Homologous Sequence in Lumican and Fibromodulin Leucine-rich Repeat 5-7 Competes for Collagen Binding*

Lumican and fibromodulin compete for collagen type I binding in vitro, and fibromodulin-deficient mice have 4-fold more lumican in tendons. These observations indicate that homologous sequences in lumican and fibromodulin bind to collagen type I. Here, we demonstrate that lumican binding to collagen type I is mediated mainly by Asp-213 in leucine-rich repeat (LRR) 7. The mutation D213N in lumican impairs interaction with collagen, and the lumican fragment spanning LRRs 5-7 is an efficient inhibitor of collagen binding. Also, the lumican LRR 7 sequence-based synthetic peptide CYLDNNKC inhibits the binding to collagen. Homologous collagen-binding site in fibromodulin, located in LRRs 5-7, inhibits the binding of lumican to collagen, and the mutation E251Q in this fibromodulin fragment does not inhibit the lumican-collagen binding. Lumican, but not the D213N mutation, lowers the melting point and affects the packing of collagen fibrils.

Lumican is expressed in many, mainly loose, connective tissues (1). It belongs to the family of small leucine-rich repeat proteoglycans (SLRPs) 2 that has a varied, often tissue-specific influence on collagen matrices, as demonstrated in several SLRP-deficient mice (2)(3)(4)(5). In particular, lumican-deficient mice have fragile skin, as well as opaque corneas with abnormally formed collagen fibrils in the posterior stroma, which results in 3-fold increased backscattering of light (6 -9). The amount of lumican is also increased in tendons of fibromodulin-deficient mice (10). Fibromodulin is structurally homologous to lumican, and their primary sequences are 47% identical. Fibromodulin-deficient mice have deviations in collagen fibril structures in tendons, which provoke secondary phenotypes like osteoarthritis of articular knee cartilage (10,11). In addition, both lumican and fibromodulin can reciprocally inhibit their collagen interaction (12). The increased deposition of lumican in fibromodulin-deficient mice suggests that lumican binds to non-occupied fibromodulin-binding sites in collagen I (12).
Some novel functions have recently been reported for lumican. In lumican-deficient mice, toll-like receptor-4 signaling by lipopolysaccharide is diminished, which leads to reduced response to septic shock (13). Because lumican is an extracellular protein and the extracellular domain of toll-like receptor-4 is made of LRR domains, these two LRR proteins could interact and cooperate in a spatial chain of events. Lumican can also prevent cleavage of collagen by MMP-13 in vitro or at least delay the reaction. In contrast to fibromodulin, lumican is not degraded by this enzyme (14,15). Lumican may also be involved in collagen calcification as it interacts with hydroxyapatite and is present in the areas of bone and tooth ossification (16,17).
Despite their close homology and similar collagen-binding site, fibromodulin and lumican do not appear to be functionally redundant. Judging from the knock-out mice phenotype, fibromodulin is primarily involved in the formation of thick collagen fibers, whereas lumican controls the early formation of thin fibrils (18). Mice deficient in both lumican and fibromodulin have a more severe tendon phenotype than fibromodulin-deficient mice (19). In addition, both SLRPs delay the turbidity of spontaneous in vitro collagen fibril formation (20,21). However, it is not known whether or how the functional differences relate to the collagen binding specificity of the two SLRPs.
In this report, we examined the lumican-collagen-binding site and how it relates to fibromodulin. We used recombinant and mutated lumican or fibromodulin fragments in direct and in competition collagen binding assays. We also determined how the lumican mutations influence collagen packing in in vitro assembled collagen fibrils.

EXPERIMENTAL PROCEDURES
Homology Modeling of Lumican-The tertiary structure of lumican was modeled with Swiss-PdbViewer software (22) using the crystal structure of decorin (Protein Data Bank (PDB) number 1xku) as template.
