Similar to most proteinases, matrix metalloproteinases (MMP) do not recognize a consensus cleavage site. Thus, it is not surprising that, in a defined in vitro reaction, most MMPs can act on a wide range of proteins, including many extracellular matrix proteins. However, the findings obtained from in vivo studies with genetic models have demonstrated that individual MMPs act on just a few extracellular protein substrates, typically not matrix proteins. The limited, precise functions of an MMP imply that mechanisms have evolved to control the specificity of proteinase:substrate interactions. We discuss the possibility that interactions with the glycosaminoglycan chains of proteoglycans may function as allosteric regulators or accessory factors directing MMP catalysis to specific substrates. We propose that understanding how the activity of specific MMPs is confined to discreet compartments and targeted to defined substrates via interactions with other macromolecules may provide a means of blocking potentially deleterious MMP-mediated processes at the same time as sparing any beneficial functions.
Metalloproteases comprise a large superfamily of endopeptidases (187 human genes; 194 mouse genes) that contain an active site Zn2+ and are divided into subfamilies based on evolutionary relationships and the structure of the catalytic domain. The metzincin subclan of metalloproteases is characterized by a 3-histidine zinc-binding motif and a conserved methionine turn following the active site . Metzincins include the serralysins, astacins, ADAMs (a disintegrin and metalloproteinase) and matrixins, otherwise known as the matrix MMPs.
Mammalian MMPs are a family of 24 related extracellular endoproteinases (Table 1). All MMPs have conserved pro- and catalytic domains (Fig. 1). MMP prodomains contain the sequence PRCXXPD, and their catalytic domains are defined by a 3-histidine motif (HEXXHXXGXXH) to which the active site Zn2+ is held. Ligation of the catalytic domain zinc ion with the free thiol of the prodomain cysteine retains the latent state of proMMPs, and this bond must be broken for an MMP to gain activity.
Table 1. MMPs
Direct GAG interactions
Mcol-A and -B are likely the murine homologues of MMP-1 . MMP-1 is not present in rodent genomes.
With the exceptions of MMP-7, MMP-23 and MMP-26, a variable, haemopexin-like domain, linked by a short hinge, makes up the C-termini of MMPs (Fig. 1). Importantly, the haemopexin domains likely serve critical roles in substrate specificity among MMPs . Other functional motifs are present within specific MMPs. For example, MMP-14, -15, -16 and -24 have cytosolic and transmembrane motifs, and MMP-2 and MMP-9 have gelatin-binding domains (Fig. 1). Overall, the three-dimensional structure of the catalytic domain of MMPs is quite similar , and thus, substrate specificity is not easily explained by the structural features within the enzymes themselves.
Although some MMPs are anchored to the cell surface (i.e. the membrane type-MMPs), most are produced as secreted proteins. However, as argued here and elsewhere , it is unlikely that MMP proteolysis could be targeted to specific substrates if the enzyme was simply released from the producing cells. Furthermore, as modelled and demonstrated by Campbell et al.  with serine proteinases released from activated neutrophils, freely-secreted proteinases would be rapidly diluted to ineffective concentrations and silenced by natural inhibitors (which can be abundant in tissue fluids) as they diffuse from the cell. Thus, it is reasonable to propose that MMP proteolysis is confined to cellular compartments, such as secretion vesicles and cell membranes, where the concentrations of both the enzyme and substrate can be maintained at catalytically favourable levels.
MMPs are called ‘matrix’ metalloproteinases because they have long been considered as the principal enzymes responsible for the turnover and degradation of extracellular matrix proteins, such as collagens, fibronectin and laminins, etc. Indeed, some MMPs do act on extracellular matrix, most notably MMP-14, which appears to be an authentic fibrillar collagenase [6-8]. However, MMPs act on a wide range of non-matrix extracellular proteins, such as cytokines, chemokines, numerous receptors and membrane proteins, and antimicrobial peptides, amongst others [9-13]. Thus, MMPs appear to function by regulating cell–cell and cell–matrix signalling typically via processing of latent proteins largely involved in repair and immune processes [14-16].
