The neural, behavioral, and epidemiological underpinnings of comorbid alcohol use disorder and post-traumatic stress disorder

Abstract

Alcohol use disorder (AUD) and (PTSD) frequently co-occur and individuals suffering from this dual diagnosis often exhibit increased symptom severity and poorer treatment outcomes than those with only one of these diseases. Although there have been significant advances in our understanding of the neurobiological mechanisms underlying each of these disorders, the neural underpinnings of the comorbid condition remain poorly understood. This chapter summarizes recent epidemiological findings on comorbid AUD and PTSD, with a focus on vulnerable populations, the temporal relationship between these disorders, and the clinical consequences associated with the dual diagnosis. We then review animal models of the comorbid condition and emerging human and non-human animal research that is beginning to identify maladaptive neural changes common to both disorders, primarily involving functional changes in brain reward and stress networks. We end by proposing a neural framework, based on the emerging field of affective valence encoding, that may better explain the epidemiological and neural findings on AUD and PTSD.

1. Introduction

The co-occurrence of alcohol use disorder (AUD) and post-traumatic stress disorder (PTSD) represents one of the most common neuropsychiatric dual diagnoses (Smith & Cottler, 2018). Individuals afflicted with both disorders frequently experience increased symptom severity and poorer treatment outcomes than those diagnosed with only one of these illnesses. Despite the major global socio-economic and public health impact of comorbid AUD and PTSD, our understanding of the neural substrates responsible for the frequent co-occurrence of these disorders is relatively limited and effective treatments are lacking. The goal of this article is to review the epidemiology and major risk factors associated with comorbid AUD and PTSD. We then discuss animal models of these disorders and outline the maladaptive changes in key neural circuits and networks that may contribute to this dual diagnosis, centered primarily on data from human functional imaging of brain reward and stress networks. Finally, we outline a more holistic neural framework for the comorbid condition based on emerging findings from both human imaging work and non-human animal circuit-mapping research.

2. The diagnosis of AUD and PTSD

AUD and PTSD are clinically defined by the Diagnostic and Statistical Manual of Mental Disorders (DSM) and the International Classification of Disease (ICD) (American Psychiatric Association, 2013; World Health Organization, 2018). In both classifications, AUD is defined by the presence of a series of diagnostic criteria or guidelines focused primarily on harmful drinking patterns, loss of control over drinking behavior, and the emergence of a negative affective state during withdrawal, with symptoms persisting for at least 12 months. The DSM-V describes AUD as a single condition that varies in severity based on the number of criteria exhibited (mild, moderate, severe). Similarly, the ICD-11 differentiates alcohol dependence (three or more symptom clusters) from harmful patterns of alcohol use (less than three symptom clusters).

With regard to PTSD, both classification systems require acute and/or chronic trauma exposure and also employ a core symptom cluster-based diagnostic approach. These latest iterations of both diagnostic systems have taken steps to distinguish PTSD from other affective disorders. In fact, the DSM-V no longer classifies PTSD as an anxiety disorder but rather incorporates it into a new class of trauma and stressor-related disorders that also includes Adjustment Disorders. There are now 20 PTSD symptoms contained within four core categories: (1) intrusion or re-experiencing, (2) avoidant symptoms, (3) negative alterations in mood or cognition, and (4) alterations in arousal and reactivity. Similarly, the ICD11 includes PTSD in a category for “Disorders specifically associated with stress.” In this system, PTSD is diagnosed by three symptom clusters including (1) re-experiencing of the trauma, (2) avoidance of trauma reminders, and (3) a persistent sense of current threat that manifests as hypervigilance or enhanced startle reactivity. The ICD11 also includes a complex PTSD diagnosis with three additional symptom clusters that reflect “disturbances in self-organization; (1) affect dysregulation, (2) negative self-concept and (3) disturbances in relationships.”

Although there are important differences in the symptom clusters used to diagnose AUD and PTSD, it is hard to overlook important similarities in the behavioral phenotypes associated with both disorders. For example, many of the symptoms used to diagnose AUD are centered around alcohol itself (e.g., loss of control over alcohol intake, persistent and unsuccessful attempts at reducing alcohol consumption, tolerance to behavioral and cognitive effects of alcohol) whereas with PTSD, exposure to a traumatic or stressful event (e.g., death, threatened death, actual/threatened serious injury of sexual violence) is a requisite criterion. However, both disorders are strongly associated with the emergence of negative affective mood and behaviors, a profound decrease in natural interest in social, occupational, and/or recreational activities, as well as major alterations in normal sleep patterns.

AUD is often considered a chronic, relapsing disorder characterized by significant fluctuations in symptom severity over the course of the illness (NIAAA, 2020). While PTSD is generally not viewed in a similar manner, it is worth noting that individuals with this disorder often experience a similar variation in symptoms over the course of their recovery. For example, a 5-year longitudinal study of PTSD in primary care patients found that the likelihood of PTSD recovery was 38% but that symptoms reoccurred in almost 30% of these individuals (Perez Benitez et al., 2012). There may also be coordinated fluctuations in the symptoms of both disorders. One study found that, in a cohort of individuals with the comorbid condition receiving outpatient treatment with sertraline, AUD symptoms improved before or simultaneously with PTSD symptoms (Back, Brady, Sonne, & Verduin, 2006). Similarly, an inpatient study with Vietnam veterans being treated for PTSD found that the onset of alcohol use coincided with the development of PTSD symptoms and the alcohol consumption levels increased as did PTSD symptoms (Bremner, Southwick, Darnell, & Charney, 1996). If one begins to consider the severity of the stress experienced by many individuals living with AUD, the overlap in both the symptoms of these disorders, and their frequent recurrence over the trajectory of these illnesses, becomes easier to understand.

3. Epidemiology of comorbid AUD and PTSD in the general population

Despite the fact that the diagnostic criteria for AUD and PTSD have evolved considerably, and that there are significant criterion differences between the DSM and ICD, epidemiological evidence for the frequent comorbidity between these two disorders has remained robust. Although clinicians have anecdotally recognized the frequent co-occurrence of AUD and PTSD for many years, it was only after the Vietnam War and the recognition of PTSD as a unique affective disorder, that the frequency of the comorbid condition began to be formally documented. In the United States, three major surveys have tracked data on neuropsychiatric illnesses in the general population, the Epidemiological Catchment Area (ECA) program, the National Comorbidity Survey (NCS), and the National Epidemiological Survey on Alcohol and Related Conditions (NESARC) and all three have provided data highlighting the frequent co-occurrence of AUD and PTSD. For example, some of the earliest evidence linking AUD and PTSD came from an analysis of ECA data, finding that a DSM-III diagnosis of PTSD was associated with a near twofold increase in risk of developing alcoholism in men and an almost threefold increase in women (Helzer, Robins, & McEvoy, 1987). Similarly, a large review of NCS data that also used DSM-III diagnostic criteria found that a diagnosis of PTSD was associated with a two- to threefold increase in risk of developing DSM-III alcohol abuse or dependence (Kessler, Sonnega, Bromet, Hughes, & Nelson, 1995). More recent studies, using data from the NESARC also found significant, albeit more modest, associations between lifetime PTSD and DSM-IV alcohol abuse/alcohol dependence (1.7 odds ratio) (Pietrzak, Goldstein, Southwick, & Grant, 2011) and DSM-V AUD (1.2 odds ratio) (Goldstein et al., 2016).

Despite major differences in how AUD and PTSD are diagnosed around the world, the striking frequency of this dual diagnosis has been reported in numerous countries. For instance, in Europe, where the prevalence of PTSD in the general population is in the range of 0.6–7% (Burri & Maercker, 2014), these numbers jump to 15–25% in treatment-seeking alcohol dependent patients (Preuss et al., 2018). Similarly, in Australia, where lifetime prevalence of PTSD is around 4–5%, one study reported that 34% of individuals with PTSD were also diagnosed with a substance use disorder, most commonly AUD (24%) (Mills, Teesson, Ross, & Peters, 2006). Even in Brazil, where the overall PTSD rate is already relatively high (8–9%) and likely underestimated (da Silva et al., 2019), the prevalence of PTSD is in excess of 30% in those with an AUD diagnosis (Castillo-Carniglia, Keyes, Hasin, & Cerda, 2019). Collectively, these epidemiological studies provide compelling evidence for the frequent comorbidity between AUD and PTSD in the United States and many other countries.

4. Epidemiology of AUD and PTSD in vulnerable populations

4.1. Military and veterans

Although the incidence of AUD and PTSD in the general population is high, there are certain populations that are particularly vulnerable to develop both disorders. One such group are members of the military and veterans. Indeed, as already noted, clinical recognition of PTSD and the comorbid condition emerged from epidemiological data collected on veterans of the Vietnam War. It is now well established that the prevalence of comorbid AUD and PTSD is significantly higher among members of the military, relative to civilians, and therefore it is not surprising that the comorbid condition is also more prevalent in military personnel (Dworkin, Bergman, Walton, Walker, & Kaysen, 2018; Richardson et al., 2017; Stein et al., 2017). For example, a recent study on mental disorders in a representative sample of National Guard members found that AUD was the most common disorder in this cohort, with a lifetime prevalence of 44% (Fink et al., 2016) and another found almost the same incidence in US veterans (Fuehrlein et al., 2016). With regard to PTSD, one study that examined the 30-day prevalence of DSM-IV mental disorders among non-deployed US soldiers found that 15% met criteria for an affective disorder, with PTSD being the most common (Kessler et al., 2014). Another study on past year prevalence of mental disorders in members of the Canadian Armed Forces reported a similar 16.5% had at least one of six disorders examined, with major depressive disorder and PTSD being most common. This study also reported a significant increase in the incidence of PTSD and other anxiety disorders between 2002 and 2013 in this population (Zamorski et al., 2016). When looking at the comorbid condition among members of the military and veterans, the data are particularly striking. In a study on veterans of the Iraq and Afghanistan wars, 63% of individuals with AUD also met criteria for PTSD (Seal et al., 2011). More recent data from the National Health and Resilience in Veterans Study reported that approximately 15% of this cohort had probable current AUD or PTSD. Of those with AUD, 20% met criteria for PTSD and for those with PTSD, approximately 17% also met criteria for AUD (Norman, Haller, Hamblen, Southwick, & Pietrzak, 2018). Another recent study on new U.S. recruits found that binge drinking, heavy drinking, and AUD increased the odds of meeting criteria for all mental disorders, with lifetime prevalence of co-occurring PTSD being one of the highest (23%) (Stein et al., 2017).

4.2. Early-life trauma and stress

Another demographic group that is highly vulnerable to AUD, PTSD, and their co-occurrence, are individuals exposed to early life stressors (ELS) such as sexual, physical, and emotional abuse or neglect (Nemeroff, 2016). Over the past three decades, a vast body of epidemiological data has shown an almost dose-response like relationship between the number of childhood/adolescent adverse life events and the likelihood of developing AUD and/or other affective disorders, including PTSD (Dube, Anda, Felitti, Edwards, & Croft, 2002; Felitti et al., 1998; Kaysen, Rosen, Bowman, & Resick, 2010; Khoury, Tang, Bradley, Cubells, & Ressler, 2010; Weber et al., 2008). This relationship is particularly disconcerting given the very high prevalence of ELS. For example, a large epidemiological study found that almost 65% of people in the United States experience at least one, and upward of 12% experience as many as four ELS events (Middlebrooks & Audage, 2008). The vulnerability for developing these disorders conferred by ELS likely stems from the fact that the nervous system undergoes profound developmental changes not only during childhood but throughout adolescence and well into early adulthood. Exposure to trauma and stress during these critical periods can alter normal neuronal maturation in key stress and reward circuits (Arain et al., 2013; Gogtay et al., 2004), leading to a marked increase in risk of developing AUD and a wide range of neuropsychiatric disorders, including PTSD (Nemeroff, 2016).

As just a few striking examples of the strong association between ELS and comorbid AUD/PTSD, one study on children removed from their parents’ care due to allegations of abuse or neglect (average age: 12.5 years) reported that PTSD was the most common neuropsychiatric diagnosis at baseline, with 50% meeting criteria for this disorder. Moreover, early onset alcohol use is a major predictor of AUD and almost 30% of maltreated children reported using alcohol, a rate sevenfold higher than demographically matched controls (Kaufman et al., 2007). Indeed, another study of 587 patients recruited from an urban primary care population found a very high rate of alcohol dependence in this cohort (39%) and notably, the number of childhood traumas experienced was significantly correlated with alcohol use and the severity of PTSD symptoms (Khoury et al., 2010).

Although prospective studies documenting the impact of ELS on AUD and PTSD are less common, data from the 10-year longitudinal Youth Emotion Project revealed that the severity of adolescent (but not childhood) ELS exposure predicted poorer health outcomes and onset of substance use disorders (including alcohol) in late adolescence/early adulthood (Wolitzky-Taylor et al., 2017). Interestingly, the relationship between adolescent ELS and substance use disorder in this study was uniquely driven by emotional abuse symptoms and not those associated with sexual or physical abuse or loss/separation of caregiver. Finally, data from a large population-based birth cohort study in Australia found that approximately 10% of participants had a history of substantiated ELS and that both internalizing and externalizing disorders were strongly associated with ELS in adolescence (Mills et al., 2013) and early adulthood (Kisely et al., 2018), with PTSD being the most strongly associated affective disorder.

4.3. Impact of gender

There is an extensive and growing body of literature suggesting that gender is an important moderating factor in the comorbid condition. In the general U.S. population, being female is associated with a twofold increase in the risk of developing PTSD (Goldstein et al., 2016; Kilpatrick et al., 2013; Pietrzak et al., 2011) whereas being male increases the likelihood of an AUD diagnosis by a similar margin (Grant et al., 2015). Reasons for these prevalence differences are complex and likely involve both biological and sociological factors. For example, the impact of gender on the prevalence of AUD is likely influenced by the fact that men typically consume more alcohol than women (Wilsnack, Wilsnack, Kristjanson, Vogeltanz-Holm, & Gmel, 2009). Interestingly, although it is often suggested that the higher prevalence of PTSD in women can be attributed to gender differences in the likelihood of being exposed to traumatic events, one recent study that sought to control for this variable actually concluded that women may also be more vulnerable to the neurological consequences of traumatic events (Blanco et al., 2018). Other data suggest that women may also be more sensitive to the deleterious effects of alcohol on the brain (although evidence is mixed) (Verplaetse, Cosgrove, Tanabe, & McKee, 2020), a finding that is particularly disconcerting given that the longstanding gender gap in alcohol consumption (Grant et al., 2017; Keyes, Grant, & Hasin, 2008; McHugh, Votaw, Sugarman, & Greenfield, 2018) and AUD (Grant et al., 2017) appears to be shrinking. Together, these studies reveal important gender differences in comorbid PTSD and AUD. The complexity of these associations highlights the need for further studies as a deeper understanding of the neurobiology underlying these differences will likely have significant implications with regard to the development of effective treatments for these disorders in women and men.

