Chondroitin sulfate enhances the barrier function of basement membrane assembled by heparan sulfate

Glycosaminoglycans (GAGs) are a family of ubiquitously expressed polysaccharides that play important roles in development and physiology (Bishop et al., 2007; Mikami and Kitagawa, 2013; Mizumoto et al., 2013). Based on their composition, these macromolecules are classified into heparan sulfate (HS), chondroitin sulfate (CS), dermatan sulfate (DS) and keratan sulfate (KS). GAG synthesis requires cascades of enzymes, including enzymes involved in the generation of nucleotide sugar precursors and enzymes, such as Chsy/Chpf and Ext1/2 that assemble the backbone of CS/DS and HS chains, respectively (Fig. 1A; Mikami and Kitagawa, 2013; Mizumoto et al., 2013). During assembly of these chains, GAGs undergo various epimerization and sulfation reactions. In CS/DS, the sulfation reactions are catalyzed by members of the Chst family of enzymes, which add sulfate groups to the C4 and/or C6 positions of N-acetylgalactosamine residues and to iduronic acid residues in DS. In HS, Ndst isozymes catalyze the initial N-deacetylation and N-sulfation of the N-acetylglucosamine residues, which is then followed by the addition of sulfate groups to the C6 and C3 positions of glucosamine residues and to the C2 of uronic acids. The extent of these modifications varies, creating enormous structural complexity, which in turn tunes the selectivity of protein binding and subsequent biological activity.

Retinal angiogenesis is a prime example that illuminates the diverse functions of GAGs in development. In mice, formation of the retinal vasculature requires sequential migration of astrocytes and endothelial cells into the neonatal retina; both processes are dependent on retinal GAGs (Tao and Zhang, 2014). We have previously examined the impact of altering the enzyme Ugdh, which is responsible for the formation of UDP-glucuronic acid required for the assembly of CS/DS, HS and hyaluronan. Inactivation of Ugdh disrupted the inner limiting membrane (ILM) of the retina, which serves as the substratum for astrocyte migration (Tao and Zhang, 2016). HS has also been shown to interact with vascular endothelial growth factor (Vegf) produced by astrocytes, generating localized Vegf concentration gradients to attract the migration of endothelial cells into the retina (Ruhrberg et al., 2002; Stalmans et al., 2002). Mutation of the HS interaction motif in Vegfa leads to severe defects in vascular outgrowth and patterning. Therefore, migration of retinal astrocytes and endothelial cells requires the distinct functions of GAGs in basement membrane and VEGF signaling.

In this study, we further explored the roles of GAGs in retinal angiogenesis and astrocyte migration by genetically disrupting the biosynthesis of HS and CS. Surprisingly, unlike the profound angiogenesis defect observed in Vegfa mutants lacking HS interactions, deletion of the HS polymerase Ext1 in the retina failed to perturb the initial expansion of the vasculature. Instead, loss of retinal HS impaired only vessel remodeling and maturation. On the other hand, the ILM was disrupted by deletion of Ext1 in a time-dependent manner, suggesting that HS is required for the initial assembly but not for the maturation of the ILM. Lastly, we show that loss of the CS sulfation enzyme C4st1 (Chst11) did not cause an obvious retinal phenotype, but it synergized with Ext1 deletion in breaching the barrier function of the ILM and retinal glycocalyx. These results reveal the distinct requirements of HS and CS in the assembly and function of the retinal extracellular matrix (ECM).

