Astrocytes regulate heterogeneity of presynaptic strengths ... · counterbalances input strengths requires N-methyl-D-aspartate re-ceptors (NMDARs) and astrocytes. Importantly, this

Astrocytes regulate heterogeneity of presynapticstrengths in hippocampal networksMathieu Letelliera,1,2,3, Yun Kyung Parka, Thomas E. Chatera,4, Peter H. Chipmana,4, Sunita Ghimire Gautama,Tomoko Oshima-Takagoa, and Yukiko Godaa,b,c,1,3

aBrain Science Institute, RIKEN, Saitama 351-0198, Japan; bSaitama University Brain Science Institute, Saitama University, Saitama 338-8570, Japan; andcDepartment of Life Sciences, Graduate School of Arts and Sciences, University of Tokyo, Tokyo 153-8902, Japan

Edited by Mu-ming Poo, Institute of Neuroscience and Key Laboratory of Primate Neurobiology, State Key Laboratory of Neuroscience, Chinese Academy ofSciences Center for Excellence in Brain Science, Shanghai Institutes for Biological Sciences, Shanghai, China, and approved March 28, 2016 (received for reviewDecember 1, 2015)

Dendrites are neuronal structures specialized for receiving andprocessing information through their many synaptic inputs. Howinput strengths are modified across dendrites in ways that arecrucial for synaptic integration and plasticity remains unclear. Weexamined in single hippocampal neurons the mechanism of hetero-synaptic interactions and the heterogeneity of synaptic strengths ofpyramidal cell inputs. Heterosynaptic presynaptic plasticity thatcounterbalances input strengths requires N-methyl-D-aspartate re-ceptors (NMDARs) and astrocytes. Importantly, this mechanism isshared with the mechanism for maintaining highly heterogeneousbasal presynaptic strengths, which requires astrocyte Ca2+ signalinginvolving NMDAR activation, astrocyte membrane depolarization,and L-type Ca2+ channels. Intracellular infusion of NMDARs or Ca2+

-channel blockers into astrocytes, conditionally ablating the GluN1NMDAR subunit, or optogenetically hyperpolarizing astrocytes witharchaerhodopsin promotes homogenization of convergent presyn-aptic inputs. Our findings support the presence of an astrocyte-dependent cellular mechanism that enhances the heterogeneity ofpresynaptic strengths of convergent connections, which may helpboost the computational power of dendrites.

synapse heterogeneity | synaptic strength | astrocyte | hippocampalneuron | heterosynaptic plasticity

An enduring challenge in neurobiology is to understand howneurons set the strengths of their numerous synapses to

efficiently process and store different information while main-taining network homeostasis. Electrophysiology and imagingapproaches have revealed that synapses display a high degree offunctional heterogeneity, even for those sharing the same axonor dendrite (1–3). The observation that synaptic strengths areheterogeneous, in turn, suggests that synapses can operate in-dependently from one another. Accordingly, many studies havedemonstrated the input-specificity of Hebbian and also of ho-meostatic forms of synaptic plasticity, where synaptic changes arerestricted to inputs whose activity is altered (4–6). Nevertheless,such a synapse-autonomous behavior could potentially compro-mise the global network homeostasis by biasing the overall ac-tivity toward excitation or depression, and to overcome this issue,it has been proposed that distinct inputs cooperate by coordinatingtheir relative strengths through heterosynaptic interactions (7–9).In support of the idea that synapses behave as interdependentrather than isolated functional units, the restriction of synapticstrength changes to active inputs has been demonstrated to breakdown at times, with the induction of synaptic plasticity in thestimulated input accompanying either synaptic depression or po-tentiation of the nonstimulated inputs (10–13). In a highly studiedplasticity paradigm of long-term potentiation (LTP) at hippocam-pal Schaffer collateral–CA1 synapses, tetanic stimulation that in-duces LTP is often accompanied by presynaptic long-term depression(LTD) of nonstimulated Schaffer collateral–CA1 synapses (11, 13,14). Heterosynaptic LTD might be advantageous for sharpening thedifference between the strengths of active and inactive inputs re-

ceived by the postsynaptic CA1 neuron, and by counterbalancingpotentiation, heterosynaptic LTD might help promote the stability ofthe overall CA3–CA1 connection by placing limits on excitation.Among the proposed mechanisms for mediating heterosynaptic

interactions, the astrocyte network has recently emerged as a keyregulator (15, 16). Astrocytes display fine processes that are prox-imal to synapses, and they can detect and integrate local synapticactivity (17–21). Moreover, astrocytes are coupled to each otherthrough gap junctions, and by forming a network, they are thoughtto be capable of modulating the efficacy of a population of synapsesby coordinately releasing diffusible gliotransmitters such asglutamate, endocannabinoids, ATP or D-serine that target synapticreceptors (15, 16). In the hippocampus, astrocytes respond toSchaffer collateral stimulations (14, 22–24), and they mediate tet-anus-induced heterosynaptic presynaptic LTD at CA1 synapsesthrough purinergic signaling (14, 25–27). The involvement of as-trocytes in heterosynaptic depression of nonstimulated inputs raisesthe intriguing possibility that astrocytes might actively play a role inbalancing synaptic strengths between different inputs. Whether suchastrocyte-dependent regulation is a basic mechanism that controlsbidirectional heterosynaptic interactions in spontaneously activenetworks and if such a regulation is implemented at the level ofsingle inputs are unclear.

Significance

We addressed the basic mechanisms underlying synapse het-erogeneity, and we identified a form of regulation that servesto increase the variations in the efficacy with which neuronscommunicate with each other through synapses. We demon-strate that this process requires astrocytes, which, previously,have been thought to play mostly a passive role in maintainingneuronal functions. The cellular mechanism that regulatessynaptic efficacy requires astrocyte membrane depolarization,activation of astrocyte NMDA receptors, and astrocyte calciumsignaling. The fundamental nature of the regulation is under-scored by the preservation of the mechanism from acute brainslices down to dissociated cultures that lack the native topol-ogy of brain networks.

Author contributions: M.L. and Y.G. designed research; M.L., Y.K.P., T.E.C., P.H.C., S.G.G.,and T.O.-T. performed research; M.L., Y.K.P., T.E.C., and P.H.C. analyzed data; and M.L.and Y.G. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1Previous address: Medical Research Council Laboratory for Molecular Cell Biology andCell Biology Unit, University College London, London WC1E 6BT, United Kingdom.

2Present address: Interdisciplinary Institute for Neuroscience, University of Bordeaux,UMR 5297, 33077 Bordeaux, France.

3To whom correspondence may be addressed. Email: [email protected] [email protected].

4T.E.C. and P.H.C. contributed equally to this work.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1523717113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1523717113 PNAS | Published online April 26, 2016 | E2685–E2694

NEU

ROSC

IENCE

PNASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

Here, using electrophysiology and optical methods, we addressedthe basic cellular principle governing the distinctness of individualsynaptic inputs. Specifically, we examined whether and how twocomparable inputs from a same neuron type show coordinatedregulation of their synaptic strengths when they share the targetneuron. In both hippocampal dissociated cultures and acute slicepreparation, we find that presynaptic strengths of inputs received bya postsynaptic neuron are highly heterogeneous. The differences inpresynaptic strengths are controlled and maintained by astrocyteCa2+ signaling requiring NMDA receptors (NMDARs) and L-typevoltage-gated Ca2+ channels (L-VGCCs). This mechanism is sharedat least in part, by that of activity-dependent heterosynaptic inter-actions in which the presynaptic strength of the nonstimulated inputis counterbalanced relative to the presynaptic strength of thestimulated input, and such a plasticity mechanism could furtheraugment the variations in presynaptic strengths. Importantly, underbasal conditions, the presynaptic strengths of convergent inputsbecome rapidly correlated on blocking NMDARs or L-VGCCsor by buffering Ca2+ in astrocytes, and such correlation is alsoobserved when GRIN1 is conditionally deleted in astrocytes.Furthermore, hyperpolarizing astrocytes by light activation ofarchaerhodopsin (ArchT) promotes the correlation of pre-synaptic input strengths over the time scale of minutes. Thesefindings reveal an important contribution of activity-dependentastrocyte signaling in diversifying the strengths of convergentpresynaptic connections received by the postsynaptic neuron.

