Lateral Growth Limitation of Corneal Fibrils and Their Lamellar Stacking Depend on Covalent Collagen Cross-linking by Transglutaminase-2 and Lysyl Oxidases, Respectively*

Background: Mechanisms of growth limitation and lamellar stacking of collagen fibrils in cornea remain elusive. Results: Covalent collagen cross-links are formed by catalysis involving both lysyl oxidases and tissue transglutaminase-2. Conclusion: Aldehyde-derived and isopeptide cross-linking of collagen determine lamellar stacking and lateral fibril growth, respectively. Significance: Two types of covalent collagen cross-linking are indispensable for correct corneal morphogenesis. Corneal stroma contains an extracellular matrix of orthogonal lamellae formed by parallel and equidistant fibrils with a homogeneous diameter of ∼35 nm. This is indispensable for corneal transparency and mechanical functions. However, the mechanisms controlling corneal fibrillogenesis are incompletely understood and the conditions required for lamellar stacking are essentially unknown. Under appropriate conditions, chick embryo corneal fibroblasts can produce an extracellular matrix in vitro resembling primary corneal stroma during embryonic development. Among other requirements, cross-links between fibrillar collagens, introduced by tissue transglutaminase-2, are necessary for the self-assembly of uniform, small diameter fibrils but not their lamellar stacking. By contrast, the subsequent lamellar organization into plywood-like stacks depends on lysyl aldehyde-derived cross-links introduced by lysyl oxidase activity, which, in turn, only weakly influences fibril diameters. These cross-links are introduced at early stages of fibrillogenesis. The enzymes are likely to be important for a correct matrix deposition also during repair of the cornea.

The suprastructure of the corneal stroma is well adapted to its specific functions as a robust and transparent tissue. Vulnerable structures within the eye are protected against environmental insults. At the same time, the cornea is an important optical element in that it diffracts light but does not impede its passage into the interior of the eye. These properties result from an exquisite fibrillar organization in the extracellular matrix (1)(2)(3). The human tissue, similarly to that of other vertebrate species, contains a stack of lamellae formed by parallel and evenly spaced collagen fibrils with small and uniform diameters. Within the lamellae, individual fibrils are packed into almost hexagonal arrays having a thickness of ϳ1 m, corresponding to ϳ20 collagen fibrils and their intervening spaces. The orientation of the fibrils in neighboring lamellae is almost orthogonal.
During development of the chick embryo, the corneal stroma arises from epithelial cells depositing an extracellular matrix called the primary stroma, which contains relatively few fibrils that already are arranged in patterns reminiscent of future lamellae. Neural crest-derived cells that thereafter differentiate into fibroblast-like keratocytes invade this primordial tissue model. These cells generate the secondary, mature stroma by apposition of further fibrils into the pattern already established in the primary stroma and by incorporation of hyaluronan and proteoglycans, causing swelling of the tissue by colloid osmotic water binding (4 -6).
The requirements for the formation of the uniformly thin corneal collagen fibrils have been investigated extensively by a variety of approaches, including reconstitution of fibrils from soluble molecular components in vitro matrix formation by keratocytes (7) and corneal epithelial cells (8) in culture and by histological and electron microscope analysis of corneas from transgenic and knock-out mice. The aggregated evidence suggests that several molecular species are indispensably involved in the appropriate fibrillogenesis in the cornea. Depending on the developmental stages, the fibrils not only contain several collagen types, including fibrillar types I, III, V, and XXIV (9) as well as fibril-associated collagens XII and XIV. In the primary stroma of the developing avian cornea, collagens II and IX are also found. In addition, several small proteoglycans with core proteins containing leucine-rich repeats (SLRPs), 2 particularly lumican, fibromodulin, decorin, and biglycan, keratocan, and mimecan are also essential components of corneal fibrils (for review, see Ref. 10). For example, lumican-deficient mice have structurally abnormal corneal fibrils and, as a result, opaque corneas (11)(12)(13). The same is true for decorin-or biglycan-null mice and, to a greater extent, mice with a compound deficiency of decorin and biglycan, suggesting redundant functions of the SLRPs in corneal fibrillogenesis (14,15).
