Desulfation of Heparan Sulfate by Sulf1 and Sulf2 Is Required for Corticospinal Tract Formation

Heparan sulfate (HS) has been implicated in a wide range of cell signaling. Here we report a novel mechanism in which extracellular removal of 6-O-sulfate groups from HS by the endosulfatases, Sulf1 and Sulf2, is essential for axon guidance during development. In Sulf1/2 double knockout (DKO) mice, the corticospinal tract (CST) was dorsally displaced on the midbrain surface. In utero electroporation of Sulf1/2 into radial glial cells along the third ventricle, where Sulf1/2 mRNAs are normally expressed, rescued the CST defects in the DKO mice. Proteomic analysis and functional testing identified Slit2 as the key molecule associated with the DKO phenotype. In the DKO brain, 6-O-sulfated HS was increased, leading to abnormal accumulation of Slit2 protein on the pial surface of the cerebral peduncle and hypothalamus, which caused dorsal repulsion of CST axons. Our findings indicate that postbiosynthetic desulfation of HS by Sulfs controls CST axon guidance through fine-tuning of Slit2 presentation.


Results
Sulf1 and Sulf2 Play a Role in Generating Sulfation Patterns of HS In Vivo. To assess whether the sulfation patterns of HS are changed in Sulf KO brains, we first performed HS disaccharide analysis. Because Sulf1/2 DKO mice died within a day of birth for unknown reasons, we used neonatal mice. HS was digested into disaccharides by heparin lyases, and the compositions of 8 different disaccharides possessing sulfate residues at different combinations at the 2-O-, 6-O-, and N-positions were determined by HPLC analysis. The Sulf1 KO and Sulf2 KO brains showed increases in 2-O-, 6-O-, and N-sulfated disaccharides and decreases in 2-O-and N-sulfated disaccharides (Supplementary Fig. S1a-b). In the Sulf1/2 DKO brains, the changes were more than a simple additive effect of the two single mutants ( Supplementary Fig. S1a-b), indicating redundant roles of Sulf1 and Sulf2. The sulfation profiles of chondroitin sulfate were not changed in the mutant brains (data not shown). These results demonstrate that Sulf1 and Sulf2 play a cooperative role in generating the sulfation patterns of HS in neonatal brains.

Sulf1/2 DKO Mice Have Defects in CST Axon
Guidance. Because no gross brain malformations were observed, we performed intensive histological searches for possible abnormalities in the nerve tracts of Sulf1/2 DKO mice. Consequently, we found that the cerebral peduncle was aberrant and the pyramidal tract was reduced in size in neurofilament-M-stained brain sections ( Supplementary Fig. S1n-o), indicating that the Sulf1/2 DKO mice had defects in the CST. The CST originates in layer 5 of the sensorimotor cortex, descends through the internal capsule, cerebral peduncle, and ventral medulla, and projects to the spinal cord postnatally 14,15 . To selectively examine the trajectory of the CST axons, we performed DiI tracing. When DiI was injected into the motor cortex of live neonatal mice, CST axons were labeled up to the pyramidal decussation within 10 h (Fig. 1a,c-e).
In the DKO mice, most of the labeled fibers extended dorsally towards the superior colliculus and then returned to the brainstem, whereas some fibers invaded the superior and inferior colliculi (Fig. 1b,f). The fibers that returned to the medulla were defasciculated and positioned more laterally (Fig. 1b,h), which was in contrast to the tightly fasciculated bundle in the control mice (Fig. 1a,e). As a result, the width of the CST axon bundles in the medulla was significantly greater in the DKO mice than that in the control mice (202.6 μm in control and 374.4 μm in DKO, n = 4; P = 0.002635 by Welch's t-test). Section analysis of the DiI-injected brains showed that the CST fibers of the DKO mice appeared to be almost normal until they reached the midbrain, where the misdirected fibers extended dorsolaterally along the brain surface ( Supplementary Fig. S1p-t). In the medulla of the DKO mice, the pyramidal tract became thinner and broader ( Supplementary Fig. S1u). These abnormalities were consistent with the defects detected by the previous neurofilament-M staining ( Supplementary Fig. S1m-o). Additional defects were observed inside the DiI-injected brains: a small portion of the cortical fibers projected aberrantly towards the tectum through the thalamus (n = 4/4, Supplementary Fig. S1r) and some mice showed midline-crossing defects in the corpus callosum (n = 1/4, data not shown).
