Glycosylation-dependent galectin-1/neuropilin-1 interactions promote liver fibrosis through activation of TGF-β- and PDGF-like signals in hepatic stellate cells

Concomitant expressions of glycan-binding proteins and their bound glycans regulate many pathophysiologic processes, but this issue has not been addressed in liver fibrosis. Activation of hepatic stellate cells (HSCs) is a rate-limiting step in liver fibrosis and is an important target for liver fibrosis therapy. We previously reported that galectin (Gal)-1, a β-galactoside-binding protein, regulates myofibroblast homeostasis in oral carcinoma and wound healing, but the role of Gal-1 in HSC migration and activation is unclear. Herein, we report that Gal-1 and its bound glycans were highly expressed in fibrotic livers and activated HSCs. The cell-surface glycome of activated HSCs facilitated Gal-1 binding, which upon recognition of the N-glycans on neuropilin (NRP)-1, activated platelet-derived growth factor (PDGF)- and transforming growth factor (TGF)-β-like signals to promote HSC migration and activation. In addition, blocking endogenous Gal-1 expression suppressed PDGF- and TGF-β1-induced signaling, migration, and gene expression in HSCs. Methionine and choline-deficient diet (MCD)-induced collagen deposition and HSC activation were attenuated in Gal-1-null mice compared to wild-type mice. In summary, we concluded that glycosylation-dependent Gal-1/NRP-1 interactions activate TGF-β and PDGF-like signaling to promote the migration and activation of HSCs. Therefore, targeting Gal-1/NRP-1 interactions could be developed into liver fibrosis therapy.


Results
Galectin-1 and its bound glycans are concordantly highly expressed in fibrotic livers and activated HSCs. We first examined whether Gal-1 expression is associated with liver fibrosis and HSC activation using experimental models of liver fibrosis. Gal-1 expression was upregulated in fibrotic livers which were induced by thioacetamide (TAA), carbon tetrachloride (CCl 4 ), and a methionine-and choline-deficient (MCD) diet (Fig. 1A). The serum Gal-1 concentrations of fibrotic livers were not significantly changed ( Supplementary  Fig. S1). IHC and immunofluorescence staining revealed that strong Gal-1 staining was spatially associated with dense collagen deposition and α-smooth muscle actin (α-SMA) expression in areas around the portal vein and areas with bridging fibrosis, suggesting that Gal-1 may regulate HSC activation (Fig. 1B). Gal-1 was also highly expressed in livers of patients with cirrhosis (Fig. 1C). Notably, two patterns of Gal-1 staining were observed: (1) Gal-1 is up-regulated in non-parenchymal regions (p't 1). (2) Gal-1 is up-regulated in both non-parenchymal and parenchymal regions (p't 2, 3). Immunofluorescence staining showed that Gal-1 expression correlated and co-localized with α-SMA in both patterns (Supplemental Fig. 2), indicating Gal-1 is not only highly expressed in activated HSCs but also hepatocytes. Therefore, it is believed that Gal-1 is commonly up-regulated in activated HSCs but the overexpression of Gal-1 in hepatocytes may reflect the results of long-term exposure of liver damages and the complexity of etiologies. If the prolonged liver damages continue for years, mouse livers may show a similar pattern. To understand whether cell-surface glycans in activated HSCs favor Gal-1 binding, we examined the"glycosylation signature" of LX-2 cells (immortalized and activated HSCs) using a panel of lectins that recognize specific glycan structures. N-Acetyllactosamine (lacNAc) is the minimal structure recognized by Gal-1 and can be presented as multiple units (poly-lacNAc) on N-and O-linked glycans on the cell surface through the coordinated actions of glycosyltransferases. Notably, N-acetylglucosaminyltransferase 5 (MGAT5) and core-2 β1-6-N-acetylglucosaminyltransferase 1 (GCNT1) are critical enzymes that generate the β1-6-N-acetylglucosamine (β1-6GlcNAc) as an intermediate for poly-LacNAc synthesis and extension on N-and O-glycans 19 . Intriguingly, the L-phytohemagglutinin (L-PHA)-reactive MGAT5 modified N-glycans and Lycopersicon esculentum lectin (LEL)-reactive poly-LacNAc were highly presented in LX-2 cells (Fig. 1D). Meanwhile, because galectin-1 binding is masked by the terminal sialic acid modification of poly-LacNac, we examined the amount of α2-6-linked sialic acid using Sambucus nigra agglutinin (SNA). The SNA-binding affinity was significantly lower than those of L-PHA and LEL in activated HSCs (Fig. 1D,E). Peanut agglutinin (PNA) recognizes core-1-galactosyl (β-1,3) N-acetylgalactosamine of O-glycan, which had lower affinity with LX-2 cells compared to L-PHA and LEL (Fig. 1D,E). In addition, the fibrotic livers also showed higher L-PHA binding compared to normal livers and the L-PHA binding was co-localized with α-smooth muscle actin (Supplemental Fig. S3), indicating the activated HSCs had high amount Gal-1 binding glycans in vivo. Interestingly, we also observed that carcinoma-associated fibroblasts of oral cancer (CAFs, a type of myofibroblast) showed higher L-PHA and LEL binding affinity compared with normal fibroblasts (Supplementary Fig. S4A). Accordingly, Gal-1's binding affinity was higher in CAFs compared to normal fibroblasts ( Supplementary Fig. S4B). Taken together, the results suggest that the glycome of activated HSCs favors Gal-1 binding which might facilitate HSC migration and activation.
