There and back again: fragments and BACE-1

One of the many interesting talks at FBLD 2009 was by Daniel Wyss of Schering-Plough on how fragment-based screening was used to discover potent and selective inhibitors of the Alzheimer’s disease target BACE-1. Two papers published online in J. Med. Chem. now begin to tell the full story.

BACE-1 is an aspartyl protease, a class of enzymes that has proven to be druggable, as illustrated by the number of HIV-1 protease inhibitors on the market. However, BACE-1 has an unusually shallow, flexible, and hydrophilic active site, and its location in the brain means that candidate drugs need to be particularly small with a limited number of hydrogen bond donors.

The first paper discusses how Wang and colleagues used 15N-HSQC NMR screening of a library of ~10,000 compounds, about half of them fragments, against the BACE-1 catalytic domain. This resulted in 9 distinct classes of hits, some of which were as potent as 30 micromolar as judged by NMR-based dissociation constants. The compound most extensively pursued was compound 2 (see figure), an isothiourea. Some 200 analogs of this were present in the corporate collection (an advantage of working in big pharma!), and 15 of these showed activity in an enzymatic assay, of which compound 3 was the most potent. Extensive NMR analysis and an X-ray crystal structure revealed that the isothiourea makes hydrogen-bond contacts to both catalytic aspartates and extends towards the S1 pocket and S3 subpocket (S3sp).


Isothioureas are potentially toxic and unstable, so structure-based design was applied to replace this moiety. One outcome was a series of 2-aminopyridines such as compound 4. Unfortunately, although they showed measurable binding by NMR and some could even be characterized crystallographically, most had little or no activity in a BACE-1 functional assay.

That’s where the second paper by Zhu and colleagues comes in. Careful analysis of the crystal structure of compound 3 combined with parallel synthesis led to a series of iminohydantoins (or cyclic acylguanidines) such as compound 23. Interestingly, a close analog of this compound made similar contacts but flipped to another orientation. These different binding modes complicated medicinal chemistry efforts, requiring parallel chemistry and NMR-based screening (as most early compounds were at best only weakly active). Ultimately this effort yielded potent inhibitors such as compound 39, with nanomolar biochemical activity. However, the compound also has a clogP of 7.5, a molecular weight of more than 500 Da, and only modest bioavailability, so turning this into a brain-active drug could be problematic.

Strikingly, truncating a large portion of the molecule (to generate compound 40) yielded a much smaller compound that, despite its reduced biochemical potency, had an improved ligand efficiency, as well as measurable brain penetration. Further simplification led to compound 41, which, being rule-of-three compliant, can be considered a fragment. In other words, these two papers report the optimization of a low-affinity fragment to a high-affinity ligand and on to a medium affinity fragment. Given that the second paper is subtitled “Part 1”, we can look forward to reading further chapters.

Many FBLD publications report the rapid discovery of new leads against established targets such as kinases or Hsp90. BACE-1 is just the opposite: references suggest that the program had been in place since before 2004. These papers, along with previous papers (for example here and here) on BACE-1 from AstraZeneca and Astex, illustrate that FBLD is also powerful for discovering leads against difficult targets.