Decrease of core-2 O -glycans on synovial lubricin in osteoarthritis reduces galectin-3 mediated crosslinking

The synovial fluid glycoprotein lubricin (also known as proteoglycan 4) is a mucin-type O linked glycosylated biological lubricant implicated to be involved in osteoarthritis (OA) development. Lubricin’s ability to reduce friction is related to its glycosylation consisting of sialylated and unsialylated Tn-antigens, core-1 and core-2 structures. The glycans on lubricin have also been suggested to be involved in crosslinking and stabilization of the lubricating superficial layer of cartilage by mediating interaction between lubricin and galectin-3. However, with the spectrum of glycans being found on lubricin, the glycan candidates involved in this interaction were unknown. Here, we confirm that the core-2 O -linked glycans mediate this lubricin-galectin 3 interaction, shown by surface plasmon resonance data indicating that recombinant lubricin (rhPRG4) devoid of core-2 structures did not bind to recombinant galectin-3. Conversely, transfection of Chinese hamster ovary (CHO) cells with the core 2 GlcNAc transferase acting on a mucin type O glycoprotein displayed increased galectin-3 binding. Both the level of galectin-3 and the galectin-3 interactions with synovial lubricin were found to be decreased in late-stage OA patients, coinciding with an increase in truncated and less sialylated core 1 O -glycans and Tn-antigens. These data suggest a defect in crosslinking of surface-active molecules in OA and provides novel insights into OA molecular pathology. Signature Chondrocytes, Teaming Up Galectin-1 Osteoarthritis Pathogenesis. glycoforms described elsewhere(42). PSGL-1/mIgG2b monomers were detected in the area between 268-460 kDa, showing intense staining of the PSGL-1/mIgG2b expressed in CHO cells transfected with the GCNT1 transferase but less staining with the PSGL-1/mIgG2b with core 1 glycans. The same samples (50 ng each) were also separated, blotted and stained by anti-PSGL1 (panel B) and anti-mouse IgG (mIgG-Fc) C) antibodies in separate experiments, confirming similar concentration of the two constructs. a typical under optimized conditions from a selection of 3 repeats.


INTRODUCTION
Lubricin is a large glycoprotein, which is found in synovial fluid (SF). Lubricin is produced by articular chondrocytes and synoviocytes in the cartilage surface layer, and secreted into the SF. In healthy joints, lubricin molecules are bound to the surface of articular cartilage, are chondroprotective, and provides lubrication at very low friction under high stress (1,2). The lubricating properties of lubricin are dependent, in part, on dense O-glycosylation (3). Lubricin, encoded by the Prg4 gene, has been associated with pathological conditions in the joint tissue, with mutations in Prg4 associated with camptodactyly-arthropathy-coxa varapericarditis syndrome (CACP)(4), a rare arthritis-like autosomal recessive disorder causing joint abnormalities. The lubrication properties of lubricin also suggest that the protein is associated with arthritic diseases, since it has been proposed that defective lubrication aggravates joint degradation (5). Osteoarthritis (OA) is the most common arthritic disease, with high prevalence in elderly people and involves cartilage degradation in the articular joints, leading to pain and restricted motion (6). Boundary lubrication of OA SF deficient in lubricin has been shown to be lowered in in vitro experiments (7). Different isoforms of lubricin are found throughout the body. In addition to presence in SF and the boundary articular cartilage, lubricin has also been detected in low amounts in menisci, blood, urine, and tendons (8)(9)(10)(11)(12). A possible immunological role of lubricin has been suggested, since it has been shown to be involved in sepsis in a mouse model (13), being the most upregulated protein detected in hepatic tissue. Lubricin is also associated with the plasma membrane of human neutrophils (14), and a lubricating and protective function of lubricin has been reported in ocular surfaces (15,16). The full-length isoform found in SF consists of 1,404 amino acids (aa) equivalent to 150 kDa, but is approximately 300 kDa when fully Oglycosylated. The protein contains a central mucin domain, consisting of 59 imperfect repeats of the aa sequence 'EPAPTTPK', where the threonine residues are potential sites for diverse and dense O-glycosylation. Despite its gene name, it is not a typical proteoglycan, and has merely a single proposed glycosaminoglycan site that may or may not be present/occupied (17). The mucin domain is flanked by two somatomedin Blike domains on the N-terminal side, and a hemopexin domain on the C-terminal side, domains, which have been proposed to interact with extracellular matrix (ECM) proteins such as cartilage oligomeric matrix protein (COMP), collagen II and fibronectin (18,19). The glycans on the lubricin mucin domain are the main contributors to its molecular weight, are responsible for the lubricating function of the molecule, are attracting water to the superficial layer and are providing repulsive negative charges to the cartilage superficial layer (20). Recombinant human lubricin (rhPRG4) has been evaluated for a large number of potential clinical applications as a friction reducing lubricant on biological surfaces (16). Forms of rhPRG4 have been investigated for use as an agent to prevent abdominal adhesion, treatment of bladder impermeability, as a lubricant supplement in contact lens hydrogels and as a therapeutic in patients suffering from dry eye disease (21)(22)(23)(24). In a rodent model, addition of lubricin has been shown to delay the progression of OA (25). Intraarticular injections of rhPRG4 in guinea pigs (26) and Yucatan minipigs (26) with medial meniscal destabilization, a model system for posttraumatic OA, exhibited decreased cartilage damage and inflammation. Lubricin may therefore be a potential candidate for treatment of OA. Galectins are glycan-binding proteins, with many proposed intra-and extracellular functions within cell regulation, the immune system and cancer progression (27). They bind a large range of bgalactosides present both on N-and O-linked glycoproteins, with binding affinity highly dependent on branching and terminal residues(28). Galectin-3 has been shown to be associated with chondrocytes and proposed to be involved in OA pathogenesis (29,30). It is unique among the galectins, as it is the only galectin that comprise a C-terminal carbohydrate recognition domain linked to a non-lectin N-terminal collagen-like domain making it possible for galectin-3 protein subunits to oligomerize (31,32). In the joint, galectin-3 together with lubricin have been proposed to generate a complex stable lubricating network on the articular cartilage (33). Limited studies have addressed galectin-3 binding to O-glycans. Galectin-3 displays only a weak binding affinity to the T-antigen (Galb1-3GalNAc) on synthetic MUC1 glycopeptides (34,35), which is a highly abundant O-glycan on lubricin, and proposed to be responsible for galectin-3 lubricin interaction (33). Since carbohydrates play such a crucial role in the function of lubricin, we decided to use state-ofthe-art glycomic and glycoproteomic analysis techniques to compare the different glycoforms of synovial lubricin from OA patients and control individuals, and to compare these with rhPRG4 expressed in CHO-cells. We confirmed the presence of galectin-3 in SF, and explored differences in the binding of galectin-3 to the various O-glycoforms. Our results enable us to correlate pathological glycosylation changes in OA with a functionally defective superficial layer, and provide a pathway for recombinant lubricin glyco-design for optimal galectin-3 binding.

