Fragment-Based Drug Discovery

Fragment-based drug discovery (FBDD)—also termed fragment-based lead discovery (FBLD)—is a paradigm in drug discovery and structure-based drug design that identifies small, polar chemical building blocks (fragments, molecular weight 120–300 Da) that bind with weak equilibrium dissociation constants (K_d ≈ 50 μM to 10 mM) to specific sub-pockets of a macromolecular target, and subsequently elaborates those fragments via X-ray crystallography or NMR into high-affinity nanomolar clinical candidates. Pioneered by Stephen Fesik and colleagues at Abbott Laboratories through "SAR by NMR" (Shuker et al., 1996) and industrialized at Astex Therapeutics and Plexxikon, FBDD has yielded more than 60 clinical candidates and multiple FDA-approved medicines, including vemurafenib (BRAF^V600E), venetoclax (BCL-2), erdafitinib (FGFR1–4), pexidartinib (CSF1R), sotorasib (KRAS^G12C), asciminib (BCR-ABL1 myristoyl pocket), and capivasertib (AKT1–3).

Whereas conventional high-throughput screening (HTS) tests 10⁶ complex drug-like molecules (MW 350–500 Da) that cover an infinitesimal fraction of the ~10⁶⁰ theoretical molecules in drug-like chemical space, a compact library of 1,000 to 3,000 fragments samples the much smaller chemical space of ≤11–18 heavy atoms (~10⁷ to 10⁹ structures) far more thoroughly.

Mike Hann's Molecular Complexity Model and Ligand Efficiency

In 2001, Mike Hann, Brian Hudson, and colleagues formulated a probabilistic molecular complexity model explaining why small fragments achieve higher hit rates and superior binding geometry than larger HTS molecules. As ligand complexity (number of pharmacophoric features and heavy atoms) increases, the probability that all functional groups simultaneously match complementary sub-pockets of the protein without introducing at least one steric clash or electrostatic repulsion drops rapidly according to a binomial decay distribution. A small fragment with only 2 or 3 interaction vectors can adopt an optimal orientation inside a protein sub-pocket unimpeded by extraneous substituents.

Although a fragment forms only a few non-covalent interactions—resulting in weak apparent affinity (K_d in the high micromolar to low millimolar range)—every heavy atom participates productively in binding, yielding a high Ligand Efficiency (LE):

LE = −Δ*G*° / N_heavy = (−*R* · T · ln K_d) / N_heavy ≥ 0.30 kcal·mol⁻¹·heavy atom⁻¹

Furthermore, as analyzed by Michael Page and William Jencks (1971) and Christopher Murray and Marcel Verdonk (2002), any bimolecular association in solution must pay a rigid-body translational and rotational entropic penalty (Δ*G*_rigid ≈ +3.5 to +5.0 kcal/mol at 298 K) that is largely independent of molecular weight. Consequently, a 150 Da fragment (N_heavy = 11) with a measured K_d of 1 mM (observed Δ*G*° = −4.1 kcal/mol, LE = 0.37 kcal·mol⁻¹·atom⁻¹) actually forms −7.6 to −9.1 kcal/mol of intrinsic enthalpic and hydrophobic interactions with the pocket once the rigid-body barrier is accounted for.

Astex's "Rule of Three" (Ro3) for Fragment Library Design

To ensure that fragment hits leave ample molecular weight and lipophilicity headroom during subsequent synthetic elaboration into oral drugs obeying Lipinski's rule of five, Miles Congreve and colleagues at Astex (2003) established the "Rule of Three" (Ro3) for fragment libraries:

• Molecular weight (MW) ≤ 300 Da (typically 140–250 Da, or 10–18 non-hydrogen heavy atoms)
• Calculated octanol–water partition coefficient (cLog*P*) ≤ 3 (typically 0.5–2.0 to ensure aqueous solubility ≥ 1 mM for high-concentration biophysical screening)
• Hydrogen-bond donors (HBD) ≤ 3 and Hydrogen-bond acceptors (HBA) ≤ 3
• Rotatable bonds (N_rot) ≤ 3 and Topological Polar Surface Area (TPSA) ≤ 60 Ų

