Investigations of the neural basis of pavlovian fear conditioning and its role in anxiety have suggested that the lateral nucleus of the amygdala is the site of convergence of neural pathways that carry information about conditioned stimuli (CSs) and aversive reinforcers (USs)4,6. The emotional expression of this learned association may then be mediated by neural connections from the lateral to the central nucleus of the amygdala4, which, through its projections to hypothalamic and brainstem areas, is thereby able to coordinate the behavioural, endocrine and autonomic responses that form an integrated emotional response8.

Serial processing in these regions of the amygdala is known to be involved in the acquisition and expression of conditioned fear responses to aversive CSs, such as freezing and fear-potentiated startle in animals3,4,5,6. However, the role of these regions in alternative indices of fear conditioning, including the instrumental choice responses involved in avoidance or conflict behaviour, is much less clear9. We therefore wished to investigate whether all forms of fear conditioning are mediated by this serial information flow between the lateral and basal nuclei to the central nucleus of the amygdala. To achieve this we designed a fear-conditioning procedure in rats in which the development of an aversive CS–US association could be assessed simultaneously by examination of two dissociable fear responses in the same animal. The aversive CS–US association created by this procedure would not only produce a pavlovian conditioned fear response, but would also provide the necessary information for animals to solve an operant discrimination that would lead to the avoidance of future presentations of the aversive stimulus.

Rats were trained on a concurrent conditioned-suppression and conditioned-punishment task (see Methods). Pressing one lever in an operant chamber produced an aversive conditioned stimulus (CS+), pressing the other lever produced control presentations of a neutral conditioned stimulus (CS−). In sham-operated control animals, the aversive CS+ caused a disruption of ongoing lever pressing during each of its presentations, relative to performance during the control CS− (Fig. 1, left). This conditioned-suppression effect is a well-established measure of the formation of a pavlovian aversive CS–US association. Behavioural evidence indicates that this is a pavlovian conditioned response deriving from species–species defence responses10. Sham-operated control rats simultaneously came to bias their lever-press responses away from the lever producing the CS+ and towards the lever producing the neutral CS− (Fig. 1, right). This conditioned-punishment effect also demonstrates the presence of an intact aversive CS–US association, as it is only on the basis of this association that rats may choose between the levers and modulate their actions to avoid future aversive events11. The concurrent assessment of conditioned suppression and conditioned punishment provided simultaneous, alternative measures of the establishment of the aversive CS–US association, the former dependent on a conditioned cessation of ongoing behaviour, and the latter on a choice between two available actions.

Figure 1: Left, Difference between suppression ratios during presentation of the aversive CS+ and neutral CS− in sham-operated control rats, and animals with lesions of the lateral/basolateral nuclei of the amygdala (blA), the nucleus (cnA), or with combined lesions of both (cnA + blA)) averaged over 10 sessions.
figure 1

In no group did animals show significant suppression of lever pressing during CS− relative to periods without stimulus presentation. A difference score of 0.5 represents complete suppression during CS+ with no suppression during CS− (high fear conditioning); 0.0 represents no suppression during either CS+ or CS− (low fear conditioning). Rats with cnA or cnA + blA lesions showed significantly reduced levels of fear conditioning relative to sham-operated controls and rats with lesions of just the blA. Analysis of variance with factor ‘lesion’ (sham, blA, cnA and blA + cnA) indicated a significant effect of lesion (F3,36 = 8.24, P < 0.001). Pairwise comparisons showed that performance of cnA rats did not differ from that of rats with combined lesions, but both groups differed (asterisk, P < 0.05) from blA rats and from sham-operated controls. Rats with lesions of the blA did not differ from sham-operated controls. Right, difference in rate of lever pressing on the lever producing the neutral CS− and on the level producing the aversive CS+. Lever pressing was measured when the aversive stimulus was not being presented, and is expressed relative to performance during baseline sessions in which no stimuli were presented. In no group did performance on the CS− lever differ significantly from baseline rates. A score of 1.0 represents a complete bias away from the CS+ lever while maintaining responses on the CS− lever at rates equivalent to baseline (high fear conditioning); 0.0 represents no difference between performance on the CS+ and CS− levers, with neither rate differing from baseline performance (low fear conditioning). Sham-operated controls and rats with cnA lesions showed a significant bias in responding away from the lever producing the aversive CS+; animals with blA lesions or combined lesions showed no such bias. Analysis of variance with factor ‘lesion’ indicated a significant effect (F3,36 = 3.42, P < 0.05). Pairwise comparisons revealed that sham controls and rats with cnA lesions did not differ in lever discrimination (and hence degree of conditioned fear), but both showed significantly greater levels of conditioned fear (asterisk, P < 0.05) than animals with blA lesions or combined lesions, which themselves did not differ.

