Anxiety disorders are marked by excessive fear (and avoidance), often in response to specific objects or situations and in the absence of true danger, and they are extremely common in the general population. According to a recent epidemiological study, the lifetime prevalence of any anxiety disorder is 28.8% (Kessler et al, 2005). Anxiety disorders are associated with impaired workplace performance and hefty economic costs (Greenberg et al, 1999), as well as an increased risk of cardiovascular morbidity and mortality (Albert et al, 2005; Kawachi et al, 1994; Smoller et al, 2007). Given that anxiety disorders are a significant problem in the community, recent neuroimaging research has focused on determining the brain circuits that underlie them to inform the use of existing treatments and guide the possible development of new treatments. In the future, neuroimaging studies of anxiety disorders may also prove to be clinically helpful in the prediction of treatment response.
Given that excessive fear is a key component of anxiety disorders, it is not surprising that the search for the neurocircuitry of anxiety disorders has its roots in and has been closely intertwined with studies of fear circuits in animal models. A large volume of experimental work has examined the neurocircuitry associated with fear responses, mainly in rodents, using primarily fear conditioning, inhibitory avoidance, and fear-potentiated startle models. Key components of fear circuitry including the amygdala (and its subnuclei), nucleus accumbens (including bed nucleus of stria terminalis BNST), hippocampus, ventro-medial hypothalamus, periaqueductal gray, a number of brain stem nuclei, thalamic nuclei, insular cortex, and some prefrontal regions (mainly infralimbic cortex) have been identified in these studies (for recent reviews see Davis, 2006; Maren, 2008; Quirk and Mueller, 2008). These regions have their respective roles in the various components of fear processing such as the perception of threat or of unconditioned stimuli, the pairing of an unconditioned stimulus and conditioned response (learning/conditioning), the execution of efferent components of fear response, and the modulation of fear responses through potentiation, contextual modulation, or extinction. Some key findings from animal literature, such as the central role of amygdaloid nuclei in the acquisition of fear conditioning and expression of fear responses, the involvement of the hippocampus in contextual processing, and the importance of the infralimbic cortex in extinction recall, have been replicated across different studies and laboratories. These basic components of fear circuitry are well preserved across species and likely support similar functions in humans. Animal work using in vivo electrophysiological recording, tracing and lesions/reversible inactivation techniques was indispensable in acquiring this knowledge. Some recent work had even suggested that there might be separate fear and anxiety systems orchestrated through the central nucleus of the amygdala and the BNST, respectively (Davis, 2006). These types of findings are particularly exciting as they might allow for a better focus on the neurocircuits involved in pathological anxiety.
On the other hand, other important issues, such as the exact neuroanatomical region that stores fear memory traces, or the exact role of a particular process (eg, the role of reconsolidation in fear memory, Nader and Hardt, 2009), or of a particular region (eg, the insular cortex) are intensely debated and actively studied. Nevertheless, the basic fear-related neurocircuitry identified in rodents is a useful place to start examining anxiety-related neurocircuitry in humans. It is important to note that the exact roles of many brain regions are yet to be firmly established and could differ across species. Even regions such as the amygdala, hippocampus, and nucleus accumbens might be involved in different, additional, or even unique tasks in humans (eg, the role of the hippocampus in explicit verbal memory in humans). Finally, there are major differences between human anxiety/anxiety disorders and fear conditioning models in animals. These differences include the frequent absence of clear unconditioned stimuli (US) in human anxiety disorders, and the central roles of avoidance and cognitive components (eg, anticipatory anxiety) in humans. These unique characteristics of anxiety disorders suggest potential involvement of other brain regions in addition to those identified in rodents, such as areas of prefrontal cortex that are more unique to humans. Thus, although animal studies are indispensable in understanding basic fear neurocircuitry, in vivo human studies are critical for understanding the neurocircuitry of anxiety disorders.
