Targeting Aggrecanases for Osteoarthritis Therapy: From Zinc Chelation to Exosite Inhibition

Osteoarthritis (OA) is the most common degenerative joint disease. In 1999, two members of the A Disintegrin and Metalloproteinase with Thrombospondin Motifs (ADAMTS) family of metalloproteinases, ADAMTS4 and ADAMTS5, or aggrecanases, were identified as the enzymes responsible for aggrecan degradation in cartilage. The first aggrecanase inhibitors targeted the active site by chelation of the catalytic zinc ion. Due to the generally disappointing performance of zinc-chelating inhibitors in preclinical and clinical studies, inhibition strategies tried to move away from the active-site zinc in order to improve selectivity. Exosite inhibitors bind to proteoglycan-binding residues present on the aggrecanase ancillary domains (called exosites). While exosite inhibitors are generally more selective than zinc-chelating inhibitors, they are still far from fulfilling their potential, partly due to a lack of structural and functional data on aggrecanase exosites. Filling this gap will inform the design of novel potent, selective aggrecanase inhibitors.


Aggrecanases as Targets in OA.
Osteoarthritis (OA) is the most common chronic degenerative joint disease, representing a leading cause of years lived with disability worldwide. 1 This places a large socio-economic burden on healthcare systems, with estimated medical costs between 1 and 2.5% of the gross domestic product in high-income countries. 2 OA affects predominantly the knee, hip, and hand joints. 1,3 In severely affected OA patients, joint replacement surgery is the only viable option, although is not a risk-free option. 4 Pharmacological treatment for symptomatic OA is largely palliative, being limited to steroidal and non-steroidal anti-inflammatory drugs (NSAIDs), which are unable to alter disease progression. 5 NSAIDs have also raised safety concerns, especially considering long-term administration on an aged population with multiple co-morbidities such as cardiovascular diseases, diabetes, and obesity. 6 No drugs able to slow down or halt the progression of OA, i.e., disease-modifying OA drugs (DMOADs), are currently available, and this led the U.S. Food and Drug Administration (FDA) in 2018 to label OA as a "serious disease with an unmet medical need". 7 This is not the result of a lack of efforts from pharmaceutical companies and academic institutions�quite the contrary.
OA is a complex multifactorial disease whose pathogenetic mechanisms are still not completely understood. Some promising DMOADs under development target cartilage degradation, a major hallmark of OA. 6,8,9 Since articular cartilage allows for low-friction movement between bones, its erosion is a major cause of impaired mobility and pain. Articular cartilage is composed by chondrocytes embedded in an extracellular matrix (ECM) rich in collagens (of which types II, VI, and XII are the most abundant) and proteoglycans such as aggrecan and, in low amounts, biglycan. 10 Collagens provide the tissue with tensile strength, whereas aggrecan provides compressibility through its ability to regulate osmotic pressure via the Donnan effect. 11,12 This function of aggrecan is mediated by the negatively charged glycosaminoglycan (GAG) chains attached to its protein core, which attract counterions from the interstitial fluid filling the cartilage pores. Not surprisingly, net loss of both collagens and aggrecan has a devastating effect on cartilage integrity, the latter representing an early, reversible phase of the dysregulated ECM catabolism which is typical of OA. 13,14 Perhaps because of the early failure of collagenase inhibitors in cancer clinical trials, 15 exploration of this class of molecules as DMOADs has been limited. Poor selectivity, lack of efficacy, and musculoskeletal (MSK) adverse effects such as joint stiffness and pain hampered further applications of matrix metalloproteinase inhibitors (MMPs), the class of ECM proteases endowed with collagenase activity. For example, a phase II clinical trial for knee OA with the MMP inhibitor PG-116800, developed by Procter & Gamble, was terminated due to an increased frequency of adverse MSK effects such as arthralgia (ClinicalTrials.gov Identifier: NCT00041756). 16 At a time when research on collagenase inhibitors was stalling, two distinct aggrecanase activities were isolated and identified as members of A Disintegrin and Metalloproteinase with Thrombospondin Motifs (ADAMTS) family of metalloproteinases: aggrecanase-1 (ADAMTS4) 17 and aggrecanase-2 (ADAMTS5, originally named ADAMTS11). 18 Since then, four lines of evidence have supported the choice of ADAMTS5 as a favored target in OA: 19 (1) ADAMTS5 is the most potent proteoglycanase in vitro; 20−22 (2) in contrast with Adamts4 knockout mice, 23 Adamts5 knockout mice showed protection in inflammatory or surgical OA models; 24,25 (3) anti-ADAMTS5 monoclonal antibodies (mAbs) effectively inhibited aggrecan degradation in human ex vivo OA models; 26−28 (4) ADAMTS5 accumulation is sufficient to lead to aggrecan degradation in human chondrocyte monolayer cultures. 29 Both genetic ablation 30 and selective inhibition 31 of ADAMTS5 in mice also reduced OA-related pain sensitization (allodynia); thus, ADAMTS5 inhibitors may show additional analgesic effects.
Notwithstanding the prominent role of ADAMTS5 in OA pathology, simultaneous inhibition of ADAMTS4 may not be undesirable as an OA treatment, given that ADAMTS4 expression is consistently upregulated under inflammatory conditions. 19,32 Provided that both side and off-target effects are carefully evaluated, inhibitors targeting both aggrecanases may exhibit a competitive advantage over those selectively directed against just one of them.
Aggrecanase inhibitors can be classified in two groups on the basis of their mechanism of inhibition, i.e., zinc-chelating inhibitors and exosite inhibitors ( Figure 1). Zinc-chelating inhibitors comprise mostly synthetic, low-molecular-weight molecules as well as the endogenous aggrecanase inhibitor Tissue Inhibitor of Metalloproteinase 3 (TIMP3) and its engineered variants. Exosite inhibitors interact with noncatalytic residues involved in substrate recognition and cleavage, i.e., exosites. These are defined as small clusters of non-adjacent residues in the ADAMTS ancillary domains, which are poorly conserved between the different ADAMTS family members. 22 Exosite inhibitors comprise sulfated GAGs, glycoconjugates, flavonoids, nucleic acids, peptides, and monoclonal antibodies (mAbs). Because of their ability to target non-conserved residues, exosite inhibitors are expected to be more selective than zinc-chelating inhibitors.
A comparison of zinc-chelating versus exosite inhibitors is instrumental in informing the development of potent, selective aggrecanase inhibitors. Here, we review the current literature on aggrecanase inhibitors as of April 2022. Data were obtained from different sources, including PubMed, the clinical trial database (www.clinicaltrials.gov), patents, company web sites, and abstracts from international congresses. We focus on molecules that act by directly inhibiting ADAMTS4 and -5, while molecules that interfere with their post-transcriptional regulation, such as 2-(8-methoxy-2-methyl-4-oxoquinolin-1(4H)-yl)-N-(3-methoxyphenyl)acetamide 33 and small interfering RNAs, 34,35 are outside the scope of this review. We initially present the data currently available on the structures of ADAMTS4 and ADAMTS5 and then proceed to a detailed comparison of zinc-chelating and exosite inhibitors by highlighting advantages and drawbacks of the two approaches.

