Virtual Fragment Screening Identification of a Quinoline‐5,8‐dicarboxylic Acid Derivative as a Selective JMJD3 Inhibitor

Abstract The quinoline‐5,8 dicarboxylic acid scaffold has been identified by a fragment‐based approach as new potential lead compound for the development of JMJD3 inhibitors. Among them, 3‐(2,4‐dimethoxypyrimidin‐5‐yl)quinoline‐5,8‐dicarboxylic acid (compound 3) shows low micromolar inhibitory activity against Jumonji domain‐containing protein 3 (JMJD3). The experimental evaluation of inhibitory activity against seven related isoforms of JMJD3 highlighted an unprecedented selectivity toward the biological target of interest.

Jumonji domain-containing protein 3(JMJD3), along with ubiquitously transcribedX chromosome tetratricopeptide repeat protein (UTX) constitutes the KDM6 subfamily,w hich catalyzes the demethylation of lysine27onhistone H3 (H3K27). Both enzymes play key roles in the epigenetic regulation of gene expression,a ltering cellular memory,a nd reprogramming cellular fate. These proteins share ah ighly homologous Jumonji C domain endowed with Fe 2 + and a-ketoglutarate for the demethylationo fH 3K27. [1] Overexpression of JMJD3 is correlated with inflammation, [2,3] neurologicald isorders, [4] as well as cancer pathologies [5] such as T-cell acute lymphoblastic leukemia, [6] Hodgkin's lymphoma, [7] and metastatic prostate cancer. [8] It hasa lso been recently shown that JMJD3 can be an ew targetfor pediatric brainstem glioma. [9] Because JMJD3isaninducible enzyme, its suppression could be very attractive for cancer treatment. Moreover,t he specific biological function of JMJ enzymesi nr egulating cellular processes is still poorly understood due to the absence of selective inhibitors. Despite a vast body of work investigating the function of this protein in recent years, only one JMJD3/UTX binder (GSK-J1) has been reported to date. [1] Recently,aseries of GSK-J1 derivatives was reported, showing activity similar to or lower than that of the reference compound. [10] Thus, the discovery of small molecules that are able to selectively modulate the biological functiono f JMJD3 is very attractive, in that they will shed light on itsr ole, both in normal biological processes and under diseasec onditions, expanding the cancert herapy toolkit.
In detail, by using an in silico approach, we screened af ragment library of metal chelators (Supporting Information Figure S1) that was previously proposed to develop metalloprotein inhibitors. [11,12] As protein model, we used two availableXray structures of JMJD3 (PDB IDs 4ASK as Model Aand 2XXZ as ModelB), as structural experiments revealed differents patial rearrangements of some residues and of the Fe 2 + ion upon GSK-J1 binding. [1] Following the same strategy adopted for our previousi nvestigations of metalloproteins, [13][14][15][16][17][18] the charges of iron and its coordinating amino acids (H1390, E1392, and H1470) were refined by DFT calculations (see experimental details in the Supporting Information),a nd they were subsequently used for molecular docking calculations. Based on our analysis, we selected the quinoline-8-carboxylic acid fragment (B11, Figure S1, Supporting Information), which was advantageously accommodated into the a-ketoglutarate cavity.W ith respectt ot he other fragments, its docked pose is deeply positioned, maximizing interactions with macromolecular counterparts. The docked pose suggested am odification at C8 to increase the interaction network with JMJD3;a ccordingly,w ei nserted as econd carboxylic acid group to interact with K1381, T1387 and N1400. The dockingp ose of the modified fragment (B11')s howedt he establishment of the foreseen interactions withouta ffecting its globalc onformation. We also observed that the docked pose of B11' orientst he C3 positiont oward a small cavity adjacent to the a-ketoglutarate pocket (Supporting Information FigureS2). Therefore, to identify additional possible interactions with this pocket by chemical decoration of the C3 positiono fB 11',w ep erformed an AutoLigand [19,20] analysis. This investigation (Supporting Information Figure S2) suggested the advantageous placemento fH -bond donors/ac-ceptors close to residues R1246 and N1331, as well as hydrophobic substituents to interactw ith other delimiting residues (F1328, T1330, T1387, andP1388). Thus, we designed asmall library inserting chemically diversea romatic substituents endowed with H-bond acceptor/donors (Scheme 1a nd Supporting Information Figure S3), and docked them on both protein conformations (ModelsA and B). Based on molecular docking energies and visual inspection, the docking outcomes of all tested compounds (3-68,S cheme 1a nd Supporting Information Figure S3) led to af ocusedl ibrary of quinolined erivatives (3-12,S cheme 1), useful to providei nformation for structureactivity relationships.
