GDE3 regulates oligodendrocyte precursor proliferation via release of soluble CNTFRα

Oligodendrocytes are important regulators of neuronal function and survival. Their formation during development occurs through a complex step-wise process that includes the specification of progenitors towards oligodendroglial fates, and a tightly controlled program of proliferation and differentiation (Mitew et al., 2014; Rowitch and Kriegstein, 2010). Emergent studies suggest involvement of multiple signaling pathways at these junctures, but how they work to control these aspects of oligodendroglial development is not well understood. In the developing spinal cord, oligodendrocytes are primarily derived from a ventral region in the ventricular zone (VZ) termed pMN. pMN progenitors are initially specified towards motor neuron fates, but once neurogenesis is complete, pMN progenitors undergo a transition known as the neuron/glial switch, where they give rise to oligodendrocyte precursor cells (OPCs) (Kessaris et al., 2001). Astrocytes are not generated from the pMN but are produced from progenitors located within discrete domains throughout the VZ (Muroyama et al., 2005).

OPCs within the pMN express Sox10, platelet derived growth factor receptor α (PDGFRα) and NG2, and OPCs continue to proliferate as they migrate throughout the spinal cord (Goldman and Kuypers, 2015; Mitew et al., 2014). OPCs subsequently differentiate into pre-myelinating oligodendrocytes and, in turn, myelin-producing oligodendrocytes that ensheath axons to facilitate fast, saltatory conduction of neuronal action potentials. Importantly, a subset of OPCs do not differentiate but maintain their proliferative and migratory potential (Bergles and Richardson, 2015). These resident OPCs survey the adult CNS for injury and inflammatory damage that necessitates oligodendrocyte repopulation. Many signaling factors have been identified as indispensable for oligodendrocyte proliferation, suggesting that the control of OPC proliferation is complex and requires integration of multiple signaling networks. These include PDGF, ciliary neurotrophic factor (CNTF), leukemia inhibitory factor (LIF), neuregulins and fibroblast growth factors (FGFs) (Bergles and Richardson, 2015; Calver et al., 1998; Deverman and Patterson, 2012; Mayer et al., 1994; Sleeman et al., 2000). Most signals discovered so far are found to stimulate rather than inhibit OPC proliferation (Mitew et al., 2014). However, to ensure that OPCs are generated in a timely manner without depleting the progenitor pool, it is probable that additional mechanisms operate developmentally to restrict OPC proliferation. Indeed, the existence of such mechanisms is supported by the observation that OPC proliferation rates slow as development progresses (van Heyningen et al., 2001). Detailed understanding of the mechanisms that negatively regulate OPC proliferation is lacking.

Glycerophosphodiester phosphodiesterase 3 (GDE3, also known as GDPD2) is a member of a small family of six transmembrane proteins with an external enzymatic domain related to bacterial glycerophosphodiester phosphodiesterases (GDPD) (Corda et al., 2014; Yanaka, 2007). This family includes GDE2 and GDE6, and these proteins are the only known enzymes in vertebrates that function at the cell surface to cleave the glycosylphosphatidylinositol (GPI) anchor that tethers some proteins to the cell membrane (Park et al., 2013). GDE3 is reported to use analogous PLC-type cleavage mechanisms to metabolize glycerophosphoinositol to produce inositol 1-phosphate and glycerol (Corda et al., 2009). Essential roles for GDE2 in regulating cellular differentiation have been established. GDE2 is expressed during the neurogenic period in the spinal cord, where it uses its GPI-anchor cleavage function to induce motor neuron differentiation by downregulating Notch signaling in pMN progenitors (Park et al., 2013; Rao and Sockanathan, 2005; Sabharwal et al., 2011). GDE2 also controls pancreas development and plays key roles in the differentiation of neuroblastoma through cleavage of the heparan sulfate proteoglycan GPC6 (Matas-Rico et al., 2016; van Veen et al., 2018). Functions for GDE6 have not been identified. Emerging studies suggest that GDE3 also plays roles in controlling cellular proliferation and differentiation. GDE3 negatively regulates proliferation of osteoblast cell lines and slows tumor growth in mouse xenografts through GPI-anchor cleavage of uPAR (van Veen et al., 2017; Yanaka et al., 2003). However, functions for GDE3 in the nervous system have not been addressed. Here, we examine the role and mechanism of GDE3 function in regulating OPC proliferation in the developing spinal cord.

To gain insight into GDE3 function in the spinal cord, we defined the developmental profile of Gde3 (Gdpd2) expression during embryogenesis. Analysis of Gde3 transcripts by in situ hybridization shows no Gde3 expression in mouse spinal cord sections at embryonic day (E) 10.5 when neurogenesis is occurring (Fig. 1A). Instead, Gde3 expression is first detected at E11.5 in progenitor cells located in the ventral VZ when neurogenesis is nearly complete (Fig. 1B). Gde3 remains highly expressed in the VZ from E12.5 but begins to extend more dorsally (Fig. 1C). By E14.5, Gde3 expression is detected in cells outside the VZ, and by E16.5 Gde3 expression is broadly dispersed in cells throughout the spinal cord (Fig. 1D,E). Analysis of Gde3 distribution in the chick spinal cord shows a similar pattern of expression, suggesting that GDE3 function is likely conserved between mouse and chick (Fig. S1).

The temporal and spatial profile of Gde3 expression coincides with the onset and progression of gliogenesis (Mitew et al., 2014; Rowitch and Kriegstein, 2010). Analysis of adjacent sections at E12.5 shows that Gde3 transcripts are restricted primarily to ventral progenitor domains (p0-p3) that lack Pax7 expression and are the site of OPC and astrocyte precursor (APC) formation (Fig. 1F-H). For detailed list of reagents, see Table S3. Olig2⁺ OPCs and Aldh1L1⁺ APCs begin to migrate out of the ventral VZ from ∼E12.5 (Fig. 1G,H) (Fu et al., 2002; Muroyama et al., 2005). Using fluorescent in situ hybridization, we determined that Gde3⁺ cells located outside the VZ do not express the neuronal marker NeuN (Fig. 1I-1I″), suggesting that Gde3 is not expressed in neurons. However, Gde3 transcripts colocalize with PDGFRα⁺ and Olig2⁺ cells, indicating expression in migrating OPCs (Fig. 1J-K″) (Fu et al., 2002). Furthermore, Gde3⁺ cells express NFIA, a transcription factor that is required for the cessation of neurogenesis and the initiation of gliogenesis, and which continues to be expressed by astrocytes (Fig. 1L-1L″) (Deneen et al., 2006). These observations indicate that the majority of Gde3-expressing cells comprise glial progenitors specified to astroglial or oligodendroglial lineages.

