Signalling through AMPA receptors on oligodendrocyte precursors promotes myelination by enhancing oligodendrocyte survival

Myelin, made by oligodendrocytes, is essential for rapid information transfer in the central nervous system. Oligodendrocyte precursors (OPs) receive glutamatergic synaptic input from axons but how this affects their development is unclear. Murine OPs in white matter express AMPA receptor (AMPAR) subunits GluA2, GluA3 and GluA4. We generated mice in which OPs lack both GluA2 and GluA3, or all three subunits GluA2/3/4, which respectively reduced or abolished AMPAR-mediated input to OPs. In both double- and triple-knockouts OP proliferation and number were unchanged but ~25% fewer oligodendrocytes survived in the subcortical white matter during development. In triple knockouts, this shortfall persisted into adulthood. The oligodendrocyte deficit resulted in ~20% fewer myelin sheaths but the average length, number and thickness of myelin internodes made by individual oligodendrocytes appeared normal. Thus, AMPAR-mediated signalling from active axons stimulates myelin production in developing white matter by enhancing oligodendrocyte survival, without influencing myelin synthesis per se. DOI: http://dx.doi.org/10.7554/eLife.28080.001


Introduction
Myelination of axons by oligodendrocytes (OLs), which develop postnatally from proliferating oligodendrocyte precursors (OPs), enables fast and energy-efficient propagation of electrical signals in the central nervous system (CNS). By decreasing the capacitance of axons, myelin dramatically increases the speed of action potentials and reduces the energy demands of the axons. Moreover, the ensheathing OLs can transfer energy substrates into axons (Fünfschilling et al., 2012;Lee et al., 2012b;Saab et al., 2016) possibly supporting them during periods of high metabolic demand. Synthesis of myelin requires a large investment of energy (Harris and Attwell, 2012), so it seems likely that myelination evolved to focus on active axons, rather than axons that are electrically silent because, for example, they are still elongating, or they are excess to requirements and destined for developmental pruning. This would require that developing OLs can sense the electrical activity of axons.
How might electrical activity be signalled to OL lineage cells? A likely mechanism is that neurotransmitters released from active axons modulate OL formation or myelination. OL lineage cells can sense the release of neurotransmitter directly, because they express ligand-gated channels (reviewed by Larson et al., 2016), including AMPA/kainate and NMDA receptors for the excitatory transmitter glutamate. OPs receive direct AMPA receptor (AMPAR)-mediated synaptic input from unmyelinated or partially myelinated axons in all brains regions examined so far (Bergles et al., 2000;Chittajallu et al., 2004;Lin and Bergles, 2004;Lin et al., 2005;Ge et al., 2006;Kukley et al., 2007;Ziskin et al., 2007;Káradó ttir et al., 2008;Müller et al., 2009). Synaptic input is maintained during OP division (Kukley et al., 2008;Ge et al., 2009) but is down-regulated as OPs differentiate into OLs (Cahoy et al., 2008;De Biase et al., 2010;Kukley et al., 2010), suggesting that it might be important in the control of OL development.
In support of the idea that neurotransmitters regulate myelination, blocking vesicular release from neurons in vitro (Wake et al., 2011(Wake et al., , 2015 or in genetically manipulated zebrafish (Hines et al., 2015;Mensch et al., 2015;Koudelka et al., 2016) reduced myelination without much affecting OL numbers. Conversely, blocking glutamate uptake into secretory vesicles in retinal ganglion cells in early postnatal mice increased the number of OLs that developed in the optic tract, suggesting that glutamate signalling negatively regulates OL development (Etxeberria et al., 2016). Glutamateevoked NMDA receptor (NMDAR)-mediated signalling in OL lineage cells has been suggested to accelerate the myelination of active axons (Lundgaard et al., 2013), which might occur by NMDAR activation up-regulating glucose transport into developing OLs (Saab et al., 2016). The role of synaptic AMPAR signalling is more enigmatic; pharmacological block of AMPAR-mediated signalling in OPs in vitro promotes proliferation and differentiation of OPs (Gallo et al., 1996;Yuan et al., 1998;Fannon et al., 2015) but decreases myelination (Fannon et al., 2015). These data are hard to interpret, because pharmacological manipulations of AMPARs in vitro affect not only OPs but also neurons, which might release substances other than glutamate to regulate myelination. Hence, we still do not have direct in vivo evidence for a role of AMPAR signalling in regulating OL lineage development.
To identify effects on myelination of AMPAR signalling in OL lineage cells, it is necessary to manipulate AMPARs in those cells without affecting neuronal AMPARs. We have therefore used Sox10-Cre to inactivate the Gria2 gene (encoding AMPAR subunit GluA2), or both Gria2 and Gria4, specifically in OL lineage cells on a Gria3 germline knockout background. Examination of these Gria2/3 and Gria2/3/4 compound mutant mice revealed that AMPAR signalling in OPs promotes the development of OLs and myelin sheaths by stimulating survival of newly-differentiating OLs.

Results
OPs in subcortical white matter express AMPAR subunits GluA2-4 but not GluA1 To determine which AMPAR subunits are expressed by OPs in vivo, we examined sections of postnatal mouse forebrain by in situ hybridization (ISH) for Gria1-4, which encode receptor subunits GluA1-4 (also known as GluR1-4 or GluRA-D). We focused on subcortical white matter (SCWM), which contains axons, glial cells and blood vessels but very few neuronal cell bodies. We found that Gria2, Gria3 and Gria4 are expressed by many small cells in the postnatal day 1 (P1) and P14 SCWM ( Figure 1). In contrast, Gria1 was expressed by very few cells in the SCWM, which NeuN labelling showed were neurons ( Figure 1A-C and not shown). As expected, Gria1-4 were all expressed highly by many neurons throughout the forebrain ( Figure 1A). Fluorescence ISH for individual Gria family members followed by immunolabelling for the transcription factor Olig2 confirmed that Gria2-4 were expressed by OL lineage cells in white matter; Olig2 was expressed in 88 ± 3% of Gria2 + cells (817 cells counted in four mice), 94 ± 4% of Gria3 + cells (506 cells from three mice) and 90 ± 0.3% of Gria4 + cells (560 cells from three mice) (Figure 2A-C). Double ISH for Pdgfra and Gria2 followed by immunolabelling for Olig2 indicated that GluA2 is expressed by the majority of OPs ( Figure 2D), consistent with previous electrophysiological findings (Kukley et al., 2007;Ziskin et al., 2007). Occasional large cells with strong Gria expression but no Olig2 co-labelling ( Figure 2B) were shown to be white matter neurons (NeuN + , not shown). Together, these data indicate that GluA2-4 are the main AMPAR subunits expressed by OL lineage cells in the developing mouse SCWM.
Gria3 germline knockout (KO) mice have no overt neural phenotype (Meng et al., 2003). Gria2 germline KOs have higher than normal mortality; the survivors are anatomically normal but display a range of behavioural abnormalities, including reduced exploratory and reproductive activity and impaired motor coordination (Jia et al., 1996;Gerlai et al., 1998;Jia et al., 2001;Shimshek et al., 2006a). Gria2/3 double KOs were reported also to develop tremor, starting in the second postnatal week (Meng et al., 2003). Since tremor is a hallmark of dysmyelination this suggested to us that impaired AMPAR signalling in OL lineage cells might inhibit normal myelin development. We  Fluorescence ISH for Gria2-4 followed by immunohistochemistry for Olig2 in P14 corpus callosum indicates that Gria2-4 are expressed in OL lineage cells (arrows). Occasional Gria + Olig2cells (e.g. arrowhead in middle row) are presumptive neurons. (D) Double fluorescence ISH for Gria2 and Pdgfra followed by immunolabelling for Olig2, P9 corpus callosum. All Gria2 + cells are Olig2 + OL lineage cells. Many of these are (Gria2 + , Pdgfra + , Olig2 + ) OPs. Some (Gria2 -, Pdgfra + ) cells (arrowhead) and (Gria2 + , Pdgfra -, Olig2 + ) cells (arrow) are Figure 2 continued on next page decided to test this directly by inactivating both Gria2 and Gria3 in OL lineage cells. Before attempting this we checked whether deleting either subunit on its own affects OL development.

