Amide bonds without protecting groups: de novo ligases put site choice into the binding pocket
Synopsis
Amide synthesis has long relied on stoichiometric coupling agents and protecting groups, while enzymatic routes are limited by narrow substrate scope and directed evolution. Researchers used deep-learning de novo design to build ligases with a cysteine catalytic nucleophile and donor and acceptor pockets flanking the active site, encoding regioselectivity and stereoselectivity in the design. KL37 modified only the intended lysine among 36 competitors, and LL52 preferred the (S)-amine 28-fold over the (R)-amine in a lenacapavir-forming step.
Interpretation
Amide bond formation no longer has to come from chemical coupling agents or engineered natural enzymes; computationally designed ligases can catalyse bond formation at a specified amine. Conventional amide synthesis needs protecting groups to confine the reaction to the intended site, adding steps, cost and waste, while enzymatic ligation often requires inserting a recognition sequence or evolving substrate specificity for each new target. On an MBP-peptide substrate carrying 36 lysines and a free N-terminus, replacing the target lysine with arginine gave no detectable product over 20 hours, indicating KL37 modifies the designed site.
The enzymes work through a cysteine catalytic triad: cysteine attacks the activated ester donor to form an acyl-enzyme intermediate, which the acceptor amine then attacks to give the amide. Design started from cysteine-protease-derived theozymes, used RFdiffusion models to generate scaffolds, LigandMPNN for sequences and AlphaFold3 for structure prediction, and filtered designs on catalytic geometry across states along the reaction path. NL6 converted 90% of the FSLTK substrate to labelled product within 16 hours, 4.5 turnovers per enzyme; KL37 labelled nearly 90% of the acceptor within 5 hours, about 3.6 turnovers, with donor hydrolysis indistinguishable from background.
Selectivity comes from catalytic geometry that forms only when the acceptor binds: without it the acyl group and oxyanion hole are misaligned, and water attack is poorly organised. Earlier serine-hydrolase designs also formed acyl-enzyme intermediates without efficient hydrolysis, and modelling attributed that to acyl-group reorientation disrupting productive water attack. For NL6, median distances from the acyl carbonyl oxygen to the two oxyanion-hole backbone amides fell from 3.8 to 3.0 Å and from 5.4 to 3.8 Å on acceptor binding; KL37's His119-Cys176 distance narrowed from 3.6-7.2 Å without peptide to 3.5-3.7 Å with it.
Perspective
The demonstrated scope is three reaction classes: acylation of a peptide N-terminal alpha-amine, acylation of a single lysine epsilon-amine, and a stereoselective amide-forming coupling in lenacapavir synthesis; each was run with recombinant enzymes expressed in E. coli and purified substrates, with conversion and selectivity read by in vitro HPLC, mass spectrometry and gels.
The authors propose that the amine acceptor both supplies the nucleophile and organises catalytic geometry, and note that future campaigns could add explicit negative design against preorganisation for water attack in the acceptor-free state; an initial RFdiffusion3 campaign on the lenacapavir step yielded no active designs, and activity appeared only with RFD4-Proteina.
