Introduction

Humans have not evolved to last long in nature. As the forces of natural selection disappear with the improvement of life and health conditions, average life expectancy has exceeded 80 years in many countries of the world. The concomitant consequence is the soaring of aging-related neurodegenerative disorders that cannot be cured or significantly slowed down, thereby affecting millions of patients and even more caregivers. These disorders generally start with mild symptoms that alter motor, behavioural and cognitive functions. The gradual loss of large amounts of neurons inevitably leads to a worsening of these symptoms, which often end up with a loss of global autonomy before the patient’s death. Neurodegenerative diseases are associated with genetic and environmental stimuli that act in synergy to elicit disease-specific dysfunctions, but also share fundamental commonalities: free radicals generation, mitochondrial dysfunction, immunosenescence, protein misfolding or inflammation. The confluence of these pathogenic driving forces makes it difficult to identify a single cause that could be more easily targeted. It is of the utmost importance to elucidate the underlying mechanisms of the pathology to clearly identify molecular targets and help design effective therapeutic strategies. Among the biofactors that can contribute to neurodegenerative processes, matrix metalloproteinases (MMPs) are of particular interest because: (1) MMPs are expressed by all types of central nervous system (CNS) cells; (2) MMPs expression is modulated by potentially neurotoxic proteins, which can in turn undergo cleavage and functional regulation by these same MMPs; (3) MMPs modulate chronic neuroinflammation through cleavage of inflammatory mediators (i.e., cytokines and chemokines) and tight junction proteins that ensure blood–brain barrier (BBB) impermeability; (4) MMPs control the processing of extracellular matrix (ECM) and transmembrane proteins involved in cell–cell interactions; (5) MMPs can initiate or be part of proteolytic cascades that amplify cellular responses in a web of interactions between numerous enzymes and substrates.

In this review, we will discuss the most relevant knowledge on MMPs and their physiological inhibitors, the tissue inhibitors of metalloproteinases (TIMPs), in four neurodegenerative diseases that strike in mid–late-life: Alzheimer’s disease (AD), Parkinson’s disease (PD), amyotrophic lateral sclerosis (ALS) and Huntington’s disease (HD). We will also discuss to a lesser extent the contribution of two other metalloproteinase subfamilies, which, like MMPs, belong to the superfamily of metzincins: a disintegrin and metalloproteinase (ADAM) and ADAMs with thrombospondin motifs (ADAMTS). While ADAMs are membrane-anchored enzymes, chiefly considered as sheddases of membrane proteins, ADAMTS are mostly extracellular matrix (ECM)-degrading enzymes. ADAMs and ADAMTS in the nervous system are extensively discussed in this CMLS special issue by Hsia and colleagues [1] and elsewhere [2].

Metalloproteinases and natural tissue inhibitors of metalloproteinases

Matrix metalloproteinases

Twenty-four proteinases constitute the MMP family in humans. According to their structure and substrate specificity, this multigenic family have been traditionally classified into five categories: gelatinases (-2 and -9), stromelysins (MMP-3 and -10), collagenases (MMP-1, -8 and -13), membrane-type MMPs (MT-MMPs, MMP-14, -15, -16, -17, -24 and -25) and the other MMPs, e.g., MMP-7, MMP-12 or MMP-28 (see for reviews [2,3,4]). Widely expressed in all human cells, MMPs have significant structural homology and this is one of the main reasons that hinder the development of specific inhibitors against each of them. All MMPs possess a 17–29 amino acid hydrophobic signal peptide that targets the proteinase to the secretory pathway, followed by a pro-domain sequence of 77–87 amino acids that interacts with a conserved cysteine residue and the Zn2+ cation of the catalytic site, thereby maintaining the enzyme under a zymogen inactive state. The catalytic domain of ~ 170 amino acids, harbours a Zn2+ binding consensus sequence HExxHxxGxxH with three histidine residues that confer stability to Zn2+. Redox-mediated break-up of the interaction between the cysteine and the metal cation as well as propeptide removal by autocatalysis or by other proteinases, triggers the conversion of inactive pro-MMPs into active MMPs by a mechanism known as “cysteine switch”. Downstream of the catalytic site, a short hinge region followed by the hemopexin domain (~ 200 amino acids) exhibits a relatively variable amino acid sequence, which is crucial for proteinase-substrate specificity. Of note, MMP-7, MMP-23 and MMP-26 lack the hemopexin domain. In addition to this multidomain common structure, the six members of the MT-MMP subfamily contain membrane anchoring domains; a glycosylphosphatidylinositol in the case of MMP-17 (MT4-MMP) and MMP-25 (MT6-MMP) or a transmembrane domain with a short intracytoplasmic moiety in the case of MMP-14 (MT1-MMP), MMP-15 (MT2-MMP), MMP-16 (MT3-MMP) and MMP-24 (MT5-MMP). Moreover, MT-MMPs, MMP-11, MMP-23 and MMP-28 contain a specific di-basic sequence at the end of the pro-domain, which is recognized by the intracellular Ca2+-dependent serine proteinase furin, leading to activation of these enzymes in the trans-Golgi network [4].

MMPs have been usually considered as ECM-degrading enzymes. This view is now changing with the growing number of non-matrix substrates identified among signalling molecules, trophic factors, receptors, cell adhesion molecules, and even nuclear proteins [5,6,7]. It must be emphasized that MMPs are more and more involved in the modulation of neuroinflammatory processes common to many nervous system pathologies through the regulation of: (1) ECM proteolysis [8]; (2) cell proliferation, differentiation and migration [9,10,11]; (3) activation and inactivation of inflammatory mediators [12]; (4) BBB permeability [13]. If we add that proteolytic-independent effects of MMPs are beginning to be unveiled [14], we can conclude that MMPs represent a paradigm of extraordinary and fascinating functional diversity.

Tissue inhibitors of metalloproteinases

TIMPs are pleiotropic secreted proteins, mostly known for their reversible MMP inhibitory activities. The four TIMPs (TIMP-1 to TIMP-4) share 40% of sequence homology and a backbone with twelve conserved cysteine residues and six disulfide bonds that are essential for their biological activities. Besides the signal peptide, TIMPs structure comprises a N-terminal inhibitory domain of MMPs and a C-terminal domain of interaction. TIMP-1, -2 and -4 are found as soluble or cell surface-associated proteins, whereas TIMP-3 preferentially binds to ECM proteins. In general, TIMPs bind all MMPs and some ADAMs, but only TIMP-3 appears to have inhibitory activity against ADAMTS [15]. The N-terminal domain of TIMPs is sufficient to inhibit MMPs, but in the case of ADAM10, TIMP-1 and TIMP-3 also require the interaction of their C-terminal domain with the proteinase. ADAM10 is not inhibited by TIMP-2 and -4 [16]. The binding affinities between TIMPs and MMPs or their modes of interaction differ significantly. For instance, TIMP-1 inhibits soluble MMPs and ADAM10/17, but it is a very poor inhibitor of MT-MMPs. Also, TIMP-1 is able to specifically bind the hemopexin domain of pro-MMP-9, although the biological function of this complex is unknown [17]. TIMP-2 inhibits both soluble and membrane-bound MMPs. In addition, TIMP-2 specifically binds to the hemopexin domain of pro-MMP-2, paradoxically leading to the activation of the enzyme on the plasma membrane; indeed, the N-terminal domain of TIMP-2 interacts with MT1-MMP, whereas the C-terminal domain binds to the hemopexin domain of pro-MMP-2, thus fixing the pro-MMP-2 on the membrane and allowing its activation by cleavage of the pro-peptide by an adjacent MT1-MMP [18]. TIMP-3 inhibits MMPs, ADAMs and ADAMTS. TIMP-4 has the same inhibitory profile as TIMP-2. All TIMPs are ubiquitously distributed, with the exception of TIMP-4, which is predominantly expressed in heart, ovary and brain [19].

Timp-1, Timp-3 and Timp-4 genes are nested in introns of genes coding for synaptic proteins synapsins and Timp-2 is close to the synapsin IV gene [20,21,22]. The intriguing genetic link between TIMPs and synapsins likely underlines functional relations of these proteins in the nervous system that need to be explored, as many neuropathological conditions show significant alterations of their expression.

