{"title": "Storing Covariance by the Associative Long-Term Potentiation and Depression of Synaptic Strengths in the Hippocampus", "book": "Advances in Neural Information Processing Systems", "page_first": 394, "page_last": 401, "abstract": "", "full_text": "394 \n\nSTORING COVARIANCE BY THE ASSOCIATIVE \nLONG\u00b7TERM POTENTIATION AND DEPRESSION \nOF SYNAPTIC STRENGTHS IN THE HIPPOCAMPUS \n\nPatric K. Stanton\u00b7 and Terrence J. Sejnowskit \n\nDepartment of Biophysics \nJohns Hopkins University \n\nBaltimore, MD 21218 \n\nABSTRACT \n\nIn modeling studies or memory based on neural networks, both the selective \nenhancement and depression or synaptic strengths are required ror effident storage \nor inrormation (Sejnowski, 1977a,b; Kohonen, 1984; Bienenstock et aI, 1982; \nSejnowski and Tesauro, 1989). We have tested this assumption in the hippocampus, \na cortical structure or the brain that is involved in long-term memory. A brier, \nhigh-frequency activation or excitatory synapses in the hippocampus produces an \nincrease in synaptic strength known as long-term potentiation, or L TP (BUss and \nLomo, 1973), that can last ror many days. LTP is known to be Hebbian since it \nrequires the simultaneous release or neurotransmitter from presynaptic terminals \ncoupled with postsynaptic depolarization (Kelso et al, 1986; Malinow and Miller, \n1986; Gustatrson et al, 1987). However, a mechanism ror the persistent reduction or \nsynaptic strength that could balance LTP has not yet been demonstrated. We stu(cid:173)\ndied the associative interactions between separate inputs onto the same dendritic \ntrees or hippocampal pyramidal cells or field CAl, and round that a low-frequency \ninput which, by itselr, does not persistently change synaptic strength, can either \nincrease (associative L TP) or decrease in strength (associative long-term depression \nor LTD) depending upon whether it is positively or negatively correlated in time \nwith a second, high-frequency bursting input. LTP or synaptic strength is Hebbian, \nand LTD is anti-Hebbian since it is elicited by pairing presynaptic firing with post(cid:173)\nsynaptic hyperpolarization sufficient to block postsynaptic activity. Thus, associa(cid:173)\ntive L TP and associative L TO are capable or storing inrormation contained in the \ncovariance between separate, converging hippocampal inputs \u2022 \n\n\u2022 Present address: Dep~ents of NeW'Oscience and Neurology, Albert Einstein College \nof Medicine, 1410 Pelham Parkway South, Bronx, NY 10461 USA. \n\ntPresent address: Computational Neurobiology Laboratory, The Salk Institute, P.O. Box \n85800, San Diego, CA 92138 USA. \n\n\fStoring Covariance by Synaptic Strengths in the Hippocampus \n\n395 \n\nINTRODUCTION \n\nAssociative L TP can be produced in some hippocampal neuroos when low(cid:173)\nfrequency. (Weak) and high-frequency (Strong) inputs to the same cells are simultane(cid:173)\nously activated (Levy and Steward, 1979; Levy and Steward, 1983; Barrionuevo and \nBrown, 1983). When stimulated alone, a weak input does not have a long-lasting effect \non synaptic strength; however, when paired with stimulation of a separate strong input \nsufficient to produce homo synaptic LTP of that pathway, the weak pathway is associa(cid:173)\ntively potentiated. Neural network modeling studies have predicted that, in addition to \nthis Hebbian form of plasticity, synaptic strength should be weakened when weak and \nstrong inputs are anti-correlated (Sejnowski, 1977a,b; Kohonen, 1984; Bienenstock et al, \n1982; Sejnowski and Tesauro, 1989). Evidence for heterosynaptic depression in the hip(cid:173)\npocampus has been found for inputs that are inactive (Levy and