Ingenuity Pathway Analysis.

Ingenuity Pathway Analysis (IPA) was used to further explore functional aspects of our protein dataset. We describe here the top 3 significant results (the remaining categories are described in Table S4). A total of 134 N-glycoproteins in our dataset are involved in cell-to-cell signaling and interaction (CTCSI) and 122 are involved in cellular movement (CM), the top two molecular and cellular functions (Figure 2). CTCSI represents primarily proteins involved in adhesion and activation of various cells of the immune system. Specific enriched immune functions include responses of phagocytes, neutrophils and antigen presenting cells. The main proteins that contribute to the immune responses were found to be transmembrane receptors such as toll-like receptor 4 (TLR4), beta 2 microglobulin (B2M), a component of MHC class 1 molecule, Fas (TNF receptor superfamily, member 6), IL2RA (interleukin 2 receptor alpha) and others. There are several reasons why these proteins could be more affected by polymorphism in the N-glycosylation sites. For example, it is known that N-linked glycosylation of toll-like receptors is required for their translocation to the cell surface [42]. N-linked glycosylation is known to activate NFkappaB signaling, an important pathway in the immune response [43].

In the cellular movement category, migration, invasion and movement of cancerous cells are the major enriched classes. Most of the proteins contributing to this important feature are tyrosine kinases such as PTK2B, ERBB3, EPHB4, and FGFR1. Tyrosine kinases play a vital role in many cellular functions such as cell differentiation, proliferation, motility and cell survival. Mutations in these kinases lead to deregulation of various cellular processes resulting in unbalanced proliferation and differentiation, which can result in cancerous growths [44]. Receptor tyrosine kinases have been suggested as drug targets in cancer therapy but cancer cells tend to develop resistance to these therapeutics. Targeting N-linked glycosylation, a highly regulated step for maturation of these receptors, is a novel therapeutic idea. A recent study pointed out that inhibiting N-linked glycosylation reduces receptor tyrosine kinase signaling and radiosensitizes tumor cells [45]. The authors also suggested that enzymatic steps regulating N-linked glycosylation can serve as novel targets in the treatment of malignancy in general. The next top functional category predicted by IPA was cellular function and maintenance. Most glycosylated proteins in this category are involved in cytoskeleton organization such as organization of actin filament extension of cellular protrusions; transportation through ion-channels (e.g. flux of calcium ions) and depolarization of cellular and mitochondrial membranes.

Analysis of physiological system function and development identified over-representation of proteins involved in tissue, organismal and, hematological system development (Figure 2). The tissue development category comprises development of various organs such as liver, kidney, and bone by adhesion of cells. The organismal development refers to a broad category comprising developmental stages from zygote to multicellular animal or from young adult to aged adult. All the major processes in this category are influenced by polymorphic N-glycoproteins. It is not surprising to note such a global effect of N-linked glycosylation because this pathway expanded significantly in multicellular organisms especially in the vertebrates. It has been shown that N-glycosylation influences many biological processes in an organ and species specific manner [46]. Many of the knockout mouse models targeting enzymes of the N- glycosylation enzymatic pathway are embryonic lethal since they affect organ development.

Pathway analysis by IPA identified 13 of 35 glycoproteins involved in the coagulation system among the polymorphic glycoproteins (p<0.05). Similarly, 11 out of 29 molecules in the intrinsic prothrombin activation pathway matched our dataset (p<0.05). The top two pathways, the coagulation system (extrinsic) and the intrinsic prothrombin activation pathway, refer to the coagulation cascade [47] where glycosylation is known to be important [48].

Clustering analysis.

To further understand protein families, we have clustered all the proteins based on significant end-to-end similarity. Major clusters include the G protein-coupled receptor (GPCR) cluster (42 proteins), the kinase (EGF receptor subfamily) cluster (27 proteins), the HLA class I histocompatibility antigen type proteins (17 proteins) and proteases (13 proteins) (see Table S5). This is in agreement with the results of the domain and pathway analyses.

With 42 members (expected 38; hence not over-represented) GPCRs form the largest cluster in our dataset. Even though the GPCRs are not over-represented, we provide a brief description of the cluster members as this cluster is the largest in our dataset. All the proteins in this group are membrane proteins with 20 olfactory receptors representing a major subset. It is known that glycosylation of GPCRs, the largest family of cell-surface receptors, is central to normal reproduction [49], [50]. Glycosylation of these receptors also affects stability and localization of the proteins. For example, mutation of glycosylation sites of the glucagon-like peptide 1 receptor (GLP1R) result in aberrant distribution with most of the receptor localized in the endoplasmic reticulum and golgi compartments instead of the cell surface [29]. This is one of the reasons why we include all the proteins in our analysis irrespective of their annotated location in UniProtKB/Swiss-Prot.