Alzheimer’s disease

Alzheimer’s disease (AD) is the most common type of neurodegenerative disorder for which we still lack efficient curing, slowing or preventing treatments. Prototypic symptoms of AD start with short-term mild memory deficits that deteriorate with time, progressively leading to severe memory loss, changes in mood, anxiety, depression, agitation, spatial disorientation, motor impairment, and finally dementia. AD affects 45 million people currently worldwide and this figure is expected to rise to 131 million in 2050 unless efficient treatments are found in the meantime [23]. Only a handful of available drugs can transiently and moderately limit clinical symptoms in the early stages of the disease (https://www.alz.org/alzheimers-dementia/research_progress/treatment-horizon). The difficulty in finding effective treatments for AD is due to a multitude of factors, including:

  • The multifactorial etiology of AD. Familial forms of the disease represent ~ 1% of cases and are caused by mutations in 3 genes: amyloid precursor protein (App), and presenilins 1 (Psen1) and 2 (Psen2). The overwhelming majority of patients suffer from sporadic forms of the disease of unknown etiology, where genetic and environmental risk factors likely synergize with age to prepare a fertile ground for the development of the disease.

  • The difficulty of successfully reproducing valuable preclinical data in human clinical trials.

  • Comorbidity with other disorders/injuries may add confounding factors from an etiological standpoint.

  • A relatively poor knowledge of the pathophysiological mechanisms involved. To date, most research efforts have focused on studying two pathological characteristics of the disease: the accumulation of the amyloid beta peptide (Aβ) resulting from the proteolytic processing of APP, and the hyperphosphorylation of the tau protein (p-tau). Unfortunately, the results of clinical trials targeting Aβ or p-tau have been well below expectations [24,25,26], prompting researchers to discover alternative targets.

  • Clinical trials are conducted on older people after diagnosis, but AD starts 10 or 15 years earlier, meaning that treatments against relevant targets (e.g., Aβ, p-tau) occur too late, when the aging brain has already lost much of its healing ability and resilience.

Despite the diversity of potentially pathogenic mechanisms in AD, metalloproteinases have been mainly involved in the metabolism of APP and Aβ. It is in this context that we will address the following sections.

Main pathogenic pathways linked to APP processing and amyloidogenesis

Almost 30 years after its publication, the amyloid hypothesis of AD remains a valuable tool to study the underlying pathophysiological mechanisms, despite the failure of clinical trials based on its principles. The amyloid hypothesis essentially asserts that excessive accumulation of Aβ is a major pathogenic determinant (see for an updated review [27]). Aβ results from sequential cleavage of APP by two canonical secretases; β-secretase (beta-site APP cleaving enzyme-1, BACE-1) generates a transmembrane fragment of 99 amino acids (C99), which is then cleaved by the γ-secretase complex, thus releasing 4 kDa Aβ monomers, mainly 40 (Aβ40) and 42 (Aβ42) amino acids. Aβ42 aggregates more easily than Aβ40 and it is, therefore, more toxic. Monomers can assemble into oligomers, protofibrils and fibrils, which will eventually give rise to amyloid plaques. Although the latter are a hallmark of AD and provide bona fide post mortem diagnostic, small size oligomers (e.g., dimers, trimers) are increasingly seen as the most toxic forms of Aβ [28]. APP cleavage by β- and γ-secretases occurs following APP endocytosis into the endosomes or during APP targeting to the cell surface through the trans-Golgi network, thereby highlighting the importance of APP cellular trafficking in amyloidogenesis (reviewed in [29]). Aβ accumulation results from a balance between its formation and its degradation/clearance. Several metalloproteinases, including insulin degrading enzyme (IDE), angiotensin converting enzyme, endothelin converting enzyme and neprilysin, play a prime role in Aβ degradation (reviewed in [30]). Moreover, some MMPs (i.e., MMP-2, -3, -9 and MT1-MMP) have also been reported to display Aβ-degrading activity (see [3]). Complementary to proteolytic degradation, the neurovascular hypothesis of AD suggests that tuning down Aβ levels in the brain parenchyma can also be achieved by scavenging systems mostly operating in the BBB microenvironment. Several reports indicate that one of these systems is the apolipoprotein E (APOE) receptor, low-density lipoprotein receptor (LDLR). Thus, overexpression of LDLR in the mouse brain limits amyloid plaque formation and enhances Aβ clearance from brain to blood [31, 32], as well as Aβ uptake and degradation by astrocytes through a mechanism where Aβ directly binds to LDLR, even in the absence of APOE [33]. The low-density lipoprotein receptor-related protein 1 (LRP-1), mainly expressed by glial and endothelial cells, also contributes to Aβ clearance. This is confirmed by experiments in which the decrease/abrogation of LRP-1 expression using antisense oligonucleotides [34] or LRP-1-conditional knockout [35], interferes with Aβ clearance via the BBB, leading to its accumulation in the brain and the consequent learning deficits. Selective knockdown of LRP-1 in astrocytes also reduces Aβ uptake and degradation, which correlates with a downregulation of major Aβ-degrading enzymes MMP-2, MMP-9 and IDE [36]. Unlike, LDLR and LRP-1, the receptor for advanced glycation end products (RAGE) promotes Aβ influx from blood-to-brain (reviewed in [37]). Accordingly, silencing or eliminating RAGE has been shown to preserve BBB integrity in vitro [38] and reduce brain Aβ levels in transgenic AD mice by decreasing β- and γ-secretase activities [39].

It is clear that neurotoxicity derived from APP cleavage can no longer be considered exclusively on the basis of Aβ accumulation. Other APP metabolites generated by different proteinases (e.g., MT-MMPs, δ-secretase) also contribute to Alzheimer’s pathology (reviewed in [40]), paving the way for an assessment of APP breakdown products as components of pathogenic pathways, and therefore, as potential new therapeutic targets.

Non-pathogenic pathways linked to APP processing

In physiological conditions, APP is mainly processed at the plasma membrane level by α-secretases of the ADAM family. This subfamily of metzincins present structural and functional similarities with MMPs (see for review [2]). ADAM10 is the main physiological/constitutive α-secretase [41, 42], while ADAM17 has been identified as responsible of the regulated α-secretase activity [43]. α-secretase cleavage in the middle of the Aβ sequence prevents its formation and concomitantly generates a soluble N-terminal APP fragment (sAPPα) with neurotrophic and neuroprotective properties [44, 45]. Aβ oligomers stimulate transient increase of sAPPα levels and α-secretase activity in cultured cortical neurons, which could be interpreted as a cell self-defence mechanism [46]. While there is little doubt about the neurotoxicity of Aβ accumulation, the physiological roles of the peptide are still poorly understood. Several pieces of evidence converge to show that Aβ is part of the innate immunity arsenal to fight against bacterial, fungal and viral pathogens in the brain [47,48,49,50]. Moreover, physiological Aβ is involved in maintaining basal neurotransmitter release and synaptic activity [51]. It has been shown that picomolar concentrations of Aβ are sufficient to stimulate learning and memory in healthy rodents as well as long-term potentiation (LTP)—a form of synaptic plasticity considered to be a cellular substrate for learning—through the activation of nicotinic receptors [52,53,54]. N-terminal fragments of Aβ found in the cerebrospinal fluid (CSF) of healthy and AD individuals appear to be determinant for these physiological effects, with the YEVHHQ motif located before the α-cleaving site in the Aβ sequence playing a prominent role [55]. It has also been shown that cGMP nucleotide stimulates the release of Aβ by facilitating the approximation in endolysosomal compartments between APP and BACE-1, the main β-secretase. The blockade of Aβ with antibodies in these conditions interferes with cGMP-induced enhancement of LTP and memory [56]. Taken together, these data indicate that the proteolytic regulation of Aβ transcends the idea of its neurotoxicity and raises the question of the potential clinical consequences of lowering Aβ levels below a physiological threshold.

MMPs are mainly considered as Aβ-degrading enzymes, but not only

MMPs with Aβ-degrading activity might be beneficial

Pioneer work in the 90’s linked for the first time AD with MMPs. Tökés and collaborators found upregulated expression levels of inactive pro-MMP-9 in hippocampal neurons of post mortem AD patients. They also demonstrated by mass spectrometry the ability of active MMP-9 to cleave Aβ40 on 3 sites of the membrane-spanning domain, but also at Lys16–Leu17, the so-called α-cleavage site [57]. In the first case, MMP-9 could prevent the formation of the neurotoxic β-sheet. In the second, MMP-9 could act as an α-secretase. Consistent with this idea, MMP-9 overexpressing mice exhibit increased levels of sAPPα associated with enhanced LTP and enhanced learning and memory capabilities [58]. Other authors have corroborated in vitro the ability of MMP-9 to cleave soluble Aβ at the α-site, but the enzyme seems to be more efficient at cleaving in the C-terminus with decreasing processing efficiency towards the N-terminus to generate fragments ending at positions 34, 33, 30, 23, 20 and 16 [59, 60]. In addition, MMP-9 exhibits a rather uncommon feature among Aβ-degrading enzymes because it can cleave preformed Aβ fibrils and compact amyloid plaques, as shown on brain slices from APP/PS1 transgenic mice [59]. This further adds to the possibility that promoting MMP-9 activity may hold therapeutic potential. However, this postulate is subject to caution as it cannot be excluded that proteolytic degradation of plaques could release toxic Aβ assemblages and other neurotoxins contained in their mesh. MMP-2 presents Aβ-degrading profiles roughly similar to those reported for MMP-9, although MMP-2 appears to be more efficient in cleaving Aβ42 [60].