Expression of GST-tagged Lumican and Fibromodulin Fragments in Bacteria-Human lumican cDNA (GenBank TM accession number BC007038) was used as template for PCR amplification of cDNAs of lumican fragments. The fragments spanned primary sequences: LumN-4 (amino acids 19 -152), Lum5-7 (amino acids 147-220), and Lum8 -12 (amino acids 215-338). The PCR primers are listed in supplemental Fig. 1. Next, the cDNAs were digested with restriction enzymes BamHI and SmaI and ligated into pGEX-4T-3 expression vector (Amersham Biosciences, Uppsala, Sweden). The final constructs were sequenced to confirm the identity and transfected into Rosetta-gami Escherichia coli (Novagen, Madison, WI), and proteins were expressed according to the manufacturer's instructions. Proteins were purified under native conditions by glutathione-affinity chromatography, as described previously (23), and dialyzed against PBS with 0.05% (v/v) Tween 20. The protein concentration was determined with Coomassie Blue protein assay reagent (Pierce). A similar procedure was used for making fibromodulin fragments Fmod5-7 and Fmod11-12 (24). Proteins usually co-purify with some free GST protein (around 28 kDa), but free GST does not bind to collagen (data not shown).
Expression of His-tagged Lumican and Fibromodulin in Bacteria-Wild type lumican cDNA was used as template in PCR (for primers, see supplemental Fig. 1) to amplify lumican cDNA with EcoRI and XhoI restriction enzyme flanking sites. After digestion with EcoRI and XhoI, cDNA was ligated into pET-27b(ϩ) expression vector (Novagen), and the His-tagged protein was expressed in Rosetta-gami E. coli (Novagen). Lumican was purified with nickel-nitrilotriacetic acid agarose (Qiagen, Hilden, Germany) under denaturing conditions according to the manufacturer's instructions. Prior to use, protein was dialyzed against PBS. Similar procedure was used for making fibromodulin, as described earlier (24). The proteins migrated faster on a non-reduced SDS-PAGE than on a reduced gel, indicating the presence of disulfide bonds.
Mammalian Expression of Lumican-Lumican cDNA (BC007038) was amplified by PCR (primers listed in supplemental Fig. 1), digested with XhoI and BamHI, and ligated into pCEP4 BM40-hisEK expression vector (25), attaching lumican cDNA to an N-terminal His tag. The constructs were confirmed by sequencing and transfected into human embryonic kidney 293 cells by electroporation. Cells were grown in Dulbecco's modified Eagle's medium containing 10% (v/v) fetal calf serum (Invitrogen, Lidingö, Sweden). After 24 h, hygromycin (Invivogen, Toulouse, France) was added to a concentration of 150 g/ml. After 1 week, single clones were picked, incubated in medium containing [ 35 S]sulfate (tagging tyrosine sulfate residues), and analyzed for proteoglycan expression. The proteoglycan-expressing clones were cultured in EX-Cell 325 PF Chinese hamster ovary protein-free medium (JRH Biosciences, Lenexa, KS) for 12 days with medium changes every 3 days. The medium was dialyzed against 300 mM NaCl and 20 mM NaH 2 PO 4 buffer, pH 8.0, and proteoglycans were purified in native form using nickel-nitrilotriacetic acid agarose (Qiagen). Lumicans were run on SDS-PAGE to check purity and also treated with N-glycosidase F to check for the presence of N-linked glycosaminoglycans (see Fig. 3A).
Site-directed Mutagenesis and Expression of Mutated Lumican and Fibromodulin in Bacteria-Lumican cDNA (BC007038) or fibromodulin cDNA (X16485) was used for sitedirected mutagenesis using a QuikChange II kit with pBluescript II KS (ϩ/Ϫ) cloning vector (Stratagene, Kista, Sweden) according to the manufacturer's instructions. Primers used for lumican D213N and D222N and fibromodulin E251Q mutations are listed in supplemental Fig. 1. The mutated cDNA was ligated into pET-27b(ϩ) expression vector for bacterial expression or in pCEP4-BM40 hisEK vector for mammalian expression. Proteins were expressed and purified as described above.
For mutation E251Q in the Fmod5-7 fragment, we used similar procedure, as described previously (24).
Peptide Synthesis-Linear peptide TLYLDNNKIS and the cyclic peptides CYLDNNKC and CYLKNNKC were purchased from Schafer-N (Copenhagen, Denmark) and dissolved in Trisbuffered saline with 0.05% (v/v) Tween 20.
