Redesigning an FKBP–ligand interface to generate chemical dimerizers with novel specificity

The C9 carbonyl oxygen of the α-keto-pipecolylamide core shared by FK506 and its synthetic FKBP ligand analogs packs tightly against Tyr26, Phe36 and Phe99 of human FKBP12, all three making unusual edge-on contacts. Replacing that carbonyl with a larger substituent clashes sterically with either Phe36 or Tyr26 depending on the stereochemistry at the now tetrahedral C9 carbon, so a mutation that truncates one of those side chains supplies a compensating hole that can restore binding. This is the design basis of the bumped-ligand/holed-FKBP pair.

Compound 1, a synthetic FKBP ligand with an unmodified C9 carbonyl, binds wild-type human FKBP12 with an IC50 of 86 nM in a competitive fluorescence polarization assay against a fluoresceinated FK506 probe, corresponding to a Kd of about 20 nM. Its affinity is largely unchanged on FKBP12 mutants in which Phe36 is replaced by a smaller hydrophobic residue, so truncating Phe36 does not by itself cost binding of an unbumped ligand.

Enlarging the C9 substituent of the synthetic FKBP ligand compound 1 to a methoxy group (compound 2) or an ethyl group (compound 3) reduces binding to wild-type human FKBP12 to undetectable levels, with IC50 above 10,000 nM. Binding is rescued by any of the Phe36 substitutions tested in FKBP12, giving apparent selectivity of up to 50-fold for the mutant over wild-type protein, and the ethyl bump of compound 3 binds the mutants more tightly than the methoxy bump of compound 2, indicating that nonpolar bumps are preferred. The Phe36 mutants clearly favor the S isomer of the bumped compounds over the R isomer, the preference predicted by modeling of the engineered pocket.

Retaining the C9 ethyl bump and replacing the tert-pentyl C9 bottom substituent with a phenyl group (compound 4) improves affinity for each FKBP12 Phe36 mutant by 3- to 5-fold, and a trimethoxyphenyl group (compound 5) adds a further 5- to 20-fold. Neither substitution restores detectable binding to wild-type FKBP12, and the preference for the S isomer over the R isomer grows as affinity improves, indicating that these peripheral changes do not alter binding of the engineered ethyl bump.

The ethyl-bumped, C9-trimethoxyphenyl ligand 5S pairs best with F36V-FKBP, the FKBP12 mutant carrying Phe36 replaced by Val, with an IC50 of 1.8 nM in competitive fluorescence polarization. A fluorescein conjugate of the same ligand, fluoro-5S, binds F36V-FKBP with a Kd of 0.094 nM by direct polarization titration, versus 67 nM for wild-type FKBP12, so selectivity for F36V-FKBP is almost three orders of magnitude while weak wild-type binding remains detectable. This affinity of fluoro-5S for F36V-FKBP exceeded that of any ligand known for wild-type FKBP12.

Affinities of C9-bumped FKBP ligands are largely insensitive to the size of the residue substituted for Phe36 in FKBP12: none of the bumped compounds shows a major preference for a particular Phe36 mutant. Molecular recognition in the engineered pocket is therefore loose and flexible rather than a precise lock-and-key fit.

In the 1.9-Å crystal structure of the complex between ligand 5S and F36V-FKBP (PDB 1BL4), the protein is, apart from the mutation itself, essentially superimposable on wild-type FKBP structures from unmodified complexes, with a main-chain rms deviation of 0.45 Å. The F36V substitution opens a cavity of about 90 ų that the protein does not relax to fill, apart from a 0.2-Å movement of the Val36 main chain, leaving the cavity lined by the predominantly hydrophobic side chains that normally pack against Phe36 and the Gly28 main chain exposed at its base.

The S-ethyl bump of ligand 5S inserts into the cavity created by the F36V substitution in FKBP12, penetrating what would be the surface of Phe36 in the wild-type protein and making multiple van der Waals contacts, while the atoms of the α-keto-pipecolylamide core shift by only 0.2–0.4 Å relative to a wild-type FKBP complex with a conventional ligand. The bump occupies only about 40% of the engineered hole, so a large fully enclosed cavity of about 60 ų remains in which no ordered water molecules are visible, and extending the ethyl bump of 5S to an allyl group increases affinity only about 3-fold. Affinity in this engineered pocket is thus largely insensitive to the quality of packing.

The substantial binding energy that the C9-trimethoxyphenyl group contributes to ligand 5S on F36V-FKBP probably stems from two good hydrogen bonds to Arg41 and His87 together with extensive intramolecular contacts that may preorganize the ligand and lower the entropic cost of binding. The remaining peripheral substituents of 5S pack into hydrophobic channels on FKBP essentially as they do for unbumped compounds.

AP1903 is a homodimer of the C9-ethyl-bumped FKBP ligand 5S, joined through a previously described linker, and it reproduces the binding affinity and specificity of its monomeric precursor. In a competition fluorescence polarization assay AP1903 displaces a fluoresceinated FK506 probe from F36V-FKBP with an IC50 of 5 nM, while its binding to wild-type FKBP is negligible.

HT1080 human fibrosarcoma cells stably transduced to express a fusion of a myristoylation sequence, two copies of F36V-FKBP and the human Fas intracellular domain undergo potent dose-dependent apoptotic death on overnight treatment with the bumped dimerizer AP1903, with an EC50 of about 0.1 nM. This is a 60-fold improvement in potency over AP1510, an unbumped synthetic dimer of compound 1. On an analogous HT1080 line expressing the same Fas construct built from wild-type FKBP, AP1903 activity is reduced by over two orders of magnitude, giving 200-fold discrimination in cells against the 1,000-fold discrimination measured for binding in vitro, and unmodified HT1080 cells show no response to AP1903 at concentrations up to 1 µM.

HT1080 cells expressing a fusion of the extracellular and transmembrane portions of the human low-affinity nerve growth factor receptor (amino acids 1-274), two copies of F36V-FKBP and the human Fas intracellular domain, together with constitutively secreted human growth hormone (clone HTFasGH-3), are eliminated in vivo by the bumped dimerizer AP1903. After intramuscular implantation into nude mice, a single intravenous dose of AP1903 produces a dose-dependent fall in serum hGH with a half-maximal effective dose of 0.4 ± 0.1 mg/kg, read 24 h after dosing. Serum hGH reports the number of viable implanted cells because hGH has a serum half-life of only about 3 min in mice.