Gray Matter NG2 Cells Display Multiple Ca2+-Signaling Pathways and Highly Motile Processes

NG2 cells, the fourth type of glia in the mammalian CNS, receive synaptic input from neurons. The function of this innervation is unknown yet. Postsynaptic changes in intracellular Ca2+-concentration ([Ca2+]i) might be a possible consequence. We employed transgenic mice with fluorescently labeled NG2 cells to address this issue. To identify Ca2+-signaling pathways we combined patch-clamp recordings, Ca2+-imaging, mRNA-transcript analysis and focal pressure-application of various substances to identified NG2-cells in acute hippocampal slices. We show that activation of voltage-gated Ca2+-channels, Ca2+-permeable AMPA-receptors, and group I metabotropic glutamate-receptors provoke [Ca2+]i-elevations in NG2 cells. The Ca2+-influx is amplified by Ca2+-induced Ca2+-release. Minimal electrical stimulation of presynaptic neurons caused postsynaptic currents but no somatic [Ca2+]i elevations, suggesting that [Ca2+]i elevations in NG2 cells might be restricted to their processes. Local Ca2+-signaling might provoke transmitter release or changes in cell motility. To identify structural prerequisites for such a scenario, we used electron microscopy, immunostaining, mRNA-transcript analysis, and time lapse imaging. We found that NG2 cells form symmetric and asymmetric synapses with presynaptic neurons and show immunoreactivity for vesicular glutamate transporter 1. The processes are actin-based, contain ezrin but not glial filaments, microtubules or endoplasmic reticulum. Furthermore, we demonstrate that NG2 cell processes in situ are highly motile. Our findings demonstrate that gray matter NG2 cells are endowed with the cellular machinery for two-way communication with neighboring cells.


Introduction
In addition to astrocytes, oligodendrocytes, and microglia, NG2 cells are now recognized as a fourth glial cell type in the CNS [1,2]. NG2 cells display long narrow processes and lack gap junction coupling. Fate mapping analysis has demonstrated that in white matter the majority of NG2 cells are oligodendrocyte precursors (OPCs). In contrast, gray matter NG2 glia only rarely give rise to oligodendrocytes or astrocytes but keep their phenotype throughout postnatal life [3], but see also [4,5].
NG2 cells are unique among glial cells in receiving synaptic input (reviewed by [2,6]), but the physiological impact of this innervation is unknown. Specifically, it remains unclear whether pre-synaptic transmitter release generates Ca 2+ -elevations in post-synaptic NG2 cells, which might evoke cellular motility or release of neuroactive substances. This ignorance is quite astonishing in view of the increasing knowledge of glia-mediated modulation of CNS signaling, such as astrocyte-neuron interactions which gave rise to the tripartite synapse concept [7][8][9]. Moreover, it is known for more than a decade that 'complex' glial cells [10], which display properties similar to NG2 cells, express Ca 2+ -permeable AMPA receptors [11][12][13] and voltagegated Ca 2+ -channels (Ca v s) [14]. In cultured presumed glial progenitor cells, Ca v s are activated by the depolarizing action of GABA [15]. However, despite these previous reports the presence of Ca v s in NG2 glia is still disputed. Instead, a role for the Na + -Ca 2+ exchanger (NCX) in NG2 cell Ca 2+ -signaling has recently been proposed [16,17].
There are different terms in the literature describing NG2-like cells in acute preparations of wild type or different transgenic mouse lines: complex glial cells (e.g. [10]); GluR cells (e.g. [18]), OPCs (e.g. [19]), synantocytes [20], and polydendrocytes (e.g. [21]). It is currently unknown to which degree these cellular populations overlap [6]. In the present study, we employed transgenic mice with fluorescence labeling of NG2 and GluR cells to study their process structure and Ca 2+ -signaling mechanisms. Morphological, molecular and functional analyses revealed that NG2 cells (i) generate transient elevations of the intracellular Ca 2+ -concentration ([Ca 2+ ] i ) upon different types of stimulation and (ii) display in situ highly motile actin-based processes.

Cell identification and basic electrophysiological properties
Cell identification in the hippocampus was based on EYFP or EGFP fluorescence, morphology, and physiological criteria as reported previously [18,22,23]. Cells used for Ca 2+ -imaging (n = 836; 691 of them genotyped) were EYFP positive, had an input resistance of 1936157 MV, a resting membrane potential of 27766 mV, and a membrane capacity of 3368 pF (K + -based pipette solution). All cells tested (n = 23) received glutamatergic and/or GABAergic synaptic input (not shown). EYFP positive cells from homozygous (n = 351) and heterozygous (n = 340) mice did not differ with respect to the above membrane parameters, expression of Ca v channel transcripts, and Ca 2+ -responsiveness upon somatic depolarization or high frequency stimulation of presynaptic fibers (see below for details). Therefore, data were pooled.

