Neural Cadherin Plays Distinct Roles for Neuronal Survival and Axon Growth under Different Regenerative Conditions

Abstract Growing axons in the CNS often migrate along specific pathways to reach their targets. During embryonic development, this migration is guided by different types of cell adhesion molecules (CAMs) present on the surface of glial cells or other neurons, including the neural cadherin (NCAD). Axons in the adult CNS can be stimulated to regenerate, and travel long distances. Crucially, however, while a few axons are guided effectively through the injured nerve under certain conditions, most axons never migrate properly. The molecular underpinnings of the variable growth, and the glial CAMs that are responsible for CNS axon regeneration remain unclear. Here we used optic nerve crush to demonstrate that NCAD plays multifaceted functions in facilitating CNS axon regeneration. Astrocyte-specific deletion of NCAD dramatically decreases regeneration induced by phosphatase and tensin homolog (PTEN) ablation in retinal ganglion cells (RGCs). Consistent with NCAD’s tendency to act as homodimers, deletion of NCAD in RGCs also reduces regeneration. Deletion of NCAD in astrocytes neither alters RGCs’ mammalian target of rapamycin complex 1 (mTORC1) activity nor lesion size, two factors known to affect regeneration. Unexpectedly, however, we find that NCAD deletion in RGCs reduces PTEN-deletion-induced RGC survival. We further show that NCAD deletion, in either astrocytes or RGCs, has negligible effects on the regeneration induced by ciliary neurotrophic factor (CNTF), suggesting that other CAMs are critical under this regenerative condition. Consistent with this notion, CNTF induces expression various integrins known to mediate cell adhesion. Together, our study reveals multilayered functions of NCAD and a molecular basis of variability in guided axon growth.

Growing axons in the CNS must often travel long distances and migrate along specific pathways to reach their targets. During development, these axons are guided by various cell adhesion molecules (CAMs; Walsh and Doherty, 1997;Blackmore and Letourneau, 2006;Kamiguchi, 2007), expressed on the surface of glial cells or other neurons. In the mature CNS, however, regenerating axons often travel circuitously and never reach their destinations, indicating that the cellular and molecular factors that once operated during development do not necessarily remain present or may be inadequate for the adult axons. CAMs are a large group of proteins located on the cell surface, involved in the binding of cells to glial cells or extracellular matrix (ECM) proteins (Kamiguchi, 2007;Hillen et al., 2018). CAMs are categorized broadly into four superfamilies: the immunoglobulin superfamily of CAMs (IgCAMs), cadherins, integrins, and the superfamily of C-type lectin-like domain proteins (CTLDs; Chothia and Jones, 1997;Graham and Duan, 2020). Within the cadherins, neural cadherin [NCAD; also known as cadherin-2 (CDH2)] is expressed in neuronal cells as well as in glial cells of the CNS in vertebrate embryos (Radice et al., 1997). Several studies have demonstrated that NCAD plays a critical role in guiding the migration of neurites on astrocytes during development (Hansen et al., 2008). Transfection with NCAD promotes the outgrowth of chicken embryonic optic axons on monolayer cultures (Matsunaga et al., 1988). An in vitro study has shown that the genetic deletion of NCAD in cultured astrocytes impairs the formation and extension of sensory neurons' neurites on astrocytes (Ferguson and Scherer, 2012). Other studies have shown that cadherins play a critical role in promoting axon fasciculation, allowing axons with similar cadherins on their surfaces to cluster together, and grow toward their common targets (Bastiani et al., 1987;Missaire and Hindges, 2015;Bruce et al., 2017). However, while a great amount of knowledge has been gathered about cadherins' contributions to axon growth and guidance during development, it is unclear whether the same factors control this process during axon regeneration in adulthood. In this study, we used optic nerve crush and gene knock-out (KO) strategies in adult mice to investigate the extent to which astrocytes facilitate axon adhesion and regeneration, and whether NCAD and other CAMs mediate axon regeneration under two different conditions, PTEN deletion and CNTF overexpression.

Animals
All animal experimental procedures were performed in compliance with protocols approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Miami. Animals used were C57BL/6J (The Jackson Laboratory, stock number 000664), NCAD f/f (The Jackson Laboratory, stock number 007611), GFAP-CreERT (a gift from Ken D. McCarthy, University of North Carolina; Casper et al., 2007), PTEN f/f (The Jackson Laboratory, stock number 006440), Rosa26 loxP-STOP-loxP-tdTomato (R26-tdTomato, a gift from Fan Wang, Duke University, Durham), and Rosa26-STOP-EYFP (The Jackson Laboratory, stock number 006148). All animals were housed in a viral antigen-free facility and kept under standard 12/12 h light/dark conditions. For all surgical procedures, mice were anaesthetized with ketamine and xylazine. For analgesia, buprenorphine (0.05 mg/kg) was administered postoperatively. Tamoxifen injections. Tamoxifen stock solutions (50 mg/ml) were prepared from 1 g of Tamoxifen Free Base MP Biomedicals (fisher scientific) dissolved in a sterile mixture of 10% ethanol plus 90% sunflower oil. The tamoxifen solution was placed at 55°C in a water bath during 3 h while protected from light and vortexed every hour and then stored at À20°C until use. Young adult mice (five to six weeks old) received daily intraperitoneal injections of tamoxifen (0.124 mg/g body weight) for five consecutive days.