Whereas the molecular requirements for the generation of individual, small diameter fibrils in cornea have begun to emerge, the mechanisms controlling formation of lamellae or their stacking are an essentially uncharted territory. Keratocytes in culture can deposit a matrix, in which primordia of corneal lamellae are established (6,7), and which is reminiscent of the primary or early secondary corneal stroma (2). However, the role of the macromolecular components remains to be identified. The regular arrangement of glycosaminoglycans between corneal fibrils suggested indirectly that anti-parallel association of glycosaminoglycans results in an even spacing (16). This would indicate that the essential information for lamellar stack formation already is available in the appropriate mixture of macromolecular constituents of the corneal stroma. By implication, the role of the cells consists of the biosynthesis of the macromolecules required but not necessarily in the building of the corneal architecture. However, keratocytes may well accelerate the deposition of the lamellae by forming compartments conducive to lamellar stacking. Here, we have studied further the role of keratocytes from 17-day-old chick embryos in forming lamellar primordia resembling those of primary corneal stroma. Surprisingly, we found that early formation of collagen cross-links is essential for both the limitation of fibril diameters and their organization into lamellae.

EXPERIMENTAL PROCEDURES
Three-dimensional Cell Culture-Fibroblasts were isolated from corneas of 17-day-old chicken embryos. After dissection of central regions of the corneas, tissues were washed in Krebs buffer. Subsequently, epithelia and endothelia were removed from tissue fragments by digestion in Krebs buffer containing 0.25% (w/v) trypsin and 2 mM EDTA for 15 min. After washing, tissues were minced and matrix-free fibroblasts were obtained by incubation with 1 mg/ml collagenase B from Clostridium histolyticum (0.191 units/mg, lyophilized, Roche Applied Science) in DMEM supplemented with 1% (w/v), each, of penicillin and streptomycin. Therefore, the cells were initially matrixfree. Cells were kept overnight at 37°C and 5% CO 2 in a humidified atmosphere incubator, washed three times with DMEM (10% FCS), and resuspended in DMEM (10% FCS), and distributed in 24-well plates (Nunc, Roskilde, Denmark) at 2 ϫ 10 5 cells per well. Cells were cultured overnight and then supplemented with 0.14 mM L(ϩ)-ascorbic acid, 1 mM sodium pyruvate, and 1 mM L-cysteine. In some experiments, cross-link inhibitors were added to give final concentrations of 0.2 mM ␤-aminopropionitrile (␤APN) and/or 5 M transglutaminase (TG) inhibitor (Boc-DON-Gln-Ile-Val-OMe, ZEDIRA). After 14 days of cultivation, the cells formed three-dimensional cellmatrix constructs that were used for further experiments.
Collagen Purification-17-Day-old embryonic chicken corneas were washed three times in PBS and were extracted twice in 15 volumes of 0.5 M acetic acid by stirring overnight at 4°C. The tissue fragments were removed by centrifugation, and the combined extracts were supplemented slowly with solid NaCl to give final concentrations of 25% (w/v), precipitating a mix-ture of corneal collagens and non-collagenous macromolecules. The precipitates were recovered by centrifugation, dissolved in 50 mM Tris-HCl containing 2 M urea and 200 mM NaCl, pH 7.4, dialyzed against the same buffer, and passed over a DEAE-cellulose column (3.5 ϫ 21 cm, 200 ml, DE52; Whatman, Ltd.) equilibrated in the same buffer. A mixture of corneal collagens, without proteoglycans and other acidic components, were recovered from the breakthrough fraction. Both corneal collagens obtained by DEAE chromatography and crude corneal protein (before chromatography) mixtures were then dialyzed extensively against 100 mM Tris-HCl, pH 7.4, containing 400 mM NaCl (storage buffer). Proteins were precipitated by adding solid sodium chloride to a final concentration of 4.5 M. After centrifugation, pellets were redissolved in storage buffer and clarified by centrifugation. These materials were used for in vitro fibrillogenesis as described previously (17).
