earch model and the program Phaser. [26] The resulting model was initially refined with REFMAC and subsequently by phenix.refine. [27] Atomic displacement parameters (ADPs) were refined anisotropically for all protein atoms individually.T he ligand occupancy values were determined by means of refinement by using phenix.refine from starting values of 0.75 in each case. The final model was obtained after several cycles of refinement and manual adjustments by using Coot [28] and validated. All refined coordinate sets and the corresponding structure factor amplitudes have been deposited in the Protein Data Bank. All refinement statistics are presented in Ta ble 2.

KDM4A FDH assay
The FDH-coupled demethylase activity assay was adapted from ref. [29] and performed in at otal volume of 20 mLo nw hite Opti-Plate-384 microtiter plates (PerkinElmer,W altham, MA, USA) with 50 mm HEPES buffer at pH 7.50 containing 0.01 %T ween-20. As olution of KDM4A 1-359 (0.10 mg mL À1 ,2 .4 mm)w as preincubated with compound solutions of varying concentration (0-400 mm)i n DMSO at room temperature for 10 min. As ubstrate solution containing 100 mm ascorbic acid, 10 mm FeSO 4 ,0 .001 U mL À1 FDH, 500 mm NAD + ,5 0mm 2OG, and 35 mm H3K9me3 substrate peptide ARK(me3)-STGGK-NH2 (Peptide Specialty Laboratories, Heidelberg, Germany) was added (final concentrations). The final DMSO concentration was 2% in all wells. The fluorescence intensity of the product formed, NADH, was measured at l ex = 330 nm and l em = 460 nm on aP OLARstar Optima microplate reader (BMG Labtech, Ortenberg, Germany) immediately after addition (t = 0) and after 1h incubation on ah orizontal shaker at 37 8C. Values were blankcorrected and the difference in intensity at t = 1a nd 0hwas taken as ameasurement of enzyme activity.Activity,inpercent, was compared with that of compound-free DMSO control and no-substrate negative control. Inhibition curves were analyzed by means of sigmoidal curve fitting by using GraphPad Prism 4.00 and IC 50 values calculated from the fit parameters as mean AE standard deviation (SD) from two independent experiments.

KDM4A LANCEUltra assay
The commercial antibody-based LANCE Ultra demethylase activity assay (PerkinElmer,W altham, MA, USA) was performed in at otal volume of 10 mLo nw hite OptiPlate-384 microtiter plates (PerkinElmer) with 50 mm HEPES buffer at pH 7.50 containing 0.01 %T ween-20 and 0.01 %B SA. As olution of 60 nm KDM4A 1-359 was preincubated with compound solutions of varying concentration (0-1000 mm)i nD MSO at room temperature for 10 min. As ubstrate solution containing 100 mm ascorbic acid, 5 mm FeSO 4 , 1 mm 2OG, and 400 nm biotinylated H3K9me3 substrate peptide ARTKQTARK(me 3 )-STGGKAPRKQLA-GGK(biotin) (BPS Bioscience, San Diego, CA, USA) was added (final concentrations). The final DMSO concentration was 5% in all wells. Plates were incubated on ah orizontal shaker at room temperature for 45 min. Reactions were stopped by the addition of 10 mLo fd etection mix containing 2nm europium-labeled anti-H3K9me 2 LANCE antibody (PerkinElmer), 50 nm ULight-streptavidin dye (PerkinElmer), and 1mm EDTAi n1 LANCE detection buffer (PerkinElmer;f inal concentrations). Plates were again incubated on ah orizontal shaker at room temperature for 60 min. FRET intensity was measured on an EnVision 2102 multilabel plate reader (PerkinElmer) at l ex = 340 nm and l em = 665 nm with ad elay of 100 ms. Values were blank-corrected and activity in percent was compared with compound-free DMSO control and noenzyme negative control. Inhibition curves were analyzed by means of sigmoidal curve fitting with GraphPad Prism 4.00 and IC 50 values calculated from the fit parameters as mean AE SD from two independent experiments.

FDH counterscreening assay
To evaluate the effect of test compounds on the FDH detection system, am odified assay was performed. The test was performed on white OptiPlate-384 microtiter plates (PerkinElmer,W altham, MA, USA) and the total test volume was 20 mL. The assay buffer was 50 mm HEPES at pH 7.50 and 0.01 %T ween-20. Formaldehyde (40 mm)i na ssay buffer was preincubated for 10 min with com-  (10, 100, 400 mm)i n DMSO at room temperature. Asolution containing 100 mm ascorbic acid, 10 mm FeSO 4 ,0 .001 U mL À1 FDH, 500 mm NAD + ,a nd 50 mm 2OG was added. The final DMSO concentration was 2% in all wells. The fluorescence intensity of the product NADH was measured at l ex = 330 nm and l em = 460 nm on aP OLARstar Optima microplate reader (BMG Labtech, Ortenberg, Germany) immediately after addition (t = 0) and after 1hincubation on ah orizontal shaker at 37 8C.
Values were blank-corrected and the difference in intensity at t = 1 and 0h was taken as am easurement of FDH activity.T he activity, in percent, was given relative to that of the compound-free DMSO control and no-enzyme negative control.

ITC measurements
ITC measurements on KDM4D and the respective tetrazolylhydrazide ligands were performed by using aM icroCal PEAQ-ITC microcalorimeter (Malvern Panalytical GmbH, Germany). Experiments were performed in 10 mm HEPES buffer at pH 7.5, 0.5 m NaCl, 5% (v/v)g lycerol, and 1mm TCEP at 20 8C. The protein concentration was determined spectrophotometrically at l = 280 nm by using the calculated molar extinction coefficient. The tetrazole ligands were diluted 100-250 times in buffer up to the working concentration from stock solutions in DMSO. The DMSO concentration in the resulting solutions ranged from 0.4 to 1%.S olutions of KDM4D were supplemented with the corresponding amount of DMSO to avoid thermal effects as ar esult of buffer mismatch. Te trazole ligands were titrated in 13 or 19 small steps (2-3 mL) into asolution of protein in the calorimeter cell. Other experimental settings included a spacing time of 180 sa nd af iltering period of 5s.F or each ligand, all measurements were performed two or three times. For all experiments, the instrument software (MicroCal PEAQ-ITC Analysis) was used for baseline adjustment, peak integration, and normalization of the reaction heats, with respect to the molar amount of injected ligand, as well as for data fitting and binding parameter evaluation. The thermodynamic parameters for ligands showing measurable heat signals are presented in Ta ble 3.