Behavioral Predictors of Individual Differences in Opioid Addiction Vulnerability as Measured Using i.v. Self-Administration in Rats

Abstract

Background:

Like other forms of psychopathology, vulnerability to opioid addiction is subject to wide individual differences. Animal behavioral models are valuable in advancing our understanding of mechanisms underlying vulnerability to the disorder’s development and amenability to treatment.

Methods.

This review provides an overview of preclinical work on behavioral predictors of opioid addiction vulnerability as measured using the intravenous (i.v.) self-administration (SA) model in rats. We also highlight several new approaches to studying individual differences in opioid addiction vulnerability in preclinical models that could have greater sensitivity and lead to more clinically relevant findings.

Results and Conclusions.

Evidence for the relationship between various behavioral traits and opioid SA in the preclinical literature is limited. With the possible exceptions of sensitivity to opioid agonist/withdrawal effects and stress reactivity, predictors of individual differences in SA of other drugs of abuse (e.g. sensation-seeking, impulsivity) do not predict vulnerability to opioid SA in rats. Refinement of SA measures and the use of multivariate designs and statistics could help identify predictors of opioid SA and lead to more clinically relevant studies on opioid addiction vulnerability.

1. Introduction

Characterizing personality and behavioral traits contributing to individual differences in vulnerability to addiction and other psychiatric disorders is essential for developing a greater understanding of underlying genetic and molecular mechanisms, as well as more effective preventions and treatments. Given substantial individual variability in its vulnerability and severity (American Psychiatric Association, 2013; Belin et al., 2016; Vowles et al., 2015), opioid addiction is a prime example of a disorder that could benefit from a clearer understanding of these mechanisms. Despite the enormous toll of opioid addiction on public health, most behavioral predictors of vulnerability to addiction established with other drugs of abuse are less well established when it comes to opioids, both in humans and animal models. Given differences in the neurobiological effects (e.g. receptor pharmacology and drug-induced synaptic and structural plasticity) between different classes of drugs of abuse (Badiani et al., 2011; Ettenberg et al., 1982; Pettit et al.,1984), it is important to characterize vulnerability factors that are specific to opioid addiction.

Preclinical studies provide a number of advantages over human studies for studying individual differences in opioid addiction vulnerability. First, animal behavioral studies allow examination of vulnerability factors in a controlled environment, thereby minimizing the number of extraneous variables (e.g., other mental disorders) that may confound findings. Second, researchers have control over subjects’ drug-exposure history, allowing for isolation of the factors uniquely associated with effects of opioids versus other drugs. Third, animal models allow experimental, as opposed to quasi-experimental or cross-sectional study designs, thereby shedding light on causal relationships between variables that could not be identified based on human studies alone. Finally, animal models can utilize invasive techniques to characterize neurobiological and genetic mechanisms underlying addiction vulnerability (Parker et al., 2014).

In preclinical research, various behavioral models have been developed in an effort to operationalize human personality traits implicated in addiction vulnerability (e.g., impulsivity, novelty-seeking, etc.) The purpose of this review is to evaluate the utility of such measures in predicting opioid addiction vulnerability as measured using the self-administration (SA) paradigm in rats. For several reasons, drug SA is often considered a model with an especially high degree of translational utility. First, while other animal models of addiction (e.g., conditioned place preference, locomotor sensitization) involve experimenter-administered drug, the SA model involves volitional drug-taking, as occurs in human. Second, various SA measures capture different elements of human addictive behavior such as the initiation of drug use (acquisition), loss of control over drug use (escalation), and relapse (reinstatement) (Table 1) (Belin et al., 2008; Grebenstein et al., 2013; McNamara et al., 2010; Leri et al., 2004; Sorge et al., 2005). Moreover, the SA model has some degree of face and predictive validity in modeling opioid addition. For example, there was a close correspondence between the abuse liability of 23 opioid-related drugs in the rat SA model and their positive subjective effects and/or abuse potential in humans (O’Connor et al., 2010). Therefore, despite the limitations in construct validity of any single animal model of human psychopathology (Geyer & Markou, 2000), opioid SA is often considered an appropriate model for studying opioid addiction in animals.

Table 1.

