Fast fragment discovery with protein crystals

Fragment-based drug discovery (FBDD) has become a standard approach for generating starting points in medicinal chemistry. Small fragments bind weakly but can be chemically elaborated into stronger ligands. The difficulty is that moving from weak binders to measurable leads usually involves multiple design-make-test-analyse (DMTA) cycles: each analogue must be synthesised, purified, tested biochemically, and crystallised. This is slow and leaves much chemical space unexplored.

Researchers developed a new technique called Binding-Site Purification of Actives (B-SPA) to bypass this bottleneck. Instead of purifying every product, they test crude reaction mixtures directly on protein crystals. Their study, published in Angewandte Chemie International Edition, shows how this works in practice. High-throughput macromolecular crystallography on Diamond’s I04-1 beamline enabled the team to detect which molecules bind, even if they were a minor product in a mixture. This structural “filtering” step dramatically accelerates hit-to-lead workflows.

The team focused on the second bromodomain of PHIP ( disPHIP(2)), a protein implicated in epigenetic regulation and linked to cancers. A fragment hit (compound F709) had been identified in earlier crystallographic screens, but like most fragments, its binding was weak and undetectable in solution assays. Researchers wanted to explore chemical space around this initial fragment to see which modifications improve binding.

They designed up to six independent synthetic routes, each involving multi-step reactions (up to five synthetic steps), exploring different vectors of substitution (i.e. different parts of the fragment to substitute, such as replacing a ring, modifying substituents, adding functional groups). The designs were guided by synthetic tractability: only routes that are feasible with reliable chemistry were chosen. Using a low-cost robotic liquid handler, the group performed 1,876 reactions, generating diverse libraries of potential binders.

Each crude reaction mixture was checked by LC–MS using an automated tool (MSCheck) that flags the presence of the expected molecular ion. Out of 1,876 attempted syntheses, 1,108 mixtures (59%) contained the intended product. Rather than purify, the team directly soaked PHIP(2) crystals with these crude mixtures and collected data at Diamond’s I04-1 beamline, which is optimised for high-throughput macromolecular crystallography.

The crystallographic workflow relied on automated pipelines: data was processed and electron-density maps analysed with PanDDA, a method designed to pick out weak or partial ligand densities from large datasets. This is crucial, because in many cases the active compound was only a minor fraction of the crude mixture.

The outcome was striking: 22 product structures were solved directly from crude mixtures. Of these, 19 bound in a consistent pose, reinforcing the fragment’s core binding mode, while 3 revealed stereochemical preferences—from racemic reactions, only one enantiomer produced interpretable density. When resynthesised and purified, one analogue displayed measurable activity (IC₅₀ ≈ 34 µM; K_d ≈ 50 µM), a clear step beyond the parent fragment.

In short, the Diamond beamline provided the sensitivity and throughput needed to “fish out” actives from impure mixtures, demonstrating that B-SPA can accelerate fragment elaboration while conserving resources.

The impact of B-SPA is twofold. For chemistry, it reduces purification workload, allowing chemists to run thousands of reactions in parallel. For biology, it provides structural insight earlier, clarifying which substitutions are tolerated by the protein pocket. Together, this may shorten DMTA cycles by months. Coupling B-SPA with computational design or machine learning-guided synthesis planning could reduce the number of “dead” reactions, making campaigns even more efficient.

To find out more about the I04-1 beamline please contact the Principal Beamline Scientist: Frank.von-delft@diamond.ac.uk