H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Project-Team BEAGLE INRIA Rhône-Alpes LIRIS UMR CNRS 5205 Université de Lyon, France DevLeaNN - Paris – OCT 201 Hugues BERRY Unconventional forms of plasticity: beyond the synaptic Hebbian paradigm

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Origin of synaptic plasticity: Donald Hebb 1949, Hebb's PhD Thesis: " When an axon of cell A [...] persistently takes part in firing cell B, some growth process or metabolic change takes place in one or both cells such that A's efficiency, as one of the cells firing B, is increased " Major points: –long-lasting changes (trace, memory) –associativity (A & B must both fire) –locality (depends only on A & B) Major issue: can only increase

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Evidence for long-term potentiation (LTP) Came out in the early 1970's (Bliss & Lomo J Physiol 1973) High Frequency stimulations (HFS) in the hippocampus maintains for hours-days A B From Fino et al. J Neurosci 2005 ≈200 pulses ≈100 Hz stimulation

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Evidence for long-term depression (LTD) Came out in the early 1990's (Dudek & Bear PNAS 1992) Low Frequency stimulations (LFS) in the hippocampus From Fino et al. J Neurosci 2005 maintains for hours-days A B ≈1000 pulses ≈1 Hz stimulation

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Transition to spikes and spike-timings Initiated in late 1990's ( Markram et al. Science 1997; Bi & Poo J Neurosci 1998) LTP or LTD depending on the timing between pairs of post- and pre- synaptic spikes A (pre) B (post) pre post pre post Δt = t post -t pre < 0Δt = t post -t pre > 0 Bi & Poo J Neurosci 1998 maintains for hours ≈100 paired ≈1 Hz ΔtΔt pre post 1 s stimulation

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Molecular basis glutamate synapses) from Citri & Malenka Neuropsychopharmacology 2008 V soma (mV) time V rest V thr tAtA tAtA tAtA tBtB  LTP  LTD

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Is that all our brain does? Most computational models with plastic / learning synapses use rules derived from Hebbian LTP/LTD or STDP. Yet there is more than this going on in our brains: – Anti-Hebbian plasticity

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Is that all our brain does? Most computational models with plastic / learning synapses use rules derived from Hebbian LTP/LTD or STDP. Yet there is more than this going on in our brains: – Anti-Hebbian plasticity – Relaxed locality Neuromodulation (e.g. dopamine, serotonin) Glia-neuron interactions

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Is that all our brain does? Most computational models with plastic / learning synapses use rules derived from Hebbian LTP/LTD or STDP. Yet there is more than this going on in our brains: – Anti-Hebbian plasticity – Relaxed locality Neuromodulation (e.g. dopamine, serotonin) Glia-neuron interactions – Nonassociative rules : synaptic scaling short-term facilitation

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Is that all our brain does? Most computational models with plastic / learning synapses use rules derived from Hebbian LTP/LTD or STDP. Yet there is more than this going on in our brains: – Anti-Hebbian plasticity – Relaxed locality Neuromodulation (e.g. dopamine, serotonin) Glia-neuron interactions – Nonassociative rules : synaptic scaling short-term facilitation – Intrinsic plasticity

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Outline STDP: evolution with the number of paired stimulations Intrinsic long-term plasticity of the threshold or gain –mechanical origins –competition with synaptic Hebbian learning in chaotic recurrent networks Short term plasticity –modulation by glial cells: the tripartite synapse

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT STDP : EVOLUTION WITH THE NUMBER OF PAIRED STIMULATIONS WITH B. DELORD AND L. VENANCE LABS

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT STDP in the basal ganglia BG = motor control, procedural memory, goal oriented tasks. The striatum is the major input nucleus for cortical input from Fino & Venance Front Synaptic Neurosci 2010 cortico-striatal synapses

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT STDP in the basal 100 stims STDP in at cortico-striatal synapses is Anti-Hebbian synaptic weight Fino et al. J Neurosci 2005 ≈100 paired pulses, ≈1 Hz ΔtΔt pre post 1 s Δt = t post -t pre < 0Δt = t post -t pre > 0 pre post pre post stimulation Bi & Poo J Neurosci 1998

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Decreasing the paired stimulation count Cui et al., submitted

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT LTP re-emerges at low stim. counts: model Cui et al., submitted signaling by AMPAR and NMDAR Retrograde signaling by endocannabinoids

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Low-stims LTP is endocannabinoid (eCB) Cui et al., submitted Full Model eCB-KO Model Experimental confirmation

