Animal models for addiction medicine: From vulnerable phenotypes to addicted individuals

Abstract

This chapter highlights the use of several animal models of abuse liability. The overall goal is to describe the most frequently used methods, unconditioned behaviors and conditioned behaviors, and how investigators can use these techniques to compare drugs and to understand the mechanisms of action mediating abuse liability. Thus, for each type of animal model described, research will be highlighted on three general features related to the use of the model: (1) determine abuse potential, (2) treatment efficacy, and (3) brain-related changes associated with drug administration.

1. INTRODUCTION

In his brief history of behavioral pharmacology, Roy Pickens (1977) listed two events in the 1940–1955 period that he considered the most significant advancements for the field: the discovery of the antipsychotic effects of chlorpromazine and the hallucinogenic effects of LSD. The former was significant primarily because it advanced the predictive nature of animal models, while the latter was significant for increasing attention on the relationship between biochemistry and behavior and because it led to the study of preclinical models of drug self-administration. The focus of this chapter will be on animal models of addiction and the foundation for these studies can be traced back to the preclinical work on chlorpromazine and LSD. As a minimum, animal models must be predictive of some outcome in people. This predictive nature could be related to models of abuse liability (i.e., is this novel drug reinforcing?) or to potential treatment outcomes (i.e., does drug X decrease drug addiction?); methods for both types of models will be described. In this chapter, two main strengths of animal models will be emphasized: (1) the ability to start with drug–naïve subjects and determine phenotypic/trait characteristics that are associated with addiction and (2) the ability to study the neurochemical, physiological, and pharmacodynamic consequences of chronic drug exposure. Utilizing both of these qualities of animal models is necessary to develop novel treatment strategies for drug addiction.

In animal models of addiction, the primary dependent variable is some behavioral endpoint—whether it is activity level, or time in a quadrant related to a conditioned stimulus (CS), or lever pressing or total drug intake. These dependent variables will be the focus of the studies described in this chapter. In addition, the relationship between behavior and brain will also be described. Although there are many methods used in the literature, this review will highlight the use of in vivo microdialysis, magnetic resonance imaging, and positron emission tomography (PET) in the study of brain–behavior relationships. There are excellent reviews on this topic that will not be repeated in this chapter (see Howell and Murnane, 2011; Murnane and Howell, 2011; Nader and Banks, 2014 for recent reviews). Thus, for each type of animal model described, the goal of this chapter will be to highlight three general features for the use of the model: (1) determine abuse potential, (2) treatment efficacy, and (3) brain-related changes associated with drug administration.

2. TYPES OF ANIMAL MODELS

When assessing animal models for addiction medicine, there are two general categories of models: those that utilize unconditioned behaviors and those that require the study of conditioned responses. While the majority of the chapter will be on conditioned responses, it is important to briefly describe some unconditioned models in order to give researchers a more thorough representation of the breadth of experimental techniques available.

2.1. UNCONDITIONED BEHAVIORS

Perhaps the simplest of behaviors to measure is overall activity in an enclosed environment. These measures can be used as trait markers for vulnerability or as an initial screen for “stimulant-like” drug effects. The best example of using locomotor activity as a trait marker for vulnerability to drug abuse was a study by Piazza et al. (1989) in which rats were first characterized as high responders (HRs) or low responders (LRs) in an open field. When given access to cocaine under a fixed-ratio (FR) 1 schedule of reinforcement, the locomotor HRs were more likely to acquire cocaine self-administration compared to the LRs. This behavioral phenotype has been well characterized in relation to corticosterone (Piazza and Le Moal, 1998; Piazza et al., 1991) and to dopamine (DA) D2-like receptor availability (Dalley et al., 2007). Major strengths of this unconditioned behavior are (1) it requires no training, (2) can utilize large numbers of animals, and (3) provides a quantitative measure that can be used to compare animals on other endpoints including neurochemical and behavioral (related to drug reinforcement, for example).