Solid-phase Collagen Binding Assay-Microtiter 96-well Maxisorp plates (Nunc, Roskilde, Denmark) were coated for 16 h using 100 g/ml acid-soluble collagen. The wells were washed twice with PBS and blocked for 2 h with PBS with 1 mg/ml bovine serum albumin (BSA) and 0.05% Tween 20. Next, the wells were washed with incubation buffer (20 mM Tris, pH 8.0, with 0.1% (w/v) BSA, and 0.05% (v/v) Tween 20), and the GST-tagged lumican fragments or full-length proteins (diluted in incubation buffer) were added and incubated for 4 h. Collagen binding was detected with rabbit anti-GST antibody (ab9085, AbCam, Cambridge, UK) for lumican fragments or rabbit anti-His antibody (ab9108, AbCam) for full-length lumicans, diluted 1:1000 in the incubation buffer. After 1 h, the wells were washed and then incubated for 1 h with anti-rabbit alkaline phosphatase-conjugated antibody (ab6722, AbCam) diluted 1:1000 in incubation buffer. Lastly, the wells were washed and then incubated with phosphatase substrate p-nitrophenyl phosphate (Sigma, Stockholm, Sweden), and protein binding was detected by measuring absorbance at 405 nm. K D values were calculated using non-linear regression sigmoidal dose-response formula Y ϭ Bottom ϩ (Top Ϫ Bottom)/(1 ϩ 10ˆ((LogEC 50 Ϫ X)), solving for EC 50 and assuming one-site binding. K i values were calculated using one-site competition formula Y ϭ Bottom ϩ (Top Ϫ Bottom)/(1 ϩ 10ˆ((X Ϫ LogIC 50 )), solving for IC 50 . For these calculations, GraphPad Prism (GraphPad Software, Inc.) was used.
Differential Scanning Calorimetry-Acid-soluble collagen PureCol from Nutacon, Leimuiden, Holland was dialyzed against PBS at 4°C and then incubated with mammalian-expressed lumican (wild type and D213N-mutated) at a molar ratio 1:10 (lumican:collagen) at 37°C for 16 h. The samples were then run in a VP-DSC calorimeter (MicroCal) in a 30 -60°C temperature gradient, at a scanning rate of 1°C/min. Sample thermograms were subtracted from blank (buffer only) thermograms and normalized for collagen concentration. The contribution of lumican to the melting thermogram of collagen was negligible due to its low concentration (5 g/ml). Data were analyzed with Origin software supplied with VP-DSC.

RESULTS
To study the interaction between lumican and collagen, we expressed GST-tagged lumican fragments that covered overlapping parts of the protein (Fig. 1, A and B). LumN-4, covering the N-terminal part of lumican and spanning to LRR 4, did not interact with collagen type I in a solid-phase assay, and neither did Lum8 -12 that contains LRRs 8 -12. Only Lum5-7 derived from the central part of the protein, containing LRRs 5-7, showed collagen binding (K D ϳ 45 nM) (Fig. 1C). Lum5-7 also inhibited the interaction of full-length mammalian expressed lumican (K i ϳ 250 nM), which none of the other fragments did (Fig. 2). The fragments did not interact with the BSA coat without the presence of collagen (data not shown).
We used site-directed mutagenesis of lumican cDNA and expressed full-length mutated lumicans that appear to carry short-chained N-linked glycosaminoglycans, indicating that cells produce a glycoprotein form of lumican (Fig. 3A). To select putative candidates for mutation, we used sequence alignment of lumican LRRs 5-7 from 11 species deposited in the Uni-ProtKB data base. We also compared these sequences with the corresponding fibromodulin LRRs 5-7 because Fmod5-7 inhibits the lumican-collagen interaction (supplemental Fig. 2). We searched for charged residues near the ␤-sheet loop region of the LRR domains because this region was previously implicated in the interaction of SLRPs (26). Asp-213 in human lumi-can LRR 7 (Fig. 4A) is semiconserved with Glu-251 in human fibromodulin. Consequently, full-length lumican with D213N mutation (LumD213N) and control (LumD222N) were expressed in mammalian cells and analyzed for their collagen binding. Wild type lumican and LumD222N bound collagen at K D ϳ 30 nM with a fitted sigmoidal binding curve, whereas LumD213N interacted weakly with collagen, with a linear binding curve (Fig. 3B). We also observed that a synthetic linear peptide (TLYLDNNKIS) based on lumican LRR 7 sequence did not inhibit the lumican-collagen interaction (data not shown), although the cyclic peptide (CYLDNNKC) was an efficient inhibitor at K i ϳ 2 M (Fig. 3C).