We used Cre-lox genetics to generate stable mouse lines with disrupted GDE3 function. We confirmed that this strategy effectively ablated GDE3 expression using PCR, in situ hybridization and western blot (Fig. S2). Gde3 is located on the X chromosome, and Gde3 KO animals (Gde3^−/− or Gde3^−/y) are viable, fertile and appear indistinguishable from Gde3^+/− or wild-type (Gde3^+/+ or Gde3^+/y) littermates.

To test whether GDE3 regulates the proliferation of glial progenitors in the spinal cord, we quantified the number of ventral and dorsal VZ progenitors in wild-type and Gde3 KOs that express Sox9 and NFIA (Deneen et al., 2006; Stolt et al., 2003). We found an ∼20% increase in the number of ventral Pax7⁻ Sox9⁺ progenitors and a roughly 18% increase in Pax7⁻ NFIA⁺ cells in Gde3 KO spinal cords at E12.5 compared with wild-type littermates (Fig. 2A). In contrast, the numbers of Sox9 and NFIA progenitors in dorsal Pax7⁺ progenitors that normally do not express GDE3 at this timepoint were equivalent between wild-type and Gde3 KOs (Fig. 2B). To determine whether the increase in ventral progenitors is a consequence of altered cellular proliferation, we measured the length of S-phase and the cell cycle in wild-type and Gde3 KO spinal progenitors at E12.5. We used an established protocol that involves sequential injection of the thymidine analogues BrdU and IdU at an interval of 1.5 h before harvesting of embryos (Martynoga et al., 2005). We found that ventral progenitor cells in Gde3 KO spinal cords have a shorter S phase, measuring 8.50±0.52 h compared with 11.61±0.69 h in wild-type littermates (Fig. 2C). In addition, ventral progenitors in Gde3 KOs exhibited a reduced cell-cycle length of 20.97±2.98 h compared with 27.20±2.27 h in wild-type animals (Fig. 2C). No change in S-phase of dorsal VZ progenitors that normally do not express GDE3 was detected (Fig. 2D). Thus, ventral progenitor cells lacking GDE3 have reduced S-phase and cell-cycle length, and an increased proliferative rate compared with wild-type counterparts.

Olig2⁺ progenitors in the pMN domain begin to be specified towards the oligodendroglial lineage at E12.5 and this is manifest by upregulation of the oligodendrocyte lineage determinant Sox10 (Fig. 2E,F) (Stolt et al., 2004, 2002). Consistent with increased proliferative capacity, we detected an 11% increase in the overall number of Olig2⁺ cells in Gde3 KOs compared with wild-type littermates (Fig. 2H). However, there was an ∼25% reduction in the ratio of Sox10⁺Olig2⁺:Olig2⁺ progenitors (Fig. 2G,H). The delay in Sox10 expression in Olig2⁺ cells raises the possibility that OPC specification may be delayed in the absence of GDE3.

We next determined whether GDE3 is required to regulate the proliferation of OPCs once they have been specified. We directly compared the numbers of Gde3 KO and wild-type Olig2⁺ cells in a mosaic genetic model. Gde3 is X-linked; therefore, female Gde3^+/− animals are natural mosaics, expressing either their maternal or paternal X-chromosome on a cell by cell basis due to X-inactivation. The HprtGFP mouse is an X-chromosome reporter that expresses GFP and will co-segregate with a wild-type Gde3 allele (Fig. 3A) (Wu et al., 2014). By breeding HprtGFP;Gde3^+/y males to heterozygous Gde3^+/− females, we generated female embryos that were HprtGFP;Gde3^+/+ or HprtGFP;Gde3^+/− (Fig. 3A). In HprtGFP;Gde3^+/+ embryos derived from this strategy, all Olig2⁺ cells are wild type, irrespective of GFP expression. In contrast, in HprtGFP;Gde3^+/− embryos, GFP co-expression marks wild-type cells, while Olig2⁺ cells that lack GFP correspond to Gde3 KO cells (Fig. 3B-D). At E14.5, we observed that GFP-Gde3 KO cells constitute a significantly higher proportion of the total Olig2⁺ population in HprtGFP;Gde3^+/− animals compared with GFP⁻ wild-type cells within HprtGFP;Gde3^+/+ littermates (Fig. 3E). At postnatal day 6 (P6), GFP⁻ Gde3 KO cells make up a significantly larger proportion of proliferating Ki67⁺ OPCs compared with GFP⁻ wild type (63% compared with 48%; Fig. 3F). These observations suggest that ablation of GDE3 leads to an increase in proliferating OPCs. Moreover, because Gde3 KO OPCs occupy the same microenvironment as wild-type OPCs in HprtGFP;Gde3^+/− mice, their increased proliferation suggests that normally, GDE3 acts cell-autonomously to inhibit OPC proliferation (Fig. 3F).

Pathways that stimulate OPC proliferation can inhibit oligodendrocyte differentiation (Bergles and Richardson, 2015; Mitew et al., 2014). To determine whether GDE3 ablation negatively impacts OPC differentiation, we examined HprtGFP;Gde3+/+ and HprtGFP;Gde3+/− animals at age P14, when expression of CC1 distinguishes premyelinating and myelinating oligodendrocytes from OPCs (Bin et al., 2016). No differences are detected between the GFP⁺ and GFP⁻ CC1/Olig2 populations in HprtGFP;Gde3^+/+ controls (Fig. 3H). However, we found that ∼68% of GFP- Olig2+ Gde3 KO cells expressed CC1 in contrast to only 47% of GFP⁺ Olig2⁺ wild-type cells (Fig. 3G,H) in HprtGFP;Gde3^+/− animals. The increase in differentiated CC1⁺ GFP⁻ Gde3 KO cells at P14 is proportional to the increase in proliferating GFP⁻ Gde3 KO Olig2⁺ cells at P6 (Fig. 3F), consistent with the differentiation of Gde3 KO OPCs into CC1⁺ oligodendrocytes. These observations suggest that GDE3 regulates the pace of oligodendrocyte generation through the regulation of OPC proliferation. In support of this notion, expression of myelin basic protein (MBP), a marker of mature oligodendrocytes, is elevated in Gde3 KO animals compared with wild-type littermates during the period of developmental myelination (Fig. S3). However, wild-type and Gde3 KO animals show equivalent amounts of Olig2, and the myelin-associated proteins MBP and myelin oligodendrocyte glycoprotein (MOG), and a similar density of Olig2⁺ cells and numbers of Olig2⁺ CC1⁺ oligodendrocytes at P30, indicating that myelination has normalized in Gde3 KOs by this stage (Fig. S3). This is consistent with known homeostatic control of OPC and oligodendrocyte numbers that ensure appropriate myelination (Bergles and Richardson, 2015).

