Loss of α2-6 sialylation promotes the transformation of synovial fibroblasts into a pro-inflammatory phenotype in Rheumatoid Arthritis

In healthy joints, synovial fibroblasts (SFs) provide the microenvironment required to mediate homeostasis but are recognized to adopt a pathological role in rheumatoid arthritis (RA), promoting the infiltration and activation of immune cells to perpetuate local inflammation, pain and joint destruction. Carbohydrates (glycans) attached to cell surface proteins are fundamental regulators of cellular interactions between stromal and immune cells, but very little is known about the glycome of SFs or how glycosylation regulates their biology. Here we fill these gaps in our understanding of stromal guided pathophysiology by systematically mapping glycosylation pathways in healthy and arthritic SFs. We used a combination of transcriptomic and glycomic analysis to show that transformation of fibroblasts into pro-inflammatory cells in RA is associated with profound glycan remodeling, a process that involves reduction of α2-6 terminal sialylation that is mostly mediated by TNFα-dependent inhibition of the glycosyltransferase ST6Gal1. We also show that sialylation of SFs correlates with distinct disease stages and SFs functional subsets in both human RA and models of mouse arthritis. We propose that pro-inflammatory cytokines in the joint remodel the SF-glycome, transforming a regulatory tissue intended to preserve local homeostasis, into an under-sialylated and highly pro-inflammatory microenvironment that contributes to an amplificatory inflammatory network that perpetuates chronic inflammation. These results highlight the importance of cell glycosylation in stromal immunology.


Abstract:
In healthy joints, synovial fibroblasts (SFs) provide the microenvironment required to mediate homeostasis but are recognized to adopt a pathological role in rheumatoid arthritis (RA), promoting the infiltration and activation of immune cells to perpetuate local inflammation, pain and joint destruction.
Carbohydrates (glycans) attached to cell surface proteins are fundamental regulators of cellular interactions between stromal and immune cells, but very little is known about the glycome of SFs or how glycosylation regulates their biology. Here we fill these gaps in our understanding of stromal guided pathophysiology by systematically mapping glycosylation pathways in healthy and arthritic SFs. We used a combination of transcriptomic and glycomic analysis to show that transformation of fibroblasts into pro-inflammatory cells in RA is associated with profound glycan remodeling, a process that involves reduction of a2-6 terminal sialylation that is mostly mediated by TNFa-dependent inhibition of the glycosyltransferase ST6Gal1. We also show that sialylation of SFs correlates with distinct disease stages and SFs functional subsets in both human RA and models of mouse arthritis. We propose that pro-inflammatory cytokines in the joint remodel the SF-glycome, transforming a regulatory tissue intended to preserve local homeostasis, into an under-sialylated and highly pro-inflammatory microenvironment that contributes to an amplificatory inflammatory network that perpetuates chronic inflammation. These results highlight the importance of cell glycosylation in stromal immunology. Not surprisingly, biologic Disease Modifying Anti-Rheumatic Drugs (bDMARS) inhibiting these cytokines are the treatments of choice in the clinic. Even though such gold standard treatments lead to a substantial improvement in life quality of thousands of patients 1,4 , as they are immunosuppressive agents they have been linked with serious adverse effects. Also 30-40% of RA patients still do not respond completely to them, suggesting that key events underpinning pathogenesis remain elusive. In fact, the cellular and molecular basis of why inflammation does not resolve in RA remains unanswered.
Synovial fibroblasts (SFs). are major components of the synovial membrane, a highly specialized mesenchymal tissue lining the joint cavity. In the synovium, two main micro domains can be described; the lining layer (directly exposed to the synovial space) and sublining layers. Due to their anatomical location, SFs provide the required nutritional and structural support. Initially SFs were considered as cells lacking any substantial impact on immune function. However SFs are known to adopt a immunopathological role in RA, responding to inflammatory cytokines and promoting bone and cartilage destruction 2,3,5,6 . Despite being cells of non-immune origin, SFs play a central role to perpetuate local immune responses in the synovial joint, delivering region-specific signals to infiltrating immune cells and contributing to bone and cartilage destruction 7,8 . Interestingly, it has been recently demonstrated by Single Cell Transcriptomics that SFs comprise distinct functional subsets that correlate with their anatomical location and activation of pathological pathways 9,10 , suggesting that SF-dependent immunity may be far more complex than anticipated. Because of their non-immune origin and their highly-specialized function in health and disease, interventions targeting SFs -or specific SFs subsetsmay modify disease progression without significant immunosuppression.