MMP activity is controlled at four levels: biosynthesis (transcription/translation), zymogen activation, compartmentalization and inactivation. Some members, such as MMP-2, MMP-19, MMP-28 and several membrane-type MMPs, may play a role in homeostasis because they are expressed in resting tissues. By contrast, several MMPs are produced in response to tissue injury and infection and function in a wide range of repair, inflammatory and defence processes that are, for the most part, beneficial [9, 12, 17, 18]. The fact that MMPs mediate largely favourable processes is neither surprising, nor controversial. It is unlikely that this family of proteinases expanded through evolution to cause harm. However, in a disease setting, such as chronically inflamed tissue and most cancers [9, 10, 13, 19], excess or inappropriately expressed MMPs can contribute to tissue destruction, tumour progression and invasion, as well as other deleterious processes.
Although several MMPs may be expressed within a given setting, such as a healing wound, an atherosclerotic plaque, an inflamed joint or a tumour, different cell types express different patterns of MMPs under different conditions. Even though a given MMP may be comprised of two different cells types within the same setting, such as is often seen with MMP-3, MMP-9 and MMP-14, it is possible, if not indeed likely, that the same proteinase produced by different cells acts on different substrates and hence impacts different processes. Thus, a central question in metalloproteinase biology is how specific proteinase:substrate interactions are controlled to limit MMP action to specific, targeted substrates, which, in turn, affect specific cellular processes. In this review, we discuss how allosteric interactions [with an emphasis on the glycosaminoglycan (GAG) chains of proteoglycans] can control and promote compartmentalization, proenzyme activation and substrate availability. The concept that MMP activity is confined to pericellular niches via interactions with extracellular macromolecules has been discussed elsewhere [2, 4, 20, 21].
As stated above, the thiol-Zn2+ bond must be broken for a proMMP to gain catalytically activity, and disruption of this bond is referred to as the ‘cysteine switch’ . The cysteine switch can be activated by direct cleavage of the prodomain by itself (autolysis) or another proteinase, modification of the free thiol by physiological or experimental means , or allosteric perturbation of the zymogen (Fig. 2A). Some MMPs, including all membrane-bound MMPs, contain a furin-recognition sequence between the pro- and catalytic domains (Fig. 1) and are activated in the secretion pathway . In vitro, several MMPs, as well as proteases of other classes, such as the serine proteinase plasmin, have been suggested to activate various proMMPs [24-28]. Similarly, oxidation modification of the cysteine thiol has been shown to activate several proMMPs in vitro [20, 29-33]. However, the in vivo activation mechanism of essentially all nonfurin cleaved proMMPs is unknown.
A notable exception to this caveat may be the activation of proMMP-2 at the cell surface by active MT1-MMP/MMP-14 in a complex with tissue inhibitor of metalloproteinase (TIMP)-2 [25-28, 34] or by MT3-MMP/MMP-16 . Within the ternary complex, TIMP-2 brings the zymogen and active proteinase in close proximity, allowing MMP-14 to cleave the prodomain of proMMP-2. Although the role of TIMP-2 in proMMP-2 activation is supported by a lack of zymogen activation in TIMP-2 null mice , a similar defect in MMP-14 knockout mice has not been reported and activation of proMMP-2 in fibroblasts is not markedly, if at all, affected by deficiency of MMP-14 .
A key aspect of the cysteine switch mechanism is that the only necessary step is that the zinc-thiol interaction be disrupted [4, 20]. The prodomain can still be attached, and the MMP can be active (Fig. 2). Indeed, in a substrate zymogram, SDS in the running buffer is sufficient to disrupt the zinc-thiol conferring full activity to proMMPs. ProMMPs can interact with nonsubstrate macromolecules, such as GAGs (see below), causing conformational changes that disrupt the cysteine-Zn2+ bond (Fig. 2A). Allosteric disruption of the cysteine-Zn2+ bond creates a transitional active state permitting inter- or intramolecular autolytic cleavage to yield the final active form (Fig. 2).
