The matrilin-1 A1 and A2 domains were amplified from a full-length human cDNA clone by PCR and subcloned into the pSecTagA vector between SfiI and XhoI restriction sites of the multiple cloning site. These cDNA fragments were then re-amplified by PCR. The forward primer encompassed the first 20 nucleotides (from the ATG start codon upstream of the Ig κ-chain secretion signal) found in pSecTag and included a NotI restriction site (primer 5′-gcggccgcatggagacagacacact-3′). The reverse primers also contained an engineered NotI cleavage site in addition to an in-frame FLAG tag (sequence, DYKDDDDK) (5′-gcggccgctcacttatcgtcgtcatccttgtaatcgtctgacaccacgcagaaggcc-3′ for M1A1 and 5′-gcggccgctcacttatcgtcgtcatccttgtaatcgtcttcctccacacagatctt-3′ for M1A2). PCR products were subcloned using the TA Cloning method (Invitrogen), digested with the restriction enzyme NotI, cloned into the pCEP4 expression vector (Invitrogen), and sequenced. A single correct clone for each construct (pCEP4-M1A1 and pCEP4-M1A2) was used for all subsequent experiments.

To abolish the N-glycosylation sites, the asparagine residue from the recognition motifs (NXT/S) was mutated to an alanine. PCR-based site-directed mutagenesis was performed on both wild type pCEP4-M1A1 and pCEP4-M1A2 clones using the following primer pairs: 5′-gacgtggggcccgctgccacccgggtg-3′ and 5′-cacccgggtggcagcgggccccacgtc-3′ for M1A1; 5′-gcggctgtgcgggctatgtcctacatg-3′ and 5′-catgtaggacatagcccgcacagccgc-3′ for M1A2. The relevant amino acid codons are indicated in bold and the mutated nucleotides are underlined.

Recombinant human M1A1 and M1A2 domains were expressed as secreted proteins by 293-EBNA cells, affinity purified using the incorporated FLAG tag, and subjected to size-exclusion chromatography as described previously (15). The purity of the recombinant proteins was checked by silver staining of the gel. The identity of the recombinant proteins was confirmed by peptide mass mapping of excised band using electrospray ionization (ESI) tandem mass spectrometry.

0.5 μg of recombinant proteins were treated with 0.01 unit of PNGase F (QA Bio) for 48 h at 37 °C, and the treated protein products were analyzed by SDS-PAGE.

The amino acid sequence of each A-domain from human matrilin-1 was used to search the Swiss Model data base (EXPASY). Protein sequences resulting from this search included human von Willebrand factor A1 and A3 domains and the integrin I-domain. The sequence with the highest homology was the von Willebrand factor A3 (PBD code 1ao3-A) and this structure was used for homology modeling. The secondary structural elements from the von Willebrand factor A3 and the matrilin-1 A-domains were overlaid on a sequence alignment (using PYMOL), which located the asparagine residue involved in the glycosylation site of the proteins. The models of the matrilin-1 A-domains were built based on the coordinates of von Willebrand factor A3.

Samples of purified recombinant protein were subjected to Multiangle Laser Light Scattering (MALLS) analysis and applied to a Superdex 75 gel filtration column (GE Healthcare). Light scattering intensity and eluant refractive index were analyzed using ASTRA version 5.21 software to give a weight-averaged molecular mass (M_w). The secondary structures of the proteins variants were then recorded using a Jasco-810 spectropolarimeter. Circular dichroism (CD) spectra were deconvoluted using CDSSTR software program. Both techniques have been described elsewhere in more detail (15).

The denaturation temperatures of all proteins variants were measured using differential scanning calorimetry (VP-DSC MicroCalorimeter, MicroCal Inc.), as described (15). The denaturation curves were fitted to a non-two-state model yielding the T_m.

The binding ability of recombinant matrilin-1 A-domains to type II (Calbiochem) and type IX collagen (15) was tested on a Biacore 3000 instrument using a CM5 sensor chip. In the first instance, the Biacore was used to investigate the effect of cations on binding between both M1A domains and bovine type II and human type IX collagens. Freeze-dried collagens were re-dissolved in 500 mm acetic acid at a concentration of 1 mg/ml, then diluted to a concentration of 10 μg/ml in 10 mm Tris, pH 7.4, 150 mm NaCl containing either 1 mm Zn²⁺, Mg²⁺, and Mn²⁺ or 1 mm EDTA. These samples were injected over immobilized M1A1 (∼1000 response units bound) or M1A2 (∼700 response units bound) for 3 min at a flow rate of 30 μl/min.

