The quest for substrates and binding partners: A critical barrier for understanding the role of ADAMTS proteases in musculoskeletal development and disease

Abstract

Secreted ADAMTS metalloproteases are involved in the sculpting, remodeling, and erosion of connective tissues throughout the body, including in the musculoskeletal system. ADAMTS proteases contribute to musculoskeletal development, pathological tissue destruction, and are mutated in congenital musculoskeletal disorders. Examples include versican cleavage by ADAMTS9 which is required for interdigital web regression during limb development, ADAMTS5-mediated aggrecan degradation in osteoarthritis resulting in joint erosion, and mutations in ADAMTS10 or ADAMTS17 that cause Weill-Marchesani syndrome, a short stature syndrome with bone, joint, muscle, cardiac, and eye involvement. Since the function of ADAMTS proteases and proteases in general is primarily defined by the molecular consequences of proteolysis of their respective substrates, it is paramount to identify all physiological substrates for each individual ADAMTS protease. Here, we review the current knowledge of ADAMTS proteases and their involvement in musculoskeletal development and disease, focusing on some of their known physiological substrates and the consequences of substrate cleavage. We further emphasize the critical need for the identification and validation of novel ADAMTS substrates and binding partners by describing the principles of mass spectrometry-based approaches and by emphasizing strategies that need to be considered for validating the physiological relevance for ADAMTS-mediated proteolysis of novel putative substrates.

1 |. INTRODUCTION

Musculoskeletal development and homeostasis is dependent on the proper function of connective tissues, which consist of tissue-specific extracellular matrix (ECM) and resident cells.¹ The formation of connective tissues during development and its continued physiological function under homeostatic conditions is critically dependent on dynamic sculpting and remodeling processes in accordance with developmental stages or biological and biomechanical inputs. Integral to these processes are proteolytic events mediated by secreted ECM proteases, which include matrix metalloproteases (MMPs), a disintegrin and metalloproteinases (ADAMs), and a disintegrin and metalloproteinases with thrombospondin type 1 motifs (ADAMTSs). In general, the biological sum of protease activity in tissue homeostasis coincides with the range of physiological substrates that are cleaved by that protease. The consequences of substrate proteolysis can involve pathological substrate degradation, substrate clearance in tissues to allow for the formation of a permanent ECM during tissue development and organ growth, the initiation or termination of biological functionality, such as activation of growth factors, shedding of growth factor receptors or generation of bioactive peptides, or the determination of the subcellular localization of substrates (Figure 1).^2–6 Protease activity is an essential and understudied element of the developmental narrative in the musculoskeletal system. Characterizing the substrate repertoire and binding partners of individual ADAMTS proteases promises deep insights into their physiological functions and molecular mechanisms that determine the phenotypes associated with their deletion or mutation in vivo. Here, we critically review evidence for the involvement of ADAMTS protease-mediated substrate cleavage in musculoskeletal development, homeostasis, and disease, and highlight the need to identify further, if not all, physiological substrates for these proteases as a pathway to fully understand the physiological and pathological roles of each individual ADAMTS protease.

FIGURE 1.

Consequences of substrate proteolysis. Protease-mediated substrate cleavage is an irreversible posttranslational modification that can have various biological consequence and is paramount in extracellular matrix (ECM) formation and remodeling in development and homeostasis. Proteolysis in most of these processes is required to trigger the assembly of ECM macromolecules, such as collagens and fibrillins or to remove the provisional ECM by degrading versican and other ECM and tissue templates. Proteases can also activate or inactivate substrates to elicit or terminate a biological function. Prime examples are the shedding of cell surface receptors and the activation of latent growth factors. The sum of these consequences define the physiological role of individual ADAMTS proteases

2 |. ADAMTS PROTEASES

The ADAMTS proteases constitute a 19-member family largely implicated in connective tissue formation and function, including in musculoskeletal development, homeostasis, and pathogenesis.⁷ ADAMTS proteases, together with MMPs and ADAMs, belong to the metzincin protease superfamily, characterized by the close proximity of a methionine residue and a zinc ion in their active site that together execute catalytic cleavage of the peptide bond (Figure 2A).¹⁰ Numbered 1 to 20, where ADAMTS11 is unused, having been assigned to a gene previously identified as ADAMTS5, each ADAMTS protease is encoded by its own gene, with the ADAMTS genes dispersed throughout the human genome.¹¹ ADAMTS proteases are typically found as homologous pairs with sometimes overlapping substrates and redundant functionality, and members of the family are conserved down to the worm, C. elegans, which expresses a single ADAMTS protease encoded by the GON-1 gene.^12–14 ADAMTS proteases contain two major functional domains, a protease domain at the N-terminus, which harbors the active site, and an ancillary domain at the C-terminus, which is thought to mediate substrate recognition, and which defines the individual ADAMTS proteases and protease pairs (Figure 2B).¹⁵ Included in the former is a signal peptide to target the protein for secretion, a propeptide of variable length that can be post-translationally cleaved by furin or furin-like proteases in an activation step, a metalloproteinase domain that enacts catalysis, and a disintegrin-like domain named for sequence homology with the disintegrin protein family found in viper venoms.¹⁵ The ancillary domains of all ADAMTS proteases are composed of a central thrombospondin type 1 motif (TSR), a cysteine-rich domain, and a spacer domain, which are then followed by variable and protease-specific domain arrays.¹⁶ ADAMTS-like (ADAMTSL) proteins share homology with the domain organization of ADAMTS proteases, but lack the catalytic domain and thus lack protease activity.⁷ However, ADAMTSL proteins may be involved in regulating ADAMTS protease activity or substrate accessibility.

FIGURE 2.

ADAMTS proteases and ADAMTSL proteins. A, Human proteases can be divided into five major classes.⁸ ADAMTS proteases belong to the metzincin family and are part of the metalloprotease class. Domain organization of, B, individual ADAMTS proteases and, C, ADAMTS-like proteins. The overall protein domain organization is indicated on top and the domain organization for the individual ADAMTS proteases is indicated below. The proteases are arranged based on a phylogenetic analysis published elsewhere and the branches were ommitted.⁹ ADAMTSL proteins share homology to the ancillary domain of ADAMTS proteases, based on the shared TSR1/cysteine-rich/spacer domain arrangement and the presence of arrays of TSRs and a PLAC domain.⁷ Domain size and domain spacing are not drawn to scale.