Our studies so far suggest that GDE3 negatively regulates OPC proliferation. To test whether GDE3 expression is sufficient to inhibit OPC proliferation, we transfected purified OPCs from P2-P4 wild-type cortices with a bicistronic construct that expresses GDE3 and GFP, which was introduced downstream of an IRES sequence 3′ to the GDE3-coding sequence. Transfected OPC cultures were grown in conventional growth factor-rich media containing PDGF and CNTF that maintains OPCs in a proliferative condition (Dugas and Emery, 2013). Cultures were pulsed for 2 h with IdU to label cells in S phase and stained for NG2 to confirm OPC identity. The percentage of transfected OPCs (NG2⁺GFP⁺) that had incorporated IdU (IdU⁺NG2⁺GFP⁺) was then quantified. Compared with when GFP alone is expressed, expression of GDE3 in OPCs resulted in a 20% reduction of OPCs undergoing S phase (Fig. 4A,B; Table S1). Thus, GDE3 is sufficient to negatively regulate the rate of OPC proliferation. Our studies of HprtGFP;Gde3^+/− mosaic animals suggest that GDE3 functions cell-autonomously to inhibit OPC proliferation (Fig. 3F). To directly address autonomy of GDE3 function, we quantified the percentage of non-transfected NG2⁺ cells that had incorporated IdU (IdU⁺NG2⁺GFP⁻, Fig. 4A,B; Table S1) in OPC cultures transfected with GDE3 or control GFP. In contrast to OPCs transfected with GDE3, neighboring untransfected NG2⁺ cells in control and experimental conditions showed equivalent levels of IdU incorporation. These observations suggest that GDE3 acts cell-autonomously to inhibit OPC proliferation. In support of this notion, cultured OPCs purified from Gde3 KO neonatal spinal cords showed higher proliferative activity than wild-type counterparts; ∼35% more Gde3 KO spinal OPCs incorporated IdU compared with wild-type cultures (Fig. 4C,D,G). Purified cortical OPCs from Gde3 KOs showed a comparable 27% increase in IdU incorporation compared with wild-type littermates (Fig. 4E-G) revealing shared functionality of GDE3 in controlling spinal and cortical OPC proliferation.

To gain insight into the mechanism by which GDE3 regulates OPC proliferation, we performed RNA-seq using cDNA libraries generated from OPCs purified from P0.5 wild-type and Gde3 KO spinal cords that had been grown in proliferation medium for 16 h. 50 bp paired end reads were generated yielding 50-61 million reads per sample; mapping rates to the mouse genome were 98% with 86% representing exonic reads, indicating very high-quality sequences. Ninety-six differentially regulated genes were found between wild-type and Gde3 KO OPCs (Fig. 5A; Table S2). Functional enrichment analysis of the differentially expressed gene products using the STRING database (v10.5) identified several pathways that are perturbed in Gde3 KO OPCs compared with wild type (Szklarczyk et al., 2017). In support of GDE3 function in OPC proliferation and differentiation, we identified genes encoding proteins involved in the regulation of cell differentiation (GO.0045595) and the regulation of developmental processes (GO.0050793) (Fig. 5B). Interestingly, we detected substantial changes in components of cellular response to cytokines (GO.0034097) and extracellular exosomes (GO.0070062) (Fig. 5B). We also detected an increase in immune response genes (GO.0002376, GO.0006955) in Gde3 KO OPCs (Table S2).

We examined the possibility that cytokine pathways contribute to the increased proliferative rates of Gde3 KO OPCs. Because GDE3 is an enzyme that cleaves GPI anchors and regulates GPI-anchored protein (GPI-AP) activity, we focused our attention on cytokine pathways that have regulatory components that are GPI-APs. We curated a database of annotated GPI-APs in the mouse genome, cross-referenced them to published expression datasets of genes that are enriched in OPCs, and identified candidates with known roles in cytokine signaling. We identified six GPI-APs that met these criteria: CNTF receptor α (CNTFRα), semaphorin 7a, signal transducer CD24, Rtn4r (reticulon 4 receptor), Rtn4rl1 (reticulon 4 receptor-like 1) and Bst2 (bone marrow stromal antigen 2) with CNTF receptor α (CNTFRα) having the highest expression in OPCs. CNTF binds to CNTFRα, which recruits gp130 and LIFRβ in a multiprotein receptor complex to initiate downstream signal transduction cascades that include activation of MAPK, ERK, Akt and STAT signaling pathways (Davis and Yancopoulos, 1993; Port et al., 2007; Sleeman et al., 2000). CNTF signaling is implicated in oligodendrocyte proliferation, differentiation, maturation and survival (Barres et al., 1996; Mayer et al., 1994; Sleeman et al., 2000; Stankoff et al., 2002), and CNTFRα is widely expressed in neurons and glia in the spinal cord (Gregg and Weiss, 2005; Ip et al., 1993). Surface biotinylation analysis of purified OPCs show that GDE3 and CNTFRα are co-expressed on the surface of OPCs (Fig. 5C). Thus, the CNTF/CNTFRα pathway is a promising candidate mechanism for mediating GDE3-dependent inhibition of OPC proliferation.

To test whether GDE3 regulates OPC responsiveness to CNTF, we cultured wild-type and Gde3 KO OPCs in the presence or absence of CNTF for 16 h. IdU was added to cultures 2 h before fixation to label cells in S phase (Fig. 5D-G′). In wild-type OPCs, the omission of CNTF from the media resulted in a 32% increase in IdU incorporation (Fig. 5H, Table S1). This increase is analogous to the effect of ablating Gde3 in the presence of CNTF. Furthermore, omitting CNTF from the media of Gde3 KO cultures produces no change in OPC proliferation (Fig. 5H, Table S1). CNTF can induce OPCs to differentiate into astrocytes (Hughes et al., 1988). Wild-type and Gde3 KO OPCs showed equivalent numbers of GFAP⁺ astrocytes in the presence or absence of CNTF, suggesting that CNTF does not induce astrocyte formation in our culture conditions (Fig. S4). These observations suggest that CNTF normally works to reduce OPC proliferation via GDE3, and that the hyperproliferation of Gde3 KO OPCs may result from defective CNTF signaling.