ResultsActivity-Dependent Coordinated Modulation of Presynaptic Strengthsin Hippocampal Networks. To explore whether synaptic inputsoperated in isolation or in coordination with one another and todetermine how this could impact synapse heterogeneity, we ex-amined the extent to which activity-dependent synaptic changes

were restricted to stimulated inputs. To this end, we comparedtwo independent monosynaptic inputs onto a single postsynapticneuron. We reasoned that a fundamental mechanism shouldoperate not only in a native brain circuit but it should be re-capitulated in a simplified neuronal network formed in culture.We took advantage of low density dissociated hippocampalneurons cocultured with glia, in which unitary connections formedbetween identified neurons can be stimulated and monitoredthrough multiple whole-cell patch-clamp recordings with relativeease. Triple patch-clamp experiments (Fig. 1A) were restricted tocases in which the two presynaptic neurons showed no directsynaptic coupling to each other. To induce synaptic changes at agiven input, we tested a variety of stimulation paradigms that wereused previously in dissociated cultures similar to ours (28–31); wefinally opted for a low-frequency stimulation protocol (1 Hz for3 min; Materials and Methods) that efficiently induced long-lastingchanges in excitatory postsynaptic current (EPSC) amplitude in∼80% of the recordings [n = 17 of 21; no change in n = 4 of 21(+0.99 ± 1.7%); Fig. 1 B and C]. Interestingly, the direction ofplasticity was variable, and the stimulation could produce eitherpotentiation (+25.9 ± 4.4%, n = 6 of 21 recordings) or depression(−29.9 ± 3.8%, n = 11 of 21) of the EPSC amplitude in thestimulated input (referred as “homosynapses”; Fig. 1 B and C).Furthermore, EPSC amplitude changes were inversely correlatedwith paired-pulse ratio (PPR) and CV−2 (Fig. 1 B and E and Fig.S1A), two parameters related to neurotransmitter release proba-bility (pr) (28, 32–34). When the same stimulation paradigm wasgiven to a cell pair in which the presynaptic neuron expressedVGLUT-pHluorin, a reporter of synaptic vesicle exo-endocyosis,the readily releasable vesicle pool that was proportional to pr (1)also showed potentiation (two of six pairs), depression (one of sixpairs), or no change (three of six pairs), which further confirmedthe presynaptic origin of plasticity (Fig. S2). This observation was

Fig. 1. Presynaptic interactions between conver-gent inputs in dissociated hippocampal cultures. (A)(Left) Experimental scheme. One of the two con-vergent inputs is repeatedly stimulated (homo-synapse; 180 action potentials at 1 Hz), whereas theother input is unstimulated (heterosynapse). (Right)DIC image of 12 DIV cultured hippocampal neuronsduring a triple recording. (Scale bar, 40 μm.) (B)Average homosynaptic EPSC traces to a pairedstimuli (50-ms interstimulus interval) before and af-ter the repeated stimulation. (C and D) Time courseof homosynaptic (C; n = 21 cells) and heterosynaptic(D; n = 21 cells) EPSC amplitude changes induced bythe repeated stimulation sorted into those showingpotentiation [dark circles: n = 6 (C); n = 6 (D)], nochange [medium circles: n = 4 (C); n = 8 (D)], ordepression [light circles: n = 11 (C); n = 7 (D)]. Indi-vidual recordings were sorted into three groupsaccording to the comparison of the baseline withthe poststimulation period using the Wilcoxon test.Data are expressed as mean ± SEM. (E and F) Plots ofthe changes in PPR vs. mean EPSC amplitude athomosynapses (E: r2 = 0.33, **P = 0.0061) and het-erosynapses (F: r2 = 0.48, ***P = 0.0005). Linear re-gression (black lines) and 95% CI (gray shade) areindicated. (G) Comparison of the extent changes inhomosynaptic vs. heterosynaptic EPSC amplitudes(r2 = 0.18, P = 0.0563). (H) Comparison of the ex-tent changes in homosynaptic vs. heterosynapticPPR (r2 = 0.04, P = 0.3940).

E2686 | www.pnas.org/cgi/doi/10.1073/pnas.1523717113 Letellier et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

consistent with presynaptic expression of plasticity that had beenpreviously reported for hippocampal synapses (28, 30, 33, 35–37).Importantly, EPSC amplitude change in the stimulated input

was often accompanied by a stable EPSC amplitude change inthe unstimulated input (referred as “heterosynapses”), suggestingthat the expression of synaptic plasticity was not confined to thestimulated input (Fig. 1D). Therefore, this stimulation paradigmallowed us to study the properties of heterosynaptic interactionsinvolving changes in synaptic strength. As for homosynapses,heterosynaptic EPSC amplitude changes were highly heteroge-neous (potentiation: +28.4 ± 4.0%, n = 6 of 21 recordings; de-pression: −22.4 ± 5.9%, n = 7 of 21; no change: 0.7 ± 5.2%, n = 8of 21; Fig. 1D) and correlated with changes in the presynapticparameters PPR and CV−2 (Fig. 1F and Fig. S1B). Surprisingly,the direction and the magnitude of homo- and heterosynapticchanges appeared nonetheless uncorrelated to each other(Fig. 1G), indicating that the active and inactive inputs interactednonuniformly.Given that the heterosynaptic change was induced by stimu-

lating the homosynaptic input, we asked if there was a homo-synaptic parameter that was related to the direction and themagnitude of the heterosynaptic change. We found that the ex-tent changes in heterosynaptic EPSC amplitude or PPR werecorrelated to the initial PPR of the homosynaptic input but notto the homosynaptic PPR change (Figs. 1H and 2 A and B) nor tothe initial homosynaptic amplitude (Fig. S3). The apparent in-dependence of the heterosynaptic changes to the homosynapticamplitude suggests that the heterosynaptic change does not de-pend on the number of activated synapses. Notably, the hetero-synaptic PPR change showed a significant inverse relationship tothe initial homosynaptic PPR (heterosynaptic PPR = −0.47 ×initial homosynaptic PPR + 0.45; r2 = 0.23, P = 0.0292; Fig. 2B),in that stimulating a high pr input was more likely to decrease prof nonstimulated convergent inputs and vice versa. Such anactivity-dependent coordinate modulation could play a potentialrole in sharpening the differences in the efficacies of active andinactive inputs received by the target neuron.

Heterosynaptic Presynaptic Modulation Requires NMDARs andAstrocytes but Not Postsynaptic Ca2+. To gain insights into the

mechanism by which stimulating the homosynaptic input producedthe PPR change at the nonstimulated input, we first tested theinvolvement of postsynaptic Ca2+, which was implicated in someforms of heterosynaptic plasticity (12, 38). However, after infusingthe postsynaptic neuron with 1,2-bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid (BAPTA; 30 mM) through the patchpipette, stimulating the homosynaptic input still produced changesin the heterosynaptic EPSC amplitude and PPR to extents com-parable to the control condition (range of PPR change for control:−0.48 to 1.03, CV = 19.8, n = 21; for postsynaptic BAPTA: −0.72to 0.32, CV = 21.8, n = 7; P = 0.7761, F test; Fig. 2C and Fig. S4),and the heterosynaptic PPR change remained inversely related tothe basal homosynaptic PPR (heterosynaptic PPR = −1.43 × ini-tial homosynaptic PPR + 1.04; r2 = 0.75, P = 0.0111; Fig. 2 B andC). Therefore, heterosynaptic presynaptic plasticity in our culturesystem appears not to require postsynaptic Ca2+.We next examined a role for NMDARs and astrocytes that had

been previously shown to mediate heterosynaptic LTD at hippo-campal CA3–CA1 synapses (14, 25–27, 39). Despite the apparentlack of requirement for postsynaptic Ca2+, the heterosynaptic PPRchange was fully blocked by the NMDAR antagonist D-2-amino-5-phosphonovaleric acid (AP5; 50 μM) (range of PPR change forcontrol: −0.48 to 1.03, CV = 19.8, n = 21; for AP5: −0.05 to +0.21,CV = 1.7, n = 10; P = 0.0002, F test) and unrelated to initialhomosynaptic PPR (heterosynaptic PPR = 0.03 × initial homo-synaptic PPR + 0.01, r2 = 0.05, P = 0.5431; Fig. 2D and Fig. S4),whereas the basal PPR itself was not affected by the AP5 treat-ment (control: 0.92 ± 0.07, AP5: 1.20 ± 0.17; Fig. S5A); thissuggested a possible involvement of NMDARs expressed in cellsother than the postsynaptic neuron. Also, incubating cultures for30 min with fluoroacetate (5 mM), an inhibitor of the Krebs cyclethat preferentially compromised glial cells (39–41), not only at-tenuated the heterosynaptic PPR change (range of PPR change forcontrol: −0.48 to 1.03, CV = 19.8, n = 21; for fluoroacetate: −0.33to 0.22, CV = 8.79, n = 10; P = 0.07, F test), but the PPR changewas no longer related to the initial homosynaptic PPR (hetero-synaptic PPR = −0.01 × initial homosynaptic PPR + 0.04; r2 =0.004, P = 0.8661; Fig. 2E). Fluoroacetate did not significantly alterthe mean basal PPR in our conditions (control: 0.92 ± 0.07, n = 21,fluoroacetate: 1.14 ± 0.21, n = 10; Fig. S5A). These results suggestthat NMDARs and astrocytes mediate activity-dependent hetero-synaptic interactions in cultured hippocampal networks.We also tested a possible role for L-VGCCs. The L-VGCC

antagonist nifedipine (10 μM) did not prevent the bidirectionalchange in heterosynaptic PPR per se (range of PPR change forcontrol: −0.48 to 1.03, CV = 19.8, n = 21; for nifedipine: −0.28 to0.74, CV = 5.73, n = 8; P = 0.7398, F test; Fig. 2F and Fig. S4).Strikingly, however, nifedipine reversed the polarity of the in-verse relationship between the heterosynaptic PPR change andthe initial homosynaptic PPR into a positive one (heterosynapticPPR = 0.67 × initial homosynaptic PPR – 0.85, r2 = 0.82, P =0.0019; Fig. 2F). This observation suggested that in the presenceof nifedipine stimulating an input with high pr was likely topotentiate nonstimulated inputs and vice versa. Thus, althoughL-VGCCs were not essential for homo- or heterosynaptic plas-ticity, blocking their activity had the apparent effect of equalizingpresynaptic strengths between homosynapses and heterosynapsesin active networks. L-VGCC activity could contribute to thecellular process by which inactive inputs adapt to the strength ofactive inputs.Given that NMDARs and astrocytes mediate heterosynaptic

plasticity in cultured networks similarly to tetanus-induced het-erosynaptic LTD in brain slices (14, 27), we wondered if theapparent lack of requirement for postsynaptic Ca2+ and thedependence on L-VGCCs of the polarity of heterosynapticplasticity that we observed in cultured networks could be ex-tended to brain slices. We tested this possibility by examiningheterosynaptic LTD at CA3–CA1 synapses in acute hippocampal