Gal-1 regulates HSC migration and activation through its CRD. Because there was concomitant expression of Gal-1 and its bound glycans in activated HSCs, we examined whether Gal-1 modulates HSC migration in a glycosylation-dependent manner. LX-2 cells were used to examine the biological effect of Gal-1 on HSCs. Extracellular Gal-1 stimulation induced HSC migration and activation in a dose-dependent manner ( Fig. 2A,B) and the effect was blocked by thiodigalactoside (TDG), a lactose analog, (Fig. 2C,D) which demonstrated that Gal-1-induced HSC migration was CRD-dependent. We also compared the Gal-1 induced cell migration in fibroblasts from oral carcinoma (CAFs) and their normal counterparts (NFs), because we observed that CAFs showed higher Gal-1 binding affinity compared with NFs (Supplemental Fig. S4A,B). Gal-1 induced higher migration ability in CAFs than NFs (Supplemental Fig. S4C), which indicates that the glycome of activated fibroblasts favor Gal-1 binding and Gal-1 induced migration. The Gal-1's effect on the perpetuation of HSC activation was examined by silencing endogenous Gal-1 in LX-2 cells, because they express high levels of Gal-1 and myofibroblast markers. Blocking Gal-1 expression reversed the characteristics of activated HSCs as evidenced by reduced α-SMA, collagen, FAP, and fibronectin (Fig. 2E,F). Therefore, the results demonstrate that Gal-1 is a critical regulator of HSC migration through its CRD and is required for the perpetuation of HSC activation.

Both N-and O-glycosylation is required for Gal-1-induced HSC migration. The glycans present
on the cell-surface were critical for Gal-1-induced HSC migration. Therefore, we characterized which glycan structure was critical for Gal-1 functions in activated HSCs by modifying N-and O-glycosylation. The biosynthesis of complex N-glycosylation is illustrated in Fig. 3A. Swainsonine (SW) was used to block poly-LacNAc elongation in complex N-glycans because it inhibits the activity of Golgi α-mannosidase II, a critical glycosidase for complex N-glycan formation (Fig. 3A). We found that SW treatment reduced the amount of L-PHA-reactive glycans (Fig. 3B) and accordingly, reduced Gal-1 binding and Gal-1-induced HSC migration (Fig. 3C,D). The  treatment increased PNA and Gal-1 binding in LX-2 cells (Fig. 4B,C), presumably due to increased exposure of nonsialylated core-1 ligands (low-affinity ligands) to Gal-1. However, BαG treatment decreased Gal-1-induced cell migration (Fig. 4D), which indicates that core 1 O-glycosylation is not required for Gal-1-induced HSC migration. However, we still did not know whether BαG suppressed Gal-1 induced HSC migration through blocking core-2 O-glycosylation. We thus examined whether core 2 branching poly-LacNAc is required for Gal-1-induced migration, because previous studies reported that GCNT1-mediated core 2 branching poly-LacNAc is critical for Gal-1 functions in cancer cells 20 and T cells 21 . GCNT1 shRNA treatment decreased GCNT1 expression (Fig. 4E) and inhibited Gal-1-induced migration (Fig. 4F), thus confirming that core-2 branching poly-LacNAc glycans are required for Gal-1-induced HSC migration. In summary, the results indicated that both MGAT5-and GCNT1-mediated N-and O-glycan structures are critical for Gal-1-induced HSC migration.