In Vitro Fibrillogenesis-The crude mixture of corneal collagens in storage buffer was degassed, and in vitro fibrillogenesis was carried out in a microcuvette (Multicell, light path, 1 cm, Beckman, Palo Alto, CA). Fibril formation was initiated by diluting the collagen mixtures with an equal volume of distilled water. In some experiments, tissue transglutaminase, CaCl 2 , or factor XIII and thrombin were added to the mixtures directly after dilution. The reconstitution products were examined by transmission electron microscopy and immunoelectron microscopy with antibodies to collagen I essentially as described (18).
Isolation of Fragments of Collagen Fibrils from Cornea-Human corneas were obtained from voluntary donors at autopsy and in agreement with local ethics regulations concerning recovery of human tissues. The corneas were stored in the tissue bank of the clinic of ophthalmology of the University Hospital of Münster. Pieces of corneas rejected for transplantation were homogenized in PBS, and fragments of collagen fibrils were isolated as described (19,20). Briefly, homogenates were subjected to centrifugation at low speed, and fragments of tissue suprastructures in the low speed supernatants were directly analyzed by transmission electron microscopy at appropriate dilutions.
Transmission Electron Microscopy and Immunogold Electron Microscopy-Multilayer corneal fibroblast-matrix constructs were cultured for 14 days, gently detached from the culture dishes, and subjected to fixation at 4°C overnight in 100 mM cacodylate buffer, containing 2% (v/v) formaldehyde and 0.25% (v/v) glutaraldehyde, pH 7.4. After washing in PBS, specimens were dehydrated in an ascending ethanol series from 30 to 70%, incubated overnight at 4°C in LR White (Agar Scientific, Stansted, UK) and 70% ethanol (2:1) followed by several steps of pure embedding medium. Finally, LR White was polymerized under UV light according to the manufacturer's instructions. Ultrathin sections were cut on an Ultramicrotome at random orientation of the specimen and collected on nickel grids coated with Formvar/carbon. Grids were floated on drops of 100 mM glycine in PBS for 30 s. Alternatively, 20 l of reconstitution products from in vitro fibrillogenesis or fibrils isolated from homogenates of human cornea were absorbed for 10 min to Formvar/carbon-coated grids, washed with PBS, and treated for 30 min with 2% (w/v) skim milk in PBS. The ultrathin sec-tions or absorbed materials were allowed to react for 2 h with a polyclonal antibody (AB752P, Chemicon) against murine collagen I or a monoclonal antibody (MAB3392, Chemicon) against human collagen III. After washing several times with PBS, the grids were incubated with PBS containing goat antibodies directed to rabbit immunoglobulins conjugated to 18-nm colloidal gold particles (Jackson ImmunoResearch Laboratories) for 2 h. Control experiments were done with the first antibody omitted. Finally, the grids were washed several times with distilled water and stained with 2% (w/v) uranyl acetate for 10 min. Electron micrographs were taken at 60 kV using a transmission electron microscope (Philips EM410). Images were recorded on image plates, which were subsequently scanned and digitized on a Micron Imaging Plate Scanner (Ditabis, Pforzheim, Germany) as TIFF files used without further processing.
Collagen Preparation and Fluorography-Matrix-free corneal fibroblasts were cultured for 24 h or 6 days with 1 Ci/ml 14 C-labeled proline (Hartmann analytic, Braunschweig, Germany), L(ϩ)-ascorbic acid, sodium pyruvate, L-cysteine, and, in some experiments, with cross-link inhibitors. Media were removed, and cell layers harvested in 1.5 ml of acetic acid containing 1 mg/ml of pepsin. Proteins were digested under stirring by rotation for 2 h at room temperature. After salt precipitation, cell pellets were recovered by centrifugation and resuspended in storage buffer and then precipitated by ethanol. After another step of resuspension in distilled water and ethanol precipitation, the cell pellets were resuspended in 50 l of 5-fold concentrated sample buffer without reducing agents and heated to 95°C for 5 min. 2-l aliquots of the labeled proteins or 2 l of molecular marker solution (CF626, Amersham Biosciences) were diluted in 2 ml of Lumasafe Plus (Lumac LSC, Groningen, Netherlands) and counted in a scintillation counter (Beckman). Samples were loaded with the same volume (luminescence counts varied from ϳ8000 to 25000) on 4.5-8% polyacrylamide gradient gels. Fluorograms were prepared as described by Bonner and Laskey (21), and bands were quantified according to Laskey and Mills (22).
