Every peptide sitting in a biological matrix is a substrate for something. Which bond breaks first is not random, and it is not simply a function of how flexible the chain is: it is set by the geometry of a particular enzyme's cleft and by which residues can reach into it. This guide covers how proteolytic specificity is described, why dipeptidyl peptidase-4 and neprilysin recur in peptide stability work, and how a cleavage site is located at the bench rather than guessed.

Exopeptidases and endopeptidases: two different liabilities

The oldest and still most useful split among peptide-bond hydrolases is whether the enzyme requires a free chain terminus. Exopeptidases do: they dock onto a free alpha-amino group or alpha-carboxylate and remove residues from that end, and the naming encodes how many come off at once — aminopeptidases release one residue from the N-terminus, dipeptidyl peptidases two, tripeptidyl peptidases three, and carboxypeptidases work inward from the C-terminus. Endopeptidases cut internal bonds and have no such requirement. This taxonomy is a convention rather than a mechanistic law, and some enzymes straddle it.

The two classes create different kinds of vulnerability. Exopeptidase liability is a property of the ends of a sequence and the first few residues inward; endopeptidase liability is a property of internal sequence and can appear anywhere. The distinction propagates into methodology: the proteome-derived library method described below blocks all primary amines before the enzyme is added, which makes it a reader of endoprotease preferences [1], while the synthetic-library method of O'Donoghue and colleagues was presented as applicable to any endo- or exopeptidase [2].

Subsite nomenclature: P and S, primed and unprimed

Schechter and Berger mapped the active site of papain in 1967 and found it larger than a pocket that reads one residue [3]. In a later retrospective, Schechter described how protease specificity came to be presented as an active site divided into subsites, with S1, S2 and so on numbering the subsites of the enzyme and P1, P2 and so on numbering the substrate residues that occupy them. The account notes that important interactions occur in regions remote from the catalytic site, that mapping yields both the number of subsites and their individual properties including the free energy of ligand binding, and that predictions of the large-active-site model were confirmed by X-ray crystallography in 1970 [4].

The convention is worth stating precisely, because it is easy to get backwards. The scissile bond lies between P1 and P1'. Unprimed positions (P1, P2, P3...) run toward the substrate's N-terminus; primed positions (P1', P2', P3'...) run toward its C-terminus. Numbering increases outward from the cut in both directions, and the corresponding enzyme pockets are S and S'. A specificity profile states which residues are favored or disfavored at each numbered position.

Subsites are not independent. Rawlings and Bateman, describing analyses available through MEROPS, report binding site preferences and exclusions and identify peptidases in which cooperative binding occurs between adjacent binding sites [5]. Kridel and colleagues reached a related conclusion for matrix metalloproteinase-9: most of the phage-displayed substrates they selected were selective for MMP-9 over MMP-7 and MMP-13 despite sharing a common motif, which they took to indicate that selectivity is conferred by subsite interactions at positions other than P3 and P1' [6].

DPP-4: an exopeptidase that reads position two

Dipeptidyl peptidase-4 is the exopeptidase most often invoked in peptide stability work, and two crystallographic studies explain why its preference is sharp.

Thoma and colleagues expressed the ectodomain of human DPP-IV in Pichia pastoris and solved it at 2.1 A resolution. The enzyme comprises a catalytic domain with an alpha/beta hydrolase fold and a beta propeller domain built from an eight-fold repeat of a four-strand beta sheet motif. They report that the propeller contributes an additional short helix carrying a double Glu sequence motif, and that this motif provides recognition and a binding site for the N-terminus of substrates, as revealed by a complex with diprotin A, a substrate of low turnover trapped in the tetrahedral intermediate in the crystal [7].

Aertgeerts and colleagues added the substrate view. They note that DPPIV removes dipeptides from the N-terminus of substrates, and solved the enzyme both free and in complex with the first ten residues of neuropeptide Y. The free structure reveals two potential channels through which substrates could access the active site, a propeller opening and a side opening; the complex suggests bioactive peptides use the side opening. They attribute substrate recognition to a Glu motif, a well-defined hydrophobic S1 subsite and minimal long-range interactions, and report that the decapeptide is trapped as a tetrahedral intermediate rather than cleaved, with conformational changes at Ser630 and His740 stabilizing that intermediate [8].

Read together, these describe an enzyme that anchors a free alpha-amino group at a fixed distance from the well-defined hydrophobic S1 subsite. What that means for any particular analog is a question for measurement, not an inference the authors drew.

