Proline hydroxylation in collagen supports integrin binding by two distinct mechanisms

Collagens are the most abundant extracellular matrix proteins in vertebrates and have a characteristic triple-helix structure. Hydroxylation of proline residues is critical for helix stability, and diminished prolyl hydroxylase activity causes wide-spread defects in connective tissues. Still, the role of proline hydroxylation in the binding of collagen receptors such as integrins is unclear. Here, we isolated skin collagen from genetically modified mice having reduced prolyl 4-hydroxylase activity. At room temperature, the reduced proline hydroxylation did not affect interactions with the recombinant integrin α2I domain, but at 37 °C, collagen hydroxylation correlated with the avidity of α2I domain binding. Of note, LC–MS/MS analysis of isolated skin collagens revealed no major changes in the hydroxyproline content of the main integrin-binding sites. Thus, the disrupted α2I domain binding at physiological temperatures was most likely due to structural destabilization of the collagenous helix. Integrin α2I binding to the triple-helical GFPGER motif was slightly weaker than to GFOGER (O = hydroxyproline). This phenomenon was more prominent when α1 integrin was tested. Integrin α1β1 expressed on CHO cells and recombinant α1I domain showed remarkably slower binding velocity and weaker avidity to GFPGER when compared with GFOGER. Structural modeling revealed the critical interaction between Arg-218 in α1I and the hydroxyproline residue in the integrin-binding motif. The role of Arg-218 was further validated by testing a variant R218D α1I domain in solid-phase binding assays. Thus, our results show that the lack of proline hydroxylation in collagen can affect integrin binding by a direct mechanism and via structural destabilization of the triple helix.