Cloning and generation of adeno-associated viruses (AAVs)
To suppress PTEN expression, we adopted a shRNA strategy, based on SIBR vectors in which shRNA is located in an intron and flanked by sequences derived from mir155, an endogenous intronic shRNA (Chung et al., 2006;Yungher et al., 2015). To maximize the probability of effectively targeting PTEN, four separate shRNA sequences, each targeting a different region of PTEN were concatenated in a single plasmid, which was then used to produce AAV (AAV-shPTEN). Four sequences that target both mouse and rat PTEN were designed using siDIRECT website and design rules: four targeted sequences for PTEN are: GCAGAAACAAAAGGAGATATCA;GATGATG TTTGAAACTATTCCA;GTAGAGTTCTTC CACAAACAGA; GATGAAGATCAGCATTCACAAA. Oligonucleotides encoding hairpin loops that included these sequences and deliberate mis-matches in the non-target strand were synthesized, annealed, inserted into the SIBR knockdown vector, and concatenated into a single plasmid as described. A region of the SIBR knock-down vector comprising the ubiquitin promoter, intronic sequences, knockdown cassette, and EGFP open reading frame was cloned into an AAV-compatible plasmid (AAV-MCS, Stratagene), from which the CMV promoter, intron and MCS were removed. pAAV-RC (Stratagene) that encodes the AAV2 genes (rep and cap) and the helper plasmid (Stratagene) that encodes E2A, E4, and VA were used for co-transfection in 293T cells to generate recombinant AAV. Plasmids were then used to produce AAV2 (1-4 Â 1013 particles/ ml). To construct AAV expressing a secretable form of CNTF, an AAV-compatible SIBR vector was created by PCR-amplifying the knock-down cassette of a SIBR vector with primers that created 59 Mlu1 (ACGCGTTTA AACTGGCCTCCGCGCC) and 39 ClaI (ccgccgATCGA TTCACTTGTACAGCTCGTCCA) sites (Yungher et al., 2015). This cassette was inserted into a Stratagene AAV plasmid, replacing the CMV promoter and B-globin intron. The resulting AAV-SIBR plasmid was then modified via bridge PCR to create KpnI and BglII sites to flank the EGFP open reading frame. Plasmid DNA encoding human CNTF was purchased from OpenBiosystems (accession BC068030) and the open reading frame was amplified using a forward primer that incorporated both a 59 KpnI restriction site and the NGF signal peptide sequence (GGTACCATGTCCATGTTGTTCTACACTCTGATCACAGC TTTTCTGATCGGCATACAG GCGGCTTTCACAGAGCATT CACCGC) and a reverse primer that incorporated 39 BglII site (AGATCTCTACATTTTCTTGTTGTTAGCAA). PCR-amplified CNTF was then used to replace the EGFP ORF in the AAV-SIBR vector via standard restriction digest and ligation. All enzymes were purchased from New England Biolabs. Plasmids were then used to produce AAV sero-type2 (1-4 Â 1013 particles/ml). For making AAV2 expressing Cre recombinase, the cDNA of Cre was inserted downstream of the CMV promoter/b -globin intron enhancer in the plasmid pAAV-MCS (Stratagene), containing the AAV2 inverted terminal repeats and a human growth hormone poly A signal. Plasmids were then used to produce AAV2 (1-4 Â 1013 particles/ml).

Intravitreal injection
For AAV injection, ;2-ml volume was injected using a Hamilton syringe (Hamilton 80900) coupled with a fine glass micropipette inserted into the posterior chamber of the eye and deliberately angled to avoid damage to the lens. To label regenerating axons anterogradely, we injected 2 ml of Alexa Fluor 488-conjugated or Alexa Fluor 555-conjugated cholera toxin b subunit (CTB; 2 mg/ml; ThermoFisher C22841 and C22843) 3 d before killing. AAV-CRE and AAV-shPTEN were injected two weeks before optic nerve crush to allow sufficient time for the buildup of the transgenes and subsequent gene KO. AAV-CNTF injection was done one week after the last tamoxifen injection, and optic nerve crush done 3 d after AAV-CNTF injection.

Optic nerve crush
The optic nerve was crushed unilaterally and intraorbitally using a pair of forceps (#5 Dumont, Fine Science Tools) for 10 s, ;1 mm distal to the nerve's emergence from the globe (Bray et al., 2019). At various time points after crush, animals were humanely euthanized and tissues processed for further analyses.

Imaging
Images were obtained using a Nikon Eclipse Ti fluorescent microscope or an Olympus FluoView 1000 (FV1000) confocal microscope. Confocal high-magnification 2Dprojected optic nerve images were obtained by combining individual z-stacks images using the mosaic automated stitching mode of the Olympus FV1000 software. All images were analyzed with ImageJ (NIH).

Quantification of retinal ganglion cell (RGC) survival and axon regeneration
The percentage of surviving RGCs in the injured retina was estimated by counting the number of RBPMS1 RGCs in several sections from the injured and the intact contralateral retina. RGC survival in the injured retina was presented as a percentage of the intact contralateral retina. To quantify the number of regenerating RGC axons, optic nerve sections were imaged, and the number of CTB1 axons at different distances distal to the lesion site was counted. Axon numbers are presented as number of axons per section normalized to 250-mm width. At least three to four sections were counted per animal.

Quantification of pS61 RGCs
For the determination of mammalian target of rapamycin complex 1 (mTORC1) activity in RGCs, several retina sections were immunostained with an antibody against the phosphorylated ribosomal pS6, a widely used marker for mTORC1 activity (Park et al., 2008;Yang et al., 2014;Wang et al., 2020). The number of pS61 RGCs in the ganglion cell layer (GCL) of the injured retina was expressed as a percentage of the total number obtained in the contralateral intact retina.

