Skip to main content
  • Loading metrics

Phylogenetic inference of the emergence of sequence modules and protein-protein interactions in the ADAMTS-TSL family

  • Olivier Dennler,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Software, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliations Univ Rennes, Inria, CNRS, IRISA, UMR 6074, Rennes, France, Univ Rennes, Inserm, EHESP, Irset, UMR S1085, Rennes, France

  • François Coste,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Univ Rennes, Inria, CNRS, IRISA, UMR 6074, Rennes, France

  • Samuel Blanquart,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliation Univ Rennes, Inria, CNRS, IRISA, UMR 6074, Rennes, France

  • Catherine Belleannée,

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Validation, Visualization, Writing – review & editing

    Affiliation Univ Rennes, Inria, CNRS, IRISA, UMR 6074, Rennes, France

  • Nathalie Théret

    Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing

    Affiliations Univ Rennes, Inria, CNRS, IRISA, UMR 6074, Rennes, France, Univ Rennes, Inserm, EHESP, Irset, UMR S1085, Rennes, France


Numerous computational methods based on sequences or structures have been developed for the characterization of protein function, but they are still unsatisfactory to deal with the multiple functions of multi-domain protein families. Here we propose an original approach based on 1) the detection of conserved sequence modules using partial local multiple alignment, 2) the phylogenetic inference of species/genes/modules/functions evolutionary histories, and 3) the identification of co-appearances of modules and functions. Applying our framework to the multidomain ADAMTS-TSL family including ADAMTS (A Disintegrin-like and Metalloproteinase with ThromboSpondin motif) and ADAMTS-like proteins over nine species including human, we identify 45 sequence module signatures that are associated with the occurrence of 278 Protein-Protein Interactions in ancestral genes. Some of these signatures are supported by published experimental data and the others provide new insights (e.g. ADAMTS-5). The module signatures of ADAMTS ancestors notably highlight the dual variability of the propeptide and ancillary regions suggesting the importance of these two regions in the specialization of ADAMTS during evolution. Our analyses further indicate convergent interactions of ADAMTS with COMP and CCN2 proteins. Overall, our study provides 186 sequence module signatures that discriminate distinct subgroups of ADAMTS and ADAMTSL and that may result from selective pressures on novel functions and phenotypes.

Author summary

The vast majority of proteins have a modular multidomain organization. Domains are conserved building blocks of proteins that are widely used to characterize and predict protein functions. However, the organization of multidomain proteins underlies a great complexity and their biological role is generally not the sum of the functions attributed to each domain. To address this problem, it is necessary to develop new methods to better identify functional signatures. Here, we developed a framework based on partial local multiple alignments (to find conserved sequence modules) and phylogenetic reconciliation methods (to integrate the evolution of species, genes, sequence modules, and Protein-Protein Interactions). Application of our framework to the search for functional sequence modules in extracellular matrix proteins from the ADAMTS (A Disintegrin-like and Metalloproteinase with ThromboSpondin motif) and ADAMTSL (ADAMTS-like) family, enabled to highlight sequence signatures potentially involved in Protein-Protein Interactions.


The ADAMTS (for “a disintegrin-like and metalloproteinase with thrombospondin motif”) and ADAMTSL (for ADAMTS-like) are secreted proteins involved in extracellular matrix remodelling. They are associated with a wide range of human diseases, including cancer, fibrosis, arthritis, and cardiovascular disease [1, 2]. These homologous multidomain proteins are present in all metazoan species. The number of paralogs varies considerably from 4 genes in C. elegans to 26 genes in H. sapiens, suggesting that the ADAMTS-TSL protein family undergoes a large expansion in the number of paralogs during metazoan evolution. ADAMTS-TSL proteins are usually described according to their domain/motif composition as shown for the human paralogs in Fig 1. ADAMTS sequences are composed of two regions. The proteinase region includes a signal peptide, a pro-peptide, a catalytic domain (metalloproteinase) and a disintegrin domain. The ancillary region contains a cystein-rich domain, a spacer region, a variable number of thrombospondin type 1 motifs (TSP1) and additional domains/motifs. Until now, ADAMTS proteins have been mainly studied as proteolytic enzymes and classified according to their substrates: the procollagenase (ADAMTS-2, -3, -14), the hyalectanases (ADAMTS-1, -4, -5, -8, -9, -15, -20) and ADAMTS-13 which specifically cleaves the von Willebrand factor. The remaining ADAMTS, ADAMTS-6, -7, -10, -12, -16, -17, -18, -19, have long been considered as orphan enzymes, however following the identification of numerous substrates over the last decade they have been grouped as fibrillin/fibronectin-associated ADAMTS proteases [3]. Unlike ADAMTS, ADAMTS-like proteins lack the proteinase region but their ancillary region is similar, sharing notably the first TSP1 motif, the cystein-rich domain and the spacer region. Both ADAMTS and ADAMTSL contribute to microfibril formation through interaction with matrix components [4].

Fig 1. Domain/motif organization of the 26 human ADAMTS-TSL paralogs, adapted from [5].

Because of their complex structure and their biochemical properties, the functional characterization of ADAMTS-TSL using classical experimental approaches remains a challenge [6]. In addition, performing specific predictions using computational methods is complicated by the lack of structural data, the complex multidomain organization of these proteins and the presence of paralogs. To date, functional predictions mainly consist in recognizing known conserved domains/motifs thanks to pre-computed signatures available in databases such as PROSITE [7] or Pfam [8]. While this approach provides a first approximation of ADAMTS-TSL functions, it fails to account for specific activities. For instance, the closely related paralogs ADAMTS-8 and ADAMTS-5 share the same domain organization but ADAMTS-5 cleaves aggrecan/versican highly efficiently [911] while ADAMTS-8 does not [12]. These observations suggest that more refined methods are needed to identify the specific sequence regions associated with the functional specificity of ADAMTS-TSL proteins. Instead of relying on external generic domains/motifs, we propose here to consider all local conservation of sequences within the ADAMTS-TSL family as potential evolutionary traces of their functional significance. The detection of all the conserved loci from at least two proteins of the family is made possible by partial and local multiple alignment of their sequences. Introduced by the Protomata suite [13, 14], this kind of alignment is composed of sets of local alignments involving each a subset of the sequences to align. These local alignments enable the identification of all the conserved sequence modules maintained during evolution, at least in a subset of sequences, by selective pressures, and thus likely to be important functionally. To identify which of these modules are associated with specific functions, we propose here to use phylogenetic inference methods to reconstruct ancestral histories of both the modules and the functions, and to predict their co-appearance along the ADAMTS-TSL evolutionary tree. The consideration of both sequence similarities and the evolutionary history of genes to propagate functional annotations along their phylogenetic tree was pioneered by Eisen et al. [15]. Subsequent studies focused on the inference of ancestral functions from the known functions in the extant genes, and more elaborated parsimonious, Bayesian or maximum likelihood evolutionary models of functional change over time have been proposed since [1621]. These models assume that ancestral traits are likely to be shared by descendants and allow the reconstruction of an evolutionary scenario of acquisition and loss of function from scattered observations. Recently, phylogenetic methods have been proposed to estimate multi-paralog and multi-domain gene family histories [2224]. These methods infer the so called reconciliation of the species, gene and domain phylogenies and enable to estimate ancestral domain compositions.

Here we present an innovative approach that relies on conserved sequence modules instead of domains. We apply the above mentioned methods to estimate the co-appearance of functional traits, i.e. Protein-Protein Interactions (PPIs), with conserved sequence modules, given the ADAMTS-TSL gene phylogeny. In a given ADAMTS-TSL ancestor, all the gained ancestral modules form a signature that is possibly involved in a gained ancestral function. Applying this new approach to ADAMTS-TSL proteins enabled us: 1) to identify 186 sequence module signatures that discriminate distinct sub-groups of ADAMTS-TSL; 2) to infer 45 signatures associated with the emergence of 278 PPIs in 45 ADAMTS-TSL ancestors; 3) to identify non-trivial convergent interactions of ADAMTS-TSL with COMP and CCN2 proteins; 4) to observe that module changes during evolution mainly occurred in the N-terminus part and in the ancillary domain, highlighting the importance of both regions in the divergence of ADAMTS-TSL specificities, and 5) to characterize ADAMTS-5 by a sequence signature that might explain its functional specificity with respect to other hyalectanases.

Materials and methods

The general workflow of our new approach is summarized in Fig 2. We detail in this section the different steps for identifying functional sequence modules in ADAMTS-TSL proteins.

Fig 2. Phylogenetic inference of module and phenotype appearances.

The different steps of the method, illustrated here for a dummy set of sequences containing two paralogs p1 and p2 (from one species) and their ortholog p3 (from another species), are: 1) Inference of the reference gene tree from protein sequences by a standard pipeline (PASTA, RAxML, TreeFix); 2) Identification of conserved sequence modules (i.e. sets of strongly similar segments from at least 2 protein sequences aligned in PLMA blocks by Paloma-D); 3) Inference of the module composition of ancestral genes in the reference tree (through Module-Gene-Species reconciliation by SEADOG-MD using the phylogenetic tree of each module inferred with PhyML and TreeFix); 4) Annotation of proteins with known phenotypic traits of interest (here Protein-Protein Interactions); 5) Reconstruction of the ancestral scenario of phenotype evolution across the reference gene tree (PastML); 6) Merging module and phenotype evolutionary information: each ancestral gene of the reference gene tree is then characterized by a module composition and a set of phenotypic traits (protein interactants here). The final result is the prediction of functional signatures by identification of module(s) and phenotypic trait(s) co-appearance.

Species selection and sequence retrieval

To build the set of input sequences, we selected 9 species: Homo sapiens, Mus musculus, Bos taurus, Gallus gallus, Xenopus tropicalis, Danio rerio, Ciona intestinalis, Caenorhabditis elegans, and Drosophila melanogaster, because they are representative of the evolutionary history of the ADAMTS-TSL family from the bilateria ancestor to Homo sapiens and they have complete proteome and genome annotations in the Gene Feature Format (GFF) datasets (S1 Table). Providing the whole proteomes of the nine species as input, we used the Orthofinder software [25] (v2.5.2, default settings) to cluster the sequences into 31,229 orthogroups. An orthogroup is defined as a group of sequences from different species descending from a common ancestor gene. We identified 22 orthogroups containing at least one of the 159 RefSeq sequences of known human ADAMTS (102) and ADAMTSL (57) proteins, 17 orthogroups corresponding to ADAMTS proteins and 5 to ADAMTSL proteins. We also selected the 2 orthogroups corresponding to 7 RefSeq sequences of 4 human ADAM genes to use them as an outgroup enabling to root the ADAMTS-TSL phylogeny. Like ADAMTS-TSL, the ADAM (for A Disintegrin And Metalloprotease) proteins belong to the adamalysin family. The resulting dataset, called dataset-708, is composed of 708 protein sequences (386 ADAMTS, 226 ADAMTSL, 96 ADAM) from 24 selected orthogroups and is detailed in Table 1.