Reactive astrocytes produce MMP-2 and MMP-9 and are systematically associated with Aβ deposits in transgenic mouse models of AD and in post mortem AD brains. In support of a functional link between both observations, the incubation of recombinant Aβ in conditioned media from cultured mouse astrocytes yields breakdown Aβ C-terminal fragments, which can be prevented by selective chemical inhibitors of MMP-2 and MMP-9, and by TIMP-1 [61]. Furthermore, MMP-9 and MMP-2 knockout mice exhibit higher levels of Aβ in the brain compared to wild-type mice, thereby supporting that both MMPs degrade Aβ in vivo [61]. In the same vein, a recent study conducted to investigate the impact of physical activity on Alzheimer’s pathology reported that exercise training in Tg2576 transgenic mice significantly decreases the levels of Aβ40 in the cortex, and Aβ42 in the cortex and hippocampus. Of note, decreased Aβ concentration was concomitant with a rise in the levels of five proteins involved in Aβ clearance, among which MMP-9 [62].

In addition to MMP-2 and MMP-9, other MMPs can also degrade Aβ. For instance, metal ligand clioquinol stimulates Aβ degradation in a cell culture system through a mechanism involving the activation of MMP-3 [63]. MMP-7 was recently reported to cleave Aβ40 and Aβ42, a process that is inhibited by binding of Aβ to Cu2+ [64]. Concerning membrane-anchored MMPs, it was shown that COS cells overexpressing MT1-MMP degrade exogenous soluble Aβ40 and Aβ42, and release 6 different Aβ-derived peptides with cleavage sites mostly located around the α-site, between residues 12 and 18 of the Aβ sequence. Moreover, a soluble variant of MT1-MMP lacking the transmembrane domain was able to degrade in situ parenchymal amyloid plaques in brain slices of Tg2576 mice [65], a feature shared only with MMP-9, neprilysin and IDE among metalloproteinases.

MMPs can be pathogenic in connection with inflammatory processes

Although the degrading action of MMPs on Aβ could in principle be considered beneficial, MMPs are multifunctional enzymes, whose roles may depend not only on the cell types that express them, but also on the substrates involved or the pathological and spatio-temporal context (see for review [2]). In the case of BBB disruption, MMPs have a prominent role as they can proteolyse major components of the basement membranes and tight junctions (see for review [66] in this special CMLS issue). The impermeability of BBB, which ensures selective exchanges between the central and peripheral nervous systems, is compromised in AD patients [67] and in transgenic AD mice [68]. This may contribute to inflammatory and neurodegenerative processes, probably with the participation of resident brain immune cells, as suggested by a study using a co-culture of endothelial cells and astrocytes exposed to Aβ42. In this work, astrocyte-induced BBB opening was related to the activation of MMP-9 and the concomitant decrease in claudin-5, a MMP-9 substrate that contributes to the maintenance of functional tight junctions [69]. Another important source of MMP, reactive microglia, may stimulate Aβ accumulation through a MMP-9-dependent mechanism. In this case, microglial MMP-9 generates C-terminal truncated Aβ fragments that interfere with Aβ clearance across the BBB and promote cerebral aggregation of intracranial-injected Aβ42 in wild-type mice. It is noteworthy that the seeding properties of these C-terminal fragments generated by MMP-9 could, therefore, contribute to the development of sporadic forms of the disease [70]. Recently, another report pointed out that disruption of the blood–CSF barrier 6 h after intracerebral injection of Aβ oligomers was linked to conspicuous inflammation, and was prevented by broad spectrum MMP inhibitor GM6001 and by MMP-3 deficiency in mice, suggesting that the control of MMP activity may have therapeutic potential [71]. Likewise, Aβ42 oligomers dowregulated tight junction scaffold proteins in an in vitro BBB model using bEnd.3 endothelial cells. The weakening effects of oligomers on BBB were blocked by either knocking down RAGE, which was induced by Aβ42, or by GM6001 treatment, which inhibited the activity of concomitantly upregulated MMP-2 and MMP-9 [38]. GM6001 and another MMP inhibitor, minocycline, also proved to efficiently reduce inflammation and oxidative stress associated with cerebral amyloid angiopathy (CAA) in AD mice [72].

Blood-derived immune cells (e.g., macrophages, lymphocytes, granulocytes) are suspected to influence AD pathogenesis, but the extent of their contribution to disease and the underlying mechanisms remain elusive (for review [73]). The question arises as to the role of immune cell MMPs, provided that these cells can reach the brain parenchyma by diapedesis or through a disrupted BBB. Evidences in this direction are rather limited at the moment. The monocytic cell line THP-1 treated with Aβ42 releases more MMP-9, along with tumour necrosis factor (TNF-α), interleukin-1beta (IL-1β) and monocyte chemoattractant protein-1 (MCP-1), compared to untreated cells [74]. Purified neutrophils dramatically increase the production of MMP-9 in response to incubation with highly toxic Aβ25-35 [75]. Interestingly, neutrophils infiltrate the brain of AD patients and are also observed in the brain parenchyma of transgenic AD mice in areas with Aβ deposition [76].

Regulation of MMPs in Alzheimer’s animal models and Alzheimer’s patients

The data above convincingly show that Aβ is a substrate of MMPs. In turn, Aβ modulates the expression of MMPs, probably revealing the existence of feedback regulatory mechanisms. Since early work by Gottschall and collaborators showing that Aβ40 could trigger the expression of MMP-2, -3 and -9 in rat mixed neuronal/astrocyte cultures [77], other authors have confirmed these data. Thus, astrocytes surrounding Aβ plaques in transgenic mice brains show increased levels of MMP-2 and -9 [61]. The increase in the mRNA of both gelatinases is already detected at 4 months of age in the hippocampus of 5xFAD mice, and even earlier (2 months) in some astrocytes close to incipient amyloid plaques [78]. On the other hand, MMP-9 immunostaining is upregulated in the soma and dendrites of cortical neurons in 2-month-old 5xFAD mice [78], indicating altogether that the upregulation of MMP-2 and MMP-9 in Aβ enriched areas starts at pre-symptomatic stages of the pathology and persists over the prodromal-like and symptomatic phases, as they have been defined in the 5xFAD model [79]. Whether this conveys beneficial or detrimental effects cannot be systematically anticipated, as functional duality is a characteristic of both gelatinases. These might be beneficial based on their Aβ-degrading properties, but they might also promote BBB disruption and neuroinflammation [80] or exert direct neurotoxicity in the hippocampus, as reported for MMP-9 [81]. Along this line, AβE22Q and AβL34V mutations associated with hemorrhagic symptoms in early onset CAA stimulate the expression and activation of MMP-2 by endothelial cells in a rather selective manner, which is linked to possible BBB dysfunctions [82]. Moreover, Mizoguchi and collaborators demonstrated that intracerebroventricular (ICV) injections of Aβ upregulated MMP-9 expression in neurons and astrocytes and triggered cognitive deficits. Such deficits were efficiently prevented in mice lacking MMP-9 and also after treatment with chemical MMP inhibitors [83]. In addition to the extensive work on gelatinases, other MMPs have also attracted attention in the field. MT1-MMP content was found increased in neurons and amyloid plaques of 5xFAD mice at 6 months of age—during the symptomatic phase of the pathology—this increase affecting both the active and inactive forms of MT1-MMP [78]. Other authors have shown that MT1-MMP expression is mainly associated with reactive microglial cells in 5xFAD mice [84] and reactive astrocytes in Tg-SwDI transgenic AD mice [65]. Morevoer, Aβ42 increases the mRNA levels of MMP-3, -12 and -13 in a microglial cell line and in primary murine microglia in a PI3/Akt dependent manner [85]. The expression of MMP-1, for which no involvement in APP/Aβ metabolism is known, correlates with neuronal degeneration in the rat hippocampus 1 month after ICV injection of Aβ25-35 [86]. Taken together, these data consistently indicate the upregulation of MMP levels after exposure to exogenous Aβ or in the vicinity of Aβ deposits in transgenic mice, with spatial distributions that may vary depending on the cell type or animal model studied.

