Heparan sulfate proteoglycans play an important role in the spreading of tau pathology (Figures 1Ai,Bi). HSPGs are proteoglycans characterized by one or more heparan sulfate (HS) groups linked to a protein core. HS is composed of disaccharide chains consisting mainly of glucuronic acid and N-acetyl-D-glucosamine (Sarrazin et al., 2011). HSPGs exist in different classes: transmembrane HSPGs, including glypicans and syndecans; serglycins, found in extracellular vesicles; and secreted HSPGs, such as perlecans and agrins (Sarrazin et al., 2011; Condomitti and de Wit, 2018). HSPGs bind to tau and have been shown to act at all stages of its spreading: secretion into the extracellular space (Zehe et al., 2006; Katsinelos et al., 2018), cellular uptake (Holmes et al., 2013; De La-Rocque et al., 2021) and self-assembly into higher-order states (Zhao et al., 2021). Tau binds to heparin, a densely sulfated form of HSPG, through both the N-terminus and the MTBR (Goedert et al., 1996). The sulfate moieties on HSPGs appear to be crucial for tau binding, requiring both 3-O- and 6-O-sulfation (Zhao et al., 2017, 2020; Stopschinski et al., 2018). Heparin induces a conformational change in the MTBR and its flanking region that exposes previously masked tau phosphorylation sites and can induce oligomerization (Paudel and Li, 1999; Sibille et al., 2006; Elbaum-Garfinkle and Rhoades, 2012). However, heparin and related molecules inhibit tau uptake (Zhao et al., 2017; Weisová et al., 2019; Puangmalai et al., 2020), thus preventing tau fibrils from driving intracellular aggregation (Holmes et al., 2013), as shown by the heparin mimetic F6 (Holmes et al., 2013). These results suggest that an excess of extracellular HSPGs could have similar beneficial effects. However, a synthetic heparinoid with nanomolar affinity for tau failed to show any effect on tau pathology after chronic administration in vivo (Stopschinski et al., 2020). It is important to note that most studies have concluded that cell surface-bound HSPGs are involved in tau transfer without assessing the possible contribution of extracellular HSPGs.

Agrin, a major extracellular HSPG, accumulates in AD brains and more specifically, in NFTs (Verbeek et al., 1999; Kroger and Schroder, 2002; Smith and Hilgenberg, 2002; Del Campo Milan et al., 2015). Since agrin contains protease-inhibiting domains, it has been suggested to protect protein aggregates against extracellular proteolysis, leading to the accumulation of these deposits (Verbeek et al., 1999). In addition, agrin may participate in tangle formation because sulfated glycosaminoglycans have been shown to stimulate tau phosphorylation, thus promoting the formation of paired helical filaments (PHFs) (Goedert et al., 1996; Wang et al., 1996; Hasegawa et al., 1997). However, whether or not agrin displays these activities and how it interacts with tau tangles remains to be elucidated.

The CSPG aggrecan forms perineuronal nets predominantly ensheathing fast spiking interneurons (Morawski et al., 2012; Suttkus et al., 2014; van’t Spijker and Kwok, 2017). These nets were proposed to be neuroprotective against pathological tau, since areas with high densities of perineuronal nets in post-mortem AD brains were largely spared of tangles, even at late disease stages (Brückner et al., 1999, 2008; Diamandis et al., 2000; Morawski et al., 2010a,b). Follow-up studies confirmed that the protective action of perineuronal nets was, in fact, mainly mediated by aggrecan. Most interestingly, aggrecan was shown to generate an external barrier restricting internalization and distribution of exogenous tau in organotypic slices (Nowicka et al., 2009; Suttkus et al., 2014, 2016). Indeed, aggrecan knock-out mice presented elevated tau uptake (Suttkus et al., 2016). Based on these results, it was hypothesized that these CSPGs may act by inhibiting the interaction of tau with HSPGs. However, in a recent study, mice expressing human tau carrying the FTD-linked mutation P301L crossed with heterozygous aggrecan KO mice, displayed changes in the expression and phosphorylation of tau but unaltered distribution of tau aggregates (Schmidt et al., 2021).