RESULTS
Decreased core-2 and sialylation on SF lubricin from OA patients We wanted to determine whether changes in Olinked glycosylation of synovial lubricin were associated with OA pathology. Detailed quantitative analysis of the released O-glycans was performed. Lubricin isolated from SF of controls (n = 7) was compared to lubricin isolated from late stage OA patients (n = 7). O-linked oligosaccharides from lubricin in age-and sexmatched OA patients' and controls' SF were released by reductive β-elimination and analyzed with quantitative MRM (multiple reaction monitoring) on a QTRAP 6500 triple quadrupolelinear ion trap hybrid mass spectrometer as has been described in detail previously (36). The MRM method was developed to semi-quantify a range of twenty-two core-1 and -2 and -3 glycans on native lubricin from both controls and OA patients (36). We refer to semi-quantification here as, without the appropriate standards, a technical difficulty of O-glycan research, absolute quantification is not attained. This is due to differences in ionization and stability between glycans, particularly between charged and neutral glycans. MRM is able to mitigate some of this as each glycan structure, even the sialic acid isomers, are optimized for the conditions that allow for its best quantification (36). Nevertheless, the most apt use of this approach is to compare the same glycan or type of glycans between samples.
Although the abundances of these glycans were found to differ between the two sample types, the glycan repertoire was the same (Fig. 1A CHO expressed rhPRG4 has mainly sialylated core 1 O-glycans Given the potential of rhPRG4 as a biopharmaceutical lubricant, where the function relies on the O-linked glycosylation, we analyzed the released glycans from CHO expressed rhPRG4 to identify the presence of core-1, core-2 glycans and sialylation and compare it to Oglycosylation of the native human SF lubricin. The composition of O-glycans released from rhPRG4 was analyzed using the same MRM lubricin method as for native human lubricin. These analyses were performed on two different batches of rhPRG4, which revealed a composition solely of core 1 structures ( Fig. 1C and D), the same core 1 structures observed on human synovial lubricin ( Fig. 1A and B). The most abundant oligosaccharide was the sialylated core 1 NeuAca2-3Galβ1-3GalNAc structure, which made up 85% of all glycans. A low proportion of neutral O-glycans (12% of all glycans) and a higher amount of charged glycans (88% of all glycans) were found in this rhPRG4. The glycosylation pattern of rhPRG4 is similar to what has been reported for CHO cell Oglycosylation previously, with an abundance of core 1 structures, particularly the linear monosialylated NeuAca2-3Galβ1-3GalNAc structure(37).
Glycoproteomics reveal location of Tn, sialyl-TN and core type 1 structures both within and outside the rhPRG4 mucin domain We next performed a glycoproteomics study of rhPRG4 (with and without partial deglycosylation) to identify glycosylated aa sites, and compared with glycosites previously reported for native human lubricin from pooled SF from OA and rheumatoid arthritis (RA) patients (38). In all, we detected 318 glycopeptides from rhPRG4 covering 172 glycosylated Ser/Thr sites, the monosaccharide compositions consistent with core 1 type Oglycans GalNAcα1-(Tn-antigen), Galβ1-3GalNAcα1-(T-antigen), NeuAcα2-3Galβ1-3GalNAcα1-(sialyl T) and NeuAcα2-3Galβ1-3(NeuAcα2-6)GalNAcα1-(disialylated T). The results are summarized in Fig. 2 and in Supplementary Table 2. Glycopeptide coverage in rhPRG4 was found to be similar to native human lubricin, with the majority of the glycopeptides originating from the S/T rich mucin domain (aa 232-1056; Fig. 2A). Site specificity overlapped well with native lubricin and within the mucin domain (Fig. 2B, C and Supplementary Fig. 1). Eighty-three sites were detected in rhPRG4, which had not previously been identified in native lubricin (Fig. 2B). The majority of the glycopeptides contained 1-3 glycans, although up to five glycosylation sites were detected on one peptide (TTPETTTAAPK, aa sequence 926-936). Glycopeptides generated after partial deglycosylation revealed additional site-specific information, enabling the detection of larger glycopeptides in the size range of 10-20 aa:s. In all, 31 additional glycosylated Ser/Thr were identified after partial deglycosylation (Supplementary Table 2). Approximately 77% (132 of the 172) of the glycosylated Ser/Thr were found in peptides originating from the mucin like 'repeat' region (aa 348-855; Fig. 2A) of rhPRG4. The most abundant aa sequence in this region was the 'EPAPTTPK', repeated 17 times in the molecule. Thirteen 'EPAPTTPK' derived glycopeptides were detected, glycosylated on either or both Thr residues (Supplementary Table  2). We found 19 glycosylation sites in the Nterminal region (aa 125-347) and 26 in the Cterminal region (aa 1057-1404). Most of these sites (16 and 25 sites from the N-and C-terminal region, respectively) were found in the non-repeat regions in the STP rich domain flanking the mucin domain 348-1057 ( Supplementary Fig. 1). The most intense glycopeptides identified in rhPRG4 correspond well with those identified from pooled SF from patients with RA and OA.
The main difference was that in regions with clustered Thr and to some extent also Ser, we usually found that more sites were positively identified in rhPRG4 compared to native lubricin.