Table 1. Distribution of protein sequences among the nine species.

The dataset-708 contains all sequences of the 24 selected orthogroups (several isoforms for a single gene). The dataset-214 contains one representative sequence per gene (described in S3 Table).

The orthogroups can include several isoforms from a single gene, however functional annotations are usually available at the gene level only. We thus chose to filter out isoforms in order to select one representative protein sequence per gene per species. In a given species, the products of a gene can be defined as a group of isoforms whose loci overlap on the genome (i.e. they share exons). Each protein sequence is associated with its locus according to its coordinates as described in the GFF file. In a species, two protein loci overlapping by at least one residue on the same genome strand were considered as two isoforms encoded by the same gene. The longest isoform sequence of a gene was then selected as the representative protein without significant loss of information as shown in S2 Table. The resulting dataset, called dataset-214, contains 214 representative proteins from 214 distinct genes over the 9 species: 173 ADAMTS-TSL (125 ADAMTS, 48 ADAMTSL) plus 41 ADAM. All the 26 H. sapiens ADAMTS-TSL genes are represented in the dataset (see Table 1).

Inference of the reference gene tree (step 1)

The gene phylogeny, starting point of our method, was inferred from the 214 representative protein sequences in dataset-214.

As a first step, we used the multiple sequence aligner PASTA [26] (v1.8.5, default settings) to compute the multiple alignment of the 214 proteins and the maximum likelihood phylogeny inference software RAxML [27] (version provided with PASTA, default settings) to infer a first initial phylogeny. This initial tree was corrected using TreeFix [28] (v1.1.10, 100 iterations with the PROTGAMMAJTT model) taking into account the species tree defined in the NCBI taxonomy, thus possibly changing the topology of the initial tree. The branch lengths (lost by TreeFix) were then recomputed using PhyML [29] (v3.3.20190909, optimizing only branch lengths and substitution rate parameters with amino acid default settings) and the tree was rooted using ADAM proteins as the outgroup, resulting in the final phylogeny that we named the reference gene tree. To analyse the uncertainty resulting from the taxon and gene sampling, and to identify robust bi-partitions in the reference gene tree, we analyzed several ADAMTS-TSL protein samples by varying the content of the outgroup and by adding ADAMTS-TSL sequences of species not considered in our 9 species (see S1 Appendix).

Identification of conserved sequence modules (step 2)

The strongly conserved sequence modules in ADAMTS-TSL proteins were identified by partial local multiple alignment of the 214 sequences in dataset-214.

Introduced by the Protomata suite [13, 14], a partial local multiple alignment (PLMA) consists of a set of alignment blocks containing each a set of locally aligned segments from some, but not necessarily all, of the given sequences (as in partial order alignments [30], but with the addition of a locality property, see Fig 3). The Protomata suite provides the Paloma program to build a PLMA by iterative selection and integration of pairwise local alignments under compatibility constraints. To efficiently process the large number of long and similar sequences available, we used Paloma-D (v0.1), a new and faster re-implementation in modern C++ of Paloma (see S2 Appendix).

Fig 3. Identification of modules by partial local multiple alignment.

We show here a schematic PLMA of sequences S1, …, S5 composed of the alignment blocks B1, …, B6. This example illustrates the locality of the alignments: each alignment of positions is supported by a local alignment of sequences, leaving possibly other sequence positions unaligned. For instance the alignment block B2 is local since it aligns only the two segments M2.3 and M2.4 of the sequences S3 and S4, and no other positions of these sequences. Local alignments authorize to align only subsets of adjacent positions from each sequence. In an orthogonal way, partial alignments authorize to align positions of only a subset of the sequences. This is also illustrated by B2 which is partial since it aligns only positions from S3 and S4 (and does not even align them to positions of the block B1 above B2). Partial local alignments are not limited to pairwise alignment: the block B3 is an example of partial local alignment block, aligning the segments M3.2, M3.3 and M3.4 from sequences S2, S3 and S4, that can be built from the pairwise local alignments (highlighted here in blue colors) of the segments M3.2 with M3.3, M3.3 with M3.4 and M3.2 with M3.4. The PLMA blocks align each a set of segments conserved specifically in a sequence subset. This set of segments, provided that they are long enough, is said to be a conserved sequence module. Let us assume here that all the blocks, except B6, align segments of 5 or more residues. The set of segments {M2.3, M2.4} aligned in B2 defines then for instance the module M2, while the block B3 enables to identify the module M3 = {M3.2, M3.3, M3.4}. Blocks B4, B5 and B6 illustrates how the definition of the blocks –requiring that each segment is aligned to, and only to, all the other segments of the block– enables to split possibly longer local alignments to identify segments specifically conserved in sequence subsets: even if the concatenation of M4.1 and M5.1 is locally aligned to the concatenation of M4.3 and M5.3 (this could be for instance a pairwise alignment used to build the PLMA), none of these two concatenations is locally aligned to M5.2. In this case, the maximal set of segments aligned with M5.2 is {M5.1, M5.3, M5.2} and the modules are here M4 = {M4.1, M4.3} specifically conserved in {S1, S3} and M5 = {M5.1, M5.3, M5.2} specifically conserved in {S1, S3, S2}. Similarly, the block B6 aligns the segments S6.3 and S6.2 but, in contrast with the segments aligned by B4, these segments are shorter than 5 residues and do not define a module.

The PLMA of the 214 sequences contained in the representatives.fasta file was computed with the command line paloma-d -i representatives.fasta -M 20 -t 10. The alignment procedure is detailed in S2 Appendix. With these parameters, the candidate pairwise local alignments were all the diagonals (i.e. gap-free pairwise local alignments) computed by the program dialign2–2 [31], of length smaller or equal to 20 residues and with a strong similarity weight of 10 or more (corresponding to similarities of probabilities lower than 5 × 10−5 in random sequences). The multiple sequence alignment was built from this set of pairwise alignments of interest as follows. First, the candidate diagonals were sorted by the default heuristic of Paloma-D, putting forward those with the stronger evidence of conservation across all the sequences. Then, they were iteratively added in this order to the PLMA, with transitive alignment and local extension of diagonals, if and only if the consistency of the resulting induced alignment [32] was preserved.

The conserved sequence modules were identified from the alignment blocks of the resulting PLMA. Alignment blocks are the maximal alignment of segment units verifying that each segment of a block is aligned by the PLMA over its full length to, and only to, all the other segments of the block. Let us remark that block segments can thus be shorter than the pairwise local alignments used to build the PLMA. With the chosen parameters, each PLMA block aligned without gaps a set of segments of same length which were strongly conserved in one, and only this, subset of the 214 sequences. We defined the conserved sequence modules to be all the sets of segments that were aligned by these PLMA blocks, provided that their length was greater than 5 residues (see Fig 3). The minimal length of 5 residues was required here to enable us inferring the module’s phylogenies in the following step.

Inference of the module composition of ancestral genes (step 3)

The step 3 consists in building the evolutionary history of the modules, according to the reference gene tree.

First, the phylogenetic tree of each module was inferred from the gapless alignment of its segments (as in the corresponding alignment blocks from step 2) by Maximum Likelihood software PhyML [29] (v3.3.20190909, amino acid default settings). Then, each module tree was corrected by TreeFix [28] (v1.1.10, 100 iterations with the PROTGAMMAJTT model) with respect to the reference gene tree. Of note, the final topology of the module tree might still be different and incompatible with the one of the reference gene tree. Then the module tree was mapped on the reference gene tree using the phylogenetic reconciliation software SEADOG-MD [24] (default options) with the help of the species tree (the evolutionary tree according to NCBI Taxonomy [33]). This resulted in mapping each internal node of the module tree to an internal node n of the reference gene tree, enabling us to infer an occurrence of an ancestral segment of this module at the ancestral gene n.

Mapping all the modules, each ancestral gene node n of the reference tree was assigned an ancestral module composition , so the reference gene tree is annotated with the module composition. It represents the inferred ancestral evolution of module composition.

Collecting Protein-Protein Interactions as phenotypic traits of interest (step 4)

The step 4 consists in collecting functional traits associated with sequences.

Because of their importance in the functions of proteins, we chose Protein-Protein Interactions (PPIs) as the phenotypic trait of interest. The conversion of the RefSeq ID of the 386 ADAMTS and 226 ADAMTSL sequences from dataset-708 enabled us to retrieve 494 UniprotKB ID that were used to query molecular interaction databases using the Proteomics Standard Initiative Common QUery InterfaCe (PSICQUIC) [34]. The retrieved file contained 1985 lines (S4 Table) including interactions between proteins of the same species (e.g. 1685 for H. sapiens, 28 for M. musculus and 106 for C. elegans) and interactions between proteins from different species (e.g. 65 for H. sapiens/M. musculus). The 66 interactions involving ADAMTS-TSL and complexes are excluded because of the lack of protein identification from complexes. After filtering redundant identifiers, 461 different PPIs were annotated (S5 Table). They included 395, 13 and 36 PPIs for H. sapiens, M. musculus and C. elegans, respectively. Analysis of the data in mice unsurprisingly shows redundancy with humans. Most C. elegans PPIs (32/36) involve papilin (mig-6) and the remaining four PPIs involve madd-4 (orthologs of ADAMTSL1 and 3) and adt-2 (ortholog of ADAMTS2), but there is no PPI shared with human. In addition, 8 interactions between human and mouse proteins are redundant with each species and the remaining 8 PPIs involve human proteins and proteins from other organisms such as viruses and bacilli. In conclusion, sequence queries of the nine species yielded very little information about the non-human species and we chose to retain only the PPIs from H. sapiens in the PSICQUIC results. Importantly, we observed that many recently identified substrates of ADAMTS were not described in molecular interaction databases and we decided to complete the list of PPIs by manual curation of the literature. Our review of the literature enabled us to identify an additional set of 303 PPIs (S6 Table). In this dataset, we included high throughput data from the iTRAQ-TAILS method that was used to identify novel substrates for the procollagenases ADAMTS2,-3 and -4 [35, 36] and the ADAMTS7 protein [37]. We also included four PPIs involving LRP1 which is not a substrate but interacts with ADAMTS4,-5,-9,-10 to mediate ADAMTS internalization and thus regulate their activities [38, 39]. Merging data from PSICQUIC queries and manual curation resulted in a PPIhuman set of 662 non-redundant documented interactions between one of the 26 H. sapiens ADAMTS-TSL representative protein and one of their 471 H. sapiens protein interactants (S7 Table).