GluA2 and GluA3 are individually dispensable for OL production
Gria3 is X-linked in mice. To check whether Gria3 null mice (i.e. Gria3 -/Y males or Gria3 -/females) exhibit any obvious OL lineage-related defects, we immunolabelled P14 and P21 forebrain sections with anti-Pdgfra to detect OPs or with monoclonal CC1 (anti-adenomatous polyposis coli, APC) for differentiated OLs. The density (cells per mm 2 ) of Pdgfra + OPs in the SCWM was unaltered in Gria3 null mice compared to wild type littermates at both ages (p=0.66 at P14; p=0.7 at P21, Mann-Whitney test; Figure 3A,D) and there was no change in the density of CC1 + OLs (p=0.66 at P14; p=0.7 at P21, Mann-Whitney test; Figure 3B,E). We examined the proliferation rate of OPs by administering 5'-ethynyl-2'-deoxyuridine (EdU) for 8 hr on P14. There was no significant difference in the fraction of Pdgfra + OPs that incorporated EdU in Gria3 null mice versus wild type controls (p=0.66, Mann-Whitney test; Figure 3C,F), indicating that the proliferation rate of OPs was not affected by loss of GluA3. We also counted the density of myelinated axons as well as their g-ratios in parasagittal sections of anterior SCWM in electron micrographs and found no difference between Gria3 null mice and wild type controls at P14 (p=0.4, Mann-Whitney test; Figure 3G,H). Furthermore, there was no change in the frequency distribution of either myelinated or unmyelinated axon diameters ( Figure 3I,J). Therefore, loss of GluA3 alone does not detectably affect postnatal OL production or myelination.
The GluA2 subunit is functionally distinct from other AMPAR subunits because its mRNA can be edited to change a glutamine (Q) codon to an arginine (R) codon, causing AMPAR tetramers that contain the edited 'R' version to be Ca 2+ -impermeable. To investigate the role of GluA2 in OL development we generated conditional knockouts (cKOs) by crossing Sox10-Cre (Matsuoka et al., 2005) to Gria2 flox/flox mice, in which exon 11 of Gria2 is flanked by loxP sites (Shimshek et al., 2006b), on the Rosa-YFP reporter background (Srinivas et al., 2001). In the CNS, Sox10 expression is restricted to OL lineage cells -OPs in the embryo and both OPs and OLs postnatally (Stolt et al., 2002). The above cross generated Sox10-Cre: Gria2 flox/flox cKOs alongside Sox10-Cre: Gria2 flox/+ and Sox10-Cre: Gria2 +/+ littermate controls (all including Rosa-YFP) (Figure 4-figure supplement 1A). We refer to these as Gria2 -/-, Gria2 +/and Gria2 +/+ . Most OPs (~80%) in the postnatal SCWM of wild type mice originate after birth (E21/P0) from neural stem cells in the cortical ventricular zone (VZ), with the remaining~20% of OPs migrating into the developing cortex from the ventral forebrain starting around embryonic day 16 (E16) (Kessaris et al., 2006;Tripathi et al., 2011;reviewed by Bergles and Richardson, 2015). Thus, Sox10-Cre driven recombination at the Gria2 flox locus is expected to commence in the cortex and SCWM before birth and to continue after birth.
Control Sox10-Cre: Rosa-YFP mice (without Gria2 flox ) displayed widespread expression of the YFP reporter ( Figure 4-figure supplement 2). In the SCWM at P14, 96.4% ± 1.0% of Sox10 + cells expressed YFP (5551 cells counted in four mice; Figure 4-figure supplement 2A). In the cortex, 97.1 ± 0.7% of Sox10 + cells expressed YFP (2506 Sox10 + cells counted in four mice; Supplementary Figure 4-figure supplement 2B), confirming that Cre was active in the great majority of OL lineage cells in both grey and white matter. YFP expression was restricted to Sox10 + cells in SCWM (98.1 ± 0.5% of YFP + cells were Sox10 + ) and was also predominantly in Sox10 + cells in cortical grey matter (88.1 ± 1.6% of YFP + cells were Sox10 + ; Figure 4-figure supplement 2). YFP expression was also observed in some astrocytes and neurons (NeuN + ) in the ventral forebrain and in a small fraction of cortical projection neurons (1.5 ± 0.4%; 9894 NeuN + neurons counted in three animals; Figure 4figure supplement 2B) as well as some astrocytes (not quantified).
To confirm that GluA2 was deleted in Gria2 -/-OPs, fluorescence-guided whole cell patch-clamp recordings from YFP + OPs in the SCWM of P14 mice were compared to age-matched controls. AMPAR tetramers lacking GluA2 display inwardly-rectifying current responses to kainate and AMPA, due to a voltage-dependent block of the channel by polyamines at positive potentials (Donevan and Immunolabelling for Pdgfra + OPs (A) or CC1 + OLs (B) in P14 corpus callosum of wild type (WT) and Gria3 null mice. (C) EdU was administered to P14 wild type and Gria3 null mice for 8 hr before analyzing the corpus callosum by Pdgfra immunolabelling and EdU detection. (D, E) Cell counts reveal no difference in the density of OPs at P14 and P21 (D; p=0.66 at P14, p=0.7 at P21, Mann-Whitney test; >600 cells counted in each mouse) or OLs (E; p=0.66 at P14, p=0.7 at P21, Mann-Whitney test; >900 cells counted in Figure 3 continued on next page Rogawski, 1995). When GluA2 is present (in the edited 'R' form), the current-voltage (I-V) relationship is linear, that is, non-rectifying. Application of kainate (100 mM), which activates AMPA and kainate receptors, elicited an inward current at negative potentials in all OPs examined ( Figure 4A). The I-V relationship of the kainate-evoked current in the presence in the patch pipette of spermine (100 mM), a polyamine that reduces outward current flow in AMPARs that lack GluA2, appeared inwardly-rectifying in all cells examined from Gria2 -/brain slices but was almost linear in Gria2 +/and Gria2 +/+ controls ( Figure 4B,C). The rectification index (RI), defined as (kainate-mediated current amplitude at +17 mV)/ (current amplitude at -63 mV), was significantly reduced in Gria2 -/mice compared to Gria2 +/and Gria2 +/+ (p<10 -4 , one-way ANOVA with Bonferroni post-hoc test; Figure 4D), confirming that GluA2 had been successfully deleted. Despite this, the kainate-evoked current density (current normalized to membrane capacitance, which is proportional to cell surface area) at -63 mV in Gria2 -/-OPs was not significantly reduced compared to Gria2 +/or Gria2 +/+ (p=0.73, Kruskal-Wallis test; Figure 4E). Thus, activation of AMPARs present at the cell surface in the presence or absence of GluA2 generates the same membrane current at -63 mV, presumably as a result of compensation from other AMPAR subunits, but the voltage-dependence of the current is different in Gria2 -/mice ( Figure 4C).
The density (cells per mm 2 ) of Pdgfra + OPs in SCWM was unaltered in Gria2 -/mice compared to Gria2 +/+ controls at P14 (p=0.34, Mann-Whitney test; Figure 4F,H), as was the density of CC1 + cells (p=0.40, Mann-Whitney test; Figure 4F,H). To test whether the OP proliferation rate was altered we administered EdU to P14 Gria2 -/and Gria2 +/mice ( Figure 4G,I). There was no significant difference between the EdU labelling indices of Pdgfra + OPs in Gria2 -/versus Gria2 +/-(p=0.94, Mann-Whitney test), indicating that the cell cycle time of OPs was not affected ( Figure 4G,I). These results demonstrate that OL development is normal in the absence of GluA2, despite the altered AMPAR Ca 2+ permeability.