Neprilysin: an endopeptidase that still counts from the end

Neprilysin, also called neutral endopeptidase, is a zinc metallopeptidase. Oefner and colleagues determined the structure of the extracellular domain of human NEP, residues 52 to 749, in complex with the generic metalloproteinase inhibitor phosphoramidon at 2.1 A resolution. They describe two multiply connected folding domains that embrace a large central cavity containing the active site, with the inhibitor bound to one side of that cavity, and note that its binding mode gives a detailed understanding of ligand-binding and specificity determinants. The same paper lists substrates cleaved in vitro as including the enkephalins, substance P, endothelin, bradykinin and atrial natriuretic factor [9].

The subsite behavior is where NEP becomes a useful teaching case. Quay and colleagues state that neprilysin's specificity is directed toward cleavage on the amino side of hydrophobic residues, and that an active site arginine can interact with the C-terminal carboxylate of a substrate. They then walked a single Phe along a succinylated glycine-rich peptide series and observed an approximately 25-fold increase in the specificity constant kcat/Km when the Phe sat adjacent to the C-terminal Gly. With peptides carrying two cleavable bonds separated by a residue, cleavage occurred at the more C-terminal of the two; when the two were adjacent, a tripeptide was released even when the first of the two cleavable residues was one that was cleaved poorly when presented as the sole site. The authors concluded that the enzyme prefers a hydrophobic residue in the P2' position [10].

The point survives beyond this one enzyme: being classed as an endopeptidase does not make an enzyme blind to distance from a terminus, and a specificity statement naming only P1 or P1' can be incomplete in a way that matters.

MEROPS and what a cataloged cleavage means

MEROPS is the standard reference for peptidase classification. Rawlings and Bateman describe a hierarchy of at least three levels: each peptidase or inhibitor gets a distinct identifier based on its biochemical and biological properties, homologous sequences are assembled into families, and families whose proteins share similar tertiary structures are assembled into clans. Alongside the classification sit collections of sequence homologs, peptidase-inhibitor interactions, alignments and substrate cleavage sites [5].

That last collection carries a caveat the authors state plainly: it was assembled from the literature and includes physiological, pathological and nonphysiological cleavages, in proteins, peptides and synthetic substrates [5]. A record that an enzyme cleaves a bond means someone observed the cut under some conditions, not that it happens in the matrix you care about.

Mapping cleavage sites with peptide libraries

Two library approaches recur throughout current specificity mapping, differing in where the peptides come from.

Proteome-derived libraries (PICS)

Schilling and Overall introduced human proteome-derived, database-searchable peptide libraries. Primary amines in the library are protected, which allows sequence preferences on both the nonprime and prime sides of the scissile bond to be determined simultaneously. Prime-side products are tagged with biotin, isolated and identified by tandem mass spectrometry, and the corresponding nonprime sequences are derived bioinformatically from human proteome databases. The authors report that identifying hundreds to over 1,000 individual cleaved peptides allows the consensus cleavage site and subsite cooperativity to be determined from P6 to P6'. For the highly specific GluC protease, more than 95 percent of the 558 cleavage sites identified displayed the canonical selectivity; for the broad-specificity matrix metalloproteinase 2, more than 1,200 peptidic cleavage sites were identified. They also profiled HIV protease 1, caspases 3 and 7, cathepsins K and G, elastase and thrombin, showing the approach applies across mechanistic classes of endoproteases [1].

Biniossek and colleagues applied PICS to cysteine cathepsins B, L and S. For cathepsin L they identified 845 cleavage sites, described as the most comprehensive PICS profile to that date. Cathepsin B showed strong selectivity for glycine at P3', which they attributed to an occluding loop blocking access to the primed subsites, and a partial preference for phenylalanine at P1' not found for cathepsins L and S. Two controls there are worth copying: a shorter incubation, 4 hours instead of 16, did not reveal time dependence of the individual specificity determinants for cathepsins B and L, and cathepsin S profiles generated at pH 6.0 and at pH 7.5 displayed a high degree of similarity [11]. Neither result licenses assuming time or pH invariance for a different enzyme.

Synthetic libraries (MSP-MS)

O'Donoghue and colleagues built a physicochemically diverse library by incorporating all combinations of neighbor and near-neighbor amino acid pairs into decapeptide sequences flanked by unique dipeptides at each terminus, giving tetradecapeptides read out by LC-MS/MS sequencing. They present the method as revealing the substrate specificity of any endo- or exopeptidase, generating information on prime and nonprime sites. In that paper, label-free quantitation of precursor-ion abundance allowed granzyme B substrates to be ranked by enzymatic turnover efficiency, and proteolytic secretions from a parasite larval stage and a pancreatic cancer cell line were deconvoluted using class-specific peptidase inhibitors in a subtractive strategy [2]. That design — run the mixture with and without a class-selective inhibitor, then difference the results — is the general tool for assigning activity in a matrix containing many peptidases.

