Interplay of multiple plasticities and activity dependent rules: data, models and possible impact on learning Gaetan Vignoud, Alexandre Mendes, Sylvie Perez, Jonathan Touboul*, Laurent Venance* (* equal contributions, email: fi[email protected]) College` de France, Center for Interdisciplinary Research in Biology, 11 Place Marcelin Berthelot, 75005 Paris, France Abstract the millisecond timing of the paired activities on either side of Synaptic plasticity, the activity-dependent evolution neuronal the synapse (Feldman, 2012). As such, plasticity in the stria- connectivity, is admitted to underpin learning and memory. A tum has been widely studied and a variety of spike-timing de- paradigmatic plasticity depending on the spike timings of cells pendent plasticity were observed in response to a large num- on both sides of the synapse (STDP), was experimentally ev- ber of presentations of a fixed pre- and post-synaptic stim- idenced through multiple repetitions of fixed pre- and post- ulation pattern. Those plasticities were attributed to a di- synaptic spike patterns. Theoretical and experimental com- verse neurotransmitters and pathways including for instance munities often implicitly admit that plasticity is gradually estab- endocannabinoids (eCB), NMDA receptors, AMPA receptors lished as stimulus patterns are presented. Here, we evaluate or voltage-sensitive calcium channels. Moreover, the vast ma- this hypothesis experimentally and theoretically in the stria- jority of the studies focused on cortical inputs, and much less tum, where (i) synapses may involve multiple pathways, or is known about thalamo-striatal plasticity rules. (ii) synaptic connections with multiple brain area exist. We The present contribution investigates on the possible inter- will present models and experiments leading to reevaluate play between multiple plasticities in the striatum, and consid- this hypothesis, showing that (i) multiple pathways may lead ers both the interaction between multiple pathways with var- to non-monotonic establishments of plasticity where plasticity ious dynamics and timescales at a single synapse, as well can be inverted depending on the number of stimulus presen- as multiple plasticities combining inputs from different brain tations, and (ii) that multiple connections with in particular tha- areas, here the combination of thalamo- and cortico-striatal lamus and cortex contribute to shaping their plasticities. We plasticity. The observation of complex history-dependent rules will propose a mathematical model building upon calcium tran- we lead us to discuss whether these are a negative aspect sients to precisely dissect these complex interplays on result- due to the biological substrate of plasticity, of it can lead an ing plasticities and reveal unexpected dependences on vari- advantage for learning. ables side variables (e.g. repetitions, frequency, temporal win- dow). These results invite to reevaluate how plasticity is im- Interplay of different mechanisms at the plemented as a global process, and to explore consequences cortico-striatal synapses on data processing capability and inspire new artificial intelli- gence techniques. We will start by presenting recent experimental plasticity data Keywords: Spike-Timing Dependent Plasticity; multiple path- from the dorsal striatum where the number of presentations of ways; mathematical modeling; calcium transients; a fixed timing of pre- and post-synaptic spikes is varied (Cui et al., 2015, 2016), and show that low numbers of pairing Introduction ( 5-10) induce an endocannabinoid-mediated spike-timing- The basal ganglia is a group of subcortical nuclei are involved dependent potentiation whereas larger numbers of pairings in goal-directed behavior, procedural learning, routine behav- ( 100) yield to anti-hebbian STDP, where potentiation is me- iors and cognition. Among those nuclei, the striatum con- diated by NMDA and depression is now dependent on en- stitutes the primary input layer to the basal ganglia and col- docannabinoids (see Fig. 1). This observation of the depen- lects information from both the cerebral cortex and the tha- dence of STDP to the number of pairings stands in contradic- lamus. The striatum constitutes a major site of memory for- tion with the implicit hypothesis of a monotonic establishment mation as the acquisition or extinction in the behavioral reper- of plasticity and points towards possible complex interplays of toire has been associated with cortico-striatal synaptic plas- pathways dynamics and timescales. To explore this hypothe- ticity (Koralek, Jin, Long, Costa, & Carmena, 2012). The stria- sis, we present here a new model building upon the calcium tum thus constitutes a key brain region to investigate in order hypothesis (Graupner & Brunel, 2012), generalized to com- to explore learning and cognition in the brain. bine multiple pathways within the same synapse. We show According to Hebbian theory, neural networks refine their