Glycoproteomics of SF lubricin from OA patients
Since previous glycopeptide analyses was performed on pooled SF samples (38), we pursued more sensitive MS analyses using HCD-MS and MS/MS, addressing the characterization of SF lubricin glycopeptides from two individual OA patients. This approach revealed 171 and 158 lubricin derived glycopeptides from the two patients, respectively, and the full glycopeptide list is found in Supplementary Table 2. Glycopeptide coverage did largely match what was found in the samples derived from rhPRG4 and our previous study of pooled native lubricin from OA and RA patients ( Fig. 2A and Supplementary Table 2), suggesting that many glycosylation sites coincide. To compare SF lubricin and rhPRG4 glycopeptide analyses, the peptide backbone sequence 'VLAKPTPK' (aa 327-334, Fig. 2D) will be used as an example. Seven VLAKPTPK glycopeptides were identified in each of the two sample types. This peptide backbone sequence contains only one potential O-glycosylation site (T). In addition to simple core 1 structures identified in rhPRG4 (Fig. 2E, F), we also detected more exotic type O-linked glycans attached to native lubricin. Examples of two of these spectra are shown in Fig. 2G, H. Both spectra were dominated by a major fragment ion corresponding to the peptide backbone without its glycosylation (m/z 853.55). The core 2 glycopeptide variant of this glycopeptide was identified containing two HexNAc and one Hex, consistent with a (Galb1-3(GlcNAcb1-6)GalNAca1-) core 2 sequence. Low molecular oxonium ions at m/z 138, 144, 168, 186, and 204 originated from fragmentation of monosaccharide residues and were diagnostic for glycan containing peptides. The other example from SF lubricin is a 'VLAKPTPK' glycopeptide with a T-antigen with an associated sulfate residue (Fig. 2H). We have previously shown that sulfated core 1 O-glycans exist on SF lubricin both as 3-linked and 6-linked sulfate to the nonreducing end Gal residue, but also to the 6-position of the reducing end GalNAc residue (39). The diagnostic glyco oxonium ions at m/z 204, 366 and 446 were glycan fragment ions consisting of HexNAc, HexNAc-Hex and HexNAc-Hex + sulfate, respectively. This peptide example illustrates that sulfoglycopeptides can successfully be identified amongst other glycopeptides in positive ion mode without targeted approaches that have been suggested (40). One N-glycosylated site was found on lubricin from both recombinant and OA patient derived lubricin at Asn1159. The two dominant glycoforms found in both samples were of high mannose type, Hex5HexNAc2 and Hex6HexNAc2 respectively. Annotated HCD and CID spectra of these are displayed in Supplementary Fig. 2, 3.

Lubricin glycosylation identified using lectin ELISA
We investigated the use of lectins to study lubricin glycosylation based on the knowledge generated by glycomics and glycoproteomics. With glycomics, singly and doubly sialylated core-1 and -2 structures were readily detected. Detection of Tn-antigens on lubricin was based on glycoproteomic analysis. We selected PNA (Galb1-3GalNAca1-), galectin-3 (Galb-1GlcNAcb1-), SNA (NeuAca2-6GalNAca1-) and HAA (GalNAca1-) based on detected lubricin oligosaccharides. We used them in a sandwich ELISA using anti-lubricin 9G3 antibody. Additional lectins included AAL (Fuca1-3/6) and L-selectin (sialyl 6-sulfo Lewis X), since it has been shown that lubricin contains both fucosylated and sulfated structures (14). Our lectin sandwich ELISA was used to probe the glycosylation of lubricin from OA patients' and controls' SF ( Supplementary Fig. 4). We found that lectins, corresponding to less abundant structures detected by MS glycomics, varied the most. For galectin-3, AAL and L-selectin, the percent CV were found to be 147% (OApatients)/141% (controls) and 78% (OApatients)/120% (controls), respectively. For SNA and PNA the percent CV varied less; 36%(OApatients)/74% (controls) and 25%(OApatients)/40% (controls), respectively. HAA lectin binding showed the least variation with percent CV of 37% and 34% in controls and OA-patients, respectively. No significant differences were detected in the ELISA response between different glycoforms of SF lubricin (n= 6 controls, n=8 OA-patients; Supplementary Fig.  4). The combined conclusion from MS-glycomics, glycoproteomics and lectin sandwich ELISA is that Tn, T and its sialylated version are by far the dominating structures on lubricin, while pathologic OA changes occur within the low abundant structures.