Lesion area quantifications
The area size (mm 2 ) of the optic nerve lesion from different animals was determined by manually drawing a contour around the GFAP-negative area (lesion site) ;1 mm away from the eye, using ImageJ software. At least three to four sections were analyzed for each animal. RT 2 Profiler PCR Array. Total RNA was isolated from four mouse retinas for each condition using Direct-Zol RNA Miniprep (R2050; Zymo Research) followed by cDNA synthesis with the RT2 First Strand kit (330401; QIAGEN) according to manufacturer's protocol. Real-Time PCRs for the RT2 Profiler PCR Arrays were performed using RT 2 SYBR Green qPCR Mastermix (330501; QIAGEN), Mouse Extracellular Matrix & Adhesion Molecules Arrays in 96-well format (PAMM-013ZC; QIAGEN), and QuantStudio three Real-Time PCR System (Thermo Fisher Scientific). Analysis of exported data were conducted at QIAGEN'S GeneGlobe Data Analysis Center using a software-based tool (www.qiagen.com/shop/genes-andpathways/dataanalysis-center-overview-page).

Statistical analyses
Statistical analyses were performed using GraphPad Prism software. Data were analyzed using Student's t test, one-way ANOVA, or two-way ANOVA with Tukey's or Bonferroni's post hoc test. Values of p , 0.05 were considered significant. Number of animals used for each animal group is stated in the figure legends. All error bars represent SEM.

Most regenerating RGC axons associate closely with astrocytes
To examine the roles played by astrocytes and cadherins in promoting RGC axon regeneration, we used intraorbital optic nerve crush in adult mice. To induce axon regeneration, we used shRNA against PTEN (shPTEN) to knock-down PTEN expression in adult RGCs. Adult mice received intravitreal injection of AAV-shPTEN, and two weeks later, animals received unilateral optic nerve crush. The optic nerves were assessed three weeks after crush. CTB was injected intravitreally 3 d before killing to anterogradely trace the regenerating RGC axons. Immunohistochemistry performed on the injured optic nerve using GFAP antibody showed the site of optic nerve crush. This is an area largely devoid of GFAP immunoreactivity, and located ;1 mm from the eye. As expected, regenerating RGC axons were seen beyond the lesion site in the shPTEN-treated animals. Nearly all regenerating axons were found in close proximity to the GFAP1 processes in the lesion site, and they appeared to follow the astrocyte processes through the lesion (Fig. 1A,B). Similarly, regenerating axons immediately distal to the lesion site almost always associated closely with astrocyte processes. In addition to PTEN deletion, axon regeneration can also be stimulated by increasing CNTF expression (Müller et al., 2009;Yungher et al., 2015;Luo et al., 2016;Bray et al., 2019). Therefore, to determine whether axons associate with astrocyte processes under different types of regenerative conditions, we assessed the degree to which regenerating axons associate with astrocytes in AAV-CNTF-treated animals. To this end, adult mice received intravitreal AAV-CNTF injection, followed by optic nerve crush. Similar to the shPTEN-treated animals, the regenerating axons associated closely with the GFAP1 processes ( Fig. 1C). Together, these results indicate that RGC axons under either condition regenerate into, and past, the lesion site along the astrocytic processes.