Our observation that CNTF inhibits OPC proliferation contrasts with previous studies that find CNTF stimulates OPC proliferation; however, OPCs in the latter study were isolated from P14 optic nerve (Barres et al., 1996). To determine whether the differential effects of CNTF on OPC proliferation are due to region-specific differences in OPCs, we repeated the experiment using OPCs isolated from P14 rat optic nerve, P14 rat cortex and P14 mouse cortex. Strikingly, CNTF stimulates the proliferation of optic nerve OPCs but inhibits the proliferation of rat or mouse cortical OPCs (Fig. S4). Thus, OPCs show region-specific response to CNTF application, with CNTF inhibiting proliferation of cortical OPCs.

Soluble forms of CNTFRα are detected in vivo, and soluble CNTFRα stimulates CNTF-CNTFRα-gp130 signaling (Davis et al., 1993; Davis and Yancopoulos, 1993). GDE3 releases GPI-APs from the cell surface through cleavage at the GPI-anchor (Park et al., 2013). Accordingly, GDE3 could generate soluble forms of CNTFRα that would stimulate CNTF-mediated inhibition of OPC proliferation. To determine whether GDE3 releases CNTFRα from the cell membrane, we transfected constructs expressing GDE3 and CNTFRα into HEK293FT cells and analyzed the medium after 24 h for the presence of CNTFRα by western blot. Co-expression of GDE3 and CNTFRα in HEK293FT cells resulted in robust release of CNTFRα into the overlying medium, whereas expression of CNTFRα alone displayed minimal release (Fig. 6A). Transfected cells are cultured in absence of CNTF, indicating that GDE3 release of CNTFRα is not dependent on presence of CNTF. Notably, CNTFRα released by GDE3 consisted of two closely migrating bands, suggesting that GDE3 releases two different forms of CNTFRα.

Incubation of transfected cells with bacterial PI-PLC, which cleaves GPI anchors, only releases the smaller form, suggesting that this corresponds to soluble CNTFRα released by GPI-anchor cleavage (Fig. 6A). We generated mutated versions of GDE3 that lack GPI-anchor cleavage activity. GDE3 is reported to be a seven-transmembrane protein (Yanaka et al., 2003). However, epitope-mapping shows that GDE3 shares the same six-transmembrane structure as family member GDE2, with intracellular N-terminus, extracellular enzymatic domain and intracellular C-terminal domain (Figs S5 and S6C). GPI-anchor cleavage by GDE2 is dependent on key histidine residues within the enzymatic domain (Matas-Rico et al., 2016; Park et al., 2013; Rao and Sockanathan, 2005). We mutated homologous histidine residues in GDE3 (H233A and H275A) to create an enzymatically impaired version of GDE3 (GDE3.HHAA) (Fig. 6C). Analysis of GDE3.HHAA release properties in transfected HEK293FT cells showed that GDE3.HHAA fails to release cleaved CNTFRα, confirming that GDE3 can cleave the GPI-anchor of CNTFRα (Fig. 6B). To identify a version of GDE3 that lacks the ability to release all forms of CNTFRα, we constructed a version of GDE3 lacking the intracellular N terminus (GDE3.ΔN) and a double mutant of GDE3 that is catalytically impaired and lacked the N-terminal domain (GDE3.HHAAΔN) (Fig. 6C). Surface biotinylation studies in HEK293FT cells confirmed that all mutant forms of GDE3 were stable and trafficked to the plasma membrane (Fig. S5). GDE3.ΔN fails to release the slower migrating form of CNTFRα, but still releases the cleaved form (Fig. 6B). In contrast, GDE3.HHAAΔN does not release CNTFRα in either form (Fig. 6B). These observations indicate that the N terminus and GDE3 catalytic activity are separately required to release distinct forms of CNTFRα from the cell surface.

We examined the ability of GDE3.HHAA, GDE3ΔN and GDE3.HHAAΔN to inhibit OPC proliferation compared with wild-type GDE3. We transfected constructs expressing GDE3 and GDE3 mutants into freshly isolated wild-type OPCs and performed pulse labeling with IdU after 24 h in culture. Quantification of the percentage of transfected cells that had incorporated IdU showed that GDE3.HHAA was as effective as wild-type GDE3 in suppressing OPC proliferation, causing a 17% decrease in IdU uptake, while GDE3.ΔN showed a more robust 30% reduction in proliferating OPCs (Fig. 6D, Table S1). However, expression of GDE3.HHAAΔN showed no effect on OPC proliferation rates and was equivalent to GFP control (Fig. 6D, Table S1). These observations indicate that the ability of GDE3 to inhibit OPC proliferation correlates with its capacity to release CNTFRα from the cell surface.

RNA-seq reveals that GDE3 ablation impacts extracellular vesicle (EV) pathways (Fig. 5B). We examined the possibility that the larger, slower-migrating form of GDE3-released CNTFRα is tethered to EVs. Triton X-114 detergent fractionation of medium from transfected HEK293FT cells showed that the higher form of CNTFRα released by GDE3 and GDE3.HHAA partitioned into the hydrophobic detergent phase, indicating this form is released within a membrane compartment (Fig. 6E). In addition, sequential high-speed EV purification protocols (Livshits et al., 2015) enriched for released CNTFRα in the presence of GDE3 or GDE3.HHAA, and for EV components CD9 and CD63 (Fig. S6) (van Niel et al., 2018). Confirming successful purification of EVs by this method, electron micrographs of samples revealed the presence of membrane vesicles with ‘cup’ shape morphology typical of EVs when GDE3 is expressed (Fig. S6). Thus, GDE3 expression results in the release of CNTFRα in EVs.

We noted that GDE3 itself is also released in EVs (Fig. S6). This raises the possibility that GDE3 and CNTFRα are co-released into EVs, where GDE3 might generate soluble forms of CNTFRα by cleavage at the GPI anchor. To determine whether GDE3 and CNTFRα are co-released in EVs, we isolated medium from HEK293FT cells co-transfected with myc-tagged CNTFRα and flag-tagged GDE3 or GDE3.HHAA that has impaired catalytic activity. We then performed coimmunoprecipitation with myc antibodies to isolate EVs that contain CNTFRα. Analysis of precipitates by western blot detected both GDE3 and GDE3.HHAA (Fig. 6F), suggesting that CNTFRα is co-released with GDE3 or GDE3.HHAA in the same EV.

