The brain acid–base homeostasis and serotonin: A perspective on the use of carbon dioxide as human and rodent experimental model of panic

Highlights

• CO₂ inhalation is a well-validated human experimental model for panic attacks.

• The serotonergic system has an important role in chemosensitivity and panic disorder.

• The brainstem may be one of the key brain structures involved in panic disorder.

• In panic research, rodent studies and translational approaches are still scarce.

• Cross-species research can strongly improve translation between humans and rodents.

Abstract

Panic attacks (PAs), the core feature of panic disorder, represent a common phenomenon in the general adult population and are associated with a considerable decrease in quality of life and high health care costs. To date, the underlying pathophysiology of PAs is not well understood. A unique feature of PAs is that they represent a rare example of a psychopathological phenomenon that can be reliably modeled in the laboratory in panic disorder patients and healthy volunteers. The most effective techniques to experimentally trigger PAs are those that acutely disturb the acid–base homeostasis in the brain: inhalation of carbon dioxide (CO₂), hyperventilation, and lactate infusion. This review particularly focuses on the use of CO₂ inhalation in humans and rodents as an experimental model of panic. Besides highlighting the different methodological approaches, the cardio-respiratory and the endocrine responses to CO₂ inhalation are summarized. In addition, the relationships between CO₂ level, changes in brain pH, the serotonergic system, and adaptive physiological and behavioral responses to CO₂ exposure are presented. We aim to present an integrated psychological and neurobiological perspective. Remaining gaps in the literature and future perspectives are discussed.

Introduction

Panic attacks (PAs) are common psychopathological phenomena that affect about 23% of the general population at least once in their lifetime (Kessler et al., 2006). PAs represent abrupt surges of intense fear or discomfort, even though no real danger is present, accompanied by various physical or cognitive symptoms. According to the current criteria in the Diagnostic and Statistical Manual of Mental Disorders (DSM)-5 (American Psychiatric Association, 2013), at least four out of the following thirteen symptoms have to develop abruptly with a symptomatic peak within a few minutes after onset: palpitations or pounding heart; sweating; trembling or shaking; sensation of shortness of breath or smothering; feeling of choking; chest pain or discomfort; nausea or abdominal distress; feeling of dizziness, lightheadedness or faintness; depersonalization (feeling of being detached from oneself) or derealization (feeling of unreality); fear of losing control or going crazy; fear of dying; paresthesia (numbness or tingling sensations); and chills or hot flushes. As several symptoms closely resemble those of a cardiac arrest or acute asthma, cardiac- and/or emergency departments are frequently visited. Patients often receive costly tests such as angiography and echocardiography (Zaubler and Katon, 1998), without finding an explanation for their complaints.

PAs can occur in any anxiety or mental disorder as well as in many medical conditions (American Psychiatric Association, 2013), but are most prominent in panic disorder (PD). PD occurs in about 4% in the general population (Norton et al., 2008, Pane-Farre et al., 2014), with the onset commonly between the ages of 25–34 years in women and 30–44 years in men (Wittchen and Essau, 1993). PD has a high heritability of about 40% (Hettema et al., 2001, Maron et al., 2010) and is characterized by PAs that occur more than once and unexpectedly (i.e., ‘out of the blue’ and not caused by a medical condition or the use or withdrawal of a drug) (American Psychiatric Association, 2013). The frequency of the attacks can vary widely: a few attacks a month, several attacks each week or having periods with frequent attacks separated by weeks or months with less or no attack (Faravelli and Paionni, 2001). In addition to recurrent unexpected PAs, at least one of the following criteria is required for a period of at least one month: persistent concern about having additional attacks or the implications of the attack (anticipatory anxiety), and/or a significant maladaptive change in behavior related to the attacks. Frequently, patients develop agoraphobia, the avoidance of places and situations that are associated with the occurrence of previous attacks or in which having an attack may be embarrassing or in which it may be difficult to get help (for instance, being alone outside the home, being in a crowd or traveling in a bus or car). This avoidance behavior can become so severe that patients are confined to their homes. Due to the unpredictability of PAs, avoidance behavior, and the common comorbid anxiety disorders (Tilli et al., 2012), patients experience a marked decrease in their quality of life (Mendlowicz and Stein, 2000). Therefore, costs associated with an individual having PD are substantial (Salvador-Carulla et al., 1995). At the population level, the costs associated with PD are comparable with the combined costs associated with social phobia, simple phobia, and generalized anxiety disorder (GAD) (Batelaan et al., 2007).