Reconstruction of the ancestral scenario of phenotype evolution (step 5)

The step 5 consists in building the evolutionary history of the PPI, according to the reference gene tree.

Ancestral ADAMTS-TSL Protein-Protein Interactions along the reference gene tree were inferred with PastML (v 1.9.33, default setting) [20]. Three kinds of status can be given as input to PastML on the ability of a protein at a leaf of the tree to interact with each partner: “able to interact” (1), “not able to interact” (0) or “missing annotation” (?); each missing annotation being later inferred by PastML as able to interact (1) or not able to interact (0). As stated in step 4, we focus on H. sapiens PPIs. We set the status of H. sapiens representative protein to “able to interact” (1) for each interacting partner listed in PPIhuman and “not able to interact” (0) for the other human proteins. Given the very limited amount of information on PPIs in non-human species, we chose to set the status at the leaves to “missing annotation” (?) for each PPI.

Given these leaf annotations and the reference gene tree, PastML associated to each node n (internal or leaf) with missing annotation a set of interactants . Let us note that PastML proposes different coherent evolution scenarios based on marginal probabilities of character (PPI presence in this study) that were also recorded. Here we focus only on the proposed most likely scenario.

The reference gene tree annotated with ancestral interactants at each node represents the inferred scenario of evolution of ADAMTS-TSL interactions with proteins.

Prediction of functional signatures (step 6)

By integrating the module compositions (from step 3) and ADAMTS-TSL PPI annotations (from steps 5) onto the reference gene tree, we can simultaneously examine the inferred evolutionary histories of genes, module compositions and ADAMTS-TSL protein interactions.

To focus on changes across the evolution of ADAMTS-TSL genes, we computed for each node n of the reference tree, the sets and of the gained and lost modules of the gene at n with respect to its ancestor at node a. A module was added to if it was in and not in . Conversely, it was added to if it was in and not in . The sets and of gained and lost PPI of the gene at node n were computed similarly from .

The set of modules gained at a node n, , was said to define the module signature of the node. The more significant module signatures are those co-appearing with a phenotypic trait at a node. Then we focused on the co-appearance of protein interactions and modules, i.e. on nodes n such that and were both non-empty. In that case, the set of modules was identified as a candidate functional module signature of the interaction of the ADAMTS-TSL with the members of . Note that both modules and interactions are shared by some, but not necessarily all, the descendants of the ancestral gene n.

Visualization of the labeled phylogenetic tree

An important contribution of this work is the availability of all the data with an interactive tree (automatically generated at the end of the pipeline), using the Interactive Tree Of Life software, Itol [40]. The ADAMTS-TSL Itol tree is available at An HTML popup window provides detailed information about each gene node, including the name of the gene, the nature of the gene (ancestor or leaf), the PPIs associated with the gene and PPI gain/loss with respect to the ancestor, the module composition of the gene and module gain/loss with respect to the ancestor. The Itol tree also provides annotations including the number, the composition and the transfer of modules, the domain composition and the presence of PPIs. In particular, each gene has its own module signature annotation in which “gained” modules are represented by green boxes on the descendants. All module signatures can be visualized within their domain contexts by first enabling the domain composition annotations. All the inferred module signatures for the ADAMTS-TSL family are available at (see S3 Appendix for screenshots and examples of usage).


Evolutionary histories of ADAMTS-TSL genes, modules, and PPIs

To explore the ADAMTS-TSL evolutionary history according to their diverging module compositions and PPIs, we first inferred a phylogenetic tree from ADAMTS and ADAMTSL protein sequences and rooted the tree using some ADAM protein sequences as the out-group. This reference gene tree is shown in Fig 4. We identified seven monophylies including the procollagen peptidases (ADAMTS-2, ADAMTS-3, ADAMTS-14), the hyalectanases (ADAMTS-1, ADAMTS-4, ADAMTS-8, ADAMTS-15, ADAMTS-5, ADAMTS-20, ADAMTS-9) and four groups associating pairs of ADAMTS, ADAMTS-7 and ADAMTS-12, ADAMTS-6 and ADAMTS-10, ADAMTS-16 and ADAMTS-18, ADAMTS-17 and ADAMTS-19. Note that the ADAMTSL form a separated clade that does not include papilin proteins.

Fig 4. Reference gene tree of the 125 ADAMTS, 48 ADAMTSL and 41 ADAM outgroup members (figure produced with Itol).

We then inferred the evolutionary history of the gene module composition. Using Paloma-D, the partial local alignment of the 214 sequences identified 1059 conserved sequence modules (S9 Table). As expected, some of these conserved sequence modules are shared by all sequences, while others are shared by different subsets of sequences. In particular, we identified two modules (covering a total length of 33 amino-acids) that were shared by all 26 human ADAMTS-TSL sequences and localized in the cysteine-rich and spacer domains, which are shared domains of all ADAMTS-TSL. In addition, we found 11 other modules (covering a total length of 111 amino-acids) that were specifically shared by the 19 human ADAMTS sequences, located in the catalytic domain, disintegrin, central TSP1, and cysteine-rich domains. These domains were already identified as shared among ADAMTS, which support our Paloma-D module decomposition at the ADAMTS-TSL family level. The module composition of the 26 H. sapiens ADAMTS-TSL sequences is detailed in Fig 5. In the inferred ancestral module composition, 186 ancestral gene nodes of the ADAMTS-TSL tree were associated with the gain of one or more modules, defining as many module signatures.

Fig 5. Module composition of the 26 H. sapiens ADAMTS-TSL sequences.

The phylogenetic gene tree of the 26 H. sapiens ADAMTS and ADAMTSL paralogs was extracted from the reference gene tree (Fig 4). The modules identified by the Paloma-D program are represented on the sequences with Itol, using a unique combination of form and color to designate each module. The complete list of modules is provided in the S8 Table.

In parallel, we investigated the evolution of PPIs involving the ADAMTS-TSL proteins and other proteins. A first result is the identification 662 PPIs involving at least one H. sapiens ADAMTS-TSL and one of their 471 H. sapiens partners. Among the 471 partners, 119 have interactions with at least 2 ADAMTS-TSL (Fig 6). Using the PastML tool, we inferred 471 ancestral scenarios representing the PPI evolution given the ADAMTS-TSL gene tree. We identified 48 ancestral genes in which one or several PPIs most likely appeared. These nodes indicate potential gain of function.

Fig 6. Protein-Protein Interaction networks of H. sapiens ADAMTS-TSL.

The 119 PPIs shared by the 26 human ADAMTS-TSL are visualized with Cytoscape [41]. Yellow nodes are hyalectanases, green nodes are pro-collagenases, grey nodes are ADAMTS with unspecific substrates, blue nodes are ADAMTSL and white nodes are proteins interacting with ADAMTS-TSL.

Conserved sequence module signatures associated with PPIs

Finally, we integrated in the reference gene tree the estimated evolutionary histories of the modules and the PPIs. As a result, only 3 ancestral gene nodes (G149, G217, G329) exhibited one PPI but no module gain and 45 ancestral gene nodes exhibited both module and PPI gains (Table 2). Among these 45 ancestral gene nodes, 5 were the ancestors of at least two paralogs (nodes corresponding to paralogous duplication events: G91, G96, G161, G235 and G341), while 40 were only the ancestors of orthologs (nodes corresponding to speciation events). These 45 nodes correspond to module-PPI co-appearance events and highlight gene ancestors in which a new PPI has emerged and from which ancestor segment(s) of gained module(s) have been conserved by evolution. Based on these co-appearance events, we proposed 45 module signatures (including 355 modules) as candidates for sequence regions likely involved in 278 PPIs (as shown in Table 2). The module signatures contain from 1 to 26 modules while the PPI signatures contain from 1 to 143 PPIs. Some ancestors have few PPIs associated with few modules which make them much easier to interpret. This is the case for the ancestral gene G96 for which two PPIs (ACAN, VCAN) are associated with two modules (B773, B783). This is also the case for the ancestral gene G315 for which one PPI (CCN2) is associated with the B1424 module. In contrast, the ancestral gene G341 is characterized by the association of numerous PPIs (66) and modules (22). In addition, we observed the co-appearance of multiple modules for a single PPI such as G91 for which the interaction with LRP1 is associated with 26 modules. Note that some PPIs and modules are involved in multiple co-appearances. A possible interpretation is that these PPIs were acquired multiple times during the evolution of ADAMTS-TSL, suggesting convergent evolution.

Table 2. The 45 events of module(s)-PPI(s) co-appearance.

G91, G96, G161, G235 and G341 are ancestral genes of at least 2 paralog gene.

Next, we looked at the module location of the 45 signatures along the human protein sequences descending from the corresponding ancestor genes (Fig 7). Of note, all the modules of the signatures are not necessarily found in human proteins, because some could have been subsequently lost or could have diverged in human. Over the 355 modules involved in the 45 signatures, 283 were present in the human descendants. Based on this view of the module signatures associated with PPI co-occurrence events, we observed a wide diversity in the distribution of modules that were gained either within, overlapping with, or outside the known classical domains. Moreover we showed that the variable ancillary region (colored in purple in Fig 7) contained the highest number of modules (49%, 139 among 283). This observation was consistent with the variability of the ancillary region and its known role in mediating specific interactions of ADAMTS-TSL [1]. Note that 30% (86 among 283) of the modules were present in the N-terminal region (colored in blue in Fig 7) which mainly corresponds to the propeptide, a domain known to evolve more rapidly than proper domains thus allowing specialization of function [42]. Importantly, the central region which includes the metalloprotease and disintegrin domains in ADAMTS (colored in orange and yellow in Fig 7) showed few modules 20% (58 among 283). All the variability of the central region was concentrated in the spacer that contained 26 of the 58 modules. Overall, our data supported evidence for the gain of new modules of conserved sequences, possibly associated to a gain of function in the N-terminal, C-terminal, and spacer regions of ADAMTS-TSL, which are regions known to be involved in the emergence of new functions (see our Discussion section on propeptides). These 45 module-PPI associations deserved further investigation, and we searched the literature for experimental evidence supporting putative roles in PPIs for some of our candidate modules.

Fig 7. Location in H. sapiens paralogs of the modules involved in the 45 events of module(s)-PPI(s) co-appearance.