Functional glycomics, an emerging discipline focused on defining the structures and functional roles of carbohydrates (glycans) in biological systems, could offer such fibroblast specific molecular targets 11,12 . Some elegant studies have shown the potential impact of glycan regulation in multiple aspects of RA pathophysiology, for example reduced sialylation of N-glycans is a feature of pathogenic immunoglobulins in RA patients and their glycosylation profile shows predictive potential for disease 4 progression [13][14][15][16][17] . Besides, galectins, a family of proteins that bind to galactose-containing glycans, are key modulators of synovial inflammation [18][19][20] . However, little is known about the glycosylation profile of SFs, and whether this varies in health and disease, despite the glycome is being critical for dictating interactions with galectins and other carbohydrate binding proteins, and the fact that altered glycosylation is a hallmark of chronic inflammatory conditions. To illustrate this, cytokines induce aberrant glycome changes in cancer that perpetuate local inflammation via control of cell adhesion, migration and signal transduction [21][22][23] ; mechanisms that are also associated to the pathogenicity of SFs in RA and are responsible of migration to cartilage-containing areas prior to tissue damage. Furthermore, the carbohydrate binding protein galectin-3 is up-regulated in RA 24 and galectin-3 -/mice show reduced pathology in experimental arthritis 20 . Moreover, exogenous galectin-3 significantly up-regulates CCL2, CCL3 and CCL5 in synovial but not in dermal fibroblasts 19 , suggesting that the synovium microenvironment can induce tissue-specific glycosylation in the stromal compartment.
We hypothesised that the cytokine milieu found in the inflamed joint determines distinct SFglycosylation, that in turn, regulates cell recruitment and inflammatory responses. Our aim in this study was to investigate changes of SF glycome that could be related to their inflammatory activity. By combining transcriptomics and glycomics analysis, we report that transformation of SFs into proinflammatory cells in RA is associated with glycan remodeling, involving reduction of terminal sialylation in Thy1(CD90) + sublining SFs upon TNFa stimulation. We also show that low sialylation of SFs is associated with disease remission, supporting the idea that the stromal glycome could be used for development of novel disease biomarkers or therapeutic agents.

Results.
Distinct anatomical locations and inflammatory conditions shape the glycosylation profile of human fibroblasts. To test our initial hypothesis, we isolated and expanded fibroblasts from the synovium and matched samples from skin from RA patients, as a reflection of inflammatory and non-5 inflammatory environments. We also isolated SFs from osteoarthritis (OA) patients, as an example of a less inflammatory, more destructive joint disease. General glycosylation was evaluated using lectinbinding assays by immunofluorescence (Fig. 1a) and flow cytometry (Fig. 1b). Based on the carbohydrate binding specificity for each lectin, our results provided an initial indication of differential fibroblast glycosylation under inflammatory conditions. All fibroblasts bound most of the tested lectins, confirming the presence of a rich and diverse glycocalix. These data indicate the SF-glycome was rich in galactose-containing glycans (RCA + , ECA + ) containing PolyLacNAc extensions (LEL + ) and α1-6 core fucosylated N-glycans (AAL + ), in contrast to the lack of α1,2 fucosylation on glycan antennae (UEA -). Interestingly, we observed significant differences in glycosylation in fibroblasts taken from distinct anatomical locations (dermal vs synovium: PNA, Jacaline and ECA binding) and in distinct inflammatory conditions (RA vs OA, LEL and ECA binding) ( Fig. 1a-b), supporting the link between local inflammatory mediators and glycan remodeling. To evaluate whether changes in glycosylation were due to cytokine stimulation, we checked the expression of glycan building enzymes, glycosyltransferases, in human SFs in response to pro-inflammatory TNFα [dataset generated by Slowikowski et al. 25 ]. TNFα significantly modulated glycosyltransferase genes involved in the synthesis of branched glycans such as A4GALT, GCNT1 and GCNT2, along with up-regulation of α2-3 sialyltransferases like ST3Gal1, ST3Gal2 and ST3Gal4 (Supp. Fig. 1a). No effect was observed when cells were stimulated with IL-17 (Suppl. Fig. 1b), indicating that changes in glycosylation are linked to individual cytokine signaling. To further explore these findings, naive SFs expanded from mouse synovium were stimulated with a panel of regulatory and pro-inflammatory factors and subsequently stained with PHA, lectin that binds complex branched N-glycans.