The docked poses of compounds 3-12 into Model Ah ighlighted the respecto fapattern of similar interactions by the commons tructural portion. Indeed, they coordinate the Fe 2 + ion in ab identate manner by the carboxylate group at C5, whereas the second carboxylic function establishes ionic interactions with K1381 and forms H-bonds with T1387 and N1400. Notably,t hese interactions were observed in the co-crystal structurew ith GSK-J1. [1] The quinoline ring forms p-p interactions with Y1379 and is H-bonded to T1387. Compounds 3-12 differ in terms of the interaction given by the substituent at C3 (Scheme 1). Indeed, the methoxy group at C2 of the pyrimidine ring of 3 establishes H-bonds with R1246 and N1331, and the nitrogen atom at position1is H-bonded with N1331 (Figure 1a).
An interaction is observed between the pyrimidine ring and N1246 (Figure 1a). The second methoxy group establishes van der Waals interactions with the side chains of F1328, T1387, and P1388 (Figure 1a).  (Figure S8 a). The linear chain oxygen atom of 9 accepts an Hbond from R1246, andt he 2-methoxyethylphenyl moiety undergoes van der Waals interactions with P1388 and H1390 (Figure S9 a). The substituent at C3 of 10-12 establishesv an der Waals contacts with F1328, T1387, P1388, H1390, and L1433 ( Figures S10 a-12 a). Concerning the theoretical resultso n Model B, we observed that 3 is well accommodated into the binding pocket (Figure 1b), keeping the same docked pose and interactions with protein residues found for the predicted conformation into Model A. In contrast, for compounds 4-12 (Figures S4 b-S12 b) we found only partial accommodation into the protein cavity,w here somec ontacts with T1330 and F1328 are lost, especially for 9-12,w hicha re endowed with bulky substituents at C3 (Figures S9 b-S12 b). In greater detail,w e observed that 4-12 coordinate the Fe 2 + ion in am onodentate manner (Figures S4 b-S12 b). Although 4 and 5 establish p-  cation interactions with R1246, they do not correctly orient the sulfonamide and carboxylic groups to interactw ith R1246, as observedf or Model A ( Figures S4 and S5). Compounds 6 and 7 are H-bonded with R1246, by ap yridazine ring and methoxy group, respectively ( Figures S6 and S7).
To evaluate the stabilityo ft he complexes between JMJD3 and compounds 3-12 obtained by docking analysis, we performed molecular dynamics simulations( 50 ns, 310 K; see experimental details in the Supporting Information). [21,22] The trajectory analysis revealed that 3 gives ah ighn umber of contacts with protein residues, and it maintains most of the contacts observed from the docked pose during the entire simulation (> 50 %) with both protein models (Figure 2, bottom). The heavy-atom-positional RMSD of 3 shows high stabilityd uring molecular dynamics simulations with respectt ot he protein backbone, with similarb ehavior on both protein models (Figure 2, top). The atom-relative orientation of 3 is kept during the simulationw ith ModelsAand B. The trajectory of 4 and 6 bound to Model Aw as found to be stable, but with a larger RMSD than 3-Model A( Figures S13, S15), whereas the remaining compounds showed large fluctuations during the simulation (Figures S14, S16-S21).L arge deviations from their initial positions were observed for 4-12-Model Bc omplexes (Figures S13-S21). Compounds 4-12 bound to both modelsg ive a lower number of contacts during the simulations relative to 3 ( Figures S13-S21). Overall, the comparison of docking results on both models, integrated by molecular dynamics, suggested higher activity of compound 3 with respectt o4-12 as experimentally confirmed (see below).
In conclusion, we have identified the novel quinoline-5,8 dicarboxylic acid scaffold by af ragment-based approach to develop selective JMJD3 inhibitors. This unprecedented result affords the possibility of shedding light on the role of this enzyme in normal and altered tissues related to pathological events, such as cancer andi nflammation. Moreover,t he obtained outcomes validate our in silico strategy based on molecular dockings tudies against two JMJD3 conformationsi ntegrated by molecular dynamics simulations, which allows the identification of potentialb indersf rom virtual screening. These encouraging resultsp rompt us to further explore other chelating fragments to design new and potent JMJD3l igands for safer cancert reatment, as well as therapies for inflammation and neurological disorders.

Experimental Section
Full experimental details are provided in the Supporting Information.