Neuron, Vol. 16, 587–599, March, 1996, Copyright 1996 by Cell Press

Deficient Cerebellar Long-Term Depression,Impaired Eyeblink Conditioning, andNormal Motor Coordination in GFAP Mutant Mice

Katsuei Shibuki,*† Hiroshi Gomi,†‡ Lu Chen,§ neurons appear to migrate to their final destination alongShaowen Bao,§ Jeansok J. Kim,§ astrocytic processes (Rakic, 1981). Recent studies haveHidemitsu Wakatsuki,* Toshiyuki Fujisaki,* revealed that astrocytes contain neurotransmitter re-Kazushi Fujimoto,‖ Akira Katoh,‡ ceptors and transporters (for review, see Teichberg,Toshio Ikeda,‡ Chong Chen,# 1991). Neurotransmitters released from neurons depo-Richard F. Thompson,§ larize cultured astrocytes and produce changes in intra-and Shigeyoshi Itohara‡ cellular Ca21 concentration (Dani et al., 1992; Murphy et*Institute for Virus Research al., 1993; Mennerick and Zurumski, 1994). Conversely,Kyoto University a rise of Ca21 concentration in astrocytes can cause aSyogo-in rise of Ca21 concentration in surrounding neurons (Ned-Sakyo-ku, Kyoto 606-01 ergaard, 1994; Parpura et al., 1994; for review, see Att-Japan well, 1994). These findings raise possibilities that*Department of Neurophysiology astrocytes can modulate neuronal function.Brain Research Institute Glial fibrillary acidic protein (GFAP) is an intermediateNiigata University filament protein that is specifically expressed inAsahi-machi, Niigata 951 astrocytes. Removal of GFAP in a human astrocytomaJapan cell line by transfection with an antisense construct§Program for Neural, Informational, caused an impaired ability of the cells to extend glialand Behavioral Sciences processes upon neuronal induction (Weinstein et al.,University of Southern California 1991). Thus, it hasbeen predicted that structural supportLos Angeles, California 90089-2520

of astrocytic processes by GFAP may be critical for the‖Department of Anatomymorphogenesis of the CNS. However, the mice devoid

Faculty of Medicineof GFAP do not show any developmental abnormalitiesKyoto University(Gomi et al., 1995; Pekny et al., 1995). We decided toYoshidaexamine the possibility of astrocytic modulation of neu-Sakyo-ku, Kyoto 606-01ronal function by using the GFAP mutant mice. The cere-Japanbellar cortex provides a suitable model system for this#Howard Hughes Medical Institutepurpose. GFAP is highly expressed in Bergmann glia, aCenter for Learning and Memorysubset of astrocytes, in the cerebellar cortex. PurkinjeMassachusetts Institute of Technologycells (PCs), the sole output neurons of the cerebellarCambridge, Massachusetts 02139cortex, are innervated by two afferents: parallel fibers(PFs) originating from granule cells and climbing fibers(CFs) originating from inferior olive neurons. BergmannSummaryglia ensheathe dendrites and synapses of PCs. PF–PCsynapses undergo long-term depression (LTD) followingMice devoid of glial fibrillary acidic protein (GFAP), an

intermediate filament protein specifically expressed in repetitive activation of these synapses in conjunctionastrocytes, develop normally and do not show any with CF inputs (Ito et al., 1982; Ekerot and Kano, 1985;detectable abnormalities in the anatomy of the brain. Sakurai, 1987). Induction of LTD requires Ca21 entry toIn the cerebellum, excitatory synaptic transmission PCs mediated by the activation of voltage-gated Ca21

from parallel fibers (PFs) or climbing fibers (CFs) to channels that are activated by CF stimuli (for review,Purkinje cells is unaltered, and these synapses display see Linden, 1994). PF–PC synapses are thought tonormal short-term synaptic plasticity to paired stimuli use metabotropic and ionotropic a-amino-3-hydroxy-in GFAP mutant mice. In contrast, long-term depres- 5-methyl-4-isoxazolepropionate (AMPA) receptors (Forsion (LTD) at PF–Purkinje cell synapses is clearly defi- review, see Linden, 1994). Some isoforms of thesecient. Furthermore, GFAP mutant mice exhibited a sig- receptors, such as ionotropic glutamate receptor 1nificant impairment of eyeblink conditioning without (GluR1), are known to be expressed also on Bergmannany detectable deficits in motor coordination tasks. glia (Blackstone et al., 1992; Petralia and Wenthold,These results suggest that GFAP is required for com-

1992; Baude et al., 1994). Therefore, we focus on themunications between Bergmann glia and Purkinjerole of astrocytes in this synaptic plasticity.cells during LTD induction and maintenance. The data

The cerebellum is important for the control and finesupport the notion that cerebellar LTD is a cellulartuning of movements that may occur through a motormechanism closely associated with eyeblink condi-learning process involving synaptic plasticity (Ito, 1984;tioning, but is not essential for motor coordinationThompson et al., 1993). In eyeblink conditioning (a formtasks tested.of discrete motor learning) the conditioned stimulus (CS)and the unconditioned stimulus (US) converge on PCsIntroductionthrough PFs and CFs, respectively (Thompson, 1986).Thus, LTD is suggested as a putative synaptic mecha-Astrocytes are thought tobe important for the nutritionalnism for this form of motor learning (Thompson andand structural support of neurons. During development,Krupa, 1994; Chen and Thompson, 1995). Recently, anattempt to link LTD and eyeblink conditioning has been†The first two authors contributed equally to this work.

brought to you by COREView metadata, citation and similar papers at core.ac.uk

provided by Elsevier - Publisher Connector

Neuron588

made with use of metabotropic glutamate receptor 1 parasagittal sections, immunohistochemistry showedthat GFAP was abundant in the astrocytes of the granule(mGluR1)-deficient mice. However, mGluR1 mutant

mice exhibited profound motor discoordination, ataxia, cell layer and Bergmann glia of the molecular layer inthe wild-type mice (Figures 1A and 1C), while the GFAPand tremor (Aiba et al., 1994; Conquet et al., 1994). These

results make it difficult to specify the precise role of immunoreactivity was absent in the mutant mice (Figure1B). Since Bergmann glia in the cerebellum express vi-LTD in discrete motor learning.

In the present study, we examined the detailed anat- mentin (Schnitzer et al., 1981; Pixley et al., 1984; Gomiomy and synaptic function of the cerebellum in GFAP et al., 1995) and GluR1 (Petralia and Wenthold, 1992;mutant mice. After conjunctive stimulation of PF and CF, Baude et al., 1994), we evaluated the cellular localizationLTD is clearly deficient in the GFAP mutant cerebellum, of vimentin and GluR1. Bergmann glia in the mutantwithout any detectable anatomical abnormalities. Fur- cerebellum labeled with anti-vimentin antibody showedthermore, the mutant mice are impaired in eyeblink con- no gross differences from the wild-type cells (Figuresditioning, but not in motor coordination. These results 1D, 1E, and 1G). In addition, the distribution and densitysuggest that Bergmann glia may have a crucial role in of the mutant Bergmann glia labeled with anti-GluR1synaptic plasticity mechanisms. The specific correlation antibody was similar to that of the wild-type Bergmannbetween cerebellar LTD and eyeblink conditioning glia (Figures 1F and 1H). Furthermore, normal branchesstrongly supports the notion that LTD plays an important of PC dendrites were observed in the mutant cerebel-role in the eyeblink conditioning. lum, with Golgi impregnation (data not shown). Calbin-

din-D and inositol 1,4,5-triphosphate receptor type 1Results (IP3R1) are Ca21-binding proteins that express in PCs

(Iacopino et al., 1990; Furuichi et al., 1989). We foundthat the expression level and intracellular localization ofAnatomy of GFAP Mutant Cerebellum

We have previously described that the size and cellular these proteins in the mutant PCs were similar to that ofwild-type mice (Figures 1I–1L).composition of the brain of GFAP mutant mice were

apparently normal (Gomi et al., 1995). Here, we made Bergmann glial processes surround PC dendrites,spines, and axon terminals of the PFs and CFs (Petersfurther detailed analysis of the cerebellar structure. In

Figure 1. Histological Analysis of the GFAPMutant Cerebellum: No Detectable Abnor-mality in the Cytoarchitecture of the MutantCerebellum

Wild-type (A, C, E, F, I, and J) and mutant (B,D, G, H, K, and L) mice. (A) and (C) show theprocesses of wild-type Bergmann glia la-beled for GFAP. Note that the same stainingdid not show the processes of mutant Berg-mann glia (B). (D), (E), and (G) show the pro-cesses of Bergmann glia labeled for vimentin.Arrows in (C) and (D) indicate Bergmann gliawith higher magnification. (F) and (H) showthe cerebellar cortex labeled for GluR1. Mod-erate immunoreactivities were detected inthemolecular layer, whereas the soma of Berg-mann glia showed intense immunoreactivit-ies (indicated by arrowheads). (I) and (K)showthe cerebellar cortex labeled for calbindin-D.(J) and (L) show the cerebellar cortex labeledfor IP3R. Both calbindin-D and IP3R immuno-staining show the well-branched dendriticarbors in the mutant PCs. Scale bars: 125 mm(A and B), 50 mm (C to L).