The cytokine IL-22 was chosen because it exerts both inflammatory and resolving actions in RA, depending on the local microenvironment 26 . Differential effects of IL-22 over cell glycome would therefore provide further support to the original hypothesis linking distinct glycosylation and effector responses. The combination of inflammatory factors with IL-22 exerted diverse effects on expression of branched glycans (Suppl. Fig. 1c), perhaps related to the dual effect described by IL-22 in joint 6 inflammation. Changes in PHA-binding in response to immunomodulators might be reflective of SFs activation as glycans recognized by PHA were detected in joint areas with a strong SF-mediated inflammation (Suppl. Fig. 1d). Overall, results shown in Fig. 1 and Suppl. Fig. 1 supported the hypothesis that immune mediators found in local microenvironments control SF glycosylation.
Comparative RNA-Seq analysis reveals that transformation of SFs into pathogenic mediators is associated with changes glycosylation pathways. To complement the lectin-based experiments we additionally undertook transcriptome sequencing (RNA-Seq) to fully dissect SFs glycosylation pathways at mRNA level. We also chose to work initially with the murine model of Collagen-Induced Arthritis (CIA), because human clinical samples have usually been exposed to long-term immunomodulatory treatments, which add high levels of variability and might affect the cytokine- KEGG pathway enrichment analysis showed that 'Rheumatoid Arthritis' as disease pathway was significantly up-regulated in CIA SFs (Suppl. Fig. 2b), validating the model to study SF-dependent inflammation. String Protein-Protein Interaction Networks Functional Enrichment Analysis 27 was applied to DE genes in CIA compared to healthy SFs, identifying 2 main functional networks: i) cell cycle and cell division and ii) inflammatory response (Fig. 2c), further demonstrating SF immune activation and hyperproliferation. Interestingly, GO-term analysis revealed that most proteins identified in the inflammatory network were glycoproteins and/or regulators of cell communication (Fig. 2c), 7 suggesting the potential role of glycosylation in SF activation. To further study this, we specifically searched for expression of genes involved at different steps of glycosylation biosynthesis, such as glycosyltransferases and glycosidases responsible of mannosylation, glycan chain branching and elongation, fucosylation, sialylation and glycan degradation (Fig. 2d). Unsupervised clustering based on these glycan biosynthetic pathways separated naïve and arthritic SFs (Fig. 2e), suggesting that proinflammatory SFs present an altered glycome in inflammatory conditions. The observed down- Glycomic studies reveal reduced levels of sialylation in pro-inflammatory SFs. Transcriptomic analysis provided a powerful tool to delineate potential changes in cell glycosylation. However, unlike proteins or nucleic acids complex, glycans are not assembled in a template-driven process. Rather, glycosylation in the endoplasmic reticulum and Golgi is the result of combined actions of glycosyltransferases and glycosidases (Fig. 2d,e). Consequently, prediction of structures based on transcriptomic data does not necessarily correlate with the final glycosylation profile and further structural information is needed to generate reliable information. We therefore used mass spectrometry (MS) based glycomics to define the N-glycome of murine SFs (Suppl. Fig. 3). N-glycans were isolated from cultured SFs, permethylated and subjected to MS analysis. Following annotation of peaks with most likely glycan structures based on molecular ion composition, knowledge of biosynthetic pathways and with the assistance of the bioinformatic tool glycoworkbench 28 , most structures were annotated as high mannose glycans or complex glycans, either core-fucosylated or non-fucosylated. Sequential addition of N-acetyllactosamines (LacNAcs) defined the larger structures, suggesting the presence of extended antennae and multi-branched structures. Sialylation was the most abundant capping modification of terminal galactoses. As expected in murine cells, two sialic acids were detected, Nacetylneuraminic and N-glycolylneuraminic acid. A similar structural profile was observed in human SFs taken from RA joint replacements (Suppl. Fig. 4), further validating the animal model. Still, some differences between mouse and human cells were found, like the absence of terminal Gal-α-gal and Nglycolylneuraminic acid, as these structures cannot be synthesized by the human glycosylation machinery. Nevertheless, human and mouse SF N-glycomes seemed to be well conserved.