Fig. 2. Presynaptic interactions between convergent inputs depend onNMDARs, astrocytes, and L-VGCCs but not on postsynaptic Ca2+. (A) Com-parison of the percent change in heterosynaptic mean EPSC amplitude vs.the initial homosynaptic PPR in the control condition (r2 = 0.20, P = 0.0418).(B–F) Comparison of the heterosynaptic PPR change vs. the initial homo-synaptic PPR under indicated conditions: control untreated (B: r2 = 0.23, *P =0.0292); BAPTA in the postsynaptic neuron (C: r2 = 0.75, *P = 0.0111); AP5 (D:r2 = 0.05, P = 0.5431); fluoroacetate (E: r2 = 0.004, P = 0.8661); and nifedipine(F: r2 = 0.82, **P = 0.0019). Linear regression (black lines) and 95% CI (grayshade) are shown.

Letellier et al. PNAS | Published online April 26, 2016 | E2687

NEU

ROSC

IENCE

PNASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

slices (14). Two independent Schaffer collateral inputs werestimulated with extracellular electrodes (>200 μm apart) to evokeEPSCs in CA1 pyramidal neurons (Fig. 3 A and B). Applyingthree 100-Hz–1-s tetanus at 30-s intervals to one of the two inputselicited homosynaptic LTP of EPSC amplitude (+62.9 ± 9.8%,n = 6; Fig. 3C), whereas the EPSC amplitude of the nonstimulatedinput was stably decreased (−16.6 ± 8.5%, n = 6; Fig. 3C). Unlikethe homosynaptic LTP, the heterosynaptic LTD displayed a de-crease in CV−2 and an increase in PPR (Fig. 3D), which supportedfor a decreased pr as the basis for the synaptic depression, inagreement with previous reports (14, 25, 26). Importantly, di-alyzing the postsynaptic neuron with BAPTA fully blocked thehomosynaptic LTP (−8.8 ± 7.7%; Fig. 3E) in accord with its re-quirement for postsynaptic NMDARs (42). In contrast, the het-erosynaptic LTD still occurred (−17.4 ± 4.9%, n = 10), and CV−2

and PPR showed changes similarly to those observed in the ab-sence of BAPTA (Fig. 3 E and F). Therefore, like heterosynapticplasticity in cultured neurons, heterosynaptic presynaptic LTDin acute slices did not require postsynaptic Ca2+. Interestingly,when tetanic stimulation was applied in the presence of nifedi-pine or another L-VGCC inhibitor, nimodipine (10 μM), althoughhomosynaptic LTP was elicited (nifedipine: +62.7 ± 29.3%, n = 5;nimodipine: +78.9 ± 28.7%, n = 8), heterosynaptic LTD wasblocked, and no statistically significant changes in PPR and CV−2

were observed (Fig. 3 G and H and Fig. S6). Rather, there was atendency for an increase in the heterosynaptic EPSC amplitude(nifedipine: +6.4 ± 8.9%, n = 5; nimodipine: +9.5 ± 14.5%, n = 8),which was accompanied by a decrease in PPR in the majority ofthe recorded cells (nifedipine: five of five cells; nimodipine: four

of eight cells). This increase in EPSC amplitude was reminiscentof the effect of nifedipine in reversing the direction of hetero-synaptic plasticity in dissociated cultures.Taken together, these results suggest that the mechanism of

expression of heterosynaptic presynaptic plasticity is shared be-tween hippocampal dissociated cultures and acute slices, andalthough independent of postsynaptic Ca2+, it involves NMDARsand astrocytes. Moreover, the finding that the polarity of hetero-synaptic plasticity is reversed by L-VGCC inhibitors suggests thatthe heterosynaptic plasticity might regulate the disparity of pre-synaptic strengths between inputs.

Astrocyte NMDARs and L-VGCCs Promote Differences in Basal PPR ofConvergent Inputs. In the above experiments, the counterbalancingof presynaptic strengths between convergent inputs was observedon deliberately stimulating one of the inputs to elicit homo- andheterosynaptic plasticity. We wondered whether the heterosynapticcoordination is exclusive to the induction of synaptic plasticity, oralternatively, stimulation could have simply exaggerated the pre-synaptic regulation that normally occurs under basal activity levels. Ifthe latter were the case, one would expect to observe differences in thebasal presynaptic strengths of convergent inputs, where the hetero-geneity is regulated by astrocytes, NMDARs, and also by L-VGCCs.We next sought to test this possibility by comparing the basal pre-synaptic strengths of convergent inputs in cultures and acute slices.Comparisons of PPR under basal conditions revealed that the

PPR was heterogeneous and uncorrelated between two pre-synaptic neurons targeting a common postsynaptic neuron inculture and for two independent Schaffer collateral inputs onto a

Fig. 3. L-VGCCs control heterosynaptic LTD in acute hippocampal slices independently of postsynaptic Ca2+. (A) (Left) Experimental scheme in acute hippocampalslices: two independent Shaffer collateral inputs are stimulated with extracellular electrodes (S1 and S2), and corresponding responses are recorded in a CA1 neuron. Atetanic stimulation (3 × 100 Hz) of one of the inputs elicits homosynaptic LTP along with heterosynaptic LTD of the nonstimulated input. (Right) Photomicrograph ofhippocampal area CA1 showing the stimulating and recording electrode positions: So, stratum oriens; Sp, stratum pyramidale; Sr, stratum radiatum. (Scale bar, 60 μm.)(B) Average EPSC traces evoked by a pairwise cross-stimulation of the two independent inputs show a lack of short-term plasticity. (C, E, and G) Time course ofhomosynaptic (light circles) and heterosynaptic (dark circles) EPSC amplitudes before and after the tetanic stimulation in control (C, n = 6 cells) or with intracellularpostsynaptic BAPTA (B, n= 10 cells) or bath applied nifedipine (E, n= 5 cells). Postsynaptic BAPTA but not nifedipine blocks the homosynaptic LTP (postsynaptic BAPTA,**P = 0.0015; nifedipine, P = 0.9171; two-way ANOVA followed by Fisher’s LSD test). In contrast, nifedipine but not postsynaptic BAPTA blocks the heterosynaptic LTD(nifedipine, *P = 0.0474; postsynaptic BAPTA, P = 0.9518; two-way ANOVA followed by Fisher’s LSD test). Data are expressed as mean ± SEM. (D, F, and H) (Left)Representative heterosynaptic EPSC traces of PPR and plots comparing heterosynaptic PPR before and after the tetanus in the three conditions (paired t test: control,*P = 0.0381; postsynaptic BAPTA, **P = 0.0043; nifedipine, *P = 0.0286). (Right) Plots of heterosynaptic CV−2 analysis of heterosynaptic EPSCs before and after thetetanus in the three conditions (paired t test: control, *P = 0.0378; postsynaptic BAPTA, *P = 0.0199; nifedipine: P = 0.6993).