Kinetic runs were also performed with type II and IX collagens using TBS running buffer containing 1 mm Zn²⁺. The method used was the same as described previously (15). The kinetic analysis was determined on the BIAevalution 4.1 software.

The kinetics of the binding between wild type and the unglycosylated forms of matrilin-1 and -3 A-domains and type II collagen were also determined using quartz crystal microbalance with dissipation (QCM-D) as an alternative to the surface plasmon resonance (SPR) method (BIAcore). QCM-D simultaneously monitors changes in resonance frequency (Δf) and dissipation (ΔD) in real time. Δf primarily measures changes in the mass attached to the oscillating sensor surface (a silicon dioxide crystal in this case), whereas ΔD measures properties related to the viscoelasticity of the adsorbed layer (a network of type II collagen) (16).

All QCM-D measurements were performed with a Q-Sense E1 System at a temperature of 20.0 °C. A SiO₂-coated crystal (Q-sense AB) with a fundamental resonance frequency (f5) of 5 MHz was equilibrated in the running buffer (10 mm Tris, pH 7.4, 150 mm NaCl, 1 mm Zn²⁺). 50 μg/ml of type II collagen (diluted in the same running buffer) was then adsorbed at a flow rate of 100 μl/min until saturation (∼3 h).

The kinetic evaluation of the interactions between type II collagen and the matrilin A-domains was performed by exposing the immobilized collagen to a series of successively increasing concentrations of: 1) matrilin-3 A-domain (0, 90.78, 181.56, 272.34, 363.12, 453.91, and 680.86 nm); 2) matrilin-1 A1 (0, 88.1, 176.21, 264.32, 352.42, 440.53, and 660.79 nm); and 3) matrilin-1 A2 (0, 86.6, 173.16, 259.74, 346.32, 432.9, and 649.35 nm). After each of the concentration steps, the bound proteins were eluted with TBS containing 1 mm EDTA and allowed to stabilize in the running buffer before passing over the next concentration. All the experiments were repeated three times.

The maximum association rate (or V_max) for each concentration was derived using the following Hill equation (equivalent to Langmuir equation), where y is the measured frequency change, x the time (s), and k is the Michaelis constant. In this case, n = 1, which corresponds to a non-cooperative reaction (the affinity between the two proteins is not dependent on whether or not other ligand molecules are already bound). The matrilin A-domain binding (or affinity) was then determined by plotting each calculated V_max as a function of increasing analyte concentrations (supplemental Figs. S2) and fitted to the same model generating the K_D (the equilibrium dissociation constant). This non-linear curve fitting method was performed using OriginPro8 software.

A different approach was employed to compare the behaviors of all A-domains variants following binding to type IX collagen. A solution of 10 μg/ml of type IX collagen was immobilized onto the crystal surface until it reached saturation. Then 5 μg/ml of each matrilin A-domain was sequentially injected and each experiment was repeated three times. The proteins concentrations were determined spectrophotometrically using extinction coefficients at 280 nm of 0.411 for M3A, 0.265 for M1A1, and 0.463 for M1A2.

This technique was also used to study complex formation between type VI collagen (Chemicon), biglycan (a kind gift from Prof. P. Bishop), matrilin-1 A-domains, and type II collagen in real time. Prior to type VI collagen adsorption, the crystal was brought into contact with TBS containing 1 mm Zn²⁺ to establish a baseline. 10 μg/ml of the collagen was then immobilized onto the surface until saturation. A succession of 5 μg/ml of biglycan, M1A1, M1A2, and type II collagen was then injected to form the complex. All proteins were diluted in the same running buffer (TBS + 1 mm Zn²⁺). Western blotting was performed using antibodies to type VI collagen (Chemicon), biglycan (Abcam), matrilin-1 A-domains (anti-FLAG, Sigma), and type II collagen (Chemicon).

Understanding the sequence relationship between the different A-domains of the matrilin protein family can unravel the evolutionary history of these proteins as well as provide information about their potential function. For example, it has been previously proposed by Deák et al. (8) that all of the matrilin proteins evolved from a common ancestor. This suggestion is based on the observation that they all comprise at least one vWFA domain (or A-domains), a variable number of EGF-like repeats and a single coil-coiled oligomerization domain. However, during evolution the matrilin-3 A-domain closest to the coil-coiled domain is believed to have been lost (8) (Fig. 1A). The significant sequence similarity between the single matrilin-3 A-domain and the matrilin-1 A1-domain (Fig. 1B; 62% sequence homology) implies that they have the same evolutionary origin and possibly the same structure and function. In contrast, the matrilin-3 A-domain and the matrilin-1 A2-domain are more distantly related and share only 41% sequence homology (Fig. 1B).