Fold and Functions of Aggrecanase Domains.
ADAMTS4 and ADAMTS5 belong to family M12 in clan MA of the metallopeptidases. Proteases in clan MA are collectively called "metzincins", due to the presence of a conserved signature composed of a zinc-chelating sequence (HEXXHXXG/NXXH/D) followed C-terminally by a methionine residue. 36 Other protease families in clan MA comprise the above-mentioned MMPs and A Disintegrin and Metalloproteinases (ADAMs), the latter including only transmembrane members.
The domain composition of ADAMTS4 and ADAMTS5 consists of a signal peptide, a prodomain, a metalloproteinase catalytic domain (Mp), followed by non-catalytic ancillary domains such as a disintegrin-like (Dis) domain, a central thrombospondin-type I motif (TS-1), a cysteine-rich (CysR) domain, and a spacer (Sp) domain. ADAMTS5 displays an additional TS-1 motif at the C terminus.
Both aggrecanases are expressed as inactive zymogens with a large prodomain (161 and 245 residues in ADAMTS4 and ADAMTS5, respectively) necessary to maintain latency. The mechanism behind the inhibitory function of the prodomain has not been elucidated. In related MMPs, the S γ of a conserved cysteine residue within the sequence PRCGVPD coordinates the active-site zinc, 37 but it is not known if this "cysteine switch" mechanism is also present in aggrecanases. Both prodomains contain a sequence with low homology ( 192 PMCNVKAP 199 and 207 ASCETPAS 214 in ADAMTS4 and ADAMTS5, respectively) to the MMP sequence. Unfortunately, no structure is available for the prodomain of aggrecanases, and AlphaFold predicts the structure of this domain with very low confidence (per-residue confidence score <50; IDs: AF-O75173-F1 and AF-Q9UNA0-F1 for ADAMTS4 and ADAMTS5, respectively). To the best of our knowledge, no mutations of the cysteine residues in the putative cysteine-switch sequences have been reported. ADAMTS4 and ADAMTS5 activation requires proteolytic removal of the prodomain by proprotein convertases such as furin and PACE4, 38 Figure 2). The Mp domain (residues 213−428 and 262−476 in ADAMTS4 and ADAMTS5, respectively) is characterized by the α/β structure typical of clan MA, with a central core of five-stranded β-sheet; four long strands are in parallel (I, II, III, and V), and a short fifth (IV) is in antiparallel configuration. The β-sheet is surrounded by α-helices A, A1, B,   (PDB 2RJP) complexed with compound 1; the β strand and α helix of the active site are shown in light pink, the S1′ loop in red. (B) Crystal structure of ADAMTS5 (PDB 2RJQ) in complex with Batimastat. The S1′ loop is shown in red. Zinc and calcium ion are colored in gray and in orange, respectively. C, C1, and D. While helices A and C are common to those of other MMP and ADAM structures, helix B is typical of ADAMTS4 and ADAMTS5. 41,42 The Mp domain contains the active site, a cleft parallel to helix C where a catalytic Zn 2+ ion is coordinated by three conserved His residues (ADAMTS4: His361, His365, His371; ADAMTS5: His410, His414, His420) ( Figure 3). In addition to the Zn 2+ ion, the Mp domain contains two or three Ca 2+ ions; in ADAMTS4 two Ca 2+ ions are located in a Ca-cluster flanked by disulfide bridges (Figure 2A). This Ca 2+ -cluster site is another unique aspect of ADAMTSs with respect to MMPs.
When a zinc-binding inhibitor is bound to the active site, ADAMTS4 and ADAMTS5 display a similar shape of the subsites (S1, S1′, S2, S2′, and S3, S3′, according to the Schechter and Berger nomenclature). 47 A comparison of the two Mp domains with those of MMPs suggests that the major differences are located around the S2′ and S1′ pockets. Compared to MMPs, the S2′ pocket is smaller and characterized by a unique motif sequence ( 322 CGVSTCDT 329 and 371 CGHHSCDT 378 for ADAMTS4 and ADAMTS5, respectively). The lipophilic S1′ pocket, formed by the base of strand IV, a part of helix C and an adjacent loop (amino acids 389−403 and 437−451 in ADAMTS4 and ADAMTS5, respectively), is able to assume different conformations based on the inhibitor bound ( Figure 3). Even if the active site is highly conserved in the two aggrecanases, the presence of four different residues (Ala252, Val390, Met395, and Val398 in ADAMTS4 compared to Leu301, Leu438, Leu443, and Ile446 in ADAMTS5) leads to a larger S1′ pocket in ADAMTS4. For this reason, inhibitors with bulky P1′ groups usually possess greater inhibitory activity against ADAMTS4 than ADAMTS5 (see section 2).
While the structures of ADAMTS4 and ADAMTS5 in complex with zinc-chelating hydroxamate inhibitors are very similar to each other, the uninhibited form of ADAMTS4 has a different conformation, with the carboxylic group of Asp328 in the S2′ loop coordinating the Zn 2+ ion ( Figure 4A). The global shift of the S2′ loop toward the active site in ADAMTS4 suggests an auto-inhibitory mechanism not present in MMPs, where the active site is wholly exposed in absence of any ligand, or in other known ADAMTS structures. In ADAMTS13, the best characterized ADAMTS family member, a different auto-inhibitory mechanism is in place. 48 Here, Asp181 in the S2′ loop does not interact with the catalytic zinc ( Figure 4B); instead, a non-proteolytically competent conformation is guaranteed by a "gatekeeper triad" of charged residues (Arg193, Asp217, and Asp252) that, through a hydrogen bond network, occlude the catalytic cleft. 48 A superimposition between the crystal structure of the ADAMTS4 Mp domain in its free (PDB 3B2Z) and inhibited (PDB 2RJP) forms and that of free ADAMTS13 (PDB 6QIG) shows that the conformation of the S2′ loop in the presence of hydroxamate inhibitor 1 is the one that more closely resembles ADAMTS13 ( Figure 4B).
Downstream from the zinc-binding sequence, there is also the methionine residue (Met369 and Met439 in ADAMTS4 and ADAMTS5, respectively) of the "Met-turn", a topological constraint conserved in metzincins which is required for the structural integrity of the zinc-binding site. 49 In both ADAMTS4 and ADAMTS5, the Dis domain adopts a common fold characterized by two α-helixes and two βsheets connected by several loops (Figure 2). Despite its name, this region shows no structural homology to the Dis domain typical of the disintegrins present in viper venoms and instead resembles the CysR of ADAMs. 50 The smallest recombinant fragment with detectable proteoglycanase activity consists of the Mp/Dis domains, 20,22 suggesting that these two domains compose a structural as well as functional unit. Mutagenesis studies followed by functional assays using truncated versican as a substrate identified two adjacent lysine residues ( 532 KK 533 ) in ADAMTS5 as an exosite ( Figure 5A). 51 It is not known if the homologous sequence in ADAMTS4 ( 485 KH 486 ) also represents an exosite.
As mentioned above, the structures of the domains Cterminal to the Dis have not been reported, but their fold has been predicted with reasonable confidence by AlphaFold (AF-O75173-F1 and AF-Q9UNA0-F1) ( Figure 5A).
The CysR (residues 576−685 in ADAMTS4 and 623−731 in ADAMTS5) contains 10 cysteine residues ( Figure 5B). According to the AlphaFold model (ID: AF-Q9UNA0-F1), in ADAMTS5 the CysR contains three antiparallel β-sheets and one α-helix ( Figure 5C), an arrangement which seems to be preserved in ADAMTS4. The solved crystal structure of ADAMTS13 CysR (PDB 6QIG), on the other hand, is quite divergent ( Figure 5D). Since deletion of the CysR severely reduced both aggrecanase 20 and versicanase activity, 22 this domain is likely to be involved in substrate recognition, as shown for ADAMTS13, 48 but no specific exosites have been identified so far. In ADAMTS13, the CysR contains a small hydrophobic exosite ( 472 AAV 474 ) ( Figure 5D) 52 which is not conserved between the two aggrecanases, being replaced by more hydrophilic residues ( Figure 5B).
The Sp (residues 686−837 and 732−874 in ADAMTS4 and ADAMTS5, respectively) is essentially cysteine-free and Q76LX8, residues 440−556). Alignment was performed in ClustalOmega (https://www.ebi.ac.uk/Tools/msa/clustalo/) and visualized using MView (https://www.ebi.ac.uk/Tools/msa/mview/). In ADAMTS13, the vWF-binding exosite 472 AAV 474 [52] is highlighted. Boxes indicate amino acids conserved in at least two of the three proteases and are colored according both amino acid identity and physicochemical properties (red, positively charged; purple, negatively charged; green, apolar; cyan and pink, polar), while cysteine residues are in yellow. consists of 10 β-strands in a jelly-roll topology ( Figure 5E). While residues in the beta-strands are conserved between ADAMTS4 and ADAMTS5, those in the interconnecting loops are not, 22 as shown by a superimposition of the Sp domains of ADAMTS4 and -5 (as predicted by AlphaFold) with those of ADAMTS13 (resolved by X-rays) (Uniprot ID: 3GHM and 6QIG) ( Figure 5F). 48,53 This suggests that the overall fold of the Sp domain is conserved among the three family members, whereas the exposed loops contain substratespecific exosites which can be exploited for selective inhibition. That this is indeed the case was demonstrated when loops β1-β2, β9-β10, and β3-β4 in ADAMTS4 and ADAMTS5 were swapped with those of ADAMTS13, which is unable to cleave proteoglycans. 22 Two of the resulting chimeras showed a severe reduction in versicanase activity: the exosites comprised residues 717−724 and 788−795 in ADAMTS4 (loops β3-β4 and β9-β10) and 739−744 and 837−844 in ADAMTS5 (loops β1-β2 and β9-β10). Importantly, these exosites were involved in cleavage of both versican and aggrecan (at least in the case of ADAMTS5), suggesting similarities in substrate recognition between these two proteoglycans. From these studies we can conclude that a general feature of aggrecanase exosites is a preference for hydrophilic, positively charged residues (Table  1).
Overall, the structural and functional data summarized in this section highlight the presence of distinct differences between ADAMTS4 and ADAMTS5, in particular in exosite preferences (Table 1), as well as between aggrecanases and other metalloproteinases in clan MA such as MMPs and ADAMs, that can be leveraged to achieve highly selective aggrecanase inhibitors.
1.3. Targeting Aggrecanases: Zinc Chelation versus Exosite Inhibition. As described in the previous section, the geometry of the active site, in particular that of the histidine triad coordinating the catalytic zinc, is widely conserved in metalloproteinase clan MA, while the enzyme subsites represent specificity determinants among the different members of this superfamily. Accordingly, the selectivity of an active-site inhibitor is determined by its ability to establish interactions with the enzyme subsites. If the affinity for the zinc ion is the driving force in the binding energy between enzyme and inhibitor, as is the case for hydroxamate-and carboxylatebased inhibitors, finely tuning selectivity is a daunting task.
Exosite inhibitors offer a solution to the selectivity issue by targeting highly divergent sequences. Small-molecule exosite inhibitors may suffer from their limited contact area (on average 1000 Å 2 ) 55 and therefore may show limited affinity/ inhibitory potency for their target protease if the exosite is relatively extended. As a comparison, complexes between ligands and exosites in thrombin span from 300 to 1700 Å 2 . 56 Macromolecular inhibitors are characterized by much larger contact areas (1500−3000 Å 2 ) 55 and therefore are ideally suited to target exosites. Not all non-zinc chelating inhibitors are exosite inhibitors (since they may target subsites in the aggrecanase Mp domain), but all exosite inhibitors act via a non-zinc binding mechanism (since they target substratebinding residues in the ancillary domains).
Compared to other protease families, 56 identification of exosites in the ADAMTS family is still at its infancy. So far, only in the case of ADAMTS13, have the ancillary domains been structurally resolved. 48,53 From a practical point of view, this means that rational designing of exosite inhibitors for aggrecanases have been virtually non-existent; instead, exosite inhibitors have been identified by structure−activity relationship (SAR) 51 or by relying on alternative technology platforms, such as phage display, 27 that are able to probe the 3D landscape of the target enzyme by screening large libraries of molecules. De novo protein structure prediction with AlphaFold 54 can inform the design/in silico screening of exosite inhibitors if the exosite sequences are functionally validated, for example with a quantitative substrate cleavage assay. Assays employing native or full-length substrates are ideally suited to identify exosite inhibitors, which may not be identified when short peptide substrates are used; at the same time, such assays more closely reflect the inhibitory potency of the molecule under physiological conditions, although an important caveat here is that it is very difficult to estimate physiological protein concentrations, in particular for ECM substrates such as proteoglycans. Remarkably, the distinction presented here between active-site inhibitors versus exosite inhibitors supersedes the classical classification into competitive versus non-competitive inhibitors which is substratedependent (i.e., the mechanism of inhibition may be different if a either a peptide or protein substrate is used in the assay).