Collagens are the most abundant extracellular matrix proteins in vertebrates. They have a characteristic triple-helical structure that is composed of three collagen ␣ chains. The primary structure of the collagen ␣ chain enables the formation of the collagenous triple-helix. Especially the location of glycine as every third residue and the presence of posttranslationally hydroxylated proline and lysine residues are critical factors (1). Lack of hydroxyproline dramatically decreases the melting temperature (T m ) of collagen and the T m of the triple-helical collagen molecule is directly proportional to the hydroxyproline content (2,3). Collagen prolyl 4-hydroxylases (C-P4Hs) 3 catalyze the formation of hydroxyprolines in collagens in the Y position prolines of the -X-Y-Gly-repeats through a mechanism that requires 2-oxoglutarate, Fe 2ϩ , molecular oxygen, and vitamin C (1). Scurvy, a disease caused by vitamin C deficiency, leads to broad connective tissue destruction and problems in the healing of skin wounds (4). In addition to vitamin C deficiency, C-P4H activity can be disrupted by mutations. Vertebrates have three C-P4H isoenzymes, of which C-P4H-I is expressed in all tissues studied, C-P4H-II is expressed particularly in chondrocytes, osteoblasts, and endothelial cells, and C-P4H-III has a wide expression pattern, but the expression is minor when compared with C-P4H-I and -II (1). C-P4H-I and C-P4H-II are likely to have distinct differences in their substrate-binding sites because the K m values of C-P4H-II for synthetic collagen-like (Pro-Pro-Gly) 10 and Gly-Val-Pro-Gly-Val peptides and a full-length procollagen chain are 3-6-fold higher than in the case of C-P4H-I (5,6). Furthermore, a marked difference exists between the two enzymes in the inhibition by a competitive inhibitor poly(L-proline), the K i of C-P4H-II being 200 -1000 times higher than that of C-P4H-I, depending on the length of poly(L-proline) (5,6). Studies with synthetic peptide substrates have shown that the rate of hydroxylation of a proline residue by C-P4H-I is affected by the nature of amino acids in the X position and also in other parts of the peptide substrate (3). Systematic comparison between C-P4H-I and C-P4H-II on their hydroxylation rates of such peptides is yet lacking, however. Studies with gene-modified mice have shown that homozygous inactivation of the P4ha1 gene, which encodes the catalytic ␣(I) subunit of the main isoenzyme, C-P4H-I, leads to an 80% decrease in total C-P4H activity and embryonic death at E10.5, whereas P4ha2 Ϫ/Ϫ mice lacking C-P4H-II have no obvious phenotypic abnormalities (7,8). Compound P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ mice with 65% reduction of total C-P4H activity in chondrocytes have moderate chondrodysplasia caused by defective growth plate development associated with soft extracellular matrix and reduced thermal stability of cartilage collagen (8). In humans, bi-allelic mutations of the P4HA1 gene have recently been shown to lead to a congenital connective tissue disorder with joint hypermobility, contractures, mild skeletal dysplasia, and high myopia in a compound heterozygous patient with a 50% reduction in total C-P4H activity in fibroblasts (9).
The members of the large integrin family mediate cell adhesion and are critical for survival, proliferation, differentiation, and migration, as well as tissue homeostasis. There are four integrin-type collagen receptors in mammals, namely ␣1␤1, ␣2␤1, ␣10␤1, and ␣11␤1 heterodimers (10). They bind to ligands in a metal-dependent manner via the inserted domain (I domain) that is structurally similar to von Willebrand factor type A domain (11). Consensus integrin-binding motif in collagens can be formulated as GXXGER and the best characterized high avidity site is triple-helical GFOGER (O ϭ 4-hydroxyproline) (12). Other similar motifs include GROGER (13), GLOGER (14), GMOGER (15), GLOGEN (14), GAOGER (13), GVOGEA (16), GFKGER (17), and GLQGER (17). Importantly, most of these sequences contain a hydroxyproline residue. Previously, it has been reported that the ␣2I domain can bind to GFPGER, but more weakly than GFOGER peptides (18). Furthermore, earlier papers show that the integrin ␣2␤1, but not ␣1␤1, can bind to unhydroxylated collagen I that has been produced in transgenic plants (19). However, both integrin ␣1I and integrin ␣2I domains can bind to GFPGER and GLPGER motifs when they have been inserted into Streptococcal collagen-like protein Scl2 (20). Thus, it is not clear whether prolyl hydroxylation plays a direct or indirect role in collagen-integrin interaction. Additionally, the structural basis of the putative difference between ␣1 and ␣2 integrin action is unknown.
To examine the role of collagen hydroxylation in the collagen-integrin interaction, we extracted skin collagen from P4ha1 ϩ/ϩ ;P4ha2 ϩ/Ϫ , P4ha1 ϩ/Ϫ ;P4ha2 ϩ/Ϫ , P4ha1 ϩ/ϩ ; P4ha2 Ϫ/Ϫ , and P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ mutant mice. The collagen preparations as well as triple-helical collagen mimicking peptides were tested with soluble recombinant integrin ␣I domains and CHO cells transfected to overexpress specific integrincollagen receptors. We report that hydroxyprolines generated by C-P4Hs can affect integrin binding by two distinct mechanisms: stabilization of the collagenous helix at body temperature and by strengthening the direct interaction between specific integrin ␣I domains.