The amino acid sequence alignments of M3A, M1A1, and M1A2 also reveal that all three proteins possess a signature sequence, DXSXS…(T/S)…D, referred to as the MIDAS motif as described previously (8, 17) (Fig. 1C). Furthermore, both M1A1 and M1A2 proteins contain a potential glycosylation directing sequence defined by the sequons NAT for M1A1 and NMS for M1A2 (Fig. 1B). Multiple sequence alignment of M1A1 sequences revealed conserved N-linked glycosylation sites in human, mouse, and chick. M1A2 sequences showed conserved N-linked glycosylation sites only in human and mouse but not in chick. Moreover both matrilin-1 A-domains lack the N-glycosylation site in zebrafish (Fig. 1C).

Matrilin-1 A-domains (M1A1 and M1A2) were purified as secreted proteins and appeared as doublets with an apparent molecular mass under denaturing conditions of ∼25 kDa according to SDS-PAGE (Fig. 2A). Tryptic peptide analysis by ESI tandem mass spectrometry validated the identity of both doublets as M1A1 and M1A2, respectively (not shown). Our finding that both A-domains ran as doublets prompted us to hypothesize that the lower bands were the unglycosylated forms, whereas the upper bands were the glycosylated forms of the proteins. This hypothesis was supported by the observation that both protein sequences contain a potential N-glycosylation site (Fig. 1C). A homology model of the M1A1 domain located the only possible N-glycosylation site at the opposite face to the MIDAS motif, whereas the only possible N-glycosylation site of the M1A2 domain was situated on the α2 helix (Fig. 2B).

An N-glycanase digestion performed on the purified proteins showed that both doublets became a sharper single band, confirming that a proportion of the molecules were glycosylated (Fig. 2C). Interestingly, the M1A1 domain appeared mostly glycosylated (with a relative amount of 70% of glycosylated species against 30% of unglycosylated quantified using densitometry), whereas the M1A2 domain contained a comparable proportion of glycosylated and unglycosylated protein (50% of glycosylated and 50% unglycosylated species). This suggested that both glycosylation-directing sequences of the matrilin-1 A-domains are accessible on the surface of the proteins, thereby allowing the attachment of N-glycan chains in vitro.

To determine the effect of glycosylation on the secretion of the recombinant proteins, mutant M1A1 and M1A2 domains were generated in which the N-glycosylation sites were abolished by replacing the appropriate asparagine residue with an alanine residue. The cell lysates and the conditioned medium were analyzed by SDS-PAGE and Western blotting (Fig. 2D), which confirmed that both of these unglycosylated proteins were secreted.

The wild type and unglycosylated proteins were then run on a size exclusion column coupled to a multiangle laser light scattering detector to determine their average molecular weights (M_r). The molecular mass of wild type M1A1 was calculated to be 23.4 kDa and the unglycosylated form to be 0.7 kDa smaller at 22.7 kDa (Fig. 2E; left-hand panel). The chromatograph for the M1A2 protein showed that it eluted in two peaks, which corresponded to the upper and lower bands of the doublet (Fig. 2E, right-hand panel). The first peak was defined as the glycosylated form of M1A2 and gave a molecular mass of 24.9 kDa. The elution position of the second peak of the protein overlaid perfectly with the one from the unglycosylated form and was found to be 1.8 kDa smaller at 23.1 kDa. These data derived from the light scattering experiments therefore supported our initial assessment that both domains possess one N-linked polysaccharide chain each.

To assess if glycosylation had any significant effect on the conformation of the A-domains, the secondary structure characteristics of the four proteins was determined by circular dichroism (CD). The M1A1 CD spectrum showed a slight decrease in the relative amounts of α-helix content and a subsequent increase in β-sheet elements compared with M1A2 (Fig. 3A). This was confirmed by calculating the average values for secondary structure composition following CDSSTR analysis. The M1A1 domain was found to be composed of 9% α-helix, 36% β-sheet, 24% turn, and 31% random coil and the M1A2 domain of 15% α-helix, 30% β-sheet, 24% turn, and 31% random coil. Moreover, both matrilin-1 A-domains exhibited a smaller β-strand proportion and a greater α-helical content than the matrilin-3 A-domain (15). These data also revealed that there was no significant difference in the secondary structure between the wild type and unglycosylated forms of the proteins suggesting that glycosylation does not affect the folding of the individual domains.