Galectin-3 binding to lubricin and endogenous galectin-3 in SF
We further investigated the ability of lubricin to bind galectin-3. SF from 30 OA patients was probed with recombinant galectin-3 using the lectin ELISA assay. We found that the level of galectin-3 active lubricin present in SF varied approximately 1-2 orders of magnitude between these individuals (Fig. 3A). Periodate treatment, which destroys the glycans, abolished the binding, proving that the interaction was purely dependent on glycosylation (Fig. 3A). Since our glycoproteomic analysis showed that the Tnantigen was present on lubricin, we investigated if there was a difference between OA patients and controls in how much of this biosynthetic precursor was further processed into core-2. The relative proportion of galectin-3 binding (core-2) against the binding to the HAA lectin (Tnantigen) showed that the glycosylation of lubricin from late stage OA (n=12) compared to the controls (n=13) had changed. The relative binding to galectin-3 compared to HAA decreased by almost a factor of 2 in OA patients (p<0.001; Fig. 3B). This data corroborates the shortening of glycans in OA observed by glycomics where core 2 glycans were decreased relative to T antigen ( Fig. 1). Here we show, by using HAA that core 2 glycans, measured by galectin-3 were decreased relative to the truncated Tn-antigen (Fig. 3B). Together this suggests that lubricin carried less core-2 glycans in OA on a molecular basis.
Since galectin-3 has been reported to be present in SF (41), we speculated that the difference in binding of galectin-3 to lubricin may have functional consequences. Hence, we went on to also establish the level of SF galectin-3 in our sample set. The data generated showed a significant decrease (50%) in the total amount of endogenous galectin-3 in SF of the late stage OA (n=23) compared to controls (n=14) (Fig. 3C). Our data suggest that both the level of galectin-3 and its ability to bind SF lubricin, are altered as a consequence of OA.

Galectin-3 and core 2 glycosylation
A potential functional impact of the altered OA glycosylation identified was investigated further. Lubricin glycans are essential for several critical properties of lubricin involving lubrication and binding properties and have been suggested to bind to galectin-3 to improve lubrication of the joint via the abundant T-antigen epitopes (Galb1-3GalNAc1α-) (33). However, we observed decreased lubricin binding to galectin-3 in OA in parallel with an increase of this glycan on OA lubricin, which lead us to test the hypothesis that the lubricin galectin-3 interaction may be mediated by epitopes presented on extended core 2 glycans. To address if core 2 glycans compared to the simpler core 1 glycans were responsible for the galectin-3 binding we used recombinant glycotechnology. We used a CHO-cell line stably transfected by the human core-2 GlcNAcb1-6 glycosyltransferase shown to convert the native CHO-cell core 1 glycosylation into sialylated core 2 with over 90% efficiency (42) . A reporter O-linked mucin type glycoprotein (PSGL-1/mIgG2b) expressed in these cells was shown to be able to bind recombinant galectin-3 ( Fig. 4A-E).
Without transfection of the glycosyltransferase, only low binding to PSGL-1/mIgG2b was observed ( Figure 4A, fist lane). This demonstrated that core-2 type oligosaccharides are key for recognition by galectin-3. Further support for this was provided by surface plasmon resonance (SPR). The core-1 recombinant version of lubricin CHO-rhPRG4 glycosylation was not able to bind galectin-3, while the positive control of lubricin isolated from bovine chondrocyte cell culture (bPRG4) with both core 1 and core 2 glycosylation (Supplementary Table 3) displayed significant interaction with galectin-3 with a KD calculated to 3.8 µM (Fig. 4F, G). The data together indicates that the avidity of the multimeric galectin-3 relies on numerous epitopes within lubricin and its mucin domain. The PSGL-1/mIgG2b O-linked oligosaccharides generated after transfection of CHO-cells using human core-2 GlcNAcb1-6 glycosyltransferase terminate with sialic acid after adding a LacNAc unit to the natively expressed core 1 structures (42). This is also the type of core-2 structures found on SF lubricin (36).

DISCUSSION Altered glycosylation and lubricin level in
OAThe data from lectin ELISA suggested that the glycosylation in late stage OA had changed compared to controls (Fig. 3B). However, using only 2 lectins (HAA and Gal-3) to display this change makes it difficult to pinpoint the overall nature of this glycosylation change. For this, the MRM semi-quantification of O-glycans from SF lubricin gave more information, demonstrating a 10% increase of the truncated unsialylated core 1 structures (Galb1-3GalNAcα1-) in late stage OA, accompanied by a decrease of low abundance core 2 glycans (Fig. 1A). This MRM result and the data from the lectin ELISA suggest that also the Tn (GalNAcα1-) antigen is increased, at the expense of larger structures including both core 1 and core 2. However, since the lectin ELISA is limited in its ability to quantify the level of lubricin in the sample, the absolute levels of the various Tn, core 1 and/or core 2 structures in lubricin are difficult to appreciate. The fact that glycosylation of lubricin is changing, makes the quantification of lubricin difficult. Current methods for determining lubricin concentration is often based on antibodies targeting the changing mucin domain and/or lectins for detection or capturing (43). Hence, the glycosylation change of lubricin may have contributed to the confusion in the literature, where the level of SF lubricin has been shown to be increased, decreased or unaltered in OA (43).

Functional implications of lubricin glycosylation
It has been demonstrated previously that the high number of negatively charged sialic acid residues located within the center of the lubricin molecule with positively charged lysine and arginine rich N-and C-termini, creates the amphoteric nature of the lubricin glycoprotein (38). The glycosylation, along with the organization of sialic acid within the molecule, has been associated with the effective lubrication property of lubricin (38,44). This organization creates a central hydration shell and sparingly glycosylated N-and C-termini, available to bind to other cartilage proteins including COMP and collagen II. An effective boundary lubrication model with firm adherence to the articular cartilage surface of glycomediated lubrication can be proposed (18,19,38). The reduction in charged residues in OA observed in this report may impact the overall charge and potentially the lubricating properties of the lubricin molecule in OA. Our data on SF lubricin O-glycan truncation in OA also bears implications on binding to galectin-3. Previous studies have indicated that galectin-3 exhibits only weak affinity to Oglycoproteins and O-glycopeptides expressing single core 1 O-glycans (Galb1-3GalNAc or NeuAcα2-3Galb1-3GalNAc) (35,45).
Instead, internal LacNAc glycans on extended oligosaccharides have been described to be the high affinity ligands for galectin-3 (46). Here, we suggest that interaction between galectin-3 and native lubricin is mediated by short single LacNAc extended core 2 structures. Indeed, expression of the core 2b-1,6-Nacetylglucosaminyltransferase (C2GnT) responsible for biosynthesis of core-2 O-glycans has been shown to enable binding of galectin-3 to bladder tumor cells (47). rhPRG4 has shown to have similar lubrication properties as bPRG4 in vitro (48). Hence, the proposed galectin-3 binding to lubricin (33) are likely to only fine-tune the boundary lubricating glycosurface of the joint. In regards to galectin-3 levels in SF, it has also been shown that patients with RA have increased SF galectin-3 (41,49,50), while the literature is less conclusive about the levels in OA. Our data indicate that compared to normal SF, the galectin-3 levels are diminished in late stage OA. Our hypothesis suggests that an altered ability of OA lubricin to bind galectin-3 together with decreased galectin-3 levels in OA may contribute to destabilization of boundary lubrication and contribute to joint degradation (Fig. 5). However, this notion needs to be reconciled with results from a large animal OA model where CHO cell derived rhPRG4 was shown to retard cartilage degeneration (26). Our data shows that glycomic and glycoproteomic analysis provides the ability to unravel the multifaceted enigmatic OA aetiology and pathology. Of note, this is even without taking into account that cartilage harbors the largest volume of glycans (keratan-and chondroitin sulfate) in the body.