Several proteases have been shown to cleave tau (Table 1). Multiple studies have shown elevated levels of several MMPs in tauopathies as well as demonstrated functional links between different MMPs and tau. For example, upregulated MMP-3 levels were found in the cortex of AD-like amyloidosis transgenic rat models (Pentz et al., 2021), and the concentration of MMP-3 and levels of total and phosphorylated tau positively correlate in the cerebrospinal fluid (CSF) of AD patients (Stomrud et al., 2010; Hanzel et al., 2014). Although MMP-9 is enriched in AD patients’ brains and co-localizes with tau, it is sparsely present in extracellular NFTs (Hernandes-Alejandro et al., 2020). Nübling et al. (2012) tested the effect of MMP-3 and MMP-9 on tau oligomer formation and aggregation. While MMP-3 mildly reduced tau aggregation, MMP-9 processing promoted oligomerization (Nübling et al., 2012; Wang et al., 2014). Multiple potential MMP-3 cleavage sites were identified within the MTBR of tau (Nübling et al., 2012), suggesting this protease could inhibit tau aggregation by degrading regions crucial for oligomer formation. In contrast, MMP-9 cleavage sites were mainly located either in the N-terminal region or close to the C-terminus (Nübling et al., 2012), thus sparing the MTBR and facilitating the generation of tau oligomers. More recently, docking simulations have predicted that a high-affinity complex can be formed between MMP-9 and full-length tau (Hernandes-Alejandro et al., 2020). This binding involves the catalytic domain of MMP-9, suggesting that this interaction could be initiated when MMP-9 is active. Interestingly, MMP-9 can be directly activated by MMP-3 (Ogata et al., 1992; Okada et al., 1992; Shapiro et al., 1995), implying that elevated MMP-3 levels might result in increased MMP-9 activity, indirectly facilitating tau aggregation. MMP-2 also has the capacity to cleave recombinant tau in vitro but fails to process its hyper-phosphorylated forms in NFTs (Terni and Ferrer, 2015). It is thus possible that the accumulation of MMP-2 around NFTs found in the entorhinal cortex at the early stages of AD (Terni and Ferrer, 2015) is a potential response aimed at eliminating the production of toxic fragments in AD brains. The cleavage sites of MMP-2 on tau, however, are yet to be determined. In conclusion, MMP-2, MMP-3 and MMP-9 show differential actions on tau aggregating behavior, suggesting their differential contribution to tau pathology.

Thrombin was also found to cleave tau at multiple arginine and lysine residues in the MTBR and proline-rich domains (Arai et al., 2005; Quinn et al., 2018; Zhang et al., 2021). In AD brains, thrombin was found to be upregulated and co-localized with amyloid plaques, microglia and NFTs (Arai et al., 2006). However, PHFs of tau extracted from AD brains were more resistant to thrombin cleavage compared to dephosphorylated PHFs (Arai et al., 2005). The seeding potential of tau fragments produced by thrombin is largely unknown, but tau 125–230 generated by thrombin-mediated cleavage at Gln124-Ala125 was shown to be non-toxic (Garg et al., 2011). This site is also cleaved by calpain-1 and calpain-2 (Garg et al., 2011). However, it is important to note that thrombin-cleaved tau fragments are yet to be identified in tauopathy brains, thereby questioning the pathophysiological relevance of this process.

Several cathepsins have been shown to interact with and modulate tau spreading. As previously mentioned, these proteases are mainly localized to lysosomes, but they can also be secreted in the extracellular space (Boonen et al., 2016; Vidak et al., 2019; Niemeyer et al., 2020). Whilst the majority of data on the action of cathepsins on tau mostly reflects their activity in lysosomes, independent experiments have been conducted either in vitro using recombinant proteins or using cathepsin knockout animals. Therefore, a role for extracellular cathepsins cannot be ruled out. Cathepsins B, D, L, and S have all been proposed to cleave tau. Although cathepsin B accumulates in close proximity to NFTs and amyloid plaques (Ii et al., 1993), there is no direct evidence that this protease can cleave tau. In contrast, cathepsin D cleaves recombinant tau at several sites, mostly sparing the MTBR (Kenessey et al., 1997), and as such these fragments could retain the ability to generate higher-order PHFs. At least one of these fragments was produced at neutral pH in vitro, suggesting that this cleavage could also occur in the extracellular space (Kenessey et al., 1997). Furthermore, cathepsin D upregulation was detected in an aging Drosophila model of tauopathy. The same group observed enhanced tau-induced neurodegeneration in cathepsin D-deficient flies which suggests that this protease may have a neuroprotective effect (Khurana et al., 2010). Cathepsin L has also been found to cleave tau at neutral pH. This activity targets sites within the MTBR, generating several fragments, such as tau 258–360 and tau 258–363, which enhance the aggregation of the full-length protein (García-Sierra et al., 2008; Wang et al., 2009; Zhang et al., 2021). Cathepsin S has been seen to associate with NFTs and its levels are elevated in the brain of AD patients (Lemere et al., 1995; Munger et al., 1995; Nübling et al., 2017). Treatment with cathepsin S in vitro yields a distinct cleavage pattern of tau, in which the MTBR seems to remain intact, allowing its association to NFTs. However, this study did not specify whether cleavage occurred at an acidic or neutral pH (Nübling et al., 2017). It is important to note that cathepsin-cleaved tau fragments have yet to be identified in AD brains, highlighting the need for further investigations to unravel the pathophysiological relevance of this protease in AD.