NMR Spectroscopy, SPR, and High-Throughput X-Ray Fragment Soaking

Because fragments bind with K_d values between 100 μM and 5 mM, conventional biochemical enzyme inhibition assays would require screening at millimolar concentrations where compound absorbance and DMSO cosolvent interfere. FBDD therefore relies on sensitive biophysical detection:

1. Ligand-Observed and Protein-Observed NMR: In Saturation Transfer Difference (STD) NMR, selective radiofrequency irradiation saturates protein aliphatic protons (~0 ppm); spin diffusion transfers magnetization to any transiently bound fragment (k_off ≈ 100–10,000 s⁻¹), attenuating its ¹H resonance after dissociation into solution. ¹⁹F NMR screens fluorinated fragment cocktails with zero background signal from protein or buffer. In 2D ¹H–¹⁵N HSQC (Heteronuclear Single Quantum Coherence) NMR, chemical shift perturbations (Δδ_NH) of assigned backbone amide cross-peaks directly map which amino acid residues line the fragment-binding site.
2. Surface Plasmon Resonance (SPR) and Grating-Coupled Interferometry (GCI): Immobilizing the target protein at high density on a sensor chip enables real-time optical detection of 150 Da fragments binding at 100–500 μM, while simultaneously rejecting promiscuous aggregators that exhibit super-stoichiometric response above theoretical R_max or slow non-specific dissociation (Surface Plasmon Resonance and Isothermal Titration Calorimetry).
3. High-Throughput Crystallographic Soaking and PanDDA: Synchrotron beamlines (such as Diamond Light Source XChem) soak hundreds of pre-formed apo protein crystals in 20–100 mM fragment solutions using acoustic dispensing. The PanDDA (Pan-Dataset Density Analysis) algorithm subtracts the statistical ground-state electron density averaged across dozens of apo crystals to reveal low-occupancy (15%–40%) fragment binding events at atomic resolution.

Fragment Growing, Fragment Linking, and Fragment Merging

Once a fragment's 3D binding pose is determined, medicinal chemists convert the millimolar fragment into a nanomolar lead while maintaining LE ≥ 0.30 kcal·mol⁻¹·atom⁻¹ via three structural strategies:

• Fragment Growing: Starting from a single anchored fragment, chemists iteratively extend chemical vectors into adjacent hydrophobic channels, oxyanion holes, or salt-bridge pockets. At Plexxikon, screening a 20,000-scaffold library identified 7-azaindole (MW 118 Da, IC₅₀ > 200 μM) binding to the kinase hinge region; structure-guided growing into the BRAF^V600E specificity pocket yielded vemurafenib (IC₅₀ = 31 nM) for metastatic melanoma. Similarly, growing a quinoxaline fragment yielded the pan-FGFR inhibitor erdafitinib.
• Fragment Linking: When two fragments (A and B) bind simultaneously to adjacent, non-overlapping sub-pockets (as detected by "SAR by NMR" or crystallographic cocktails), covalently tethering them with an unstrained linker produces super-additive affinity. Because fragment A and fragment B independently paid the rigid-body entropic penalty (Δ*G*_rigid ≈ +4 kcal/mol each), the linked molecule A–B pays that translational/rotational barrier only once:

Δ*G*°_AB = Δ*G*°_A + Δ*G*°_B + Δ*G*°_link Thus, linking a 1 mM fragment (K_d,A = 10⁻³ M) and a 100 μM fragment (K_d,B = 10⁻⁴ M) with a strain-free linker can yield a sub-100 nM inhibitor (K_d,AB ≤ 10⁻⁷ M). At Abbott, linking a fluoro-biaryl acid fragment (K_d = 300 μM) and a phenolic naphthol/nitrobenzene fragment (K_d = 2,000–4,300 μM) binding two adjacent pockets of the anti-apoptotic protein BCL-X_L/BCL-2 led to navitoclax and the selective BCL-2 protein-protein interaction inhibitor venetoclax (K_i < 0.01 nM).