Neuronal Plasticity and Astrocytes589

et al., 1991). We further analyzed the fine structures of expressed normally in the GFAP mutant mice by immu-nohistochemistry and Western blot analyses. Light mi-the cerebellar cortex in mutant mice with the electron

microscopy. In the cerebellar cortex, the presumptivePF croscopic immunohistochemistry with polyclonal anti-bodies (G18 and a), which are specific to an N-terminalterminals of granule cells formed asymmetric synapses

with the spines of PCs (Figures 2A and 2B). The numbers extracellular domain of all isoforms of mGluR1 and aC-terminal cytoplasmic domain of mGluR1a, respec-of the synapses per 21 mm2 counted on photographs

were 3.42 6 2.0 for a wild-type mouse (n 5 24) and tively (Shigemoto et al., 1994; S. Nakanishi, personalcommunication), showed immunoreactivities in the mo-3.49 6 1.6 for a mutant mouse (n 5 41); their difference

was not significant. The presumable CF terminals also lecular layer of mutant cerebellar cortex, indistinguish-able from those of wild-type cerebella (Figures 3A andmade synaptic contact with the multiple spines of PCs

in both wild-type and mutant mice (Figures 2C and 2D). 3B; data not shown). The specificities of the immunore-activities for mGluR were confirmed by the absence ofMoreover, the processes of mutant Bergmann glia en-

sheathed both PF and CF synaptic complexes as in antibody staining of cerebellar slices from an mGluR1-deficient mouse (Aiba et al., 1994; data not shown). Withthe case of wild-type processes (Figures 2A–2D). The

extracellular space between glial processes and a antibodies, the immunogold particles were accumu-lated in the intracellular side along the membranes ofneurons was almost constant, about 20 nm, in both

wild-type and mutant cerebella. The presynaptic and the dendritic spines of PCs (Figures 3C and 3D). Toevaluate the amount of mGluR1 expressed in the mutantpostsynaptic membrane structures, synaptic clefts, and

density of synapses in these two distinct types of syn- cerebella, we performed Western blot analysis of crudesynaptosomal membrane preparations from cerebellaapses in the mutantcerebellar cortex were indistinguish-

able from those in the wild-type cerebellar cortex. of age-matched wild-type and mutant mice. A majorband with an estimated molecular mass of 145 kDamGluR1 is known to play important roles in cerebellar

functions including LTD (Aiba et al., 1994; Conquet et (mGluR1a) was detected with a antibody in the prepara-tions (Figure 3E). With G18 antibody, an additional bandal., 1994). Thus, we examined whether mGluR1 was

Figure 2. Electron Microscopic Analysis ofPC Synapses in the GFAP Mutant Mice: Mu-tant Bergmann Glia Ensheathe PF and CFSynapses

Wild-type (A and C) and GFAP mutant (B andD) mice. (A) and (B) show the presumable PFsynapses with the spines of PCs. (C) and (D)show the axon terminals possibly belongingto the CF synapses with the multiple spines ofPCs. The processes of wild-type and mutantBergmann glia ensheathed these synapticcomplexes. The letter G indicates the pro-cesses of Bergmann glia. Arrowheads indi-cate asymmetric synapses. Scale bar,400 nm.

Neuron590

Figure 4. Effects of APV and CNQX on EPSPs in the Wild-Type andMutant PCs

(A) EPSPs evoked by PF stimulation before and during applicationof APV (50 mM) or CNQX (10 mM) in the wild-type and mutant PCs.(B) EPSPs and Ca21 spikes (asterisks) evoked by CF stimulation andeffects of APV or CNQX in the wild-type and mutant PCs.

Furthermore, the anatomical pattern of the cerebellardeep nuclei did not seem to be altered by the mutation(data not shown). In wild-type mice, astrocytes express-ing GFAP are scarce in the deep cerebellar nuclei. Insum, various histological analyses including Golgi stain-ing, electron microscopy, and immunochemical stain-ings (calbindin-D, IP3R1, GluR1, mGluR1, vimentin, andGFAP) revealed no morphological abnormalities in thecerebella of mutant mice, except for the lack of GFAPfilaments.

Excitatory Synaptic Transmission, Short-TermSynaptic Plasticity, and Dendritic Ca21 SpikesAre Normal in the Mutant PCsBy the fourth postnatal week, cerebellar developmentis completed, and thereafter, PCs receive PF inputs onthe well-developed dendritic arbor. To analyze synaptictransmission in the adult cerebellum (4–6 weeks old),we first recorded all-or-none excitatory postsynapticFigure 3. Unaltered Expression of mGluR1 in the GFAP Mutant Cer-potentials (EPSPs) from PC dendrites, which were fol-ebellalowed by Ca21 spikes, when CFs were stimulated (FigureImmunohistochemistry with anti-mGluR1a showed immunoreactivi-4B). Recordings in dendrites of PCs were identified byties in the molecular layers of wild-type (A) and GFAP mutant (B)

cerebella. Electron microscopic analysis revealed immunogold par- small (<3 mV) and spontaneous Na1 spikes (arrowheadsticles on the dendritic spines of wild-type (C) and GFAP mutant (D) in Figures 5A and 5B), which were attenuated actionPCs. The specificity of these analyses were confirmed by control potentials generated in the soma (Llinas and Sugimori,staining of a mGluR1 mutant cerebellum (Aiba et al., 1994) (data not

1980; Shibuki and Okada, 1992). We didnot use voltage–shown). Scale bars: 200 mm (A and B); 100 nm (C and D). (E) Westernclamp recording because of difficulty in space clampingblot analyses of crude synaptosomal membrane preparations fromof the intradendritic voltage. We also recorded gradedwild-type, GFAP mutant, and mGluR1 mutant cerebella. The anti-

bodies used and molecular weights in kilodaltons were indicated EPSPs evoked by PF stimulation (Figure 4A), whoseat right and left of the figures, respectively. amplitudes were gradually increased as the stimulus

intensity augmented.Synaptic responses in the wild-type and mutant PCsof 97 kDa (presumably mGluR1b and mGluR1c) was

detected (Figure 3E), as consistent with the results pub- were insensitive to D-2-amino-5-phosphonopentanoicacid (APV, 50 mM) but could be blocked by 6-cyano-7-lished previously (Shigemoto et al., 1994). We did not

observe quantitative differences of mGluR1 expression nitroquinoxaline-2,3-dione (CNQX, 10 mM; Figures 4Aand 4B), as reported previously (Llano et al., 1991; Aibabetween the preparations from mutant and wild-type

mice in the Western blot analysis. et al., 1994; Kashiwabuchi et al., 1995). No apparent

Neuronal Plasticity and Astrocytes591

showed no apparent effect on the Ca21 spikes in thewild-type and mutant cells (data not shown).

It is known that during the normal development ofthe rodent cerebellum, the adult one-to-one relationshipbetween CFs and PCs is preceded by a transient stageof multiple innervations (Crepel, 1982). Previously, othergene knockout mice including GluRd2 (Kashiwabuchi etal., 1995) and PKCg (Kano et al., 1995) were found tohave persistent multiple innervation of PCs by CFs inadulthood. In contrast, none of 164 PCs examined fromthe adult GFAP-deficient mice have shown multiple in-nervation. This indicated that the developmental transi-tion from multiple to single CF innervation of each PCis normal in GFAP mutant mice.