Analysis of Fibril Diameters-Observed fibril diameter distributions were modeled as the sum of one or more Gaussian profiles to entire experimental data sets corresponding to particular culture conditions. The custom program Chi-Square Calculator (Fourmilab, Switzerland, available on the internet) was used that successively fits one, then two, etc., Gaussians to the integral of the data using least-squares. The minimum number of Gaussians that provided an acceptable Chi-square at the 5% level of significance (i.e. Q Ͼ ϭ 0.05) is reported. Special care was taken to avoid false minima in the least-squares minimization procedure by repeating the fit many times with randomly perturbed values of the initial parameters until a global minimum was reached. Mean fibril diameter (nm), S.D. (), and fraction of total fibrils in the specimen (f), as well as Q, are reported. Subscripts denote the individual components.

RESULTS
Fibril diameters were analyzed in images obtained from ultrathin sections of corneas from 17-day-old chick embryos (Fig. 1A). A best fit to the experimental data (n ϭ 795) was obtained by two Gaussian distributions with maxima at 35 nm (86% of the fibrils) and 52 nm (14% of the fibrils; ϭ 6.0 and 8.0 nm, respectively; see Fig. 3A). The values of the larger population reasonably agree with reported diameter values determined from transmission electron micrographs for typical corneal collagen fibrils in many adult vertebrate species. Diameter values range from 25 to 40 nm (23,24), depending on the exact tissue embedding conditions and, more specifically, on the extent of dehydration. X-ray diffraction of fully hydrated corneas yielded diameter values close to 40 nm (25).
Fibrils extracted from homogenized human cornea comprised a very rare population of strongly cross-striated fibrils with variable diameters (40 -80 nm), in addition to the abundant and characteristic fibrils with small diameters described above (Fig. 1B). The large fibrils were not derived from scleral tissue because precautions were taken to use only central portions of the corneas. Interestingly, such fibrils were labeled with monoclonal antibodies to collagen III by immunogold electron microscopy (Fig. 1B), whereas the major population of small diameter fibrils was systematically devoid of such labeling even after demasking of the epitopes by treatment of the fibrils with acetic acid (26). The specificity of the commercial collagen III antibodies was suitable for this investigation, as revealed by immunoblotting on collagens I (negative) and III (positive) purified from human skin (data not shown). Keene and colleagues (27) have described previously a small population collagen III-positive fibrils in human corneas, but a preferential labeling of strongly banded, large diameter fibrils was not apparent. However, Bruns et al. (28) investigated tissue sections of embryonic and adult chick cornea and have observed large banded fibrils at very low abundance and in conjunction with microfibrillar material. Although this was not studied at the time, it is tempting to speculate that the large banded collagen fibrils in fact were similar to the collagen III containing fibrils described here. Such fibrils, and their aligned microfibrillar material, may serve to prevent mechanical stress-induced delamination of the lamellar stacks in cornea.