Steward, 1979; Lynch et \nal, 1977) or weakly active (Levy and Steward, 1983) during the stimulation of a strong \ninput, but this depression did not depend on any pattern of weak input activity and was \nnot typically as long-lasting as LTP. \n\nTherefore, we searched for conditions under which stimulation of a hippocampal \npathway, rather than its inactivity, could produce either long-term depression or potentia(cid:173)\ntion of synaptic strengths, depending on the pattern of stimulation. The stimulus para(cid:173)\ndigm that we used, illustrated in Fig. I, is based on the finding that bursts of stimuli at 5 \nHz are optimal in eliciting LTP in the hippocampus (Larson and Lynch, 1986). A high(cid:173)\nfrequency burst (S'IRONG) stimulus was applied to Schaffer collateral axons and a low(cid:173)\nfrequency (WEAK) stimulus given to a separate subicular input coming from the oppo(cid:173)\nsite side of the recording site, but terminating on dendrites of the same population of CAl \npyramidal neurons. Due to the rhythmic nature of the strong input bursts, each weak \ninput shock could be either superimposed on the middle of each burst of the strong input \n(IN PHASE), or placed symmetrically between bursts (OUT OF PHASE). \n\nRESULTS \n\nExtracellular evoked field potentials were recorded from the apical dendritic and \nsomatic layers of CAl pyramidal cells. The weak stimulus train was first applied alone \nand did not itself induce long-lasting changes. The strong site was then stimulated alone, \nwhich elicited homosynaptic LTP of the strong pathway but did not significantly alter \namplitude of responses to the weak input. When weak and strong inputs were activated \nIN PHASE, there was an associative L TP of the weak input synapses, as shown in Fig. \n2a. Both the synaptic excitatory post-synaptic potential (e.p.s.p.) (Ae.p.s.p. = +49.8 \u00b1 \n7.8%, n=20) and population action potential (&Pike = +65.4 \u00b1 16.0%, n=14) were \nsignificantly enhanced for at least 60 min up to 180 min following stimulation. \n\nIn contrast, when weak and strong inputs were applied OUT OF PHASE, they eli(cid:173)\ncited an associative long-term depression (L TO) of the weak input synapses, as shown in \nFig. 2b. There was a marked reduction in the population spike (-46.5 \u00b1 11.4%, n=10) \nwith smaller decreases in the e.p.s.p. (-13.8 \u00b1 3.5%, n=13). Note that the stimulus pat(cid:173)\nterns applied to each input were identical in these two experiments, and only the relative \n\n\f396 \n\nStanton and Sejnowski \n\nphase of the weak and strong stimuli was altered. With these stimulus patterns. synaptic \nstrength could be repeatedly enhanced and depressed in a single slice. as illustrated in Fig \n2c. As a control experiment to determine whether information concerning covariance \nbetween the inputs was actually a determinant of plasticity. we combined the in phase \nand out of phase conditions, giving both the weak input shocks superimposed on the \nbursts plus those between the bursts. for a net frequency of 10 Hz. This pattern. which \nresulted in zero covariance between weak and strong inputs. produced no net change in \nweak input synaptic strength measmed by extracellular evoked potentials. Thus. the asso-\n\na \n\nb \n\nA.SSOCIA.TIVE STIMULUS PA.RA.DIGMS \n\nPOSJTIVE.L Y CORKELA TED \u00b7 \"IN PHASE\" \n\n~K~~ _I~ __ ~I ____ ~I ____ ~I_ \nSI1IONG,NJO\\IT ..u.Jj1ll11l..