Studies in humans mostly reflect the upward trend in MMP levels, although some declines have been noted. Human cerebrovascular smooth muscle cells, constituents of the BBB, show increased expression of MMP-2 and its activating enzyme MT1-MMP after Aβ treatment in culture, which eventually results in increased MMP-2 activity and reduced cell viability [87]. Oligomers of Aβ42 positively regulate the expression of several MMPs (i.e., MMP-1, -3, -9, -10, -12 and -19) in microglia isolated from the brain of AD patients post mortem, compared to controls without dementia [88]. ELISA assay revealed MMP-1 levels significantly elevated by more than 50% in different cortical areas of AD patients [89]. Also in post mortem brain tissue, MT5-MMP immunostaining was detected in amyloid plaques [90] and MT1-MMP in microglial cells neighbouring plaques [84], while MMP-3 was found in both amyloid plaques and the interstitium between myelinated axons and astrocytes, in particular, in the parietal cortex [91]. The concentrations of MMP-3 are higher in the CSF of AD patients compared to healthy controls, in contrast with MMP-9 whose levels are lowered in the AD condition [92]. Similarly, Horstmann and collaborators reported an increase in the levels of MMP-3 in CSF and plasma of AD patients, whereas MMP-2 and MMP-9 decrease in plasma [93]. Interestingly, Lorenzl and collaborators found consistent increased MMP-9 levels in plasma in two studies [94, 95] and another study also showed increased MMP-9/TIMP-1 ratios in CSF from AD patients [96]. Noticeably, this study reported increased MMP-3 and MMP-9 levels, as well as increased MMP-3/TIMP-1 ratio, in the CSF of non-demented individuals carrying AD risk markers (e.g., p-tau, Aβ42 or APOE ε4 allele). A negative correlation was found between plasma levels of MMP-3 and the mini-mental state examination score that assesses cognitive performance [97]. Such negative correlation was also confirmed in the plasma of AD patients for MMP-3 activity measured by casein zymography. Consistent with elevated plasma levels of MMP-3, its substrate gelsolin was decreased in these samples, suggesting that the combined use of gelsolin and MMP-3 could, therefore, hold potential as a plasma biomarker in AD [98]. Even accepting that correlative data do not allow for a causal link between MMP levels and disease progression, it has been proposed that proteolysis limits the amount of circulating Aβ that could reach the brain. This idea is based on the ability of MMPs and ADAMs (e.g., MT1-MMP, MMP-9, ADAM10, ADAM12 and ADAM17) to shed RAGE or LRP-1 [99,100,101,102,103], with the consequent release of the soluble truncated receptors that can bind circulating Aβ; it is estimated that 70% of circulating Aβ40 and 90% of Aβ42 is bound to soluble LRP-1 [104].

Most reports associate MMPs to APP metabolism in AD, however, recent work also unveils the possibility of functional links between MMPs and tau protein. For example, increased levels of active MMP-2 colocalize with p-tau in neurofibrillary tangles (NFT) and dystrophic neurites in the entorhinal cortex of AD patients at early disease stages [105]. The same study reports that MMP-2 can cleave recombinant tau in vitro, but fails to cleave tau contained in NFT, leading to the speculation that inactivation of MMP-2 in NFT might prevent the formation of toxic truncated tau assemblies. MMP-9 and MMP-3 are also described as tau-degrading enzymes, but only MMP-9 can generate fragments prone to generate tau oligomers, which are increasingly considered as the most toxic forms of tau [106]. MMP-3 and MMP-10 levels correlate with tau or p-tau levels in the CSF of AD patients, further stressing the potential interest of MMPs as biomarkers of AD pathology [97, 107].

Overall, these data illustrate a wide range of mechanisms by which MMPs can have beneficial or detrimental effects on AD. Particular mention is given in the following section to MT-MMPs, which are relatively new enzymes in the field of AD and open new avenues for a better understanding of pathophysiology.

Alternative processing of APP by MT-MMPs and contribution to Alzheimer’s pathogenesis

Early work from Higashi and Miyazaki showed that MT1-MMP could process APP outside the Aβ sequence [108]. In this study, MT1-MMP activated by concanavalin A in the HT1080 fibrosarcoma cell line was shown to cleave at the Asn579 and Met580 site of APP770 (VLAN579-M580ISEPR), releasing a soluble truncated APP fragment lacking the MMP-2 inhibitory domain (ISYGNDALMP). This decapeptide was previously reported in a study where MMP-2 was originally identified as a putative α-secretase capable of hydrolizing the Lys16-Leu17 bond within the Aβ sequence [109]. The decapeptide, known as APP-derived peptide inhibitor (APP-IP), specifically inhibits MMP-2 with an IC50 value of 30 nM, well below the 2 µM for MT1-MMP and > 10 µM for MMP-3, -7 and -9. The APP-IP sequence, spanning from amino acids 579–601 (βAPP770 numbering), is present in sAPPα and sAPPβ. The authors concluded that MT1-MMP processing of APP would be a way for the tandem MMP-2/MT1-MMP to regulate ECM proteolysis [108]. Additional work on HEK293T cells demonstrated that MT1-, MT3-, and MT5-MMP could release 5 major APP fragments after being co-expressed with APP770 [110]. In this study, the cleavage sites for MT3-MMP were identified by mass spectrometry at positions Ala463-Met464, Asn579-Met580 (VLAN579-M580ISEPR), His622-Ser623 and His685-Gln686. It is noticeable that cleavage at Asn579-Met580 by MT3-MMP matches the MT1-MMP-mediated cleavage previously identified by Higashi and Miyazaki [108]. It is also interesting to note that although His685-Gln686 cleavage is one residue upstream of the α-cleavage site in the Aβ sequence, MT3-MMP does not modify Aβ levels in HEK293T cells [110]. The authors suggested that MT5-MMP could have cleavage profiles similar to those of MT1- and MT3-MMP, which has been confirmed by a recent study showing that MT5-MMP can effectively cleave the shorter neuronal isoform of APP696 at the VLAN504-M505ISEPR, also known as eta (η) site [111]. According to this study, MT5-MMP (η-secretase) generates a CTF fragment of ~ 30 kDa (CTF-30/η-CTF) that can be further cleaved by α-secretase to produce a peptide (Aη-α, for amyloid eta-alpha) of ~ 12 kDa, which impairs LTP in cultured rat neurons. On the contrary, the concerted action of η- and β-secretase generates the Aη-β peptide, surprisingly lacking neurotoxicity. Together, these data challenge the traditional view of α- and β-secretase as beneficial and detrimental enzymes, respectively. Interestingly, η-cleavage by MT5-MMP is strongly potentiated in the presence of BACE-1 inhibitors, raising concerns about possible side effects mediated by the MMP in the case of therapeutic inhibition of BACE-1. The putative Aη-α peptide was found in the CSF of Alzheimer’s patients and healthy individuals, but the concentrations did not differ between the two groups, making it difficult for the moment to assess the importance of Aη-α in AD pathology. Of note, despite that MT5- and MT1-MMP cleave at the VLAN504-M505ISEPR site, only MT5-MMP and not MT1-MMP was identified in this study as having η-secretase activity in vivo at 10 days postnatal [111].

By the same time, our group tested the implication of MT5-MMP in AD using a different approach; we crossed transgenic 5xFAD mice carrying familial AD mutations in human APP and Psen1 genes [112] with MT5-MMP knockout mice [113]. The resulting bigenic 5xFAD/MT5-MMP-/- strain exhibits dramatic reductions of Aβ burden (e.g., plaques, oligomers, soluble Aβ40 and Aβ42) and C99 levels in cortical and hippocampal regions [114]. The fall in Aβ/C99 levels is concomitant with reduced glial reactivity and reduced levels of IL-1β and TNF-α in different brain areas [114, 115]. Functionally, 5xFAD mice exhibit impaired LTP [116, 117] and impaired learning and memory [118,119,120]. Both deficits are reversed in 5xFAD/MT5-MMP-/- bigenic mice, which show preservation of LTP and olfactory and spatial learning and memory, compared to 5xFAD controls [114, 115]. It is also important to note that the lack of MT5-MMP strongly decreases the release of a soluble N-terminal APP fragment of ~ 95 kDa (sAPP95), complementary of the transmembrane CTF-30 fragment [114], supporting the notion that APP is an in vivo substrate of MT5-MMP. In addition to these in vivo data, MT5-MMP co-immunoprecipitates with APP following MT5-MMP overexpression in HEK293 cells expressing App with the Swedish mutation (HEKswe), resulting in the release of sAPP95 and increases in Aβ40 levels [114, 115]. Although the precise molecular mechanism behind MT5-MMP actions is still elusive, it is known that the proteinase is internalized in the endosomal system [121] and favours APP sorting in endosomes [115], a major site of Aβ formation. In addition, it has been shown that N2a Swedish cells release exosomes containing η-CTF that can be uptaken by neurons [122]. Overall, MT5-MMP could promote trafficking of APP or its metabolites (i.e., CTF-30/η-CTF) into intracellular organelles for more efficient amyloidogenic processing by β- and γ-secretases. This way of cooperative cleavage has been recently shown for asparagine endopeptidase (δ-secretase), whose cleavage of APP upstream the β-cleavage site favours subsequent processing by β-secretase [123]. Not exclusively, it is also conceivable that η-CTF could be used as an exosomal cargo for intercellular propagation (see Fig. 1 for schematic representation of MT5-MMP interactions with APP).

