Control of Synapse Number by Glia

See allHide authors and affiliations

Science  26 Jan 2001:
Vol. 291, Issue 5504, pp. 657-661
DOI: 10.1126/science.291.5504.657


Although astrocytes constitute nearly half of the cells in our brain, their function is a long-standing neurobiological mystery. Here we show by quantal analyses, FM1-43 imaging, immunostaining, and electron microscopy that few synapses form in the absence of glial cells and that the few synapses that do form are functionally immature. Astrocytes increase the number of mature, functional synapses on central nervous system (CNS) neurons by sevenfold and are required for synaptic maintenance in vitro. We also show that most synapses are generated concurrently with the development of glia in vivo. These data demonstrate a previously unknown function for glia in inducing and stabilizing CNS synapses, show that CNS synapse number can be profoundly regulated by nonneuronal signals, and raise the possibility that glia may actively participate in synaptic plasticity.

Astrocytes, which ensheathe synapses throughout the CNS, are presently thought of as synaptic support cells, clearing ions and neurotransmitters from the synaptic cleft, but accumulating evidence suggests the possibility of more active roles as well (1–3). The role of glia at CNS synapses has been difficult to study because glial cells are present in CNS cultures and are often necessary for neuronal survival. In this study, we have taken advantage of recently developed methods to isolate a defined type of CNS neuron, retinal ganglion cells (RGCs), by immunopanning the cells to greater than 99.5% purity (4, 5) and then culturing them in a defined serum-free medium (5). Using these methods, we have previously shown that RGCs cultured in the presence of astrocyte-conditioned medium have 10 times higher levels of synaptic activity (6). This glial effect is not simply due to enhancement of RGC maturation, because RGCs cultured either in the presence or absence of glia are electrically excitable, develop comparable numbers of dendrites and axons, and have identical survival rates. In addition, RGCs cultured under both conditions synthesize, secrete, and respond to glutamate (6) and express nearly identical levels of mRNAs for a large battery of neurotransmitter receptors, voltage-dependent ion channels, and synaptic proteins, as assessed by gene chip analysis (7). In this study, we examine whether the increased synaptic activity in glia is due to an increase in the number of synapses or to the efficacy of synaptic transmission.

To address this question, we cultured purified RGCs in the presence or absence of astrocytes. RGCs were cultured for 1 week followed by the addition of astrocytes either in contact with RGCs or as a feeding layer for RGCs (8). Similar results were obtained under both conditions. After 2 weeks in culture, we used whole-cell patch-clamp recording to measure the significant enhancement of spontaneous synaptic activity by astrocytes (Fig. 1, A and B) (9). In the absence of astroglia, little synaptic activity occurred even when RGCs were cultured with their tectal target cells (6). This difference in synaptic activity persisted even after a month in culture, indicating that it is not accounted for by a maturational delay. To examine whether glia regulate synaptic activity by enhancing postsynaptic responsiveness, we measured the response of RGCs to pulses of l-glutamate applied to the cell somas. Regardless of the presence of glia, RGCs responded with inward currents (Fig. 1C) that were blocked by glutamate receptor antagonists (10). The size of the postsynaptic response, however, was threefold larger in RGC neurons cultured with glia (Fig. 1D), consistent with the increased amplitude of spontaneous synaptic miniature events (miniature excitatory postsynaptic currents, or mEPSCs) recorded in the presence of glia (Fig. 2G). To determine whether astrocytes also enhanced presynaptic function, we measured quantal content (QC) of evoked responses in autaptic neurons cultured with a feeding layer of astrocytes (Fig. 1, E and F) (11, 12), thus allowing us to measure presynaptic differences independently of differences in postsynaptic responsiveness. Glia increased QC by sevenfold (Fig. 1J). In addition, RGCs cultured without glia produced EPSCs only half as often as RGCs cultured with glia (Fig. 1I), and the EPSCs detected without glia had higher depression rates (Fig. 1, G and H). These findings indicate that glia enhance synaptic efficacy by both a postsynaptic and a presynaptic mechanism.

Figure 1

Glial astrocytes (glia) enhance synaptic efficacy by both presynaptic and postsynaptic mechanisms. (A and B) In the absence of glia, only a low frequency of small spontaneous synaptic currents is observed (A), whereas in the presence of glia large spontaneous synaptic currents are frequently observed (B). All synaptic currents were CNQX-sensitive (10). (C) Application of 50-ms pulses of 100 μM l-glutamate to RGC cell bodies at a holding potential of −70 mV results in sustained inward currents. (D) Current densities of RGCs cultured in the absence or presence of glia. Whole-cell currents: RGCs without glia, 15.7 ± 4.7 pA/pF; RGCs with glia, 45 ± 8.6 pA/pF (P = 0.015, Student's t test). (E and F) Autaptic EPSCs from RGCs cultured in the absence (E) or presence (F) of glia. (G and H) Higher synaptic failure rates are seen in RGCs without glia (G) than in RGCs with glia (H). (I) Without glia, 44 ± 18% autaptic neurons have detectable EPSCs, whereas 100% have EPSCS in the presence of glia (Mann-Whitney U test, P < 0.05). (J) Quantal content of RGCs cultured in the absence of glia is 7 ± 4 compared with 47 ± 10 for RGCs cultured with glia (including failures, P = 0.004). Quantal content was determined with the direct method.