Ultrastructure of neuron-NG2 cell synapses in the hippocampus
Applying correlated light and electron microscopy, we investigated synapses onto NG2 cells in the CA1 region. The typical current pattern and light microscopic morphology of the filled cells analyzed ultrastructurally (n = 3) are shown in Figs. 1A, B. Axon terminals form synapses with processes of all three NG2 cells (Fig. 1D, E). This confirms earlier findings demonstrating synapses on processes of NG2 cells in the hippocampus [6,[23][24][25]. However, only 3, 6, and 8 synapses, respectively, were found on the three cells analyzed, (Table 1), although all serial sections from a given biocytin filled NG2 cell were examined over its full process extent. The total number of synapses on the three cells was estimated to be 30 (as described above; Table 1). These synapses were very similar in structure to neuron-neuron synapses, displaying pre-synaptic vesicles, post-synaptic density and cleft material (Figs. 1C, D 1 , E 1 ). In several axon terminals, docked vesicles were observed at the presynaptic membrane (Figs. 1D 1,2 , E 1 ). In some cases, the DAB reaction product was faint enough to reveal distinct post-synaptic detail, which was indistinguishable from neuron-neuron synapses. Thus, several neuron-NG2 cell synapses could be unequivocally classified as either asymmetric (7/17) or symmetric (1/17) (see Table 1, Figs. 1D-F). All synapses were on the processes of NG2 cells, none on the soma. The post-synaptic NG2 cell process was frequently conspicuously thin, measuring 0.2-0.5 mm (Fig. 1C, E), but in several instances 1-2 mm (Fig. 1D). Thus, in contrast to earlier studies in adult rats [24], we found only few synapses per cell, and morphology in our material was indistinguishable from classical synapses between neurons.
The physiological properties of these neuron-NG2 cell-synapses are characterized in some detail [6]. So far, however, it is largely unclear whether neuronal innervation initiates Ca 2+ -signaling in post-synaptic NG2 cells. Therefore, we tested for potential pathways provoking [Ca 2+ ] i elevation in NG2 cells, which might be activated by the synaptic input.