5. Which comes first?

It is important to understand the temporal relationship between PTSD and AUD as such knowledge may provide mechanistic insight into the underlying causes of the comorbid condition. Four models have been proposed to explain the functional associations between these disorders (Straus, Haller, Lyons, & Norman, 2018). The Self-Medication model posits that PTSD precedes AUD and that pathological alcohol use develops in an attempt to mitigate the negative affective symptoms of PTSD. The High Risk and Susceptibility models propose that excessive alcohol use can increase the likelihood of experiencing traumatic events or exacerbate emotional and/or neurobiological consequences associated with experiencing a traumatic event. Finally, the Shared Vulnerability model argues that both disorders arise as a result of common biological and/or environmental risk factors. In general, the majority of epidemiological studies suggest that PTSD most often develops before AUD, and, not surprisingly, there is a large body of evidence in support of the Self-Medication model (see Gilpin & Weiner, 2017; Straus et al., 2018). For example, a comprehensive review of large cross-sectional, prospective, and survival analysis datasets found consistent evidence that pre-existing PTSD increased the risk of drug abuse/dependence with little evidence in support of the “High Risk” hypothesis (Chilcoat & Breslau, 1998). A more recent study that sought to evaluate all four hypotheses using longitudinal data from a community sample also found the strongest support for the Self-Medication hypothesis, with PTSD symptoms being the strongest predictor of higher levels of alcohol and drug problems, when compared with pre-trauma family risk factors, pre-trauma substance use problems, trauma exposure, and demographic variables (Haller & Chassin, 2014). This study also found no support for the Susceptibility or Shared Vulnerability models but did show that adolescent substance use problems had a modest but significant effect on risk of assaultive violence exposure, but not overall risk for trauma, partially supporting the High Risk hypothesis. Finally, a study using data from the NESARC found that PTSD may be the anxiety disorder with the highest rates of self-medication with either alcohol or other drugs and that men were twice as likely to seek out alcohol or drugs to alleviate their PTSD symptoms (Leeies, Pagura, Sareen, & Bolton, 2010).

One final comment on the temporal patterns associated with this dual diagnosis is that far fewer studies have actually sought out evidence for the high risk or susceptibility models in which AUD precedes PTSD. One recent study that also used NESARC data and directly addressed the order of onset between these disorders actually found that the pathway from PTSD to AUD was significantly stronger than the reverse, but only in women (Berenz et al., 2017). Moreover, another NESARC study that used wave I and II data to conduct what may be the first longitudinal investigation for the direct effects of past-year AUD on the development of PTSD, found that prior alcohol dependence significantly increased risk of trauma exposure and the development of PTSD, relative to matched controls (Balachandran, Cohen, Le Foll, Rehm, & Hassan, 2020).

6. Clinical consequences of comorbid AUD and PTSD

Although many factors influence the temporal relationship between AUD and PTSD, the physical and psychological burden associated with this dual diagnosis is significantly greater than that of either disorder alone. To that end, a large number of studies that have compared individuals with either AUD or PTSD alone with subjects diagnosed with both disorders have found that the latter group often experiences greater symptom severity of both disorders (Dworkin, Wanklyn, Stasiewicz, & Coffey, 2018; Longo et al., 2020; Somohano, Rehder, Dingle, Shank, & Bowen, 2019), exhibits higher rates of other psychiatric comorbidities (e.g., bipolar disorder, depression, anxiety, schizophrenia) (Bowe & Rosenheck, 2015; Norman et al., 2018; Norman, Tate, Anderson, & Brown, 2007; Read, Brown, & Kahler, 2004; Terhaag et al., 2019), suffers greater physical health and psychosocial problems (Bowe & Rosenheck, 2015; Hoge, Terhakopian, Castro, Messer, & Engel, 2007; Ouimette, Goodwin, & Brown, 2006; Tate, Norman, McQuaid, & Brown, 2007), as well as higher rates of homelessness and poor life conditions (Bowe & Rosenheck, 2015; Najavits, Weiss, & Shaw, 1999). For example, a study of a nationally representative sample of U.S. veterans with AUD, PTSD, or the comorbid condition found that, relative to those with AUD alone, the dual diagnosis cohort scored significantly lower on measures of cognitive, mental and physical functioning as well as quality of life (Norman et al., 2018). The comorbid group also had a 5–10-fold increased incidence of major depression, generalized anxiety disorder, suicidal ideation and suicide attempts. Indeed, many other studies have reported very high levels of suicidal ideation, suicide attempts and increased mortality among the comorbid population (Afzali, Sunderland, Batterham, Carragher, & Slade, 2017; Kachadourian, Gandelman, Ralevski, & Petrakis, 2018; Shorter, Hsieh, & Kosten, 2015; Stefanovics & Rosenheck, 2019) and, although literature on gender differences is sparse, the dual diagnosis may represent an even greater risk factor for suicide attempts in women (Gradus et al., 2017). Not surprisingly, there is also extensive evidence that the dual diagnosis has a profoundly negative impact on treatment and recovery (Flanagan, Jones, Jarnecke, & Back, 2018; Kingston, Marel, & Mills, 2017; Verplaetse, McKee, & Petrakis, 2018).

7. Animal models of comorbid AUD and PTSD

The preceding epidemiological review clearly indicates that comorbid AUD and PTSD represents a significant global public health concern. The co-occurrence of these disorders is high in many countries and there is compelling evidence that this dual diagnosis represents a major treatment challenge. Despite overwhelming data highlighting the clinical significance of this dual diagnosis, there has been surprisingly little human research directed at the neurobiological mechanisms responsible for the comorbid condition. As reviewed in subsequent sections of this chapter, there is a rich literature on the neural mechanisms that are thought to contribute to each disorder alone. However, studies on the neurobiology underlying the comorbid condition in human subjects is sparse. To address this gap in our knowledge, a number of preclinical animal models have been developed that engender symptoms of both disorders. To provide a background for a discussion of some of the neurobiological findings obtained using these paradigms, in this section we provide a brief description of the more commonly used models. Although numerous animal regimens have been developed to model each disorder (for recent reviews see: Deslauriers, Toth, Der-Avakian, & Risbrough, 2018; Flandreau & Toth, 2018; Goltseker, Hopf, & Barak, 2019; Kenny, Hoyer, & Koob, 2018; Pinna, 2019; Richter-Levin, Stork, & Schmidt, 2019; Torok, Sipos, Pivac, & Zelena, 2019), here we focus exclusively on studies that have explicitly sought to model the comorbid condition. All of these models involve some element of stress-induced escalation of alcohol consumption. Briefly, subjects are exposed to some form of stress, such as predator odor (Edwards et al., 2013) or adolescent social isolation (Locci & Pinna, 2019), among others. Then, alcohol intake in stress-exposed animals can be compared to control subjects. Alternatively, the animals may be trained to drink prior to stressor presentation, to monitor for individual changes.

The efficacy of an animal model is based on its adherence to certain metrics of validity (Willner, 1986). Predictive validity refers to the propensity of the dependent variable of the model to predict translational outcomes; in other words, using animal performance to predict human outcomes. These predicable outcomes may be treatments or behavioral or physiological markers of the disorder. Face validity refers to the extent to which the model appears to measure what it purports to. For example, a model of AUD would achieve face validity if it were capable of producing behavioral signs of dependence, like tolerance, withdrawal, craving, or binging. Finally, construct validity involves the relationship between the model and the theory underlying its conception, namely the extent to which a given model represents the construct it is intended to. This representation should be both empirical and theoretical. For example, if traumatic experience increases alcohol use in the human population, a translational model with construct validity would have to demonstrate a similar pattern in animals. Furthermore, the underlying cause of the change in alcohol drinking in this example should be homologous. In other words, if the traumatic condition in the animal model were prolonged water restriction, a subsequent increase in drinking cannot be attributed without doubt to trauma, because thirst is also likely a cause. Opinions differ as to which of these criteria are primary, or even necessary, for a viable animal model (Geyer & Markou, 1995; Koob, Heinrichs, & Britton, 1998; Robbins, 1998; Sarter & Bruno, 2002).

In terms of creating and using valid animal models of PTSD, it is crucial to consider the differences between modeling stress and modeling trauma. Though Generalized Anxiety Disorder (GAD) and PTSD do frequently co-occur in humans and are both highly comorbid with AUD, they are distinct diagnoses. The aforementioned stress-induced drinking models involve acute, low level stressors and may produce a phenotype more consistent with GAD than PTSD. Indeed, GAD may be a risk factor for the development of PTSD following exposure to a traumatic event. However, symptoms that differentiate PTSD involve re-experiencing of the specific trauma, which includes flashbacks, avoidance of related stimuli, and hyper rousal. Models of PTSD are characterized by high intensity stressors and often aim to mirror the conditions in which traumatic events occur in the human population.

Every animal model that aims to represent the comorbid condition of AUD and PTSD necessitates application of a stressor and access to alcohol. A multitude of stressors have been utilized, simulating the variety of events in the human population which may lead to a PTSD diagnosis. In animal models, the stressors may be physical or non-physical. A physical stressor entails physiological threat like shock, restraint, or forced submersion in water. Non-physical stressors includes both social/psychological and pharmacological, where psychological involves manipulation of an external factor, like social isolation or vicarious trauma, and pharmacological involves the induction of autonomic stress reactivity with a drug compound, like yohimbine (Lamb, 1979; Verbitsky, Dopfel, & Zhang, 2020). The vast breadth of these models highlights an especially important consideration: it may not be feasible to use a single model to reliably reproduce a human condition that can be induced by qualitatively distinct stressors. Importantly, though no existing model captures the full symptomatology of both disorders, those reviewed here are capable of reliably producing escalation in alcohol intake as well as at least two core PTSD symptoms.

One early translational model of PTSD is the single prolonged stress (SPS) procedure, which has been performed in many iterations and involves a series of different assortments and schedules of stressor application in a single day (Dal-Zotto, Marti, & Armario, 2003; Lisieski, Eagle, Conti, Liberzon, & Perrine, 2018; Richter-Levin, 1998). Briefly, SPS involves a single period of intense stress exposure, often in the form of multiple physical stressors like restraint, forced swim, and foot shock, exposure to ether until loss of consciousness. The amalgamation of multiple, varied stressors occurring over a relatively short (less than a day) period of time is intended to simulate a translationally relevant traumatic event. Such models do reliably produce behavioral symptoms associated with PTSD, including reduced fear extinction, enhanced startle response, hyperarousal, and heightened anxiety-like behavior (mood alteration), as well as an escalation of ethanol consumption indicative of AUD (Knox et al., 2012; Lisieski et al., 2018; Souza, Noble, & McIntyre, 2017).

Similar models, involving varied schedules of repeated stress have also been performed: these include chronic variable stress (Lopez, Doremus-Fitzwater, & Becker, 2011; McGuire, Herman, Horn, Sallee, & Sah, 2010; Rompala, Simons, Kihle, & Homanics, 2018) and intermittent unpredictable stress (Peay et al., 2020; Sequeira-Cordero, Salas-Bastos, Fornaguera, & Brenes, 2019). These models capitalize on mercurial stressor application, in order to simulate a context in which the stressor may appear at any time. If the stressor is unpredictable, it can neither be avoided nor adapted to, which leaves the animal in a state of hypervigilance or prolonged anxiety, which may disrupt extinction of conditioned fear. Another related model that reliably produces alterations in fear learning is aptly referred to as Stress-Induced Enhancement of Fear Learning (SEFL) (Rajbhandari, Gonzalez, & Fanselow, 2018). In this procedure, prior exposure to repeated footshock is sufficient to produce fear generalization. In other words, animals that have been subjected to electric footshock in the past are more sensitive to similar, but less intense stressors at a later time point. Translationally, this phenotype tracks with resurgence of symptoms in response to related, but nonspecific stressors as well as enhanced acquisition of novel fears in patients with PTSD. Interestingly, this model increases voluntary alcohol drinking in alcohol-naïve rats but not in those with a prior history of alcohol self-administration (Meyer, Long, Fanselow, & Spigelman, 2013).

Given the prevalence of PTSD in combat-exposed populations, some models have sought to more directly mimic the conditions of war. One recent model employed a combination of physical and psychological stressors, including repeated footshock, distressing auditory tones, sleep deprivation and forced exercise to represent the onslaught of stressors experienced by soldiers (Kim, Kim, Koo, Heo, & Cheon, 2017). This model did increase anxiety-like behavior, as well as producing meaningful alterations in cognitive and immune functions. However, because exposed animals displayed no alteration of hypothalamic-pituitary-adrenal (HPA) function, the authors reasoned that the model may not fully represent PTSD. Yet, further validation of this model may prove fruitful, given its robust behavioral effects, despite the lack of histopathological changes. If future studies were to demonstrate either a subsequent increase in voluntary alcohol intake or alterations in reward/stress circuitry, the combat-exposure animal model may have significant construct validity for the comorbid condition, given that HPA-specific changes in PTSD are often inconclusive (de Kloet et al., 2006).

Another popular model involves exposing animal subjects to the odor of a natural predator in order to simulate a life threatening situation, without causing actual harm to the animal (Edwards et al., 2013; Manjoch et al., 2016). In such models, adult rodents will be trained to self-administer alcohol prior to exposure to predator odor (commonly bobcat, feline, or fox urine). Subjects are then categorized by their immediate response to the given stressor and once again allowed to consume alcohol. Rodents that displayed high reactivity to the predator odor display behavioral phenotypes associated with alcohol-dependence, including escalated drinking post-stress, compulsive-like responding, and aversion resistant drinking. This model is unique in its replication of the avoidance symptom of PTSD.