Glycosylation-dependent Gal-1 and NRP-1 interactions promote HSC migration. We previously reported that NRP-1 is a critical glycosylated receptor of Gal-1 in endothelial cells and dermal fibroblasts 13,17 , and recent studies demonstrated that NRP-1 is highly expressed in activated HSCs and regulates PDGF and TGF-β signaling 22,23 . The evidence indicates that NRP-1 could be a glycosylated receptor for Gal-1 in HSCs. Therefore, we investigated whether N-and O-glycosylation of NRP-1 is required for Gal-1's functions. Using a Western blot analysis, we found two isoforms of NRP1 in LX-2 cells, and both of their expressions could be suppressed by NRP-1 shRNA (Fig. 5A). A flow cytometric analysis revealed that knockdown of NRP-1 reduced Gal-1 binding to LX-2 cells (Fig. 5B) and significantly inhibited Gal-1-induced migration (Fig. 5C), indicating that NRP-1 is essential for Gal-1-induced HSC migration. To understand how glycosylation regulates the interaction between NRP-1 and Gal-1, LX-2 cells were treated with SW and BαG to modify the N-and O-glycan structures of NRP-1. The low-molecular-weight (LMW) NRP1 (~130 kDa) is presumably the N-and O-GalNAc (N-acetylgalactosamine) glycosylated NRP-1, and the high-molecular-weight (HMW) NRP-1 was presumed to be a glycosaminoglycan (GS) modification of NRP-1 as previously reported 24,25 . Using Far-Western blotting, we found that one major band (~130 kD) interacted with Gal-1, presumably LMW NRP-1 and knockdown of NRP-1 suppressed the binding (Fig. 5D). Furthermore, we found that blocking the N-glycosylation of NRP-1 inhibited Gal-1 binding while blocking O-GalNAc glycosylation did not (Fig. 5E), which suggested that O-GalNAc glycosylation of other low-affinity receptors could be critical for Gal-1 induced migration and signaling because blocking the core-2 O-glycans was able to suppress Gal-1 induced migration (Fig. 4D). In summary, the results indicate that the N-glycosylated NRP-1 is the major receptor of Gal-1 and is critical for Gal-1-induced HSC migration. Although O-GalNAc glycosylation of NRP-1 is not required for Gal-1 binding, the O-GalNAc glycosylation of other low-affinity receptors, such as PDGF or TGF-β receptors (PDGFRs or TGF-βRs), may be required for Gal-1 induced HSC signaling and migration. It has been reported that galectin-3 (Gal-3) promotes HSC activation and liver fibrosis but Gal-3 did not interact with NRP-1 (Supplemental Fig. S5), suggesting Gal-3 regulates HSC homeostasis through distinct mechanisms compared to Gal-1.

Loss of Gal-1 attenuates methionine-and choline-deficient diet (MCD)-induced liver fibrosis.
To examine the functional relevance of Gal-1-induced HSC migration and activation in vivo, liver fibrosis was induced in mice by feeding them an MCD for 8 weeks 29 . Picrosirius red and Masson's trichrome staining revealed reduced collagen deposition in Gal-1-null mice compared to wild-type mice (Fig. 8A,B and Supplemental Fig. S6). Western blotting also showed a decreased α-SMA expression in Gal-1-null mice compared to wild-type mice. However, the amount of pro-inflammatory cytokine, serum AST (aspartate aminotransferase) and ALT (alanine transaminase) were similar between Gal-1 KO and wild-type mice feed with MCD diet (Fig. 8E,F). We thus examined whether the macrophages express Gal-1 in MCD-fed mice using anti-CD68 (a macrophage marker) and anti-Gal-1 antibodies. The CD68 staining was co-localized with Gal-1 indicating that macrophages expressed Gal-1 (Supplemental Fig. S7). However, there were few CD68-positive signals in the livers indicating the MCD feeding did not induce strong inflammatory responses in the livers. Therefore, it is suggested that Gal-1 mainly affects fibrogenesis in the MCD-fed mice, and the inflammatory effect of Gal-1 may not be significantly detected because of the mild inflammation in this model.

