Peptide Modification & Drug Design

The evolution from natural peptide hormones to optimized therapeutic peptides, the journey from GLP-1 (half-life 2 minutes, no oral bioavailability) to semaglutide (half-life 7 days, oral formulation available), encapsulates the art and science of peptide drug design. This progression did not happen by serendipity: it required systematic application of medicinal chemistry principles to identify and address the specific pharmacological limitations of the natural peptide lead compound, while preserving and enhancing its receptor engagement and selectivity.

The drug design process for a therapeutic peptide typically begins with identifying a naturally occurring peptide (or peptide-related protein domain) with the desired biological activity. The natural peptide serves as the "lead compound", a biologically validated starting point with the right receptor specificity and mechanism of action, but inadequate pharmaceutical properties. Systematic SAR studies (alanine scans, truncation, extension, single amino acid substitutions) identify the minimal pharmacophore, the smallest structural unit retaining full activity, and the "hot spot" residues that make critical receptor contacts. This information guides design, residues predicted to destabilize the interface significantly when mutated are the hot spots worth preserving or optimizing, while energetically neutral positions can be varied or removed (PMID 14872095).

The "rule of five" from Lipinski and colleagues famously describes properties associated with good oral bioavailability for small-molecule drugs: MW <500 Da, logP <5, ≤5 H-bond donors, ≤10 H-bond acceptors. Most therapeutic peptides violate multiple rule-of-five criteria, even very short peptides (>4-5 residues) exceed 500 Da, and hydrogen-bond donor/acceptor counts from backbone amides and carboxyls far exceed the limits. This violation explains why oral bioavailability is so challenging for peptides. Peptides larger than the Lipinski rule-of-five limits can still be made orally available through design strategies such as cyclization and backbone N-methylation, although no simple rule reliably predicts which modifications will work (PMID 28886995). Cyclosporin A, a cyclic peptide far larger than the rule-of-five limits, is absorbed well enough to be given orally, showing that molecular design can achieve oral availability beyond those limits (PMID 28886995).

The iterative nature of drug design requires accepting failure as information: each design hypothesis is tested experimentally, failures reveal the structure's sensitivities and the receptor's geometric requirements, and these insights guide the next design iteration. Machine learning and artificial intelligence are increasingly applied to peptide drug design to accelerate these cycles: AI models trained on existing SAR data can predict the activity of proposed modifications before synthesis, dramatically reducing the number of compounds that need to be physically made and tested. AlphaFold2-predicted receptor structures enable structure-based peptide design that was previously dependent on the rare availability of experimentally determined receptor-peptide co-crystal structures (PMID 34265844).