It has been hypothesized that for stimulant drugs (e.g., cocaine, amphetamine, nicotine) increases in locomotor behavior represent an initial screen for potential abuse liability. These simple procedures involving unconditioned behaviors can be used to better understand the potential mechanisms of action related to drugs of abuse, but they are not models of abuse potential. For example, within the DA D2 receptor family, drugs that act at different subtypes have been identified and these subtypes (D2, D3, and D4) have implications for drugs of abuse. Li et al. (2010) used drug-elicited yawning and locomotor activity in mice to better understand the roles of DA D3 versus D2 receptors, respectively, with the goal of identifying in vivo screens for each receptor subtype that could ultimately lead to medications for drug abuse. Interestingly, Collins et al. (2008) showed that food restriction altered these unconditioned behaviors suggesting an interaction between diet composition and brain function, which could lead to increased or decreased vulnerability to drug abuse. With regard to the DA D4 receptor subtype, Katz et al. (2003) examined the effects of cocaine (1.0–10 mg/kg, i.p.) administered to wild-type (WT) and DA D4 receptor knockout (KO) mice in order to better understand the role of this receptor subtype in the behavioral effects of cocaine. While the two groups did not differ in baseline measures of activity, cocaine administration resulted in significantly larger increases in locomotor activity in the D4 KO mice compared to WT animals. Katz et al. (2003) also found that D4 KO mice were more sensitive to the discriminative stimulus effects of cocaine compared to WT littermates.

Typically, in studies that utilize locomotor activity, other behavioral or physiological measures are examined to more thoroughly characterize the behavioral effects of drugs. For example, Miller et al. (2013) used an immunotherapeutic approach to attenuate the behavioral effects of methamphetamine and examined multiple dependent variables. These investigators reported that vaccination against methamphetamine blocked the effects on locomotor activity, as well as wheel running (another measure of activity) and changes in body temperature, suggesting protection against physiological and behavioral disruptions induced by methamphetamine.

In a recent study, Vanhille et al. (2015) characterized rats using two unconditioned behaviors, novelty-induced locomotor activity and open-arm access in an elevated plus maze, and a conditioned behavior in which lever pressing and head entries into the food magazine during presentation of the CS were used to assess sign tracking and goal tracking, respectively. Interestingly, all the behaviors were characterized as normally distributed but not correlated with each other, indicating independent constructs being assessed. When used to phenotypically characterize vulnerability to cocaine abuse, neither elevated plus maze (high vs. low anxiety) nor sign tracking versus goal tracking (i.e., CS vs. food-maintained lever pressing) was related to a rat’s propensity to acquire cocaine self-administration. However, HRs in the locomotor assay were more likely to choose saccharin over cocaine than LRs, who primarily chose cocaine over saccharin. This interesting finding is at odds with earlier work showing HR rats more vulnerable to cocaine reinforcement when available under an FR 1 schedule of reinforcement (Piazza et al., 1989). Vanhille et al. (2015) suggest that the difference is due to the importance of environmental context in which drugs are self-administered; environmental context has been shown to influence the behavioral effects of drugs under many conditions (see Barrett and Katz, 1981 for review; e.g., Barrett and Stanley, 1980; McKearney and Barrett, 1975). There were important methodological differences in the Vanhille et al. (2015) study compared to earlier saccharin–cocaine choice studies (see Ahmed, 2010) that may have biased initial choice toward cocaine (see comments from Ahmed, 2014). Certainly, there needs to be standardization of protocols in order to better compare between studies, as has been pointed out earlier (Katz, 1990).

Other investigators have also used locomotor activity as a trait marker to identify or “unmask” some other predisposition. For example, Hamilton et al. (2010, 2011) studied two groups of adult rhesus monkeys—one group was prenatally exposed to cocaine and the other group was control monkeys. When they were approximately 12–14 years old, each monkey was assessed in an open field for locomotor activity, along with other unconditioned behaviors including approaching a novel object. Hamilton et al. (2011) reported that there were no differences in locomotor activity or approaching a novel object between prenatally cocaine-exposed and control monkeys, even though other behaviors (e.g., drug-elicited yawning, resistance to extinction, and cocaine self-administration) were different between groups (Brutcher and Nader, 2012; Hamilton et al., 2010, 2011). This suggests that some characteristics that are hypothesized to influence vulnerability to drug abuse (e.g., in utero cocaine exposure) may not be amenable to the predictive validity of behavioral assays hypothesized to measure “anxiety-like” behaviors, like locomotor activity in an open-field apparatus.