Measures of SA

Stage of Addiction	SA Model	Operational Measure	Example Study
Initiation of drug use	Acquisition	Average number of infusions earned during first days of drug SA	Belin et al., 2008; Nishida et a., 2016; Smith et al., 2015; Suto et al., 2001
Reinforcing efficacy of drug	Progressive ratio schedule of reinforcement; Behavioral economics	Breakpoint, or the highest fixed ratio at which the animal maintains responding for drug; Elasticity of demand or essential value	Hodos, 1961; Katz, 1990; Richardson & Roberts, 1996; Grebenstein et al., 2013; Swain et al., 2018; Stafford et al., 2019
Loss of control over drug use	Escalation	Increase in number of infusions earned after duration of daily access to drug is extended	Kitamura et al., 2006; Edwards & Koob, 2013; Ahmed & Koob, 1999
Drug use despite negative consequences	Resistance to punishment	Reduction in drug SA when infusions are accompanied by aversive consequence (e.g., foot shock)	Deroche-Gamonet et al., 2004 Belin et al., 2008
Relapse to drug use following exposure to drug-associated environmental cues, stress, or the drug itself	Cue-/stress-/drug-induced reinstatement	Increase in drug-seeking (active lever pressing) following extinction of SA and exposure to drug-associated cue stimuli, stress (e.g., foot shock), or non-contingent injection of previously self-administered drug	Childress et al., 1993; Epstein et al., 2006; McNamara et al., 2010; de Wit, 1996; Banna et al., 2010; Sinha, 2001;

Our primary focus will be on studies of outbred rats, which have been most commonly used and which show significant individual variability in both drug SA itself and in its behavioral predictors (Parker et al., 2014). Studies of inbred or selectively bred strains will also be discussed to provide further insights on the relationships between certain behavioral predictors and opioid SA propensity. We conclude that few reliable behavioral predictors of opioid SA have been identified. We therefore propose several strategies for assessing and analyzing the relationship between predictor variables and the severity of opioid SA that may help uncover more robust behavioral phenotypes for elucidating the substrates of opioid addiction in people. Such approaches could also help improve the validity and sensitivity of the opioid SA model in general.

2. Behavioral Predictors of Opioid Addiction Vulnerability

As described below, a number of traits have been evaluated as putative predictors of vulnerability to opioid SA in rats.

2.1. Impulsivity

Impulsivity refers to the tendency to engage in premature and suboptimal behaviors (Bardo, 2013; Kurth-Nelson & Redish, 2010). Most facets of impulsivity can be categorized as forms of either impulsive action (difficulty inhibiting or controlling behavior), or impulsive choice (preference for small, immediate rewards over larger, delayed rewards) (Baldacchino et al., 2015; Swann et al., 2009).

2.1.1. Clinical findings

Higher impulsivity has been associated with a higher risk of opioid addiction in some clinical studies (Marino et al., 2013; Nielsen et al., 2012; Vest et al., 2016), while another found no relationship between impulsivity and heroin use (Ahn & Vassileva, 2016). These inconsistent findings may stem from a variety of factors including differences in the population studied, the measure or subscales of impulsivity evaluated (e.g., attentional versus motoric impulsivity), and/or history of opioid and other drug use. Additionally, high impulsivity could be a consequence of opioid exposure rather than a predisposing trait (Baldacchino et al., 2015). As such, the extent to which impulsivity predisposes an individual to opioid addiction is unclear.

2.1.2. Preclinical findings

Preclinical studies have not found an association between trait impulsivity and opioid SA. Individual differences in impulsive action measured using the 5-choice serial reaction time task (5-CSRTT, see Table 2 for a description of this and other behavioral measures evaluated as predictors of opioid SA) did not predict subsequent acquisition, escalation, or cue-induced reinstatement of heroin SA (40 μg/100 μl) under a fixed ratio (FR) schedule of reinforcement in rats (McNamara et al., 2010). Similarly, there was no relationship between impulsive choice in a delayed reward procedure (see Table 2a) and several measures of heroin SA (100 μg/kg/infusion; FR1, 2 and 4) in rats including acquisition, breakpoint during progressive ratio testing (i.e., reinforcing efficacy), or drug seeking during extinction or cue- or drug-induced reinstatement (Schippers et al., 2012). This contrasts with the positive relationships between impulsivity in 5-CSRTT and delayed-reward procedures and SA of other drugs of abuse such as cocaine and nicotine (Belin et al. 2008; Diegaarde et al., 2008; Perry et al., 2005; Anker et al., 2009). Nevertheless, when rats in the Schippers et al. (2012) study were tested on the delayed-reward task again after completion of heroin SA, those with a history of heroin SA showed increased impulsivity compared to baseline. Another study found no effects of experimenter-administered heroin on impulsivity (Harty et al., 2011). However, a more recent study found that experimenter-administered morphine increased short-term motor impulsivity in adolescent, young adult, and adult rats, and increased long-term motor impulsivity (i.e, following a 25 day drug-free period) in adolescents (Moazen et al., 2018).