In this review, we will discuss three main topics: (1) fear neurocircuitry in healthy humans; (2) stress as a normal response to internal and external stimuli, and (3) anxiety disorders as defined in human psychopathology. The first of these topics will include a discussion of Pavlovian fear conditioning and extinction, pharmacologically induced fear and anxiety states, and the assessment of emotional stimuli in humans. In the third topic, we review the role of brain regions such as the amygdala, medial prefrontal cortex (including the rostral anterior cingulate cortex (rACC) and dorsal anterior cingulate cortex (dACC)), hippocampus, and insular cortex in anxiety disorders (Figure 1). Finally, we discuss some of the limitations of neuroimaging studies of anxiety disorders as well as the directions that we expect the field to take in the near future.
Magnetic Resonance Images (Sagittal Slices) Showing the Structures of Interest in This Review: (a) the Hippocampus and the Amygdala; (b) the Dorsal Anterior Cingulate Cortex (dACC) and the Rostral Anterior Cinulate Cortex (rACC); and (c) the Insular Cortex.
Pavlovian fear conditioning and extinction
In its most basic form, Pavlovian fear conditioning involves repeatedly presenting a previously neutral conditioned stimulus (CS; eg, a tone) with an aversive unconditioned stimulus (US; eg, a shock). After repeated pairings, the CS alone comes to elicit a conditioned fear response (eg, increased freezing, fear-potentiated startle, or skin conductance responses). Pavlovian fear conditioning has been used as a testable and translational, though admittedly simplistic, model of the acquisition of fears that might be relevant to some anxiety disorders like phobias and possibly to some aspects of post-traumatic stress disorder (PTSD).
Studies of Pavlovian fear conditioning in non-humans have highlighted the importance of the amygdala in the acquisition of fear conditioning (LeDoux, 2000; LeDoux et al, 1990; Pare et al, 2004; Sananes and Davis, 1992). Similarly, functional neuroimaging studies in humans have reported amygdala activation during fear conditioning (Alvarez et al, 2008; Barrett and Armony, 2009; Buchel et al, 1998, 1999; Cheng et al, 2003, 2006; Gottfried and Dolan, 2004; Knight et al, 2004, 2005; LaBar et al, 1998; Milad et al, 2007b; Morris and Dolan, 2004; Pine et al, 2001; Tabbert et al, 2006), even when the CS is presented below perceptual thresholds (Critchley et al, 2002; Knight et al, 2009; Morris et al, 2001) and even when more complex USs are used (Doronbekov et al, 2005; Klucken et al, 2009). In addition, amygdala activity has been associated with skin conductance changes during fear conditioning (Cheng et al, 2006; Furmark et al, 1997; LaBar et al, 1998; Phelps et al, 2001). Interestingly, amygdala activation in humans also has been observed in response to cues following (1) verbal instructions that discriminate between cues that predict shock vs safety (even though no shock was actually administered) (Phelps et al, 2001), and (2) observational fear learning, whereby participants watch a video of another person experiencing a Pavlovian fear-conditioning paradigm (Olsson et al, 2007). What exactly amygdaloid activation represents in these latter paradigms is not entirely clear. It could suggest for example that: (1) higher order centers that decipher the anticipated predictive value of the cue, or that learn from observation using empathy, convey information to the amygdala, or (2) alternatively, that the human amygdala is less specific in its responses and is more sensitive to contextual modulation in the absence of a US. These interpretations could have potentially different implications for the understanding of the role of the amygdala in anxiety disorders.
Fear conditioning is also associated with increased activation in the dACC and rACC (Alvarez et al, 2008; Buchel et al, 1998, 1999; Dunsmoor et al, 2007; Klucken et al, 2009; LaBar et al, 1998; Marschner et al, 2008; Milad et al, 2007a,b; Morris and Dolan, 2004; Phelps et al, 2004). Activation in the dACC and rACC also occurs during observational fear learning (Olsson et al, 2007). In addition, dACC activation is positively correlated with differential skin conductance responses (Milad et al, 2007a). Fear conditioning studies (involving both specific CSs and contexts) also commonly report insular cortex activation (Alvarez et al, 2008; Buchel et al, 1999; Buchel et al, 1998; Critchley et al, 2002; Dunsmoor et al, 2007; Gottfried and Dolan, 2004; Klucken et al, 2009; Knight et al, 2009; Marschner et al, 2008; Morris and Dolan, 2004; Phelps et al, 2001, 2004) and hippocampal activation (Alvarez et al, 2008; Buchel et al, 1999; Knight et al, 2004, 2009; Lang et al, 2009; Marschner et al, 2008).