Because the striking abnormalities of the CST axons were present on the brain surface and could be detected by neurofilament antibody, we performed whole-mount neurofilament-M staining at embryonic day 18.5 (E18.5), when the CST axons reach the medulla. In the control brain, the CST axons were seen as a large ventral bundle (the cerebral peduncle): they emerged onto the brain surface from the posterolateral side of the hypothalamus and immediately turned medially towards the pons (Fig. 1a,i-j). In contrast, in the Sulf1/2 DKO mice, the CST axons extended dorsally and were defasciculated on the lateral surface of the midbrain (Fig. 1b,k). When viewed from the ventral aspect, the CST fibers turned laterally after exiting the hypothalamus (Fig. 1b,l). Thus, this method can assess the CST defects in DKO mice clearly.
Because whole-mount neurofilament staining is more useful to examine the overall changes in axonal trajectories on the brain surface without experimental bias and because many samples can be evaluated without sectioning, we adopted this method in the subsequent experiments to determine the presence of CST abnormalities. When a number of Sulf1/2 DKO mice were examined, the CST defects were somewhat variable between individuals. Most of the DKO mice showed an aberrant CST trajectory in the midbrain, as shown in Fig. 1k,l (62 of 70 tracts examined, Supplementary Fig. S1y and S1y'). In one severe case, all the CST axons turned dorsally and projected to the superior colliculus (1/70, Supplementary Fig. S1z and S1z'). In less severe cases, the CST axons turned medially towards the pons after exiting the hypothalamus but were located lateral to the medial lemniscus (7/70, Supplementary Fig. S1x and S1x'), whereas they overrode the medial lemniscus in the control brains ( Supplementary Fig. S1w'). In summary, all the Sulf1/2 DKO mice showed CST axon guidance defects, whereas no CST abnormalities were observed in the Sulf1 −/− , Sulf2 −/− , Sulf1 −/− ;Sulf2 +/− , or Sulf1 +/− ;Sulf2 −/− mice (data not shown). We thus focused on this robust defect and explored the underlying mechanism.

Electroporation of Sulf Genes Into the Hypothalamus and Midbrain Rescues the CST Defects in
Sulf1/2 DKO Mice. To examine the Sulf expression potentially relevant to CST development, we performed in situ hybridization. Because CST axons pass through the cerebral peduncle at E15 and reach the medulla at E17-18 in mice 14,15 , we examined Sulf expression at E15.5. Sulf1 mRNA showed relatively restricted expression in the choroid plexus, the cortical hem, and the ventricular zone of the third ventricle ( Fig. 2a-f), whereas Sulf2 mRNA showed broader expression in the brain (Fig. 2a'-f '). In the cortical plate, Sulf1 and Sulf2 mRNAs were observed in the presumptive layer 6 and in layers 5-6, respectively ( Fig. 2a and a'). Outside the cerebral cortex, Sulf1 and Sulf2 showed strong and overlapping expression in the ventricular zone of the third ventricle and aqueduct ( Fig. 2c-f and 2c'-f ').