Calpains are proteases mainly found in the cell cytosol (Goll et al., 2003), albeit secretion of calpain has also been observed (Frangié et al., 2006; Letavernier et al., 2012; McDougall et al., 2021), including from isolated rat brain synaptosomes (Pestereva et al., 2021). Calpain has been detected in the CSF, though it remains unclear whether its presence is due to leakage from dying cells (Laske et al., 2015). This dual localization makes investigations on the role of calpain in the context of tau spreading and seeding challenging and as such, the action of extracellular calpains has not been directly investigated as yet. Both calpain-1 and -2 cleave tau (Chesser et al., 2013), and play opposing roles in inducing neurodegeneration (Baudry and Bi, 2016). Calpain-1 activity is significantly upregulated in AD cortical brain tissue from Braak stage III (Kurbatskaya et al., 2016). Much of the interest surrounding the role of calpain in tau pathology has centered on the 17 kDa N-terminal to mid-domain tau fragment (45–230), which is generated through cleavage by calpain-1 at Lys44-Glu45 (Yang and Ksiezak-Reding, 1995) and by calpain-1 (Park and Ferreira, 2005) or -2 (Garg et al., 2011) at Arg230-Thr231. These events have been observed in vitro, suggesting calpain-1 and -2 could also exert their activity in the extracellular space, although direct evidence is still lacking. This fragment has been observed in patients affected by AD and other tauopathies, such as progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD) (Ferreira and Bigio, 2011; Garg et al., 2011). However, the specific effect of this tau fragment remains controversial as it was found to induce apoptosis (Park and Ferreira, 2005; Amadoro et al., 2006; Park et al., 2007; Reinecke et al., 2011; Lang et al., 2014) or morphological changes in cell lines and primary neurons (Chen et al., 2021), whereas other groups have observed no such alterations (Garg et al., 2011). Nonetheless, both Drosophila and mouse models overexpressing tau 45–230 showed accelerated hippocampal degeneration and synaptic defects (Reinecke et al., 2011; Lang et al., 2014). However, when this 17 kDa fragment was incubated with full-length tau, it significantly lowered aggregate formation, suggesting a possible protective effect (Ferreira and Bigio, 2011). Fascinatingly, aggregation and phosphorylation of full-length tau protected it from calpain cleavage (Ferreira and Bigio, 2011), which is in line with previous data indicating aggregated tau is less susceptible to protease cleavage (Yang and Ksiezak-Reding, 1995; Yang et al., 1997). In a recent study investigating stroke biomarkers, this 17 kDa tau fragment accumulated in primary neurons and their media upon hypoxic treatment, and was decreased by calpain inhibitors (Chen et al., 2021). These findings suggest that calpain-mediated cleavage is specific and can be initiated under different pathological conditions. Similarly, the AD-relevant fragment, tau 26–230, is generated by calpain-1 (Park and Ferreira, 2005) and calpain-2 (Garg et al., 2011) cleavage at Arg230-Thr231. In contrast, Matsumoto and colleagues (Matsumoto et al., 2015) have demonstrated that tau is cleaved in vitro by calpain-1 at Arg242-Lys243 producing a 24 kDa C-terminal fragment lacking the N-terminal projection domain (aa 243–441; CTF24). This truncated tau accelerated heparin-induced aggregation and was unable to support microtubule assembly. Furthermore, CTF24 efficiently propagated to other tau-expressing cells, where it displayed higher aggregation and seeding activity than full-length tau (Matsumoto et al., 2015). Interestingly, active calpain-2 co-localizes with tau filaments in AD, Down syndrome and FTD brains (Adamec et al., 2002a,b). A recent study by Cicognola et al. (2020) identified that calpain-2, but not calpain-1, cleaves tau at Lys224, generating an N-terminal fragment previously found to be enriched in CSF in tauopathies (Cicognola et al., 2019). Knockdown of the calpain-2 catalytic subunit gene caused a significant reduction of this N-244 tau fragment in cell-conditioned media (Cicognola et al., 2019). Overall, calpain cleavage appears to promote tau aggregation, thus potentially enhancing its spreading.