Hydroxylation of proline residues in skin collagens is essential for the intact binding function of integrin ␣2I domain at physiological temperatures
To study if reduction in C-P4H activity affects the collagenintegrin interaction, we extracted bulk collagen from mouse skin using a acetic acid/pepsin digestion method followed by NaCl precipitation. The binding of soluble human recombinant integrin ␣2I domain to isolated collagen was measured by a solid-phase binding assay. We could not detect any significant differences between binding to skin collagen derived from the WT mice (n ϭ 5) and collagen from P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ mice (n ϭ 6), when the experiments were performed at room temperature (Fig. 1A). However, when the binding measurements were repeated at 37°C, the ␣2I domain showed ϳ30% weaker binding to collagen derived from the mice with deficient C-P4H activity (Experiment 1, p ϭ 0.004; Experiment 2, p ϭ 0.0004; Fig. 1B).
Next, we tested binding of the integrin ␣2I domain to collagenous triple-helical peptides, which harbor the well-described hydroxyproline containing an integrin-binding motif (GFOGER) or the same site with proline (GFPGER) at room temperature and at 37°C. The ␣2I domain binding was slightly weaker to the proline containing peptide, but the difference was the same in both temperatures ( Fig. 1, C and D). These results indicate that there are temperature-dependent and temperature-independent mechanisms by which the hydroxylation state of proline residues in collagen can affect the function of collagen receptor integrin.
To confirm our findings, we isolated skin collagen from WT mice (n ϭ 5) and mutant mice with different genetic combinations leading to C-P4H deficiency: P4ha1 ϩ/ϩ ;P4ha2 ϩ/Ϫ (n ϭ 4), P4ha1 ϩ/Ϫ ;P4ha2 ϩ/Ϫ (n ϭ 5), P4ha1 ϩ/ϩ ;P4ha2 Ϫ/Ϫ (n ϭ 7), and P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ (n ϭ 6). The LC-MS/MS shotgun proteomics analysis revealed that the skin collagen preparations almost exclusively contained collagen I and collagen III. On average, 72.3% of the detected peptide spectrum matches belonged to the Col␣1(I) or Col␣2(I) chains and 26.5% to the Col␣1(III) chain. The average fraction of spectra matched to the other 29 proteins was only 1.2% showing the high purity level of our collagen preparations. The hydroxylation frequency of prolines in -X-P-G-sequences in collagen was progressively and significantly diminished when the number of gene alleles coding for C-P4H-I and -II was reduced ( Fig. 2A). At room temperature binding of the soluble ␣2I domain did not correlate with the hydroxyproline content in collagen (Fig. 2B), whereas there was a strong correlation at 37°C (r ϭ 0.647, p ϭ 0.0005) (Fig. 2C). We used circular dichroism (CD) and trypsin/ chymotrypsin digestion analysis to study the thermal stability of the triple-helical conformation of collagen isolated from WT and P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ mice. CD analysis of two individual skin collagen samples from both genotypes gave melting temperature (T m ) values of 39.3 and 39.9°C for the WT collagen samples, whereas the T m of P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ collagen samples was markedly lower, 36.3 and 36.4°C. Similarly, at 37°C collagen derived from P4ha1 ϩ/Ϫ ;P4ha2 Ϫ/Ϫ mice was remarkably less trypsin/chymotrypsin resistant, which indicates opening of the triple-helical conformation, than collagen derived from WT animals (Fig. 2D). In conclusion, the temperature-dependent changes in integrin binding seem to be due to the altered hydroxylation state of the proline residues in collagen.

Proline hydroxylation and integrin binding to collagen Partial genetic inactivation of C-P4Hs in vivo does not affect the main integrin-binding motifs
To study further the hypohydroxylation of proline residues in collagen and its temperature-dependent consequences, we analyzed the LC-MS/MS data for the hydroxylation status of prolines in the known integrin ␣2␤1-binding motifs.
The proline residues in the GLOGER, GMOGER, GROGER, and GFOGER motifs of the Col␣1(I) chain and the GROGER motif of the Col␣2(I) chain were always (100%) hydroxylated both in the WT and mutant mice (Fig. 3A). We detected both proline and hydroxyproline containing forms of the GLOGER motif in the Col␣2(I) chain and the GAOGER motif in the Col␣1(III) chain (Fig. 3B). The hydroxyproline content of GAOGER decreased from 100% in WT mice to 85% in P4ha1 ϩ/Ϫ ; P4ha2 Ϫ/Ϫ mice. In the case of GLOGER of the Col␣2(I) chain, the hydroxyproline content decreased from 70 to 40%, correspondingly. The relative abundance of the detected peptides containing the first triplet of individual hydroxylated integrinbinding motifs (GRO, GAO, GLO, GFO, and GMO) was equal in collagen samples derived from the different mouse lines (Fig. 3C).
The changes detected in the hydroxyproline content of GLOGER of the Col␣2(I) chain and GAOGER of the Col␣1(III) chain are not sufficient to explain the affected binding of the integrin ␣2I domain. Giving also the fact that the ␣2I domain bound only slightly weaker to GFPGER than to GFOGER containing triple-helical peptides and that the difference was not temperature sensitive ( Fig. 1, C and D), the most likely explanation for the decreased ␣2I binding to hypohydroxylated skin collagen at 37°C is a partial or local destabilization of the triple helix.