Lesions of the lateral and basal amygdala (Figs 2 and 3) did not affect the conditioned suppression of responses elicited by the CS+ (Fig. 1, left), but significantly impaired the ability of the animals to bias their choice of action away from the lever that produced the aversive, punishing stimulus (Fig. 1, right). In contrast, lesions of the central nucleus of the amygdala (Figs 2 and 3) produced severe impairments in the rats' suppression of baseline responses during presentations of the CS+ (Fig. 1, left), but their ability to direct their responses away from the lever that led to these presentations was unaffected (Fig. 1, right). Animals with lesions to both structures, as expected, showed a blockade of both conditioned-punishment and conditioned-suppression effects (Fig. 1).

Figure 2: Representation of largest (pale shading) and smallest (dark shading) lesions of: a, lateral/basolateral nuclei; b, central nucleus; and c, both basolateral and central nuclei of the amygdala.
figure 2

Outlines are reproduced from ref. 30 and represent sections ranging from 1.33 to 3.9 mm posterior to Bregma.

Figure 3: Photomicrographs of sections from three representative brains showing an intact brain (a, b) and excitotoxic lesions of the basolateral (c, d) and central (e,f) amygdala. a, b, A section (right side) through the amygdala of a sham-operated control at low (a) and higher (b) magnification.
figure 3

b The appearance of the lateral (la) and basal magnocellular (bm) nuclei of the amygdala, which are together referred to as the basolateral amygdala, are visible, along with the central nucleus (c). The lateral and medial boundaries of the amygdala are formed by the external capsule (ec) and optic tract (ot), respectively. c, d, A section (left side) through the brain of a subject with a lesion of the basolateral amygdala at the same magnifications as a, and b. The distinctive magnocellular neurons of the basal magnocellular nucleus have been destroyed, so only glial nuclei are visible; the central nucleus has been spared (compare c with a and d with b). e, f, A section (left side) through the brain of a subject with a lesion of the central nucleus, at the same magnifications as a and b. The distinctive neurons of the central nucleus can no longer be seen; only glial nuclei remain stained in this area). The basolateral parts of the amygdala have been spared (compare e with a and f with b).

This double dissociation of the effects of lesions of different amygdaloid nuclei on the behavioural expressions of fear conditioning has important implications for our understanding of the role of the amygdala in the formation of aversive CS–US associations. This is particularly important because the pattern of results cannot be explained by gross changes in primary motivation, such as sensitivity to footshock12, or by deficits in stimulus discrimination (as both conditioned suppression and conditioned punishment are assessed using measures that contrast behaviour associated with the CS+ with that associated with the CS−). According to the hypothesis that the lateral nucleus of the amygdala is the site of formation of the CS–US association, which then achieves behavioural output through projections to the central nucleus of the amygdala4,13, lesions of either the central nucleus or the basolateral amygdala should have produced complete and profound deficits in both conditioned suppression and conditioned punishment. Thus lesions of the basolateral amygdala should have prevented any formation of CS–US associations, whereas lesions of the central nucleus of the amygdala should have prevented such associations from being expressed in the behaviour. The double dissociation shown here demonstrates that both the basolateral amygdala and the central nucleus are not only independently capable of supporting the formation of CS–US associations, but also independently mediate the different forms of behavioural expression of defensive responses that are dependent on these associations.

Two strong conclusions follow from this dissociation. First, the lateral nucleus cannot be the only site at which pavlovian CS–US associations are stored, as rats with extensive lesions of the basolateral amygdala, which included the lateral nucleus, were shown to have intact CS–US associations as assessed by conditioned suppression. Although it has been shown that excitotoxic lesions of the basolateral amygdala produce deficits in the acquisition of conditioned fear responses, as assessed by freezing14, fear-potentiated startle13 and lick suppression12, other findings have suggested that this deficit is abolished following more extended training15. Our rats with lesions in the basolateral amygdala also exhibited in the first session of training a transitory deficit in suppression that was not apparent in any subsequent sessions, where the level of suppression did not differ from that of control animals. In contrast, animals with lesions of the central nucleus of the amygdala or combined lesions showed a persistent deficit across all sessions. Hence, although the basolateral amygdala may contribute to the formation of CS–US associations, the convergence of sensory information mediating conditioned fear can clearly also occur independently of this region, for example within the central nucleus of the amygdala or its afferent circuitry, including direct projections from midline thalamic nuclei and the sensory posterior thalamus16.

Second, our results are not compatible with theories suggesting that all manifestations of aversively motivated conditioned fear behaviour, including conflict and avoidance responses, are mediated by the various descending brainstem and hypothalamic projections of the central nucleus3,5,6. This important conclusion follows because rats with complete lesions of the amygdaloid central nucleus, although impaired in pavlovian conditioned suppression, showed normal formation of an aversive CS–US association as assessed by instrumental conditioned punishment behaviour. Thus the neural basis of any aversive CS–US association and its behavioural expression is distributed more widely in the brain. This finding is in line with previous data showing that, although important in freezing and startle behaviour, the central nucleus of the amygdala is unlikely to be important in active avoidance, conflict or punishment behaviours, or in aversive eyelid conditioning in the rabbit (which seems to depend on the deep nuclei of the cerebellum and midbrain, including the interpositus and red nuclei17). Serial information transfer from the basolateral to the central nuclei of the amygdala seems to be critical only for certain classes of conditioned fear responses, assessed using specific behavioural output systems (indeed, the basolateral amygdala has been shown to be important in the unconditioned, as well as the conditioned, modulation of startle responding18), and does not seem to provide a general mechanism underlying fear conditioning. Similarly, just as different forms of conditioned fear responses seem to depend on dissociable systems in the amygdala, assessments of pavlovian appetitive responding have failed to demonstrate a unitary role for amygdala nuclei in the development of simple appetitive CS–US associations. Rather, they implicate the basolateral amygdala in the control of behaviour by secondary reinforcers9, and they implicate the central nucleus of the amygdala in behaviour dependent on the predictive status of the CS itself, regardless of its association with primary reinforcement8.