Extinction learning occurs when a CS that previously predicted a US no longer does so, and over time, the conditioned response (eg, freezing or elevated skin conductance responses) decreases. Extinction learning or, more likely, the later recall of this learning involves the ventromedial prefrontal cortex (vmPFC) (Milad and Quirk, 2002; Morgan et al, 1993; Quirk et al, 2000, 2003, 2006) in rodents. Functional neuroimaging studies of healthy humans have reported vmPFC activation during extinction (Barrett and Armony, 2009; Gottfried and Dolan, 2004; Kalisch et al, 2006; Milad et al, 2007b) and the later recall of extinction (Milad et al, 2007b; Phelps et al, 2004). Skin conductance measures of extinction memory are positively correlated with vmPFC activation (Milad et al, 2007b; Phelps et al, 2004) and vmPFC cortical thickness (Milad et al, 2005). Activation of the amygdala and insular cortex also may occur during extinction learning or recall (Gottfried and Dolan, 2004; LaBar et al, 1998; Milad et al, 2007b; Phelps et al, 2004), and greater amygdala responses during extinction have been associated with higher trait anxiety (Barrett and Armony, 2009). Finally, extinction can be modulated by context (ie, the surroundings in which extinction takes place), and the hippocampus has a role in this process. In rodents, dorsal hippocampal lesions reduce the context-dependence of extinction (Bouton et al, 2006). In a recent fMRI study, hippocampal activation to the CS+ occurred in the extinction context but not in the conditioning context (Kalisch et al, 2006). Hippocampal activation was also positively correlated with vmPFC activation in this study (Kalisch et al, 2006), suggesting that hippocampal-vmPFC interactions may be important for the contextual modulation of extinction.
Fear states and responses to emotional stimuli
Another way to examine the mediating functional neuroanatomy of fear or anxiety is to use specific pharmacological agents to provoke such states in healthy individuals during PET or fMRI scanning. For example, cholecystokinin-4 (CCK-4) is associated with increases in subjective states of fear and anxiety, as well as increased activation in the amygdala, insular cortex, claustrum, cerebellum, brain stem, and the ACC (Benkelfat et al, 1995; Eser et al, 2009; Javanmard et al, 1999; Schunck et al, 2006). In addition, two studies reported dACC increases during anticipatory anxiety preceding the CCK administration (Eser et al, 2009; Javanmard et al, 1999). It is important to keep in mind, however, that CCK-B receptor agonists like pentagastrin also have direct effects on stress axis stimulation independent of their effects on subjective experience of distress/fear (Abelson et al, 2005, 2008). Procaine administration has been associated with elevated subjective ratings of fear/anxiety, activation of the amygdala, ACC, and insular cortex (Ketter et al, 1996; Servan-Schreiber et al, 1998), and deactivation of neocortical structures (Servan-Schreiber et al, 1998). Furthermore, amygdala activity was positively correlated with subjective reports of anxiety (Ketter et al, 1996; Servan-Schreiber et al, 1998). Interestingly, those subjects who did not have a panic attack in response to procaine had greater activation in the rACC compared with those who did have a panic attack (Servan-Schreiber et al, 1998), consistent with the idea that the rACC may perform a regulatory or inhibitory function (Mayberg, 1997). The alpha-2 adrenergic antagonist yohimbine has likewise been associated with increased normalized blood flow in medial prefrontal cortex, insular cortex, and cerebellum in healthy individuals (Cameron et al, 2000). A major caveat in the interpretation of pharmacological challenge studies, however, is the difficulty in disentangling the effects that are specific to fear induction from the direct effect of a pharmacological agent on regional brain activity and from the non-specific effects of the pharmacological agent.