There are some limitations to the use of locomotor activity to understand factors related to abuse liability. One major limitation is that while behavioral sensitization to locomotor stimulation frequently occurs, this does not necessarily translate into sensitization to the reinforcing effects of cocaine, and vice versa (e.g., Ahmed and Koob, 1998; Lack et al., 2008). It is also the case that characterizing animals as “high” and “low” responders does not necessarily translate into more or less vulnerable individuals, respectively (e.g., Dalley et al., 2007). Thus, while the behavior is amenable to pharmacological manipulations, and the combination of other unconditioned behaviors allows for rapid screening, some caution should be used when these are the primary behaviors under investigation.

2.2. CONDITIONED PLACE PREFERENCE

Conditioned place preference (CPP) studies are most frequently conducted in rodents and are said to involve “reward.” CPP involves classical conditioning in which stimuli associated with one quadrant are paired with a drug dose, while stimuli associated with a distinctly different quadrant are paired with the drug vehicle (see Mucha et al., 1982); the two compartments are separated by a neutral space. CPP (i.e., reward) is said to occur if the animal spends more time in the drug-paired side compared to the vehicle-paired side (e.g., Bali et al., 2015; see reviews by Wise, 1989 and by Sanchis-Segura and Spanagel, 2006). In training a CPP, many investigators use an unbiased procedure in which the initial phase consists of giving the animal access to both compartments of the apparatus. If the animal spends significantly more time in one compartment over the other (e.g., some investigators use 80% vs. 20%, others 67% vs. 33% as criterion), then they are not used in the conditioning phase of the study. In the conditioning phases, drug is paired with one compartment and drug vehicle with the other compartment; these compartments and drug/vehicle are typically counterbalanced across subjects. Most drugs of abuse can produce CPP and recent literature indicates that this methodology is frequently used to study drugs of abuse from all classes, including stimulants (e.g., Aguilar et al., 2015), opiates (e.g., Wang et al., 2015), alcohol (e.g., Gubner et al., 2015), and Δ⁹-THC, the active ingredient in marijuana (e.g., Manwell et al., 2014). Time spent on the drug-paired side is typically represented as an inverted U-shaped function of dose; very high doses can induce a conditioned place aversion (e.g., Kirkpatrick and Bryant, 2015).

In addition to examining abuse liability, CPP can be used to better understand the neurochemical and neuropharmacological mechanisms of action for drugs of abuse. Two examples will be provided here, one involving systemic drug administration and the other central administration. Northcutt et al. (2015) trained rats using an unbiased CPP protocol with 10 mg/kg cocaine and saline in the different compartments over 4 training days. For one group, during conditioning they received 10 mg/kg cocaine plus (+)-naloxone. When place preference was determined on Day 5, 10 mg/kg cocaine induced a CPP, but the group that was coadministered (+)-naloxone did not show a preference. Through in silico computer modeling and in vitro assays, the investigators hypothesized that cocaine and (+)-naloxone were binding to the same proinflammatory central immune signaling cascade; the CPP data suggested a functional consequence to these in vitro findings.

Using a slightly different version of CPP, Galaj et al. (2014) first trained the CPP with cocaine (10 mg/kg) and then examined the effects of a DA D1 receptor antagonist, SCH23390, administered via microinjection directly into the ventral tegmental area. The investigators found that SCH23390 (0, 2.0, 4.0, and 8.0 μg/0.5 μl) dose dependently reduced cocaine CPP. The difference between the results of this study and the previous one is related to neurochemical mediation involving acquisition (Northcutt et al., 2015) and expression (Galaj et al., 2014). In the latter case, the model addresses issues related to treatment efficacy, since conditioning had already taken place, while in the former study, neuropharmacological considerations related to vulnerability were addressed.