Table 2a.

Measures of behavioral traits as predictors of opioid SA vulnerability

Predictor	Behavioral Model	Description	Study on Opioids	Conclusion
Impulsivity	5-Choice Serial Reaction Time Task (5-CSRTT)	Response to light signal to obtain food after an inter-trial interval. Premature responses are punished	McNamara et al., 2010	No difference between high-and low-impulsivity rats in heroin SA
	Delayed Reward Training	Choice between small, immediate reward or delayed, bigger reward	Schippers et al., 2012	No difference between high- and low-impulsivity rats in heroin SA. Heroin SA increased impulsive responses
Sensation-seeking	Spontaneous Locomotor Activity	Amount of exploratory activity in a novel open-field chamber	Ambrosio et al., 1995	Inbred rat strains with higher activity levels showed greater morphine SA
			Swain et al., 2018, 2020	No relationship between activity levels and morphine SA in outbred rats
Anxiety	Elevated Plus Maze (EPM)	Amount of time spent on the open arms (no walls) of the maze	Dilleen et al., 2012	No difference between high-and low-anxiety rats in heroin SA
	Thigmotaxis	Amount of time spent in the periphery of an open-field chamber	Swain et al., 2018, 2020	Thigmotaxis did not correlate with later morphine SA
Stress reactivity	Forced Swim Test (FST)	Number of attempts to climb out of the testing container	Stafford et al., 2019	Climbing behavior positively predicted subsequent demand for heroin SA

Overall, these preclinical studies suggest that impulsive behavior may be an effect of opioid exposure rather than a preexisting vulnerability trait for addiction, as has been suggested in humans (Baldacchino et al. 2015). The findings of Moazen et al. (2018) further suggest that the effects of adolescent opioid exposure on impulsivity may be long-lasting. Such enduring effects could contribute to the difficulty in parsing cause from effect in human studies evaluating the role of impulsivity in opioid addiction vulnerability.

2.2. Sensation seeking

Sensation seeking refers to the tendency to attain novel and intense experiences despite risks (Zuckerman, 1994). Sensation seeking has been associated with other addiction-related traits such as impulsivity (Hur & Bouchard,1997; Krueger et al., 2002), and there is some overlap in how these traits are defined (Ahn & Vassileva, 2016; Whiteside & Lynam, 2001).

2.2.1. Clinical findings

Despite some human studies showing a positive relationship between sensation seeking and opioid addiction vulnerability (Franques et al., 2003; Kosten et al., 1994; Vest et al., 2016), others have shown either no relationship (Conrod et al., 2000; Marino et al., 2013; Nielsen et al., 2012) or a negative relationship (Ahn & Vassileva, 2016). These inconsistencies may reflect the same general limitations of human studies described above.

2.2.2. Preclinical findings

Two tests have been developed to model sensation seeking in rats. The first uses spontaneous locomotor activity in a novel environment as a measure of novelty seeking — a dimension of sensation seeking (Blanchard et al., 2009; Pawlak et al., 2008; Piazza et al., 1989). Higher locomotor activity reliably predicts greater SA of psychostimulants (e.g., cocaine, amphetamine), particularly in terms of acquisition (Piazza et al., 1989; Piazza et al., 2000). These findings are consistent with studies showing a positive relationship between sensation seeking and psychostimulant use in humans (Ahn & Vassileva, 2016; Nielsen et al., 2012).