Replacement of hydroxyproline with proline in the GFOGER integrin-binding motif leads to affected binding by integrins ␣1␤1 and ␣11␤1
To investigate the role of the direct interaction between integrin ␣I domain and the hydroxyproline residue in the integrin-binding motif, we analyzed, in addition to the ␣2I domain, the binding of the soluble recombinant integrin ␣1I domain to GFOGER and GFPGER containing peptides in solid-phase binding assays. Furthermore, we also used CHO cells transfected to express ␣1, ␣2, or ␣11 integrins. The adhesion of these cells to the collagenous peptides was measured by impedancebased xCELLigence technology that monitors cell-substrate interaction at real-time. WT CHO cells do not express collagen receptor integrins on their cell surface and the adhesion of the CHO transfectants to collagen is solely mediated by the overexpressed integrin ␣ subunit that has formed a heterodimer with CHO cell's own ␤1 subunit. The protein-binding assays were performed at room temperature and the cell-adhesion assays at 37°C. The CD-spectra of GFOGER and GFPGER indicated a small difference in melting temperature (48.9 versus 46.9°C) (Fig. 4A). However, at room temperature and 37°C it was possible to study the direct interaction between  integrin and hydroxyproline independently of triple-helix stability.

Proline hydroxylation and integrin binding to collagen
As reported earlier, the effect of prolyl hydroxylation on ␣1I binding was substantial (19). K d of ␣1I domain binding to GFOGER was about 125 Ϯ 4 nM, when it was 670 Ϯ 130 nM in binding to GFPGER (Fig. 4B). In accordance with the data in Fig. 1C, ␣2I binding was only slightly weaker to GFPGER (K d 30 Ϯ 2 nM) than to GFOGER (K d 23 Ϯ 1 nM) (Fig. 4C). Batch-to-batch variation in the recombinant integrin ␣11I domain prevented valid interpretations concerning the binding of this ␣I domain. In accordance with the ␣I domain data, CHO-␣1 integrin cells spread remarkably slower on GFPGER peptide when compared with GFOGER peptide (Fig. 4, D and F), whereas in the case of CHO-␣2 integrin cells only a minor difference was detected (Fig. 4, E and F). Interestingly, CHO-␣11 integrin cells strongly favored the adhesion to GFOGER peptide when compared with GFPGER (Fig. 4, G and H).
Further analyses of the real-time cell spreading revealed that CHO-␣1 integrin cells adhered to nonhydroxylated peptide with slower velocity than to GFOGER peptide ( Fig.  5A), whereas in the case of CHO-␣2 integrin cells no difference was detected (Fig. 5B). Similar analysis of the integrin ␣I domain binding supported these observations (Fig. 5, C  and D). Binding of the ␣2I domain to GFOGER and GFPGER similarly reached a steady binding state within the first 10 min (Fig. 5D), whereas ␣1I domain reached a steady binding state very slowly and there was a remarkable difference between GFOGER and GFPGER (Fig. 5C). Based on the observations that indicated the slow binding velocity, we hypothesized that hydroxylation of the proline residue in the GFOGER motif may be needed for proper activation of the integrin ␣1I domain. To test this possibility, we measured the binding velocity using a recombinant ␣1I harboring gain-offunction mutation E317A, which is known to pre-activate the integrin (21). The mutation did not abolish the difference between GFOGER and GFPGER (Fig. 5E). The observation suggested that the hydroxyproline-integrin interaction is essential during the initial interaction rather than at the later activation stages.