ZINC-CHELATING INHIBITORS
The high structural homology among the Mp domains of clan MA metalloproteinases is one of the factors that have hampered the development of selective aggrecanase inhibitors. Nevertheless, as discussed in the previous section, some specific structural features such as the shape of S1′ specificity pocket or the conformation of S2′ loop, offer some opportunity for the design of small molecules with a biased if not selective inhibitory profile.
The classical approach to design metzincin inhibitors relied on the use of zinc metal chelating groups such as hydroxamates and carboxylates. As a result, inhibitors with activity in the nanomolar and picomolar ranges have been identified. Unfortunately, often these molecules were broad-spectrum inhibitors, active also against MMPs and ADAMs, and responsible for off-target toxicity.
Zinc-chelating inhibitors of aggrecanases can be classified as either small molecules or endogenous protein inhibitors such as TIMP3 (Figure 1).

Small-Molecule Inhibitors.
The first aggrecanase inhibitors were inspired by the classical structure of metzincin inhibitors, constituted by an aromatic backbone, able to interact with the S1′ and/or S2′ pockets of the enzyme, and a zinc-binding group (ZBG) able to coordinate the catalytic zinc ion. The most used ZBG is the hydroxamic acid. The high affinity for the catalytic zinc (up to picomolar) combined with the conserved geometry of the active site in clan MA of metalloproteinases, often results in a poor selectivity of zincbinding inhibitors. For example, GM6001 (Ilomastat) is also a  Figure 6) as an inhibitor of partially purified aggrecanase activity using a structure-based approach. 58 The succinate-derived peptidomimetic structure of 2 was inspired by substrate specificity of MMP8 which is endowed with limited aggrecanolytic activity. 59 The introduction of a Tyr residue in P1′ position of the peptide hydroxamate scaffold and the shift of the pseudotyrosine hydroxyl group from para to meta position improved the inhibitor potency as well as selectivity over MMPs. Moreover, in P2′ position a rigid structure was introduced in compound 3, resulting in increased potency and selectivity over MMP8. Minor modifications of the P1 side chain also affected selectivity. Compounds 2 and 3 showed good inhibitory potency against isolated ADAMTS4 and -5; in particular, compound 3 displayed lower IC 50 values than 2 ( Figure 6). 60 Crystal structures of compounds 2 and 3 in complex with the ADAMTS5 Mp domain showed that the ligands bound to the active site in a similar manner. 60 The hydroxamate group coordinated the catalytic Zn 2+ in a standard geometry, thus orienting the phenolic ring into the small S1′ pocket and locating the 2-indanol ring in a specific position further stabilized by several hydrogen bonds ( Figure 7). This conformation may justify the selectivity profile of compounds 2 and 3 ( Figure 6). The higher inhibitory potency of inhibitor 3 could be explained by an additional hydrogen bond between the −NH group of the cyclopropyl-N-methyl methanamine chain and a water molecule connected to Thr378 (Figure 7).
Since replacing the aromatic ring with a biphenyl moiety in P1 did not result in any improvement in activity and selectivity profiles, 61 Cherney et al. inserted cyclic P1 groups, identifying the N-methansulfonyl piperidine 4 ( Figure 6) as the most potent aggrecanase inhibitor of the series with selectivity over MMP9. 62 The first sulfonamido-based aggrecanase inhibitors containing a pipecolic scaffold were reported by Noe et al. in 2005 in two papers exploring different series of hydroxamate-based inhibitors: the 3,3-dimethyl-5-hydroxypipecolic and the 3-OH-3-methylpipecolic series ( Figure 6). 63,64 Dimethyl-5-hydroxypipecolic inhibitors, selective for aggrecanases and the collagenase MMP13, were inspired by a screening on previously published ADAM17 inhibitors. The best inhibitor was compound 5 ( Figure 6), for its excellent inhibitory activity on aggrecanases and MMP13, sparing MMP1. 63 In the 3-OH-3-methylpipecolic series, the best inhibitor was 6 ( Figure 6) presenting a 2-chloro-4-fluorobenzyloxyphenyl function in P1′ with good inhibitory activity for the aggrecanases and MMP13, but poorly selective over MMPs. 64 The exploration of different structures by Cappelli et al. in 2010 led to the design, synthesis, and biological evaluation of a small series of aggrecanase inhibitors, based on a central planar scaffold containing oxoisoindoline or pyrrolo [3,4-c]quinolin-1one, bearing a 4-(benzyloxy)phenyl substituent and different ZBGs. 65 Derivatives 7 and 8 ( Figure 6) exhibited the highest activity against the two aggrecanases. Interestingly, the simplified structure of oxoisoindoline derivative 8 lacked inhibitory activity against ADAMTS4, while maintaining micromolar activity for ADAMTS5. Unfortunately, no selectivity profile over MMPs/ADAMs was reported for this series.
A series of N-hydroxyformamide inhibitors was investigated as ADAMTS4 inhibitors. 66 Starting from a screening of previously published MMP13 inhibitors, the N-hydroxyformamide group was identified as a key structural element for ADAMTS4 inhibition. This led to the synthesis of two series of compounds, functionalized by either a phenylpiperazine or a benzyloxypiperidine group. The best compound was the dimethylisoxazolyl derivative 9 ( Figure 6), displaying picomolar activity for ADAMTS4 and good selectivity over MMPs. No selectivity data for ADAMTS5 were reported. Compound 9 was crystallized in complex with the Mp domain of ADAMTS1, here chosen as a proxy for ADAMTS4. By combining the results from the crystallographic analysis with a homology model of the ADAMTS4 active site, the orthomethyl substituent on the aromatic ring of P1′ was identified as a crucial moiety for ADAMTS4 inhibition. Later, the P1′ group of 9 was further modified to improve its bioavailability. 67 The best compound of this series was 10, being selective for ADAMTS4 over ADAMTS5/MMPs and showing good pharmacokinetic properties as well as in vivo efficacy in a spontaneous OA model. In 2013, the arylsulfonamido- hydroxamate 11 ( Figure 6) was identified as an inhibitor of aggrecanases and MMP13, with high selectivity over other MMPs. 68 The inhibitory activity against ADAMTS5, initially tested using a quenched fluorescent (QF) peptide substrate, was further confirmed using purified aggrecan. Inhibition of aggrecan cleavage was significantly decreased (∼2-fold) compared with that of the peptide substrate, a phenomenon frequently observed with small-molecule inhibitors. Compound 11 was able to inhibit aggrecan breakdown in porcine cartilage explants stimulated with interleukin (IL)-1α with almost complete inhibition observed at 10 μM, and with no toxicity effects. 68 2.1.2. Carboxylate Inhibitors. The carboxylate is a viable option as a ZBG since its lower affinity for Zn 2+ compared to the hydroxamate provides more opportunities for selectivity, 69 given that the binding energy of the interaction with its target protease will be more evenly distributed between the ZBG and the P substituents. In 2006, researchers at Wyeth reported the first aggrecanase inhibitors bearing a carboxylic acid as a ZBG. 70 This series presented a biphenylsulfonamido-3methylbutanoic acid scaffold and was designed on the basis of high-throughput screening (HTS) results and a homology model of ADAMTS4 Mp domain derived from the structure of metalloprotease Atrolysin C. The broad-spectrum MMP inhibitor CGS27023A (Novartis) was docked into the ADAMTS4 active site. In the following SAR analysis, carboxylate 12 ( Figure 8) was identified as the best ADAMTS4 inhibitor, sparing MMP1 and MMP14, but still inhibiting MMP2 and MMP13. No data were reported for ADAMTS5, although the parental compound CGS27023A was inactive against this aggrecanase at concentrations up to 25 μM. Compound 12 showed promising pharmacokinetics properties, with a good oral bioavailability and dose−response inhibition of aggrecan degradation in bovine IL-1α-stimulated cartilage explants.
Investigation on the SAR of a variety of substituted aromatic systems, particularly on the para position of the biphenyl ring of biphenyl-4-sulfonamido carboxylates, identified the 4isobutyryl derivative 1 (Figure 2A and Figure 8) as a nanomolar inhibitor of ADAMTS4 and MMP13, with good selectivity over ADAMTS5, MMP14, and MMP1. 71 Starting from the structure of 1, different substitutions to replace the second ring of the biphenyl moiety were investigated. 72 As a first step, heterocycles were employed such as pyridine, furan, and tetrazole, but the result was a loss of activity against ADAMTS4. Functional groups containing hydrogen bond acceptors and donors were subsequently inserted in the meta and para positions of the biphenyl ring. The acetamido derivative 13 ( Figure 8) was identified as a potent inhibitor of ADAMTS4. The last modification was the insertion of a substituted benzyloxy functionality. The best compound was the trifluoromethyl derivative 14 (Figure 8) which displayed nanomolar activity against ADAMTS4 and MMP13 and 87% inhibition of aggrecan degradation at 10 μg/ mL.