(A) ADAMTS-TSL phylogeny indicating the 45 ancestral nodes (labels in white boxes) corresponding to the 45 module(s)-PPI(s) co-appearance events. TS, ADAMTS; TSL, ADAMTSL; Hs, Homo sapiens; Dr, Danio rerio; Xt, Xenopus Tropicalis; Gg, Gallus gallus; Mm, Mus musculus; Bt, Boss Taurus. (B) Each line corresponds to a H. sapiens protein chosen as representative of the ancestors, and onto which the ancestral module signature is reported. The Pfam domains are represented as grey boxes and the gained modules as green marks. Each sequence is divided into 3 regions; 1) the N-terminal region that contain the propetide in ADAMTS (blue), 2) the central region including the catalytic domain and the disintegrin domain (orange) and the central TSP1, the cys-rich domain and the spacer (yellow) and 3) the variable ancillary region from the end of the spacer to the C-terminal end (purple).

To further explore the ability of our model to predict associations between conserved sequence modules and PPIs, we focused on two specific cases: first, ADAMTS-7/ADAMTS-12 and their common substrates CCN2 and COMP and, second, the hyalectanase group. In both cases, we compared the model predictions with experimental data to validate the method as well as to propose new insights into the underlying evolutionary processes. The CCN2/COMP PPI case highlights the convergent evolution of PPIs during ADAMTS-TSL evolution while the hyalectanases case explores the sequence evolution of an entire gene subgroup up to ADAMTS-5, a member of this subgroup that shows significant interaction specificity.

Convergent evolution of COMP and CCN2 interactions with ADAMTS

Among the 26 H. sapiens ADAMTS-TSL proteins, the literature curation (step 4) identified ADAMTS-3, ADAMTS-4, ADAMTS-7 and ADAMTS-12 as interacting with the cartilage oligomeric matrix protein COMP and/or the connective tissue growth factor (CTGF alias CCN2). The role of ADAMTS-7 and ADAMTS-12 in the pathogenesis of arthritis was firstly supported by their ability to cleave COMP [43, 44]. More recently, CCN2 was also identified as a substrate of ADAMTS-7 and ADAMTS-12 [45, 46]. Since the structures of ADAMTS-7 and ADAMTS-12 are closely related, they form their own subgroup and it is not surprising that they share common substrates [47]. However, the identification of CCN2 and COMP as substrates of the distant ADAMTS, ADAMTS-3 [35] and ADAMTS-4 [48], respectively is less expected. As shown in Fig 8 and consistent with the PPIs inference mediated by PastML, the interaction with CCN2 or COMP most likely occurred twice during the evolution of ADAMTS-TSL. On the one hand, both PPIs (CCN2 and COMP) appeared in G161, the last common ancestral gene of ADAMTS-7 and ADAMTS-12. On the other hand, the COMP PPI was gained at G15, the last common ancestral gene of H. sapiens, B. taurus, M. musculus ADAMTS-4 and the CCN2 PPI was gained at G315, the last common ancestral gene of the H. sapiens, B. taurus, M. musculus, G. gallus ADAMTS-3.

Fig 8. Convergent evolution of COMP and CCN2 interactions with ADAMTS-TSL.

COMP and CCN2 interactions with ADAMTS-TSL are associated with independent module signatures acquired during evolution. (A) Phylogenetic tree of ADAMTSs with magnification of phylogenetic ADAMTS subtrees involved in COMP and CCN2 PPIs. (B) Heatmap: the published interactions of H. sapiens ADAMTS-3, ADAMTS-4, ADAMTS-7 and ADAMTS-12 with COMP and CCN2 are represented as dark blue boxes. The gains of the PPIs (inferred by PastML) are represented by the internal nodes: G315 (the ADAMTS-3 Amniota ancestor), G161 (the ADAMTS-7 and ADAMTS-12 paralogs ancestor) and G15 (the ADAMTS-4 mammalian ancestor) for CCN2, CCN2/COMP and COMP respectively. The PPIs inferred by PastML are represented as light blue boxes. At the opposite the absence of interaction (missing information) between ADAMTS and CCN2 and COMP is represented as dark orange boxes (for human proteins) and the absence of interaction inferred by PastML is represented as light orange boxes (for non human proteins).

To characterize the PPI changes associated with the ancestral genes G161, G15 and G315, we focused on the modules predicted to appear in those ancestors. We first investigated the module signature at G161 and we observed that the 25 modules gained were primarily localized in the TSP1-repeat of the ancillary region (Fig 9A). Importantly the four C-terminal TSP1 motifs corresponded to a region previously shown to be sufficient and necessary for the interaction of ADAMTS-7 and ADAMTS-12 with COMP [43, 44] and CCN2 [45, 46]. However, COMP was also identified as a substrate of ADAMTS-4 that lacks the 4 C-terminal TSP1 motifs suggesting that the interaction with COMP was associated with different sequences. Consistent with this hypothesis, we observed 8 gained modules at G15, the Mammalian ancestor of ADAMTS-4 (Fig 9B), that were localized at both the N-terminal (including peptide signal and part of the propeptide) and the C-terminal parts of the sequence. In addition, no module of the G15 signature was shared with the G161 signature. Similarly, we observed a specific gained module at G315, the Amniota ancestor of ADAMTS-3 (Fig 9C) that localized in the end of the ancillary region and this unique module was not found in the G161 signature. These analyses revealed that, despite similar localization in the ancillary region, the three module signatures (G161, G15, G315) were completely different in both their module sequences and their domain contexts. Overall, our data support the hypothesis that the interactions with COMP and CCN2 were each acquired independently at two distinct time points during ADAMTS-TSL evolution. These two cases of convergent evolution allowed us to propose 3 distinct module signatures, each characterized by specific modules in the C-terminal part of the ancillary region. In these particular cases, when convergent PPI evolution occurred, each independent PPI acquisition could be associated with a different ADAMTS-TSL interaction site.

Fig 9. Three module signatures are associated with COMP and/or CCN2 PPIs.

The ancestral module signatures reported on the descendant proteins in H. sapiens: (A) ADAMTS-7, (B) ADAMTS-4 and (C) ADAMTS-3. The location of the modules is represented as green boxes along with the location of the Pfam domain represented as grey boxes (top panel) while the content of the modules is represented by raw sequence logos [49] using the Protomata visualization [14], which displays also the chaining of the modules with arrows labeled by the minimal and maximal distances between the modules in the sequences of descendants (bottom panels). Because of the size of the G161 module signature, only the modules in the region of interaction with COMP are shown in (A).

The paralog expansion of hyalectanases highlights the functional specificity of ADAMTS-5

The hyalectanase family is composed of 7 members (ADAMTS-1, ADAMTS-4, ADAMTS-5, ADAMTS-8, ADAMTS-9, ADAMTS-15 and ADAMTS-20) which degrade proteoglycans such as aggrecan (ACAN), versican (VCAN), brevican and neurocan [50]. The last common ancestor of the hyalectanase genes was identified as the G96 gene for which the phenotype of interaction with ACAN and VCAN was gained (Fig 10A). Note that some hyalectanases shared other interactants such as LRP1 which was identified as partner of ADAMTS-4, ADAMTS-5, ADAMTS-9 and ADAMTS-20 (Fig 10B).

Fig 10. Evolutionary histories of hyalectanases PPIs.

(A) Hyalectanase tree. The last common ancestor of the human hyalectanases is the G96 gene node. The gain of the ACAN and VCAN PPIs was inferred at the G96 gene node. (B) Heatmap: the published interactions of H. sapiens hyalectanases with ACAN, VCAN, LRP1, TIMP3 and COMP are represented as dark blue boxes. The PPIs inferred by PastML tool are represented as light blue boxes. At the opposite the absence of interaction (missing information) between hyalectanases and ACAN, VCAN, LRP1, TIMP3 and COMP is represented as dark orange boxes (for human proteins) and the absence of interaction inferred by PastML is represented as light orange boxes (for non human proteins).

According to the literature, ADAMTS-4 (aggrecanase-1) and ADAMTS-5 (aggrecanase-2) are the main hyalectanases involved in the degradation of aggrecan (ACAN), the major component of articular cartilage. While ADAMTS-1 [51] and ADAMTS-9 [52] interact with aggrecan, they have lesser degrading activity. Although the activity of recombinant ADAMTS-5 was first shown to be twice as slow as that of ADAMTS-4 in vitro [53], ADAMTS-5 is the major aggrecanase in physiological conditions [911]. The interaction site with aggrecan involves the spacer domain for ADAMTS-4 [54] and both the cystein-rich and spacer domains for ADAMTS-5 [11]. Other substrates of ADAMTS-4 and ADAMTS-5 have been identified and some are common such as VCAN [51, 55] and others are different such as COMP which is only cleaved by ADAMTS-4 and not by ADAMTS-5 [48]. In addition, ADAMTS-4 and ADAMTS-5 share interactions with other proteins such as TIMP3 through their C-terminal domains [56] or such as LRP1 through different domains. Indeed, the cystein-rich and spacer domains of ADAMTS-4 were involved in the interaction with LRP1 whereas the thrombospondin 1 and spacer domains of ADAMTS-5 were involved in the interaction with LRP1 [57]. It is important to note that unlike other hyalectanases, the propeptide of ADAMTS-5 is cleaved extracellularly suggesting specific features [58].

These experimental observations suggest a functional specificity of ADAMTS-5 compared with ADAMTS-4 and other hyalectanases. Therefore, we examined the module signature in the ancestor of ADAMTS-5 proteins that may account for its uniqueness. At G65, the last common ancestor of all ADAMTS-5 orthologs (Fig 10A), 12 modules were gained, 11 of them being present in ADAMTS-5 H. sapiens (green boxes in Fig 11A). These modules are mainly localized in the propeptide domain, the spacer and the C-terminal TSP1, the two other domains being in the catalytic site and the cystein-rich domain. Looking at the ancestor G96 of the hyalectanases, the 2 gained modules are localized in the cystein-rich domain and the N-ter of the spacer domain (purple boxes in Fig 11A).

Fig 11. Module signatures of ADAMTS-5 and hyalectanase proteins.

(A) Location of G65 and G96 signature modules on H. sapiens ADAMTS-5 (NP_008969.2). The protein domains are shown in grey, the modules gained at the ADAMTS-5 ancestral gene G65 are shown in green and the modules gained at the G96 hyalectanase ancestral gene are shown in purple. (B) Excerpt of the PLMA restricted to the sequences of H. sapiens hyalectanases and three non hyalectanases (ADAMTS-6, ADAMTS-10 and ADAMTSL-4) in the spacer domain. PLMA blocks defining modules are shown as boxes containing the module segments. The succession of the module segments of each sequence is indicated by arrows labeled by a numbering of the sequences displayed (from 1 to 10 here), completed with the interval of sequence positions skipped if the segments are not contiguous. (C) Sequence logos as in Fig 9 of the modules gained at G96 (left) and G65 (right) in the spacer. (D) All the segments of H. sapiens ADAMTS-5 (NP_008969.2) in G65 and G96 signature modules (E) Predicted structures of the H. sapiens ADAMTS-5 protein with and without propeptide, colored with G65 and G96 modules. The three hypervariable loops, β1-β2, β3-β4 and β9-β10 previously described in Santamaria et al, 2019 [59] are marked by *.