NG2 cells express functional voltage-gated Ca 2+ -channels
Previous work has demonstrated that complex glial cells in wild type mice express different types of Ca v [14], although later on its presence in NG2 cells has been disputed [16,17]. To reinvestigate this issue in NG2/EYFP positive cells, putative Ca v currents were isolated using Na + -and K + -free bath and pipette solutions. In addition, solutions were supplemented with Na v and K v channel blockers, and [Ca 2+ ] in the bath was increased to 5 mM (see Materials and Methods and [14]). To remove steady-state inactivation from putative Ca v channels, conditioning pre-pulses to 2110 mV and 210 mV were applied for 1.5 s, respectively. Afterwards, current families were subtracted at corresponding membrane potentials. This procedure isolated transient membrane currents in NG2 cells (peak amplitudes 100630 pA at 220 mV, n = 14) (Fig. 2B 1 ). Plotting the I/V relationship of the evoked currents revealed a threshold potential of 260 mV, while peak inward currents occurred at about 220 mV (Fig. 2B 2 ). The L-type channel blocker Verapamil (100 mM) reduced the maximum inward currents from 167635 pA to 85633 pA (n = 9, Fig. 2C 2 ) and significantly shifted the half maximum voltage of the steady state inactivation curve (from 286.367.2 mV to 264.364.5 mV, n = 4, paired T-test, Fig. 2C 1 ). Coapplication of the T-type channel blocker Mibefradil (50 mM) further diminished Ca v currents in 4/5 cells tested (to 25610 pA). These properties resemble Ca v currents in complex glial cells of the hippocampal CA1 region [14].
To identify the subtype(s) of Ca v s expressed by NG2/EYFP positive cells, transcript analysis was performed employing single cell RT-PCR (Tab. S1). We found predominant expression of mRNA encoding the L-type channel isoforms Ca v 1.2 and Ca v 1.3 (Fig. 2D 1 ) and the T-type channels Ca v 3.1 and Ca v 3.2. Transcripts for P/Q and N-type channels, Ca v 2.1 and Ca v 2.2, were less abundant, while mRNAs for Ca v 1.4, Ca v 2.3 and Ca v 3.3 were never detected (Fig. 2D 2 ). Interestingly, the majority of NG2 cells tested (n = 39/46) expressed mRNA for the glial marker S100b This is in line with our previous data showing that some of the NG2/EYFP positive cells express S100b while the astrocytic marker GFAP was consistently lacking (Karam et al., 2008).
To further confirm the presence of functional Ca v s in NG2 cells of the hippocampus, Ca 2+ -imaging was combined with patchclamp recording in the whole-cell mode. Train stimulation via the patch-pipette (15 consecutive depolarizing voltage steps (100 ms) from 2100 mV to +20 mV, see lower traces in Fig. 3B and 3C 1 ) produced reversible elevations of [Ca 2+ ] i in NG2/EYFP cells (Fig. 3A 1 ). It is important to note that in the same cell, several [Ca 2+ ] i elevations could be elicited up to 30 min after establishing the whole-cell configuration (Fig. 3A 2 ). Next, we tested the sensitivity of the [Ca 2+ ] i elevations to Ni 2+ . At high concentrations Ni 2+ is known to non-specifically block Ca v s [26,27]. Indeed, application of 200 mM Ni 2+ abolished the [Ca 2+ ] i elevations in the NG2/EYFP cells tested (n = 4) (Fig. 3B).
At these high concentrations, Ni 2+ might also inhibit the NCX [28]. To exclude that the observed block of [Ca 2+ ] i elevations by Ni 2+ was due to its action on NCX rather than Ca v s, we tested the sensitivity of evoked [Ca 2+ ] i elevations to the NCX inhibitor SN-6. SN-6 has no effect on Ca v s while blocking NCX operating in the Ca 2+ -influx mode [29]. The amplitudes (103634 pA vs. 86622 pA, n = 5) and decay time-constants (39.466.8 ms vs. 39.263.6 ms, monoexponential fit, n = 4) of depolarizationinduced Ca v currents (at 210 mV) were not affected by SN-6 (10 mM; paired Student's T-test, p.0.05; not shown; but see Fig. 2B). Together, these data demonstrate functional expression of Ca v s by NG2 cells in the hippocampus, corroborating previous findings in complex glial cells of wild type mice [14].
We further analyzed the kinetics and amplitudes of depolarizationinduced [ To improve time resolution of Ca 2+ -imaging we also performed LSM based x-t line scans. Therefore, individual NG2/EYFP cells were loaded with 400 mM Fluo-4 via the patch-pipette (Fig. 4A).  Second, the [Ca 2+ ] i elevation outlasted channel open time more than tenfold but the binding kinetics of the Ca 2+ -indicator dyes used are in the range of microseconds [30]. Therefore, this can not account for the phenomenon.
Recently, it was suggested that in NG2 cells [Ca 2+ ] i elevation evoked by depolarization is mainly due to NCX operating in the Ca 2+ -influx mode in a tetrodotoxin (TTX) sensitive manner [17]. In our hands, TTX (1 mM, n = 13) neither affected the amplitudes nor the kinetics of depolarization-induced [Ca 2+ ] i elevations in NG2/EYFP cells (n = 13, Fig. 4D). This goes in line with our finding, that Ca v channels were not influenced by the specific NCX reverse mode blocker, SN-6 ( Fig. 2B).
Ca 2+ -influx through Ca v s evokes Ca 2+ -induced Ca 2+ -release in NG2 cells Ca 2+ -influx through the plasma membrane may evoke further increase in [Ca 2+ ] i by triggering Ca 2+ -release from intracellular stores [31], which might account for the observed saturation and prolonged kinetics of [Ca 2+ ] i elevations. To investigate whether Ca 2+ -induced Ca 2+ -release (CICR) is operative in NG2 cells we performed recordings in nominal Ca 2+ -free bath solution supplemented with 2 mM EDTA. Under these conditions no [Ca 2+ ] i elevation could be elicited by train stimulation. The same individual cells showed strong increases in [Ca 2+ ] i after switching to artificial cerebrospinal fluid (aCSF) bath solution containing 2 mM Ca 2+ (Fura-2/CCD recording, n = 5; Fluo-4/LSM recording, n = 5) ( Fig 5A). Hence, depolarization per se was insufficient to increase [Ca 2+ ] i . This indicated that Ca v s mediated the initial phase of the [Ca 2+ ] i elevations in NG2 cells while CICR was responsible for the late phase. To test this hypothesis, single pulses were applied before and after depletion of intracellular Ca 2+ -stores. Depletion was achieved by train stimulation in the presence of thapsigargin (1 mM), a blocker of sarco/endoplasmic reticulum Ca 2+ -ATPase [32]. Under these conditions, single pulse [Ca 2+ ] i elevations declined to 16% of the control value (n = 5) (Fig. 5B). This suggests that the depolarization-induced [Ca 2+ ] i elevations in NG2 cells are due to initial influx of Ca 2+ through Ca v s, followed by CICR.

NG2 cells express functional group I metabotropic glutamate receptors
Next, we tested whether NG2 cells express metabotropic glutamate receptors (mGluRs). The group I mGluR-specific agonist 3,5-DHPG was focally applied, while membrane currents and [Ca 2+ ] i were monitored by simultaneous patch-clamp recording in the whole cell mode and line scan imaging. All cells tested responded to 3,5-DHPG with [Ca 2+ ] i elevation (DF/F 0 = 1.1760.66, n = 7, 100 mM; DF/F 0 = 1.1460.79, n = 6, 10 mM). This was never accompanied by current responses (Fig. 7A 1 ). The delay between substance arrival and the onset of [Ca 2+ ] i rises (see [Material and Methods] for details) varied among cells (3.463.3 s, n = 7, range between 0.6 and 9.4 s), but not between multiple 3,5-DHPG applications to the same individual cell.