Next, we prepared suspensions of keratocytes obtained from corneas of 17 day-old chick embryos by enzymatic digestion of the tissue with collagenase. The cells were plated at high density and were cultured for 14 days in medium containing fetal calf serum. Ultrathin sections of the cultures containing the newly synthesized matrix were analyzed by transmission electron microscopy after negative staining. As shown in Fig. 2A, the organization of the fibrils was highly ordered and resembled that of the primary corneal stroma during development. Bundles of parallel, thin fibrils were deposited in orthogonal arrays. The fibrils contained collagen I, the quantitatively major constituent of corneal collagen fibrils, as revealed by immunogold electron microscopy (data not shown). Fibril diameters (n ϭ 758) had a uniform Gaussian distribution centered at 45 nm ( ϭ 12 nm, Fig. 3B). Our observations are consistent with those reported previously on cultures of keratocytes emigrating from fragments of human corneal stroma (7). However, the experimental protocol employed here excludes the possibility that authentic tissue fragments influence or control the organization of fibrils deposited de novo. In our experiments, the cells were devoid of extracellular matrix at the time of initiation of the cultures. However, the cells still generated a de novo matrix with corneal characteristics.
The culture media were then supplemented with ␤APN, a natural drug inhibiting the activity of all isoforms of lysyl oxidase, the enzymes catalyzing post-translational oxidative deamination of (hydroxy)lysyl residues to yield (hydroxy)lysyl aldehydes. In the absence of ␤APN, the aldehydes spontaneously react and mature into several bivalent, trivalent, and oligomeric molecular species that covalently cross-link corneal fibrillar components, including collagens (29). Strikingly, the culture regimen with ␤APN completely abrogated the formation of lamellae. Only residual assemblies sometimes remained of parallel and approximately equidistant fibrils alternating with non-fibrillar material as typically seen within corneal lamellae. By contrast, abolition of lysyl oxidase activity with ␤APN had no major influence on the diameter of most fibrils (Fig. 2B). However, as revealed by the occurrence of clusters of gold particles in immunogold electron microscopy, the fibrils were identified as collagen I-containing when sectioned in oblique or longitudinal orientation (Fig. 2B, inset, arrows). Sections not exposed to antibodies to collagen I, were essentially devoid of gold particles and clusters were never observed (data not shown). 62% of the fibrils had diameters with a distribution centered at 44 nm ( ϭ 10 nm), which closely corresponds to the values observed in control cultures without ␤APN. However, several minor populations of thicker fibrils also occurred, which resulted in a tailing of the distribution toward higher values (Fig. 3C).
We observed the formation of divalent and tetravalent but not trivalent cross-linked species labeled with 14 C-labeled proline during early phases of our cultures. A fluorogram of a SDS-PAGE gel of collagen metabolically labeled during the first 24 h is shown in Fig. 4A, lane 1. By contrast, trivalent species appeared at the expense of dimer and tetramer species on a fluorogram of newly synthesized collagens labeled for 6 days (Fig. 4B, lane 1).
As shown in Fig. 4, A and B, lane 3, addition of ␤APN to the culture medium does not entirely suppress the formation of cross-linked collagenous polypeptides at any time during culture. However, it is well known that ␤APN, at the concentrations employed, inactivates all lysyl oxidase isoenzymes and, hence, eliminates all cross-links of the aldehyde variety (30).   between appropriately spaced side chains of glutamine and lysyl residues. Tissue transglutaminase (transglutaminase-2, TG-2) is present in cornea (31,32) and is up-regulated after injury, particularly in the corneal epithelium but also in the stroma (32) and, possibly, also during corneal development because similarities are frequently observed in wound healing and development (10). Therefore, we treated our cultures with a synthetic inhibitor specifically and irreversibly inactivating TG-2. The lamellar organization of the collagen fibrils essentially remained normal under these conditions (Fig. 2C). However, the fibrils were much thicker and had an obvious D-periodic banding pattern (Fig. 2C, inset). Diameters fell into two Gaussian distributions centered at 107 nm ( ϭ 13 nm, 87% of the fibrils) and 134 nm ( ϭ 8 nm, 13%; Fig. 3D). Therefore, the average fibril width was more than doubled in comparison with control cultures without TG-2 inhibitor, and the diameter distributions overlapped only slightly (compare Fig. 3, B and C, with 3D). Metabolically labeled collagens were then isolated from the cultures with the TG2 inhibitor and were analyzed by SDS-PAGE followed by fluorography. As shown in Fig. 5B, lane 2, the formation of trimeric species at late time points was almost unaffected by the inhibitor, whereas tetramers were reduced in quantity to about one-half. When both inhibitors were applied, fibrillogenesis was severely compromised. In general, fibrils were scarce and both the diameter control and the lamellar organization were lost (Fig. 2, D and E). The few fibrils, which still were formed, contained collagen I as revealed by immunogold labeling (highlighted by arrows in Fig. 2E). They also had irregular crosssections that were similar to those of tendon or skin fibrils of decorin-deficient mice (33) because smaller fibrils appeared to fuse into larger and less organized fibril-like suprastructures. This event does not take place in corneal stroma under normal conditions because it would strongly impair tissue transparency. In addition, it prevented us from determining meaningful diameters of the thicker fibrils. However, there was an additional population of thin fibrils with a mean diameter of 25 nm ( ϭ 4 nm, Fig. 3E), which represented ϳ14% of the fibrils. Such fibrils were not found in our cultures with or without addition of the single inhibitors.