To determine which brain regions of Sulf expression are required for CST formation, we performed in vivo rescue of the Sulf1/2 DKO phenotype by local introduction of Sulf genes. For this purpose, we electroporated Sulf1 and Sulf2 with EGFP into various brain areas at E12.5 and examined the EGFP-positive area and CST trajectory at E18.5. We performed electroporation at E12.5 because it was before the CST axons began to extend and because the ventricles of the embryonic brains were widely open, enabling easy introducion of exogenous genes into various regions. To induce strong and ubiquitous expression, we used a pCX vector carrying the CAG promoter 16 . By holding embryos with electrodes at different angles, exogenous genes could be electroporated into restricted brain regions of interest ( Fig. 2g-t). Because the coelectroporated genes were coexpressed in the same brain region (Fig. 2u-w and Supplementary Fig. S2a-d), the EGFP-positive areas represented the brain regions where Sulf genes had been introduced. The dashed lines indicate the midline. The average intensity of fluorescence signals in the midbrain area was significantly higher in the DKO mice than that in the control mice (18.388 arbitrary unit in control and 29.096 in DKO; n = 4, P = 0.018637 by Welch's t-test). (i-l) Whole-mount neurofilament staining of the E18.5 brain. Lateral views (i,k) and ventral views (j,l) of control (Sulf1 −/− ; i-j) and Sulf1/2 DKO brains (k,l) are shown. The cerebral cortices were removed. Abnormal fibers (closed arrowheads) were observed in the Sulf1/2 DKO brain (k,l). Open arrowheads in (i,j) indicate the normal cerebral peduncle. Statistical analyses of the neurofilamentpositive fibers in the midbrain (i,k) are shown in Supplementary Table 1  Electroporation of pCX-Sulf1/2 into the cerebral cortex (5 mice; Fig. 2g-j) and into the area from the superior colliculus to the cerebellum (3 mice; Fig. 2k-n) failed to rescue the DKO phenotype. In contrast, when Sulf1/2 were electroporated into the area including the hypothalamus and midbrain, the CST defects were completely restored to normal only on the electroporated side (6 mice; Fig. 2o-t). Moreover, a mutant Sulf1 lacking enzyme activity could not rescue the phenotype (4 mice; Supplementary Fig. S2e-h), whereas wild-type Sulf1 or Sulf2 could rescue the defects (4 mice for Sulf1 and 3 mice for Sulf2; Supplementary Fig. S2i-p). These data indicate that Sulf activities are required not in the cortical axons, but in the middle of the CST trajectory.
Sulf Expression in the Radial Glial Cells Along the Third Ventricle Is Required for Navigating CST Axons. Because the CST defects were observed only when all 4 Sulf alleles were lost, it is plausible that the brain areas expressing both Sulf1 and Sulf2 are associated with the defects. Given that electroporation of Sulf genes into the hypothalamus and the midbrain rescued the DKO phenotype, it is likely that strong overlapping expression of Sulf1/2 in the ventricular zone of the third ventricle plays a critical role. We thus wondered whether selective electroporation of Sulf genes into these cells could rescue the DKO phenotype. Conveniently, we found that the pEF-BOS vector 17 can induce relatively specific expression in these cells, although the underlying mechanism is unknown. When electroporated into the hypothalamus and midbrain, pEF-BOS-EGFP induced weak but restricted expression in the cells in the ventricular zone ( Fig. 3a,b,e), whereas pCX-EGFP induced widespread expression (Fig. 2o,p and Supplementary Fig. S2a,c). We therefore electroporated pEF-BOS-Sulf1 and pEF-BOS-Sulf2 into the Sulf1/2 DKO brain and found that the CST defects were completely rescued (3 mice; Fig. 3c,d). These data indicate that Sulf expression in the ventricular zone of the third ventricle was sufficient to steer CST axons properly.
To examine Sulf protein localization in the brain, we performed immunostaining. Both Sulf1 and Sulf2 proteins were detected in the cells in the ventricular zone along the third ventricle ( Fig. 3f 4  To test whether disruption of Sulf genes alters HS sulfation patterns locally where Sulf proteins are present, we performed immunostaining of HS using several anti-HS phage display antibodies at E15.5, even though our disaccharide analysis had already revealed that trisulfated HS disaccharides were increased in the whole brains of the neonatal DKO mice ( Supplementary Fig. S1a,b). The signals with AO4B08 and RB4CD12 antibodies, which recognize HS disaccharides containing the 2-O-, 6-O-, and N-sulfate groups 18,19 , were stronger on the brain surface of the DKO mice than on that of the control mice, whereas the signals in the blood vessels were comparable between the DKO and the control mice (Fig. 3j,k and Supplementary Fig. S2q-r). The increases in AO4B08 and RB4CD12 signals were particularly prominent in the cerebral peduncle and hypothalamus. Conversely, the signals with HS4E4, which recognizes HS disaccharides containing the 2-O-and N-sulfate groups 18 , were weaker in the DKO mice (Fig. 3l,m). These findings indicate that the proportion of trisulfated HS disaccharides was increased on the pial surface as a result of Sulf1/2 disruption. Because the changes were most robust around the CST trajectory, it is likely that this local change in HS contributes to the emergence of abnormalities in CST axons.

Slit Overexpression Mimics the Axon Guidance Defects in Sulf1/2 DKO Mice.