As alluded to previously, some animal models of the comorbid condition have focused more so on socially-derived stressors that may act as a correlate for early life adversity, parental neglect, and/or psychosocial trauma. One such model is adolescent social isolation, during which rodents are separated early after weaning (typically PD 28) into two conditions: socially isolated animals housed singly in standard cages and group housed animals, housed 3–4 in larger, guinea pig cages for 3–6 weeks. Following this housing manipulation, socially isolated rats display enduring increases in alcohol intake and preference, increased anxiety-like behavior, and notably, impaired fear extinction (Skelly, Chappell, Carter, & Weiner, 2015) compared to their group housed counterparts. Importantly, this manipulation only produces the aforementioned phenotypes when performed during adolescence and in males; animals that reach adulthood before being separated do not develop AUD/PTSD-like symptoms, nor do female rodents (Butler, Carter, & Weiner, 2014). Because this housing manipulation leads to the development of both AUD- and PTSD-like symptoms, it has been suggested as a viable model of the comorbid condition (Butler, Karkhanis, Jones, & Weiner, 2016; Locci & Pinna, 2019; Pinna, 2019). This procedure may model early life adversity or neglect, which have been explored as a potential cause or correlate of PTSD and AUD in human populations.

Other social stressor models rely not on the anxiogenic effect of isolation, but on the propensity of negative social interaction to produce trauma. Indeed, psychosocial stress in various forms has been linked to PTSD. Social defeat paradigms are commonly used as models of depressive-like behavior. However, they may also produce behavioral phenotypes that encapsulate the depression, anxiety, and social avoidance symptomology of PTSD and have gained popularity in this field (Ishikawa, Uchida, Kitaoka, Furuyashiki, & Kida, 2019; Sial, Warren, Alcantara, Parise, & Bolanos-Guzman, 2016; Verbitsky et al., 2020). In resident-intruder (RIP) procedures, a male “intruder” rodent is introduced to the homecage of an older, dominant “resident” rodent. The resident acts aggressively, in defense of the homecage, by attacking the intruder. In RIP models, the intruder animal is assessed for symptoms of trauma exposure. Some variations have been assessed in recent years. In a study of adolescent animals, Long-Evans residents were socially defeated by Sprague Dawley intruders (Manz, Levine, Seckler, Iskander, & Reich, 2018). This finding, a reversal of the classic outcome in which the resident is socially dominant, prompted the use of reverse RIP (rRIP) as a model of adolescent stress exposure or bullying. Another variation involves assessment not of the intruder animal, but of a spectator animal (Sial et al., 2016). This model, termed “vicarious” social defeat, seeks to characterize the impact of witnessing trauma. A major limitation for social defeat models is sex specificity; resident-intruder pairings typically involve animals of the same sex. Aggression is differentially expressed in male and female rodents, females tending to attack less frequently than males. Furthermore, males tend not to attack female intruders as readily as other males, which limits the available data on the impact of social defeat on behavior in females (Verbitsky et al., 2020). However, one model had adolescent rodents as the spectator of social defeat of their mother (Liu, Patki, Salvi, Kelly, & Salim, 2018) and did see that both male and female pups developed depressive-like behaviors after exposure to vicarious maternal trauma. All of the social defeat procedures are capable of producing PTSD-like symptoms in the target rodent, including heightened anxiety-like behavior, social avoidance, and alterations of stress-related biomarkers including corticosterone (CORT) as well as increased in alcohol self-administration (Deal, Konstantopoulos, Weiner, & Budygin, 2018; Macedo et al., 2018; Newman, Albrechet-Souza, et al., 2018).

Finally, a necessary component of any model for comorbid PTSD/AUD is an escalation of alcohol drinking-related behaviors. Several of the models discussed above have produced considerable effects on varied aspects of alcohol-directed behavior. The adolescent social isolation model produces a robust drinking phenotype in males (Butler et al., 2016; McCool & Chappell, 2009). SI animals consume more alcohol during intermittent two bottle choice procedures, where bottles containing 10% ethanol or water are offered side by side in the home cage for 12h periods three times per week. The aSI manipulation also produces behavioral changes in operant alcohol self administration. Importantly, aSI animals display a higher level of both appetitive (seeking) and consummatory (drinking) behavior compared to GH rats. Importantly, this effect is alcohol-specific; responding is not similarly altered toward sucrose. Predator odor also engenders vulnerability to drinking in a subset of animals, which is representative of the clinical condition in that some but not all individuals will be affected. Specifically, animals that avoided the odor paired chamber (Avoiders) exhibited prolonged increases in operant responding and compulsive-like responding for alcohol compared to non-Avoiders (Edwards et al., 2013).

A critical consideration in the stress-induction models discussed above is whether they actually succeed in traumatizing the animal. As discussed previously, a diagnostic criterion for PTSD is direct or indirect exposure to a traumatic event leading to symptoms that persist for at least 1-month. Thus, enduring changes in core PTSD-like phenotypes may be the most viable indicator that animals have experienced actual trauma. Ethologically relevant stressors, like predator odor, are perhaps the most appropriate given these criteria. Social defeat paradigms have successfully produced behavioral alteration up to a month following exposure. While some animal models of the comorbid condition, like adolescent social isolation, engender escalations in alcohol drinking and anxiety-like behaviors lasting up to 8 weeks (Skelly et al., 2015; Yorgason, Espana, Konstantopoulos, Weiner, & Jones, 2013), it is not clear if all of the symptoms that manifest early after the housing manipulation last as long. Additionally, paradigms like adolescent social isolation, that focus on stress exposure during specific developmental periods, may not model PTSD that results from trauma exposure during adulthood. An ideal model would involve exposure to a stressor that is capable of producing AUD- and PTSD-like symptoms when applied at any time during the life course. Finally, in an ideal model, only a subset of the animals exposed to the traumatic stimulus or event should manifest symptoms associated with PTSD. This criterion is met in models like the predator odor procedure in which only animals that avoid odor-paired stimuli go on to develop AUD/PTSD-like phenotypes (Edwards et al., 2013).

8. Neurobiological substrates of comorbid AUD and PTSD

As reviewed above, there is extensive epidemiological evidence documenting both the frequent co-occurrence of AUD and PTSD as well as the many negative clinical consequences associated with this dual diagnosis. This evidence has led to the hypothesis that these disorders may involve maladaptive changes in common neural substrates. Surprisingly, relatively few studies have directly examined the neurobiology of the comorbid condition. However, there is a rich literature on the biological basis of each disorder (e.g., Fenster, Lebois, Ressler, & Suh, 2018; Fritz, Klawonn, & Zahr, 2019; Koob, 2014; Koob & Volkow, 2016; Malikowska-Racia & Salat, 2019; Yehuda et al., 2015). Indeed, the available evidence strongly supports the notion that many of the behavioral phenotypes associated with both disorders manifest as a result of a profound dysregulation of brain networks that process reward and stress. In this section, we review recent findings from human studies that highlight the maladaptive effects of AUD and PTSD on these canonical brain reward and stress circuits, with an emphasis on the impact of these illnesses on measures of functional connectivity. We also review findings from preclinical studies using animal models of the comorbid condition that provide convergent evidence that alterations in reward and stress pathways may play a causal role in the AUD/PTSD-like phenotypes promoted by these models. We then advance a novel conceptual framework for the comorbid condition that borrows concepts from the emerging field of affective valence processing. This framework proposes that reward and stress circuits may be viewed as a functionally integrated valence network that assigns some degree of positivity or negativity to all sensory and emotional stimuli, encodes the relative salience or value of this information, and then orchestrates appropriate behavioral responding to these stimuli. Using this model, we propose that AUD and PTSD be viewed as examples of valence network imbalance disorders and outline how this heuristic may provide a better framework for interpreting the epidemiological and neurobiological findings on the comorbid condition. Readers interested in reviews focused on more traditional anatomical, structural, and functional adaptations and treatments for the comorbid condition are encouraged to read these other recent reviews (Blaine & Sinha, 2017; Dunlop & Wong, 2019; Fritz et al., 2019; Gilpin & Weiner, 2017; Goode & Maren, 2019; Suh & Ressler, 2018).

9. Reward circuitry

Many studies over the years have identified neural systems that comprise a “brain reward” circuitry, with the dopaminergic (DAergic) projection from the ventral tegmental area (VTA) to the nucleus accumbens (NAc) receiving the most attention (Fig. 1). Indeed, it is now well established that acute exposure to all drugs of abuse, including alcohol, engages the VTA-NAc pathway to increase striatal DA levels and that activation of this pathway reinforces drug/alcohol taking behaviors (Cooper, Robison, & Mazei-Robison, 2017; Kim, Ham, Hong, Moon, & Im, 2016; Volkow & Morales, 2015). Moreover, as reviewed below, dysregulation of this pathway, and many other elements of the reward network, play an integral role in the development and progression of AUD and PTSD.

Fig. 1.

Brain reward and stress circuitry in the human brain. Diagrammatic illustration of the predominant brain regions and circuits generally though to mediate the processing of reward (green) and stress/fear (red) in the human brain. The nucleus accumbens (NAc) is usually featured as the central node of reward circuit diagrams, which also typically include the DAergic input from the VTA and excitatory input from other limbic structures, like the anterior hippocampus (aHPC). The stress circuit is generally centered around the basolateral AMY (BLA) which has dense, and often reciprocal connectivity with numerous cortical structures including the medial prefrontal cortex, anterior cingulate cortex, insula, the thalamus and aHC as well as nodes of the extended amygdala (CEA, BNST). Illustrated by dashed lines are some of the many additional connections that link elements of both the reward and stress networks that are often omitted from these circuit diagrams.

9.1. Reward circuit and AUD—Human studies

Numerous imaging studies have demonstrated that AUD is associated with profound alterations in VTA-NAc DA signaling. Early PET studies from Volkow and colleagues reported that individuals diagnosed with AUD had lower D2 receptor availability in the striatum, relative to healthy controls (Volkow et al., 1996), a finding also observed in people with SUDs (Volkow et al., 1990; Wang et al., 1997). These findings, coupled with evidence of decreased methamphetamine-stimulated striatal DA release in AUD (Volkow et al., 2007), led to the hypothesis that AUD was associated with a hypoDAergic state and decreased function of the brain reward circuitry (although see Hirth et al., 2016 for a contrarian view).

While it remains well-established that AUD dampens mesolimbic DA signaling, and that this maladaptive change contributes to the negative affective symptoms of this disorder, more recent work using functional imaging techniques, like fMRI, suggests that AUD has more complex effects on the canonical VTA-NAc DA pathway and many other elements of the reward circuitry. For example, several studies have shown that, relative to healthy controls, individuals with AUD respond more to alcohol-associated cues in the ventral striatum and other reward-related areas like the amygdala (AMY), anterior cingulate cortex (ACC), and medial prefrontal cortex (mPFC) (Huang, Mohan, De Ridder, Sunaert, & Vanneste, 2018; Schacht, Anton, & Myrick, 2013; Vollstadt-Klein et al., 2012). One recent study found that binge drinkers with a heightened risk of developing AUD display heightened NAc BOLD responses to monetary rewards (Crane et al., 2017). These findings are consistent with the incentive salience theory of addiction which suggests that repeated drug exposure actually sensitizes the brain reward system to drug-associated cues, leading to increased craving and relapse (Robinson & Berridge, 1993). However, other fMRI studies have shown that positive cue-evoked BOLD responses in the ventral striatum are lower in individuals with AUD, particularly for non-alcohol-related cues (Fritz et al., 2019). For example, one study found that, while visual alcohol-associated cue reactivity in the NAc was elevated in AUD subjects, relative to controls, the AUD cohort actually had lower NAc BOLD responses to non-alcohol-related monetary cues (Wrase et al., 2007). Notably, both of these changes were correlated with alcohol craving in the AUD group but not the controls.

Over the past decade or so, many groups have begun to use functional imaging to go beyond measurements of task-relate activation of discrete brain regions to examine alterations in functional connectivity across complex brain networks. Most early work focused on task-free assessments of resting state functional connectivity and these studies identified a distributed default mode network (DMN) that was preferentially activated in the resting state. Although there is no universally accepted map of the DMN, numerous regions are consistently identified and can be separated into three interconnected subnetworks: a midline core DMN, a medial temporal DMN, and a dorsal medial prefrontal DMN (Zhang & Volkow, 2019). It is now appreciated that, while these DMN networks are active at rest, they can also be engaged by a variety of tasks, included some that are related to emotional processing. Importantly, the DMN is integrally associated with brain reward circuitry and many studies have now shown that key elements of the DMN are disrupted in AUD. Although a detailed review of this literature is beyond the scope of this chapter (please see Zhang & Volkow, 2019 for a comprehensive recent review), AUD has been shown to be associated with a shift in DMN connectivity from higher order structures (e.g., posterior cingulate cortex to middle cingulate cortex, (MCC)) to more cue-reactive circuits (e.g., increased connectivity between midbrain and MCC). Such changes could contribute to alterations in reward processing for alcohol and non-alcohol-related cues.

Interestingly a few recent studies have explicitly examined NAc connectivity in “at risk” populations and AUD subjects. Given that early onset drinking is a risk factor for AUD, one study explored the relationship between NAc resting state functional connectivity (RSFC) and lifetime drinking behavior in 199 healthy non-alcohol dependent male adolescents and asked whether this measure could predict alcohol consumption during a 1-year follow-up (Veer et al., 2019). The authors reported that greater past alcohol consumption, particularly during the most recent year, was associated with a modest but significant weakening of functional connectivity between the left NAc and dlPFC and bilateral IFF, extending into the left dorsal anterior insula. Most notably, lower connectivity between the NAc and these cortical regions predicted alcohol drinking behavior during the 1-year follow-up. Another study examined NAc RSFC in cohorts of adolescents with a family history of AUD (FHP), another well-established risk factor for this disorder (Dawson, Harford, & Grant, 1992; Sher, 1991), with a control group of family history negative individuals (FHN). Numerous group differences in resting state synchrony were observed, including decreased segregation between the NAc and fronto-parietal networks as well as less integration of the right NAc and left OFC, both changes that may be related to impaired reward processing (Cservenka, Casimo, Fair, & Nagel, 2014). Finally, another group combined resting state EEG and task-based fMRI to examine neural correlates of craving in the NAc and other elements of the reward circuit in a cohort of 11 AUD patients 24h into abstinence (Huang et al., 2018). This study found convergent evidence, using both imaging techniques, of a “craving network” characterized by increased BOLD activation in response to visual alcohol cues in the NAc, AMY, dACC, pgACC, PCC and parahippocampus as well as increased beta-band activity in the dACC and pgACC in resting-state EEG. These changes were accompanied by hyperconnectivity in resting-state EEG functional connectivity and dysconnectivity in cue-based fMRI functional connectivity across this densely connected network of alcohol-related brain areas. It will be important, in future studies, to conduct similar assessments in healthy controls and AUD patients in long-term abstinence to gain a better understanding of the specific role of these measures of network connectivity in the etiology of the disorder.

