Prdx4 is a compartment-specific H2O2 sensor that regulates neurogenesis by controlling surface expression of GDE2

Neural progenitors and terminally differentiated neurons show distinct redox profiles, suggesting that coupled-redox cascades regulate the initiation and progression of neuronal differentiation. Discrete cellular compartments have different redox environments and how they contribute to differentiation is unclear. Here we show that Prdx4, an endoplasmic reticulum (ER) enzyme that metabolizes H2O2, acts as a tunable regulator of neurogenesis via its compartmentalized thiol-oxidative function. Prdx4 ablation causes premature motor neuron differentiation and progenitor depletion, leading to imbalances in subtype-specific motor neurons. GDE2, a six-transmembrane protein that induces differentiation by downregulating Notch signalling through surface cleavage of GPI-anchored proteins, is targeted by Prdx4 oxidative activity. Prdx4 dimers generated by H2O2 metabolism oxidize two cysteine residues within the GDE2 enzymatic domain, which blocks GDE2 trafficking to the plasma membrane and prevents GDE2 neurogeneic function. Thus, Prdx4 oxidative activity acts as a sensor to directly couple neuronal differentiation with redox environments in the ER.

C ellular H 2 O 2 levels in the nervous system show marked correlations with cellular proliferative states, with higher oxidative conditions in cycling progenitors compared with differentiated neurons 1,2 . Subtle redox changes can alter the thiol-redox states of proteins to regulate their function, and redox-dependent effects are known to impact fundamental cellular functions such as signal transduction, DNA and RNA synthesis, protein synthesis and cell cycle regulation [3][4][5][6] . These observations have led to the proposal that redox-coupled molecular cascades act as a series of 'nano-switches' to regulate the timing and progression of neuronal differentiation 3 . Distinct cellular compartments such as the cytosol, nucleus, mitochondria and the endoplasmic reticulum (ER) have different redox environments 6 ; thus, such a model would require that compartment-specific redox events individually or collectively synergize to control the transition from progenitor proliferation to neuronal differentiation. However, the identities of such redox pathways and how they might regulate the progression of neuronal differentiation are unclear.
Here, we identify the antioxidant enzyme Prdx4 as a tunable H 2 O 2 sensor in the lumen of the ER that controls the timing of neurogenesis. Prdx4 is a 2-cysteine peroxiredoxin that is a major component of the ER oxidative protein folding pathway [7][8][9][10] . It metabolizes and removes H 2 O 2 by-products generated by enzymes such as protein disulfide isomerase (PDI) and Ero1 during disulfide bond formation. After H 2 O 2 removal, resultant Prdx4 dimers oxidize PDIs, which participate in the oxidative folding of new client proteins [9][10][11] . In times of oxidative stress, Prdx4 redox activity can target proteins such as the G-CSF (granulocyte-colony stimulating factor) receptor or the thromboxane A 2 receptor for degradation 12,13 . Of note, Prdx4 reactive cysteines are highly susceptible to overoxidation, and accordingly, Prdx4 is inactive when H 2 O 2 levels are high 14,15 .
Here we identify a new function for Prdx4 thiol-oxidative activity in controlling the surface expression of GDE2, a sixtransmembrane glycosylphosphatidylinositol (GPI) anchorcleaving enzyme, with prominent roles in regulating neuronal differentiation [16][17][18][19][20] . Oxidized Prdx4 dimers prevent surface trafficking of GDE2 by oxidizing two cysteine residues within the GDE2 enzymatic domain, thereby abolishing its neurogeneic function. Prdx4 thus functions as a compartmentalized redox sensor that controls the timing of neurogenesis by coupling GDE2 surface expression in response to redox environments in the ER.

Results
Prdx4 expression in the developing spinal cord. In the developing spinal cord, cycling progenitors are distinguished from terminally differentiated neurons by cell-body position and molecular marker expression (Supplementary Fig. 1; refs 16,21,22). Prdx4 transcripts were detected in cycling progenitors localized within the medial ventricular zone (VZ) and in newly differentiating neurons positioned more laterally within the intermediate zone (IZ) of the chick spinal cord (Fig. 1a-c). Similar distribution of Prdx4 protein was found in the mouse: Prdx4 is expressed in cycling motor neuron progenitors within the motor neuron progenitor domain (pMN) and in newly differentiating Isl1/2 þ motor neurons (Fig. 1d,e). At E10.5, Prdx4 continues to be expressed in VZ and IZ cells, but is not expressed by fully differentiated Isl1/2 þ motor neurons in the lateral marginal zone (Fig. 1g,h).