Because the closest homologue of lumican is fibromodulin and both bind to the same region of collagen, we also tested whether any fibromodulin fragment used in our previous study (24) could inhibit the lumican-collagen interaction. Fmod5-7, containing LRRs 5-7, impaired the binding of lumican to collagen (K i ϳ 250 nM), whereas other fragments were ineffective. Fmod8 -12, the previously identified main collagen-binding fragment of fibromodulin, did not contribute to the inhibitory activity of Fmod5-7 (Fig. 4B). To further explore the competition between lumican and fibromodulin for collagen binding, we introduced E251Q in full-length fibromodulin and in fibromodulin fragment Fmod5-7. This mutation reduced the inhibitory activity of Fmod5-7 but not of full-length fibromodulin (Fig. 4C).
Finally, we studied the effect of the D213N mutation on lumican function during collagen fibrillogenesis. We used differential scanning calorimetry to analyze the denaturation thermograms of collagen preincubated with mammalian-expressed lumican (wild type and LumD213N) in PBS, at 37°C for 16 h. We used the molar ratio lumican:collagen 1:10, at which the binding of lumican appears to be saturated, because higher concentrations do not further affect the collagen denaturation thermograms. In this functional assay, two melting points are observed; an early 41°C-peak represents the denaturation of FIGURE 1. Interactions of lumican fragments with collagen. A, lumican was modeled after the decorin crystal structure (1xku.pdb) and energy-minimized. Fragments expressed as GST fusion proteins are depicted. B, expression of lumican fragments as GST fusion proteins. SDS-PAGE on bacterial total lysate (left panel) and purified (right panel) lumican fragments. C, solid-phase collagen binding assay. 96-well plates were coated with acid-soluble collagen type I, blocked with BSA, and incubated with LumN-4 (f), Lum5-7 (OE), or Lum8 -12 (B), and the binding was detected with rabbit anti-GST followed by alkaline phosphatase-conjugated anti-rabbit antibody (A 405 nm ). Protein concentration is plotted on a semilog scale. Three replicates of each sample were used to calculate means and standard deviations (error bars).

FIGURE 2. Inhibitions of lumican-collagen interaction by lumican fragments.
A solid-phase collagen binding assay was performed. 96-well plates were coated with acid-soluble collagen type I, blocked with BSA, and incubated with 100 ng of full-length mammalian-expressed lumican together with increasing concentrations of lumican fragments: LumN-4 (f), Lum5-7 (OE), or Lum8 -12 (ƒ). Binding of lumican was detected by rabbit anti-His followed by alkaline phosphatase-conjugated anti-rabbit antibody (A 405 nm ). Protein concentration is plotted on a semilog scale. Three replicates of each sample were used to calculate means and standard deviations (error bars). monomeric collagen, whereas the late 52°C-peak corresponds to the denaturation of the fibrils. In this system, wild type lumican reduces the melting point of collagen fibrils by about 1°C, whereas LumD213N produces a thermogram similar to the control without lumican (Fig. 5).

DISCUSSION
We demonstrate that lumican binding to collagen type I is mediated mainly by Asp-213 in LRR 7. This maps the interaction to the ␤-sheet loop region of the LRR domain, which is similar to other LRR proteins including the close homologues  (PNGaseF). B, solid-phase collagen binding assay. 96-well plates were coated with acid-soluble collagen type I, blocked with BSA, and incubated with increasing concentrations of full-length wild type (Wt) (f) or mutated lumicans: LumD213N (OE) or LumD222N (ƒ). Binding of lumican was detected by rabbit anti-His followed by alkaline phosphatase-conjugated anti-rabbit antibody (A 405 nm ). Protein concentration is plotted on a semilog scale. Three replicates of each sample were used to calculate means and standard deviations (error bars). C, inhibition of the lumican-collagen interaction by synthetic peptides on solid-phase assay. 96-well plates were coated with acid-soluble collagen type I, blocked with BSA, and incubated with 100 ng of mammalian-expressed lumican and increasing concentrations of peptide CYLDNNKC (Ⅺ) or control peptide CYLKNNKC (OE). The binding of lumican was detected with anti-His antibody followed by alkaline phosphatase-conjugated anti-rabbit antibody (A 405 nm ). Protein concentration is plotted on a semilog scale. Three replicates of each sample were used to calculate means and standard deviations (error bars).
fibromodulin and decorin (24,27). In general, this implies that LRR domains contain interaction sites near the distal parts of the conserved ␤-sheet structures and include charged amino acids. In many cases, the full-length SLRPs bind more strongly than the fragments of the protein to collagen (24). The characteristic curved shape of LRR proteins may engage additional weak collagen interactions mediated by several LRR domains, thus creating a cooperative high affinity collagen binding.