Glycosylation can also influence the behavior of a protein whereby the addition of the N-glycans can decrease the amino acid core flexibility resulting in an increased thermal stability (18), perhaps in an attempt to increase protein longevity. To determine thermal stability of the matrilin-1 A-domains, differential scanning calorimetry was used. The denaturation temperature of the wild type M1A1 domain was within the temperature ranges of 56–57 and ∼49 °C for the unglycosylated form (Fig. 3B). The thermal stability of the wild type and unglycosylated M1A2 domains was also assessed and the denaturation temperatures obtained were ∼54 and 52 °C, respectively. These data revealed that the wild type forms of the proteins were indeed more resistant to thermal unfolding than the unglycosylated forms, which provides evidence that glycosylation has evolved to give some protection and durability to the proteins.

We have recently shown that the matrilin-3 A-domain bound with high affinity to type II and type IX collagens in the presence of Zn²⁺ (15). We therefore compared the role of different cations (1 mm each of Zn²⁺, Mn²⁺, and Mg²⁺) on the binding properties of the matrilin-1 A-domains using SPR (Fig. 4A). These studies demonstrated that whereas 1 mm Zn²⁺ supported strong binding of type II and IX collagens to the wild type M1A1 and M1A2 domains, in contrast, binding in the presence of 1 mm Mn²⁺ and Mg²⁺, or 1 mm EDTA (used as control) was negligible or completely abolished. This suggests that all three individual A-domains (M3A, M1A1, and M1A2) have acquired specificity for Zn²⁺ binding.

The binding kinetics of the wild type and unglycosylated matrilin-1 A-domains to type II collagen was studied further by QCM-D to determine whether glycosylation influenced molecular associations. During the adsorption phase type II collagen caused an initial rapid frequency decrease (mass increase) followed by a slower frequency decrease as the crystal saturated (supplemental Fig. S1, phase 1). By using this technique we were able to form a stable layer-by-layer architecture of type II collagen molecules without loosing any activity (which was the case when using SPR). Type II collagen showed evidence of oligomerization on the surface of the chip in a similar way to collagen type I shown by Johansson et al. (21), with collagen II forming fibrils before laying flat on the surface with a layer depth of 85 nm (similar to that for a single fibril). The fibrillar collagen surface could be repeatedly regenerated by removal of bound proteins using 1 mm EDTA (supplemental Fig. S1, phase 5), thus enabling multiple matrilin A-domain binding experiments (supplemental Fig. S1, phase 4). Successive incubations with increasing amounts of the different matrilin A-domains (M3A, M1A1/M1A2 both wild type and unglycosylated) allowed us to determine their relative affinities for type II collagen.

The M1A2 domain showed the strongest affinity with a K_D of ∼50 nm compared with M3A (∼91 nm) and M1A1 (∼104 nm) (Fig. 5A). This trend was in agreement with the calculated K_D as determined by SPR analyses, which were 2.13 nm for M1A2, 6.2 nm for M3A, and 8.8 nm for M1A1 (Fig. 4B, top panel, and table). The difference in binding between M1A2 and M3A was also statistically significant (p < 0.001). However, no difference was detected between wild type M1A2 and the unglycosylated form (probably due to the fact that wild type M1A2 contains a high proportion of unglycosylated protein resulting in a masking effect). An interesting finding was that the unglycosylated form of M1A1 bound more strongly to type II collagen than the wild type molecule. This suggests that the flexible N-glycan chains might partially hide the binding site for type II collagen, thus narrowing the location of the binding site for this molecule to the opposite face of the MIDAS motif (Fig. 2D).

Subsequent data analysis by plotting frequency and dissipation shifts enabled us to better understand the nature of the interaction between the matrilin-1 and -3 A-domains and type II collagen. A comparison of the change in resonance frequency (bound mass) versus time revealed a fast binding of all three matrilin A-domains to type II collagen (Fig. 5B, top left panel). Measurements of the damping effect (dissipation factor) during binding experiments between the matrilin A-domains and type II collagen suggested that the matrilin-1 A-domains formed a more rigid structure with type II collagen than M3A did (Fig. 5B, bottom left panel). By combining f and D data we were able to obtain information about structural changes of the collagen surface following the binding of the matrilin A-domains. It is evident from the plot (Fig. 5B, right panel) that the M1 A-domains and M3A affected the immobilized collagen layer in a very different way. When M3A was added it bound to type II collagen in such a way that the resulting complex was more flexible, which was reflected by a higher ΔD/Δf ratio than the M1 A-domains. In contrast, the lower ΔD/Δf ratio measured for the four M1 A-domains indicated a more rigid complex.

