In four experiments, rats that received lesions of BLA after either Pavlovian or instrumental training with two reinforcers, but before devaluation of one of those reinforcers by either selective satiation or taste aversion training, failed to selectively reduce responding associated with the devalued reinforcer. By contrast, sham-lesioned rats showed highly selective reinforcer-specific reductions in learned responding in all cases. At the same time, the devaluation procedures were equally effective at reducing consumption of the reinforcers themselves in BLA- and sham-lesioned rats.
Previous studies that examined the effects of post-training disruptions in BLA function on devaluation performance produced contrasting results. Pickens et al. (2003)
found that rats trained with single-reinforcer Pavlovian procedures showed normal devaluation effects in testing if BLA function was disrupted by neurotoxic lesions made after that training, but before reinforcer devaluation by taste aversion training. Similarly, we recently replicated Pickens et al.’s (2003)
results, using procedures identical to those of the present Experiment 3, except that a single reinforcer was delivered after each cue in training, and fewer trials were required to establish non-discriminative taste aversion. Rats with post-training lesions of BLA showed normal devaluation effects (unpublished findings). By contrast, Ostlund and Balleine (2008)
found that rats trained with multiple-reinforcer instrumental procedures failed to show devaluation effects in testing if BLA lesions were made after training but before reinforcer devaluation by selective satiation. Similar to Ostlund and Balleine’s (2008)
findings, Wellman et al. (2005)
found that monkeys trained with a multiple-reinforcer instrumental object discrimination procedure failed to show devaluation effects in testing if BLA function was depressed by muscimol prior to the selective satiation treatment used to devalue one of the reinforcers.
The present results indicate that those earlier conflicting results are not attributable to differences in species (rats or monkeys), method of disrupting BLA function (lesions or transient inactivation), training contingency (Pavlovian or operant), or devaluation procedure (associative or motivational). Thus, some previously-suggested accounts for these discrepancies can be rejected. For example, Ostlund and Balleine (2008)
suggested that BLA might only be required for the expression of stimulus-outcome learning when these associations are needed to guide instrumental action selection. However, Experiments 3 and 4 showed deficits in devaluation performance with post-training BLA lesions made after Pavlovian training procedures. Similarly, Wellman et al. (2005)
suggested that the BLA serves an “amplification function” necessary to register and encode changes in sensory-specific incentive value in other brain regions, such as orbitofrontal cortex (OFC) during selective satiation, whereas when the reinforcer value is made negative, as in taste aversion training, this amplification function is bypassed. However, Experiments 2 and 3 showed deficits in devaluation performance with post-training BLA lesions made prior to taste aversion training procedures.
Ostlund and Balleine (2008)
suggested that successful devaluation performance after multiple-but not single-reinforcer training might depend on the ability to generate outcome representations detailed enough to be discriminated from those of other available outcomes at the time of action. Such an ability may require intact BLA function, whereas the ability to retrieve or modify less-detailed representations may not (Balleine & Killcross, 2006
; Blundell et al., 2001
). For example, some theorists have conceptualized outcome representations as involving multiple parallel associations that can separately encode motivational and sensory properties of the outcome (e.g., Wagner & Brandon, 1989
; Konorski, 1967
). Within this account, variations in training conditions may differentially encourage coding of these outcome properties. For example, the use of multiple reinforcers might especially encourage the formation of associations between cues or responses and detailed sensory properties of reinforcers. Such sensory representations may be maintained and processed further in the BLA, whereas less-detailed motivational representations, once established, may be processed elsewhere, for example in the OFC. Notably, lesions of OFC made after the completion of even single-outcome Pavlovian training prevent the expression of normal Pavlovian devaluation performance (Pickens et al., 2003
The pattern of test responding of BLA-lesioned rats deserves further comment. The devaluation impairment was revealed as either lower responding to the non-devalued CS (Experiments 3 & 4) or a general reduction in responding on both levers (Experiments 1& 2), as if the post-lesion reinforcer devaluation procedure successfully altered the motivational value of the reinforcer representation, but left the rats unable able to correctly distinguish between the devalued and non-devalued representations. Notably, Ostlund and Balleine (2008)
observed the same pattern of general reductions in instrumental test responding in their study. Thus, consistent with Pickens et al.’s (2003)
suggestion, BLA function may be unnecessary for updating previously-established representations of reinforcer value after food-illness or satiation procedures, but as suggested by Ostlund and Balleine (2008)
, it may be critical for the maintenance of more detailed sensory-specific reinforcer representations that would permit integrating new information about reinforcer value selectively into existing associative structures. Even from this perspective, the significantly greater responding to the devalued cues than to the non-devalued cues in the lesioned rats in Experiment 3 remains puzzling. However, it is notable that in a Pavlovian devaluation experiment, Kerfoot et al. (2007)
found that presentation of a cue for a devalued reinforcer produced greater FOS expression in BLA, OFC, gustatory cortex, and portions of the accumbens shell than presentation of a nondevalued cue. Perhaps in the absence of moderating influences from BLA (e.g., Arana et al., 2003
; Baxter & Murray, 2002
), these greater neural responses in other brain regions may be reflected in more vigorous conditioned responding.