Most recently, CPP has been used to investigate environmental and social variables that influence vulnerability to drug abuse. One hypothesis is that when combined with social enrichment, lower drug doses induce a CPP (e.g., Thiel et al., 2008, 2009; see review by Trezza et al., 2010). For example, Watanabe (2011) studied three groups of mice in a methamphetamine CPP study: (1) individually housed animals in standard CPP training with a low methamphetamine dose of 2.0 mg/kg; (2) paired animals in which both mice received the exact treatment (i.e., saline on one side and 2.0 mg/kg methamphetamine on the other); and (3) control pairs in which CPP training was reversed such that when one animal received methamphetamine the other received saline. The pair group, in which both animals received the identical treatment, resulted in greater CPP than the individually housed and control pairs, indicating an enhancement of methamphetamine reward when the cage mate also received methamphetamine. It is important to note that merely the presence of a cage mate did not enhance CPP, but rather only when both animals received drug together was there evidence of methamphetamine reward. Interestingly, when time spent on the nondrug side was examined, the control pairs showed a profound place aversion. That is, when one animal received methamphetamine and the partner received saline, there was a place aversion on the saline side, perhaps indicating a negative consequence on social behavior related to drug use.

One final example to close out this section involves using CPP in combination with in vivo brain imaging to better understand the neurochemical consequences associated with drug use. Schiffer et al. (2009) first trained rats in CPP using 5.0 mg/kg (i.v.) cocaine and saline. This dose of cocaine was chosen because this group had previously shown, using in vivo microdialysis, that the cocaine-paired side would elicit increases in extracellular DA in the ventral striatum (Gerasimov et al., 2001). After the CPP was established, each rat underwent two PET scans using [¹¹C]raclopride. The investigators hypothesized that if the cocaine-paired side elicited DA release, the [¹¹C]raclopride binding potential would be significantly reduced compared to the PET signal when rats were placed on the saline-paired side. In fact, Schiffer et al. (2009) found an approximate 20% lower [¹¹C]raclopride binding potential in the dorsal and ventral striatum on the cocaine-paired side relative to the saline-associated side and a direct relationship between changes in binding potential and cocaine preference. These findings highlight the amenability of CPP to in vivo imaging studies.

There are some limitations to the use of CPP as a model to understand factors related to abuse liability. As mentioned above, CPP does not measure “drug seeking” or “drug taking,” two hallmarks of addiction. A second limitation is the ability to study multiple pharmacological manipulations—once the conditioning has been established, any tests without the drug of abuse can decrease the effectiveness of the CS, thereby making repeated, longitudinal studies more challenging. In general, these models are good initial screens that can lead to follow-up studies involving drug discrimination (DD) and drug self-administration procedures.

2.3. DRUG DISCRIMINATION

By definition, a discriminative stimulus “sets the occasion” for responding by providing information related to the contingencies mediating stimulus–response relationships. In models of DD, the discriminative stimulus is the presence or absence of the training drug. Training a discriminative stimulus in animal models typically involves two operant responses in which responding on one manipulandum (e.g., lever, key, nose poke, finger poke) is reinforced following administration of the training drug while responding on the other manipulandum is reinforced following administration of the drug vehicle. For example, when the subject is administered a dose of 0.2 mg/kg cocaine (the training drug and dose), responding on the left lever results in food reinforcement; responding on the right lever would have no scheduled consequence (or may reset the FR value on the correct lever). When the subject is administered saline, responding on the right lever would be reinforced, but left-lever responding would not. It has been hypothesized that the “interoceptive” discriminative stimulus effects of a drug in an animal, model the subjective effects in humans. A particular strength of DD procedures is that the behavioral effects of drugs are thought to be mediated centrally (i.e., receptor changes in the brain; see Carter and Griffiths, 2009 and Stolerman et al., 2011 for reviews). In addition to understanding the mechanisms of action mediating the discriminative stimulus effects of a drug, substitution studies are also used as an index of the abuse liability of compounds and impact the scheduling of drugs by the US Food and Drug Administration (FDA; see Nader et al., 2015 for examples).

