Emotion regulation has been defined as those processes involved in changing the onset, duration, intensity or content of an emotional response (Gross, 1998
; Gross, 2008
). Emotion regulation processes range from actions taken long before an emotion arises, such as situation selection, to those processes engaged either just prior to or once an emotion has begun to emerge, such as attention deployment or cognitive reappraisal (Gross, 1998
). It is in these latter types of strategies that investigations into the relationship between regions associated with the cognitive control of emotion and those associated with the emotional response become of greatest interest. These investigations either implicitly or explicitly describe emotion regulation as the deployment of top-down, ‘cold’ cognitive control regions of the PFC to down regulate bottom-up, ‘hot’ reactive processes involving the subcortical limbic regions like the amygdala. Failures in the successful deployment of PFC top-down cognitive control mechanisms or overactive bottom-up amygdala processes have been proposed to contribute to several forms of psychopathology (Rottenberg & Gross, 2003
; Rottenberg & Johnson, 2007
The emotion regulation strategy that has received the most attention in the neuroimaging literature is cognitive reappraisal. This regulation strategy involves cognitively reinterpreting emotional information in order to change an emotional response (Gross, 1998
). Reappraisal encompasses a broad class of related processes. For example, a reappraisal can focus on the reinterpretation of the personal meaning of the emotional object to make it more or less self-relevant. Alternatively, a reappraisal can focus on reinterpreting the cause, consequence, or the reality of emotional stimuli without changing one’s relationship to the stimuli. For example, one could reappraise a car accident on the side of the road as probably ending with all parties walking away from the incident safely. A number of functional neuroimaging studies have now been performed during reappraisal tasks, and are listed in , with the location of PFC activations displayed in . Using the key words emotion regulation, distraction and reappraisal, empirical articles measuring voluntary emotion regulation were included. These fMRI studies consisted of instructed cognitive reappraisal, emotion suppression and distraction studies in non-clinical populations. This list of emotion regulation studies is not exhaustive; for example, it does not include related concepts like mood regulation. We note that in all Tables we have retained the nomenclature (applied Brodmann labels, or topographical/regional descriptions) used by the authors of the original papers. There are some cases where questions could be raised about the specific application of labels, but lacking a published “gold standard” coordinate system for most prefrontal regions, we have not generally changed labels, with the exception that in the text we specifically note VLPFC activations that are consistent with the posterior part of BA 47/12. Lacking a clear demarcation of the portion of BA 47/12 with significant amygdala connections in humans, we consider the portion of the region that is posterior to y = 32 as generally representing posterior BA 47/12. We also indicate in text when OFC foci are consistent with the location of BA 13 (regardless of their original designation).
Prefrontal Regions Recruited During Emotion Regulation
Figure 11 Areas activated during emotional regulation of negative emotions. The cyan markers are surface renderings of coordinates reported as more engaged in reappraisal to decrease negative emotion than a non-regulated condition. The blue markers are coordinates (more ...)
The most common paradigm for studying reappraisal asks participants to view primarily negatively valenced, highly arousing, static images (e.g., mutilation, assault, decay and defecation) and compares neural activation during trials cued for cognitive reappraisal with trials cued for passive viewing (Eippert et al., 2006; Kim et al., 2007
; Ochsner et al., 2002
; Ochsner et al., 2004
; Phan et al., 2005
; Urry et al., 2006
; Van Reekum et al., 2007
). While there are variations in the details of the reappraisal instructions from study to study, they consistently require participants to create a new interpretation of the meaning, cause, consequence or the personal significance of the image during the reappraisal trials. Reappraisal contrasted with unregulated viewing of negative images recruits broad areas of the PFC, including bilateral DLPFC and VLPFC (often more heavily left sided), and regions of the dorsal ACC and/or medial PFC as supporting the cognitive control aspects of reappraisal. displays the location of reappraisal related activations (cyan markers for decreasing negatively valenced stimuli, and yellow for decreasing positively arousing stimuli) from the above cited studies.