Fig. 1
figure 1

Schematic model summarizing the potential roles of MT1- and MT5-MMP in APP processing and the amyloidogenic pathway. 1: Canonical amyloid precursor protein (APP) and MT-MMPs are targeted from the trans-Golgi network (TGN) to the plasma membrane or directly to the endosomal pathway. 2: On the plasma membrane, MT-MMPs generate a soluble APP fragment of 95 kDa (sAPP95), either alone or in concert with MMP-2, and the residual transmembrane fragment of 30 kDa (CTF-30/η-CTF). 3: Also at the membrane level, APP can be processed by α-secretase through the canonical pathway that generates sAPPα and the transmembrane α-CTF/C83. In addition, the combined action of MT-MMPs and α-secretase could generate a synaptotoxic peptide (Aη-α). 4: MT-MMPs could promote the internalization of CTF-30/η-CTF or unprocessed APP in early endosomes (5), where β-secretase generates sAPPβ, β-CTF/C99 or Aη-β. β-secretase processing of APP may also occur in multivesicular bodies (MVB) (6), followed by γ-secretase cleavage (7) to generate beta amyloid peptide (Aβ) and APP intracellular domain (AICD). The latter may be also generated in the lysosomes (8) and translocated into the nucleus (9). Aβ and sAPPβ can be degraded by lysosomes (10) or released into the extracellular space (11). CTF-30/η-CTF and Aβ can also be released in exosomes (12). Note that generation of η-fragments has only been reported for MT5-MMP so far. The contribution of MMP-2 to sAPP95 production has been described in association with MT1-MMP, and it remains to be determined whether this could also be the case for MT5-MMP. For the sake of simplicity, the cartoon does not show late and recycling endosomes, the possible location of γ-secretase on the plasma membrane or APP trafficking between the endosomal system and the TGN. This scheme includes freely available objects from Servier Medical Art templates (https://smart.servier.com)

Full size image

The structurally close MT1-MMP shares about 60% sequence homology with MT5-MMP. While MT5-MMP is primarily expressed in the nervous system, MT1-MMP is ubiquitously distributed in the organism, including the brain. We found that MT1-MMP expression is upregulated in the brains of 5xFAD mice at symptomatic stages of the pathology and its overexpression in HEKswe cells stimulates C99 and Aβ production [78]. Moreover, MT1-MMP interacts with APP in HEKswe cells, and such interaction generates sAPP95 and the complementary transmembrane CTF-30, eventually leading to twofold increase of Aβ40 levels [124]. Consistent with its prominent β-secretase role, it was shown that BACE-1 inhibition with C3 almost suppresses basal Aβ production. Interestingly, such drastic effect of C3 is prevented in the presence of MT1-MMP, which restores Aβ basal levels, suggesting that MT1-MMP may function as a surrogate β-secretase upon BACE-1 inhibition [124]. The β-secretase-like activity of MT1-MMP may reveal functional redundancies with BACE-1, which could be useful for maintaining the physiological levels of Aβ in situations where, for example, BACE-1 would be therapeutically inhibited. From these studies, it appears that MT5-MMP and MT1-MMP behave differently in response to BACE-1 inhibition; while MT5-MMP would generate synaptotoxic peptides (i.e., Aη-α) leveraging on increased APP bioavailability for α-secretase, MT1-MMP would rather compensate for a physiological deficiency in Aβ. Another interesting feature is that sAPP95 release by MT1-MMP partially involves MMP-2 when activated by MT1-MMP. In the absence of MT1-MMP, MMP-2 loses its ability to generate sAPP95, indicating that MT1-MMP is the step-limiting enzyme [124]. It remains to be elucidated whether MMP-2 can also cleave at VLAN579-M580ISEPR, which would indicate the existence of a consensus cleavage site for APP shared by several MT-MMPs and MMP-2. These data highlight a new function for MMP-2—in tandem with MT1-MMP—in addition to its Aβ-degrading properties, thus providing a new example of the dual/multifunctional nature of MMPs.

Like MT5-MMP, MT1-MMP also favours endosomal APP sorting [124]. Gaining insight into endocytosis and recycling of MT-MMPs between the endosomal system and the membrane may be essential to understand the pathophysiological features of these enzymes. In this context, both MT1-MMP [125] and APP [126] are internalized through a clathrin-dependent mechanism that involves specific short sequences of their cytoplasmic tails. MT5-MMP internalization and recycling is mediated by interactions between the last three residues at carboxyl terminus and Mint3, a protein that contains two type III PDZ domains [121]. Mint3 has also been identified as an adaptor protein that promotes the export of APP from the Golgi complex [127]. Together, these data may reflect interdependent trafficking mechanisms between MT-MMPs and APP that enable functional interactions eventually leading to C99/Aβ production (summarized in Fig. 1).

Overall, MT-MMPs appear as new actors in AD pathogenesis, with multiple influences on pathways where proteolytic and non-proteolytic interactions with APP can coexist. A better understanding of such interactions is needed to shed light on the underlying pathogenic mechanisms.

Singular contributions of ADAMs to Alzheimer’s disease

Although the importance of ADAMs in AD has long been recognized for their contribution to α-secretase activity [128, 129], recent research extends their scope to mechanisms other than those involving a direct cleavage of APP. For instance, it has been shown that ADAM30 stimulates APP sorting to lysosomes and its degradation via the activation of cathepsin D [130]. It is also noteworthy that ADAM10 [131, 132] and ADAM17 [133] can shed the triggering receptor expressed on myeloid cells 2 (TREM2), a protein for which some genetic variants are linked to an increased risk of AD (see [134] for review). Mechanistically, lowering cell surface TREM2 results in reduction of microglia phagocytic activity, thereby possibly contributing to Aβ accumulation [131]. Similarly, the shedding of LRP-1 by different ADAMs (e.g., ADAM12) [99] might affect the ability of this receptor to promote endocytic-mediated clearance of Aβ. Finally, ADAMs could also contribute to the control of chronic inflammation by regulating cytokine activity. This could be the case of ADAM17, which sheds membrane-bound TNF-α and releases its soluble form [135]. Increasing experimental evidence suggests that TNF-α contributes to AD pathogenesis and TNF-α inhibitors, some of which are in clinical trials, may improve the pathological and cognitive outcome (reviewed in [136]).

TIMPs in Alzheimer’s disease

Along with the accumulation of data demonstrating the involvement of metalloproteinases in neurodegenerative processes , the interest in their inhibitors has increased in parallel. The list of endogenous MMP inhibitors and their physiological implications in the nervous system is rather extensive and has already been examined elsewhere [3]. We will focus on TIMPs in the context of this review.

Early work reported the presence of TIMP-1 in neuritic lesions of AD and colocalization with APP, as well as with abnormally phosphorylated and truncated tau [137]. Since then, numerous attempts to describe a clear profile of expression for TIMPs have been undertaken with eventually somehow discordant results. It has been shown that TIMP-1 and TIMP-2 levels remain stable when comparing patients suffering from AD or mild cognitive impairment (MCI) with non-demented controls [92]. While plasma levels of MMP-9 were higher in AD, PD and ALS patients compared to healthy individuals, no changes in TIMP-1 and TIMP-2 levels were observed in the same work [138]. A pilot study on a small cohort of AD patients reported moderately elevated levels of TIMP-1 in plasma [139]. The comparison of different types of dementia reveals that plasma levels of TIMP-1 and TIMP-2 respectively decrease in vascular and frontotemporal dementia, compared to AD and MCI [95]. However, TIMP-2 levels are increased in the CSF of Alzheimer patients [138]. Another study showed increased levels of TIMP-1 in bigenic APPSwDI/NOS2-/- mice compared to APPSwDI controls, along with decreased MMP activity and increased Aβ levels, suggesting that TIMP-1 could negatively influence the pathological outcome as a consequence of lowered nitric oxide production [140]. It has been proposed that the concentration of TIMP-1 in the CSF is a distinctive biomarker to differentiate AD patients from those with subcortical vascular disease, the latter having higher levels than the former [107]. In the same vein, Duits and collaborators reported decreased levels of TIMP-1 and TIMP-2 in the CSF of AD patients with cerebral microbleeds, arguing for an imbalance in favour of MMP proteolytic activity, which could eventually contribute to the generation of such lesions  [141].