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT eCB-Low stims LTP also exists in the cortex Cui et al., submitted

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT A novel form of LTP "Classical" NMDAR-mediated LTP and LTD disappear with fewer stimulations But an endocannabinoid-mediated LTP emerges for 5-20 pairings First evidence for endocannabinoid-mediated potentiation Ongoing work: –variations of the frequency and stochasticity of Δt –effects in a network (# stims, time scales) A possible cellular support for rapid learning: –Associative memories and behavioral rules can be learned within a few or even a single trial (Pasupathy et al. Nature 2005; Tse et al. Science 2007). –This would represent as few as 2-50 spikes – eCB-low stims LTP may code such rapid learnings

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT INTRINSIC PLASTICITY OF THE THRESHOLD OR GAIN WITH B. DELORD LAB

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Plasticity of the f-I curve in the cortex firing frequency f Injected current I "Transfert function" Paz et al. J Physiol 2009

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Plasticity of the f-I curve in the cortex firing frequency f Injected current I "Transfert function" Paz et al. J Physiol 2009 stimulation

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Plasticity of the f-I curve in the cortex No change in synaptic weights : INTRINSIC PLASTICITY Paz et al. J Physiol 2009

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Modulations of voltage-gated channels What happense if other voltage-dependent ionic channels (ie not synaptic AMPAR and NMDAR) are also modulated by neural activity? Synaptic plasticity I NMDA I AMPA

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT generic voltage-gated ion channel X A generic Hodgkin-Huxley model leakspikeinjected current (soma) 0 injected current I inj Firing frequency of the neuron f Naudé et al., submitted g X = total quantity of X: ?? I X properties 1/

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Increasing g X changes the f-I Naudé et al., submitted IP expected with modulation of stiff- activating channels Channels activating before spike: plasticity of the f-I threshold θ Channels activating at or after spike: plasticity of the f-I inverse slope ε

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Homeostatic IP of the threshold A single neuron with Homeostatic IP of the threshold firing frequency f i Input of neuron i Input from the reservoir 0 firing frequency Naudé et al., submitted Desaturation of neuron i

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT H-IP of the threshold in mixed networks Recurrent neural networks (firing rate) with initial chaotic dynamics + Hebbian synaptic learning (Siri et al. J Physiol 2007; Neural Computation 2008) +/- H-IP of the threshold ? Network-averaged firing Network 10 3 learning steps SP (no IP) SP+IP Largest Lyapunov exponent Learning steps Naudé et al., submitted chaotic non chaotic

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT H-IP of the threshold in mixed networks Functional property: sensitivity to the input : Δ[x] = [network dynamics w/ input - network dynamics w/o input] Naudé et al., submitted SP (no IP) SP+IP Learning steps Δ[x] Network-averaged firing

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT H-IP of the threshold in mixed networks Functional property: sensitivity to the input : Δ[x] = [network dynamics w/ input - network dynamics w/o input] Naudé et al., submitted SP (no IP) SP+IP Learning steps Δ[x] SP+IP Network-averaged firing

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Conclusions Intrinsic plasticity – is to be expected when the number of stiff-activating voltage- dependent ionic channels is regulated by the neuron activity –nature of changes (θ or ε) depends on if the channel opens before or during the spike. – H-IP may prevents some of the runaway effects of synaptic plasticity (neuronal saturation, simplification of dynamics) Ongoing work –Non-homeostatic IP ? + - -

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT MODULATION OF SHORT-TERM PLASTICITY BY GLIAL CELLS WITH E. BEN JACOB LAB

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Short-term plasticity (STP) Recycling glutamate takes time: τ rec ≈ sec Restoring basal Ca levels after a presyn spike takes time: τ f ≈ 1-2 sec Maintains short-term only (0.1 – few sec) STP crucially conditions information transfert from pre to post (Tsodyks & Markram PNAS 1997; Abbott et al Science 1997) Ca 2+ + glutamate presyn spike frequency synaptic weight (% control) facilitation depression 100 Paired-pulse plasticity

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Glial cells modulate STP Glial cells (astrocytes) ≈ 50 % of human brain Not only mechanical/feeding support to neurons: –communicate with each other over long distances through slow calcium waves ( Goldberg et al. PLoS Comp Biol 2010; De Pittà et al. J Biol Phys 2009 ) – associate with synapses into tripartite synapses –bidirectional comunication between presynaptic, postynaptic and astrocytes Modulation of STP by glial cells has been evidenced but confusing: facilitation, depression, or both (Robitaille Neuron 1998; Jourdain et al., Nature Neurosci 2007) ? Haydon Nature Rev Neurosci 2001