Double deletion of GluA2/ GluA3 in the OL lineage
To generate mice in which OL lineage cells lack both GluA2 and GluA3, we deleted Gria2 flox conditionally, using Sox10-Cre, in the Gria3 germline KO background. We chose this approach rather than attempting a double Gria2/Gria3 cKO because it simplifies the breeding program and it might increase the probability of Cre-mediated deletion at both Gria2 flox/flox alleles, by eliminating potential competition with Gria3 flox .
By appropriate breeding ( Figure 4-figure supplement 1B) we generated mice with the genotype Sox10-Cre: Gria3 null : Gria2 flox/flox : Rosa-YFP (referred to as Gria3 null 2 -/-). Control mice were Gria3 null with two functional copies of Gria2, either because they carried wild type alleles of Gria2 (no loxP sites) or because they did not carry the Sox10-Cre transgene. In some experiments, mice with one functional copy of Gria2 were also used as controls (Figure 4-figure supplement 1B). We designate these controls Gria3 null 2 +/+ and Gria3 null 2 +/-, respectively. As observed for Gria2 -/-(see above), the kainate-evoked current in Gria3 null 2 -/was inwardly-rectifying in all YFP + OPs that we examined, whereas Gria3 null 2 +/+ and Gria3 null 2 +/controls had an almost linear I-V relationship . The rectification index was significantly reduced in Gria3 null 2 -/mice compared to Gria3 null 2 +/and Gria3 null 2 +/+ controls (p<10 À4 , Kruskal-Wallis with Dunn's post-hoc test; Figure 5-figure supplement 1B). These data confirm that, as for Gria2 -/single cKOs, our genetic strategy had deleted GluA2 efficiently in OPs of Gria3 null 2 -/mice. each mouse) in Gria3 null versus WT corpus callosum. (F) Cell counts reveal no difference between the EdU labelling indices of Pdgfra + OPs in WT versus Gria null corpus callosum (p=0.66, Mann-Whitney test). (G) There was no change in the number of cross-sectional myelin figures per unit area in P14 Gria3 null compared to WT littermate controls (p=0.4, Mann-Whitney test). (H) Scatter plot of g-ratios as a function of axon diameter in P14 Gria3 null and wild type mice (>130 axons measured in each of three mice per group). (I) Frequency distribution of myelinated axon diameters from P14 Gria3 null and WT mice (>800 axons measured in each of three mice per group). (J) Frequency distribution of unmyelinated axon diameters from P14 Gria3 null and WT mice (>800 axons measured in each of three mice per group). Numbers of mice analyzed are indicated in (D), (E) and (F). Scale bar: 50 mm. DOI: 10.7554/eLife.28080.004 AMPAR-mediated currents are reduced in Gria3 null 2 -/-OPs Gria3 null 2 -/mice were viable and survived into adulthood without neurological signs, suggesting that the tremor reported in Gria2/3 germline KO mice (Meng et al., 2003) was not related to impaired AMPAR signalling in OL lineage cells. Since Gria3 null 2 -/-OPs lack both GluA2 and GluA3, and GluA1 appears not to be expressed in the SCWM of wild type mice (see above, first paragraph of Results; Figure 1), this implies that Gria3 null 2 -/-OPs express GluA4 alone. Thus, it might be expected that fewer AMPARs would be assembled in OPs from Gria3 null 2 -/mice relative to controls.
To test this, we whole-cell patch-clamped YFP + OPs in the SCWM of P14 Gria3 null 2 -/and Gria3 null 2 +/+ mice and measured kainate-evoked currents at -63 mV. In both genotypes, bath application of 100 mM kainate elicited inward currents, indicating that AMPA/kainate receptors were present ( Figure 5A). Normalizing the amplitude of the maximum kainate-evoked current at -63 mV to the cell size (assessed as membrane capacitance) revealed a significant (~47%) reduction in the current density for Gria3 null 2 -/compared to Gria3 null 2 +/+ controls (p<10 -4 , Student's t-test; Figure 5B), indicating that the number of AMPARs expressed on the cell surface was decreased in Gria3 null 2 -/mice. The alternative explanation, that the single-channel current passed by the remaining AMPARs was reduced, is less likely because GluA4 homo-tetramers are reported to have a higher conductance than GluA2/GluA4 hetero-dimers, for example (Swanson et al., 1997).
Application of kainate activates both AMPA and kainate receptors. When a fast application system is used it is possible to distinguish the activation of each receptor type based on the kinetics of the response (Patneau et al., 1994); kainate receptors desensitize quickly to kainate while AMPARs desensitize slowly or not at all. With the slow application method that we employed, most of the kainate-evoked current should therefore be AMPAR-mediated. To confirm this, we examined the responses to kainate in the presence of 50 mM GYKI-52466 (GYKI), a selective AMPAR antagonist. In both Gria3 null 2 -/and Gria3 null 2 +/+ OPs, kainate-evoked currents were blocked by~80% in the presence of GYKI ( Figure 5A,C), confirming that most of the kainate-evoked current is AMPAR-mediated under our experimental conditions. Thus, in Gria3 null 2 -/mice, activation of AMPARs in OPs generates less membrane current (and hence less depolarization) than in Gria3 null 2 +/+ controls.
The fact that a significant fraction (~53%) of the normal kainate current is still observed in the Gria3 null 2 -/-OPs ( Figure 5B) indicates that AMPARs are still formed and trafficked to the surface membrane. It is possible either that the remaining GluA4 subunits form functional homo-tetrameric receptors, as they can do when expressed on their own in cultured mammalian cells (Keinänen et al., 1990), or that GluA1 is up-regulated in response to loss of GluA2 and GluA3 and forms GluA1/GluA4 heteromeric or GluA1 homomeric channels. However, by ISH we found no evidence for up-regulation of Gria1 mRNA in Gria3 null 2 -/-SCWM relative to Gria3 null 2 +/+ (Figure 5figure supplement 2). Thus, the majority of AMPAR formed in Gria3 null 2 -/-OPs are probably homotetramers of GluA4. With 100 mM spermine included in the patch pipette, inward rectification of the current was observed in Gria2 -/but not in Gria2 +/or Gria2 +/+ . (D) Rectification index (RI; current at +17 mV/ current at -63 mV) of cells from (C). RI is greatly reduced in Gria2 -/mice (p<10 -4 , one-way ANOVA with Bonferroni post-hoc test). (E) Mean kainate-evoked current density (current/capacitance, pA/pF) is the same in P14 Gria2 -/-, Gria2 +/and Gria2 +/+ OPs (p=0.73, Kruskal-Wallis with Dunn's post-hoc test). (F) Immunolabelling for Pdgfra + OPs (top) and CC1 + OLs (bottom) in P14 Gria2 -/and Gria2 +/+ corpus callosum. (G) EdU incorporation by proliferating OPs (Pdgfra + ) in the P14 corpus callosum of Gria2 -/and Gria2 +/mice. The bottom panels show higher-magnification images of the cells indicated on the top; all Pdgfra + OPs co-expressed YFP confirming that Sox10-Cre is expressed in all these cells. (H) There were no significant changes in the densities of either Pdgfra + OPs (p=0.34, Mann-Whitney test; >600 cells counted per mouse) or CC1 + oligodendrocytes (p=0.40, Mann-Whitney test; >900 cells counted per mouse) at P14. (I) There were no significant differences in the fractions (%) of EdU + Pdgfra + OPs in Gria2 -/versus Gria2 +/-(p=0.89, Mann-Whitney test). Numbers of cells and mice analyzed are indicated in (D, E) and (H, I), respectively. Scale bars: 50 mm. The kainate-evoked current is blocked reversibly by the AMPAR antagonist GYKI-52466 (GYKI, 50 mM). (B) Mean kainate-evoked current density is 47% less in Gria3 null 2 -/-OPs compared to Gria3 null 2 +/+ (p<10 À4 , Student's t-test). (C) GYKI blocks the kainateevoked current in both Gria3 null 2 -/and Gria3 null 2 +/+ mice (normalized to 100% before GYKI application). (D, E) No Figure 5 continued on next page To determine whether deletion of GluA2/3 from OL lineage cells altered their general membrane properties, the electrophysiological properties of YFP + OPs from SCWM of P14 Gria3 null 2 -/and Gria3 null 2 +/+ mice were compared ( Figure 5D-G). Neither the membrane resistance (R m ) nor the capacitance (C m ) of Gria3 null 2 -/-OPs at -63 mV were different from Gria3 null 2 +/+ controls (for R m , p=0.16, Mann-Whitney test and for C m , p=0.13, Student's t-test; Figure 5D,E). In addition, when 20 mV voltage steps were applied from a holding potential of -63 mV, membrane depolarization evoked a transient inward Na + current (I Na ; Figure 5F). Subtraction of the linearly scaled capacitive transient and ohmic leak current evoked by a 20 mV hyperpolarizing step allowed us to estimate the amplitude of I Na ( Figure 5F). There was no significant difference in the I Na density between Gria3 null 2 -/and Gria3 null 2 +/+ controls (p=0.72, Mann-Whitney test; Figure 5G), suggesting that altering AMPAR signalling does not regulate the expression of voltage-gated sodium channels. We conclude that AMPAR signalling does not much affect the membrane properties of OPs.
Spontaneous synaptic currents occur at very low frequency at neuron-OP synapses but their frequency can be increased by application of the secretagogue Ruthenium Red (RR) (Lin and Bergles, 2004;Kukley et al., 2007). RR induces vesicular release at the synapse, increasing the frequency of miniature EPSCs without affecting their amplitude (Sciancalepore et al., 1998). Bath application of 100 mM RR caused a marked increase in the frequencies of spontaneous EPSCs recorded from OPs of Gria3 null 2 +/+ , Gria3 null 2 +/and Gria3 null 2 -/mice. Consistent with the minimal stimulation-evoked EPSC data, the mean amplitude of individual RR-evoked EPSCs was not reduced in Gria3 null 2 -/compared to Gria3 null 2 +/controls or Gria3 null 2 +/+ controls (p=0.32, one-way ANOVA; Figure 6C,D) -suggesting that, at a given synapse, post-synaptic AMPARs generate a similar current in Gria3 null 2 -/as in controls. However, the frequency of RR-evoked EPSCs was~70% less in Gria3 null 2 -/than in Gria3 null 2 +/+ controls, and was reduced to an intermediate frequency in Gria3 null 2 +/controls (p=2Â10 -4 , Kruskal-Wallis with Dunn's post-hoc test; Figure 6E). Consistent with previous reports (Ziskin et al., 2007), EPSCs were strongly inhibited by the application of 50 mM GYKI (by 88.8 ± 6.5%, n = 3 for Gria3 null 2 -/-, 83.3 ± 3.8%, n = 2 for Gria3 null 2 +/and 93.8 ± 2.4%, n = 2 for Gria3 null 2 +/+ ), provving that they are AMPAR-mediated. The reduction in frequency of EPSCs suggests either that the probability of RR-evoked presynaptic vesicle release is decreased in Gria3 null 2 -/or, more likely, that fewer synapses are made between neurons and OPs in Gria3 null 2 -/mice. significant changes in membrane resistance Rm (D), p=0.16, Mann-Whitney test) or capacitance Cm (E), p=0.13, Student's t-test) of OPs from Gria3 null 2 -/compared to Gria3 null 2 +/+ . (F) I Na recorded from P14 Gria3 null 2 +/+ OPs. Top current traces are responses to voltage steps in 20 mV increments from À63 mV, showing the capacity current and the subsequent (much smaller) ionic current changes evoked by the voltage steps. Bottom trace is after subtraction of the linearly scaled response to a 20 mV hyperpolarizing step. (G) There was no significant change in I Na density in Gria3 null 2 -/compared to Gria3 null 2 +/+ controls (p=0.72, Mann-Whitney test). Numbers of cells analyzed are indicated in (B-E, G). DOI: 10.7554/eLife.28080.008 The following figure supplements are available for figure 5: To distinguish between these possibilities we estimated the paired pulse ratio (PPR) at neuronal-OP synapses in SCWM. The PPR was calculated from two consecutive evoked EPSCs 25 ms apart, recorded from patch-clamped OPs. There was no difference in PPR of Gria3 null 2 -/-OPs versus Gria3 null 2 +/+ OPs (p=0.95, Student's t-test; Figure 6F). This suggests that the reduction in EPSC frequency is not due to a change in presynaptic glutamate release but rather because fewer active synapses are formed or maintained in Gria3 null 2 -/mice.
OL survival is reduced, and OL accumulation delayed, in Gria3 null 2 -/white matter Having established that OPs in Gria3 null 2 -/mice receive reduced synaptic input, we asked whether the development of OL lineage cells was affected. We counted CC1 + differentiated OLs in the SCWM of early postnatal (P7, P14, P21) and adult (P70) Gria3 null 2 -/and Gria3 null 2 +/+ mice. There was a~22% reduction in the density of OLs at P7 and a~27% reduction at P14 in Gria3 null 2 -/compared to Gria3 null 2 +/+ controls (p=0.018 at P7; p=0.001 at P14, t-tests with Holm-Bonferroni correction; Figure 7A,B). By P21 (p=0.14) and P70 (p=0.48, Mann-Whitney test with Holm-Bonferroni correction) this difference was no longer significant. These data suggest that production or survival of OLs is impaired in Gria3 null 2 -/mice in the early postnatal period.
Enpp6 (ectonucleotide pyrophosphatase/phosphodiesterase family member 6; an ecto-enzyme with choline-specific glycerophosphodiester phosphodiesterase activity: Morita et al., 2016) is a recently-identified marker of newly-differentiating OLs that is strongly down-regulated in mature OLs (Zhang et al., 2014;Xiao et al., 2016). The number of Enpp6 + cells present at any given time therefore provides a measure of the rate of formation of OLs from their precursors (OPs). We visualized Enpp6 + cells in the SCWM of P14 Gria3 null 2 -/and Gria3 null 2 +/+ mice by ISH. This revealed a~26% reduction of Enpp6 + newly-differentiated OLs in Gria3 null 2 -/mice compared to their Gria3 null 2 +/+ littermates (p=0.004, Mann-Whitney test; Figure 7C,D), reinforcing our conclusion that the rate of generation or survival of newly-forming OLs is reduced by loss of GluA2/GluA3. That is, glutamate signalling through AMPARs in OPs normally stimulates OPs to exit the cell cycle and differentiate into OLs, or else enhances the survival of immature OLs once they have formed.
We attempted to distinguish between these two scenarios by comparing the density of Pdgfra + OPs in Gria3 null 2 -/and control mice. If OPs continue to divide but are unable to differentiate into OLs in the Gria3 null 2 -/mice we might expect the density of OPs to increase. We found no significant differences in the density of OPs in Gria3 null 2 -/-SCWM compared to Gria3 null 2 +/+ controls (p=0.6 at P3; p=0.85 at P7; p=0.07 at P14; p>0.99 at P21 and p=0.76 at P70, t-test or Mann-Whitney test with Holm-Bonferroni correction; Figure 7E,F). This suggests that reduced accumulation of OLs in the SCWM of Gria3 null 2 -/mice is not caused by a reduction in the rate of differentiation of OPs. However, the reduction in OL density might have resulted from a reduced proliferation rate of OPsthat is, more slowly dividing OPs produce fewer OLs. To assess OP division rates we administered EdU to mice at P3, P7, P14 and P70 (see Materials and methods). The fraction of Pdgfra + OPs that incorporated EdU ('labelling index') was not detectably different in Gria3 null 2 -/versus Gria3 null 2 +/+ SCWM at any age examined (p=0.69 at P3; p=0.69 at P7; p=0.7 at P14; p=0.34 at P70, Mann-Whitney tests; Figure 7-figure supplement 1A,B), nor was there any difference in the fraction of Pdgfra + OPs in the P14 SCWM that immunolabelled for Ki67, a marker of actively dividing cells (p=0.78, Student's t-test; Figure 7-figure supplement 1C,D). These data indicate that the OP division rate is not affected by loss of GluA2 and GluA3.
Reduced total myelin in Gria3 null 2 -/white matter A reduction in the number of OLs would be expected to lead to a reduction in myelination. To test this, we examined P14 Gria3 null 2 -/and Gria3 null 2 +/+ brains by electron microscopy (EM) and counted the number of transverse myelin figures viewed in parasagittal sections of SCWM ( Figure 8A,B). This revealed a~20% reduction in the number of myelin cross-sections in Gria3 null 2 -/normalized to littermate controls at P14 (p=0.02, Mann-Whitney test), consistent with the~27% reduction in the number of OLs in Gria3 null 2 -/mice at the same age (see previous section; Figure 7). Therefore, the most parsimonious explanation for the myelin deficit is that the number of OLs is decreased without a compensatory change in the number or length of myelin internodes made by individual OLs. To test this, we dye-filled OLs from P14 Gria3 null 2 -/and Gria3 null 2 +/+ brain slices and measured the lengths and numbers of internodes associated with individual OLs. There was no change in the average length of internodes (p=0.47, Student's t-test) or the average number of internodes per OL (p=0.26, Mann-Whitney test; Figure 8C,D). In addition, by EM we found no change in the range or frequency distributions of myelinated ( Figure 8E) or unmyelinated ( Figure 8F) axon diameters at P14, arguing against an indirect effect of the Gria3 null and/or Gria2 -/mutations on OLs and myelin via axons -for example, because the fraction of axons that exceeded the size threshold for myelination might have been reduced in the Gria mutants. We also measured myelin thickness, expressed as g-ratio, and found no significant change (p=0.29, Mann-Whitney test; Figure 8G).
At P70 there was no longer any significant difference in the number of myelin cross-sections per unit area observed by EM (p=0.57, Mann-Whitney test; Figure 8B), consistent with the fact that the numbers of OLs identified by CC1 immunolabelling were similar in Gria3 null and Gria3 null 2 -/at this age ( Figure 7B). At P70, as at P14, there was no change in the g-ratios of myelinated axons in Gria3 null versus Gria3 null 2 -/corpus callosum (p=0.29, Mann-Whitney test; Figure 8H). Our data are consistent with a model in which each OL synthesizes a similar total length of myelin sheath in Gria3 null 2 -/mice as in controls and the overall synthesis of myelin is reduced in proportion to the reduced number of OLs.
We counted CC1 + OLs and Pdgfra + OPs in the SCWM of Gria3 null 2 -/-4 -/mice at P14 and P53. At P14 there was a~22% reduction in the density of OLs relative to Gria3 null controls and at p53 a~26% reduction (p=0.026 at P14; p=0.029 at P53, Mann-Whitney test, Figure 10A,B) without a concomitant change in the density of OPs (p=0.11 at P14; p=0.7 at P53, Mann-Whitney test) ( Figure 10C,D). The fraction of Olig2 + cells that expressed cleaved Caspase-3 in SCWM of P14 Gria3 null controls was 24.8 ± 0.8% (n = 5 mice) and this increased in Gria3 null 2 -/-4 -/mice by a factor of~1.24 (increased by 24.2 ± 4.8%, p=0.004, Mann-Whitney test; Figure 10E,F). As observed for Gria3 null 2 -/mice, there was no significant change in the average length (p=0.93, Student's t-test) or number (p=0.41, Student's t-test) of internodes per individual Gria3 null 2 -/-4 -/-OL, analyzed by dyefilling OLs in brain slices ( Figure 10G,H). Taken together, our data demonstrate that signalling through AMPAR stimulates OL production and myelination in the white matter by enhancing survival of newly-formed OLs, without a detectable effect on OL morphology.