In the years from 2009 to 2011, sulfonamido-based cyclopropane carboxylates were investigated as ADAMTS5 inhibitors. These compounds were characterized by a specific P1′ group with novel piperidine or piperazine-based heterocycles connected to a cyclopropane amino acid scaffold via a sulfonamide linkage. The first series of N-substituted 2-phenyl-1-sulfonylaminocyclopropane carboxylates was reported with the specific enantiomeric configuration 1R,2S. The best compound of this series was 15 ( Figure 8) with an IC 50 value of 73 nM against ADAMTS5. 73 A SAR of non-Nsubstituted 2-phenyl-1-sulfonylaminocyclopropane carboxylates identified compound 16 ( Figure 8) with an IC 50 value of 84 nM against ADAMTS5. In sharp contrast to the previous series, the preferred cyclopropane configuration for the ADAMTS5 activity of compound 16, and in general of the non-N-substituted series, was 1S,2R. The key points for stereochemical activity were the different orientation of the sulfonamide nitrogen toward the solvent (compound 15) or a hydrogen bond to the backbone carbonyl of Gly380 residue in the absence of N-substitution (compound 16). A further hit optimization based on the structure of compound 16 was undertaken by modification of the arylsulfonyl moiety and the cyclopropane core. The best compound was 17 (Figure 8), which presented a chloro-imidazole phenyl ring on the sulfonyl group and a cis-3-methyl substitution on the cyclopropane. Compound 17 was a potent inhibitor of both ADAMTS4 and ADAMTS5, but, notwithstanding a good selectivity over MMP1 and ADAM17, was equally potent against MMP13 and MMP14. In order to improve the selectivity of 17, the authors explored the effects of different substituents on thiophene and pyrazole rings and then replaced them with a condensed tricyclic scaffold. 74 The most promising compound, 18, contained a methyl group at the 2-position of the cyclopropane ring and a novel P1′ heterotricycle sulfamidebased scaffold (1,2,3,4-tetrahydropyrido-(3′,4′:4,5)imidazo-[1,2-a]pyridine). Carboxylate 18 showed IC 50 values of 23 and 8.4 nM against ADAMTS4 and ADAMTS5, respectively, and an improved selectivity over other MMPs (>1000-fold). Docking of 18 into ADAMTS5 and MMP14 Mp domains provided an explanation for this remarkable selectivity. While the cyclopropane ring interacted favorably with Thr378 of ADAMTS5, the 2-methyl substituent provided steric repulsion with Phe198 of MMP14.
Following the design of the P1′ substituted bicyclic ring, Peng et al. reported a series of 4-(benzamido)-4-(1,3,4oxadiazol-2-yl)butanoic acids as aggrecanase inhibitors. 75 In this series, a highly rigid 1,3,4-oxadiazol-2-yl ring was introduced as a linker between the scaffold (composed by the carboxylic acid ZBG and the biphenyl P1′ group) and the aromatic P2′ group. The best compound was the biphenyl derivative 19 ( Figure 8) with a trimethoxy phenyl moiety as a P2′ interacting group and inhibitory activity in the low micromolar range against ADAMTS4 and ADAMTS5. No selectivity data for MMPs were reported.
Another glutamate-like compound, 20 (AGG-523, US Patent WO2007008994) (Figure 8), developed by structurebased drug design by Wyeth (now Pfizer) and moderately selective for ADAMTS4 and ADAMTS5 over MMPs, is so far one of the few aggrecanase inhibitors reaching clinical trials. Notwithstanding its protective effect in a rat model of surgeryinduced OA, 76 development of AGG-523 was halted following phase I clinical trials in patients with mild to moderate (Clinical Trials ID: NCT00427687) and severe (NCT00454298) knee OA. The two studies were completed in 2008, but no results were reported. Sadly, the inconsistency between the performance of aggrecanase inhibitors in in vivo models and clinical trials is a common setback in the pharmaceutical field and highlights once again the need for improved preclinical models and a better understanding on the pathogenesis of OA (see section 4).
An alternative scaffold containing a central thienosultam (1,1-dioxothieno[2,3-d]isothiazole) was reported by Atobe et al. 77 These compounds presented different aromatic, polyaromatic, biphenyl, and alkyne substituents in P1′. The best inhibitors were the N-benzyl derivative 21 and the alkyne derivative 22 (Figure 8), which showed good selectivity for ADAMTS5 over ADAMTS4 and MMPs. The best oral bioavailability in rats was reported for carboxylate 21.

Hydantoin Inhibitors.
In order to improve both the selectivity and the pharmacokinetic profile of aggrecanase inhibitors, novel ZBGs alternative to classical hydroxamate and carboxylate were explored.
After HTS of more than 80 000 structurally different compounds, researchers at Eli Lilly identified hydantoin 23 ( Figure 9) as an alternative ZBG to develop aggrecanase inhibitors. 78 The X-ray structure of compound 23 in complex with ADAMTS4 showed that the hydantoin ring coordinates the Zn 2+ ion while the amide linker established hydrogen bonds with Leu330 and Pro393, thus orienting the aromatic ring into the S1′ pocket ( Figure 10A). This crystallographic analysis provided fundamental information to address the P1 substitution using structure-based drug design to improve selectivity. Modifying P1 from methyl (compound 23) to thiazole or imidazole group (compounds 24 and 25, respectively, Figure 9) resulted in increased selectivity for  (24) and imidazole (25) derivatives in complex with ADAMTS4 showed that they bound to the active site in a similar manner, a slight difference being detectable only around the imidazole ring that was rotated 45°out of the plane occupied by the thiazole ( Figure 10B). Benzofuran derivative 26 ( Figure 9) was identified as the best inhibitor of this series, with an IC 50 value of 4 nM for ADAMTS4 and ADAMTS5 and good efficacy in a rat model of inflammatory OA.
On the basis of these results, Eli Lilly's researchers further optimized the benzofuran hydantoin scaffold by introducing a cyclopropyl substituent in P1′ position to obtain compound 27 ( Figure 9). 79,80 Compound 27 revealed a good projected human pharmacokinetic profile but a significant, undesired glutathione conjugation in rats. With the aim of minimizing glutathione conjugation and lowering the projected human dose, the structure of 27 was further modified by replacing the benzofuran moiety. para-Trifluoromethyl benzyl derivative 28 ( Figure 9) was finally identified as the most promising aggrecanase inhibitor with nanomolar activity against ADAMTS4 and ADAMTS5, good selectivity profile over MMPs, good pharmacokinetic profile, and efficacy in a rat model of inflammatory OA.
In 2021, a new hydantoin-based ADAMTS5 inhibitor, GLPG1972/S201086 (29, Figure 9), bearing a difluorophenyl-piperazine as P1′ group, was co-developed by Galapagos and Servier. 81 The crystal structure of 29 in complex with the ADAMTS5 Mp domain showed, in agreement with other similar derivatives, that the hydantoin ring coordinated the Zn 2+ ion thus orienting the cyclopropyl ring toward the S1 pocket while the difluorophenyl ring perfectly fitted the S1′ pocket. The specific conformation of the methyl group, axial to the piperazine ring, established hydrophobic contacts with Leu443 ( Figure 11). GLPG1972 had IC 50 values of 19 and 156 nM against ADAMTS5 and ADAMT4, respectively, and good selectivity over MMPs and ADAM17. In mouse cartilage explant assays, the IC 50 value increased 100-fold (10 μM), 81 most likely reflecting reduced target engagement and/or competition with aggrecan. This reduced efficacy in cartilage explant assays compared with pure component assays has been frequently observed for aggrecanase inhibitors. 82 No inhibition was observed on type II collagenolysis in both mouse and human cartilage explants or on MMP-mediated aggrecan degradation, thus confirming GLPG1972 selectivity over MMPs. 83 In a mouse model of surgery-induced OA, GLPG1972 at 30−120 mg/kg reduced femorotibial aggrecan loss, cartilage structural damage, and subchondral bone sclerosis (20−40% compared to vehicle controls). Doubleblind, placebo-controlled phase I trials were then conducted in Belgium (NCT02612246), USA (NCT03311009), and Japan. GLPG1972 was safely tolerated in healthy adult men (of both white and Japanese origin) and in male and female participants with OA. 84 In OA patients, once-daily dosing for 14 days significantly reduced levels of ADAMTS-generated aggrecan cleavage (ARGS) fragments in plasma compared with placebo in healthy volunteers. Once GLPG1972 administration was stopped, ARGS levels returned to baseline within 14 days, remaining stable until day 50, suggesting that the interaction between GLPG1972 and ADAMTS5 was reversible. In the light of these promising results, GLPG1972 was evaluated in 932 patients with symptomatic knee OA in a double-blind placebo-controlled randomized phase II clinical trial (NCT03595618). GLPG1972 was given orally at 3 different doses (75,150, and 300 mg), once daily for 52 weeks and was well tolerated, with no increased risk of adverse MSK events compared with placebo. However, GLPG1972 did not meet its primary end point of change from baseline in cartilage thickness of the medial tibiofemoral compartment, as measured by magnetic resonance imaging at week 52. All  secondary outcomes, both structural and pain-related, were not met in this trial. The causes of this lack of efficacy are currently unknown, and GLPG1972 did not progress into phase III trials.