Cerebellar LTD Is Deficient in the Mutant MiceTo evoke LTD in the PF EPSPs, PFs and CFs werestimulated simultaneously (Ito et al., 1982; Sakurai, 1987)by pulse trains (five pulses at a 20 ms interval) deliveredevery 2 s for 10 min. This protocol of short train pulseswas chosen to allow maximal Ca21 influx by CF inputs(Shibuki and Okada, 1991; Aiba et al., 1994). During thestimulation, Ca21 spikes in PCs were monitored andwere confirmed to follow the stimuli without failure (Fig-ures 5C and 5D). Since the mutant mice were indistin-guishable from the wild-type mice in appearance, wewere able to carry out the experiments blind. Figure 6represents the data acquired from blind (wild type, n 57; mutant, n 5 5) and nonblind (wild type, n 5 7; mutant,Figure 5. Dendritic Ca21 Spikes in the Wild-Type and Mutant PCsn 5 5) experiments. The conjunctive (CJ) stimulation(A and B) Dendritic Ca21 spikes in the wild-type (A) and mutant (B)produced transient depression of PF EPSPs in the wild-cells evoked by depolarizing current (0.4 nA, 200 ms). Somatic Na1

spikes were indicated by arrowheads. type cells, and the depression was followed by LTD(C and D) Dendritic Ca21 spikes in the wild-type (C) and mutant (D) (Figures 6A and 6C; closed circles). Although the CJcells during the first 10 (top) and last 10 (bottom) conjunctive (CJ) stimulation transiently produced moderate depressiontrains of PF and CF used for LTD induction.

of PF EPSPs in the mutant cells, the depression lastedonly 10 min after the CJ stimulation. In contrast with LTD

difference was found in CF–PC or PF–PC EPSPs be- in wild-type mice, the mutant cells underwent reboundtween the mutant and wild-type littermates, suggesting potentiation following the initial depression (Figures 6Bthat postsynaptic responses were unaltered in the GFAP and 6D, closed circles). The relativechanges in the risingmutant mice. slope of PF EPSPs 30 min after the CJ stimulation were

Paired pulse facilitation and depression are a charac- 220.0% 6 5.6% (mean 6 SEM in 14 experiments) in theteristic feature of PF–PC and CF–PC synapses, respec- wild-type and 122.7% 6 8.3% (n 5 10) in the mutanttively (Konnerth et al., 1990; Aiba et al., 1994; Kashiwa- cells. The difference was significant (Mann–Whitney Ubuchi et al., 1995). These two types of short-term test, P < 0.01 for blind data, P < 0.01 for nonblind data,synaptic plasticity are believed to be caused by presyn- and P < 0.001 for both data). Between the 14 wild-typeaptic mechanisms (Zucker, 1989). Paired pulses at a 50 and 10 mutant cells, no significant difference was foundms interval potentiated the rising slope of PF EPSPs by in resting membrane potentials (-52 6 1 versus 250 642 6 5 % (n 5 6) in the wild-type and 46% 6 5% (n 5 1 mV), membrane resistance (65 6 6 versus 68 66) in the mutant cells. The slope of CF EPSPs, on the 7 MV), and series resistance of recording (65 6 6 versusother hand, was depressed by 26% 6 2% (n 5 6) in the 66 6 9 MV).wild-type and 26% 6 1% (n 5 6) in the mutant cells. No It is possible that the LTD deficiency observed maysignificant difference was found in these forms of short- be caused by instability of the mutant slices during re-term plasticity between the wild-type and mutant cells, cording. To test this possibility, we monitored PF–PCsuggesting that presynaptic mechanisms were not af- EPSPs with the conjunctive protocol but without PFfected in the GFAP mutant mice. stimulation. The CF stimulation alone did not induce any

Dendritic Ca21 spikes followed by CF EPSPs are the significant change of PF–PC EPSPs in either wild-typemain mechanisms of intracellular Ca21 increase required or mutant cells (Figures 6C and 6D, open circles). Thefor LTD (Sakurai, 1990; Konnerth et al., 1992). These relative changes in the rising slope of PF EPSPs 30 minCa21 spikes evoked by CF inputs (Figure 4B) or by depo- after the CF stimulation were 20.9% 6 3.1% (n 5 7) inlarizing current injection were normal in the mutant PCs the wild-type and 20.1% 6 7.6% (n 5 7) in the mutant(Figures 5A and 5B). Application of tetrodotoxin (0.5 cells. Thus, the stability of long-term recording in mutant

slices did not seem to differ from that in wild-type slices.mM), which blocks spontaneous somatic Na1 spikes,

Neuron592

Figure 6. Cerebellar LTD Is Deficient in theGFAP Mutant Mice

(A and B) EPSPs evoked by PF stimulation inthe wild-type (A) and mutant (B) mice. Stimu-lus pulses were delivered at a 12 s interval,and five traces were averaged for the mea-surement of the rising slope of EPSPs. Threesuccessive averaged traces immediately be-fore CJ and other 3 successive traces 30 minafter CJ (*) were superimposed for com-parison.(C and D) The time course of changes in theEPSP slope before and after CJ (closed cir-cles) or CF (open circles) stimulation in thewild-type (C) and mutant (D) PCs. Bars repre-sent the standard error of the mean. The ordi-nate shows the relative EPSP slope normal-ized by the averaged slope in threesuccessive traces immediately before CJ orCF stimulation. There were significant differ-ences between the two CJ groups (MannWhitney U-test, p<0.001).

Hippocampal LTP Is Normally Evokedin the Mutant MiceSince GFAP-positive astrocytes are also abundant inthe hippocampus (Gomi et al., 1995) and since synapticplasticity in thehippocampus has beenextensively stud-ied (Bliss and Collingridge, 1993), we analyzed long-term potentiation (LTP) of field EPSPs in the CA1 region.Schaffer collateral stimulation produced very similarfield EPSPs in the wild-type and mutant slices. LTP wasinduced by u-burst stimulation in both wild-type andmutant mice (data not shown). The amplitude of LTP 40min after u-burst stimulation was 39% 6 3% (n 5 39)in the wild-type and 48% 6 4% (n 5 39) in the mutantslices. Although the amplitude of LTP in the mutant micewas slightly larger than that in the wild-type mice, thedifference was not significant. Application of APV (25 or100 mM) blocked LTP in both the wild-type and mutantslices (data not shown). Thus, the deficiency of synapticplasticity in GFAP mutant mice appeared to be specificto cerebellar LTD but not to hippocampal CA1 LTP.

Associative Learning of EyeblinkResponse Is ImpairedThe study of mGluR1-deficient mice suggested that cer-ebellar LTD may be important for the associative learn-ing of the eyeblink response (Aiba et al., 1994). There-fore, we tested the capability of GFAP mutant mice ineyeblink conditioning. During eyeblink conditioning, aCS (tone 1 kHz, 352 ms, 80 dB) was paired with a perior-bital shock (100 ms, 100 Hz pulses) as a US. In each

Figure 7. Eyeblink Conditioning Is Partially Impaired in the GFAPsession (per day), an animal received 100 paired trials.Mutant Mice

Wild-type mice exhibited a gradual increase in the(A) Percent CRs of wild-type (n 5 11) and mutant (n 5 10) micefrequency of conditioned responses or CRs (percent CRduring 8 days of paired CS–US training followed by 4 days of CS-

or CR%) until the seventh session (approximately 70%). alone extinction training. For each paired training session, animalsIn contrast, mutant mice showed only a moderate in- were given 100 trials. A trial consisted of a 352 ms tone (CS) that

coterminated with a 100 ms periorbital shock (US) except that thecrease of CR% for first 3 sessions and plateaued (ap-first trial in each block was tone-alone trial and the sixth trial wasproximately 30%–40%) for subsequent sessions (Figureshock alone. During each extinction session, each animal was given7A). Although CRs in the first session did not seem to100 tone-alone trials.differ between the wild-type (n 5 11) and mutant (n 5(B) There is no group difference in shock sensitivity between wild-

10) mice, CRs in the seventh and eighth sessions were type and mutant mice. The level of the shock intensity (volts) wassignificantly lower in mutant mice (p < 0.05). The CR adjusted to elicit a minimally detectable head movement for each

animal before each session.disappeared during extinction trials, in which the CS

Neuronal Plasticity and Astrocytes593

Figure 8. Temporal Properties Are Normal inthe GFAP Mutant Mice

Accumulative histograms of the CR onset la-tency (A) and the CR peak latency (B). Boththe onset and peak latencies of CRs werecomputed using the criteria detailed in thematerials and methods. There were no statis-tical differences between the groups in bothtemporal measures of the CR.

alone was repeatedly presented to trained animals (Fig- then eventually learned to stay on the rod. By the fifthsession, most mice (16 of 22 wild type, and 16 of 23ure 7A; sessions 9–12), indicating the associative nature

of CRs. We did not observe significant differences be- mutant) learned to retain themselves on the rod for theentire observation period (Figure 9A). In the second pro-tween the wild-type and mutant mice in sensitivity to

the US (Figure 7B) as measured by the eyeblink/head tocol, we evaluated the number of trials required to learnthe task. In this protocol, we terminated the observationmovement UR during each training session. Thus, the

UR appears to be unaffected by the mutation. at 60 s. Trials were limited at most to seven times perThe wild-type and mutant mice showed similar modes

both in onset latencies of the CR, between 60–88 ms(Figure 8A), and in peak latencies of the CR, between60–228 ms (Figure 8B) after the onset of CS. We did notobserve significant differences between wild-type andmutant mice in these measures. This suggests that thekinetics of CRs was unaffected in the mutant mice.

Motor Coordination Is NormalDespite the apparent deficit in cerebellar LTD, the GFAPmutant mice did not show obvious cerebellar symptomssuch as ataxic gait and intention tremor. Spontaneousactivities of the adult wild-type and mutant mice in anew environment were recorded in an activity-tracingchamber that used infrared photobeams to monitor hori-zontal and vertical movements. We did not observe sig-nificant differences between the groups in any aspects,including horizontal and vertical movements (data notshown).

