Pathological lymphangiogenesis is modulated by galectin-8-dependent crosstalk between podoplanin and integrin-associated VEGFR-3

Lymphangiogenesis plays a pivotal role in diverse pathological conditions. Here, we demonstrate that a carbohydrate-binding protein, galectin-8, promotes pathological lymphangiogenesis. Galectin-8 is markedly upregulated in inflamed human and mouse corneas, and galectin-8 inhibitors reduce inflammatory lymphangiogenesis. In the mouse model of corneal allogeneic transplantation, galectin-8-induced lymphangiogenesis is associated with an increased rate of corneal graft rejection. Further, in the murine model of herpes simplex virus keratitis, corneal pathology and lymphangiogenesis are ameliorated in Lgals8−/− mice. Mechanistically, VEGF-C-induced lymphangiogenesis is significantly reduced in the Lgals8−/− and Pdpn−/− mice; likewise, galectin-8-induced lymphangiogenesis is reduced in Pdpn−/− mice. Interestingly, knockdown of VEGFR-3 does not affect galectin-8-mediated lymphatic endothelial cell (LEC) sprouting. Instead, inhibiting integrins α1β1 and α5β1 curtails both galectin-8- and VEGF-C-mediated LEC sprouting. Together, this study uncovers a unique molecular mechanism of lymphangiogenesis in which galectin-8-dependent crosstalk among VEGF-C, podoplanin and integrin pathways plays a key role.

L ymphatic vessels are essential for preservation of fluid balance, nutrient absorption and immune surveillance. Lymphangiogenesis is associated with diverse pathological conditions including metastatic dissemination, solid organ graft rejection, type 2 diabetes, obesity, hypertension, lymphedema and chronic wound healing 1 . Pathological lymphangiogenesis is also associated with various diseases of the eye including corneal graft rejection, herpetic keratitis, dry eye disease, ocular allergy and glaucoma [2][3][4][5][6][7] . Indeed, recent studies suggest that lymphatic vessel invasion in and around primary tumours rather than invasion of blood vessels, is the key prognostic marker of the aggressiveness of various types of cancers 8 , and that the growth of lymphatic vessels is also the major reason of corneal graft rejection 3 . The key regulatory signalling axis that induces lymphangiogenesis is vascular endothelial growth factor receptor-3 (VEGFR-3) and its ligand, VEGF-C.
Recent studies have demonstrated that the members of the galectin family of mammalian lectins characterized by a carbohydrate recognition domain (CRD) with affinity for b-galactoside-containing glycans, play a critical role in hemangiogenesis. In this respect, we have shown that a member of galectin family, galectin-3, modulates VEGF-A-induced angiogenic response by binding via its CRD to the N-glycans of integrin avb3 and VEGFR-2 and subsequently activating angiogenic signalling pathways 9,10 . Galectin-8 is a tandem-repeat type member of the galectin family. It contains two different CRDs. The N-terminal CRD prefers a2,3-sialyl glycans and mainly contributes to its unique carbohydrate-binding specificity [11][12][13] . In vitro studies have shown that galectin-8 binds to podoplanin (PDPN) and that the lectin promotes adhesion and haptotaxis of lymphatic endothelial cells (LECs) 14 . However, the direct evidence that galectin-8 exerts its biological functions through PDPN is lacking. In fact, Cueni and Detmar 14 speculated that contribution of the interaction of galectin-8 with PDPN in the modulation of LEC migration and adhesion is most likely minor. Also, based on the findings that both unglycosylated and extensively glycosylated PDPN-Fc inhibit LEC adhesion and migration in vitro, it has been suggested that the interactions with PDPN ligands on the surface of LECs do not depend on PDPN glycosylation 15 . To date, the biological relevance of carbohydratedependent galectin-8/PDPN interactions is still elusive and more direct studies involving the use of galectin-8 knockout (KO) and PDPN KO mice have not been reported.
PDPN is a unique transmembrane receptor protein 16 . It is expressed by LECs but not blood ECs and promotes blood-lymph separation. Mice lacking PDPN have leaky lymphatic vessels and congenital lymphedema 17,18 . In vitro studies have shown that PDPN expression in LECs is required for lymphatic capillary tube formation as well as VEGF-A-induced cell migration 19,20 . The critical role of extracellular domain of PDPN in lymphangiogenesis has been demonstrated by studies showing that PDPN-Fc and the functional blocking antibody against extracellular domain of PDPN inhibit LEC migration and tube formation in vitro and suppress lymphangiogenesis in inflamed mouse corneas in vivo 15,21 . The extracellular domain of PDPN is heavily glycosylated, and O-glycosylation as well as sialylation are critical for PDPN-mediated blood-lymph separation and platelet aggregation 18,22,23 .
Integrins are a major class of adhesion receptors. It is well-established that angiogenic signals induced by VEGF and the interplay between VEGF receptors and integrins expressed on endothelial cells play an important role in the process of angiogenesis. Integrins that are well-defined for their role in lymphangiogenesis include a1b1, a2b1, a4b1, a5b1 and a9b1 (reviewed in Chen et al. 24 ). Activation of integrin b1 by collagen and fibronectin promotes VEGFR-3 activation 25 . Conversely, inhibition of integrin a5b1 but not avb3 attenuates VEGF-Cinduced VEGFR-3 activation 26 . In addition to extracellular matrix proteins and growth factor receptors, integrins are glycosylated and interact with distinct members of galectin family in a glycan-dependent manner 10,27 . Although integrins, VEGFR-3 and PDPN are glycosylated like most cell surface receptors, more direct studies on the role of carbohydratedependent function of integrins, PDPN and VEGF-C/VEGFR-3 in the regulation of lymphangiogenesis have not been reported.