Only limited data speak to the relationship between spontaneous locomotor activity and opioid SA vulnerability. Inbred rat strains with higher locomotor activity also exhibited greater acquisition of morphine SA (1 mg/kg, FR 1) under certain conditions compared to other strains (Ambrosio et al., 1995). However, we found that locomotor activity did not predict individual differences in morphine SA vulnerability in outbred rats using two different morphine doses (0.5 or 0.2 mg/kg/infusion) and durations of daily access (4 or 2 hours) (Swain et al., 2018; Swain et al., 2020). The cause of this discrepancy may be related to genetic differences between inbred and outbred rats (Cadoni et al., 2015; Chaouloff et al.,1995; Meyer et al., 2010).

Another model for sensation seeking in rats focuses on preference for novelty, measured using a choice task, rather than reactivity to novelty (Belin et al., 2008; Belin et al., 2011; Belin & Deroche-Gamonet, 2012). However, this factor has not been studied in the context of opioid SA vulnerability. Thus, the relationship between opioid SA vulnerability and sensation-seeking, measured using either spontaneous activity or novelty preference, has received only limited attention and requires further elaboration.

2.3. Anxiety

2.3.1. Clinical findings

The self-medication hypothesis of addiction (Khantzian, 1987) posits that individuals experiencing greater anxiety are more likely to choose addictive drugs with anxiolytic properties, such as opioids (Khantzian, 1987; Markou et al.,1998). The fact that anxiety has been linked to opioid addiction vulnerability in humans is consistent with this hypothesis (Lejuez et al., 2008; Martins et al., 2012; Norton, 2001; Rogers et al., 2018). However, the experience of negative affect including anxiety is common during withdrawal from opioids and other drugs (Koob & LeMoal, 1997). Therefore, as with other putative predictive traits, it is unclear whether anxiety is a predictor of opioid addiction vulnerability, a consequence of chronic opioid use, or both.

2.3.2. Preclinical findings

The relationship between anxiety and opioid addiction has not been well established in animal models. Rats categorized as showing High- versus Low-anxiety based on time spent in the open arms of an elevated plus-maze (EPM; see Table 2) did not differ in their subsequent escalation of heroin SA (40 μg/100 μl/infusion; FR1), and time spent on the open arms of the EPM did not correlate with heroin SA escalation (Dileen et al 2012). Similarly, we have found that thigmotaxis (time spent in the periphery of an open field; Table 2a) does not predict various measures of morphine SA (e.g., acquisition, reinstatement, etc.) using two different morphine unit doses (Swain et al., 2018; 2020). In contrast, anxiety-like behavior in rodents predicts individual differences in SA of drugs other than opioids, such as cocaine (Dilleen et al., 2012; Pelloux et al., 2009; Walker et al., 2009). The role of anxiety in individual differences in opioid SA has not yet been studied via measurement of additional behaviors that more fully capture the multifaceted nature of anxiety and the heterogeneity of anxiety disorders (e.g., conflict or defensive behavior, conditioned fear)(Blanchard et al., 1993; Shekhar et al., 2001).

2.4. Stress Reactivity

2.4.1. Clinical findings

Stress is associated with opioid use both mechanistically and epidemiologically. First, opioids suppress activity of the hypothalamo-pituitary-adrenal (HPA) axis (Goeders, 2007; Facchinetti et a., 1985; Kreek et al., 2005), whereas opioid withdrawal activates the HPA axis (Li et al., 2008). Second, post-traumatic stress disorder (PTSD) and opioid addiction share certain symptoms and are frequently comorbid (Fareed et al., 2013). The link between vulnerability to the effects of stress and to opioid addiction is further supported by clinical studies showing a positive relationship between stress reactivity and opioid use (Back et al., 2015; McHugh et al., 2016). For example, patients with prescription opioid dependence exhibited higher reactivity than controls to acute social stress (Back et al., 2015).

2.4.2. Preclinical findings

Numerous preclinical studies have found that exposure to stressors can elicit or exacerbate opioid addiction-related behavior (e.g., Shaham & Stewart, 1994; Shaham, 1993). However, in the only study to evaluate stress reactivity as a predictor of opioid addiction vulnerability (Stafford et al., 2019), higher behavioral (open-field activity, forced swim test) and hormonal (corticosterone) reactivity to intermittent swim stress were predictive of higher reinforcing efficacy for heroin SA (0.05 mg/kg/infusion) in rats measured using a behavioral economic approach (see below for further discussion of behavioral economics). The involvement of the endogenous opioid system in vulnerability to both stress and drug addiction may underlie this positive relationship (Koob, 2013; Kreek et al., 2005; Piazza & Le Moal, 1996).