Mapping cleavage in a real proteome: N-terminomics

Library methods report what an enzyme can do to short peptides. N-terminomics asks what it did to intact proteins. Prudova and colleagues developed whole-protein iTRAQ labeling that simultaneously labels and blocks all primary amines, including protein N-termini and lysine side chains. Blocking lysines limits subsequent trypsin cleavage to arginine, elongating the truncated peptides and improving MS/MS identification; negative selection then enriches neo-N-termini. Comparing the substrate degradomes of MMP-2 and MMP-9 in fibroblast secreted proteins, they identified 3,152 unique N-terminal peptides corresponding to 1,054 proteins, and detected 201 cleavage products for MMP-2 but, in their word, unexpectedly only 19 for the homologous MMP-9 under identical conditions [12].

That asymmetry is the most useful cautionary datum here: two close homologs, one experiment, one set of conditions, an order-of-magnitude difference in detected cleavage products. Similarity of catalytic clefts did not predict substrate repertoire.

Zelanis and colleagues ran both methods on a single snake venom metalloprotease and cross-checked them. TAILS on a mouse embryonic fibroblast secretome identified more than 2,000 N-termini and indicated a preference for Leu at P1'; PICS with a tryptic library from a monocytic cell line detected 799 unique cleavage sites and returned the same P1' preference. They also noted that 190 unique peptides matched annotated cleavage sites in the TopFIND database, and suggested these occurred either at sites prone to cleavage or were generated by other proteases activated during the incubation, naming caspases, cathepsins, granzyme B and MMPs among the possibilities [13]. Secondary proteolysis during a long incubation is a documented confounder, not a hypothetical one.

What a specificity profile does not tell you

A consensus motif is a summary statistic over a library, and the gap between that statistic and a list of real substrates can be wide. Kridel and colleagues profiled MMP-9 by phage display, identified three groups of substrates each occupying a distinct set of subsites, and found the most prevalent motif to be Pro-X-X-Hy-(Ser/Thr) spanning P3 through P2', similar to MMP cleavage sites within the collagens; they also report a unique preference for Arg at both P2 and P1 and for Ser/Thr at P2'. When they queried protein databases with the consensus, the result was what they called a surprisingly limited list of putative physiologic substrates, framed as testable hypotheses rather than conclusions [6].

Three limits are worth carrying forward. A profile ranks preferences; it does not give a half-life, because turnover depends on enzyme concentration, accessibility and competing substrates as well as sequence fit. The library bounds what can be observed — contexts absent from it cannot be scored, and a proteome-derived library inherits the composition of the digest that made it. And an in vitro cleavage says nothing about whether the same cut occurs in a different matrix; that is a separate experiment with its own controls.

Designing a cleavage-mapping experiment

The following are standard bench practice rather than findings from any particular paper.

  • Run a time-zero aliquot and a heat-inactivated matrix alongside the live incubation. Fragments present at time zero are sample handling, not enzymology.
  • Run a buffer-only blank of the substrate at the same temperature and pH, so chemical hydrolysis is separated from enzymatic cleavage.
  • Quench reproducibly, and check that the quench itself does not generate new cleavage — strong acid at elevated temperature can.
  • Identify products by accurate mass and MS/MS sequence, not by disappearance of the parent alone. Loss of parent tells you something happened; it does not tell you where.
  • Look for complementary fragment pairs. Hydrolysis of one backbone bond consumes one water, so the two product masses should sum to the parent mass plus about 18.011 Da at monoisotopic resolution. A fragment with no partner deserves scrutiny.
  • State the incubation time, temperature, pH and matrix with every profile. All four change what you see.

What is not known

Specificity mapping is mature as a descriptive science and weak as a predictive one. No general method takes a sequence and returns a ranked list of the bonds a given matrix will cut, with rates. Subsite contributions have been mapped for only a handful of positions in most peptidases, and cooperativity between adjacent subsites is documented for some enzymes and uncharacterized for most. Where a laboratory reports a peptide as stable or unstable, the informative statement is the one naming the enzyme or matrix, the conditions and the observed cleavage position — not the adjective.