In the current study, using multiple approaches involving the use of galectin-8 mutants lacking carbohydrate-binding activity, KO mice, specific sugar inhibitors of galectin-8, and siRNA knockdown of key players of lymphangiogenesis, we establish a critical role of galectin-8 and carbohydrate-mediated recognition in the process of lymphangiogenesis. We demonstrate that galectin-8 expression is markedly upregulated in inflamed corneas, and that galectin-8 is a potent lymphangiogenic factor and a key mediator of VEGF-C signalling. Further, we show that PDPN is a key player in VEGF-C-induced lymphangiogenesis, that knockdown of PDPN interferes with integrin signalling cascades in LECs, and that galectin-8 is a critical mediator of crosstalk among VEGF-C, PDPN and integrin lymphangiogenic pathways. In addition, herpes simplex virus (HSV)-1 infectioninduced pathological lymphangiogenesis is reduced in galectin-8 KO mice, and galectin-8-induced lymphangiogenesis is associated with an increased rate of graft rejection in a mouse model of allogeneic corneal transplantation. More importantly, we show here that inhibitors of galectin-8 decrease lymphangiogenesis in inflamed mouse corneas in vivo. This is significant considering that there is much interest in finding ways to inhibit the activities of pro-lymphangiogenic factors for preventing graft rejection, tumour metastasis and other inflammatory conditions.

Results
Galectin-8 is upregulated in inflamed human and mouse corneas. In corneas of patients with graft failure and bacterial keratitis, numerous inflammatory cells were detected in the stroma as highlighted by periodic acid Schiff (PAS) staining (Fig. 1a). Normal corneas expressed little galectin-8 (Fig. 1a). In contrast, robust galectin-8 immunoreactivity was detected in corneas of patients with graft failure and bacterial keratitis. Similarly, in mouse corneas treated with thermal cautery or AgNO 3 cautery, intense galectin-8 immunoreactivity was detected in the stromal matrix, whereas in untreated control mouse corneas, galectin-8 expression was minimal (Fig. 1b). In some areas, particularly in the anterior stroma, galectin-8 and type I collagen immunoreactivity colocalized (Fig. 1c). While strong galectin-8 immunoreactivity was detected in lymphatic vessels (CD31 þ LYVE-1 þ ), weak immunoreactivity of galectin-8 was observed in blood vessels (CD31 þ LYVE-1 À , Fig. 1d). These findings are consistent with a published study showing that the mRNA and protein levels of galectin-8 are higher in LECs than in blood ECs 14 .
In inflamed mouse corneas, galectin-8 immunoreactivity was detected in macrophages (F4/80 þ CD11b þ , Fig. 1e) and CD4 þ T cells (CD4 þ CD45 þ , Fig. 1f). Interestingly, some F4/80 þ cells in the posterior corneal stroma were galectin-8 - (Fig. 1e), suggesting that either a subset of F4/80 þ cells express galectin-8, or the cells need to be activated to express galectin-8. While it is reasonable to suggest that cells stained positively may be the possible source of the lectin, we note that paracrine actions of galectins have been reported. In this respect, galectins secreted by one cell type may bind to the glycan receptors on the adjacent cells. Therefore, the cells that exhibit immunoreactivity with galectin-8 may not necessarily be the cells that produce the lectin. Taken together, this study demonstrates that galectin-8 is upregulated in inflamed human and mouse corneas.
Galectin-8 promotes lymphangiogenesis in vivo. The normally avascular cornea has been extensively used as the in vivo model to investigate the molecular mechanism of hemangiogenesis and to examine the efficacy of the inhibitors and activators of hemangiogenesis. In recent years, cornea has also proven to be an invaluable model for defining general mechanisms of lymphangiogenesis. To determine whether galectin-8 promotes lymphangiogenesis, we used the mouse corneal micropocket assay. The vessel area, representing the extent of lymphangiogenesis, was calculated 1 week after galectin-8 pellets were implanted in mouse corneas. The extent of galectin-8-mediated lymphangiogenesis increased in a dose-dependent manner, whereas control pellets had no effect (Fig. 2a,b). To further demonstrate the pro-lymphangiogenic capacity of galectin-8 in vivo, we employed another well-established method, the Matrigel plug assay. Matrigels containing galectin-8 or VEGF-C were injected in mice subcutaneously, where they solidified to form plugs. On day 7 post injection, the Matrigel plugs were harvested and frozen sections of the plugs were stained with anti-LYVE-1 to visualize lymphatic vessels. As expected, VEGF-C stimulated the growth of new lymphatic vessels in Matrigel, whereas PBS (control) did not (Supplementary Fig. 1a  Supplementary Methods). Similar to VEGF-C, galectin-8 also promoted robust lymphangiogenesis in Matrigel ( Supplementary  Fig. 1a). To determine the mitogenic effect of galectin-8 on LECs in vivo, corneal micropocket assays were performed using VEGF-C, galectin-8 or control pellets in Prox1-EGFP (enhanced green fluorescent protein) reporter mice. On day 7 post surgery, corneas implanted with various pellets were stained with anti-Ki67. In control corneas, o4% of LEC (Prox1 þ cells) were Ki67 þ . In contrast, more than 35% of LECs were Ki67 þ in both VEGF-C-and galectin-8-induced lymphangiogenic areas (Supplementary Fig. 1b and Supplementary Methods). These data lead us to conclude that galectin-8 is a pro-lymphangiogenic factor as assessed by two independent in vivo methods.