TIMP-1 was found to be secreted by neurons in response to MMP-9 release by mixed neuron/astrocyte cultures when exposed to neurotoxic Aβ25-35, and to stimulate astrocyte proliferation [142], in agreement with our previous work [143]. Coincident with gliosis, TIMP-1 mRNA expression is strongly upregulated at early stages (between 4 and 6 months) of the pathology in the brain of 5xFAD mice, whereas TIMP-2 and TIMP-3 mRNA levels show only moderate changes during the same period [78]. It remains to be assessed whether TIMP-1-mediated gliosis represents a self-defence mechanism to eliminate Aβ and/or a consequence of chronic inflammation. Increased levels of TIMP-1, along with the absence of MMP-9, have been observed in microvessels of AD compared to non-demented controls. The authors correlated this observation with the ability of thrombin to induce TIMP-1 expression in cultured brain endothelial cells. Altogether, these data suggest that changes in angiogenesis in AD brains may involve the modulation of the MMP-9/TIMP-1 balance [144].

Although not directly associated with AD, TIMP-2 mRNA and protein have been proposed as biomarkers of aging in the heart and cerebellum [145]. These results are particularly significant in the context of recent findings from Wyss–Coray’s group showing that TIMP-2 is an essential anti-aging factor as it promotes neuroplasticity, including LTP and hippocampal-dependent cognition. These conclusions were drawn after elegant parabiotic experiments in which plasma from young donor mice enriched with TIMP-2 compensated for cognitive deficits in elderly recipient mice, evoking the effect of an “elixir of eternal youth” [146]. With aging being the main risk factor for AD, TIMP-2, as a rejuvenating factor, could certainly play a role in the prevention of AD and other neurodegenerative diseases.

TIMP-3 inhibits APP cleavage by α-secretases ADAM10 and ADAM17, resulting in increased levels of C99 and Aβ, probably due to increased cleavage of APP by β-secretase [147]. The proposed mechanism implies that α-secretase inhibition by TIMP-3 could stimulate APP trafficking to endosomes, where β- and γ-secretases convert APP to Aβ. Consitent with this hypothesis, TIMP-3 levels are upregulated in brains from both AD patients and transgenic AD mice [147]. In addition, TIMP-3 expression is correlated with NFT in AD patients compared to non-demented individuals [148].

TIMP-4 has been less studied in neurodegenerative diseases than the other three TIMPs. Only one study reports that plasma levels of TIMP-4 are higher in AD patients than in healthy controls and TIMP-4 levels positively correlate with the severity of the disease [149]. While we are still waiting for clear evidence of TIMP-4 functions in the CNS, its inhibitory potential on MMPs suggests that it could contribute to modulate the progression of neurodegenerative diseases.

Parkinson’s disease

PD is the second most common neurodegenerative disorder after AD,  characterised by motor symptoms, including rest tremor, bradykinesia, muscle stiffness, impaired posture and balance, and loss of automatic movement. Motor impairment is associated with selective loss of dopaminergic neurons in the substantia nigra and the degeneration of their striatal projections. Although PD is an incurable disease, a number of drugs and surgical interventions have shown to be effective and significantly improve symptoms for a limited period of time. A major feature of PD is the presence of fibrillar intracellular aggregates called Lewy bodies (LBs) in different brain areas. LBs are mainly composed of α-synuclein (α-Syn), a presynaptic protein of 140 amino acids for which three missense mutations (A30P, E46K and A53T) as well as wild-type gene multiplications, have been associated with familial forms of PD [150,151,152,153,154]. α-Syn aggregation has also been reported in sporadic forms of PD of unknown etiology, which constitute the majority of cases [155, 156]. It should be noted that α-Syn neuronal aggregates released into the extracellular space can be internalized by other neurons, reminiscent of a prion-like mechanism that can promote PD progression through the self-propagation of α-Syn [157]. In addition to α-Syn, ninety other molecules have been identified in LBs, including parkin, tau, leucine-rich repeat kinase 2 (LRRK2), DJ-1 and APP [158]. Even though a clear correlation exists between PD and the presence of LBs, the causal role of these aggregates has been questioned by recent studies in which early signs of neuronal dysfunction/death in the substantia nigra precede LBs formation [159]. In this context, other authors have suggested that α-Syn oligomers and protofibrils are true toxic assemblies as opposed to fibrillar structures contained in LBs, which would rather convey neuroprotective roles [158, 160,161,162]. Alternatively, there is also experimental evidence that truncated forms of α-Syn contribute to PD and dementia with Lewy bodies (DLB) pathology. Early work reported the presence of truncated α-Syn in isolated LBs from DLB brains [163] and in transgenic mice harbouring the A53T familial mutation [164]. Subsequently, 3 other C-terminal truncated major peptides (residues 1-78, 1-91 and 1-93) were identified, with the particularity that they were more abundant after digestion of A53T α-Syn compared to wild-type  [165]. When the 1-93 peptide was overexpressed in the mouse substantia nigra it selectively killed dopaminergic neurons independently of LB-like formation. Crowther and colleagues demonstrated in vitro the ability of these truncated α-Syn forms to aggregate into filaments [166], while other authors found that stable expression of C-terminal truncated α-Syn increases neuroblastoma cell vulnerability to oxidative stress [167]. These data clearly indicate that α-Syn fragments contribute to PD and DLB pathogenesis, and several reports involve MMPs in α-Syn truncation. One study showed that, in this order of efficacy, MMP-3, MT1-MMP, MMP-2, MMP-1 and MMP-9 cleaved recombinant α-Syn in vitro [168], a cleavage that was confirmed in another study for MMP-1, MMP-3 and MMP-9 [169]. α-Syn fragments generated by MMP-3 are prone to form amyloidogenic aggregates that exhibit higher cytotoxicity compared to intact α-Syn. Further support for a pathogenic role of MMP-3 in PD comes from reports showing elevated expression of the enzyme in dopaminergic neurons of the substantia nigra in a rat model of PD induced by 6-hydroxydopamine (6-OHDA) [168], as well as in the brains of PD patients [165]. In addition, in another PD model induced by the injection of 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), the production of reactive oxygen species and the degeneration of dopamine neurons are attenuated in MMP-3-deficient mice [170], these effects being associated with the prevention of BBB disruption and the improvement of locomotor activity [171]. Along this line, intracellular activation of MMP-3 triggers neuronal apoptosis [172] and the subsequent release of active MMP-3, which stimulates the production of inflammatory cytokines by neighbouring microglia and eventually the exacerbation of dopaminergic neuron death [173]. Cultured microglia undergoes activation in response to α-Syn by a mechanism that involves the upregulation of MMP expression (i.e., MMP-1, -3, -8 and -9), and the consequent proteolytic activation of protease-activated receptor-1 [174]. Under these conditions, the inhibition of MMP-8 effectively reduces the levels of reactive oxygen species and inflammatory mediators such as TNF-α and IL-1β. MMP-3 implication in PD pathology has also been related to its ability to cleave DJ-1. The latter is a protein deglycase with protective antioxidant properties that mediates pleiotropic cellular responses, and whose mutation is associated with autosomal recessive early-onset PD [175]. MMP-3-mediated cleavage of DJ-1 impairs its antioxidant function in cultured dopamine-producing CATH.a cells upon oxidative stress [176]. The same study provided ground for in vivo cleavage, as the decrease in DJ-1 content observed after MPTP treatment was prevented in MMP-3 knockout mice. Interestingly, one of the DJ-1 mutations (L166P) prevents its dimerization and causes its proteolytic digestion by the proteasome [177, 178]. Also noteworthy, the L1669 mutation abrogates the anti-aggregation properties of wild-type DJ-1 with respect to α-Syn [179]. From these data, it follows that MMP-3 processing of DJ-1 could harm its anti-oxidant and anti-α-Syn aggregation properties. In addition, DJ-1 can be cleaved effectively by MT1-MMP as well as by MMP-1, -2, -8 and -9 [180], indicating that DJ-1 is a substrate of several MMPs and that these interactions may have pathophysiological significance in PD.