Inhibitors with Sulfur-Based
ZBGs. An alternative and less explored ZBG is the thioxothiazolidinone, investigated in a class of rhodanine-based ADAMTS5 inhibitors developed by Wyeth. 85,86 The 5-((3-(trifluoromethyl)-1H-pyrazol-4-yl)methylene)-2-thioxothiazolidin-4-one derivative 30 ( Figure  12) was identified as the best compound of these series with inhibitory activity in the micromolar range for ADAMTS5 and a modest selectivity over ADAMTS4. 86 No selectivity profile over MMPs/ADAMs was reported for this series.
Another series of derivatives bearing a thiazolidin-4-one emerged via HTS and led to the identification of the pyridine derivative 31 ( Figure 12) as a promising ADAMTS5 inhibitor with a good selectivity profile over ADAMTS4, ADAM17, and MMPs. 87 Compound 31 was described as a non-competitive inhibitor in a QF-peptide cleavage assay following Lineweaver−Burk plot analysis. 87 However, given the magnification To address this issue, the structure of compound 31 was modified by removing the carboxylic acid alkyl chain, which was considered responsible for the low membrane permeability, while maintaining the thiazolidinone as ZBG. 89 The 2-pyridyl thiazole central core was then replaced by various heterocyclic systems (monocycle, bicycle, or tricycle) to investigate the effect on ADAMTS5 inhibition. The benzimidazole derivative 32 ( Figure 12) showed improved membrane permeability compared to 31, but this was achieved at a cost of a loss in selectivity over MMPs and ADAM17. An alternative ZBG is the 1,2,4-triazole-3-thiol scaffold where the exocyclic sulfur atom coordinates the zinc-ion. From a focused library of 500 differently substituted 1,2,4-triazole-3thiols, the 3-(N-imidazolyl)propyl derivative 33 ( Figure 12) emerged as the best ADAMTS5 inhibitor, with good selectivity over ADAMTS4. 90 Based on the inhibitory activity of the synthetic intermediate acylthiosemicarbazide 34 (Figure 12), a library of 920 analogues with this ZBG was designed. 91 Different modifications of acetylthiosemicarbazide were explored, the SAR analysis and docking study revealing three fundamental interactions of acylthiosemicarbazide and the ADAMTS5 active site. The best inhibitor was 35 (Figure 12), with nanomolar activity against ADAMTS5 and good selectivity profile over a panel of MMPs, probably caused by an optimized interaction between its cyclohexylbutyl group and the S1′ pocket.