A related paradigm uses dynamic movie images instead of static pictures. These studies also demonstrate recruitment of bilateral DLPFC during cognitive reappraisal but vary as to whether regions of ACC and medial PFC are additionally recruited to decrease sadness, disgust or sexual arousal (Beauregard et al., 2001
; Goldin et al., 2008
; Levesque et al., 2003
In several reappraisal studies utilizing either static or dynamic images, amygdala decreases were used as a proxy for change in negative valence and arousal along with decreases in insula recruitment in some studies (Goldin et al., 2008
; Levesque et al., 2003
; Ochsner et al., 2002
; Phan et al., 2005
). We note that a simple equating of amygdalar activity with negative affect is problematic, given that 1) the amygdala becomes active in situations that are not negative, and 2) negative affective experiences involve cortical and subcortical components that extend beyond the amygdala. However, given our interest in regional brain interactions, the down-regulation of the amygdalar activity provides a useful index for measuring prefrontal-limbic interactions regardless of the extent to which its activity correlates with negative affect. Most of the studies find decreases in the left amygdala, and often bilateral amygdalae, when utilizing reappraisal to down regulate negative affect. Only a couple of studies have examined reappraisal of positively valenced stimuli. When asked to reappraise or down regulate positive or sexually arousing stimuli, the level of right amygdala activation to the stimuli decreased (Beauregard et al., 2001
; Kim & Hamann, 2007
). This may raise speculation as to the laterality of emotion regulation, but in general, studies testing for formal interactions with amygdala laterality are lacking.
Another emotion regulation strategy involves bringing to mind positive or soothing images either from nature or from one’s past either to replace or counteract negative affect. Behavioral experiments demonstrate that recalling mood incongruent memories or images decreases negative affect (Erber & Erber, 1994
; Joormann, Seimer & Gotlib, 2007
; Parrott & Sabini, 1990
; Rusting & DeHart, 2000
). Two neuroimaging studies compared regulating one’s affect by calling to mind a calming image or memory to the unregulated anticipation of shock. Kalisch and colleagues (2005)
cued trials with tones indicating whether there was a probability of shock on those trials or not. In the regulation trials, participants were instructed to detach from their feelings of anxiety and think of a special place identified earlier. In the non-regulation trials, participants were instructed to engage with their emotional responses. ROI analyses showed that this form of regulation recruited a region of right anterolateral frontal cortex (MNI: 42, 48, 18) and regulation in the presence of anxiety recruited regions of the medial PFC and rostral ACC (−4, 46, 28). In a similar study, Delgado and colleagues (2008b)
used colored blocks to designate trials in which shock was possible, and asked participants to regulate their anxiety by calling to mind one of two pre-identified places in nature. Their ROI analyses show that calling to mind nature images when anticipating shock recruits the left middle frontal gyrus (Talairach: −43, 28, 30). The amplitude of which was associated with regulation success. Regulation also resulted in activation in the ventral medial wall and subgenual cingulate (BA 32; −3 36, −8 and BA 25; 0, 14, −11), which the authors point out has been associated with extinction (Phelps et al., 2004
) and decreases in left amygdalar activity. While both of these studies employ similar paradigms, their analytic approaches including choice of ROIs and modeling of tonic versus phasic effects may be responsible for some of the differences in regions reported for drawing upon positive or soothing images to counter the anxiety associated with waiting for possible shock.
Similar to the prior emotion regulation strategy, distraction involves holding neutral and irrelevant information in one’s working memory. Behavioral research shows that doing so decreases negative affect in both dysphoric and nondysphoric individuals (Fennell, Teasdale, Jones, & Damle, 1987
; Lyubomirsky, Caldwell, & Nolen-Hoeksema, 1998
; Teasdale & Rezin, 1978
). By taking up working memory capacity with mood incongruent cognitions, mood congruent thoughts are prevented from gaining access to attentional resources (Siemer, 2005
). Neuroimaging studies of distraction have utilized two different paradigms. The first, employed by Kalisch et al. (2006)
, utilized the anticipation of shock paradigm, except instead of having the participant recall a pleasant or safe memory, there was an open distraction instruction in which the participant was encouraged to think of anything other than the possible shock. This paradigm identified a region of the left PFC (MNI: −56, 30, 22) that was more active in trials in which participants were instructed to distract themselves than in the no distraction trials. The second distraction paradigm involved an assigned distraction task (Sternberg working memory task) in which the participant holds a series of letters in working memory while viewing negative or neutral static images and then following the picture offset has to respond to whether a single letter was in the set they were holding in mind. McRae et al. (2009) report that engaging in a working memory task while viewing negative slides as compared to passive viewing increases the BOLD response in left and right superior and middle frontal gyri (MNI: BA6; −6, 10, 62 and −56, −4, 48 and 48, 42, 32; BA 9; −42, 22, 30 and 42, 30, 34; BA 10; −36, 62, 12 and 38, 64, 14) as well as right inferior PFC (BA47/12p; 36, 20, −4).