In C17.2 mouse-derived neural stem cells, internalized α-synuclein fibrils colocalize with HSPGs (Holmes et al., 2013). Similar to tau, α-synuclein endocytosis in cell lines is inhibited by co-application of heparin in a dose-dependent manner (Holmes et al., 2013; Ihse et al., 2017). Heparin acts competitively by binding α-synuclein at sites responsible for its interaction with HSPGs, thus blocking its internalization. However, heparin has been found to be less efficient in blocking internalization of monomeric and oligomeric forms of α-synuclein (Ihse et al., 2017). A neuroblastoma cell line exposed to heparin lyases I, II, and III showed reduced endocytosis of α-synuclein fibrils and similarly, cells lacking enzymes responsible for HPSGs biogenesis fail to internalize α-synuclein (Ihse et al., 2017). Additionally, HSPG sulfation was found to be important, as treatment with chlorate, which inhibits sulfation, reduces α-synuclein fibril internalization (Holmes et al., 2013; Hudák et al., 2019). Further analyses on the sulfation requirements of HSPGs was carried out using differentially sulfated heparin. When applied to C17.2 cells, 2-O, 6-O and N-desulphated heparin showed lower efficiency in inhibiting α-synuclein uptake and seeded aggregation compared to standard heparin. Shorter heparin chains are also less efficient in inhibiting α-synuclein uptake and seeding. Accordingly, a CRISPR/Cas9 genetic screen in HEK293 cells identified EXT1, 2, and 3, which mediate the initiation and elongation of the glycosaminoglycan chain in HSPGs, to be involved in α-synuclein uptake and seeding. Similarly, NDTS1, which encodes an enzyme responsible for N-deacetylation and N-sulfation of HSPGs, also appeared to play a role, as its knockout reduced uptake and seeding of α-synuclein. Although the ablation of HS6ST2, an enzyme involved in the 2-O sulfation of HSPGs, did not show overt effects on α-synuclein fibril uptake, its overexpression decreased α-synuclein internalization but increased seeding. This suggests a more complex role of 2-O sulfation on α-synuclein spreading (Stopschinski et al., 2018). Zhang et al. (2020) recently analyzed the interaction between α-synuclein and HSPGs by molecular modeling, suggesting that α-synuclein fibrils display more stable binding to HSPGs compared to its monomeric or dimeric forms, possibly explaining the stronger inhibitory effect of heparin on fibril internalization (Zhang et al., 2020).

In a follow up study, treatment of neuroblastoma cells or primary neurons with heparinase failed to inhibit internalization of N-terminal acetylated α-synuclein monomers or fibrils. Since this modification is present at high levels in vivo (Burré et al., 2013), this result questions the physiological importance of HSPGs in this process. In contrast, peptide-N-glycosidase F (PNGase F), which cleaves complex N-linked glycans from glycoproteins, reduced endocytosis of both fibrillar and monomeric N-acetylated α-synuclein but did not affect non-acetylated α-synuclein. Interestingly, acetylated α-synuclein was found to interact with glycans in the absence of their protein core (Birol et al., 2019). Of note, surface membrane proteins as well as secreted proteins and ECM components are heavily glycosylated (Scott and Panin, 2014), creating an ideal multivalent binding environment for pathological α-synuclein. These data indicate that HSPGs are fundamental players in the internalization of α-synuclein fibrils.

CSPGs have also been implicated in α-synuclein aggregation and spreading (Lehri-Boufala et al., 2015; Mehra et al., 2018). Accordingly, incubation of α-synuclein with chondroitin sulfate A and B in vitro enhances the formation of aggregates able to enter SH-SY5Y cells (Mehra et al., 2018). Interestingly, chondroitin sulfate, and glycosaminoglycans in general, inhibit cathepsin D, suggesting that the endocytosis of an α-synuclein-chondroitin sulfate complex could induce higher seeding effects due to lysosomal inhibition (Lehri-Boufala et al., 2015). However, co-injection of α-synuclein aggregates with chondroitinase in mice did not change the spread of pathology. On the other hand, degradation of hyaluronan, another major component of the ECM, reduced α-synuclein pathology in the same model, although the mechanism at the basis of this effect is currently unclear (Soria et al., 2020).

It is generally assumed that fragments of α-synuclein must contain the NAC region to propagate pathology. In vitro studies using recombinant α-synuclein found that negative charges at the α-synuclein C-terminus counteracts the aggregation propensity of the NAC domain (Izawa et al., 2012). As such, cleavage of the C-terminus by proteases increases aggregation and seeding (Sorrentino et al., 2018, 2020; Chakroun et al., 2020; Sorrentino and Giasson, 2020). Interestingly, α-synuclein carrying PD-linked mutations is more efficiently cleaved at the C-terminus than the wildtype protein (Li et al., 2005).