Despite the widely accepted role of HS in VEGF signaling, the function of retinal HS in angiogenesis has not been directly tested. To this end, we abolished HS synthesis by ablating the HS polymerase gene Ext1 using α-Cre, which is expressed in the peripheral retina beginning at embryonic day (E) 10.5 (Cai et al., 2011; Marquardt et al., 2001). We first performed the ligand and carbohydrate engagement (LACE) assay to confirm the loss of HS. In this experiment, the presence of HS on tissue sections was detected by exogenously applied FGF10 and FGFR2B protein, which together bind avidly and specifically to sulfated HS (Pan et al., 2008). As expected, the LACE signal was observed throughout the control eye at postnatal day (P) 0, most prominently in the ILM overlying the retina (Fig. 1B, arrowheads). In contrast, α-Cre;Ext1^flox/flox (Ext1^ΔRet) mutants displayed a specific loss of LACE signals in peripheral retinae (Fig. 1B, arrow), which corresponded to the known pattern of α-Cre expression. Interestingly, the LACE staining was reduced but not eliminated in the ILM above the area lacking Ext1 expression (Fig. 1B, arrowheads), whereas the staining of perlecan, a major heparan sulfate proteoglycan (HSPG) found in the basement membrane, was unaffected. The integrity of the ILM was further confirmed by the continuous staining of laminin in Ext1^ΔRet mutants. Even in P30 Ext1^ΔRet mutants, we failed to observe retinal ectopias, which are usually associated with retinal cells bulging through defective ILM into the vitreous (Fig. 1C, arrows and arrowheads). These results suggest that genetic ablation of Ext1 abrogated HS synthesis in the retina without abolishing ILM formation.

We next examined whether depletion of HS affected retinal angiogenesis, which begins with IB4-positive endothelial cells emerging from the optic disc in the neonatal mouse before spreading to the periphery of the retina (Fig. 2A). Between control and Ext1^ΔRet mutants, we did not observe any statistically significant difference in the outgrowth of retinal vasculature. Nevertheless, the pruning of nascent vessels was significantly altered in Ext1^ΔRet mutants, as shown by the increasing numbers of empty vessel sheaths that were positive for the basement membrane marker collagen type IV (Col IV), but negative for the endothelial marker IB4 (Fig. 2B, arrows). Smooth muscle actin (SMA) staining further revealed fewer arterioles and venules in 3-week-old Ext1^ΔRet mutant retinae compared with wild-type retinae (Fig. 2B). These results showed that the retinal Ext1 was dispensable for initial retinal angiogenesis but necessary for remodeling the vasculature.

The relatively mild vasculature defects in Ext1^ΔRet mutant retinae raise questions as to the role of HS in VEGF signaling. In addition to inducing angiogenesis, VEGF signaling also promotes maturation of blood vessels, making them less vulnerable to a hyperoxic environment. To test this function of VEGF signaling in Ext1^ΔRet mutants, we moved P9 pups from the room atmosphere and housed them in a 75% O₂ environment for 7 days (Fig. 3A). In control animals, this treatment led to an expected loss of capillaries in central retinae, leaving major vessels intact (Fig. 3B). In contrast, Ext1^ΔRet mutants exhibited a far more severe vessel reduction, extending to include the peripheral retinae, and even included arterioles and venules. However, when we postponed the hyperoxia treatment until P24, both control and Ext1^ΔRet mutant retinae were able to maintain an intact vessel plexus, consistent with the eventual maturation of their retinal vasculatures. Therefore, the retina-specific deletion of Ext1 delayed but did not prevent vessel maturation. We reasoned that if the delayed vessel maturation in Ext1^ΔRet mutants was due to impaired VEGF signaling, it may be ameliorated by exogenous VEGFA. To test this hypothesis, we injected VEGFA into the vitreous of the eye at P9 prior to the hyperoxia treatment, which, unlike VEGFA injection at P7, did not substantially affect the extent of vessel ablation in wild-type animals (Alon et al., 1995). In Ext1^ΔRet mutants, however, intravitreal injection of VEGFA led to a significant rescue of microvasculature compared with the saline-injected controls (Fig. 3B). These results support that subtle impairment of VEGF signaling leads to the vasculature defects in Ext1^ΔRet mutants.

We have previously reported severe astrocyte migration defects in α-Cre;Ugdh^flox/flox mutants, in which biosynthesis of HS and CS was abolished (Tao and Zhang, 2016). This prompted us to examine whether similar defects could be observed in HS mutants. By P4, astrocytes had reached the peripheries of control retinae, but the migratory front of astrocytes appeared uneven in Ext1^ΔRet mutants (Fig. 4A, dotted lines). As shown by Pax2 staining in control retinae, the nuclei of migrating astrocytes typically displayed spindle shapes; however, in Ext1^ΔRet mutants, astrocyte nuclei were more rounded (Fig. 4A, insets). In contrast, the cell bodies of astrocytes labeled by Pdgfra were hyperplastic in Ext1^ΔRet mutants, resulting in a denser astrocytic network. At the end of migration, maturing astrocytes wrap around IB4-positive blood vessels and express the glial cell marker GFAP, but heightened expression of GFAP is also a hallmark of gliosis and Müller cell activation (Yang and Wang, 2015). Compared with the uniform expression of GFAP in the control retina at P17, Ext1^ΔRet mutants displayed significant increases in GFAP expression at the peripheries of retinae (Fig. 4A), suggesting that astrocytes in these regions were under stress.

