Last week we highlighted the construction
of a covalent fragment library at AstraZeneca. The first fruits of this library
have recently been published as a pair of papers.
The protein BFL1 (or Bfl-1) is a
member of the BCL2 family and blocks apoptosis by binding to pro-apoptotic
proteins such as BIM, BID, and Noxa. Blocking these types of protein-protein
interactions should increase apoptosis in cancer cells. Indeed, BCL2 itself is
the target of the approved fragment-derived drug
venetoclax, which took heroic
measures to discover.
Finding noncovalent inhibitors of
BFL1 was also expected to be
difficult, but fortunately the protein contains a
unique cysteine (C55) in the protein-protein binding site, facilitating both
covalent attachment and selectivity. As we mentioned last week, the protein was
screened against the emerging AstraZeneca covalent library, resulting in the
discovery of several hits, including compound 8. Its optimization is described
by Simon Lucas and colleagues in the first
J. Med. Chem. paper.
Compound 8 showed promising
kinact/KI
for BFL1 as well as micromolar inhibition in a TR-FRET assay using a
BIM-derived peptide. Crystallography was initially unsuccessful, but synthesis
of close analogs led to compound 13, which is slightly more potent and could be
co-crystallized with the protein. The structure confirmed covalent binding and
revealed that one of the phenyl rings binds in a lipophilic pocket created by
movement of a phenylalanine side chain.
To explore more regions of the
protein-protein binding site, the researchers performed a high-concentration
crystallographic screen with 384 non-covalent fragments. This yielded nine
hits, four of which made hydrogen bonds with a glutamic acid side chain (E78)
that had previously been targeted by others. To try to engage with this
residue, the researchers modeled and synthesized a series of amine-containing
molecules. Happily, one of the highest priority compounds gave a ten-fold boost
in potency. Adding a methyl to the benzylic position and tweaking substituents
around one of the phenyl rings ultimately led to compound (
R,
R,
S)-26,
the best molecule in this paper.
Because C55 is unique to BFL1,
the hope was that compounds would be selective against other BCL2 family
members, and indeed (R,R,S)-26 showed no activity against
BCL-xl, BCL2, or MCL1. In vitro ADME parameters were encouraging, and the
molecule also showed moderate bioavailability in mice. (R,R,S)-26
showed some cellular activity, though a mass-spectrometry assay showed only
~50% target engagement in cells after treatment at 10 µM for five hours.
The second
J. Med. Chem. paper, by Adeline
Palisse and colleagues, describes further optimization. Structure-based design
was supported by “multiple X-ray cocrystal structures,” and as in the first
paper the researchers consistently measured the half-life of new molecules
against the cellularly abundant thiol glutathione to ensure they were not
simply optimizing non-specific reactivity. The paper is an excellent
blow-by-blow account of some of the challenges of medicinal chemistry:
improving activity at the expense of stability or permeability, for example.
The most potent compound has
kinact/KI = 120,000 M
-1s
-1,
but the hepatocyte stability data suggested it would be rapidly cleared.
In the end, compound 20 was
chosen as the best overall molecule, with a
kinact/KI
comparable to that of the approved drug
sotorasib. As with (
R,
R,
S)-26,
it showed no activity against BCL-xl, BCL2, or MCL1, and it was also clean
against a panel of 48 kinases and fairly clean against a panel of other
potential off-target proteins.
Among the several BCL2 family
members, the protein MCL1 can also bind to BIM, thereby blunting the effects of
inhibiting BFL1. Thus, the researchers performed cell assays in the presence of
the MCL1 inhibitor AZD5991, whose discovery we wrote about
here. In the
presence of 0.5 µM AZD5991, compound 20 had an EC
50 = 350 nM in a
cell viability assay and also activated caspase 3, as expected in apoptosis. A
similar effect is also seen in combination with venetoclax.
Pharmacokinetic studies in mice
revealed that compound 20 is 55% orally bioavailable, and this combined with
the other properties suggest this molecule will be a useful
chemical probe for
exploring the biology of BIM.