-1 ---1&1111 ..... \n11 ---,I~IIII \nNEGATIVELY CORRELATED\u00b7 'our OF PHASE\" \n\n11 ---1&1111 ..... \n\nW[AKIN'lTf \n\nSTIONG 'N'''' ~I \n\n11111 \n--,-; \n\n11111 \n\n11111 \n\nFigure 1. Hippocampal slice preparation and stimulus paradigms. a: The in vitro hippo(cid:173)\ncampal slice showing recording sites in CAl pyramidal cell somatic (stratum pyrami(cid:173)\ndale) and dendritic (stratum radiatum) layers. and stimulus sites activating Schaffer col(cid:173)\nlateral (STRONG) and commissural (WEAK) afferents. Hippocampal slices (400 Jlm \nthick) were incubated in an interface slice chamber at 34-35 0 C. Extracellular (1-5 M!l \nresistance, 2M NaCI filled) and intracellular (70-120 M n. 2M K-acetate filled) record(cid:173)\ning electrodes. and bipolar glass-insulated platinum wire stimulating electrodes (50 Jlm \ntip diameter). were prepared by standard methods (Mody et al, 1988). b: Stimulus para(cid:173)\ndigms used. Strong input stimuli (STRONG INPUT) were four trains of 100 Hz bursts. \nEach burst had 5 stimuli and the interburst interval was 200 msec. Each train lasted 2 \nseconds for a total of 50 stimuli. Weak input stimuli (WEAK INPUT) were four trains of \nshocks at 5 Hz frequency. each train lasting for 2 seconds. When these inputs were IN \nPHASE. the weak single shocks were superimposed on the middle of each burst of the \nstrong input. When the weak input was OUT OF PHASE. the single shocks were placed \nsymmetrically between the bursts. \n\n\fStoring Covariance by Synaptic Strengths in the Hippocampus \n\n397 \n\nciative LTP and LTD mechanisms appear to be balanced in a manner ideal for the \nstorage of temporal covariance relations. \n\nThe simultaneous depolarization of the postsynaptic membrane and activation of \n\nglutamate receptors of the N-methyl-D-aspartate (NMDA) subtype appears to be neces(cid:173)\nsary for LTP induction (Collingridge et ai, 1983; Harris et al, 1984; Wigstrom and Gus(cid:173)\ntaffson, 1984). The SJ?read of current from strong to weak synapses in the dendritic tree, \n\nd \n\nASSOCIATIVE LON(;.TE~ I'OTENTIATION \n\n!!Ll!!!!. ---\n\nb \n\nASSOCIATIVE LONG-TE~ DE,/tESSION \n\nI \n\n\u2022 \n\n11111 \n\n11111. \nI \n\nc \n\ne... \n\nI \n\nI \n\nI \n\nI \n\nFigure 2. mustration of associative long-term potentiation (LTP) and associative long(cid:173)\nterm depression (LTD) using extracellular recordings. a: Associative LTP of evoked \nexcitatory postsynaptic potentials (e.p.s.p.'s) and population action potential responses in \nthe weak inpuL Test responses are shown before (Pre) and 30 min after (post) applica(cid:173)\ntion of weak stimuli in phase with the coactive strong input. b: Associative LTD of \nevoked e.p.s.p.'s and population spike responses in the weak input. Test responses are \nshown before (Pre) and 30 min after (post) application of weak stimuli out of phase with \nthe coactive strong input. c: Time course of the changes in population spike amplitude \nobserved at each input for a typical experiment. Test responses from the strong input (S, \nopen circles), show that the high-frequency bursts (5 pulses/l00 Hz, 200 msec interburst \ninterval as in Fig. 1) elicited synapse-specific LTP independent of other input activity. \nTest responses from the weak input (W. filled circles) show that stimulation of the weak \npathway out of phase with the strong one produced associative LTD (Assoc LTD) of this \ninput. Associative LTP (Assoc LTP) of the same pathway was then elicited following in \nphase stimulation. Amplitude and duration of associative LTD or L TP could be increased \nby stimulating input pathways with more trains of shocks. \n\n\f398 \n\nStanton and Sejnowski \n\ncoupled with release of glutamate from the weak inputs, could account for the ability of \nthe strong pathway to associatively potentiate a weak one (Kelso et al, 1986; Malinow \nand Miller, 1986; Gustaffson et al, 1987). Consistent with this hypothesis, we find that \nthe NMDA receptor antagonist 