Discussion
Previous studies have shown that signalling between electrically active axons and OL lineage cells can influence several aspects of OL development, including proliferation and differentiation of OPs and the initiation and stabilization of myelin internodes. We have now provided the first direct evidence that synaptic signalling through AMPARs in OPs and/or early OLs modulates OL development in vivo. Knock-out of Gria2 or Gria3 alone had no discernible effect on OL development but double-KO of Gria2/3 in the OL lineage reduced, and triple-KO of Gria2/3/4 abolished, AMPAR-mediated signalling in OPs and led to reduced accumulation of mature OLs and myelin without affecting the proliferation or number of OPs. The presence of GluA2 subunits (encoded by Gria2) is known to Figure 8 continued (p=0.26, Mann-Whitney test) in Gria3 null 2 -/compared to Gria3 null 2 +/+ mice. (E) Frequency distribution of myelinated axon diameters from P14 Gria3 null 2 +/+ and Gria3 null 2 -/mice (100-350 axons measured in each of three mice per group). (F) Frequency distribution of unmyelinated axon diameters from P14 Gria3 null 2 +/+ and Gria3 null 2 -/mice (>300 axons measured in each of three mice per group). (G) Scatter plots of g-ratios as a function of axon diameter in P14 Gria3 null 2 +/+ and Gria3 null 2 -/mice (>260 axons measured from four mice per group). (H) Scatter plots of g-ratios as a function of axon diameter in P70 Gria3 null 2 +/+ and Gria3 null 2 -/mice (>150 axons measured from each of three mice per group). Number of mice analyzed is indicated in (B) and number of cells analyzed in (D). DOI: 10.7554/eLife.28080.015 Figure 9. Gria3 null 2 -/-4 -/mice lack AMPAR-mediated inputs. (A) The kainate-evoked current is greatly reduced in OPs of P14 Gria3 null 2 -/-4 -/mice compared to Gria3 null 2 -/littermates and is less effectively blocked by the AMPAR antagonist GYKI-52466 (GYKI, 50 mM). (B) Mean kainate-evoked current density in P14 Gria3 null 2 -/-4 -/-OPs is~24% that of Gria3 null 2 -/littermates (p<10 -4 , Student's t-test with Welch correction), corresponding tõ 13% that of Gria3 null OPs ( Figure 5B). (C) GYKI blocks the kainate-evoked current by~80% in Gria3 null 2 -/-, but only by~39% in. Gria3 null 2 -/-4 -/mice Figure 9 continued on next page suppress the Ca 2+ permeability of AMPARs (provided GluA2 is the Q -> R edited isoform). The fact that knockout of Gria2 alone did not affect OL development implies that the reduction in OL number observed in the Gria2/3 and Gria2/3/4 mutants results from an aspect of AMPAR signalling other than Ca 2+ permeability, for example the induced depolarization or the Na + influx that it produces. Thus, AMPAR signalling in OPs and/or early-differentiating OLs is a positive regulator of OL development; we speculate that this favours myelination of active axons that are releasing glutamate, thus accelerating development of a functioning CNS.