To compare further the motor coordination ability, weemployed the rotating rod task with two experimentalprotocols. In one experiment, the littermates (4–6 monthold) were given single trial per day for 5 consecutivedays (Figure 9A). The retention time of the mice on therotating rod was automatically recorded. The observa-

Figure 9. Motor Coordination Is Normal in GFAP Mutant Micetion was terminated at 300 s. Most mice either fromData shown is from the rotating rod task.mutant (male, n 5 138, female, n 5 10) or wild-type (male,(A) Wild-type (n 5 22) and mutant (n 5 23) were trained once a dayn 5 128, female, n 5 10) groups stayed on the rod nofor 5 consecutive days. Maximal allowed time was 300 s.

longer than 100 s at the first session (Figure 9A). Only (B) Wild-type (n 5 30) and mutant (n 5 30) mice were trained repeat-four from each group remained on the rod for 300 s. edly until they stayed on the rod longer than 60 s. Horizontal axis

represents the numbers of trials required.These mice initially rotated together with the rod and

Neuron594

day. Most mice required multiple trials on the first day which is known to play a crucial role in cerebellar func-tions (Aiba et al., 1994; Conquet et al., 1994), was nor-and a single trial on the third day. Both mutant (n 5 30)

and wild-type (n 5 30) female mice apparently learned mally expressed in the mutant cerebella (Figure 3). Thedata of Western blot analyses for synaptic moleculesthe task (Figure 9B). We did not observe any difference

between the two groups with both protocols. Since al- including mGluR1 (Figure 3E) strengthened the conclu-sion made from morphological analyses. Bergmann gliamost all mice who learned the task had experienced

rotating with the rod, it is obvious that their muscle were found to ensheathe these synapses in both thewild-type (Figures 2A and 2C) and mutant cerebella (Fig-strength influenced the results. Therefore, we carried

out a traction test. The mice used for the rotating rod ures 2B and 2D). These data strongly argue against theprevious notion that GFAP is crucial for the structuraltask were suspended by the front paws from a horizontal

wire. All mice were able to touch the wire with at least support of astrocytes. This conclusion is consistent withthe previous finding made on vimentin, another interme-one hind paw within a few seconds. We did not observe

any difference between the groups (data not shown). diate filament protein expressed in many types of cells(Colucci-Guyon et al., 1994).We further applied two motor tasks (rope climbing

and running along an elevated runway) that were suc-cessfully used to assess quantitatively the deficits in Cerebellar LTD Deficiency Is the Majormotor performance of the lurcher mutant mice (Goldo- Electrophysiological Abnormalitywitz et al., 1992) and GluRd2 mutant mice (Kashiwabuchi in the Mutant Cerebellumet al., 1995). In the rope climbing task, mice were placed In addition to normal cerebellar anatomy, we found thatat the base of a suspended piece of 50 cm twine with various electrophysiological parameters including CF orknots at z15 mm intervals, and the time required to PF EPSPs, pharmacological properties of the EPSPs,climb to the top was recorded. Mice were given three paired pulse facilitation and depression, and dendritictrials in a day. The mutant mice climbed the rope as fast Ca21 spikes were normal in the mutant PCs. In contrast,as wild-type mice (wild type [n 5 10]; 30.0 6 13.6 s; cerebellar LTD was clearly deficient: simultaneous stim-mutant [n 5 9]; 27.5 6 10.6 s). In the runway task, mice ulation of PFs and CFs failed to evoke LTD, but insteadrun along an elevated runway (100 cm long and 2 cm a slight potentiation was observed at the mutant PF–PCwide) with low obstacles at 6 cm intervals intended to synapses (Figures 6A–6D). However, in the hippocam-impede the progress of mice (Goldowitz et al., 1992). pus, where GFAP-positive astrocytes are abundant, LTPThe number of slips caused by slight lateral missteps induction was normal in the mutant mice. These resultswas counted during each trial. Mice were given four strongly suggest that communications between glia andtrials in a day. The mutant and wild-type mice quickly neurons, which are specifically required for cerebellarran through with few slips (wild type [n 5 9]; 22.4 6 8.8 LTD, are disrupted by the GFAP mutation. Since LTD iss and 1.8 6 1.6 slips/trial; mutant [n 5 9]; 19.5 6 9.9 s and evoked in the molecular layer of the cerebellar cortex,1.1 6 1.4 slips/trial]. We did not observe any difference the deficit of LTD in the GFAP mutant mice might bebetween mutant and wild-type mice in these tasks. primarily attributable to the defects in Bergmann cells.

These results indicated that the motor coordination, It has been previously shown that PF–PC synapsesas defined by the tasks employed here, is not impaired also show LTP when PFs were stimulated without CFin the mutant mice. stimulation (Sakurai et al., 1987). Thus, one can imagine

that the LTD deficiency in GFAP mutant mice is causedby augmented LTP as a result of PF stimulation. To testDiscussionthis possibility, we recorded PF–PC EPSPs under the PFstimulation without CF stimulation. The PF stimulationCerebellar Architecture Is Normalapplied in this study did not induce significant LTP inin GFAP Mutant Micethe mutant cells. The relative changes in the rising slopeIn the light microscopic analysis, no obvious abnormali-of PF EPSPs 30 min after the PF stimulation were 24.3%ties in cytoarchitecture were observed in the cerebellum6 3.7% (n 5 8) in the wild-type and 15.1% 6 7.5%of GFAP mutant mice. In the cerebellar cortex, the Berg-(n 5 10) in the mutant cells. The difference was notmann glia is the most characteristic subtype of astro-statistically significant. Thus, the results suggested thatcytes (Figures 1A–1D). This type of cell has its cell bodyLTP triggered by PF stimulation may only partly underliein the PC layer and has long processes that extendthe deficiency of LTD, if at all.through the molecular layer to the pia (Polak et al., 1982).

In immunohistochemical labeling of vimentin in the mu-tant Bergmann glia, their processes were observed to Cerebellar LTD and Eyeblink Conditioning

The importance of the cerebellum for classical eyeblinkspan upward through the molecular layer similar to thewild-type processes (Figures 1C, 1D, 1E, and 1G). This conditioning, an associative learning of discrete motor

responses, has been extensively characterized (for re-result suggests that the architecture of Bergmann gliais completed in the mutant mice without GFAP. views, see Thompson, 1986; Thompson and Krupa,

1994; Thompson et al., 1993). Two afferents to the cere-Golgi impregnation and immunohistochemical analy-sis of the mutant PCs showed well-branched dendritic bellum, mossy fibers originating from the pontine nu-

cleus and CFs emanating from the inferior olive, appeararbors (Figures 1I–1L). Furthermore, electron micro-scopic analysis of PC synapses revealed that the forma- to convey CS and US information, respectively, to the

cerebellum. Mossy fibers make excitatory synapses totion of the synapses with PFs and CFs was not alteredin the mutant cerebellum (Figures 2A–2D). mGluR1, granule cells whose axons are PFs that in turn make

Neuronal Plasticity and Astrocytes595

numerous excitatory synapses to PCs. The PCs are also 1994; Kashiwabuchi et al., 1995; for reviews, see Ma-lenka, 1994; Lisberger, 1995). These mutant mice bothinnervated by a single CF. In Larsell’s hemisphere lobule

VI (HVI), many PCs exhibit evoked responses to both exhibited deficient cerebellar LTD and severe cerebellarsymptoms, such as ataxic gait and intention tremor. Inthe tone CS and the corneal airpuff US (Thompson and

Krupa, 1994; Krupa, 1993). The convergence of CS and the rotating rod task, these mice were unable to keepthemselves on the rods (Aiba et al., 1994; KashiwabuchiUS information on PCs fulfills the condition required for

the induction of LTD (see Chen and Thompson, 1995). et al., 1995). However, our results strongly argue againstthis hypothesis. The GFAP mutant mice were indistin-Recently, Aiba et al. (1994) found that the training-re-

lated increase in the CR amplitude was markedly im- guishable from the wild-type mice in the rotating rod(Figures 9A and 9B), rope climbing, and runway tasks.paired in the mGluR1 mutant mice in comparison to

wild-type mice. Since some degree of eyeblink condi- The GFAP mutant mice learned the rotating rod task asefficiently as the wild-type mice did. So far, we have nottioning was still present in the mGluR1 mice, they con-

cluded that the LTD in the PF–PC synapses is a cortical noticed any deficit in their motor activities, except forthe eyeblink conditioning.mechanism important for the acquisition of the condi-

tioned eyeblink response but that it may not be es- It is interesting that adult GluRd2 mutant mice containmany PCs multiply innervated with CFs (Kashiwabuchisential.