Additional study is needed to further refine our understanding of the post-training role of BLA in multiple-reinforcer devaluation experiments. For example, is BLA required simply to maintain sensory-specific reinforcer representations, or is its function more proscribed, for example, to integrate new information about reinforcer value into those representations, or to use that information in guiding behavior (or both)? Notably, Wellman et al. (2005)
found that although inactivation of BLA throughout both selective satiation procedures and response testing eliminated accurate devaluation performance, BLA inactivation only at the time of response testing left performance intact. Thus, Wellman et al. (2005)
concluded that BLA was necessary for registering the changed reinforcer value but not for expressing that devaluation in choice performance. However, Wellman et al.’s (2005)
training and testing procedures were considerably different from those used here, including presentation of both reinforcers during response testing. It remains to be seen whether similar outcomes would be observed in rats after training and testing procedures more like those used in the present studies.
Regardless of the precise nature of the BLA lesion deficit in devaluation noted after multiple-reinforcer training, it is notable that data from other experimental paradigms also support the assertion that the roles of BLA differ depending on whether task performance involves detailed sensory representations. For example, considerable data, beyond the results of single-reinforcer devaluation studies already discussed, also indicate that BLA function is critical to the acquisition of associations with more generic motivational information about the reinforcer, but not to the maintenance or subsequent use of that information in guiding behavior. For example, whereas intact BLA function is needed for a first-order CS paired with food to acquire the ability to serve as a reinforcer for subsequent second-order conditioning of another cue (Hatfield et al., 1996
; Setlow et al., 2002
), once the first-order CS has acquired its conditioned reinforcement power as a result of CS-food pairings, BLA lesions have no effect on its ability to establish second-order conditioning (Lindgren et al., 2004; Setlow et al., 2002
). It would be of interest to determine if BLA function is required for expression of reinforcer-selective conditioned reinforcement (Burke et al., 2008
). Similarly, although BLA function is not critical to the acquisition or display of single-outcome Pavlovian-instrumental transfer (Corbit & Balleine, 2005
; Hall et al., 2001
; Holland & Gallagher, 2003
), it is needed for transfer when multiple outcomes are involved (Blundell et al., 2001
; Corbit & Balleine, 2005
Finally, it is notable that even in intact animals, the extent to which learned responding is ultimately governed by sensory-specific reinforcer representations may vary as a function of the use of single or multiple reinforcers (e.g., Adams, 1982
; Colwill & Recorla, 1985
; Holland, 2004
). After extensive single-reinforcer instrumental training, responding often loses its sensitivity to changes in reinforcer value (Adams, 1982
; Dickinson et al., 1998
), whereas performance under multiple-reinforcer conditions does not appear to be susceptible to such a transition to more habitual modes of responding (Colwill & Rescorla, 1985
; Holland, 2004
The use of outcome representations to guide behavior provides the flexibility needed to adapt efficiently to changing environmental conditions. Patients with damage to the amygdala and other, especially prefrontal brain regions, often have difficulty adjusting their behavior according to the consequences of their actions (e.g., Adolphs et al., 1998
; Bechara et al., 1999
; Tranel & Hyman, 1990
; Weller et al., 2007
). A better understanding of the conditions under which various sorts of outcome representations are formed, maintained, and used to guide behavior, and the brain mechanisms underlying those capacities may contribute to the understanding and treatment of such pathologies.