Perirhinal-amygdala circuit-level computational model of temporal encoding in fear conditioning

Abstract

Here we present a real-time model of fear conditioning in which the functional anatomy and neurophysiology of the lateral amygdala and perirhinal cortex provide a mechanism for temporal learning during Pavlovian conditioning. The model uses realistic neuronal and circuit dynamics to map time onto space and relies on a conventional Hebbian learning rule that requires strict temporal contiguity for synaptic modification. The input-output relationships of the model neurons simulate our physiological recordings with respect to latency to fIre, firing frequency, and accommodation tendency. Chains of these neurons form a spectrum of activity windows delayed by various amounts from the conditioned stimulus onset. Simulations reveal that learning occurs only when the conditioned and unconditioned stimuli are explicitly paired, that the interstimulus interval (lSn is accurately learned over a time range from 0.5 to 16 sec, and that low-frequency noise causes the accuracy of temporal learning to decrease as the lSI increases, in accordance with a Weber-type law.

Temporal Encoding in Fear Conditioning Revealed Through Associative Reflex Facilitation.

“…Simulations of perirhinal-amygdala circuits (Faulkner et al, 1997;Tieu et al, 1999) show that relatively small networks containing Hebb-type synapses (T. H. Brown, Furtak, & Lindquist, in press;T. H. Brown, Ganong, Kairiss, & Keenan, 1990) and late-spiking neurons can easily encode the ISIs explored in the present experiments.…”

Neurophysiological theory of Kamin blocking in fear conditioning.

“…Our guesses at the minimum number of neurons are probably high, depending on the criteria, but we have learned that using too few neurons can lead to unpredicted time-domain problems that depend on the exact noise levels, a complexity that we wanted to avoid. Whereas approximately 4,000 neurons were used to represent US-anticipating blocking, which requires ISI encoding (as in Tieu et al, 1999), only about 500 neurons were used for CS ϩ -onset-triggered blocking and continuous blocking, which do not require ISI encoding.…”

“…The YNET neurons were connected by 2,010 -83,388 synapses (again, depending on the type of blocking being analyzed), each of which functioned probabilistically in that repetitively evoked synaptic responses fluctuated randomly (via added Gaussian noise) around a mean value (Tieu et al, 1999). Because of random synaptic amplitude fluctuations, successive CS presentations produced different postsynaptic response amplitudes.…”

“…Postsynaptic firing thresholds for synaptic potentiation ( P ) and depression ( D ) determine whether synapses undergo long-term potentiation (LTP), long-term depression (LTD), or no change (see Figure 2), as described elsewhere (Brown, Byrne, et al, 2004;Tieu et al, 1999). The standard BCM learning rule was modified (Tieu et al, 1999) so that different kinetics controlled 1 All three basic cell types were represented as previously described Tieu et al, 1999), except for the ES cells. The latter were modified "fast-spiking" (FS) neurons (Tieu et al, 1999) in which the charging time constant t m was increased to 18.2 ms and the maximum firing frequency was reduced to 40 Hz (McCreless, McGann, & Brown, 2000) to match reported properties in ACe (Schiess et al, 1993(Schiess et al, , 1999.…”

Time course of cardiac conditioned responses in restrained rats as a function of the trace CS–US interval.

“…For instance, a small number of trace components are assumed by the model of Buhusi & Schmajuk (B&S;. Some models do not account for trace conditioning in their present form (e.g., Tieu et al, 1999). Other models account for trace conditioning by assuming a decaying eligibility trace following CS offset (Sutton & Barto, 1990).…”