Psychological and cognitive theories and experimental evidence that implicated various aspects of reinforcement-biased cognition in the development, maintenance, and recurrence of alcohol addiction
Several psychological theories sought to describe reinforcement-biased processes that are implicated in the development and progression of AUD (Skinner & Aubin, 2010). According to the allostatic model (Koob, 2003; Koob & Volkow, 2010), the early stages of alcohol addiction are primarily driven by positive reinforcement, which generally refers to the pleasant feelings and social enhancement effects that alcohol provides, while the later and more severe stages are driven by negative reinforcement, which encompasses relief from negative affective states such as stress or anxiety. The second model, the incentive-sensitization theory (Robinson & Berridge, 1993), highlighted increasing subjective ‘wanting’ for alcohol that is driven by hypersensitization of the dopaminergic system, while a related theory postulated by Field & Cox (2008) proposed that sensitization of the dopaminergic system drives alcohol attentional biases related to alcohol-related cues, leading to an increased craving. Finally, according to Cox & Klinger (2011), some individuals drink alcohol for its positive reinforcing properties to increase their positive affective experience (e.g., mood enhancement), while others drink alcohol for its negative reinforcing properties to dampen their negative emotions and cope with distress and anxiety (tension reduction). The former, whose drinking is maintained by the rewarding effects of alcohol, are called ‘reward drinkers’, while the latter are called ‘relief drinkers’ (Ooteman, Koeter, Verheul, Schippers & Van den Brink, 2006). Based on accumulating experimental evidence, it seems that there may be distinct types of AUD in which alcohol abuse is differentially driven by various mixtures of the above-mentioned reinforcement-based processes.
Individual differences in vulnerability to substance use disorders, including AUD, may be explained in light of reinforcement sensitivity theory (Corr, 2004). According to this hypothesis, every behaviour is governed by two conceptual brain systems: the behavioural approach system (BAS) and the behavioural inhibition system (BIS). These two systems differentially respond to the rewarding and punishing effects of alcohol. The BAS underlies approach motivation, positive affect, and reward learning processes and may contribute to the acquisition of AUD, while the BIS inhibits behaviour in response to stimuli signalling the loss of expected reward, uncertainty, and goal conflict and may support the maintenance of AUD (Corr, 2008). The BIS also gives rise to emotional distress and has been associated with trait negative affect, especially anxiety (McNaughton & Gray, 2000). Research has suggested that the BAS can be further subdivided into three different types of reward sensitivity: (a) reward drive, relating to persistent pursuit of desired goals; (b) fun-seeking, relating to simple biological reinforcers that do not require planning; and (c) reward responsiveness, encompassing responses related to receiving or anticipating reward (Carver & White, 1994; Corr, 2008). As alcohol consumption represents a high reward value (Everitt & Robbins, 2005), it is not surprising that several studies demonstrated a relationship between positive reinforcement sensitivity, particularly in relation to fun-seeking, and higher alcohol intake, particularly in the form of binge drinking (Feil & Hasking, 2008; Franken & Muris, 2006; Loxton & Dawe, 2001; O’Connor & Paley, 2009; Voigt et al., 2009). It was also found that BAS sensitivity was related to both desire and negatively reinforcing aspects of alcohol craving. Subjects with high BAS sensitivity scores experienced significantly stronger desires and intention to drink alcohol and negative reinforcement craving during exposure to alcohol-related cues than subjects with low BAS sensitivity scores (Franken, 2002). Furthermore, persons with high BAS sensitivity experienced high negative reinforcement craving during this exposure (Franken, 2002). The relationship between BIS sensitivity and alcohol consumption is less clear. Although several investigators reported null associations between BIS and AUD (Hundt, Kimbrel, Mitchell & Nelson-Gray, 2008; Kambouropoulos & Staiger, 2001; O’Connor & Colder, 2005), other studies suggested that the BIS, instead of inhibiting behaviour, draws attention to the potential dangers of a situation and functions as a conflict resolution system (Corr, 2008)). As a result, a more sensitive BIS leads to high anxiety, which could be reduced by indulging in drinking (Hasking, 2006). Indeed, it has been explicitly proposed that a sensitive BIS acts anxiogenically in response to health information, which may motivate protective health-related behaviours (Norman, Boer & Seydel, 2005). By the same token, a