E2688 | www.pnas.org/cgi/doi/10.1073/pnas.1523717113 Letellier et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

CA1 neuron in acute slices (Fig. 4 A and B and Fig. S7 A and B).In both cultures and acute slices, bath applying AP5 (50 μM) ornifedipine (or nimodipine, 10 μM) in the absence of any condi-tioning stimulation decreased the PPR difference between theconvergent inputs, and basal PPR became more correlated acrossexperiments (Fig. 4 C–E and Fig. S7 C and D). Moreover, in cul-tures, 30-min treatment with fluoroacetate (5 mM) that inhibitedheterosynaptic plasticity also promoted the correlation of PPR be-tween inputs (Fig. S7E). Whereas none of the drug treatments af-fected the average PPR compared with the control condition(control, 1.90 ± 0.12, n = 29; AP5, 1.86 ± 0.11, n = 11; nifedipine,1.74 ± 0.10, n = 17, nimodipine, 2.21 ± 0.11, n = 11; Fig. S5 B andD), strikingly, in all drug treatment conditions, the PPR differencebetween the two inputs (the PPR disparity) was decreased by morethan 50% in slices (control: 0.59 ± 0.13, n = 29; AP5: 0.23 ± 0.06,n = 9; nifedipine: 0.19 ± 0.06, n = 17; nimodipine: 0.23 ± 0.07, n =11) and in cultures (control: 0.66 ± 0.17, n = 10; AP5: 0.20 ± 0.06,n = 9; fluoroacetate: 0.18 ± 0.05, n = 9; nifedipine: 0.28 ± 0.06, n =7; Fig. 4F and Fig. S7F). Therefore, conditions that compromisedheterosynaptic plasticity resulted in increased correlation of basalPPR. The increase in correlation occurred relatively rapidly, in thatnimodipine application in acute slices decreased the PPR disparitybetween the two inputs to its full extent within 10–15 min (−48.2 ±9.3%, n = 6; Fig. 4G). Altogether, these results suggest the existenceof a cellular process involving astrocytes, NMDARs, and L-VGCCsthat maintain variations in presynaptic strengths under basal con-ditions, which appears to be a basic process that is not strictly de-pendent on the native hippocampal circuit but is reproduced in a

simplified culture network, and that this process is embedded in themechanism of heterosynaptic plasticity.We next sought to obtain further insights into how astrocytes

regulated the presynaptic strength heterogeneity and the re-lationship between astrocytes, NMDARs, and L-VGCCs in thismechanism. Having confirmed that the basic properties of het-erosynaptic coordination of presynaptic strengths were sharedbetween cultures and acute slices, we examined more detailedmechanisms primarily in acute hippocampal slices. We first de-termined a requirement for astrocyte Ca2+ signaling by whole-cell patch clamping an astrocyte in the area CA1 with a pipettecontaining BAPTA (30 mM) and waiting for >15 min. Duringthis time, Alexa dye that was also included in the patch pipettespread across the astrocyte network that suggested of extensivegap junctional coupling (Fig. 4H); in turn, this observation in-dicated that BAPTA could also spread across the astrocytenetwork. Recording from a CA1 neuron close to the patchedastrocyte, a comparison of the PPR of two independent Schaffercollateral inputs showed an ∼50% reduction in the PPR disparitycompared with control recordings in which an astrocyte waspatched without BAPTA for the same duration (control: 0.42 ±0.11, n = 12; BAPTA: 0.20 ± 0.04, n = 12; Fig. 4 I, J, and N).Thus, astrocyte Ca2+ plays a role in decorrelating convergentpresynaptic strengths.We next sought to clarify the cellular location of NMDARs

and L-VGCCs with respect to their involvement in the astrocyte-dependent mechanism of presynaptic strength regulation. Pre-vious reports have suggested that both NMDARs and L-VGCCsare expressed in hippocampal astrocytes (43–48, but see ref. 49).

Fig. 4. NMDARs and L-VGCCs activity in astrocytesdecorrelate presynaptic strengths of convergent in-puts in hippocampal slices. (A) Scheme in acute hip-pocampal slices. (B–E) Scatter plots comparing basalPPR of the two independent inputs normalized tothe fist EPSC amplitude for control (n = 29 cells, r2 =0.06, P = 0.1822), AP5 (n = 9 cells, r2 = 0.26, P = 0.1998),nifedipine (n = 17 cells, r2 = 0.57, ***P = 0.0004),and nimodipine (n = 11 cells, r2 = 0.67, **P = 0.0022).(Inset) Average traces from representative recordings.(F) Summary of the average basal PPR difference forthe same conditions (one-way ANOVA followedby Holm–Sidak’s multiple comparison test, *P < 0.05,**P < 0.01). Data are expressed as mean ± SEM.(G) Time course of PPR disparity between two inputswhen nimodipine is added to the bath (n = 6) or not(n = 7) (two-sample t test, *P < 0.05, **P < 0.01).(H) (Upper) Experimental scheme as in A with concur-rent patch-clamping of astrocytes that are intracellularlycoupled via GAP junctions. (Lower) Fluorescence imageof an experiment showing a CA1 neuron filled with agreen dye and the intracellular spread of a red dyeacross the astrocyte syncytium. (Scale bar, 30 μm.) (I–M)Scatter plots comparing basal PPR of the two indepen-dent inputs when astrocytes are dialyzed with controlintracellular solution (n = 12 cells, r2 = 0.16, P =0.1994), MK-801 (n = 11 cells, r2 = 0.63, **P = 0.0038),BAPTA (n = 12 cells, r2 = 0.75, ***P = 0.0003), D890(n = 9 cells, r2 = 0.66, **P = 0.0077), or QX-314 (n =11 cells, r2 = 0.90, ****P < 0.0001). (Inset) Averagetraces from representative recordings. (N) Summaryof the average basal PPR difference for the sameconditions (one-way ANOVA followed by Holm–Sidak’smultiple comparison test, *P < 0.05, **P < 0.01). Dataare expressed as mean ± SEM. (Scale bar for insets ofnormalized PPR traces, 25 ms.)

Letellier et al. PNAS | Published online April 26, 2016 | E2689

NEU

ROSC

IENCE

PNASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

Therefore, we tested whether NMDARs and L-VGCCs on as-trocytes contributed to the observed PPR disparity. Analogousto the BAPTA experiment described above, the channels wereblocked intracellularly by perfusing drugs into astrocytes via thepatch pipette. We tested the following inhibitors: MK-801 (1 mM),a noncompetitive antagonist of NMDARs that was shown to blockastrocyte NMDARs intracellularly (50); D890 (2 mM), a mem-brane impermeant verapamil derivative that blocks L-VGCCs (51,52); and QX-314 (10 mM), a lidocaine derivative that has beenshown to block voltage-gated channels including L-VGCCs (53).All three inhibitors reduced the PPR disparity between two con-vergent inputs by >50% compared with the control (control: 0.42 ±0.11, n = 12; MK-801: 0.12 ± 0.03, n = 12; D890: 0.16 ± 0.06, n = 9;QX-314: 0.11 ± 0.03, n = 11; Fig. 4 K–N).Collectively, these results implicate a Ca2+-dependent signaling

mechanism involving NMDAR and L-VGCC activities in astrocytesin decorrelating presynaptic strengths of convergent inputs.

Activation of Astrocyte NMDARs Depolarizes Astrocyte Membraneand Opens L-VGCCs. To further clarify the contribution ofNMDARs and L-VGCCs expressed in astrocytes to the observedregulation of PPR, we investigated whether astrocytes could di-rectly respond to NMDA application and if this response requiredNMDAR or L-VGCC activity specifically in astrocytes. We per-formed whole-cell patch-clamp recordings from astrocytes in thearea CA1 of hippocampal slices (Fig. 5A). Astrocytes showed alinear I-V relationship with a low resting membrane potentialunder I-clamp (−85.2 ± 0.89 mV, n = 12; Fig. 5B), and a dyeincluded in the patch pipette efficiently diffused throughout theastrocyte network (Fig. 4H). On bath applying NMDA and glycine(20 μM each for 3 min) in the presence of TTX (0.5 μM) to blockaction potentials (Fig. 5A), we observed a large and slow depo-larization of the astrocyte membrane potential (peak depolariza-tion, 30.67 ± 0.59 mV, n = 12; Fig. 5 C and D), which was inagreement with previous studies (46, 50). In addition, observationof the NMDA response despite of the low resting astrocyte mem-brane potential and in the presence of 1.5 mM extracellular Mg2+

was reminiscent of the weak Mg2+ block found for NMDARsexpressed by cortical astrocytes (54). Lowering extracellular Ca2+

(from 2 to 0.25 mM) or dialyzing astrocytes with MK-801 (1 mM)for 15 min before applying NMDA+glycine significantly re-duced the peak depolarization by ∼45% and ∼65%, respectively(low Ca2+: 16.52 ± 4.36 mV, n = 4; intracellular MK-801: 11.63 ±2.97 mV, n = 10); moreover, blocking L-VGCCs with bath appliednifedipine (10 μM) or nimodipine (10 μM), or intracellularlyloading D890 (2 mM) or QX-314 (10 mM) also attenuated thepeak depolarization by 25–45% (nifedipine: 17.12 ± 3.36 mV,n = 8; nimodipine: 21.02 ± 3.11 mV, n = 9; D890: 22.82 ± 2.2 mV,n = 9; QX-314: 23.59 ± 1.85 mV, n = 9; Fig. 5 C andD). Together,these results indicate that NMDAR activation in astrocytes trig-gers astrocyte membrane depolarization that is strongly dependenton Ca2+-influx, and the depolarization, in turn, triggers voltage-gated conductances, including those mediated by L-VGCCs.