9.2. Reward circuit and PTSD—Human studies

As with AUD, negative alterations in mood and anhedonia are core features of PTSD (Eskelund, Karstoft, & Andersen, 2018; Fenster et al., 2018; Vujanovic, Wardle, Smith, & Berenz, 2017). For example, symptoms of anhedonia are observed in approximately two-thirds of PTSD patients, independent of comorbid major depressive disorder (Carmassi et al., 2014; Franklin & Zimmerman, 2001). Moreover, a recent systematic review of 29 studies comparing measures of reward functioning in healthy controls and individuals with PTSD found that most reported decreases in reward anticipation, motivation to obtain rewards, and hedonic responding to reward in PTSD patients (Nawijn et al., 2015). Despite these findings, PET imaging studies of the DAergic system in PTSD are surprisingly scarce. A meta-analysis of PET/SPECT data from individuals with OCD, GAD, phobia, and PTSD found a significant decrease in D2 receptor binding in the ventral striatum, similar to that observed in individuals with AUD (Nikolaus, Antke, Beu, & Muller, 2010). One other study used SPECT to assess striatal DAT levels in asymptomatic trauma-exposed subjects and PTSD patients who had been exposed to urban violence. It reported significantly greater DAT binding density in both the left and right striatum in the PTSD cohort (Hoexter et al., 2012). The authors suggested that this change may reflect a compensation for increased dopamine (DA) signaling in fear circuits. However, an alternative interpretation is that higher striatal DAT may lead to lower DA levels which could contribute to the anhedonia and negative mood associated with this disorder.

Although direct evidence of DA receptor changes in PTSD is limited, numerous functional imaging studies have demonstrated that dysregulation of brain reward circuitry is a central feature of this disease. One early fMRI study examined decision-making in trauma-exposed women with PTSD and healthy controls using a reward processing task in which subjects could maximize their gains by learning an advantageous response pattern (Sailer et al., 2008). While both groups increased their performance and reduced their reaction time across this task, controls improved faster and performed better in the late phase, relative to PTSD patients. These behavioral differences were associated with a significant decrease in activation of the NAc and PFC during the late phase of the task in the PTSD cohort. Another study examined BOLD activation of reward circuit responsivity using a Wheel-of-Fortune-like monetary reward task in a cohort of 28 civilians diagnosed with PTSD and mentally healthy controls (Elman et al., 2009). Using both an unbiased voxel-wise analysis and region-of-interest approach, PTSD subjects exhibited smaller bilateral striatal activation during the passive experiencing of monetary gains vs. losses. Moreover, in the PTSD group, striatal activation was negatively correlated with self-reported motivational and social deficits. Interestingly, other reward-related regions, like the AMY, pallidum, insula, cingulate, orbitofrontal, and frontal cortices, were all activated during the task in control subjects but not those with PTSD. These findings suggest not only a blunting but a narrowing of the elements of the reward circuitry in individuals with PTSD. It should be noted that most, if not all, of these brain regions are also involved in the processing of stressful and aversive stimuli. To that end, one recent study sought to examine the intriguing possibility that PTSD-associated deficits in reward processing may involve over-recruitment of reward-related brain regions by aversive stimuli, making them less available to respond to stimuli with positive valence (Elman et al., 2018). In one experiment, PTSD patients and healthy controls were exposed to pleasant, aversive, or neutral visual stimuli. Individuals with PTSD exhibited decreased subjective positive ratings of pleasant images and decreased activation of the striatum and several cortical regions. Interestingly, although no group differences in subjective ratings of unpleasant visual stimuli were observed, PTSD subjects showed decreased activation in the AMY and thalamus. In a second experiment, the same subjects were exposed to a pain-inducing thermal stimulus. Although PTSD patients and controls reported similar subjective pain ratings in response to this aversive sensory stimulus, this stimulus was associated with increased activation of the striatum, AMY, HC, and mPFC in PTSD patients. These findings reveal that PTSD may be associated with decreased activation of the striatum and other areas that process rewarding (and aversive) psychosocial stimuli but that these same brain regions may be hyper-responsive to aversive sensory stimuli.

As observed in patients with AUD, there is considerable evidence that PTSD has profound effects on brain-wide network connectivity. While most of these studies have focused on network connectivity and fear/stress-related symptoms of PTSD, some alterations in functional connectivity likely impact reward processing and may contribute to the anhedonic symptoms of PTSD. There is a large circuitry called the salience network (SN) which is comprised of numerous cortical and subcortical structures that serve to integrate emotional and sensory input, particularly in the context of threatening stimuli (Peters, Dunlop, & Downar, 2016; Szeszko & Yehuda, 2019). However, many of the brain regions that comprise the SN are also known to play an integral role in reward processing, such as the insula, the AMY, the dorsal ACC, and the OFC (Koch et al., 2016; Kunimatsu, Yasaka, Akai, Kunimatsu, & Abe, 2020) as well as the striatum and brainstem DAergic nuclei (Peters et al., 2016). Although the SN has received relatively little attention in the AUD field (Peters et al., 2016; Zhang & Volkow, 2019), numerous studies have examined intrinsic and extrinsic changes between DMN and SN connectivity in PTSD patients (for reviews see: Koch et al., 2016; Kunimatsu et al., 2020; Peters et al., 2016; Szeszko & Yehuda, 2019). One of the most consistent findings has been a decrease in intrinsic connectivity within nodes of the DMN in PTSD (Akiki et al., 2018; Bluhm et al., 2009; Chen & Etkin, 2013; DiGangi et al., 2016; Sripada, King, Garfinkel, et al., 2012). In contrast, most studies have observed increased intrinsic SN connectivity in PTSD patients between the insula and both the AMY and HC (Chen & Etkin, 2013; Nicholson et al., 2015; Rabinak et al., 2011; Sripada, King, Welsh, et al., 2012). Although, both increases (Brown et al., 2014) and decreases (Sripada, King, Garfinkel, et al., 2012) in connectivity have been reported between the AMY and dACC. Interestingly, increased cross-network connectivity has been seen in combat veterans with PTSD (Sripada, King, Welsh, et al., 2012) but decreased coupling was observed in PTSD patients with early-life trauma (Bluhm et al., 2009), indicating that the timing and nature of the traumatic event(s) that trigger PTSD may have a profound impact on how this disorder disrupts brain-wide functional connectivity. Nevertheless, these findings seem to suggest that PTSD may lead to a relative dominance of the SN over the DMN in the resting state. While this maladaptive change likely plays an integral role in the hypervigilance and hyperarousal symptoms of PTSD, it may also contribute to the anhedonia and dysregulation of reward processing that is so prevalent in this disorder.

9.3. Reward circuits—Preclinical animal studies

Although human data on the neurobiology of comorbid AUD and PTSD are sparse, findings in animal models that generate symptoms associated with the comorbid condition lend support to the notion that both disorders result in profound dysregulation of key nodes of the reward circuitry. As observed in human studies that have characterized each disorder alone, there is convergent evidence that animal models that promote symptoms of both AUD and PTSD can lead to a sensitization of phasic mesolimbic DA release, particularly in response to alcohol and other drugs of abuse. These adaptations would be expected to enhance the reinforcing properties of these drugs, particularly during the early stages of self-administration. For example, male Sprague Dawley rats exposed to predator odor for a single 15min session that go on to develop PTSD-like behaviors (e.g., increased anxiety measures on the elevated plus-maze, odor context avoidance) exhibit greater cocaine-stimulated NAc DA release than control subjects as well as odor-exposed rats that did not develop PTSD-like phenotypes (Brodnik et al., 2017). Moreover, susceptible odor-exposed rats also showed greater motoric responses to cocaine and heightened motivation to self-administer this drug, relative to the resilient and control groups. Similarly, another study in male Sprague Dawley rats found that repeated social defeat stress led to an enduring NMDA receptor-dependent LTP of glutamatergic synaptic transmission onto VTA DA cells (Stelly, Pomrenze, Cook, & Morikawa, 2016). They went on to show that this stress-induced plasticity was mediated by a PKA-dependent enhancement in the potency of inositol 1,4,5-triphosphate-induced Ca²⁺ signaling in VTA DA cells. Moreover, blockade of glucocorticoids with mifepristone mitigated these neurobiological adaptations as well as a sensitization of cocaine conditioned place preference that developed after the social defeat paradigm. Another study from our group used ex vivo, fast-scan cyclic voltammetry to evaluate DA release dynamics in DA terminal fields in the NAc of Long-Evans rats (Deal et al., 2018). We found that a relatively mild (3-day) social defeat procedure led to a significant increase in the magnitude of electrically evoked DA transients as well as an increase in the rate of DA uptake. Interestingly, both of these effects were occluded in a separate cohort of rats with a history of intermittent ethanol self-administration, possibly because the drinking procedure may have already engaged the stress systems underlying these changes in DA release dynamics. Finally, a fascinating study revealed that VTA DA neurons in mice susceptible to social defeat stress show increased intrinsic excitability relative to controls, due to an increase in the hyperpolarization-activated cation channel (I_h) conductance (Friedman et al., 2014). Surprisingly, when stress-resilient mice were examined, these subjects had an even strong I_h current. Through an elegant series of experiments, the authors discovered that, in resilient mice, the elevated levels of I_h drove a compensatory increase in an inhibitory potassium channel conductance that normalized VTA DA cell firing. Remarkably, pharmacologically enhancing I_h or optogenetically stimulating DA cells in susceptible mice reversed the maladaptive behavioral phenotypes promoted by this model.

We have also conducted a series of studies characterizing adaptations in the mesolimbic DA system using an adolescent social isolation model of the comorbid condition. As observed following social defeat, relative to rats that were group housed during adolescence, we have consistently observed increases in measures of DA release and uptake in socially isolated rats in adulthood. These changes have been observed in the NAc core, NAc shell, and dorsal striatum and are accompanied by an increase in the expression of the DAT (Karkhanis et al., 2019; Yorgason et al., 2016). Using in vivo microdialysis, we have also reported that adolescent social isolation increases ethanol-stimulated DA release in the NAc (Karkhanis, Locke, McCool, Weiner, & Jones, 2014) and BLA (Karkhanis, Alexander, McCool, Weiner, & Jones, 2015). Notably, we also examined NE release in response to acute ethanol in these studies. Surprisingly, acute ethanol administration had no effect on NE release in either the NAc or BLA but in socially isolated rats, this same treatment resulted in a robust increase in NE release in both brain regions (Karkhanis et al., 2014, 2015).

In more recent work, we have sought to explore possible mechanisms underlying these dramatic effects of adolescent social isolation on the mesolimbic DA system. In the NAc, kappa opioid receptors (KORs) are expressed on presynaptic terminals of DAergic synapses and activation of these receptors inhibits DA release (Svingos, Chavkin, Colago, & Pickel, 2001; Werling, Frattali, Portoghese, Takemori, & Cox, 1988). Since a variety of stressors increase KOR expression in the NAc and KOR system dysregulation has been implicated in the etiology of AUD and PTSD (Karkhanis & Al-Hasani, 2020; Van’t Veer & Carlezon Jr, 2013), we examined the effect of adolescent social isolation on KOR signaling in this brain region (Karkhanis, Rose, Weiner, & Jones, 2016). We confirmed that adolescent social isolation led to an increase in both electrically-evoked and ethanol-stimulated DA release and also discovered that, despite this sensitization of DA release dynamics, baseline levels of DA were lower in these animals. This decrease in ambient DA is similar to the hypoDAergic state observed in individuals with AUD and likely contributes to the negative affective state promoted by this model of the comorbid condition (Butler et al., 2016). These changes were accompanied by a significant increase in KOR-mediated inhibition of DA release and pretreatment with a KOR antagonist, normalized baseline DA levels, blocked the increase in ethanol-evoked DA release, and reduced the escalation in alcohol intake observed in socially isolated animals. Collectively, these findings suggest that increased mesolimbic KOR signaling may contribute to the DAergic dysregulation and maladaptive behaviors promoted by this model of the comorbid condition.

10. Stress circuitry

The other major brain system that underlies many of the symptoms of AUD and PTSD is a complex network of cortical and subcortical structures that translates stress-associated stimuli into appropriate behavioral and emotional responses to stressors. Key nodes of this stress circuitry include the hypothalamus as well as the PFC, the basolateral AMY, the ventral hippocampus, the locus coeruleus, and the extended AMY (e.g., central AMY, BNST, nucleus accumbens; Flook et al., 2020; Koob, 2008; Koob & Volkow, 2010) (Fig. 1). An astute reader will immediately appreciate that there is considerable overlap between the core regions of the stress and reward circuits, a critical issue that will be addressed in the final section of this chapter. While we have already discussed how AUD and PTSD are associated with reward-related dysregulation of some of these brain regions (e.g., NAc, AMY, PFC, ACC), in this section, we focus on functional human imaging studies that have examined these, and other key nodes of this circuitry in the context of stress/fear-related emotional processing. A growing literature has revealed that many brain regions associated with aberrant reward processing in AUD and PTSD are also dysregulated during the processing of negative/aversive stimuli and data from animal models of the comorbid condition suggests that these pathways may play a causal role in many of the behavioral phenotypes associated with both disorders (Gilpin & Weiner, 2017; Suh & Ressler, 2018; Whitaker, Gilpin, & Edwards, 2014).