Galectin-8 promotes LEC sprouting in vitro. To characterize the role of galectin-8 in the regulation of phenotypic behaviour of LECs in vitro, we examined the effect of galectin-8 on LEC proliferation, migration, tube formation and sprouting. In contrast to in vivo results, galectin-8 treatment had no effect on LEC proliferation in vitro (Supplementary Fig. 1c and Supplementary Methods). We reason that continuously produced galectin-8 may be required to stimulate LEC proliferation in vitro, as has been observed for other (lymph)angiogenic factors such as angiopoitin-1 (refs 28,29) and sphingosine-1-phosphate 30 . Galectin-8, however, promoted cell migration and tube formation ( Supplementary Fig. 1d,e and Supplementary Methods). To better characterize the molecular mechanism by which galectin-8 mediates lymphangiogenesis, we utilized an in vitro threedimensional LEC sprouting assay. In the sprouting assay, galectin-8, but not galectins-1, 3 or 7, promoted LEC sprouting (Fig. 2c). The stimulatory effect of galectin-8 on LEC sprouting was concentration-dependent ( Fig. 2d,e). Next, we tested whether the stimulatory effect of galectin-8 on LEC sprouting was carbohydrate-dependent.
Next, we tested whether N-CRD can serve as a dominant negative inhibitor of galectin-8. Di/multivalent property of galectins allow them to crosslink many cell surface and extracellular matrix glycoproteins, such as integrins and growth factor receptors, to regulate signal transduction pathways 32 . Isolated CRDs, which retain their carbohydrate-binding ability but are unable to dimerize or oligomerize and crosslink cell surface receptors, may compete with the carbohydrate-binding ability of the endogenous galectins and, hence, act as a dominant negative inhibitor 10,33 . Published studies have shown that isolated CRDs of galectin-8 retain the carbohydrate-binding activity but manifest impaired biological activity 13,34 , suggesting that the biological function of the lectin is dependent on cooperative interactions of the two CRDs. As described before, N-CRD of galectin-8 (Gal-8N) is unique among galectins in exhibiting a very high affinity for a2,3-sialyl glycans [11][12][13] . To determine whether the pro-lymphangiogenic property of galectin-8 is dependent on the cooperative action of both CRDs, we tested whether N-CRD is able to promote LEC sprouting. Unlike full-length galectin-8, Gal-8N failed to induce LEC sprouting (Fig. 2h). Moreover, Gal-8N effectively inhibited galectin-8-induced LEC sprouting ( Fig. 2i and Supplementary Fig. 2d-f). These results suggest that Gal-8N serves as a dominant negative inhibitor of galectin-8 and that a2,3-sialyl glycans recognized by Gal-8 N as well as cooperative action of both CRDs are required for galectin-8induced LEC sprouting.

Galectin-8 modulates pathological lymphangiogenesis in vivo.
Since lymphangiogenesis contributes to corneal graft rejection 3,37,38 , we sought to determine if galectin-8-induced lymphangiogenesis promotes graft rejection in a mouse model of corneal transplantation. To test this, we performed allogeneic corneal transplantation (donor, male C57BL/6 mice; recipient, male BALB/c mice; both 10 weeks old) and treated the recipient mice with recombinant galectin-8 or PBS (subconjunctivally and intraperitoneally, twice a week beginning at day 7). On postoperative week 4, graft survival rate was markedly decreased in galectin-8-treated mice compared with PBS-treated mice (survival rates: PBS, 66.7%, 14 out of 21; galectin-8, 18.5%, 5 out of 27) (Fig. 3a). Excised corneas from each group were assessed for the extent of corneal lymphangiogenesis on postoperative week 4. The extent of lymphangiogenesis in galectin-8-treated corneas was significantly higher than that of control corneas (Fig. 3b). These results support the notion that galectin-8-induced lymphangiogenesis promotes graft rejection.
Next, to determine the role of endogenous galectin-8 in lymphangiogenesis in vivo, we utilized two different mouse models of pathological lymphangiogenesis in WT and galectin-8 KO mice. HSV-1 keratitis is one of the most common ocular infections. It is characterized by recurrent episodes and is the leading cause of infectious corneal blindness in the developed countries 39 . HSV-1 infection drives corneal lymphangiogenesis 7 and inhibiting lymphangiogenesis is thought to be of potential therapeutic value to alleviate corneal pathology caused by HSV-1 infection 40 . To determine if endogenous galectin-8 plays a role in HSV-induced pathology and lymphangiogenesis, we infected mouse corneas of WT and galectin-8 KO mice with a clinical isolate of HSV-1. As expected, on day 8 post infection, HSV-1 infection caused corneal opacification (Fig. 3c) and induced corneal lymphangiogenesis (Fig. 3d). In contrast, galectin-8 deficiency not only ameliorated corneal opacity ( Fig. 3c) but also reduced corneal lymphangiogenesis (Fig. 3d), suggesting that galectin-8 is involved in the pathogenesis of HSV keratitis, at least partly, by regulating pathological lymphangiogenesis.