While the most basic organization of the ancillary domain is found in ADAMTS4, diverse structural variations occur across the ADAMTS family at the C-terminus, with combinations of up to 14 consecutive TSRs and a range of possible additional domains (Figure 2B).¹⁵ ADAMTS5, a protease phenotypically linked to ADAMTS4 in its musculoskeletal implications, is structurally identical except from one additional TSR that follows the spacer domain. Conversely, ADAMTS7 and 12 show vastly different C-termini, with a total of eight TSRs and a mucin-/proteoglycan-rich domain interposed between the fourth and fifth TSR, which makes ADAMTS7 and 12 putative proteoglycans as well.¹⁶ All four aforementioned ADAMTS proteases are implicated in the pathogenesis of arthritis, and yet, their distinct substrate profiles may be reflected in the structural and functional differences in the ancillary domain. ADAMTS4 and 5 cleave proteoglycans, such as aggrecan and versican in the pericellular matrix and the ECM.¹⁷ ADAMTS7 and 12 cleave the cartilage oligomeric matrix protein (COMP), and their unique C-termini are implicated in their particular substrate interactivity; the final four TSRs of ADAMTS7 and 12 are both necessary and sufficient for interaction with COMP.^18,19

Evolutionary relationships amongst the ADAMTS proteases are reflected in the similarity of their substrates and ADAMTS proteases can accordingly be divided into four subgroups: proteoglycanases, procollagen N-propeptidases, the von Willebrand factor protease ADAMTS13, and a middle clade of ADAMTS proteases for which substrates began to emerge only very recently.^20–22 The proteoglycanases include ADAMTS1, 4, 5, 8, 9, 15, and 20, with such ECM substrates as aggrecan, versican, or brevican.⁹ The procollagen N-propeptidases include ADAMTS2, 3, and 14, and they primarily process the N-terminal propeptide of procollagens to promote their maturation into tropocollagen, where ADAMTS2 may in fact be the only physiologically relevant procollagen N-propeptidase, at least in embryonic development.^23,24 Procollagen processing represents one of the prime examples for the cooperation of multiple proteases in the same biological process, that is, the formation of collagen fibrils. The N-terminal propeptide of procollagen type I can not only be processed by ADAMTS2, 3, or 14, but also by the metalloproteases meprin α and β.^25,26 In addition, the C-terminal procollagen propeptide is cleaved by BMP1 and meprin α and β^{27, 28}. As such, procollagen processing requires the proteolytic cleavage at two sites, which is executed by combinations of at least four proteases and probably dependent on the specific tissue context. ADAMTS13 seems to be uniquely specialized since the only known substrate is von Willebrand factor, a blood clotting factor involved in hemostasis.^29,30 Lastly, no substrates are known for several of the ADAMTS proteases, some of which are mutated in congenital musculoskeletal disorders and these orphan enzymes represent a fertile field of study for substrate identification.

3 |. ADAMTS PROTEASE AND THEIR SUBSTRATES IN MUSCULOSKELETAL DEVELOPMENT AND PATHOLOGY

3.1 |. Proteoglycanases

The substrate profiles of ADAMTS proteases together with some of the musculoskeletal manifestations in mouse models and human diseases are listed in Table 1. The most documented and expansive substrate repertoires are found amongst the proteoglycanase subfamily, particularly for ADAMTS1, 4, and 5, supporting their prominent roles in proteoglycan turnover within the ECM and their disease relevance. The proteoglycan substrates identified so far reside primarily in two classes: the hyalectans, such as aggrecan and versican, and the small leucine-rich proteoglycans (SLRPs), such as biglycan and fibromodulin.⁹

TABLE 1.

Musculoskeletal implications of disorders associated with ADAMTS proteases and ADAMTS substrates

ADAMTS protease	Involvement in human disorders	Musculoskeletal phenotype in mouse knockout	Substrates
ADAMTS1	implicated in intervertebral disc (IVD) degeneration	growth retardation without apparent musculoskeletal phenotypes	aggrecan, versican, syndecan 4, TFPI-2, semaphorin 3C, nidogen-1, −2, desmocollin-3, dystroglycan, mac-2, gelatin, amphiregulin, TGF-α, heparin-binding EGF
ADAMTS2	dermatosparaxis Ehlers-Danlos Syndrome (extreme skin fragility, short stature, short hand an feet, joint laxity, craniofacial abnormalities)	dermatosparaxis Ehlers-Danlos syndrome	procollagen type I, II, III, V, Dickkopf-related protein 3, fibronectin, TGF-β RIII, reelin
ADAMTS3	Hennekam lymphangiectasia-lymphedema syndrome	aberrant lymphatic vessel development	procollagen type II, biglycan, LTBP1, fibronectin, TGF-β RIII, reelin, VEGF-C
ADAMTS4	Osteoarthritis (OA) onset, correlated with arthritic severity and progression	not protected from OA	aggrecan, versican, reelin, biglycan, brevican, matrilin-3, α2-macroglobulin, COMP, decorin
ADAMTS5	OA onset, correlated with arthritic severity and progression, implicated in IVD degeneration, suggested role in myoblast fusion	delayed onset of arthritis, decreased arthritic severity and progression	aggrecan, versican, reelin, biglycan, matrilin-4, brevican, α2-macroglobulin, decorin
ADAMTS6	—	—	LTBP1, syndecan 4,
ADAMTS7	tendon maintenance, implicated in rheumatoid and OA	heterotopic ossification in tendons, meniscus, and ligaments	COMP, LTBP3, LTBP4
ADAMTS8	implicated in OA	—	aggrecan
ADAMTS9	nephronophthisis-related ciliopathy (short stature, renal disease) insulin sensitivity in skeletal muscle	soft tissue syndactyly	aggrecan, versican, fibronectin
ADAMTS10	Weill-Marchesani syndrome 1 (short stature, brachydactyly, joint stiffness, hypermuscularity)	shorter long bones, growth plate abnormalities, increased skeletal muscle mass	fibrillin-1, fibrillin-2
ADAMTS12	tendon maintenance, implicated in rheumatoid arthritis and OA	heterotopic ossification in tendons, meniscus, and ligaments	COMP, neurocan
ADAMTS13	thrombotic thrombocytopenic purpura, Upshaw-Schulman syndrome	—	von Willebrand factor
ADAMTS14	—	—	procollagen type I, DKK3, fibronectin, TGF-β RIII
ADAMTS15	implicated in myoblast fusion	—	aggrecan, versican
ADAMTS16	—	—	fibronectin
ADAMTS17	Weill-Marchesani syndrome 4 (short stature, brachydactyly, hypermuscularity)	shorter long bones, growth plate abnormalities, brachydactyly	ADAMTS17
ADAMTS18	microcornea, myopic chorioretinal atrophy and telecanthus, variation in bone mineral density	transient growth delay	—
ADAMTS19	non-syndromic heart valve disease	heart valve malformation	—
ADAMTS20	—	soft tissue syndactyly	versican

