Neuronal Cell Death and Synaptic Pruning driven by Spike-Timing - PowerPoint PPT Presentation

Neuronal Cell Death and Synaptic Pruning driven by Spike-Timing Dependent Plasticity Javier Iglesias 1 , 2 and Alessandro E.P . Villa 2 ICANN 2006 Athens, 2006-09-11 1 Information Systems Department, University of Lausanne, Switzerland 2 Laboratory of Neurobiophysics, University Joseph-Fourier, France < javier.iglesias@unil.ch >

introduction: synaptogenesis and synaptic pruning 1 12 4 neurons / mm x10 10 3 8 6 4 20 2 8 synapses / mm x10 15 NB 0.5 1 2 5 10 adult aged 3 years (74-90) 10 5 NB 0.5 1 5 10 15 20 40 60 80 100 years modified from Huttenlocher, Synaptic density in human frontal cortex – developmental changes and effects of aging , Brain Research, 163:195–205, 1979

introduction: existing work 2 • The memory performance of a network is optimally maximized if, under limited metabolic energy resources restricting their number and strength, synapses are first overgrown and then pruned. Chechik et al. , Synaptic pruning in development: A computational account , Neural Computation, 10(7):1759–77, 1998 • Neuronal regulation might maintain the memory performance of networks undergoing synaptic degradation. Horn et al. , Memory maintenance via neuronal regulation Neural Computation, 10(1):1–18, 1998 • STDP has been shown to maintain the postsynaptic input field. Abbott et al. , Synaptic plasticity: taming the beast Nature Neuroscience, 3:1178–83, 2000

model: network 3 a b c d probability +50 +50 500 0.6 e i 0.4 cell count 0.2 0 0 250 y y 0.0 e e 50 -50 0 y -50 0 -50 0 -50 0 +50 -50 0 +50 0 200 400 600 x -50 50 connection count x x e f g h probability +50 +50 500 0.6 0.4 cell count 0.2 0 250 y 0 y 0.0 i i 50 i e -50 0 -50 0 y -50 0 -50 0 +50 -50 0 +50 0 200 400 600 x -50 50 connection count x x Iglesias et al. , Dynamics of Pruning in Simulated Large-Scale Spiking Neural Networks , Biosystems, 79(1-3):11-20, 2005

introduction: leaky integrate and fire neuromimetic model 4 Type I = excitatory 80% ~190 excitations Type II = inhibitory 20% B(t) = -76 [mV] V rest = -40 [mV] θ i S(t) w(t) V(t) = 15 [ms] τ mem = exc:3 inh:2 [ms] t refract = 5 [spikes/s] λ i n = 50 ~115 inhibitions � V i ( t +1) = V rest[ q ] +(1 − S i ( t )) · (( V i ( t ) − V rest[ q ] ) · k mem[ q ] )+ w ji ( t )+ B i ( t ) j S i ( t ) = H ( V i ( t ) − θ q i ) w ji ( t + 1) = S j ( t ) · A ji ( t ) · P [ q j ,q i ] B i ( t + 1) = P reject ( λ q i ) · n · P [ q 1 ,q i ]

model: STDP and pruning 5 time S (t) i post L ji ( t + 1) = L ji ( t ) · k act[ q j ,q i ] +( S i ( t ) · M j ( t )) − ( S j ( t ) · M i ( t )) pre S (t) time j

model: STDP and pruning 5 time time S (t) S (t) i i post post -M (t) i L ji ( t + 1) = L ji ( t ) · k act[ q j ,q i ] +( S i ( t ) · M j ( t )) M (t) − ( S j ( t ) · M i ( t )) j pre pre S (t) S (t) time time j j

model: STDP and pruning 5 time time time S (t) S (t) S (t) i i i post post post -M (t) -M (t) i i L ji ( t + 1) = L ji ( t ) · k act[ q j ,q i ] +( S i ( t ) · M j ( t )) M (t) M (t) − ( S j ( t ) · M i ( t )) j j pre pre pre S (t) S (t) S (t) time time time j j j L (t) ji

model: STDP and pruning 5 time time time S (t) S (t) S (t) i i i post post post -M (t) -M (t) i i L ji ( t + 1) = L ji ( t ) · k act[ q j ,q i ] +( S i ( t ) · M j ( t )) M (t) M (t) − ( S j ( t ) · M i ( t )) j j pre pre pre S (t) S (t) S (t) time time time j j j L (t) ji L ji (t) 4 A ji (t) 2 1 0 0 50 100 150 time [s] w ji ( t + 1) = S j ( t ) · A ji ( t ) · P [ q j ,q i ]

model: synaptic adaptation examples 6 a b c L 4 L 4 L 4 A 4 L 3 L 3 L 3 A 3 L 2 L 2 L 2 A 2 L 1 L 1 L 1 A 1 L 0 L 0 L 0 time time time

model: graph considerations 7 Strongly Interconnected (SI) units at the end of the simulation set of cells (discarding input units) maintaining k out ≥ 3 and k in ≥ 3 with strongest activation level ( A ji ( t ) = 4 ) with units with the same properties. Neighbourhood all excitatory units (including input units) with at least one projection to or from SI-units.