Because the spacer domain at G65 and G96 genes accumulated many changes, we further analyzed the sequences in this domain which has been shown to be involved in the interaction with aggrecan. The Fig 11B shows all the modules found in the spacer by the PLMA on sequences of the H. sapiens hyalectanases as well as three additional non hyalectanase sequences (ADAMTS-6,-10 and ADAMTSL-4) to illustrate the block conservation with regard to close and distant paralogs of hyalectanases. The B781 block (purple box in Fig 11B and 11C) is the specific block of hyalectanases in the spacer region. After this block, the only conservation blocks detected in hyalactanase paralogs localized between ADAMTS-1, ADAMTS-8 and ADAMTS-15 and between ADAMTS-9 and ADAMTS-20. It is important to note that the two close aggrecanases ADAMTS-4 and ADAMTS-5 displayed distinct conservation blocks in this spacer region, ADAMTS-5 being characterized by three specific blocks, B1603, B1605 and B1607 (green boxes in Fig 11B and 11C) shared only with its orthologs. The sequences of B1603 and B1607 blocks overlapped with the hypervariable loops β3-β4 and β9-β10 that were previously reported to mediate the specificity of ADAMTS interaction [59]. Note that no block was identified in sequences corresponding to the third hypervariable loop β1-β2 described by Santamaria et al [59]. The G65 specific conservation identified by block B1605 matched instead with the β6-β7-β8 region and the surrounding loops (using beta strand numbering from [59]).

The full set of segments of H. sapiens ADAMTS-5 belonging to G65 and G96 module signatures are shown in Fig 11D. We crossed also our inference of G65 and G96 module signatures with the prediction of their spatial localization in the H. sapiens ADAMTS-5 protein. AlphaFold 2.1.14 [60] was used in the ColabFold 1.3.0 local environment [61] to predict the 3D structure of H. sapiens ADAMTS-5. Since ADAMTS-5 is activated in the extracellular space [58], we explored the structures of ADAMTS-5 with and without the propeptide, the peptide signal being removed from the sequence. The prediction was made from both sequences by launching the command colabfold_batch with the parameters --num-recycle 3 --num-models = 5 --amber. The two resulting first ranked relaxed models colored by the location of G65 and G96 modules are shown in Fig 11E. The predicted structures are available as S1 and S2 Files and the corresponding pLDDT and PAE plots are presented in S1 and S2 Figs. It is important to note that removing the propeptide does not affect the prediction much, the main difference being in the movement of the C-terminal TSP1 domain towards the place left empty by the propeptide. The pLDDT and pTM scores which are respectively 90.3 and 0.79 without the propeptide changed to 78.3 and 0.592 with the propeptide, showing a stronger confidence in the prediction of the model without the propeptide.

Interestingly, although module segments may be distant in the sequence, they are all located in the same area of the predicted 3D structures at the interface of the protein, around a conserved ADAMTS-TSL core consisting of the central TSP1 domain and the C-term of the cystein-rich domain, with the exception of the module segment in the catalytic domain. In particular, the segments of the G65 and G96 signature modules from the cystein rich and spacer domains seem to form a localized sub-unit made of a beta sheet in the spacer facing the conserved segments in the cystein rich domain. The segment of the G65 module in the cystein rich domain is also interestingly predicted in a loop with an orientation similar to the β1-β2, β3-β4 and β9-β10 hypervariable loops identified by Santamaria et al [59] and the β6-β7-β8 loops from the spacer domain. Note that the specific G65 module segments present in the C-terminal TSP1 and the propeptide are co-localized near the central TSP1 domain in the predicted structure suggesting a regulatory role for the propeptide.

Together our data summarized in Fig 11 allow us to propose a scenario for the evolution of the spacer domain leading to a specialization of hyalectanases/ADAMTS-5. According to this scenario, a first conserved module appeared in the hyalectanase ancestor (positions 748–762 in Fig 11D) prior to the specialization of the three specific ADAMTS-5 conservation modules (positions 764–773, 795–825 and 827–855 in Fig 11D). Two of these three specific modules coincided with already identified regions by Santamaria et al [59]. The third one, appearing in an interesting location of the predicted structures, and the absence of conservation in the β1-β2 region of ADAMTS-5 provided novel insights into the spacer domain. Moreover, these conserved modules in the spacer are likely to be functionally linked to the specific modules in the cystein rich domain from their respective orientation in the predicted 3D structures. Besides, our data highlight also specific conservations predicted to be spatially co-localized in the propeptide and C-term TSP1 regions, which are challenging to interpret and are of interest to guide further studies.


Emerging new combinations of domains plays a major role in the evolution of modular organization of proteins, and the emergence of adaptive traits. This phenomenon has been shown to be predominant in the evolution of proteins involved in the extracellular matrix, an interconnected network composed of hundreds of macromolecules that includes the multidomain protein family ADAMTS-TSL [6266]. Here we investigated the ADAMTS-TSL family in order to identify functional sequence regions in their modular structure based on two widely used methods for functional prediction [67], namely sequence conservation analysis [68] and phylogenomic inference [15]. We provided an extension of phylogenomics inference to modules using Domain-Gene-Species (DGS) reconciliation [24]. This task is challenging because of the large paralogous expansion in that family (26 paralogs in human) and the large number and diversity of annotated protein domains (from 5 to 9 in human paralogs).

Inference of the ADAMTS-TSL phylogeny

The evolutionary history of the ADAMTS-TSL protein family has previously been studied [6972]. Using ADAM proteins as the outgroup, the present work provided a rooted phylogeny of 173 ADAMTS and ADAMTSL from 9 bilateria species that was consistent with the main conclusions of these previous studies, and especially with a family expansion in vertebrates, possibly driven through whole genome duplication events [72]. Note that similarly to the rooting using one Porifera ADAMTS sequence [72], our rooted phylogeny identified the procollagen peptidases as the first to diverge within the ADAMTS and ADAMTSL. Importantly and unlike other studies, we used both a reconciliation method that identified paralogous gene duplication and speciation events, and a correction method taking into account the species tree, that corrected stochastic errors in the gene tree. As previously described by Brunet et al. [72], we identified a monophyly of the procollagen peptidases (ADAMTS-2, -3 and -14) and a monophyly of the hyalectanases (ADAMTS-1, -4, -5, -8, -9, -15 and -20) with consistent relations within these two clades. In adition, we also identified as monophyletic the COMP proteases (ADAMTS-7 and ADAMTS-12) and the other pairs of enzymes known as closely related paralogs (i.e. ADAMTS-6 and ADAMTS-10, ADAMTS-16 and ADAMTS-18, ADAMTS-17 and ADAMTS-19). Importantly these groups are found robust to the choice of the considered sequence sets, as shown by gene re-sampling assays (See S1 Appendix).

An interesting new finding is the monophyly of the ADAMTSL group while papilin, a member of the ADAMTSL protein family, clustered in a distinct group related to ADAMTS-6 and ADAMTS-10. Note that these two distinct groups were not supported in all the gene re-sampling assays (See S1 Appendix). However our results were in agreement with an origin of ADAMTSL and papilin within the ADAMTS and a loss of the catalytic domain [7375], even if it remains unclear whether this event has happened once or more.

Sequence decomposition into modules and evolution of module composition

Many studies have used domain composition as a proxy for protein function, as they represent independent structural and functional units (see Cromar et al 2014 [62], for such a study focusing on ECM proteins). Instead, in this work we focused on highly conserved protein modules as a proxy for functional prediction. High local conservation reveals regions that likely confer functional features conserved since a common ancestor [67, 68]. The rational for this assumption is that negative natural selection applying to maintain an important phenotype (including functions) therefore applies at the genotype level, avoiding sequence divergence.

Sequence conservation is generally identified using global alignments, protein structures or features and motifs identified a priori [76]. Few alignment methods allow the identification of locally conserved sequences without a priori. Here, we used a new implementation of the alignment program Paloma from the Protomata suite (See S2 Appendix). Segments within the same module are strongly similar, have the same size and can only differ through substitutions. Thanks to this method, we identified modules representing highly conserved segments shared by some of, and not necessarily all, the protein sequences. By applying Domain-Gene-Species (DGS) reconciliation to module trees, we next identified the gene ancestors which acquired new modules. This provides a phylogenetic characterization of the module. Proteins derived from the ancestral gene tend to possess this module, but not the more distantly related proteins.

The modules are gapless: the strongly conserved segments have no visible traces of insertion or deletion events. Hence, a conserved region with insertions and deletions will be modeled by several modules. The number of modules will increase as the number of insertion and deletion augments, and the size of the modules will decrease, leading to a more fragmented characterization of the region focusing on the conserved parts only.

The value of the conservation threshold parameter can be particularly decisive for the results. Actually, relaxing this threshold, it might be possible to get longer modules involving more genes and older common ancestors. Thus a module gain or loss does not necessarily account for the ancestral locus gain or loss, but indicates that significant differences exist between subsets of segments originating from that locus. For instance, the gain of a conserved module in a given lineage would correspond to a sequence divergence above the chosen conservation threshold. In such a scenario, two different modules may correspond to homologous sequences that diverged from an older ancestral locus. In this study, the modules were identified on the basis of a highly conservative diagonal’s similarity threshold. Let us emphasize that our module decomposition identified segments conserved in all the ADAMTS and belonging, as expected, to shared domains such as the catalytic domain, suggesting that the approach and the chosen threshold are adapted to our study case.

Our innovative approach of combining partial local multiple alignment with the DGS reconciliation method allowed us to define module signatures (i.e., modules acquired simultaneously during gene evolution). A module signature thus consists of one or several independent modules, so that one of their distinguishing feature is the possibility of being discontinuous along the sequences.