What are the mechanisms underlying the CST defects in Sulf1/2 DKO mice? We hypothesized that increases in trisulfated HS altered the amount or localization of some axon guidance protein(s). To determine the responsible molecule(s), we first investigated all the known axon guidance proteins expressed in the hypothalamus and midbrain at E15.5 by an unbiased approach using proteomic analysis. Because Sulfs modified HS in the basement membrane, the meninges (enriched in the basement membrane) were analyzed using liquid chromatography-ion trap mass spectrometry (LC-MS/MS). The table in Supplementary Fig. S3a shows the list of axon guidance proteins detected by this method. Slit2, Sema3E, Sema4G, Sema5B, and Sema6B were detected only in the DKO brain, suggesting that they were more abundant in the DKO mice. In contrast, Sema4D and EphrinB2 were detected only in the control brain, suggesting that they were less abundant in the DKO mice. Slit1 showed the same spectral counts in the control and the DKO mice.
Sema3e mRNA was expressed near the cerebral peduncle (Fig. 4a-c), suggesting possible involvement in CST formation. However, Sema3e overexpression did not elicit such CST defects as observed in the Sulf1/2 DKO brain, indicating that excess Sema3E alone cannot account for the DKO phenotype. Electroporation of pCX-Slit2 at E12.5 caused very strong repulsion of the CST axons (Fig. 4m), similar to the severe DKO phenotype and consistent with previous observations 20, 21 . Induction of weaker Slit2 expression by electroporation of pEF-BOS-Slit2 (pEF-BOS induces weaker expression than pCX) at E13.0 (lower electroporation efficiency than at E12.5) led ). (f-m) Immunohistochemistry of the E15.5 brain with anti-Sulf or anti-HS antibodies. The signals with anti-Sulf1 in the Sulf2 KO brain (f) were abolished in the Sulf1/2 DKO brain (g), whereas the signals with anti-Sulf2 in the wild-type brain (h) were abolished in the Sulf2 KO (i) brain. Sulf2 KO and Sulf1/2 DKO brains were used for anti-Sulf1 staining because anti-Sulf1 antibody weakly crossreacts with Sulf2 protein. The signals with anti-HS AO4B08 (j,k) and anti-HS HS4E4 (l,m) in the control (Sulf1 −/− ) brain (j,l) were increased and decreased in the Sulf1/2 DKO brain (k,m), respectively. Pictures  to CST defects that were indistinguishable from those of the moderate DKO phenotype (Fig. 4p). Slit2 mRNA showed high expression in the ventricular zone of the third ventricle, especially in the posterior hypothalamus (Fig. 4d-f and Supplementary Fig. S3p-s), suggesting colocalization and possible collaboration with Sulf1/2. By contrast, Slit1 expression was not seen in the ventricular zone of the ventral third ventricle ( Supplementary  Fig. S3l-o), and therefore, Slit1 was thought to be irrelevant to the CST defects although its overexpression can induce repulsion of CST axons ( Supplementary Fig. S3t-u). Electroporation of other candidate genes did not induce any CST abnormalities in the wild-type brains, nor did it rescue the DKO phenotype ( Supplementary  Fig. S3c,e,g,i,k).
Therefore, Slit2 turned out to be the best candidate that accounted for the CST defects observed in the Sulf1/2 DKO mice. Given that Slit2 binds to HS with high affinity 22 in a 6-O-sulfate-dependent manner 23 , increased trisulfated HS is expected to bind more Slit2 protein in Sulf1/2 DKO mice. Consistent with this prediction, the N-terminal fragment (an active form) of Slit2 protein was more abundant in the meninges of the DKO mice than in those of the control mice ( Fig. 5i and Supplementary Fig. S4s). We thus hypothesized that increases in Slit2 protein in the ventral brain region caused dorsal repulsion of the CST axons in the DKO mice.

Abnormally High Amounts of Slit2 Protein Accumulate on the Pial Surface of Sulf1/2 DKO Mice.
To test this possibility, we wished to detect Slit2 protein in embryonic brains. However, because no antibodies for immunostaining Slit2 were available, we used the extracellular portion of Robo2, a Slit receptor, which was tagged with the Fc region of human IgG (Robo2-Fc; ref. 24 ). First, we confirmed that Robo2-Fc bound to Slit2 in the transfected cells ( Supplementary Fig. S4a-d,f). Next, we tested Robo2-Fc binding on brain sections that were electroporated with pCX-FLAG-Slit2. When the brain sections were incubated with Robo2-Fc, the binding was clearly detected only on the electroporated side and was colocalized with the FLAG epitope ( Supplementary  Fig. S4m-r), indicating that Robo2-Fc can detect Slit2 protein in brain sections. Robo2-Fc binding was strongly detected on the brain surface in addition to the cell bodies of the electroporated cells in the ventricular zone and brain parenchyma.
