Oligodendrocyte Precursor Cells Modulate the Neuronal Network by Activity-Dependent Ectodomain Cleavage of Glial NG2

This study shows that the activity of neurons can trigger shedding of a protein, NG2, from the surface of oligodendrocyte precursor cells; this protein in turn modulates synaptic transmission, revealing a two-way conversation between neurons and glia.

These NG2+ OPC represent 5%-8% of total cells in the adult brain [6,7] and are ubiquitously spread throughout the grey and white matter: they are unique among glia in forming glutamatergic and GABAergic synapses with neurons [8,9]. These neuron-OPC synapses are present in all major brain areas including hippocampus, cerebellum, corpus callosum, and cortex [10][11][12][13][14]. Differentiation of OPC into oligodendrocytes is associated with a down-NG2 cleavage in isolated primary OPC. Inhibition of ADAM10 blocks both this activity-dependent release and also the constitutive cleavage of the ectodomain, identifying ADAM10 as the major protease cleaving NG2 under physiological conditions. To elucidate the functional role of NG2 cleavage and the OPCderived NG2 ectodomain, we studied the electrophysiology of pyramidal neurons in the somatosensory cortex in mice lacking the NG2 protein (NG2 2/2 mice, NG2-knockout). We observed a strong impairment of N-methyl-D-aspartate receptor (NMDAR)dependent long-term potentiation (LTP) and reduced Glut receptor currents at these neuronal synapses. Furthermore, we observed that inhibition of ADAM10 in control mice mimics the NG2 2/2 phenotype, reflected by a reduction in NMDARdependent LTP. Neuronal AMPAR-and NMDAR-mediated currents are reduced in NG2 2/2 mice, while AMPAR but not NMDAR reveal an altered subunit composition at neuronal postsynapses in NG2 2/2 mice: strikingly altered AMPAR characteristics could be restored to the wild-type (WT) pattern by addition of recombinant protein comprising the conserved Nterminal LNS domains of the NG2 ectodomain to the slices. These NG2 domains also increased levels of c-fos in Glut-stimulated slices of the somatosensory cortex, demonstrating that they affect activity-dependent gene expression in target neurons. NG2 2/2 mice reveal altered behavior in tests measuring sensorimotor function.
Taken together, our data demonstrate for the first time that OPC not only respond to, but are also able to modulate the neuronal network. They show that the NG2 protein itself is part of this regulatory mechanism, via activity-dependent cleavage and release of the NG2 ectodomain from OPC. Furthermore, we identify the two conserved LNS domains of the NG2 ectodomain as potent modulators of neuronal activity and gene expression.

ADAM10 and c-Secretase Dependent Cleavage of NG2
To analyze NG2 processing, we used a well-characterized cell line of murine OPC (Oli-neu) [38]. These cells, in contrast to primary OPC (pOPC), can be readily transfected and represent a good model system, since all cells express endogenous NG2 and are arrested in their differentiation. Overexpression of the asecretases ADAM10 (A10) and ADAM17 (A17, TACE) in OPC reduced by over 2-fold the levels of full-length (FL) NG2 (300 kD, cyto AB) (see Figure 1D) in post nuclear (PN) cell-lysates after 48 h ( Figure 1A). Culture of OPC for 6 h with the specific ADAM10 inhibitor GI254023X (GI) resulted in a 2.5-fold increase in levels of the FL protein ( Figure 1B).
To visualize size differences due to a-secretase-dependent shedding of NG2 more easily, a shortened recombinant form of NG2, NG2_del with a Flag tag in the extracellular domain was expressed (Flag_NG2_del) ( Figure 1C) [39]. Lysates analyzed for the largest protein fragments, revealed the FL (66 kD) and the asecretase cleavage product, an ectodomain of 54 kD ( Figure 1C), analogous to the 300 kD FL and 290 kD ectodomain of endogenous NG2.
Next we investigated the generation of a membrane bound CTF of NG2 and its cleavage by the c-secretase complex. In OPC expressing NG2_del ( Figure 1D), levels of the corresponding CTF fragment increased 4 times compared to endogenous CTF in control cells with no NG2_del expression ( Figure 1E). Additionally the endogenous NG2 CTF runs at exactly the same size as the CTF from NG2_del (,12 kD), again verifying an a-secretasemediated cleavage of NG2 and NG2_del at the same site and the creation of an 290 (or 54) kD ectodomain.