We examined whether GDE3 is capable of releasing CNTFRα from released EVs by GPI-anchor cleavage. We isolated media from HEK293FT cells expressing CNTFRα and GDE3 or GDE3.HHAA. EVs were immediately purified from one aliquot of the medium by ultracentrifugation, while remaining aliquots were incubated for 1.5 h and 24 h at 37°C prior to EV purification. The amount of CNTFRα in purified EVs was then quantified by western blot. In the presence of wild-type GDE3, levels of EV-bound CNTFRα were decreased by ∼71% of the starting level; however, no reduction was observed in the catalytically impaired GDE3.HHAA condition (Fig. 6G,H). After 24 h, we observed a further reduction of EV-bound CNTFRα in the presence of wild-type GDE3 but not GDE3.HHAA, indicating continued release of CNTFRα from EVs by GDE3 GPI-anchor cleavage (Fig. 6G,H). Addition of PI-PLC to cleave the remaining EV-bound CNTFRα before EV purification at 24 h caused a robust reduction of CNTFRα levels from EVs containing GDE3.HHAA but did not reduce levels of CNTFRα in EVs expressing wild-type GDE3 (Fig. 6G,H). This suggests minimal retention of membrane-bound CNTFRα on EVs when GDE3 is enzymatically active. Thus, GDE3 is co-released with CNTFRα in EVs, where it can release CNTFRα from EV membrane by GPI-anchor cleavage.

Our studies in heterologous cells indicate that GDE3 releases soluble forms of CNTFRα in two ways; by GPI-anchor cleavage from the plasma membrane and from EVs. To examine whether GDE3 is required to release soluble CNTFRα in vivo, we fractionated cortical extracts prepared from wild-type and Gde3 KO animals with Triton X-114. Cleaved, soluble GPI-APs are enriched in the detergent poor fraction (T2), while uncleaved membrane-bound GPI-APs are localized to the detergent-rich phase (Park et al., 2013). We observed a significant reduction of released CNTFRα in absence of GDE3, confirming that GDE3 is required to release soluble CNTFRα in vivo (Fig. 7A).

To test whether the lack of soluble CNTFRα generated by GDE3 is responsible for the increased proliferation of Gde3 KO OPCs, we supplied soluble recombinant (r) CNTFRα to Gde3 KO OPCs and examined effects on OPC proliferation. We incubated cultured Gde3 knockout and wild-type OPCs with rCNTFRα in presence of CNTF and PDGF for 18 h, and evaluated OPC proliferation by pulse labeling cultures with IdU. Addition of 1 µg/ml rCNTFRα did not alter the proliferation of wild-type OPC cultures. In contrast, addition of the same amount of rCNTFRα to Gde3 KO cultures reduced OPC proliferation to wild-type levels (Fig. 7B, Table S1). To test the sensitivity of Gde3 KO OPCs to soluble CNTFRα, we repeated the experiment but reduced the concentration of CNTFRα 100 times to 0.01 µg/ml. Here too, Gde3 KO OPCs exhibited a significant decrease in OPC proliferation that was comparable with wild-type levels. No effect was observed in wild-type controls (Fig. 7B, Table S1). These observations indicate that the regulation of OPC proliferation is exquisitely sensitive to cleaved, soluble CNTFRα and suggest that the proliferative defects in Gde3 KO OPCs are due to impaired release of soluble CNTFRα.

We show here that the six-transmembrane protein GDE3 is an essential regulator of gliogenic progenitor proliferation. GDE3 negatively controls OPC proliferation by stimulating CNTF-mediated signaling through release of soluble CNTFRα (Fig. 7C). GDE3 releases CNTFRα by two distinct mechanisms that use specific protein domains within GDE3. GDE3 cleaves CNTFRα at the GPI anchor to release it directly from the cell surface; in addition, GDE3 recruits CNTFRα through its N-terminal domain in EVs, where it is capable of cleaving and releasing CNTFRα from EV membrane. This study identifies a new physiological pathway driven by GDE3 that is essential for regulating OPC development. In addition, it uncovers new biochemical functions for GDE3 in the co-recruitment and release of GPI-anchored proteins through EV-mediated mechanisms.

Ablation of GDE3 leads to a reduction in S phase, a shortened cell cycle and increased numbers of glial precursors. Although the increase in glial progenitors is likely to contribute to elevated numbers of OPCs, our studies using the Hprt mosaic model and cultured OPCs indicate that OPCs lacking GDE3 are hyperproliferative, indicating that GDE3 normally acts as a ‘brake’ on OPC proliferation. Notably, the lack of GDE3 does not impair OPC differentiation and early postnatal Gde3 KOs show elevated levels of MBP compared with wild type, suggestive of increased numbers of mature oligodendrocytes. These observations suggest that GDE3 normally regulates the pace of OPC proliferation and the rate of oligodendrocyte generation during development. Previous studies have identified multiple factors that impact OPC proliferation, including PDGF-A, LIF, Shh, BDNF and FGFs (Bergles and Richardson, 2015; Mitew et al., 2014). In contrast to GDE3, most factors stimulate OPC proliferation and, in general, such factors also increase OPC survival and inhibit their differentiation into oligodendrocytes (Bergles and Richardson, 2015; Mitew et al., 2014). Our discovery that GDE3 acts to slow the pace of OPC proliferation provides a new window into regulatory mechanisms that prevent excessive OPC proliferation to coordinate the timing of OPC development and ensure appropriate myelination of target axons. The ability of GDE3 to negatively regulate glial progenitor and OPC proliferation is consistent with GDE3 function in other cellular contexts. For example, GDE3 is sufficient to inhibit osteoblast proliferation and initiate differentiation, and GDE3 overexpression slows tumor growth in xenograft mouse models, consistent with inhibition of proliferation (van Veen et al., 2017; Yanaka et al., 2003). Our study, combined with these previous observations, reinforces the concept that GDE3 is an important physiological regulator of cellular proliferation in multiple contexts. Here, we have focused on developmental functions of GDE3 in OPC proliferation. In the adult, OPCs form a quiescent population that undergoes active proliferation and differentiation in response to injury or insult (Bergles and Richardson, 2015). It would be interesting to determine whether GDE3 inhibitory function extends to maintain homeostasis of resident OPCs in the adult nervous system.