Effect of AMPAR subunit knockouts on synaptic input to OPs
OPs in developing white matter receive glutamatergic synaptic input from unmyelinated axons in white matter. Most glutamatergic synaptic current in OPs is via AMPAR (Lin and Bergles, 2004; reviewed by Bergles and Richardson, 2015). We have shown that OL lineage cells in the SCWM express mRNA encoding AMPAR subunits GluA2, GluA3 and GluA4 but not GluA1 -the same spectrum of AMPAR subunits reported for rat and mouse OPs in culture (Patneau et al., 1994;Zhang et al., 2014). Knocking out GluA2 and GluA3 in OPs reduced the AMPAR-mediated current to~53% of normal, while knocking out all three subunits GluA2/3/4 reduced it to~13% of normal. Surprisingly, we found that in OPs lacking GluA2 and GluA3 the amplitude of RR-induced synaptic events was unchanged while their frequency was reduced to <30% of normal, while in OPs lacking all AMPAR subunits their frequency was reduced to only~1% of control. The simplest interpretation is that there are fewer AMPAR-containing synapses in double-KO OPs (and almost none in triple-KOs), but that those synapses that survive contain a normal number of AMPARs. In support of this it has been shown in neurons that, when the number of post-synaptic AMPARs is increased experimentally, an increase in frequency of EPSPs with no change in their amplitude is observed (Sinnen et al., 2017), suggesting that the number of functioning synaptic connections is determined, at least in part, by the number of postsynaptic AMPARs available, and that quantal amplitude is regulated independently of the number of synaptic connections. In the GluA2/3 double-KO the surviving synapses must contain exclusively or predominantly GluA4 subunits, which are known to form functional homo-tetramers (Keinänen et al., 1990). We cannot tell whether separate 'GluA4-only' and 'GluA4lacking' synapses are normally present on OPs, although distinct sets of GluA2/GluA3-containing and GluA2/GluA3-lacking synapses have been described for neurons in the carp retina (Schultz et al., 2001). In any case our data provide striking genetic confirmation that almost all glutamatergic synaptic input to OPs is via AMPARs composed of GluA2, GluA3 and GluA4 subunits.