Suture placement in the mouse cornea is a commonly used technique to determine inflammatory lymphangiogenesis in the setting of sterile condition (in contrast to infectious agent-induced lymphangiogenesis) 41,42 . Clinical relevance of the method stems from the fact that suture placement is routinely used in many corneal surgeries. In this model also, the extent of suture-induced lymphangiogenesis was reduced in the galectin-8 KO mice (Fig. 3e). Together, these findings conclusively demonstrate that galectin-8 is required for robust pathological lymphangiogenesis.
Effect of galectin-8 on VEGF-C-induced LEC sprouting and lymphangiogenesis. To gain mechanistic insight on galectin-8mediated LEC sprouting and lymphangiogenesis, first we tested whether the lectin influences the function of a well-known lymphangiogenic molecule, VEGF-C. VEGF-C-induced LEC sprouting was inhibited by galectin-8 inhibitors: TDG (a pan inhibitor of galectins, Fig. 4a   To further determine the effect of galectin-8 on VEGF-C-induced LEC sprouting, we tested the effect of exogenous galectin-8 on VEGF-C-induced LEC sprouting. Galectin-8, but not galectins-1, 3 and 7, markedly enhanced VEGF-C-induced LEC sprouting (Fig. 4g). In addition, at 0.75 mM of galectin-8, VEGF-C-induced LEC sprouting was five times higher than that seen by VEGF-C alone or galectin-8 alone (Fig. 4h), indicating that galectin-8 has a synergistic effect on VEGF-C-induced LEC sprouting. In the in vivo micropocket assays also, galectin-8 collaboratively augmented VEGF-Cinduced lymphangiogenesis (Fig. 4i). Moreover, in galectin-8 KO mice, the extent of VEGF-C-induced lymphangiogenesis was significantly reduced (Fig. 4j).
Galectin-8-and VEGF-C-induced LEC sprouting is dependent on PDPN. It has been reported that cell surface receptor clustering by the galectin-glycan lattices increases the magnitude or duration of signalling from the cell surface 43,44 . Therefore, in an effort to characterize the mechanism by which galectin-8 modulates VEGF-C-induced lymphangiogenesis, we first conducted a study to determine whether the lectin modulates VEGFR-3, the predominant VEGF-C receptor. This study revealed that VEGFR-3, but not VEGF-C, is a galectin-8binding protein ( Supplementary Fig. 4a,b), and that galectin-8 clusters VEGFR-3 on cell surface ( Supplementary Fig. 4c). However, surprisingly, knockdown of VEGFR-3 had little effect on galectin-8-induced LEC sprouting ( Supplementary Fig. 5a-c), suggesting that molecules besides VEGFR-3 are involved in galectin-8-induced LEC sprouting. Therefore, we sought to determine whether galectin-8-mediated LEC sprouting involves other receptors for VEGF-C. In this respect, it is known that VEGFR-2, which also binds VEGF-C, is not involved in VEGF-Cinduced sprouting 45 Next we performed studies to determine whether galectin-8induced LEC sprouting is dependent on PDPN. PDPN is thought to play a role in lymphangiogenesis 19,20 . However, the role of galectin-8 in the modulation of PDPN has thus far not been fully investigated and virtually nothing is known about the role of PDPN in VEGF-C-induced lymphangiogenesis. Here, we first showed that PDPN expressed in LECs interacts with galectin-8, but not galectins-1, 3 or 7 (Fig. 5a), and the binding of PDPN to galectin-8 was carbohydrate-dependent (Fig. 5b). In addition, PDPN expressed in LECs contains a2,3-sialylated glycans (Fig. 5b). Removal of a2,3-sialylated glycans by treatment with a2-3 neuraminidase abrogated the interaction of PDPN and galectin-8, suggesting that galectin-8 binds a2,3-sialylated glycans of PDPN ( Supplementary Fig. 6 and Supplementary Methods). Secondly, to determine whether PDPN plays a role in galectin-8and/or VEGF-C-induced LEC sprouting, spheroids prepared using primary LECs transfected with control or PDPN siRNA were treated with galectin-8 or VEGF-C. The expression of PDPN was reduced by 82% in the siRNA transfected LECs (Fig. 5c). PDPN knockdown not only markedly inhibited galectin-8induced LEC sprouting, but also substantially reduced VEGF-C-induced LEC sprouting (Fig. 5d,e). Thirdly, PDPN knockdown in LECs substantially reduced galectin-8-and VEGF-C-induced activation of AKT but not ERK (Fig. 5f), suggesting that PDPN modulates galectin-8-and VEGF-C-induced LEC sprouting largely by activation of AKT pathway.
Lastly, to determine whether PDPN plays a role in galectin-8and/or VEGF-C-induced lymphangiogenesis in vivo, we used mice with tamoxifen-inducible global deletion of PDPN (Pdpn f/f ;CagCre) 49 to perform the corneal micropocket assays. As expected, VEGF-C pellets markedly induced both haemangiogenesis and lymphangiogenesis in WT mice. In contrast, VEGF-C-induced lymphangiogenesis was significantly reduced in the PDPN-deficient mice (Fig. 5g). Likewise, galectin-8-induced lymphangiogenesis (Fig. 5h) was reduced in PDPN-deficient mice. Together, the data suggest that PDPN is a key player in, not only galectin-8-induced lymphangiogenesis, but also VEGF-C-induced lymphangiogenesis.