Deleting NCAD in astrocytes decreases PTEN-knockdown-induced RGC axon regeneration
There are over 100 different types of cadherins found in vertebrates, which can be classified into four groups: classical, desmosomal, protocadherins, and unconventional (Maître and Heisenberg, 2013). NCAD belongs to the classical group and is found predominantly in neurons. To determine whether NCAD expression in astrocytes mediates RGC axon regeneration, we deleted NCAD specifically in adult astrocytes using the inducible Cre driver line, GFAP-CreERT (Casper et al., 2007). To first validate Cre recombination in the optic nerves of GFAP-CreERT mice, we crossed a GFAP-CreERT mouse to a Rosa26-loxp-stop-loxp-YFP reporter line (Rosa26-YFP) to generate GFAP-CreERT; Rosa26-YFP mice. Tamoxifen was administered daily for five consecutive days in these mice. Consistent with the previous reports, tamoxifen injection led to recombination in the majority of astrocytes; most GFAP immunoreactive cells co-expressed YFP ( Fig. 2A). We did not quantify the percentage of astrocytes that express YFP. Nonetheless, previous reports have shown that ;80-90% of astrocytes successfully express the transgene (Casper et al., 2007;Park et al., 2018). To confirm that Cre recombination does not occur in RGCs, we performed immunohistochemistry on the sectioned retinas of GFAPCreERT; Rosa26-YFP mice.
We observed that the YFP1 cells were located predominantly in the inner nuclear layer where GFAP1 Müller cells are normally located (Fig. 2B). Some YFP immunoreactivity was also seen in the GCL, but these YFP1 cells did not co-localize with RBPMS immunoreactivity (i.e., a marker for RGCs), indicating that Cre recombination does not occur in RGCs. We crossed GFAPCreERT; Rosa26-YFP to NCAD f/f mice to generate GFAP-CreERT; Rosa26-YFP; NCAD f/f (hereafter referred to as GFAPCreERT; YFP; NCAD f/f ). These animals received intravitreal AAV- Figure 1. Regenerating RGC axons associate closely with GFAP1 processes. A, Representative confocal image of an optic nerve section from an AAV-shPTEN-injected mouse three weeks after crush injury. Red, CTB555. White, GFAP immunoreactivity. Broken yellow line marks the boundary of the glial scar (i.e., lesioned area). Image on the right, higher magnification of the lesioned area. Yellow arrow points to an axon which appears to follow the GFAP1 processes through the lesion. B, An optic nerve section from another AAV-shPTEN-treated mouse with a lesion larger than the animal shown in A. Broken yellow line marks the boundary of the lesioned area. Image on the right, higher magnification of the lesioned area. Regenerating axons, indicated by the yellow arrowheads, seem to follow the GFAP1 processes through the lesion. C, Representative confocal image of an optic nerve section from an AAV-CNTF-injected mouse three weeks after crush injury. Red, CTB555. White, GFAP immunoreactivity. Right panel, higher magnification of the boxed area in the left panel. The regenerating RGC axons (indicated by yellow arrows) appear to grow along the astrocyte processes. Scale bars, 100 mm. Figure 2. Deletion of NCAD in astrocytes prevents PTEN knock-down-induced axon regeneration in adult mice. A, Cre recombination in the GFAPCreERT; Rosa26-YFP mouse. Representative optic nerve section from an adult GFAPCreERT; Rosa26-YFP mouse (eight weeks old) shows that most of the GFAP immunoreactivity is accompanied by YFP immunoreactivity, indicating high degree of Cre recombination in the astrocytes. Animal received five daily intraperitoneal injections of tamoxifen. Animals were perfused 2 d after the last tamoxifen injection. B, Representative images of retinal sections (intact uninjured) of a GFAPCreERT; Rosa26-GFP mouse. Adult animals received tamoxifen injection for five consecutive days and perfused six weeks after the first tamoxifen injection. RBPMS in white, YFP in green, and DAPI in blue. YFP expression is predominantly in the inner nuclear layer. Some YFP immunoreactivity is also seen in the GCL. However, YFP immunoreactivity does not co-localize with RBPMS1 cells, indicating that there is no or minimal Cre expression in the RGCs. C, Timeline of the astrocyte-specific NCAD KO experiment. D, Representative optic nerve sections from AAV-shPTEN-injected WT (i.e., GFAPCreERT; YFP mice with tamoxifen injections) and AAV-shPTEN-injected astrocyte-specific KO mice (i.e., GFAPCreERT; YFP; NCAD f/f , or also referred in the figure as NCAD aKO) mice. Red asterisks, shPTEN injection, followed by tamoxifen injection daily for five consecutive days. Two weeks after the AAV injection, animals received unilateral optic nerve crush. We injected CTB to label the regenerating axons (Fig. 2C). GFAPCreERT; YFP mice with the same treatment served as control animals (i.e., wild-type animals). Notably, we observed that axon regeneration was markedly reduced in the NCAD-deleted mice. There was approximately a 50% decrease in the number of regenerating axons at 250 mm distal to the lesion site (Fig. 2D,E). More distally, the reduction was even more obvious; in the NCAD-deleted mice, virtually no regenerating axons were seen beyond the 1000-mm mark. It is known that, even without regenerative treatment, a few RGC axons can regenerate spontaneously (Yin et al., 2006;Benowitz and Yin, 2010;Bray et al., 2019). To examine whether NCAD deletion has any effects on the spontaneous regeneration, we compared axon regeneration between the WT and NCAD-deleted animal groups, both without an AAV-shPTEN injection. There was no difference in the number of axons between these groups (Fig. 2F,G). These results demonstrate that, while astrocyte expression of NCAD is required for PTEN-knock-down-induced axon regeneration, it plays no role in promoting spontaneous RGC axon regeneration. Additionally, since the death of RGCs would affect axon regeneration, we also assessed RGC survival. Immunohistochemistry on retinas using an antibody against RBPMS (i.e., a marker for RGCs; Rodriguez et al., 2014) showed that RGC survival is unchanged after NCAD deletion in astrocytes (Fig. 2H).

Deleting NCAD in RGCs decreases PTEN-deletioninduced axon regeneration
The functionality of cadherins depends largely on the formation of homodimers. The homodimeric cadherins, including NCAD, create surface adhesion with cadherins present in the membranes of other cells, through changing conformation from cis-dimers to trans-dimers (Shapiro et al., 1995;Vendome et al., 2011). NCAD is expressed on apposing cell membranes of both axons and glial cells (Redies and Takeichi, 1996;Redies, 2000). In our previous study, we performed RNA-sequencing on murine RGCs and observed that Cdh2 is highly expressed in at least two RGC subtypes (i.e., intrinsically-photosensitive RGCs and direction-selective RGCs; Bray et al., 2019). Other studies have reported that RGCs of different subtypes also express cdh2 (Rheaume et al., 2018;Tran et al., 2019). Thus, we reasoned that NCAD is expressed on the RGC axons and promotes adhesion to astrocytes, thereby supporting axon regeneration. If this is true, then deleting NCAD in RGCs would reduce axon regeneration. To test this idea, we generated NCAD f/f ; PTEN f/f double floxed mice. In these mice, intravitreal injection of AAV2-Cre will result in deletion of both NCAD and PTEN in RGCs. Adult animals received intravitreal AAV2-Cre injection, followed by optic nerve crush two weeks later. The results showed that axon regeneration was drastically reduced in the NCAD f/f ; PTEN f/f double KO compared with the PTEN f/f single KO mice (Fig. 3A,B).

Deleting NCAD in RGCs eliminates PTEN-deletioninduced RGC survival
To assess whether deletion of NCAD in RGCs affects RGC survival, we performed immunohistochemistry using the RBPMS antibody. NCAD deletion alone had a statistically insignificant decrease in the RGC survival when compared with the WT mice. Previous studies have demonstrated that PTEN deletion in neurons promotes cell survival after injury or in neurodegenerative conditions (Park et al., 2008;Domanskyi et al., 2011). As shown in Figure 3, there was a difference in the number of surviving RGCs between the PTEN f/f mice and the PTEN f/f ; NCAD f/f mice; three weeks after optic nerve crush, there was approximately a 50% reduction in RGC survival in the double KO mice (Fig. 3C) compared with the PTEN f/f single KO mice.
Deleting NCAD in astrocytes has no effect on CNTFinduced axon regeneration Unlike PTEN, which regulates the activity of mTORC1 and glycogen synthase kinase 3 b (GSK3b ; Park et al., 2008;Leibinger et al., 2019), CNTF promotes regeneration primarily through activation of STAT3 (Sun et al., 2011;Leibinger et al., 2013). To examine whether astrocytic NCAD plays a general role under different regenerative conditions, we assessed the effects of deleting NCAD in AAV-CNTF-treated animals. As expected, AAV-CNTF promoted axon regeneration in the GFAPCreERT; YFP (i. e., WT) mice. A similar level of CNTF-induced regeneration was seen in the GFAPCreERT; YFP; NCAD f/f mice (Fig. 4A,B).