Convergent Protein-Protein Interactions to increase ECM complexity

Protein-Protein Interactions play fundamental role in the structure of the ECM [62, 66] and are key features to understand the functions of ADAMTS-TSL as demonstrated for the formation of microfibril and its deregulation in Acromelic dysplasias [4, 77]. Here, we reconstructed 278 ancestral PPIs along the ADAMTS-TSL phylogeny and we identified the convergent appearance of ADAMTS interactions with COMP and CCN2. Because the interaction with COMP involves four TSP1 repeats in ADAMTS-7 and ADAMTS-12 [43, 44] that are not present in the ADAMTS-4, interaction sites are different. We propose that such convergences are favored by selective pressures leading to the increase of the ECM complexity by increasing the possible interactions among ECM proteins. Importantly, ADAMTS4, ADAMTS7, and ADAMTS12 degrade the COMP protein that must interact with the catalytic site in the metalloprotease domain. However, the enzyme-substrate interaction is transient in nature and requires the substrate to be “fixed” in the vicinity of the catalytic site. This local organization depends on the molecular networks where the enzyme-substrate complex takes place and could explain the involvement of different module sequences according to ADAMTS. Cromar et al (2014) [62] demonstrated that novel domain arrangements have contributed to increased ECM complexity in vertebrates, multiple possible domain combinations contributing to the emergence of new functions. Similarly, increasing the possible ADAMTS-TSL interactions may contribute to increase ECM complexity. Hence, whereas domain combination, appearance and shuffling received attention in understanding the functions of proteins involved in the ECM [64], our hypothesis suggests that convergent interactions between ECM proteins, such as observed here for the ADAMTS, might constitute another mechanism to diversify the functions of ECM proteins.

Ancestral module signatures as putative functional motifs

The identification of functional sites in a protein is a prerequisite to understand its molecular action and to develop efficient therapeutic approaches [76]. Numerous computational prediction methods have been developed [78] providing a large diversity of tools for biologists. Among them, methods based on conserved sequence analysis are widely used. As mentioned above they are relying on the principle that proteins inherit their phenotypes and genotypes from a common ancestor. However having similar domains in common does not necessarily indicate that proteins have similar functions especially in the case of multidomain proteins such as ADAMTS-TSL. Consistent with this is the failure to target the activities of MMPs via their catalytic domain [79]. By combining PPI annotations and sequence conservation analysis, we developed an original approach that predicts module signatures composed of non linear sub-sequences sets associated with the function of Protein-Protein Interactions. Note here that any phenotype of interest might be associated with the ancestral module signature. In the present work, we chose PPIs as the phenotype of interest and we identified 45 module signatures as putative functional motifs.

Our method is supported by the identification of module signatures that matched functional motifs previously identified, such as PPIs involving COMP and ADAMTS-7/ADAMTS-12 [43, 44], or PPIs involving ACAN and the spacer domain of aggrecanases [80]. In addition the module signatures of the hyalectanases revealed differences that might explain the differences of activity against the substrate ACAN [912, 51, 52, 52]. The physiological aggrecanase ADAMTS-5 showed a unique signature containing 12 modules spread over the sequence but spatially very close in the structure of the protein, thereby defining a putative specific area of interaction with ACAN. Importantly two hypervariable loops of ADAMTS-5, shown to be important for its aggrecanase activity, overlapped with our predicted signature supporting its reliability [81]. Although monoclonal antibodies and small molecules have been developed to inhibit the deleterious effect of ADAMTS-5 in osteoarthritis [82], side effects and lack of specificity have prevented their progression towards clinical trials. In this context, our specific sequence signature could be valuable for the design of novel inhibitory molecules.

The propeptide of ADAMTS as a source of divergence

Like other ADAMTSs, ADAMTS-5 shows high variability in the propeptide domain (Fig 7) that has been previously described as an evolutionary module and transient protein domain [42]. Propeptides evolve more rapidly than other domains allowing diversification of gene functions. Our observations are consistent with the specificity of these propeptides while suggesting module signatures specific to functional dissimilarities within the ADAMTS-TSL family. The appearance of modules in the N-terminal part of ADAMTS (signal peptide and propeptide) is intriguing since the signal peptide and the amino-propeptide are usually processed for protease secretion and activation thus acting as sorting signal [83] and inhibitory [84] peptides, respectively. However, other functions have been attributed to propeptides including chaperone-assisted intramolecular protein folding, protein sorting, and mediation of precursor interaction with its partners [42, 84]. Most ADAMTS contain a convertase cleavage site, however processing of ADAMTSs affects the localization and activity of the protein differently. For ADAMTS9, the propeptide acts as both an intramolecular chaperone and a sorting protein to target ADAMTS9 to the membrane where the protease is then processed by furin [85, 86]. Unlike most ADAMTSs, furin processing of ADAMTS9 leads to an inactive enzyme. Similar to ADAMTS9, the propeptide of ADAMTS7 acts as a chaperone and ADAMTS17 undergoes autocatalytic processing independent of propeptide cleavage [87]. ADAMTS13 is also characterized by propeptide independent regulation. The functional specificity of ADAMTS13 may be related to its very short propeptide that does not affect protease activity against Von Willebrand factor [88]. Consistent with the heterogeneity of propeptide functions in ADAMTS, Kutz et al [89] demonstrate that the propeptide of ADAMTS10 is processed by convertases with very low efficiency and remains associated with the catalytically active protein. The associated propeptide may induce a conformational change that affect the interaction with fibrillin 1, contributing to microfibril biogenesis rather than in fibrillin turnover. In contrast, its closest homolog, ADAMTS6, showed a functional furin consensus sequence [90] and no interaction with fibrillin or contribution to microfibril biogenesis has been demonstrated to date [91]. Specific regulation was also demonstrated for ADAMTS5 which is characterized by a unique extracellular processing [92]. While processing of ADAMTS5 is required for its activity against versican and aggrecan, cleavage of the propeptide is mediated by extracellular proprotein convertases, suggesting that the propeptide contributes to adequate localization in the extracellular matrix where processing takes place. This specific mechanism may be associated to the high efficiency of ADAMTS5 activity compared to all other hyalectanases. All of these observations highlight the diversity in the regulation of ADAMTS activity by the N-terminal peptide and are consistent with the known variability of propeptides with respect to the conserved catalytic domain making propeptides “evolutionary specific factors” [42]. This suggests that future development of inhibitory molecules could take into account the specificity of propeptides that regulate ADAMTS activity.


We propose here a new approach bringing together advanced sequence analysis and phylogenetic methods for identifying conserved sequence modules associated with protein function. This approach is based on an original integration of the evolutionary histories of species, genes, modules and on the prediction of modules and phenotype co-appearances to reveal sets of (non-contiguous) conserved sequence regions defining phenotype specific signatures. We applied this strategy to search for conserved modules in the ADAMTS-TSL family using Protein-Protein Interaction data as phenotypes. Based on a set of experimentally supported ADAMTS-TSL protein interactions we provide novel insight into the evolution and specialization of ADAMTS-TSL interactions. Our framework is generalizable to other protein families as well as to other phenotypes, giving new perspectives to the functional characterization of other multi-functional, multi-domain protein families.

Supporting information

S1 Appendix. Uncertainty analysis of the ADAMTS-TSL phylogenetic tree.


S2 Appendix. Partial local multiple alignment with Paloma-D.


S3 Appendix. Browsing the ADAMTS-TSL Itol tree.

Itol tree screenshots and usage examples.


S1 Table. Description of the species, identifiers and assemblies.


S2 Table. Quantification of amino acids in longest isoforms compared to splice graph.

The whole isoform sequences were examined as splicing graphs and the unconsidered information for all isoforms was quantified.


S3 Table. The 214 representative protein sequences dataset: Sequence RefSeq ids and species taxid.


S4 Table. List of PPIs identified by PSICQUIC analyses.


S5 Table. Distribution of PPIs identified by PSCICQUIC analyses according to species.


S6 Table. List of ADAMTS substrates identified by manual literature curation.


S7 Table. List of Protein-Protein Interactions for ADAMTS-TSL.


S8 Table. Module composition of the 26 H. sapiens ADAMTS-TSL.


S9 Table. List of the 1059 conserved sequence modules.


S1 Fig. pLDDT and PAE scores of the ADAMTS-5 AlphaFold’s predicted structure without peptide signal and with propeptide.


S2 Fig. pLDDT and PAE scores of the ADAMTS-5 AlphaFold’s predicted structure without both peptide signal and propeptide.


S1 File. The ADAMTS-5 AlphaFold’s predicted structure without peptide signal and with propeptide.


S2 File. The ADAMTS-5 AlphaFold’s predicted structure without both peptide signal and propeptide.



The authors thank the excellent support of the GenOuest bioinformatics core facility, Nicolas Buton (Univ Rennes, Inria, CNRS, IRISA, UMR 6074, Rennes, France) for predicting protein structures using AlphaFold 2 method and Dr Catherine Lavau (Univ Rennes, Inserm, EHESP, Irset, UMR S1085, Rennes, France) for proofreading the manuscript.