Our results support the hypothesis3,4,5,6,7 that projections from the central nucleus of the amygdala are involved in conditioned responses elicited by aversive CSs. However, they also suggest that different projections from the basolateral amygdala are important in the production and direction of instrumental actions in response to fearful stimuli9. The demonstration that these dissociable systems can support different forms of fear-related behaviour, each dependent on the formation of CS–US associations, suggests that aversive responding in general may comprise aspects of behaviour derived from both systems. It is well established that projections from the central nucleus are important in reflexive emotional responding3,4,5,6. However, the role of direct projections from the basolateral amygdala to areas such as the ventral striatum and medial prefrontal cortex19 is not fully understood. These regions are important elements in the limbic cortico-ventral striatopallidal circuitry that provides an interface between the processing of emotionally salient stimuli and intentional action6,9. For example, in humans, areas of the orbital prefrontal cortex that have significant reciprocal connections with the basolateral amygdala have been implicated in the assignment of affective or ‘somatic’ markers that inform choice behaviour20,21. Moreover, our previous work on the neural mechanisms underlying the control of instrumental behaviour by appetitive CSs has demonstrated that interactions between the basolateral amygdala and the ventral striatum provide an important route by which associative processes gain access to instrumental response mechanisms9,22. Our present results suggest that comparable interactions may underlie the effects of aversive conditioned punishers on aversive instrumental discrimination. Indeed, one way in which the present results may be interpreted is to regard the basolateral amygdala as part of a system responsible for voluntary or intentional instrumental choice behaviour based on emotional events, whereas the central nucleus of the amygdala is involved more closely in the reflexive, automatic pavlovian conditioned responses evoked by motivationally salient stimuli.

These results also have more direct implications for clinical research into the function of the amygdala in fear, anxiety and emotional behaviour in general. The amygdala in humans has been shown to be involved in varied aspects of fear conditioning23, emotional memory24,25 and the recognition of emotion in facial expressions26,27 or vocal intonation28. For example, our findings may be seen to provide neuroanatomical support for theories of anxiety that explain the heterogeneity of disorders by appealing to interactions between the dual mechanisms of reflexive pavlovian conditioned fear and voluntary avoidance behaviour, mediated by related neural systems29. In the analysis of the effects of damage to the amygdala in humans23, it may be important to consider the dissociable roles in different aspects of fear-related behaviour of nuclei of the amygdala and the subsystems of which they are part.


Behavioural procedures. We trained 42 rats in operant chambers to maintain pressing on two levers for food on independent variable interval 60-s schedules. After lesions were made the rats were trained on a concurrent conditioned punishment and suppression task. Superimposed on the schedules of food reinforcement were two further independent variable interval 120-s schedules of a response-contingent 10-s auditory CS+ (3-kHz tone or 10-Hz clicks at 80 dB, counterbalanced) terminated by mild footshock (0.2 mA for 0.5 s) on one lever, and a matched neutral CS− (clicks or tone) on the other. Even with a low response rate, the number of stimulus presentations on such schedules is effectively independent of response rate, so all animals received equivalent numbers of CS–US pairings and control CS presentations, regardless of their actual level of responding. In each of 10 30-min sessions the rats received an average of about 12 presentations of the CS+ and CS−. Rates of pressing on the two levers were recorded during baseline sessions in which no stimuli were presented, and in test sessions during the CS+ and CS− and in the absence of stimulus presentation, during the inter-trial interval.

Surgical procedures. The rats were anaesthetized using Avertin and received either bilateral excitotoxic lesions of the central nucleus of the amygdala with 0.063 M ibotenic acid (AP −2.2, −2.7, L ±4.0, V −7.8 (from Bregma), 0.2 µl per infusion site), or bilateral excitotoxic lesions of the basolateral amygdala with 0.09 M quinolinic acid (AP −2.3, −3.0, L ±4.6, V −7.3 (from Bregma), 0.3 µl per infusion site). Sham-lesioned animals received identical surgical treatment except for the infusion of excitotoxin. The choice of neurotoxins was based on extensive pilot experiments achieving discrete lesions of the two areas. Subsequent histological analysis revealed selective lesions of the central nucleus (n = 6), the lateral and basal magnocellular nuclei (n = 9), along with lesions of both structures (n = 9) and sham controls (n = 18).