Hyalectans are a proteoglycan subfamily defined by their tri-domain structure, with an N-terminus that contains an immunoglobulin-like domain and binds to hyaluronan, a central domain with glycosaminoglycan side chains, and a multifunctional C-terminus that binds to lectins, fibronectin, and other ECM and cell surface proteins.^31,32 The hyalectan family includes four proteoglycans: aggrecan, versican, brevican, and neurocan, and the first two are most efficiently cleaved by several ADAMTS proteases.⁹ Nearly all proteoglycanase members of the ADAMTS family can digest aggrecan (ADAMTS1, 4, 5, 8, 9, 15) and versican (ADAMTS1, 4, 5, 9, 15, and 20), both of which are expressed in musculoskeletal tissues.^33–36 Expression of brevican and neurocan remains primarily confined to the central nervous system. Whereas ADAMTS4 and 5 show brevicanase activity, neurocan proteolysis is associated most with ADAMTS12, a COMP-degrading ADAMTS protease with little prior evidence of proteoglycanase activity.^37–39 The small leucine-rich proteoglycans are a broad selection of 17 short ECM proteins that are structurally defined by a central domain with a variable number of tandem leucine-rich repeats, which is flanked by two cysteine-rich regions.^40,41 SLRPs are marked by functional diversity, endowed in part by N-terminal variability and enormous variability in glycosaminoglycan content. Recombinant ADAMTS4 and ADAMTS5 can cleave biglycan and both can digest bovine fibromodulin, but only ADAMTS4 appears to cleave fibromodulin in human articular cartilage.^42,43

The presence of aggrecan, versican, and a number of SLRPs in cartilage or tendon suggests the importance of ADAMTS protease activity in maintaining the ECM to preserve musculoskeletal health. Substrate activity in the ADAMTS proteoglycanases is perhaps most overtly correlated with onset of osteoarthritis (OA). ADAMTS4 and 5 are critical indicators of arthritis, induced by aggrecan proteolysis, and this discovery was essential to the early characterization of what was then a novel protein family.^44,45 Aggrecan is an integral ECM protein in cartilaginous tissue, particularly in articular cartilage, providing structure and allowing cartilage to resist compression via its massive and charged glycosaminoglycan side chains.⁴⁶ Aggrecan loss can lead to injurious results in the hips and knees, frequently requiring surgical intervention or joint arthroplasty.⁴⁷

OA, the most common form of arthritis, is characterized by degeneration of articular cartilage, due primarily to poorly regulated degradation of ECM proteins.⁴⁸ Aggrecan is one of the first proteins that is degraded in the trajectory of cartilage functional loss, eventually followed by collagen degradation.⁴⁹ The preventative homeostasis of cartilage is disrupted when the balance between production and destruction of the ECM is lost, a possible consequence of improper regulation of proteases that degrade aggrecan and type II collagen.⁴⁸ Inhibition of ADAMTS4 or ADAMTS5 expression can reduce OA onset and joint destruction via attenuation of aggrecan degradation, as shown in human cartilage explants and in vivo.^17,50

There are differences between ADAMTS4 and 5 in arthritic manifestations, particularly in mice, where the ADAMTS4 knockout shows no effect on aggrecan degradation rates or OA progression and severity and ADAMTS5 ablation protects from joint destruction in arthritis.^50,51 These differences may lie largely with their in vivo regulation. While human ADAMTS5 is still 1000-fold more proteolytically potent under physiological conditions, such factors as alternative splicing, promoter activity, epigenetic modifications, and noncoding mRNA regulation may influence comparative phenotypic significance of the two aggrecanases.^52,53 ADAMTS5 activity is reduced by C-terminal processing, whereas ADAMTS4 activity is enhanced by removal of its C-terminal spacer region.^52,53 Across species, the roles and indications of proinflammatory cytokines may account for some phenotypic difference. In human OA synovium, select cytokines such as interleukin (IL-1) and tumor necrosis factor (TNF) α can induce upregulation of ADAMTS4 alone, while ADAMTS5 activity seems to be constitutive.^54–56 Conversely, in mice, expression of murine ADAMTS5 is not constitutive, with potent upregulation by IL-1 and TNFα in models of OA.^50,57

Other ADAMTS proteases with differential aggrecanase activity have varying effects in OA. ADAMTS1 can cleave aggrecan and versican and is expressed in cartilage. Whereas several studies of OA document increased ADAMTS1 expression in human cartilage, other studies describe a reduction in ADAMTS1 protease activity in later stages of joint degeneration.⁵⁸ Since Adamts1 knockout mice were not protected from cartilage erosion in models of inflammatory arthritis, ADAMTS1 may not be a major player in the onset or progression of arthritis.⁵⁹ ADAMTS8 demonstrates weak aggrecanase activity, reported to be as much as 3000-fold lower than ADAMTS4.⁶⁰ ADAMTS8 is expressed in both healthy and OA cartilage, but is detected at increased levels in OA synovium.⁶¹ Based on the requisite C-terminal removal to promote ADAMTS4 activity, studies have probed C-terminal truncation in ADAMTS8 and ADAMTS1, but unexpectedly in both instances, aggrecan degradation decreased.^36,60 Questions of proper truncation position still cloud the effects of C-terminal processing in regulating aggrecan degradation activity and further questions remain about the full substrate profile of ADAMTS8, which along with the profiles of the hyalectanases ADAMTS9, 15, and 20, remains relatively underdeveloped.