Author Summary
Although glial cells substantially outnumber neurons in the mammalian brain, much remains to be discovered regarding their functions. Among glial cells, oligodendrocyte precursors differentiate into oligodendrocytes, whose function is to enwrap nerves with myelin to ensure proper impulse conduction. However, oligodendrocyte precursors also comprise a stable population in all major regions of the adult brain, making up around 5% of the total number of neurons and glia. Synapses are classically formed between neurons. Nonetheless, oligodendrocyte precursors are unique among glial cells in that they receive direct synaptic input from different types of neurons; whether OPC also send signals to neurons is still unknown. Here we show a bidirectional communication between neurons and oligodendrocyte precursors: neuronal activity regulates the cleavage of a glial membrane protein and the release of an extracellular domain that in turn modulates synaptic transmission between neurons. Our data thus show that a particular subtype of glial cells, oligodendrocyte precursors, functionally integrate into the neuronal network and we link this bidirectional signaling to mouse behavior and disease.  Figure S1). (A) Reduction of NG2 FL protein levels after overexpression of ADAM10 or 17. (B) NG2 FL levels were increased if endogenous ADAM10 was inhibited by GI. (C) Immunoblot analysis of the Flag_NG2_del construct after expression in an OPC cell line using an antibody against the Flag tag within the ectodomain. Major protein bands were detected at 66 kD (FL protein) and 54 kD (largest ectodomain). (D) Schematic illustration of the murine NG2 FL protein and NG2_del expression constructs used. Cleavage sites, fragments, and epitopes recognized by antibodies (AB) are indicated. (E) Immunoblot analysis of endogenous compared to the NG2_del derived CTF. NG2 CTF levels are increased after expression of the NG2_del construct in an OPC cell line, both CTF run at the same molecular weight of 12 kD. (F) Immunoblot analysis of NG2_del_Myc expression in three different cell lines. Application of the c-secretase inhibitor DAPT highly increased Myc-tagged CTF in OPC and HEK cells. In MEF cells that are lacking presenilin 1 and 2, DAPT application was without effect. CTF levels of APP were used as a control to show the inhibition of c-secretase activity by DAPT. (G) Immunoblot analysis of endogenous NG2 CTF levels in an OPC cell line after inhibition of the c-secretase with DAPT. CTF levels were increased when DAPT was present and normalized against the FL protein levels. (In (A), n represents four experiments each performed in duplicate, the WB shows one experiment. In (B) and (G) each experiment consists of triplicates. Data represent the mean 6 SEM (paired Student's t test). doi:10.1371/journal.pbio.1001993.g001 To distinguish overexpressed from endogenous NG2 CTF, expression of NG2_del with an intracellular c-terminal Myc tag ( Figure 1D) was analyzed after expression in three different cell lines. c-secretase activity was inhibited by application of DAPT for 24 h, resulting in a further increase of Myc tagged NG2 CTF ( Figure 1F). In OPC, levels of the Myc tagged NG2 CTF increased 6-fold when the c-secretase was inhibited. In human embryonic kidney 293 (HEK) cells the CTF levels increased 12fold in the presence of DAPT. Incubation of the c-secretase inhibitor with a mouse embryo fibroblast (MEF) line, which lacks c-secretase activity owing to a genetic deletion of the presenilin 1 and 2 genes, did not affect levels of the NG2 CTF. Increased CTF levels of amyloid precursor protein (APP) were used as a control to demonstrate the inhibition of c-secretase activity by DAPT ( Figure 1F and 1G).
These results demonstrate c-secretase-mediated cleavage of the NG2_del derived CTF. Similarly, culture of untransfected OPC for 24 h in the presence of DAPT resulted in a 2-fold increase of CTF levels derived from the endogenous NG2 ( Figure 1G). As reported originally by the Barres group [40], we also observed that culture in the presence of DAPT caused an accelerated differentiation of OPC as indicated by a down-regulation of NG2 protein expression ( Figure 1G).

Sequential Cleavage Generates the NG2 CTF and ICD
To further analyze the sequential cleavage of NG2 by aand c-secretases we used a cell-free assay. Purified crude membranes (CMs) from OPC were incubated for 2 h with different combinations of specific protease inhibitors (GI, ADAM10 inhibitor; GW, A17/10 inhibitor and DAPT, c-secretase inhibitor) and compared to the control with standard protein inhibitors (PIs; 16 control). Membrane associated proteins (P100) were subsequently separated from soluble proteins (S100) generated within the 2 h of incubation by ultracentrifugation ( Figure 2A). A protein band at 300/290 kD was detectable in P100 representing full length NG2 protein and the 290 kD ectodomain (mc AB) (see Figure 2D). Additional major bands of truncated but high molecular proteins were detectable at 275 and 190 kD, which are likely to be additional ectodomains. In S100 all three theoretical ectodomain fragments were detected, although the levels were lower compared to forms of similar size in P100, suggesting that substantial amounts of the NG2 ectodomain bind to the membrane. The ectodomain origin of these cleavage forms was supported by their failure to react with the NG2 cyto AB ( Figure S6A). Application of a-secretase inhibitors (GI or GW) alone or in combination with the csecretase inhibitor (GI and DAPT), led to a significant reduction of the 290 kD ectodomain in S100 ( Figure 2B). Incubation with DAPT alone had no effect. The generation of the NG2 CTF (present in the P100 fraction only) (Figure 2A) was also influenced by the inhibitors: GI or GW as well as GI and DAPT reduced CTF levels while DAPT alone increased the CTF levels ( Figure 2B). These results demonstrate that the asecretase ADAM10 is responsible for the majority of the shedding in this paradigm, and that the ectodomain shedding is independent of c-secretase activity. Furthermore, a-secretase cleavage is mandatory for subsequent cleavage by the csecretase. Incubation of the membranes with inhibitors of the b-secretase had no effect, excluding the activity of the bsecretase in this process ( Figure S6B). The NG2 ICD was enriched in a crude-membrane fraction from Oli-neu (Figure 2C) and levels were decreased by c-secretase inhibition, suggesting a role in a membrane-associated complex ( Figure 2C and 2D).