The HS polymer generated by the Ext1 enzyme can be further modified by a number of HS sulfotransferases at N-, 2-O, 3-O or 6-O positions of the glucosamine residues (Bishop et al., 2007). To investigate the role of HS sulfation in the development and distribution of retinal astrocytes, we first genetically ablated the HS N-deacetylase and N-sulfation genes Ndst1 and Ndst2 in the retina. As shown by the LACE assay, sulfated HS was indeed depleted in α-Cre;Ndst1^flox/flox;Ndst2^−/− mutants, leading to an uneven migration of astrocytes in peripheral retinae similar to that seen in Ext1^ΔRet mutants (Fig. 4B, dotted lines). Hyaloid vessels constitute the embryonic vasculature in the vitreous that normally regresses after birth (Lang, 1997). However, α-Cre;Ndst1^flox/flox;Ndst2^−/− mutants exhibited persistent hyaloid vessels, marked by strong LACE staining, that became attached to HS-deficient retinae (Fig. 4B, arrows). We next deleted the HS uronyl 2-O sulfotransferase gene Hs2st (also known as Hs2st1) and HS glucosaminyl 6-O sulfation genes Hs6st1 and Hs6st2. Similar to α-Cre;Ndst1^flox/flox;Ndst2^−/− mutants, we observed an uneven migration of astrocytes and persistent hyaloid vessels in the peripheral retinae of α-Cre;Hs2st^flox/flox;Hs6st1^flox/flox;Hs6st2^−/− mutants. These subtle but consistent defects suggest that sulfated HS in the retina regulates astrocyte development and migration, as well as hyaloid vessel clearance.

The ILM is essential for astrocyte migration and preventing abnormal penetration of hyaloid vessels into the retina (Edwards et al., 2010; Gnanaguru et al., 2013). Although the immunostaining of retinal sections did not reveal obvious defects in the ILM (Fig. 1B), we re-examined ILM integrity more thoroughly by whole-mount staining. In the majority of the P7 Ext1^ΔRet mutant retinal regions, the ILM appeared as a smooth layer of laminin overlying the IB4-positive blood vessels (Fig. 5A). In the far periphery of mutant retinae, however, we observed scattered holes in the ILM (Fig. 5A, arrowheads). Moreover, these holes were often penetrated by residual hyaloid vessels (Fig. 5A, arrow), similar to what we observed in the peripheral retinae from mutants lacking the HS N-sulfation and O-sulfation genes. The hyaloid vessel persistence and ILM fragmentation remained in P28 Ext1^ΔRet mutant retinae, demonstrating that the defective ILM was not repaired in adult animals.

Our data revealed that the ILM defects in Ext1^ΔRet mutants were restricted to a narrow band at the end of the retinae, whereas a wider region of HS-depleted retinae showed an intact ILM. To reconcile the discrepancy between the restricted ILM defect and the extensive HS depletion, we considered the fact that retinal development follows a central-to-peripheral pattern. We hypothesized that because of the relatively late onset of α-Cre activity and persistent activity of previously expressed enzymes, much of the HS depletion may have occurred after the formation of the ILM in the mutant retinae. If this model is correct, an early Cre driver may induce much more severe disruption of the ILM. To test this hypothesis, we replaced α-Cre with Six3-Cre, which acts 1 day earlier in the central retina at E9.5 (Cai et al., 2013; Furuta et al., 2000). In Six3-Cre;Ndst1^flox/flox;Ndst2^−/− mutants, there were indeed much larger holes in the ILM in the retina (Fig. 5B, boxes). As a result, astrocyte migration was mostly aborted, leaving only a few escaped astrocytes scattered in the retina (Fig. 5B, arrowheads). A caveat of this experiment is that HS also acts as the co-receptor for FGF signaling, which is required for development of the optic disc. However, we have previously shown that optic disc defects in both FGF receptor and Ndst mutants could be rescued by constitutively active Kras signaling (Cai et al., 2014). We therefore stimulated Kras signaling by inducing the oncogenic Kras^LSL-G12D allele and again observed extensive ILM breaches in Six3-Cre;Ndst1^flox/flox;Ndst2^−/−;Kras^LSL-G12D mutants (Fig. 5B). There were only a small number of astrocytes and associated endothelial cells in the retina, consistent with the essential roles of the ILM and HS in astrocyte migration and angiogenesis.