Discussion
Previous studies reported glycosylation alterations in HSC activation and liver fibrosis using high-throughput analyses. For example, using an oligonucleotide and lectin microarray, Yu et al. reported that both glycan-modification gene and glycan profiles changed in fibrotic livers compared to normal livers 30 . Mondal et al. reported distinct glycosylation of serum proteins in HBV-and HCV-related cirrhotic patients using 2D gel electrophoresis and liquid chromatography-mass spectroscopy (LC-MS) 31 . However, it is unclear whether altered glycosylation affects HSC homeostasis and liver fibrosis. In this study, we highlighted the coordinated effect of cell-surface glycans and Gal-1 on HSC homeostasis and integrated glycosylation-dependent Gal-1/NRP-1 interactions into canonical PDGF-and TGF-β-induced signaling pathways. Based on the results, we proposed a model in which co-evolution of the HSC glycome and Gal-1 promotes HSC activation and migration (Fig. 9). First, Gal-1 is highly expressed in activated HSCs compared to quiescent HSCs. Second, the "Gal-1-permissive" glycan repertoire (greater poly-LacNAc-modified glycans and less terminal sialic acid modification) in activated HSCs facilitates Gal-1 to induce PDGF-and TGF-β-like signals through the co-clustering of NRP-1/PDGFRs and NRP-1/TGF-βRs. In contrast, the "Gal-1 non-permissive" glycan repertoire in quiescent HSCs provides low-affinity ligands for Gal-1 binding, which induces weak PDGF and TGF-β signals. Our results not only demonstrated the cooperative effect of Gal-1 and glycosylation changes in HSC activation but also implied that   MCD diet-induced collagen deposition is reduced in Gal-1 null mice (Gal-1-KO) (n = 6) compared to wild-type mice (WT) (n = 6). Liver fibrosis was induced by feeding mice with an MCD diet for 8 weeks. Collagen was detected using picrosirius red staining. The left panel shows the representative images. The right panel shows the quantitative data of picrosirius red staining using ImageJ software. (C,D) HSC activation is reduced in Gal-1 null mice compared to wild-type mice. HSC activation was examined by measuring α-smooth muscle actin (α-SMA) expression using Western blotting (left panels) which confirmed that loss of Gal-1 suppresses HSC activation and extracellular matrix production in livers of MCD diet-fed mice. (E) Serum AST and ALT amount are similar between Gal-1 null mice and wild-type mice. Serum AST and ALT were measured using a VetTest ® Chemistry Analyzer. (F) Loss of Gal-1 does not change the RNA expression levels of proinflammatory cytokines (IL-1β, TNF-α, CCL2) compared to wild-type mice. The RNA was extracted from mouse livers and was converted to cDNA. Proinflammatory cytokines expression was measured using RT-qPCR. Relative expression levels were calculated by comparing ∆CT values of each group to those of wild-type mice without treatment. Data are shown as folds of change.
We found that Gal-1 interacts with NRP-1 in an N-glycosylation-dependent manner which is consistent with a previous finding 33 . However, the glycosylation status of NRP-1 is poorly understood. Several N-and O-glycosylation sites of NRP-1 has been identified 24, 34-36 , but there are few publications addressing the biological effects of N-and O-glycosylation on NRP-1's functions. Glycosaminoglycan (GAG) modification was the best-characterized O-glycosylation of NRP-1, which is characterized by heparin or chondroitin attached to ser 612 (refs 24 and 25) and a HMW NRP-1 (>180 kDa). GAG-modified NRP-1 regulated VEGF-and PDGF-induced intracellular signaling and biological functions in smooth muscle cells 24,37 and glioblastoma cells 25 . However, in our results, Gal-1 interacted with the LMW NRP-1 (~130 kDa) instead of GAG-modified NRP-1. Furthermore, blocking N-glycosylation reduced Gal-1/NRP-1 interactions and suppressed Gal-1-induced HSC migration, which highlights the critical role of N-glycosylation in modulating NRP-1's functions. Meanwhile, we found that O-GalNAc glycosylation of NRP-1 was not required for Gal-1 binding but blocking core-2 O-glycosylation suppresses Gal-1 induced migration. The contradictory results may be explained by that O-GalNAc glycosylated PDGFRs and TGF-βRs are low-affinity receptors for Gal-1 but they are required for signal transduction. Furthermore, since it has been known that NRP-1 forms a complex with PDGFRs and TGF-βRs in HSCs 22, 23 , we proposed the N-glycan of NRP-1 is responsive for Gal-1 binding and the O-GalNAc glycosylation of PDGFRs and TGF-βRs are responsive for Gal-1 to transduce signaling. Interestingly, another study proposed that GAG-modified NRP-1 increased in activated HSCs, which forms a "molecular net" to sequester local growth factors such as PDGF and TGF-β 22 . Therefore, these results indicated that both GAG and N-glycosylation of NRP-1 could influence PDGF-and TGF-β-induced HSC activation and migration. Full characterization of NRP-1 glycan structures in HSCs will be important to understand the mode of the Gal-1/NRP-1 interaction and its role in HSC activation and migration.