more sensitive BIS may act as a protective factor due to the avoidance of potentially risky situations or aversive consequences (e.g., hangovers). This theory has been supported by several other studies (Kimbrel, Nelson-Gray & Mitchell, 2007; O’Connor, Stewart & Watt, 2009; Pardo, Aguilar, Molinuevo & Torrubia, 2007) who found a negative correlation between the sensitivity of the BIS and the frequency and quantity of alcohol consumption. Similar results were reported by Knyazev and collaborators (2004), who found that a more sensitive BIS protects against substance abuse among youths. Taken together, it seems that the effects of BAS and BIS on the trajectories of alcohol addiction might be interactive (Figure 1). Indeed, several experiments demonstrated that high BAS sensitivity coupled with low BIS sensitivity was associated with significantly increased alcohol use (Kellough, Beevers, Ellis & Wells, 2008; Wardell, O’Connor, Read & Colder, 2011). Moreover, Wardell and colleagues (2011) demonstrated high BIS sensitivity as a risk factor for subsequent problematic drinking, but only when combined with high BAS sensitivity. When the sensitivity of BAS was low, BIS sensitivity was demonstrated to be protective against subsequent drinking. The above-mentioned studies suggested that the relationship between BIS and problematic drinking is moderated by BAS: in the absence of a sensitive BAS to shift attention towards the rewarding, tension-reducing properties of alcohol, the aversive consequences of drinking are more salient, leading to alcohol avoidance (Wardell, O’Connor, Read & Colder, 2011).
The objective investigation of BAS and BIS sensitivity in people abusing alcohol can be conducted using a framework of neurophysiological and neuropsychological measures. The neurophysiological measures are mostly based on the peripheral assessment of cardiac and electrodermal reactivity during appetitive responding for reward and during extinction of appetitive responding (i.e., frustrative non-reward), respectively (Beauchaine, 2001; Fowles, 1988; Iaboni, Douglas & Ditto, 1997; Tranel, 1983). In turn, the neuropsychological evaluation is usually conducted using self-report questionnaires that assess individual differences in BAS and BIS sensitivity by the degree to which respondents endorse prototypical approach- and avoidance-related behaviours. This strategy is exemplified in the BIS/BAS scales of Carver and White (1994), which were developed with the explicit purpose of assessing individual differences in state reactivity of these systems. Recently, investigators have developed computerized neuropsychological tests to experimentally measure sensitivity to reinforcement (Cools, Clark, Owen & Robbins, 2002; Evers et al., 2005; Paulus, Hozack, Frank & Brown, 2002; Paulus, Hozack, Frank, Brown & Schuckit, 2003; Slaney, Hinchcliffe & Robinson, 2018). These tests offer many advantages over conventional pencil-and-paper testing because they standardize aspects of administration and automate data collection and analysis. One of these tests is the probabilistic reversal learning (PRL) test, offering an effective way of measuring an individual’s sensitivity to reinforcement by assessing win-stay and lose-shift behaviours (WSLS) following rewarding and punishing feedback, respectively. In this paradigm, subjects are presented with two (sometimes more) stimuli on each trial and using trial-and-error feedback after each response, learn to select the stimulus that is usually correct (rewarded on a majority or punished on a minority of trials) and to avoid the stimulus that is usually incorrect (punished on a majority or rewarded on a minority of trials). This rule intermittently reverses such that the stimulus that was usually rewarded becomes usually punished/unrewarded and vice versa. Consequently, responding must be adjusted to gain the reward and avoid punishment. Rewarded outcomes followed by a decision to stay with the response that delivered them (win–stays) constitute a measure of sensitivity to positive reinforcement. Conversely, lose–shift ratios, calculated by dividing punishing outcomes after which the subject changed the choice by the total number of punished trials on a given stimulus, represent a measure of sensitivity to negative reinforcement. The use of probabilistic reinforcement increases the complexity of the task in such a way that the information from any given choice is insufficient to guide future behaviour, and subjects must engage cognitive functions to track the reward and punishment history for both stimuli to determine the stimulus that is more beneficial overall. The PRL paradigm has been recently successfully applied in a number of studies used to investigate the neuroanatomical and neurochemical correlates of reinforcement sensitivity in humans and non-human animals (Rygula, Noworyta-Sokolowska, Drozd & Kozub, 2018). These will be discussed in subsequent sections of this paper.