Our results support the notion that LTD is a cellular et al., 1995). Thus, the regression of the multiple in-nervations, which is established during the normalsubstrate underlying eyeblink conditioning. We ob-

served that the conditioned eyeblink response in the course of development of the cerebellum, is apparentlyimpaired in the GluRd2 mice. These mutant mice alsoGFAP mutant mice is significantly impaired compared

with normal littermates (Figure 7A). Analogous to the contained fewer intact PF–PC synapses (Kashiwabuchiet al., 1995). Recently, two other strains of mutant micemGluR1 mutant mice (Aiba et al., 1994), some degree

of eyeblink learning appeared in the GFAP mutant mice deficient in mGluR1 and PKCg were found to have per-sistent multiple CF innervation as well as impaired motor(Figure 7A). The mutant mice displayed significant in-

crease in the conditioned eyeblink response over days, coordination (Kano et al., 1995; Chen et al., 1995). Incontrast, we did not observe any of these abnormalitiesbut the asymptote of conditioning was significantly

lower than that of wild-type mice. Thus, it appears that in the GFAP mutant mice. Perhaps CF innervation playsa key role in the deficits of motor coordination observedthe maximal level rather than the rate of learning is im-

paired in the GFAP mutant mice. The difference in eye- in other mutants (Chen et al., 1995).blink conditioning is not due to either sensory or motorperformance deficits, since the GFAP mutant mice did Putative Roles of Bergmann Glianot differ from the wild-type mice in periorbital shock on Neuronal Plasticitysensitivity (Figure 7B) nor residual UR to the shock (data Considering the cell-type specificity of GFAP and thenot shown). Furthermore, temporal properties of the CR lack of detectable developmental deficits of neurons,onset and peak latencies (Figures 8A and 8B) suggested we propose that the disrupted communication betweenthat the CS, US, and CR circuitry itself was functional PCs and Bergmann glia causes the LTD deficiency inin the mutant mice. GFAP mutant mice. As mentioned, expression of the

Substantial but variable, and generally not complete, GluR1 subunit of the AMPA-type receptors was demon-impairment in eyeblink conditioning has been reported strated in the membrane of Bergmann glia of rodentsfor rabbits with lesions limited to cerebellar cortex (La- (Blackstone et al., 1992; Petralia and Wenthold, 1992;vond et al., 1987; Lavond and Steinmetz, 1989; Yeo et Baude et al., 1994; Figures 1F and 1H). The Bergmannal., 1985b). In contrast, appropriate lesions or reversible glia also express GluR3 and GluR4 subunits (Burnashevinactivations of the interpositus nucleus completely and et al., 1992) and mGluR3 (Ohishi et al., 1993, 1994). Thus,permanently prevent acquisition and abolish retention it is possible that the excitatory transmitter glutamateof the conditioned eyeblink response (Clark et al., 1992; released from axon terminals diffuses out of synapticKrupa et al., 1993; Lincoln et al., 1982; McCormick et clefts to activate some of these receptors on Bergmannal., 1982; Steinmetz et al., 1992; Welsh and Harvey, 1989; glia. It has been shown that the activation of AMPA-Yeo et al., 1985a). One possible interpretation of these type GluR channels on cultured Bergmann glial cellsdata is that memory traces for eyeblink conditioning are exhibit high Ca21 permeability (Burnashev et al., 1992).formed and stored in both cerebellar cortex (LTD) and The Ca21 influx in Bergmann glia may trigger the releaseinterpositus nucleus (perhaps LTP, as in Racine et al., of some diffusible factors, such as homocysteate, an1986) and that the cortical memory trace, while not es- endogenous transmitter-like substance present in Berg-sential for some degree of learning, plays a critically mann glia (Cuenod et al., 1990; Grandes et al., 1991). Ifimportant role in normal learning (Thompson and Krupa, so, these factors released from Bergmann glia may act1994). on PCs or other neurons and, hence, modulate neuronal

functions such as LTD.

Cerebellar LTD and Motor CoordinationExperimental ProceduresCerebellar LTD has also been suggested to mediate

other aspects of motor learning, such as the develop-Production of GFAP Mutant Mice

ment of motor coordination (Ito, 1984). This hypothesis Mutant mice were produced as described previously (Gomi et al.,is strengthened by the studies of mGluR1-deficient and 1995). Mutant and wild-type mice used were the littermates inter-

crossed between male and female heterozygotes that had beenGluRd2-deficient mice (Aiba et al., 1994; Conquet et al.,

Neuron596

backcrossed to C57BL/6J mice for one to three generations. The Reichert Optishe AG). The sections were transferred to Formvar-coated grids, stained with rabbit anti-mGluR1 antibody a (1:450)genotypes of the mice were determined by Southern blot analysis

of DNA prepared from tails, as described previously (Gomi et al., overnight at 48C. After washing three times in PBS, the sectionswere incubated with 15 nm gold-conjugated goat anti-rabbit IgG1995).(Amersham; 1:20) for 1 hr at room temperature. After immunolabel-ing, they were postfixed with 1% glutaraldehyde, stained with uranylLight Microscopyacetate, and embedded in methylcellulose. Observations wereAnimals were deeply anesthetized with Nembutal, and brains weremade with a JEM 2000EX.fixed by intracardiac perfusion with neutral buffered formalin, or

fixed by immersion with 96% ethanol. After the fixation, sampleswere dehydrated and embedded in paraffin. Sections (4–8 mm) were Immunoblot Analysis

Crude synaptosomal membranes were prepared according toused for Nissl or immunohistochemical staining. For mGluR1 immu-nohistochemistry, brains were perfused with 3.5% paraformalde- Blackstone et al. (1992) and analyzed by immunoblotting. The mice

cerebella were homogenized in 10 vol of a solution containing 4 mMhyde, 1% picric acid, and 0.1% glutaraldehyde in 0.1 M phosphatebuffer (PB; pH 7.3), cryoprotected in 20% sucrose in PB overnight HEPES–NaOH (pH 7.4), 0.32 M sucrose, 1 mM EDTA, 5 mM EGTA,

0.1 mM PMSF, and 10 mg/ml leupeptin. The homogenates wereat 48C, and cut on a freezing microtome into 40 mm sections, asdescribed previously (Shigemoto et al., 1994). centrifuged at 800 3 g for 10 min at 48C, and the supernatants

were then centrifuged at 9000 3 g for 15 min at 48C. The pelletsFor Nissl staining, deparaffinized slides were immersed in cresylfast violet solution at 378C for 30 min. The immunohistochemical were washed with the homogenized solution, and the final pellets

were suspended in distilled water, stored on ice for 15 min, anddetection was performed as described previously (Gomi et al., 1995),with an avidin–biotin–peroxidase technique using the Vectastain added equal volume of 8 M urea. After incubation at 378C for 15

min, samples were mixed with an equal volume of 2 3 SDS sampleABC kit (Vector Laboratory), except for the immunohistochemistryof IP3R1. Anti-GFAP (GF 12.24) monoclonal antibody was purchased buffer (0.25 M Tris–HCl (pH 6.8), 10% SDS, 40% glycerol, and 10%

2-mercaptoethanol, and 5 mg of protein per well was separated byfrom Progen Biotechnic. Anti-vimentin (VIM 3B4) monoclonal anti-body was purchased from Cymbus Bioscience Limited. Anti-calbin- 7.5% or 12% SDS–polyacrylamide gel electrophoresis (Laemmli,

1970) and transferred to Immobilon-P membrane (Millipore). Afterdin-D (CL-300) monoclonal antibody was purchased from Sigma.Anti-GluR1 (AB 1504) polyclonal antisera were purchased from blocking the membrane with 5% skim milk in 20 mM Tris- buffered

saline (TBS; pH 7.4), the blots were immersed in mouse monoclonalChemicon International, Incorporated. Rabbit anti-mGluR1 antibod-ies a and G18, which are specific to the C-terminal domain of antibodies to synaptophysin (Cymbus Bioscience Limited; 1:100)

and syntaxin (Sigma; 1:2000), or in rabbit anti-mGluR1 antibodies amGluR1a and the N-terminal domain of all isoforms of mGluR1,respectively (Shigemoto et al., 1994; S. Nakanishi, personal commu- (1:1000) and G18 (1 mg/ml). After washing three times in TBS con-

taining 0.05% Tween 20, the blots were immersed in horseradishnication), were provided by S. Nakanishi. Anti-IP3R1 monoclonalantibody 18A10 (Maeda et al., 1988) were provided by K. Mikoshiba. peroxidase (HRP)-conjugated rat anti-mouse IgG (Zymed Labora-

tories Incorporated; 1:5000) or goat anti-rabbit IgG (Vector Labora-For the detection of IP3R, FITC-labeled goat anti-rat IgG (Caltag,R40101) was used as the secondary antibody. tory; 1:5000). Reacted signals were developed in enhanced chemilu-