Compartmentalization refers to a regulated process that functions to confine where, in the pericellular environment, an MMP (or any protein) is released and held and is arguably the essential mechanism for regulating the substrate specificity of any proteinase. After all, undirected release of an active proteinase could lead to unwanted degradation of any protein (or proteins) that comes into the path of the enzyme. When left to themselves, most proteinase are promiscuous as to which substrates they cleave. Thus, targeted extracellular proteolysis would require that the enzyme be anchored to the cell membrane or matrix or held within multi-component complexes, ensuring a sufficiently high enzyme concentration within the locale of the desired substrate and, in turn, barring off target effects. Indeed, a number of mechanisms have been described or proposed to compartmentalize MMPs , such as the binding of MMP-2 to the αvβ3 integrin , MMP-1 to the α2β1 integrin [38, 39], MMP-9 to CD44  and MMP-7 cholesterol  or CD151 . As suggested for CD44  and the α2β1 integrin , these attachment factors could facilitate autolytic proenzyme activation and the interaction with and, hence, cleavage of the target substrate. Overall, such a process would increase the probability that proteolysis would be highly regulated and specific. Such as is typical of just about any protein–protein interactions, MMPs are likely attached to cells or matrix via precise interactions.
An interaction between a proMMP with a macromolecule on the cell surface or in the matrix could promote autolysis of the proenzyme by allosteric interference of the cysteine-switch (Fig. 2A). In support of this idea, the tetraspanin CD151, a transmembrane protein, ligates and promotes the activation of proMMP-7 . It was proposed that ligation with CD151 mediates a structural change in proMMP-7 favouring autolytic activation of the zymogen. Interestingly, activated MMP-7 was only found if an in vitro substrate of MMP-7 (carboxymethylated transferrin) was added to the culture medium . Hence, ligation of both substrate and zymogen to a membrane attachment factor, such as an integrin, tetraspanin or the GAG chains of a syndecan, could achieve all of this. That is, when held within such an organized complex, conformational changes in the zymogen could favour disruption of the repressive thiol-Zn bond and, in turn, promote autolytic activation and substrate cleavage (Fig. 2B).
In vitro, a proteolytic reaction requires only a proteinase and a substrate (or more minimally, just a proteinase if the enzyme can act on itself, as many MMPs can). However, in vivo, specificity would require one further (if not more) component. Such accessory or attachment factors or anchors could be proteins, either membrane-associated or extracellular matrix, lipids, or GAGs . One class of molecules that has emerged as being central to the regulation and activation of MMPs comprises the GAGs, either free as heparin and hyaluronan or as chains of proteoglycans. GAGs interact with a wide variety of effector proteins, such as chemokines and growth factors, to elicit various processes , and they have a well-established ability to control serine proteinases activities. For example, heparin, a highly sulfated mast cell GAG, affects the activity of several coagulation cascade proteinases and inhibitors, such as thrombin, factor Xa and antithrombin III [44, 45], as well as tryptase, chymase and cathepsin B [46-49].
Several studies have investigated how interactions between GAGs and proteinases impact physiological and disease processes and enzyme function [35, 50-54]. These studies found that GAGs control the cellular localization of certain proteinases and can modulate their conformation, activity, substrate preference and stability. Furthermore, GAG:proteinase interactions tend to be highly specific, and GAGs with different degrees or patterns of sulfation mediated vastly different effects on the same proteinase [51, 54, 55].
Overall, the ability of GAGs to affect proteinase activity is influenced by the degree of sulfation and chain size [55, 56]. For example, both zymogen activation and catalytic activity of cathepsin D, an aspartic proteinase, are promoted by heparin, and these affects are dependent on GAG sulfation and proportional to chain length . Li et al.  reported that chondroitin-4-sulfate (aka CS-A) stabilizes cathepsin K, a cysteine proteinase, and expands its pH optimum resulting in a markedly enhanced collagenolytic activity. In addition, although long considered to be strictly lysosomal enzymes that cannot function in the extracellular space, many cysteine proteinase are secreted from cells, including macrophages , and are active within the neutral pH environment of tissue fluid when associated with extracellular GAGs [17, 18]. Furthermore, procathepsin L, a cysteine proteinase secreted by macrophages, gains catalytic activity against matrix proteins when it is bound to proteoglycans , including chondroitin sulfate proteoglycans. By contrast, although chondroitin-4-sulfate augments the ability of cathepsin K ability to cleave fibrillar collagens , this GAG, as well as dermatan sulfate, chondroitin-6-sulfate and heparin, inhibits the potent elastolytic activity of cathepsins K and V . Because many of the enzymes discussed above are macrophage proteinases implicated in inflammatory tissue destruction, the local GAG content and composition could markedly impact upon their ability to cause matrix destruction in conditions such as emphysema, aneurysms, and atherosclerosis, amongst others.