Interplay among integrins, galectin-8 and PDPN. In addition to a9b1 integrin, several other integrins including a1b1, a4b1 and a5b1 are involved in the process of lymphangiogenesis 24 and the interplay between integrin b1 and VEGFR-3 has been reported 25,26 . Therefore, to determine whether galectin-8-PDPNinduced LEC sprouting involves specific integrins; first, LEC spheroids were treated with galectin-8 or VEGF-C in the presence or absence of blocking antibodies and peptides against a panel of integrins. In this study, only blocking of integrins a1b1 (by obtustatin) and a5b1 (by the neutralizing antibody) inhibited both VEGF-C-and galectin-8-induced LEC sprouting (Fig. 6a). Secondly, to determine whether PDPN indirectly regulates the functions of VEGF-C/VEGFR-3 through controlling the function of integrins a1b1 and a5b1 in galectin-8-dependent manner, we performed studies to determine whether: (i) PDPN inhibition attenuates matrix-mediated LEC migration; (ii) PDPN interacts with integrins a5b1 in a galectin-8-dependent manner; and (iii) knockdown PDPN impedes integrin-mediated signalling cascades. In this study, blocking the function of PDPN by antibodies as well as siRNA knockdown attenuated both fibronectin-and galectin-8-promoted cell migration (Fig. 6b). These data in conjunction with a published study 15 showing that PDPN-Fc inhibits type I collagen-mediated LEC migration, suggest that PDPN is involved in not only galectin-8 but also fibronectin-and type I collagen-mediated LEC migration, a process in which integrins are well-known to play a key role. To assess the galectin-8-dependent interaction between integrin b1 and PDPN, primary LECs were treated with galectin-8 for 15 min, fixed without permeabilization, stained with antibodies against integrin b1, PDPN and galectin-8, and examined by confocal microscopy. In untreated control cells, galectin-8, integrin b1 and PDPN were homogenously distributed all over the LECs (Fig. 6c). Since we showed that galectin-8 was upregulated in inflamed corneas (Fig. 1), we added the exogenous galectin-8 to see whether galectin-8 changes the distribution of PDPN and/or integrin b1 on LECs. Addition of galectin-8 caused dramatic redistribution and clustering of PDPN and integrin b1 on LEC plasma membrane (Fig. 6d). To more directly assess the association between integrins and PDPN, lysates from untreated or galectin-8-treated LECs were incubated with anti-PDPN antibody, and immunoprecipitated proteins were examined by western blotting using antibodies against specific integrins, galectin-8 and PDPN. In untreated cell lysates, immunoprecipitation with anti-PDPN co-immunoprecipitated endogenous galectin-8 and specific integrins (a1, a5, av, b1, but not a9 or b3), indicating that PDPN interacted with endogenous galectin-8 and the association between PDPN and specific integrins (a1, a5, av and b1) was constitutive (Fig. 6e). When cells were treated with exogenous galectin-8, there was an increased association between PDPN and integrins a5, av, b1, while the association between PDPN and integrin a1 remained similar (Fig. 6e).
Next, to assess the role of PDPN on integrin-mediated signalling, LECs transfected with control or PDPN siRNA were  seeded on collagen I-or fibronectin-coated wells and allowed to adhere for 15 min at 37°C. Cell lysates from attached cells on collagen I or fibronectin were analysed with western blotting using phospho-specific integrin and focal adhesion kinase (FAK) antibodies. Phosphorylation of integrin b1 was reduced in PDPN knockdown cells (Supplementary Fig. 7), suggesting that the activation of integrin b1 is reduced in the absence of PDPN. Also, phosphorylation of FAK was decreased in the PDPN knockdown cells seeded on collagen I-but not fibronectin-coated wells ( Supplementary Fig. 7), whereas phosphorylation of ERK was markedly reduced in the PDPN knockdown cells seeded on both matrix proteins (Supplementary Fig. 7). To eliminate the possibility that the loss of integrin activation and signalling in the PDPN knockdown cells was not due to altered expression and cellular distribution of integrins and VEGFR-3, we performed studies to determine whether PDPN knockdown alters the cell surface expression of integrins and/or VEGFR-3, using two different approaches: cell surface biotinylation and flow cytometry analysis. In the cell surface biotinylation approach, GAPDH and Prox1 were not detected in the streptavidin pulldown cell lysates, and cell surface PDPN was markedly reduced in the PDPN knockdown cells (Supplementary Fig. 8 and Supplementary Methods). Expression of cell surface VEGFR-3, integrins a5 and b1 was similar in the PDPN knockdown cells and control cells (Supplementary Fig. 8). Similarly, in the flow cytometry analysis approach, cell surface expression of PDPN was reduced, whereas that of VEGFR-3, integrins a5 and b1 did not change in the PDPN knockdown cells (Fig. 6f). Thus, PDPN deficiency does not change the cell surface expression of integrins a5/b1 and VEGFR-3 of LECs. Furthermore, transcriptome analysis of PDPN knockdown and control cells revealed that the coefficient of determination (R 2 ) of the linear curve fitting was more than 0.9 ( Supplementary Fig. 9, Supplementary Data 1  ARTICLE differentially expressed. Taken together, these data lead us to conclude that PDPN regulates the functions of integrin b1 complexes, specifically of integrins a1b1 and a5b1 in LECs, and that this function is galectin-8-dependent.