The lateral and basolateral nuclei of the AMY (BLA) are arguably the central hub of the stress network. The BLA is largely comprised of excitatory pyramidal neurons which are the primary target of all sensory input into the amygdala complex, which mostly originates from the thalamus and cortex (Davis & Whelan, 2001; Paxinos, 2003). These cells in turn project back to many cortical regions as well as numerous elements of the extended AMY and reciprocal connections between these brain regions are thought to encode our behavioral and emotional responses to sensory and emotional stimuli (Janak & Tye, 2015; Sharp, 2017; Zhang et al., 2018). The BLA also receives strong glucocorticoid input and serves as a key interface between the hypothalamic-pituitary-adrenal axis and the stress network (Duvarci & Pare, 2007; Nathan, Griffith, McReynolds, Hahn, & Roozendaal, 2004; Roozendaal, Griffith, Buranday, De Quervain, & McGaugh, 2003; Segev & Akirav, 2016; Stringfield, Higginbotham, & Fuchs, 2016; Taslimi, Sarihi, & Haghparast, 2018). It should also be noted that the afferent and efferent circuitry of the BLA are highly conserved across species (Lanuza, Belekhova, Martinez-Marcos, Font, & Martinez-Garcia, 1998; LeDoux, 2012; Moreno & Gonzalez, 2007; Pabba, 2013). Moreover, the BLA is the largest subregion of the AMY complex, and its relative contribution to the overall size of this structure, both in terms of volume and number of neurons, has increased more than other major AMY nuclei across evolution (Avino et al., 2018; Barger et al., 2012; Chareyron, Banta Lavenex, Amaral, & Lavenex, 2011). For example, the volume of the BLA is 30–40 times larger in monkeys than rats whereas the central and medial nuclei are only 4–8 times larger in monkeys (Chareyron et al., 2011). The fact that the BLA is the primary site of sensory input into the AMY complex and that it represents that largest AMY subregion are important to keep in mind while reviewing human AMY functional imaging studies. While most of these studies do not distinguish between AMY nuclei, given its relatively large size and that it receives the majority of afferent input into the AMY complex, it seems reasonable to assume that the BLA is the predominant contributor to AMY BOLD signals.

10.1. Stress circuitry and AUD—Human studies

Although the stress network plays an integral role in many behavioral and emotional processes, it is most frequently studied in the context of fear and anxiety. For the purposes of this review, fear is defined as an adaptive emotional response to a real or perceived threat that typically dissipates once the threat is no longer present whereas anxiety is a more temporally diffuse, negative emotional state, often evoking arousal and vigilance, that manifests in anticipation of these stressors (Davis, Walker, Miles, & Grillon, 2010; Dias, Banerjee, Goodman, & Ressler, 2013). One of the most influential conceptual frameworks for the neurobiology of AUD posits that, while activation of the canonical reward circuitry likely drives the early stages of this disorder, the stress network plays an increasingly important role in its latter stages (Koob, 2013a, 2013b; Koob et al., 2014). The idea is that, during the initial stages of AUD, alcohol is consumed primarily for its rewarding or pleasurable effects and that abstinence periods are generally associated with a neutral affective state. However, after repeated periods of binge intoxication and withdrawal, a negative affective state develops during abstinence and it is relief from this negative affective state that becomes the primary driver of the maladaptive drinking behaviors associated with AUD. While anhedonia is a key element of this aversive state, excessive anxiety is also an integral component. This anxiety is often linked with craving and together these affective states represent a major trigger for relapse (Huang et al., 2018; Karch et al., 2015; Seo et al., 2013; Sinha, 2012; Tanabe, Regner, Sakai, Martinez, & Gowin, 2019). Not surprisingly, numerous human imaging studies have demonstrated that AUD leads to maladaptive alterations in many elements of the stress circuitry. Moreover, more recent human and animal research has revealed that strikingly similar maladaptive changes in the stress network are associated with relapse in AUD and deficits in fear extinction, a hallmark phenotype of PTSD (Goode & Maren, 2019; Peters, Kalivas, & Quirk, 2009).

Over the past decade, high resolution functional imaging studies have demonstrated that AMY hyperactivity during the processing of anxiogenic or fearful stimuli is a highly reproducible core feature of anxiety and stressor-related disorders (Etkin, Prater, Hoeft, Menon, & Schatzberg, 2010; Etkin & Wager, 2007; Nitschke et al., 2009). While data on AUD and AMY activity has been more sparse, there is now a growing body of evidence revealing that similar alterations may be linked with AUD. One of the first of these studies used fMRI to compare AMY reactivity in individuals with AUD and healthy controls in response to positively and negatively arousing images and found that the AUD cohort showed greater overall AMY activation to negative vs. positive stimuli. While AMY reactivity was greater in AUD subjects, this was the case for both positive and negative images (Gilman & Hommer, 2008). Other early functional imaging studies found mixed evidence of a positive link between AMY hyperactivity and AUD. For example, one study concluded that low AMY responsiveness may be a risk factor for AUD, as they reported decreased threat-related AMY reactivity in individuals at high familial risk for AUD (Glahn, Lovallo, & Fox, 2007) whereas another study found that AMY hyperactivity may be protective against stress-related problem drinking in at least some university students (Nikolova & Hariri, 2012). Despite these seemingly disparate findings, recent evidence suggests that both hypo- and hyperreactivity of the AMY may be linked with AUD, and that these contrasting findings may be explained by examining the balance between overlapping elements of the reward and stress networks.

For example, one large study of 759 undergraduate students used fMRI to image individual differences in reward and threat processing and examined whether these measures predicted stress-related problem drinking (Nikolova, Knodt, Radtke, & Hariri, 2016). Surprisingly, they found that stress-related problem drinking was associated with one of two neural phenotypes: relatively high reward-related ventral striatum (VS) activation and relatively low threat-related AMY activation or relatively low reward-related VS activation and high threat-associated AMY activation. Importantly, they also showed that the relationship between stress and problem drinking was mediated by impulsivity (steeper delayed discounting curve) for individuals with high VS-low AMY activation and by anxious/depressive symptoms for those with low VS-high AMY activation. Across both neural phenotypes, they discovered that, regardless of the direction, the greater the VS-AMY imbalance, the higher the risk for problem drinking. Moreover, for the low VS-high AMY risk phenotype, stress not only predicted the presence of AUD at the time the study was conducted but also self-reported problem drinking at a 3-month follow-up.

Although the AMY is a central element of the stress network, its activity is strongly regulated by numerous cortical structures, particularly the insula, PFC, ACC and OFC. There are numerous reciprocal connections between the AMY and many of these cortical and subcortical (e.g., straitum, hippocampus, extended AMY) regions. Emerging network connectivity studies are beginning to shed light on how dysregulation of this cortico-AMY-subcortical stress circuitry contributes to the etiology of AUD. Notably, a series of seminal studies by Rajita Sinha and colleagues have used individualized guided imagery scripts to study stress-related engagement of the stress network in subjects with AUD. Their initial studies, conducted in healthy controls, used fMRI to measure differences in brain-wide activation in response to stressful vs. neutral imagery scripts and found increased stress-related activation in the mPFC and vACC as well as numerous mesolimbic structures (striatum, thalamus, hippocampus), all regions intimately connected with the AMY (Sinha, Lacadie, Skudlarski, & Wexler, 2004). Further analyses were conducted to identify brain regions whose activity correlated with subjective stress ratings. A positive correlation between self-reported distress ratings and caudate and thalamic regions was observed and a negative relationship emerged between distress ratings and the ACC and insula. Together, these data provided early empirical evidence that many elements of the stress network are indeed engaged by stressful stimuli and provided critical insight into how these circuits could modulate negative emotional processing. As just a couple of examples, activation of areas like the ACC may serve to reduce emotional distress whereas thalamic and striatal regions may actually be engaged during the subjective experience of stress.

Guided by these and other findings, the Sinha group, and others, have conducted numerous additional studies in subjects with AUD (for reviews see: Blaine, Milivojevic, Fox, & Sinha, 2016; Sinha, 2012; Sinha & Li, 2007; Tanabe et al., 2019). One important finding that has emerged from these studies is that many nodes of the stress circuitry (e.g., mPFC, OFC, insula) are not only engaged by stressful stimuli but also by alcohol-associated visual cues. These findings suggest that, although alcohol-associated cues clearly engage elements of the reward circuitry, they also activate nodes that are more often considered part of the stress circuitry, likely due to stress-evoked craving. These findings also serve to reinforce the inarguable fact that the reward and stress networks are highly inter-related and largely comprised of overlapping brain regions.

One particularly notable study by the Sinha group used the customized script-driven imagery to examine how engagement of stress and reward circuits by stress and alcohol cues influenced craving and relapse in 45 treatment engaged individuals at 4–8 weeks of abstinence (Seo et al., 2013). Relative to neutral scripts, both the stress and alcohol cues significantly increased self-reported anxiety and craving ratings in the AUD group. Similarly, whole-brain analyses revealed that AUD subjects showed increased activation in corticolimbic and striatal regions during stress and alcohol cues relative to neutral stimuli, including the mPFC, ACC, AMY, and HC. Importantly both stress and alcohol cue-induced craving scores were positively correlated with hyperactivity in the vmPFC and ACC during neutral-relaxing trials with stress-induced craving also being correlated with hyperactivity of the VS and precuneus. Furthermore, using a covariate analysis to exclude behavioral factors that predict relapse (e.g., years of alcohol use, higher stress-induced alcohol craving), they found that vmPFC/ACC, VS, and precuneus hyperactivation during neutral trials was actually predictive of a shorter time to relapse. Notably, a secondary analysis revealed that vmPFC/ACC hyperactivity on neutral trials was the most accurate classifier of relapsers vs. non-relapsers. These findings reveal that key elements of the stress network are profoundly disrupted during early abstinence and suggest that these alterations may contribute to relapse. The mPFC and ACC provide powerful top-down control of many limbic elements of the stress network, including the AMY, and play a key role in emotional processing. The observed blunting of mPFC/ACC engagement in response to stress and alcohol cues in the AUD cohort is consistent with affect dysregulation associated with these brain areas. Moreover, these measures may serve as potential biomarkers for relapse and may represent viable targets for novel treatments, like TMS, that could potentially restore normal neural activity in these circuits.

A number of imaging studies have examined functional connectivity among elements of the stress network in individuals with AUD. One common finding has been that connectivity between the AMY and prefrontal regions is weakened in individuals with AUD, both in the resting state and during the processing of aversive stimuli (Muller-Oehring, Jung, Pfefferbaum, Sullivan, & Schulte, 2015; O’Daly et al., 2012; Wade et al., 2017). One recent study conducted a whole-brain analysis, using the AMY as a seed, to examine the relationship between resting state AMY connectivity and problem drinking in 83 non-dependent alcohol drinkers (Hu et al., 2018). This brain-wide analysis found that, in the resting state, the AMY is positively connected to many cortical and subcortical structures including the HC, insula, and vmPFC. Notably, they reported only one significant relationship between problem drinking and resting state AMY connectivity to any of these nodes, a negative correlation between problem drinking scores and AMY-dACC connectivity and this correlation remained significant even when controlling for depression and other substance use. In addition, they examined this relationship for the four major AMY subregions, including the BLA and CeA. Only weakened BLA-dACC connectivity was significantly correlated with problem drinking scores.

Another study examined task-related functional connectivity using a voxel-wise method, called the intrinsic connectivity distribution, which permits an unbiased analysis of brain-wide changes in connectivity without the need for a priori seeds or regions of interest. The first part of this study examined 45 recovering AUD patients and examined whole-brain connectivity during presentation of scripted stress and alcohol imagery cues (Zakiniaeiz, Scheinost, Seo, Sinha, & Constable, 2017). Interestingly, this unbiased approach also identified cingulate cortex as a major locus of connectivity changes. They found reduced voxel-wise connectivity to stress vs. neutral cues in the ACC, MCC, and PCC and reduced PCC connectivity to alcohol cues. They also showed that weaker PCC response to alcohol vs. neutral cues was associated with a longer time to relapse and greater connectivity of the PCC during alcohol- vs. stress-cues was also associated with better recovery outcomes. In the second phase of the study, a subset of 30 AUD patients were demographically matched with 30 healthy controls to permit group comparisons. Using the same unbiased analytical approach, they found reduced MCC and PCC connectivity in the AUD cohort during stress and alcohol cues compared to the healthy controls. These findings extend those from Hu et al., revealing that excessive alcohol drinking is associated with decreased cingulate cortex connectivity in the resting state and during presentation of aversive stimuli known to elicit craving and anxiety. The finding that cingulate cortex connectivity changes correlated with recovery outcomes suggests that this alteration may represent yet another potential biomarker for relapse.

Finally, a related paper by the Sinha group used a seed-based approach to examine neural correlates and connectivity patterns associated with stress-related impulse control problems in 37 recovering AUD patients and 37 demographically matched healthy controls (Seo, Lacadie, & Sinha, 2016). This study employed their alcohol and stress cue imagery scripts and also assessed impulse control difficulties using IMPULSE, a scale previously shown by these investigators to be predictive of addictive disorders and relapse vulnerability (Fox, Axelrod, Paliwal, Sleeper, & Sinha, 2007; Fox, Hong, & Sinha, 2008). As in their prior studies, AUD patients reported higher anxiety ratings during alcohol cues and elevated craving during both stress and alcohol conditions. Interestingly, impulsivity measures did not correlate with anxiety or craving ratings in either the AUD or control groups. A whole-brain analysis of IMPULSE scores during stress vs. neutral and alcohol cue vs. neutral conditions revealed a number of significant correlations in the AUD cohort, but only for the stress vs. neutral comparison. For that condition, activity in the vmPFC, caudate, and dlPFC was negatively correlated with impulsivity ratings. In the healthy control group, no significant correlations emerged between impulsivity and neural activity in any brain region. However, there were also no significant differences in any regional correlational coefficients between the AUD and control groups, suggesting that the neural network encoding impulsivity may be similar in both cohorts but that elements of this network may be strengthened in individuals with AUD. In a final analysis, the authors used the three brain regions identified above as being involved in impulse control difficulties as seeds to compare functional connectivity patterns in AUD and control subjects. A number of significant group differences in connectivity were observed for each seed (both increases and decreases). Perhaps most notably, with the vmPFC as a seed, AUD patients showed significantly reduced connectivity with the rACC compared with the control group. When an analysis of the relationships behavioral ratings of anxiety and craving and all of these functional connectivity patterns was conducted, the only significant finding was a negative correlation between stress-induced anxiety and functional connectivity between the dlPFC and dmPFC in AUD subjects. No correlations were detected between behavioral ratings and functional connectivity patterns in controls. Together, these findings reveal that AUD is associated with reduced stress reactivity of numerous nodes of the stress network and that altered connectivity between several cortical nodes of this network may contribute to greater stress-related impulse control problems in these individuals.