3.2 |. COMP-degrading proteases

ADAMTS proteases, other than the hyalectanases that are implicated in arthritis have different substrates with varied molecular indications. Cartilage oligomeric matrix protein (COMP) is the primary substrate identified for ADAMTS7 and 12.^18,19 COMP is an ECM protein that is compromised in pseudoachondrodysplasia, a short stature syndrome.^62,63 The interactions of COMP with collagen fibrils and aggrecan suggest involvement in the assembly and structure of the ECM, perhaps with modulation of cartilage ECM components to allow for loadbearing.⁶⁴ Indeed, COMP has been shown to influence organization and to enhance fibrillogenesis of collagens type I and II by promoting early fibril association.⁶⁵

ADAMTS7 and 12 can bind and degrade COMP in vitro.^18,19 Further validation in cartilage explants and chondrocyte cell models showed that silencing of both proteases with neutralizing antibodies or siRNA led to reduced levels of proteolytic COMP fragments when treated with cytokines that normally induce ADAMTS7 and 12 expression.⁶⁶ Only undigested 524 kDa COMP, but not the ~110 kDa COMP fragment was detected when both proteases were silenced.⁶⁶ COMP is a marker for the presence of arthritis, is elevated in patients at levels that often correlate with disease severity, and proteolytic COMP fragments can be observed in arthritic cartilage, synovial fluid, and serum.^67,68 In patients with OA, these fragments are ~110 kDa in size, similar to the peptides generated by in vitro digestion of COMP with ADAMTS7 and 12.⁶⁶ Gene expression of both of these ADAMTS proteases is also elevated in the cartilage and synovium of patients with rheumatoid arthritis, where COMP levels are correlated positively with disease severity and negatively with cartilage thickness.^18,69

The function of the COMP-degrading proteases in musculoskeletal development extends beyond their indications in arthritis. ADAMTS7 and 12 may regulate tendon development and prevent premature aging, co-expressed to preserve tendon cell identity.¹² Mice express both, ADAMTS7 and 12 in hind limb tendons, skeletal muscle, and knee meniscal fibrocartilage, but not in cartilage or bone, and Adamts7 × Adamts12 double knockout mice suffer from heterotopic ossification. These findings suggest an expedition of later onset chondrogenesis and ossification. This premature state results when prochondrogenic factors prevail beyond the profibrogenic factors like ADAMTS7 and 12 that maintain phenotypic stability of tenocytes.¹²

The developmental role of most ADAMTS proteins is not limited to musculoskeletal tissues. For example, ADAMTS7 plays a distinctive role in the maintenance of vascular integrity. Overexpression of ADAMTS7 in rat results in increased cell migration and proliferation, the consequence of which is significant neointimal thickening.⁷⁰ These results underscore possible correlations with atherosclerosis and restenosis, which involve the hardening, thickening, and narrowing of arteries. ADAMTS7 knockdown in rat via siRNA reduces migration and proliferation of vascular smooth muscle cells, and this reduction can be ameliorated with increased COMP expression, suggesting a role for ADAMTS7 proteolytic activity in COMP degradation.⁷⁰ ADAMTS7 degradation of COMP also mediates vascular smooth muscle cell calcification. A rat model showed that overexpression of ADAMTS7 increases calcium deposition via enhanced degradation of COMP, which is accompanied by an increase in bone morphogenetic protein (BMP) 2 signaling.^71,72 Conversely, knockdown of ADAMTS7 via siRNA leads to attenuated COMP processing, decreased BMP2 signaling, and reduced vascular smooth muscle cell calcification, though again, overexpression of COMP can ameliorate this effect.⁷¹ Why ADAMTS7 would promote calcification in vascular smooth muscle cells but, together with ADAMTS12, protect from calcification in tendon, remains to be clarified. However, it may reflect the consequences of ADAMTS7 cleaving different substrates in vascular smooth muscle cells when compared to tendon, a different role for ADAMTS7 compared to ADAMTS7 co-expressed with ADAMTS12, or possibly differences in the calcification mechanisms in these two tissues.

3.3 |. Procollagen N-propeptidases

Several of the ADAMTS proteases provide essential proteolytic activity in the formation of collagen fibrils, the most abundant component of connective tissues. As procollagen N-propeptidases, ADAMTS2, 3, and 14 extracellularly process the N-terminal propeptide of several procollagens to allow assembly of the processed procollagen into tropocollagen triple helices. These tropocollagen triple helices then join to form collagen fibrils.^73–76 ADAMTS2 seems to be the physiologically relevant procollagen N-propeptidase and can remove the propeptide of the fibrillar collagen types I, II, III, and V, and accordingly shows high gene expression in collagen-rich tissues such as skin, bone, tendon, and the aorta.^23,73 ADAMTS3 can cleave procollagen type II and was proposed to be the protease that excises the N-propeptide of procollagen type II in cartilage.⁷⁴ However, the Adamts3 knockout mouse showed no defect in embryonic cartilage development or procollagen type II processing.²⁴ Instead, the physiological roles for ADAMTS3 appear to be processing of reelin, a large neuronal ECM glycoprotein, and VEGF-C during the formation of lymphatic vessels.^77,78 Adamts3 knockout mice die of lymphedema during embryonic development due to lack of lymphatic vessels.²⁴ In concordance, a mutation in human ADAMTS3 was recently implicated in the third form of a lymphatic disorder, Hennekam lymphangiectasia-lymphedema syndrome (HKLLS) 3.⁷⁷ An autosomal recessive disorder, HKLLS 3 results in lymphedema, lymphangiectasia, and distinct facial features, and the two ADAMTS3 mutations identified in the compound heterozygous individual prevented post-translational processing and inhibited secretion resulting in the loss of ADAMTS3 protease activity.^77,79 The phenotype appears as a consequence of the lack of the necessary proteolytic activation of VEGF-C performed by secreted ADAMTS3.⁷⁷ The fact that patients with HKLLS 3 do not develop characteristic features of dermatosparaxis Ehlers Danlos syndrome (EDS), supports a physiological function for ADAMTS3 other than processing N-propetides from procollagens. If ADAMTS3 plays a role in procollagen processing in postnatal or adult cartilage needs to be established or ruled out. ADAMTS14 is typically co-expressed, though at a lower level, with ADAMTS2 and can process type I procollagen, suggesting possible redundancy.⁷⁵ Polymorphisms in ADAMTS14 have been associated with delayed onset of Achilles tendon pathology and increased risk for OA.^80–82 In a recent study in zebrafish, it was demonstrated that both, ADAMTS3 and ADAMTS14 can process VEGF-C and that inactivation of both proteases was required to elicit a lymphatic phenotype.⁸³ This suggests that, in contrast to mice and humans, ADAMTS14 can cleave VEGF-C and compensate for the loss of ADAMTS3 in zebrafish. .