2.5. Sensitivity to Acute Drug Effects

2.5.1. Clinical findings

Sensitivity to the initial acute effects of drugs (e.g., euphoria, aversion) has long been recognized as a key predictor of addiction vulnerability to drugs other than opioids (DiFranza et al., 2007; O’Loughlin et al., 2003; Schuckit et al., 2004). For instance, sensitivity to the relaxing effects of tobacco during first exposure is a robust predictor of subsequent nicotine addiction (DiFranza et al. 2007). However, these relationships have not yet been examined in clinical studies on opioids.

2.5.2. Preclinical findings

2.5.2.1. Acute opioid effects

Only one preclinical study has evaluated the relationship between the acute effects of opioids and subsequent opioid SA. That study found that rats with lower sensitivity to the antinociceptive effects of morphine subsequently exhibited greater acquisition of morphine SA at a unit dose of 0.5 mg/kg/infusion under a FR 1 schedule of reinforcement (Nishida et al., 2016). This suggests that reduced sensitivity to the initial analgesic effect of opioids may predict greater opioid addiction vulnerability.

2.5.2.2. Withdrawal effects

In addition to acute effects, opioid injections can also result in opioid withdrawal in both humans and animals. These withdrawal effects, characterized by negative affective (emotional) states such as anhedonia or diminished reward sensitivity, can be induced after only a single opioid exposure (“acute” dependence) (Harris & Gewirtz, 2004; Schulteis et al., 2004) and often become more severe with repeated drug exposures (Engelmann et al., 2009; Harris et al., 2004; Schulteis et al., 2004). Avoidance of severe withdrawal effects following prolonged drug exposure may serve as a key motivational force driving compulsive drug-taking (Koob & Le Moal 1997). In contrast, it has been proposed that greater sensitivity to the aversive effects of withdrawal may be a protective trait against addiction to opioids and other drugs (Carroll et al., 2008; Dess et al., 2005; Holtz et al., 2015; O’Dell et al., 2006; O’Dell, 2009). Moreover, anhedonia during opioid withdrawal could potentially reduce the motivation for reward-seeking (Wise, 2004). Consistent with these views, saccharin-preferring rats, which exhibit greater SA of opioids and other drugs compared to saccharin non-preferring rats (Carroll et al., 2002), exhibit lower anhedonia during withdrawal from acute morphine injections as measured by increases in intracranial self-stimulation (ICSS) thresholds (i.e., withdrawal-induced anhedonia, WIA; Table 2b). In outbred rats, we found that greater antagonist-precipitated and spontaneous WIA following acute morphine injections were associated with lower addiction-like behavior on multiple morphine SA (0.2 mg/kg/infusion) measures (e.g., demand, morphine-induced reinstatement) (Swain et al., 2020). Somatic (i.e. physical) withdrawal signs (see Table 2b), in contrast, did not predict any primary morphine SA measure, consistent with the idea that affective rather than somatic withdrawal signs are critical in addiction vulnerability (Schulteis et al., 1994; Baker et al., 2004; Koob & Le Moal, 2005).

Table 2b,

Measures of opioid sensitivity as predictors of opioid SA vulnerability

Sensitivity to acute agonist effects	Hot Plate Test (Analgesic Effect)	Nociception after acute opioid injection	Nishida et al., 2016	Lower morphine-induced antinociception predicted greater morphine SA
Sensitivity to acute withdrawal effects	Somatic Withdrawal Signs	Score on severity of somatic (physical) withdrawal signs (e.g. writhing, wet-dog shakes)	Swain et al., 2020	Somatic withdrawal signs did not predict subsequent morphine SA
	Intracranial Self-Stimulation (ICSS; anhedonia)	Lowest electrical brain stimulation intensity that maintains operant responding	Holtz et al., 2015	Saccharin-preferring rats (high vulnerability) show lower withdrawal-induced anhedonia
			Swain et al., 2020	Higher elevation in ICSS thresholds correlate with lower morphine SA
	Startle Response (Anxiety)	Acoustic startle response during withdrawal	Radke et al., 2013	Saccharin-preferring rats exhibit higher anxiety during withdrawal
	Conditioned Place Aversion (CPA)	Amount of time spent in an environment associated with withdrawal	Radke et al., 2013	Only saccharin-preferring rats develop CPA to morphine withdrawal