10.2. Stress circuit and PTSD—Human studies

As discussed earlier, hypervigilance, hyperarousal, and deficits in fear learning represent core symptoms of PTSD and it is now widely appreciated that dysregulation of the brain stress network contributes to these maladaptive phenotypes (Akiki, Averill, & Abdallah, 2017; Dossi, Delvecchio, Prunas, Soares, & Brambilla, 2020; Kunimatsu et al., 2020; Prasad, Chaichi, Kelley, Francis, & Gartia, 2019; Shalev, Liberzon, & Marmar, 2017). Indeed, it is perhaps a review of the imaging literature on PTSD that provides the greatest empirical support for considering the AMY (and arguably the BLA) as a core node of the stress circuitry. Decades of animal research has revealed that fear learning is mediated by a strengthening of synaptic activity in the BLA and that fear memories can reside as engrams in this brain region (Goode & Maren, 2019; Ressler & Maren, 2019; Sun, Gooch, & Sah, 2020). Moreover, as reviewed below, AMY hyperexcitability is, perhaps, one of the most replicable neural phenotypes associated with PTSD.

As with most nascent fields, initial functional imaging studies examining AMY responsivity in PTSD were somewhat mixed, with some reporting increases (Rauch et al., 2000; Shin et al., 2005; Williams et al., 2006), decreases (Britton, Phan, Taylor, Fig, & Liberzon, 2005; Phan, Britton, Taylor, Fig, & Liberzon, 2006), or no change in activity (Bremner et al., 1999; Lanius et al., 2001) relative to controls. However, one of the first large meta-analyses of this literature provided compelling evidence that increased AMY reactivity to aversive stimuli represents one of the most robust and reproducible neural phenotypes of PTSD. This review employed an elegant voxel-wise approach to examine fMRI and PET studies of PTSD, social anxiety, and specific phobia (Etkin & Wager, 2007). Their analysis included 40 studies that contrasted a negative emotional condition with a neutral or positive emotional condition or a resting baseline. Importantly, they also included data on fear conditioning in healthy subjects. The AMY was one of only two brain regions, along with the insula, that showed hyperactivity across all three disorders and these two areas were also the only structures that showed increased activity during fear conditioning in controls. This study also identified numerous areas of hypoactivation within the stress network, but only in PTSD subjects, including the mPFC, ACC, MCC, OFC, and anterior hippocampus (aHC). Of note, lower mPFC activity was associated with greater PTSD symptom severity. This was also one of the first studies to examine co-activation across brain regions of interest. They reported consistent co-hypoactivation among cortical areas and a significant negative co-activation between the ACC and the AMY, drawing attention to the now well-established link between hypoactivity of frontal regions and AMY hyperactivity. The authors also speculated that AMY hyperactivity was like driven by the BLA, as it receives the majority of sensory cortical input and they actually observed a PTSD-associated decrease in dorsal AMY activity, which comprises the central AMY.

In the past decade, a number of additional meta-analyses have confirmed and extended these findings, showing that AMY hyperactivity and dysregulation of many nodes of the stress network at rest, and during the processing of fear/stress-related stimuli, are not only a common feature of PTSD but can also be associated with vulnerability to the disorder, symptom severity and treatment outcome (Akiki et al., 2017; Hayes, Hayes, & Mikedis, 2012; Koch et al., 2016; Patel, Spreng, Shin, & Girard, 2012; Prasad et al., 2019; Wang et al., 2016).

One major advance in functional imaging over the past decade has been that improvements in fMRI resolution have made it possible to parse subregions of heterogeneous brain regions like the AMY. In fact, a number of recent studies have taken advantage of these imaging advances to analyze reactivity of the BLA, CEA, and other large AMY subregions in PTSD patients and controls. For example, one study examined AMY subregion activation in response to brief presentations of consciously or unconsciously viewed fearful faces, using a backward masking procedure (Neumeister et al., 2018). This study included three cohorts of patients diagnosed with PTSD, Generalized Anxiety Disorder (GAD) or Panic Disorder (PD) as well as a large sample of healthy controls. When viewing conscious stimuli, all participants (including healthy controls), showed increased activity in several AMY nuclei, including the BLA. However, hyper-responsivity to these stimuli was not significantly greater in any of the patient cohorts and healthy controls. In marked contrast, a similar analysis during the unconscious condition revealed a selective increase in BLA activity in the PTSD patients relative to the PD and GAD groups as well as the healthy controls. These striking findings highlight the central role of BLA dysfunction in the etiology of PTSD and provide evidence of differential neural processing of fear-related stimuli in individuals with stressor and anxiety disorders.

Many recent resting state fMRI studies of PTSD have also focused on differentiating changes in functional connectivity between distinct AMY subregions and other nodes of the stress circuitry. These studies have all shown that the BLA and CEA, the two largest AMY nuclei, have distinct patterns of functional connectivity within the stress network (Aghajani et al., 2016; Brown et al., 2014; Nicholson et al., 2015) and, although not all data concur (Brown et al., 2014), most support the idea, first proposed by Etkin and Wager, that the frequently reported hypoconnectivity between the AMY and cortical areas like the PFC and ACC, likely reflects a reduction in normal cortical regulation of BLA activity. For example, a study of adolescent sexually abused PTSD patients and health controls used intrinsic functional connectivity analysis to examine the BLA and CEA (Aghajani et al., 2016). They found that adolescent PTSD patients showed decreased connectivity between the BLA and a cluster of cortical areas that included the ACC and mPFC. In contrast, they reported increased CEA connectivity with another cortical cluster comprised of the OFC and subcallosal cortices, with the latter change being correlated to increased stress and anxiety symptoms in PTSD patients. Another interesting study examined patients with PTSD alone, PTSD and comorbid MDD, and healthy controls that were all recruited from a cohort who survived the same large earthquake (Yuan et al., 2019). A seed-based approach was employed to compare BLA and CEA functional connectivity across groups. Major findings from this study were that PTSD+MDD patients had lower BLA connectivity with the ACC, SMA and bilateral putamen/pallidum, with the latter negatively correlating with depressive symptoms. Again, functional CEA connectivity, which was predominantly to the ACC/SMA in this study, was actually increased in the PTSD+MDD group. Notably, an analysis of functional connectivity that merged both AMY subregions found no group differences. Collectively, these findings are consistent with the notion that PTSD is associated with a loss of top-down cortical control of BLA activity and that, perhaps, cortical connectivity with the CEA may be increased in this disorder. These findings underscore the distinct connectivity patterns of key AMY subregions and the need to consider, where possible, the possibility that PTSD may have distinct effects on activity patterns between these subregions and other nodes of the stress circuitry.

Finally, several exciting advances suggest that our growing understanding of stress network dysregulation in PTSD may be starting to inform on novel treatments for this disorder. For example, several studies have explored the utility of using real-time fMRI neurofeedback (rt-fMRI NF) to normalize PTSD-associated perturbations in stress network functional connectivity. Rt-fMRI NF is a relatively new technique in which participants receive contingent visual or auditory feedback about ongoing activity in a specific brain region. Remarkably, numerous studies have shown that this procedure allows individuals to exert volitional control over neurophysiological activity within discrete brain regions and networks and that this neurofeedback can be harnessed to alleviate behavioral and neurophysiological symptoms of a number of psychiatric and neurological conditions (Linhartova et al., 2019; Watanabe, Sasaki, Shibata, & Kawato, 2017; Young et al., 2018).

To date, three studies have been published using rt-fMRI NF to target the AMY in PTSD patients. The first conducted an unblinded pilot feasibility study on three veterans with chronic PTSD (Gerin et al., 2016). After a baseline fMRI scan, participants were trained over three sessions to suppress AMY activity after exposure to personalized trauma scripts. All three of the subjects showed an improvement of PTSD symptoms and a normalization of changes in AMY resting-state functional connectivity, with a strengthening of AMY-ACC and AMY-OFC connectivity and a dampening of AMY connectivity to elements of the salience network, including the insula. A slightly larger study on 10 PTSD patients used a similar approach to train subjects to suppress AMY activity across three training sessions (Nicholson et al., 2017). A final scan was also conducted without rt-fMRI NF to determine if the training had any lasting effects. Notably, subjects were able to significantly reduce AMY activation, even during the final scan without neurofeedback. In addition, activation in the PFC, ACC, and insula were negatively correlated with dissociative PTSD symptoms during the final scan. Task-based functional connectivity was also increased between the AMY and dmPFC and ACC during the training sessions.

The largest rt-fMRI NF PTSD study to date was conducted on 40 male veterans with PTSD and 22 veteran controls (Misaki et al., 2018). This study actually sought to increase AMY responsivity in the participants. While this may seem paradoxical, the authors noted that the AMY is activated by both positive and negative emotional stimuli and that training the AMY to increase its activity in response to positive stimuli can actual decrease responsivity to negative cues. Thus, they used rt-fMRI NF to train 25 of the PTSD patients and all of the control subjects to increase AMY responsivity during recall of positive autobiographical memories. The other 15 PTSD patients were also asked to increase an fMRI signal but one derived from activity within the intraparietal sulcus, a region not thought to be involved in emotion regulation. Subjects received three training sessions followed by a final session without feedback approximately 1-week later. PTSD symptoms were significantly improved in the AMY rt-fMRI cohort but not the group receiving feedback from the intraparietal sulcus. Importantly, PTSD subjects had significantly lower AMY-PFC connectivity than controls at baseline and the PTSD group receiving AMY rt-fMRI NF showed an increase in this connectivity over training sessions. Interestingly, this change did not correlate with PTSD symptom improvement. Rather, a connectome-wide analysis revealed that neurofeeback directed at the AMY normalized other PTSD-associated alterations in functional connectivity. For example, baseline connectivity between the ACC and supplementary motor area (SMA) was lower in PTSD patients vs. controls and AMY rt-fMRI NF increased ACC-SMA connectivity in PTSD patients. Moreover, a positive correlation was noted between the strengthening of this connectivity and an improvement of depressive symptoms in the PTSD group. These findings demonstrate that PTSD patients can be trained to modulate activity in key brain areas impacted by PTSD. In addition, these data reveal that rt-fMRI NF can have widespread effects on network connectivity beyond the targeted region. Though fairly preliminary, the therapeutic implications of these findings are bolstered by other evidence demonstrating that some the beneficial effects of rt-fMRI NF, like the strengthening of AMY-PFC connectivity, are also associated with resilience to PTSD (Bolsinger, Seifritz, Kleim, & Manoliu, 2018) and positive outcomes with other neuromodulatory treatments like transcranial magnetic stimulation (Philip et al., 2018).

10.3. Stress circuitry—Preclinical animal studies

As noted in our review of the reward circuitry in AUD and PTSD, few human studies have directly addressed stress network dysregulation associated with this dual diagnosis. However, animal studies have provided compelling evidence that key nodes of this circuitry likely mediate the stress-related behavioral phenotypes associated with these disorders. Moreover, studies using animal models of the comorbid condition have identified several stress circuit adaptations that are similar to those observed in human subjects with each disorder.

For example, optogenetic circuit manipulations have been elegantly employed to demonstrate that central nodes of the stress network, like the BLA and elements of the extended AMY, can bidirectionally modulate anxiety-like behaviors in mice and rats (Daviu, Bruchas, Moghaddam, Sandi, & Beyeler, 2019; Janak & Tye, 2015; Lalumiere, 2014). In addition, deficits in associative fear learning, particularly extinction of fear-related memories, are a hallmark feature of PTSD (Norrholm & Jovanovic, 2018). While there is only limited evidence linking AUD and alterations in fear learning (Muench et al., 2019; Stephens et al., 2005), extensive studies in animals have shown that stress network regions like the BLA, CeA, BNST, and HC are critical for the acquisition, consolidation, expression and extinction of fear memories (for reviews see: Fenster et al., 2018; Ressler & Maren, 2019). Early rodent work revealed that the acquisition of fear memories was associated with a strengthening of glutamatergic synapses on BLA projection neurons (Maren & Fanselow, 1996). More recent work, integrating elegant molecular and optogenetic approaches, has demonstrated that there are sparse clusters of cells in the BLA and other stress network nodes like the HC, termed engrams, that are activated during the formation and extinction of fear memories (Davis & Reijmers, 2018; Josselyn, Kohler, & Frankland, 2015; Zhang, Kim, & Tonegawa, 2020). Optogenetic activation of these engrams can elicit freezing behavior or induce extinction in a context-specific fashion. It is also worth noting that animal models of AUD and vulnerability to AUD and PTSD both engender deficits in extinction of fear memories, therefore these cellular populations could play an important role in the disorders (Holmes et al., 2012; Skelly et al., 2015).

Studies in our lab have shown that adolescent social isolation leads to an increase in the intrinsic excitability of glutamatergic projection neurons in the BLA (Rau, Chappell, Butler, Ariwodola, & Weiner, 2015). Similar increases in the excitability of these cells are seen following chronic stress exposure in preweaning (Guadagno, Wong, & Walker, 2018) and adult (Rosenkranz, Venheim, & Padival, 2010) rats. This enhanced BLA excitability is mediated, at least in part, by an increase in the expression and function of calcium-activated small conductance (SK) potassium channels (Atchley, Hankosky, Gasparotto, & Rosenkranz, 2012; Rau et al., 2015) and inhibition of BLA SK channels can significantly attenuates anxiogenic phenotypes promoted by this model (Atchley et al., 2012; Rau et al., 2015). More recently, we have shown that adolescent social isolation is also associated with an increase in synaptic excitation in the ventral hippocampus (analogous to the human aHC), a brain region that receives strong innervation from the BLA (Almonte et al., 2017). In addition, we have confirmed, using chemogenetic approaches, that silencing BLA-vHC synapses reduces anxiety-like behaviors and have also discovered that inhibiting this input reduces alcohol seeking-related behaviors (Ewin et al., 2019).

The Gilpin laboratory has established a predator odor model of comorbid AUD/PTSD in which a subset of rats exposed to bobcat urine exhibit enduring avoidance of odor-paired stimuli (Albrechet-Souza & Gilpin, 2019). This model is particularly powerful as it mirrors findings in humans that only a subset of trauma-exposed subjects go on to develop PTSD. Relative to non-avoiders and control subjects not exposed to the predator odor, avoiders show increased operant alcohol self-administration and aversion resistant drinking. Neurobiological studies using this model have shown that avoiders exhibit greater increases in c-Fos expression in the CeA after odor exposure than non-avoiders and, notably, higher avoidance behavior correlates with increased c-Fos and CRF immunoreactivity in the CeA. In addition, infusion of a CRF1 antagonist into the CeA reduced alcohol self-administration and avoidance behavior in avoiders, while having no effect on these measures in non-avoiders (Weera, Schreiber, Avegno, & Gilpin, 2020).