results: circuit 8 Iglesias et al. , Emergence of Oriented Cell Assemblies Associated with Spike-Timing-Dependent Plasticity , LNCS 3696:127-132, 2005 4695 4761 time = 500 s 5326 5550 1435 8467 992 490 490 490 2800 7794 492 492 1472 1472 1472 2279 931 931 1234 2848 804 7297 7235 8863 2666 4547 2427 2427 1539 8120 8120 2267 1845 8022 1842 4622 8267 8267 8267 9983 9661 3027 2007 7928 2532 7199 1048 1608 5697 1202 8889 1049 1152 1151 8402 8300 5724 5724 layer 1 layer 2 layer 3 layer 4 layer 5

results: circuit 8 Iglesias et al. , Emergence of Oriented Cell Assemblies Associated with Spike-Timing-Dependent Plasticity , LNCS 3696:127-132, 2005 4695 4695 4761 4761 time = 500 s time = 0 s 5326 5326 5550 5550 1435 1435 8467 8467 992 992 490 490 490 490 490 490 2800 2800 7794 7794 492 492 492 492 1472 1472 1472 1472 1472 1472 2279 2279 931 931 931 931 1234 1234 2848 2848 804 804 7297 7297 7235 7235 8863 8863 2666 2666 4547 4547 2427 2427 2427 2427 1539 1539 8120 8120 8120 8120 2267 2267 1845 1845 8022 8022 1842 1842 4622 4622 8267 8267 8267 8267 8267 8267 9983 9983 9661 9661 3027 3027 2007 2007 7928 7928 2532 2532 7199 7199 1048 1048 1608 1608 5697 5697 1202 1202 8889 8889 1049 1049 1152 1152 1151 1151 8402 8402 8300 8300 5724 5724 5724 5724 layer 1 layer 1 layer 2 layer 2 layer 3 layer 3 layer 4 layer 4 layer 5 layer 5

model: stimulus 9 t = 5 t = 6 t = 1 t = 2 t = 3 t = 4 t=duration stimulation (50, 100 onset or 200 ms) A [...] B [...] every 2 seconds 10 groups of 40 units activated in sequence (10% input units) during 100 time steps (5 × A + 5 × B or 5 × B + 5 × A) Animated sequences are available for both stimuli A and B.

model: stimulus (cont.) 10 1 ordered sequence of set A 1 ordered sequence of set B A 1 A 10 A 1 A 10 B 1 B 10 B 1 B 10 A 1 1 10 11 50 51 60 61 100 2001 time steps [ms] stimulus stimulus stimulus onset offset onset

model: cell death mechanisms 11 • apoptosis induced by excessive firing rate 50 ms running window firing rate threshold: θ exc = 245 sp/s for excitatory units ν θ inh = 250 sp/s for inhibitory units ν 0 . 5 · t 2 − 4 . 5 · 10 − 6 · t 3 death probability function: P apopt ( t ) = 44 · (2 . 5 · 10 6 +6 · 10 − 3 · t 2 ) P apopt ( t = 100) = 4 . 5 · 10 − 5 P apopt ( t = 700) = 2 . 2 · 10 − 3 P apopt ( t = 800) = 2 . 9 · 10 − 3 • apoptosis induced by lack of excitatory afferents loss of all excitatory inputs due to: apoptosis of pre-synaptic unit STDP driving to A ji ( t ) = 0

model: simulation layout 12 • 0 ≤ t < { 700 , 800 } ms (initial phase) apoptosis induced by excessive firing rate • { 700 , 800 } ≤ t < 10 5 ms Spike-Timing Dependent Plasticity ⇒ synaptic pruning ⇒ apoptosis induced by lack of excitatory inputs • t = { 1000 , 3000 , 5000 , . . . } ms 100 ms lasting stimuli 50 presentations (random mix: 50% AB and 50% BA)

results: excitatory vs. inhibitory cell death 13 Lag ≈ 120 ms inh apoptosis phase STDP phase Lag ≈ 190 ms exc 100 95 surviving units [%] 90 85 80 excitatory units inhibitory units 75 0 250 500 750 800 1000 time [ms]

introduction: detection of spatiotemporal patterns of activity 14 a simultaneous recording of spike trains A B C time (ms)

introduction: detection of spatiotemporal patterns of activity 14 a simultaneous recording of spike trains A B C time (ms) b detection of statistically significant spatiotemporal firing patterns <A,C,B; ∆t 1 ,∆t 2 > pa tte rn s found n=3 cell # A cell # B e xp e c te d coun t N=0.02 cell # C si gn i f i canc e of t h is pa tte rn -6 < 0.00 1 time (ms) pr ( 3 , 0.02 ) ≈ 1 .3 · 1 0 Δ t 2 Δ t 1

introduction: detection of spatiotemporal patterns of activity 14 a simultaneous recording of spike trains A B C time (ms) b detection of statistically significant spatiotemporal firing patterns <A,C,B; ∆t 1 ,∆t 2 > pa tte rn s found n=3 cell # A cell # B e xp e c te d coun t N=0.02 cell # C si gn i f i canc e of t h is pa tte rn -6 < 0.00 1 time (ms) pr ( 3 , 0.02 ) ≈ 1 .3 · 1 0 Δ t 2 Δ t 1 for methods, see Villa et al. , 1999; Tetko and Villa, 2001

B B A B C simultaneous฀recording฀of฀spike฀trains time฀(ms) A time฀(ms) C rasters฀of฀spikes฀aligned฀on฀pattern฀start A C introduction: representation of spatiotemporal patterns of activity 15 a c

lag฀[ms] time฀[s] lag฀[ms] results: spatio-temporal pattern of activity 16 stopping firing rate-induced apoptosis at t=700ms < 79, 79, 79; 453 ± 3.5, 542 ± 2.5 > a -400 +1600 b +453 +542 -400 +1600 c 0 25 50 75 100

lag฀[ms] time฀[s] lag฀[ms] results: spatio-temporal pattern of activity 17 stopping firing rate-induced apoptosis at t=800ms < 13, 13, 13; 234 ± 3.5, 466 ± 4.5 > a -400 +1600 b +234 +466 -400 +1600 c 0 25 50 75 100