When a discrete stimulus (conditional stimulus [CS]) is paired with a biologically significant stimulus (unconditional stimulus [US]), an organism will learn an association between the two, as exhibited by subsequent, conditional responding (CR) to the CS alone (Pavlov, 1927
). For example, if a light presentation is followed by a brief footshock, an animal will subsequently display fear-related behaviors in response to that light. However, if an additional cue, such as a noise stimulus, is presented in conjunction with the light, the degree to which that light can be fear conditioned is reduced (Mackintosh, 1971
). In other words, the noise overshadows
the light during conditioning. Like Kamin’s blocking effect (Kamin, 1968
) – wherein conditioning of a stimulus is blocked by being presented in compound with a previously conditioned stimulus, overshadowing seems to rest on the degree to which the US is surprising.
This idea of US-surprisingness was elegantly captured by Rescorla and Wagner (1972)
, whom formulated a mathematical model to describe the course of conditioning. According to their model, the less predictable (i.e. more surprising) a US is, the more power it has as a reinforcer. The Rescorla-Wagner model states:
Where ΔV refers to the change in associative strength between a particular CS and US on a given trial. The term α is a learning rate parameter determined by the salience of the CS and accounts for the more rapid learning that accrues to attention-grabbing CS’s. The term λ refers to US intensity and VΣ refers to the associative strength of all stimuli present on that trial. Put simply, learning about a CS on a given trial is determined by how surprising the US is on that trial. Thus, surprise is essentially an error term – the difference between the predicted and the actual value of a reinforcing stimulus on a given conditioning trial (λ − VΣ). It is this element of surprise, or prediction error, which regulates conditioning.
A strength of the Rescorla-Wagner model is that it can be used to perform calculations that describe the course of conditioning on a trial-by-trial basis. In particular, the model is excellent at predicting US-limited phenomena such as blocking and overshadowing, because the surprise term (λ − VΣ) can clearly be lowered for a blocked or overshadowed stimulus because the other cue in the compound contributes to VΣ. In the case of overshadowing, the more salient CS and the US build an association early on in conditioning, rapidly bringing the error term (λ − V) to near zero and thereby limiting further learning at later trials for either CS. Consequently, the Rescorla-Wagner model makes the prediction that if there weren’t US limitations on conditioning (VΣ), then eventually all CS’s could fully condition to asymptote.
Current knowledge of the neural mechanisms responsible for Rescorla-Wagner-like calculations makes empirical investigations of its predictions quite feasible (Fanselow, 1986
; Kim, Krupa, & Thompson, 1998
). For example, pairing an initially neutral CS with an aversive foot-shock-US conditions an endogenous opioid-mediated analgesic response (Fanselow & Baackes, 1982
; Fanselow & Bolles, 1979a
). Thereafter, this conditional analgesia modulates the painful impact of the shock-US, thereby diminishing the effectiveness of the US as a reinforcer and providing negative feedback on the acquisition of fear. In essence, this “negative feedback model” () provides a neural mechanism by which calculations of the Rescorla-Wagner kind are automatically performed and US-limited phenomena are predicted (Fanselow, 1981
). In a similar vein, Schull (1979)
described how the acquisition of a conditional opioid response could provide an “opponent process” that antagonized a US’s ability to condition.
Figure 1 Schematic diagram of the negative feedback model of Pavlovian fear conditioning. The model describes how Rescorla-Wagner calculations may be made (the symbols adjacent to the arms refer to the Rescorla-Wagner model). The diagram displays how input from (more ...)
The discovery of negative feedback-type mechanisms in other Pavlovian preparations has helped to support their ubiquity. For example, negative-feedback type mechanisms have been established in eyeblink conditioning via GABAergic inhibitory feedback onto olivary neurons conveying airpuff-US information to the cerebellum (Kim, et al., 1998
). Further support for this view comes from findings that specific groups of neurons fire in proportion to predicted magnitudes of the error signal (Kim, et al., 1998
; see Schultz, 2006
for review). Systematic studies from Schultz and colleagues found that in appetitive conditioning, midbrain dopamine neurons fire in a manner generally consistent with the formation of a prediction error for signaling reward (e.g. Fiorillo, Newsome, & Schultz, 2008
; Fiorillo, Tobler, & Schultz, 2003
; Hollerman & Schultz, 1998
; Tobler, Dickinson, & Schultz, 2003
; Tobler, Fiorillo, & Schultz, 2005
One benefit of such models is that they are readily testable. For instance, in fear learning, conditional analgesia is mediated by endogenous opioids, which can be blocked by opioid antagonists. Indeed, a number of fear conditioning studies have done just that. The administration of an opioid antagonist has been shown to eliminate blocking (Fanselow & Bolles, 1979a
; McNally, Pigg, & Weidemann, 2004
; Schull, 1979
), attenuate unblocking (McNally, et al., 2004
), lift the limits on conditional asymptotes (Young & Fanselow, 1992
), eliminate Hall-Pearce negative transfer (Young & Fanselow, 1992
), and prevent over-expectation of fear (McNally, et al., 2004
While a number of studies have offered strong support for the role of endogenous opioids in regulating US-limited Pavlovian phenomena such as blocking, none have determined whether or not they play a role in Pavlovian overshadowing. Given that overshadowing, like blocking, can be thought of as due to US-limitations, it should similarly be abolished by administration of an opioid antagonist. However, because overshadowing, unlike blocking, may allow for direct sensory competition between novel CS’s, it may be regulated by distinguishable mechanism(s). Thus, in the experiments presented below, we investigated whether or not overshadowing, like blocking, could be explained by an endogenous opioid-mediated negative feedback circuit. In Experiment 1, we tested the effects of the opioid antagonist, naltrexone, on eight-trial overshadowing. Experiments 2 and 3 investigated whether one-trial or two-trial overshadowing could provide a possible, “pre” conditional analgesia, contribution to overshadowing. Lastly, Experiment 3 investigated whether dopamine-mediated, like opioid-mediated processes, contribute to overshadowing.