Figure 2

Glial astrocytes (glia) increase the total number of synaptic vesicles released. (A to C) Labeling of synaptic puncta with FM1-43 in high K+. In RGCs cultured in the absence of glia (A, left), FM1-43 weakly labeled puncta; pseudocolor inset shows weakly labeled areas along neurites. In RGCs cultured with glia (A, right), FM1-43 strongly labeled many puncta. (B) Normalized data indicate that puncta destain in the presence or absence of glia with similar time courses. (C) Comparison of the difference in fluorescence intensity (ΔF) between fully stained and fully destained puncta in cells cultured with or without glia. RGCs with glia take up seven times more FM1-43: without glia, mean ΔF = 216 ± 9; with glia, mean ΔF = 1429 ± 100 for RGCs [arbitrary fluorescence units, Kruskal-Wallis analysis of variance (ANOVA),P < 0.001]. Bar, 50 μm. (D) RGCs cultured with glia display a two- to threefold increase in Ca2+ current density at a holding potential of −90 mV (left). Peak current: −33.4 pA/pF without glia, −86.7 pA/pF with glia. RGCs cultured with glia display a nearly threefold increase in Ca2+ current density at a holding potential of −50 mV (right). Peak currents: −20.7 pA/pF without glia, −87.1 pA/pF with glia. (E) Ionomycin (5 μM) in the presence of TTX (1 μM) and picrotoxin (50 μM) increases the spontaneous rate of miniature synaptic currents (mEPSCs) in RGCs grown both in the absence (left) and presence (right) of glia. (F) Comparison of the number of mEPSCs detected after treatment with ionomycin indicates that RGCs cultured in the presence of glia have more vesicles available for release. Number of mEPSCs: RGC with glia, 3928 ± 307 events/min; RGCs without glia, 250 ± 217 events/min (ANOVA,P = 0.0006). (G) Cumulative amplitude histogram of mEPSC recorded in RGCs in the presence and absence of glia. Mean mEPSC amplitude: with glia, 33 ± 4 pA; without glia, 21 ± 2 pA (P < 0.05, Student'st test). Maximum mEPSC amplitude: with glia, 151 pA; without glia, 60 pA. (H and I) Amount of presynaptic vesicle turnover caused by calcium influx with ionomycin treatment visualized with FM1-43. In RGCs cultured without glia, little labeling of presynaptic puncta occurs (H, left) compared with RGCs cultured with glia (H, right). (I) There is no difference between the ΔF values for puncta loaded with ionomycin or 90 mM K+ in RGCs without glia (ANOVA, P > 0.5). Bar, 50 μm.

To directly measure presynaptic function without using postsynaptic responsiveness as an indicator, we next examined the ability of RGCs to take up the fluorescent stryryl dye FM1-43 into synaptic vesicles during activity (13–16). Despite a prolonged depolarization, RGCs that had been cultured in the absence of glia took up little FM1-43 compared with RGCs that had been cultured with glia (Fig. 2A). In both cases, FM1-43 was observed in synaptic punctalike spots along dendrites and somas (Fig. 2A), and these puncta destained with similar kinetics (Fig. 2B). However, glia significantly increased the fluorescence difference ΔF, measured from FM1-43–loaded puncta before and after destaining, by about sevenfold (Fig. 2C). These findings indicate that glia strongly enhance presynaptic function (17).

To determine whether glia enhance presynaptic efficacy by increasing calcium influx, and thereby triggering synaptic vesicle release (18), we next compared the density of voltage-dependent calcium currents in RGCs cultured in the presence and absence of glia by measuring whole-cell calcium-current densities at two different holding potentials (−90 and −50 mV). At both holding potentials, glia increased calcium current density by about two- to threefold (Fig. 2D), suggesting that an increased presynaptic calcium influx might account for the glial enhancement of presynaptic function. To test this possibility directly, we bypassed the endogenous presynaptic voltage-gated calcium channels in RGCs by using the calcium ionophore ionomycin to produce a large presynaptic calcium influx (19,20). The number of ionomycin-induced mEPSCs recorded was still 10-fold greater in RGCs cultured with glia than without (Fig. 2, E and F). Comparable results were obtained when 500 mM sucrose or 0.3 nM α-latrotoxin were used to trigger vesicular release rather than ionomycin (10). Similarly, when we used ionomycin to load and unload presynaptic terminals with FM1-43, vesicular recycling was still greater in RGCs cultured with glia (Fig. 2H). Notably, the ΔF values measured with ionomycin in RGCs in the absence of glia were indistinguishable from the values measured with high-K+ stimulation of these neurons (Fig. 2I). These data indicate that glia do not increase presynaptic function solely by increasing presynaptic calcium influx; rather, glia increase the total number of vesicles released.