Activity-Dependent Proteolytic Cleavage of NG2 Results in Increased Levels of Ectodomain Associated with the ECM
It has recently been reported that cleavage of the synaptic cell adhesion molecule neuroligin1, expressed by neurons, is under the control of electrical activity with modulatory effects on synaptic function [34,41]. To investigate if NG2 cleavage is similarly regulated, primary OPC from P9 mouse brain were isolated by magnetic beads coupled to monoclonal antibody recognizing NG2. After 1 day of culture, the cells were incubated with (I) picrotoxin, forskolin, and rolipram (PFR) inducing chemical LTP (cLTP) for 20 min; (II) 4-aminopyridine and bicuculline (4AP + BCC), a weak potassium and a GABA A channel blocker, known to depolarize neurons, for 20 min; (III) Glut for 10 min. In some cases the cells were pre-incubated with the NMDAR antagonist MK801 prior to incubation with Glut, or with the ADAM10 inhibitor GI for 15 min prior to stimulation with PFR or Glut. We observed a significantly reduced level of NG2 FL protein in PN cell lysates after incubation with 4AP + BCC, and after incubation with Glut or Glut+MK801 ( Figure 3A). PFR treatment did not alter NG2 FL levels. After pre-incubation with the ADAM10 inhibitor GI, neither Glut nor PFR altered NG2 FL levels. These results imply that ADAM10 cleavage-dependent reduction of NG2 FL protein can be increased by activity directly in primary OPC.
The NG2 ectodomain of 290 kD can be extracted with PBS from the ECM of the rodent brain (Figures 3B and S1) [31,32]. To analyze the effect of activity-dependent effects of the whole neural network on NG2 cleavage within OPC, we subjected acute hippocampal slices to cLTP (PFR incubation) or incubation with 4AP + BCC and analyzed levels of the released NG2 ectodomain.
Here we extracted ECM proteins by 30 min incubation with chondroitinase ABC, which removes exclusively sugar side chains from proteins, subsequent to the stimulation. The resulting samples contain only soluble, extracellular proteins as membrane proteins are not extracted in the absence of detergent, similar to the extraction of the ectodomain from tissue with PBS described above but with less mechanical stress. These samples thus contain only shedded soluble NG2 ectodomain. Strikingly, levels of the NG2 ectodomain were increased 2.5-fold after 15 min of PFR ( Figure 3C). Levels slightly decreased with ongoing recovery time and returned to starting levels after 180 min of recovery time ( Figure 3C). In order to identify the protease responsible for the activity-dependent ectodomain shedding, we incubated the slices in the presence of the metalloprotease inhibitor GM6001 or the ADAM10 inhibitor GI. Both compounds completely inhibited the NG2 ectodomain shedding induced by PFR ( Figure 3D). In the presence of the ADAM10 inhibitor GI, the ECM-associated levels of the NG2 ectodomain were reduced to below those of the control, demonstrating that basal levels of NG2 cleavage in OPC are ADAM10-dependent. Incubation of acute slices with 4AP + BCC, which depolarizes neurons and affects OPC as described above, increased NG2 ectodomain levels in the ECM in a comparable fashion to incubation with PFR ( Figure 3D).

Mice Lacking NG2 in OPC Exhibit Impaired NMDAR-Dependent LTP at Neuronal Synapses
Next we investigated a potential effect of the cleaved NG2 ectodomain on neuronal synaptic transmission. To address this question, we used transgenic mice with ablation of NG2 (NG2 2/2 ). In these mice OPC do not express NG2 (NG2 2/2 OPC) [42]. Whole-cell recordings were made from L2/3 pyramidal neurons in the somatosensory cortex; these are contacted by NG2 2/2 -OPC that express enhanced yellow fluorescent protein (EYFP) (green in the Figure 4A). We analyzed NMDAR-dependent LTP as a typical form of synaptic long-term plasticity.
Cortical pyramidal neurons were current-clamped at a membrane-potential of 280 mV, and excitatory postsynaptic potentials (EPSPs) were evoked by electrical stimulation in layer IV that stimulated ascending afferent fibers projecting onto these pyramidal neurons. Baseline EPSPs were initially recorded for 10 min before applying an LTP inducing stimulation protocol that consisted of presynaptic high-frequency theta-burst stimulus (TBS) paired with postsynaptic depolarization. This protocol is known to induce high presynaptic glutamate release as well as an enhanced opening of postsynaptic glutamate receptors due to the membrane depolarization [43].
This   . Activity-dependent NG2 cleavage and release of the ectodomain into the ECM. (A) Primary OPC were isolated from total mouse brain of postnatal day 9 (P9) by magnetic cell sorting (MACS). After 1 day of culture (90%-95% OPC), the cells were incubated with PFR or 4AP + BCC for 20 min or with Glut for 10 min. To test the involvement of NMDARs, in some experiments cells were pre-incubated for 15 min with the NMDAR inhibitor MK801 (Glut+MK801) and to confirm shedding by ADAM10 with its inhibitor GI (PFR + GI, Glut + GI). Cells were harvested and PN cell-lysates were obtained by differential centrifugation prior to subjecting to Western blot. Western blot quantification (bar graphs) revealed a reduction in the NG2 FL protein for 4AP + BCC, Glut, and Glut+MK801 stimulation, no change was observed if the cells were stimulated with PFR or if the cells were pre-incubated with the ADAM10 inhibitor GI (unpaired Student's t test). (B) Soluble (saline) and membrane fractions (triton) were obtained from forebrains of adult WT and NG2 2/2 mice. Only a small amount of the NG2 290 kD ectodomain is soluble in the saline-fraction in the WT (see Figure  S1) as reported before. (C) Western blot of ECM extracts of acute hippocampal slices from P70 rats by digestion of the glycosaminoglycan chains by chondroitinase ABC without detergent. An increase of the NG2 ectodomain was detected after chemical LTP (PFR) stimulation for 15 min (159). The NG2 ectodomain levels within the ECM return to starting levels after 180 min of recovery time. NG2 ectodomain levels were normalized against total protein from a matching coomassie gel ( Figure S5). (D) NG2 ectodomain levels increase 1.5-fold 15 min after stimulation with PFR and 40 min recovery time detected by Western blot from ECM extracts of acute hippocampal slices. Inhibition of metalloproteases (GM6001), or ADAM10 (GI), prevented the increase of ECM-associated NG2 ectodomain. Incubation with GI reduced levels of ECM-associated NG2 ectodomain to below the control value. Treatment of the slices with 4AP + BCC for 10 min with 20 min recovery time showed the same increase in levels of ECM-associated NG2 ectodomain as seen with the PFR treatment (one way ANNOVA with Dunnett's multiple comparison test). doi:10.1371/journal.pbio.1001993.g003 Inhibition of ADAM10 Activity Results in a Similar Impairment of NMDAR-Dependent LTP at Neuronal Synapses to That Seen in NG2 2/2 Mice ADAM10 is the major metalloprotease responsible for NG2 cleavage, whose activity is promoted by stimulation of the neural network. To examine the physiological relevance of NG2 ectodomain processing by ADAM10 for neuronal activity, we used the metalloprotease blocker GM6001 and the ADAM10specific inhibitor GI, which we had shown prevents activitydependent release of the NG2 ectodomain ( Figure 3A and 3D) and repeated LTP recordings in cortical slices from control (WT) and NG2 2/2 mice. Following a pre-incubation period of 1 h, GM6001 abolished the induction of NMDAR-dependent LTP in controls and mimicked a NG2 2/2 -like phenotype, as reflected by a small EPSP increase of 9.8%63.3% after 20-30 min post LTP induction by TBS ( Figure 4C). GM6001 did not alter the level of EPSPs in neurons of NG2 2/2 mice (20-30 min post-TBS: 6%62.3%) ( Figure 4C). Similar results were obtained when the experiments were repeated substituting GI for GM6001, which also prevented LTP induction in neurons of control mice (5.6%61.3%), but did not affect neurons of NG2 2/2 mice (2.5%62.0%) ( Figure 4D). These data suggest that the ectodomain of NG2 is released by the action of the metalloprotease ADAM10 in the vicinity of neuronal synapses and can modulate glutamatergic transmission between cortical pyramidal neurons. Subsequently we examined intrinsic membrane properties of L2/ 3 pyramidal neurons and found no differences in membrane capacitance and input resistance after NG2 deletion ( Figure S2), thereby most likely excluding influences of altered ion channel conductance on the cellular excitability in NG2 2/2 mice. Interestingly pyramidal neurons of NG2 2/2 mice displayed a slightly hyperpolarized resting membrane potential (DV m = 23.360.1 mV) ( Figure S2).

