In both the v4 protoarray and a custom extracellular nerve-related protoarray, corneal KS bound the greatest number of proteins. In contrast, sturgeon notochord CSA bound significantly fewer v4 proteins and extracellular nerve-related proteins than did KS. Nonsulfated cockscomb HA bound the fewest v4 proteins in comparison to corneal KS and sturgeon notochord CSA, and no extracellular nerve-related proteins at all. Mimecan, included in this study because it is the KS core protein about which the least protein-binding information is known, bound slightly more v4 proteins than did sturgeon notochord CSA and was not tested with our custom extracellular nerve-related protein array. However, binding strengths of 9 of the 23 sturgeon notochord CSA-binding candidates were higher than the binding strength of the strongest corneal KS-binding candidate. In contrast, binding strengths of five of the six nonsulfated HA-binders were lower than the strengths of any positive highly sulfated corneal KS- or highly sulfated CSA-binding proteins. Both corneal KS and sturgeon notochord CSA bound preferentially with the biologically active (IgG1
-tagged) form of almost all extracellular nerve-related epitope pairs, suggesting that the GAG-protein interactions observed in this study have potential in vivo relevance. Although generally regarded as extracellular, KS24,62–66
have also been detected intracellularly. These GAG–protein binding profiles demonstrate that GAGs found in the corneal stroma may interact with many proteins, both intracellularly and extracellularly, and thus could influence the availability and/or the configuration of these proteins for subsequent interaction with their target cellular ligands/receptors.
Our study is the first to examine binding of corneal KS, sturgeon notochord CSA, or nonsulfated HA to a wide variety of cellular and ECM proteins using high-throughput protoarray techniques with the proteins immobilized on a fixed surface and the GAGs free in solution. Other protoarray studies examining GAG–protein interactions have immobilized naturally occurring and synthetic GAGs on a fixed surface and probed them with solutions containing a limited number of known proteins.67,68
That the protoarray format can reveal new protein–GAG interactions is substantiated by the fact that in a study by Wang et al.68
dermatan sulfate GAG was found to bind to an antibody not previously thought to bind it. In our study the protoarray format first revealed KS binding to a SLIT family member, which was then further defined for KS binding to SLIT2 by SPR studies. It is not clear from our and other comparative protein-GAG interaction studies whether it is better to attach the GAG or the protein to the fixed matrix, and what is optimal may differ for different GAGs and different proteins. Attaching some GAGs69
to a matrix has been found to impose structural constraints that have diminished their binding to some proteins. Also, some GAGs67,71
and some proteins may attach better to a given matrix than others. In our study, HA in solution did not bind the known HA-binders aggrecan, link protein, and fibronectin affixed to a matrix, but it has been shown to bind link module when HA is the fixed moiety.45,72
In our study, KS in solution bound SEMA3A in the protoarray format where SEM3A was the fixed moiety, but not in the SPR format where KS was the fixed moiety.
Although there are several indicators of reproducibility for the protein binders for KS, CSA, and HA detected in our protoarray protocol, our protocol failed to detect binding of some of these GAGs to proteins they have been reported to bind in other studies, such that these known GAG binders did not serve as positive controls in our protoarray study. The most probable reason that the v4 versions of aggrecan, link protein, or fibronectin did not serve as positive controls for HA binding in our proteomics study is that the v4 versions of these proteins are produced in insect cells and are therefore not glycosylated in the ways that these molecules are when produced under their usual human cell biosynthetic conditions. Deglycosylated aggrecan cannot bind HA,73
and it is likely that the same is true for nonglycosylated link protein. In addition, fibronectin binding of HA requires that the fibronectin be both glycosylated and aggregated, and nonaggregated fibronectin bound to an affinity column does not bind HA.74
In the same vein, although KS had been shown to bind IGFBP2 before this proteomics study, IGFBP2 could not serve as a positive control for KS binding in our proteomics screening format because IGFBP2 binds GAGs only if IGF-I or -II are present.75
These considerations suggest that the proteins identified as GAG binders in our protoarray format bind those GAGs without being posttranslationally glycosylated or the participation of additional cofactors.
