Logo: to the web site of Uppsala University

SMYD3 (SET- and MYND-domain containing protein 3) is an epigenetic enzyme with lysine methyl transferase activity and multiple protein binding partners. It is implicated in cancer development and active site inhibitors with antitumor activity have been developed. We have previously discovered that diperodon is an allosteric SMYD3 ligand and are interested in developing ligands that can interfere with non-catalytic functions of SMYD3, while avoiding conceivable draw-backs of targeting a conserved site in an enzyme with several close family members. Herein, the features of the diperodon site were explored via computational modelling and served as a basis for identifying analogues in commercial compound space, thus avoiding the need for in-house compound synthesis. Time-resolved grating coupled interferometry (GCI) biosensor analysis confirmed that two out of 21 acquired analogues interacted with SMYD3, with similar affinities as diperodon (KD ∼ 180 and 210 vs. ∼200 µM). As a second approach, fragmentation of diperodon followed by growing of fragments identified an additional 11 compounds in commercial compound space. GCI analysis confirmed that N-phenylformamide and three compounds evolved from this fragment interacted with SMYD3. These four ligands varied structurally from diperodon and had higher affinities (KD = 0.4–130 µM) and superior ligand efficiencies. However, all ligands interacted with rapid kinetics and weak affinities, indicating that the site had poor ligandability, possibly a result of its extremely flexible structure. Difficulties in protein production and the overall flexible structure of SMYD3, prevented NMR experiments and X-ray crystallography. Nevertheless, the combination of computational ligand design supported by biosensor-based analyses resulted in new allosteric ligands with minimal resources in a short time.

Cancer is a common disease and diagnosis frequency correlates with population age. Though many cancers can be cured today, numerous types remain difficult to treat. Treatments can evoke side effects and often don’t reach the clinic due to inefficacy. Thus, better targeted anti-cancer therapies and candidate drugs are required. This thesis focusses on initial stages of drug discovery where we sought to identify ligands specifically targeting SET and MYND domain containing protein 3 (SMYD3), and Cullin3 associated adaptor proteins: Kelch-like protein 12 (KLHL12) and 20 (KLHL20). These targets are all associated with cancer although their biological mechanisms remain elusive. The targets were challenging from a biochemical perspective, nevertheless via robust expression and purification methods, an amalgamation of biochemical techniques and computational methods were used to identify, characterize, and evolve fragment and peptide-based ligands. Sensitive multiplexed screening assays enabled selection of specific hits. A grating coupled interferometry-based biosensor assay implemented a kinetic criterion for fragment hit identification against SMYD3. Four fragments from a library containing 1056 fragments had their binding site and orientation established using X-ray crystallography. Fragment evolution based on the SMYD3 allosteric ligand diperodon encompassed a structure-affinity-relationship (SAR)-based approach, and a deconstruction-and-growth method wherein ligands with K_D of 0.4-180 μM were attained. Structure prediction complemented a surface plasmon resonance (SPR) biosensor-driven approach to develop a stapled peptide ligand against the Kelch domain of KLHL20, derived from zinc finger translocation associated protein (ZFTA). This peptide had K_D of 1.14 mM and alanine scanning revealed aspartate as vital for interaction. Multiplexed fragment-based SPR biosensor screening assays against the Kelch domains of KLHL12 and 20 identified 237 and 266 hits from a library containing 3000+ fragments. Hit selection was based on preference for folded protein and dose-response analysis was conducted for validation and hit reduction to 24 and 21. Hit SAR was probed using modelling and fragment analogues. NMR confirmed fragment-protein binding. All targets studied herein were concluded as poorly druggable, however using multiple experimental approaches alongside computational methodologies permitted hit identification, validation, and a further understanding of poorly druggable targets. The validated hits presented are befitting for evolution by medicinal chemistry.

A 1056-membered fragment library has been screened against SMYD3 using a novel multiplexed experimental design implemented in a grating coupled interferometry (GCI)-based biosensor. SMYD3 is a prospective target for anticancer drugs and the focus has initially been on discovery of inhibitors of its lysine methyl transferase activity. However, it has multiple protein interaction partners and several potential roles in carcinogenesis. It therefore remains unclear what mode of action ligands targeting the protein should have. Our goal was therefore to identify new ligands and discriminate hits that interact with the active site and those that interact with other sites. In addition, we were interested in selecting hits based on kinetic features rather than affinity. Screening was done in parallel against SMYD3 alone or SMYD3 with the active site blocked by a tight binding inhibitor. Hit selection was primarily based on dissociation rates. In total, 20 fragments were selected as hits, of which half apparently targeted the active site and half targeted other sites. Twelve of the hits were selected for structural analysis using X-ray crystallography in order to identify binding sites and modes of binding. Four of the hits were successfully identified in crystal structures with SMYD3; the others did not show any electron densities for ligands in the crystals. Although it might be possible to optimize the crystallography approach for a better success rate, it was clear that the sensitivity and time resolution of the biosensor assay was exceptional and enabled kinetic rate constants to be estimated for fragments. Fragments are typically considered to interact too rapidly for such quantification to be possible. This approach consequently represents a paradigm shift. In addition, the multiplexed approach allows ligands targeting different sites to be rationally selected already in the fragment library screening stage.