Decreased Neuronal Excitability Results from Altered Composition of Neuronal Glutamate Receptors
We examined basal glutamatergic neurotransmission at L4to-L2/3 synapses in NG2 2/2 mice and recorded spontaneous excitatory postsynaptic currents (sEPSCs). These signals were exclusively mediated by glutamatergic AMPA receptors, since we blocked the activation of GABA-and NMDARs by bath application of PTX (50 mM) and D-AP5 (25 mM), respectively. We found no differences in NG2 2/2 mice compared to WT, the mean frequency and peak amplitude of sEPSCs were not altered in L2/3 pyramidal neurons ( Figure S3). Next we assessed paired pulse facilitation of synaptic currents to detect potential changes of presynaptic glutamate release. The paired pulse ratio, which was calculated from two consecutive evoked EPSCs, was similar in NG2 2/2 and control mice ( Figure S4). Thus, together with unchanged sEPSCs, these data suggest that presynaptic glutamate release is unaffected in neurons of NG2 2/2 mice. NMDAR-dependent LTP is often associated with a reduction in the number or changes in the subunit composition of postsynaptic glutamate receptors [42,43]. To examine the functional properties of glutamatergic AMPARs and NMDARs at excitatory synapses, we recorded AMPAR-and NMDARcurrents within the same neuron. AMPAR-mediated currents were recorded at 280 mV, a potential where NMDARs are rarely open, followed by detection of NMDAR currents at + 60 mV in the presence of DNQX (20 mM), which blocks AMPAR activity. Strikingly, both types of receptor currents were reduced in their amplitudes in NG2 2/2 mice, with AMPAR currents being impaired by 24.5%62% compared to controls and the NMDAR current component being decreased by 43.7%62.2% ( Figure 5A and 5B). Accordingly, we observed a shift in the ratio of NMDARs to AMPARs (NG2 2/2 : 0.860.07 versus control: 1.060.1) ( Figure 5C). However, despite the drop in the amplitude of NMDAR currents, the kinetic properties ( Figure 5F and 5G) and the current-voltage relation of NMDAR currents ( Figure 5H) remained unchanged with NG2 deletion, suggesting no alteration of subunit composition. In contrast, one kinetic parameter of AMPAR currents ( Figure 5D and 5E To further investigate the functional properties of AMPARs at L4-to-L2/3 synapses of pyramidal neurons, we assessed a typical current-voltage (I-V) relationship of AMPAR-mediated currents. Strikingly, we found a strongly decreased slope, also called inwardrectification, of the resulting I-V curve at positive potentials in the NG2 2/2 mice, while WT controls displayed a nearly linear I-V relationship ( Figure 6A). The rectification index (RI) was significantly smaller in pyramidal neurons of NG2 2/2 mice (0.460.03; range: 0.2-0.8) compared to controls (1.060.07; range: 0.6-1.6) ( Figure 6B). These results suggest the presence of Ca 2+ -permeable (CP)-AMPARs at L4-to-L2/3 synapses of NG2 2/2 mice, which was further confirmed by repeating the recordings of AMPAR currents in the presence of NASPM (250 mM), a selective inhibitor of CP-AMPAR channels [44] at a holding potential of 280 mV. Bath application of NASPM reduced the initial peak amplitude of AMPAR currents by 26.6%68.0% in the NG2 2/2 , while AMPAR currents in controls remained unaffected (4.8%62.1%; n = 6) ( Figure 6C and 6D). These data suggest that both CP and Ca 2+ -impermeable AMPARs are present in the postsynaptic membrane of L2/3 pyramidal neurons from NG2 2/2 mice. The nearly linear I-V relation of AMPAR currents in the presence of NASPM further substantiates this assumption (Figure 6C and 6D). Moreover, subsequent bath application of GYKI53655 (25 mM), an antagonist of all AMPAR, blocked effectively all current responses in both NG2 2/2 (88.8%61.8%) and control groups (91.9%61.7%) ( Figure 6D). These data confirm that the remaining current components in the presence of NASPM were mediated by AMPARs and not by other glutamate receptors like kainate receptors.
In order to investigate a potential direct influence of the NG2 ectodomain, in particular the LNS domains, on the observed AMPAR I-V relation in NG2 2/2 mice, we pre-incubated cortical slices for 1 h with purified recombinant protein comprising alkaline phosphatase (AP)-tagged LNS domains of the NG2 ectodomain (LNS-AP). Strikingly, application of the NG2 LNS domains restored the AMPAR currents in NG2 2/2 mice to WTlevels, as reflected by the presence of a normal linear I-V relationship ( Figure 6E and 6F). In contrast a pre-incubation with the AP tag alone (control) did not alter the I-V relation of the AMPAR currents.
Finally we also tested the effects of the NG2-LNS domains on neuronal activity in an independent approach. The expression level of c-fos, an immediate early gene and a marker for neuronal activity [45] was analyzed in cortical slices of NG2 2/2 tissue after incubation of the slices with 1 mM Glut for 10 min to activate the neuronal network. Pre-incubation of slices with the LNS protein domains resulted in more c-fos+ neurons compared to slices preincubated with AP alone (Figure 7).