Taken together, the results described above are consistent with the notion that the accurate deposition of collagen fibrils in cornea is specified, at least in part, by appropriate mixtures of macromolecules and their post-translational modifications, including intermolecular cross-linking by isopeptide bonds  shown. Note that no diameter control is apparent in the absence of transglutaminase activity (A). Inset, a strong D-periodic banding pattern of collagen fibrils is visible at high magnification. Fibril diameters are strongly reduced by transglutaminase activity, and non-collagenous material is diffusely associated with and cross-linked to the fibrils (B). Fibril diameters after reconstitution in vitro from corneal collagens were purified by chromatography on DEAE-cellulose in the presence or absence of activated recombinant tissue transglutaminase-2 (C). Note that fibril diameters are reduced in the presence of active enzyme in a dose-dependent manner. Factor XIII had no effect. Bars, 200 nm (B, inset, 100 nm).
introduced by transglutaminase activity. To test this hypothesis, we generated crude extracts from chick embryo corneas comprising neutral salt-soluble collagens as wells as acidic macromolecules normally eliminated by chromatography on DEAE-cellulose. These mixtures were subjected to reconstitution of fibrils in vitro (34) from buffers with or without Ca 2ϩ . In some experiments, recombinant tissue transglutaminase-2 was also added. As shown in Fig. 5A, a heterogeneous population of fibrils as well as finely contoured aggregates without defined fibrillar features was formed when tissue transglutaminase-2 was not added. Therefore, a strict diameter control was not apparent under these conditions. By contrast, in the presence of tissue transglutaminase-2, thinner fibrils were formed, which massively accumulated non-fibrillar material on their surface (Fig. 5B). This effect was dose-dependent (Fig. 6). Fibrils were also reconstituted from collagens without material binding to DEAE-cellulose, i.e. without acidic components, including proteoglycans and glycosaminoglycans. As shown in Fig. 5C, average fibril diameters were also reduced in a dose-dependent manner when tissue transglutaminase-2 activated by Ca 2ϩ was added to the reconstitution mixtures. The blood-clotting factor XIII, which is known to occur in cornea (35), was ineffective although the enzyme was active (transfer of dansyl cadaverine to standard proteins; data not shown). Therefore, the restriction of lateral growth of fibrils during reconstitution in vitro is specified by the introduction of cross-links by tissue transglutaminase and not by factor XIII.