Contrary to our findings using a measure of anhedonia, we found that saccharin-preferring (i.e., addiction-vulnerable) rats display higher anxiety during morphine withdrawal compared to saccharin non-preferring rats as measured by potentiated acoustic startle responding (Radke et al., 2013; Table 2b). In addition, only the saccharin-preferring rats developed a morphine withdrawal-induced conditioned place aversion (Radke et al., 2013). Together, these findings suggest that early-stage negative affective withdrawal signs predict opioid addiction vulnerability, although the direction of the relationship depends on the withdrawal sign measured (e.g., measures of anhedonia vs. anxiety). While these early withdrawal signs could not be measured in established drug users to predict treatment efficacy, their further investigation in animal models may be valuable in identifying genetic and neurobiological mechanisms underlying vulnerability to opioid addiction.

2.6. Conclusion

Evidence for the relationship between specific behavioral indices and opioid addiction vulnerability is mixed in human studies, and sparse in preclinical studies. With the possible exception of sensitivity to agonist/withdrawal effects and stress reactivity, none of the aforementioned behavioral traits reliably predict opioid SA in rats. Furthermore, most of the studies described above used only a single opioid SA unit dose and only one schedule of reinforcement (typically FR 1). These limitations, along with the fact that most of these behavioral phenotypes were exclusively examined in male rats, may limit the generalizability of these findings.

3. Translation of Preclinical Research

The fact that behavioral traits associated with addiction to other drugs of abuse (e.g., psychostimulants) have not reliably predicted opioid SA in preclinical studies raises the possibility that unique behavioral phenotypes, such as those associated with opioid exposure (e.g., WIA, see above), predict individual differences in opioid SA. It is also possible that other factors implicated in individual differences in vulnerability to SA of other drugs, such as incentive salience (i.e., the tendency to attribute incentive value to drug-associated cues), could prove to be stronger predictors of opioid SA (Beckmann et al., 2011; Flagel et al., 2009). Alternatively, the issue could be one not of searching for additional measures but of refining existing ones. That is, our metrics for measuring addiction vulnerability using opioid SA, and our means of statistically evaluating their relationship to specific vulnerability factors, may underestimate those factors’ predictive value. We propose that employing one or more novel approaches to behavioral measurement and statistical analysis, some of which have already shown their worth in human studies, may enable us to establish relationships between predictors and outcomes of opioid SA with greater construct and face validity (Geyer & Markou, 1995; Markou et al. 2009; Smith, 2020). These approaches may ultimately also prove beneficial in furthering our understanding of individual differences in addiction to other drugs of abuse, such as stimulants.

3.1. Further Refining the SA Paradigm

3.1.1. Behavioral Economics

Behavioral economics quantifies the extent to which consumption of a reinforcer (e.g., drug) is maintained following increases in its “unit price.” In drug SA models, unit price is operationalized as the cost-benefit ratio of response requirement and unit dose (Bickel et al., 2000; Hursh, 1991; Hursh & Silberberg, 2008). A more rapid decrease in consumption following increases in unit price (lower demand) indicates lower abuse liability, while a slower decrease (higher demand) indicates higher abuse liability. Behavioral economics provides an operationalized and quantifiable measure of reinforcement efficacy that can be used in humans and animals, and has been useful for studying individual differences in demand for numerous addictive drugs (e.g., nicotine) in both species (Diergaarde et al., 2008; Grebenstein et al., 2013; Chase et al., 2013; Hursh & Silberberg, 2008).