NG2 2/2 Mice Show Impaired Sensorimotor Gating
To investigate possible effects of the altered electrophysiological properties of cortical neurons in NG2 2/2 mice on mouse behavior, we measured pre-pulse inhibition (PPI) of the acoustic startle response in these animals. In this assay, the startle response   subsequently stimulated with glutamate for 10 min. Immunofluorescence staining was performed for the activity-dependent early gene c-fos (in red) and the neuronal marker NeuN+ (green). Incubation with the NG2 LNS domains resulted in more c-fos+ neurons compared to slices incubated with the control AP protein (C). c-fos+ cell numbers were related to NeuN+ cell numbers (c-fos+/NeuN+ cell numbers) from each double stained slice. Cell numbers of representative cortical areas as shown in (A) and (B) were counted from seven different animals (n = 7) for each incubation condition. The incubation conditions (LNS-AP and AP) were compared from the same individual mouse. c-fos/NeuN numbers were compared between the two experimental conditions (paired Student's t test). doi:10.1371/journal.pbio.1001993.g007 of the animals is measured after initial presentation of a weaker sensory stimulus (the pre-pulse). Normally, the response is diminished after a pre-pulse and it is thought to reveal sensorimotor gating, i.e., the ability to filter sensory input [46]. We found a significant reduction of PPI in NG2 2/2 animals in comparison to WT littermates ( Figure 8A), where the NG2 2/2 mice still reacted strongly to the stimulus in spite of the pre-pulse. As a test for sensory input, the olfactory habituation/dishabituation test was performed. Olfactory sensations are integrated in the olfactory cortex and are a part of the somatosensory cortex ( Figure 8B). The NG2 2/2 mice spent less time sniffing the two presented odors compared to WT animals. Notably, motor cortex-associated behavior assessed by the Rotarod or CatWalk tests ( Figure 8C and 8D) was unchanged in NG2 2/2 mice. Learning and memory was tested by contextual fear conditioning (Figure 8E) and the Morris Water Maze test ( Figure 8F). In both tests, no differences were observed between genotypes.

aand c-Secretase-Mediated Cleavage of NG2
We show here that NG2 can be cleaved by aand c-secretases in a sequential manner similar to other type-1 membrane proteins such as Notch, APP, L1, N-cadherin, and neuroligin1 [33,[47][48][49][50]. We identified ADAM10 as the major constitutively acting asecretase, similar to the cleavage of Notch [51]. ADAM10 cleavage of proteins generally results in two fragments, a released ectodomain and a membrane bound CTF. In the case of NG2, we found an ectodomain of ,290 kD that is ,10 kD smaller than the FL protein and a CTF of 12 kD. These findings match the reported NG2 ectodomain fragment of 290 kD described previously [31]. So far, CTFs or ICDs have not been reported for NG2 in contrast to the proteins described above [52,53]. The NG2 CTF fragments were detected in membrane fractions and were more abundant when the c-secretase was inhibited. These results suggest a functional role for the CTF fragment and demonstrate constitutive processing by the c-secretase leading to the creation of an NG2 ICD, which was often found associated with the cellmembrane similar to the ICD of APP [54]. FL NG2 protein is expressed by OPC within the CNS and the ectodomain is likely to be contributing to the ECM [31,55]. Interaction with collagen V&VI and b1 integrin has been shown, compatible with this concept [56,57]. Interestingly, we found the NG2 ectodomain to be highly adhesive for OPC membranes (Figure 2A) and membrane fractions from total forebrain ( Figure 3B). We also observed as reported previously, that a fraction of the 290 kD ectodomain is soluble in aqueous buffers.
Our results using isolated OPC show that NG2 is processed on OPC by endogenous ADAM10 and c-secretase in a constitutive manner and confirm that all cleavage fragments identified originate from OPC membranes.