2-amino-S-phosphonovaleric acid (APS, 10 J.1M) blocks \ninduction of associative LTP in CAl pyramidal neurons (data not shown, n=S). In con(cid:173)\ntrast, the application of APS to the bathing solution at this same concentration had no \nsignificant effect on associative LTD (data not shown, n=6). Thus, the induction of LTD \nseems to involve cellular mechanisms different from associative LTP. \n\nThe conditions necessary for LTD induction were explored in another series of \nexperiments using intracellular recordings from CAl pyramidal neurons made using \nstandard techniques (Mody et al, 1988). Induction of associative L TP (Fig 3; WEAK \nS+W IN PHASE) produced an increase in amplitude of the single cell evoked e.p.s.p. and \na lowered action potential threshold in the weak pathway, as reported previously (Bar(cid:173)\nrionuevo and Brown, 1983). Conversely, the induction of associative LTD (Fig. 3; \nWEAK S+W OUT OF PHASE) was accompanied by a long-lasting reduction of e.p.s.p. \namplitude and reduced ability to elicit action potential firing. As in control extracellular \nexperiments, the weak input alone produced no long-lasting alterations in intracellular \ne.p.s.p.'s or firing properties, while the strong input alone yielded specific increases of \nthe strong pathway e.p.s.p. without altering e.p.s.p. 's elicited by weak input stimulation. \n\nPRE \n\n30 min POST \n\nS+W OUT OF PHASE \n\n30 min POST \n\nS+W IN PHASE \n\nFigure 3. Demonstration of associative L TP and LTD using intracellular recordings from \na CAl pyramidal neuron. Intracellular e.p.s.p.'s prior to repetitive stimulation (pre), 30 \nmin after out of phase stimulation (S+ W OUT OF PHASE), and 30 min after subse(cid:173)\nquent in phase stimuli (S+ W IN PHASE). The strong input (Schaffer collateral side, \nlower traces) exhibited LTP of the evoked e.p.s.p. independent of weak input activity. \nOut of phase stimulation of the weak (Subicular side, upper traces) pathway produced a \nmarked, persistent reduction in e.p.s.p. amplitude. In the same cell, subsequent in phase \nstimuli resulted in associative L TP of the weak input that reversed the LTD and enhanced \namplitude of the e.p.s.p. past the original baseline. (RMP = -62 mY, RN = 30 MO) \n\n\fStoring Covariance by Synaptic Strengths in the Hippocampus \n\n399 \n\nA weak stimulus that is out of phase with a strong one anives when the postsynap(cid:173)\ntic neuron is hyperpolarized as a consequence of inhibitory postsynaptic potentials and \nafterhyperpolarization from mechanisms intrinsic to pyramidal neurons. This suggests \nthat postsynaptic hyperpolarization coupled with presynaptic activation may trigger L'ID. \nTo test this hypothesis, we injected current with intracellular microelectrodes to hyperpo(cid:173)\nlarize or depolarize the cell while stimulating a synaptic input. Pairing the injection of \ndepolarizing current with the weak input led to L TP of those synapses (Fig. 4a; STIM; \n\na \n\nPRE \n\n\u2022 \u2022 IDPOST \n\nS'I1M \u2022 DEPOL \n\nr \n\" , i \n-Jj \n\nCOI'ITROL \n(W.c:ULVllj \n\nb \n\nPRE \n\n~l\"V \n\nlS.,.c \n\nI \n\n--\" \\ \"----\n\nlOlIIin POST \n\nSTlM \u2022 HYPERPOL \n\nFigure 4. Pairing of postsynaptic hyperpolarization with stimulation of synapses on CAl \nhippocampal pyramidal neurons produces L'ID specific to the activated pathway, while \npairing of postsynaptic depolarization with synaptic stimulation produces synapse(cid:173)\nspecific LTP. a: Intracellular evoked e.p.s.p.'s are shown at stimulated (STIM) and \nunstimulated (CONTROL) pathway synapses before (Pre) and 30 min after (post) pair(cid:173)\ning a 20 mY depolarization (constant current +2.0 nA) with 5 Hz synaptic stimulation. \nThe stimulated pathway exhibited associative LTP of the e.p.s.p., while the control, \nunstimulated input showed no change in synaptic strength. (RMP = -65 mY; RN = 35 \nMfl) b: Intracellular e.p.s.p. 