NG2 cells express vesicular glutamate transporters
The observation of stimulus-induced [Ca 2+ ] i elevations prompted us to search for potential downstream signaling mechanisms in NG2 cells. Astrocytes express vesicular glutamate transporters (vGLUTs) in their distal processes, and were reported to communicate with neurons by Ca 2+ -dependent release of vesicular glutamate [39][40][41]. To investigate whether vGLUTs may also be expressed by NG2 cells, transcript analyses were performed. vGLUT1 and vGLUT2, but not vGLUT3 could be detected by post-recording single cell RT-PCR from NG2 cells of hGFAP/ EGFP mice (p9-15). Gene transcripts for vGLUT1 were detected in 6/25 NG2 cells, resembling its prevalence in astrocytes [39]. vGLUT2 was co-expressed in 1/25 cells (not shown). As a positive control for cell type specificity, mRNA of the NG2 cell-specific PDGFa-receptor was co-amplified (n = 22). We further investigated presence and localization of vGLUT1 and vGLUT2 protein in gray matter NG2 cells in hippocampal slices by applying high resolution fluorescence microscopy, subsequent to patch-clamp recording and biocytin filling. Staining was observed for vGLUT1 (2/3 cells) and vGLUT2 (2/2 cells). Larger vGLUT1 positive puncta, putative vesicle groups, were found in the fine NG2 cell processes (Fig. 9). The inclusion of vGLUT-immunoreactivity (vGLUT-IR) within NG2 cell profiles was verified at high magnification by 3D inspection (Fig. 9A), and by increasing the opacity of surface-rendered, 3D-reconstructed NG2 cells (Fig. 9B, Video S1). Based on the rigorous thresholding, we assume that in our analysis the amount of vGLUT-IR in NG2 cells is underestimated. vGLUT1 or vGLUT2 positive puncta did not display a preference for the varicosities of NG2 cell processes but occurred all over the process tree, also at any proximo-distal distance. The immunhistochemical and RT-PCR data indicate heterogeneity among NG2 cells with regard to expression of vGLUTs.
NG2 cell processes are motile and display actin and ezrin, but not tubulin Recent reports suggested a link between [Ca 2+ ] i elevation and migration of NG2 cells in vitro [17]. To investigate the possibility of process motility in situ, we performed time-lapse recordings in acute hippocampal slices. We detected process motility in 5 out of 11 dye-labeled NG2/EYFP cells (Fig 10A). At least three types of process motility were observed; including elongation ( Fig. 10B) and retraction (Fig. 10C) of processes (see also Videos S2, S3).
Additionally, we observed that strongly dye-labeled varicosities, which are characteristic of NG2 cells, move along the processes (Fig. 10D). The varicosities traveled up to 2.9 mm within 6 min (Fig. 10D). Some varicosities showed bi-directional motility. Thus, NG2 cell processes and their varicosities exhibit motility on a minute time range.
Next, we investigated cytoskeletal constituents potentially relevant to motility of NG2 cells. Therefore, cells were freshly isolated from tg(hGFAP/EGFP) mice and selected according to their characteristic morphology and specific immunolabeling (GFP positive, GFAP negative) [18]. Antibodies against a-tubulin, bactin, ezrin (a microvillus-associated, actin-binding protein [42]), or protein disulfide isomerase (PDI) were combined with both, anti-GFP and anti-GFAP staining. Noteworthy, a-tubulin (6/6) was not present in the processes but restricted to the soma and in a few cases to the proximal portion of processes (Fig. 10E). At the same time, the processes of nearby astrocytes were positive for atubulin (Fig. S1). b-actin (10/10) and ezrin (10/10) were distributed all over the cell including the fine NG2 cell processes (Fig. 10F,G). GFAP was detected in astrocytes but not in NG2 cells (36/36 cells, not shown). In the context of CICR mentioned above, we also studied the localization of endoplasmic reticulum, applying anti-PDI as a marker [43,44]. PDI-IR (10/10) was restricted to the soma and never detected in the NG2 cell processes (Fig. 10H).