In this series of experiments we used QCM-D to investigate changes in frequency and energy dissipation during adsorption of the different matrilin A-domains to a SiO₂ surface covered with a stable film of type IX collagen with a thickness of 6.5 nm. From the Δf versus time plot it can be seen that the M3A domain had a faster association rate than the M1A1 and M1A2 domains (Fig. 5C, top left panel) (i.e. following addition of the M3A solution a rapid frequency decrease (mass increase) occurred). The adsorption kinetics of the wild type M1A1 and its unglycosylated form exhibited a difference in mass uptake as opposed to both variants of M1A2, which showed no difference. The ΔD versus time plot showed that dissipation increased with time meaning more energy was dissipated as more proteins were adsorbed (Fig. 5C, bottom left panel). This was especially the case for the M3A domain, but no significant differences in viscoelasticity of the different matrilin-1 A-domains could be observed. To correlate the observed changes in f and D, we plotted ΔD versus Δf, which does not take into account the time parameter (Fig. 5C, right panel). These data revealed differences between the adsorption behavior of the matrilin-3 A-domain and both of the matrilin-1 A-domains, which was not directly observed on the other plots. The different ΔD/Δf ratio can be divided into two groups; one for all of the matrilin-1 A-domains and the second one for the matrilin-3 A-domain alone, indicating at least two different kinetic processes occur during adsorption. The low ΔD/Δf value for M3A suggests a strong binding to type IX collagen and also indicates that the addition of this molecule triggers a more rigid compact conformation of the type IX collagen layer. All four matrilin-1 A-domains share a similar, but higher ratio, suggesting that they form a more flexible structure and have a lower affinity for the layer below (i.e. type IX collagen). Further analysis of the curves for both M1A1 domains indicated that the wild type protein binds with slightly higher affinity (lower ratio), but again no difference could be detected between the two M1A2 proteins.

Type VI collagen has previously been shown to bind a wide range of proteins in vitro, such as the fibrillar type II collagen, and the proteoglycans biglycan and decorin (19). More recently Wiberg et al. (20) described the interaction between biglycan and matrilin-1 acting as a bridging complex with type II and VI collagens. In our study we monitored in real time the complex formation between type VI collagen, biglycan, the two matrilin-1 A-domains and type II collagen using QCM-D. In the first instance a layer of type VI collagen was deposited on the sensor surface in a random fashion corresponding to the sudden decrease in frequency (or increase in mass). Once a critical density had been reached, the collagen molecules most likely organized themselves into a stable and compact network, which was reflected by the small increase in frequency and decrease in dissipation (Fig. 6A, phase 1) as described elsewhere (21). The type VI collagen showed evidence of some oligomerization before forming a flat layer of 15 nm, which is consistent with an arrangement of triple helical molecules into filaments. Biglycan was then passed over the type VI collagen layer followed by the individual matrilin-1 A-domains (M1A1 first followed by M1A2; see Fig. 6A, phase 2 and 3, respectively). All 3 proteins showed the characteristics of binding (i.e. an increase in mass with a corresponding increase in dissipation). Finally, type II collagen was injected and bound to this complex (Fig. 6A, phase 4), which was characterized by a large shift in frequency and dissipation (due to the size of the collagen molecules).

At this stage in the experiment the protein complex was eluted from the crystal surface using SDS-PAGE loading buffer. The presence of the individual proteins in the complex was then confirmed by Western blotting using specific antibodies against type VI collagen, biglycan, matrilin-1 A-domains, and type II collagen (Fig. 6B). All of the individual proteins of the complex could be detected (i.e. types II and VI collagen and matrilin-1 A1-domain) with the exception of biglycan and matrilin-1 A2-domain. We subsequently detected biglycan following silver staining; however, M1A2 was not detectable, which led us to conclude that it was not part of the stable complex. We therefore performed the subsequent experiments with M1A1 alone.

To determine specificity of this protein assembly and complex formation, each protein of the complex was individually tested for binding against each of the other components (supplemental Fig. S3A). Biglycan and the matrilin-1 A1-domain individually bound to types II and VI collagen. Moreover, a pre-assembled type II collagen and matrilin-1 A1-domain complex interacted with biglycan, but not with type VI collagen. In a similar way, type II collagen and the biglycan complex associated with the matrilin-1 A1-domain, but not with type VI collagen. Interestingly, when type II collagen was absorbed onto the crystal no detectable direct binding with type VI collagen could be seen. However, in the reverse orientation we could observe a slight binding (3.5-fold less than in the complex) between type II collagen and type VI collagen. We therefore concluded from these data that both biglycan and the matrilin-1 A1 domain are required for the association of type VI collagen with type II collagen (supplemental Fig. 3B).