Thus far, study of these HS biosynthesis mutants has revealed a narrow time window for HS to affect ILM development, suggesting that HS may be required for the initial assembly of the ILM but not its later maintenance. To test this model, we turned to an in vitro basement membrane assembly assay (Colognato et al., 1999). In this experiment, C2C12 myoblast cells were induced to form myotubes after replacing fetal bovine serum (FBS) with horse serum in the media (Fig. 5C, arrows). Although myotubes themselves expressed little endogenous laminin, they were able to assemble de novo basement membrane using exogenously provided laminin (Fig. 5C, arrowheads). However, if the cell surfaces of these myotubes were first stripped of HS by heparin lyase treatment, laminin polymerization was no longer detected. This result shows that the cell-surface HS plays an important role in the assembly of the basement membrane.

In comparison with the HS biosynthesis mutants described above, our previously studied α-Cre;Ugdh^flox/flox mutants displayed more extensive ILM defects (Tao and Zhang, 2016). One possible reason is Ugdh knockout abolished biosynthesis of both HS and CS, which may have a synergistic effect on the integrity of the ILM. To investigate this possibility, we ablated the CS 4-O-sulfation enzyme C4st1 (also known as Chst11). α-Cre;C4st1^flox/flox (C4st1^ΔRet) mutants did not show any obvious abnormalities in astrocyte migration and angiogenesis (Fig. 6A). As shown by laminin staining, even the combined deletion of Ext1 and C4st1 in α-Cre;Ext1^flox/flox;C4st1^flox/flox (Ext1;C4st1^ΔRet) mutants failed to generate more severe ILM defects compared with those of Ext1^ΔRet mutants (Fig. 6A, arrows).

As the deletion of C4st1 is not expected to disrupt CS function entirely, we considered the possibility that its effect on the ILM might be relatively subtle. A sensitive test for the integrity of the ILM is intravitreal injection of adeno-associated virus AAV5, which can robustly infect the retina if there are any breaches in the ILM (Dalkara et al., 2009). To detect the virally infected regions, we used a pseudotyped AAV2/5 virus that manifested the tropism of AAV5 but carried an AAV2 genome encoding the mCherry reporter. After injection into the vitreous, we detected the viral mCherry only at injection sites in both control and C4st1^ΔRet animals (Fig. 6B, arrows), demonstrating that the barrier function of their ILMs was intact. In contrast, Ext1^ΔRet mutants displayed mCherry expression in peripheral retinae (Fig. 6B, dotted lines), where we previously detected obvious holes in the ILM by laminin staining. Importantly, mCherry-expressing domains were significantly enlarged in Ext1;C4st1^ΔRet mutants compared with those in Ext1^ΔRet mutants, which matches the α-Cre deletion pattern in the retina. These results showed that Ext1 and C4st1 both contribute to the barrier function of the ILM.

We next asked whether the increased permeability in Ext1;C4st1^ΔRet mutant retinae was caused by structural abnormalities in the ILM. Transmission electron microscopy (EM) showed a continuous sheet of basement membrane beneath the control retinae at P4 (Fig. 7A, arrowheads). In contrast, there were occasional gaps in the basement membranes of the Ext1;C4st1^ΔRet mutant (Fig. 7A, arrow), which was consistent with sporadic holes in the ILM revealed by our previous laminin staining (Fig. 6A). Interestingly, lower magnification images also revealed a layer of intermediate electron density, which separated the basement membrane from the vitreous (Fig. 7A, blue bars). In Ext1;C4st1^ΔRet mutants, this band was not only reduced in width, but also appeared paler and less dense. Because we used an en bloc fixation and staining protocol that also highlighted glycoproteins in our EM analysis (Chuang et al., 2015; Tapia et al., 2012), this result suggested that retinal deletion of Ext1 and C4st1 may have led to the depletion of proteoglycans in this vitreoretinal interface.