We then tested whether the disruption of Sulf genes altered Slit localization. Robo2-Fc binding was hardly detectable in the control brains except for weak signals on the pial surface of the superior colliculus (Fig. 5a). In contrast, in Sulf1/2 DKO mice, strong Robo2-Fc binding was detected on the pial surface, especially in the areas of the cerebral peduncle and hypothalamus (Fig. 5b). Quantitative comparison revealed that the signal intensity in the Sulf1/2 DKO mice was significantly higher in the cerebral peduncle and hypothalamus than that in the controls but was comparable in the superior colliculus (Fig. 5f). Robo2-Fc binding was colocalized with laminin ( Fig. 5g-h), suggesting that Slit protein was associated with the basement membrane. Furthermore, Robo2-Fc binding in the DKO brain was completely abolished by deletion of the Slit2 gene (namely in the triple KO brain), except for the weak signals in the superior colliculus (Fig. 5j-k and Supplementary Fig. S5a-e), suggesting that the Robo2-Fc binding signals in the cerebral peduncle and hypothalamus are indicative of Slit2 protein localization. These data clearly showed that Slit2 protein accumulated excessively in the ventral brain region of Sulf1/2 DKO mice.
The elevated levels of Robo2-Fc binding in Sulf1/2 DKO mice were restored to the control levels by electroporation of Sulf1/2, which was performed under the same conditions as those used for the phenotype rescue experiments, whereas electroporation of the mutant Sulf1 was ineffective (Fig. 5c-e). The morphology of the basement membrane and radial glial cells, as assessed by laminin and nestin immunostaining, was normal in the Sulf1/2 DKO mice ( Supplementary Fig. S5f-i). In addition, no increase in Slit2 mRNA expression was observed in the Sulf1/2 DKO mice (data not shown). Taken together, these data indicate that abnormally high amounts of Slit2 protein accumulated in the ventral brain region, causing CST axon guidance defects in Sulf1/2 DKO mice (Fig. 6).

Abnormal Slit2 Protein Accumulation is Responsible for the CST Defects in Sulf1/2 DKO Mice.
Finally, we tested whether reduction of Slit2 levels in Sulf1/2 DKO mice could restore the CST defects. First, we electroporated Slit2 siRNA into DKO brains but failed to reduce the Slit2 levels significantly and the DKO phenotype was not rescued (data not shown). Next, we produced Sulf1/2 DKO mice with heterozygous deletion of Slit2 and found that the CST defects were not restored to normal. This was consistent with the finding that Robo2-Fc binding signals were still detected after deleting one allele of Slit2 ( Supplementary Fig. S5b,e). Finally, we tried to reduce Slit activity by means of Robo2-Fc overexpression, which was shown to be effective in neutralizing Slit activity in vitro 25,26 . Strong overexpression of Robo2-Fc by pCX-Robo2-Fc electroporation restored the DKO phenotype on the electroporated side (n = 2/3 in Sulf1/2 DKO mice; n = 3/6 in Sulf1 −/− ;Sulf2 −/− ;Slit2 +/− mice; Fig. 5l-m). However, even in the rescued mice, the restoration was not complete: the CST axons turned medially towards the pons but were located laterally to the medial lemniscus, resembling the weak phenotype of Sulf1/2 DKO mice. The incidence of phenotype recovery in pCX-Robo2-Fc-electroporated embryos was significantly higher than the appearance frequency of the weak phenotype in nonelectroporated Sulf1/2 DKO mice (5/9 and 7/70 for electroporated and nonelectroporated mice, respectively; P = 0.003, Fisher exact test). These data indicate that suppression of Slit activity by Robo2-Fc overexpression led to the recovery of the CST defects in Sulf1/2 DKO mice. The observation that Robo2-Fc did not completely rescue the CST defects might be explained by the weaker effects of other axon guidance molecules (for example, Sema3E). Taken together, our results demonstrate that abnormal accumulation of Slit2 protein is mainly responsible for the CST axon guidance defects in Sulf1/2 DKO mice.