In many GAG-protein interaction studies, biotinylation is a common method of labeling either the proteins or the GAGs. In some protoarray studies, soluble protein ligands are labeled with biotin for detection after GAG binding.67,68
We used immobilized proteins with solubilized bGAGs for detection. Others have also used bGAGs to study GAG interactions with immobilized proteins.71–72,75–78
However, biotinylation procedures can biotinylate different positions within a GAG,69
as occurred in our study with biotinylation of KS at only one end of its chain and biotinylation of CSA and HA all along their GAG backbones. In addition, biotinylation can have different efficiencies with different GAGs, again as it did in our study. These differences in biotinylation may have different effects on the functional properties of GAGs.69,71,79
Similar position, efficiency, and steric hindrance effects may perturb the binding properties of proteins if they are the biotinylated moieties in GAG–protein binding studies. Also, many GAG–protein interaction studies using column chromatography or SPR protocols use the biotin of the bGAG to attach the GAG to the fixed substrate,32,69,80
as we did in our SPR experiments. This method could magnify structural aberrations in a GAG to a greater or lesser extent, depending on where the GAG is biotinylated. Thus, protoarray techniques can suggest false GAG–protein interactions, and proteins identified by protoarray studies as GAG–binding proteins should be considered candidate GAG–binders until additional techniques confirm the GAG-protein binding.
GAGs are defined by their primary disaccharide structure, and therefore KS, CS, and HA have the same primary structure from one species to another. However, the extent and positions of sulfation of GAG disaccharides can vary considerably for any one kind of GAG, both from species to species, and also from tissue to tissue within a given species. Bovine and human corneal KS chains have similar charge densities and chain lengths, and the three most prevalent capping structures at the nonreducing termini are identical in relative proportions.81
Therefore, it is likely that human corneal KS chains would bind v4 human proteins in a manner very similar to that displayed by bovine corneal KS. In contrast, sturgeon notochord CSA is 100% sulfated82
(every disaccharide is monosulfated at C-4 of GalNAc), which is a higher percentage of disaccharide sulfation than is found in most human83
corneal CSA. The extent and positions of sulfate groups within CS chains have been shown to significantly affect CS binding to protein targets,33,34,84
leading to the proposal of a sulfation code85
for GAGs that governs their ability to interact with proteins. It is not known whether human corneal CSA chains may be asymmetrically sulfated, such that they have regions of high (100%) disaccharide sulfation contiguous with regions of low disaccharide sulfation. If they do, 100% sulfated regions (sulfation hot spots) of such human corneal CSA would be expected to show protein interactions similar to those seen with sturgeon notochord CSA, whereas the low sulfated regions may bind very different proteins. HA is never sulfated, so HA from any species or tissue should be equivalent in its binding properties with human proteins, as long as their average chain lengths are similar.
In our GAG protoarray studies, Invitrogen optimized conditions for KS binding in their protoarray protocol and then used the same general conditions for CSA and HA binding. It is possible that under other binding conditions, highly (or asymmetrically) sulfated CSA and/or HA would react with more candidate proteins. Alternatively, spotting complete posttranslationally modified forms of proteins such as aggrecan, link protein, or fibronectin on microarray plates may also increase the number and kinds of proteins that bind GAGs in a protoarray format. No mimecan binding proteins were known before our study, so it is difficult to assess whether the binding conditions used in this study were optimal for mimecan.
