From Active-Site Mapping to Lead Discovery using Fragment-based Approaches on the Aspartic protease Endothiapepsin

The work focuses on the evaluation and comparison of different fragment-based approaches, for which purpose the model system Endothiapepsin (EP) has been used. The enzyme belongs to the family of aspartic proteases. We used the 361-entry in-house fragment library compiled with physico-chemical properties similar to the rule-of-three. We applied six different techniques to screen the fragment library: saturation-transfer difference NMR (STD-NMR), thermal shift assay (TSA), reporter-displacement assay (RDA), microscale thermophoresis (MST), fluorescence-based biochemical assay (HCS), and native mass spectrometry (MS). Alarmingly, the overlap between the hits produced by all six methods was very low. While only 41 out of all 361 fragment library entries could be identified by at least two methods, only 3 were identified by five methods, and no single fragment was identified as hit by all six methods taken together (chapter 2). We thus performed X-ray crystallography with individual fragments soaked to beforehand prepared apo EP crystals. For this purpose very high fragment concentrations of 90 mM were used. Intriguingly, we were able to identify 71 fragment hits as bound to EP, corresponding to 20% hit rate. Worryingly, only 30% of the 71 hits were predicted by only one of the beforehand applied screening techniques, clearly emphasizing that any screening strategy comprising at least two prescreening methods would have been able to identify only 19 (27%) of the 71 crystallographic hits. We divided the hits into two main groups: catalytic dyad binders and remote binders. Chapter 3 deals with the catalytic dyad binders. Therein we describe a variety of warheads which address the catalytic aspartates either directly or via the catalytic water molecule W501, while occupying the S1 and S1’ binding pockets. Furthermore, we found fragments, which occupied both pockets simultaneously, offering an optimal platform for further optimization strategies into both directions. The very useful information regarding preferred spots of binding remote from the catalytic dyad is described in detail in chapter 4. The so called hot spots represent key pocket residues, which have to be addressed in an optimization procedure in order to achieve optimal ligand binding. The high-throughput potential of X-ray crystallography (chapters 3 and 4) compared to the other six prescreening techniques presented in chapter 2 is still very low. Because of this, a lot of effort has been invested in academia and industry to extend the method’s applicability as a primary screening technique. Whenever used as such, practitioners exposed protein crystals to a mixture of compounds to accelerate the hit identification. However, we cannot recommend the use of cocktail experiments with clear conscience as we believe that parameters such as compound solubility, chemical reactivity, and crystal damage, often resulting in reduced ligand occupancy, diffuse electron density, or deviating fragment binding poses are provoked by the presence of multiple compounds in a mixture. These issues are exemplified in chapter 5 where we directly compared the difference electron density in the binding pocket of EP between singly soaked fragments and cocktails prepared those. For example, many of the fragments used in a mixture of two failed to be identified as hits, whereas clear density around those fragments could be observed in the single soaking experiments. Moreover, the clear assignment of a fragment to diffuse electron density usually requires additional experiments, which confirm the binding site. When considering this, even more time has to be spent, which extremely slows down the hit identification procedure. The only plausible reason to use several compounds in a mixture is when the reaction between those is aimed. This is an approach named dynamic combinatorial chemistry (DCC) and described in chapter 6. The compounds, hydrazides and aldehydes, chosen for the mixtures react with one another with dehydration to larger acylhydrazones. The formation of the strongest binders among others, acylhydrazones (S)-H4-A4 and (R)-H3-A5, was induced by the natural selection by the target protein EP. Moreover, in a follow-up optimization project, which started based on the binding modes of the two acylhydrazones, the combination of a bis-aldehyde with hydrazides resulted in the natural selection of a bis-acylhydrazone by EP, with an 240-fold improved potency compared to the initial acylhydrazones