minescence (ECL) Western blotting detection reagents (Amersham).For Golgi impregnation, the fresh cerebellar tissues were im-mersed in Golgi-Cox solution (solution A, 16 ml of 5% potassiumchromate and 40 ml of distilled water; solution B, 20 ml of 5% Electrophysiology in Cerebellar Slicespotassium dichromate and 20 ml of 5% mercuric chloride; solutions Cerebellar slices were prepared from the wild-type and mutant miceA and B are combined immediately prior to immersion). The tissues (4–6 weeks old). Animals were anesthetized with ether, and im-were hardened in the solution in the dark for 1 day. The solution mersed in ice-cooled water for 1.5–2.5 min to reduce brain tempera-was changed and stored at 378C for 2 months. Alternatively, the ture. Immediately after decapitation, the cerebellum was isolatedformalin-fixed brain samples were refixed with chromate solution and placed into an ice-cooled artificial medium bubbled with 95%consisting of 30 ml of 3% potassium dichromate, 10 ml of 96% O2 and 5% CO2. The composition of the medium was: 124 mM NaCl,ethanol, and 10 ml of acetic acid at 378C for 1 day. After rinsing 5 mM KCl, 1.24 mM NaH2PO4, 1.3 mM MgSO4, 2.4 mM CaCl2, 26with distilled water, samples were further immersed in chromate mM NaHCO3, and 10 mM glucose. Parasaggital slices (thickness;solution at room temperature for 4 days and with 3% potassium 400 mm) of the vermis were prepared with a microslicer (Dosaka,dichromate for 5 days. Samples were impregnated with 0.5% silver DTK-2000). Slices were incubated at room temperature or 308C fornitrate solution for 1 hr, with 1.0% silver nitrate solution for 4 days, at least 30 min before recording.and with 2.0% silver nitrate solution for 5 days. The impregnated Voltage recording from PC dendrites was done at 308C 6 18C intissues were embedded in 10% gelatin and cut for 20–50 mm cryo- the presence of picrotoxin (0.1 mM) by using blind and perforatedsection. slice patch technique (Rae et al., 1991). As recording electrodes,

glass micropipettes (8–12 MV) were filled with a medium containing92 mM K2SO4, 31 mM KCl, 1 mM MgCl2, 10 mM HEPES (pH 7.4 withElectron Microscopy

To perform electron microscopy, animals were anesthetized with KOH). Amphotericin B (Wako) was added to the filling medium toachieve a final concentration of 50 mg/ml in 0.5% dimethylsulfoxide,Nembutal and fixed by perfusion with 100 ml of 2% paraformalde-

hyde and 2.5% glutaraldehyde (2PA/2.5GA) in 0.1 M PB (pH 7.3) at although the tip of micropipettes was filled with a small amountof amphotericin B–free medium to facilitate giga sealing. In the48C, followed by perfusion with 0.9% saline solution (10 ml). After

perfusion, the brains were dissected to small blocks and postfixed experiments shown in Figures 4A and 4B, recording pipettes werefilled with a medium containing 140 mM CsCl, 2 mM MgCl2, 0.5 mMin 2PA/2.5GA in 0.1 M PB for 1 hr at 48C. Tissue blocks were washed

with 0.1 M phosphate buffered saline (PBS) at 48C three times for EGTA, 4 mM NaATP, 0.3 mM GTP, 10 mM HEPES (pH 7.3; Konnerthet al., 1992), and whole-cell recording was done instead of20 min, and embedded in 2% low melting agarose in 0.1 M PBS.

Parasagittal sections (100 mm) were made on a microslicer (Dosaka, perforated patch technique.To stimulate CFs, a cut surface of Teflon-coated platinum wireDTK-1000) and postfixed with 2% osmium tetroxide in 0.1 M PBS

at 48C for 1 hr. The sections weredehydrated, infiltrated, and embed- (metal diameter; 75 mm) was placed on the white matter, and nega-tive pulses (duration, 300 ms) were applied. To stimulate PFs, weded in plastic resin. Ultrathin sections were cut on an LKB ultrami-

crotome, stained with uranyl acetate and lead citrate, and subjected filled glass micropipettes (resistance <1 MV) with 2 M NaCl andinserted them into the molecular layer so that the tip was 300 mmto electron microscopic analysis in a JEM 2000EX (JEOL Company).

For mGluR1 cryomicrotomy and immunolabeling, cerebellar sam- deeper than the surface. Biphasic stimulus pulses (each phase ofa pulse; 100 ms) were used. CFs and PFs were stimulated simultane-ples fixed with 3.5% paraformaldehyde, 1% picric acid, and 0.1%

glutaraldehyde in 0.1 M PB were cryoprotected with 2.3 M sucrose, ously by pulse trains (5 pulses at a 20 ms interval) delivered every2 s for 10 min to evoke LTD. This protocol of short train pulses wasplaced on specimen stubs, and frozen in liquid nitrogen. Ultrathin

cryosections (100 nm thick) were cut with a Richert ultramicrotome chosen to allow maximal Ca21 influx by CF inputs (Shibuki andOkada, 1991; Aiba et al., 1994). The intensity of CF stimulation (<30Ultracut E with a low temperature sectioning system FC-4D (C.

Neuronal Plasticity and Astrocytes597

mA) was adjusted to be 10 mA larger than the threshold of CF re- training, the CR percentage was calculated using both paired andtone-alone trials.sponses. During the stimulation, Ca21 spikes in PCs were monitored

and were confirmed to follow the stimuli without failure. The intensity For all CR trials, the onset and peak latencies of CRs were calcu-lated in the following manner: the CR onset was determined as theof PF stimulation (<20 mA) was adjusted so that the rising slope of

EPSPs in PCs was stabilized between 4 and 17 V/s. first time, during the period between the CR onset and the US onset,at which the average unit count of any consecutive seven bins (28ms) was higher than the average 1 SD 11 of pre-CS period; the

Electrophysiology in Hippocampal Slices CR peak latency was computed as the time, between the CS andThe condition of the hippocampal slice experiments was similar to the US onsets, at which the average unit count of any consecutivethat of the cerebellar slice experiments unless otherwise specified. seven bins was the highest. Accumulative histograms of the onsetSlices were prepared from mice between 6 and 12 weeks old. No and peak latencies in the last paired training session for both mutantpicrotoxin was used during recording. Field EPSPs were recorded and wild-type groups are shown in Figure 8.from the CA1 region through electrolytically polished silver elec-trodes that were insulated with polyvinyl chloride except at the tip. Motorod TestSchaffer collaterals were stimulated with biphasic pulses (duration Running rotorod tests were performed on the 4–6 months-old ani-of each phase, 200 ms) through glass micropipettes filled with 2 M mals. The rotorod (Ugo Basile Biological Research Apparatus) con-NaCl. The pulse intensity (<35 mA) was adjusted so that the ampli- sists of a gritted plastic roller (3 cm diameter, 6 cm long) flankedtude of field EPSPs were between 1 and 2 mV. LTP was evoked by by two round plates (30 cm in diameter) to prevent animals fromthe theta burst stimulation (TB, Larson et al., 1986) in which Schaffer escaping. The roller is driven at 12 rpm by a motor. A mouse wascollaterals were stimulated with pulse trains (5 pulses at a 10 ms placed on the roller, and the time it remained on the roller wasinterval) delivered every 200 ms for 20 times. automatically measured. All experiments were conducted during

the light portion of a 12 hr/12 hr light–dark cycle.

Eyeblink ConditioningAcknowledgmentsMouse eyeblink conditioning was adapted from a method for condi-

tioning freely moving rats (Weiss and Thompson, 1991). Male wild-We wish to thank all the members of the Itohara lab for their encour-type and GFAP mutant mice (3 month old) were individually housed,agement, and for maintaining mice; T. Kaneko, A. Aiba, and H. Ohishiwith food and water available ad libitum. All experiments were con-for their helpful discussions, and T. Hirano and A. Silva for theirducted during the light portion of a 12 hr/12 hr light–dark cycle.comments on the manuscripts. We wish to thank K. Mikoshiba, S.Under ketamine (80 mg/kg; intraperitoneal) and xylazine (20 mg/Nakanishi, and A. Aiba for providing monoclonal antibody to IP3R1,kg) anesthesia, animals were implanted with four wires (0.0045“polyclonal antibodies to mGluR1, and mGluR1-deficient mice, re-teflon-coated stainless steel, 0.003” bare) subcutaneously to thespectively. H. G., T.I., S. I., and C. C. wish to thank S. Tonegawa forleft upper eyelid. The wire tips were exposed and two wires werehis invaluable help and continuous encouragement. This work wasused to record differential electromyograph (EMG) from obicularissupported by grants from the Shionogi Institute for Medical Science,oculi, and the other two wires were used to deliver periorbital shock.the Japanese Ministry of Education, Science,and Culture, and Japa-A four pin strip connector, to which the wires were soldered, wasnese Ministry of Public Welfare to S. I.; grants from Japanese Minis-cemented to skull with dental acrylic.try of Education, Science, and Culture, Japanese Ministry of PublicDuring habituation and training, the subjects were placed in aWelfare, Uehara Memorial Foundation, Toyota RIKEN, and Nissansmall cylindrical plexiglas (diameter, 3.75“) that was placed insideScience foundation to K. S., grants from the Howard HughesMedicala sound- and light-attenuating chamber. The strip connector wasInstitute to C. C.; and grants from the National Science Foundation,attached to a mating plug of the tether commutator that has chan-the National Institute on Aging, and Sankyo to R. F. T.nels to relay EMG signals and deliver shocks. After 1 day of habitua-