Discussion
In this study, we have demonstrated that Sulf-mediated trimming of 6-O-sulfate groups in HS is required for CST axon guidance. Normally, Sulfs produced by radial glial cells modify HS sulfation patterns in the pial basement  Figures (a-c) and Figures (d,e) show opposite sides of the brain sections. Con, control side; Ep, electroporated side. (f) Quantitative analysis of Robo2-Fc binding. Average of the fluorescence intensity (arbitrary unit; a.u.) on the pial surface obtained by confocal microscopy in control mice (n = 3, 2 Sulf1 −/− ;Sulf2 +/− and 1 Sulf1 −/− ), Sulf1/2 DKO mice (n = 3), and Sulf1/2 DKO mice electroporated with pCX-Sulf1/2 (n = 4) or pCX-Sulf1(C87A) (n = 4) are shown. Statistical significance was calculated using ANOVA with a Tukey-Kramer post hoc test (**P < 0.01; ***P < 0.001; ****P < 0.0001). (g,h) Colocalization of Robo2-Fc binding and anti-laminin immunostaining in the cerebral peduncle (g) and hypothalamus (h) of Sulf1/2 DKO mice. (i) Western blot analysis of Slit2 protein in the meninges isolated from the hypothalamus and midbrain of membrane. Loss of Sulfs results in increases in 6-O-sulfated HS, leading to accumulation of Slit2 protein and dorsal displacement of the CST. This contrasts sharply with the phenotype of Slit/Robo KO mice, in which CST axons invade the ventral brain owing to the lack of repulsion 21,27 (Fig. 6). We propose a novel regulatory mechanism by which HS desulfation secures accurate axon guidance by tempering Slit2 protein accumulation. Although we have presented unequivocal evidence that Slit2 is a major cause for this particular CST phenotype, possible involvement of other molecules in other axon guidance defects is worthy of investigation in future.
Genetic studies in mice and zebrafish demonstrated requirement of HS biosynthesis in axon guidance 2,4,28-30 . More specifically, HS modification by specific sulfotransferases has also been implicated in axon guidance 2,4,[31][32][33][34][35][36] . These studies indicate that presence of HS and its sulfate groups is essential for mediating axon guidance signals. Given that sulfate groups at specific positions in HS are required for the interaction between HS and signaling molecules 23 , desulfation is thought to be of the same importance as sulfation in the regulation of HS functions. Correspondingly, our present study has demonstrated that shaping of HS by removing sulfate groups at specific positions is important in axon guidance. Sulfs thus can act as a dynamic regulator that controls HS-dependent signaling in a spatiotemporal manner.
Although many lines of evidence indicate HS dependence of Slit/Robo signaling, how HS interacts with Slit/ Robo and regulates their signaling in vivo remains largely elusive, particularly in the vertebrate brain. We devised a method to detect Slit2 protein on brain sections and clearly demonstrated the localization of Slit2 protein on the pial basement membrane in the Sulf1/2 DKO brain. Moreover, we showed that Slit2 accumulation was associated with trisulfated HS disaccharides. Given that Slit2 binds strongly to 6-O-sulfated heparin 23 , it is likely that Slit2 accumulation in the DKO brain was due to increased stability of Slit2 by 6-O-sulfated HS. Interestingly, Robo2-Fc binding in the control brain was below the detection level in the examined area, suggesting rapid diffusion and/ or degradation of Slit2 protein in vivo. This means that dissociation of Slit2 from HS by means of desulfation contributes to fine-tuning of Slit2 activity. In this context, it is noteworthy that in our proteomic analysis of the control brain, a Slit2 fragment was detected in a small-sized fraction (53-65 kDa), but not in the large-sized fraction corresponding to an active N-terminal Slit2 fragment (100-140 kDa).
Precise patterning of Slit localization is indispensable for accurate guidance of growing axons. In addition to HS, glycosylated dystroglycan and type IV collagen were implicated in the interaction with Slit protein and Slit-mediated axon guidance 37,38 . In particular, in zebrafish, Slit1 protein was shown to be translocated to the surface of the tectum by the actions of the radial glial cells and anchored to the basement membrane via interaction with type IV collagen 37 . Combined with these data, our findings suggest that transport of Slit protein to the brain surface by radial glial cells is a generalized mechanism controlling axon guidance. Two recent papers reported a similar mechanism in which netrin-1 protein produced by ventricular zone progenitor cells is transported to the pial surface to guide commissural axons in the spinal cord and hindbrain 39,40 , although the involvement of HS remains to be examined. Future studies are required to clarify how Sulfs play cooperative roles in anchoring guidance proteins in the basement membrane with dystroglycan and type IV collagen.