Many neuroimaging reports of emotion regulation explicitly present DLPFC regions as being engaged in some kind of cognitive control and are cautious about attributing concurrent decreases in amygdala responses to direct connections with the amygdala. In the case of reappraisal and distraction, this caution is particularly warranted since these processes produce foci that are distributed across the PFC (). As mentioned earlier, the pattern of anatomical projections from the cortex suggest that direct paths from regions of DLPFC are unlikely to exert strong control of amygdala processing. Areas of the PFC with moderately dense projections in the lateral PFC are only found in a small portion of the VLPFC, specifically in the more posterior regions of BA 47/12. Unfortunately, as mentioned earlier, the nomenclature used to report activations in this region in most studies creates ambiguity when it comes to questions of connectivity with the amygdala. Studies of reappraisal, positive memory or image engagement and distraction commonly report activations in the general regions of VLPFC and medial OFC (Eippert et al., 2007
; Goldin et al., 2008
; Kim & Hamann, 2007
; Lieberman et al., 2007; McRae et al., 2009; Ochsner, Ray et al., 2004
). Specifically, many of the reappraisal studies report bilateral activations of BA 47/12 when decreasing negative or positive emotion. As noted above, BA 47/12 is a large and heterogeneous area and only posterior regions of BA 47/12 are sites of significant amygdalar projections. Therefore, strong statements about direct cognitive influence on the amygdala become more plausible in those studies with activations in this specific segment of BA 47/12.
Medial regions of the PFC are often treated as having privileged access to subcortical regions such as the amygdala. However, according to the mapped medial direct connections to the amygdala, only those regions of subgenual cingulate (BA 25) and dorsal ACC (BA 24) have dense direct connections with the amygdala. Only the studies by Delgado and colleagues (2008a
) report foci on the medial surface that are in regions positioned to broadly impact the amygdala. Given the anatomical data, it may seem surprising that activations of BA25 does not arise more frequently in these studies. However, it is plausible that signal drop out in the posterior VMPFC has prevented studies from demonstrating more consistent activation in this region. More frequently, studies of inhibition/suppression, distraction and reappraisal only report foci in BA 32, which may reflect a more specific modulation of the amygdala, given the more circumscribed nature of BA 32 input to the amygdala.
Correlational Studies of amygdala deactivation
In order to understand in more detail how top-down emotion regulation interacts with the amygdala, a subset of emotion regulation studies have gone further than task versus control contrasts to investigate the specific correlates of decreases in amygdala activity (See ). That is to say, instead of asking what areas are engaged in a task known to down regulate amygdala activity, they explicitly tested the correlation or functional/effective connectivity between the amygdala and the whole brain during emotional regulation performance. Alternatively, some studies correlated amygdala decreases with already identified prefrontal regions from the main regulation contrasts. These studies indicate that amygdalar decreases are negatively correlated with many areas of PFC activity. Of particular note are the activations in the VMPFC, including BA 11m/14r (5, 37, −12; −6, 46, −20: Urry et al., 2006
, Ochsner et al., 2002
respectively). Additionally, subgenual and pregenual cingulate regions were observed to be negatively correlated with amygdala activity during regulation. For instance, Urry and colleagues (2006)
reported a region of BA 32/10 (maximum at −23, 43, −10) that extended ventrally and medially. Delgado et al. (2008b)
also report an inverse correlation between BA 32 (0, 35, −8) activity and amygdala decreases. Posterior (BA 13) areas of the OFC also negatively correlated with amygdala deactivation (−24, 28, −14; 26, 24, −22: Banks et al, 2007: −30, 22, −16; 34, 24, −16: Ochsner, Ray et al., 2004
). Less ventral areas of the PFC in BA 47 (34, 54, 12) and BA46 (−54, 12, 12: Urry et al., 2006
; −54, 42, 12: Ochsner et al., 2002
), also arose in these studies. Two studies statistically linked specific DLPFC regions to medial regions, which then corresponded to decreases in amygdala response. In a study by Urry et al. (2006)
, a mediation analysis demonstrated the connection between the amygdala, BA 10 (3, 63, 18) and a DLPFC region (−50, 23, 19). Delgado et al. (2008b)
alternatively used the medial BA 32 region as the seed for their PPI analysis that then identified a left amygdala region and a DLPFC region. Importantly, these studies identify regions corresponding to amygdala decreases that have also been noted above as projecting to the amygdala such as the dorsal anterior cingulate, subgenual cingulate and posterior orbitofrontal cortex.