GluN1 Expressed in Astrocytes Promotes the Differences in PPR ofConvergent Inputs. We next used a genetic approach to confirmthe contribution of astrocyte NMDARs in controlling the PPRdisparity between convergent inputs. To impair NMDAR activityspecifically in astrocytes, we used mice homozygous for thefloxed GRIN1 gene encoding GluN1, the obligatory NMDARsubunit (55). Adeno-associated virus (AAV DJ/8) carrying eitherGFP or Cre-mCherry under the shortened version of the humanglial fibrillary acidic protein (hGFAP) promoter (56) wasmicroinjected bilaterally into the hippocampal area CA1 of 5- to6-wk-old GRIN1 floxed mice. Two weeks after the injection, Cre-mCherry was expressed in GFAP-positive cells that showed ahighly ramified morphology typical of astrocytes but not in cellslabeled for NeuN, a neuronal marker (Fig. 5E; 0.7% of mCherry

positive cells colabeled with NeuN; n = 138 cells from two slices).The robust neuronal expression of NMDARs might interferewith assessing the efficacy of GluN1 conditional KO by a bio-chemical or immunolabeling approach. Therefore, we used afunctional assay by whole-cell patch-clamping astrocytes in hip-pocampal slices prepared from virus injected mice and monitoringmembrane depolarization induced by NMDA+Gly application asdescribed above except, in addition to TTX, Cd2+ (100 μM), andCNQX (10 μM) were also present in the bath to block VGCCsand AMPARs, respectively. In slices from mice injected with theCre AAV, peak membrane depolarization induced by NMDA+Gly was reduced by 60% relative to recordings from control AAVslices (control: 10.4 ± 1.4 mV, n = 14; Cre: 4.4 ± 1.7 mV, n = 14;Fig. 5F); the decrease was similar to the extent inhibition achievedby intracellular perfusion of MK-801 into astrocytes via the patchelectrode (Fig. 5 C and D). Under these conditions, recordingsfrom a CA1 neuron showed that PPR disparity of convergentSchaffer collateral inputs was reduced by ∼50% (control: 0.39 ±0.07, n = 23; Cre: 0.19 ± 0.03, n = 23; Fig. 5 G and H). Theseobservations provide further support to the key role for astro-cyte NMDA receptors in promoting basal synaptic strengthheterogeneity.

Optogenetic Hyperpolarization of Astrocytes Correlates Basal PPR ofConvergent Inputs. Taken together, the above experiments sug-gested that directly activating astrocyte NMDARs induced de-polarization of the astrocyte membrane that was needed toactivate L-VGCCs, which, along with NMDARs, were requiredfor the observed decorrelation of basal presynaptic strengths.If this were the case, then preventing astrocyte membrane de-polarization should reverse the decorrelation (similar to theL-VGCC blockade), and the presynaptic strengths of convergentinputs should become more similar. Given the highly ramifiedstructure of astrocytes and their low input resistance that madecontrolling their membrane potential difficult by standard elec-trophysiology methods, we took an optogenetic approach andused light-induced hyperpolarization of astrocyte membranepotential using ArchT (57) to test the effect on the PPR differencesbetween convergent inputs.AAV (DJ/8) carrying either GFP or ArchT-GFP under the

full-length hGFAP promoter was microinjected bilaterally intothe hippocampal area CA1 of 3-wk-old mice. After 6 d, GFP orArchT-GFP was reliably expressed in GFAP-positive cells (Fig. 6A and B). The expression was highly specific to astrocytes in thatcells double labeled for GFP and NeuN were limited to 0.8% ofneurons in area CA1 (5 of 666 NeuN-positive cells showedcolabeling with GFP; n = 4 slices). Using acute hippocampalslices prepared from virus-injected animals, we performedwhole-cell patch-clamp recordings from ArchT-GFP–expressingastrocytes. Light stimulation (Materials and Methods) produced areversible hyperpolarization of the astrocyte membrane potentialunder I-clamp (from −78.36 ± 0.35 to −102.51 ± 3.58 mV, n = 7;Fig. 6C) and outward currents were detected in V-clamp (Fig.S8) similarly to a recent study that expressed ArchT in astrocytesin the cerebellum (58).We then monitored EPSC amplitude and PPR of two indepen-

dent inputs received by a CA1 neuron and tested the effect of acontinuous, 10-min light activation of astrocyte ArchT. Light stim-ulation, which hyperpolarized astrocytes, rapidly and reversiblyincreased the EPSC amplitude relative to the baseline (ArchT:+27.8 ± 11.8%, n = 7; control: −6.01 ± 6.04%, n = 6; Fig. 6D). Thisincrease in EPSC amplitude indicated that basal depolarization ofastrocytes provided an inhibitory tone on excitatory synaptictransmission. Remarkably, compared with the baseline, lightstimulation decreased the PPR disparity between the two inputsin slices expressing ArchT but not in GFP control (at time35 min, control: +33.3 ± 20.4%, n = 6; ArchT: −64.6 ± 20.5%,n = 7; Fig. 6 E–G). Moreover, reminiscent of the time course of

E2690 | www.pnas.org/cgi/doi/10.1073/pnas.1523717113 Letellier et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

the effect of blocking L-VGCCs (Fig. 4G), on light stimulation ofArchT, the PPR disparity was decreased by greater than 50%over several minutes; this effect was reversible as the level ofdisparity returned to the baseline 20–25 min after shutting off thelight (Fig. 6E). Therefore, our finding strongly implicates astro-cyte membrane depolarization in the mechanism that promotesthe decorrelation of PPR.

DiscussionHere we identified an astrocyte-dependent cellular process thatserves to enhance the heterogeneity of presynaptic strengths byincreasing the PPR disparity between inputs targeting the sameneuron. This mechanism shares its key properties with themechanism of heterosynaptic plasticity that we have also studiedhere, which counterbalances PPR of the nonstimulated inputrelative to that of the stimulated input. Furthermore, this as-trocyte-dependent presynaptic regulation is observed in acutehippocampal slices and is recapitulated in dissociated cultures, asimplified system without the native topological organization ofthe hippocampal circuit. That the features of this presynapticregulation are conserved in different conditions—i.e., betweenthe basal state and during synaptic plasticity—and across dif-ferent experimental preparations underscores the fundamentalnature of the underlying mechanism.The mechanism for generating presynaptic strength hetero-

geneity, which also supports heterosynaptic plasticity, may playan important role in the developing nervous system. For in-stance, this mechanism may represent a powerful way to or-chestrate synapse competition where the relative strengths ofcompeting terminals biases the winner by favoring stabilizationof strong synapses at the expense of weaker synapses (18, 59, 60).

Specifically, the mechanism we describe here could promote atwo-step process where first it will help create differences ininput strengths to ensure variability; subsequently, it will facili-tate the selective stabilization of the strongest input (59, 60).Importantly, the astrocyte-dependent mechanism that controlsthe disparity or the decorrelation of PPR we report here is notlimited to the developing brain. Acute hippocampal slices fromadult mice, which have been used for GluN1 conditional KO andArchT experiments, show the dependence on astrocyte NMDARsand membrane depolarization for maintaining the PPR decorre-lation (Figs. 5 E–H and 6). Therefore, the mechanism that pro-motes the presynaptic strength heterogeneity is likely to functionin concert with the various rules of synaptic plasticity that operatesin the adult hippocampal circuit. For example, decorrelated pre-synaptic strengths may boost the circuit responsiveness undersparse activity conditions (61) or facilitate the reorganization ofcorrelated networks that is associated with learning (62). More-over, its dysregulation might culminate in diseased states; for in-stance, unrestrained correlation may nucleate synchronization ofactivity across synaptic networks, which is a hallmark feature ofepileptic disorders (63).Mounting evidence supports a role for astrocytes in regulating

synaptic transmission and synaptic plasticity (15, 64). In general,astrocyte influence on neurons has been thought to be global,a view established by the prominent slow Ca2+ transients thatspread through the astrocyte network. However, recent studieshave identified fast and local Ca2+ signals that regulate synapsefunction in astrocyte processes, which are in physical proximity ofsynapses (17, 19–21). Our finding of the role for astrocytes incontrolling presynaptic strengths is compatible with such locallygenerated signals. Astrocytes must be capable of deciphering

Fig. 5. Astrocyte NMDAR activation depolarizes astrocyte membrane and promotes PPR decorrelation. (A) Experimental scheme in acute hippocampal slices.TTX is present throughout, and astrocyte membrane potential change is measured from whole-cell patch-clamp recordings. (B) Astrocyte identification basedon the typical linear I-V relationship (Right) and the morphology visualized by infusing a dye via the patch pipette (Left), which spreads across other cells(arrows) of the syncytium. (C) Representative traces of astrocyte whole-cell patch clamp recordings showing membrane depolarization induced by bathapplied NMDA+glycine in control, in low extracellular Ca2+, in nifedipine or nimodipine, or on infusing astrocytes intracellularly with MK-801, D890, or QX-314. (D) Summary data of the effect of different blockers on astrocyte depolarization induced by NMDA+glycine (one-way ANOVA followed by Holm–Sidak’smultiple comparison test, *P < 0.05, ***P < 0.001, ****P < 0.0001). (E) (Left) Experimental strategy for GRIN1 genetic deletion in CA1 astrocytes. AAV DJ/8carrying Cre-mCherry under hGFAP promoter is microinjected into the hippocampal area CA1 of GRIN1 floxed mice. (Right) Confocal section showing specificexpression of Cre (red, arrows) in GFAP-expressing astrocytes (green) but not in NeuN-expressing neurons (blue). (F) (Left) Average traces from patch-clamprecordings of Cre-mCherry (Cre, red) or GFP-expressing (Control, black) astrocytes showing depolarization induced by NMDA+glycine in presence of CNQXand CdCl2. (Right) Summary data of the average peak depolarization in EGFP or Cre-mCherry expressing astrocytes (unpaired t test, P = 0.0006). (G) Scatterplots comparing basal PPR of two independent inputs when astrocytes express EGFP (Control, n = 23 cells, r2 = 0.16, P = 0.1223) or Cre-mCherry (Cre, n = 23cells, r2 = 0.47, P = 0.0003). (Inset) Average traces from representative recordings. (Scale bar, 20 ms.) (H) Summary of the average basal PPR difference (Left;unpaired t test, P = 0.0188) and average PPR (Right; unpaired t test, P = 0.1494) for the same conditions. Data are expressed as mean ± SEM.