DISCUSSION
The corneal stroma is a particularly exquisite example for the adaptation of the suprastructural organization of the extracellular matrix to functional tissue requirements. The significance of the uniformity and of the small diameters of corneal fibrils has been studied extensively, and the aggregated evidence suggests that the control of fibril shapes in the cornea is rather complex. In cartilage, the correct stoichiometry of collagens II and XI (8:1) is sufficient for the rigorous diameter control of prototypic small diameter fibrils (17) that are typically found in immature cartilage or in territorial zones of adult cartilaginous tissues (36). In the corneal stroma, important roles in controlling collagen fibrillogenesis have been assigned to quantitatively minor collagen V, a structurally close relative of collagen XI, and to several SLRPs, including lumican (11), decorin, and biglycan (37) acting in a partially redundant manner (10). Here, we have discovered two additional factors contributing to the control of fibril shape and the formation of lamella. The diameter control depended on isopeptide bond formation catalyzed by transglutaminases, specifically tissue transglutaminase-2, whereas collagen cross-links of the classical, aldehyde variety were necessary for the correct organization of the fibrillar corneal lamellae. Both findings are surprising because the event of covalent cross-linking of collagen molecules is usually thought to occur after rather than during fibrillogenesis. In addition, cross-linking is generally associated with the stabilization of fibrils against their unraveling after assembly. Early isopeptide cross-link formation results in a limited lateral growth of the future fibrils. Presumably, this process also requires the integration of several types of SLRPs with partially redundant roles at later stages of fibrillogenesis. In addition, further cross-linking of the isopeptide variety introduced at later stages may serve to stabilize fibrils after their assembly is completed.
Less is known about the origins of the regular lateral spacing of parallel collagen fibrils and the formation of fibrillar lamellae in the cornea. It has been suggested that dermatan and/or keratan sulfate chains of SLRPs on the surfaces of neighboring fibrils serve as spacers by interacting in an anti-parallel direction (16). Based on this notion, one may predict that crude corneal extracts can be reconstituted in vitro into arrays of parallel collagen fibrils with uniform spacing. Our reconstitution experiments, however, have not yet revealed any indication for the establishment of this type of higher-order suprastructural organization. This may well be due to inappropriate molecular mixtures in our crude extracts or to still insufficient control of the experimental conditions, although it was possible to reduce fibril diameters by providing transglutaminase activity to the reconstitution mixtures. Therefore, further details of the exact mechanism controlling the spacing of corneal fibrils still remain to be elucidated.
It has been shown previously (for review, see Ref. 6) that keratocytes in culture can elaborate, at least in principle, the lamellar organization of collagen fibrils resembling that of early developmental stages in cornea. In addition, the cells retain this capacity even if their authentic extracellular matrix is entirely eliminated. However, the mechanistic origins of the formation of lamellae, their size limitation, and their orthogonal stacking remained obscure. Here, we observed that covalent collagen cross-linking is essential for lamellar stacking. In this context, the chemistry of cross-linking is oxidative deamination of (hydroxy)lysyl residues by lysyl oxidases, followed by spontaneous condensation reactions between the side chains containing the enzymatically introduced aldehyde groups. In this way, aldehyde-type collagen cross-links formed early during fibrillogenesis appear to lay the ground for the generation of lamellae. In an earlier study (38), the formation of orthogonal lamellae was observed even though lysyl oxidase activity was inhibited starting at day 8 of in ovo development, i.e. 3 days after the deposition of the primary corneal stroma. Therefore, we surmise that the aldehyde-type cross-linking triggering orthogonal lamellar stacking must occur at early stages of fibrillogenesis long before matrix deposition is complete. Supporting this notion, we found that dimeric and tetrameric collagen polypeptides are formed in the matrix deposited by keratocytes. Tetramers are generated by covalently connecting two dimers. Trimers are also formed, but only at later stages of the cultures. It is tempting to speculate that the trimers have a cross-link chemistry that differs from that of the dimers and tetramers even if they both originate from oxidative deamination. It also remains to be established by future studies whether the trimers only serve to stabilize the fibrils or whether they are also required for lamellar stack formation.
The principle that early cross-linking of collagens is required for appropriate tissue morphogenesis may well apply in general. Indeed, our preliminary experiments with cultured chick embryo tendon fibroblasts analogous to the cultures described here indicate that bundle formation of collagenous fibrils also depends on cross-linking of the aldehyde-derived variety. We conclude that the important roles of cross-link formation not only include the structural stabilization of mature connective tissues but also to achieve appropriate tissue organization.