The Salk Institute MNL/S
Synaptic transmission in the hippocampus is unreliable: the probability of neurotransmitter release following synaptic stimulation is usually between 0.05 and 0.50. What are the implications of this unreliability for neural encoding? Using whole cell recording in rat hippocampal slices, we measured EPSPs, EPSCs, and spikes in CA1 cells; constant amplitude synaptic stimulation was presented via the Schaffer collaterals. We presented temporally irregular sequences of shocks that mimicked trains of action potentials recorded from the hippocampus of a freely moving rat. The irregular stimuli consisted of bursts with short interstimulus intervals (ISIs) between 60-250 msec separated by longer ISIs of up to 10 sec.
These stimuli triggered sequences of spikes in which the spike latency --the time between the stimulus and the spike--was highly variable within each trial. Nevertheless, when the same stimulus was applied in multiple trials, the pattern of spike latencies across trials was rather repeatable. Furthermore, the latency was not a simple function of the preceding ISIs. Thus while the transformation from input ISIs to spike latency is rather complex, it is also surprisingly deterministic.
What is the mechanism underlying the high variability of spike latency within trials? The variability is not due to variability in EPSC or EPSP timing: the time to peak for both EPSCs and EPSPs is quite consistent, independent of ISI. By contrast, synaptic strength (as assessed by either EPSC or EPSP amplitude) shows high variability within trials, comparable to that of spike latency. This variability of synaptic strength is presumably due to rapid modulation of release probability through the interaction of short term plasticity mechanisms. Further, spike latency is strongly correlated with synaptic strength. Thus the within-trial variability of spike latency appears largely due to the temporal modulation in the synaptic strength.
The repeatability of spike latencies across trials indicates that unreliable synapses can encode reliably. This suggests that within-trial variability should be viewed not as an unfortunate consequence of stochastic release, but rather as an indication that synapses operate over a large dynamic range on short time scales.