This experiment investigated the impact of expectations on subsequent neural responses to aversive pictures in 2 contexts: uncertainty signaled by a cue that could be followed by either an aversive or neutral picture, and certainty signaled by a cue that was always followed by an aversive picture. Expectations produced by these cues modulated mid-insula and amygdala responses to aversion, with greater activation observed in response to aversive pictures preceded by an uncertain cue than aversive pictures preceded by a certain cue. Individual differences in anticipatory ACC activity were inversely related to these bilateral mid-insula and amygdala responses, such that individuals with greater ACC activation to the uncertain than certain cue showed the smallest uncertainty-related enhancement of mid-insula and amygdala activation to aversive pictures. In addition, ACC coupling with the mid-insula and amygdala was greater for aversive pictures preceded by an uncertain cue than aversive pictures preceded by a certain cue. Finally, the ACC and insula were correlated with overestimations of the frequency of aversive pictures following the uncertain cue, providing the first evidence of brain mechanisms contributing to covariation bias.
The impact of uncertainty on insula and amygdala responses to aversion likely affects emotion-related functions of these brain areas. Current conceptualizations of insula function emphasize interoception and the detailed representation of the body's internal state, including peripheral signals relevant for emotion (Craig 2002
; Critchley 2004
; Critchley et al. 2004
). In a recent review, Craig (2009)
argued that the insula is centrally involved in self-awareness more broadly, generating a stable representation of the self over time and allowing for the representation and awareness of current feelings as well as predictions about future feelings. Findings here indicate that uncertainty results in modulation of the mid-insula, which in Craig's model integrates emotionally salient environment stimuli such as uncertainty with the representation of bodily signals previously mapped in the posterior insula. His model posits that this integrative process culminates in a final representation of a given moment in time in the anterior insula, which is the predominant location of the aversion effects here and in other reports (Wicker et al. 2003
; Kong et al. 2006
; Nitschke, Sarinopoulos, et al. 2006
; Craig 2009
). Moreover, the left insula effects for uncertainty and aversion, but not the right insula effects, were asymmetric, as was the greater left anterior insula anticipatory activity to the certain aversive than uncertain cue. The proposed asymmetry of parasympathetic influences to the left insula (Craig 2005
) suggests that certainty about an upcoming aversive stimulus may activate the parasympathetic system, as might the presentation of aversive pictures after uncertain cues have been shown. Associations of the left insula with positive affect predicted by Craig's model were not found, although the expected associations for the right insula with negative affect were observed (supplementary Fig. S7A–C
). Formal tests of asymmetry, such as those used in the present report, have rarely been applied to the insula and are needed to further address Craig's proposed roles for each insula in autonomic and emotional behavior.
To investigate neural mechanisms predicting the observed insula and amygdala modulation, we examined activity during the anticipatory phase that defined the context on any given trial as either certain or uncertain. The findings of an inverse association between anticipatory ACC activity on uncertain versus certain trials and bilateral mid-insula and amygdala responses to subsequent aversive pictures are consistent with prior reports implicating the ACC in anticipation-driven modulatory functions (Ploghaus et al. 2003
; Petrovic et al. 2005
; Phan et al. 2005
; Sarinopoulos et al. 2006
) and top-down modulation of insula and amygdala responsivity (Etkin et al. 2006
; Sarinopoulos et al. 2006
; Egner et al. 2008
). Anticipatory ACC activity may play a causal role in modulating subsequent neural firing in the insula and amygdala during the perception of aversive pictures. Extensive monosynaptic white matter tracts connect the ACC to both the insula and the amygdala (Augustine 1996
; Roberts et al. 2007
; Bissière et al. 2008
; Johansen-Berg et al. 2008
; Beckmann et al. 2009
). However, the correlational nature of fMRI data precludes definitive conclusions about causation. Alternatively, the insula and amygdala responses to aversive pictures may influence anticipatory ACC activity on subsequent trials. Causality could be further assessed using the same paradigm in cingulotomy patients or people with ACC damage. Absent insula and amygdala modulation in such individuals, in conjunction with the findings here, would provide support for causative effects of the ACC on mid-insula and amygdala responses to aversion.
An important question following from the findings of anticipatory ACC activity is whether the ACC continues to show evidence of regulatory functions after picture presentation. Indeed, individual differences in ACC responses to the aversive pictures showed the same inverse association with the left mid-insula and bilateral amygdala, replicating recent findings from an emotion regulation paradigm (Urry et al. 2006
; Johnstone et al. 2007
) and an emotional Stroop task (Etkin et al. 2006
; Egner et al. 2008
). Complementing these cross-subject connectivity analyses that average ACC activity over all trials of each condition, we also conducted within-subject connectivity analyses examining intraindividual variation on a trial-by-trial basis. These functional connectivity analyses indicated that ACC coupling with the mid-insula and amygdala was consistently stronger in response to aversive pictures following the uncertain cue than aversive pictures following the certain cue. These data suggest that uncertainty results in tighter coupling of the ACC to insula and amygdala responses to negative outcome.