The functional and regulatory interplay between various MMPs and PD markers further strengthens the implication of MMPs in PD pathogenesis. Thus, α-Syn stimulates the expression of MMP-9 by cultured microglia and astrocytes [181]. Moreover, α-Syn promotes the expression of MT1-MMP and cell adhesion molecule CD44 in microglial cells. Shedding of CD44 by MT1-MMP facilitates microglia detachment from the ECM and promotes its migration in vitro [182]. Accordingly, siRNA-mediated silencing of MT1-MMP dramatically decreases the migration of the microglial cell line BV-2 when transplanted in the injured substantia nigra of 6-OHDA treated mice, suggesting that MT1-MMP stimulates microglia recruitment into injured brain areas and thus accelerates the degenerative process [182]. Mutations in the LRRK2 gene are the most common cause of autosomal dominantly inherited PD [183, 184]. One study on microglia and leukocytes from LRRK2 mutant mice showed that increased levels of inflammatory mediators and decreased levels of anti-inflammatory progranulin (PGRN) were associated with increased and decreased levels of MMP-2 and MMP-9, respectively. Also interesting, hippocampal neurons of LRRK2 mutant mice show high levels of MMP-12 [185], an enzyme that cleaves microglial PGRN, thereby possibly interfering with its anti-inflammatory properties [186]. Together, these data are particularly sound because microglia is a critical regulator of the inflammatory process that accompanies PD pathogenesis (reviewed in [187]).

In order to implement cell-based neuroprotective strategies against PD, conditioned media from cultured mesenchymal stem cells (MSC) have been used to disaggregate preformed α-Syn fibrils in vitro, with the consequent increase in cell viability. Conditioned media of cells treated with MMP-2 siRNA prevented the beneficial effects of normal conditioned media, suggesting that MMP-2 is one of the soluble factors that confer neuroprotection [188]. In line with these observations, reduced levels of MMP-2 in MPTP-treated mice can be restored by the inoculation of MSC. In addition, injection of MMP-2 in the brain of mice pre-inoculated with α-Syn degrades extracellular α-Syn aggregates and oligomers [188]. Also consistent with these data, MMP-2 levels were decreased in the substantia nigra of post mortem PD brains, while those of TIMP-1 were increased [189]. Whether the MMP/TIMP ratio may be used as a marker of PD progression is still unclear, despite studies associating reduced TIMP-2 levels and increased vulnerability of neuroblastoma cells to MPTP or 6-OHDA, which is consistent with TIMP-2 preventing the effects of these neurotoxins [190]. It appears, therefore, that different MMPs display opposite actions in PD; while MMP-3 cleavage of α-Syn could promote its aggregation into neurotoxic fibrils, MMP-2 could rather degrade them, thus providing neuroprotection. Therefore, the distinction between both MMPs should be considered in prospective anti-PD therapies with MMPs as potential targets in PD. In this line, it has been shown that the inhibition of MMP-8 in the LRRK2G2019S mouse model reduces the levels of inflammatory mediators and improves locomotor activity during systemic inflammatory LPS challenge [191]. Likewise, hydroxamate-based MMP inhibitor Ro28-2653, which targets MMP-2, -9 and MT1-MMP, significantly reduces neuronal loss and MMP-9 levels in the substantia nigra of MPTP-treated mice [192].

There are no reports yet linking ADAM function to PD. However, since the contribution of reactive microglia to pathology is increasingly supported by experimental data, it would be justified to explore in this context the modulation by ADAMs of proteins involved in microglial function, such as TREM2. Of note, some authors have reported association between TREM2 variants and increased risk of PD, although the data are not conclusive for the moment [134].

Amyotrophic lateral sclerosis

Amyotrophic lateral sclerosis (ALS) is a rare rapidly progressive neurodegenerative disease that selectively affects motor neurons in the brain, brainstem and spinal cord, leading to severe disability, muscle atrophy, and also cognitive deficits in about 40% of ALS patients. Poor prognosis is a distinctive clinical feature and life expectancy after the diagnosis is less than 20 months in 50% of patients, who often die from ventilatory failure (see for review [193]). Currently, there are no drugs that can cure or prevent ALS and the rapid course of the disease, as well as the lack of reliable biomarkers, make therapeutic strategies particularly challenging. As with many other degenerative diseases, aging is a major risk factor, with a rapid increase in prevalence after age 55 [193]. While the majority of patients have sporadic ALS of unknown multifactorial origin, genetically linked ALS accounts for between 5 and 10% of cases. Deleterious genetic mutations affect genes coding for proteins implicated in RNA processing, and protein transport and clearance, like C9orf72, SOD1, TDP-43 and FUS (reviewed in [194]).

The pathogenesis of ALS is likely the result of a combination of genetic and environmental factors that may influence key biological processes such as excitatory/inhibitory neurotransmitter balance, BBB integrity, protein folding, oxidative stress or neuroinflammation. All of these events are commonly linked with MMP activity, and a number of evidences discussed below support the role of MMPs in the disease.

Many studies on ALS have used transgenic mice expressing Cu2+/Zn2+ superoxide dismutase (SOD1) carrying the G93A familial mutation (SOD1G93A). In these mice, the involvement of MMP-9 in ALS pathogenesis is still under debate because different laboratories have reported conflicting data. One study shows that MMP-9 deletion in SOD1G93A mice accelerates motor neuron disease and increases cell death [195], whereas another study reports decreased neuron loss and greater survival in MMP-9 knockout mice in correlation with decreased TNF-α and FasL immunostaining [196]. Selective expression of MMP-9 in fast-fatigable motor neurons, which are highly vulnerable in SOD1G93A mice compared to the fast-fatigue resistant and slow motor neuron subtypes, is concomitant with the activation of the unfolded protein response and endoplasmic reticulum stress mark characteristic of early pathological stages of SOD1G93A. Genetic deletion of MMP-9 in these mice, as well as pharmacological inhibition or viral gene therapy, delays muscle denervation [197]. In an attempt to interpret selective motor neuron vulnerability, it has been shown that osteopontin, an ECM protein, is selectively expressed in fast-fatigue resistant and slow motor neurons, and stimulates late expression of MMP-9 in these neuronal subtypes, which could contribute to the second wave of degeneration in ALS by a mechanism controlled by MMP-9 [198]. Another study showed that early administration of MMP inhibitor Ro28-2653 increases the survival of SOD1G93A mice [199], suggesting a powerful detrimental activity of MMPs in ALS. in vitro studies further support the possible contribution of MMP-9 to ALS, as MMP-9 inhibition with glycoursodeoxycholic acid in a cell model of SOD1G93A prevents cell degeneration [200]. Although these data suggest that MMP-9 can be a candidate for therapeutic intervention in ALS, caution is required; indeed, in a sporadic model of ALS using rNLS8 mice, MMP-9 knockdown leads to the preservation of motor neurons and muscle function, even if in the long run the reduction of MMP-9 results in premature death of a subset of rNLS8 mice [201]. It is then possible that targeting MMP-9 outside the neuromotor system may cause adverse effects due to the inhibition of more general physiological processes controlled by this MMP. In addition, MMP-9 could also be considered as an early biomarker of the disease. MMP-9 (and also MMP-2) levels measured by gelatin zymography were found to increase in the serum of SOD1G93A mice at the beginning of the symptomatic phases of the pathology, as opposed to significantly lower levels observed during the pre-symptomatic and late-end stages [202].

Beyond a neurocentric interpretation of ALS pathogenesis, astrocytes could also play a deleterious role, thereby justifying attempts to replace diseased astrocytes by healthy astrocytes. Izrael and colleagues intrathecally injected astrocytes derived from human embryonic stem cells into transgenic SOD1G93A mice and rats, and observed an improvement of motor performance and delayed disease onset linked to astrocytic secretion of TIMP-1 and TIMP-2 [203]. On the contrary, a positive correlation has been put forward between TIMP-3 levels and neurodegeneration. Degenerating neurons of SOD1G93A mice exhibit augmented TIMP-3 levels in parallel with increased interactions between Fas and Fas-associated protein with death domain (FADD). Mechanistically, TIMP-3 inhibition of FADD cleavage by MMP-3 would stabilize Fas–FADD interactions, thereby promoting Fas-mediated apoptosis [204].