Letellier et al. PNAS | Published online April 26, 2016 | E2691

NEU

ROSC

IENCE

PNASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

signals associated to different inputs, and the decoding mecha-nism could engage local Ca2+ signals whose magnitudes might berelated to the strength of synaptic inputs, as recently reported atthe neuromuscular junction (18).We find that the decorrelation of presynaptic strengths re-

quires membrane depolarization, NMDARs and L-VGCCs, andCa2+ signaling, all within astrocytes. Whereas the expression ofNMDARs and L-VGCCs in astrocytes has been debated, ourstudy is in line with previous studies reporting of the presence ofthese channels in astrocytes (43–48, 50). With respect to astro-cyte NMDARs, we find that applying NMDA in the presence ofTTX to prevent synaptic transmission directly triggers astrocytemembrane depolarization, and this depolarization is compro-mised by intracellularly infusing MK-801 or deleting GRIN1 inastrocytes. That we could obtain astrocyte-dependent NMDARresponses reliably in the presence of 1.5 mM extracellular Mg2+

even though astrocytes have low resting membrane potential iscompatible with the presence of NR3 containing NMDARs withreduced Mg2+ block reported for cortical astrocytes (54). Fur-thermore, a function for astrocyte NMDARs in modulatingpresynaptic release is supported by observations in which in-tracellularly infusing MK-801 or conditionally deleting GRIN1 inastrocytes reduces the PPR disparity between convergent inputs.Similarly, the expression of L-VGCCs in astrocytes is supportedby the decrease in PPR disparity of convergent inputs on in-tracellularly delivering D890 into astrocytes.

How are these astrocyte-specific components—membranedepolarization, activation of NMDARs and L-VGCCs, and as-trocyte Ca2+ signaling—orchestrated to impose the presynapticstrength differences between inputs? Whereas we sampled, forevery experiment, a set of two presynaptic inputs that convergeonto the postsynaptic target neuron, the postsynaptic neuronreceives synaptic inputs from numerous other afferents. Onewould then expect that the presynaptic strengths of other non-sampled inputs might be similarly decorrelated. Given such asituation, it is doubtful that an astrocyte mechanism exists thatcan specifically compare and counterbalance the strengths be-tween a pair of presynaptic inputs. Rather, the decorrelatedpresynaptic strengths might emerge out of a local control ofsingle synapses that is embedded within a global form of synapseregulation. Following experimental observations support the in-volvement of per synapse basis regulation. First, in heterosynapticplasticity experiments, we find that, although the heterosynapticEPSC amplitude and PPR changes are inversely related to theinitial PPR of the stimulated input (Fig. 2 A and B), they do notshow a relationship to the initial EPSC amplitude of the stimu-lated input (Fig. S3). This lack of relationship to the basal EPSCamplitude indicates that the presynaptic strength change at thenonstimulated input does not necessarily depend on the totalnumber of active, stimulated synapses, and suggests that the pre-synaptic control could be executed independently of the co-incident activation of many synapses. Second, the preservation of

Fig. 6. Optogenetic hyperpolarization of astrocytes produces the correlation of PPR of convergent inputs reversibly. (A) Confocal section showing specificexpression of EGFP in GFAP-expressing astrocytes (red) but not in NeuN-expressing neurons (blue). (B) Astrocyte-specific expression of ArchT-GFP (green)throughout the GFAP-positive processes (red) with stereotyped astrocyte morphology. (C) Membrane potential hyperpolarization induced by light (n = 7 cells,paired t test, ***P = 0.0005) and representative membrane potential trace acquired in I-clamp in acute hippocampal slices. (D) Time course of the EPSCamplitude changes recorded from CA1 neurons near astrocytes infected with EGFP (n = 6 cells) or ArchT-GFP (n = 7 cells) in acute slices before, during, andafter 10-min light exposure (two-sample t test with Welch correction, *P = 0.00826). [Scale bars for representative traces (Upper), 50 pA, 20 ms.] (E) Timecourse of PPR disparity for the recordings in D (two sample-t test with Welch correction, **P = 0.00609). (F and G) Scatter plots comparing basal PPR of twoindependent inputs normalized to the first EPSC amplitude in control (Left) or with light exposure (Right) when nearby astrocytes are infected with EGFP(F: no light, n = 17 cells, r2 = 0.000003, P = 0.9950; light, n = 9 cells, r2 = 0.01, P = 0.7897) or ArchT-GFP (G: no light, n = 11 cells, r2 = 0.0007, P = 0.9392; light,n = 7 cells, r2 = 0.89, **P = 0.0015). (Inset) Representative PPR traces. (Scale bar, 40 ms.)

E2692 | www.pnas.org/cgi/doi/10.1073/pnas.1523717113 Letellier et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

the astrocyte-dependent decorrelation of PPR in dissociatedcultures indicates that the underlying mechanism does not relyon the unique topological organization of hippocampal areaCA1 astrocytes and pyramidal neurons (65). A regulatoryprocess that operates on the basis of individual synapses thatfunctions in concert with a global form of regulation is com-patible with a simple system in which the spatial organizationof the synaptic connections is variable. Although a single as-trocyte in area CA1 is likely to contact more than 105 synapses(65), given the extensive gap junctional coupling of astrocytes,the spatial extent to which the present form of regulationoperates remains to be addressed.The mechanism that ensures presynaptic strength heteroge-

neity we describe here highlights a form of homeostatic functionfor the postsynaptic neuron. That is, by compensating high prsynaptic inputs by weakening of other synaptic inputs received bythe postsynaptic neuron, it provides a boundary to the total ex-citatory drive. This compensatory activity functions over the timescale of minutes as illustrated by the rapid loss of decorrelationon blocking L-VGCCs or hyperpolarizing of astrocyte membraneby ArchT (Figs. 4G and 6 E and G). Notably, the overall averagepresynaptic strength received by the neuron is not altered underconditions when the decorrelation of presynaptic strengths be-tween convergent inputs is decreased (Fig. S5). This lack ofchange in average synaptic strength suggests that the mechanismthat decorrelates presynaptic strengths itself is not setting thelimits for the range of excitatory drive but functions in compli-ance with the homeostat to produce the variability.Based on our findings, we propose the following scheme for

initiating the decorrelation of presynaptic strengths (Fig. S9). Wepostulate that astrocytes impart a suppressive tone on excitatorysynaptic transmission (27, 66, 67); this could be mediated via thetonic release of inhibitory gliotransmitters such as ATP that ishydrolyzed extracellularly into adenosine or endocannabinoids(15, 64, 68). Glutamate released from active synapses (with ahigh pr) activates NMDARs on the surface of a nearby astrocyteprocess, causing a local depolarization of the astrocyte mem-brane. The membrane depolarization in turn, opens L-VGCCsthat engage intracellular Ca2+ signaling within astrocytes. We pro-pose that the change in global Ca2+ signaling serves to strengthenthe tonic inhibitory tone on excitatory synapses, whereas locally, theactive synapses are protected from the inhibition. The model in-volves a global signal that acts in concert with a local signal, and ittakes into account the astrocyte membrane depolarization, NMDARsand L-VGCCs on astrocytes, and astrocyte Ca2+ signaling thatare required for the observed PPR decorrelation. Notably, themodel also explains the reversal of the polarity of heterosynapticdepression by the L-VGCC inhibitor, which compromises theglobal Ca2+ signaling that is proposed to stimulate the inhibitorytone on excitatory synapses. That astrocytes in general exert aninhibitory tone on excitatory synapses is supported by the ArchTexperiments in which light-induced hyperpolarization of the as-trocyte membrane potential is accompanied by a rapid increaseof the EPSC amplitude (Fig. 6D) (66). At present we do notknow the nature of the protective signal that spares active syn-apses from the inhibitory tone. The apparent protection frominhibition could at least in part result from a local positivefeedback regulation by the active synapse (20) and likely in-volving astrocyte Ca2+ signaling that shows a highly complexspatial and temporal dynamics (15, 64, 69). Future studies war-rant a careful examination of the various synaptically evokedsignals to tease apart those crucial for controlling the presynapticstrength decorrelation.