Glycomap of native lubricin
A previous map of native lubricin was performed on tryptic and Lys-C digested O-glycopeptides of intact and partially de-glycosylated glycopeptides from a pooled sample (38). Here, we included sites identified on native lubricin from individual patients using a fast scanning Orbitrap mass spectrometer to make the glycosite map more complete. Due to fast scanning and sensitivity, low abundant glycoforms could be selected for MS/MS, without any prefractionation and multiple digestion strategies. We identified 171 and 158 glycopeptides from the two lubricin tryptic digests from OA patients, many of which have not been detected before. This sensitive approach allowed the first detection of a sulfated O-glycopeptide from lubricin (Fig. 2), and evidence for an N-glycosylation site on lubricin ( Supplementary Fig. 1, 2). In our former study of pooled lubricin from both OA and RA patients, we reported 185 O-glycopeptides, yet after excluding those glycopeptides that were enzymatically de-glycosylated, this number dropped down to 122. However, partial deglycosylation may still need to be carried out to identify sites within the mucin domain. For example, the glycopeptides with aa sequences 'EPAPTTTK', 'KPAPTTPK' and 'SAPTTPK', which are found repeated eight, four, and four times in the protein respectively, were either not detected or poorly represented in the analyses of individual OA patients, but were found in both rhPRG4 and in our previous study after deglycosylation. In this paper we identified eight sites in native lubricin in addition to already recorded sites (38).

The rhPRG4 molecule
The analyses here showed the detailed glycomic and glycoproteomic analyses of the heavily Oglycosylated CHO expressed protein lubricin giving glycosite-specific information as well as detailed O-glycan identification and quantification. Obtained glycopeptides from rhPRG4 digested with trypsin or Lys-C, were analyzed with LC/MS/MS and CID/ETD fragmentation before and after partial deglycosylation. In total, 172 glycosylation sites (Ser/Thr) were identified compared to the previously reported 107 for native lubricin obtained with the same approach (38). The majority of the glycosylation sites of rhPRG4 were found on peptides that were repeated between 2-17 times in lubricin mucin domain. The dominant glycans found corresponded to simple core 1 type O-glycosylation whereas native lubricin also contained core 2. The difference in glycosylation sites is probably a reflection of differences in specificity and level of GalNAca1-polypeptide transferases. Human synovial cells are dominated by T1-, T2-, T5-and T15-GalNAc transferase (38), while CHO cell glycosylation predominantly express the hamster equivalent of human T2-transferase (51). We may also attribute differences in detected glycosylation sites to the increased glycosylation complexity of native lubricin compared to rhPRG4. In addition, the decreased complexity of glycans in rhPRG4 would cause decreased glycopeptide diversity leading to increased molar amounts and detectability of individual glycopeptides. In addition, the differences in the level of glycosylation could also contribute to differences in efficiency in the digestion of the mucin domain. Our glycoproteomic data shows that glycans on rhPRG4 are positioned in the same central domain as in synovial lubricin. Lubricin (both native and recombinant) contains one of the largest mucin domains where glycosylation sites have been mapped. In synovial lubricin, 37% of Ser/Thr-residues were glycosylated in the STP rich region aa 232-1056, whereas in rhPRG4, 60% of Ser/Thr-residues were found to be glycosylated. However, we need to state that the site localization relies on positive identification, so absence of detection does not mean that remaining sites are not glycosylated. Differences in glycosylation between native lubricin and rhPRG4 invoke a question about the impact of glycosylation on the N-and C-termini involved in interactions. Lubricin has, for example, been shown to bind non-covalently to COMP within aa105-202 and covalently by specific disulfide bonds between aa64-86 (19).
Glycosylation sites were identified just adjacent or within these regions (Supplementary Table 2).
The O-glycomic data shows that rhPRG4 glycosylation is dominated by the linear monosialylated core 1 structure, similar to native synovial lubricin. rhPRG4 as such contains the main molecular functions attributed to native lubricin. Although there is a reduction in disialylated structures, there is an overall increase in sialylation giving the rhPRG4 molecule a more negatively charged central domain similar to SF lubricin from control individuals rather than lubricin found in OA patients (Fig. 1). However, rhPRG4 does have a more limited range of glycans with no core 2 or extended core 1 structures present. These structures may be involved in other glycan-glycan interactions with the ECM, interacting with galectin-3 or involved in the immune system (14). Overall, it is clear that both the type and the specific pattern of glycosylation are quality attributes to the lubrication and binding function of this very heavily glycosylated protein. The data indicates that glyco-design could help to further improve functionality of rhPRG4. Active modulation of O-linked glycosylation for biologics is an unexplored area. As of yet, it is unknown whether these changes of lubricin in SF are deleterious to the function of the molecule in vivo. It is also unknown whether alterations in lubricin glycosylation are a biological response to OA conditions, where the change in negative charges is an attempt to alter the lubricating function of the expressed lubricin. This could be due to a metabolic response to simplify and maintain expression of a less energetically expensive molecule in the face of exogenous stress, or to potentiate downstream signaling through altered galectin-3 release. Given the complexity of timing and severity impacts on biological functions within the OA joint, it can be summarized that there are important alterations in the amount and type of glycosylation in a critical lubricating system, and that functional and inflammatory aspects of the disease are likely regulated by the attached sugars, all of which warrant additional research and focus for heretofore unexplored pathogenic mechanisms in a very serious disease.