Next, in order to identify specific glycan changes that could contribute to SF activation, the relative expression of individual N-glycans structures in healthy and CIA SFs were compared. Structures whose relative expression was below 2% of total were ruled out, since the low expressed forms could easily add artefactual results, leaving 43 glycans to analyze. This dataset was then used to perform unsupervised hierarchical clustering to uncover expression patterns characteristic of inflammatory CIA SFs (Fig. 3). Interestingly, N-glycans were clustered into 6 discrete groups containing similar structural features: i) cluster 1987: high mannose, ii) cluster 2285: LacNAc extended, iii) cluster 3026: sialylatedfucosylated, iv) cluster 3271: sialylated-LacNAc v) cluster 3258: sialylated-LacNAc-fucosylated, and vi) cluster 2069: simple non-sialylated. Three of these clusters contained glycans that were significantly down-regulated in CIA SFs: clusters 3026, 3271 and 3258, which had one characteristic in common: a significant high proportion of sialic acid containing glycans. In fact, 96% of all the sialylated glycans analyzed were found within these three clusters, strongly suggesting that reduced sialylation is associated with activated CIA SFs.
SFs from arthritic mice exhibit down-regulated a2-6-linked sialic acid. Comparative glycomic analysis ( Fig. 3) indicated that experimental arthritis was strongly associated with a down-regulation in sialic-acid-containing N-glycans in SFs. Retrospective examination of the data related to transcriptomic regulation reveals a significant down-regulation of ST3Gal6, ST3Gal2 and ST6Gal1 [fold-increase 0.73, 9 0.64 and 0.60 respectively] (Fig. 2e), supporting the reduced presence of sialylated glycans observed by MS (Fig. 3). However, it also showed a significant up-regulation of ST3Gal4 mRNA [1.49 fold-increase ( Fig. 2e)]. These apparently conflicting results could be explained by differential regulation of sialic acid linkages, differences that would not be detected in our initial MS-based studies. In the structures depicted in Fig.ure 3, two major types of glycosidic bonds can be formed between sialic acid and terminal galactose, namely, sialic acid-α2-3Gal and sialic acid-α2-6Gal, synthesized by six ST3 betagalactoside alpha-2,3-sialyltransferases (ST3Gal1-6) and two ST6 beta-galactoside alpha-2,6sialyltransferases (ST6Gal1-2). To evaluate the type of sialylation present in arthritic SFs, we used SNA and MAAII, lectins that specifically recognize sialic acid in a2-6 and a2-3-linkages. Results showed that SNA binding was reduced in CIA SFs compared to naïve SFs, whereas MAAII binding was not affected (Fig. 4a). These results indicate that the differential sialylation profile observed in CIA SFs is due to specific reduction in a2-6 sialylation, presumably because of lower ST6-sialyltransferases expression. Furthermore, binding of galactose-recognizing lectins such as PNA and SBA was upregulated in arthritic SFs (Fig. 4a), probably reflecting increased terminal galactose in the undersialylated glycome. Since inflammation does not change a2-3-sialylation of SFs, a reduced ratio of a2,6/a2,3-linked sialic acid might constitute a hallmark of inflammatory SFs. Supporting these findings, synovial membranes of healthy mouse joints had a a2-6 > a2-3 sialylation profile, as observed by immunofluorescence with SNA and MAAII, whilst inflamed joints in CIA joints show comparable levels of both sialic acid linkages (Fig. 4b).
Down-regulation of a2-6 sialylation in SFs is a consequence of TNFa-dependent inhibition of the sialyltransferase ST6Gal1. The results from transcriptomics, MS-based glycomics and lectin-binding experiments all concluded that reduced a2-6-sialylation is associated with inflammatory SFs. We therefore next sought to identify the molecular mechanisms underlying this shift in the SF-glycome.