The costs of publication of this article were defrayed in part bytion, mice were given 8 days of CS–US paired training followed bythe payment of page charges. This article must therefore be hereby4 days of CS alone extinction training. The daily paired trainingmarked “advertisement” in accordance with 18 USC Section 1734session consisted of 100 trials arranged into 10 blocks. Each blocksolely to indicate this fact.included 1 CS alone (the first trial), 1 US alone (the sixth trial), and

8 CS–US paired trials. The CS and US were a 352 ms tone (1000Received October 26, 1995; revised December 21, 1995.Hz, 80 dB with 5 ms rise/fall time) and a 100 ms shock (100 Hz

biphasic square pulses), respectively, with a 250 ms interstimulusReferencesinterval (ISI) and randomized intertrial interval between 20–40 s

(mean, 30 s). The US intensity, to elicit a robust eyeblink and mini-Aiba, A., Kano, M., Chen, C., Stanton, M.E., Fox, G.D., Herrup, K.,mally detectable head movements for each animal (Skelton, 1988),Zwingmann, T.A., and Tonegawa, S. (1994). Deficient cerebellarwas adjusted daily (mean values ranging from 10 to 35 V) becauselong-term depression and impaired motor learning in mGluR1 mu-the effectiveness of the US appears to decline over days (e.g., duetant mice. Cell 79, 377–388.to habituation, increases in electrode impedance, or both). For ex-Attwell, D. (1994). Glia and neurons in dialogue.Nature 369, 707–708.tinction, each animal was presented with 100 tone-alone trials in

each session. Baude, A., Molnar, E., Latawiec, D., Mclhinney, R.A.J., and Somogyi,The EMG signal was amplified in the band of 300–5000 Hz and P. (1994). Synaptic andnonsynaptic localization of the GluR1subunit

sent to the window discriminator in which the number of pulses of the AMPA-type excitatory amino acid receptor in the rat cerebel-above the noise envelope was sampled and stored by a computer lum. J. Neurosci. 14, 2830–2843.into 4 ms bin. CRs and bad trials were determined trial by trial using Blackstone, C.D., Moss, S.J., Martin, L.J., Levey, A.I., Price, D.L.,following criteria: the trial was considered bad and excluded from and Huganir, R.L. (1992). Biochemical characterization and localiza-the analysis if, during the 252 ms pre-CS period, the average unit tion of a non-N-methyl-D-aspartate glutamate receptor in rat brain.count per bin is higher than 2 (suggesting high activity), the standard J. Neurochem. 58, 1118–1126.deviation (SD) of unit counts per bin is larger than the average unit

Bliss, T.V., and Collingridge, G.L. (1993). A synaptic model of mem-count per bin (suggesting unstable EMG activity), or the average

ory: long-term potentiation in the hippocampus. Nature 361, 31–39.unit count within 28 ms (7 bins) after CS onset is higher than the

Burnashev, N., Khodorova, A., Jonas, P., Helm, P.J., Wisden, W.,average 1 SD of pre-CS period (suggesting a short latency startleMonyer, H., Seeburg, P.H., andSakmann, B. (1992). Calcium-perme-response); the CR was defined as the average unit count of anyable AMPA-kainate receptors in fusiform cerebellar glial cells. Sci-consecutive seven bins during the period from 84 ms after toneence 256, 1566–1570.onset to shock onset that is higher than the average 1 SD 1 1 of

pre-CS period. In tone alone trials, the period that is used to score Chen, C., and Thompson, R.F. (1995). Temporal specificity of long-term depression in parallel fiber–Purkinje synapses in rat cerebellarthe CR is extended to the offset of the tone. In extinction, the CR

percentage was calculated using tone-alone trials, while in paired slice. Learning Memory 185, 185–198.

Neuron598

Chen, C., Kano, M., Abeliovich, A., Chen, L., Bao, S., Kim, J.J., Krupa, D.J. (1993). Localization of the essential memory trace for aclassically conditioned behavior. PhD thesis. University of SouthernHashimoto, K., Thompson, R.F., and Tonegawa, S. (1995). Impaired

motor coordination correlates with persistent multiple climbing fiber California, Los Angeles, California.innervation in PKCg mutant mice. Cell 83, 1233-1242. Krupa, D.J., Thompson, J.K., and Thompson, R.F. (1993). Localiza-

tion of a memory trace in the mammalian brain. Science 260,Clark, R.E., Zhang, A.A., and Lavond, D.G. (1992). Reversible lesions989–991.of the cerebellar interpositus nucleus during acquisition and reten-

tion of a classically conditioned behavior. Behav. Neurosci. 106, Laemmli, U.K. (1970). Cleavage of structural proteins during theassembly of the head of bacteriophage T4. Nature 227, 680–685.879–888.

Larson, J., Wong, D., and Lynch, G. (1986). Patterned stimulationColucci-Guyon, E., Portier, M.-M., Dunia, I., Paulin, D., Pournin, S.,at the theta frequency is optimal for the induction of hippocampaland Babinet, C. (1994).Mice lacking vimentin develop and reproducelong-term potentiation. Brain Res. 368, 347–350.without an obvious phenotype. Cell 79, 679–694.Lavond, D.G., and Steinmetz, J.E. (1989). Acquisition of classicalConquet, F., Bashir, Z.I., Davies, C.H., Daniel, H., Ferraguti, F., Bordi,conditioning without cerebellar cortex. Behav. Brain Res. 33,F., Franz-Bacon, K., Reggiani, A., Matarese, V., Conde, F., Colling-113–164.ridge, G.L., and Crepel, F. (1994). Motor deficit and impairment ofLavond, D.G., Steinmetz, J.E., Yokaitis, M.H., and Thompson, R.F.synaptic plasticity in mice lacking mGluR1. Nature 372, 237–243.(1987). Reacquisition of classical conditioning after removal of cere-

Crepel, F. (1982). Regression of functional synapses in the immature bellar cortex. Exp. Brain Res. 67, 569–593.mammalian cerebellum. Trends Neurosci. 5, 266–269.

Lincoln, J.S., McCormick, D.A., and Thompson, R.F. (1982). Ipsilat-Cuenod, M., Do, K.Q., and Streit, P. (1990). Homocysteic acid as eral cerebellar lesionsprevent learning of the classically conditionedan endogenous excitatory amino acid. Trends Pharmacol. Sci. 11, nictitating membrane/eyelid response. Brain Res. 242, 190–193.477–478. Linden, D.J. (1994). Long-term synaptic depression in the mamma-

lian brain. Neuron 7, 81–89.Dani, J.W., Chernjavsky, A., and Smith, S.J. (1992). Neuronal activitytriggers calcium waves in hippocampal astrocyte networks. Neuron Lisberger, S.G. (1995). A mechanism of learning found? Curr. Biol.8, 429–440. 5, 221–224.

Llano, I., Marty, A., Armstrong, C.M., and Konnerth, A. (1991). Synap-Ekerot, C.F., and Kano, M. (1985). Long-term depression of paralleltic- and agonist-induced excitatory currents of Purkinje cells in ratfibre synapses following stimulation of climbing fibres. Brain Res.cerebellar slices. J. Physiol. 434, 183–213.342, 357–360.

Llinas, R., and Sugimori, M. (1980). Electrophysiological propertiesFuruichi, T., Yoshikawa, S., Miyawaki, A., Wada, K., Maeda, N., andof in vitro Purkinje cell dendrites in mammalian cerebellar slices. J.Mikoshiba, K. (1989). Primary structure and functional expressionPhysiol. 305, 197–213.of the inositol 1,4,5-triphosphate-binding protein P400. Nature 342,Maeda, N., Niinobe, M., Nakahira, K., and Mikoshiba, K. (1988).32–38.Purification and characterization of P400 protein, a glycoproteinGoldowitz, D., Moran, T.H., and Wetts, R. (1992). Mouse chimerascharacteristic of Purkinje cell, from mouse cerebellum. J. Neuro-in the study of genetic and structural determinants of behavior.chem. 51, 1724–1730.In Techniques for the Genetic Analysis of Brain and Behavior, D.Malenka, R.C. (1994). Mucking up movements. Nature 372, 218–219.Goldowitz, D. Wahlsten, and R.E. Wimer, eds. (Amsterdam: Elsevier),

pp. 271–290. McCormick, D.A., Clark, G.A., Lavond, D.G., and Thompson, R.F.(1982). Initial localization of the memory trace for a basic form ofGomi, H., Yokoyama, T., Fujimoto, K., Ikeda, T., Katoh, A., Itoh, T.,learning. Proc. Natl. Acad. Sci. USA 79, 2731–2735.and Itohara, S. (1995). Mice devoid of the glial fibrillary acidic proteinMennerick,S., andZorumski, C.F. (1994). Glialcontributions to excit-develop normally and are susceptible to scrapie prions. Neuron 14,atory neurotransmission in cultured hippocampal cells. Nature 368,29–41.59–62.