Studies that report correlations between decreased amygdala activity and prefrontal region increases during emotion regulation tasks.
Of the regions reported from these correlational analyses or multiple regression analyses, a limited number of them have plausible direct connections into the amygdala. The most common regions that are negatively correlated with amygdala response are regions of the posterior OFC and subgenual cingulate and VLPFC (). Of the lateral prefrontal regions only the posterior lateral portion of BA 47/12 has strong projections to the amygdala. Regions of anterior BA 32 are also identified in correlational analyses, which could reflect projections to the assessory and basal lateral nucleus of the amygdala (Cheba et al, 2001
Figure 12 Coordinates identified in as correlated with deactivations in the amygdala during emotion regulation plotted on the surface of a template brain (top left and right) and rendered on a glass brain (bottom view and left view). The cyan markers are (more ...)
Models of emotion regulation
To date, the most sophisticated data driven model of emotion regulation comes from a study of positive reappraisal by Wager and colleagues (2008)
. The outcome variable of interest is change in self-reported negative affect. A structural equation methodology was applied to a neuroimaging dataset from a reappraisal paradigm similar to the ones used by Ochsner et al. (2002
. The right VLPFC was chosen as the starting point for the analyses, with coordinates centered in an area that plausibly includes the posterior portion of area 47/12 with projections to the amygdala. The authors first used an ROI approach to test the role of the amygdala and nucleus accumbens as mediators between the right VLPFC and decreased negative affect which was identified as the primary metric of reappraisal success. In this ROI analysis both structures were shown to mediate the relationship between the right VLPFC and self-reported decrease in negative affect (see ).
Figure 13 A diagram of the mediation analysis testing the relationship between the right VLPFC and decreases in negative affect mediated by activation in the amygdala and nucleus accumbens. Figure adapted with permission from Wager, Davidson, Hughes, Lindquist, (more ...)
The authors then used whole brain cluster analysis and nonparametric inference to identify two networks as possible mediators of the relationship between the VLPFC and changes in self-reported negative affect (see ). One network has an indirect positive bias towards increasing the change in negative affect. This network includes regions of nucleus accumbens, subgenual cingulate (BA 25), pre-SMA, precuneus, DMPFC (MNI: 24, 41, 40), and superior frontal gyrus (24, 21, 58). Amongst these regions, the nucleus accumbens and subgenual cingulate have the most interconnection with the amygdala. The second network identified has an indirect negative bias towards decreasing the change in negative affect and reducing reappraisal success. This network includes the rostral dorsal ACC, amygdala (bilateral) and posterior-lateral OFC (48, 24, −18). Future work will have to elucidate how the components of the networks interact and whether these networks are specific to this particular type of emotion regulation strategy.
Figure 14 Path model of the positively biased network in yellow and negatively biased network in blue mediating the relationship between the VLPFC and the decrease in self-reported negative affect. Figure adapted with permission from Wager, Davidson, Hughes, Lindquist, (more ...)