According to the current DSM-5 (American Psychiatric Association, 2013), PD is classified as an anxiety disorder and is characterized by unexpected PAs (including intense fear) as well as anticipatory anxiety. Research has indicated that anxiety, fear, and panic are distinct entities involving divergent brain structures and behaviors. The main factor that determines the specific behavioral response is the ‘defensive distance’ to the threat (e.g., a predator). According to this concept, first introduced by Blanchard and Blanchard (1990), at a large defensive distance, i.e., in the presence of a potential threat, anxiety predominates and the individual displays non-defensive behavior such as complex risk assessment and approach behavior (defensive approach). In contrast, at a small defensive distance, i.e., in the direct presence of a threat, fear will be predominant, leading to a fast primitive response to avoid or leave the dangerous situation (i.e., fight-or-flight response, defensive avoidance).

Graeff (1994) mapped distinct neural structures to these concepts, primarily based on his research in small mammals. These structures were later integrated into a two-dimensional behavioral defense system model by McNaughton and Gray (2000) and McNaughton and Corr (2004). A large distance to a threat is associated with more rostral, cortical brain structures, including the prefrontal cortex, to evaluate the threat. The smaller the defensive distance, the more important is a fast, primitive response, which is linked to a caudal shift to subcortical, primordial structures. Extrapolating this concept to the clinic, a large defensive distance is associated with rostral neural structures and complex anxiety as seen in obsessive compulsions and in social anxiety. In contrast, with more proximal threats, a fear response is induced, which is associated with limbic and brainstem structures. The smallest defensive distance possible, which is from within a subject's own body, is related to PAs and is likely associated with primordial brainstem structures (Fig. 1).

Recently, this concept was confirmed in a seminal functional imaging study by Mobbs et al. (2007). Participants were chased in a virtual predator paradigm, in which they received either no electrical shock or a low or high intensity shock, when captured. Participants were informed about the condition before the trial. Results showed that as the distance to the predator shortened, brain activity shifted from the prefrontal cortex to the periaqueductal gray (PAG), particularly when a high intensity electrical shock was expected. In addition, PAG activity correlated with an increased degree of dread of being chased and decreased confidence of escape. In spite of the sophisticated approach, conceptually, this study did not model the smallest defensive distance possible and the within-body origin of PAs.

From a psychological point of view, there are at least two prominent theories regarding panic: the cognitive theory and the learning theory.

In cognitive theories such as Clark's model (1986) PAs are the result of the catastrophic misinterpretation of somatic sensations believed to signify an imminent harmful consequence. For instance, converging evidence exists for an increased interoceptive sensitivity to stimuli arising from the cardiac system in anxiety disorders (for comprehensive review see Domschke et al., 2010). The misinterpretation of harmless heart pounding as a heart attack can further intensify the symptoms and reinforce misinterpretation, resulting in a full-blown PA. As a consequence, the person becomes chronically hypervigilant for bodily sensations. This attentional bias toward somatic cues associated with the occurrence of a PA can make someone more prone to misinterpret somatic sensations, eventually leading to a vicious circle of misinterpretations culminating in PAs. Spontaneous catastrophic misinterpretations of bodily sensations are therefore assumed to play a major role in the transition of a PA into PD and are considered as a hallmark of PD. Several studies provide support for catastrophic misinterpretations of bodily sensations in PD, however, most of those studies are based on self-reports (Austin and Kiropoulos, 2008) that do not capture the spontaneous nature of the misinterpretations. In this view, indirect measures, such as priming paradigms, have been suggested to provide a good alternative. In such a priming paradigm, the response time can be measured to, for instance, word pairs that are either panic-related (e.g., breathlessness — suffocate; bodily PA symptom — catastrophic misinterpretation) or neutral (e.g., flowers — picking). However, the results of those studies have been inconsistent (Hermans et al., 2010, Schneider and Schulte, 2007). In a study from our group (Hermans et al., 2010), significant priming effects were not only observed in PD patients, but also in a group of mental health professionals without a history of PD. This observation raises doubts about the pathogenic nature of catastrophic misinterpretations. In addition, an important critique on cognitive theory is the observation that PAs also occur in situations in which no catastrophic misinterpretation can be identified, such as during sleep (Craske and Tsao, 2005, Freire et al., 2007, Mellman and Uhde, 1989, Sarisoy et al., 2008). Importantly, there is currently no experimental support for the hypothesized causal role of catastrophic misinterpretations in PD (De Cort et al., 2013).