Social defeat stress, although often used as a model of depression (Knowland & Lim, 2018), also leads to a number of behavioral phenotypes associated with AUD and PTSD, including increases in anxiety-like behaviors, deficits in fear extinction, and escalations in alcohol intake (Lisboa et al., 2018; Narayanan et al., 2011; Newman, Leonard, Arena, de Almeida, & Miczek, 2018). Many studies have shown that this model strongly activates the BLA and dysregulates numerous nodes of the stress network, including the PFC and extended AMY (Diaz & Lin, 2020; Patel, Kas, Chattarji, & Buwalda, 2019). Finally single prolonged stress, in which subjects are exposed to a series of stressors across a single day, followed by a week-long incubation period, also engenders many key phenotypes of AUD and PTSD (Lisieski et al., 2018; Souza et al., 2017). One recent study employed longitudinal magnetic imaging approaches to scan subjects before and after this procedure (Piggott et al., 2019). Their findings revealed a significant decreased in PFC glutamate levels as well as decreased neural activity in the infralimbic PFC region, along with increased activity in the BLA following sPS with no changes in the control group. These findings are remarkably similar to the commonly reported decreases in PFC-AMY functional connectivity reported in both AUD and PTSD.

11. A novel neural framework for comorbid AUD and PTSD

The preceding section sought to provide a comprehensive review of reward and fear circuit dysregulation in AUD and PTSD. Collectively, this literature reveals striking similarities in many of the functional connections within these networks that are altered in both conditions. Focusing first on the reward circuitry, both disorders are associated with a profound dysregulation of ventral striatal activity and connectivity, at rest and during the processing of positive and negative emotional imagery. However, the direction of these alterations is complex and likely dependent on the nature of the stimuli used to experimentally engage this network. For example, relative to healthy controls, individuals with AUD exhibit stronger behavioral reactivity to alcohol-related cues (Huang et al., 2018; Schacht et al., 2013; Vollstadt-Klein et al., 2012) whereas PTSD patients respond more to trauma-related stimuli (Vujanovic et al., 2017) and both conditions are often associated with decreased reactivity to non-alcohol-related positive stimuli (Fritz et al., 2019; Nawijn et al., 2015; Wrase et al., 2007). These behavioral findings are generally in line with functional imaging data showing that AUD subjects exhibit higher VS activity in response to alcohol cues but often lower VS responsivity to the presentation of non-alcohol-related positive stimuli (Schacht et al., 2013; Wrase et al., 2007). Similarly, PTSD patients exhibit lower VS activation during the presentation of positive cues but display heightened VS activation in response to aversive cues (Elman et al., 2009, 2018).

There are also several convergent adaptations within the stress circuitry in both AUD and PTSD. Notably, extensive fMRI findings in individuals with PTSD, along with more recent evidence from functional imaging studies in AUD cohorts, suggest that both disorders involve a loss of top-down cortical control of AMY activity, mainly due to a weakening of functional connectivity between the mPFC-AMY and ACC-AMY. In PTSD patients, these maladaptive changes are generally associated with AMY hyperactivity at rest and during emotional tasks (Etkin & Wager, 2007). Interestingly, as reviewed earlier, in AUD subjects, increased AMY reactivity to stressful stimuli was only seen in individuals with low VS reactivity to rewarding cues and this relationship was mediated by anxious/depressive symptoms, which are also core symptoms of PTSD (Nikolova et al., 2016). Collectively, these findings are consistent with the notion that individuals who exhibit behavioral phenotypes common to AUD and PTSD share a loss of cortical regulation of AMY reactivity that may contribute to these maladaptive behaviors.

Neurobiological studies in animal models of the comorbid condition are generally supportive of the human literature. For example, adolescent social isolation, which promotes symptoms of AUD and PTSD, results in an enduring dysregulation of the mesolimbic DA system similar to that observed in patients with these disorders (Butler, Ariwodola, & Weiner, 2014). In addition, many animal models of AUD and PTSD engender increases in BLA excitability that are comparable to the AMY hyperexcitability that is frequently observed in individuals afflicted with these disorders (Gilpin & Weiner, 2017). Moreover, emerging preclinical data suggest that normalizing DA signaling or AMY activity may reverse some AUD/PTSD-like symptoms (Karkhanis et al., 2016; Rau et al., 2015).

Together, these findings provide compelling evidence that the development and progression of AUD and PTSD both involve common maladaptive changes in brain reward and stress networks. These shared neural alterations provide a simple, biological explanation for the overlap in many of the core symptoms of these disorders as well as their frequent co-occurrence. Unfortunately, despite these impressive advances in our understanding of the neural substrates that may contribute to the comorbid condition, this knowledge has not yet led to the development of novel treatments for the dual diagnosis (or either disorder alone) or a significant improvement in patient outcomes. Thus, a fundamental and, as yet, unresolved question is: what is holding us back? Notwithstanding the large volume of data showing that AUD and PTSD are associated with a profound dysregulation of similar nodes of the reward and stress circuits, we suggest that viewing these circuits as discrete networks that independently process rewarding and stressful stimuli does not actually reflect how these networks function. We end this review with a brief proposal that an over-reliance on this dichotomous framework to interpret emerging findings on the neural substrates of AUD and PTSD may be hampering efforts to translate these advances into better treatments and patient outcomes and offer an alternative, based on sensory/affective valence encoding.

Perhaps the most important limitation with the current conceptual framework of reward and stress networks is that they do not depict the extensive structural and functional connectivity between many of the brain regions within these circuits. Canonical diagrams of reward and stress circuitry also mask the fact that many, if not all, brain regions within these circuits are comprised of heterogeneous cell populations that play an integral role in the processing of both reward and stress-related stimuli. For example, as illustrated in Fig. 1, while the reward circuit is usually centered around the VTA-NAc DA projection, the VTA contains a diverse array of non-DA cells (Morales & Margolis, 2017) and the NAc also receives dense excitatory input from many other regions, like the PFC and aHC, and all of these pathways are involved in the processing of both reward- and stress-related information (Britt et al., 2012; Daviu et al., 2019; Francis & Lobo, 2017; MacAskill, Little, Cassel, & Carter, 2012). Moreover, DA can be released by both aversive and rewarding stimuli (for reviews see: Morales & Margolis, 2017; Thibeault, Kutlu, Sanders, & Calipari, 2019). Similarly, the BLA, which is commonly illustrated as the core node of stress/fear circuits, is inextricably interconnected with many elements of the reward circuitry, including the NAc, and there is now compelling opto- and chemo-genetic evidence that many of these pathways process reward- as well as stress-related stimuli (Daviu et al., 2019; Janak & Tye, 2015; Sharp, 2017). While this overlap between the brain reward and stress systems is widely appreciated in the literature (Centanni, Bedse, Patel, & Winder, 2019; Koob & Schulkin, 2019; Koob & Volkow, 2016), these heuristics are nevertheless still commonly used to interpret functional connectivity data in the AUD and PTSD fields and their continued use may be limiting our ability to truly appreciate how dysregulation of these two networks, which are actually highly functionally connected, contributes to the etiology of these disorders.

How then might we provide a conceptual framework that more accurately illustrates the integration of canonical reward and stress circuits and, perhaps, better explains how maladaptive changes in these functionally inter-related networks leads to the frequent co-occurrence of AUD and PTSD? The answer may lie in a rapidly growing field of neuroscience that is seeking to understand the critical brain circuitry that allows all organisms to perform the essential function of responding to motivationally salient internal and external stimuli in a manner that is congruent with this input. In this field, the term valence is used to define an internal state that is generated by integrating all relevant incoming sensory and/or emotional stimuli. Internal states with positive valence are generally perceived as pleasurable and usually promote behaviors that seek to maintain these conditions while those with negative valence are typically discerned as aversive and tend to encourage behaviors that relieve these unpleasant states. Thus, fundamentally, this circuitry serves the essential function of reinforcing behaviors that are life-sustaining while discouraging behaviors that may be hazardous to one’s health and well-being. It is important to note that, in humans, valence can be measured in the absence of overt behavior, by asking a subject to rate the relative pleasantness or aversiveness of a stimulus. In contrast, valence can only be inferred from behavior in non-human animals.

As just one example of valence circuits, a recent study identified excitatory projections from regions of gustatory cortex that encode sweet and bitter taste which project onto distinct clusters of cells in the AMY (Wang et al., 2018). Using elegant optogenetic approaches, this study demonstrated that these cortico-AMY circuits do not encode sweet or bitter taste but rather the hedonic valence of these tastants. In other words, photoactivation of the sweet pathway could promote appetitive behavior directed at a neutral stimulus or override avoidance directed at a bitter tastant and optical stimulation of the bitter circuit could do the opposite. In the last decade, similar approaches have been used to identify neural circuits that encode the valence of other sensory stimuli as well as many emotion-like states, like anhedonia (Hoflich, Michenthaler, Kasper, & Lanzenberger, 2019) and fear (Tovote, Fadok, & Luthi, 2015).

Although this is a nascent field of neuroscience, the brain regions that have already been shown to contribute to the encoding of hedonic valence include all the major nodes that comprise traditional reward and stress circuits that are known to be dysregulated in AUD and PTSD (Fig. 2A). However, a major advantage of this conceptual framework is that it does not assign discrete brain regions for the processing of positive and negative stimuli. Rather, it envisions this functionally cohesive network working in a highly integrated fashion to process the valence of all sensory and emotional information, thus allowing the common maladaptive alterations in functional connectivity associated with AUD and PTSD to be visualized within a single cohesive circuitry (Fig. 2B). Moreover, as alluded to above, this heuristic is supported by a large and growing volume of preclinical circuit-mapping data demonstrating that there are numerous monosynaptic projections that connect many cortical and subcortical structures within the reward and stress circuits (Beyeler et al., 2018; Namburi, Al-Hasani, Calhoon, Bruchas, & Tye, 2016; Tye, 2018). Together, these circuits form a valence network (VN) that we operationally define as a highly distributed circuitry tasked with assigning some degree of positivity or negativity to all sensory and/or affective input, encoding the relative salience or importance of this input (see below), in order to orchestrate behavioral and/or cognitive changes that are congruent with the incoming stimuli.

Fig. 2.

Using a human valence network model to visualize the primary alterations in functional connectivity that are common to AUD and PTSD. (A) The major cortical and subcortical brain regions and functional connections involved in assigning a degree of positivity or negativity to all sensory and emotional stimuli, encoding the salience of this input, and orchestrating appropriate behavioral and cognitive responses to this information. (B) Illustration of major maladaptive alterations in valence network connectivity common to both AUD and PTSD, based primarily on data from human functional neuroimaging studies. These changes include an increase in BLA excitability driven, in part, by decreased mPFC and ACC connectivity (thinner line), and increased insula connectivity to this brain region (thicker line).

Importantly, these preclinical animal findings are actually supported by human functional imaging data. A recent meta-analysis of almost 20 years of human affective valence neuroimaging studies sought to test three competing hypotheses about the brain basis of hedonic valence encoding (Lindquist, Satpute, Wager, Weber, & Barrett, 2016). This study found little evidence for a bipolarity hypothesis, in which a single brain network encodes valence by monotonically increasing or decreasing along the spectrum of positive to negative affect. It also did not support a bivalent hypothesis, analogous with brain reward and stress circuits, in which independent brain networks encode positive and negative affect. Instead, it found evidence for what the authors termed an “affective workspace” hypothesis, in which both positive and negative affect are encoded by a single, flexible network. In other words, at the level of brain regional activity, there does not appear to be a single brain region or even voxel that uniquely represents positive or negative affect. Rather, there appears to be a valence-general network that engages in the processing of all affective stimuli, regardless of their hedonic valence.

At first, these findings may seem at odds with the preclinical data demonstrating that there do appear to be dedicated monosynaptic circuits within the valence-general network that do indeed encode either positive or negative valence. However, as noted above, many of these studies have also shown that there are direct, non-overlapping connections between elements of the VN that assign both positive and negative valence to incoming stimuli. This can be illustrated by examining the role of the BLA in valence processing. This brain region receives dense input from many sensory and affective brain regions. These inputs synapse primarily onto excitatory BLA pyramidal neurons which encode the positive or negative valence of these stimuli and generate valence-appropriate behavioral and cognitive responses via monosynaptic projections to downstream structures like the CeA, NAc and aHC (Janak & Tye, 2015). Importantly, many studies have demonstrated that discrete projections from the BLA to each of these regions can elicit both positive and negative valence-related behaviors (Kim, Pignatelli, Xu, Itohara, & Tonegawa, 2016; Pi et al., 2020; Zhang et al., 2020). As just one recent example, CCK+ and CCK− BLA pyramidal cells are monosynaptically connected to distinct populations of NAc cells and photoactivation of these circuits can elicit positive or negative affective states (Shen et al., 2019). The reason that dedicated positive and negative valence circuits were not revealed in the human valence processing meta-analysis likely reflects the fact that these neuroimaging studies capture functional, and not structural, connectivity on a time-scale that is orders of magnitude slower than the monosynaptic communication identified in preclinical circuit-mapping studies. Despite these methodological issues, it is noteworthy that the brain regions that comprise the human VN are essentially identical to those identified in animal valence circuit-mapping studies and, perhaps not surprisingly, include many of the structures and functional networks that are altered in AUD and PTSD, including cortical areas like the PFC, ACC, and insula as well as subcortical regions like the AMY and VS (Fig. 2B).

It also warrants discussion that many of the brain regions and functional connections that we refer to as a VN also comprise what is commonly referred to as the salience network (SN) in the human functional imaging literature (Seeley, 2019). The primary function of the SN is to assess the importance (salience) of sensory and/or emotional stimuli, independent of valence, to inform on ongoing behavioral or cognitive processes, often related to threat detection (Kolesar, Bilevicius, Wilson, & Kornelsen, 2019; Szeszko & Yehuda, 2019). However, the authors of the valence functional connectivity meta-analysis correctly note that all of our conscious experience is simultaneously assigned with valence as well as the relative importance of this information. Thus, we suggest that there are two core processing functions of this network: (1) to assign a hedonic valence to all sensory and emotional information and (2) to encode the salience of this input. Since the primary goal of this circuitry is to decipher internal states in manner that promotes the viability of an organism, the term VN may better reflect this role. So, while we refer to this circuitry as the VN, we operationally define it as a distributed brain network that encodes both the valence of all sensory/emotional input as well as the relative salience or importance of this information in order to invigorate behavioral responding for rewarding stimuli and dampen responding for aversive stimuli.