Grandes, P., Quang, Do, K., Morino, P., Cuenod, M., and Streit, P.Murphy, T.H., Blatter, L.A., Wier, W.G., and Baraban, J.M. (1993).

(1991). Homocysteate, an excitatory transmitter candidate localizedRapid communication between neurons and astrocytes in primary

in glia. Eur. J. Neurosci. 3, 1370–1373.cortical cultures. J. Neurosci. 13, 2672–2679.

Iacopino, A.M., Rhoten, W.B., and Christakos, S. (1990). Calcium- Nedergaard, M. (1994). Direct signaling from astrocytes to neuronsbinding protein (calbindin-D 28K) gene expression in the developing in cultures of mammalian brain cells. Science 263, 1768–1771.and aging mouse cerebellum. Mol. Brain Res. 8, 283–290. Ohishi, H., Shigemoto, R., Nakanishi, S., and Mizuno, N. (1993).Ito, M. (1984). The Cerebellum and Neural Control (New York: Raven Distribution of the mRNA for a metabotropic glutamate receptorPress). (mGluR3) in the rat brain: an in situ hybridization study. J. Comp.

Neurol. 335, 252–266.Ito, M., Sakurai, M., and Tongroach, P. (1982). Climbing fibre-in-Ohishi, H., Ogawa-Meguro, R., Shigemoto, R., Kaneko, T., Nakanishi,duced depression of both mossy fibre responsiveness and gluta-S., and Mizuno, N. (1994). Immunohistochemical localization of met-mate sensitivity of cerebellar Purkinje cells. J. Physiol. 324, 113–134.abotropic glutamate receptors, mGluR2 and mGluR3, in rat cerebel-

Kano, M., Hishimoto, K., Chen, C., Abeliovich, A., Aiba, A., Kurihara, lar cortex. Neuron 13, 55–66.H., Watanabe, M., Inoue, Y., and Tonegawa, S. (1995). Impaired

Parpura, V., Basarsky, T.A., Liu, F., Jeftinija, K., Jeftinija, S., andsynapse elimination during cerebellar development in PKCg mutant

Haydon, P.G. (1994). Glutamate-mediated astrocyte-neuron signal-mice. Cell 83, 1223-1231.

ling. Nature 369, 744–747.Kashiwabuchi, N., Ikeda, K., Araki, K., Hirano, T., Shibuki, K., Taka- Pekny, M., Leveen, P., Pekna, M., Eliasson, C., Berthold, C.H., Wet-yama, C., Inoue, Y., Kutsuwada, T., Yagi, T., Kang, Y., Aizawa, S., ermark, B., and Betsholtz, C. (1995). Mice lacking glial fibrillary acidicand Mishina, M. (1995). Impairment of motor coordination, Purkinje protein display astrocytes devoid of intermediate filaments but de-cell synapse formation and cerebellar long-term depression in velop and reproduce normally. EMBO J. 14, 1590–1598.GluRd2 mutant mice. Cell 81, 245–252.

Peters, A., Palay, S.L., and Webster, H. de F. (1991). The Fine Struc-Konnerth, A., Llano, I., and Armstrong, C.M. (1990). Synaptic cur- ture of the Nervous System: Neurons and Their Supporting Cells,rents in cerebellar Purkinje cells. Proc. Natl. Acad. Sci. USA 87, Third Edition (New York: Oxford University Press).2662–2665. Petralia, R.S., and Wenthold, R.J., (1992). Light and electron immu-

nocytochemical localization of AMPA-selective glutamate receptorsKonnerth, A., Dreessen, J., and Augustine,G.J. (1992). Briefdendriticin the rat brain. J. Comp. Neurol. 318, 329–354.calcium signals initiate long-lasting synaptic depression in cerebel-

lar Purkinje cells. Proc. Natl. Acad. Sci. USA 89, 7051–7055. Pixley, S.K.R., Kobayashi, Y., and de Vellis, J. (1984). A monoclonal

Neuronal Plasticity and Astrocytes599

antibody against vimentin: characterization. Dev. Brain Res. 15,185– Zucker, R.S. (1989). Short-term synaptic plasticity. Annu. Rev. Neu-rosci. 12, 13–31.199.

Polak, M., Haymaker, W., Johnson, J.E. Jr., D’Amelio, F., and Hager,H. (1982). Neuroglia and their reactions. In Histology and Histopa-thology of the Nervous System, W. Haymaker and R. Adams, eds.(Springfield: Charles C. Thomas Publishers), pp. 363–480.

Racine, R.J., Wilson, D.A., Gingell, R., and Sunderland, D. (1986).Long-term potentiation in the interpositus and vestibular nuclei inthe rat. Exp. Brain Res. 63, 158–162.

Rae, J., Cooper, K., Gates, P., and Watsky, M. (1991). Low accessresistance perforated patch recording using amphotericin B. J. Neu-rosci. Meth 37, 15–26.

Rakic, P. (1981). Neuronal-glial interaction during brain develop-ment. Trends Neurosci. 4, 184–187.

Sakurai, M. (1987). Synaptic modification of parallel fibre–Purkinjecell transmission in in vitro guinea-pig cerebellar slices. J. Physiol.394, 463–480.

Sakurai, M. (1990). Calcium is an intracellular mediator of the climb-ing fiber in induction of cerebellar long-term depression. Proc. Natl.Acad. Sci. USA 87, 3383–3385.

Schnitzer, J., Franke, W.W., and Schachner, M. (1981). Immunocyto-chemical demonstration of vimentin in astrocytes and ependymalcells of developing and adult mouse nervous system. J. Cell Biol.90, 435–447.

Shibuki, K., and Okada, D. (1991). Endogenous nitric oxide releaserequired for long-term synaptic depression in the cerebellum. Nature349, 326–328.

Shibuki, K., and Okada, D. (1992). Cerebellar long-term potentiationunder suppressed postsynaptic Ca21 activity. Neuroreport 3, 231–234.

Shigemoto, R., Abe, T., Nomura, S., Nakanishi, S., and Hirano, T.(1994). Antibodies inactivating mGluR1 metabotropic glutamate re-ceptor block long-term depression in cultured Purkinje cells. Neuron12, 1245–1255.

Skelton, R.W. (1988). Bilateral cerebellar lesions disrupt conditionedeyelid responses in unrestrained rats. Behav. Neurosci. 102,586–590.

Steinmetz, J.E., Lavond, D.G., Ivkovich, D., Logan, C.G., and Thomp-son, R.F. (1992). Disruption of classical eyelid conditioning aftercerebellar lesions: damage to a memory trace system or a simpleperformance deficit? J. Neurosci. 12, 4403–4426.

Teichberg, V.I. (1991). Glial glutamate receptors: likely actors in brainsignaling. FASEB J. 5, 3086–3091.

Thompson, R.F. (1986). The neurobiology of learning and memory.Science 233, 941–947.

Thompson, R.F., and Krupa, D.J. (1994). Organization of memorytraces in the mammalian brain. Annu. Rev. Neurosci. 17, 519–549.

Thompson, R.F., Thompson, J.K., Kim, J.J., Krupa, D.J., Nordholm,A. F., and Chen, C. (1993). Synaptic plasticity and memory storage.In Synaptic Plasticity: Molecular, Cellular, and Functional Aspects.M. Baudry, R.F. Thompson, and J.L. Davis, eds. (Massachusetts:MIT Press).

Weinstein, D.E., Shelanski, M.L., and Liem, R.K.H. (1991). Suppres-sion by antisense mRNA demonstrates a requirement for the glialfibrillary acidic protein in the formation of stable astrocytic pro-cesses in response to neurons. J. Cell Biol. 112, 1205–1213.

Weiss, C., and Thompson R.F. (1991). The effects of age on eyeblinkconditioning in the freely moving Fischer-344 rat. Neurobiol. Aging12, 249–254.

Welsh, J.P., and Harvey, J.A. (1989). Cerebella lesions and the nicti-tating membrane reflex: performance deficits of the conditioned andnonconditioned response. J. Neurosci. 9, 299–311.

Yeo, C.H., Hardiman, M.J., and Glickstein, M. (1985a). Classicalconditioning of the nictitating membrane response of the rabbit. I.Lesions of the cerebellar nuclei. Exp. Brain Res. 60, 87–98.

Yeo, C.H., Hardiman, M.J., and Glickstein, M. (1985b). Classicalconditioning of the nictitating membrane response of the rabbit IV.Lesions of the cerebellar cortex. Exp. Brain Res. 60, 99–113.