Our studies suggest that GDE3 mediates CNTF-dependent inhibition of OPC proliferation through release of soluble forms of CNTFRα. The ability of cleaved CNTFRα to stimulate CNTF signaling is in line with previous studies that demonstrate that soluble CNTFRα potently stimulates CNTF signaling (Davis et al., 1993; Davis and Yancopoulos, 1993). It is notable that many receptors are found in soluble forms and these forms usually exhibit dominant-negative activity; in contrast, soluble CNTFRα promotes ligand-dependent receptor activation. While the importance of soluble CNTFRα has been known for decades, the enzymes responsible for soluble CNTFRα generation have not been identified. Our studies identify GDE3 as a long-sought-after factor that releases soluble CNTFRα through GPI-anchor cleavage.

In addition to release by GPI-anchor cleavage at the plasma membrane, GDE3 recruits CNTFRα into EVs. This observation shows for the first time that CNTFRα can be released by this mechanism, and further suggests that GDE3 can regulate surface GPI-anchored protein levels through two different mechanisms. GDE3 is itself trafficked into EVs where it can release CNTFRα by GPI-anchor cleavage, consistent with the formation of a signaling EV. EVs containing GDE3 and CNTFRα provide further benefits in addition to cleavage and release from the cell surface. EV cargoes are highly stable. Accordingly, the signaling EV could constitute a stable reservoir of GDE3 and CNTFRα that could be rapidly activated to provide a source of released CNTFRα to bind CNTF and initiate co-receptor activation on the plasma membrane. This mechanism could be particularly relevant to pathways dependent on surface GPI-anchored proteins such as CNTFRα, because GPI-anchored proteins on the plasma membrane are constitutively endocytosed by the CLIC/GEEC pathway (Fujita and Kinoshita, 2012). GDE3.HHAA, which retains negligible GPI-anchor cleaving ability but still releases EVs, can inhibit OPC proliferation in a gain-of-function setting. This observation raises the possibility that CNTFRα in EVs could also be involved in delivery/maintenance of GPI-anchored CNTFRα surface expression that is necessary for CNTF signal transduction.

Our analyses of GDE3 overexpression in OPC cultures and Gde3 KO OPCs in HprtGFP animals indicate that GDE3 functions cell autonomously to regulate OPC proliferation. OPCs transfected with GDE3 show reduced proliferation while neighboring untransfected OPCs do not. Similarly, Gde3 KO OPCs in the mosaic HprtGFP model show increased proliferation compared with neighboring wild-type OPCs that occupy the same microenvironment. Our model posits that GDE3 inhibits OPC proliferation by releasing CNTFRα, which stimulates CNTF signaling. The cell autonomy of GDE3 function suggests that released CNTFRα activity is local. By the same token, the increased proliferation of Gde3 KO OPCs in HprtGFP mosaic animals rules out major contributions of CNTFRα released by GDE3 in astrocytes. If astrocytes were a major source of released CNTFRα, then the proliferative response of OPCs to this signal would be equivalent between wild-type and Gde3 KO OPCs in HprtGFP mosaic mice. How released CNTFRα mediates its effects in a local cell-autonomous fashion warrants further investigation.

CNTF is reported to have different effects on cellular proliferation. CNTF can inhibit proliferation of chick sympathetic neural progenitors, osteoblasts and neural crest cells, but can stimulate proliferation of myoblasts and Muller glia-derived progenitors (DeWitt et al., 2014; Ernsberger et al., 1989; Johnson et al., 2014; Todd et al., 2016; Wang et al., 2017). Here, we show that CNTF has markedly different effects on OPCs isolated from different regions of the nervous system. Consistent with previous reports, CNTF stimulates proliferation of OPCs purified from rat optic nerves (Barres et al., 1996; Power et al., 2002); however, CNTF inhibits proliferation of OPCs isolated from cortex. This is in line with emergent appreciation for region-specific properties of OPCs (Dimou and Simons, 2017). How might the divergent effects of CNTF on cellular proliferation be explained? Studies in cell lines suggest that CNTF effects on proliferation can be biphasic, where stimulation or inhibition of cellular proliferation is dependent on ligand concentration (Sun and Maxwell, 1994). We speculate that GDE3 release of CNTFRα could increase ligand capture and/or increase accessibility to remote gp130/LIFRβ complexes distributed throughout the cell membrane, which together might tune OPC responses towards inhibition of proliferation. Thus, CNTF-mediated effects on OPC proliferation likely reflect a highly tuned response that integrates multiple components that include cellular subtype and ligand availability.

In summary, we show that GDE3 release of CNTFRα through EVs and GPI-anchor cleavage constitutes a unique and versatile signaling axis that governs the pace of OPC proliferation during nervous system development. This study further suggests that bimodal release of GPI-anchored proteins is central to six-transmembrane GDE protein function. When considered with the gamut of GPI-anchored substrates, this work provides an important framework for GPI-anchor signaling events that operate in developmental and disease settings.

HEK293FT cells were maintained in DMEM (Invitrogen) plus 10% fetal bovine serum (Sigma) and 1% penicillin-streptomycin (Life Technologies) in a 37°C incubator with 5% CO₂. For transient transfections, 12-well plates were coated with 25 µg/ml polyethyleneimine (PEI) in 150 mM NaCl for 1 h at 37°C and then washed three times with PBS. HEK293FT cells were plated on PEI coated plates at a density of 150,000 cells per well (37,500 cells per cm²) in 1 ml of DMEM+FBS. One day after plating, cells were transfected with FuGENE HD (Promega) following the manufacturer's instructions with the indicated plasmids. One day after transfection, medium was replaced with fresh DMEM and conditioned for 16 h unless otherwise noted. Where indicated, PI-PLC (Invitrogen) was added to media 1 h prior to analysis. For details of the reagents, see Table S3.

For media fractionation using ultracentrifugation, conditioned medium plus protease inhibitors were spun at 3000 g for 10 min at 4°C to remove cellular debris. Extracellular vesicles were pelleted at 100,000 g for 70 min at 4°C in a fixed angle TLA 100.4 rotor. The pellet was resuspended in SDS loading buffer for SDS PAGE and immunoblotting. Triton X-114 phase partitioning was performed as previously described (Park et al., 2013). Briefly, 2% Triton X-114 (Sigma) was purified by mixing 1:1 with Tris buffer [100 mM Tris-HCl (pH 7.4), 150 mM NaCl]. This solution was phase separated by raising the temperature to 30°C for 5 min. The buffer was aspirated, and this process was repeated twice. For media fractionation, purified 2% Triton X-114 was added 1:1 with the medium and incubated at 4°C for 10 min with occasional vortexing. For tissue fractionation, cortex of P2-P4 mice was sonicated in purified 2% Triton X-114 and spun at 4°C at 21,000 g to remove debris. An aliquot of input was saved, and the rest was used for detergent partitioning. The detergent-rich pellet and detergent-free supernatant were separated by incubating at 30°C for 5 min followed by centrifugation at 3000 g for 3 min. For details of the reagents, see Table S3.