Several investigators have put forth theoretical models as to the neural mechanisms behind emotion regulation. The simplest of these models proposes that a limited number of areas exert a direct influence on the amygdala. Delgado et al. (2008b)
, Hansel and von Kanel (2008)
and Quirk and Beer (2006)
each propose that the ventromedial PFC down regulates regions of the amygdala. These models importantly attempt to ground our understanding of the neuroanatomical bases of human emotion regulation in the extensive animal literature on extinction and the ventromedial PFC’s connections to the intercalcated masses in the basolateral amygdala (Morgan, Romanski & LeDoux, 1993
; Likhtik et al., 2005
; Quirk et al., 2000
). Quirk and Beer (2006)
build on the presence of both excitatory and inhibitory effects of the “ventral” medial PFC projections to the amygdala in humans and rats. The subgenual cingulate region, BA 25, is argued to be more inhibitory whereas the more dorsal and anterior BA 32 is proposed to have excitatory connections with the amygdala. Both BA 25 and 32 have connections with the amygdala. BA 32, however, has much more limited connections.
Phillips et al (2008)
have developed a circuit model that attempts to explain the neural underpinnings of multiple types of emotion regulation (see ). The model contains component regions of the DLPFC, OFC, VLPFC, DMPFC and ACC. Of particular interest, the authors distinguish between areas involved in automatic emotion regulation (in subgenual and rostral ACC) and regions that are recruited for voluntary emotion regulation (DLPFC and VLPFC). They characterize these latter regions as phylogenetically newer and providing feedback to the older emotion generation processes. The OFC, DMPFC and ACC, on the other hand are phylogenetically older regions that are described as operating through feedforward processes to relay internal state information to the DLPFC and VLPFC. The authors place the DMPFC as the conduit through which the OFC feeds value information forward to neocortical regions of the brain for decision processes.
Figure 15 Phillips et al. (2008) model of prefrontal amygdala interactions a) The OFC, subgenual ACG (ACC), and rostral ACG (ACC) feedforward information to the MdPFC and then to the lateral PFC regions for decision and action. B) The feedback processes from the (more ...)
One unique aspect to this model is the explicit articulation of the processes of feedforward and feedback. The model is intuitively appealing and clearly fits with traditional ideas about the DLPFC exerting top-down control over more “emotional” regions. However, it is difficult to reconcile this conceptualization with the structural model, given the laminar distribution of PFC connections (Barbas & Rempel-Clower, 1997
; Barbas, 2000
). Indeed, the structural model suggests that the information flow between the DLPFC and the OFC is actually in the opposite direction with processes originating in the OFC and going to the DLPFC characterized predominantly as feedback, and those arising in the DLPFC and going to the OFC predominantly characterized as feedforward.
The Phillips et al. model is also notable in its placement of so called “automatic regulation” regions such as the subgenual cingulate and OFC as the primary route through which more phylogenetically newer regions impact limbic areas such as the amygdala. This is largely consistent (particularly the subgenual cingulate region) with the network arrangements described above. It may be speculated, however, that there may be more than one route through which voluntary emotion control areas can impact amygdala processing. In particular, the posterior VLPFC may be able to directly impact amygdala processes without requiring engagement of one of the more medial “automatic regulation” regions, given its direct inputs to amygdala nuclei.
In summary, a wealth of data indicate the engagement of PFC regions during emotion regulation tasks, with activity in a more select group of areas (BA 47/12, BA25 and BA 32) showing associations with the ability to down-regulate amygdala activity. Increasingly sophisticated models have been proposed to explain these data. The emergence of these models is appealing, as is the concern shown by their authors for the plausibility of the proposed connectional networks. We do note, however, that no models to date have explicitly acknowledged the laminar pattern of connections between different PFC regions. For instance, Wager et al (2008)
provides the most complicated model for a particular emotion regulation strategy, but does not address the nature of the information flow between the component regions. Phillips et al. more explicitly incorporate the concept of feedforward and feedback information, but do not reconcile these ideas with the observed pattern of feedback and feedforward projections in the regions in question. We believe that reconciling these issues provides one of the key challenges for researchers attempting to understand the neural substrates of emotion regulation.