Several sialytransferases (Sia-Ts) are expressed in SFs (Fig. 5a), including ST3Gal enzymes [ST3Gal1>>ST3Gal2>ST3Gal4>ST3Gal3] and only one involved in a2-6-Sialylation, ST6Gal1, with some significant gene regulation during CIA as explained before. ST8Sia-T involved in polysialic synthesis were only marginally expressed, in agreement with MS glycomics data (Suppl. Fig. 3). In addition to Sia-Ts, we studied expression of enzymes synthesizing their substrate, CMP-Neu5Ac ( Fig.   5a), since a reduction in its intracellular concentration would also lead to hyposialylated glycomes, even if the expression of Sia-Ts remained at normal levels. As expected, given the abundant sialic acid content of SFs glycoproteins, these enzymes were highly expressed in both healthy and CIA SFs (Fig. 5a), suggesting that intracellular availability of Sialic acid donors might not restrain glycan biosynthesis.
However, and rather unexpectedly, neither RNA-Seq analysis nor qPCR approaches detected expression of N-Acetylneuraminic Acid Phosphatase (NANP), an enzyme that dephosphorylates sialic acid 9phosphate to free sialic acid as part of the required intermediate biosynthetic metabolites. As sialylated glycans are undeniably present in SFs, alternative biosynthesis pathways might be operating in these cells, perhaps as in recent observations in CHO cells 29 . To identify possible mechanisms behind changes in sialylation, we used qPCR to evaluate expression of selected genes (as in Fig. 5a) in response to IL-1b, IL-17 and TNFa, key cytokines known to initiate inflammatory pathways in RA. IL-6 mRNA, included as a positive control, was significantly up-regulated in response to all cytokines tested (Fig.   5b), confirming cell activation. No difference was observed in GNE, NANS, CMAS or SLC35A1 expression, corroborating our findings that sialic acid biosynthesis is not the cause for the reduced sialylation observed in CIA SFs. Regarding SiaTs expression, no effect was seen in response to IL-17, in line with the human observations in Suppl. Fig. 2b. IL-1b increased expression of ST6Gal1 (in Naïve SFs) and ST3Gal4 (in CIA SFs), with an approximately 2-fold increase. However, these changes were only mild compared with the significant down-regulation of ST6Gal1 in response to TNFa, that induced a strong down-regulation of ST6Gal1 (8-fold) in both naïve and CIA SFs (Fig. 5b). Since ST6Gal2 was not detected in SFs, TNFa-mediated down-regulation of ST6Gal1 seems to a key mechanism responsible for reduced SF sialylation in CIA arthritic mice. To confirm the functional effect over the SF-glycome, we repeated the experiment evaluating in parallel the expression of a2-6-sialic acid (SNA + binding). Again, only TNFa decreased ST6Gal1 mRNA and such effect was accompanied by reduction of a2-6-Sialic acid on the cell surface (Fig. 5c). Finally, cells were stimulated with TNFa for 48 hours to allow glycan turnover, and N-glycans were then isolated to conduct MS-based glycomic analysis. 12 out of 16 sialylated N-glycans showed a reduced expression upon TNFa stimulation, providing further support to our findings (Fig. 5d).

Fibroblast populations are differentially sialylated in homeostatic and inflammatory conditions.
