Screening of a Novel Fragment Library with Functional Complexity against Mycobacterium tuberculosis InhA

Abstract Our findings reported herein provide support for the benefits of including functional group complexity (FGC) within fragments when screening against protein targets such as Mycobacterium tuberculosis InhA. We show that InhA fragment actives with FGC maintained their binding pose during elaboration. Furthermore, weak fragment hits with functional group handles also allowed for facile fragment elaboration to afford novel and potent InhA inhibitors with good ligand efficiency metrics for optimization.

Our findings reportedh erein provide support for the benefits of including functional group complexity (FGC) within fragments when screening against protein targetss uch as Mycobacteriumt uberculosis InhA. We show that InhA fragment actives with FGC maintained their binding pose during elaboration. Furthermore, weak fragment hits with functional group handles also allowed for facile fragment elaboration to afford novel and potent InhA inhibitors with good ligand efficiency metrics for optimization.
We recently reported on the design and synthesiso fanovel fragment library of diversef ragments which include functional group complexity (FGC). [1] Functional groups, [2] or "chemical handles",w ere incorporated onto diverse scaffolds to allow for additional interactions with target proteins with the aim of maintaining binding poses during optimization, as well as aid fragment elaboration. [1] Howeverc are has to be taken because increasing complexity in af ragment decreases the probability of it achieving optimal ligand-protein interactions. [3] Conversely,t oo little complexity can lead to interesting interactions being missed. [4] In this respect, fragment deconstruction studies on b-lactamase inhibitors, [5] suggest that small fragments with "minimal complexity" derived from the deconstructiono f potent inhibitors do not always retain the bindingp oses of the parentm olecules. Whereas, fragments with built in FGC recapitulatet he larger potent inhibitor binding mode. Therefore, a careful balance is required to identify as weet spot of complexity where detectable, single-mode binding of al igand to a target is most probable. There is further support for this concept based on other fragment deconstruction case studies. [6] Despite the molecular complexity model and its putative applications in fragment screening were introduced and refined by Hann [3] and others over nearly two decades ago, to date the authors are not aware of any reported fragment screening against an ovel diverse fragment library designed to include FGC. Our recently reported FGC libraryi sb ased on synthetic chemistry toward selected functionalg roups. [1] However, recent reports of an algorithm to identify all functional groups in organic molecules, allows for the analysiso fF GC in large chemicaldatabases of commercial fragments. [7] Herein we describet he screening of such al ibrary,i nc omparisonw ith other fragment sets, against the highly validated Mycobacterium tuberculosis (Mtb) target InhA. [8] Mtb is the causativea gento ft uberculosis (TB), which is currentlyt he leadingi nfectious disease killer worldwide. [9] Isoniazid (INH, Figure 1), as uccessful frontline TB drug for more than 50 years, targets the NADH-dependent2 -trans enoyl-acyl carrierp rotein   (ACP) reductase InhA. This is ak ey enzyme in the Mtb cell wall synthesis pathway [10] and does not have ah uman orthologue. The development of Mtb resistant strains to standard anti-TB drugs, [11] includingI NH, necessitates the need for novel Mtb targeted therapies. Resistance to INH developsm ainly via mutations in the Mtb KatG enzyme, which converts INH into an acyl radical, which covalently binds to NADH and the resulting adduct inhibits InhA. [12] Direct InhA inhibitors are envisaged to bypassthis resistance mechanism andmaintain clinicalefficacy. Accordingly,t here hasb een widespreadr esearch in this field [13] and whilst limited set of potent direct InhA inhibitors with activity against INH-resistant strains have been identified (1-3 [14] in Figure 1), none have been progressed into clinicald evelopment. Hence, there remains an eed to identify novel direct InhA inhibitor scaffolds.
InhA inhibitors are known to modulate the tertiarys tructure of the InhA protein binding pocket, in particular the substrate binding loop (SBL). [15] In this respect, af ragment based (FB) approach [16] was considered appealing in order to assesst he InhA protein conformations for fragment actives and the structural requirementsf or their optimizationi nto potent InhA inhibitors.