EXPERIMENTAL PROCEDURES
Samples, human tissues and cells rhPRG4 expressed from CHO cells was purified as described elsewhere (2,52).
bPRG4 was purified using CsCl gradient and anion exchange chromatography, and purity confirmed with SDS-PAGE/western blot and mass spectrometry as described elsewhere (53). Recombinant galectin-3 used in all experiments unless stated elsewise was provided by Professor Haakon Leffler, Lund University, Sweden. SF samples from OA patients that were scheduled for total knee replacement (n=2, age 76 and 84 years old) for the glycopeptide analyses were collected during therapeutic joint aspiration at Sahlgrenska University Hospital (Gothenburg, Sweden). SF samples for O-glycomic analyses were collected from patients with symptomatic chronic knee OA requiring aspiration (OA patients n = 7, average age 56 years, range 34-72 years). The patients for O-glycomics were diagnosed as having knee OA by 2 sports medicine physicians following a review of the patient's symptoms, a physical examination, and plain-film radiography. Control samples (n = 7, average age 56 years, range 31-72 years old) were collected post mortem within 4 hr. Samples were collected from the knee by aspiration and clarified (3,000 g for 30 min at 4°C) and stored at -80°C. SF samples for galectin-3 studies were obtained from late-stage OA patients prior to total knee replacement, mean age 69 years (n=30, range 58-78 years old) at Sahlgrenska University Hospital and at Danderyds' Hospital (Stockholm, Sweden). Control SF samples were collected as described above, mean age 65 years (n=18, range 47-76 years old). All samples were collected after written consent from patients and controls and the procedure was approved by the Conjoint Health Research Ethics Board of the University of Calgary (REB15-0880) or by the Regional Ethical Review Board in Gothenburg (ethical application 172-15) and the study abide by the Declaration of Helsinki principles.

Release
and LC-MRM analysis of oligosaccharides from rhPRG4 and synovial lubricin Lubricin from SF of OA patients (n=7) and controls (n=7) were isolated via a multistage process. The same amount of SF (500 µL) was used for each sample to allow for consistent MRM analyses on the resultant released glycans. The acidic SF proteins were isolated from SF samples by DEAE chromatography as previously described (54). Samples were reduced and alkylated and separated by SDS-PAGE using NuPAGE 3-8% Tris-acetate gels to allow isolation of the lubricin. Gels were transferred to PVDF membrane by semi-dry transfer blotting, stained with Alcian blue and destained with methanol (55). The lubricin band was excised from each sample individually and the O-linked oligosaccharides were released by reductive βelimination (50 mM sodium hydroxide, 0.50 M sodium borohydride) overnight followed by cleanup on cation exchange columns (AG50WX8, BioRad, Hercules, CA, USA) in C18-ZipTips (Millipore) (56). Oligosaccharides with reducing end (alditols) were separated on porous graphitized carbon columns (PGC, 5 µm particles, Hypercarb, ThermoFisher Scientific, Waltham, MA, USA) prepared in-house, with the dimensions of 10 cm (length) and 250 µm (inner diameter). The gradient, after 5 min of 98% solvent A (10 mM ammonium bicarbonate), increased solvent B (80% acetonitrile in 10 mM ammonium bicarbonate) from 2% to 45% in 41 min, then to 95% solvent B in 4 min. Solvent B (95%) was held for 5 min before re-equilibration at 98% A for 35 min. The flow rate was kept at a constant 10 µL/min using an ekspert microLC 200 HPLC system (Eksigent, AB Sciex, Framingham, MA, USA). The oligosaccharides were analyzed on a QTRAP 6500 ESI-triple quadrupole linear ion trap hybrid mass spectrometer (AB Sciex) in negative ion, high mass mode. A turbo V ion source was used with a 25 µm electrode. An ion spray voltage of 4200 was used. CE, DP and CXP were optimized for each transition. The final method included transitions to quantify and identify all 24 identified structures on lubricin. All transition information has been published in detail previously (36).

Relative quantitation and statistical analyses of MRM data
MultiQuant 3.0 (Sciex) was used for MRM peak integration. Area under the curve was determined for all structures and combined for total glycan abundance. Each glycan was then represented as a relative percentage of the total glycan abundance. Statistical analyses, two-tailed t-tests, were carried out to compare the normal and OA synovial lubricin glycan relative abundances. GraphPad Prism 5.00 for windows (GraphPad Software, Inc, San Diego, CA, USA) was used to carry out statistical analyses and to draw graphs.

Release and analysis of oligosaccharides from bPRG4
Oligosaccharides from bPRG4 (60 μg) were released in 1.0 M NaBH4/0.10 M NaOH 50°C (100 μL) overnight, followed by neutralization with concentrated acetic acid, desalted with 150 μL cation exchange media AG50WX8 in 100mg C18 Strata SPE column (Phenomenex). The glycans with analyzed with LC/MS and MS/MS using PGC chromatography as described above, connected to an LTQ XL ion trap mass spectrometer (ThermoFisher Scientific) as described elswhere (57). Structures were assigned based on MS/MS fragmentation by comparison with fragment spectra available in UniCarb-DB(58) and -DR(59) and by manual interpretation aided by using GlycoWorkbench software(60,61). Assignment of structural data based on MS interpretation was recorded according to reporting guidelines (MIRAGE) for glycomics sample preparation (62) and MS (59,63).