The discovery of distinct subsets of SFs in different microdomains of the synovium has revolutionized our understanding of fibroblast biology in RA. Single cell transcriptomic experiments have shown that fibroblast subsets localize to specific regions in the synovium contributing to different aspects of disease pathogenesis 9,10 . For example, CD90+FAPα+ fibroblasts located in the synovial sub-lining are essential for perpetuation of the inflammatory response, whereas CD90-FAPα+ in the lining membrane are responsible for bone and cartilage damage 9 . Understanding which functional SF subset(s) lose sialic acid during inflammation would allow us to infer the dominant functional relevance of this glycan modification in SF-mediated pathogenesis. Therefore, we evaluated expression of ST6Gal1 and α2-6-Sialylation in CD90-and CD90+ SFs, representative of lining and sublining SFs respectively. As described previously, SNA and MAAII lectins were used to detect α2-6 and α2-3 sialic acid in SFs directly isolated from digested mouse synovium. PNA was used as a control, because sialic acid prevents its binding. SFs from healthy joints showed higher affinity for SNA than MAAII, supporting the homeostatic role of α2-6 sialylation. Interestingly, naïve, unstimulated CD90+ SFs had a higher basal α2-6 sialylation than CD90-SFs (Fig. 6a), perhaps suggesting that sialic acid content is related to functional SF specialization dependent on their anatomical location. We therefore FACS sorted CD90+ and CD90-SFs from naïve and CIA joints and levels of ST6Gal1, IL-6 and MMP3 mRNA were evaluated in both subsets (Fig. 6b). As expected, and further corroborating our results, ST6Gal1 was down-regulated in CIA SFs, but interestingly, ST6Gal1 was differentially down-regulated only in CD90+ SFs (Fig. 6b). MMP3 and IL-6 were also differentially expressed in both subsets (Fig. 6b), upregulated during arthritis and preferentially expressed by CD90-SFs, consistent with findings from Croft et al 9 and further demonstrating that CD90 discriminates functional subsets of SFs.
ST6Gal1 mediated sialylation of the SFs glycome is associated with disease remission. CD90+ SFs are differentially found in the sublining layer synovium, and are expanded in RA synovia immune responses 9,10,30 . As we observed that CD90+ SFs downregulate ST6Gal1 in inflammatory conditions, we postulated that a low sialylated SF-glycome would be a characteristic signature of environments undergoing strong immune reactions, as a consequence of high TNFα concentration in the synovial space. To test this hypothesis, we compared the N-glycome of SFs expanded from CIA animals, separating them into cells from non-affected paws (low scores, LS-SFs) and very inflamed paws (high scores, HS-SFs), taking advantage of the CIA model being asymmetrical, where not all limbs develop the same degree of inflammation. In addition, we included SFs from naïve non-arthritic mice.
Unsupervised clustering of N-glycan relative expression, assessed by MS as before, revealed that one group of N-glycans that was strongly overexpressed in LS-SFs (cluster 3619, Fig. 7a). Strikingly, sialic acid was present in 11 out of 13 structures found within this cluster, consistent with idea that highly sialylated SFs is linked to a non-inflammatory phenotype. By contrast, cluster 2080 including almost exclusively non-sialylated glycans was down-regulated (Fig. 7a). Because α2-6-linked sialic acid is a blocker for galectin binding and function 31,32 , these findings suggested that a hypo-sialylated glycome, present in CD90+ SFs would be more sensitive to interactions with galectins by virtue of a higher exposure of galactosides, their natural ligand, being. Thus, we decided to stimulate naïve, LS-and HS-SFs (as in Fig. 7a) with recombinant galectin-3, described as a pro-inflammatory factor in RA 33 , and IL-17 and TNFα as inflammatory mediators that do not bind glycans. Expression of IL-6 was used as a positive control to confirm SF activation, since it is a cytokine produced by SFs in response to the inflammatory cytokines used 19 . Indeed, all SF groups show some ability to respond to IL-17 and TNFα, being their response in direct correlation to their inflammatory status (Fig. 7b-c). However, LS-SFs did not respond to galectin-3, in clear contrast to HS-SFs (Fig. 7c) which were easily activated. This might be explained by the protective coating of sialic acid in the LS-SFs-glycome compared to HS-SFs, preventing galectin-3 binding and consequent inflammatory response. Nevertheless, LS-SFs still expressed elevated IL-6 expression compared with naïve SFs, and their glycome was significantly 13 different to the naïve cells indicating that we cannot rule out that highly sialylated LS-SFs are in transition to become more inflammatory.