For the above reasons, we screenedt he recently reported FGC fragment library (FGC-FRAG), [1] as well as an informed InhA commercial fragment set (InhA-INF-FRAG), which was compiled based on the known direct InhA inhibitors in the public domain (see Supporting Information). The above libraries were screened alongside an historical commercial fragment library (HIST-FRAG), ar eported 3D fragment library (3D-FRAG) [17] and fragments derived from inventory (INV-FRAG)a nd project (PROJ-FRAG)s ources. The overall library constituted 1360 fragments (Figure 2A), which were screened against the NADH bound form of the InhA, using saturation transfer difference (STD) 1 HNMR (complete results in SupportingI nformation).
STD-NMRt ypicallyi dentifies ligandst hat bind weakly to moderately to protein targets. [18] The criteria forabinding event used here was ap ositive STD signal intensity whichw as decreased by at least 50 %o nt he addition of the known inhibitor 1 (R = Me). [14a] This resulted in 149 hits (11% hit rate). A breakdown of these hits based on their source is given in Figure 2B.D ue to its binding affinity being in the suitable range (K d % 5 mm), [19] NADH binding was also observed in the STD-NMR spectra.I tw as noted that the stronger binders 1 (R = Me) and 3 (R = CH 2 iPr) [14b] decreased the STD-NMR intensities for the NADH co-factor peaks. Therefore, greateri mportance was given to those fragments which also caused ad ecrease in the NADH STD peak intensities, as this was considered as evidence of stronger binding. This furthers election step decreased the number of hits to 32 (4-35 in Figure 3; 2.4 %h it rate). The pie chart for the source of these 32 hits is given in Figure 2C.T his process increased the fraction of hits from the FGC-FRAG set (29 %t o4 1%). These data are interesting considering the FGC-FRAG set only constituted 24 %o ft he whole screening library. The initial hit rate for the InhA-INF-FRAG set was low,a lthough the size of the library was small. This may be the result of a lack of InhA fragment inhibitors that can be purchased from vendors, as observed for deconstruction of kinase inhibitors from the public domain. [20] The two hits derived from this library did, however,s urvivet he second selection step and could also be classified as FGC fragments. Ah igh proportion of project, historical deriveda nd 3D fragment actives were also noticeably enriched with FGC.
Surface plasmon resonance (SPR) equilibrium dissociation binding constants( p K d )w ere also determined, and found to correlate with InhA biochemical pIC 50 values (Table 1). The InhA-NADH co-crystal structuresf or FGC fragments 22 (pIC 50 = 3.1) and 24 (pIC 50 < 3) were found to be in accordance with the FGC-FRAG set design principle, in that additional interactions were observed for their FG with the InhA protein ( Figures S1 and 4A). The co-crystal structure of 24 contains a unique tetramer with four slightly different binding sites. The followingd escription is for one of these, details of the others are shown in Figure S2). The amide carbonyl FG of 24 forms an H-bond interaction with the InhA Tyr158 residue and the NADH ribosyl-20-OHg roup, while the NH H-bonds with Met199. Furthermore,t he pyrimidine ring of 24 picks up an additional H-bond interaction with the backbone NH of Met98, a feature also observedf or the thiadiazole core of ac lose analogue (PDB ID:4 BQP) of advanced lead 2,w hich,h owever, does not interactw ith Tyr158. [19] Whereas lead 3a (PDB ID: 4R9S) has as imilar H-bond interaction with Tyr158 and the NADH ribosyl-20-OHgroup. FGC fragment 24 is able to identify distincti nteractions associated with both leads 2 and 3a (Figure 4A).
