Information Transfer at Dynamic Synapses: Effects of Short-Term Plasticity Patrick Scott 1 Anna Cowan 1 Andrew Walker 1 Christian Stricker 1,2 1 Division.

Information Transfer at Dynamic Synapses: Effects of

Short-Term Plasticity Patrick Scott1

Anna Cowan1

Andrew Walker1

Christian Stricker1,2

1 Division of Neuroscience, John Curtin School of Medical Research, ANU, Canberra, ACT.

2 ANU Medical School

Background

• Probability of neurotransmitter release changes according to previous activity

• Four major short-term effects:– Release-dependent depression (depletion; RDD)– Release-independent depression (RID)– Facilitation– Frequency-dependent recovery (FDR)

• How do they affect information transfer? Nobody knows… (yet)

ModellingExtended mathematical model of short-term plasticity

• Phenomenological response to AP

• success/failure to release

• changes in probability of subsequent release

• no channels, Ca2+, etc.

• Previously

• RDD+F, deterministic (Fuhrmann et al 2002, J Neurophysiol 87:140)

• RDD+RID+FDR, quasi-stochastic (Fuhrmann et al 2004, J Physiol 557:415)

• Now

• RDD+RID+FDR+F, fully stochastic

• 4 coupled 1st-order ordinary differential equations, with an explicit (iterative) solution

Parameter Estimation

• Fitted models to EPSCs from paired recordings in Layers IV/V of rat somatosensory cortex (N = 11)

• Simultaneous fits to different stimuli, EPSCs/variances

• Defined typical ‘facilitating’ and ‘depressing’ connection parameters

• Reduced 2 values all < 1 (i.e. good fits)

Information Measurement

• Generated 5.4 hours of synthetic data for each parameter combination, as postsynaptic APs with an integrate-and-fire model

• Measured information transfer using information theory; entropy (Strong et al 1998, Phys Rev Lett 80:197)

• Includes extrapolations to infinite data size and window length

• For single vesicle and network configurations => spike timing and rate-coding dominated.

Results – RDD & RID

rec = recovery timescale from RDD U1R = strength of RID

RDD, spike timing RID, spike timing

RDD, rate coding RID, rate coding

Results – Facilitation & FDR

U1F = strength of facilitation 1 = strength of FDR

Facilitation, spike timing FDR, spike timing

Facilitation, rate coding FDR, rate coding

Results - ExampleRDD-dominated, no FDR RID-dominated, with FDR

U0 0.4

U1R 0

U1F 0

0 1 s

1 0.2

fac 0 s

FDR2 s

rec 500 ms

U0 0.4

U1R 0.2

U1F 0

0 1 s

1 0.2

fac 0 s

FDR2 s

rec 5 ms

Spike Timing:11.49 bits/s

Rate Coding: 1.84 bits/s

Spike Timing:27.31 bits/s

Rate Coding: 1.86 bits/s

Outcomes

• Information transfer by spike timing goes with release probability, not so for rate-coded information.–RDD: spike timing ↓, rate unaffected

–RID: spike timing ↓, rate ↓–Facilitation: spike timing ↑, rate ↓–FDR: spike timing ↑, rate ↑ or ↓ with other

parameters

Speculation

• Shows how brain can use alternative coding schemes and different dynamic processes to achieve varying goals at different network levels.

• Possible applications to neural prosthetics, neural electronics and artificial neural networks.