  1. 1. Kelwick R, Desanlis I, Wheeler GN, Edwards DR. The ADAMTS (A Disintegrin and Metalloproteinase with Thrombospondin motifs) family. Genome Biology. 2015;16:113. pmid:26025392
  2. 2. Mead TJ, Apte SS. ADAMTS proteins in human disorders. Matrix biology: journal of the International Society for Matrix Biology. 2018;71-72:225–239. pmid:29885460
  3. 3. Rose KWJ, Taye N, Karoulias SZ, Hubmacher D. Regulation of ADAMTS Proteases. Frontiers in Molecular Biosciences. 2021;8:701959. pmid:34268335
  4. 4. Hubmacher D, Apte SS. ADAMTS proteins as modulators of microfibril formation and function. Matrix Biol. 2015;47:34–43. pmid:25957949
  5. 5. Théret N, Bouezzedine F, Azar F, Diab-Assaf M, Legagneux V. ADAM and ADAMTS Proteins, New Players in the Regulation of Hepatocellular Carcinoma Microenvironment. Cancers. 2021;13(7). pmid:33805340
  6. 6. Apte SS. ADAMTS Proteins: Concepts, Challenges, and Prospects. Methods in Molecular Biology (Clifton, NJ). 2020;2043:1–12. pmid:31463898
  7. 7. Sigrist CJA, Cerutti L, Hulo N, Gattiker A, Falquet L, Pagni M, et al. PROSITE: a documented database using patterns and profiles as motif descriptors. Briefings in Bioinformatics. 2002;3(3):265–274. pmid:12230035
  8. 8. Mistry J, Chuguransky S, Williams L, Qureshi M, Salazar G, Sonnhammer ELL, et al. Pfam: The protein families database in 2021. Nucleic Acids Research. 2021;49(D1):D412–D419. pmid:33125078
  9. 9. Stanton H, Rogerson FM, East CJ, Golub SB, Lawlor KE, Meeker CT, et al. ADAMTS5 is the major aggrecanase in mouse cartilage in vivo and in vitro. Nature. 2005;434(7033):648–652. pmid:15800625
  10. 10. Glasson SS, Askew R, Sheppard B, Carito B, Blanchet T, Ma HL, et al. Deletion of active ADAMTS5 prevents cartilage degradation in a murine model of osteoarthritis. Nature. 2005;434(7033):644–648. pmid:15800624
  11. 11. Gendron C, Kashiwagi M, Lim NH, Enghild JJ, Thøgersen IB, Hughes C, et al. Proteolytic activities of human ADAMTS-5: comparative studies with ADAMTS-4. The Journal of Biological Chemistry. 2007;282(25):18294–18306. pmid:17430884
  12. 12. Santamaria S, Martin DR, Dong X, Yamamoto K, Apte SS, Ahnström J. Post-translational regulation and proteolytic activity of the metalloproteinase ADAMTS8. The Journal of Biological Chemistry. 2021;297(5):101323. pmid:34687701
  13. 13. Kerbellec G. Apprentissage d’automates modélisant des familles de séquences protéiques. (Learning automata modelling families of protein sequences) [PhD Thesis]. University of Rennes 1, France; 2008. Available from:
  14. 14. Protomata 2.0;.
  15. 15. Eisen J. Phylogenomics: Improving Functional Predictions for Uncharacterized Genes by Evolutionary Analysis. Genome research. 1998;8:163–7. pmid:9521918
  16. 16. Engelhardt BE, Jordan MI, Muratore KE, Brenner SE. Protein Molecular Function Prediction by Bayesian Phylogenomics. PLOS Computational Biology. 2005;1(5):e45. pmid:16217548
  17. 17. Engelhardt BE, Jordan MI, Srouji JR, Brenner SE. Genome-scale phylogenetic function annotation of large and diverse protein families. Genome Research. 2011;21(11):1969–1980. pmid:21784873
  18. 18. Gaudet P, Livstone MS, Lewis SE, Thomas PD. Phylogenetic-based propagation of functional annotations within the Gene Ontology consortium. Briefings in Bioinformatics. 2011;12(5):449–462. pmid:21873635
  19. 19. Sahraeian SM, Luo KR, Brenner SE. SIFTER search: a web server for accurate phylogeny-based protein function prediction. Nucleic Acids Research. 2015;43(W1):W141–W147. pmid:25979264
  20. 20. Ishikawa SA, Zhukova A, Iwasaki W, Gascuel O. A Fast Likelihood Method to Reconstruct and Visualize Ancestral Scenarios. Molecular Biology and Evolution. 2019;36(9):2069–2085. pmid:31127303
  21. 21. Yon GGV, Thomas DC, Morrison J, Mi H, Thomas PD, Marjoram P. On the automatic annotation of gene functions using observational data and phylogenetic trees. bioRxiv. 2020; p. 2020.05.14.095687.
  22. 22. Stolzer M, Siewert K, Lai H, Xu M, Durand D. Event inference in multidomain families with phylogenetic reconciliation. BMC Bioinformatics. 2015;16 Suppl 14:S8. pmid:26451642
  23. 23. Li L, Bansal MS. An Integrated Reconciliation Framework for Domain, Gene, and Species Level Evolution. IEEE/ACM Trans Comput Biol Bioinform. 2019;16(1):63–76. pmid:29994126
  24. 24. Li L, Bansal M. Simultaneous Multi-Domain-Multi-Gene Reconciliation Under the Domain-Gene-Species Reconciliation Model; 2019. p. 73–86.
  25. 25. Emms DM, Kelly S. OrthoFinder: phylogenetic orthology inference for comparative genomics. Genome Biology. 2019;20(1):238. pmid:31727128
  26. 26. Mirarab S, Nguyen N, Guo S, Wang LS, Kim J, Warnow T. PASTA: Ultra-Large Multiple Sequence Alignment for Nucleotide and Amino-Acid Sequences. Journal of Computational Biology. 2015;22(5):377–386. pmid:25549288
  27. 27. Stamatakis A. RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies. Bioinformatics. 2014;30(9):1312–1313. pmid:24451623
  28. 28. Wu YC, Rasmussen MD, Bansal MS, Kellis M. TreeFix: Statistically Informed Gene Tree Error Correction Using Species Trees. Systematic Biology. 2013;62(1):110–120. pmid:22949484
  29. 29. Guindon S, Dufayard JF, Lefort V, Anisimova M, Hordijk W, Gascuel O. New Algorithms and Methods to Estimate Maximum-Likelihood Phylogenies: Assessing the Performance of PhyML 3.0. Systematic Biology. 2010;59(3):307–321. pmid:20525638
  30. 30. Lee C, Grasso C, Sharlow MF. Multiple sequence alignment using partial order graphs. Bioinformatics. 2002;18(3):452–464. pmid:11934745
  31. 31. Morgenstern B. DIALIGN 2: improvement of the segment-to-segment approach to multiple sequence alignment. Bioinform. 1999;15(3):211–218. pmid:10222408
  32. 32. Abdeddaïm S, Morgenstern B. Speeding Up the DIALIGN Multiple Alignment Program by Using the ‘Greedy Alignment of BIOlogical Sequences LIBrary’ (GABIOS-LIB). In: Gascuel O, Sagot MF, editors. Computational Biology. Berlin, Heidelberg: Springer Berlin Heidelberg; 2001. p. 1–11.
  33. 33. Schoch CL, Ciufo S, Domrachev M, Hotton CL, Kannan S, Khovanskaya R, et al. NCBI Taxonomy: a comprehensive update on curation, resources and tools. Database: The Journal of Biological Databases and Curation. 2020;2020. pmid:32761142
  34. 34. Aranda B, Blankenburg H, Kerrien S, Brinkman FSL, Ceol A, Chautard E, et al. PSICQUIC and PSISCORE: accessing and scoring molecular interactions. Nature Methods. 2011;8(7):528–529. pmid:21716279
  35. 35. Bekhouche M, Leduc C, Dupont L, Janssen L, Delolme F, Vadon-Le Goff S, et al. Determination of the substrate repertoire of ADAMTS2, 3 and 14 significantly broadens their functions and identifies extracellular matrix organization and TGF-β signaling as primary targets. FASEB Journal. 2016;5. pmid:26740262
  36. 36. Leduc C, Dupont L, Joannes L, Monseur C, Baiwir D, Mazzucchelli G, et al. In vivo N-Terminomics Highlights Novel Functions of ADAMTS2 and ADAMTS14 in Skin Collagen Matrix Building. Frontiers in Molecular Biosciences. 2021;8:643178. pmid:33816558
  37. 37. Colige A, Monseur C, Crawley JTB, Santamaria S, de Groot R. Proteomic discovery of substrates of the cardiovascular protease ADAMTS7. The Journal of Biological Chemistry. 2019;294(20):8037–8045. pmid:30926607
  38. 38. Yamamoto K, Owen K, Parker AE, Scilabra SD, Dudhia J, Strickland DK, et al. Low density lipoprotein receptor-related protein 1 (LRP1)-mediated endocytic clearance of a disintegrin and metalloproteinase with thrombospondin motifs-4 (ADAMTS-4): functional differences of non-catalytic domains of ADAMTS-4 and ADAMTS-5 in LRP1 binding. J Biol Chem. 2014;289(10):6462–6474. pmid:24474687
  39. 39. Nandadasa S, Kraft CM, Wang LW, O’Donnell A, Patel R, Gee HY, et al. Secreted metalloproteases ADAMTS9 and ADAMTS20 have a non-canonical role in ciliary vesicle growth during ciliogenesis. Nat Commun. 2019;10(1):953. pmid:30814516
  40. 40. Letunic I, Bork P. Interactive Tree Of Life (iTOL) v4: recent updates and new developments. Nucleic Acids Research. 2019;47(W1):W256–W259. pmid:30931475
  41. 41. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, et al. Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003;13(11):2498–2504. pmid:14597658
  42. 42. Demidyuk IV, Shubin AV, Gasanov EV, Kostrov SV. Propeptides as modulators of functional activity of proteases. Biomolecular Concepts. 2010;1(3-4):305–322. pmid:25962005
  43. 43. Liu CJ, Kong W, Ilalov K, Yu S, Xu K, Prazak L, et al. ADAMTS-7: a metalloproteinase that directly binds to and degrades cartilage oligomeric matrix protein. FASEB journal: official publication of the Federation of American Societies for Experimental Biology. 2006;20(7):988–990. pmid:16585064
  44. 44. Liu Cj, Kong W, Xu K, Luan Y, Ilalov K, Sehgal B, et al. ADAMTS-12 Associates with and Degrades Cartilage Oligomeric Matrix Protein. The Journal of biological chemistry. 2006;281(23):15800–15808.
  45. 45. Pi L, Jorgensen M, Oh SH, Protopapadakis Y, Gjymishka A, Brown A, et al. A Disintegrin and Metalloprotease with Thrombospondin Type I Motif 7. The American Journal of Pathology. 2015;185(6):1552–1563.
  46. 46. Wei Jl, Fu W, Hettinghouse A, He Wj, Lipson KE, Liu Cj. ADAMTS-12 protects against inflammatory arthritis through interacting with and inactivating proinflammatory CTGF. Arthritis & rheumatology (Hoboken, NJ). 2018;70(11):1745–1756.
  47. 47. Liu CJ. The role of ADAMTS-7 and ADAMTS-12 in the pathogenesis of arthritis. Nat Clin Pract Rheumatol. 2009;5(1):38–45. pmid:19098927
  48. 48. Dickinson SC, Vankemmelbeke MN, Buttle DJ, Rosenberg K, Heinegård D, Hollander AP. Cleavage of cartilage oligomeric matrix protein (thrombospondin-5) by matrix metalloproteinases and a disintegrin and metalloproteinase with thrombospondin motifs. Matrix Biology: Journal of the International Society for Matrix Biology. 2003;22(3):267–278. pmid:12853037