AMPAR signalling does not affect OP proliferation
We found that OP proliferation and OP number were normal in Gria3 null 2 -/and Gria3 null 2 -/-4 -/mice. Previous studies disagree over whether activity-driven glutamate release stimulates, inhibits or has no effect on proliferation of OPs. Experiments with organotypic rat or mouse cerebellar slice cultures found that adding AMPA/ kainate receptor agonists inhibited proliferation of OPs (Yuan et al., 1998;Fannon et al., 2015). However, direct electrical stimulation of glutamatergic pyramidal neurons in the rat motor cortex was found to stimulate OP proliferation in the cerebro-spinal tract (Li et al., 2010), and optogenetic stimulation of projection neurons in mouse premotor cortex stimulated OP proliferation in the premotor pathway, including the SCWM (Gibson et al., 2014). Moreover, blocking the activity of retinal ganglion cells (RGCs) and their axons in the mouse optic nerve by intra-ocular injection of tetrodotoxin inhibited OP proliferation . On the other hand, sensory deprivation of mouse barrel cortex by whisker removal (Mangin et al., 2012;Hill et al., 2014), which presumably restricts glutamate release in the cortex, increased OP proliferation in the barrels, in keeping with the data from brain slice cultures (Yuan et al., 1998; The density of CC1 + OLs is significantly less (by~22% at P14 and~26% at P53) in Gria3 null 2 -/-4 -/mice relative to Gria3 null controls (p=0.026 at P14, p=0.029 at P53, Mann-Whitney test; >900 cells were counted per mouse at all ages). (C) Pdgfra + OPs in Gria3 null and Gria3 null 2 -/-4 -/mice. (D) There were no significant differences in the density of Pdgfra + OPs at P14 or P53 in Gria3 null 2 -/-4 -/versus Gria3 null mice Figure 10 continued on next page Fannon et al., 2015). Against all of the above, a recent in vivo study, in which glutamate release from the axons of mouse RGCs was genetically attenuated, found no effect on OP proliferation in the optic tract (Etxeberria et al., 2016). Genetic block of vesicular release from longitudinally-projecting axons in the zebrafish spinal cord also had no effect on OP proliferation (Hines el al., 2015;Mensch et al., 2015), although numbers of OL lineage cells were reduced by~10%, apparently because of reduced specification of OPs from their stem cell precursors in the embryonic ventricular zone (Mensch et al., 2015). It is unlikely that we would have detected an effect of this size, even assuming it would not have been overridden by homeostatic controls on cell number in the perinatal mouse brain.
The data cited above are contradictory and confusing. However, the experiments are not directly comparable to each other. For example, pharmacological manipulation of slice cultures is not targeted to specific cells and opens the possibility that there might be indirect effects mediated via neurons, some of which also express AMPA/ kainate receptors. The same reservation might apply to some of the in vivo experiments cited above -on top of which, stimulating or inhibiting electrical activity or vesicular release in vivo is likely to affect more than just AMPAR-mediated signalling. For example, electrically active neurons release Neuroligin-3 (NLGN3), a mitogen for malignant glioma cells, which are believed to be derived from and closely related to OPs (Venkatesh et al., 2015). Our conditional knockout approach is tightly focused on AMPAR-mediated effects in OL lineage cells and avoids confounding influences by other signalling pathways, direct or indirect. We conclude, as discussed below, that signalling through AMPAR in OL lineage cells promotes survival of newly-differentiating OLs.

AMPAR signalling promotes the survival of differentiating OLs
Reduced AMPAR signalling in the SCWM of our Gria3 null 2 -/mice resulted in a modest (~27%) reduction in OL accumulation in the early postnatal period (around P14), but OL numbers recovered to normal levels in adult mice. Additional deletion of Gria4 did not reduce the number of OLs further, but the reduction now persisted into adulthood. We assume that these changes reflect the loss of synaptic currents that we have documented above. The relatively modest effect on OL dynamics of ablating AMPAR might reflect the fact that glutamate can signal to OPs through receptors other than AMPAR (including NMDAR, KAR or mGluR), and that glutamate is only one of a number of signalling molecules that communicate axonal activity to OPs.
Previous studies into the effects of glutamatergic signalling on OL differentiation have been conflicting. Experiments in dissociated cell cultures and slice cultures from rat or mouse indicated that blocking AMPA/ kainate receptors promoted OL differentiation (Gallo et al., 1996;Yuan et al., 1998;Fannon et al., 2015) while preventing their further maturation and expression of myelin proteins (Fannon et al., 2015). In vivo, manipulating electrical activity pharmacologically (Demerens et al., 1996), electrically (Li et al., 2010), optogenetically (Gibson et al., 2014) or physiologically (Simon et al., 2011;McKenzie et al., 2014;Xiao et al., 2016) have suggested that axonal activity promotes OL generation and myelin formation in rats and mice. However, inhibiting vesicular release from axons in the developing zebrafish spinal cord by pan-neuronal expression of tetanus toxin had no effect on OL differentiation, but reduced the number and length of myelin sheaths synthesized by individual OLs (Hines et al., 2015;Mensch et al., 2015). This effect was observed for OLs that myelinated some but not all classes of axon (Koudelka et al., 2016) -consistent with the idea that there are both activity-dependent and -independent mechanisms of myelination (Lundgaard et al., 2013). However, tetanus toxin inhibits all SNARE-dependent vesicular release, including release of neurotransmitters other than glutamate (e.g. ATP and GABA) that have also 10 continued (p=0.11 at P14; p=0.7 at P53, Mann-Whitney test; >400 cells were counted per mouse at all ages). (E) Cleaved Caspase-3 + , Olig2 + OL lineage cells in Gria3 null 2 -/-4 -/mice. Cells in the rectangle (dotted line) are shown on the right at higher magnification. (F) There was a~24% increase in the fraction of Olig2 + cells that expressed cleaved Caspase-3 in Gria3 null 2 -/-4 -/compared to Gria3 null mice (p=0.004, Mann-Whitney test; >1500 Olig2 + cells were counted in each mouse). (G) Dye-filled OLs in P14 corpus callosum of Gria3 null and Gria3 null 2 -/-4 -/mice. (H) There was no change in the length of the internodes (p=0.93, Student's t-test) or the number of internodes per OL (p=0.41, Student's t-test) in Gria3 null 2 -/-4 -/compared to Gria3 null mice. Numbers of mice analyzed are indicated in (B), (D) and (F) and numbers of cells in (H). Scale bars: 50 mm. DOI: 10.7554/eLife.28080.017 been implicated in the control of myelination (Stevens et al., 2002;Ishibashi et al., 2006;Zonouzi et al., 2015;Hamilton et al., 2017). Further complicating the picture, inhibition of glutamate release from the axons of RGCs increased OL number in the mouse optic tract (Etxeberria et al., 2016). Although the latter study preferentially affected glutamate signalling it did not separate AMPAR-mediated effects from those mediated by e.g. NMDAR or mGluR.
We found that the reduced number of differentiated OLs in the double-and triple-KOs was a result of increased death of OL lineage cells -most likely newly-formed, pre-myelinating OLs. In keeping with this, we found that Gria2-4 continue to be expressed in Enpp6 + early-differentiating OLs, suggesting that some AMPAR-mediated signalling can still occur during the early stages of OL differentiation, even though AMPAR-mediated EPSCs are down-regulated in maturing OLs (Kukley et al., 2010;De Biase et al., 2010). Synaptic input -and activation of AMPARs in particular -has been shown to promote neuronal survival during early postnatal development (Heck et al., 2008) and a recent study showed that sensory input from the whiskers to somatosensory barrel cortex promotes the survival of newly-differentiating OLs (Hill et al., 2014). Thus, AMPAR signalling in OL lineage cells could be one of the mechanisms ensuring that only pre-myelinating OLs that make contact with active axons survive to myelinate in the longer term (Barres et al., 1992Trapp et al., 1997).
Electrical activity is also known to affect the terminal stages of OL differentiation, that is, myelin synthesis (Demerens et al., 1996;Wake et al., 2011;Lundgaard et al., 2013;Gibson et al., 2014;Wake et al., 2015). Glutamate receptors other than AMPAR, that is, NMDAR and mGluR, have been suggested to regulate this process (Wake et al., 2011;Lundgaard et al., 2013), possibly by regulating the energy supply to developing OLs (Saab et al., 2016) -although a major effect of NMDARs in vivo has been disputed (De Biase et al., 2011;Guo et al., 2012). In our present study we found that the amount of myelin in the SCWM of Gria3 null 2 -/mice was reduced in proportion to the reduced number of OLs. The g-ratio, as well as the internode number and length, were unchanged, suggesting that once a pre-myelinating OL makes a commitment to myelinate an axon, the process continues until a normal internode number and length, and a normal number of myelin wraps are achieved. Therefore, the activity-dependent regulation of internode length that has been observed in vitro (Wake et al., 2015) and in vivo (Koudelka et al., 2016) is presumably mediated by glutamate receptors other than AMPAR, or by signals other than glutamate. In short, we have shown that AMPAR signalling stimulates myelin formation in mice by increasing the number of myelinating OLs that develop from OPs, not by regulating myelin synthesis per se.
In summary, our study demonstrates that AMPAR signalling in OL lineage cells promotes OL development and myelination during postnatal development. Elucidating further mechanisms of signalling from neurons to OL lineage cells will be crucial for understanding developmental myelination and for devising therapies to promote remyelination in diseases of the white matter.