Activity-Dependent Cleavage Results in Altered ECM Levels of the NG2 Ectodomain
We observed that after cleavage, part of the NG2 ectodomain is deposited in the ECM and can be extracted with chondroitinase ABC, which cleaves glycosaminogycan chains thus freeing proteoglycans from their potential association with the ECM, allowing their extraction in a soluble fraction without detergent. In our experiments ECM proteins are extracted from acute slices after incubation with activity-modulating chemicals. The amount of detergent-free extractable ectodomain rapidly increases after incubation of the slice (and thus the network), with PFR, which is known to induce cLTP by increasing cAMP levels in neurons [58]. Interestingly, ectodomain levels in this fraction returned to starting levels after 180 min, demonstrating that subsequent to shedding a fast removal of the ectodomain fragment ensues. Removal of the released ectodomain from the ECM fraction may be due to binding to cell-surface receptors and subsequent internalization or further proteolytic processing. ADAM10 appears to be the protease responsible for both the PFR-induced and constitutive cleavage in vivo, as ADAM10 inhibition reduced NG2 ectodomain cleavage to below control levels in the presence of PFR. Thus ADAM10 can be considered as the constitutively active protease releasing NG2 ectodomain and CTF, and neuronal activity further increases this cleavage. The activity-dependence of NG2 ectodomain cleavage can furthermore be shown by stimulation of slices with 4AP + BCC, which is reported to inhibit repolarization by blocking voltage-dependent potassium channels [59], and also increases NG2 ectodomain levels in the ECM.
Since in these treatments of acute slices we are potentially affecting all neural cell types, we also repeated the experiments with isolated primary OPC. These were subjected to the same activity-inducing chemical treatments as the slices and in addition we tested the effect of incubation with glutamate. We observed that 4AP + BCC but not the PFR treatment stimulated NG2 cleavage on isolated OPC. Incubation of OPC with the neurotransmitter glutamate additionally increased ADAM10dependent cleavage of NG2 independent of NMDAR activity.
The combination of the effects we observed in an OPC cell line, primary OPC and the results in acute slices, suggest the following scenario. OPC have high densities of AMPAR at their synapses, which enable direct synaptic response to glutamate in vivo [8,10], thus presenting a rapid signaling mechanism that potentiates ADAM10-dependent NG2 ectodomain cleavage in addition to the constitutive ADAM10-dependent cleavage. With respect to the mechanism of activity-dependent ADAM10 cleavage, we show that induction of chemical LTP (with PFR), a known stimulator of neuronal activity [58], is likely to function via the neuronal network as it did not increase NG2 ectodomain cleavage in isolated OPC. OPC exhibit voltage-dependent potassium channels [60], which could explain why 4AP + BCC incubation of primary OPC also caused an increased release of NG2 ectodomain, as blocking both voltage-dependent potassium channels and GABA A receptors may lead to depolarization of the OPC, which is required for activity-dependent NG2 shedding.
In neurons, ADAM10 is located at excitatory postsynapses [61] and we have confirmed that OPC also express ADAM10. NG2 binds the PDZ domain protein GRIP in OPC, which itself binds to AMPAR forming a tripartite complex in OPC [30]. Thus an activity-dependent release of the NG2 ectodomain is likely to occur at neuron-OPC synapses resulting in increased levels of NG2 ectodomain in the ECM. These findings are similar to activity-dependent cleavage by ADAM10 of neuronal adhesion molecules such as neuroligin-1 and N-cadherin [34,62].

Modulation of the Neuronal Network by NG2 and ADAM10 Activity: Alterations in NMDAR-Dependent LTP
As an effect of NG2 ablation in OPC (NG2 2/2 OPC), we found a diminished NMDAR-dependent LTP at neuron-to-neuron synapses of pyramidal neurons in the somatosensory cortex. We observed a similar lack of neuronal LTP in neurons of control mice 45 min after blocking NG2 cleavage by ADAM10 with metalloprotease or specific ADAM10 inhibitors, while the overall small potentiation in pyramidal neurons of NG2 2/2 mice remained unaffected by these inhibitors. These results demonstrate a major contribution of activity-dependent ADAM10 cleavage to the observed LTP changes. The neuronal cell adhesion molecule (CAM) neuroligin has recently been shown to be processed by ADAM10 in an activitydependent manner at neuronal synapses, leading to alterations of the spine structure [34] and synaptic function. Additionally, other proteoglycans that are associated with the ECM, such as brevican [63] and tenascin R and C [64,65], or the structural integrity of the ECM itself [66] influence long or short term potentiation at neuronal synapses. It is feasible that the released NG2 ectodomain directly interacts with a neuronal receptor, leading to modulation of glutamatergic receptor composition and glutamergic transmission at neuronal synapses. These findings are in accordance with previous results showing that processing of synaptic cell adhesion molecules such as neuroligin1 by ADAM10 correlates with efficient synaptic transmission [67,68]. These results suggest that ADAM10-mediated activity-dependent cleavage underlies the LTP alterations that we observe and also strongly argue that ADAM10 dependent ectodomain shedding of NG2 is a major contributory factor.