Our ultrastructure analysis showed that the thinning of the vitreoretinal interface was much more prevalent than the sporadic breaches in the basement membrane, resembling the continuous pattern of viral infection in Ext1;C4st1^ΔRet mutant retina. This suggested that the increased permeability in the Ext1;C4st1^ΔRet mutant may primarily be caused by disruption of this vitreoretinal interface, not just breaches in the basement membrane. We predicted that if we removed GAGs after the formation of the ILM, this increased permeability would still promote virus infection in the retina. To test this hypothesis, we injected the AAV2/5 virus with heparin lyase and chondroitin ABC lyase (ChABC), which have been shown to degrade HS and CS in the vitreous without breaking the retinal basement membrane (Balasubramani et al., 2010), into the eyes of 1-month-old animals (Fig. 7B). Consistent with our previous results, injection of the AAV2/5 virus alone induced mCherry expression only in the injection site (Fig. 7B, arrow), but when the AAV2/5 virus was co-injected with glycolytic enzymes, mCherry expression was observed in the entire retina. These data support an important role for GAGs in maintaining the retinal barrier.

In this study, we have explored the distinctive functions of HS and CS in the neural retina. By deleting the HS polymerase Ext1 in retinal progenitor cells, we demonstrate that retinal HS is dispensable for VEGF-dependent migration of endothelial cells, but it regulates the subsequent pruning and maturation of the retinal vasculature. We also show that retinal HS is required in a narrow time window for the assembly of the ILM, which is necessary for the migration of astrocytes into the retina. Lastly, we present evidence that deficient CS sulfation enhances the retinal permeability caused by HS depletion, demonstrating an important role for retina-derived CS in maintaining the barrier function. These results shed light on the multifaceted function of GAGs in growth factor signaling and ECM integrity in the retina.

It is notable that our genetic ablation of Ext1 failed to perturb the progression of the vascular front during postnatal retinal development, yet led to a more severe oxygen-induced retinopathy (OIR) phenotype with extensive capillary loss in response to hyperoxia. At first glance, this result may appear to contradict the widely accepted model that Vegfa requires interaction with the cell-surface HS to maintain a chemo-attractive gradient for angiogenesis (Ruhrberg et al., 2002; Stalmans et al., 2002). However, it is important to consider that our conditional knockout of Ext1 was restricted to the neural retina, leaving the lens and the optic stalk intact as potential sources of exogenous HS. Halfter and colleagues have previously performed the elegant mouse/chick transplant experiment to show that the lens synthesizes many crucial components of the ILM, including HSPGs, perlecan and collagen XVIII, whereas only agrin is expressed solely by the neural retina (Halfter et al., 2000). This explains why reduced HS staining persisted in the ILM above Ext1 mutant retinae. In addition, the optic stalk is the source of retinal astrocytes, which migrate ahead of endothelial cells on top of the retina and express their own proteoglycans. In fact, it has previously been reported that an astrocyte-specific knockout of Ext1 slowed down the radial expansion of the endothelial network by dampening VEGF signaling (Stenzel et al., 2011). This is in contrast to our retina-specific Ext1 knockout, which instead disrupted the VEGF-dependent pruning and maturation of the vasculature. Therefore, the HS expressed by different ocular tissues cooperates to regulate distinctive aspects of VEGF function, orchestrating the fine-tuning of vascular development. Our findings may also have clinical significance in terms of the stability of the nascent retinal vasculature, and suggest that defects in HS expression in the retina may lead to increased vulnerability to retinopathy in premature infants (e.g. retinopathy of prematurity) without showing other obvious abnormalities in vascular development.