Proline hydroxylation and integrin binding to collagen Arginine 218 in ␣1I domain is critical for the interaction between integrin and hydroxyproline residue in the integrin-binding site
To analyze the role of the hydroxyprolines in the integrin ␣I-peptide interactions, a structural model for integrin ␣1I-GFOFER was created and compared with the X-ray structures of integrin ␣2I-GFOGER (PDB code 1DZI) (11) and the ␣1I-GLOGEN complex (PDB code 2M32) (22) (Fig. 6, A-C). Interestingly, the interactions between the hydroxyprolines and ␣I are similar in the ␣2I-GFOGER (Fig. 6A) and ␣1I-GLOGEN (Fig. 6C) complexes and these interactions are conserved also in the ␣1I-GFOGER model (Fig. 6B), which was created assuming that ␣1I interacts with the GFOGER peptide similarly as ␣2I. The conserved residues that form hydrogen bonds with the hydroxyprolines are Ile-156, Ans-154, and His-258 in the integrin ␣2I domain and corresponding Ile-156, Asn-153, and His-257 in the integrin ␣1I domain (Fig. 6, A-C). The hydroxyproline in the trailing strand of the collagen (Fig. 6A, orange) forms hydrogen bonds with the carbonyl oxygen of Ile-156 and Asn-154, whose side chain nitrogen in turn makes a hydrogen bond with the carbonyl of a middle strand hydroxyproline (Fig. 6A,  green). On the other side of the metal ion-dependent adhesion site (MIDAS), the carbonyl group of His-258 forms a hydrogen bond with the hydroxyl group of the middle strand hydroxyproline and the side chain of His-258 with the carbonyl of the middle strand Arg of GFOGER. All the integrin ␣2I domain residues (Asn-154, Tyr-157, and His-258) that form direct hydrogen bonds with the collagenous GFOGER peptide (11)

Proline hydroxylation and integrin binding to collagen
are very conserved in the integrin ␣1I domain. However, the salt bridge between the middle strand Arg of the peptide and Asp-219 in ␣2I domain (Fig. 6D) is not conserved, as the corresponding residue in the ␣1I domain is a positively charged Arg-218, which cannot form a similar interaction. Based on the ␣1I-GFOGER model, Arg-218 presumably stacks with the middle strand Arg of GFOGER, which forms a hydrogen bond with Gln-219 (Fig. 6E). The position of Arg-218 in the ␣1I domain is stabilized by a salt bridge with Glu-188, whereas Asp-219 in the ␣2I domain is positioned by a hydrogen bond with Asn-189 (Fig. 6, D and E).
As the differing interactions did not directly involve collagen hydroxyprolines and, thus, did not explain why the integrin ␣1I domain binds more weakly to GFPGER than GFOGER, whereas the ␣2I domain binds them equally well, we next analyzed the water-mediated hydrogen bonds formed by the hydroxyprolines within GFOGER (Fig. 6, D and E, waters are shown as red spheres and hydrogen bonds with red dashes). Based on the analysis, all these bonds are formed by the hydroxyprolines in the trailing strand and mainly stabilize its structure, but one of the waters connects the trailing and middle strands (Fig. 6, D and E, circles in red). Intriguingly, the same hydroxyproline makes conserved interactions with the ␣I domain (Fig. 6, A-C). As the GFPGER peptide lacks the hydroxyprolines involved in the stabilizing water-mediated hydrogen bond network, its triple-helical structure is presumably more flexible compared with GFOGER. The lack of the water-mediated bonds is likely to weaken the positioning of the Arg in the middle strand. Thus, Arg might not be optimally oriented for the hydrogen bonding and stacking interactions in the ␣1I-GFPGER complex (Fig. 6C), which require more precise positioning than the stronger Arg-Asp-219 salt bridge within the ␣2I-GFPGER complex (Fig. 6A). Additionally, the differences in the residues in the vicinity of Asp-219 of the ␣2I domain (Leu-220; Asn-189) and Arg-218 of the ␣1I domain (Gln-219; Glu-188) and the formed interaction networks likely contribute to ␣I-peptide interactions (Fig. 6, D and E, circles in black).
The structural model was validated by solid-phase binding assays using variant domain ␣1I R218D. Domain ␣1I R218D bound very weakly to GFOGER peptide confirming the central role of Arg-218 in the interaction between integrin and hydroxyproline (Fig. 6F). In addition, ␣1I R218D binding to GFOGER and GFPGER was equally poor supporting the idea (Fig. 6F). In accordance with own previous data (23), binding of ␣1I R218D domain to collagen I was not affected, but actually slightly improved (Fig. 6G). Here, this observation serves as an important control showing that the ␣1I R218D domain is folded properly and is fully functional. In general, the result indicates that in collagen I there must be other high aviditybinding motifs than GFOGER for ␣1␤1 integrin.