's are shown evoked at stimulated and control pathway \nsynapses before (Pre) and 30 min after (post) pairing a 20 mV hyperpolarization (con(cid:173)\nstant current -1.0 nA) with 5 Hz synaptic stimulation. The input (STIM) activated during \nthe hyperpolarization showed associative LTD of synaptic evoked e.p.s.p.'s, while \nsynaptic strength of the silent input (CONTROL) was unaltered. (RMP = -62 m V; RN = \n38M!l) \n\n\f400 \n\nStanton and Sejnowski \n\n+64.0 -9.7%, n=4), while a control input inactive during the stimulation did not change \n(CONTROL), as reported previously (Kelso et al, 1986; Malinow and Miller, 1986; Gus(cid:173)\ntaffson et al, 1987). Conversely, prolonged hyperpolarizing current injection paired with \nthe same low-frequency stimuli led to induction of LTD in the stimulated pathway (Fig. \n4b; STIM; -40.3 \u00b1 6.3%, n=6). but not in the unstimulated pathway (CONTROL). The \napplication of either depolarizing current, hyperpolarizing current, or the weak 5 Hz \nsynaptic stimulation alone did not induce long-term alterations in synaptic strengths. \nThus. hyperpolarization and simultaneous presynaptic activity supply sufficient condi(cid:173)\ntions for the induction of LTD in CAl pyramidal neurons. \n\nCONCLUSIONS \n\nThese experiments identify a novel fono of anti-Hebbian synaptic plasticity in the \nhippocampus and confirm predictions made from modeling studies of information storage \nin neural networks. Unlike previous reports of synaptic depression in the hippocampus, \nthe plasticity is associative, long-lasting, and is produced when presynaptic activity \noccurs while the postsynaptic membrane is hyperpolarized. In combination with Hebbian \nmechanisms also present at hippocampal synapses. associative LTP and associative LTD \nmay allow neurons in the hippocampus to compute and store covariance between inputs \n(Sejnowski, 1977a,b; Stanton and Sejnowski. 1989). These finding make temporal as \nwell as spatial context an important feature of memory mechanisms in the hippocampus. \nElsewhere in the brain, the receptive field properties of cells in cat visual cortex \ncan be altered by visual experience paired with iontophoretic excitation or depression of \ncellular activity (Fregnac et al, 1988; Greuel et al, 1988). In particular, the chronic hyper(cid:173)\npolarization of neurons in visual cortex coupled with presynaptic transmitter release leads \nto a long-teno depression of the active. but not inactive, inputs from the lateral geniculate \nnucleus (Reiter and Stryker, 1988). Thus. both Hebbian and anti-Hebbian mechanisms \nfound in the hippocampus seem to also be present in other brain areas, and covariance of \nfiring patterns between converging inputs a likely key to understanding higher cognitive \nfunction. \n\nThis research was supported by grants from the National Science Foundation and \nthe Office of Naval research to TJS. We thank Drs. Charles Stevens and Richard Morris \nfor discussions about related experiments. \n\nRererences \n\nBienenstock, E., Cooper. LN. and Munro. P. Theory for the development of neuron \nselectivity: orientation specificity and binocular interaction in visual cortex. J. Neu(cid:173)\nrosci. 2. 32-48 (1982). \n\nBarrionuevo, G. and Brown, T.H. Associative long-teno potentiation in hippocampal \n\nslices. Proc. Nat. Acad. Sci. (USA) 80, 7347-7351 (1983). \n\nBliss. T.V.P. and Lomo, T. Long-lasting potentiation of synaptic ttansmission in the den(cid:173)\ntate area of the anaesthetized rabbit following stimulation of the perforant path. J. \nPhysiol. (Lond.) 232. 