Another intriguing finding in our study is that α-Cre-induced abrogation of the HS polymerase gene Ext1 disrupted ILM formation, but the breach was restricted to the far periphery of the retina. This led to abnormal migration of astrocytes in this region and penetration of hyaloid vessels into the retina, phenotypes also observed in HS N- and 6-O sulfation mutants. The limited scope of the ILM defect was not due to the lack of gene inactivation, as LACE staining on sections revealed a much larger region of HS depletion in Ext1^ΔRet mutant retinae. Instead, we note that retinal development occurs in a center-to-periphery pattern. As a result, when Halfter and colleagues performed a pulse-chase experiment to study ILM regeneration in vivo, they found that the newly synthesized ILM was present only in the distal retina (Halfter et al., 2000). It is conceivable that, owing to the relatively late onset of the α-Cre driver and the persistence of residual Ext1 protein, the eventual arrest of HS biosynthesis in our Ext1^ΔRet mutants coincided with the time of ILM assembly only at the far periphery of the retina. In this scenario, even if HS was lost in the more central retina already covered by the ILM, only the nascent ILM above the distal retina would be affected. This suggests that the HS derived from the neural retina is required only for the initial assembly of the ILM and not for its later maintenance. This model further predicts that the ILM could also become vulnerable to HS depletion in the central retina, provided that the onset of genetic knockout was early enough. Indeed, when we switched to an earlier retinal driver (Six3-Cre) to delete Ndst genes, we observed much larger and more numerous ILM holes in the central retina. Moreover, we showed in an in vitro assay that the degradation of HS prevented myotubes from promoting laminin polymerization, suggesting a general role of cell-surface HS in basement membrane assembly. This role of cell surface HS is in contrast with the roles of the secreted HS proteoglycans perlecan and collagen XVIII, which are constitutional components of the basement membrane necessary for its stability (Yurchenco and Patton, 2009). These results highlight that HS plays tissue-specific roles in the assembly and maintenance of the basement membrane.

Although we have thus far attributed the phenotype of Ext1 knockout to the loss of HS, we also need to consider the potential gain-of-function effect of CS. It has been shown that CS biosynthesis may be upregulated in the absence of HS (Bachvarova et al., 2020). In addition, we have previously reported that deletion of the GAG biosynthesis gene Ugdh resulted in more severe ILM defects than Ext1 knockout, which could be due either to differences in the enzymatic kinetics between Ugdh and Ext1 or to compensation by hyaluronan and CS. To test the latter possibility, the ideal approach would be genetic ablation of CS biosynthesis. However, unlike HS in which chain elongation can be blocked by the removal of a single enzyme (Ext1), there are four CS co-polymerase genes in the mouse genome that act redundantly, making it daunting to abrogate CS biosynthesis genetically (Bishop et al., 2007; Mikami and Kitagawa, 2013; Mizumoto et al., 2013). Thus, we opted to remove C4st1, which catalyzes the important 4-O sulfation of CS, with the understanding that biosynthesis of DS will be unaffected and that the function of CS will be at most partially impaired because C4st1 is not the only 4-O-sulfotransferase acting on CS. These reasons may explain the lack of an obvious phenotype in our C4st1 mutants. Although the compensatory roles of hyaluronan, DS and other C4st genes await further study, using the sensitive virus infection assay, we found that loss of C4st1 enhanced the permeability of the ILM caused by Ext1 deficiency, revealing an important role for HS and CS in establishing the neuroretinal barrier.

How do HS and CS regulate the retinal permeability? As integral parts of the basement membrane, GAGs may play structural roles in stabilizing the ILM. However, ultrastructure analysis of our GAG mutants also revealed that a band of intermediate electron-dense materials above the basement membrane was significantly reduced. Previous studies have shown that the retinal ECM in this region is very labile (Heegaard et al., 1986; Rhodes, 1982), which may explain why it is usually missed by regular EM analysis. In contrast, we readily detected it using an EM fixation and staining protocol optimized for glycoproteins, suggesting that it is enriched in proteoglycans. This region of the retinal ECM thus resembles the glycocalyx covering endothelial cells, a gel-like coating of carbohydrates that functions as a barrier for the vasculature. Indeed, previous studies have shown that intravitreal injections of heparin and chondroitin lyases significantly increased transduction of AAV2 into the retina (Cehajic-Kapetanovic et al., 2018), although the interpretation of this result was complicated by the fact that AAV2 uses HSPG as the receptor for cellular entry. Using the pseudotyped AAV2/5 virus, we confirmed that degradation of HS and CS enhanced retinal permeability. Together, these results showed that both the basement membrane and the glycocalyx are important for maintaining the retinal barrier. Breach of the retinal barrier is the hallmark of retinopathy of prematurity, diabetic retinopathy and proliferative vitreoretinopathy. With deeper understanding of the biological function of the retinal barrier and innovative strategies to repair or stabilize its structure, the retinal basement membrane and glycocalyx may present promising therapeutic targets for preventing or curbing the harmful effects of these diseases.