Discussion
Mammalian cells can directly recognize collagens by using a specific subgroup of integrins that, in complex with the ␤1 subunit, contain one of the four collagen-binding ␣ subunits. These ␣I domain-harboring collagen receptor integrins can be found in vertebrates and lampreys but not in other multicellular animals (24). The co-evolution of collagen receptor integrins and collagens has led to the development of high affinity integrin recognition motifs in collagens. These sites, including the prototype motif GFOGER, typically contain one hydroxyproline residue. The biological role of direct cell adhesion to collagen is unclear. In some in vitro models integrin ␣2␤1 and ␣2I domains can bind to collagen I fibrils (25), whereas other lines of experimentation suggest that cartilage type, collagen II containing fibrils may not be direct targets of the collagen receptors or their ␣I domains (26). Some researchers have suggested that the collagen receptors participate in the organization of the collagen fibrils (27,28), but so far the described phenotypes of the ␣1, ␣2, and ␣11 integrin-deficient mice have not supported this idea. In general the phenotypes of collagen receptor knockouts are mild and show relatively small defects in immunoresponse, platelet function, wound healing, angiogenesis, bone fracture healing, or tooth development (29 -34). Phenotypes of double knockouts have not revealed significant overlap or compensation in the functions of these receptors (35,36). Studies on integrin ␣10-deficient mice have, however, revealed a reduced density of the collagen fibrillar network matrix in the growth plates (37). In general, the alterations in cartilage of the ␣10 knockout mice are similar as the defects in the P4ha1 ϩ/Ϫ ; P4ha2 Ϫ/Ϫ mice (8), including abnormal arrangement of growth plate chondrocytes and an elevated number of apoptotic chondrocytes (8,37).

Proline hydroxylation and integrin binding to collagen
Here, we made an interesting observation that after concomitant inactivation of both alleles of P4ha2 and one allele of the P4ha1 gene all the major integrin-binding motifs were completely hydroxylated, despite the fact that hydroxylation of proline residues in -X-P-G-sequences dropped significantly (from 53 to 36%). There are two possible explanations for the result. It can be hypothesized that the site specificity of C-P4H-I has evolved to favor the integrin-recognition motifs over other sites in collagen or that only the tropocollagens that have intact integrin-binding sites end up to skin collagen fibrils. We cannot exclude either one of the two hypotheses, but they both stress the importance of direct integrin-collagen interaction.
Atomic structures of integrin ␣1I-GLOGEN and ␣2I-GFOGER complexes suggest that there is a direct interaction between the integrin and hydroxyproline residue (11,22). Previous studies have, however, indicated that recognition of the GFOGER motif by the major epithelial cell collagen receptor, ␣2␤1 integrin, is not dependent on the hydroxylated proline residue (19). Our observations are in accordance with this result. Still, in vivo hypohydroxylation of proline residues results in collagen that cannot effectively interact with the ␣2I domain, when the test is performed in a physiological temperature. Thus, the structural stability of the collagen triple-helix may be more important regulator of integrin binding than the proline hydroxylation in the individual recognition motifs.
In agreement with a previous paper (19) we report here that the high avidity binding of ␣1I domain, and similarly ␣1␤1 integrin, requires the presence of hydroxyproline as part of the GFOGER motif. To study further the mechanism of this interaction we used molecular modeling and binding assay utilizing variant recombinant ␣1I domains. Based on the structural analysis, we were able to identify differences in the ␣I domainpeptide interactions that may result in the observed differences in the binding assay, despite the interactions that hydroxyprolines form with the ␣I domains being totally conserved in the X-ray structures of ␣2I-GFOGER (PDB code 1DZI) (11) and the ␣1I-GLOGEN complex (PDB code 2M32) (22), as well as in the created structural model of the ␣1I-GFOFER complex. Intriguingly, the water-mediated hydrogen bonds formed by one of the hydroxyprolines seem to be the key element for the different binding characteristics of ␣1I and ␣2I domains. Due to the lack of these bonds and the increased flexibility of the GFPGER triple-helix, the resulting decreased binding of ␣1I compared with ␣2I was predicted to originate from the Asp-219 ␣2I /Arg-218 ␣1I and Leu-220 ␣2I /Gln-219 ␣1I differences, which allows the aspartate in the ␣2I domain to form a salt bridge with the Arg of GFPGER, whereas the interactions formed by the ␣1I domain are weaker hydrogen bond and stacking interactions. In general, the interactions formed by these ␣I domain residues to the Arg and their positioning by the interacting residues (Asn-189 ␣2I /Glu-188 ␣1I ) seems to be crucial for the collagen peptide binding properties.
The experiments with the variant ␣1I R218D domain revealed the importance of Arg-218 in the interaction between the ␣1I domain and GFOGER. Importantly, here and also in our previous paper (23) we have shown that the same mutation actually improves the interaction between the ␣1I domain and collagen I. Thus, ␣1␤1 integrin must also have other binding sites in collagen I. Collagen I is not a homotrimer, but composed of two ␣1(I) chains and one ␣2(I) chain. The presence of multiple ␣ chains has complicated the search for integrin-binding sites in collagen I and also in other heterotrimeric collagens (23), whereas integrin recognition motifs have systematically been searched in homotrimeric collagens II and III, only. Indeed, there is evidence that in collagen IV all three ␣ chains may participate in the formation of an ␣1␤1 integrin-binding site, in which the critical residues include one arginine and two aspartic acid residues (38 -40). The presence of similar motifs in collagen I would also explain our results.
To conclude, our results show that intact collagen receptor function requires proper hydroxylation of collagen, which can partially explain the phenotypes of C-P4H mutant mice. We also provide two distinct molecular mechanisms, an indirect and a direct one, by which proline hydroxylation modulates integrin function.