Individual differences in reinforcement sensitivity may influence the acquisition and maintenance of positively and negatively biased alcohol-drinking outcome expectancies, which can be defined as positively or negatively inflated beliefs about the effects of alcohol on behaviour, cognition, moods, and emotions (Leigh, 1989). These expectations, in turn, have a pivotal role in determining decisions about alcohol drinking and trajectories of its use (Jones, Corbin & Fromme, 2001). Beginning with the seminal work of Brown and her colleagues (Brown, Goldman, Inn & Anderson, 1980; Goldman, 1994), many studies have documented positively biased outcome expectancies for engaging in alcohol drinking, together with minimized negative expectancies and poor self-efficacy or beliefs about one’s ability to cope without alcohol, can maintain addictive behaviours and predict relapse (Brown, Christiansen & Goldman, 1987; Christiansen & Goldman, 1983; Floyd & Widaman, 1995; Fromme & D’Amico, 2000; Fromme, Stroot & Kaplan, 1993). In these experiments, alcohol-drinking outcome expectancies were measured using structured interviews of focus groups and the self-report questionnaires in which individuals endorse each questionnaire item as to whether or not they hold that particular expectancy. A sum of the expectancy endorsements representing the individual’s overall alcohol outcome expectancy is further positively associated with alcohol consumption, e.g., subjects who drank in a frequent, social manner, expected alcohol to enhance their social behaviour (Christiansen & Goldman, 1983; Lee, Maggs, Neighbors & Patrick, 2011). Other studies have investigated the potential associations of different categories of alcohol outcome expectancies with subsequent drinking behaviour. A study by Leeman and collaborators (2009) found significant associations of euphoria and social enhancement expectancies with binge alcohol use, while the study by Pabst and colleagues (2010) found the same association concerning sexual and social relationship enhancement expectancies. Other cross-sectional and longitudinal studies have demonstrated that greater numbers of positive alcohol outcome expectancies are associated with greater numbers of negative consequences independent of the level of consumption (Blume & Blume, 2014; Blume, Lostutter, Schmaling & Marlatt, 2003).
Abnormal perceptions of risks associated with alcohol-drinking outcomes can also be considered within the concept of unrealistic optimism bias (Sharot, Korn & Dolan, 2011), which encompasses two different phenomena: unrealistic comparative optimism and unrealistic absolute optimism (Shepperd, Klein, Waters & Weinstein, 2013). The former refers to people’s tendency to view the risks as lower for themselves than for the others, while the latter refers to unrealistically positive risk assessment when compared to an objective criterion, such as an actual risk assessment. The mechanism that allows individuals to maintain or arrive at unrealistically positive beliefs in the face of disconfirming evidence was described for the first time by Sharot, Korn and Dolan (2011). The results of this study demonstrated that when people update their initial risk estimates, they tend to incorporate desirable information (i.e., information that risks are lower than expected) to a greater extent than undesirable information (i.e., information that risks are higher than expected). In those with AUD, the asymmetric reliance on information about alcohol-drinking outcomes, dependent on their valence, may result in larger updates after desirable/positive information than after undesirable/negative information and in this way affect drinking trajectories. When considering expectancies and beliefs about the effects of alcohol, it is also important to mention the interpretation of ambiguity in drinking outcomes. Indeed, positive/appetitive alcohol-related expectations are often associated with more negative/inhibitory information arising from memories. This conflict and the bias in its interpretation can be assessed using various types of ambiguous-cue interpretation (ACI) paradigms. In this experimental and fully translational approach, participants initially learn to discriminate two stimuli (e.g., tones of different frequencies), which acquire emotional and motivational value due to subsequent feedback (monetary gain or avoidance of monetary loss). After such an acquisition phase, the test phase introduces ambiguous stimuli (e.g., tones of intermediate frequencies) that serve as a measure of interpretation bias since the response to these stimuli indicates the participants’ expectation of rewarding or potentially punishing effects of their decision (Papciak & Rygula, 2017; Schick, Wessa, Vollmayr, Kuehner & Kanske, 2013). Other ways of assessing biased interpretations of ambiguity include using single ambiguous words individually presented to the participant, which are typically either homophones (words that sound the same yet have different spellings and meanings, e.g., meat/meet or sea/see) or homographs (words with identical spelling yet distinct meanings, e.g., bat—animal or wooden club; change—to alter or money) (Drury, 1969; Gorfein & Weingartner, 2008). Homophones and homographs used in these studies usually have both threatening/emotional and neutral interpretations, and an interpretation bias is established as the number or proportion of each type of interpretation made by participants from a list of such words (Hindash & Amir, 2012; Warren, Warren, Green & Bresnick, 1978). Ambiguity interpretation biases can also be investigated using ambiguous images, often with emotional and neutral faces (Beevers, Wells, Ellis & Fischer, 2009; Schaefer, Baumann, Rich, Luckenbaugh & Zarate Jr, 2010).