Altered Composition of AMAPR: A Response to Decreased NMDAR Density in NG2 2/2 Neurons
To gain insight into the synaptic mechanism leading to the LTP phenotype in the somatosensory cortex of NG2 2/2 mice, we analyzed potential pre-and postsynaptic contributions. Basal synaptic transmission mediated by AMPARs was not altered at L4-to-L2/3 connections in NG2 2/2 mice. This result, together with an unchanged short-term plasticity as addressed by pairedpulse facilitation of excitatory synaptic transmission, makes impairments of glutamate release rather unlikely. Nevertheless, a lack of LTP induction is often associated with reduced density or changed subunit composition of glutamatergic receptors at these neuronal synapses [69,70]. In response to a constant stimulus strength, both NMDAR-and AMPAR currents were significantly reduced in L2/3 pyramidal neurons from NG2 2/2 mice compared to WT controls. However, alterations of NMDAR subunit composition could be excluded as the respective kinetics and the current voltage relationship remained unaffected by NG2 deficiency, hence suggesting a down-regulation of NMDARs per se.
In contrast, AMPAR currents revealed remarkable differences between neurons of NG2 2/2 and WT mice, including shorter decay time constants and an inward rectification of their I-V curve at positive holding potentials. The latter is characteristic for CP AMPARs lacking the GluR2 subunit, where it is caused by a polyamine-dependent block [71,72]. Application of two selective antagonists, NASPM (GluR2 lacking AMPAR) and GYKI53655 (total AMPAR), confirmed the functional expression of these ion channels at excitatory synaptic inputs onto L2/3 pyramidal neurons of NG2 2/2 mice. The presence of CP-AMPAR can explain the observed slightly hyperpolarized resting potential in the NG2 2/2 neurons (Figure S2), since a more hyperpolarized resting potential has been described previously in neurons expressing CP-AMPAR [71]. CP-AMPARs are normally downregulated during development in cortical pyramidal neurons, but retained in certain types of GABAergic interneurons in the adult brain [73,74]. Here, similar to our studies, the insertion of CP-AMPARs is negatively correlated with the functional expression of CP-NMDARs in the postsynaptic membrane [75,76], which influences the increase of intracellular Ca 2+ and hence affects information processing within the somatosensory cortex.
Our results thus show that shedding of the NG2 ectodomain regulated by neuronal activity has a profound impact on neuronal glutamate receptor currents and long-term synaptic plasticity in the somatosensory cortex.
The LNS Domains of the NG2 Ectodomain Restore the Impaired AMPAR Currents in NG2 2/2 Neurons Strikingly, the observed inward rectification in the I-V relationship at positive holding potentials in NG2 2/2 mice could be restored to the WT phenotype by application of the N-terminal LNS domains of the NG2 ectodomain. In support of this, preincubation of cortical slices with the NG2 LNS domains prior to brief stimulation with glutamate increased activity-dependent neuronal c-fos expression in somatosensory neurons. These results demonstrate a direct influence of these specific domains [27] on the functional properties of AMPARs discussed above. Pyramidal neurons of NG2 2/2 mice have more functional CP-AMPAR at postsynaptic membranes, which supports the concept of a stabilizing effect of the NG2-LNS domains on AMPAR containing the GluR2 subunit, which are impermeable to calcium. The LNS domains have been postulated by in silico experiments to belong to the same superfamily as pentraxins [77,78]. Neuronal pentraxins such as Nptx2 (Narp) have been shown to bind to and stabilize GluR2 subunit-containing AMPAR at the neuronal plasma membrane [79]. Furthermore, application of the single LNS domain of b-neurexin 1 or b-neurexin 3, respectively, is sufficient to restore heterologous synapse formation in knock-out animals for each of these neurexins [29]. An alternative splicing variant of a-neurexin 3, resulting in an altered structure of one specific LNS domain, has recently been shown to impair postsynaptic AMPAR currents and LTP when introduced genetically [28].
The rescue in our experiments by recombinant NG2 LNS domains shows that these can diffuse within the slice to reach the sites of action on neurons. It is thus likely that endogenous NG2 ectodomain (containing these LNS domains) exhibits a limited diffusion within the ECM in vivo. The major 290 kD form of the soluble (in detergent-free tissue extract) NG2 ectodomain found by us and others is most likely mediating the endogenous NG2 signaling effects on neurons.
Sensorimotor-Gating Behavior Is Impaired in NG2 2/2 Mice We observed altered NMDAR and AMPAR properties of L2/3 neurons of the somatosensory system in NG2 2/2 mice. Functional integrity of the motorcortex (which is adjacent to the somatosensory cortex) was tested by two motor tests, Rotarod and CatWalk, and revealed no differences between the genotypes. A test assessing olfaction showed a reduced response to defined odors in the NG2 2/2 mice, which could reflect a changed integration of sensory input in the olfactory cortex, as part of the somatosensory cortex. NG2 2/2 mice also show impaired PPI in behavioral assays. Reduced PPI is associated with human diseases, such as schizophrenia, Huntington disease, Tourette syndrome, and obsessive-compulsive disorder [46]-diseases associated with an impaired integration of sensory data. Furthermore, receptors for three different neurotransmitters have been shown by pharmacological studies to participate in a reduction of PPI in rodents: stimulation of dopamine or serotonin receptors, or inhibition of NMDAR [46,80,81]. Two widely used learning and memory tasks (contextual fear conditioning and Morris Water Maze) showed no differences between NG2 2/2 and WT control animals. These results suggest that the observed NG2-dependent alterations of the neuronal system affect in particular the neuronal networks of the somatosensory cortex (as tested by PPI), rather than those of the hippocampus-amygdala circuits (as tested by contextual fear conditioning and Morris Water Maze). This issue has to be addressed by further behavioral and molecular studies, to reveal if neuronal network activity is most strongly affected in the somatosensory cortex by the reported synaptic phenotype in NG2 2/2 mice. We propose that synaptic signaling of the NG2 ectodomain influencing AMPAR/NMDR and LTP is likely to occur throughout the brain, as NG2 shedding on OPC seems to be a general phenomenon. On a molecular level, this could mean that in other areas of the brain such as the hippocampus, other regulators modulate LTP more strongly than the NG2 ectodomain.
Taken together, our data provide compelling evidence that OPCs in the adult brain are functionally integrated in the neuronal network and not only respond to but modulate neuronal activity via newly identified mechanisms. We show that NG2 is cleaved by ADAM10 and the c-secretase on OPC. Furthermore, cleavage by ADAM10 releases the NG2 ectodomain into the ECM in an activity-dependent mechanism, where it influences postsynaptic glutamate receptor currents and neuronal LTP in the somatosensory cortex. We identify the LNS domains of the NG2 ectodomain as potent modulators of neuronal synapses, as they rapidly restore the altered AMPAR kinetics observed in the NG2 2/2 mice to the WT pattern. These changes in sensorimotor processing are likely to underlie the behavioral alterations exhibited by NG2 2/2 mice. Our results demonstrate a bidirectional cross-talk between OPC and the surrounding neuronal network, and demonstrate a novel physiological role for NG2-expressing OPC in regulating information processing. They also suggest a possible approach for modulating neuronal network function in disease.