Deleting NCAD in RGCs neither affects CNTF-induced axon regeneration nor RGC survival
As NCAD expression in RGCs was critical for axon regeneration and cell survival in the background of PTEN deletion, we examined whether NCAD in RGCs plays similar roles for CNTF-induced regeneration and RGC survival. NCAD f/f mice received intravitreal AAV-Cre continued crush site. E, Quantification of the number of regenerating axons per section at different distances from the lesion site. N = 5 mice for AAV-shPTEN; WT and N = 4 for AAV-shPTEN; NCAD aKO mice. Two-way ANOVA with Bonferroni's post hoc test. ns, not significant; **p , 0.01, ***p , 0.001. F, Representative optic nerve sections from WT (i.e., GFAPCreERT; YFP mice with tamoxifen injections) and NCAD aKO mice without an AAV-shPTEN injection. Red asterisks, crush site. G, Quantification of the number of regenerating axons for the WT and NCAD aKO groups without AAV-shPTEN injection. N = 4 for WT and N = 7 for NCAD aKO. Twoway ANOVA with Bonferroni's post hoc test. ns, not significant. H, Quantification of RGC survival for the AAV-shPTEN-injected WT and AAV-shPTEN-injected NCAD aKO mice shown in C. N = 4 for each group. ns, not significant. Scale bars, 100 mm. injection, followed by intravitreal AAV-CNTF injection two weeks later. Three days after the AAV-CNTF injection, animals received optic nerve crush. In contrast to the results seen in the PTEN-deleted animals, NCAD deletion in RGCs did not alter CNTF-induced regeneration or RGC survival (Fig. 5).

Deleting NCAD in astrocytes does not affect the size of the lesion
Since the size of the lesion negatively correlates with axon regeneration, we also examined whether the deletion of NCAD in astrocytes increases the size of the lesion after optic nerve crush. Immunohistochemistry using GFAP antibody showed that the lesion size was unaffected by NCAD deletion in astrocytes (Fig. 6).

NCAD deletion in astrocytes does not change the level of mTORC1 activity in RGCs
Several studies have demonstrated that mTORC1 acts downstream of PTEN, and mediates axon regeneration (Park et al., 2008;Yang et al., 2014;. To further examine the mechanisms by which NCAD deletion reduces axon regeneration, we assessed the level of mTORC1 activity in the RGCs of shPTEN-treated animals. To this end, we performed immunohistochemistry on sectioned retinas using an antibody against the phosphorylated ribosomal pS6, a known indicator of mTORC1 activity (Park et al., 2008;Yang et al., 2014;Wang et al., 2020). The numbers of pS61 RGCs after injury were similar between the WT and NCAD-deleted animals (Fig. 7). Thus, these results indicate that the reduced regeneration and cell survival seen after NCAD deletion is unlikely to be caused by changes in mTORC1 activity.

Distinct CAMs are differentially expressed in the PTEN-deleted and CNTF-treated RGCs
Our results show that NCAD is required for the PTENdeletion induced but not for the CNTF-induced regeneration. What accounts for this difference? One possibility is that CAMs other than NCAD are induced in the CNTFtreated RGCs, and that these CAMs compensate for the loss of NCAD. This explanation is supported by the fact that (1) there are many different types of CAMs that can promote cell adhesion and growth, and that (2) cytokines are known to induce expression of various types of CAMs (Meager, 1999). To examine whether CNTF results in differential expression of CAMs in RGCs, we used RT 2 Profiler PCR Array, and compared the expression of 84 CAMs and ECM genes (Table 1) between the AAV-CNTFtreated and the AAV-shPTEN-treated animals. Both groups received optic nerve crush 14 d after AAV injection, and retinas were removed 3 d after optic nerve crush. We extracted total RNAs from the whole retinas. As shown in Figure 8, several genes were differentially expressed between the two animal groups. Notably, nine genes were expressed at least two-fold higher in the CNTF animals. Of these 9, 5 were integrins, namely Itgae, Itgal, Itgam, Itgax, and Itgb2. On the other hand, there were five genes expressed at least two-fold higher in the PTEN-deletion animals, namely Col4a3, Mmp15, Mmp7, Mmp9, and Postn (Fig. 8).
Integrins play key roles in promoting cell adhesion and neurite growth (Tomaselli et al., 1988;Izumi et al., 2017;Nieuwenhuis et al., 2018). High expression of integrins in the CNTF-treated animals supports the notion that the axons in these animals may use integrins for regeneration. To determine whether integrins are, in fact, expressed higher in the CNTF-treated RGCs, we performed immunohistochemistry on sectioned retinas. Animals received either AAV-CNTF or AAV-shPTEN injection, followed by crush two weeks later. Three days after crush, animals were perfused and eyes processed for cryosection. We tested antibodies against ITGAX and ITGAE. These antibodies were selected because several studies in the past successfully used these antibodies for immunohistochemistry. However, the ITGAX antibody failed to show obvious difference in the immunoreactive signals between the two animal groups (data not shown). On the other hand, we observed a striking difference in the intensity of ITGAE immunoreactivity between the two treatment groups. In the CNTF-treated RGCs, intense ITGAE immunoreactivity was observed, while the signals in the PTEN-deleted or uninjured RGCs were very weak (Fig. 9). We also observed that ITGAE expression was higher in layers other than the GCL, indicating that CNTF results in induction of ITGAE not only in RGCs, but also in other retinal cell types. Overall, these results show that various integrins were highly expressed after CNTF treatment, which could explain why NCAD deletion had a minimal effect on axon regeneration in the CNTF-treated animals. Figure 5. Deletion of NCAD in RGCs does not affect CNTF-induced axon regeneration or RGC survival. A, Representative optic nerve sections from WT (C57BL/6J) and NCAD f/f mice. AAV-Cre was injected intravitreally in both animal groups, followed by AAV-CNTF injection two weeks later. Animals received optic nerve crush 3 d after AAV-CNTF and survived for another three weeks after crush. Red asterisks, crush site. B, Quantification of the number of regenerating axons per section at different distances from the lesion site. N = 9 mice/group. Two-way ANOVA with Bonferroni's post hoc test; ns, not significant. C, Quantification of RGC survival. Unpaired Student's t test; ns, not significant. N = 5 mice/group; Scale bars, 100 mm.