that the model is able to reproduce accurately the data, and connectivity by patterned firing of action potentials in pre- predict a variety of response as a function of the number of and postsynaptic neurons. Spike-timing-dependent plasticity presentations but also their frequency. We develop this model (STDP) is a synaptic Hebbian learning rule that has been the to predict possible outcomes of Hebbian learning rules when focus of considerable attention in experimental and computa- those rely on multiple pathways, and will show that diverse tional neuroscience. STDP relies on the precise order and (but not any) type of plasticity profiles may emerge as a func- a b Cortex Synaptic strength Pre-synaptic element CB1R Χ Thalamus Glu 3.0 eCB-LTD 3.0 2.5 eCB-LTP 2.5 2.0 AMPAR mGluR 2.0 1.5 1.5 1.0 1.0 NMDAR 0.5 0.5 0.0 −0.10 −0.05 0.00 0.05 0.10 0.0 Δt (s) −0.10 −0.05 0.00 0.05 0.10 eCB Δt (s) synthesis - - + LTD + + LTD + 2+ LTP > θ LTP LTP > θ LTP Ca eCB,LTD eCB,LTD Striatum NMDAR-LTP > θeCB,LTP > θeCB,LTP > θNMDAR,LTP > θNMDAR,LTP VSCC Ca CamKII Calcium Calcium Cortex - Striatum Thalamus - Striatum Synapse 1 Synapse N Post-synaptic element Change in synaptic strength c 0.04 0.04 1.5 2.5 0.02 0.02 ) s Δt < 0 ( S T t 0.00 0.00 1.0 Δt > 0 Δ Model Mean of 2.0 experimental data −0.02 −0.02 0.5 −0.04 −0.04 1.5 −0.04 −0.02 0.00 0.02 0.04 −0.04 −0.02 0.00 0.02 0.04 Δt (s) ΔtCS (s) CS 1.0 Change in synaptic strength synaptic in Change 0.5 Figure 2: Model for the coupled cortico- and thalamo-striatal 0 20 40 60 80 100 Number of pairings STDP: top: schematic description of the cortical (green) and thalamic (purple) input to the a striatal cell (pink), with distinct types of plas- ticities (green +: experimental points, fitted with the calcium model - Figure 1: Model for the cortico-striatal STDP dark blue curve). Bottom: predictions of the interplay of Cortico and (a-b) Synapse with the three distinct plasticity pathways, (c) interplay thalamo-striatal synaptic efficacy together with experimental points of eCB and NMDA plasticity in the dorsal striatum as a function of (circles) for double stimulations. the number of pairings: model (curves) and experimental points from (Cui et al., 2015, 2016). Acknowledgments This project was funded in part by Inserm, FRM, Inria Paris- tion of the number of stimulus presentations, and propose a Rocquencourt, College` de France and CNRS Mission pour simple methodology to distinguish experimentally those sce- l’interdisciplinarite´ (project Big Data for the brain). narios. References Thalamic vs cortical inputs on striatum Cui, Y., Paille,´ V., Xu, H., Genet, S., Delord, B., Fino, E., . Venance, L. (2015). Endocannabinoids mediate bidirec- Using ex vivo electrophysiology (patch-clamp recordings in tional striatal spike-timing-dependent plasticity. J. Physiol. acute rodent brain slices), we show new experimental data Lond., 593(13), 2833–2849. revealing bidirectional STDP at both thalamic and cortical in- Cui, Y., Prokin, I., Xu, H., Delord, B., Genet, S., Venance, puts, driving opposing changes at striatal synapses, respec- L., & Berry, H. (2016). Endocannabinoid dynamics gate tively Hebbian and anti-Hebbian, see Fig. 2. Using a double- spike-timing dependent depression and potentiation. eLife, stimulation protocol, we also observed that thalamo-striatal 5, e13185. STDP produced a heterosynaptic plasticity at non-stimulated Feldman, D. E. (2012, August). The spike-timing de- cortico-striatal synapses, and conversely at thalamo-striatal pendence of plasticity. Neuron, 75(4), 556–571. doi: synapses. These findings highlight the major impact of pre- 10.1016/j.neuron.2012.08.001 cise timing in cortical and thalamic activity for the memory Graupner, M., & Brunel, N. (2012). Calcium-based plastic- engram at striatal synapses. Via this heterosynaptic plas- ity model explains sensitivity of synaptic changes to spike ticity, thalamus efficiently shapes the cortico-striatal efficacy pattern, rate, and dendritic location. PNAS, 109(10), 3991– changes. Thus, these findings highlight the major impact of 3996. precise timing in cortical and thalamic activity for the mem- Koralek, A. C., Jin, X., Long, J. D., Costa, R. M., & Car- ory engram at striatal synapses. We further investigate ex mena, J. M. (2012). Corticostriatal plasticity is necessary vivo and in vivo the functional consequences of such mir- for learning intentional neuroprosthetic skills. Nature, 483, rored plasticity. Moreover, we will further develop our calcium 331–335. based-model to explore the possible origins of these mirror plasticities, and particularly the possible role of calcium diffu- sion across distinct compartments within the cell. This model, fitted to the data, allows to predict a number of unexpected plasticities for specific combinations of timings of thalamic and cortical spikes relative to striatal spikes, both on the thalamo- striatal and cortico-striatal synapses, that we are currently val- idating experimentally..