NMDA RECEPTOR-DEPENDENT LTP OF iClOSSY FIBRE GRANII!,E CELL EPSCS IN THE RAT CEREBEL1,URl. 1.: II (11 l ~ i s i ~ d ~ ~ g i o lX4iigc~lo, I' /?(>\.\I, .S .4r1~1iriw, 1'. Y > I , ~ / I ~ I/\IIIII~(I (;eiwrci/e, ~ / t i ~ w r w c01 i ' l'uv1u 1-271110, /'flv!u, 1 1 d ~ Long-tenn ct\a~igcs III syiiapt~c t~wsmmsslol~ac thought to plaj' an tmportant role m bram Icmimg and compotmon lti part~cular.long-tenn potenhation (LI'P) o f synaptlc h a l s m ~ s s ~ o nhas been repotted at glutamaterg~c synapses explessmg N-rnetl~yl-D-aspartate (NMDA) receptors m the h ~ p p o c a m p u sand neocorres NMDA receptors are actwated durmg rnosny fibre - grarlule cell synaptic transmmlon In the cerebellum T o inbe<t?$atr whethcr t h ~ ssytla?sr could delelop !.'rP, n r pe:rfonned whole-cell patch-clamp ~ e c o r d ~ n gfrom s gr;umle ells In rat cerebella^ s l ~ c e s iP19-P23) A b r ~ e fIiiyh-ficquency mossy fibre s h m u l a t ~ o r (theta-burst) ~ pawed w ~ t hmembrane depolarlsatton (-40 mV) led to a stable EPSC potentlatton for 30 rmnutes Pairmg w t h a lowel potential (-60 inV). o~ In the presence o f the NMDA receptor blocker. APV. prevented LTP LTP was expressed t h o u g h an increase III both the NMDA and non-NCID;\ receptor-mediated current (32113% and 47-tl3"o Increase. rcspect~vely: n=1 O), atid potentlation of NMDA receptor-dependent tra~lsinlssioowas not prevented by blockmg non-NMDA receptors with C N Q X These obsemations are consistent w ~ t ha voltage-dependent mduction mechrui~sm based o n NMDA receptor actlvatlon. LTP at the mossy fibre - gramlle cell relay may have important implications for cerebellw computation and learning. < S I I ~ ~ I WbyI U hl~1l<.Yl, ~ ( ' N / <i i d IN/.;i/ o/ 11d.y CI SE LONG-TERM SYNAPTIC PLASTICITY I S RELATED TO SPIKE FLRING AT A CEREBELLAR INHIBITORY SYNAPSE. T . Sejnowski, Computational Neurobiology I,aboratory, Salk Institnle, Sail Diego CA 92186-5800 We previously demonstrated that coniputational models of pyramidal neurons are capable of a simplified version of indirect, or model-based, reinforcement learning. They can detect temporal correlations in their excitatory input, use them to predict future inputs, infer the impact of their present firing on future inputs, detect which excilatory inputs are followcd by a reward signal like dopamine, and change their firing patterns so thitl they lhave a positive impact o n the occurence of rewards A major limitation of this learning algorithm is t h a t neurons can not easily detect delayed impacts, hke those produced by the motor output of the neocortex on its sensory input. Here we demonstrate computat~onallyhow th!s problem can bc solved by a layered network of similar, but more specialized neurons, whose connectloos resemble those of layers 2&3, 5 , and 6 of the neocoriex. Layers 2k3 in our model detect t e n ~ p o r a linvariances on nhich predictions of the sensory input can be based. Neurons of layer 6 learn to predict the sensory input t o the tbalamus. This works besl if the netwark goes through a sleep-wake cylcle. with layer 6 driving the thalamus during sleep. Neurons of layer 5 learn to detect how their motor output affects the predictions made by layer 6 neurons. T h e learning rule in basal dendrites is similar t o temporal difference learnmg, but the apical learning rule is quite different, as it is specialized in detecting the feedback from other, nearby neurons. Together, these unsupervised learning processes produce a network that call perform efiicient, rnodel-based reinforcement learning in layer 5 , A few reward signals are enough t o change the motor output in the right way, even if ~ t impact s on sensory input is quite delayed. Supported by the german DFG grant E l 37311.2 ACTIVATION OF MUSCARlNlC MI RECEPTORS FACILITATES SYNAPTIC PLASTICITY IN LAYER 213 01;KI7TEN VISUAL CORTEX. L. Kolic*. 0. Gu. R.M. Douelas. M. Cvnader. Dept. ofOphthalniology, Univ. of British Colunibia, 2550 W~llonSt Vancouver, B C , Canada VSL 3N9 Both acetylchol~ne-~ned~ated rleurotransmlssion and modulat~onhave been reported In the vlsual cortex Muscarlnrc MI, M2 and M3 receptors are trans~entlyexpressed In layer 4 ~n the k~ttcn v~sual cortex duriilg the ii~st fen weeks ot postnetal develop~nent AAer postnatal da) 70 an adult lanimar d~stribut~on develops wltll the most dense labelmg In the superficial lavers 111 a d d ~ t ~ o n ,agon~stsof chol~nerglc receptols have been dcmonstraled to fac~l~late the lnduct~onof LTP In vlsual cortlcal sl~cessuggestmg a role of both nlcotmlc and muscar~n~c receptors In the control d actwty-dependent plast~c~ty In develop~ngv~soalLoRex To test the role of MI and M2 muscarlntc receptors in v~sualcort~calplast~c~ty, we recorded field potentials 111 v~sualcortex sl~cesfrom 70-90 day old kittens r ~ e l dpotentials were ~ e c o ~ d ehd m layers 4 and 213 after wh~tematter stnnulat~on before and after a perlod of low frequency (LFS, IHz, 15 mm), h ~ g h frequency (HFS), or theta-burst (TBS) st~mulation At these ages, the protocols used almost never Induced LTD or LTP In mductmn of LTP 111 layers 213 or 4 The MI receptor agon~stMcN-A-343 fac~l~tated layer 213 (LTP) ~nducedw ~ t hHFS (8117 sl~ces)and TBS st~niulat~on (418 sllces), was largely blocked by the MI receptor with no effm ~n layer 4 T h ~ sfacll~tat~on antagonist plrenzepnie There was no facil~tatoryeffect of MI receptor actlvatlon wlth LFS st~rnulat~onO\otlemor~ne M, the M2 receptor agonlst dld not induce any facll~latolye f f ~ w t ~ t hall three stmulat~oa protocols The receptor subtype- and lam~nar-speclficfac~l~tatoly effects suggest that MI receptors might play a permlsslvc role in evperlence dependent changes ofsynapt~ceflicacy, ~ n e d ~ a t ~the r l g gatlng effects of the cholinergic nonret~nalprolection system The pattern and the level of thea expressloll may detemune the probab~lityand the level of actwry dependent changes durmg development of kmen v w a l cortex Supported by Med~calResearch Counc~lof Canada