NG2 cells display several mechanisms of intracellular Ca 2+ -elevation
Our data demonstrate the capability of gray matter NG2 cells to increase [Ca 2+ ] i via several independent pathways: G-protein coupled receptors, as well as ligand-and voltage-gated ionchannels. While the presence of mGluRs in NG2 cells represents a new finding, expression of Ca v s is under discussion. Recently, it was reported that NG2 cells in the hippocampus lack Ca v s [16,17]. In contrast, earlier work on complex glial cells in the hippocampus described low-and high-threshold activated Ca v s which were sensitive to Cd 2+ or dihydropyridines and omega-conotoxin GIVA, respectively [14]. Here, we confirm the presence of Ca v s in identified NG2/EYFP cells. This discrepancy with the former data may be due to different recording conditions. Ca 2+ -currents in NG2 cells are small in amplitude, compared with the dominating K + currents. Its reliable separation requires use of Na + -and K + -free solutions, elevated [Ca 2+ ] in the bath solution and application of conditioning pre-pulses.
The small amplitudes and high activation threshold of the Ca 2+ -currents through NG2 cell Ca v s raise the question of its physiological relevance. To tackle this question, we employed Ca 2+ -imaging. Using aCSF, depolarization evoked reversible [Ca 2+ ] i elevations in NG2 cells. This was due to influx of Ca 2+ through Ca v s, but not to the activation of NCXs, as recently suggested [17]. A possible explanation for this conflicting finding might be that in the latter study, KB-R 7943 was used as an inhibitor of NCX, which blocks Ca v s with almost the same affinity [45]. Similarly, Ni 2+ does not only block Ca v s but also NXCs [28]. SN-6, on the other hand antagonizes with high affinity only the Ca 2+ -influx mode of NCXs, preferentially of NCX1, while not interfering with Ca v s at the concentration used here [29]. Because (i) SN-6 did not affect the electrophysiologically recorded Ca 2+ -currents (Fig. 2B) and (ii) TTX did not diminish the voltage-step induced [Ca 2+ ] i elevations (Fig. 4D) we believe that in NG2 cells Ca 2+ -influx through NCXs plays only a minor role, if any. The functional characterization of the NG2 cell Ca v subtypes is a challenging task for future studies. The transcript data reported here together with the pharmacological findings by Akopian [14] might provide first clues.
[Ca 2+ ] i elevation through Ca v activation was almost doubled due to CICR. Notably, this led also to a significant prolongation of the [Ca 2+ ] i elevations. Thus, CICR represents a powerful mechanism to amplify small inward currents through Ca v s in NG2 cells. The observed saturation effect (Fig. 4C) suggests the involvement of Ca 2+ binding sites with low affinity acting as intracellular Ca 2+ sensors, analogously to myocardial cells (e.g. [46]). This may regulate the gain of CICR depending on ambient [Ca 2+ ] i levels. Currently, we do not know whether Ca 2+ amplification exists in NG2 cell processes. The absence of PDI-IR from processes (Fig. 10H) precludes CICR in these structures, and potential amplification mechanisms would have to be independent of endoplasmic reticulum.
In agreement with previous findings [6] our data suggest the presence of Ca 2+ -permeable AMPA/kainate and GABA A receptors in NG2/EYFP cells. Activation of the latter receptors depolarizes NG2 cells, which might trigger the activation of Ca v s. Such indirect GABA receptor-mediated [Ca 2+ ] i elevations have been observed in cultured OPCs [15]. Depolarizations induced by AMPA/kainate receptor activation might have similar effects, although we can not exclude a contribution of metabotropic kainate receptors to the [Ca 2+ ] i elevations [34]. It will be a challenge to determine whether in the fine processes, receptor activation produces depolarization sufficient for Ca v activation in NG2 cells under physiological conditions. We further report that NG2 cells in the hippocampus express functional group I mGluRs. Pharmacological analysis indicated elevations. Experiments shown in (C 1 -C 3 ) were performed in the presence of the blocking cocktail described in the text. Each row represents one individual brain slice. doi:10.1371/journal.pone.0017575.g007

NG2 cell processes are highly motile, actin-based surface extensions
Our live microscopic data demonstrate, for the first time, motility of NG2 cell processes in situ. We investigated the presence of cytoskeletal proteins in NG2 cell processes to test for prerequisites of process motility. The cytoskeleton of NG2 cell processes is found to be actin-based, since GFAP-positive glial (intermediate) filaments or microtubules were not observed by immunolabeling and electron microscopy. This appears astonishing in respect of their length (30-50 mm) and small diameter (0.2-1 mm) in between the varicose expansions. Of the many actinbinding proteins ezrin was chosen as a further marker, because its (de)phosphorylation-based mode of membrane-to-cytoskeleton linking enables rapid shape changes [47]. Ezrin, and its close relatives, radixin and moesin (the ERM protein family), are typically involved in establishing highly motile and very narrow structures in the CNS, such as neuronal growth cone filopodia [47,48] or peripheral astrocyte processes [49,50]. Also, ERM proteins are required for maintaining stereocilia integrity in cochlear and vestibular hair cells [51]. Altogether, the set of features displayed by NG2 cell processes classifies them as actinbased stereocilia and surface extensions. They constitute a rare example of an actin-based surface extension that is directly involved in synaptic signaling.