In DD studies, the two primary dependent variables are % responding on the drug-associated lever and overall response rates. Most investigators operationally define substitution as occurring when at least 80% of the total responses occurred on the drug-appropriate lever. Including response rate data is important for several reasons. If a test drug substitutes for a drug of abuse, but only at doses that result in significant rate-decreasing effects, that may suggest less abuse liability because doses that disrupt ongoing behavior are required to produce subjective-like effects similar to the drug of abuse. Conversely, if a novel drug is studied and that drug does not substitute for the training drug, it may not be clear that high enough doses were tested unless response rates were altered. Related to both substitution and response rate effects of test drugs is the dose of the drug used to train the discriminative stimulus. As pointed out by Stolerman et al. (2011), “… training dose may show an impact on qualitative aspects of a discrimination, as defined by changes in the drugs to which generalization occurs, and sensitivity to antagonists” (p. 415). One example will be given in order to demonstrate the types of questions that can be addressed by manipulating the training dose.

2.3.1. Influence of training dose

Grant et al. (2000) trained male (n=8) and female (n=10) cynomolgus monkeys to discriminate either 1.0 g/kg ethanol from water or 2.0 g/kg ethanol from water (all solutions were administered intragastrically) in a two-lever, food-reinforced operant procedure. In addition to determining an ethanol dose–response curve, pentobarbital, midazolam, muscimol, and morphine dose–response curves were determined. Not surprisingly, the training dose influenced the ED50 values for ethanol substitution, with ethanol being more potent in the 1.0 g/kg training groups compared to the 2.0 g/kg groups. Pentobarbital and midazolam, two GABA_A agonists, substituted for 1.0 g/kg and 2.0 g/kg ethanol, but only the potency of pentobarbital was influenced by training dose. Grant et al. (2000) did not observe sex differences with regard to any manipulation in the study. These findings were extended to N-methyl-d-aspartate (NMDA) glutamate receptors by examining PCP, ketamine, and dizocilpine (Vivian et al., 2002). At the low training dose condition (1.0 g/kg ethanol), all three NMDA receptor compounds substituted for ethanol in both males and females. In contrast, at the 2.0 g/kg ethanol training dose, PCP, ketamine, and dizocilpine did not substitute for ethanol in the males. One possible mechanism for these sex differences was the greater sensitivity to the rate-decreasing effects of NMDA receptor antagonists in males compared to females; these sex differences were only apparent when the higher ethanol training dose was studied. Taken together, using different ethanol training doses, Grant and colleagues concluded that the NMDA receptor system is less prominent than the GABA_A receptors in mediating the discriminative stimulus effects of ethanol in nonhuman primates, especially with higher ethanol training doses. Such mechanistic understanding of ethanol’s effects would not have been obtained if only one training dose had been studied.

2.3.2. Other methodological considerations

In addition to the importance of training dose and sex, there are other independent variables that have recently been identified that can impact the substitution profile of drugs. In many DD studies, the subjects are modestly food restricted in order to study food-maintained operant responding. Depending on the drugs under study, this may influence the outcome of substitution studies (e.g., Baladi and France, 2010). For example, the DA D2/D3 receptor agonist quinpirole can be trained as a discriminative stimulus, and this typically involves food-restricted animals (e.g., Katz and Alling, 2000). Baladi et al. (2010) trained free-feeding rats to discriminate quinpirole from saline under a schedule of stimulus–shock termination. DA D2/D3 receptor agonists apomorphine and lisuride substituted for quinpirole and, as reported by Baladi et al. (2010), similar findings have been reported in food-restricted animals. However, using DA receptor antagonists, differences between free-feeding and food-restricted animals became apparent. In free-feeding rats, a D2/D3 receptor antagonist (raclopride) and a D3 receptor-selective antagonist (PG01037), but not a D2 receptor-selective antagonist (L-741,626), blocked the discriminative stimulus effects of quinpirole, shifting the quinpirole dose–response curve to the right. These findings suggest that the discriminative stimulus effects of quinpirole in free-feeding animals are primarily D3 receptor mediated, while in food-restricted animals, quinpirole’s discriminative stimulus effects are thought to be mediated by D2 receptors (cf. Baladi et al., 2010).

