Proteases hydrolyse peptide bonds because water is the nucleophile that reaches the acyl-enzyme intermediate first. Give the same active site an activated ester as the acyl donor and a peptide whose free α-amine can compete against water for that intermediate, and the enzyme forms a bond instead of breaking one [1, 2]. Every practical decision in enzymatic ligation — which enzyme, which leaving group, which solvent, which temperature — is an argument about that one competition, and about what the enzyme does to the product once the bond exists.
Kinetic control and thermodynamic control are not the same protocol
Krieg and Wolf ran both modes on a single enzyme and the contrast is clean. With an activated peptide ester as the acyl component, proline-specific endopeptidase catalysed kinetically controlled aminolysis onto amino acid amides; with a peptide amide as the acyl component it catalysed transpeptidation; and with short peptides carrying a free C-terminal carboxylate it catalysed thermodynamically controlled condensation with the same class of acceptors [1].
The distinction matters because the two modes fail differently. Thermodynamic condensation is an approach to equilibrium: the catalyst changes how fast you arrive, not where the endpoint sits, so the only levers are the ones that move the equilibrium itself. Kinetic control spends the chemical potential stored in the activated ester, so it can reach yields the equilibrium would forbid — but it buys that with a race. The acyl-enzyme must be aminolysed before it is hydrolysed, and the product amide must then survive an enzyme that is, by ancestry, an amidase.
The same report shows that the central engineering move generalises beyond subtilisin: converting the catalytic Ser556 of the endopeptidase to Cys gave a variant with dramatically increased peptide ligase activity in aqueous solution [1]. Swapping the catalytic oxygen nucleophile for sulfur is not a subtilisin trick.
Why subtiligase needed two mutations
Thiolsubtilisin — subtilisin BPN′ with the catalytic Ser221 converted to Cys — had already been shown to have some peptide ligase activity [2]. It was not enough to make a reagent. Abrahmsén and colleagues added a second substitution, Pro225 to Ala, and the resulting double mutant S221C/P225A has 10-fold higher peptide ligase activity and at least 100-fold lower amidase activity than the S221C single mutant [2].
The design argument was steric. A 1.5 Å crystal structure of an oxidised derivative of the double mutant showed that the P225A change partly relieves the crowding expected from the S221C substitution, which the authors took as accounting for the improved catalytic efficiency [2]. Read that as support for the design rationale rather than as a picture of the working catalyst: the crystallised species was an oxidised derivative, not the thiol form that turns over.
Two further findings in the same paper set the operating envelope. Stable, synthetically reasonable alkyl ester peptide substrates were prepared that rapidly acylate the S221C/P225A enzyme, and aminolysis of the resulting thioacyl-enzyme intermediate by various peptides is strongly preferred over hydrolysis [2]. The acyl donor therefore has to be made as an ester. You do not hand this enzyme a free acid and expect a bond.
What the enzyme was shown to build
Jackson and colleagues assembled ribonuclease A, 124 residues long, by stepwise ligation of six esterified peptide fragments of 12 to 30 residues each, at yields averaging 70% per ligation, producing fully active enzyme in milligram quantities [3]. Six fragments means five junctions; at 70% apiece, under a fifth of the starting material reaches full length, which is the honest arithmetic of blockwise enzymatic assembly. The payoff was a substitution that recombinant expression could not deliver at the time: the catalytic histidines at positions 12 and 119 were replaced with 4-fluorohistidine, which has a pKa of 3.5 against 6.8 for histidine. That produced large changes in the pH–rate profile with surprisingly little change in kcat for either the RNA cleavage or the hydrolysis step, from which the authors concluded that these imidazoles function as general acids and bases but that the proton transfer steps are not rate-limiting when the imidazoles are correctly protonated [3].
Semisynthesis was the other early application. Chang and colleagues used subtiligase to ligate esterified peptides site-specifically onto the N termini of proteins or peptides in aqueous solution and in high yield, making biotinylated and heavy-atom derivatives of methionyl-extended human growth hormone and a PEG-modified atrial natriuretic peptide, and establishing the N-terminal sequence requirements for efficient ligation using both synthetic substrates and pools of filamentous phage displaying Met-hGH with randomised N-terminal sequences [4]. For acceptors whose N termini are highly structured or buried, they designed a more stable subtiligase that still ligates peptides onto Met-hGH in 4 M guanidine hydrochloride [4]. That is the practical answer when the α-amine you want is not solvent-exposed: denature the acceptor and use a ligase that tolerates the denaturant.
Improving the catalyst itself was done by selection rather than design. Atwell and Wells displayed subtiligase on phage, randomly mutated 25 active-site residues in groups of four or five to yield six independently sorted libraries drawn from more than 109 independently derived mutants, and selectively captured variants that ligated a biotin peptide onto their own extended N termini [5]. The screen returned mutants with increased ligase activity and, unbidden, residues already known to improve the stability and oxidative resistance of wild-type subtilisin [5].