Inhibitors with Non-canonical ZBGs.
Following HTS, researchers at Wyeth reported preliminary data on two different series of ADAMTS5 inhibitors using as a scaffold either 5′-phenyl-3′H-spiro[indoline-3,2′-[1,3,4]thiadiazol]-2one 92 or hydroxyquinoline. 93 These two series have been investigated through a wide SAR analysis and led to the identification of several ADAMTS5 inhibitors with submicromolar potency characterized by a good selectivity over ADAMTS4, MMP12, and MMP13. The best compounds of each series were the spiroindoline 36 and the 8-hydroxychlorochine 37, respectively ( Figure 13).
Researchers at GlaxoSmithKline (GSK) reported the identification from a four-billion-member DNA-encoded 1,3,5-triazine library (Encoded Library Technology) of sulfonamide 38 ( Figure 13) as a potent ADAMTS5 inhibitor (IC 50 value: 30 nM) presenting a >50-fold selectivity over ADAMTS4 and an impressive >1000-fold selectivity over ADAMTS1, MMP13, and ADAM17. 94 Compound 38 was able to inhibit the release of ARGS aggrecan fragments and GAGs in response to IL-1β/oncostatin M (OSM) stimulation in human OA cartilage explants. No binding/functional experiments were carried out to assess the mechanism of inhibition of 38. Analysis of the literature allowed El Bakali et al. to define the amino-triazine ring as the ZBG, either via the exocyclic NH group or one of the triazine nitrogen. 95 The same approach was used to identify a potent, highly selective ADAMTS4 inhibitor, the 3,4-dihydroisoquinoline derivative 39 ( Figure 13). 96 2.2. Tissue Inhibitor of Metalloproteinase 3. The proteolytic activity of aggrecanases is regulated by TIMPs. TIMPs act as endogenous, ECM-associated inhibitors of several MA families such as MMPs, ADAMs, and ADAMTSs. 97 TIMPs are 4 small (21−28 kDa) proteins residues and a C-terminal domain of about 65 residues, each stabilized by 3 disulfide bonds. 98 The N-terminal domain is a fully active metalloproteinase inhibitor. The mechanism of inhibition involves the insertion of the ridge comprising the Nterminal five residues into the metalloprotease active site in such a way that the first amino acid, which is invariably a cysteine, coordinates the active-site zinc through its α-amino and carbonyl groups. 97 TIMP3 is the only vertebrate TIMP that is bound to the ECM through electrostatic interactions with sulfated GAGs. 99,100 TIMP3 expression is post-translationally downregulated in OA. 97 That a decreased TIMP3 expression may contribute to the proteolytic imbalance typical of the disease was corroborated by the phenotype of Timp3 null mice, which exhibited mild cartilage degradation in the absence of inflammatory or mechanical insults. 101 Among the 4 TIMPs, TIMP3 is also the most potent aggrecanase inhibitor (Table  2). 102,103 A truncated TIMP3 variant containing only the Nterminal domain (N-TIMP3) inhibited the activity of ADAMTS4 and ADAMTS5 against native bovine aggrecan with IC 50 values of 3.3 and 0.66 nM, respectively. 103 Because of its sub-nanomolar affinity, TIMP3 is appealing as a DMOAD. Unfortunately, two factors prevented the use of TIMP3 as a therapeutic agent, i.e., its broad-spectrum inhibitory activity as well as its low half-life. TIMP3 inhibits the majority of MMPs, several ADAMs as well as ADAMTS2. 97,104 Promiscuous metalloproteinase inhibition has been frequently associated with undesired MSK effects such as arthralgia, myalgia, joint stiffness, and tendinitis. 105 TIMP3 half-life is negatively regulated by its endocytosis and subsequent lysosomal degradation via the low-density lipoprotein receptor-related protein 1 (LRP1) receptor. 106 Therefore, it might be desirable to increase TIMP3 half-life to improve efficacy and reduce the dose or frequency of administration under a therapeutic regime. Strategies aiming to engineer TIMP3 as a DMOAD should aim to increase both TIMP3 selectivity and half-life.
Despite its mechanism of inhibition, N-TIMP3 is a more potent inhibitor of ADAMs and ADAMTSs compared with MMPs ( Table 2). Introduction of an extra alanine residue at N-terminus of N-TIMP3 further increases this bias by disturbing the interaction between Cys1 and the active-site Zn 2+ (Table 2). 107 The resulting variant, [-1A]N-TIMP3, was a nanomolar inhibitor of ADAMTS5, being 13-fold selective over ADAMTS4 and 20-fold selective over ADAM17, while almost sparing MMPs. 108 [-1A]N-TIMP3 potently inhibited GAG release from knee OA cartilage stimulated with IL1α/ OSM (although not as potently as TIMP3), while not affecting MMP-mediated collagen release. 108 Full-length [1-A]TIMP3 was tested in vivo using a model of spontaneous OA, the STR/ Ort mice. Transgenic STR/Ort mice over-expressing [1A]TIMP3 either under the elongation factor EF1α promoter (ubiquitous expression) or the Col2a1 promoter (chondrocytes-specific expression) were protected from cartilage degradation compared to wild-type mice at 40 weeks. 109 In addition, these transgenic mice showed increased trabecular bone mass, suggesting that administration of [1-A]TIMP3 may prevent osteoporotic bone loss, particularly in female mice. [1-A]TIMP3 was also tested in a surgical OA model. Transgenic C57BL/10 mice overexpressing [1-A]TIMP-3 under the Col2a1 promoter showed increased protection following destabilization of medial meniscus (DMM) as compared to wild-type. 110 Overexpression of [1-A]TIMP3 was more efficient in protecting from cartilage degradation than that of TIMP3 8 weeks post-DMM, a time point mimicking late-stage OA. Importantly, while mice over-expressing TIMP3 showed a significant decrease in trabecular bone volume, number, and thickness, neither WT mice nor [-1]TIMP3 showed these changes, thus demonstrating that selective inhibition of aggrecanases may prevent unwanted effects on bone integrity.
Recombinant TIMP3 (rTIMP3) had a short half-life (3.6 h) when added to HTB94 chondrosarcoma cells, due to its rapid uptake and degradation by the LRP1 receptor. 106,111 Based on the notion that LRP1 ligands are characterized by a positively charged cluster composed by two lysine residues 21 Å apart which bind to negatively charged residues on LRP1, 112 the Troeberg's group analyzed a panel of TIMP3 variants where pairs of lysine residues predicted to be separated by 21 Å were mutated to alanine to increase TIMP3 half-life. 111 They  111 Importantly, the two variants maintained the inhibitory profile of the parental TIMP3 molecule against several metzincins ( Table 2). Most likely due to their resistance to LRP1mediated endocytosis, TIMP3 variants K26A/K45A and K42A/K110A were more effective than wild type TIMP3 in inhibiting GAG release from porcine cartilage explants following a 3-days pre-incubation period. 111 Mutations aiming to prevent LRP1 binding exert also positive effects on TIMP3 expression levels, 111,113 an important factor in view of a future scale-up for industrial production. If systemic administration of rTIMP3 is attempted, another issue is the short half-life of the molecule in serum. The molecular weight cutoff for glomerular filtration is 30−50 kDa, 114 well above TIMP3 molecular weight of ∼22 kDa. 97 Fusion with a human Fc antibody region can extend half-life through the interaction with the immunoglobulin salvage receptor FcRn; the Fc region itself can be engineered to strengthen further this interaction. 115 A similar effect is produced by fusing with a serum protein with extended halflife such as albumin or by increasing the molecular mass of TIMP3 above the glomerular filtration cut-off, for example by conjugation with polyethylene glycol (PEG) or introduction of additional glycosylation sites. These strategies have been extensively explored by Chintalgattu et al. 113 Table  2). This is quite unfortunate since it is likely that extended glycosylation/PEGylation will affect their inhibitory profile. Another approach involved N-terminal fusion of TIMP3 with the latency-associated peptide from the cytokine Transforming growth factor β, which can be removed in situ by MMP1. 116 The resulting activated TIMP3 molecule has an extra leucine at the N-terminus and, similarly to [-1A]TIMP3, showed higher selectivity for aggrecanases over MMPs, although no inhibition constants have been reported so far (estimated IC 50 value for ADAMTS4 inhibition from Figure 1E in ref 116 is ∼10 nM, i.e., considerably higher than wild-type TIMP3).
Taken all together, these studies highlighted the feasibility of improving TIMP3 selectivity and pharmacokinetics. The next step will be combining the selectivity profile of [1-A]TIMP3 with the increased half-life of the K26A/K45A and K42A/ K110A variants. So far, administration of recombinant TIMP3 has not been tested in mouse models of OA, which have focused on transgenic expression. Therefore, there is an important piece of information missing along the pathway to the therapeutic application of TIMP3 as DMOAD. However, this approach has been investigated in the context of cardiovascular diseases. 117 For example, to test the protective effect of TIMP3 on myocardial infarction, rTIMP3 has been directly injected into the myocardium of pigs subjected to coronary ligation. 118,119 In this case rTIMP-3 was administered in a hyaluronan-rich hydrogel, mimicking binding of TIMP3 to GAGs, to extend its half-life. 118

EXOSITE INHIBITORS
Avoiding the chelation of the zinc atom, common to all the metalloproteases belonging to clan MA, could be an important factor for improving the selectivity profile and avoiding offtarget toxicity. 120,121 The inhibitors discussed in this section are devoid of a ZBG; some of them have been further characterized as exosite inhibitors.
3.1. Sulfated Glycosaminoglycans. Sulfated GAGs represent a promising opportunity to achieve exosite inhibition. For example, heparin (Figure 14), a heterogeneous preparation of linear, highly sulfated GAGs, inhibited ADAMTS5 aggrecanase activity with an IC 50 value of 20 μg/ mL. 122 Since the average molecular weight for porcine heparin is 17.5 kDa, 123 this translates to an IC 50 of 1.14 μM. Unfortunately, due to its anticoagulant properties and associated side effects, such as thrombocytopenia, 124 heparin itself is not suitable as a DMOAD.
An alternative to heparin may be Calcium Pentosan Polysulfate (CaPPS) (Figure 14), a calcium salt form of chemically sulfated molecule produced from beechwood (Fagus sylvatica) consisting of a β-1,4-linked polymer of xylose with β 4-methyl glucuronic acid residues attached to the 2-OH of every 10th xylose. CaPPS has been shown to effectively inhibit aggrecan degradation in human OA cartilage explants under inflammatory conditions. 122 122 In cell culture, the mechanism of inhibition of CaPPS is quite complex. By blocking the endocytosis of TIMP3 via the LRP1 receptor, CaPPS increased extracellular TIMP3 levels; it also enhanced the affinity of TIMP3 for ADAMTS4 and ADAMTS5 (>100 fold). 122,125,126 Although CaPPs has been shown to be effective in some OA clinical trials, 127−129 it has not been yet approved as a DMOAD. Further clinical trials are under way (NCT04814719, NCT04809376).

Glycoconjugates.
GAGs can be successfully linked to canonical metalloproteinase inhibitory scaffolds, such as the arylsufonamide, and ZBGs, thus generating glycoconjugates. 130 By screening a series of glycoconjugate MMP12 inhibitors, 131,132 Santamaria et al. 51 identified carboxylic acid 40 (Figure 15), where a β-N-acetyl-D-glucosamine monosaccharide is linked to the arylsulfonamide scaffold, as an ADAMTS5 inhibitor with activity in the micromolar range. Removal of the ZBG resulted in compound 41 (Figure 15), which inhibited ADAMTS5 cleavage of both versican and aggrecan with IC 50 values in the micromolar range, but spared ADAMTS4. No significant inhibition was observed on QF peptide cleavage assays; moreover, 41 enhanced the inhibitory activity of the broad-spectrum zinc-binding MMP inhibitor GM6011 against ADAMTS5. These results suggested the possibility that 41 targets an exosite. Docking calculations combined with molecular dynamics simulations demonstrated that 41 targets the interface of the Mp and Di domains. The combination of kinetic and in silico study demonstrated that 41 is an exosite cross-domain inhibitor, acting by an unprecedented mecha-

Journal of Medicinal Chemistry
pubs.acs.org/jmc Perspective nism where the S1′ pocket is occupied by the arylsulfonamide scaffold, whereas the sugar moiety interacts with a positively charged cluster ( 532 KK 533 ) in the ADAMTS5 Dis domain ( Figure 5A). Site-directed mutagenesis confirmed that this region represents a previously unknown exosite which is critical for substrate recognition and can therefore be targeted for the development of selective ADAMTS5 inhibitors.