ADAMTS2 deficiency causes the rare dermatosparaxis form of the collagen-based disorder EDS showing soft and extremely fragile skin, short stature, short hands and feet, craniofacial abnormalities, and mild to severe joint hypermobility.^84,85 In support of the human phenoptype, Adamts2 knockout mice demonstrate skin fragility evident in human dermatosparaxis EDS, but with less severity in the joint phenotype.⁸⁶ While ADAMTS2 gene expression is 30-fold higher in human skin, ADAMTS3 gene expression is near 5-fold higher in human cartilage, and this disparity is thought to account for the reduced phenotypic severity presented in the cartilage of dermatosparaxis EDS patients.⁷⁴ Examinations of skin obtained from dermatosparaxis EDS patients revealed collagen type I deposition in ribbon-like sheets that resemble hieroglyphics attributed to the absence of ADAMTS2 pro-tease activity.^23,87 However, the fact that the ultrastructural appearance of collagen fibrils in arthrochalasia, where the N-propeptide cleavage site in collagen type 1a1 or 1a2 is mutated and the N-propeptide also remains associated with the collagen chains, is different, raises the possibility that the ultrastructural changes observed in collagen fibrils in dermatosparaxis EDS could be induced by a function of ADAMTS2 other than cleaving the N-propeptide of procollagen type I⁸⁸.

3.4 |. ADAMTS10 and 17 in Weill-Marchesani syndrome

Most of the eight ADAMTS proteases, ADAMTS6, 7, 10, 12, 16, 17, 18 and 19, that cluster in the middle clade of the evolutionary tree possibly suggesting an overlap in substrates, were lacking substrates until very recently. However, a surge of publications in the past years identified fibrillin-1 and −2, as well as proteins related to fibrillin microfibril assembly and homeostasis, such as fibronectin, as binding partners and putative substrates for several of these ADAMTS proteases, suggesting that this group of proteases may be involved in regulating fibrillin microfibril formation, function, and/or turnover in the ECM.^20,21,89,90

Identifying the complete substrate profiles of these ADAMTS proteases can broadly clarify their functionality. Conversely, oftentimes investigation of phenotypic data in mice and human disorders can predate substrate elucidation, offering clues to that end as it is the case for ADAMTS10 and ADAMTS17, in which autosomal recessive mutations cause Weill-Marchesani syndrome (WMS).^91,92 WMS can also be caused by dominant mutations in FBN1, the gene encoding fibrillin-1, lending support to the notion that ADAMTS10, ADAMTS17, and FBN1 cooperate in the formation and homeostasis of connective tissues, compromised in WMS.⁹² WMS represent one of several Mendelian disorders associated with fibrillin dysfunction and is phenotypically characterized by short stature, brachydactyly (short fingers and toes), stiffness of the distal joints, hypermuscularity, cardiac valve anomalies, and severe ocular defects, namely spherophakia (small and spherical lenses), and ectopia lentis (subluxation of the ocular lens).⁹³ Fibrillin-1 forms microfibrils in the ECM which lend structural integrity to tissues and organs, and have roles in elastin deposition and elastic fiber homeostasis and the modulation of extracellular transforming growth factor (TGF) β and BMP bioavailability.⁹⁴

The role of ADAMTS10 and its substrate interactions in WMS 1 is partly revealed. The first discovered WMS 1 variants of the ADAMTS10 gene encoded a truncated protein.⁹¹ Some variants resulted in transcripts lacking the catalytic domain, while others included this domain but lacked C-terminal TSR motifs.⁹² Immunogoldlabeling and electron microscopy showed colocalization of ADAMTS10 to fibrillin microfibrils in tissues, and cell culture experiments showed ADAMTS10 binding to fibrillin-1 and −2 and promoting fibrillin microfibril assembly in the ECM.⁸⁹ Interestingly, ADAMTS10 demonstrates limited capability to cleave fibrillin-1 and fibrillin-2 without furin-mediated propeptide removal, which cannot be achieved in wild type ADAMTS10.^20,89 Only restoration of a furin-processing consensus sequence found in other ADAMTS proteases allowed for significant fibrillinase activity. This dichotomy poses a question of proteolysis-independent ADAMTS10 functionality, whether ADAMTS10 is activated in vivo, and whether ADAMTS10 performs proteolysis or promotes microfibril assembly via other mechanisms.⁹⁵ ADAMTS10 was localized to the ciliary zonule, a microfibril-rich cell-free anchorage between the ciliary body and the lens, and an irregularity of microfibril assembly here likely produces the severe ocular complications of WMS.^89,96–98 While ADAMTS10 deficiency in the ciliary zonule in mice resulted in higher immune-reactivity of fibrillin-2 microfibrils postnatally, suggestive of ADAMTS10 proteolytic function, these data could also indicate decreased fibrillin-1 microfibril assembly, which would otherwise mask epitopes in fibrillin-2 microfibrils and reduce their immunoreactivity.^20,89,90,99 In one mouse model of ADAMTS10 deficiency, muscle mass was increased and bone length was reduced; the latter was attributed to a shortened resting zone and an enlarged hypertrophic zone in the growth plate.⁹⁰ Thus, the mechanistic interactions of ADAMTS10 and fibrillin substrates that influence microfibril assembly remain unclear and the question of whether ADAMTS10-mediated proteolysis occurs in vivo remains unanswered.

WMS 4, previously denominated as WMS-like syndrome, is a connective tissue disorder caused by mutations in ADAMTS17.^91,99 WMS 4 manifests in short stature, brachydactyly, hypermuscularity, and the ocular abnormalities characteristic for WMS, but somehow lacks joint stiffness and cardiac valve abnormalities typically associated with classical WMS caused by ADAMTS10 or FBN1 mutations. Due to the rarity of the disorder, the distinction between differences in phenotypic manifestations and the existence of two clinically distinct forms of WMS is difficult to ascertain. In missenses mutations described in individuals and families with WMS4, ADAMTS17 remained in the cell at its perimeter and was absent in the culture medium, suggesting defective secretion.^100,101 Here, fibrillin-1 and collagen type I were both retained in the secretory pathway, leading again to a question of proteolysis-independent activity of ADAMTS17. Fibrillin-1 could be a substrate for ADAMTS17, processed by the protease for deposition in the ECM, or ADAMTS17 could be a chaperone for fibrillin-1, necessary for its secretion, which would account for the co-retention of the two proteins.¹⁰⁰ A recently published ADAMTS17-deficient mouse model showed bone shortening due to alterations in the growth plate and aberrant BMP signaling.¹⁰² In contrast to the ADAMTS10-deficient growth plate, a shortened hypertrophic zone was reported.