Possible impact of the synaptic input onto NG2 cells
Recent findings suggest a role of neuron-NG2 cell synapses in migration. Thus, in the corpus callosum adult-born migrating NG2 cells receive glutamatergic synaptic input from demyelinated axons [52], and GABA-mediated [Ca 2+ ] i elevation is essential for migration of subventricular zone NG2 cells to and within white matter in vitro [17]. Ca v s might be important in this context as they have been reported to govern migration in newborn neurons, e.g. in the postnatal olfactory bulb [53]. However, the reported data relate to lesioned white matter, where neuron-glia synapses are transient [52]. In contrast, gray matter NG2 cell synapses are lesion independent and functional under physiological conditions. An alternative function of synaptic input on NG2 cells in gray  This hypothesis would be in line with the finding that synapses were exclusively found on NG2 cell processes but not at somata.
Synaptic activation may cause small [Ca 2+ ] i elevations through the Ca 2+ -signaling pathways reported here. However, because the processes are devoid of endoplasmic reticulum, these [Ca 2+ ] i elevations are unlikely to be amplified by CICR and might occur locally confined. Local [Ca 2+ ] i elevations might play a role in regulation of process motility. In addition, restricted Ca 2+ -signaling might be interesting in the light of the demonstrated vGLUT expression. In neurons, vGLUT expression is sufficient for defining a glutamatergic phenotype [54]. In astrocytes vGLUTs mediate vesicular transmitter release, at least in the cell culture [39][40][41]. The scattered vGLUT organelles within NG2 cell processes might serve a similar function. The intriguing perspective that NG2 cells might signal to neighboring cells in a Ca 2+ -dependent manner remains to be addressed in future studies.

Materials and Methods
Maintenance and handling of animals used in this study was according to local government regulations. Experiments have been approved by the State Office of North Rhine-Westphalia, Department of Nature, Environment and Consumerism (LANUV NRW, approval number 9.93.2.10.31.07.139). All measures were taken to minimize the number of animals used.
Recordings were monitored with TIDA software (HEKA). Series and membrane resistance were checked in constant intervals with self-customized macros using Igor Pro 6 software (WaveMetrix Inc., Lake Oswedo, USA). Visual control was achieved by a microscope equipped with an infrared DIC system (Leica DM6000, Leica, Mannheim, Germany) and an IR objective (HCX APO L 20x/1.0 W; Leica). Infrared and epifluorescence images were captured with a digital CCD camera (DFC350FX R2; Leica).
Membrane currents were compensated offline for stimulus artifacts using Igor Pro 6 software according to the following procedure: Ten traces evoked by voltage steps from 280 to 270 mV were averaged and fitted monoexponentially. Compensated current traces were obtained by multiplying the fitted curve with the respective factors and subsequent subtraction from the original current traces at different membrane potentials.
Evoked post-synaptic currents in NG2 cells were compensated for stimulus artifacts by subtracting averaged failure traces.
Substances were pressure-applied focally using a multichannel Octaflow superfusion system (ALA Scientific Instruments, Farmingdale, USA). The 20-80% rise time of agonist concentration amounted to ,100 ms. Short test pulses of GABA were used to assess the delay between valve opening and arrival of the substance at the recorded cell, which ranged between 0.4 and 0.8 s. All agonist responses were corrected for this delay. In some cases, substances were applied by changing the bath solution. All statistical data are given as mean 6 SD.

Two-photon time-lapse imaging
Individual NG2/EYFP-positive cells were filled for 2 min with Alexa-594 (Invitrogen, Karlsruhe, Germany) via the patch-pipette [56]. Dye was allowed to diffuse for .30 min before imaging. Subsequent two-photon imaging was performed on a confocal laser scanning microscope (LSM)(SP5, Leica) equipped with a mode-locked infrared laser (MaiTai BB, Newport/Spectra Physics, Irvine, USA). The dye was excited at 810 nm and emitted light was detected with built-in non-descan detectors below 680 nm. These experiments were performed at 35uC to increase process motility. The bicarbonate concentration of aCSF was reduced to 20 mM to achieve correct pH values. Image stacks of up to 60 optical planes were acquired for 20 to 60 min (z-step distance 250 nm, aCSF). We assured by inspection of all optical planes that the observed cellular motility was not caused by drift of slices, recording chamber, or microscope.
Ca 2+ -imaging NG2/EYFP cells in the stratum radiatum of the CA1 area were used for Ca 2+ -imaging. To determine absolute [Ca 2+ ] i and achieve a high time resolution of Ca 2+ -transients two different methods were applied. non-identified, nearby cell (H, overview) and in the soma of the NG2 cell, but not in its processes (H, red, merge). The same is observed for a-tubulin (E red, merge). Note that a-tubulin/microtubules are well-preserved in the processes of nearby non-identified cells (E, merge) and of GFAP positive astrocytes (cf. Fig. S1). Scale bar 5 mm. doi:10.1371/journal.pone.0017575.g010 (i) Changes in [Ca 2+ ] i were monitored by a CCD camera (SensiCam; TILL photonics, Martinsried, Germany) mounted on a wide-field epifluorescence system (Polychrome II, TILL photonics). It was attached to an upright microscope (Axioskop FS2, Zeiss, Oberkochen, Germany) equipped with a 60x LUMPlan FI/IR objective (Olympus Optical Co., Hamburg, Germany). Fluorescence excitation was achieved by a monochromator. Individual cells in acute hippocampal slices were loaded via the patch-pipette with Fura-2 (200 mM; Invitrogen). Dye filling lasted $5 min before Ca 2+ -imaging was started. If not stated otherwise, Fura-2 was excited at 380 or 340 nm for 40 ms and emission was detected at an acquisition rate of 25 Hz during, and 3 Hz after depolarization. Single frames were recorded at the isosbestic point (362 nm) before and after each sequence. This allowed offline calculation of pseudo-ratiometric images to correct for bleaching. The latter was assumed to be proportional to exposure time. A linear function was calculated from the first and the last 362 nm frame of each of the 380 or 340 nm sequences. This function was used to determine the 362 nm values for each recorded frame. Pseudo-ratios F 380 or F 340 /F 362 were calculated from the measured F 380 or F 340 and the extrapolated F 362 values for each time point. F 380 /F 362 pseudo-ratios were inversely plotted so that [Ca 2+ ] i elevations are always indicated by upward deflections.
Absolute [Ca 2+ ] i was estimated through calibration according to Grynkiewicz et al. [57]: (c f : concentration ofCa 2z unbound; c b : concentration of Ca 2z bound Fura-2) R min and R max were determined with 10 mM BAPTA or 10 mM CaCl 2 in the pipette solution, respectively. K d was determined with a pipette solution buffered to 11 nM free Ca 2+ and amounted to 51 nM. R(t) curves were calculated from two successive recordings at 380 nm and 340 nm. F 380 (t) and F 340 (t) were corrected for bleaching using the pseudo-ratio method described above. Calibration was performed using self-customized IGOR 6 functions.
(ii) Alternatively, an LSM (Leica) was used for Ca 2+ -imaging, allowing for higher time resolution. Individual NG2/EYFP positive cells were loaded with Fluo-4 (400 mM, Invitrogen) via the patch-pipette. Subsequent line-scans, taken at the soma, were recorded with an excitation at 488 nm. Emission was detected between 500 and 650 nm. Signals were sampled at 1-0.