Hence, to assess whether high content of sialic acid is a protective marker, we used SFs expanded from either untreated early RA patients (<12 months from joint symptoms beginning), or RA patients in sustained remission under c-DMARDs+TNF-inhibitor combination therapy (Fig. 8). These samples were obtained from US-guided minimally invasive synovial tissue biopsies, which limited the amount of biological material and conditioned the coverage of MS-based studies. Nevertheless, we could still detect the most abundant structures, including high mannose and core fucosylated and non-fucosylated biantennary N-glycans. The ratio of mono and bi-sialylated versus their corresponding non-sialylated core provided us with the degree of sialylation in both clinical groups. Results showed that RA patients in sustained remission had a higher ratio of sialic-acid containing N-glycans, differences that were not seen when other non-sialylated structures were compared, such as ratios of fucosylated versus nonfucosylated or different mannose-containing glycans, suggesting that high sialic acid content is a distinctive feature of SF-glycome in RA patients in sustained remission (Fig. 8a). In line with this, SNA binding revealed a higher expression of a2-6 sialic acid in the synovium of RA patients in remission after c-DMARDs + TNF-inhibitor treatment (Fig. 8b) and in less inflammatory OA, supporting our conclusion of sialylation being an anti-inflammatory factor not only in the mouse model, but also in human RA. Therefore, it is likely that a better understanding of SF-glycobiology aids the identification of novel glycan-based therapeutics to target SFs, mirroring recent advances in cancer and opening new therapeutic opportunities for patients that are refractory to immunosuppressive drugs.

Discussion
We combined transcriptomic and MS-based glycomic analysis to define changes in SF-glycome with unprecedented detail in synovial inflammation. Our results show that N-glycome from healthy SFs comprises LacNAc-containing structures with high levels of core fucosylation and terminal sialylation.
Thus, because of a desialylated glycome, galectin-3 could induce secretion of inflammatory mediators such as IL-6 or Ccl2 in SFs, increasing local concentrations of TNFα that would further reduce ST6Gal-1 and α2-6 sialylation, establishing an autocrine pro-inflammatory loop that could explain perpetuation of disease in a similar fashion shown by members of the IL6 family 43 . By contrast, healthy synovium shows higher levels of α2-6-Sialylation in SFs, perhaps masking LacNAc repeats and preventing inflammatory actions of galectins. Thus, we propose that sialylation might be a homeostatic mechanism that is lost during RA progression because of TNFα overexposure.
To understand the physiological role of sialic acid in SFs, the role of carbohydrate binding proteins (CBPs) requires consideration. CBPs are expressed either in SFs, or in other immune cells, promoting trans or cis interactions. Siglecs (sialic acid-binding immunoglobulin-type lectins) are a family of immune regulatory proteins primarily found on hematopoietic cells. Consistent with our proposed homeostatic role for sialic acid in the synovium, Siglec-9 protected mice against experimental arthritis, although authors reported that it had no effect on SFs 44 . Instead, Siglec-9 inhibited NF-kB activation in human RA macrophages. However, effects on SFs cannot be ruled out, as following cytokine stimulation during disease, RASFs might already have a reduced sialylation profile which would make them unresponsive to Siglec-mediated actions. It might therefore be possible that siglecs mediate regulatory effects in synovial self-tolerance under healthy conditions, or at very early disease stages. The mechanism could involve Siglec-sialic acid trans interactions between SF and immune cells. In line with this, B cells lacking Siglec-2 (CD22) and Siglec-G develop spontaneous autoimmunity 45 and Siglec-G -/lupus-prone MRL/lpr mice exhibit increased severity and early onset of arthritis 46 . Besides, most siglecs (2,3,5,6-11) show inhibitory effects on TLR-dependent activation and mediate immunosuppression in the tumor microenvironment because of local hypersialylation 47 . By contrast, Siglec-1 (sialoadhesin) shows pro-inflammatory actions, it is up-regulated in activated macrophages in RA 48 and suppresses Tregs 49 . Interestingly, the anti-inflammatory Siglec-2 has predilection for α2-6 sialic acid, but the pro-inflammatory Siglec-1 binds preferentially to α2-3 50 , thereby supporting our conclusion that pro-inflammatory SFs diminish α2-6 sialylation but not α2-3. Thus, homeostatic α2- These cellular networks will be further defined by inclusion of additional immune cells in the equation, but also by the SFs anatomical and functional heterogeneity that has been recently revealed 10,30 . Distinct CD90+ SFs subsets found in the sublining synovial areas are expanded in RA synovium and are indeed the main contributors to inflammatory responses, results that have been confirmed in animal models 9 .