Fragment 24 was found to overlay well with HIST-FRAG urea 12 (pIC 50 < 3.0), with the embedded urea FG having similari nteractions with Tyr158, Met199 and the NADH ribosyl hydroxy group ( Figure 4A). Furthermore, the pyridyl group of 12 occupies the same space as the lipophilic di-methyl cyclohexane group in lead 3a ( Figure S3), however,i ts nitrogen is predicted to be protonated and binds to the carboxylicm oiety of Glu219.O verlay of the urea 12 with the amide 24 led to synthesis of the merged amide 36 and urea 37 (Figure 5a nd SchemeS1). Amide 36 was inactive in the biochemical screen (pIC 50 < 3), butu rea 37 had improved InhA biochemical poten-  (Figure 5a nd Scheme S1). Advanced lead 3a was overlaid with urea 37 (Figure S4), and the GSK inventory searched to identify core fragment replacements of 3a.T his led to the identification of pyridinone 40 [21] (Figure 5), with good biochemical potency (pIC 50 = 4.8). The co-crystal structure of 40 showed the conserved functional groups overlaid with both the advanced leads 3a and urea 37 ( Figure S5). Replacement of the phenylg roup of 40 with the novel pyrimidine fragment afforded compound 41 (Figure 5a nd SchemeS2), with improved enzymatic activity (pIC 50 = 6.0). Once again, the overall pose of the individual FGC fragment components of 40 and 41 was conserved ( Figure 4C). The novel pyrimidines 42 and 43 (Figure5 and Scheme S3) were also prepared, based on 3a (pIC 50 = 6.2), and showedg ood InhAb iochemical potency (pIC 50 = 6.5 and 5.2, respectively).
The interaction with Tyr158, previously described for the urea and pyridone derivatives, is not present in the crystal structures obtained for pyrazolef ragment hits 4, 9 ( Figure 6A) and 34 (Figure 6B), as the Tyr158 side chain forms aw atermediated bridge with NADH. However,t he pyrazoler ings occupy the same sub-pocket, stack againstt he nicotinamide ring of NADH, and the 2-N-pyrazole providesa nH -bond interaction with the 20-OH group of NADH. Fragment 9 is more potent in the biochemical screen (pIC 50 = 3.4) and its 5-NH 2 group forms water-bridged interactions with the NADH phosphate group asw ell as Met199 and Thr196( Figure 6A). In contrast, pyrazole 4 (pIC 50 < 3) has a slightly alteredb inding mode, where the hydroxy group has an additional interaction with Met98 backboneN H( Figure 6A). In accordance with previous fragment deconstruction studies, [2][3] the functionally complex InhA informedf ragment 34 (pIC 50 < 3) retains the binding pose observed for published ad-vancedI nhA lead 45 [10c] (PDB ID:5 JFO, Figure 6B). The 3N of the thiadiazole ring and NH side group of 34 and 45 H-bonds with Met98 backbone.
In conclusion, herein we report on the identification of novel InhA fragment hits using STD-NMR screening,a sw ella s orthogonal InhA biochemical and SPR assays.H igh hit rates were obtained from screening ar ecently reported novel fragment set with built FGC versus other fragment sets. Notably, startingf rom weakly active FGC fragments facilitated rapid fragment based lead generation (FBLG) due to 1) easy chemical tractability and derivatization, 2) retention of the functional group binding pose during fragment evolution througha dditional interactionsw ith the target, which was confirmed by Xray studies. Conversely,e laboration of weakly bound InhA frag-     ments with minimal FGC, relies on FGC implementation,w hich is likely to alter bindingc onformation. [23] This is ac ommon issue associated with fragment optimization, resulting in structure-activity relationship disconnections which are often difficult to interpret. Our findings are also in agreement with the molecular complexity theory by Hann et al., [3,24] for which moderately complex ligands, like the identified FGC fragment hits, have ah igher probability of a" usefule vent", that is the detection of aunique binding pose. These resultsr eportedh ereprovide support for the rational design, synthesis ands creening of novel diversef ragments with built in functional groups. The described InhA FB-leads showed good InhA enzymatic activity as well as ligand efficiency (LE) metrics. [25] Additional optimization effortshave resulted in furtherimprovedInhA biochemical as well as Mtb whole cell potency, which will be reported elsewhere.