Regarding MMPs and other metalloproteinases, the autolytic activation of proMMP-2 and the activity of MMP-1 are enhanced by heparin [58, 59]. Iida et al.  reported that the chondroitin-4-sulfate chains of melanoma-specific chondroitin sulfate proteoglycan bind to the C terminus of pro-MMP2 to facilitate its activation by cell-bound MT3-MMP/MMP-16. Cell surface activation of ADAMTS4 by MT4-MMP involves an interaction with the GAG chains of syndecan-1  and activation of ADAMTS5 is driven by binding syndecan-4 . Similarly, pro-ADAM12 binds to a cell surface heparan sulfate, possibly a syndecan, which functions to limit its catalytic activity, a property dependent on sulfation . For ADAM12, Sorenson et al.  proposed the interesting idea that this inhibition is a result of the binding of the GAG to basic amino acids in the pro- and catalytic domains of ADAM12 creating a unique molecular switch.
Serglycin is a chondroitin-4,6-sulfate proteoglycan that is found in the secretion vesicles of a variety cell types [62, 63]. Mice with targeted disruption of the serglycin gene (Srgn−/−) are overtly normal, yet their mast cell proteases have reduced activity and are missorted . As a result of its location within the secretion pathway and its known role in serine proteinase regulation, serglycin is a good candidate for regulating MMP zymogen activation. In THP-1 cells, a macrophage cell line, proMMP-9 is covalently linked to chondroitin sulfate proteoglycans, most likely serglycin, which induces the processing of proMMP-9 at both its N- and C-termini [64, 65]. However, proMMP-9 activation is not reliant on the chondroitin sulfate chains.
GAGs and MMP7
MMP-7 is a product of nonskin (i.e. mucosal) epithelia and functions as a key effector of repair and immunity [9, 17, 66]. The established functions of MMP-7 include elastolysis by human macrophages  and promoting epithelial apoptosis by cleaving Fas ligand [68, 69]. MMP-7 is induced in response to epithelial injury and, at early stages, sheds syndecan-1 to control both neutrophil influx  and re-epithelialization[71-73] and, subsequently, sheds E-cadherin to apparently promote the homing and activation of immunosuppressive dendritic cells [72, 74].
In mice, MMP-7 activates pro-α-defensins (procryptdins), a family of 3–4-kDa antimicrobial peptides found in the granules of Paneth cells that lie at the base of the crypts of Lieberkühn . As a result of a lack of active α-defensins, Mmp7−/− mice are susceptible to gut pathogens that are readily cleared by wild-type mice . α-Defensins are packaged as pro-proteins and cleaved by MMP-7 within the secretion vesicles [76, 77]. In granules of resting Paneth cells, the steady-state levels of pro- and activated cryptdins are approximately equivalent and, upon stimulation, the unprocessed procryptdins are rapidly activated, indicating efficient proteolysis by MMP-7 within the secretory pathway [77, 78]. However, in defined in vitro reactions containing just substrate and proteinase, the activation of procryptdins by MMP-7 is slow, with only 50% of the precursor cleaved within 8 h or longer [55, 76]. Furthermore, both pro- and active MMP-7 are present in Paneth cells granules , indicating that this MMP is activated by prodomain cleavage within the secretion pathway. The inefficient cleavage of procryptdins in vitro, their rapid processing in vivo, and the presence of active MMP-7 in Paneth cell granules led us to hypothesize that other factors regulate both the activation of proMMP-7 and its activity against physiological substrates.