2.3.3. DD in combination with brain imaging

As mentioned above, it is believed that the discriminative stimulus effects of drugs are centrally mediated. Studies have been conducted that combine DD techniques with in vivo microdialysis to study how drugs that share discriminative stimulus effects influence neurotransmitter concentrations (e.g., Czoty et al., 2000; Kimmel et al., 2012). In one study, Czoty et al. (2004) trained monkeys to discriminate 0.32 mg/kg methamphetamine from saline under an FR 10 schedule of stimulus–shock termination. Monkeys were also implanted with guide cannulae above the caudate nucleus and microdialysis experiments were conducted in the same operant chambers as the DD procedures. The investigators found that methamphetamine, as well as cocaine and methylphenidate, resulted in dose-dependent increases in methamphetamine-appropriate responding when studied in the DD protocol. Doses that occasioned 100% methamphetamine responding produced similar increases in extracellular DA concentrations. Interestingly, the time course for elevations in DA and substitution in DD was not identical, indicating the involvement of other neurotransmitter systems in mediating the discriminative stimulus effects of methamphetamine.

There are some considerations regarding the use of DD that investigators should address. In terms of scheduling of drugs, the FDA suggests that if a novel drug substitutes for a drug of abuse, it has abuse liability, but if it does not substitute it may still have abuse liability. Considering the examples provided in this section on how dose, environmental context, and sex can influence these profiles, the use of DD in scheduling of drugs appears less than straightforward. Also of relevance for the development of treatment agents is the time course of substitution. The FDA does not distinguish the importance of time course, so if a novel drug does not substitute for cocaine (for example) until 2 h after administration, this information does not factor into “abuse liability,” but it should. If pharmacological agonists become a treatment strategy, a profile in which the subjective effects occur at a later time after administration and last longer than the drug of abuse, should positively impact compliance and reduce drug taking.

2.4. DRUG SELF-ADMINISTRATION MODELS

There is probably no behavioral model that is more predictive of human disease than animal drug self-administration models of abuse liability. Readers interested in the history of drug self-administration are referred to the original pioneering studies of Spragg (1940), Weeks (1962), Thompson and Schuster (1964), and Deneau et al. (1969); see also Griffiths et al. (1980). The behavioral process mediating drug self-administration is reinforcement, which can be either positive reinforcement or negative reinforcement. Positive reinforcement is defined as response-contingent presentation of a stimulus (e.g., drug) increases the probability of the response that produced the stimulus. Negative reinforcement is also an increase in responding, but in this case it is based on the response contingency of removing a stimulus (e.g., withdrawal symptoms). In the initial work (Spragg, 1940; Thompson and Schuster, 1964; Weeks, 1962), animals were made physically dependent on morphine and the drug self-administration behavior was thought to be mediated by negative reinforcement. That is, responding leading to morphine presentation was believed to be maintained by removing withdrawal symptoms. However, for all drugs of abuse, there are positive reinforcing effects and, most likely, negative reinforcing effects (see Czoty et al., 2015 for more discussion of this distinction). Drugs are self-administered by animals using the same routes of administration as humans including oral (e.g., Baker et al., 2014; Carroll and Meisch, 1978; Grant and Samson, 1985), inhalation (e.g., Carroll et al., 1990; Evans et al., 2003; Newman and Carroll, 2006; Pickens et al., 1973), and intravenous (some examples provided below). For the purposes of this chapter, basic information regarding schedules of reinforcement will be provided, as well as some examples involving different drug classes using the intravenous route. Because much work has been done with intravenous stimulants, especially cocaine, that will be the most frequently described drug in this section.