Physiological differences in redox can be detected by ARE (antioxidant response element)-luciferase reporters, which express luciferase in oxidative conditions via the binding and activity of the transcription factor Nrf2 (ref. 23). Electroporation of ARE-luciferase into chick spinal cords showed that Olig2 þ progenitors in the VZ coexpressed luciferase as did a small number of cells in the IZ that express NeuroM, a transcription factor transiently expressed in newly differentiating neurons (Supplementary Fig. 1; ref. 24). Postmitotic neurons located in the marginal zone did not express luciferase. Thus, Prdx4 is expressed in oxidative zones in the spinal cord, and these correspond to VZ progenitors and newly differentiating neurons in the IZ.
Prdx4 loss induces premature differentiation. To investigate the physiological function of Prdx4, we ablated Prdx4 expression by electroporating plasmids expressing short hairpin RNAs (shRNAs) engineered against the Prdx4 3 0 UTR (untranslated region) and coding sequence into chick spinal cords before motor neuron differentiation (Fig. 2a). Olig2 is expressed at high levels by motor neuron progenitors but is decreased in newly differentiating cells in the IZ. Spinal cords lacking Prdx4 showed no changes in total numbers of Olig2 þ cells; however, Olig2 expression was decreased in subsets of progenitors within the pMN, indicative of premature differentiation ( Supplementary  Fig. 2a-d). Consistent with this, Prdx4 shRNA embryos showed a marked increase in cells expressing NeuroM (Fig. 2b-e), and additional NeuroM þ cells were located primarily in the VZ of Prdx4 shRNA embryos (Bin 1; Fig. 2d,e). The increase in newly differentiating cells did not expand total Isl1/2 þ motor neuron numbers but resulted in elevated cell death and a net reduction in motor neurons ( Fig. 2c and Supplementary Fig. 2b,d,e). Mouse embryos lacking Prdx4 showed similar phenotypes (Fig. 2f,i; Prdx4 À /y or Prdx4 À / À ; ref. 25). Newly differentiating motor neurons in the IZ transiently coexpress Olig2 and Isl1/2 (refs 21,22). Prdx4 null embryos showed a roughly 30% increase of Olig2 þ Isl1/2 þ cells (Fig. 2g,j,l) consistent with the premature onset of differentiation. In contrast to the chick, Prdx4 null mutants exhibited a concomitant 20-40% increase in Isl1/2 þ and HB9 þ motor neurons ( Fig. 2f-l), and no changes in cell death were detected between Prdx4 mutants and wild-type (WT) animals ( Supplementary Fig. 2f,g). These observations suggest that motor neuron progenitors lacking Prdx4 differentiate into postmitotic motor neurons ahead of schedule.
Prdx4 loss alters cell cycle exit and motor neuron identity. To determine how the loss of Prdx4 affects cell cycle progression, we performed pulse labelling studies using the nucleotide analogue 5-bromodeoxyuridine (BrdU) in Prdx4 null mice and WT littermates. No changes in the proportion of cells in S-phase after a 30-min BrdU pulse (BrdU þ /Ki67 þ ; Ki67 marks all cycling cells; Fig. 3c,d,g) or in M-phase (Mpm2 þ /Ki67 þ ; Fig. 3b,d,f) were observed between Prdx4 nulls and controls, implying similar cell cycle lengths in both groups (Fig. 3d). However, the total number of BrdU þ Ki67 þ cells in Prdx4 mutants was reduced by 27% compared with WT siblings when a 16-h BrdU pulse was administered (Fig. 3a,d,e). This observation suggests that the overall number of cycling cells is decreased when Prdx4 is ablated. In addition, cell cycle exit indices (BrdU þ Ki67 À : BrdU þ Ki67 þ ) were increased by B60% in Prdx4 null animals compared with WT littermates (Fig. 3a,d,e). These collective observations indicate that Prdx4 ablation forces premature cell cycle exit of progenitors resulting in increased postmitotic motor neuron numbers.