Given that calcium influx is not sufficient to explain the glial increase in synaptic efficacy, we next examined presynaptic proteins such as the putative calcium sensor synaptotagmin (21). To determine if glia affect synaptic protein levels, we performed immunoblots on extracts of RGCs cultured with or without a feeding layer of astrocytes (22). These analyses show that glia do not appreciably alter the levels of several major presynaptic vesicular proteins including synaptotagmin and synaptophysin (Fig. 3A), as well as SV2 (10). To determine if glia affect the localization of these proteins, we immunostained RGCs that had been cultured with or without astrocytes (23). In the absence of glia, only a diffuse, nonpunctate pattern of presynaptic protein immunoreactivity was observed when the cultures were stained with specific antibodies for synaptotagmin (Fig. 3B, left), synaptophysin, or SV2 (10). Only occasional small immunoreactive puncta were observed. In contrast, in the presence of glia, nearly all RGCs displayed numerous large, brightly immunoreactive puncta along their somas and dendrites (Fig. 3B, right). This analysis revealed that, although the absolute levels of synaptic proteins are unchanged, glia have a pronounced effect on the localization of these proteins, promoting their aggregation into synaptic puncta.

Figure 3

Glial astrocytes (glia) markedly increase the number of puncta containing both pre- and postsynaptic proteins. (A) Immunoblot analysis of extracts of RGCs cultured without glia or with glia indicates no change in total synaptic synaptotagmin and synaptophysin protein levels. p115 is a control for total protein. (B) Synaptic protein immunoreactivity in RGCs cultured in the presence and absence of glia. Cultured RGCs were fixed with paraformaldehyde and immunostained with antibodies against pre- and postsynaptic markers. Staining of RGCs with an antibody against the presynaptic marker synaptotagmin shows that in the absence of glia presynaptic markers are diffuse (left), whereas in the presence of glia presynaptic markers are discretely clustered (right). (C) Staining of RGCs with an antibody to the postsynaptic marker PSD-95 reveals that in the absence of glia there are relatively few PSD-95 puncta, whereas in the presence of glia numerous puncta are apparent. (D) In the absence of glia there is little overlap between pre- and postsynaptic markers, whereas in the presence of glia there is a high degree of overlap. Bar, 50 μm. (E) Double labeling of autaptic neurons indicates that fewer synaptic puncta are present in the absence of glia than in the presence of glia. Bar, 50 μm. (F) About six times as many synaptic puncta are present in RGCs cultured in the presence of glia as in the absence of glia: with glia, 34.5 ± 4.3 GluR2/3 puncta per cell; without glia, 6.1 ± 0.9 GluR2/3 puncta per cell.

To confirm that the large immunoreactive puncta correspond to synapses, we examined whether these puncta that stained brightly with antibodies to presynaptic proteins such as synaptotagmin colocalized with postsynaptic antigens. As shown in Fig. 3, B to D, puncta of presynaptic synaptotagmin were almost always colocalized with puncta of postsynaptic PSD-95 immunoreactivity in the presence of glia. In contrast, few puncta of postsynaptic PSD-95 or GluR2/3 formed in the absence of glia, and there was little colocalization with the few presynaptic puncta that formed. These data show that presynaptic and postsynaptic markers colocalize in puncta that are structural counterparts of synapses in the presence of glia, but rarely in their absence. The glial-dependent clustering and colocalization of synaptic markers was also observed when RGCs were cocultured with their highly purified target neurons from the superior colliculus (24), indicating that the inability of RGCs to form synapses without glia cannot be accounted for by the absence of their normal target neurons [see supplemental data (25)].