• Fragment Merging: When two fragments bind to overlapping regions of the same pocket (or when a fragment overlaps with a known substrate or HTS hit), their core pharmacophores are hybridized into a single consolidated scaffold.

Why is fragment linking rarely able to achieve the full theoretical multiplicative affinity (K_d,AB < K_d,A · K_d,B) in practice?

Although eliminating one rigid-body entropic barrier (Δ*G*_rigid ≈ +3.5 to +5.0 kcal/mol) theoretically boosts affinity beyond simple additivity (Δ*G*°_A + Δ*G*°_B), the covalent linker connecting fragment A and fragment B must adopt an exact bond length and dihedral angle geometry that allows both fragments to retain their optimal native hydrogen-bonding and van der Waals orientations in their respective sub-pockets. If the linker introduces even 1–2 kcal/mol of conformational torsional strain or forces either fragment out of its energy minimum by 0.3 Å, that unfavorable linker energy (Δ*G*°_link > 0) erodes the entropic gain.

Why do fragments maintain a stable binding pose during fragment growing instead of flipping orientation?

High-efficiency fragments (LE ≥ 0.35 kcal·mol⁻¹·atom⁻¹) typically anchor to an enthalpically dominant "hot spot" within the protein pocket—such as forming a bidentate donor–acceptor hydrogen-bond pair with the backbone amide of a kinase hinge region or coordinating a catalytic Zn²⁺ ion. Because breaking those directional polar interactions in a low-dielectric protein cavity incurs a steep enthalpic penalty (+3 to +6 kcal/mol), the core fragment remains locked in its crystallographically observed orientation as substituents are grown outward into secondary sub-pockets.

How did covalent fragment screening enable the discovery of KRAS^G12C inhibitors like sotorasib?

For over three decades, oncogenic RAS GTPases were considered "undruggable" because RAS binds guanosine triphosphate (GTP) with picomolar affinity (K_d ≈ 20 pM) against millimolar intracellular GTP concentrations and lacked any obvious allosteric pocket in apo crystal structures. In 2013, Kevan Shokat and colleagues used disulphide tethering and electrophilic acrylamide covalent fragment screening targeting the mutant cysteine of KRAS^G12C. Fragment binding induced the opening of a previously unseen cryptic allosteric pocket (Switch-II pocket) beneath the Switch-II loop, locking KRAS^G12C in its inactive GDP-bound state and enabling fragment growing into the approved drugs sotorasib and adagrasib (covalent drug design).

1. Shuker, S. B., Hajduk, P. J., Meadows, R. P., & Fesik, S. W. (1996). "Discovering High-Affinity Ligands for Proteins: SAR by NMR." Science, 274(5292), 1531–1534. https://doi.org/10.1126/science.274.5292.1531
2. Hann, M. M., Leach, A. R., & Harper, G. (2001). "Molecular Complexity and Its Impact on the Probability of Finding Leads for Drug Discovery." Journal of Chemical Information and Computer Sciences, 41(3), 856–864. https://doi.org/10.1021/ci000403i
3. Congreve, M., Carr, R., Murray, C., & Jhoti, H. (2003). "A 'Rule of Three' for Fragment-Based Lead Discovery?" Drug Discovery Today, 8(19), 876–877. https://doi.org/10.1016/S1359-6446(03)02831-902831-9)
4. Erlanson, D. A., Fesik, S. W., Hubbard, R. E., Jahnke, W., & Jhoti, H. (2016). "Twenty Years On: The Impact of Fragments on Drug Discovery." Nature Reviews Drug Discovery, 15(9), 605–619. https://doi.org/10.1038/nrd.2016.109