Over the past two decades, functional neuroimaging studies have shown that a core set of brain regions mediate responses to emotional stimuli in healthy humans. (For reviews, see Phan et al, 2002; Phan et al, 2004b). The relevance of these studies to fear/anxiety circuitry is two-fold: (1) A significant number of these emotional activation paradigms utilize stimuli that depict and/or elicit fear, and (2) these studies shed light on more general emotion-generating neurocircuitry. PET and fMRI studies have reported amygdala activation in response to emotionally negative photographs (Britton et al, 2006; Hariri et al, 2002; Irwin et al, 1996; Lane et al, 1997a; Paradiso et al, 1999; Phan et al, 2003b; Reiman et al, 1997; Taylor et al, 1998), odors (Zald and Pardo, 1997) and tastes (Zald et al, 1998). Several studies have reported amygdala activation to positive stimuli as well (Garavan et al, 2001; Hamann and Mao, 2002; Hamann et al, 1999, 2002; Liberzon et al, 2003), which suggests that the amygdala responds more broadly to emotionally arousing and/or salient stimuli (Phan et al, 2004b). Reappraisal of emotionally negative photographs is associated with reduced amygdala activation (Ochsner et al, 2002) and increased ventromedial prefrontal cortex activation (Urry et al, 2006). Finally, amygdala activation during encoding of emotionally arousing stimuli is correlated with the subsequent recollection of those stimuli (Cahill et al, 1996; Dolcos et al, 2004, 2005; Hamann et al, 1999).
Medial prefrontal cortex, including the rACC, also activates in response to emotional pictures (Lane et al, 1997a,b; Phan et al, 2003a,b, 2004a; Reiman et al, 1997) and may mediate self-referential processing (Kelley et al, 2002; Lane et al, 1997a; Zysset et al, 2002). Although the medial prefrontal cortex may activate regardless of task or valence, the rACC may be more likely to activate when a cognitive task is performed during scanning (Phan et al, 2002). Ventromedial PFC responses to fear-related images have been negatively associated with cortisol reactivity (Root et al, 2009). The dACC also activates in response to emotional photographs (Britton et al, 2006; Teasdale et al, 1999) and aversive tastes (Zald et al, 1998). Finally, the insular cortex is responsive to aversive stimuli (Phan et al, 2004a), internally generated sadness (Lane et al, 1997b; Reiman et al, 1997) and disgust-related stimuli (Britton et al, 2006).
Emotional facial expressions.
Interestingly, the same neurocircuitry that has been implicated in fear/anxiety responses in humans is readily activated by stimuli that are not intrinsically threatening, but may convey information regarding the presence of threat in the environment or about the fearful emotional state of others. Responses in the amygdala are readily elicited by photographs of facial expressions, especially those of fear (Breiter et al, 1996a; Davis and Whalen, 2001; Fitzgerald et al, 2006; Morris et al, 1996; Vuilleumier and Pourtois, 2007; Whalen et al, 2001), even when presented below conscious awareness (Morris et al, 1998; Whalen et al, 1998, 2004). Emotional facial expressions have also been associated with activation in the dACC, rACC, medial frontal gyrus, and insular cortex (Fitzgerald et al, 2006; Gorno-Tempini et al, 2001; Morris et al, 1996; Phillips et al, 1997, 2004; Sabatini et al, 2009; Sprengelmeyer et al, 1998).
Brain responses to the relatively ambiguous facial expression of surprise have been shown, in some studies, to depend on the extent to which individual subjects interpreted these expressions as positive or negative; more negative interpretations were associated with greater amygdala and lower ventral medial prefrontal cortex activation (Kim et al, 2003). These findings are consistent with the notion that the amygdala and medial prefrontal cortex are reciprocally modulated (eg, Garcia et al, 1999). Furthermore, the experimental manipulation of the context in which surprise facial expressions are presented alters brain activation patterns in a similar way: surprise expressions associated with a negative context elicited more amygdala activation than those associated with a positive context (Kim et al, 2004). These amygdala activations were positively correlated with activation in the dACC (Kim et al, 2004).