Conversely, the effect of N-terminal truncation is less clear. Most studies report no change (Luk et al., 2009; Volpicelli-Daley et al., 2011) or a decrease (Luk et al., 2016) in seeded aggregation, even though removal of the first two apolipoprotein binding motifs can lead to an increase (Kessler et al., 2003; Sorrentino and Giasson, 2020). Increased propagation of α-synuclein pathology after brain injection of recombinant α-synuclein lacking the first 10 or 30 residues, compared to full length fibrils has been observed (Terada et al., 2018), suggesting that the N-terminus of α-synuclein can also modulate aggregation. Therefore, the action of extracellular proteases has the potential to both negatively and positively modulate α-synuclein propagation.

Different MMPs such as MMP-1, -2, -3, -9, and -14 cleave α-synuclein, with MMP-3 proven to be the most effective (Levin et al., 2009). In vitro tests have shown cleavage at multiple sites at the N-terminus of the NAC (Sung et al., 2005). These cleavage products were observed in the extracellular media of neuroblastoma SK-N-BE cells overexpressing human α-synuclein. MMP-3-cleaved α-synuclein had an elevated propensity to aggregate in vitro and displayed increased toxicity compared to full length α-synuclein when applied to SK-N-BE cells (Sung et al., 2005). Interestingly, MMP-3 is elevated in rat brains exposed to 6-hydroxydopamine or 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine, two classical models of parkinsonism (Sung et al., 2005; Leem et al., 2020). Accordingly, MMP-3 localization in Lewy bodies was detected in the substantia nigra of PD brains (Choi et al., 2011). Choi et al. (2011) investigated the MMP-3-dependent cleavage of α-synuclein harboring mutations associated with PD, and observed that in vitro, the A53T mutant is processed more efficiently than the wildtype and A30P mutant proteins, especially at sites 78–79 and 91–93. In contrast, the overexpression of MMP-3 and α-synuclein in COS cells produced lower levels of insoluble α-synuclein aggregates. However, overexpression of the A53T-containing 1–93 α-synuclein fragment in mice induced increased toxicity and formation of Lewy body-like structures in dopaminergic neurons of the substantia nigra, both at the site of injection of the α-synuclein-encoding adeno-associated virus and contralateral side (Choi et al., 2011). These results imply that the 1–93 fragment, which contains the full NAC domain and lacks the C-terminus, is prone to aggregation and spreading. Nevertheless, this is not the only fragment produced by MMP-3 cleavage and the overall effect of MMP-3-cleaved α-synuclein has yet to be conclusively evaluated in vivo.

MMP-2 has also been suggested to degrade α-synuclein fibrils. Injection of MMP-2 into animals inoculated with α-synuclein in the cortex and striatum led to reduced levels of insoluble and oligomeric α-synuclein. In addition, human α-synuclein positivity was limited to the injection site in MMP-2 treated animals, suggesting reduced spreading (Oh et al., 2017).

Plasmin is a serine protease that degrades fibrin blood clots, and is also involved in inflammation, collagenase activation and synaptic plasticity (Pang et al., 2004). Application of recombinant plasmin to α-synuclein monomers, oligomers or fibrils resulted in their degradation, including when they harbored several PD-linked mutations, including A53T. Interestingly, α-synuclein found in the media of SH-SY5Y cells was also cleavable by plasmin at different sites spanning the N-terminal domain and the NAC. Using a propagation model utilizing SH-SY5Y cells expressing α-synuclein, co-cultured with the microglia-like cell line BV2 lacking α-synuclein, Kim et al. (2012) showed that exogenous plasmin added to the media could reduce spreading between these two cell types. However, further studies are necessary to confirm the importance of plasmin cleavage on α-synuclein aggregation and spreading in vivo.

α-Synuclein has been found to be degraded by neurosin, also called kallikrein 6. Similar to other members of the kallikrein family, neurosin is a secreted trypsin-like serine protease that is activated extracellularly by sequential cleavage (Yoon et al., 2008) and can be found in human CSF (Diamandis et al., 2000). In vitro treatment of recombinant α-synuclein with neurosin generates several fragments by cleavage within and in the proximity of the NAC domain. Neurosin digestion was found to inhibit α-synuclein aggregation in vitro (Iwata et al., 2003), and treatment of α-synuclein oligomers with neurosin induced their almost complete degradation (Spencer et al., 2013). Pro-aggregation variants such as α-synuclein phosphorylated at Ser129 (Kasai et al., 2008) or carrying the PD-linked mutations A30P, A53T or E46K showed reduced cleavage by neurosin (Iwata et al., 2003; Kasai et al., 2008; Spencer et al., 2013). Of note, neurosin levels are lower in brains of patients affected by dementia with Lewy bodies and in α-synuclein transgenic mouse models (Spencer et al., 2013). Furthermore, overexpression of neurosin in HEK293 cells or in cortical neurons induced degradation of α-synuclein in the culture media, whereas primary neurons from neurosin knockout mice showed increased α-synuclein internalization and aggregation (Tatebe et al., 2010; Pampalakis et al., 2017). In addition, neurosin can also activate proMMP-2 (Pampalakis et al., 2017) and an unidentified extracellular protease (Ximerakis et al., 2014), further promoting α-synuclein degradation. Although it is unclear whether neurosin acts directly or through activation of a downstream protease, its activity appears to strongly reduce both aggregated and monomeric forms of α-synuclein.