Flavonoids.
Several natural compounds such as flavonoids are known to possess metzincin inhibitory activity, and some of them have been investigated as aggrecanase inhibitors.
A series of green tea catechin gallate esters have been reported as ADAMTS4, ADAMTS5, and ADAMTS1 inhibitors. In particular, (−)-epigallocatechin-3-gallate (EGCG) and (−)-epicatechin gallate (ECG) and piceatannol ( Figure 16) showed IC 50 values for the two aggrecanases in the range of 100−150 nM, although they were poorly selective over MMPs and ADAMs. 133 In 2009 Cudic et al. 134 evaluated molecules that are structural components or structurally related to EGCG, ECG, and piceatannol such as resveratrol, transstilbene, cis-stilbene, deoxyhapontin, rhapontin, pyrocathecol, and pyrogallol ( Figure 16). These molecules inhibited aggrecanase cleavage of QF triple-helical peptides with IC 50 values in the low micromolar range. A poorer inhibition on short (≤10 amino acids) compared with long (>20 amino acids) substrates suggested that pyrogallol and luteolin may bind to exosites which are not engaged by the former, although their broad inhibition of MMPs contradicts this hypothesis. It is likely that these molecules, devoid of an obvious ZBG, bind to subsites within the Mp domain of ADAMTS4 and ADAMTS5 rather than bona f ide exosites.
Luteolin (Figure 16), a flavonoid widely distributed in plants, especially in celery and green pepper, inhibited aggrecanase activity of ADAMTS4 and ADAMTS5, although was only modestly selective over MMPs. 135 Luteolin effectively inhibited the release of GAGs and ARGS-aggrecan fragments in mouse chondrogenic ATDC5 cells and in murine cartilage explants stimulated with IL-1α/retinoic acid, while MMP aggrecanolytic activity was not affected. Interestingly, this inhibitory effect was partly caused by a transcriptional downregulation of Adamts4 and Adamts5 expression, another example of a dual mode of inhibition.
Overall, these findings suggest that the structure of flavonoids should be further modified to improve inhibitory potency and selectivity before these molecules could be tested in clinical studies.

Aptamers.
Nucleic acid aptamers, often termed "chemical antibodies", are short, single-stranded DNA or RNA molecules (20−100 nucleotides in length) that share with antibodies the ability to recognize their targets with exquisite affinity and selectivity. 136 Complementary base pairing allows the formation of unique 3D folds that can be selected for their ability to bind a specific target through in vitro selection methods such as systemic evolution of ligands by exponential enrichment (SELEX). Compared to mAbs, aptamers have theoretically a competitive advantage for therapeutic purposes due to their smaller size (6−30 kDa), lower manufacturing costs, and lower immunogenicity, although they suffer from limited half-life in vivo (∼10 min in the absence of specific modifications). 136 RBM-010 (patents US20210246451A1, WO2019093497) is the first RNA aptamer-based ADAMTS5 inhibitor developed by Ribomic Inc. and is currently in preclinical evaluations.
DNA aptamers are more stable and easier to synthesize compared with RNA aptamers, while RNA aptamers are typically endowed with higher affinity and selectivity. 137 Yu et al. used SELEX to isolate two DNA aptamers, apt21 and apt25, against ADAMTS5. 138 Although the two aptamers had affinities in the low nanomolar range (1.54 and 1.79 nM, respectively) they exhibited a poor inhibitory activity in a QFpeptide cleavage assay (52.76 and 61.14 μM, respectively). Inhibition of proteoglycan cleavage was not tested.
Although aptamers have been so far superseded by mAbs in therapeutic applications, 136 it is likely that more of them will reach the clinic, therefore we expect that R&D investments in aptamer-based aggrecanase inhibitors will grow, albeit at a slow pace.

Peptide-Based Inhibitors.
Like protein-based inhibitors, peptide-based inhibitors bind to their targets with an extended surface of interaction, thus generally achieving higher selectivity. However, like small molecules, peptides can Note that different forms of enzymes were tested in the different studies. K d values (in μM) are reported within parentheses and were measured by fluorescence polarization. Abbreviations: ND, not determined; NA, not applicable; Ac, acetyl. Unless indicated differently, all sequences are reported from the N-terminus to the C-terminus. b Values determined using bovine aggrecan (Glu392-Ala393 cleavage site). c Values determined using a QF-peptide.
be synthesized chemically and are thus cheaper to produce than recombinant proteins. Other advantages include low toxicity and reduced antigenicity. 139 Therefore, peptide-based inhibitors are potentially endowed with the advantages of the two different classes of molecules. However, due to their small size, peptide-based inhibitors have reduced half-life, an issue that can be addressed in a similar way as TIMPs. So far, few peptide-based aggrecanase inhibitors have been reported, all of them targeting ADAMTS4 (Table 3). Unfortunately, none of them has been tested neither against ADAMTS5, nor against any other metalloproteinase.
Following the observation that removal of the TS-1 motif greatly reduced the aggrecanase activity of ADAMTS4, Tortorella et al. hypothesized that this domain was involved in aggrecan binding. 140 They then tested a series of overlapping peptides based on the TS-1 sequence for their ability to inhibit ADAMTS4 aggrecanase activity. These peptides inhibited ADAMTS4 with IC 50 values in the micromolar range, presumably by competing with ADAMTS4 for binding to aggrecan (Table 3). 140 Hills et al. reported several peptides inhibiting ADAMTS4 peptidolytic activity with IC 50 values in the micromolar range (Table 3). 141 The sequences of these peptides were based on peptide substrates identified by phage display selection of a library of 10 8 random 13-amino-acid peptides. The amino acid composition of these peptides was equimolar for all 20 amino acids except cysteine.
In an alternative approach, Zhang et al. generated disulfidebonded cyclic peptides based on the sequence of a short inhibitory loop 85 EASESLC 91 (Uniprot ID P35625) of TIMP3. 142 While the linear peptide bound ADAMTS4 with a weak affinity (74 μM), cyclization improved considerably the affinity by minimizing the entropy penalty of the interaction (Table 3).
Overall, from the few examples reported in the literature it seems that the pharmacological potential of peptide-based inhibitors is far from being unlocked.
3.6. Monoclonal Antibodies. mAbs are potent and selective binders of many biologically relevant targets. For this reason, they are well established as therapeutic agents for several diseases including cancer, autoimmune disorders, and infectious diseases (the 100th mAb was approved by the U.S. Food and Drug Administration in 2021). 143 In 1975, Koḧler and Milstein described hybridoma technology, a method to generate mAbs based on the fusion of B-lymphocytes from an immunized animal with immortal myeloma cells. 144 Soon this method became popular for generation of mAbs for a variety of applications. An alternative way to generate mAbs is phage display, which has superseded hybridoma technology through the creation of large natural and synthetic in vitro repertoires of antibody fragments. 145 Both approaches have been used to generate potent and selective inhibitors of aggrecanases. 146 The versatility of phage display offers the opportunity to isolate mAbs with desired properties. For example, phage display selections where the active site of ADAMTS5 was blocked with the zinc-chelating inhibitor GM6001 have been used to obtain mAbs targeting ADAMTS5 exosites. 27 The two most potent inhibitors, 2D3 and 2B9, bound to the Mp/Dis and Sp domains, respectively (Table 4). Competition surface plasmon resonance experiments with TIMP3 and GM6001 confirmed that all these mAbs recognized epitopes outside the active-site cleft. Remarkably, the anti-Sp mAb 2B9 showed inhibitory activity on protein substrates such as aggrecan 27 and versican 22 but was unable to inhibit cleavage of a QF-peptide (a clear-cut example of exosite inhibition), while the anti-Mp/Dis mAb 2D3 was able to inhibit efficiently cleavage of both protein and peptide substrates by targeting an epitope in the Dis domain. 2D3 showed potent inhibitory activity of aggrecanase activity in unstimulated human chondrocyte monolayer cultures from healthy donors 27 and OA cartilage explants. 28 These mAbs showed exquisite selectivity, with no inhibition observed on ADAMTS4 at concentrations up to 500 nM. 27 Another anti-ADAMTS5 Sp mAb, CRB0017, developed by Rottapharm using a proprietary selection technology, was effective in delaying cartilage degradation in STR/Ort mice. 147 Phage display was instead used to isolate an anti-ADAMTS4/ ADAMTS5 inhibitory Fab fragment, 237-53, binding to an epitope in the central TS-1 motif of both aggrecanases. 148 This mAb completely inhibited ADAMTS4 but showed only partial inhibition of ADAMTS5 at a 1:5 enzyme/mAb ratio.
Several mAbs have been generated by GSK against ADAMTS4 and ADAMTS5. These mAbs showed subnanomolar affinity and recognized different domains on their target proteases (Table 4). 26 Both anti-ADAMTS4 and anti-ADAMTS5 mAbs (670 nM) effectively inhibited the release of aggrecan ARGS-fragments from IL-1β/OSM stimulated human OA cartilage explants, while in the absence of inflammatory stimuli only the anti-ADAMTS5 mAbs were effective. At 10−16 mg/kg, anti-ADAMTS5 mAbs conferred significant protection in the DMM mouse model. 26,31  Remarkably, intense knee staining was observed 4 days after administration via intraperitoneal injection, thus demonstrating high target engagement. 26 Prophylactic or therapeutic treatment (10 mg/kg) also protected from mechanical allodynia. 26,31 These promising results prompted further investigations in a non-human primate model. Administration in cynomolgus monkeys of anti-ADAMTS5 mAb GSK2394002 significantly decreased serum aggrecan ARGS levels. However, sub-endocardial hemorrhage as well a sustained increase in mean arterial pressure and ST segment elevation were observed with doses from 3 to >30 mg/kg, and these side effects were sustained for up to 8 months following a single dose of mAb. 149 It has been suggested that these cardiovascular effects may be due to inhibition of ADAMTS5 versicanase activity. 149 Although ADAMTS5 is ∼18-fold more potent than ADAMTS4 as a versicanase in vitro, 22 no mechanistic link between the cardiovascular anomalies elicited by GSK2394002 and ADAMTS5 versicanase activity has been reported so far. However, there are indications that these potentially concerning side effects may be mAb-specific, since another anti-ADAMTS5 mAb, M6495, was safely tolerated in phase I clinical trials. M6495 is a bivalent nanobody developed by Nordic Bioscience, Merck, and Ablynx, comprising two variable domains sequences derived from llama antibodies separated by a flexible glicyine-serine linker: an N-terminal sequence recognizing ADAMTS5 and a C-terminal sequence binding to human serum albumin to increase its half-life. 150 M6495 not only inhibited aggrecan degradation in OA synovial membranes, but also decreased Toll-like receptor 2 activation, suggesting a potential application as a painkiller. 151 Inhibition of ADAMTS5 activity by M6495 decreased the release of a 32mer aggrecan fragment (generated following independent cleavage by aggrecanases at Glu392-Ala393 and MMPs at N360-F361) which acts as a matrikine by exciting dorsal root ganglion nociceptive neurons in chondrocytes. 152 Two phase I clinical trials (NCT03583346 and NCT03224702) have been completed for M6495; at least in one of them (NCT03224702), M6495 was safely tolerated at doses up to 300 mg: a single dose of 300 mg resulted in a 45% decrease in circulating ARGS aggrecan levels that was maintained up to 74 days. 153