Single cell RT-PCR
After electrophysiological characterization in situ, the cytoplasm of individual cells was harvested under microscopic control as reported previously [18]. Reverse transcription (RT) was started after addition of RT-buffer, 10 mM DTT (final concentration; Invitrogen), 46250 mM dNTPs (Applied Biosystems, Darmstadt, Germany), 50 mM random hexamer primer (Roche, Mannheim, Germany), 20 U RNase inhibitor (Promega, Madison, USA), and 100 U SuperscriptIII reverse transcriptase (Invitrogen). Final volume was ,10 ml. A multiplex two-round PCR with single-cell cytosol was performed with primers for the Ca v 1, Ca v 2 and Ca v 3 families or vesicular glutamate transporters (vGLUT) 1/2 and vGLUT3, respectively (Table S1). Primers were located in conserved regions to amplify all members of the respective family. The first PCR was performed after adding PCR buffer, MgCl 2 (2.5 mM), and primers (200 nM each) to the reverse transcription product (final volume 50 ml). Taq polymerase (3.5 U; Invitrogen) was added after denaturation. 45 cycles were performed (denaturation at 94uC, 25 s; annealing at 49uC, first five cycles: 2 min, remaining cycles: 45 s; extension at 72uC, 25 s; final elongation at 72uC, 7 min). An aliquot (2 ml) of the PCR product was used as a template for the second PCR (35 cycles; annealing at 54uC, first five cycles: 2 min, remaining cycles: 45 s) using nested, subunit-specific primers (Table S1). The conditions were the same as described for the first PCR-round, but dNTPs (4650 mM) and Platinum Taq polymerase (2.5 U; Invitrogen) were added. Products were identified by gel electrophoresis using a molecular weight marker (Phi X174 HincII digest; Eurogentec, Seraing, Belgium).
Primer specificity was tested with total RNA from freshly isolated mouse brain (p20). For optimization, a two-round RT-PCR was performed with 2 ng of total RNA and primers as described above. Subsequent gel analysis did not detect unspecific products. The primers for different targets were located on different exons to prevent amplification of genomic DNA. Omission of the RT-enzyme and substitution of template by bath solution served as negative controls for reverse transcription and PCR amplification and confirmed the specificity of the reaction.

Electron microscopy
Acute hippocampal slices were prepared from juvenile (p9-12) hGFAP-EGFP mice. Weakly fluorescent cells with a typical electrophysiological current-pattern (previously termed GluR cells; [18]) were filled with biocytin (0.5%) via the patch-pipette during whole-cell recording. Slices were then fixed for 2 h in a solution containing paraformaldehyde (PFA) and glutaraldehyde (2% each in 0.1 M phosphate buffer, PB). Fixation delay after decapitation ranged from 45-120 min. Slices containing a biocytin-filled cell were rinsed, cryoprotected in sucrose solution (30% in PB), snapfrozen in liquid nitrogen and thawed [58]. Cells were visualized for correlating light and electron microscopy by overnight incubation in a combination of avidin-biotin complex (1:100, Vector, Burlingame, USA; [59]) and streptavidin-CY3 (1:1,000, Vector). After rinsing, the biocytin-filled cells were coverslipped in PB and documented by recording image z-stacks under a fluorescence microscope. Subsequently, the peroxidase was developed by diaminobenzidine (DAB) and 0.07% H 2 O 2, for ultrastructural staining. Sections were osmicated (1% OsO4), block stained (1% uranyl acetate in 70% ethanol), dehydrated and flat embedded in Araldite. Ultrathin sections were contrasted with lead citrate and uranyl acetate. To analyze overall synaptic contacts on NG2 cells at the ultrastructural level, these flat embedded cells were completely sectioned. Inspecting all ultrathin sections from a given cell, the complete process tree was scanned for synapses on DAB-containing profiles. Most synapses found in one section could also be documented in subsequent sections. To estimate the total number of synapses, the observed number of synapses was documented (Table 1) and then multiplied by 1.75 (1+0.5+0. 25). An estimated factor of 0.5 was introduced to account for the missed, nearly tangentially sectioned synapses above and below a DAB-labeled profile. This corresponds to missing unrecognized synaptic profiles which are obliquely sectioned between 30 and 0 degrees (tangential). Further, we amply estimated to have overlooked J of the NG2 cell profiles, because most synapsebearing profiles were below 0.3 mm (comp. Figs. 1 C, E), which was corrected for by a factor 0.25.