How can this VN framework be used to inform our understanding of the neural network dysregulation in AUD and PTSD? Most importantly, as illustrated in Fig. 2B, the VN provides a way to view the pathophysiology of both disorders within a single integrated circuitry that more accurately depicts the functional role of the brain regions that comprise it. We suggest that, rather than thinking of AUD and PTSD as disorders of reward and/or stress, that they both be considered disorders of hedonic valence processing or VN imbalance disorders. We suggest that, in healthy individuals, the VN is able to reliably encode both the valence of incoming sensory and emotional information as well as the relative salience of that information to appropriately guide behavior and cognition. However, if elements within the VN circuitry are dysregulated, perhaps by exposure to a traumatic event and/or excessive alcohol consumption, incongruous valence encoding may occur which could then lead to the behavioral and cognitive symptoms of disorders like AUD and PTSD.

Indeed, all of the maladaptive behaviors that are central to the diagnosis of AUD and PTSD can be mediated, at least in part, by alterations in VN encoding. For example, many of the symptoms of AUD are associated with excessive alcohol consumption, loss of control over this behavior, and craving. These behaviors could be driven by an increase in the salience of the VN circuitry that encodes the rewarding properties of this drug and/or an amplification of the salience of the negative affect engendered by withdrawal. Similarly, core symptoms of PTSD, like hypervigilance, may arise due to heightened responsivity of VN circuitry to anxiety-provoking stimuli whereas anhedonia and negative mood may reflect hypoactivity of VN nodes that normally encode the positive value of naturally rewarding input.

This VN framework is particularly helpful when trying to interpret the epidemiological and neurobiological data on comorbid AUD and PTSD. First, as already noted, virtually all of the major alterations in functional connectivity that are common to both disorders involve elements of the VN, thus providing a biological basis for the overlap in the many symptoms shared by these diseases. This model may also explain why most epidemiological studies suggest that PTSD more frequently precedes AUD and generally support the self-medication hypothesis for the comorbid condition (Straus et al., 2018). Acute alcohol likely normalizes activity in the valence circuits that are dysregulated and drive the maladaptive symptoms of PTSD. However, repeated cycles of excessive alcohol consumption and withdrawal may actually exacerbate the imbalance in these valence circuits, thus contributing to a worsening of PTSD symptoms and an increased likelihood of developing AUD. For example, animal studies have shown that alcohol effectively decreases BLA excitability and reduces anxiety-like behaviors (McCool, Christian, Diaz, & Lack, 2010; Weiner & Valenzuela, 2006), which may initially quell the negative affective symptoms associated with PTSD. However, repeated cycles of intermittent alcohol exposure result in robust increases in BLA activity that play a causal role in the anxiogenic behaviors that develops during withdrawal (McGinnis, Parrish, Chappell, Alexander, & McCool, 2020; McGinnis, Parrish, & McCool, 2020). These maladaptive changes in the circuits that encode this negative valence information would then promote an even stronger negative affective state during periods of withdrawal, thus exacerbating PTSD symptoms and strengthening motivation to drink to relieve this aversive state.

The VN model also provides a neural framework that may explain the profound heterogeneity in symptom profiles of people with AUD and PTSD as well as individual differences in the severity of these disorders. Given that the VN encodes the valence and salience of all sensory and affective stimuli, there are likely distinct circuits within this network that are responsible for specific symptom clusters of AUD and PTSD, such as anhedonia and anxiety-like behaviors. Indeed, preclinical animal studies have actually identified circuits that encode these behaviors (Hoflich et al., 2019; Janak & Tye, 2015). The clinical presentation of these disorders likely depends on the specific connections within the VN that are dysregulated and may vary depending on the life experience and genetic makeup of each individual. This model can also provide a neural basis for differences in vulnerability and resilience to these disorders. Environmental influences, like early life stress, as well as genetic factors, may compromise or strengthen the resilience of the VN, making it more or less susceptible to the deleterious neural adaptations associated with subsequent exposure to trauma and/or excessive alcohol drinking.

The VN can also reconcile functional neuroimaging datasets that are difficult to interpret using traditional reward and stress circuit constructs. For example, as noted earlier, individuals with AUD often exhibit stronger VS activation in response to alcohol-related cues but show blunted VS reactivity to non-alcohol-related positive stimuli (Schacht et al., 2013; Wrase et al., 2007). Similarly, PTSD patients also commonly show dampened VS responses to cues with positive valence but in some cases, these subjects actually show heightened VS reactivity to aversive stimuli (Elman et al., 2009, 2018). The VN provides a framework to outline simple and testable hypotheses to account for these seemingly discordant findings. While some VN circuits seem to broadly encode the valence of sensory or affective stimuli, others may process a narrower range of this information. It seems reasonable to hypothesize that distinct positive valence circuits may encode the positivity of alcohol-related cues vs. other positive cues that are unrelated to alcohol. If so, AUD may increase salience encoding within the circuitry that processes the positive valence of alcohol-related stimuli while simultaneously decreasing salience encoding in pathways that process non-alcohol-related positively valenced stimuli. Similarly, since it is well-established that many nodes of the VN receive convergent, non-overlapping positive and negative valence information (Tye, 2018), it seems likely that PTSD could diminish the salience of a reward-related signal, while simultaneously increasing the salience of a pathway that encodes aversion. Indeed, recent data suggest that VN neurons that process positive and negative valence are functionally connected by GABAergic interneurons, and thus mutually antagonistic (Kim, Pignatelli, et al., 2016; Zhang et al., 2020).

Finally, we suggest that the VN framework may explain some of the difficulties associated with the development of better treatments for AUD and PTSD and offer new insight into the reasons why novel therapeutic approaches, such as neuromodulation, may prove to be particularly effective in treating these disorders. First, given the complexity of the VN and the broad range of sensory and affective VN circuit imbalances that can potentially give rise to these disorders, it seems highly unlikely that any single treatment could prove effective in all cases. Rather, by gaining a better understanding of the VN elements responsible for specific AUD/PTSD symptoms, we may be able to better tailor treatments to mitigate these distinct neural alterations. More importantly, functional imaging studies have clearly established that the VN processes valence in a highly distributed manner (Lindquist et al., 2016). Although preclinical studies focus primarily on monosynaptic valence circuits, it is clear that these are simply components of a much broader, highly integrated network. Given that AUD and PTSD are associated with widespread alterations in functional connectivity throughout the VN, treatments that could have a distributed effect on VN circuit function may prove to be particularly effective in treating these disorders. As noted earlier, several studies have already demonstrated that real-time fMRI neurofeedback can be used to modulate critical VN nodes, like the AMY, to relieve PTSD symptoms. Indeed, the VN heuristic can readily account for the findings that both decreasing AMY reactivity to aversive stimuli and increasing AMY activity in response to positively valenced cues led to an improvement in PTSD symptoms (Gerin et al., 2016; Misaki et al., 2018; Nicholson et al., 2017). Moreover, the observations that targeting one node of the VN could have distributed effects on VN functional connectivity independent of that node, and that some of these changes actually correlated with improvement of depressive and hyperarousal PTSD symptoms (Misaki et al., 2018), provides encouraging support for the notion that rt-fMRI neurofeedback, and other non-invasive neuromodulation techniques, like exposure therapy and transcranial magnetic stimulation (TMS), may prove particularly effective in treating VN circuit dysregulation in AUD and PTSD. Indeed, TMS is already FDA-approved to treat other affective disorders, like depression and OCD, and there is growing evidence that it may also be an effective treatment for AUD (Diana et al., 2017; Philip, Sorensen, McCalley, & Hanlon, 2020) and PTSD (Kozel, 2018; Trevizol et al., 2016). Future studies are certainly warranted to explore the efficacy of these neuromodulation therapies in the treatment of the comorbid condition.

Despite the advantages that the VN framework offers, there are several major challenges that must be addressed before we are likely to see recent advances in our understanding of the neurobiology of AUD and PTSD translated into better treatments for these disorders. First, there are many discordant findings in the preclinical literature regarding the anatomical segregation of the cells and circuits that encode the valence and salience of sensory and emotional stimuli. While there is a general consensus about the specific brain regions responsible for these functions (Beyeler et al., 2018; Namburi et al., 2016; Peters et al., 2016; Thibeault et al., 2019; Tye, 2018; Zhang & Volkow, 2019), there is considerable disagreement as to the anatomical localization of positive and negative valence encoding cells within these structures. For example, several BLA valence circuit-mapping studies have shown that there are distinct populations of cells that encode positive and negative valence that are largely segregated along the anterior/posterior axis of this brain region (Kim, Pignatelli, et al., 2016; Pi et al., 2020; Wang et al., 2018; Zhang et al., 2020). However, other studies, using similar sensory or affective stimuli, suggest that positive and negative valence cells are intermixed throughout the BLA (Beyeler et al., 2018; Shen et al., 2019). Even more baffling, studies that have reported anatomical segregation of BLA valence encoding cells do not always agree on which subregion encodes positive vs. negative valence (Kim, Pignatelli, et al., 2016; Wang et al., 2018). While there are likely many reasons underlying these disparate findings, such as methodological differences and the inherent difficulty associated with the study of emotion-like behaviors in animals, it will be important to resolve the actual anatomical location of the neurons and circuits responsible for processing sensory and affective valence.

A second challenge is that, although significant progress has been made in identifying changes in VN functional connectivity that contribute to the etiology of AUD and PTSD, our understanding of the neurobiological mechanisms responsible for these alterations is far from complete. While we have advanced the idea that this circuitry be viewed as a VN, we have emphasized that this network not only encodes the valence of sensory and emotional stimuli but also determines the relative salience or importance of this input, likely via modulation of neural activity within these circuits. We have also noted that incongruent encoding of the salience of valence signals likely plays an integral role in the pathophysiology of both disorders. While animal research has identified numerous neurotransmitters and neuropeptides that can modulate the signaling strength of many of the circuits within the VN (Thibeault et al., 2019; Tye, 2018), how such changes actually contribute to alterations in VN functional connectivity in AUD and PTSD is largely unknown. These neuromodulatory systems that govern the salience of VN signals may well represent some of the most promising targets for the development of novel treatments for these disorders.

Finally, for these challenges to be overcome, it is imperative that we improve scientific interaction and communication between researchers that employ human and non-human subjects to elucidate the specific changes in VN functional connectivity that give rise to AUD and PTSD. As alluded to above, animal researchers typically focus on monosynaptic projections that function in the millisecond to second range whereas human neuroimaging researchers study a much slower timescale and characterize networks that are functionally, but not necessarily structurally, connected. While the alterations in monosynaptic circuits being examined in animal models of AUD and PTSD likely give rise to the changes in functional connectivity observed in human studies of individuals afflicted with these conditions, better techniques and methods will be needed to bridge these fields.

12. Conclusions

In summary, this chapter has provided a review of recent literature on the epidemiology and neurobiology of comorbid AUD and PTSD. Collectively, the available data continue to show that there is considerable overlap in the core symptoms associated with these disorders, that these diseases frequently co-occur, and that this dual diagnosis is associated with increased symptom severity and poor treatment outcomes. Although the comorbid condition remains a major global public health concern, groups such as members of the military, war veterans, and individuals exposed to early life stress are particularly vulnerable to these disorders. Focusing primarily on evidence from human functional imaging studies, we have highlighted numerous maladaptive changes in brain reward and stress networks that are common to both disorders. We note, however, that these recent advances have not led to better treatments for the comorbid condition. We suggest that canonical brain reward and stress circuits do not accurately depict the strong, functional integration between these networks and that the continued use these frameworks to interpret clinical and preclinical findings on AUD and PTSD may be limiting our ability to translate recent discoveries into more effective therapies. Instead, we suggest that reward and stress circuits be viewed as an integrated valence network that functions in a cohesive manner to assign a degree of positivity or negativity to all sensory and emotional stimuli. This network also encodes the relative salience or level of importance of these hedonic signals in order to guide behavioral responding that is congruent with this input. Using this heuristic, we suggest that AUD and PTSD may be viewed as valence network imbalance disorders. We highlight a large volume of data from both human functional imaging and animal circuit-mapping studies that provide empirical support for this framework and discuss why this model provides a more parsimonious way of interpreting much of the recent epidemiological and neurobiological data on the comorbid condition. Finally, we propose that emerging noninvasive neuromodulatory therapies, like rt-fMRI NF and TMS, may be particularly effective at restoring the functional perturbations in valence network connectivity that give rise to these disorders.

Abbreviations

AMY

amygdala

ACC

anterior cingulate cortex

aHC

anterior hippocampus

AUD

alcohol use disorder

BLA

basolateral amygdala

BOLD

blood-oxygen-dependent-level

CEA

central amygdala

Cort

corticosterone

DMN

default mode network

DSM

diagnostic and statistical manual of mental disorders

DA

dopamine

dACC

dorsal anterior cingulate cortex

dlPFC

dorsal lateral prefrontal cortex

dmPFC

dorsal medial prefrontal cortex

ELS

early life stress

EEG

electroencephalography

ECA

epidemiological catchment area

FHN

family history negative

FHP

family history positive

fMRI

functional magnetic resonance imaging

GAD

generalized anxiety disorder

HC

hippocampus

ICD

international classification of disease

KOR

kappa opioid receptor

LTP

long term potentiation

MDD

major depressive disorder

mPFC

medial prefrontal cortex

mCC

middle cingulate cortex

NESARC

national epidemiological survey on alcohol and related conditions

NMDA

n-methyl-d-aspartate

NE

norepinephrine

NAc

nucleus accumbens

OCD

obsessive compulsive disorder

OFC

orbital frontal cortex

PET

positron emission tomography

pACC

posterior anterior cingulate cortex

PFC

prefrontal cortex

pgACC

pregenual anterior cingulate cortex

PTSD

post-traumatic stress disorder

PKA

protein kinase A

Rt-fMRI NF

real-time fMRI neurofeedback

RSFC

resting state functional connectivity

RIP

resident intruder procedures

SEFL

Stress enhanced fear learning

SN

salience network

SPECT

single-photon emission computed tomography

SPS

single prolonged stress

TMS

transcranial magnetic stimulation

VN

valence network

vHC

ventral hippocampus

vmPFC

ventral medial prefrontal cortex

VS

ventral striatum

VTA

ventral tegmental area