In contrast to the cognitive theory that mainly explains the transition of a PA into PD, the learning theory is the first that attempts to explain the unexpected ‘non-clinical’ PAs in the general population. It is based on the proposition that individuals can ‘learn’ to develop certain symptoms. An early theory by Barlow (1988) is known as the ‘alarm theory’ and distinguishes three types of alarms: true, false, and learned. A true alarm is an unconditioned fear reaction to an actual threat, whereas a false alarm is an alarm that occurs in the absence of a real threat. PAs can be true or false alarms, depending on the actual presence of a threat. The occurrence of an alarm, whether true or false, can become associated with an external or internal cue. The individuals ‘learn’ that a specific cue is connected to the alarm and become fearful of that cue because it may be a predictor for the onset of another alarm (i.e., a PA). Consequently, they may develop anxiety and avoidance behavior regarding future attacks. In a modern learning theory account, Bouton et al. (2001) further specified that a false, learned alarm is not necessarily pathological. The key process between experiencing a PA and the progression to PD is the development of anxiety that is focused on the occurrence of the next potential PA. In this respect, it is hypothesized that interoceptive conditioning is of particular importance in the pathogenesis of PD. For instance, a bodily stimulus that preceded the onset of a previous PA (e.g., slight increase in heart rate) becomes associated with the occurrence of the PA, thereby becoming a learned stimulus that predicts intense arousal and induces anxiety. Experienced anxiety, in turn, generates more intense and additional stimuli (e.g., a further increase in heart rate, sweating), which, in turn, intensifies anxiety. This spiraling process can lead to a panic state. To date, a role of learning mechanisms in PD is supported by various experimental studies. Support is found for enhanced resistance to extinction (Michael et al., 2007), impaired discriminatory learning, and overgeneralization in PD patients (Lissek et al., 2009, Lissek et al., 2010) during fear conditioning paradigms. Recently, neural substrates were provided for alterations in PD patients during fear conditioning. Activations were found in the dorsal inferior frontal gyrus during differential conditioning and in the midbrain during safety signal processing. These results were interpreted as altered top-down and bottom-up processes, reflecting changes in risk assessment/behavioral inhibition and defensive reactivity, respectively (Lueken et al., 2014). Further, increased baseline activations were found in structures such as hippocampus and amygdala in patients that did not respond to cognitive behavioral therapy. Treatment response was related to an intact inhibitory functional anterior cingulate cortex-amygdala coupling (Lueken et al., 2013). A limitation of the current studies is that they employ external cues, which do not reflect the within-body nature of the sensations that are associated with PAs. Despite their clear relevance, interoceptive conditioning studies in the framework of PD are rather scarce. Interoceptive conditioning has recently been established in healthy volunteers through respiratory loads/breathing occlusions (Pappens et al., 2012), and inhalations of 20% (Acheson et al., 2007) or 35% CO₂ (De Cort et al., 2012). Yet again, there is a paucity of interoceptive studies with PD patients or persons at risk for developing PD. Recently, we (De Cort et al., 2014) applied a differential interoceptive conditioning paradigm using pentagastrin and 35% CO₂ in subjects scoring low or high on anxiety sensitivity, a risk factor for PD (Naragon-Gainey, 2010). Anxiety sensitivity refers to, partially heritable (Stein et al., 1999), inter-individual differences in anxiety-related sensations that are associated with harmful consequences (McNally, 2002, Reiss, 1997). Discriminatory learning was impaired in subjects with high anxiety sensitivity, causing them to overpredict danger. Therefore, the progression from the initial insult toward the development of PD may be due to impairment in discriminatory learning and overgeneralization rather than excessive conditioning.