Because corneal KS, sturgeon notochord CSA, and HA are large linear polyanionic molecules, it is probable that electrostatic interactions are part of the mechanism involved in the protein-KS, -CSA, and -HA binding observed in our protoarray studies, as has been demonstrated in many heparin- and heparan sulfate-protein interactions.86
Recent studies of reactions of yeast87
protein arrays with a range of biotinylated linear polyanionic macromolecules (actin, tubulin, heparin, heparan sulfate, and DNA) have defined a wide array of polyanionic binding proteins (PABPs), with sulfated GAGs binding the greatest number of different PABPs in each study. In addition, lack of reduction in the number of PABPs bound by increasing binding solution salt concentrations suggests that there are also noncoulumbic (e.g., hydrophobic and hydrogen binding) molecular interactions involved in protein-GAG binding.88,89
Moreover, reacting proteins did not need to be positively charged overall, nor to have a specific GAG-binding signature sequence motif,18
but only needed to have some positive subdomain along their amino acid sequence to be capable of polyanionic binding.86–89
The authors argue that many proteins contain subdomains of amino acids that are probably unfolded in solution90
and that these disordered regions may allow them to interact with many different binding partners with various degrees of strength or specificity by mechanisms such as electrostatic interaction that do not require tertiary structure complementarity. Similar considerations also apply to interactions of proteins like kinases with smaller anionic molecules such as single nucleotides.89
In our study, 75 of the 217 KS-candidate v4 protein binders and 9 of the 23 CSA-candidate v4 protein binders were kinases. The catalytic domains of all kinases contain two invariant lysines and an arginine, with the serine/threonine kinases containing an additional invariant lysine, and the tyrosine kinases containing additional histidine, arginine, and lysine residues,91
suggesting that kinase catalytic domains may have an overall basic charge that would allow them to interact electrostatically with polyanionic GAGs. Viewed more broadly, 28 of the top 50 KS-candidate binders and 11 of the 23 CSA-candidate binders could be classified as PABPs on the basis of their known binding to actin, microtubules, DNA, RNA, nucleotides, or nucleosides. HA's top protein interactants could also be viewed as potential PABPs. Keratin-associated protein 13-1 (KRTAP13-1), contains one partial (-RPR-) basic amino-acid–rich GAG-binding motif18
in its short sequence.92
Insulin receptor substrate 1 (IRS1), possesses numerous basic amino-acid–rich segments in its primary amino acid sequence and two ATP-binding sites.93,94
The nature of the interactions between mimecan and its 15 candidate v4 protein binding partners remains unresolved. The revelation that mimecan binds SLAIN2, a β-tubulin-like protein, and three kinases that regulate microtubule behavior suggests a possible role for mimecan in microtubule function in vivo. It is probable that the protein-binding behavior of mimecan core protein is altered when it is glycosylated with KS chains. Although all three corneal KS core proteins are small leucine-rich repeat proteins, mimecan is smaller than the other two proteins.4–6
It would be interesting to know whether the protein-binding properties of KS are affected differently when it is attached to different core proteins.
SLIT-1, -2, and -3 orthologues usually function as chemorepellants of migrating neuronal growth cones.95
However, an N-terminal fragment of SLIT2, purified from rat brain, causes DRG sensory axons to elongate and branch.96
Full-length neurorepellant SLIT2 can be transformed into neuroattractant and bifurcation N-terminal SLIT2 by interaction with ECM components,61
including rat brain heparan sulfate proteoglycan glypican-1(GPC1).97
Heparan sulfate O
-sulfate groups are critical for GPC1-SLIT2 binding.98
The mechanism of the proteolytic SLIT cleavage that results from GPC1-SLIT binding is not yet known. Our protoarray and SPR demonstrations that SLIT2 binds significantly to highly sulfated corneal KS suggests that polyanionic, highly O
-sulfated ECM KSPG may function as a SLIT2-cleavage facilitator in the cornea, thus converting SLIT2 into its neuroattractant/neurobifurcating form in the area of the developing cornea where corneal nerves are extending and branching extensively.53,54
This possible chaperone role for polyanionic KS in SLIT2 transformation indeed would mimic the role of polyanionic regions in other better-known chaperone–protein interactions.99–101
In addition, under protoarray conditions, ECM neurorepellant SEMA3A was identified as a candidate KS binder. Perhaps association and dissociation constants for SEMA3A with KS could not be determined by SPR because of increased steric hindrance in KS function when it is the moiety attached to the fixed substrate. Our observations suggest that corneal KS is very influential in regulation of corneal innervation during development.