Time of cell cycle exit is linked to the acquisition of specific motor neuron fates, thus loss of Prdx4 could cause imbalances in motor columns and pools. We compared the numbers of medial median motor column (MMCm) and medial and lateral divisions of lateral motor column (LMC) neurons at Prdx4 null and WT hindlimb level spinal cords using distinguishing molecular markers ( Supplementary Fig. 3a-f,k; refs 21,22). No changes in MMCm motor column motor neuron numbers were detected between Prdx4 nulls and WT littermates ( Fig. 3h and Supplementary Fig. 3a,d). However, early-born medial LMC motor neurons were increased by 20% with a concomitant decrease in late-born lateral LMC neurons ( Fig. 3h and Supplementary Fig. 3b-f). Motor neurons within the medial and lateral LMC that project to the same muscles are clustered into motor pools and are molecularly distinct (Supplementary Fig. 3g-k). Consistent with increases in early-born motor neuron subtype populations, Prdx4 mutants showed an increase in early-born adductor longus, adductor magnus and gracilis posterior motor pools and a corresponding decrease in late-born adductor brevis and vasti motor pools ( Fig. 3i  Prdx4 and GDE2 interact to regulate neurogenesis. What is the mechanism that underlies Prdx4 temporal control of neurogenesis? We noted that Prdx4 null phenotypes are directly opposite to phenotypes exhibited by embryos lacking GDE2, a six-transmembrane protein that specifically regulates the timing of LMC motor neuron differentiation in the spinal cord [16][17][18][19][20] . GDE2 and Prdx4 are expressed in the IZ where newly differentiating motor neurons are found ( Fig. 1; refs 16,18). We hypothesized that Prdx4 null animals have increased GDE2 activity, and that, normally, Prdx4 inhibits GDE2-dependent stimulation of neurogenesis. We therefore compared the extent of premature differentiation in Prdx4 null mutants (Prdx4 À /y or Prdx4 À / À ) that lack one allele of Gde2. We found that the increase in newly differentiating Olig2 þ Isl þ cells and Isl1/2 motor neurons normally observed in Prdx4 null animals was substantially reduced in Prdx4 À /y ;Gde2 þ / À or Prdx4 À / À ; Gde2 þ / À embryos ( Fig. 3j-l). The suppression of premature differentiation in Prdx4 nulls by genetic removal of one allele of Gde2 suggests that Prdx4 negatively regulates GDE2 activity to control the timing of neurogenesis.
Prdx4 thiol-redox activity inhibits GDE2 function. Co-immunoprecipitation (co-IP) assays using extracts prepared from transfected HEK293T cells showed that Prdx4 and GDE2 are capable of forming protein complexes (Fig. 4a). Thus, Prdx4 regulation of GDE2 activity is likely to be direct. To test whether Prdx4 inhibits GDE2 activity, we utilized established in vivo assays of GDE2 function in electroporated chick embryos 16,17 . Unilateral GDE2 overexpression in embryonic chick spinal cords triggers premature differentiation manifest by the formation of many ectopic Isl2 þ motor neurons in the electroporated VZ but not in the contralateral unelectroporated VZ (Fig. 4b,e). Strikingly, coexpression of Prdx4 with GDE2 suppressed the ability of GDE2 to induce Isl2 þ motor neuron differentiation (Fig. 4c,e). Coexpression of GDE2 and a redox inactive version of Prdx4 that substitutes its reactive cysteine residues with serine (Prdx4C118.239S) resulted in many ectopic Isl2 þ motor neurons, equivalent to when GDE2 is overexpressed alone (Fig. 4d,e; refs 9,27,28). Co-IP assays confirmed that Prdx4C118.239S retained the ability to interact with GDE2 ( Supplementary Fig. 4b). These observations suggest that Prdx4 interacts with GDE2 and utilizes its thiol-redox activity to inhibit GDE2-dependent induction of motor neuron differentiation.
To define the sites of Prdx4 redox activity on GDE2, we mutated individual and paired cysteine (C) residues within GDE2 to serine (S). With the exception of GDE2 carrying a double mutation at C340 and C467 (GDE2C340.467S), all GDE2 variants showed appropriate glycosylation and surface expression, and maintained their abilities to interact with Prdx4 in co-IP assays ( Supplementary Fig. 4a-c). GDE2C25.576S, GDE2C25S, GDE2C15.18S, GDE2C340S and GDE2C467S readily induced motor neuron differentiation in the VZ of electroporated embryonic chick spinal cords ( Fig. 4f-i; data not shown). Prdx4 coexpression with the GDE2 variants effectively blocked GDE2C25.576S, GDE2C25S and GDE2C15.18S activity ( Fig. 4f-h). However, Prdx4 failed to inhibit the inductive activities of GDE2C340S or GDE2C467S, suggesting that the thiol residues located within the extracellular GDPD domain are the sites of Prdx4 inhibitory redox activity ( Fig. 4i; data not shown). Consistent with this observation, co-IP assays demonstrate that Prdx4 can interact with GDE2 via the GDPD region where C340 and C476 are located (Fig. 4a).