3.5 |. Orphan ADAMTS proteases

Remaining orphan enzymes, such as ADAMTS6, 18, and 19, still maintain their status of enigma. Mutations in ADAMTS6 and ADAMTS19 are linked to modulation of cardiac conduction and non-syndromic heart valve disease, respectively.^103,104 However, no substrates have been reported for both proteases. ADAMTS18 shows widespread tissue expression and seems to be involved in eye development as evidenced by mutations in ADAMTS18 that cause Knoblauch syndrome.¹⁰⁵ ADAMTS18 expression is documented in the lung, liver, and kidney of fetal tissues and the brain, prostate, sub-maxillary gland, endothelium, and lens and retina of adult tissues.^105–108 In musculoskeletal tissues, ADAMTS18 is most associated with bone remodeling and it may play a role in regulating bone mineral density. Genome-wide association studies identified associations between single nucleotide polymorphisms in ADAMTS18 and decreased hip bone mineral density across several ethnic groups.¹⁰⁹ Replications studies for three of these polymorphisms among cohorts from the United States, China, and Tobago underscored their associations with bone mineral density.^109–111 Decreased bone mineral density is the characteristic sign of osteoporosis and a major risk factor for bone fractures.

4 |. ADAMTS-LIKE PROTEINS AS MODULATORS OF ADAMTS PROTEASE FUNCTION?

The seven genes that encode ADAMTSL proteins (ADAMTSL1–6 and papilin) present an ensuing question regarding alternative function of ADAMTS proteins beyond proteolytic activity.^96,112 ADAMTSL proteins lack the catalytic domain characteristic for ADAMTS proteases, but contain the central TSR, the spacer domain, and the cysteine-rich domain that constitute the core of the ADAMTS ancillary domain (Figure 2C).⁷ All ADAMTSL proteins possess a C-terminal protease and lacunin (PLAC) domain that is also found in members of the ADAMTS family.^7,113,114 The functions of ADAMTSL proteins remain largely uncharted, particularly regarding any associations with or possible modulation of ADAMTS proteases. One possibility could be that ADAMTSL proteins facilitate or block access of ADAMTS proteases to their substrates due to their similarities in the ancillary domain, which mediates substrates recognition of ADAMTS proteases. However, direct testing of interactions between ADAMTS and ADAMTSL proteins remains absent, and evidence of incidental interactions is sparse, the clearest example being an in vitro observation that Drosophila papilin non-competitively inhibited bovine ADAMTS2.^115,116 Still, the commonality between certain interaction partners for ADAMTSL proteins and ADAMTS protease substrates, particularly fibrillin-1 and other TGFβ regulators, expands the known functionality of ADAMTS proteins beyond proteolytic activity to include modulation of developmental signaling pathways and ECM homoeostasis.^{22,89,117,118} ADAMTSL2, 4, 5, and 6 have all been shown to bind to fibrillin-1 and/or promote fibrillin microfibril assembly.^118–122 In addition, commonality between phenotypes due to mutations in ADAMTS proteases, ADAMTSL proteins, and FBN1, such as in acromelic dysplasias and ectopia lentis suggests linkage in relevant human disorders and extends the demarcation of ADAMTS substrate interactions to possibly include non-cleaved substrates.^96,112 Research of ADAMTSL protein functions suggests predominantly ECM localization and relevance in ECM growth factor regulation.^{117,123–126}

5 |. HOW CAN NOVEL ADAMTS SUBSTRATES BE IDENTIFIED AND VALIDATED?

Under the premise that the biology of (ADAMTS) proteases is determined by the biology and fate of their respective substrates, substrate identification is crucial for elucidating the functionality of orphan proteases, such as ADAMTS18 or 19, and for the identification of all physiological functions for other ADAMTS proteases. The developmental interconnectivity of various proteins, especially throughout the ECM, demands the need for rigorous testing and stringent criteria for validation of candidate substrates. In its most reductive approach, substrate identification begins with exposure of individual ADAMTS proteases to libraries of potential substrates and concludes with bioinformatic analyses to identify putative ADAMTS substrates that must then undergo subsequent validation (Figure 3). Most fundamental to this process are the quality of the proteome subjected to proteolysis and resulting in the degradome and the use of unbiased mass spectrometry (MS)-based proteomics approaches for substrate and cleavage site identification.^127,128 In addition, the repertoire of ADAMTS binding partners that are not direct protease substrates should be considered. Such putative ADAMTS binding partners may mediate substrate binding, modulate protease activity, or may point towards potential non-canonical functions of ADAMTS proteases.

FIGURE 3.

Workflow for the identification and validation of putative substrates for ADAMTS proteases. Substrate identification requires the generation of a degradome from cell culture or tissue sources and its analysis by mass spectrometry. Following substrate identification, validation of putative substrates is required and can include biochemical, cell culture and in vivo approaches or likely a combination thereof. Not all validation steps may be achieved for each individual protease-substrate pair and the physiological relevance for a candidate proteolytic event may not be immediately evident. In addition, major challenges in assigning ADAMTS protease cleavage events in vivo are related to compensation by other ADAMTS proteases and overlapping substrate repertoires. Identification of the ADAMTS interactome can also result in the assignment of putative substrates or point to non-canonical functions for individual ADAMTS proteases

One of the most important considerations on the outset of the quest to identify novel ADAMTS substrates, is the choice of the proteome library to be subjected to proteolysis. The proteome could be derived from cells in culture that secrete and deposit ECM, such as fibroblasts, smooth muscle cells, and others, or the proteome could be extracted from tissues derived from mice or from human biopsies. The protease, on the other hand, can be introduced by adding a purified, active ADAMTS protease or by exogenous overexpression of recombinant pro-teases. Alternatively, expression of endogenous ADAMTS proteases in cells and animals can be modulated by CRISPR/Cas9-mediated activation or inactivation, RNA interference-mediated knockdown, or traditional gene knockout. Despite the critical importance of this initial choice of proteome and protease source for the outcome of the experiment, each approach has its intrinsic challenges, such as the lack of expression or deposition of in vivo substrates in cell culture systems, the challenges to recreate the complex tissue-specific in vivo environment required for ECM deposition in cell culture, the lack of efficient extraction of protease substrates from tissues, which is especially relevant for musculoskeletal tissues, such as bone, cartilage and tendon, the challenges in expressing recombinant active ADAMTS proteases, or the substrate overlap between proteases. Given these shortcomings, the analysis of a combination of proteome sources may be necessary to achieve the ultimate goal to fully define the physiological substrate repertoire for each ADAMTS protease.