Our results in the mouse model indicate that only sublining CD90+ SFs reduce ST6-Gal1 expression, compared to the lining CD90-SF suggesting that sialic acid would exert different effects on the SFs subsets interacting with other immune cells, which could offer potential pathways to modulate specific immune networks in the synovium. We can also hypothesize that these glycan-dependent interactions could explain the observed disease tissue heterogeneity described in RA patients, including lymphoid, myeloid and fibrotic phenotypes 55,56 . Because synovial phenotypes observed in RA have been associated to distinct pathways (myeloid-IL-1β/TNF, lymphoid-IL-17 55 , it would be relevant to study ST6Gal1 expression to assess whether pro-inflammatory sialylation correlate with synovium RA phenotypes.
In conclusion, in the present report, we revealed that the reduction of α2,6-linked terminal sialylation constitutes a signature of the inflammatory, vs damage status of synovial fibroblasts. This altered phenotype seems to be induced by TNFα, in contrast with IL-1 or IL-17 that had           RNA-Seq preparation, sequencing and data analysis. Total RNA from sorted synovial fibroblasts was isolated immediately post sorting using RNeasy Micro kit (Qiagen, Germany). RNA integrity was checked with the Agilent 2100 Bioanalyzer System. All purified RNA had a RIN value >9. Libraries were prepared using the TruSeq mRNA stranded library preparation method. Samples were sequenced 2x75bp to an average of more than 30 million reads. Kallisto was used to quantify the transcript abundance of RNA-seq reads. All RNAseq reads were then aligned to mouse reference genome (GRCM38) using Hisat2 version 2.1.0. Featurecounts version 1.4.6 was used to quantify reads counts.
Data quality control, non-expressed gene filtering, median ratio normalization (MRN) implemented in DESeq2 package and identification of differentially expressed (DE) genes were done using the R bioconductor project DEbrowser 62 . Genes that passed a threshold of padj < 0.01 and log2foldChange > 2 in DE analysis were considered for further analysis. Gene Ontology (GO) enrichment, KEGG pathway enrichment and UniProt Keywords enrichment were performed in String (https://string-db.org) based on statistically significantly DE genes.

Mass spectrometric analysis of synovial fibroblasts N-glycans and unsupervised hierarchical clustering.
Synovial fibroblasts cells were scraped off tissue culture plates and suspended in iced-cold ultrapure water before homogenization and sonication were performed. Cells protein extract was precipitated in a methanol/chloroform extraction, Cell extracts were reduced and carboxymethylated, using Dithiothreitol and Iodoacetic acid, and then treated with trypsin. The treated samples were purified using a C18 cartridge (Oasis HLB Plus Waters) prior to the release of N-glycans by PNGase F (recombinant from Escherichia coli, Roche) digestion. Released N-glycans were permethylated and then purified using a Sep-Pak C18 cartridge (Waters) prior to MS analysis. The resulting enzyme-treated samples were lyophilized and permethylated prior to MS analysis. Purified permethylated glycans were dissolved in 27 10 µl methanol and 1 µl of the sample was mixed with 1 µl of matrix, 20 mg/ml 2,5-dihydroxybenzoic acid (DHB) in 70% (v/v) aqueous methanol and loaded on to a metal target plate. 4800 MALDI-TOF/TOF mass spectrometer (AB SCIEX) was run in the reflectron positive ion mode to acquire data.
MS spectra were annotated manually with the assistance of the glycobioinformatics tool GlycoWorkBench (GWB). All N-glycans were assumed to have a Manα1-6(Manα1-3)Manβ1- CarboFree solution (Vector Laboratories) was used to block non-specific interactions, following incubation with 1 µg/mL of biotinylated lectin (Vector Laboratories, Burlingame), for 30 min. Cells were washed with PBS and incubated with HRP conjugated streptavidin for 20 minutes, and washed again with PBS-Tween. A reaction was induced in the cells with developing solution, consisting of 1 mg/mL p-nitrophenyl phosphate in 0.5 mmol/L. The reaction was allowed to proceed in the dark and plates were read at 405 nm using a microplate spectrophotometer. Interleukin-6 (IL-6) expression was measured by ELISA in SF-supernatants according to the manufacturer's instructions (BD Biosciences, Oxford, UK).