Prdx4 inhibits GDE2 through thiol oxidation. To examine how Prdx4 alters thiol-redox states of GDE2, we utilized biotinlabelling assays to quantify the levels of oxidized thiol groups in GDE2 in the absence or presence of Prdx4 in HEK293T cell extracts 17 . Free thiol groups were first blocked by addition of N-ethylmaleimide (NEM) to cell extracts, and existing disulfide bonds were reduced by tris (2-carboxyethyl) phosphine (TCEP). Resultant free thiols were labelled with Biotin-HPDP (N-[6-(Biotinamido)hexyl]-3-(2-pyridyldithio)propionamide) and quantified by densitometric analysis of IP-ed (immunoprecipitated) GDE2 on western blots. This method effectively blocked free thiol groups in GDE2 as IP-ed GDE2 was not labelled by biotin in the absence of the reducing agent TCEP (Fig. 5a). Addition of TCEP led to GDE2 biotinylation consistent with earlier reports that GDE2 contains endogenous disulfide bonds ( Fig. 5a; ref. 17). Coexpression of Prdx4 with GDE2 increased levels of biotinylated GDE2, suggesting that Prdx4 oxidizes GDE2 (Fig. 5a). The extent of biotin increase is variable and likely reflects the redox sensitivity of the extracellular GDE2 Cys residues ( Fig. 5a-c). Nonetheless, when redox inactive Prdx4C118.239S was coexpressed with GDE2, no quantitative changes in GDE2 biotinylation were observed (Fig. 5b,c). We repeated the biotinlabelling experiments, this time coexpressing Prdx4 with either GDE2C340S or GDE2C467S. The amounts of biotinylated GDE2C340S and GDE2C467S remained similar whether Prdx4 was present or absent, which is in agreement with the extracellular cysteine residues being the sites of Prdx4 oxidative function (Fig. 5d,e). Tandem liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis of GDE2 in HEK293T cells confirmed that the extracellular C340 and C467 residues are capable of being oxidized 17 . Here, free thiol groups were blocked by incubation with NEM, existing oxidized thiols were reduced with dithiothreitol and newly reduced thiols were modified with iodoacetamide (IAM). LC-MS/MS analysis of IP-ed GDE2 showed that peptides containing C340 and C467 were labelled by NEM and IAM, respectively, while LC-MS analysis identified peptides containing C340 labelled with IAM ( Fig. 5f). These collective observations suggest that Prdx4 inhibits GDE2 activity by oxidation of the C340 and C467 residues located in the GDE2 extracellular enzymatic domain.
Prdx4 inhibits GDE2 surface trafficking. We confirmed that Prdx4 is localized to the ER in motor neurons, as Prdx4 colocalized with the ER marker KDEL in sectioned mouse spinal cords and in primary motor neuron cultures (Supplementary Fig. 5; refs 29,30). Moreover, Triton X-114 extraction of transfected HEK293T cells consistently detected mPrdx4 in the detergentfree (DT-free) hydrophilic phase where cytoplasmic and luminal proteins accumulate and not in the detergent-rich (DT-rich) membrane fraction ( Supplementary Fig. 5). Membrane-bound proteins are normally trafficked to the cell surface from the ER, raising the possibility that Prdx4 oxidizes GDE2 and prevents its transport to the plasma membrane. If so, then GDE2 surface expression should be redox sensitive. Treatment of primary motor neuron cultures with H 2 O 2 significantly decreased levels of surface biotinylated GDE2, whereas incubation of cultures with established reactive oxygen species (ROS) scavengers, Tempol and Edaravone, elevated GDE2 surface expression (Fig. 6a,

31)
. Consistent with changes in cellular oxidative states upon addition of these compounds, Prdx4 dimer to monomer ratios were increased upon H 2 O 2 addition and decreased when incubated with Tempol and Edaravone ( Supplementary Fig. 6).