Cathepsins B, D, E, G, K, L, S and V have been shown to cleave α-synuclein, both in its monomeric and fibrillar forms. In particular, both cathepsins L and K were shown to completely ablate α-synuclein fibrils, whereas cathepsins B, D, E, G, S and V generate small α-synuclein fragments (McGlinchey and Lee, 2015; McGlinchey et al., 2017, 2020). Importantly, degradation of α-synuclein by cathepsins L and K was achieved upon long incubation (16 h) and at an acidic pH, which would be typically found in lysosomes. Shorter incubation times or treatment at a neutral pH mimicking that of the extracellular space, generated fragments truncated at the N- and C-termini (McGlinchey et al., 2017, 2020). As discussed above, these fragments display an increased aggregation propensity, albeit this was not tested experimentally. Cathepsin B was found to cleave both α-synuclein monomers and fibrils in vitro upon incubation at low pH. It is thus unclear whether this activity could be retained at neutral pH in the extracellular space (McGlinchey and Lee, 2015; Tsujimura et al., 2015). C-terminal cleaved fragments were also observed upon addition of recombinant cathepsin B to lysates of 3D5 cells expressing human α-synuclein, or to mouse and human brain extracts (Sevlever et al., 2008; Tsujimura et al., 2015). Cathepsin D was also found to cleave recombinant α-synuclein in vitro at residues 116, 124 and 132, thus potentially generating aggregating fragments (Hossain et al., 2001; Sevlever et al., 2008). Since the proteolytic activity of both cathepsin B and D on α-synuclein was only observed at acidic pH, it is questionable whether these events occur in the extracellular space. However, cathepsin D knockout mice display insoluble α-synuclein in brain extracts, even in the absence of overexpression. A similar result was observed in human post-mortem brains affected by mutations in the CTSD gene (Cullen et al., 2009). Accordingly, knockdown of the C. elegans Ctsd ortholog caused increased aggregation of overexpressed human α-synuclein, whereas overexpression of Ctsd in α-synuclein expressing worms increased survival of DA neurons, an effect that was not present upon expression of cathepsins B and L (Qiao et al., 2008).

The effects of calpain-1 on α-synuclein have been studied in detail. However, similar to the action of calpain on tau, its activity on α-synuclein has been assumed to occur intracellularly. Calpain cleaves monomeric α-synuclein mostly within the N-terminal or central region (Mishizen-Eberz et al., 2003; Greenbaum et al., 2005; Kim H. J. et al., 2006; Diepenbroek et al., 2014). The aggregation potential of monomeric α-synuclein treated with calpain is still controversial (Mishizen-Eberz et al., 2005; Dufty et al., 2007). In vitro, calpain-1 cleaves both wildtype and PD mutant forms of fibrillar α-synuclein within the C-terminus (Mishizen-Eberz et al., 2003; Diepenbroek et al., 2014), which may increase the aggregating potential of α-synuclein. Diepenbroek and colleagues evaluated the role of calpain in vivo by crossing mice expressing human α-synuclein carrying the A30P mutation with a calpastatin knockout or overexpressing mouse model. Overexpression of calpastatin, which acts as a calpain inhibitor, reduced α-synuclein aggregates in the brain, whereas calpastatin knockout mice showed the opposite effect (Diepenbroek et al., 2014). Similar findings were obtained by treating mice expressing human wildtype α-synuclein with the calpain inhibitors gabadur and neurodur. This treatment decreased insoluble α-synuclein aggregates with a concomitant reduction in gliosis, neurodegeneration and hyperactivity (Hassen et al., 2018). Therefore, calpain processing of α-synuclein appears to increase its aggregation potential.