2.4.1. Use of simple schedules of reinforcement

Depending on whether the investigator is simply examining a drug for abuse liability or wanting to compare it to other drugs, different schedules of reinforcement are used. For example, answering the question “does the drug have reinforcing effects?” most investigators use an FR schedule of reinforcement in which a particular number of responses are required for each drug injection. For example, an FR 30 schedule, the thirtieth response results in drug presentation. If it is a within-subject design, behavior is compared to when saline is self-administered and if it is significantly higher, the drug has abuse potential. Less ideal is the use of an “inactive” lever in the chamber—responding that is higher on the “active,” drug-contingent lever relative to the inactive lever also represents reinforcement. Some investigators may use a fixed-interval (FI) schedule of reinforcement, in which a response after a specific period of time has elapsed results in drug presentation. For example, under an FI 3-min schedule, the first response after 3 min results in drug presentation; responding during the interval has no scheduled consequence. If the drug under investigation has substantial response rate-decreasing effects, this may be a better schedule than FR schedules because only one response is required after the interval has timed out. Irrespective of the schedule of reinforcement, behavior (response rates or number of injections) is represented as an inverted U-shaped function of dose (e.g., Pickens and Thompson, 1968; Weeks, 1962). The shape of this curve is influenced by several factors (Zernig et al., 2004), including reinforcing effects (increasing the probability of future responding) and rate-decreasing effects (decreasing likely responding). For this reason, it is not appropriate to compare drugs and rank them in terms of abuse potential using simple schedules of reinforcement. Later in this section, measures of reinforcing strength will be described; these models can be used to directly compare and rank drugs.

The use of animals allows investigators to begin with drug–naïve subjects and study vulnerability to drug abuse. As described earlier with high and low locomotor responders, phenotypic characteristics can be used to identify more or less vulnerable individuals. Others have shown that a particular drug history is needed for certain drugs to function as reinforcers. For example, Nader and Mach (1996) and Collins and Woods (2007) showed that monkeys and rats required a cocaine self-administration history before DA D3 receptor agonists would function as reinforcers, implying that a cocaine history alters DA D3 receptor function. Investigators frequently operationally define acquisition of some performance criterion (e.g., number of sessions needed to earn 30 injections) or acquisition of reinforcement. The latter implies a within-subject design and compares self-administration of a drug to vehicle-contingent responding. In order to show reinforcement, responding contingent on administration of a drug dose needs to be higher than responding leading to drug vehicle administration.

In a recent study, Gill et al. (2012) tested the hypothesis that adolescent exposure to methylphenidate would increase vulnerability to cocaine abuse. For this experiment, adolescent rhesus monkeys (~30 months old) were treated with extended-release methylphenidate or vehicle for 12 months. At the end of that treatment period, monkeys were trained to respond under an FR 30 schedule of food presentation (methylphenidate treatment had terminated and there was a 3- to 5-month washout). When responding was deemed stable, saline was substituted for the food pellets until responding declined to less than 20% of baseline for three consecutive sessions. There was a return to food-reinforced baseline and then ascending doses of cocaine were made available for at least the same number of sessions as was required for saline extinction, beginning at a very low cocaine dose (0.001 mg/kg per injection) and making half log unit higher doses available until cocaine functioned as reinforcer. There was a return to food-reinforced baseline before different cocaine doses were tested. This procedure allowed for a quantitative measure of cocaine acquisition—defined as the dose that maintained higher responding than when saline was available. Survival curves were generated for both groups and compared statistically. Gill et al. (2012) did not find any differences in vulnerability (i.e., cocaine acquisition) in the group treated with methylphenidate and controls. A similar procedure has been used and shown to differentiate female monkeys based on their social rank (Nader et al., 2012b).