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT A model of glia-modulated STP De Pitta et al. PLoS Comp Biol 2011

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Mean-field theory of the system De Pitta et al. PLoS Comp Biol 2011 syn. weight potentiation depression

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT syn. weight Confirmation by simulations De Pitta et al. PLoS Comp Biol 2011 = =

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Conclusion Astrocytes dynamically switch synapses between depressing and potentiating modes Astrocytes can decrease global synaptic weight while increasing paired-pulse potentiation (and vice-versa) (cf Jourdain et al., Nature Neurosci 2007) The plasticity characteristics of a synapse may not be fixed but could be modulated by associated astrocytes. Ongoing work: –effect on the postsynaptic terminal –modulation of long-term plasticity (STDP) Perspectives: assembly of a mixed glia/neuron network neuron glia ≈10 m/s ≈10 µm/s

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT SUMMARY

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Unconventional plasticity The "Unconventional computation" CS community: –"enrich or go beyond the standard models, such as the von-Neumann computer architecture and the Turing machine, which have dominated computer science for more than half a century" Unconventional plasticity: even though the synaptic Hebbian framework prevails, outside-the-box plasticity forms do exist and it may be worth looking at them for learning & memory. Expands the (limited?) framework of Hebbian synaptic plasticity: –time-scale-dependent plasticity ("quick learning" LTP) –homeostatic maintenance of dynamic regimes (intrinsic plasticity) –dynamic modulation of the synaptic operating regime (depressing/potentiating)

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Thanks!!! Funding : INSERM U1050 Collège de France, Paris Y. Cui E. Fino L. Venance ISIR, Univ P&M Curie, Paris France J. Naudé B. Delord S. Genet Contributions: for computer ressources Sch. Physics & Astronomy, Tel Aviv Univ, Israel M. Goldberg M. De Pittà E. Ben Jacob Project-Team Beagle, INRIA, Lyon, France J. Lallouette J.M. Gomès H. Berry Salk Institute, San Diego, USA V. Volman

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT More information 2011 Meeting of the Society for Neuroscience (Washington, Nov ) –"Sub-second induction unveils a switch from NMDA- to endocannabinoid-LTP", abstract # –"Astrocyte regulation of presynaptic plasticity", abstract # Published paper: –De Pittà, Volman, Berry & Ben Jacob (2011) A tale of two stories: astrocyte regulation of synaptic depression and facilitation, PLoS Comput Biol (in press). Questions?

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Is that all our brain does? Most computational models with plastic / learning synapses use rules derived from Hebbian LTP/LTD or STDP. Yet there is more than this going on in our brains: – Anti-Hebbian plasticity: – Relaxed locality Neuromodulation (e.g. dopamine, serotonin) Glia-neuron interactions – Nonassociative rules : synaptic scaling short-term facilitation A1 B A2 A1+A2=3 A1/A2=1/3 LTP A1+A2=6 A1/A2=2 Scaling A1+A2=3 A1/A2=2

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Short-term plasticity (STP) Recycling glutamate takes time: τ rec ≈ sec Restoring basal Ca levels after a presyn spike takes time: τ f ≈ 1-2 sec Maintains short-term only (0.1 – few sec) Ca 2+ + glutamate time presyn spikes presyn Ca postsyn potential τfτf

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Short-term plasticity (STP) Recycling glutamate takes time: τ rec ≈ sec Restoring basal Ca levels after a presyn spike takes time: τ f ≈ 1-2 sec Maintains short-term only (0.1 – few sec) Ca 2+ + glutamate time presyn spikes presyn Ca postsyn potential τfτf potentiation (facilitation)

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Short-term plasticity (STP) Recycling glutamate takes time: τ rec ≈ sec Restoring basal Ca levels after a presyn spike takes time: τ f ≈ 1-2 sec Maintains short-term only (0.1 – few sec) Ca 2+ + glutamate time presyn spikes presyn Ca postsyn potential τ rec depression

H. BERRY - Unconventional forms of plasticityDevLeaNN - Paris – OCT Short-term plasticity (STP) STP crucially conditions information transfert from pre to post (Tsodyks & Markram PNAS 1997; Abbott et al Science 1997) Ca 2+ + glutamate presyn spike frequency synaptic weight (% control) facilitation depression 100