Synthesis against hydrolysis: the ratio you are optimising
A parallel subtilisin-derived lineage treats the partition explicitly as a design variable. Starting from the peptiligase variant M222P/L217H, Toplak and colleagues introduced A225N in the S1′ pocket and F189W in the S2′ pocket, both of which increased the synthesis-to-hydrolysis (S/H) ratio and overall coupling efficiency, and added I107V in the S4 pocket to increase the reaction rate; the final omniligase variants coupled more than 250 peptides in a 400-member library of acyl acceptors as read out by a high-throughput FRET assay, with crystal structures and computational modelling used to rationalise the behaviour [6].
Note where the productive mutations sit. In Schechter–Berger nomenclature the primed subsites bind residues on the far side of the scissile bond, which in a ligation is the incoming acceptor peptide; the S/H gains came from S1′ and S2′, while the rate increase came separately from S4 on the donor side [6]. If a ligation is turning over but hydrolysing, the acceptor-side subsites are where the problem is likely to live.
The other lever is the medium. Jönsson and colleagues immobilised α-chymotrypsin on celite and ran the aminolysis of N-acetyl-L-phenylalanine ethyl ester with L-alaninamide in four water-miscible solvents containing small amounts of water, between −30 °C and 25 °C. Low temperature favoured the nucleophile over water in the competition for the acyl-enzyme, the effect was greatest in acetonitrile and tetrahydrofuran, and secondary hydrolysis was reduced at subzero temperatures because ester hydrolysis was favoured relative to dipeptide hydrolysis; in acetonitrile containing 10% water the maximal yield was 99% at −20 °C compared with 84% at 25 °C [7].
That is the water-activity argument in measurable form, and it comes with a boundary. The measurement was a dipeptide, an immobilised chymotrypsin and a water-miscible organic medium. Subtiligase and its relatives are deployed on folded proteins in aqueous buffer, where water activity cannot be dropped far without unfolding the acceptor. The transferable lesson is directional — cold, drier, and an acceptor that competes well all shift the partition toward synthesis — not numerical.
Specificity at the junction, measured rather than assumed
The original characterisation reported that the efficiency of aminolysis is relatively insensitive to the sequence of the first two residues in the acyl acceptor peptide whose α-amino group attacks the thioacyl-enzyme, and that three additional ligases carrying mutations known to change subtilisin’s preference at P1 showed specificity properties paralleling those of the wild-type enzyme and its corresponding specificity variants [2]. That is a real result on the peptides tested, and it is why subtiligase became a general reagent rather than a sequence-specific one.
It is not flat, and the size of the deviation only became visible when the measurement got large. Weeks and Wells built proteome-derived peptide libraries and characterised ligation efficiency for more than 25,000 enzyme–substrate pairs, identifying a family of 72 mutant subtiligases with activity toward N-terminal sequences that had previously been recalcitrant to modification [8]. They then used those mutants individually for site-specific bioconjugation of purified proteins, including antibodies, and in algorithmically selected combinations for sequencing of the cellular N terminome with reduced sequence bias, along with a web application that selects the most efficient variant or variants for a user-defined sequence [8].
The practical consequence is a change in how a ligation gets planned. A single subtiligase is a reagent with a preference; a panel chosen against the actual N-terminal sequence is a reagent with coverage. When a target refuses to ligate, the first hypothesis should be the acceptor sequence rather than the enzyme loading.
N-terminal selectivity and N-terminomics
What makes subtiligase a proteomics reagent as well as a synthesis reagent is that the attacking nucleophile is the α-amino group of the acceptor [2]. Mahrus and colleagues built a global method on exactly that: an engineered enzyme selectively biotinylates free protein N termini, which permits positive enrichment of the corresponding N-terminal peptides rather than depletion of everything else [9].
Applied to apoptosis, the method sequenced 333 caspase-like cleavage sites distributed among 292 protein substrates [9]. Two features of that dataset should carry into any N-terminomics design. The sites are generally not predicted by in vitro caspase substrate specificity, though they can be used to predict other physiological caspase cleavage sites; and structural bioinformatic analysis placed them often in surface-accessible loops and even occasionally in helical regions [9]. A captured neo-N terminus therefore establishes that a free α-amine existed at that position in that sample. It does not, from sequence alone, identify the protease that made it — the in vitro specificity you would use for that inference is the thing the data argued against.
Compartment-targeted variants extend the same logic to places where abundance defeats a bulk experiment. Cell surface proteolysis had been hard to profile because plasma membrane proteins are a low relative fraction of the proteome; Weeks and colleagues applied plasma membrane-targeted subtiligase variants (subtiligase-TM) to capture cell surface N termini in live cells, sequenced 807 of them, and quantified changes in their abundance in response to stimuli that induce proteolytic remodelling of the cell surface proteome [10]. The comprehensive account of the technology — its development, its applications, its stated advantages and limitations, and how it compares with other peptide ligase enzymes — is Weeks and Wells’s review [11].