Because members of the galectin family recognize galactoside, it was reported that galectins have redundant roles in regulating biological functions such as plasma cell formation 38 , tumor angiogenesis 39 , T cell apoptosis 40 , and wound healing 41 . Intriguingly, previous studies also reported that Gal-3 regulates rat HSC proliferation 42 and activation, and liver fibrosis 43 . Therefore, the glycan repertoire of activated HSCs may facilitate Gal-3 binding, which explains why there is a redundancy of Gal-1 and Gal-3 in HSC activation and proliferation. However, we speculated that Gal-1 and Gal-3 stimulate partly distinct signaling pathways in HSCs, because a microarray analysis of Gal-1-and Gal-3-knockdown HSCs showed distinct gene expression profiles (our unpublished data). Differences may have resulted from the distinct protein structures of Gal-1 and Gal-3, which may selectively facilitate their recognition of distinct glycoproteins. For instance, we found that Gal-1 binds to NRP-1 but Gal-3 does not (Supplemental Fig. S4). Recently, two galectin inhibitors, GR-MD-02 and GM-CT-01, showed therapeutic effects on thioacetamide (TAA)-and non-alcoholic steatohepatitis (NASH)-induced liver fibrosis 44,45 , but the underlying mechanisms are largely unknown. Although both GR-MD-02 and GM-CT-01 can bind Gal-1 and Figure 9. A proposed model: the co-evolution of the HSC glycome and Gal-1 promotes HSC activation and migration. First, Gal-1 is highly expressed in activated HSCs compared to quiescent HSCs. Second, the "Gal-1permissive" glycan repertoire (greater poly-LacNAc-modified glycans and less terminal sialic acid modification) in activated HSCs facilitates Gal-1 to induce PDGF-and TGF-β-like signals through the co-clustering of NRP-1/PDGFRs and NRP-1/TGF-βRs. In contrast, the "Gal-1 non-permissive" glycan repertoire in quiescent HSCs provides low-affinity ligands for Gal-1 binding, which induces weak PDGF and TGF-β signals.
Immunohistochemical (IHC) and immunofluorescence analysis. Serial 5-µm histological sections were deparaffinized in xylene and rehydrated. After blocking of endogenous peroxidase by incubation with 3% hydrogen peroxide, slides were incubated with anti-Gal-1 and anti-α-SMA antibodies overnight at 4 °C, and the signal was visualized by applying the PolyDetector HRP Detection System (Bio-SB). Slides were counterstained with hematoxylin and mounted with mounting solution. Picrosirius red and Masson's trichrome stains were used to visualize collagen using two kits (Polysciences and Sigma). For immunofluorescence analysis, the slides were incubated with anti-Gal-1, anti-CD68 and anti-α-SMA antibodies overnight at 4 °C, and the signal was visualized by Alexa Fluor-488 and -594 secondary antibodies. For L-PHA binding, the slides were incubated with biotinylated L-PHA for 30 min followed by DyLight ® 488 streptavidin staining for 30 min. The slides were mounted with DAPI solution and were visualized using a fluorescence microscopy.
Luciferase assay. LX-2 cells (2 × 10 5 ) were seeded in 24-well dishes and were transfected with 0.4 µg of the SBE4-Luc plasmid (luciferase reporter containing four copies of the Smad-binding site), and 0.1 µg of the pCMV-LacZ plasmid (Clontech) was co-transfected in all transfection experiments as an internal control. Cells were starved for 24 h followed by TGF-β (1 ng/ml) treatment for 24 h, and luciferase activities were measured with the Luciferase Assay system (Promega). All luciferase activities were normalized to β-galactosidase activity.

Statistical analysis.
All experiments were performed in duplicate. Results are shown as the mean ±SEM of three independent assays. Results were assessed using an unpaired Student's t-test or 1-way analysis of variance (ANOVA). Statistical significance was set at p < 0.05.