The kinase (EGF receptor subfamily) cluster is over-represented for the tyrosine kinases (expected 6, observed 21) including receptor tyrosine-protein kinase erbB-4 (ERBB4), fibroblast growth factor receptor 2 (FGFR2) and tyrosine-protein kinase Tec (TEC). All of the proteins in this cluster are ATP-binding proteins and the majority of them (21 out of 27) are known membrane/secreted proteins. 17 of the 21 tyrosine kinases in our list are annotated as receptors with a ligand-binding extracellular domain and a catalytic intracellular kinase domain. Analysis of this subset showed that 15 proteins contain a new NXS/T in the extracellular region and are expected to be functionally affected by N-glycosylation [24], [44]. We further explored the connection of these proteins with diseases and, as expected, several map to cancer diseases, with the majority implicated in colorectal and breast cancers. For example, the variat T140I in ERBB4, leads to loss of a glycosylation motif and altered development of colorectal adenocarcinoma [51].

HLA class I histocompatibility antigens (17 proteins) consist of membrane/secreted proteins including HLA class I histocompatibility antigen, beta-2-microglobulin, and T-cell surface glycoprotein CD1c. The majority of the proteins are involved in immune responses as part of the MHC protein complex. This group of proteins is also over-represented (expected 7, observed 17) in our dataset. It is known that glycosylation plays an important role in molecules of the MHC that present antigens to T cells [52] as part of the adaptive immune response [53].

Serine proteases represent another set of proteins with over-represented NXS/T variants (expected 7, observed 13). N-glycosylation of the proteases and their substrates plays an important role in protease activity. The serine proteases participate in the processes of blood coagulation, apoptosis and inflammation [54]. All of the processes described above represent a recurrent theme in our analysis and point out the overall importance of N-glycosylation in these pathophysiological pathways. Overall the protein families that are over-represented in our dataset based on cluster analysis show that protein involved in immunity, blood coagulation and kinase activity are the major types that undergo changes in glycosylation. A survey of nsSNV in these proteins reveals that there is an abundance of nsSNVs in the blood coagulation pathway proteins and serine proteases (number of amino acids in the human proteome 11293923, total nsSNVs 494983; number of nsSNVs inside data set 1959, expected 1335, p-value 0.05). The difference between the expected and observed is less for HLA class I histocompatibility antigens and related proteins with the MHC domain (number of nsSNVs inside data set 1952, expected 2070, p-value 0.40) and for the kinases with the protein tyrosine kinase domain (number of nsSNVs inside data set 5670, expected 4749, p-value 0.09). It is possible that the blood coagulation group of proteins indeed has a high level of variation or there is a distinct bias in the research community towards these groups of proteins. Availability of more data will allow us to identify if there is indeed a bias in the research focus, which subsequently results in more depositions of SNVs for some of these functional groups.

In some rare cases, such as HLA class I histocompatibility antigens (HLA-B and HLA-C) and tubulins (TBA1C_HUMAN and TBA1B_HUMAN), the change that leads to gain of glycosylation occurs in the same N-glycosylation motif in the related proteins and could indicate a common functional effect in the gene family. Unlike the HLA class I histocompatibility antigens, which are membrane proteins, the tubulins are found in the cytoplasm and cytoskeleton and therefore are not expected to be glycosylated. There are reports on tubulin glycosylation and it has been suggested that some portions of the α and β tubulin are translated in the endoplasmic reticulum and hence it is possible they might be glycosylated [55], [56].

Structural analysis.

We performed structural analysis to elucidate where the modifications are located. We found that 277 out of 1091 proteins have structures available in the Protein Data Bank (PDB) [57] at the position of variation. It was previously believed that the glycosylated region has to adopt beta-turn conformation [58]. Later it was found that many turns or loop conformations are suitable to produce an exposed glycosylation site [59]. Petrescu et al. analyzed the secondary structures at the N-glycosylation sites and found that turns and bends are favored with ∼27% and 18% of occupied sites on turns and bends with the lowest glycosylation site incidence (11%) on helices [60]. We notice that the majority of the variations are within the loop and bend regions (32% are in loop regions, 16% in bends, 10% in the strands and 22% in the helices). Zielinska et al. predicted the secondary structure localizations in mouse glycoproteins and suggested that 71% of the N-linked asparagines are located in the loop regions. [9]. A survey of the distribution of SNPs in the different secondary structures shows that there is no bias in the occurrence of nsSNPs in the different secondary structural elements (number of SNPs inside loops/turns: 1385, expected 1305; strands: 9374 expected: 8355; helix: 14118, expected: 13376). These findings indicate that although the nsSNPs that affect N-glycosylation motifs are enriched in the loop and bend region (48%) and under-represented in the other secondary structural elements, the enrichment is not as high as expected based on the fact that 71% of the aspargines are located on loops. This can be attributed to the fact that N-glycosylation sites in loops are functionally important and hence, less likely to contain variation.