Few studies have been conducted in patients to explore possible changes in MMP content. Pioneer work in CNS tissue from ALS patients localized increased immunostaining of MMP-2 in astrocytes and MMP-9 in pyramidal neurons of the motor cortex and in spinal motor neurons [205]. In addition, one study reported increased levels of MMP-9 in plasma of ALS patients but not in CSF [206], while another study also reported an increase in MMP-9 levels in CSF [207]. It has been shown by ELISA assay that the concentration of MT1-MMP, MMP-2 and TIMP-1 increases in the CSF of ALS patients compared to healthy controls [208], thus confirming previous work showing increased TIMP-1 levels in the CSF as well [138].

There is scarce information on ADAMs in ALS. One study described enhanced secretion of TNF-α by microglia from SOD1G93A, SOD1L8Q, or SOD1G10V mutants upon shedding by ADAM10 and ADAM17, thereby contributing to microglia inflammatory response [209]. In the same vein, it was shown that iron accumulation in ventral motor neurons of SOD1G93A mice induced oxidative stress and the secretion of TNF-α mediated by increased ADAM17 activity [210]. It has been suggested that the reduction in ADAMTS-4 observed in the spinal cord of SOD1G93A mice may be an adaptive mechanism to counteract the potential neurotoxic effect of this metalloproteinase [211]. The authors reached this conclusion after observing significant motor neuron death following the injection of recombinant ADAMTS-4 in the spinal cord of SOD1G93A mice. Collectively, different ADAMs could, therefore, contribute to the degenerative process that takes place in ALS.

Huntington’s disease

HD (also known as Huntington’s chorea) is the most frequent among inherited neurodegenerative polyglutamine disorders. It is still an incurable disease that causes changes in mood and mental abilities, as well as deficits in motor coordination and ultimately dementia. HD is not properly speaking an aging-related disease since symptoms may appear at the age of 40 or even earlier. The cause of the disease lies on the mutation of the gene coding for huntingtin (Htt), which presents an expanded CAG triplet repeat in exon 1 that encodes a chain of polyglutamine (polyQ) residues close to the N-terminus of Htt. This results in the aggregation of Htt in the nucleus and cytoplasm of neurons and the consequent cell demise that affects mainly the striatum and cortex [212,213,214]. The underlying neurotoxic mechanism is still elusive, but several lines of evidence point to Htt proteolysis by caspases and calpains, which releases N-terminal small fragments thought to be particularly toxic [215,216,217]. More recently, Miller and collaborators expanded the repertoire of proteinases involved in this process to MMPs [218]. They screened 514 siRNA targeting all the proteinases encoded by the human genome to determine those whose inhibition would reduce Htt processing, and more precisely the production of small N-terminal fragments. Among 11 confirmed proteinases, they found MMP-10, MT1-MMP and MMP-23B. They confirmed the upregulation of MMP-10 and MT1-MMP expression in striatal Hdh111Q/111Q cells as compared to Hdh7Q/7Q cells, and established a good correlation between the knockdown of these MMPs and a significant reduction of caspase activation in Hdh111Q/111Q cells. In the same study it was shown that TIMP-1 and TIMP-3 inhibited Htt cytotoxicity in Hdh111Q/111Q cells. Moreover, striatal MMP activity was also increased in Htt YAC128 (glutamine repeat expansion model) and R/62 (N-terminal fragment model) mouse models of HD. While MT1-MMP and MMP-2 do not directly cleave Htt, MMP-10 generates a N-terminal fragment of 48 kDa after cleavage at residue 402, very close to the N-terminus. Importantly, Hdh111Q/111Q cell death was prevented upon knockdown of MMP-10. Further work from the same laboratory confirmed significant alterations of MMPs expression in induced pluripotent stem cell-derived neural stem cells from HD patients (HD-NSCs) with an expanded CAG repeat, compared to the isogenic corrected control [219]. The study shows that the expression of MMP-3 and -10 increases in HD-NSCs compared to control NSCs, in contrast with the decrease observed in MT1-MMP, TIMP-1 and TIMP-2. TIMP-1 levels were restored by TGF-β treatment, in pace with reduced neurotoxicity of Htt, suggesting overall a beneficial effect of TIMP-1-mediated inhibition of Htt proteolysis by MMP-3 and -10 [219]. However, this idea is questionable, as other authors have reported increased levels of TIMP-1 and -2 in the CSF of HD patients [138].

MMP-9 levels are increased in post mortem tissues of HD patients, as well as the content of pro-inflammatory mediators MCP-1, IL-6, IL-8 and the anti-inflammatory IL-10 [220]. Likewise, plasma MMP-9 and IL-6 levels are higher in HD patients compared to healthy controls and this is also observed in R6/2 HD mice compared to wild-type littermates [221]. Moreover, MMP-3 and MMP-9 levels increase in CSF from HD patients in correlation with disease severity [222]. The same authors showed that cultured microglia from YAC128 mice (transgenic mouse model with a 128-glutamine stretch) exhibits increased secretion of cytokines upon stimulation with MMP-3, an activator of microglia. Interestingly, the pro-inflammatory effect of MMP-3 is equivalent to that caused by LPS treatment, linking the action of MMPs in HD to their pro-inflammatory properties. Seemingly, in a rat model of HD induced by intraperitoneal administration of a natural toxin, 3-Nitropropionic acid, MMP-9 expression and gelatinolytic activity are upregulated in injured striatal blood vessels, suggesting the possible implication of MMP-9 in BBB disruption in HD. Noticeably, MMP-9 expression has been reported to be upregulated in subsets of astrocytes and microglial cells in the vicinity of lesions [223]. In contrast with the aforementioned studies, other authors reported recently in HD patients no changes in CSF MMP-9 levels and the lack of neuroinflammation at premanifest stages of the pathology [224].

With regard to ADAMs, one study reports the physiological interaction between ADAM10 and Htt during neurulation [225]. The authors found that Htt inhibits N-cadherin shedding by ADAM10, which promotes neurulation and rosette formation. They also found that ADAM10 activity increased in the Htt-deficient adult brain, suggesting a role for Htt in synapse plasticity and remodelling. The question arises as to whether deficient regulation of ADAM10/N-cadherin in HD brains with dysfunctional Htt could possibly contribute to the disease.

Concluding remarks and perspectives

The traditional view of metalloproteinases as merely non-specific degrading enzymes is definitely obsolete. Their implications in homeostatic or detrimental pathways in many neurodegenerative diseases mobilize finely tuned proteolytic interactions with substrates and the generation of biologically active cleavage products. A plethora of substrates has been identified and we can anticipate that their number will grow in the coming years in the light of new high-throughput proteomic techniques. Table 1 summarizes, in a non-exhaustive manner, some of principal substrates that have been cited in this review.

Table 1 Substrates that undergo direct or indirect cleavage under the influence of metalloproteinase actions
Full size table

In some cases, metalloproteinases have a dual effect, as it is the case for MT1-MMP in AD experimental models by degrading Aβ and also contributing to its formation. Likewise, α-secretases (e.g., ADAM10) may generate neuroprotective sAPPα and give rise to synaptotoxic C-terminal fragments in synergy with MT5-MMP. We are witnessing the expansion of APP processing beyond prototypical control by canonical secretases. In this scenario, MT-MMPs appear to play an important role in the generation of new truncated APP fragments, whose functions are still poorly understood. Similar challenges arise when considering the physiological and pathological roles of breakdown products resulting from the metalloproteinase-mediated processing of tau, α-Syn or Htt in AD, PD or HD, respectively. Although proteolysis is clearly the main functional characteristic of metalloproteinases, the field now opens up to study proteolysis-independent protein–protein interactions, such as those reported for ADAM30, which bring to light new ways in which metalloproteinases can influence disease outcomes. Therefore, it is plausible that new strategies will emerge based on the modulation non-proteolytic interactions between metalloproteinases and partner proteins. In addition, this could offer alternatives to targeting the proteolytic activity of metalloproteinases, which often suffers from the lack of specificity due to highly conserved catalytic domains between family members.

Inflammation is a common event in neurodegenerative diseases and many metalloproteinases are produced by brain and peripheral immune cells. Gaining insight into the disease will also require a better understanding of how the metalloproteinases released by these cells contribute to pathology in a given spatio-temporal context and/or how metalloproteinases can activate/inactivate immune cells by controlling the cleavage of key receptors and cytokines (e.g., TREM2, TNF-α, IL-1β, etc).

From all the above, it appears that the study of metalloproteinases warrants a more global approach in which we can learn from their conserved functions as well as their specificities. These concepts should inspire future research on neurodegenerative diseases, where a thorough knowledge of biological processes should pave the way for exciting discoveries that will help to design more effective diagnostic and therapeutic strategies.