Our results also have an important implication in that Slit2 activity can be regulated by changing the HS microstructure in the extracellular environment without altering Slit2 expression. Manipulation of the localization and activity of axon guidance molecules by Sulfs or heparin-related chemicals 41 may be useful to control the action of chemorepellents for promoting regeneration of CST axons.

Methods
Knockout Mice. Sulf knockout mice were generated as described previously 19 . Offspring of mice backcrossed to C57BL/6N for 5 generations (N5 generation) were used. Slit2 knockout mice were obtained from Mutant Mouse Regional Resource Centers. Noon of the day on which a vaginal plug was observed was taken as embryonic day 0.5 (E0.5). All animal experiments were approved by and performed according to the guidelines of the Animal Care and Use Committee of the University of Tsukuba.
Disaccharide Analysis of Heparan Sulfate and Chondroitin Sulfate. Disaccharide analysis was performed as described previously 19 by using neonatal brains because most of the DKO mice died within a day of birth. In brief, acetone-extracted neonatal mouse brains were treated with 0.8 mg/ml protease (P5147; Sigma) at 55 °C overnight, and then with 125 U benzonase (Sigma) at 37 °C for 2 h. After centrifugation, supernatants were purified with Vivapure D Mini M (Vivascience) and concentrated with Ultrafree-MC Biomax-5 (Millipore). The purified glycosaminoglycans were treated at 37 °C overnight with a mixture of 1 mIU of heparinase I (Sigma), 1 mIU of heparitinase I (Seikagaku), and 1 mIU of heparitinase II (Seikagaku) or with a mixture of 50 mIU of chondroitinase ABC (Seikagaku) and 50 mIU of chondroitinase ACII (Seikagaku). Unsaturated disaccharides were analyzed by ion-pair reversed-phase chromatography. An NaCl gradient (2-106 mM) in 1.2 mM E15.5 mice. Slit2 was more abundant in Sulf1/2 DKO mice than in the control (Sulf2 −/− ) mice. FL and N indicate the full-length (~180 kDa) and N-terminal fragments (~130 kDa), respectively. Addition of blocking peptide abolished the bands (right panel), indicating the specificity of the antibody. Full-length blots that were used for this picture are shown in Supplementary Fig. S7. (j,k) Robo2-Fc binding was completely abolished in the Sulf1/ Sulf2/Slit2 triple knockout (TKO) brain. (l,m) Neutralization of Slit rescues the CST defects in Sulf1/2 DKO mice. The plasmids pCX-Robo2-Fc and pCX-EGFP were electroporated into E12.5 Sulf1/2 DKO brains, and the CST trajectory was examined at E18.5. The CST was nearly normalized by electroporation (open arrowheads) but lay slightly lateral to the medial lemniscus (ml). Anterior-posterior (A-P), dorsal-ventral (D-V), and mediallateral (M-L) body axes are shown. Scale bars indicate 100 μm (a-e,j,k), 50 μm (g,h), and 500 μm (l,m). See also Supplementary Fig. S4-7.
Plasmid Constructs for Electroporation. The plasmids used in this study were obtained as follows. EGFP was derived from pEGFP-N3 (Clontech). Rat Sulf1 (SulfFP1) and Sulf2 (SulfFP2) were previously described 19,45 . The point mutation in Sulf1(C87A) was introduced by PCR. Mouse cDNAs Slit1 (BC062091), Sema3e (IMAGE:5357516), Sema4d (IMAGE:6509473), Sema4g (IMAGE:6395010), Sema5b (IMAGE:5719939), and Efnb2 (IMAGE:6827408) were obtained from the IMAGE consortium clone collection. Mouse Slit2 (G730002D07) and Sema6b (G730036J03) were obtained from RIKEN Mouse FANTOM Clones (DNAFORM). Because both the Slit1 and the Slit2 clones contained short deletions, the incorrect sequences were replaced with the correct sequences that were amplified by RT-PCR, followed by sequence confirmation. Rat Robo1 and Robo2 cDNAs were amplified by RT-PCR, and their whole sequences determined to confirm that they contained the correct sequences.