331-356 (1973). \n\n\fStoring Covariance by Synaptic Strengths in the Hippocampus \n\n401 \n\nCollingridge, GL., Kehl, SJ. and McLennan, H. Excitatory amino acids in synaptic \ntransmission in the Schaffer collateral-commissural pathway of the rat hippocampus. J. \nPhysiol. (Lond.) 334, 33-46 (1983). \n\nFregnac, Y., Shulz, D., Thorpe, S. and Bienenstock, E. A cellular analogue of visual cort(cid:173)\n\nical plasticity. Nature (Lond.) 333, 367-370 (1988). \n\nGreuel. J .M.. Luhmann. H.J. and Singer. W. Pharmacological induction of use(cid:173)\n\ndependent receptive field modifications in visual cortex. Science 242,74-77 (1988). \n\nGustafsson, B., Wigstrom, H., Abraham, W.C. and Huang. Y.Y. Long-term potentiation \nin the hippocampus using depolarizing current pulses as the conditioning stimulus to \nsingle volley synaptic potentials. J. Neurosci. 7, 774-780 (1987). \n\nHarris. E.W., Ganong, A.H. and Cotman, C.W. Long-term potentiation in the hippo(cid:173)\ncampus involves activation of N-metbyl-D-aspartate receptors. Brain Res. 323, 132-\n137 (1984). \n\nKelso, S.R.. Ganong, A.H. and Brown, T.H. Hebbian synapses in hippocampus. Proc. \n\nNatl. Acad. Sci. USA 83, 5326-5330 (1986). \n\nKohonen. T. Self-Organization and Associative Memory. (Springer-Verlag. Heidelberg, \n\n389 (1977). \n\nStanton, P.K. and Sejnowski, TJ. Associative long-term depression in the hippocampus: \n\nEvidence for anti-Hebbian synaptic plasticity. Nature (Lond.), in review. \n\nWigstrom, H. and Gustafsson, B. A possible correlate of the postsynaptic condition for \nlong-lasting potentiation in the guinea pig hippocampus in vitro. Neurosci. Lett. 44, \n327\u00b7332 (1984). \n\n1984). \n\nLarson. J. and Lynch. G. Synaptic potentiation in hippocampus by patterned stimulation \n\ninvolves two events. Science 232, 985-988 (1986). \n\nLevy. W.B. and Steward, O. Synapses as associative memory elements in the hippocam(cid:173)\n\npal formation. Brain Res. 175,233-245 (1979). \n\nLevy. W.B. and Steward, O. Temporal contiguity requirements for long-term associative \n\npotentiation/depression in the hippocampus. Neuroscience 8, 791-797 (1983). \n\nLynch. G.S., Dunwiddie. T. and Gribkoff. V. Heterosynaptic depression: a postsynaptic \n\ncorrelate oflong-term potentiation. Nature (Lond.) 266. 737-739 (1977). \n\nMalinow. R. and Miller, J.P. Postsynaptic hyperpolarization during conditioning reversi(cid:173)\n\nbly blocks induction of long-term potentiation Nature (Lond.)32.0. 529-530 (1986). \n\nMody. I.. Stanton. P K. and Heinemann. U. Activation of N-methyl-D-aspartate \n(NMDA) receptors parallels changes in cellular and synaptic properties of dentate \ngyrus granule cells after kindling. J. Neurophysiol. 59. 1033-1054 (1988). \n\nReiter, H.O. and Stryker, M.P. Neural plasticity without postsynaptic action potentials: \nLess-active inputs become dominant when kitten visual cortical cells are pharmacolog(cid:173)\nically inhibited. Proc. Natl. Acad. Sci. USA 85, 3623-3627 (1988). \n\nSejnowski, T J. and Tesauro, G. Building network learning algorithms from Hebbian \nsynapses, in: Brain Organization and Memory J L. McGaugh, N.M. Weinberger, and \nG. Lynch, Eds. (Oxford Univ. Press, New York, in press). \n\nSejnowski, T J. Storing covariance with nonlinearly interacting neurons. J. Math. Biol(cid:173)\n\nogy 4, 303-321 (1977). \n\nSejnowski, T. J. Statistical constraints on synaptic plasticity. J. Theor. Biology 69, 385-\n\n\f", "award": [], "sourceid": 100, "authors": [{"given_name": "Patric", "family_name": "Stanton", "institution": null}, {"given_name": "Terrence", "family_name": "Sejnowski", "institution": null}]}