Isolated OPCs or transfected HEK293FT cells were chilled on ice and all solutions were cooled on ice. Cells were washed with PBS three times. Freshly made sulfo-NHS-SS-biotin solution was added for 30 min with gentle rocking every 5 min. Cells were then washed once with PBS and then quenched with 100 mM glycine solution. Cells were washed with PBS and lysed with RIPA (PBS, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS) buffer containing protease inhibitors and rocked for 20 min at 4°C. The lysate was sonicated and spun at 21,000 g for 20 min at 4°C to remove any debris. A fraction was saved as input and the remainder of the supernatant was mixed with avidin-agarose beards and rotated at 4°C overnight. The beads were washed with RIPA buffer, and proteins were eluted by addition of SDS-loading buffer plus 10% β-mercaptoethanol and heated at 60°C for 15 min. For details of the reagents, see Table S3.

HEK293FT cells were plated on 100 mm PEI-coated plates at a density of 1.5×10⁶ cells per plate. Cells were transfected with FuGENE HD following manufacturer's instructions. Following transfection, medium was switched to DMEM and conditioned for 3 h. Medium was transferred to a tube containing protease inhibitors and spun at 3000 g for 10 min at 4°C to remove cellular debris. The medium was then aliquoted into individual tubes and incubated in a 37°C water bath for the specified amounts of time, followed by centrifugation. Where indicated, PI-PLC (Invitrogen) was added for 1 h prior to ultracentrifugation. Pellets were resuspended in SDS-containing loading buffer. For co-immunoprecipitation experiments, HEK293FT cell-conditioned medium was incubated with anti-myc magnetic beads (Thermo) to isolate myc-CNTFRα. Beads were washed with TBS+0.15% Tween-20 prior to elution with SDS-containing loading buffer. For details of the reagents, see Table S3.

HEK293FT were transfected with FungeneHD as described. The following day, the media was replaced with serum-free DMEM and incubated overnight. Cellular debris was removed through two consecutive 3000 g centrifugations. Exosomes were then purified with two consecutive 10,000 g centrifugations followed by two consecutive 100,000 g (1 h at 4°C) centrifugations onto 30% sucrose cushions. Samples were adsorbed to glow discharged (EMS GloQube) carbon-coated 400 mesh copper grids (EMS), by floatation for 2 min. Grids were quickly blotted then rinsed in three drops (1 min each) of TBS. Grids were negatively stained in two consecutive drops of 1% uranyl acetate with tylose (1% UAT, double filtered, 0.22 um filter), blotted then quickly aspirated to get a thin layer of stain covering the sample. Grids were imaged on a Phillips CM-120 TEM operating at 80 kV with an AMT XR80 CCD (8 megapixel). For details of the reagents, see Table S3.

OPCs were isolated from P2-4 cortices or spinal cords, as indicated using Miltenyi Biotec magnetic beads for A2B5⁺ cells according to manufacturer's recommendations. OPCs were plated on poly-L-lysine (0.1 mg/ml) and laminin (10 µg/ml)-coated plates at a density of 150,000 cells per cm² in proliferative media containing 1 mM sodium pyruvate, 5 µM forskolin, 10 ng/ml PDGF, 10 ng/ml CNTF, B27, N2 and antibiotic unless otherwise specified. OPC medium was made with freshly reconstituted growth factors, 0.22 µm filtered and stored at 4°C for no longer than 3 days. For P14 cortical OPC purification, triturated cells were resuspended in 3 ml of 20% Optiprep in HBSS, layered with 3 ml of 9% Optiprep, and centrifuged at 670 g for 20 min to remove myelin debris. Cells at the interface were collected and diluted with PBS containing 0.5% BSA, centrifuged at 300 g and used for magnetic antibody incubation step. For transfections, Lipofectamine 2000 (Invitrogen) was used according to manufacturer's recommendations. Briefly, OPCs were cultured in proliferative medium for at least 16 h prior to transfection. Transfection was performed for 1 h in MEM (Gibco). Cells were washed once with OPC proliferation medium without antibiotic and incubated for at least 16 h prior to the proliferation assay.

For all OPC proliferation assays, IdU was added (20 µg/ml) for 2 h prior to fixation with 4% paraformaldehyde in 0.1 M phosphate buffer (PB) for 15 min at room temperature. To visualize IdU, antigen retrieval was performed by treating the cells with 2N HCl for 7 min at 37°C. Cells were washed with PBS and neutralized in 0.1 M borate buffer. For details of the reagents, see Table S3.

Animals were maintained and used in accordance with approved Johns Hopkins University IACUC protocols. Mice were anesthetized and transcardially perfused with 0.1 M PB followed by 4% paraformaldehyde (PFA) in 0.1 M PB. Tissue samples were post-fixed in 4% PFA for 1 h at 4°C, washed in PBS and incubated in 30% sucrose for 24 h. Samples were embedded in OCT compound (Tissue-Tek 62550-12) and flash frozen in a dry ice ethanol bath. Cryomolds were stored at −80°C and sectioned on an UltraPro 5000 Cryostat (Vibratome). For details of the reagents, see Table S3.

Tissue sections were air dried for at least 20 min then washed twice in PBS. Both tissue sections and cultured cells were blocked in PBS with 0.3% Triton X-100 and 10% heat-inactivated normal goat serum (HINGS) for 30 min. Primary antibody incubation was performed in PBS with 0.3% Triton X-100 and 1% HINGS at 4°C overnight. Primary antibodies used were as follows: rabbit anti-Ki67 (1:1000, Abcam, ab15580), guinea pig anti-Nkx6.1 (1:4000, from T. M. Jessell, Columbia University, NY, USA), guinea pig anti-Olig2 (1:20,000, from B. Novitch, UCLA, USA), rabbit anti-Sox10 (1:1000, Abcam), rabbit anti-NFIA (1:500, ActiveMotif), guinea pig anti-NG2 (1:500, from D. Bergles, Johns Hopkins University, Baltimore, MD, USA), mouse anti-NeuN (1:500, Chemicon, MAB377), rat anti-PDGFRa (1:200, BD Sciences, 558774), rabbit anti-NFIA (1:500, Active Motif, 39398), Isl1/2 (1:1000, K5), mouse anti-Pax7 (1:50, DHSB), rabbit anti-Sox9 (1:3000, M. Wegner, Erlangen, Germany), rabbit anti-Aldh1L1 (1:500, B. Barres, Stanford University, CA, USA), rabbit anti-Olig2 (1:1000, Millipore), mouse anti-GFAP (1:1000, BD Pharmingen). Samples were washed in PBS and incubated with secondary antibodies (Jackson ImmunoResearch) for 1 h at room temperature. Samples were stained using DAPI (Invitrogen R37606). Slides were coverslipped with Vectashield (Vector Laboratories) mounting medium and imaged using a Zeiss LSM 700 or Keyence BZ-X710 microscope. Brightness and contrast are adjusted evenly between experimental groups. For details of the reagents, see Table S3.