Transpeptidases: a different bargain
Sortase A occupies the same conceptual slot with different chemistry and a different cost. It cleaves polypeptides at the threonine of an LPXTG motif; purified sortase catalyses the hydroxylaminolysis of LPXTG-bearing peptides, the reaction is inhibited by sulfhydryl-modifying reagents, and replacing the single conserved cysteine at position 184 with alanine abolishes activity — which is what pointed to a thioester enzyme intermediate [12]. As a ligation tool, a protein or peptide bearing an LPXTG motif can be specifically joined through an amide bond to an aminoglycine partner, and the same activity conjugates D-peptides, synthetic branched peptides and aminoglycine-derivatised small molecules to the C terminus of a recombinant protein [13].
The difference from subtiligase is where the recognition lives. Sortase reads a defined motif on the donor and rebuilds an LPXT–glycine junction in the product, so the recognition sequence is present in the product by construction [12, 13]. Subtiligase reads a preference rather than a motif, and the junction it creates is an ordinary peptide bond whose sequence you chose [2, 8]. Which of those is the advantage depends entirely on whether a fixed insertion at the joint is acceptable in the construct.
The plant asparaginyl ligases sit between the two. Butelase 1, isolated from Clitoria ternatea, shares 71% sequence identity and the same catalytic triad with legumain proteases yet does not hydrolyse the protease substrate of legumain; it cyclises various peptides of plant and animal origin in yields greater than 95%, with kcat values up to 17 s−1 and catalytic efficiencies as high as 542,000 M−1 s−1, and it displays broad specificity for the N-terminal amino acids of the peptide substrate, which is what enables C terminus-specific intermolecular ligation [14]. The constraint sits on the donor: the enzyme is Asx-specific.
OaAEP1 is the recombinant answer to a supply problem. The wild-type enzyme from Oldenlandia affinis displayed only weak peptide cyclase activity, and its protein–protein ligation activity was described as extremely weak; a 2.56 Å structure of the preactivation form guided mutagenesis that produced an ultrafast variant with catalytic kinetics hundreds of times faster, able to ligate well-folded protein substrates using only submicromolar enzyme [15]. Mind the analyte class and the comparator: that demonstration was on folded proteins, and the rate figure is against the wild-type enzyme, not against subtiligase.
Against the chemical ligations
Native chemical ligation solves the same assembly problem without an enzyme. Two unprotected peptide segments react chemoselectively to give an initial thioester-linked species, which rearranges spontaneously to a full-length product carrying a native peptide bond at the ligation site; the demonstration was a one-step preparation of a cytokine containing multiple disulfides, subsequently folded and oxidised [16]. With no protein catalyst there is no active site to be recalcitrant — but the junction is not free either. The reaction is defined at Xaa-Cys, and its extension to Xaa-Phe, Xaa-Val, Xaa-Lys, Xaa-Leu, Xaa-Thr, Xaa-Pro and, later, Xaa-Gln junctions came from post-ligation desulfurisation of thiol-bearing residue surrogates rather than from relaxing the ligation chemistry itself [17].
Both routes are constrained at the junction; the constraints differ in kind. The chemical constraint is paid off with synthesis — build the thiol surrogate, remove it afterwards [17]. The enzymatic constraint is paid off with protein engineering — select a variant from a characterised panel [8]. Native chemical ligation needs a synthetic thioester segment [16]; subtiligase needs a synthetic ester acyl donor [2] and an acceptor whose α-amine is accessible, which for a folded protein may mean running the ligation under denaturant [4].
What a successful ligation licenses you to conclude
A mass matching the ligated product establishes that a species of the right composition formed. It does not establish that the bond formed where it was intended to. For subtiligase the regiochemical argument rests on the α-amino group being the attacking nucleophile [2] and on the enzyme selectively acylating free protein N termini in a whole proteome [9]; for sortase it rests on the motif [12]. In both cases the claim is about the enzyme’s demonstrated selectivity in the systems tested, not about your particular substrate, and the way to close that gap is a proteolytic digest and sequencing of the junction peptide.
Three checks are worth building in from the start. Run a no-enzyme control with the same acyl donor and acceptor, so anything you attribute to the enzyme is attributed by difference. Take a time course rather than a single endpoint: kinetic control means the product coexists with a catalyst whose amidase activity was reduced by engineering rather than abolished, and the quantity that matters — synthesis against hydrolysis — is a ratio of rates, not a state [2, 6]. And when a sequence refuses to ligate, treat that as a specificity measurement rather than a failure; the tools now exist to say which variant, if any, will accept that N terminus [8].