To determine if these puncta correspond to functional synapses, we counted synaptic puncta on autaptic neurons using immunostaining (Fig. 3E). Glia increased the number of GluR2/3-positive synaptic puncta by sixfold (Fig. 3F). The average number of immunoreactive puncta that formed on RGCs in the presence and absence of glia was 35 ± 4 and 6 ± 1, respectively. By comparison, the average number of quanta released in response to a strong stimulation in autaptic neurons cultured in the presence and absence of glia was 47 ± 10 and 7 ± 4, respectively. The similarity of these values suggests that the immunoreactive puncta that form in the presence of glia correspond to functional synapses. They also indicate that few synapses form between RGCs in the absence of glia and that glia increase the total number of synapses per RGC by six- to sevenfold. The few synapses that do form in the absence of glia are functionally immature, as indicated by their high failure rates and low uptake of FM1-43. Although functional synaptic activity has been reported within minutes of two excitable cells coming into contact, a mature ultrastructural appearance takes days to develop (26–28). To determine the effect of glia on the ultrastructural characteristics of synapses, we performed electron microscopy (EM) (29,30). RGC neurons were cultured for 14 days in the presence or absence of glial feeding layers, and then fixed and processed for EM (Fig. 4, A and B). In both cases, synapses were present and displayed the normal ultrastructural characteristics of asymmetric synapses formed by RGCs in vivo (31, 32). RGCs cultured in the presence of astrocyte-conditioned medium, however, had sevenfold more synapses than RGCs cultured in the absence of glia (Fig. 4C). Both the numbers of total vesicles and docked vesicles per synapse were unaffected by glia (Fig. 4D). These data demonstrate that glia significantly increase the average number of synapses per cultured neuron.

Figure 4

Glial astrocytes (glia) increase the number of synapses per neuron. (A and B) Electron micrographs of synapses between RGC neurons cultured in the absence of glia (A, left and right) and presence of glia (B, left and right). No difference is apparent in synaptic ultrastructure. Bar, 200 nm. (C) The number of synapses detected by EM increased sevenfold in the presence of glia: without glia, 0.4 ± 0.4; with glia, 2.8 ± 0.7 per cell body and proximal neurite (P = 0.013, Student's t test). (D) There are no differences in the total number of vesicles (34 ± 6 without glia, 42 ± 13 with glia) or docked vesicles (6.8 ± 2 without glia, 8.7 ± 0.8 with glia) per synapse. (E) Double labeling of superior colliculus (SC) reveals that synaptic puncta (green) appear around postnatal day 6 (P6), concurrently with the appearance of S100β-positive astrocytes (red, arrows). Bar, 50 μm.

To determine whether astrocytes increase synaptic stability, we tested whether the glial effect was reversible. We allowed large numbers of synapses to form by culturing autaptic RGCs with astrocyte-conditioned medium for 5 days, then removed the glia and cultured the RGCs for another 6 days. We found a significant reduction in both the QC and number of synaptic puncta after removal of glia. (33). The QC was 47 ± 12 before removal of glia and 14 ± 8 after removal of glia (P < 0.05). Postsynaptic GluR2/3 puncta that colocalized with presynaptic SV2 immunoreactivity on autaptic neurons were counted blind. The number of synaptic puncta was 41 ± 4 before removal of glia and 13 ± 1 after removal of glia (P < 0.001, Mann-Whitney U test). This fourfold reduction in the number of synapses upon removal of glia demonstrates that astrocytes significantly increase synaptic stability and are necessary for synapse maintenance.

These findings raise the question of whether glia also regulate synapse number in vivo. RGC axons reach their targets in the superior colliculus (SC) around embryonic day 16, yet form few synapses, as assessed by both ultrastructural and physiological studies, until around the end of the first postnatal week, as is true in most brain regions (31, 32). To determine whether there is a correspondence between the appearance of synapses and astrocyte development, we immunostained cryosections of the SC at various times during development. The appearance of SV2-positive synaptic puncta in the SC corresponds closely to the appearance and proliferation of S100β-positive astrocytes at the end of the first postnatal week (Fig. 4E). Taken together with our culture findings, the late appearance of synapses, only after glial generation, strongly suggests that astrocytes trigger the increase in synapse number within the CNS.

In summary, these results show that the total number of synapses on a neuron is not an intrinsic property of that neuron, but can be profoundly regulated by extrinsic signals. In the absence of glia, neurons in culture have only a limited ability to form synapses. Astrocytes greatly increase the number of structurally mature, functional synapses and are necessary to maintain synaptic stability. These findings have implications for the possible roles of glia in vivo. They suggest that during normal embryonic development, initially formed synapses may be immature and highly plastic, with the postnatal appearance of astrocytes serving to increase synapse number and lock synaptic circuitry in place. In addition, glia may play important and unexpected roles in adult neural plasticity underlying learning and memory. Finally, our results raise the question of whether the gliosis produced by brain injury increases synapse number, potentially leading to epilepsy and excitotoxic neurodegenerative changes.

  • * To whom correspondence should be addressed at Stanford University School of Medicine, Department of Neurobiology, Fairchild Science Building, Stanford, CA 94305–5125, USA. E-mail: emu{at}


View Abstract

Navigate This Article