The advent of an N-terminomics approach followed the drive to clarify proteolytic activity, which necessarily generates new (neo-) N- or C-termini upon cleavage of substrates.^128,129 N-terminomics involves identification of neo-N-termini in proteolyzed substrates compared to undigested controls via MS-based proteomics approaches. These approaches enable determination of both substrate identity and the possible protease cleavage site(s).¹³⁰ Terminal amine isotopic labeling of substrates (TAILS) emerged as a main N-terminomics protocol used for substrate identification of proteases, and it utilizes a particular isotopic labeling methodology to distinguish cleaved peptides from the uncleaved substrate.¹³¹ A side effect of this approach is that trypsin cleavage is blocked at labelled lysine residues, which may aid in the identification of the longer peptides, but at the same time renders these peptides less susceptible for MS-based proteomic analysis. Importantly, the fact that many internal peptides are labelled can be exploited for peptide quantification. This differential N-terminal labeling approach employs stable heavy and light isotopes. In their original protocol, Kleifeld et al. used heavy (¹³CD₂O) and light (¹²CH₂O) formaldehyde to label protease-treated and control samples, respectively.¹³² More recent developments using isobaric labeling strategies (iTRAQ-TAILS or tandem mass tags) currently allow for up to 16-plexing.^133,134 After digestion with trypsin, many new internal peptides are generated that bind via their amine-reactivity to a hyperbranched polyglycerol polymer with aldehydes (HPG-ALD) and can be depleted by negative selection to reduce the complexity of the degradome prior to MS-based analysis.¹³⁵ MS-based analysis can then characterize the remaining peptides, most of which should represent the native N-termini of the full-length proteins and protease-generated neo N-termini in semitryptic peptides. Subsequent bioinformatics analysis can then identify the peptides and quantitation of relative levels of native N-terminal and neo-N-terminal peptides derived from putative substrates is possible. In addition to TAILS, other technologies can be used to identify ADAMTS protease substrates and cleavage sites, such as combined fractional diagonal chromatography (COFRADIC), the use of subtiligase as an alternative for N-terminal labeling, or approaches to identify C-termini, including the use of LysargiNase or C-terminal COFRADIC.^136–140 Most of these approaches differ in how peptides are generated and how they are enriched, rather than in the downstream MS modality or bioinformatic analyses.

Identification of a putative ADAMTS substrate then requires verification in vitro and in vivo, with three corresponding levels of validation.¹⁴¹ First and most rudimentary is to obtain direct biochemical evidence for proteolysis. This in vitro evidence requires observation of substrate cleavage, ideally after interaction between the purified ADAMTS protease and the purified putative substrate.¹⁴² If no direct cleavage is observed, the possibility of indirect proteolysis remains where the protease under investigation activates a second protease in vivo that then would cleave the candidate substrate. If another unrelated protease cleaves the same substrate at a matching site, assays can include inhibitors and enhancers of that alternative protease to rule out any minor levels due to co-purification. Gel assays with staining such as SDS-polyacrylamide gel electrophoresis can offer validation of cleavage via comparison of the digested and undigested substrate, based on peptide size and migration patterns. Additional validation might include extraction of the cleaved substrate from the gel for direct analysis by MS, MS-based selected and/or parallel reaction monitoring or N-terminal sequencing via Edman degradation.¹²⁸ Use of multiple techniques offers the most comprehensive biochemical evidence for a candidate substrate. However, in many cases the purification of active ADAMTS proteases and/or candidate substrates cannot be easily achieved, and approaches that rely on exposure of the substrate to cells or conditioned medium containing the protease of interest may need to be employed.²¹ The active site mutant form of the respective ADAMTS protease should always be analyzed as a negative control in parallel.

The following two steps of substrate verification adopt a physiologically relevant approach toward protease/substrate analysis. The first step involves removal of proteolytic activity and observation of lack of substrate cleavage or substrate accumulation in cell culture or in tissues. Protease deficiency may be achieved in cell culture systems or model organisms via RNA interference, CRISPR/Cas9 genome editing, or gene knockout, and it aims to demonstrate eliminated cleavage of the candidate substrate in the absence of the protease. Through the analysis of protease-deficient mice, both phenotypic data and ADAMTS protease activity can additionally be correlated. The presence of uncleaved substrate implies that the protease in question is necessary for proteolysis of the candidate substrate in vivo. However, it is important to note, that ADAMTS proteases can compensate for each other, which was demonstrated for ADAMTS4 and 5 in spinal cord injury, ADAMTS9 and 20 in interdigital web regression, and ADAMTS7 and 12 in tendon homeostasis.^{12,13,143,144} Therefore, compensatory upregulation of related ADAMTS proteases in knockout models should be ruled out to unequivocally assign a substrate cleavage event to a specific ADAMTS protease. The interpretation of in vivo findings can be further complicated by the fact that some substrates, including procollagen, aggrecan, versican, and fibronectin, can be cleaved by more than one protease, sometimes even at the same site.^{21,57,145–147} Therefore, candidate substrate cleavage in a specific ADAMTS protease knockout cell line or animal model may still occur and does not necessarily invalidate the prior assignment of this substrate cleavage event to a specific ADAMTS protease.

The third and final step of validation offers the most rigorous proof and extends from the approach mentioned before. Here, the candidate substrate will be modified while keeping protease production intact. Mutating the scissile bond or deleting the site where cleavage occurs can mimic the effects of inhibited proteolysis and would presumably result in a phenotype not unlike that of the protease mutant. This ultimate level of evidence, while challenging to achieve, demonstrates the necessity of substrate cleavage in the presence of the active protease by showing that inhibition of either produces mirrored molecular and phenotypic results.