A growing body of clinical and preclinical evidence supports the utility of behavioral economics in the study of opioid addiction. For instance, greater demand predicted poorer treatment outcomes for prescription opioid dependence (Worley et al., 2015). Additionally, an exponential demand function closely approximated opioid demand in current and previous opioid users (Strickland et al., 2019), as well as in animals (Stafford et al., 2019; Swain et al., 2018; Swain et al.,2020). Several recent preclinical studies have examined the relationships between predictors of opioid addiction vulnerability (sensation-seeking, withdrawal sensitivity and stress reactivity) and demand, demonstrating the feasibility and utility of this analytical approach in preclinical opioid addiction research (Swain et al., 2018; Stafford et al., 2019; Swain et al., 2020). Since most rodent opioid addiction measures are not directly comparable to measures used in humans, behavioral economics could prove to be a powerful translational tool (Bentzley et al, 2013).

3.1.2. Alternative Addiction Models

Developing behavioral measures that more closely resemble those used to define addiction in humans may also be useful for establishing reliable predictors of opioid SA in preclinical models (Belin-Rauscent et al., 2016). Much as the Diagnostic and Statistical Manual of Mental Disorders (DSM) is utilized to measure addiction in humans (American Psychiatric Association, 2013), Belin and colleagues proposed a checklist approach to classifying addiction behavior in rats (Belin & Deroche-Gamonet, 2012). Three measures of drug SA were selected, each corresponding to a DSM diagnostic criterion, with the total score taken as the addiction score. In theory, other criteria, such as tolerance and withdrawal severity, could be similarly incorporated into such a model.

Alternatively, a number of researchers have modeled addiction in rodents by construing it as a disorder of “choice”; namely, the choice between drugs and non-drug reinforcers such as social interaction, financial stability, etc. (Heyman, 2009; Townsend et al., 2019). This approach emphasizes the availability of non-drug rewards that compete with drugs for the individual’s attention and efforts to obtain. With the flexibility of SA, one can establish reliable procedures to assess individual differences in choice between drug and food or between drug and social interaction (Banks & Negus, 2017; Venniro & Shaham, 2020). In a recent study utilizing these procedures, Townsend and colleagues found that although female rats self-administered more fentanyl than males when it was the only reinforcer, males self-administered more fentanyl when the drug was available with an alternative food reinforcer (Townsend et al., 2019). By establishing choice procedures and examining the relationship between various predictors of individual differences in choice between drug and ethologically valid, non-drug reinforcers, researchers could add to the existing preclinical literature that views addiction primarily as a disorder of disinhibition (Belin et al., 2016; Townsend et al., 2019; Smith & Pitts, 2012).

3.2. Multivariate Designs and Statistics

3.2.1. Utilizing Multivariate Designs in Addiction Research

Studies in which more than one outcome is simultaneously observed and analyzed provide unique information about the clustering and interactions of different factors contributing to the behavioral outcomes of interest. Human studies have long incorporated multivariate designs and corresponding statistical methods to reveal complex relationships between large numbers of predictors and outcome measures of addiction (Krueger et al., 2002; Ahn & Vassileva, 2016; Lynskey & Agrawal, 2007). For example, using elastic net regression, a machine-learning multivariate statistical method, Ahn and Vassileva (2016) identified distinct groups of personality traits associated exclusively with amphetamine or heroin use.

Multivariate study designs also allow us to examine the potential role of a reduced set of unobservable “latent” variables in accounting for variability among a much larger number of related variables. Such variables have greater statistical reliability than individual measures and have been valuable in characterizing core dimensions contributing to addiction vulnerability (Krueger et al., 2002; Lynskey & Agrawal, 2007; Monga et al., 2007). In the human literature, some exploratory factor analysis (EFA) studies have suggested a single latent variable underlying many aspects of opioid addiction (Lynskey & Agrawal, 2007), while others have indicated 2 or 3 factors, each underlying different aspects of the disorder (Monga et al., 2007).

In contrast to EFA, confirmatory factor analysis (CFA) provides a method for testing a priori theory-driven models of factor structure (Schmitt, 2011). This approach has shown that alcohol abuse risk reflects both a general liability to abuse of differing substances along with distinctive influences that give rise to alcohol abuse specifically (e.g., genetic variants in alcohol metabolizing enzymes) (Krueger et al., 2002; Tsuang et al., 1998; Luczak et al., 2006). By adopting EFA and CFA methods, preclinical studies could complement human opioid addiction research in identifying dimensions underlying addiction vulnerability and their biological substrates.