Samples were resolved using SDS-PAGE, transferred to a PVDF membrane, blocked with 5% milk and probed with indicated antibodies overnight at 4°C: anti-CD9 (1:500, Biolegend, 312102), anti-CD63 (1:500 Santa Cruz sc-5275), anti-FLAG (1:10,000, Sigma, F7425), anti-Myc (1:10,000, Cell Signaling Technologies, 2276), anti-HA (1:1000, Cell Signaling Technologies, 2367S), anti-MBP (1:2000, Biolegend), anti-MOG (1:2000, Millipore), anti-Olig2 (1:2000, Millipore), anti-Actin (1:20,000, Millipore), anti-GAPDH (1:2000, Cell Signaling) and HRP-conjugated secondary antibodies (Jackson ImmunoResearch). Samples probed for CD9 and CD63 were lysed in non-reducing conditions. Membranes were developed with Western Blot Detection Kit (Kindle Biosciences) and imaged on autoradiography paper or digital imaging (KwikQuant, Kindle Biosciences). Bands were quantified using ImageJ. For details of the reagents, see Table S3.

Spinal cords from P0.5 pups were dissected and the OPCs were isolated and cultured as described above. One day after isolation, RNA was extracted from cells using RNeasy Plus Micro Kit (Qiagen). cDNA libraries were prepared using the Illumina TruSeq Stranded mRNA Library Prep Kit (Illumina, RS-122-2101). Paired-end reads, 50 bp in length, were generated on an Illumina HiSeq 2500 system. To analyze the RNA-seq data, reads were quality checked and trimmed using the programs fastqc and fqtrim. Reads were then mapped to the mouse genome mm10 using the spliced alignment program Tophat2 v2.1.1, and assembled into transcripts using Cufflinks v2.2.1. Transcript assemblies across all samples were merged with Cuffcompare v.2.2.1, using GENCODE v.M5 as reference, to create a set of gene and transcript annotations that was later used in the differential analyses. Finally, Cuffdiff v2.2.1 was run on each pairwise comparison to determine statistically significant differentially expressed genes (significance cutoffs: P-value <=0.05, q-value <=0.05). For details of the reagents, see Table S3.

Mouse and chick spinal cords were harvested at defined developmental time points and fixed in 4% PFA. Gde3 probes targeting mouse Gde3 (1888 bp) and chicken Gde3 (1781 bp) were derived from a full-length coding region sequence. In situ hybridization was performed as previously described (Yan et al. 2015). For details of the reagents, see Table S3.

Using 129S7/AB2.2 BAC library, bMQ-312G21 was found to contain the Gdpd2 locus (also known as Gde3). Using bacterial recombination, two mini-targeting vectors targeting the intron between exons 7 and 8 and the intron between exons 10 and 11 were generated by PCR amplification. The final targeting vector backbone is PL253 with Kanamycin resistance. Purified and linearized product was supplied to the Transgenic Core Facility at Johns Hopkins University which was electroporated into Sv129 ES cells to generate ∼200 neomycin-resistant clones. Initial screening was carried out by PCR and further karyotyping identified two clones for C57BL/6J blastocyst injection. More than 20 chimeric founders were obtained and crossed with C57BL/6J animals to generate F1 mice. Stable heterozygous Gde3^FRTneolox mouse line was generated and confirmed by nested PCR. Gde3^FRTneolox mice are indistinguishable from wild-type littermates. This mouse line was crossed with B6.C-Tg (CMV-cre)1Cgn/J mice obtained from transgenic core at Johns Hopkins Medical Institutions to generate Gde3 KO mice. Further outcrosses to C57BL/6J wild types were performed and all analyses were carried out on animals outcrossed at least five times.

Validation of genomic recombination was performed with a PCR reaction targeting the deleted region of the Gde3 locus. Furthermore, in situ hybridization using a probe that targets the deleted region shows no mRNA hybridization in Gde3 KO tissues. Last, to confirm protein ablation, a custom antibody against mouse GDE3 (amino acids 21 through 35) was raised in rabbits. Western blotting with wild-type brain and spinal cord samples reveals a band at the expected size of ∼100 kDa that is fully absent in Gde3 KO lysates. See Fig. S2. For details of the reagents, see Table S3.

Image quantification was performed with ImageJ. All quantification was performed using raw data while blind to the experimental condition. For in vivo tissue sections, a minimum of 10 sections were quantified per embryo. Embryos were generated from a minimum of two litters for each experiment. Regions of interest were determined with the relevant counterstain, either DAPI or indicated VZ marker. For in vitro cell counts, a minimum of three wells were quantified per animal and at least three animals were used for each experiment. The total number of cells quantified for the in vitro experiments is documented in Table S1. S-phase measurements and calculation of cell cycle length was performed as previously described (Martynoga et al., 2005). Briefly, BrdU was injected (70 mg/kg by weight) intraperitoneally into pregnant dams 2 h prior to sacrifice. IdU was injected using an equimolar dose 30 min before sacrifice. Counts were performed within the ventral region of the spinal cord identified by Nkx6.1 positive cells. Data are mean±s.e.m. The reported n number refers to individual animals, processed uniformly across experimental conditions. Statistical significance was determined using a two-tailed unpaired Student's t-test, except in Fig. 6H where the analysis was paired. Data are considered significant when P<0.05.

We thank Dwight Bergles and Alex Kolodkin for discussions; Paul Worley for rat tissue; Jeremy Nathans for Hprt:GFP mouse lines; Cindy Wladyka and Yuhan Li for technical assistance; Lilliana Florea for bioinformatics expertise; Barbara Smith from the Microscope Core Facility (JHUSOM) for electron microscopy; and the Multiphoton Imaging Core of the Johns Hopkins P30 Center for Neuroscience Research (NS050274).