Materials and MethodsCell Culture Preparations and Transfection. Hippocampal cultures were pre-pared from P0–P1 rats and plated at low density onto an astrocyte monolayer.The cultures were maintained as described previously (1) and used for elec-

trophysiology experiments at days in vitro (DIV) 9–14. To probe synapticvesicles dynamics, neurons were transfected with a plasmid encoding VGLUT1-pHluorin (kindly provided by Robert Edwards, University of California, SanFrancisco) at DIV 6 using Lipofectamine 2000 (Invitrogen). Cultures were usedfor imaging/electrophysiology experiments at DIV 9–14.

In Vivo Virus Injections. Infection of astrocytes in vivo was performed as de-scribed previously (70). For ArchT experiments, full-length hGFAP promoterconstructs were used, and 700 nL AAV-DJ/8 virus solutions were bilaterally in-jected into brains of 3-wk-old C57BL/6J mice. Targeting coordinates for dorsalhippocampus area CA1 were +1.9 mm anteroposterior from bregma, ±1.7 mmmediolateral, and +1.6 mm dorsoventral. Concentrations of recombinant AAV,determined by real-time quantitative PCR, were 1.0 × 1013 viral particles (vp)/mLfor both AAV-DJ/8–hGFAP-GFP and AAV-DJ/8–hGFAP-ArchT-EGFP. For GluN1conditional KO experiments, Cre and control AAV constructs were based on ashortened hGFAP promoter (56), and virus solutions were injected into brains of5- to 6-wk-old mice homozygous for the floxed gene encoding GluN1 (55). Thestereotaxic coordinates were as follows: X (anteroposterior from bregma),+1.9 mm; Y (mediolateral), ±1.9; Z (dorsoventral), +1.4 mm; and X, +3 mm; Y,±2.7; Z, +1.4 mm in both hemispheres. Recombinant AAV concentrations were9.0 × 1012 vp/mL for AAV-DJ/8-GFAP104-EGFP and 5.9 × 1012 vp/mL for AAV-DJ/8-GFAP104-nlsCre-mCherry, respectively.

Hippocampal Slice Preparations. Transverse hippocampal slices were obtainedfrom young (P14–P21) male Sprague–Dawley rats. For experiments involvingastrocyte expression of control GFP or ArchT, acute hippocampal sliceswere prepared from 4-wk-old mice at least 6 d after virus injection, andfor conditional GluN1 KO experiments, slices were made from 7- to 8-wk-oldmice at 2 wk after virus injection (above). For details, see SI Materialsand Methods.

Electrophysiology. Patch-clamp recordings from dissociated cultures andacute hippocampal slices were performed at room temperature using Axo-patch 200B and Multiclamp 700B amplifiers (Axon Instruments). For details,see SI Materials and Methods.

Live Cell Imaging. For estimating the RRP size at single boutons in dissociatedcultures, VGLUT1-pHluorin (VGLUT1-pH)–transfected neuron was patch-clamped along with a postsynaptic neuron filled with 100 μM AF 594 dye.The presynaptic neuron was stimulated at 20 Hz for 2 s (100 mV, 1- to 2-msstep depolarization) under V-clamp. Time-lapse VGLUT1-pH images wereacquired at 1Hz on an iXon EMCCD camera (Andor Technology) driven byMetamorph software (Universal Imaging). ΔF/F for identified active boutonswas measured after subtracting local background, where F is the initialfluorescence. This measurement was repeated before and 20 min after the1-Hz, 3-min stimulation of the presynaptic neuron.

Immunohistochemistry. Brain sections from mice injected with AAV wereprepared and processed for immunohistochemistry using standard proce-dures. For details, see SI Materials and Methods.

Light Stimulation. The light at excitation wavelength for red fluorescenceprobes from Spectra X light engine (Lumencor) was passed through a neuraldensity filter and a 560/40-nm filter, and delivered through a 40×, 0.8 N.A.water-immersion objective. Under this condition, the power density mea-sured at 560 nm was 2 mW/mm2.

Statistics. For normally distributed data (as determined by the d’Agostino–Pearson normality test), differences were tested using the paired or un-paired two-tailed Student t test or one-way ANOVA. The Mann–Whitneytest, the Wilcoxon test, or the Kruskal–Wallis test was used when criteria fornormality were not met. Graphpad Prism software was used for statisticalanalysis. Data are expressed as mean ± SEM.

ACKNOWLEDGMENTS. We thank Ayumu Konno, Hirokazu Hirai, HajimeHirase, Eunmi Hwang, and Joshua Johansen for DNA constructs; ThomasMcHugh for DNA constructs and GRIN1 floxed mice; Olivier Thoumine, AudePanatier, Hajime Hirase, Yasunori Hayashi, and Taro Toyoizumi for discussions;David Elliott, Izumi Kono, andMizuki Kurosawa for expert technical assistance;and Rachel Wong, Yasunori Hayashi, and Charles Yokoyama for comments onan earlier version of the manuscript. This work was supported by the MedicalResearch Council UK, European Union 7th Framework Program Grant HEALTH-F2-2009-241498 (“EUROSPIN” project), RIKEN Brain Science Institute, JSPSCore-to-Core Program, Grants-in-Aid for Scientific Research (15H04280) fromthe MEXT, and the Brain/MINDS from the Japan AMED.

Letellier et al. PNAS | Published online April 26, 2016 | E2693

NEU

ROSC

IENCE

PNASPL

US

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0

1. Branco T, Staras K, Darcy KJ, Goda Y (2008) Local dendritic activity sets releaseprobability at hippocampal synapses. Neuron 59(3):475–485.

2. Dobrunz LE, Stevens CF (1997) Heterogeneity of release probability, facilitation, anddepletion at central synapses. Neuron 18(6):995–1008.

3. Hanse E, Gustafsson B (2001) Factors explaining heterogeneity in short-term synapticdynamics of hippocampal glutamatergic synapses in the neonatal rat. J Physiol537(Pt 1):141–149.

4. Béïque JC, Na Y, Kuhl D, Worley PF, Huganir RL (2011) Arc-dependent synapse-specifichomeostatic plasticity. Proc Natl Acad Sci USA 108(2):816–821.

5. Andersen P, Sundberg SH, Sveen O, Wigström H (1977) Specific long-lasting poten-tiation of synaptic transmission in hippocampal slices. Nature 266(5604):736–737.

6. Lee S-JR, Escobedo-Lozoya Y, Szatmari EM, Yasuda R (2009) Activation ofCaMKII in single dendritic spines during long-term potentiation. Nature 458(7236):299–304.

7. Rabinowitch I, Segev I (2008) Two opposing plasticity mechanisms pulling a singlesynapse. Trends Neurosci 31(8):377–383.

8. Vitureira N, Goda Y (2013) Cell biology in neuroscience: The interplay betweenHebbian and homeostatic synaptic plasticity. J Cell Biol 203(2):175–186.

9. Vitureira N, Letellier M, Goda Y (2012) Homeostatic synaptic plasticity: From singlesynapses to neural circuits. Curr Opin Neurobiol 22(3):516–521.

10. Bonhoeffer T, Staiger V, Aertsen A (1989) Synaptic plasticity in rat hippocampal slicecultures: local “Hebbian” conjunction of pre- and postsynaptic stimulation leads todistributed synaptic enhancement. Proc Natl Acad Sci USA 86(20):8113–8117.

11. Lynch GS, Dunwiddie T, Gribkoff V (1977) Heterosynaptic depression: A postsynapticcorrelate of long-term potentiation. Nature 266(5604):737–739.

12. Royer S, Paré D (2003) Conservation of total synaptic weight through balanced syn-aptic depression and potentiation. Nature 422(6931):518–522.

13. Scanziani M, Malenka RC, Nicoll RA (1996) Role of intercellular interactions in het-erosynaptic long-term depression. Nature 380(6573):446–450.

14. Serrano A, Haddjeri N, Lacaille J-C, Robitaille R (2006) GABAergic network activationof glial cells underlies hippocampal heterosynaptic depression. J Neurosci 26(20):5370–5382.

15. Araque A, et al. (2014) Gliotransmitters travel in time and space. Neuron 81(4):728–739.

16. Min R, Santello M, Nevian T (2012) The computational power of astrocyte mediatedsynaptic plasticity. Front Comput Neurosci 6:93.

17. Di Castro MA, et al. (2011) Local Ca2+ detection and modulation of synaptic releaseby astrocytes. Nat Neurosci 14(10):1276–1284.

18. Darabid H, Arbour D, Robitaille R (2013) Glial cells decipher synaptic competition atthe mammalian neuromuscular junction. J Neurosci 33(4):1297–1313.

19. Kanemaru K, et al. (2014) In vivo visualization of subtle, transient, and local activity ofastrocytes using an ultrasensitive Ca(2+) indicator. Cell Reports 8(1):311–318.

20. Panatier A, et al. (2011) Astrocytes are endogenous regulators of basal transmission atcentral synapses. Cell 146(5):785–798.

21. Shigetomi E, Tong X, Kwan KY, Corey DP, Khakh BS (2011) TRPA1 channels regulateastrocyte resting calcium and inhibitory synapse efficacy through GAT-3. Nat Neurosci15(1):70–80.