Discussion
Despite the decades' long demonstration of lengthy CNS axon regeneration, our knowledge of the glial factors that support the migration of these axons after injury is rather limited. We do know that, depending on the types of regenerative stimuli, some axons appear to be able to migrate successfully toward their targets. In many cases, however, axons take circuitous routes back toward the cell body (Luo et al., 2013;Pernet and Schwab, 2014;Yang et al., 2020). As functional regeneration requires innervation of axons to their correct targets, we are becoming increasingly aware of the importance of defining the cellular and molecular factors that regulate this process in adulthood.
Axons cannot simply elongate in space. Rather, they normally bind to substrates. Several studies have demonstrated that regenerating axons follow astrocytes. In the mature CNS, there are two main types of astrocytes, the protoplasmic and the fibrous. The former resides mainly in the gray matter, whereas the latter resides in the white matter. Thus, in the optic nerve, which is a white matter tract, the astrocytes consist predominantly of the fibrous type. These fibrous astrocytes are further classified into three subtypes based on their morphology: the transverse subtype, with processes projecting mostly perpendicular to the optic nerve; the longitudinal subtype, with processes projecting mostly parallel to the optic nerve; and the random subtype, with processes projecting in both directions (Butt et al., 1994). These subtypes make up ;34%, 48%, and 18% of the total optic nerve astrocytes in mice, respectively (Butt et al., 1994). It is quite possible that regenerating RGC axons simply follow whichever astrocytes they come in contact with. If this is the case, then these axons will travel stochastically in all different directions, perpendicular or in parallel to the optic nerve. In support of this notion, our studies and others have shown that, in the case of regenerative treatments involving CNTF, many RGC axons in the optic nerve meander in all different directions (Luo et al., 2013;Pernet et al., 2013;Pernet and Schwab, 2014). Others have shown that some RGCs can grow toward the brain more effectively under different regenerative conditions (de Lima et al., 2012;Lim et al., 2016).
In addition to the astrocytes, other environmental factors can affect the direction of RGC axons, including immune cells, oligodendrocytes, myelin, and various growth factors and ECM proteins secreted by the glial cells (Duffy et al., 2012;Fawcett et al., 2012;Sharma et al., 2012;Geoffroy and Zheng, 2014). The myelin and axons distal to the injury site undergo Wallerian degeneration. Macrophages and, to some extent, the local microglial cells migrate throughout the injured nerve, and engulf the degenerated residues. The myelin debris and the immune cells not only physically affect the path of axons, but they also affect axon trajectories through their expression of various repulsive molecules on their surfaces, including Nogo, myelin associated glycoprotein, oligodendrocyte myelin glycoprotein, and PirB (Atwal et al., 2008;Pernet and Schwab, 2012;McKerracher and Rosen, 2015).
In this study, we focused on examining NCAD's roles on axon regeneration in vivo. It has long been known that one of NCAD's primary functions in the CNS is promoting cell-cell adhesion. In line with this function, deleting NCAD in astrocytes decreased axon regeneration, at least Figure 6. NCAD deletion in astrocytes does not alter the lesion size. A, Representative optic nerve sections from AAV-shPTEN-injected WT (i.e., GFAPCreERT; YFP mice with tamoxifen injections) and AAV-shPTEN-injected astrocyte-specific KO mice (i.e., GFAPCreERT; YFP; NCAD f/f , or also referred in the figure as NCAD aKO) mice. Red, GFAP immunoreactivity; White, CTB-Alexa555. Broken yellow line marks the boundary of the glial scar (i.e., lesioned area). B, Quantification of the lesion size. N = 5 for AAV-shPTEN; WT and N = 4 for AAV-shPTEN; NCAD aKO. Unpaired Student's t test; ns, not significant. Scale bar, 100 mm.
in the background of PTEN deletion. While it is likely that the loss of regeneration is because of the loss of NCAD's homophilic adhesive action in the KO mice, we cannot rule out the possibilities that other mechanisms contributed to this loss. The activity of mTORC1 in RGCs is an important neuron intrinsic factor for RGC axon regeneration after PTEN deletion. However, NCAD deletion did not affect mTORC1 activity, which suggests that this mechanism is not involved. It remains to be seen whether NCAD controls other signaling pathways in RGCs relevant to axon regeneration, including KLFs and HDACs. It seems unlikely that NCAD deletion in astrocytes will cause changes in neuron-intrinsic signaling factors, but nonetheless, this possibility has not been examined. Figure 7. NCAD deletion in astrocytes does not alter the number of pS6 immunoreactive RGCs. A, Representative images taken from retinal sections of contralateral intact, AAV-shPTEN-injected WT (i.e., GFAPCreERT; YFP mice with tamoxifen injections), and AAV-shPTEN-injected astrocyte-specific KO mice (i.e., GFAPCreERT; YFP; NCAD f/f, or also referred in the figure as NCAD aKO) mice. Eyes were dissected out three weeks after optic nerve crush. B, A higher magnification image of the boxed area in A. Yellow arrows point to pS61 RGCs. C, Quantification of pS61 RGCs, presented as percentage of the contralateral eye. Unpaired Student's t test; ns, not significant. Scale bars, 50 mm.