Sulf1 was tagged with either a FLAG peptide (DYKDDDDK) or a Myc peptide (EQKLISEEDL) plus the 6xHis tag at its C-terminus. Sulf2 was tagged with the Myc peptide at its C-terminus. Sulf cDNAs were subcloned into the pCX vector 16 and pEF-BOS vector 17 . For the rescue experiments, pCX-Sulf1-FLAG, pCX-Sulf2-Myc, pEF-BOS-Sulf1-FLAG, and pEF-BOS-Sulf2-Myc were used; for the sake of simplicity, they are denoted in the text as pCX-Sulf1, pCX-Sulf2, pEF-BOS-Sulf1, and pEF-BOS-Sulf2, respectively. In the experiments in which Sulf protein localization was examined, Sulf1-Myc-His and Sulf2-Myc were subcloned into a pCX vector containing a Kozak sequence (GCCACCATG; the underline is the initiation codon), a signal sequence from the human CD8A gene (ALPVTALLLPLALLLHAARP), the FLAG peptide, and the 6xHis tag. Sulf1 and Sulf2 are connected to the above sequence from the positions of the 20th and 25th amino acids, respectively (both are the N-termini of the mature protein predicted by the SignalP 3.0 algorithm [http://www.cbs.dtu.dk/services/ SignalP/]). For simplicity, the expression constructs created, pCX-Kozak-SS-FLAG-His-Sulf1-Myc-His and pCX-Kozak-SS-FLAG-His-Sulf2-Myc, are denoted in the text as pCX-FLAG-Sulf1 and pCX-FLAG-Sulf2, respectively.
The mouse cDNAs Slit1, Slit2, Sema3e, Sema4d, Sema4g, Sema5b, Sema6b, and Efnb2 were subcloned into a pCX vector (and a pEF-BOS vector for Slits) that contained the Kozak sequence, the signal sequence from the human CD8A gene, the FLAG peptide, and the 6xHis tag. Slit1, Slit2, Sema3E, Sema4D, Sema4G, Sema5B, Sema6B, and EphrinB2 were connected to the above sequence from the positions of the 33 rd , 26 th , 26 th , 24 th , 18 th , 27 th , 17 th , and 30 th amino acids, respectively (all are the N-termini of the mature protein predicted by the SignalP 3.0 algorithm). For simplicity, they are denoted in the text as pCX-Slit1, pCX-Slit2, pCX-Sema3e, pCX-Sema4d, pCX-Sema4g, pCX-Sema5b, pCX-Sema6b, and pCX-Efnb2, respectively. In the in vitro Robo2-Fc binding experiments, rat Slit2 tagged with an HA sequence (YPYDVPDYA) was subcloned into a pCEP4 vector (Invitrogen) containing the Kozak sequence, the signal sequence from the human CD8A gene, the FLAG peptide, and the 6xHis tag.
Technologies) at room temperature for 15 min. In this preparation, the fluorescence signal of elecroporated EGFP was eliminated, and thereby the immunostaining was not disturbed.
A series of z-stack fluorescence images (1024 × 1024 pixels) was obtained using a laser-scanning microscope, the LSM510 with a 10x objective lens (Carl Zeiss). For quantitative comparison of the fluorescence intensity of the samples, the same parameters (pinhole, 80.3 μm; optimal interval, 6.54 μm; detector gain, 1000; amplifier offset, −0.1; amplifier gain, 1; scan speed, 9; average number, 2) were used for all scans. Three sections were used for each experimental condition. In each section, the single optical slice with the strongest intensity was selected for measurement. An outline of the brain surface was traced, and the fluorescence intensity in the traced area was measured.

Statistics.
No statistical methods were used to predetermine sample sizes, but our sample sizes are similar to those generally employed in the field. Statistical analyses for the data of Robo2-Fc binding and disaccharide analysis were done with ANOVA with Tukey-Kramer and Bonferroni post hoc tests, respectively. The Fisher exact test was used to analyze the incidence of phenotype recovery in pCX-Robo2-Fc-electroporated embryos. Welch's t-test and a paired t-test were used to analyze the average fluorescence signal intensity, the width of the axon bundles, and the area of the neurofilament positive area (details of the analysis are described in Supplementary Table 1).

Data availability.
The data that support the findings of this study are availiable from the corresponding author upon reasonable request.