Latent variable analyses would also help in improving the ability of animal models to capture important facets of human behavior and psychopathology. Most animal studies select one or two variables to examine while controlling all others. This approach provides important insight on cause and effect. However, broader application of such insights may be limited by sources of variability specific to each individual trait measure or outcome measure. Such idiosyncrasies may be associated with a specific measure or paradigm (e.g. speed of learning in a conditioning paradigm), or the manner in which that paradigm is conducted in a specific lab. This variability can make it difficult to generalize across different preclinical studies, or from preclinical to clinical studies. Since latent variables capture only the commonality between a number of observed measures (e.g., demand and choice, measured in the same animals), they are relatively impervious to the idiosyncratic and unique variability of each observed measure. Therefore, latent variable analyses can provide more robust information about the relationships between observed measures and underlying constructs.

3.2.2. Multivariate Designs in Preclinical Studies

A recent study used linear mixed-effects modeling to determine the predictability of demand for heroin as a function of one or multiple stress reactivity measures (Stafford et al., 2019). The results showed that models including multiple predictors explained a greater proportion of variance in heroin demand than did any bivariate model, demonstrating the value of incorporating multiple measures. Linear model and factor analytic approaches have begun to be used to identify predictors of individual differences in psychostimulant SA (e.g., Dickson et al., 2015; Marusich et al., 2011; Belin et al., 2008; Deroch-Gamonet et al., 2014). Nevertheless, multivariate designs and statistical methods have been underutilized in preclinical studies of addictive drugs in general and of opioids in particular. In addition to conducting a priori multivariate analyses on novel datasets, existing or published datasets could also be re-analyzed with multivariate statistics when applicable, to further establish latent factors underlying addiction severity, as well as help identify clusters of behavioral traits that are most closely associated with these latent factors.

3.2.3. Small Sample Size Multivariate Analyses

One of the biggest challenges to incorporating multivariate statistics methods in preclinical research is a common reliance on small sample sizes. Estimates for the minimum sample size required for a factor analysis have ranged widely e.g., N = 100–250 (Cattell, 1978; Gorsuch, 1983; Guilford, 1954; Kline, 1979). Most of these recommended sample sizes are impractical for preclinical behavioral studies. This may be a primary reason why between-group approaches have been favored over within-subject, factor-analytic approaches in the preclinical literature. It is encouraging, therefore, to note the incorporation of much larger sample sizes (e.g., >1000 rats) in several recent or ongoing studies in outbred Heterogeneous Stock (HS) rats in order to conduct genomic analyses of complex traits (Gileta et al., 2018; Hughson et al., 2019).

While such studies are unlikely to become commonplace, it is now feasible with advances in quantitative psychology and statistics to conduct factor analysis and structure equation modeling with much smaller sample sizes. For example, the regularization method involves shrinking or adding penalties to specific parameters within the statistical models. The viability of this approach for conducting factor analysis and structure equation modeling using small sample sizes has been demonstrated in both simulation and human studies (Jacobucci et al., 2016; Jung & Takane, 2007; Jung & Lee, 2011). In fact, regularization provides reasonable factor recovery (i.e., how close the sample factor loadings are to population factor loadings) with sample sizes as small as N=5 and N=10 (albeit, less than can be achieved with much larger sample sizes), and yields stable factor loadings when the number of measured variables is large (Jung & Lee, 2011). Greater use of these methods could enable preclinical studies with smaller sample sizes to implement complex models and test hypotheses more comparable and relevant to those tested in clinical addiction research.

4. Overall Conclusions

Despite the prevalence of opioid addiction and the societal burden it imposes, few factors have been associated with opioid addiction vulnerability in humans or animals. By utilizing more clinically relevant measures of drug addiction, and by further adopting and adapting data analytic tools commonly used in human studies, preclinical behavioral studies may further increase the construct and predictive validity of opioid SA and provide new insights into factors associated with vulnerability to opioid addiction, their genotypic and neurobiological basis, and their potential role in prevention and treatment.

Highlights.

• Animal models can provide insights into vulnerability to drug addiction.

• Preclinical evidence for predictors of opioid self-administration (SA) is limited.

• Most behavioral predictors of SA of other drugs do not predict opioid SA.

• Novel behavioral and statistical approaches could identify predictors of opioid SA.