Of relevance to our later discussion of anxiety disorders are findings that suggest that healthy individuals with high scores on anxiety measures have greater amygdala and insular cortex responses to emotional (angry, fearful, and happy) faces and less rACC activation than participants with normative scores on these measures (Bishop et al, 2004a,b; Stein et al, 2007). Similarly, trait anxiety has been positively correlated with amygdala responses to neutral faces (Somerville et al, 2004).
Studies of fear conditioning, pharmacologically induced fear, and responses to emotional stimuli and facial expressions have provided evidence that the human amygdala, although responsive to multiple salient stimuli, responds reliably and potentially preferably to stimuli that predict threat and can be involved in mediating fear/anxiety states. Given that patients with anxiety disorders experience fear and distress in response to possible predictors of threat, the amygdala has been hypothesized to be hyperresponsive in some anxiety disorders. In the next section, we will review the evidence related to this hypothesis.
Studies of extinction have highlighted the potential involvement of the vmPFC and hippocampus in the process of learning and remembering that stimuli that used to predict threat no longer do. One possible reason for exaggerated fear, anxiety, and distress in patients with anxiety disorders is that these emotional responses fail to extinguish or that extinction learning is not recalled even when specific cues no longer predict threat. Indeed, some studies have reported impaired extinction in several anxiety disorders, such as PTSD (Blechert et al, 2007; Milad et al, 2008; Orr et al, 2000; Peri et al, 2000; Wessa and Flor, 2007). Thus, the vmPFC and hippocampus are clear regions of interest in functional neuroimaging studies of anxiety disorders.
Finally, both the animal literature and studies reviewed above suggest that the dACC (and its likely homolog prelimbic cortex) and insular cortex respond to emotional stimuli or those that predict threat. As with the amygdala, hippocampus, and vmPFC, these regions are involved in multiple other functions; however, they might also have important roles in specific aspects of anxiety. For example, the insular cortex is thought to mediate the monitoring of internal body states, and has been found to be hyperresponsive in anxiety-prone individuals (Paulus and Stein, 2006). In summary, research on healthy individuals has suggested that all of these brain regions are prime candidates to examine in patients with anxiety disorders.
An important and often overlooked aspect of the fear/anxiety neurocircuitry is its overlap and interaction with the neurocircuitry that orchestrates the stress response. It is important to note that the concept of ‘stress’ used here is relatively specific. It does not encompass general concepts of ‘subjective distress’ or ‘performance load.’ Although these are useful concepts, they are heterogeneous by nature and are not likely to be associated with a specific neurocircuitry. On the other hand, the concept of a stress system that leads to activation of limbic-hypothalamopituitary-adrenal axis (LHPA) and secretion of stress hormones like corticotropin-releasing hormone (CRH), adrenocorticotropic hormone, and cortisol is quite specific and is likely to be highly relevant to the neurocircuitry of fear and anxiety. The neurocircuitry governing LHPA activation has been the focus of intense studies in rodents, primates, and humans because it has been repeatedly linked to the neurobiology of mood disorders (which is addressed in detail elsewhere in this volume), but the evidence linking LHPA axis abnormalities to anxiety disorders has been less consistent, sometimes confusing, and often oversimplified. At the same time, some of the same brain regions are implicated in both anxiety and stress responses, suggesting that these responses are interrelated and can influence each other. Furthermore, anxiety and mood disorders are highly comorbid, suggesting that some common abnormalities in neurocircuitry might be present in both disorders. In the following few paragraphs, we briefly address only the structural overlap in neurocircuits and the effects of stress system activation (or stress hormones) on anxiety/fear neurocircuitry.