Ndst1^flox and Hs2st^flox mice have been previously reported (Grobe et al., 2005; Stanford et al., 2010). Chst11^f/f mice were generated from Chst11^{tm1a(KOMP)Wtsi} chimeric mice (Mutant Mouse Regional Resource Centers, MMRRC) after crossing to Flp mice. Hs6st1^flox mice were a kind gift from Dr Wellington V. Cardoso (Columbia University, New York, NY, USA) (Izvolsky et al., 2008). Ndst2^KO mice were a kind gift from Dr Lena Kjellén (University of Uppsala, Sweden) (Forsberg et al., 1999). α-Cre and Six3-Cre mice were kindly provided by Drs Ruth Ashery-Padan (Tel Aviv University, Israel) and Yasuhide Furuta (M.D. Anderson Cancer Center, Houston, TX, USA), respectively (Furuta et al., 2000; Marquardt et al., 2001). Hs6st2^KO and LSL-Kras^G12D mice were obtained from MMRRC and the Mouse Models of Human Cancers Consortium (MMHCC) repository at the National Cancer Institute, respectively (Tuveson et al., 2004). All mice were maintained in mixed genetic background. The floxed animals that do not carry Cre transgene were used as controls. All animal procedures were performed according to the protocols approved by the Columbia University's Institutional Animal Care and Use Committee and conform to the relevant regulatory standards.

Neonatal mice and their nursing mother were kept in the room atmosphere from birth through P9 or P24 before being placed in a hyperoxia chamber and exposed to 75% oxygen for 7 days. Throughout this period, the P9 pups were housed with their nursing mother and the oxygen levels were monitored electronically. Neonatal pups (P9) were anesthetized by hypothermia over ice for a few minutes. The skin over the eye was cleaned with 70% ethanol using a cotton swab and the eyelid was opened by gently cutting along the scar tissue of the prospective eyelid with a sharp 30-gauge needle. The skin was pushed away from the eye using sterile forceps, and 0.7 μl of VEGFA (100 ng/μl, #293-VE, R&D Systems) solution was injected slowly into the vitreous by inserting the glass needle at a 45° angle though the corneoscleral boundary. The eyelids were closed using cotton buds and the pups were placed on a heating mat at 37°C to recover, before they were returned to their mother.

Adult mice were anesthetized by intraperitoneal injection of ketamine/xylasine (071069 and 061035, Covetrus) and the breathing rate and toe-pinch reflex were monitored to ensure full anesthesia. Sterile tropicamide (070498, Covetrus) and phenylephrine hydrochloride (068882, Covetrus) drops were applied to the eye, followed by proparacaine hydrochloride eye drops (068926, Covetrus) for topical anesthesia. Lubricant eye gel was applied on each eye to prevent corneal ulcers. Using a sterile acupuncture needle, a small hole was created at the corneoscleral boundary and 0.5 μl of the AAV2/5-CMV-mCherry (1×10¹² vg/ml, VVC-U of Iowa-4220, University of Iowa Viral Vector Core) solution was injected into the eye vitreous using a sharp microinjection glass needle. For co-injection with glycosidic enzymes, each eye also received one unit of heparin lyase (H3917, Sigma-Aldrich) and one unit of ChABC (C0773, Sigma-Aldrich) dissolved in 4 nM CaCl₂ and 20 mM Tris-HCl (pH 7.5). Proparacaine hydrochloride eye drops and lubricant eye gel were applied again on each eye and the mice were placed on a heating pad to recover.