Dissociation of NG2 cells
Unequivocal determination of antigen presence in the NG2 cell processes is hampered by light microscopic resolution because they are frequently only 200-500 nm thick. We either studied freshly dissociated NG2 cells by conventional immunofluorescence or NG2 cells in brain slices using deconvolution microscopy with higher resolution.
Detection of vGLUT-IR in NG2 cells is challenging because it is abundant and dense in brain, and NG2 cell processes are frequently thinner than 0.5 mm, as observed in the electron microscope (cf. Fig. 1C). We carried out subresolution microscopy on an appropriate microscopy setup (Zeiss 200M; Orca AG camera, Hamamatsu, Hamamatsu City, Shizuoka, Japan; Openlab software, Improvision, Coventry, UK; 4061.3, 6361.4, 10061.45 oil immersion lenses, Zeiss). We applied on-chip magnification (100-160x), imaging the cells at 50-100 nm steps in two fluorescence channels (filter sets (I) ex 475/20, bp 495, em 513/17 and (II) 632/22, 660, 700/75). The resulting image stacks underwent iterative deconvolution (Openlabs) based on calculated point spread function that has previously been applied and validated for antigen colocalization in single vesicles [40,64]. Image analysis and 3D reconstruction (Openlabs) included intensity thresholding in both channels. In particular, intensity thresholding in the vGLUT channel was rigorous and led to disappearance of most smaller vGLUT-positive puncta, with many false negatives to avoid false positives. Thresholding in the GFP channel frequently resulted in discontinuous glial cell processes. Post hoc exclusion of all vGLUT-IR outside the cell facilitated visualization. All instances of vGLUT-IR within in the glial cells were checked for full inclusion in 3D cardbox view (see Fig. 9). No vGLUT-IR was detected in controls without primary antibody. Further processing of electron or light microscopic images was done with Photoshop (Adobe Systems), and comprised only linear operations for optimizing brightness and contrast, but no selective processing of image detail. Figure S1 Microtubules are well-preserved in the processes of freshly dissociated, identified astrocytes. Labeling for both, cell nuclei (bisbenzimidine) and glial filaments (GFAP, Alexa 360) is revealed in the blue channel. An astrocyte (center) and two unidentified cells (right) are displayed. Microtubules (a-tubulin, red) are obvious in the astrocyte processes demonstrating that the dissociation method does not interfere with microtubule integrity even in the processes. (TIF) Figure S2 Exemplary agarose gels of mRNA-transcripts for Ca v channel family and S100b.

(DOC)
Video S1 Demonstration of full inclusion of vGLUT1 positive objects in NG2 cell processes (3D reconstruction). The cell is the one shown in Fig. 9. NG2 cells from hippocampus (CA1) were identified by electrophysiology, biocytinfilled, fixed and visualized by streptavidin CY3 (red channel). The green channel displays immunocytochemical detection of vGLUT1. For clarity, all vGLUT staining outside the cells has been removed. After deconvolution of 75 nm optical sections, the cells (n = 5) were 3D reconstructed and isosurface-rendered. Due to high magnification, a frame displays only parts of a cell. By 3D rotating the reconstruction and changing its transparency, the movies demonstrate full inclusion of the vGLUT1 objects in the small processes (,0.5 mm, often 0.2 mm). Unit of the 3D grid scale: 5.5 mm.

(AVI)
Video S2 Elongation of an NG2 cell process. (cf. Fig. 10B). Two-photon time-lapse video was obtained from Alexa-594 dyeloaded NG2/EYFP cell processes located in an acute brain slice. Optical stacks of 20 planes were recorded every 34 s. Maximum zprojections are shown with 1 frame per second (volume 1661465 mm, total time 330 s, aCSF, 35uC).

(AVI)
Video S3 Retraction of an NG2 cell process and movement of intracellular varicosities. (cf. Fig. 10C, D). Similar recording parameters as in Video S2 were used. (AVI)