A previous study has shown that blocking NCAD functions in the spinal cord using an NCAD antibody reduces the formation of inhibitory fibrotic scar, thus resulting in enhanced axon regeneration (Hara et al., 2017). It has also widely been shown that lesion size negatively correlates with axon regeneration (i.e., the larger the lesion, the smaller degree of axon regeneration). In our model, where NCAD was deleted specifically in astrocytes, we did not  Figure 8, the gene symbol and value of log2 fold change of mRNA expression (shPTEN compared with CNTF) Layout 1 2 3 4 5 6 7 8 9 10 11 12 Figure 8. PCR array expression analysis shows differential expression of integrins between the shPTEN-treated and the CNTF-treated retinas. RT 2 Profiler PCR Array was used to profile 84 genes on eight samples (i.e., four AAV-shPTEN-injected retinas and four AAV-CNTF-injected retinas). A, The volcano plot identifies significant gene expression changes. Red circles, upregulated genes from shPTEN versus CNTF (i.e., genes expressed higher in shPTEN compared with CNTF). Green circle, downregulated genes from shPTEN versus CNTF (i.e., genes expressed lower in shPTEN compared with CNTF). Plotted are the log2 of the fold changes in gene expression on the x-axis versus their statistical significance on the y-axis. The center vertical line indicates unchanged gene expression, while the two outer vertical lines indicate the selected fold regulation threshold. The horizontal line indicates the selected p value threshold. B, Heat map of the CAM and ECM RT 2 Profiler PCR Array. The magnitude of the log2 fold change in mRNA expression of each gene is represented by the color of each square. Yellow indicates mRNA overexpression (i.e., genes expressed higher in shPTEN compared with CNTF). Purple indicates reduced mRNA expression (i.e., genes expressed lower in shPTEN compared with CNTF). Of the nine genes underexpressed in the shPTEN retinas (with fold regulation cut off of À2.0), five were integrin genes.
observe obvious changes in lesion size. However, NCAD deletion could have altered the lesion environment in different ways, and caused a decrease in regeneration. For instance, astrocytes are known to secret various regeneration promoting factors, including cytokines and trophic factors (Wiese et al., 2012;Anderson et al., 2016). Thus, another possibility is that NCAD deletion may have reduced the release of such factors in the injured optic nerve. Alternatively, considering that NCAD would have been deleted in the astrocytes and the Müller cells in the retinas of GFAPCreERT; NCAD f/f mice, it is also possible that the reduced regeneration may have resulted from changes that took place in these cells locally in the retina. It is also worth noting that several studies have described the existence of multiple transcript variants of NCAD resulting from alternative splicing. In fact, the expression patterns of distinct NCAD isoforms and their roles have been studied extensively in the invertebrate species including the Drosophila (Yonekura et al., 2006;Schwabe et al., 2014). In vitro cell aggregation assays performed on Drosophila cell lines revealed that all NCAD isoforms mediate homophilic interactions, but the isoforms encoded by different exons exhibit different adhesive activity (Yonekura et al., 2006). These studies have suggested that NCAD alternative splicing might provide a mechanism to fine-tune its adhesive activity at different developmental stages. However, the extent to which different NCAD isoforms are expressed in the adult mice, and whether they contribute to RGC axon regeneration and neuronal survival in the mammals remain unknown.
Integrins are a/b heterodimeric adhesion glycoprotein receptors that regulate critical cellular processes including cell adhesion, and growth. Integrins bind ECM proteins through their ectodomain, and the presence of an appropriate integrin heterodimer on the neuronal surface determines whether a regenerating neuron possesses the ability to grow on a particular ECM molecule (Tan et al., 2011;Eva and Fawcett, 2014;Fawcett, 2017). Integrins are known to interact with many astrocyte-expressed factors that can directly or indirectly support cell adhesion. For example, osteopontin is recognized by a4b 1, a5b 1, aVb 1 and aVb 5, while tenascin-C is recognized by a8b 1 and aVb 3. Moreover, integrin-aVb 3 forms complexes with syndecan-4, which in turn interacts with neuronal Thy-1 receptors (Avalos et al., 2009;Hillen et al., 2018). Moreover, in vitro studies have demonstrated that, in addition to NCAD, integrins are a major surface system that mediates neurite outgrowth on astrocyte surfaces; NCAD functions prominently in the outgrowth of neurites on astrocytes by E8 and E14 chick ciliary ganglion (CC) neurons. NCAD and integrin b 1 antibodies together eliminates CG neurite outgrowth on cultured astrocytes. Thus, it was proposed that NCAD and integrin receptor systems are important in regulating axon growth on astroglia in vivo (Tomaselli et al., 1988).
In contrast to the results seen in the PTEN-deleted mice, we observed that NCAD deletion, either in RGCs or astrocytes, does not reduce CNTF-induced regeneration. These results suggest that the manner by which regenerating axons adhere to glial cells might be quite distinct for axons under different conditions. Since different combinations of integrin heterodimers can promote adhesion to astrocytes, and thus they have the ability to compensate for NCAD deletion, integrins may have seized a dominant role and propelled axon regeneration in the CNTF-treated animals. Furthermore, our studies and others support a model in which CNTF overexpression leading to induction of differential expression of integrins (and possibly other CAMs) allows axons to adhere on the various astrocyte subtypes more extensively. This, in turn, will result in circuitous axon growth in the injured optic nerve. In our unpublished work, we applied 3D imaging and traced single regenerated RGC axons (Bray et al., 2017) in PTENdeleted mice. In these mice, we observed less aberrant RGC growth (i.e., transverse and circular growth) in the optic nerve compared with the AAV-CNTF-injected mice, supporting the proposed model. Having said that, we did not functionally examine whether the various integrins expressed highly in CNTF animals in fact mediate RGC axon adhesion to astrocytes and promote regeneration. Furthermore, as mentioned above, numerous factors can affect the path of axons. Thus, it remains rather speculative that differential expression of CAMs in RGCs causes the various growth patterns seen under different regenerative conditions.