22. Latour I, Gee CE, Robitaille R, Lacaille J-C (2001) Differential mechanisms of Ca2+responses in glial cells evoked by exogenous and endogenous glutamate in rat hip-pocampus. Hippocampus 11(2):132–145.

23. Perea G, Araque A (2005) Properties of synaptically evoked astrocyte calcium signalreveal synaptic information processing by astrocytes. J Neurosci 25(9):2192–2203.

24. Porter JT, McCarthy KD (1996) Hippocampal astrocytes in situ respond to glutamatereleased from synaptic terminals. J Neurosci 16(16):5073–5081.

25. Chen J, et al. (2013) Heterosynaptic long-term depression mediated by ATP releasedfrom astrocytes. Glia 61(2):178–191.

26. Manzoni OJ, Manabe T, Nicoll RA (1994) Release of adenosine by activation of NMDAreceptors in the hippocampus. Science 265(5181):2098–2101.

27. Pascual O, et al. (2005) Astrocytic purinergic signaling coordinates synaptic networks.Science 310(5745):113–116.

28. Bekkers JM, Stevens CF (1990) Presynaptic mechanism for long-term potentiation inthe hippocampus. Nature 346(6286):724–729.

29. Deisseroth K, Bito H, Tsien RW (1996) Signaling from synapse to nucleus: PostsynapticCREB phosphorylation during multiple forms of hippocampal synaptic plasticity.Neuron 16(1):89–101.

30. Goda Y, Stevens CF (1996) Long-term depression properties in a simple system.Neuron 16(1):103–111.

31. Bi GQ, Poo MM (1998) Synaptic modifications in cultured hippocampal neurons: De-pendence on spike timing, synaptic strength, and postsynaptic cell type. J Neurosci18(24):10464–10472.

32. Branco T, Staras K (2009) The probability of neurotransmitter release: Variability andfeedback control at single synapses. Nat Rev Neurosci 10(5):373–383.

33. Malinow R, Tsien RW (1990) Presynaptic enhancement shown by whole-cell record-ings of long-term potentiation in hippocampal slices. Nature 346(6280):177–180.

34. Mallart A, Martin AR (1967) An analysis of facilitation of transmitter release at theneuromuscular junction of the frog. J Physiol 193(3):679–694.

35. Ryan TA, Ziv NE, Smith SJ (1996) Potentiation of evoked vesicle turnover at in-dividually resolved synaptic boutons. Neuron 17(1):125–134.

36. Antonova I, et al. (2001) Rapid increase in clusters of presynaptic proteins at onset oflong-lasting potentiation. Science 294(5546):1547–1550.

37. McGuinness L, et al. (2010) Presynaptic NMDARs in the hippocampus facilitatetransmitter release at theta frequency. Neuron 68(6):1109–1127.

38. Nishiyama M, Hong K, Mikoshiba K, Poo MM, Kato K (2000) Calcium stores regulatethe polarity and input specificity of synaptic modification. Nature 408(6812):584–588.

39. Zhang JM, et al. (2003) ATP released by astrocytes mediates glutamatergic activity-dependent heterosynaptic suppression. Neuron 40(5):971–982.

40. Henneberger C, Papouin T, Oliet SHR, Rusakov DA (2010) Long-term potentiationdepends on release of D-serine from astrocytes. Nature 463(7278):232–236.

41. Pougnet J-T, et al. (2014) ATP P2X receptors downregulate AMPA receptor traffickingand postsynaptic efficacy in hippocampal neurons. Neuron 83(2):417–430.

42. Nicoll RA, Roche KW (2013) Long-term potentiation: Peeling the onion. Neuropharmacology74:18–22.

43. Barres BA, Chun LL, Corey DP (1989) Calcium current in cortical astrocytes: Inductionby cAMP and neurotransmitters and permissive effect of serum factors. J Neurosci9(9):3169–3175.

44. D’Ascenzo M, et al. (2004) Electrophysiological and molecular evidence of L-(Cav1),N- (Cav2.2), and R- (Cav2.3) type Ca2+ channels in rat cortical astrocytes. Glia 45(4):354–363.

45. Latour I, Hamid J, Beedle AM, Zamponi GW, Macvicar BA (2003) Expression of volt-age-gated Ca2+ channel subtypes in cultured astrocytes. Glia 41(4):347–353.

46. Serrano A, Robitaille R, Lacaille J-C (2008) Differential NMDA-dependent activation ofglial cells in mouse hippocampus. Glia 56(15):1648–1663.

47. Tippens AL, et al. (2008) Ultrastructural evidence for pre- and postsynaptic localiza-tion of Cav1.2 L-type Ca2+ channels in the rat hippocampus. J Comp Neurol 506(4):569–583.

48. Yan E, Li B, Gu L, Hertz L, Peng L (2013) Mechanisms for L-channel-mediated increasein [Ca(2+)]i and its reduction by anti-bipolar drugs in cultured astrocytes combinedwith its mRNA expression in freshly isolated cells support the importance of astrocyticL-channels. Cell Calcium 54(5):335–342.

49. Carmignoto G, Pasti L, Pozzan T (1998) On the role of voltage-dependent calciumchannels in calcium signaling of astrocytes in situ. J Neurosci 18(12):4637–4645.

50. Morquette P, et al. (2015) An astrocyte-dependent mechanism for neuronal rhyth-mogenesis. Nat Neurosci 18(6):844–854.

51. Hescheler J, Pelzer D, Trube G, Trautwein W (1982) Does the organic calcium channelblocker D600 act from inside or outside on the cardiac cell membrane? Pflugers Arch393(4):287–291.

52. Kovalchuk Y, Eilers J, Lisman J, Konnerth A (2000) NMDA receptor-mediatedsubthreshold Ca(2+) signals in spines of hippocampal neurons. J Neurosci 20(5):1791–1799.

53. Talbot MJ, Sayer RJ (1996) Intracellular QX-314 inhibits calcium currents in hippo-campal CA1 pyramidal neurons. J Neurophysiol 76(3):2120–2124.

54. Palygin O, Lalo U, Pankratov Y (2011) Distinct pharmacological and functionalproperties of NMDA receptors in mouse cortical astrocytes. Br J Pharmacol 163(8):1755–1766.

55. Tsien JZ, Huerta PT, Tonegawa S (1996) The essential role of hippocampal CA1 NMDAreceptor-dependent synaptic plasticity in spatial memory. Cell 87(7):1327–1338.

56. Ortinski PI, et al. (2010) Selective induction of astrocytic gliosis generates deficits inneuronal inhibition. Nat Neurosci 13(5):584–591.

57. Chow BY, et al. (2010) High-performance genetically targetable optical neural si-lencing by light-driven proton pumps. Nature 463(7277):98–102.

58. Beppu K, et al. (2014) Optogenetic countering of glial acidosis suppresses glial glu-tamate release and ischemic brain damage. Neuron 81(2):314–320.

59. Lichtman JW, Colman H (2000) Synapse elimination and indelible memory. Neuron25(2):269–278.

60. Clarke LE, Barres BA (2013) Emerging roles of astrocytes in neural circuit develop-ment. Nat Rev Neurosci 14(5):311–321.

61. Ikegaya Y, et al. (2013) Interpyramid spike transmission stabilizes the sparseness ofrecurrent network activity. Cereb Cortex 23(2):293–304.

62. Modi MN, Dhawale AK, Bhalla US (2014) CA1 cell activity sequences emerge afterreorganization of network correlation structure during associative learning. eLife3:e01982–e01982.

63. Carmignoto G, Haydon PG (2012) Astrocyte calcium signaling and epilepsy. Glia 60(8):1227–1233.

64. Rusakov DA, Bard L, Stewart MG, Henneberger C (2014) Diversity of astroglial func-tions alludes to subcellular specialisation. Trends Neurosci 37(4):228–242.

65. Khakh BS, Sofroniew MV (2015) Diversity of astrocyte functions and phenotypes inneural circuits. Nat Neurosci 18(7):942–952.

66. Sibille J, Zapata J, Teillon J, Rouach N (2015) Astroglial calcium signaling displaysshort-term plasticity and adjusts synaptic efficacy. Front Cell Neurosci 9:189.

67. Martín ED, et al. (2007) Adenosine released by astrocytes contributes to hypoxia-induced modulation of synaptic transmission. Glia 55(1):36–45.

68. Cunha RA (2008) Different cellular sources and different roles of adenosine: A1 receptor-mediated inhibition through astrocytic-driven volume transmission and synapse-restricted A2A receptor-mediated facilitation of plasticity. Neurochem Int 52(1-2):65–72.

69. Khakh BS, McCarthy KD (2015) Astrocyte calcium signaling: From observations tofunctions and the challenges therein. Cold Spring Harb Perspect Biol 7(4):a020404.

70. Cetin A, Komai S, Eliava M, Seeburg PH, Osten P (2006) Stereotaxic gene delivery inthe rodent brain. Nat Protoc 1(6):3166–3173.

E2694 | www.pnas.org/cgi/doi/10.1073/pnas.1523717113 Letellier et al.

Dow

nloa

ded

by g

uest

on

Dec

embe

r 1,

202

0