  49. 49. Crooks GE, Hon G, Chandonia JM, Brenner SE. WebLogo: A Sequence Logo Generator. Genome Research. 2004;14(6):1188–1190. pmid:15173120
  50. 50. Fontanil T, Mohamedi Y, Espina-Casado J, Obaya AJ, Cobo T, Cal S. Hyalectanase Activities by the ADAMTS Metalloproteases. International Journal of Molecular Sciences. 2021;22(6):2988. pmid:33804223
  51. 51. Sandy JD, Westling J, Kenagy RD, Iruela-Arispe ML, Verscharen C, Rodriguez-Mazaneque JC, et al. Versican V1 proteolysis in human aorta in vivo occurs at the Glu441-Ala442 bond, a site that is cleaved by recombinant ADAMTS-1 and ADAMTS-4. The Journal of Biological Chemistry. 2001;276(16):13372–13378. pmid:11278559
  52. 52. Somerville RPT, Longpre JM, Jungers KA, Engle JM, Ross M, Evanko S, et al. Characterization of ADAMTS-9 and ADAMTS-20 as a distinct ADAMTS subfamily related to Caenorhabditis elegans GON-1. The Journal of Biological Chemistry. 2003;278(11):9503–9513. pmid:12514189
  53. 53. Tortorella MD, Liu RQ, Burn T, Newton RC, Arner E. Characterization of human aggrecanase 2 (ADAM-TS5): substrate specificity studies and comparison with aggrecanase 1 (ADAM-TS4). Matrix Biology: Journal of the International Society for Matrix Biology. 2002;21(6):499–511. pmid:12392761
  54. 54. Kashiwagi M, Enghild JJ, Gendron C, Hughes C, Caterson B, Itoh Y, et al. Altered proteolytic activities of ADAMTS-4 expressed by C-terminal processing. The Journal of Biological Chemistry. 2004;279(11):10109–10119. pmid:14662755
  55. 55. Foulcer SJ, Nelson CM, Quintero MV, Kuberan B, Larkin J, Dours-Zimmermann MT, et al. Determinants of versican-V1 proteoglycan processing by the metalloproteinase ADAMTS5. The Journal of Biological Chemistry. 2014;289(40):27859–27873. pmid:25122765
  56. 56. Abou-Elhamd A, Cooper O, Münsterberg A. Klhl31 is associated with skeletal myogenesis and its expression is regulated by myogenic signals and Myf-5. Mechanisms of Development. 2009;126(10):852–862. pmid:19643178
  57. 57. Yamamoto K, Owen K, Parker AE, Scilabra SD, Dudhia J, Strickland DK, et al. Low density lipoprotein receptor-related protein 1 (LRP1)-mediated endocytic clearance of a disintegrin and metalloproteinase with thrombospondin motifs-4 (ADAMTS-4): functional differences of non-catalytic domains of ADAMTS-4 and ADAMTS-5 in LRP1 binding. The Journal of Biological Chemistry. 2014;289(10):6462–6474. pmid:24474687
  58. 58. Characterization of proADAMTS5 processing by proprotein convertases—ScienceDirect;. Available from:
  59. 59. Santamaria S, Yamamoto K, Teraz-Orosz A, Koch C, Apte SS, de Groot R, et al. Exosites in Hypervariable Loops of ADAMTS Spacer Domains control Substrate Recognition and Proteolysis. Scientific Reports. 2019;9:10914. pmid:31358852
  60. 60. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;596(7873):583–589. pmid:34265844
  61. 61. Mirdita M, tze K, Moriwaki Y, Heo L, Ovchinnikov S, Steinegger M. ColabFold: making protein folding accessible to all. Nat Methods. 2022;19(6):679–682. pmid:35637307
  62. 62. Cromar G, Wong KC, Loughran N, On T, Song H, Xiong X, et al. New Tricks for “Old” Domains: How Novel Architectures and Promiscuous Hubs Contributed to the Organization and Evolution of the ECM. Genome Biology and Evolution. 2014;6(10):2897–2917. pmid:25323955
  63. 63. Hynes RO. The extracellular matrix: not just pretty fibrils. Science. 2009;326(5957):1216–1219. pmid:19965464
  64. 64. Hynes RO. The evolution of metazoan extracellular matrix. J Cell Biol. 2012;196(6):671–679. pmid:22431747
  65. 65. Reuten R, Mayorca-Guiliani AE, Erler JT. Matritecture: Mapping the extracellular matrix architecture during health and disease. Matrix Biol Plus. 2022;14:100102. pmid:35243299
  66. 66. Karamanos NK, Theocharis AD, Piperigkou Z, Manou D, Passi A, Skandalis SS, et al. A guide to the composition and functions of the extracellular matrix. FEBS J. 2021;288(24):6850–6912. pmid:33605520
  67. 67. Zhou N, Jiang Y, Bergquist TR, Lee AJ, Kacsoh BZ, Crocker AW, et al. The CAFA challenge reports improved protein function prediction and new functional annotations for hundreds of genes through experimental screens. Genome Biology. 2019;20. pmid:31744546
  68. 68. Capra JA, Singh M. Predicting functionally important residues from sequence conservation. Bioinformatics. 2007;23(15):1875–1882. pmid:17519246
  69. 69. Huxley-Jones J, Apte SS, Robertson DL, Boot-Handford RP. The characterisation of six ADAMTS proteases in the basal chordate Ciona intestinalis provides new insights into the vertebrate ADAMTS family. The International Journal of Biochemistry & Cell Biology. 2005;37(9):1838–1845. pmid:15899586
  70. 70. Nicholson AC, Malik SB, Logsdon JM, Van Meir EG. Functional evolution of ADAMTS genes: Evidence from analyses of phylogeny and gene organization. BMC Evolutionary Biology. 2005;5(1):11. pmid:15693998
  71. 71. Huxley-Jones J, Clarke TK, Beck C, Toubaris G, Robertson DL, Boot-Handford RP. The evolution of the vertebrate metzincins; insights from Ciona intestinalis and Danio rerio. BMC Evolutionary Biology. 2007;7(1):63. pmid:17439641
  72. 72. Brunet FG, Fraser FW, Binder MJ, Smith AD, Kintakas C, Dancevic CM, et al. The evolutionary conservation of the A Disintegrin-like and Metalloproteinase domain with Thrombospondin-1 motif metzincins across vertebrate species and their expression in teleost zebrafish. BMC Evolutionary Biology. 2015;15(1):22. pmid:25879701
  73. 73. Todd AE, Orengo CA, Thornton JM. Sequence and Structural Differences between Enzyme and Nonenzyme Homologs. Structure. 2002;10(10):1435–1451. pmid:12377129
  74. 74. Murphy JM, Farhan H, Eyers PA. Bio-Zombie: the rise of pseudoenzymes in biology. Biochemical Society Transactions. 2017;45(2):537–544. pmid:28408493
  75. 75. Ribeiro AJM, Das S, Dawson N, Zaru R, Orchard S, Thornton JM, et al. Emerging concepts in pseudoenzyme classification, evolution, and signaling. Science Signaling. 2019;12(594):eaat9797. pmid:31409758
  76. 76. Pazos F. Computational prediction of protein functional sites-Applications in biotechnology and biomedicine. Adv Protein Chem Struct Biol. 2022;130:39–57. pmid:35534114
  77. 77. Stanley S, Balic Z, Hubmacher D. Acromelic dysplasias: how rare musculoskeletal disorders reveal biological functions of extracellular matrix proteins. Ann N Y Acad Sci. 2021;1490(1):57–76. pmid:32880985
  78. 78. Rauer C, Sen N, Waman VP, Abbasian M, Orengo CA. Computational approaches to predict protein functional families and functional sites. Curr Opin Struct Biol. 2021;70:108–122. pmid:34225010
  79. 79. Levin M, Udi Y, Solomonov I, Sagi I. Next generation matrix metalloproteinase inhibitors—Novel strategies bring new prospects. Biochim Biophys Acta Mol Cell Res. 2017;1864(11 Pt A):1927–1939. pmid:28636874
  80. 80. ADAMTS-1 cleaves a cartilage proteoglycan, aggrecan—Kuno—2000—FEBS Letters—Wiley Online Library;. Available from:
  81. 81. Santamaria S, Yamamoto K, Teraz-Orosz A, Koch C, Apte SS, de Groot R, et al. Exosites in Hypervariable Loops of ADAMTS Spacer Domains control Substrate Recognition and Proteolysis. Sci Rep. 2019;9(1):10914. pmid:31358852
  82. 82. Jiang L, Lin J, Zhao S, Wu J, Jin Y, Yu L, et al. ADAMTS5 in Osteoarthritis: Biological Functions, Regulatory Network, and Potential Targeting Therapies. Front Mol Biosci. 2021;8:703110. pmid:34434966
  83. 83. Owji H., Nezafat N., Negahdaripour M., Hajiebrahimi A. & Ghasemi Y. A comprehensive review of signal peptides: Structure, roles, and applications. Eur J Cell Biol. 2018;97(6):422–441. pmid:29958716
  84. 84. Boon L, Ugarte-Berzal E, Vandooren J, Opdenakker G. Protease propeptide structures, mechanisms of activation, and functions. Crit Rev Biochem Mol Biol. 2020;55(2):111–165. pmid:32290726
  85. 85. Koo BH, é JM, Somerville RP, Alexander JP, Leduc R, Apte SS. Regulation of ADAMTS9 secretion and enzymatic activity by its propeptide. J Biol Chem. 2007;282(22):16146–16154. pmid:17403680
  86. 86. Koo BH, Apte SS. Cell-surface processing of the metalloprotease pro-ADAMTS9 is influenced by the chaperone GRP94/gp96. J Biol Chem. 2010;285(1):197–205. pmid:19875450
  87. 87. Hubmacher D, Schneider M, Berardinelli SJ, Takeuchi H, Willard B, Reinhardt DP, et al. Unusual life cycle and impact on microfibril assembly of ADAMTS17, a secreted metalloprotease mutated in genetic eye disease. Sci Rep. 2017;7:41871. pmid:28176809
  88. 88. Majerus EM, Zheng X, Tuley EA, Sadler JE. Cleavage of the ADAMTS13 propeptide is not required for protease activity. J Biol Chem. 2003;278(47):46643–46648. pmid:12975358
  89. 89. Kutz WE, Wang LW, Bader HL, Majors AK, Iwata K, Traboulsi EI, et al. ADAMTS10 protein interacts with fibrillin-1 and promotes its deposition in extracellular matrix of cultured fibroblasts. J Biol Chem. 2011;286(19):17156–17167. pmid:21402694
  90. 90. Turpeinen H, Kukkurainen S, Pulkkinen K, Kauppila T, Ojala K, nen VP, et al. Identification of proprotein convertase substrates using genome-wide expression correlation analysis. BMC Genomics. 2011;12:618. pmid:22185580
  91. 91. Karoulias SZ, Taye N, Stanley S, Hubmacher D. The ADAMTS/Fibrillin Connection: Insights into the Biological Functions of ADAMTS10 and ADAMTS17 and Their Respective Sister Proteases. Biomolecules. 2020;10(4). pmid:32290605
  92. 92. Longpré JM, McCulloch DR, Koo BH, Alexander JP, Apte SS, Leduc R. Characterization of proADAMTS5 processing by proprotein convertases. Int J Biochem Cell Biol. 2009;41(5):1116–1126. pmid:18992360