Under different regenerative conditions, it is also possible that CAMs and guidance factors are differently expressed in the astrocytes (or in other cell types) in the injured optic nerve, thus contributing to the difference seen between the PTEN deleted and the CNTF-treated animals. In a previous study, RNA-sequencing performed on the spinal cord astrocytes after spinal cord injury revealed changes in the expression of several genes whose products are known to mediate cell adhesion or axon-glial interaction (Anderson et al., 2016). For example, Anderson et al., has shown that the expression of Cntn2, Itgam, Itgb2, Ephb1, and Vcan were upregulated more than twofold in the astrocytes after spinal cord injury. On the other hand, there was ;40% reduction in the expression of Cdh2 in the astrocytes after SCI (Anderson et al., 2016). In an unpublished study, we performed RNA-sequencing on the optic nerve astrocytes of adult mice. We observed that the expression of Cdh2 in the optic nerve astrocytes is unchanged after optic nerve crush. Nonetheless, we observed that numerous genes that encode CAMs and guidance factors are differentially expressed in the optic nerve astrocytes after injury. For example, Cdh15, Icam1, Sdc4, and Epha3 were upregulated more than twofold after optic nerve crush injury (unpublished observation). However, it remains to be seen whether the products of these genes in fact modulate the glial environment at the lesion site, or mediate axon-glial interaction, and promote RGC axon regeneration.
Cadherin-mediated cell-cell adhesions have been implicated in regulating cell survival during development and preserving tissue homeostasis. Indeed, studies have reported that interfering with cadherin adhesion triggers apoptosis in various non-neuronal cells, including tumor cells (Nguyen et al., 2018). However, NCAD's specific role for neuronal survival is not well understood. One in vitro study reported that NCAD contributes to the survival of dissociated neurons (Lelièvre et al., 2012); plating spinal or hippocampal neurons on NCAD recombinant substrate enhanced neuronal survival compared with non-specific adhesion on poly-L lysine. NCAD engagement, in the absence of other survival factors (cell-matrix interactions and serum) was shown to protect neuronal cells against apoptosis. Nonetheless, to our knowledge, our study is the first to demonstrate that NCAD contributes to the survival of CNS neurons after injury. The mechanism by which NCAD provides neuroprotection is unclear. The in vitro study did examine the potential signaling pathways involved; the phosphoinositide 3-kinase (PI3K)/Akt survival pathway and its downstream effector BAD were not involved in the NCAD-mediated survival of GT1-7 neuronal cells. In contrast, NCAD activated the Erk1/2 MAP kinase pathway, reducing the level of the proapoptotic protein Bim-extralong (EL) in these cells (Lelièvre et al., 2012). Another study using a prostate carcinoma line demonstrated that NCAD-catenin adhesion complex results in up-regulation of the anti-apoptotic protein Bcl-2, whereas the level of the proapoptotic protein Bax remained constant. NCAD homophilic ligation initiated PI3K-dependent activation of Akt resulting in Akt phosphorylation of Bad (Tran et al., 2002). Together, these studies indicate that interruption of NCAD-mediated cellcell interactions can induce an increase in the expression of the proapoptotic Bcl-2 family of proteins and tip the balance toward programmed cell death. If so, it is unclear why NCAD deletion affected RGC survival in the PTENdeleted, but not in the CNTF-treated animals. It seems plausible, however, that CNTF induces signaling pathways, including the JAK/STAT3 pathway that can counterbalance the detrimental effects of NCAD deletion.
Several studies have documented that different RGC subtypes respond differently to an injury with some RGC subtypes more resilient than others (Duan et al., 2015;Bray et al., 2019;Tran et al., 2019). We note that the AAV-Cre used in this study deletes NCAD in whole RGCs, and not in specific RGC subtypes. Our results showed that NCAD deletion in RGCs reduces the overall number of surviving RGCs after PTEN deletion. However, some RGC subtypes may express NCAD and rely on this protein for survival and/or regeneration while other subtypes might be NCAD independent. Alternatively, NCAD may in fact function in a detrimental manner in some RGC subtypes after injury, and deletion of NCAD in these RGCs may have actually resulted in increase in their survival and regeneration. In this sense, defining NCAD's general role in the CNS neurons using the AAV-Cre approach is not without a caveat. While it is beyond the scope of this current study, it might be interesting in the future to use Cre driver lines that allow deletion of NCAD in specific RGC subtypes, and comprehensively determine NCAD's role in RGCs.
In summary, we show that NCAD can engage in promoting the survival of CNS neurons and axon regeneration. However, its contributions are dispensable, and likely compensated for by other factors under certain stimulating conditions. The last decade has unveiled multiple strategies that promote lengthy axon regeneration. However, the propelled growth is accompanied by a lack of direction in the degenerated tract, warranting investigations into the cellular and molecular guidance factors that determine the fate of growing axons. Dissecting the mechanisms by which NCAD and other CAMs shape the paths of axons could help devise better ways to promote functional axon regeneration.