Collagen extraction and precipitation from the mouse skin
Fibrillar collagen was purified as described earlier (41), with slight modifications. Briefly, to extract collagen, skin punches were minced and incubated in 0.5 M acetic acid overnight at 4°C. The tissue was minced further by vigorous shaking with stainless steel beads using TissueLyser LT (Qiagen) for 3 ϫ 30 s, 50 Hz. After this, a pepsin-soluble collagen fraction was obtained by adding 1 mg/ml of pepsin for 1 week with constant mixing at 4°C and collecting the supernatant after centrifugation at 13,000 ϫ g for 30 min. Type I collagen was precipitated by adding 5 M NaCl in 0.05 M Tris-HCl, pH 7.4, to a final concentration of 0.7 M. Collagen containing pellet was obtained after overnight incubation at 4°C followed by 21,000 ϫ g centrifugation for 20 min. The pellet was washed by resuspension in 4 M NaCl and overnight incubation at 4°C, followed by 21,000 ϫ g centrifugation for 20 min. The wash was repeated once and the final collagen pellet was suspended in 0.1 M acetic acid.

Proline hydroxylation and integrin binding to collagen Thermal stability analysis of collagen
The native triple-helical conformation of a collagen molecule is resistant to most proteinases and becomes susceptible for digestion only after unfolding of the triple helix, for example, by thermal denaturation, and proteolytic enzymes are used as probes for quantitative assay of fully aligned triple-helical collagen molecules (52). The thermal stability of the precipitated skin collagen was analyzed by trypsin/chymotrypsin digestion performed at various temperatures (52). Collagen concentration was measured with Direct Detect IR spectrometer (Millipore Corporation) and 25-l aliquots were neutralized and preincubated at 35°C for 5 min. 2.5 l of trypsin/chymotrypsin solution (0.1 mg/ml of trypsin (Sigma) and 0.25 mg/ml of chymotrypsin (Sigma) in 0.2 M NaCl, 50 mM Tris-HCl buffer, pH 7.4) was added to the samples, which were then incubated at 36, 37, 38, 39, 40,

CD
Circular dichroism (CD) spectroscopy was performed using Chirasca CD Spectrometer (Applied Photophysics, Leatherhead, UK). The data were collected between 190 and 260 nm at 22°C using a 0.1-cm path length quartz cuvette triplicate samples with 0.1 mg/ml of protein concentration. CD measurements were acquired every 1 nm with 1 s as an integration time and repeated three times with baseline correction. The data were analyzed with Pro-Data Viewer (Applied Photophysics). Thermal unfolding was recorded between 190 and 260 nm with a 2°C step size at 1°C/min ramp rate with Ϯ 0.2°C tolerance. The melting temperature was analyzed with Global3 (Applied Photophysics).