Yu et al. [79, 80] reported that MMP-7 colocalizes with heparan sulfate and that heparin increases MMP7 activity by approximately two to fourfold in a transferrin zymogram assay. These findings, along with our identification of syndecan-1 as an in vivo substrate of MMP-7 , led us to investigate whether GAGs had a direct effect on MMP-7 activity. Indeed, we found that both zymogen activation and activity against specific substrates are markedly enhanced by highly sulfated molecules .
We reported that highly sulfated GAGs, such as heparin, chondroitin-4,6-sulfate and dermatan sulfate, potently augment (> 50-fold) the intermolecular autolytic activation of proMMP-7, as well as the activity of fully active MMP-7 to cleave specific physiological substrates (i.e. pro-α-defensins) . By contrast, heparan sulfate and less sulfated forms of chondroitin sulfate did not promote the activation of proMMP-7 or the activity of the mature proteinase. Surface plasmon resonance demonstrated that proMMP-7 binds with nanomolar affinity to chondroitin-4,6-sulfate, whereas active MMP-7 has a much weaker interaction with this GAG. These observations demonstrate that sulfated GAGs regulate MMP-7 activation and its activity against specific substrates. In addition, we have observed serglycin in the same secretion vesicles containing proMMP-7 (W.C. Parks, unpublished observations), and these findings suggest that its chondroitin-4,6-sulfate chains may function as allosteric, physiological activators of proMMP-7.
Based on our findings, we propose that GAGs serve two broad functions in regulating MMP-7 activity and possibly that of other MMPs. First, they act as allosteric modulators of MMP-7 catalysis, particularly in promoting zymogen activation via autolytic (either inter- or intramolecular) cleavage of the prodomain (Fig. 2A). Second, GAGs provide an anchor on the cell surface, within secretion granules, or the pericellular environment to compartmentalize MMP-7 proteolysis to specific substrates within defined locations. In the case of procryptdin activation, our binding data suggest that GAGs, and specifically chondroitin-4,6-sulfate, can also interact with the substrate, possibly in a trimeric complex (Fig. 2B).
Because MMPs (at least some MMPs) can degrade matrix proteins, it was commonly assumed that these proteases contributed to tumour progression by facilitating metastasis through tissue barriers, and also contributed to diseases characterized by matrix destruction, such as osteoarthritis [10, 13, 81]. As a result, much effort was made to design a means of blocking MMP activity. Essentially all of the inhibitors that were developed, including all of those that made it into phase III clinical trials, bonded to the conserved, structurally similar catalytic site and effectively silenced the activity of all MMPs. Once given to patients with advance stage adenocarcinomas, however, the inhibitors did not improve outcomes, and some made things worse .
What proved to be the Achilles' heal of the MMP inhibitor trials was that these drugs lack specificity [9, 81]. The compounds used block the activity of all MMPs, as well as ADAMs, ADAMTS and other metalloenzymes. Notwithstanding the outcomes of the clinical trials, many groups have shown that specific MMPs are key mediators of specific steps of cancer progression and invasion [8, 13, 81-83]. In a complex, multifaceted setting such as a tumour, multiple cell types (resident, inflammatory, tumour) are present and essentially all of these express different patterns of MMPs. Furthermore, and quite importantly, each of these cells types use their proteinases to either block some processes or facilitate others, and the processes so effected by metalloproteinase activity may contribute to or act against disease progression [9, 10, 81]. MMP-7 provides an example of how a given metalloproteinase can carry out disparate processes in the same setting. Epithelial-derived MMP-7 is required for wound closure and the generation of antimicrobial activity [71, 76], which are clearly desirable outcomes. However, when produced in the injured epithelium, MMP-7 also drives neutrophil activation, which, in excess, causes tissue destruction and, potentially, death [70, 72]. We propose that these distinct MMP-7-mediated processed are spatially controlled by an interaction with different accessory/anchoring factors. Hence, uncovering the identity of the allosteric mediators, as well as the structural details of how they interact with zymogens and active proteinases and their substrates, would lead to new ways of targetting specific processes at the same time as sparing others. Furthermore, exploring metalloproteinase biology with a focus on allosteric anchors will advance our understanding and identification of activation mechanisms and substrates.