The altered ratios of Prdx4 dimers:monomers further indicate that the levels of oxidized Prdx4 dimers capable of oxidizing GDE2 were increased by H 2 O 2 treatment in motor neuron cultures and decreased by ROS scavengers (Supplementary Fig. 6).   surface biotinylated GDE2 protein in primary motor neuron cultures prepared from WT and Prdx4 null spinal cords. Surface expression of GDE2 was increased by B20% in the absence of mPrdx4 (Fig. 6c). This degree of change in GDE2 surface expression is consistent with the restriction of Prdx4 and GDE2 coexpression to newly born motor neuron populations and the lack of Prdx4 expression in terminally differentiated motor neurons (Fig. 1).
To determine whether Prdx4 is sufficient to prevent GDE2 surface expression, we performed surface biotinylation assays in HEK293T cells cotransfected with GDE2 and tagged versions of Prdx4. Levels of surface biotinylated GDE2 were markedly reduced by Prdx4, consistent with our studies in motor neurons, but not by redox inactive Prdx4C118.239S (Fig. 7a). Similar results were obtained with mPrdx4 consistent with conserved functions of Prdx4 in mouse and chick. In accordance with our earlier analyses, Prdx4 failed to reduce the levels of surface biotinylated GDE2C340S and GDE2C467S, but effectively inhibited surface accumulation of GDE2C25S and GDE2C15.18S (Fig. 7b,c).
Prdx4 is localized to the ER in motor neurons but is also reported to be secreted 32 . We fused Prdx4 to a secretory signal that targets proteins to the ER lumen (SecMycmPrdx4) and confirmed its partitioning to DT-free fractions after Triton X-114 extraction of transfected HEK293T cells, consistent with localization of SecMycmPrdx4 to the ER lumen. SecMycmPrdx4 effectively reduced levels of biotinylated GDE2; reinforcing the model that Prdx4 inhibition of GDE2 occurs within the ER (Fig. 7a and Supplementary Fig. 5). If Prdx4 function occurs within the ER, we would expect Prdx4 suppression of GDE2 surface expression to be cell autonomous. Plasmids expressing human influenza hemagglutinin (HA) Prdx4 and GDE2.FLAG were transfected into HEK293T cells, and the distribution of GDE2 was compared with Prdx4. Live-cell staining to detect surface GDE2 using a polyclonal antibody (GDE2Loop) against the GDE2 extracellular enzymatic domain was performed before permeabilization and visualization of intracellular FLAG and HA epitopes on GDE2 and Prdx4, respectively (Fig. 7i). Many transfected cells coexpressed GDE2 (purple) and Prdx4 (green) as visualized by FLAG and HA immunoreactivity (Fig. 7d-h); however, the majority of HA þ FLAG þ cells did not express GDE2 on the cell surface (red, Fig. 7f). In rare cases, where GDE2 surface expression was detected in Prdx4 þ cells, minor GDE2Loop immunoreactivity was detected, indicative of very low levels of GDE2 at the plasma membrane (asterisks, Fig. 7e,f,h)  ARTICLE surface (GDE2Loop þ ) did not coexpress Prdx4 (arrows) even when adjacent to cells expressing high levels of Prdx4. These observations indicate that Prdx4 inhibition of GDE2 is cell autonomous, and reinforces the model that Prdx4 function in the ER inhibits surface expression of GDE2.