Truncated huntingtin species are generated through cleavage by different proteases that can be found in the extracellular space such as MMPs, calpains and cathepsins (Table 1). MMP-10, -14 and -23B were identified in an shRNA-based knockdown screen in HEK293 cells looking for a reduction in huntingtin fragments. Knockdown or pharmacological inhibition of these MMPs reduced toxicity in a striatal cell line expressing huntingtin with 111 glutamines, as well as in Drosophila lines expressing the first 336 amino acids of the human huntingtin protein with 128 glutamines. MMP-10 was confirmed to cleave recombinant huntingtin both in vitro and in cell lysates, mostly producing a 45 kDa fragment which could also be observed in brain extracts from HD patients. MMP-10 and -14 activity was increased in a striatal cell line expressing huntingtin with 111 glutamines compared to a non-pathogenic variant with only 7 (Miller et al., 2010). Furthermore, increased MMP-2, -3 and -10 activity was observed in neural stem cells from HD patients compared with their isogenic lines, where the polyQ expansion was corrected. In contrast to this, the activities of MMP-9 and -14 were found to be reduced in HD (Naphade et al., 2017).

Calpain also mediates proteolytic processing of huntingtin. Application of recombinant calpain-1 to recombinant huntingtin or to mouse brain extracts caused the appearance of fragments with molecular weights between 45 and 72 kDa, which could also be observed in post-mortem brain samples of both healthy and HD-affected subjects (Kim et al., 2001, 2003; Gafni and Ellerby, 2002; Lunkes et al., 2002). However, high-polyQ huntingtin appeared to be more resistant to calpain cleavage compared to lower repeat mutants in vitro (Gafni and Ellerby, 2002). Calpain-resistant variants showed reduced aggregation and toxicity in HEK293T cells compared to calpain-cleavable huntingtin (Gafni et al., 2004). In addition, calpain-derived fragments have been observed both in mouse models expressing human huntingtin with 150 polyQ (Gafni et al., 2004; Landles et al., 2010), and in post-mortem caudate samples from HD patients which also presented increased calpain levels (Gafni and Ellerby, 2002; Gafni et al., 2004). These results were confirmed in Drosophila, where knocking out calpain reverted the toxic effects of huntingtin overexpression, although this effect appeared to be due to the intracellular activity of calpain, since a double knockout for calpain and the autophagic protein Atg8 failed to show the same rescue effects (Menzies et al., 2015). Furthermore, crossing calpastatin knockout mice with HD mice with 111 polyQ showed increased production of huntingtin N-terminal fragments and subsequent aggregation (Weber et al., 2018). Similar to tau and α-synuclein, a direct demonstration of calpain-mediated cleavage of huntingtin in the extracellular space is still lacking.

Cathepsins also cleave huntingtin. In particular, cathepsins B, D and L were found to generate two 55–60 kDa fragments when incubated with striatal cell lysates. Interestingly, cathepsin D only generated an N-terminal fragment from wildtype huntingtin, whereas up to four different fragments were detected when huntingtin containing 100 polyQ was exposed to cathepsin D. This cleavage was carried out at a neutral pH, suggesting that it could occur in the extracellular space (Kim J. et al., 2006).

A class of extracellular proteins which is found to play extensive roles in Aβ uptake, aggregation and deposition is the HSPG family (Figures 1Aiii,Biii). Amyloid plaques were first found to contain HSPGs by Snow and colleagues (Snow et al., 1988), followed by the observation that there is a direct interaction between HSPGs and both APP and Aβ (Narindrasorasak et al., 1991; Brunden et al., 1993). Cell lines deficient in HS biosynthesis, or treated with heparin, show a decrease in the binding of Aβ peptides to the cell membrane and subsequent internalization (Kanekiyo et al., 2011). An AD mouse model with a neuron-specific impairment in HS biosynthesis exhibited decreased plaque burden and Aβ oligomerization (Liu et al., 2016). Interestingly, the low-density lipoprotein-related receptor (LRP1)-dependent endocytosis of Aβ42 relies on HSPGs, since the addition of heparin abrogated the increase in Aβ42 uptake observed by the overexpression of an LRP1 mini-receptor (Kanekiyo et al., 2011). In conjunction with this, the Bu group has observed that ApoE-immunoisolated extracellular vesicles, derived from astrocytes, suppress Aβ binding and uptake in mouse cortical neurons, an effect abrogated by heparin (Fu et al., 2016). These findings suggest that HSPGs are involved in Aβ42 internalization. Although mostly localized at the plasma membrane, syndecans and glypican-1 show extensive co-localization with extracellular amyloid plaques in the brain, with the neuron-specific syndecan-3 being most effective at increasing Aβ42 uptake (van Horssen et al., 2002). By virtue of their HS chains, syndecan-3 and -4 also trigger formation of Aβ42 fibrillar assemblies, underscoring their relevance in plaque formation (Letoha et al., 2019). Glypican-1 interacts with higher-order Aβ structures and reduces SH-SY5Y viability when overexpressing APP (Watanabe et al., 2004). However, its function in plaque seeding or aggregation is currently not known.

