Fragment-based discovery of novel thymidylate synthase leads by NMR screening and group epitope mapping

Abstract

Solution-state nuclear magnetic resonance (NMR) is a versatile tool for the study of binding interactions between small molecules and macromolecular targets. We applied ligand-based NMR techniques to the study of human thymidylate synthase (hTS) using known nanomolar inhibitors and a library of small molecule fragments. Screening by NMR led to the rapid identification of ligand pairs that bind in proximal sites within the cofactor-binding pocket of hTS. Screening hits were used as search criteria within commercially available sources, and a subset of catalog analogs were tested for potency by in vitro assay and binding affinity by quantitative saturation transfer difference (STD)-NMR titration. Two compounds identified by this approach possess low micromolar affinity and potency, as well as excellent binding efficiency against hTS. Relative binding orientations for both leads were modeled using AutoDock, and the most likely bound conformations were validated using experimentally derived STD-NMR binding epitope data. These ligands represent novel starting points for fragment-based drug design of non-canonical TS inhibitors, and their binding epitopes highlight important and previously unexploited interactions with conserved residues in the cofactor-binding site.

Figure 1

Structures of AG331 (1), AG337 (5) and related structures, with proton assignments characterized by ¹H NMR.

Figure 2

STD amplification factor (SAF) values for resolvable protons in STD-NMR spectra for compounds 3 and 4 (Figure 1) in aqueous buffer with hTS. (above): Addition of 3 does not significantly alter SAF values for 4, and addition of 4 does not significantly impact 3. (below): Addition of the native substrate dUMP does not impact binding of 3 or 4, while addition of methotrexate effectively competes with both fragments in binding hTS. Data acquired for 1.0 mM (above) and 0.5 mM (below) ligand solutions at 280K on a 750 MHz spectrometer. See Experimental Methods for calculation of SAF.

Figure 3

(above): Overlaid 2D NOESY spectra and 1D ¹H reference spectra for compounds 3 and 4 prepared individually in the presence of hTS (see Figure 1 for proton labeling scheme). This spectral region reveals positive intra-molecular cross-peaks for individual samples of 3 (black) and 4 (blue), indicative of independent binding to hTS, and an absence of peaks where indicated. (below): When 3 and 4 are mixed together in solution with hTS, intermolecular (ILOE) cross-peaks are generated (arrows) in the same experiment, indicative of proximal pairwise binding. Acquired on 750 MHz at 280K with 750 ms mixing time (see Experimental Methods).

Figure 4

Reference (red), saturation (black) and difference (maroon) 1D ¹H STD-NMR spectra shown above the 2D ¹H-¹H NOESY spectrum for screening mixture 8 and hTS. Positive cross-peaks (red) correlate to negative NOEs and strong STD intensities for binding fragments; non-binders generate null or positive NOEs, resulting in negative cross-peaks (cyan). Resonances from three different molecules (MV1331, MV1370 and MV1491) are highlighted, showing both intra-molecular transfer NOEs and inter-ligand NOEs (ILOEs). Sample contains 1.0 mM ligands with 25 μM monomeric hTS in aqueous buffer. Data collected at 280K on a 750 MHz spectrometer using 750 ms mixing time (see Experimental Methods).

Figure 5

Fragment binding “interactome” connecting the top ten folate-site binding ILOE fragments. Ten fragment hits exhibited strong ILOE cross-peak intensities (arrows) when tested in randomized pairs for hTS binding, and were used for catalog analog searches. Other preliminary hits rejected due to weak or null ILOE signals (wavy lines) are denoted by italicized labels. Data collected at 280K on a 750 MHz spectrometer using 250 ms mixing time (see Experimental Methods).

Figure 6

Lowest energy AutoDock model of compound 6 docked into hTS (green) with the substrate dUMP (cyan) bound. Key protein side chains are labeled, and the last three C-terminal residues have been removed for clarity. Ligand protons resolved by ¹H NMR are color-coded according to normalized STD_FIT values (95 - 100% = red; 85 - 95% = orange; 75 - 85% = yellow). Figure generated using PyMOL (81) from PDB ID 1hvy (57).

Figure 7

AutoDock conformations of MV1555 from two well-resolved clusters with nearly identical binding energy scores, with the last five C-terminal residues removed for clarity. Relative STD_FIT values for MV1555 protons acquired in solution with hTS (green) and the substrate dUMP (yellow) are color-coded (see legend). One conformation correlates well with STD-NMR-derived binding proximities (above), compared with the next best conformation which does not (below). Figure generated using PyMOL (81) from PDB ID 1ju6 (58).

Figure 8

STD-NMR binding epitope-validated AutoDock conformations of AG337 (top left), AG331 (top right), MV1565 (bottom left) and MV1555 (bottom right) and hTS (green) with the substrate dUMP bound (yellow), highlighting polar contacts between ligands and the protein (black dashed lines). Conformations for AG331 and AG337 are comparable to known crystallographic data on those complexes. Figures generated using PyMOL(81) from PDB ID 1ju6 (58).

Figure 9

Flowchart depicting the progression from primary fragment screening to final lead compounds carried out in this study.