Materials and methods
Transgenic mice AMPAR subunits GluA1-4 are encoded by separate genes Gria1-4. In the Gria2 flox/flox line, exon 11 of Gria2 is flanked by loxP sites (Shimshek et al., 2006a). Gria4 flox/flox mice (Gria4tm1Mony; MGI:3798453) also have exon 11 flanked by loxP sites (Fuchs et al., 2007). Gria3 null mice (Gria3 -/or Gria3 -/Y ) have exon 11 deleted in the germ line (Sanchis-Segura et al., 2006). Rosa-YFP reporter mice (Srinivas et al., 2001) allow Cre-induced expression of the enhanced yellow fluorescent protein (YFP). Most transgenic lines were maintained as homozygous breeding colonies. Sox10-Cre mice (line 22-7) (Matsuoka et al., 2005) express Cre recombinase under Sox10 transcriptional control in a phage artificial chromosome (PAC). They were maintained as a heterozygous breeding colony and only female carriers were used for experiments, because Sox10-Cre is expressed and causes recombination in the male germ line. Our breeding schemes are shown in Figure 4-figure supplement 1A-C. Mice were maintained on a 12 hr light-dark cycle. The day of birth was designated postnatal day 0 (P0). For histological experiments the age of the animals is stated; for electrophysiological experiments a range of ages P13-P16 was used, referred to collectively as P14.
Patch pipettes were pulled from thick-walled borosilicate glass capillaries (type GC150F-10; Harvard Apparatus) to a resistance of 3-5 MW. For RR experiments the intracellular solution contained 140 mM CsCl, 10 mM HEPES, 2 mM MgATP, 0.5 mM CaCl 2 , 4 mM NaCl, 5 mM EGTA, 0.052 mM Alexa Fluor 568, pH 7.3 (adjusted with CsOH). For the rest of the experiments the intracellular solution contained CsCl 145 mM, 10 mM HEPES, 4 mM MgATP, 2.5 mM NaCl, 1 mM EGTA, 0.052 mM Alexa Fluor 568, pH 7.2 (adjusted with CsOH). When indicated, spermine (Sigma) was added to the intracellular solution at a final concentration of 100 mM. For both solutions the junction potential of -3 mV has been corrected for.
YFP-expressing OPs in the corpus callosum were selected for whole-cell patch-clamping based on their size and shape in the fluorescence microscope, and their membrane properties. Cells with voltage-gated Na + currents, !1 GW membrane resistance and 35 pF capacitance were considered to be OPs. Due to the high membrane resistance of OPs, recordings were made without series resistance compensation. In some experiments the following drugs were applied by adding to the superfusing extracellular solution: kainic acid monohydrate (100 mM; Sigma) to activate AMPAR and kainate receptors, GYKI52466 (GYKI; 50 mM; Sigma) to block AMPARs, or RR (100 mM; Sigma) to evoke transmitter release. To record the OP's voltage-gated currents and the I-V relationship of drug-evoked currents, 20 mV voltage steps of 200 ms duration were applied from a holding potential of -63 mV (voltage range from -103 mV to +17 mV). For RR-evoked EPSCs the holding potential was -83 mV to increase the driving force for excitatory synaptic events and thus increase the chance of detecting the events.
For evoked synaptic responses cells were clamped at -63 mV and the responses were elicited by stimulation (Digitimer Limited) of callosal axons using a glass micropipette filled with external solution (resistance 1-2 MW) placed~150-200 mm away from the recorded cell in the corpus callosum. For minimal stimulation, activating only one axon with input to the recorded cell, the stimulus strength was gradually reduced until no post-synaptic current was detected in the patch-clamped OP. Then the stimulus was gradually increased until a post-synaptic response was detected. In each cell 50-80 trials under these conditions were recorded and analyzed. The amplitude of the unitary EPSC was obtained by averaging those events where an EPSC was observed. For estimation of paired-pulse ratio (PPR), two strong (90 V, 800 ms) stimulation pulses, 25 ms apart, were delivered ten times, responses were averaged and the PPR was calculated by dividing the peak current generated by the second pulse by the peak current generated by the first pulse.
Responses were recorded using an Axopatch 200B amplifier with pClamp 9 (Axon Instruments) and Axoscope 9 (Axon Instruments) software. The majority of membrane currents were filtered at 5-10 kHz, except for spontaneous synaptic currents, which were filtered at 1 kHz. The sampling rate was always at least three times the filtering frequency.
Data were analyzed off-line using Clampfit 9 (Axon Instruments). Membrane resistance and capacitance were estimated by fitting the transient current elicited by a 5 mV voltage step to a single exponential. Series resistance was monitored and was typically 10-20 MW. Cells were excluded from the analysis when the resistance changed by >50% during the recording and/or was >30 MW (unless otherwise stated). The amplitude of voltage-gated Na + currents was calculated by subtracting the passive components of current responses evoked by voltage steps. In particular, the current response to a 20 mV hyperpolarizing voltage step with no active components, comprising the capacitive current and an ohmic leak conductance (which are proportional to the size of the voltage step) was linearly scaled according to the size of the voltage step applied. This linearly scaled passive component was then subtracted from all of the voltage-gated Na + current responses evoked by depolarizing voltage steps.
For detection of RR-evoked EPSCs, traces were high pass filtered at 2 Hz to eliminate drift in the baseline over time. The events were automatically detected using a threshold search with the following criteria: amplitude threshold set at three times the s.d. of the noise in regions of the trace lacking obvious EPSCs, minimum event duration set to 1 ms and rising phase faster than the decay. The detected events were then either accepted or rejected by the experimenter. The frequency was calculated for each minute of drug application. For the majority of cells, the maximum frequency was reached after 3 min of RR application. Therefore, the frequency of the events recorded between 3 and 6 min after application of the drug was used for comparison between Gria3 null 2 -/and control mice. Event frequency during GYKI application was estimated over the last 3 min of drug application.
To quantify kainate-evoked currents, the peak kainate-evoked current was measured at a nominal holding potential of -63 mV. Kainate-evoked current magnitudes were then corrected after the experiment for voltage errors generated by series resistance, using the value of Rs, a reversal potential of 0 mV and the current flowing before and after applying kainate. All cells, irrespective of their series resistance, were included in this analysis.
Numbers of cells analyzed are indicated in the main text. For most experiments, cells from more than three mice of a given genotype were analysed, apart from the PPR experiment ( Figure 6F), the Gria3 null 2 +/+ group in the experiments of Figure 5B,C and the experiments in Figure 9 where two animals were used. Only one cell was recorded from in each slice, so the numbers of cells given are also the numbers of slices.