Purification, digestion and LC/MS analysis of glycopeptides from rhPRG4
Aliquots of purified proteins from rhPRG4 were treated with or without glycosidases (sialidase A and/or O-glycanase), followed by SDS-PAGE (3-8% Tris acetate NuPage gels (Invitrogen, Stockholm, Sweden) and tryptic digestion, or digested in-solution with either trypsin or Lys-C as described previously (38). The peptides were desalted with C18 stagetips(64), followed by glycopeptide enrichment using HILIC tips prepared in-house using cottonwool (38,65) or commercially available columns (ThermoFisher Scientific). For LC/MS analyses, C18 columns were prepared in-house and used with formic acid: acetonitrile gradients. The glycopeptides were analyzed on an LTQ Orbitrap XL mass spectrometer (ThermoFisher Scientific) equipped with CID and ETD fragmentation as described (38).

Purification, digestion and LC/MS analysis of glycopeptides from SF lubricin from two OA patients
SF lubricin from patients with OA was purified as described elsewhere (54). Briefly, SF (0.5 mL) was subjected to anion exchange chromatography, the acidic protein fraction was collected and precipitated in ethanol. The acidic proteins were reduced in 10 mM DTT (70°C, 1 hour) and alkylated in 50 mM iodoacetamide, 30 min at room temperature in the dark. Salts were removed and buffer exchanged to 50 mM ammonium bicarbonate (Sigma-Aldrich) using 30 kDa cutoff filters (Merck Millipore, Billerica, MA), followed by lyophilization. The reduced and alkylated protein containing gel bands were digested with trypsin, and peptides extracted and desalted with StageTip C18 columns (64). Peptides were separated using in-house packed C18 columns at 200 nL/min using a 120 min gradient of 5-40% buffer B (A: 0.1% formic acid, B:0.1% formic acid, 80% acetonitrile). The column was connected to an Easy-nLC 1000 system (Thermo Fisher Scientific), a nanoelectrospray ion source and a Q-Exactive Hybrid Quadrupole-Orbitrap Mass Spectrometer (ThermoFisher Scientific). For full scan MS, the instrument was scanned m/z 350-2000, resolution 70000 (at m/z 200), AGC target 1x10 6 , max IT 120 ms, dynamic exclusion was auto or 20 s. The twelve most intense peaks (charge states 2, 3, 4) were selected for fragmentation. For MS/MS, resolution was set to 17500 (at m/z 200), AGC target to 5x10 5 , max IT 256 ms, and collision energy normalized collision energy (NCE) =30 with stepped NCE of 25%.

Quantification of PSGL-1/mIgG2b by ELISA
A sandwich ELISA method was used for the quantitative measurement of PSGL-1/mIgG2b concentrations in supernatants. The 96-well ELISA plates (Costar 3590; Corning Inc., Corning city, NY, USA) were coated with an affinity-purified polyclonal goat anti-mIgG Fcspecific antibody (Sigma-Aldrich) at a concentration of 10 µg/mL in 50 mM carbonate buffer, pH 9.6, at 4°C overnight. The plates were blocked with 1% (w/v) bovine serum albumin (BSA) in PBS to minimize non-specific binding. This buffer was also used as antibody dilution buffer. The plates were sealed and incubated at room temperature for 2 hours. Supernatants were incubated in the 96-well plate in triplicates for 2 hours before incubating with a peroxidaseconjugated anti-mIgG Fc-specific antibody (Sigma-Aldrich) diluted 1:2 000 for 2 hours at room temperature. PBS containing 0.05% (v/v) Tween-20 (Sigma-Aldrich) and 0.15 M sodium chloride was used as the wash buffer between incubations. A bound peroxidase-conjugated antibody was visualized with 3, 3′, 5, 5′tetramethylbenzidine dihydrochloride (Sigma-Aldrich) as a substrate. The color reaction was stopped by the addition of 2.0 M H2SO4 and the optical density of the samples was read at a wavelength of 450 nm in a microplate reader (Synergy 2, Biotek instruments Inc, Winooski, USA). The PSGL-1/mIgG2b concentration was estimated using a dilution series of purified mIgG2b (AbD Serotec, Oxford, UK) in blocking buffer as a standard.