Another conceptual approach focuses not on expectancies regarding alcohol-drinking outcomes per se but on the mechanisms that underlie them. These mechanisms have been defined as attributional styles, which Alloy and colleagues (1984) described as the manner in which a person explains the causes of prior outcomes. Indeed, following positive or negative experiences with alcohol, people often wonder why the event occurred (DeJoy, 1989; Jessup et al., 2014). To answer this question, they make causal attributions based upon different dimensions: internal-external (whether the event was caused by themselves (internal) or by the situation (external)), stable-unstable (whether the cause of an experience is constant and likely to happen again (stable) or pliable and unlikely to reoccur (unstable)), and global-specific (whether the cause may affect many areas of one’s life (global) or only one area (specific)) (Weiner, 1985). From this perspective, attributing negative outcomes to causes that are internal to the person, stable and global can reflect pessimism, learned helplessness, and addiction vulnerability (Scheier & Carver, 2018). Following the above-mentioned framework, Weiner (1985) and Marlatt & Donovan (2005) proposed that a certain pattern of attribution promotes relapse in substance use. They argued that a combination of internal, stable, and global attributions contributes to maladaptive thought patterns concerning addiction and promotes relapse. This pattern, often experienced by an individual after a return to substance use following a period of self-imposed abstinence from substances, has been named the abstinence violation effect (Curry, Marlatt & Gordon, 1987). Within this model, bleak attributions regarding future reactions to alcohol exposure lead individuals to believe that fighting addiction is pointless and, in turn, to relapse rather than returning to abstinence (Walton, Castro & Barrington, 1994). This conceptual orientation has given rise to assessment techniques that focus on measuring the causes that people identify for the outcomes that they experience. Typically, respondents are asked to imagine being in outcome scenarios and to indicate how they would explain the described outcomes (Peterson, Semmel, Von Baeyer, Abramson, Metalsky & Seligman, 1982) or to analyse naturally occurring verbatim materials (e.g., newspaper articles or speeches) for explanatory style (Schulman, Castellon & Seligman, 1989).
Taken together, these studies consistently support the notion that various types of RBCBs, such as abnormal sensitivity to performance feedback, biased interpretation of ambiguity, inflated expectations, and asymmetrical belief updating, can be differentially associated with drinking at different stages of AUD (Figure 2). Understanding the underlying mechanisms could provide useful information for tailoring prevention and intervention efforts. For example, medications used in the early stages of addiction should target positive reinforcement sensitivity, while treatments that target alcohol-dependent individuals in the late stages of addiction may need to simultaneously target positive and negative reinforcement processes. In other words, understanding individuals’ positive and negative reinforcement profiles and resulting cognitive biases can provide important information for personalized medicine. Indeed, there is preliminary evidence for the effectiveness of tailoring pharmacotherapies to AUD patients based on their positive and negative reinforcement tendencies (Mann et al., 2018; Roos, Mann & Witkiewitz, 2017). For example, an opioid receptor antagonist, naltrexone, has been found to be particularly effective for reward drinkers, while acamprosate, which has been shown to down-regulate the glutamatergic system, is particularly effective for relief drinkers (Roos, Mann & Witkiewitz, 2017). Specific neurochemical correlates of the interactions between RBCBs and AUD, as well as potential future treatment targets, are described in the next sections.