Discussion
Prdx4 is the only peroxiredoxin located within the ER, where it metabolizes H 2 O 2 and oxidizes PDIs that play central roles in oxidative protein folding [9][10][11] . Our study identifies a new role for Prdx4 oxidative properties in regulating the timing of neurogenesis through redox-dependent control of GDE2 surface expression (Fig. 7j). Prdx4 regulation of GDE2 trafficking is not due to altered folding of GDE2, as GDE2, GDE2C340S and GDE2C467S proteins are stable and normally glycosylated in the presence of Prdx4, and GDE2C340S and GDE2C467S exhibit appropriate surface localization and are physiologically active in terms of inducing motor neuron differentiation. Instead, H 2 O 2 levels in the ER modulate the availability of Prdx4 dimers that can oxidize GDE2 and regulate its surface expression, enabling Prdx4 to serve as a finely tuned regulator of GDE2 function in response to subtle physiological changes in the ER redox environment. The susceptibility of GDE2 Cys residues to oxidation raises the question whether ROS can directly modify these Cys residues independently of Prdx4 to inhibit GDE2 surface localization. Arguing against this possibility is our observation that application of H 2 O 2 to primary motor neuron cultures derived from Prdx4 null mice does not decrease GDE2 surface expression ( Supplementary Fig. 7). Instead, GDE2 surface expression is further increased under these conditions, suggesting that Prdx4 negatively regulates additional redox-dependent mechanisms in the ER that promote GDE2 surface localization. Taken together, these observations indicate that GDE2 trafficking is regulated through complex redox-dependent pathways in the ER, and suggest that Prdx4 occupies central roles in the redox-dependent control of GDE2 surface expression. Prdx4 and GDE2 are coexpressed in newly differentiating neurons in the IZ. Our analysis of ARE-luciferase reporter expression shows that IZ cells exhibit mosaic luciferase expression, suggesting that newly differentiating neurons in the IZ display a range of redox levels as they transition towards a terminally differentiated state. These differences in IZ redox levels would vary surface levels of GDE2 within this region through Prdx4-dependent regulation of GDE2 trafficking. GDE2 surface expression is critical for its ability to induce differentiation; it utilizes its extracellular enzymatic function to cleave and inactivate the GPI-anchored protein RECK on the surface of IZ     ARTICLE cells 20 . RECK inactivation releases inhibition of ADAM10 that stimulates shedding of the Notch ligand from the cell membrane. This in turn downregulates Notch signalling in neighbouring progenitors and induces differentiation 20 . Thus, attenuation of GDE2 surface expression by Prdx4 will significantly impact the ability of GDE2 to initiate progenitor differentiation. Importantly, GDE2 activity at the cell surface is also subject to redox control through Prdx4-independent mechanisms 17 . GDE2 cell surface activity is gated by an intracellular disulfide bond formed between Cys residues in the N-and C-terminal domains; this inhibition is released by the thiol-redox activity of Prdx1, a cytoplasmic 2-Cys peroxiredoxin related to Prdx4 that is expressed in IZ cells.
Taking this into consideration, we propose a model where GDE2 activity is regulated in two different cellular compartments by opposing redox-dependent mechanisms; within the ER, Prdx4 dimers generated after metabolizing H 2 O 2 , oxidizes GDE2 within its enzymatic domain to inhibit its trafficking to the cell surface, whereas Prdx1 reduces an intracellular disulfide bond to activate GDE2 at the plasma membrane. Thus, redox-dependent regulation of GDE2 activity in discrete cellular compartments by Prdx proteins requires integration to ensure the appropriate temporal progression of neurogenesis. GDE3 and GDE6 are in the same family as GDE2 and share conserved cysteine residues 16,20,33 . Thus, compartmentalized thiol-oxidative regulation of six-transmembrane GDE surface expression by Prdx4 could constitute a key mechanism to attenuate GDE enzymatic control of GPI-anchored protein function.

Methods
Expression constructs. Chick Prdx4 ORF was PCR cloned using cDNA plasmid from University of Delaware (clone ID: pgm2n.pk011.C15) as template. The same sequence was also obtained by reverse transcription (RT)-PCR from HH St21-23 chicken spinal cord. The nucleotide sequence was identical to XM_416800.2 (updated to XM_416800.4) except two nucleotide difference (G139 to A and A487 to G), which did not change the protein sequence (same as NP_416800.2). Mouse Prdx4 ORF was PCR cloned using cDNA plasmid from Open Biosystem (Image ID: 3481827). All chick Prdx4 expression constructs were cloned in the chick bactin promoter vector (pCAGGS) and CMV IE94 promoter vector (pCS2) with three HA tag on the N terminus unless otherwise specified. Mouse Prdx4 expression constructs were cloned in the pCAGGS vector with no tag or SecMyc tag (secretory signal with 6 myc) on the N terminus 20 . Chick GDE2 expression constructs were cloned in pCAGGS vector with no tag, one FLAG tag on the N terminus or three FLAG tag on C terminus 17,20 . GDE2 loop 5 (GDE2L5) was cloned in frame into the N terminus of the myc-tagged platelet-derived growth factor (PDGF) receptor transmembrane domain of pDisplay vector. All mutant constructs were generated by site-directed (Quickchange kit, Stratagene) or PCRbased mutagenesis.