Several of these techniques of substrate identification were demonstrated in an exemplary study that examined the substrate profiles of the procollagen N-peptidases ADAMTS2, 3, and 14 beyond their canonical procollagen substrates.¹⁴⁸ This investigation was the first application of TAILS toward ADAMTS proteases and the researchers used the secreted proteins of human dermatosparaxis EDS dermal fibroblasts as a proteome in coculture with HEK293 cells expressing recombinant ADAMTS2, 3, or 14 as the protease source. Controls and protease-treated samples were distinguished using isobaric labeling and analyzed using liquid chromatography-MS/MS, leading to the discovery of 8, 17, and 22 candidate substrates for ADAMTS2, 3, and 14, respectively. Unexpected results included detection of a C-terminal cleavage site in collagen III for ADAMTS2 and identification of several TGFβ regulators as putative substrates, namely latent TGFβ-binding protein (LTBP) 1, TGFβ receptor type III, and Dickkopf homolog 3 (DKK3), an antagonist of canonical Wnt signaling. To validate these substrates biochemically, the investigators examined incubation of several putative substrates individually with purified recombinant ADAMTS2 and 14. This biochemical validation was followed by some cell culture-based analyses, where siRNA was used to silence ADAMTS2 expression in human dermal fibroblasts. Gene expression analysis showed that after TGFβ treatment, the level of downstream effector α-smooth muscle actin was reduced where ADAMTS2 silencing occurred, suggesting that ADAMTS2 may typically process TGFβ regulators responsible for sequestering the TGFβ ligand in the ECM to reduce downstream effector activity.

The most challenging in vivo proof of relevant substrate proteolysis was achieved by generating ADAMTS-cleavage-resistant aggrecan and versican knock-in mice.^149,150 Mutation of the canonical ADAMTS5 cleavage site in aggrecan resulted in diminished aggrecan loss in mouse models of OA and even enhanced cartilage repair after acute joint inflammation.¹⁴⁹ When a canonical ADAMTS cleavage site in versican was mutated in mice, some of the mice developed syndactyly, a feature also found in some versicanase deficient mice.^150,151 The versican cleavage-resistant mice showed accelerated wound healing and versican accumulation in the initial phases of wound healing, coinciding with elevated TGFβ signaling.¹⁵⁰ Notably, not all phenotypes previously associated with versican cleavage in ADAMTS-deficient mice could be identified in the versican cleavage-site knock-in mouse, suggesting a role for secondary cleavage sites in versican.¹⁵¹ Both studies showed that mutation of the scissile bond in candidate substrates mimicked the protease-deficient phenotypes in arthritis and wound healing, respectively. Such studies also have the potential to reveal secondary substrate processing sites and thus additional mechanisms of ADAMTS-mediated substrate cleavage and clearance.^150,152

6 |. CONCLUSIONS AND FUTURE STUDIES

Substrate identification is a crucial component of ADAMTS protease characterization, revealing not only the ubiquitous prominence of ADAMTS protease activity, but also the diversity of enzyme functions, including those that may extend beyond proteolysis. The ADAMTS protease family is large, with a variety of validated and putative substrates among the 19 family members. Though many ADAMTS proteases, including members of the proteoglycanases, procollagen N-propeptidases, and COMP-degrading proteases, are implicated in such well-characterized musculoskeletal pathologies as OA or rheumatoid arthritis, others are associated with rare Mendelian disorders such as ADAMTS3 in HKLLS3 or ADAMTS10 and ADAMTS17 in WMS. The substrates in these latter examples include not only structural components of the ECM but also key regulators in essential signaling pathways, including members from the VEGF and TGFβ families. Then, there remain several orphan enzymes, such as ADAMTS19, for which substrate identification has not yet commenced. Substrate identification among all of these proteases can uncover key interactions in biological pathways to offer insights on tissue development and tissue homeostasis.

Since there is no reported consensus sequence for ADAMTS cleavage sites, protein sequence-based substrate identification is not feasible. Therefore, protocol refinement and expansion of the degradomics tool kit can provide greater efficiency in identifying the substrate profiles of both orphan and non-orphan members of the ADAMTS family. iTRAQ-TAILS represents a substantial refinement within the proteomics catalogue. As described in the aforementioned study, TAILS technology revealed the unexpected insight of associations between procollagen N-peptidases and TGFβ regulation when overviewing their substrate profiles. In a recent publication, TAILS was utilized to identify LTBP3 and 4 as proteolytic substrates of ADAMTS7; the LTPB4 cleavage sites were in a linker domain that bridged fibrillin-1 binding and fibulin-5/tropoelastin binding.¹⁵³ Accordingly, a key future direction may be investigation of ADAMTS activity in modulation of TGFβ signaling, possibly involving fibrillin microfibrils or LTBP complexes. While substrate profiles are highly individualized, ADAMTS proteases and ADAMTSL proteins currently associated with the TGFβ pathway, either through proteolysis or protein-protein interactions, include ADAMTS2, 3, 6, 7, 10, 14, and 17, and ADAMTSL 2, 4, 5, and 6, and relevant substrates include LTBP1, 2, 3, and 4, TGFβ RIII, DKK3, fibrillin-1 and fibrillin-2. These discoveries expand conceptions of ADAMTS substrate profiles beyond the subcategories currently assigned and demonstrate the potential of overlap within the ADAMTS family. In addition to MS-based degradomics approaches, other modes of substrate identification can also be valuable in revealing candidate substrates that can be validated biochemically or in vivo. One such approach is yeast-two-hybrid screening using the ancillary domains of ADAMTS proteases as baits. Under the premise that the ancillary domain mediates substrate recognition, proteins that interact with the C-terminus of ADAMTS proteases are potential substrates and can be verified and validated in vitro and in vivo.¹⁴⁵

Continued study of ADAMTS proteases and their substrate activity must not be confined along a singular avenue lest they overlook large sectors of biological activity. Both proteolytic and alternative functions must be considered, and the substrate identification for orphan ADAMTS proteases remains significant. Examination here may follow lines of homology between these proteases and those for which substrates have already been characterized. All findings require stringent verification to reach evidential validation, and in addition to biochemical evidence, these criteria should involve in vivo examinations, beginning with ADAMTS protease knockout and ideally concluding with modification of the substrate cleavage site to show phenotypic results mirroring those of protease inhibition. We anticipate that new technologies and the expanded use of CRISPR/Cas9 gene editing, specifically to introduce active site mutations in ADAMTS proteases and mutations of scissile bonds in substrates, will allow for a more efficient validation processes. ADAMTS protease substrate identification is a progress of persistence. Continued study of ADAMTS proteases and their physiological substrates will greatly enhance our understandings of musculoskeletal development, elucidate the molecular origins of diseases, and uncover new insights into biological development and tissue homeostasis, from embryogenesis to adulthood.