Galectin-8 inhibitors decrease inflammatory lymphangiogenesis.
To determine whether galectins can be targeted to control lymphangiogenesis, two in vivo models of lymphangiogenesis were used. After suture placement and AgNO 3 cauterization in the corneas of Prox1-EGFP reporter mice to induce inflammation, the mice were treated with TDG (200 mM, a pan inhibitor of galectins) or Gal-8N (15 mg, the dominant negative inhibitor of galectin-8) by subconjunctival injections on days 0, 2, 4 and 6 post surgery. At the end of the treatment period, lymphatic vessel areas were quantified. Treatment with both TDG (Fig. 7a,c,g,i) as well as Gal-8N (Fig. 7d,f,j,l) significantly suppressed suture-and cautery-induced corneal lymphangiogenesis. These data suggest a promising new mechanism for the modulation of pathological lymphangiogenesis by targeting galectin-8.

Discussion
We demonstrate here that galectin-8 is highly upregulated in pathological corneas and plays a critical role in the process of lymphangiogenesis. The striking finding that several other members of galectin family including galectins-1 and 3, which are known to promote hemangiogenesis, did not promote LEC sprouting suggests that galectin-8-mediated LEC sprouting involves the affinity of N-CRD of galectin-8 for 3 0 -sialylated galactosides that is unique among animal galectins [11][12][13] . In support of this notion, specific inhibition of the N-CRD of galectin-8 with 3 0 -SL reduced LEC sprouting and a galectin-8 mutant, Gal-8Q47A, which has lost its ability to bind to a2,3-sialyl glycans, did not promote LEC sprouting. Together, these data establish that galectin-8 promotes LEC sprouting in a carbohydrate-dependent manner and that N-CRD of galectin-8 was directly involved in the stimulatory effect of galectin-8 on LEC sprouting. A major finding of the current study is that galectin-8 modulates VEGF-C-mediated lymphangiogenesis. Our studies

Boost
Signalling transduction (FAK, ERK, AKT) Promotes lymphangiogenesis via LEC migration and sprouting. show that Gal-8N (the dominant-negative inhibitor of galectin-8) and 3 0 -SL (a competing disaccharide) ameliorated VEGF-Cinduced LEC sprouting. Furthermore, in the in vivo corneal micropocket assay, the extent of lymphangiogenesis induced by VEGF-C was significantly less in galectin-8 KO mice compared with the WT mice. To our knowledge, this is the first demonstration of a defect in lymphangiogenic response of galectin-8 KO mice. In addition, exogenous galectin-8 markedly enhanced VEGF-C-induced LEC sprouting in vitro and lymphangiogenesis in vivo in a carbohydrate-dependent manner. Together, these data conclusively establish that galectin-8 significantly influences VEGF-C-mediated lymphangiogenesis. Of note, the inhibitory effect of Gal-8N on VEGF-C-induced LEC sprouting is bell-shaped, which is similar to several other anti-(lymph)angiogenic molecules such as RGD-mimetic integrin inhibitors (ref. 50). Not surprisingly, much higher concentration of galectin-8 (0.75 mM), compared with VEGF-C (2.38 nM), was required to produce equivalent LEC sprouting. This is because generally, the affinity of CRD of galectins towards their glycan ligands is lower (dissociation constant:BmM) compared with typical protein À protein interaction (dissociation constant: B10 nM) 51 . Despite the weak affinity of their CRD, galectins achieve a stable interaction with their ligands due to their multivalency that results in overall high avidity 52 . Therefore, even if the affinity of one galectin-8 molecule for one PDPN molecule is weak, the overall high avidity is able to activate lymphangiogenesis pathway. Our findings that PDPN binds to galectin-8 in a carbohydratedependent manner, that it contains the high affinity glycans of galectin-8 (a2,3-sialylated glycans), that galectin-8 clusters PDPN on cell surface, and that unlike the knockdown of VEGFR-3, knockdown of PDPN abrogates galectin-8-induced LEC sprouting suggest that PDPN is a key player in the mechanism of galectin-8-induced LEC sprouting. Another major finding of the current study is that PDPN plays a critical role in VEGF-Cmediated lymphangiogenesis. Thus far, VEGF-C-and PDPNmediated pathways have been independently shown to promote lymphangiogenesis, but the relationship in the molecular mechanism of the two pathways has not been demonstrated. Overall, our findings suggest that a galectin-8-dependent cross-talk among VEGF-C, PDPN, and integrin pathways plays a critical role in lymphangiogenesis. This is an important conceptual advance in the understanding of the molecular mechanism of a well-known VEGF-C lymphangiogenic pathway.