CONCLUSIONS AND PERSPECTIVES
The socioeconomic burden of OA is likely to increase, given the combined trends of aging and rising epidemic of obesity. Despite massive efforts in R&D pipelines, approval of a DMOAD is still far away. More than 20 years after the identification of ADAMTS4 and ADAMTS5 as the aggrecanases involved in cartilage degradation, 17,18 no molecule able to inhibit their activity has reached the clinic ( Table 5).
The development of many aggrecanase inhibitors was terminated owing to a lack of efficacy in animal models, which incompletely recapitulate human OA and are therefore poorly predictive of its progression. The majority of these preclinical studies used rodents as model organisms and mimicked mechanical loading/trauma (surgical models such as the DMM model), inflammation (such as the antigen-induced arthritis models) or genetically susceptible joint degeneration (the STR/Ort model). 154 As discussed in the previous sections, the converse is also true, with several inhibitors showing efficacy in animal models being terminated because of a lack of efficacy in clinical trials. Rodents differ from humans in articular cartilage physiology, weight bearing, gait, and sexdependent responses to catabolic stimuli and pain. 154 Developing animal models able to capture more closely the complexity of human OA will help focusing drug development efforts on bona f ide DMOAD candidates. Another factor that may affect the outcome of clinical trials is the choice of primary end point for the study. Radiographic joint space narrowing (i.e., the decrease in joint space width) is the primary structural end point accepted by the European Medicines Agency and the FDA to prove effectiveness of DMOAD candidates, but it suffers from many limitations, such as the need for long-term follow-up to observe changes in disease progression and its poor applicability to early OA. 155 To address this, composite end point approaches have been proposed. 156 However, as noted in a recent FDA draft guidance, "the ability of treatment effects on common measures of structural progression to Finally, the delivery route for the candidate DMOAD should be carefully selected. Intra-articular injections are inconvenient, uncomfortable to the patient and require trained healthcare staff. For these reasons, the ultimate goal for DMOADs has been oral administration. To enhance cartilage penetration, candidate DMOADs can be designed to target chondrocytes or the cartilage ECM. For example, addition of positively charged groups may increase the affinity for negatively charged ECM components such as aggrecan. However, even in the event that these modifications do not severely affect the physicochemical properties of the drug or its ability to effectively engage the target, levels of many proteoglycans such as aggrecan are known to increase in a variety of cardiovascular diseases, 157 which represent common co-morbidities for OA patients. 158 Targeting proteoglycans to deliver candidate DMOADs specifically to the cartilage will be a daunting task.
Initially, synthetic chemistry approaches followed in the path marked by MMP inhibitors, a choice that with the hindsight was doomed to failure. These first generation aggrecanase inhibitors predominantly contained the hydroxamate as a ZBG and peptide/peptidomimetic backbone which made them poorly selective. The MSK symptoms exhibited by this class of molecules during previous clinical trials for cancer therapy was a cause of concern. 15 Because of chronic administration in an older population with multiple co-morbidities, DMOADs must be able to demonstrate utmost safety. Molecules with alternative ZBGs, such as hydantoin GLPG1972/S201086 (29), 81 showed improved selectivity which can be reflected in their safe profile in preclinical and clinical trials (Table 5). Given these first results, further studies involving smallmolecule inhibitors bearing this ZBG may be considered as a promising strategy to develop new chemical probes or therapeutic agents.
So far, no zinc-binding inhibitor has received approval from regulatory bodies. There is still the possibility that the presence of ZBG may confer an intrinsic disadvantage to this class of molecules, for example by binding other metzincins. On the other hand, exosite molecules such as sulfated GAGs, flavonoids, and glycoconjugates at the moment do not have the potency required for being tested as feasible DMOADs. It seems obvious that an aggrecanase inhibitor must be endowed with the right combination of selectivity and potency in order to be developed as a DMOAD. Another consideration is that targeting either ADAMTS4 or ADAMTS5 may be preferable to avoid unwanted systemic effects. On this regard, the biological function of ADAMTS5 in tissues other than cartilage still needs to be fully elucidated. The highly selective anti-ADAMTS5 mAb GSK2394002 showed cardiovascular side effects in a non-human primate model 149 that arrested its progress to the clinic. OA is associated with a slightly increased risk of cardiovascular death compared with non-OA controls, 158,159 therefore cardiovascular integrity upon DMOAD administration must be preserved. As discussed in section 3.6, significant side effects were not observed for another anti-ADAMTS5 mAb, M6495, 153 thus further highlighting the extremely complex drug-specific pathways of target engagement. Although classical immunoglobulins remain promising DMOADs, antibody fragments such as nanobodies may have a competitive advantage in terms of target engagement and side effects.
It is worth highlighting that the majority of the antiaggrecanase mAbs reported so far either block access of substrates to the active-site cleft, for example by "freezing" their target protease in a closed conformation (GSK2394002) 26 or by binding to exosites in the Dis domain (2D3). 27 Alternatively, mAbs can block exosites in distal ancillary domains (2B9); 27 the modality of action of these mAbs resembles those of autoantibodies against ADAMTS13 that mainly target the Sp domain. 160 Unfortunately, exosite inhibitors have not yet fulfilled their mission. The fact that the anti-Sp mAb 2B9 inhibits not only the aggrecanase, but also the versicanase activity of ADAMTS5 22 may be a potential red flag for those who hope that these molecules may be able to achieve substrate-specific inhibition, even if such an effect is desirable from a therapeutic point of view.
Another unexplored mechanism by which mAbs can inhibit aggrecanases is by targeting their zymogen activation, as demonstrated with a mAb inhibiting activation of urokinasetype plasminogen activator (uPA). 161 Exploring alternative approaches is essential not only to enhance the chances of success in our quest for a clinically approved DMOAD, but also to deepen our understanding of aggrecanase biology. Engineering the structure of the endogenous aggrecanase inhibitor TIMP3 has resulted in a number of recombinant variants with increased selectivity and half-life (Table 2). An alternative scaffold for the generation of protein-based inhibitors may be α 2 -macroglobulin (α2M), a 720 kDa homotetrameric plasma inhibitor of a variety of proteases. 162 α2M is characterized by a unique mechanism of inhibition. Proteolytic cleavage within a bait region 39 residues-long triggers a conformational change resulting in sequestration of the target protein. 162 Engineering of the bait region generated α2M variants selective for MMP2, 163 therefore it may be feasible to fine-tune this sequence to target either one or both aggrecanases. Peptide and aptamer inhibitors can also probe unexplored regions in the 3D landscape of ADAMTS4/ ADAMTS5 binding sequences, although the reduced number of FDA-approved drugs falling within these categories definitely provides an obstacle for pharmaceutical investments in this field.
Notwithstanding the recent frustrating outcomes of antiaggrecanase clinical trials, there is room for optimism. Compared to 20 years ago, our knowledge of aggrecanase biology has vastly improved. Not only has the cardiovascular role of these proteases been uncovered 164,165 but the first exosite sequences have been identified. 22,51 Further research in the structure and function of aggrecanases will definitely improve our chances to target these elusive proteases for OA therapy.