Recruitment of the ACC and insula nodes of this modulatory network was reliably associated with uncertainty-related covariation bias. ACC and insula responses to uncertainty during the scan predicted individual differences in covariation estimates provided after the scan about the percentage of uncertain cues followed by aversive pictures. Individuals with greater anticipatory activity in the supragenual ACC following the uncertain than certain cue showed biased covariation estimates. Subgenual ACC responses to the aversive pictures showed the same pattern. Both of these areas overlapped with the respective ACC areas found to be associated with insula responses to the aversive pictures (, supplementary Fig. S4
), suggesting that although ACC activation may be adaptive in downregulating insula and amygdala responses to aversion, this comes at the cost of distorting perceptions about aversive experiences in these individuals. This downregulation of the insula would be expected to compromise its function in updating bodily representations in these individuals (Craig 2002
; Damasio 2003
; Critchley 2004
), thereby resulting in less accurate perceptions about their exposure to aversion when faced with uncertainty. Supporting this interpretation, bilateral anterior insula and mid-insula responses to the pictures were inversely associated with covariation bias. Supragenual ACC responses to the aversive pictures showed the same pattern as these insula areas, consistent with the tight supragenual ACC coupling with mid-insula responses to aversive pictures following the uncertain cue.
The findings for covariation bias were not explained by negative affect. Instead of showing uncertainty-related effects (e.g., uncertain cue–certain aversive cue), negative affect was positively associated with greater supragenual ACC, right anterior insula, and left mid-insula activity for certain and uncertain cues as well as for the aversive pictures following each cue. The indiscriminate recruitment of the supragenual ACC and insula across these different contexts in individuals endorsing higher levels of negative affect may be pathognomic in patients with anxiety and mood disorders (Nitschke and Heller 2005
; Paulus and Stein 2006
). Unlike the insula findings for covariation bias, these right anterior insula and left mid-insula findings for negative affect were asymmetric. These effects are consistent with findings for the right anterior insula in low trait resilience (Waugh et al. 2008
) and for the left mid-insula in state negative affect (Mériau et al. 2009
), although asymmetry was not statistically evaluated in those reports. Because negative affect and a general negativity bias (estimated proportion of certain aversive versus neutral cues) did not account for the associations of uncertainty-related covariation bias to ACC and insula activity, research is needed to address other potential contributing factors, such as expectancy bias (Davey 1992
), memory bias for emotional events (Christianson 1992
; Phelps 2006
; Mackiewicz et al. 2006
), or cardiorespiratory activity (Rainville et al. 2005
The current study can be grounded in the context of reinforcement learning models, which posit that learning is a result of discrepancies between an organism's predictions about future events and the occurrence of such events (Rescorla and Wagner 1972
; Schultz et al. 1997
; Niv and Schoenbaum 2008
). Research into the neural basis of such discrepancies, termed prediction errors, has predominantly focused on reward prediction errors generated by dopaminergic neurons (Schultz et al. 1997
; Niv and Schoenbaum 2008
). There have been recent proposals that the anterior insula plays an analogous role in signaling probabilities of risk and risk prediction errors, thus promoting learning in the context of risky or unpredictable circumstances (Paulus and Stein 2006
; Preuschoff et al. 2008
). In the current study, it might be hypothesized that increased mid-insula activity to aversive pictures following uncertain cues reflects risk prediction errors. The inverse relationship between covariation estimates and insula activation to aversive pictures following uncertain cues would support this hypothesis if higher covariation estimates after
the scan reflected greater risk predictions following uncertain cues during
the scan. If this were the case, a stronger covariation bias would indicate greater risk predictions and smaller risk prediction errors for aversive pictures. Of note, differences in brain activation to expected versus unexpected outcomes may occur for reasons other than prediction error signaling, such as attention modulation or the encoding of stimulus value (Belova et al. 2007
; Niv and Schoenbaum 2008
). Activity associated with prediction errors could be assessed in future work with our anticipation paradigm by monitoring expectancies/predictions of aversion on a trial-by-trial basis. Indeed, a limitation of the present study is that this version of our paradigm did not include on-line ratings of expectancies that could be used to assess expectancy bias and prediction errors, or on-line affective ratings that could be used to assess subjective responses to the aversive pictures.
In sum, the impact of aversion on the brain is contingent upon expectations about whether the aversive outcome is a definite certainty or only a possibility. The neural mechanisms of this phenomenon include the ACC, insula, and amygdala. Extending prior work demonstrating recruitment of these regions in response to a wide range of aversive stimuli, mid-insula and amygdala activation to aversive pictures was greater after an uncertain cue that was sometimes followed by an aversive picture as compared with a certain cue that was always followed by an aversive picture. Anticipatory activity in the supragenual ACC predicted this impact of uncertainty on mid-insula responses to aversion, whereas anticipatory activity in the subgenual and pregenual ACC predicted this impact of uncertainty on amygdala responses to aversion. Findings for covariation bias suggest that multiple ACC and insula areas influence perceptions of cue–outcome contingencies, resulting in the observed overestimations of aversive pictures following the uncertain cue. This study highlights the importance of chronometry in the context of uncertainty and aversion by dissociating anticipatory ACC effects from neural responses to aversive stimuli that follow. Individual differences in ACC and insula activation have behavioral significance of particular relevance to anxiety disorders, both in terms of disposition toward negative affect and a covariation bias that accompanies uncertainty about aversive outcomes.