PSGL-1 constructs analyzed by SDS-PAGE/Western blot
For sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE)/Western blotting, PSGL-1/mIgG2b was purified from clarified supernatants on goat anti-mIgG agarose beads by rotated at 4 °C overnight. The beads with captured fusion proteins were washed three times in PBS, re-suspended in 4×SDS sample buffer without reducing agent (Invitrogen), and incubated at 95°C for 5 min for protein denaturation. The samples were analyzed by SDS-PAGE using 3-8% Tris-acetate gradient gels and Tris-acetate SDS running buffer (Invitrogen). Precision protein standard (Hi-Mark, Invitrogen) was applied as reference for protein molecular weight determination. Separated proteins were blotted using iBlot (Invitrogen) onto nitrocellulose membranes. The membranes were blocked with 1x carbo-free Blocking solution (Vector laboratories, Burlingame, CA), which was also used for dilution of antibodies and lectins. A mouse antihuman CD162 was used to detect the N-terminal part of PSGL-1 (1:1000 dilution; BD Pharmingen, San Diego, CA, USA) together with an HRP-conjugated polyclonal goat anti-mouse IgG (Fab-specific, 1:10000 dilution; Sigma Aldrich) used as secondary antibody. Biotinylated galectin-3 (diluted 1:100) was used to detect galectin-3 together with streptavidin (diluted 1:10 000; Sigma Aldrich). Bound antibody and lectins were visualized by chemiluminescence using the ECL kit according to the manufacturer's instructions (GE Healthcare, Uppsala, Sweden). The presence of glycosylated and non-glycosylated proteins was assessed by staining of the gels using a glycoprotein staining kit and the Ruby stain (Thermo Fisher Scientific, San Jose, CA), respectively. Candy Cane markers (Thermo Fisher) was used for molecular weight reference. PSGL-contructs from transient transfection of C2GnT-1 of CHO-cells (69) was also subjected to the procedure of SDS-PAGE, blotting and staining as described above.
Lectin ELISA of Lubricin from SF SF Lubricin was probed using an in-house sandwich lectin ELISA method. 96-well Nunc-Immuno maxisorp plates (ThermoFisher Scientific) were coated with 1 µg/mL monoclonal anti-lubricin antibody clone 9G3 against rhPRG4 mucin domain (ThermoFisher Scientific) in PBS at 4°C over-night. After blocking with 3% BSA in TBS-Tween (TBS pH 7.4 + 0.05% Tween-20), same volume of SF samples (duplicate of each sample) that were pre-diluted 1 in 50 in assay buffer (1% BSA in TBS-Tween), were loaded (100 µL) and incubated for 90 minutes. The captured lubricin was incubated with 2 µg/mL biotinylated Helix Aspersa Agglutinin (HAA) (Sigma-Aldrich St. Louis, MO, USA), Peanut Agglutinin PNA (Vector laboratories), L-selectin(Sigma-Aldrich) Aleuria Aurantia Lectin (AAL) (Vector laboratories), Sambucus nigra Aggutinin (SNA) (Vector laboratories), or biotinylated galectin-3 for 1 hour, followed by a 1-hour incubation with 0.2 µg/mL horseradish peroxidase (HRP) conjugated streptavidin (Vector Laboratories). The plate was washed 3 times with TBS-Tween between each incubation procedure, and 5 times before the staining step. After staining with 1-Step™ Ultra TMB-ELISA Substrate Solution (ThermoFisher Scientific), absorbance was read at 450 nm. The ratios of the average absorbance of galectin-3 ad HAA staining for each individual sample were calculated to estimate the level of extension of GalNAca1-Ser/Thr (HAA) into Galectin-3 reactive glycans on lubricin. To test the glycodependency of galectin-3 binding to lubricin, 6 µL of SF was incubated with 40 µL of 1% periodic acid for 1 hour on shaker, and the system was then neutralized with 40 µL TBS for 20 minutes. Samples were diluted in assay buffer and subject to protocol as described above. Statistical evaluation of lectin sandwich ELISA as described in Figure 3 and Supplementary  Figure 4. Relative differences between OA and control of Galectin-3 binding was compared using the lectin ELISA absorbance ratio of Galectin-3 versus HAA.

Endogenous synovial galectin-3
Endogenous galectin-3 levels in OA and control SF were measured by the Galectin-3 test ELISA kit (BG Medicine Inc. Waltham, MA, US) according to manufacturer's instructions. OA samples were measured in duplicates and control samples were measured as singlets due to limited amounts of material. Statistical evaluation of ELISAs were performed by GraphPad Prism 8 (GraphPad Software, LLC, USA). Statistically significant differences were calculated by Nonparametric t-test using Mann-Whitney test.

Surface plasmon resonance (SPR)
Binding of rhPRG4 and bPRG4(53) to galectin-3 was assessed using a Biacore X100 SPR instrument (GE Healthcare, Pittsburg PA). Galectin-3 (R&D Systems Inc., Minneapolis, MN, USA) was covalently attached to the CM5 chip by running 50 mg/ml galectin-3 solution in 20 mM sodium acetate solution at pH 5 giving a response unit of 3500. Injection of lubricin was performed as described (70). Lactose was used as a regeneration buffer after lubricin injections (71).

ABBREVIATIONS
OA -osteoarthritis; RArheumatoid arthritis; rhPRG4recombinant lubricin; bPRG4bovine chondrocyte lubricin; SFsynovial fluid; CHO -Chinese hamster ovary; MRMmultiple reaction monitoring; SEM -standard error of the mean; Tn-antigen -GalNAcα1-; T-antigen -Galβ1-3GalNAcα1-; ECMextracellular matrix; COMP -cartilage oligomeric matrix protein; HCDhigher-energy collisional dissociation; CID -collision induced fragmentation; SPR -surface plasmon resonance; DEAE -diethylaminoethyl; PVDF -polyvinylidene fluoride; Hex -hexose; HexNAc -Nacetylhexosamine; Sulf -sulfate   Galectin-3 binding to lubricin from SF of late stage OA patients (n=30) was measured using a sandwich lectin ELISA assay (see Materials and Methods). Treatment of OA SF (n=8) with periodic acid shows that the binding to galectin-3 was glycosylation dependent. (B) Increase of extended structures binding to galectin-3 relative to un-extended GalNAca1-Ser/Thr (measured by HAA-lectin) present on lubricin from late stage OA patients (n=12, red) and control SF (n=13, blue). Ratio of the absorbances at 450 nm were calculated for each sample. Samples were measured in duplicates. (C) Endogenous galectin-3 in SF. Galectin-3 concentrations in SF were measured in samples from 23 late stage OA patients (red) and 14 controls (blue). Data in this figure is represented as median with interquartile range, and significant difference was calculated by non-parametric Mann-Whitney test with ***p<0.001.  ) that allows galectin-3 interaction and crosslinking of lubricin molecules. (B) In OA, lubricin with changed glycosylation due to an altered glycosylation machinery in synoviocytes and/or chondrocytes is still capable of adhering to the cartilage and provide boundary lubrication, but the decreased amount of galectin-3 and its decreased interaction with lubricin with lower amount of LacNAc destabilize the glycosylated articular surface.