Mass spectrometry
The collagen samples were vacuum-dried and dissolved in 6 M urea in 25 mM ammonium bicarbonate. The cysteines were reduced by 10 mM DTT at 37°C for 1 h and alkylated by 40 mM iodoacetamide at room temperature for 30 min. The solution was diluted to 1:10 with 25 mM ammonium bicarbonate, and the proteins were digested with 0.5 g of trypsin (Promega, Madison, WI) per 10 g of protein at 37°C overnight. The peptides were desalted by StageTips (42), vacuum dried, dissolved in 1% formic acid, and loaded on a nanoflow HPLC system (Easy-nLC1000, Thermo Fisher Scientific) coupled to the Q Exactive mass spectrometer (Thermo Fisher Scientific) equipped with a nano-electrospray ionization source. The peptides were first loaded on a trapping column and subsequently separated in-line on a 15-cm C18 column (75 m ϫ 15 cm, Magic 5 m 200 Å C18, Michrom BioResources Inc., Sacramento, CA). The mobile phase consisted of water/acetonitrile (98:2 (v/v)) with 0.2% formic acid (solvent A) and acetonitrile/ water (95:5 (v/v)) with 0.2% formic acid (solvent B). The pep-tides were separated with a 10-min gradient from 5 to 35% of solvent B. Before the end of the run, the percentage of solvent B was raised to 100% in 2 min and kept there for 8 min. Data-dependent acquisition was enabled and higher energy collisional dissociation of the top 5 ions from the survey scan over the mass-to-charge (m/z) range 300 -2000 with dynamic exclusion time of 10 s was performed in each cycle. Two repeated runs per sample were performed. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (43) partner repository with the dataset identifier PXD008802.

Data analysis
Tandem mass spectra were searched against mouse Swissprot sequences (release 2014_08) with Proteome Discoverer software, version 1.4 (Thermo Fischer Scientific) using the Mascot search engine (Matrix Science, version 2.4) allowing for 5 ppm precursor mass tolerance and 0.02 Da fragment mass tolerance. Carbamidomethyl (C) as a fixed modification, and oxidation (M, K, P) as dynamic modifications were included. Maximum of two missed cleavages were allowed. Decoy database search using reversed mouse SwissProt sequences was used to assess the false discovery rate. Only the peptides with false discovery rate Ͻ0.05 and determined as "rank 1" by Proteome Discoverer software were accepted for further analysis. All statistical tests were done by IBM SPSS Statistics, version 23.

Real-time cell adhesion
Plates (96-well E-Plate 96; ACEA Biosciences Inc.) were coated with synthesized triple-helical collagen mimetic peptides GFOGER or GFPGER (see above, 10 g/ml in PBS) or BSA (1 mg/ml in PBS) overnight at ϩ4°C. The nonspecific binding sites were blocked with BSA (1 mg/ml, 37°C, 1 h). 30,000 CHO cells/well were seeded in serum-free ␣-minimum Eagle's medium. Cell adhesion and spreading was followed for at least 24 h with the xCELLigence RTCA (ACEA Biosciences Inc.). The velocity of cell spreading was calculated by the following equation: (C2 Ϫ C1)/(t2 Ϫ t1), where C is a cell index value at a specific time point and t is time.

3D modeling and structural analysis
A 3D model for the integrin ␣1I domain (UniProt Knowledgebase (UniProtKB) P56199) in complex with the GFOGER peptide was constructed using the ␣2I-GFOGER peptide complex as a structural template (PDB code 1DZI) (11), as the binding studies suggest that the binding mode of the ␣1I domain is similar to that of ␣2I (18). The model with the lowest value of MODELLER objective function (49) of the 10 created models was chosen for structural analysis and visualization. The intermolecular interactions were studied by the "Protein interfaces, surfaces and assemblies" service (PISA) at the European Bioinformatics Institute (50). The figures and structural analysis were done with PyMOL (51).