We found that 189 proteins in our dataset are secreted based on UniProtKB/Swiss-prot annotation and prediction results from WoLF PSORT [61]. Network analysis followed by Gene Ontology enrichment analysis of these 189 proteins using STRING [62] reveals that a large number of these proteins are involved in blood coagulation. Response to wounding, platelet activation, blood coagulation and hemostasis belong to the top terms, ranked based on p-value. We also saw blood coagulation pathway as one of the top canonical pathways predicted by IPA and over-representation of this set in our clustering and domain analysis was also observed. Antithrombin III, a protein which loses an NXS/T site due the N167T substitution, is found to interact with ten proteins of which 9 are involved in blood coagulation pathway based on STRING network analysis (Figure 3). We have selected this protein to carry out structural and functional analysis as described below. Availability of structure in PDB (2ANT) of antithrombin III allowed us to model the variation and see the effect of loss of glycosylation motif (UniProtKB accession P01008, position 167NKS169; PDB id 2ANT, position 135NKS137). As there is no carbohydrate attached to the glycosylation motif 167NKS169, this is an example of changes can be observed at the structural level due the loss of the glycosylation motif. Modeling results show that the loss of glycosylation motif does not result in significant changes to the structure when compared with 2ANT (0.42Å RMSD). Our model also shows that the variation increases the relative solvent accessibility (RSA) of that region from 0.69 to 1.00, which could have an effect on heparin binding (heparin binding site: F156– G180 [63]). Bayston et al. [64] investigated the role of loss of glycosylation on antithrombin deficiency in an asymptomatic individual (and family) with borderline levels of antithrombin. Direct sequencing confirmed the mutation that leads to N167T substitution and loss of glycosylation near the heparin interaction site which results in deficiency of total antithrombin in the plasma of the asymptomatic individual and her family. The author’s hypothesize that the loss of glycosylation results in more conformational flexibility which in turn results in lower affinity binding to heparin. Our modeling studies do not show any conformational flexibility but it is possible that the carbohydrate assists in heparin binding and its loss could lead to lower binding affinity to heparin.

In the second example we chose a protein that has a carbohydrate attached to the glycosylation motif. Based on our functional analysis using IPA, we had observed that cathepsin D, an acid protease, is in the list of top cellular, molecular functions and diseases such as metabolic, genetic and developmental disorders and has a structure available with a carbohydrate attached at the variant glycosylation sequon. The loss of the N-glycosylation motif is caused by the variation T136A in this protein (UniProtKB accession P07339, position: 134NGT136; PDB id 1LYA, position: 70NGT72). While there were no major structural changes, we did observe a loss of a hydrogen bond that originally existed between the threonine residue and the sugar (Figure 4a and 4b). This also changed the charge distribution at the site making it more neutral with the alanine present. Also, there was an increase in solvent accessibility of the asparagine from 0.45 to 0.62 and a decrease in the accessibility at the site of the variation from 0.53 for threonine to 0.45 for alanine. There is evidence in the literature for the role of glycosylation in the regulation of cathepsin D in cancers [65]. Cathepsin D contains two N-glycosylation sites at positions 134 and 263 and both sites are important for targeting to lysosomes [66]. While we cannot test this with our data, it is tempting to speculate that the occurrence of the nsSNV in one of the sites interferes with its stability and lysosomal targeting. The example demonstrates the potential of such analysis to stimulate laboratory verification of these interesting functional characteristics.

To summarize, the results obtained from this study provide for the first time, not only a comprehensive survey of the effect of nsSNVs on the human glycoproteome, but also a framework that will assist glycobiologists with sorting and ranking of the datasets available in Tables S1, S2, S3, S4, and S5 to identify proteins of interest. These candidates can be further analyzed in the laboratory to test functional predictions. The structural analysis of antithrombin-III and cathepsin D provides a method that can be applied to study the functional implications of variation that lead to change of glycosylation in other proteins. The requirement is that these proteins either have a solved 3D structure or have closely related proteins with solved structures. The amount of information available on the 1091 proteins described in this paper is vast and is constantly being updated by the UniProtKB/Swiss-Prot curation team. A brief description is provided here to facilitate browsing the datasets in UniProt. Copy accessions from Tables S1, S2, S3, S4, and S5, paste them in the UniProt identifiers input box at http://www.uniprot.org/batch and click on retrieve; in the intermediate window click on UniProtKB to view the results in tabular format; next, one can either browse individual proteins by clicking on the accession numbers or browse by Gene Ontology or UniProtKB/Swiss-Prot keywords by clicking. Availability of more SNV data from ongoing genome sequencing projects, validation of N-glycosylation sites, and availability of more functional data on individual proteins will undoubtedly unravel how variation in this important post-translation modification affects human biology.