Modern biotech drug discovery often starts with the observation that the activity of a specific gene or protein is critical to the progression of a particular disease. Once confirmed in knock-down studies, hits can be identified by testing a small library of perhaps 500 compounds in an assay for activity against that target.
While these hits may have some activity, they are unlikely to have the mix of activity, specificity, and ADME/toxicity properties required to make a drug, and so the hits progress to the medicinal chemistry teams for hit to lead, lead optimization, and candidate selection. In smaller companies, these medicinal chemistry teams are often outsourced.
After receiving the preliminary hits, a medicinal chemist will take the 2-D structures of the hits in those screens and their activity data and create detailed structure activity relationships that attempt to explain the activity of the hits in terms of changes to their 2-D structure and other properties.
They will then attempt to predict, given knowledge gained from years of experience, which small evolutionary changes to that 2-D structure will enhance the affinity and specificity of the lead compound while overcoming any ADME and toxicity issues that might be observed. They will undertake iterative cycles of design, synthesis, and testing in an attempt to improve the molecular properties of the lead until it is optimized. This process can be time-consuming and expensive. During this process, the medicinal chemist’s reliance on insights gleaned from the 2-D structure may appear counterintuitive and confusing to many biologists—and they may be right to have reservations.
It has been known since the early days of drug discovery that molecules with different structural types (chemotypes) can have the same biological activity and properties. Such molecules are called bioisosteres. Moreover, x-ray crystallography has shown that these structurally diverse bioisosteres will bind to their common protein target in the same way. This similarity can be entirely independent of 2-D molecular structure, and the structural differences between bioisosteres can be so profound that it is impossible to predict by looking at them that their activities are related (Figure 1).
Because a protein target does not “see” the atoms and bonds of a drug, but instead interacts with the electron cloud around the molecule and the physicochemical properties at the surface of the compound, the 2-D structure is often a poor indicator of the biological activity and properties of a molecule.
Many bioisosteres have diverse chemotypes, well beyond the small degrees of 2-D structural difference that a medicinal chemist could possibly feel comfortable in suggesting as a change to a lead. This can create costly dead-ends or chemotype traps where the inability to predict a bioisosteric alternative to a given chemotype prevents the right mix of biological activity and properties from being achievable in a lead series.