Various studies have demonstrated the pathological contribution of lymphangiogenesis to diseases of the eye 2-7 . Specifically, corneal lymphatics play a vital role in the pathogenesis of graft rejection 3,37,38,53-55 , herpetic keratitis 7,56 , dry eye disease 4 , ocular allergy 5 and wound healing 57 . In the present study, we demonstrated that: (i) a dominant negative inhibitor of galectin-8 as well as the pan inhibitor of galectins dampen lymphangiogenesis; (ii) in the mouse model of corneal allogeneic transplantation, galectin-8-induced lymphangiogenesis is associated with an increase in corneal graft rejection; and that (iii) in the mouse model of HSV keratitis, corneal pathology and lymphangiogenesis is ameliorated in galectin-8 knockout mice. These findings have broad implications for developing novel therapeutic strategies for conditions resulting from pathological lymphangiogenesis in both ocular diseases mentioned above as well as nonocular diseases such as cancer metastasis and solid organ transplant rejection. Indeed, the avascular cornea serves as an excellent in vivo model to study mechanisms of hem-and lymph-angiogenesis that are also relevant to nonocular diseases. For example, based on the findings that lymphatics in the cornea promotes graft rejection, subsequent studies reported that lymphangiogenesis also occurs in solid organ grafting, such as renal and cardiac transplantation [58][59][60] . Undeniably, corneal micropocket assays have proven to be extremely valuable in revealing the mechanism of angiogenesis in cancer and many other nonocular tissues. In conclusion, our study offers a new perspective on how glycans of the cell surface receptors can be exploited to understand and modulate the process of lymphangiogenesis.

Methods
Study approval. All animal procedures were approved by the Institutional Animal Care and Use Committee at Tufts University and were performed in accordance with the regulations of Association for Research in Vision and Ophthalmology Resolution on the Use of Animals in Vision Research and recommendations of the National Institutes of Health Guide for the Care and Use of Laboratory Animals. Tufts University Institutional Review Board/ethics committee approval was obtained for human specimen acquirement for this study.
Mice. Lymphatic-specific Prox1-EGFP reporter mice (FVB background) 61 were purchased from Mutant Mouse Regional Resource Centers, FVB/NCrl mice were purchased from Charles River Laboratories, and C57BL/6 mice were purchased from Jackson Laboratory. Mice with inducible deletion of PDPN (Pdpn f/f ;CagCre) and WT littermates (Pdpn f/w ;CagCre) in mixed background (C57BL/6 and 129/Sv) were generated as previously described 49 . PDPN deletion was accomplished by administering tamoxifen orally (20 mg each day) from P1 to P6. After weaning, the mice were orally administered 1 mg tamoxifen weekly. The Lgals8 KO mouse strain used for this study was created from embryonic stem cell clone (14305A-F8), obtained from the KOMP Repository (www.komp.org) and generated by Regeneron Pharmaceuticals, Inc 62 . The Lgals8 coding region deletion was achieved by LacZ (bacterial b-galactosidase) reporter gene replacement of chromosome 13 from 12,440,786 (deletion start) to 12,459,212 (deletion end). Details of the primer sequences and predicted PCR products are available at the Velocigene website (www.velocigene.com/komp/detail/14305). The galectin-8 null status of the KO mice is confirmed by western blotting (Supplementary Fig. 11). The Lgals8 KO mice have no obvious defects in lymphatic vessel development examined by gross morphological analysis.
Expression and purification of recombinant proteins. All recombinant human galectins were expressed in the Escherichia coli expression system.
Mouse models of corneal chemical injuries. To determine the expression pattern of galectin-8 in chemical injury, C57BL/6 mice were injured by silver nitrate and sodium hydroxide solution, and frozen sections of the eyes were subjected to immunofluorescence staining using anti-galectin-8 antibody as described above. For silver nitrate cautery, silver nitrate applicators (Grafco) were applied on the central cornea of the right eye for 5 s under a surgical microscope. The corneas were rinsed with 2 ml of PBS, and ophthalmic antibiotics were topically applied to the operated eyes. For alkaline burn, sodium hydroxide solution (1.5 ml of 0.15N) was applied to the central corneas of the right eye of each animal for 1.5 min and was immediately rinsed away with PBS 64 . The corneal and limbal epithelium was removed by an Algerbrush (Ambler surgical, Exton, PA). Ophthalmic antibiotic ointment was topically applied to the operated eyes to prevent infection. At the end of the experiment (day 7 post-silver nitrate cautery and day 14 post-alkaline burn), mouse corneas were harvested and embedded in OCT compound.
Knockdown of target genes. ON-TARGET plus human VEGFR-3, PDPN and Nrp2 siRNA SMART pool was purchased from GE Dharmacon. Oligonucleotide siRNA duplexes targeting PDPN and AllStars Negative control siRNA were purchased from Qiagen. Hs_PDPN_1 and Hs_T1A-2_7 siRNA were designated as PDPN siRNA1 and siRNA4, respectively. Similar results were obtained using PDPN siRNA from GE Dharmacon and Qiagen. The transfection of siRNA in primary LECs (Lonza) was carried out with the Lipofectamine 2000 reagent (Invitrogen). Briefly, 3 ml of Lipofectamine 2000 in 250 ml of Opti-MEM medium (Invitrogen) and 3 ml of siRNA (20 mM) in 250 ml of Opti-MEM medium were incubated separately for 5 min at 25°C, and the two mixtures were combined and incubated for an additional 20 min to form Lipofectamine 2000-siRNA complexes. At the end of the incubation period, serum-free Opti-MEM (1.5 ml) was added to each well of a six-well plate and 500 ml aliquots of Lipofectamine 2000-siRNA complex were added to each well. Final concentration of siRNA was 30 nM. After 3 h incubation, media were replaced with complete EGM-2MV medium (Lonza).
The same procedure was repeated on the next day and knockdown efficiency was assessed by Western blot after 48 h transfection.
