Nathaniel Corley

Amide bond formation is of central importance in the chemical industry, underpinning the synthesis of small bioactive molecules, macrocycles, peptides and their conjugates. Conventional methods of amide synthesis rely on stoichiometric coupling agents and costly protecting group manipulations. While biocatalysis can alleviate some of these challenges, enzymes typically operate across a narrow range of substrates and their adaptation to new coupling partners requires extensive engineering. Here we show that de novo amide ligases featuring a cysteine catalytic triad can be programmed computationally to operate across diverse substrate classes with high selectivity. We first designed complementary peptide ligases to selectively functionalise either the N-terminus or a single lysine residue of a polypeptide sequence containing 36 competing lysines. Crystal structures of both ligases show excellent agreement with the designs, with structural modelling suggesting that amidation outcompetes hydrolysis because the catalytically competent geometry forms only upon acceptor binding. We then extended our approach to the synthesis of small bioactive molecules by designing stereoselective ligases that mediate a key amide bonding forming step in the synthesis of the preventative HIV therapeutic lenacapavir. These results establish de novo enzyme design as a programmable platform for achieving selective amide bond formations across a wide range of molecular targets.

Alzheimer’s disease (AD) is characterized by heterogeneous Tau phosphorylation and aberrant amyloid-β (Aβ) processing, yet molecular tools that distinguish closely related molecular states of tau and Aβ with predefined specificity remain difficult to develop. Here, we designed de novo protein readers targeting three disease-associated phospho-Tau regions and N-terminal pyroglutamate-modified Aβ (N3pG-Aβ). Screening 192 designs per target identified 51 modification-selective hits among 768 candidates (6.6%). Representative readers exhibited nanomolar to subnanomolar affinities for modified peptides, discriminated authentic modifications from unmodified or substitution-based controls, and preferentially recognized modified full-length targets. We further programmed local modification specificity by interface redesign, generating pT231 readers conditional on the phosphorylation state of neighboring S235 and S237. A 2.0 Å crystal structure of an N3pG-Aβ reader closely matched the computational design model (interface Cα RMSD = 0.52 Å), and redesign of pT231 readers tuned apparent binding affinities from greater than 1 μM to below 1 nM. In a human 3D neuronal culture with both pathological Aβ and tau, selective readers targeting N3pG-Aβ, Tau pT231, and Tau pT217 recognized disease-associated signals, exhibiting colocalization with related antibodies and much higher signal in the AD line than the control. Tau pT231 readers further quantitatively distinguish AD patients and control human brain lysates in dot-blot assays. Together, these results establish genetically encoded readers with programmable specificity for defined Tau and Aβ modification states, enabling molecular-state-resolved detection in human disease models and biological samples, and lay a foundation for controlled pathological molecular perturbation.

The catalytic and other sophisticated functions of biomolecules arise from the placement of many atoms in three-dimensional space. Here, we describe a generative artificial intelligence method for biopolymers, RFdiffusion3 (RFD3), which employs free atom diffusion to satisfy complex sets of atomic constraints more rapidly and accurately than previous methods across several challenges from enzyme design to RNA structure design. We demonstrate experimentally the power of the RFD3 pipeline for enzyme design by designing de novo phosphotriesterases with a complex active site architecture incorporating a carbamylated lysine coordinating two metal ions and zinc proteases specific for the transthyretin protein involved in amyloid diseases; in both cases we obtain active catalysts in zero shot 96 construct design campaigns.

Accurately predicting the phenotypic consequences of genetic variation is a major challenge for precision medicine. The problem is exacerbated by epistatic interactions, nonadditive effects between genetic variants that produce unexpected phenotypes. Here, we explore an understudied form of positive epistasis: intragenic complementation, in which pairs of loss-of-function variants restore near wild-type protein function. Using mutational scanning in yeast, we identify thousands of such interactions in a clinically important enzyme, human argininosuccinate lyase (ASL). Restoration of protein function is not due to the biochemical properties of the substituted amino acids, but rather to a structural feature of the protein, the active site assembly. We develop a machine learning algorithm that uses protein language model embeddings to predict intragenic complementation in ASL with 99.6% accuracy. Additionally, the model trained on ASL generalizes to a structurally related but sequence-divergent enzyme, fumarase, with accuracy over 90%. Our findings reveal a structural basis for this form of epistasis and provide a predictive framework that could extend to at least 4% of human proteins.

Deep learning has accelerated protein design, but most existing methods are restricted to generating protein backbone coordinates and often neglect interactions with other biomolecules. We present RFdiffusion3 (RFD3), a diffusion model that generates protein structures in the context of ligands, nucleic acids and other non-protein constellations of atoms. Because all polymer atoms are modeled explicitly, conditioning the model on complex sets of atom-level constraints for enzyme design and other challenges is both simpler and more effective than previous approaches. RFD3 achieves improved performance compared to prior approaches on a range of in silico benchmarks with one tenth the computational cost. Finally, we demonstrate the broad applicability of RFD3 by designing and experimentally characterizing DNA binding proteins and cysteine hydrolases. The ability to rapidly generate protein structures guided by complex sets of atom-level constraints in the context of arbitrary non-protein atoms should further expand the range of functions attainable through protein design.

Understanding the protein sequence-function relationship is essential for advancing protein biology and engineering. However, fewer than 1% of known protein sequences have human-verified functions. While deep learning methods have demonstrated promise for protein function prediction, current models are limited to predicting only those functions on which they were trained. Here, we introduce ProtNote, a multimodal deep learning model that leverages free-form text to enable both supervised and zero-shot protein function prediction. ProtNote not only maintains near state-of-the-art performance for annotations in its train set, but also generalizes to unseen and novel functions in zero-shot test settings. We envision that ProtNote will enhance protein function discovery by enabling scientists to use free text inputs, without restriction to predefined labels – a necessary capability for navigating the dynamic landscape of protein biology.