Apart from their roles in mediating cell surface binding and endocytosis of APP/Aβ, HSPGs, along with CSPGs, also play a role in clearance and degradation of Aβ peptides (Gupta-Bansal et al., 1995). While post-mortem AD brain samples contain higher than normal levels of HSPGs, clearance of soluble Aβ in the mouse hippocampus is increased upon HSPG depletion (Liu et al., 2016). It is currently postulated that HSPGs function to promote Aβ fibrillation or act as protective chaperones, thus inhibiting Aβ degradation (van Horssen et al., 2003). Both agrin and perlecan, like other HSPGs, bind to Aβ40 fibrils through glycosaminoglycan side chains, protecting them from degradation and accelerating the fibrillation of monomeric Aβ40 (Castillo et al., 1997; Cotman et al., 2000). This role of agrin was recapitulated in an AD mouse model crossed with a conditional Agrn KO line where deletion was driven by an endothelial cell-specific promoter. These mice exhibited increased Aβ generation and deposition (Rauch et al., 2011). However, this effect was not observed when neuronal agrin was deleted, pointing to a non-cell autonomous effect on plaque formation.

Proteolytic degradation contributes to the regulation of extracellular Aβ levels and deposition in amyloid plaques. Several extracellular proteases have been implicated in the degradation of Aβ40/42, including neprilysin (NEP) and the insulin-degrading enzyme (IDE) (Table 1; Saido and Leissring, 2012).

Neprilysin is a zinc metalloprotease with broad substrate specificity and cell surface localization. Demonstrating an inverse correlation with age, it is reported to cleave only Aβ monomers along the axon and at synaptic sites, and has negligible enzymatic activity for higher-order Aβ structures (Fukami et al., 2002; Saido and Leissring, 2012). Consistent with its role in the reduction of Aβ42 levels, brain tissues of AD patients exhibit a decrease in NEP expression in AD-vulnerable areas, including the hippocampus, cortex and temporal gyrus (Yasojima et al., 2001). The genetic deletion of NEP or its pharmacological inhibition results in a two-fold increase in Aβ levels in the brain as well as increased hippocampal plaque burden (Marr et al., 2004; Farris et al., 2007). Conversely, overexpression of neuronal NEP results in reduced Aβ levels and deposition, along with a reduction in associated pathology (Leissring et al., 2003). Additionally, its homolog NEP2 also contributes to decreasing Aβ levels (Saido and Leissring, 2012).

The zinc metalloprotease IDE also displays a broad distribution in the extracellular space, cytosol and mitochondria. Similar to NEP, IDE catalyzes the degradation of monomeric Aβ in the brain and its ablation causes an increase in Aβ levels in primary neurons as well as in vivo (Farris et al., 2003). Other extracellular proteases processing Aβ include MMP-2, MMP-9, and MMP-14, cathepsin B and plasmin (Saido and Leissring, 2012; Hernandez-Guillamon et al., 2015). Interestingly, these enzymes are capable of catabolizing Aβ fibrils, in contrast to NEP and IDE (Saido and Leissring, 2012). Their investigation in vivo remains limited, though cathepsin B has been observed in amyloid plaques (Mueller-Steiner et al., 2006).

In contrast to HSPGs and extracellular proteases, the role of ECM core proteins in Aβ production, seeding or plaque deposition is less clear. Several ECM proteins, including laminin, collagens and fibronectin co-localize with senile plaques in AD brains (Perlmutter et al., 1991). Proteomic profiling of AD hippocampi has revealed an upregulation of several ECM proteins during all disease stages (Hondius et al., 2016). However, whether these observations are causative, or correlative of amyloid formation, is currently unknown.

Laminin1 binds to soluble Aβ40 through its α-chain and inhibits its fibrillogenesis in a time- and dose-dependent manner, thus influencing the survival of cortical neurons (Drouet et al., 1999). While it increases the amyloidogenic fragment production, laminin1 also promotes Aβ40 depolymerization when incubated with pre-formed fibrils in vitro (Bronfman et al., 1996; Castillo et al., 2000).

Total collagen levels are upregulated in AD; however, the roles of individual isoforms are currently unclear. The Mucke group has shown that synthetic Aβ42 oligomers induce the transcription of the Col6a1 gene in hippocampal and cortical neurons and that collagen VI decreases the interaction of Aβ42 oligomers with the neuronal cell surface, ultimately leading to a reduction in Aβ42 neurotoxicity (Cheng et al., 2009).

In addition, when the 7PA2 cell model of AD is treated with peptides derived from elastin, it shows an elevation in Aβ40 and Aβ42 production. This effect was dependent on the length of the individual peptides, since longer peptides were more potent in Aβ generation and led to AD-related behavior in mice. Significantly, these changes were pinned down to an increased expression of β- and γ-secretase mRNAs (Ma et al., 2019).