Epidemiologically, major depression is highly comorbid with anxiety disorders like PTSD, panic disorder, and social phobia (Reiger et al, 1990), and anxiety symptoms are highly prevalent in depression (Frances et al, 1990). Furthermore, major subcortical components of the LHPA axis (eg, hypothalamus, hippocampus, amygdala, and BNST) have also been identified as key components of anxiety/fear neurocircuitry (albeit sometimes involving different subnuclei, for example paraventricular nuclei vs ventromedial hypothalamus for LHPA and fear neurocircuitry, respectively). More recently with the introduction of in vivo imaging methodologies in LHPA/stress research, the role of cortical structures like the insula and dorsal mPFC in the activation and inhibition of stress response, respectively, has been reported (Liberzon and Martis, 2006) as well as the role of subgenual ACC in self-induced sadness and depression (Mayberg et al, 1999). Together, these findings suggest a significant overlap in structures involved in the stress response and those involved in fear/anxiety responses (eg medial prefrontal cortex, insula, amygdala, hippocampus, and BNST). Finally, with respect to neurotransmitters involved, CRH is likely involved in the orchestration of both LHPA axis activity and many anxiety/fear responses. (For a review see Heim and Nemeroff, 2001.)
The activation of these overlapping regions in functional neuroimaging studies does not necessarily signify, however, activation of both the fear/anxiety response and the LHPA axis. As a matter of fact, activation of fear/anxiety does not necessarily activate an LHPA stress response, even in highly fearful (phobic) individuals (Curtis et al, 1976). In turn, activation of LHPA axis is not necessarily experienced subjectively as fear or anxiety. For example, morning awakening, food intake, and nausea all lead to LHPA axis activation without notable increases in subjective sense of fear. It is becoming increasingly clear that specific characteristics of experience (novelty, control, social support, etc.) are more salient for LHPA axis activation than degree of subjective distress or fear (Abelson et al, 2007). These facts help to better understand the findings of non-specific, or even sometimes counterintuitive, findings regarding the LHPA and stress responses in anxiety disorders such as panic disorder (Abelson et al, 2007) and PTSD (Yehuda, 2006; Yehuda et al, 1991). This also suggests that activation in specific cortical regions like mPFC or insula cannot be readily interpreted as a component of the fear response, and has to be considered within a context of a specific experiment, subjective report, symptoms, neuroendocrine profile, etc.
With these caveats in mind, important findings about stress exposure and LHPA axis activation affecting fear/anxiety responses have been accumulating. These can be seen in two general categories: (1) the immediate effects of stress or of stress hormones on fear/anxiety responses (eg, stress or stress hormone exposure immediately precedes, or is present during the fear/anxiety responses), and (2) delayed or developmental effects, (eg, stress exposure during developmentally sensitive periods, like early childhood, modulates fear/anxiety responses later in life). In the former category, it has been reported that exposure to acute stress in healthy individuals potentiates the anxiety response (Grillon et al, 2007). In addition, stress exposure (Trier Social Stress Test) led to enhanced galvanic skin responses to conditioned stimuli (CS+) during fear conditioning (Jackson et al, 2006). Interestingly, stress modulates fear responses differentially in men and women. Differential effects of stress on fear/anxiety in females vs males also have been demonstrated in animal studies. Chronic stress exposure led to impaired extinction recall of fear conditioning in male but not female rats (Baran et al, 2009). Stress exposure in animal studies also led to enhancement in contextual fear conditioning (Cordero et al, 2003). The effects of stress hormone exposure are somewhat more difficult to interpret because higher endogenous cortisol levels were associated with higher skin conductance responses (SCR) (Jackson et al, 2006), whereas administration of exogenous cortisol led to decreased SCR (Stark et al, 2006).
With respect to delayed effects of stress during the vulnerable developmental period, the findings are somewhat complex. Studies of early maternal separation in rodents (Plotsky and Meaney, 1993) and variable foraging in primates (Coplan et al, 1996) have revealed long-term alteration in