The mutation D213N in lumican LRR 7 impaired the interaction with collagen, and the fragment spanning LRRs 5-7, as well as the cyclic peptide CYLDNNKC based on the LRR 7 sequence, were both inhibitors of the lumican-collagen interaction. Linear peptide TLYLDNNKIS did not inhibit the interaction, suggesting that the interaction depends on a specific conformation.
Interestingly, fibromodulin does not inhibit the lumican-collagen interaction via the high affinity collagen-binding site located in LRR 11 because Fmod8 -12 or the peptide (RLDG-NEIKR) are both inactive as inhibitors (24). The low affinity collagen-binding site of Fmod5-7 (LRRs 5-7) contains Glu-251, homologous and topologically related to lumican Asp-213, and inhibits the binding of lumican to collagen (Fig. 4A). The fragment Fmod5-7 with an E251Q mutation does not inhibit lumican binding. In contrast, the full-length fibromodulin with the same E251Q mutation inhibits lumican collagen binding. This indicates a close proximity of the Fmod/Lum LRRs 5-7 and Fmod LRR 11 collagen-binding sites. The full-length fibromodulin, with a functional LRR 11 collagen binding site, may sterically hinder the binding of lumican despite the mutated LRRs 5-7 collagen-binding site.
The alternate collagen binding of fibromodulin and lumican may have a role during the development of collagen matrices. Tendons of fibromodulin-deficient mice have an altered colla-gen cross-linking pattern, 3 and the amount of lumican is increased some 4-fold in tendons from these mice (10). The absence of fibromodulin most likely leads to more binding of lumican to unoccupied collagen-binding sites in fibromodulindeficient tendons. During tendon development, both lumican and fibromodulin have a role in the initial assembly of intermediates, whereas fibromodulin facilitates growth steps leading to mature fibrils (18). It has also been demonstrated that the fibromodulin mRNA level is increased 6 -8-fold in chicken tendons between day 14 and day 19 of development, which coincide with a dramatic increase in collagen fibril length (28). Increased expression of fibromodulin with a higher affinity for collagen may displace lumican from collagen and appears to have a role in tendon developmental transitions. We propose that the initial cross-linking of collagen can be regulated by both lumican and fibromodulin, but fibromodulin can regulate the fibril assembly further by connecting two assembling collagen units through its two collagen-binding sites. This would later result in regulation of the mature cross-link formation (Fig. 6).
The binding of combinations different SLRPs may shape collagen molecules into specialized matrix networks, such as tendon and cornea. This may occur by binding to different or the same region of collagen. Decorin and fibromodulin bind to separate collagen sites, but here, we show that fibromodulin and lumican can bind to and compete for the same collagen-binding site. This SLRP-collagen binding could determine how collagen monomers interact and juxtapose lysine/hydroxylysine residues involved in cross-link formation. SLRP-collagen molecules may thus be aligned or shielded to allow the formation of 3 S. Kalamajski and Å. Oldberg, unpublished observation.  . The proposed mechanism of lumican and fibromodulin regulation of collagen fibrillogenesis. A, lumican (green) binds to collagen monomers through LRR 7 and regulates their initial cross-linking (red). B, fibromodulin (yellow) binds to collagen monomers preferentially with its high affinity site in LRR 11, regulates the initial cross-linking, and can compete with collagen binding with the collagen-bound lumican. C, as lumican is dislodged from the collagen by fibromodulin, with the homologous collagenbinding sequence in LRR 7, the two units of initially cross-linked collagens can assemble further in third dimension, and the mature cross-links can be formed between the collagens (red). a specific cross-linking pattern. In this manner, the differential tissue expression of SLRPs could regulate the mechanical properties of collagen matrices.
We also show that lumican binding to collagen regulates the hydration of the fibrils, which can be determined by differential scanning calorimetry (29 -31). The bound lumican changes the hydration of the polymer by sterical hindrance, leading to a lower melting temperature of the fibrils assembled without lumican. In contrast, LumD213N (not binding collagen) does not influence the denaturation of the fibrils.