Histology and immunohistochemistry were performed on the paraffin wax-embedded sections and cryosections as previously described (Carbe et al., 2013; Carbe and Zhang, 2011). Whole retina fixed in 4% paraformaldehyde (PFA) or 10 μm rehydrated cryosections were blocked with 10% normal goat serum for 1 h at room temperature and incubated with primary antibody overnight at 4°C. After washing with PBS, samples were incubated with the secondary fluorophore-conjugated antibody in 2% bovine serum albumin for 1 h at room temperature in the dark. Isolectin GS-IB₄ (IB4) conjugated with Alexa Fluor 488 (I21411, Thermo Fisher Scientific) was applied to visualize the vasculature. Samples were washed and mounted with n-propyl gallate anti-fading reagent (P3130, Sigma-Aldrich) and examined under a Leica DM5000-B fluorescence microscope. The following antibodies were used: anti-Col IV (1:1000, AB756P, Merck Millipore), anti-GFAP (1:1000, Z0334, Agilent Dako), anti-laminin (1:1000, L9393, Sigma-Aldrich), anti-Pax2 (1:200, PRB-276P, Covance), anti-Pdgfra (1:1000, 558774, BD Pharmingen) and anti-SMA (1:1000, M085129-2, Agilent). The anti-perlecan antibody (1:1000) was a kind gift from Dr Peter Yurchenco (Rutgers University, Piscataway, NJ, USA). The following secondary antibodies were used: Alexa Fluor 488 AffiniPure donkey anti-mouse IgG (715-545-151, 1:500) and Cy3 AffiniPure donkey anti-rabbit IgG (711-165-152, 1:1000), both from Jackson ImmunoResearch. All commercial antibodies are validated by vendors. At least three embryos of each genotype were stained for each marker.

The LACE assay was used to probe the in situ binding affinity of FGF-FGFR complexes to heparan sulfate on retina sections as previously described (Pan et al., 2006). Recombinant FGF10 and FGFR2B were obtained from R&D Systems.

C2C12 cells (CRL-1772, American Type Culture Collection) were authenticated and tested for contamination by the vendor and cultured in Dulbecco's Modified Eagle Medium (DMEM) with high glucose (10-013-CV, Corning) containing 10% FBS. After switching to 5% horse serum (16050130, Thermo Fisher Scientific), laminin (CB-40232, Thermo Fisher Scientific) was added at 100 µg/ml to trigger basement membrane assembly. The surface HS was removed by 1 U/ml Heparitinase I and III Blend from Flavobacterium heparinum (H3917, Sigma-Aldrich).

For the EM experiments, eyeballs were enucleated from P4 mice under room light and were fixed in 4% PFA/0.1 M cacodylic acid overnight with their corneas partially cut open. Vibratome sections (∼120 μm) were prepared from small pieces of eyecup (∼2 mm×∼3 mm) embedded in 5% agarose II/0.1 M cacodylic acid and post-fixed in 2.5% glutaraldehyde with 4% PFA/0.1 M cacodylic acid for 72 h. Sections were processed for en bloc fixation and staining, as previously described (Chuang et al., 2015). Briefly, after several washes in ice-cold 0.15 M cacodylate buffer containing 2 mM CaCl₂, the specimens were incubated with 1.5% potassium ferrocyanide, 2 mM CaCl₂ and 2% osmium tetroxide in 0.15 M cacodylate buffer (pH 7.4), for 1 h on ice, followed by treatment with 10 mg/ml of thiocarbohydrazide solution (01211, Polysciences) for 20 min at room temperature, and then fixation with 2% osmium tetroxide for 30 min at room temperature. The en bloc-stained tissues were dehydrated with graded ethanol and embedded in Epon. Ultrathin sections (72 nm) were collected on G400-Cu grids (Electron Microscopy Sciences) and were examined under a Hitachi HT7800 electron microscope for conventional transmission EM analysis.

The migratory lengths of blood vessels and the areas of retinae were measured using ImageJ. For quantification of staining, fluorescent images were converted to grayscale and mean intensities were measured using ImageJ. Statistical analysis was performed using GraphPad Prism 7. Sample sizes were not predetermined. Data represent mean±s.d. Unpaired two-tailed Student's t-test was used for comparing two conditions and one-way ANOVA with Tukey's multi-comparison test for three or more conditions.

The authors thank Peter Yurchenco for the perlecan antibody and Drs Wellington V. Cardoso, Yasuhide Furuta, Lena Kjellén, Ruth Ashery-Padan for mice. We thank one of the anonymous reviewers for suggesting EM analysis. We also thank members of Zhang lab for discussion.

The peer review history is available online at https://journals.biologists.com/dev/article-lookup/doi/10.1242/dev.200569.