In situ hybridization and immunofluorescence. For immunostaining analyses, 12.5 mm sections were generated from fixed embryos and slides were air-dried for 20 min. Slides were blocked in PBT (PBS, 0.1% Triton X-100, 1% BSA) for 1 h at room temperature (RT) in a humidified chamber and then incubated in PBT with the relevant concentration of antibody overnight at 4°C. Slides were washed in PBS (3 Â 10 min) and incubated in PBT with secondary antibody for 40 min at RT and washed again (3 Â 10 min PBS). Slides were mounted with coverslips in Vectashield (Vector Laboratories). For BrdU staining, slides were incubated in 600 ml of 10 mM sodium citrate, pH 6.0, and microwaved for 20 min, replacing evaporated water after 10 min. Slides were cooled for 20 min, washed 3 Â 5 min in water followed by 1 Â 5 min in PBS. Slides were processed for immunostaining as described above.
Antibodies used were as follows: mouse anti-HA antibody (12CA5, 1 Fig. 2a for example of section designated as 80-85% Prdx4 knockdown). Sections spanning the rostral-caudal axis were assayed for Prdx4 expression in each embryo to evaluate Prdx4 knockdown efficiency. For Gde2 þ / À and Prdx4 null mutant analysis, 10-20 sections from each of 6-8 E9.5 mouse embryos were counted. Cell cycle analyses were performed as follows 17 . BrdU (100 mg kg À 1 body weight) was injected intraperitoneally into pregnant females 30 min and 16 h before embryo harvest to calculate S-phase and cell cycle exit indices, respectively. Sections were stained for BrdU, Ki67 and Mpm2, and the following was calculated: S-phase: BrdU þ /Ki67 þ cells, cell cycle exit: BrdU þ Ki67 À /BrdU þ Ki67 þ and M-phase: Mpm2 þ /Ki67 þ . Columnar counts of motor neurons span the rostro-caudal extent of the hindlimb identified by Isl2 þ Lhx1 þ cells 18 . For gain-of-function experiments, chick embryos were labelled with BrdU for 30 min, 30 h after electroporation to delineate the VZ/IZ boundary 17 . Images of 10-20 sections per embryo from 4 to 10 embryos were scored. Sections were chosen to ensure equivalent electroporation efficiency by LacZ (GDE2) and HA (Prdx4) staining.
All work involving animals was performed in accordance with National Institutes of Health (NIH) regulations, and approved by Institutional Animal Care and Use Committees (IACUC) at the Johns Hopkins School of Medicine.
Primary spinal motor neuron culture. Primary motor neuron cultures from spinal cords were prepared as follows 34 . Spinal cords from the brachial region were dissected from E11.5 mouse embryos in cold Hank's balanced salt solution (Invitrogen). Spinal cords were cut into small pieces and treated with dissociation solution containing l Â Trypsin-EDTA (Gibco) and 0.2 mg ml À 1 DNase I (Roche) for 15 min at 37°C. Heat-inactivated horse serum (HIHS) and pre-culture medium (DMEM containing 10% fetal bovine serum (FBS)) were then added to stop the reaction. Cells were completely dissociated with gentle titration by 1 ml pipettes, centrifuged and resuspended with fresh pre-culture medium. Cells were then cultured in uncoated plastic dishes for 1 h at 37°C to allow non-neuronal cells to adhere to the plastic dish. The neuron-enriched supernatant was collected and centrifuged, and the pellet was resuspended in MMTN medium (DMEM, 4% HIHS, 2% B27 (Gibco) and 1% N2 (Gibco) with fresh added growth factors (10 ng ml À 1 glial cell line-derived neurotrophic factor (GDNF), brain-derived neurotrophic factor (BDNF), ciliary neurotrophic factor (CNTF) and neurotrophin-3 (NT3). Cells were then plated to poly-D-lysine (PDL)/laminin-coated glass chamber slides (Nalge Nunc International). After 2 days, cultured neurons were fixed in 4% paraformaldehyde (PFA) for 10 min at RT, washed with PBS and used for immunohistochemical analysis.
Uncropped versions of key western blots are included in Supplementary  Figures 8-10.
Surface biotinylation and biotin-labelling assays. Surface biotinylation assay was performed as follows 20 . Briefly, 40 h after transfection, cells were cooled on ice, washed twice with ice-cold PBS þ þ (PBS with 1 mM CaCl 2 and 0.5 mM MgCl 2 , pH 7.4) and incubated with PBS þ þ containing 1 mg ml À 1 Sulfo-NHS-SS-Biotin (Pierce) for 30 min at 4°C. Unreacted biotin was quenched by PBS þ þ containing 100 mM glycine. Cells were lysed in RIPA buffer, sonicated and centrifuged for 20 min at 4°C. The supernatant was incubated with neutravidin-agarose beads (Pierce) for 2 h at 4°C. Biotinylated protein bound by neutravidin were washed with RIPA buffer and eluted with SDS sample buffer.