Galectin-8 immunohistochemistry staining of corneas. Paraffin sections of normal and inflamed corneas were obtained from the archived tissues of the Tufts Medical Center Ophthalmic Pathology Laboratory Tissue sections of: (i) normal human corneas were from eyes which were enucleated due to choroidal melanoma (N ¼ 2) and uveal malignant melanoma (N ¼ 2); and (ii) inflamed human corneas were obtained at the time of keratoplasty from patients with failed corneal graft (N ¼ 6), bacterial keratitis (N ¼ 4) and Acanthamoeba keratitis (N ¼ 2). Tissue sections of normal and inflamed corneas were processed for immunolocalization of galectin-8 using a procedure we have described earlier 65 . Briefly, paraffinembedded sections were deparaffinized, rehydrated and incubated with rabbit anti-galectin-8 primary antibody (1:100 in 1% BSA/PBS, 1 h, 37°C, Novus) and a biotinylated secondary antibody (1:300 in 1% BSA/PBS, 30 min, 37°C, R&D Systems). Sections were subsequently incubated with HSS-HRP (30 min, 25°C, R&D Systems), and a DAB/AEC chromogen solution (37°C, R&D Systems). Images were acquired by EVOS XL Core cell imaging system (Invitrogen).
Corneal mouse micropocket lymphangiogenesis assay. The corneal micropocket lymphangiogenesis assay was performed as described previously using implants containing a test agent, hydron and sucralfate 66,67 . Test agents included full-length galectin-8 (40-320 ng per pellet) and VEGF-C (160 ng per pellet). Implants containing hydron and sucralfate alone served as negative controls. The mice were anaesthetized by intraperitoneal injection of a cocktail of ketamine (90-120 mg kg À 1 ) and xylazine (10 mg kg À 1 ). The eyes were topically anaesthetized with proparacaine and were gently proptosed with forceps. Using a corneal blade and a stereoscope, intrastromal linear keratotomy was performed about 2 mm from the limbus. Using a von Graefe knife (Miltex), a pocket was extended towards the limbus, and the pellet was manoeuvered into the pocket. The wound was coated with a veterinary ophthalmic ointment (Akorn) to prevent infection. Mouse corneas were harvested 7 days after pellet implantation, fixed in 4% paraformaldehyde/PBS (1 h at 4°C), washed with PBS, and fixed again in iced acetone (15 min at À 20°C). To quantitate the extent of lymphangiogenesis in Prox1-EGFP reporter mice, flat mounts of the dissected corneas were evaluated by fluorescence microscopy. Fluorescent images were acquired by EVOS FL cell imaging system (Invitrogen), and vessel areas were calculated using the formula 67 : vessel area ¼ pellet distance Â vessel length Â clock hours Â 0.2p. To quantitate the extent of lymphangiogenesis in WT and PDPN-deficient mice, corneas were stained with eFluor 570-anti-mouse LYVE-1 (clone ALY7, eBioScience, and 1:75) in 10% goat serum/0.2% Triton X-100/PBS overnight, 4°C. After several washes with 0.2% Triton X-100/PBS, the corneas were flattened and mounted with VECTASHIELD mounting medium (Vector Laboratories) and evaluated by fluorescence microscopy. In some experiments, for comparison purposes, corneas were stained with Alexa Fluor 488-anti-mouse CD31 (clone MEC13.3, 1:100, BioLegend) to visualize blood vessels.
Corneal transplantation. The mouse model of corneal transplantation was used as previously described 68 . In brief, male BALB/c mice (10-week old) were used as graft recipients and male C57BL/6 mice (10-week old) were used as donors. The donor corneal grafts (2.0 mm) were sutured into 1.5 mm diameter corneal host beds of BALB/c mice recipients with 8-10 interrupted 11-0 nylon sutures. At the conclusion of surgery, the viscoelastic material was exchanged for PBS. Neomycin and polymyxin B sulfates and bacitracin Zinc Ophthalmic Ointment USP (Akorn) were topically applied, and a single eyelid suture (8-0 nylon, Surgical Specialties) was performed. Both eyelid and cornea sutures were removed 1 week after surgery. The mice were treated with PBS or galectin-8 (10 mg for subconjunctival injection and 50 mg for intraperitoneal injection) twice a week beginning postoperative day 7. Grafts were evaluated for signs of rejection by slit-lamp biomicroscopy (SL-D7, Topcon, Tokyo, Japan) once per week. A previously reported scoring system 68,69 was used to grade the degree of corneal clarity, ranging from 0 to 5 þ (0, clear graft; 1 þ , minimal superficial opacity; 2 þ , mild stromal opacity with pupil margin and iris vessels visible; 3 þ , moderate stromal opacity with only pupil visible; 4 þ , intense stromal opacity with the anterior chamber visible; 5 þ , severe stromal opacity with total obscuration of the anterior chamber). Grafts with an opacity score of 3 þ or greater on postoperative week 4 were considered rejected; grafts with opacity scores of 3 þ or greater on postoperative week 2 which never cleared were also considered rejected. Mice were killed on postoperative week 4 and corneas were excised for whole-mount staining to visualize and quantify lymphatic and blood vessels.
Statistics. Data in all figures are presented as mean ± s.e.m.. All results were confirmed in 2 or more independent experiments. Data were analysed using paired two-tailed Student's t-test or one-way analysis of variance in Prism 6 (GraphPad) as indicated in figure legends. P value o0.05 was considered statistically significant.