Cell Lysis and Fragmentation
Cells were lysed in Triton X-100, after removal of nuclei by centrifugation, supernatants were defined as PN lysates. Alternatively cells were homogenized without detergent with a Potter S teflon pestle, nuclei were removed by centrifugation, and after high-speed centrifugation fractions were defined as cytoplasmic, or crude membrane (CM) pellet fractions. For the cell free protease assays, CMs were obtained and incubated with different combinations of specific protease inhibitors (GI254023X [GI], GW280264X [GW]) [83]; DMSO and DAPT (Sigma), b-secretase inhibitor II (Calbiochem), Roche complete (PI) (Roche). After ultracentrifugation, supernatant (S100) and pellet (P100) were taken for Western blot.

SDS-PAGE and Western Blotting
Equal volumes or amounts of total protein were separated on 4%-12% NuPage BisTris gradient gels (Invitrogen). Proteins were blotted on a PVDF membrane (Millipore). HRP-conjugated secondary antibodies were used with hyper films (GE) for detection. Protein levels were normalized against those of GAPDH.

NG2 2/2 Mice
Homozygous NG2-EYFP mice, (NG2 2/2 ), lack NG2 protein expression and were previously described [42]. All experiments were carried out in strict accordance with protocols approved by local Animal Care and Use Committee of Johannes Gutenberg University of Mainz.

LNS-AP and AP Purification
The two N-terminal LNS domains of NG2 (amino acids  were cloned in the pAPtag-5 (GenHunter) plasmid with a Cterminal AP, a Myc and hexaHis Tag; the construct was named LNS-AP. The original plasmid coding only for AP with Myc hexaHis Tag was used as control (AP). Proteins were purified from the cell and serum free medium of HEK over a nickel column (Sigma). Proteins were pre-incubated for 1 h (10 mg/ml) with acute slices for electrophysiology or immunocytochemistry (c-fos).

Activity-Dependent Neuronal Gene-Expression
For investigation of c-fos expression in acute slices of the somatosensory cortex, slices were stimulated with glutamate (1 mM) for 10 min after pre-incubation (1 h) with purified LNS-AP or AP protein (10 mg/ml each). Experimental conditions (LNS-AP and AP) were compared from the same individual mouse and of a total of seven different mice were analyzed.

Behavioral Tests
All behavioral tests were performed with the same cohort of mice between 5-8 weeks of age. Only male littermates were used (11 NG2 2/2 and 10 WT animals). The tests performed were: Rota-Rod, CatWalk, test for olfaction, PPI, fear conditioning, Morris Water Maze.

Statistics
Statistical analysis was done by using SPSS or MS Excel. Data were tested for normal distribution with non-parametric onesample Kolmogorov-Smirnov test. All quantitative data are expressed as mean 6 standard error of the mean (SEM). Significance was classified as follows: *, p#0.05; **, p,0.01; ***, p,0.001; n.s. p.0.05. Classified by Student's t test or ANOVA with Dunnett's Multiple Comparison test.
For detailed description of methods refer to Text S1.

Supporting Information
Figure S1 NG2 ectodomain and LNS constructs. (A) Immunoblot of saline (S1) and Triton X-100 (T1, membrane bound/associated proteins) soluble proteins from forebrains of adult WT and NG2 2/2 mice. The T1 fraction contains all membrane bound/associated forms of NG2 including the FL protein (300 kD) and a small fraction of the ectodomain. In the S1 fraction the 290 kD ectodomain was detected as a major fragment, in the T1 fraction additionally a 275 and a 190 kD fragment and the FL protein were detectable. Full length NG2 (dashed rectangle) is exclusively present in the T1 fraction, detected by the NG2 cyto AB, which shows no band at 300 kD in the NG2 2/2 . an epitope in the NG2 extracellular domain (NG2 mc AB), detected bands of around 300 kD on a Western blot of the P100 and S100 fraction. In contrast, an antibody recognizing the ICD (NG2 cyto AB) only detected one major band in P100 but did not react with proteins of around 300 kD in the S100 fraction. (The lane marker 2 contains hand-drawn lines on the film marking the positions of the MW markers). (B) An inhibitor of the b-secretase (b-secretase inhibitor II, 25 mM) had no effect on levels of the protein band of around 300 kD in the S100 fraction. In contrast, a-secretase inhibitors (GI, GW) caused a decrease in levels of the protein band of around 300 kD in comparison to the control (PI 16). (TIF) Data S1 Rawdata. (XLSX) Text S1 Supplemental materials and methods. (DOC)