GDE2 protein was similarly labelled for LC-MS/MS; however, 10 mM dithiothreitol was used instead of TCEP and 40 mM IAM was used instead of biotin-HPDP. GDE2 was immunoprecipitated by anti-FLAG M2 Affinity gel, resolved by SDS-PAGE and stained with Coomassie blue. Cysteines were identified by LC-MS/MS (Taplin Biological Mass Spectrometry Facility).
Triton X-114 partitioning. Triton X-114 (Sigma) partitioning was performed as follows 20 . Triton X-114 (2%) was pre-conditioned in 100 mM Tris-HCl (pH 7.4) and 150 mM NaCl buffer. Cells were washed twice with PBS, incubated with 1% Triton X-114 buffer (10 min at 4°C), and collected and iced for 10 min with occasional mixing. After centrifugation at 16,000g for 15 min at 4°C, supernatants were diluted threefold with Triton X-114 buffer (final 1%). Two phases were separated by incubation at 30°C for 5 min, followed by centrifugation at 3,000g for 3 min at RT. DT-rich pellet contained membrane-bound proteins, and DT-free supernatant was repeatedly extracted and analysed.
Assays for motor neuron culture. Primary motor neuron culture was performed as described in the preceding section, except that the dissected spinal cords were treated with dissociation solution containing l Â Accumax (Sigma) and 0.2 mg ml À 1 DNase I (Roche) for 15 min at 37°C. HIHS and DMEM medium (containing 10% FBS) were then added to stop the reaction. Cells were completely dissociated with gentle titration, centrifuged and resuspended with fresh MMTN medium. Cells were then plated onto poly-L-lysine (PLL) (Sigma, 0.2 mg ml À 1 ) plus laminin (Sigma, 0.01 mg ml À 1 )-coated six-well plates. The density for plating is B1 embryo per well. Culture medium was changed 2 h after plating and half changed every 2 days.
At (days in vitro) DIV5, 200 mM H 2 O 2 was added to the medium for 30 min, repeated four times in 2 h before harvesting cells. For ROS scavenger treatment, 50 mM Edaravone or 200 mM Tempol were added to growth medium for 24 h before harvesting cells for surface biotylation 31 . Surface biotinylation was performed as described in the preceding section 20 , except after labelling with Sulfo-NHS-SS-Biotin, cells were lysed in 1% Triton X-114 buffer with 100 mM maleimide and subjected to partitioning. DT-rich pellet were resuspended in RIPA buffer without 1% Triton X-100 and then incubated with neutravidin-agarose beads (Pierce) for 2 h at 4°C. Biotinylated protein bound by neutravidin was washed with RIPA buffer and eluted with SDS sample buffer. Western blotting was performed using rabbit anti-mGDE2 antibody (1:4,000), rabbit anti-mPrdx4 antibody (1:1,000) and mouse anti-actin antibody (C4, 1:5,000, Millipore, cat. no. MAB1501).
Live-cell staining. HEK293 cells were transfected with mGDE2FLAG and HAPrdx4 constructs. After 24 h, the culture medium was changed to serum-free DMEM. Forty to forty-eight hours after transfection, cells were pre-incubated with 5 mg ml À 1 of chlorpromazine in Hank's balanced salt solution for 30 min at 4°C, followed by deglycosylation with PNGase F at 37°C for 1 h. After washing with PBS, cells were incubated with rabbit anti-mGDE2 loop antibody (1:2,000) in DMEM plus 10% FBS at 37°C for 1 h, followed by secondary antibody incubation at RT for 45 min. The surface staining was fixed with 4% paraformaldehyde for 10 min before permeabilization and staining with FLAG and HA antibodies.
Statistical analyses. All statistical tests utilized two-tailed, unpaired Student's t-test. All data are shown as mean ± s.e.m. A value of Po0.05 was considered significant. The numbers analysed in each experiment (n) are either derived from individual experiments or individual embryos as stated in the figure legends. Analysis of embryos utilized B10-20 sections for each embryo.