PAs can be studied in two ways, i.e., by means of ambulatory assessment of naturally occurring PAs, or by provoking the sensations from within the body using experimental stimuli. Given the spontaneous occurrence of PA, which makes ambulatory assessment challenging, the possibility to trigger PAs in the lab has been tested using a large number of different techniques (for review see Esquivel et al., 2008b). These include CO₂ exposure; hyperventilation; or administration of lactate, cholecystokinin (CCK, a peptide hormone activating CCK-B receptors), flumazenil (a benzodiazepine receptor antagonist), caffeine (an adenosine receptor antagonist), agents affecting the serotonergic (5-HT) system such as fenfluramine (acute 5-HT enhancer) and metachlorophenylpiperazine (m-CPP, primarily a 5-HT_2B/2C receptor agonist), and agents with effects on the noradrenergic system such as yohimbine (an alpha2-antagonist) and isoproterenol (a beta-agonist).

Five criteria have been proposed that an ideal experimental model should fulfill (Guttmacher et al., 1983). First, safety is an essential consideration in any experiment involving humans. The evoked symptoms should be temporary and readily reversible, and without any predictable health risks. Second, there should be convergence (i.e., the induced symptoms should reflect the real-life ones). Third, the model should be specific and should differentiate between those with and without the disease (discrimination). Fourth, the effects should be replicable when consecutive trials are performed. Fifth, after a clinically effective treatment (e.g., drugs or cognitive behavioral therapy) individuals should respond markedly less than individuals who did not receive that treatment (clinical validation). Based on these criteria, to date, there are three models that are relatively well validated: CO₂ exposure, hyperventilation, and lactate administration. All three have in common that they alter the pH within the body.

We have previously proposed that an acutely disturbed acid–base homeostasis, involving an entire network within the brain, underlies the occurrence of PAs (Esquivel et al., 2009). Under normal circumstances, the acid–base homeostasis is tightly regulated within a narrow range around a pH of 7.4. A shift out of this normal range can have a severe impact on chemical reactions and enzymatic functions, with potentially fatal consequences. Therefore, sensing a change in pH and triggering adaptive responses are of pivotal importance for survival. Neurons sensitive to CO₂/H⁺ and thus to changes in pH have been found in many brain regions such as the amygdala (Ziemann et al., 2009), the hypothalamus (Johnson et al., 2012a, Williams et al., 2007), and various brainstem nuclei (Biancardi et al., 2008, Dean et al., 1990, Dias et al., 2007, Mulkey et al., 2004, Putnam et al., 2004, Richerson, 2004, Severson et al., 2003). Combined with the defensive distance model proposed by McNaughton and Gray (2000) and McNaughton and Corr (2004) and the functional imaging predator study by Mobbs et al. (2007), these chemosensitive brainstem regions would be of particular interest in the pathophysiology of PAs. The only missing component to validate the brainstem model in humans was evidence of a threat stimulus originating from within the body, which would represent the most proximal defensive distance possible. We recently provided this missing component by applying CO₂ inhalations to human subjects (Goossens et al., 2014), showing that breathing CO₂ resulted in brainstem activation in PD patients in comparison with healthy controls (Fig. 2).

To unravel the molecular mechanisms underlying such brainstem activation as well as the properties of neurons that sense changes in pH, the number of rodent studies on panic increased over the past few years, as this concept is difficult to study in humans. Important progress has been made with regard to the PAG (Johnson et al., 2014, Paul et al., 2014), the raphe nuclei (Richerson, 2004, Severson et al., 2003), and the role of the 5-HT system (Richerson, 2004). As the role of the PAG in relationship to panic has been elaborated extensively elsewhere (Johnson et al., 2014, Paul et al., 2014), it will not be discussed in this review.

In the present review, we focus on CO₂ as panic-provoking method in both humans and animals to discuss the relationship between CO₂, changes in brain pH, the 5-HT system, and associated adaptive physiological and behavioral responses.