Internode measurements of dye-filled OLs
Coronal slices (225 mm thick) of P13-P15 mouse forebrain were prepared as described above (section on Electrophysiology). The slices were superfused with bicarbonate-buffered solution containing 124 mM NaCl, 2.5 mM KCl, 26 mM NaHCO 3 , 1 mM NaH 2 PO 4 , 2 CaCl 2 , 1 mM MgCl 2 , 10 mM glucose, pH 7.14, bubbled with a 95% O 2 /5% CO 2 mixture. Mature OLs in the corpus callosum were identified by their location and morphology as revealed by expression of YFP. The cells were whole-cell voltage clamped at -75 mV for 8 min to allow dye to enter the internodes fully. The pipettes had a series resistance of 8-35 MW and the internal solution was 145 mM CsCl, 10 mM HEPES, 4 mM MgATP, 2.5 mM NaCl, 1 mM EGTA and 0.5 mM Alexa Fluor 594, adjusted to pH 7.4 with CsOH. After dye filling, the slices were immediately fixed in 4% PFA for 1-4 hr, washed three times in PBS, mounted on slides under coverslips and imaged in a Zeiss LSM700 confocal microscope. Confocal stacks (between 17 and 80 images separated by 1 mm steps) were made of each OL and internode lengths were measured in three dimensions using the Fiji (ImageJ) simple neurite tracer.
In vivo EdU labelling 5'-ethynyl-2'-deoxyuridine (EdU, Invitrogen, 5 mg/ml in phosphate-buffered saline) was administered to P3 mice by a single 30 ml subcutaneous injection (in the scruff of the neck) and the mice were perfused 4 hr later. P7 mice were injected subcutaneously twice with 60 ml of EdU solution, 3 hr apart, and perfused 3 hr after the second injection. EdU was administered to P14 mice by intra-peritoneal injection (5 mg per kg body weight) 3 times at 2.5 hr intervals and the mice were perfused 2.5 hr after the final injection. EdU was provided to P70 mice via their drinking water (0.2 mg/ml EdU) with free access for 6 days. These regimens were chosen in an attempt to label a comparable proportion of OPs at each age, given that the cell cycle time is known to increase from less than a day to >10 days over this age range (Rivers et al., 2008;Psachoulia et al., 2009;Young et al., 2013).

Tissue processing
Mice were intracardially perfusion-fixed with 4% (w/v) paraformaldehyde (PFA; Sigma) in PBS. Brain tissue was dissected and post-fixed in 4% PFA overnight at 4˚C. Tissue was cryoprotected in diethylpyrocarbonate (DEPC)-treated 20% (w/v) sucrose (Sigma) for 24-48 hr in PBS. It was then frozen in Optimal Cutting Temperature (OCT) medium (Tissue Tek) on the surface of dry ice and stored at -80˚C until needed.

Image acquisition and analysis
Images were captured with either a light microscope (Zeiss Axioplan) with a Hamamatsu digital camera (10x or 20x objective) or with a laser scanning confocal microscope (Leica TCS SPE; 20x objective), using standard excitation and emission filters for Hoechst 33258 and Alexa fluorophores. Confocal Z-stacks were captured for each section with 1.5 mm steps.
Cells were counted in images of non-overlapping confocal sections of corpus callosum (3-4 fields covering the entire length between the dorsolateral corners of the lateral ventricles at bregma 0-1 mm) or motor cortex (3-4 fields covering all cortical layers in at least three sections per mouse; numbers of mice are indicated in Results). Images were displayed on-screen in Adobe Photoshop and cells were marked as they were counted, to avoid double-counting. All experiments were counted with the experimenter blinded as to genotype.

Electron microscopy
Mice were perfused at UCL with PBS followed by 2.5% (v/v) glutaraldehyde and 2% (w/v) paraformaldehyde in 0.1 M cacodylate buffer (pH 7.4). Brains were removed into the same fixative, kept at 4˚C for 48 hr, washed with 0.1 M cacodylate buffer then rinsed briefly with distilled water and shipped to Japan in PBS at ambient temperature. Sagitally-sectioned brains were processed for electron microscopy (EM). After immersion in a 1% (w/v) osmium tetroxide solution for 2 hr at 4˚C, the specimens were dehydrated through a graded alcohol series and embedded in Epon 812 (TAAB Laboratories, UK). The required area was trimmed and ultrathin sections were cut, collected on a platinum-coated glass slide, stained with uranyl acetate and lead citrate and imaged in a scanning electron microscope equipped with a back-scattered electron beam detector (Hitachi SU8010) at 1.5 kV accelerating voltage. EM images of sagittal sections of corpus callosum were taken at 2000x or 4000x magnification at P14, and 15000x at P70 starting 100 mm posterior to the anterior-most edge of the corpus callosum and spanning its entire dorso-ventral width, resulting in tiled images~210 mm x 160 mm for P14 and~25 mm x 170 mm for P70. Numbers of cross-sectional views of myelinated axons were counted in one such image per animal (~2000 axons counted in each image, n = 5 mice per genotype at P14 and >5000 axons counted in each image, n = 3 or four mice per genotype at P70). To estimate g-ratios, the circumference of the axon as well as the external circumference of the myelin sheath were measured using ImageJ, avoiding paranodal areas. The diameters of axon and myelin sheath were calculated from these circumferential measurements, assuming a circular profile, and the g-ratio was calculated. 50-100 axons were counted in each of 4 mice per genotype.

Statistics
Prism 6.0 (GraphPad) was used for statistical analysis. Data are presented as mean ± s.e.m. The Shapiro-Wilk test was used to test for normality of data distribution and p-values were determined using a two-tailed Student's t-test (when the distribution was consistent with normality) or non-parametric Mann-Whitney U test, as appropriate. One-way ANOVA with Bonferroni post-hoc test or Kruskal-Wallis with Dunn post hoc test were used for comparing multiple groups, as appropriate. When multiple t-tests or Mann-Whitney tests were performed, p-values were corrected for multiple comparison using a procedure equivalent to the Holm-Bonferroni method (for N comparisons, the most significant P value is multiplied by N, the second most significant by (N À 1), the third most significant by (N À 2), etc.; corrected p-values<0.05 are considered significant). Numbers of animals and cells studied are stated for each experiment in the text.
We analyzed the Caspase-3 immunolabelling data as follows. First, we calculated for control conditions the average fraction of Olig2 + cells that also expressed cleaved Caspase-3 (and hence were undergoing apoptosis), by averaging across control litters (these numbers are stated in the main text). Second, within each litter we normalized the data from experimental mice (double-or tripleknockouts) and their Gria3 null controls to the average of the controls in that same litter, setting the latter to 1. Then we averaged these normalized data across all litters to obtain the mean percentage increase in the fraction of Olig2 + cells that were apoptotic in experimental mice compared to control mice.
We did not perform a priori power analyses to calculate the sample size for the experiments, because the size of the effects expected was unknown before doing the experiments. We did however perform sample size calculation post hoc for experiments where group differences were close to significant, using the observed mean ± s.d. of the control group, an effect size of 20-25%, a power of 80% and p-value<0.05. With n = 3 for both groups, we should be able to detect a difference of~23% in CC1 + OLs in Gria2 -/-( Figure 4H). With n = 9 for control and n = 7 for the knockout group we should be able to detect a~20% reduction in CC1 + OLs in P21 Gria3 null 2 -/-( Figure 7B). With n = 14 for both groups, we should be able to detect a difference of~20% in P14 Pdgfra + OPs in Gria3 null 2 -/-( Figure 7F).