As mentioned above, most drugs that humans abuse, animals will self-administer. One drug class that has proven challenging is marijuana or THC, the nonselective partial cannabinoid agonist. One of the first efforts to maintain THC self-administration in monkeys was reported by Harris et al. (1974). In that study, rhesus monkeys were given access to THC (0.025–0.3 mg/kg/injection over 10 s) under an FR 1 schedule of reinforcement during daily 12 h sessions. No dose maintained responding higher than vehicle-contingent behavior. Next, the investigators gave monkeys noncontingent THC in an effort to make them physically dependent and studied 0.025 mg/kg THC self-administration (perhaps as a negative reinforcer). Again, the behavior was not maintained above response rates leading to vehicle injections. Others have also reported negative results (Li et al., 2012; Mansbach et al., 1994). However, Tanda et al. (2000) and Justinova et al. (2003, 2008) reported THC self-administration in squirrel monkeys responding under an FR 10 schedule of reinforcement. There are several possibilities for the different outcomes including the species used (squirrel monkeys vs. rhesus monkeys), the drug vehicle, the pump duration, and the schedule of reinforcement. Clearly, much additional work is required (see Panagis et al., 2008) since recreational marijuana use continues to increase across the world.

In addition to acquisition (vulnerability), simple schedules of drug self-administration have also been modified so as to assess other phases of addiction including “loss of control,” by studying long-access conditions (e.g., Ahmed and Koob, 1998), long-term consequences during maintenance of drug self-administration (e.g., Nader et al., 2006), and relapse/reinstatement (e.g., Achat-Mendes et al., 2012; de Wit and Stewart, 1981), including the study of “incubation” (see reviews by Lu et al., 2004; Weiss, 2010). A recent series of studies have examined the powerful role of environment on drug self-administration, including alternative physical activities (e.g., Smith and Lynch, 2011) and social variables (e.g., Morgan et al., 2002; Nader et al., 2012b; Smith, 2012; Yap et al., 2015; see also Smith et al., 2014). Finally, it should be mentioned that the use of simple schedules of reinforcement and drug self-administration has recently been used to better understand the role of specific brain regions related to drug addiction, using in vivo imaging, such as PET, in vitro imaging using receptor autoradiography, optogenetics, and DREADDS. A full description of these protocols is beyond the scope of this chapter, but it is relevant that investigators studying the neurochemistry associated with addiction utilize self-administration models rather than noncontingent drug administration.

2.4.2. Use of complex schedules of reinforcement

Several investigators have suggested that more complex schedules of reinforcement that measure reinforcing strength (efficacy) are a better model of the human condition than simple schedules of reinforcement (Ahmed, 2010; Ahmed et al., 2013; Badiani, 2013; Banks and Negus, 2012; Banks et al., 2015). The two most frequently used models of reinforcing strength are the progressive-ratio (PR) schedule and drug choice procedures (either drug vs. drug or food vs. drug). For responding maintained under PR schedules, the number of responses required for a drug injection increase with each injection; this may occur within the same session (e.g., Czoty et al., 2010a; Kimmel et al., 2008) or across sessions (e.g., Griffiths et al., 1978; see also Rowlett et al., 1996). For these studies, the primary dependent variable is the final ratio completed, termed the break point (BP), when no injections have been received after a specified period of time (termed the limited hold) or at the end of the session. As with all schedules of drug self-administration involving reinforcement, the shape of the dose–response curve is an inverted U-shaped function; for PR studies, BPs for different drugs can be compared statistically (see Stafford et al., 1998 for review).

PR schedules are quite amenable to examining the effects of treatments on drug self-administration, including cocaine self-administration (e.g., Czoty et al., 2006, 2010b, 2013). As an example, the effects of d-amphetamine on cocaine BP will be described. Amphetamine has been shown to have efficacy as a cocaine pharmacotherapy (Grabowski et al., 2001; Negus and Mello, 2003a,b). In one study, Czoty et al. (2011) had monkeys self-administering cocaine under a PR schedule; the dose of cocaine was on the ascending limb of the dose–response curve. Monkeys received a continuous infusion of d-amphetamine at a rate of 0.4 ml/h and every 7 days they were given access to cocaine. If the amphetamine treatment decreased the cocaine BP, they were retested 1 week later to examine for tolerance to these effects; if tolerance developed or if the initial amphetamine dose had no effect on cocaine BP, the daily amphetamine dose was increased. In this study, d-amphetamine decreased the BP for cocaine and, importantly from a clinical perspective, tolerance did not develop to these effects. Also of relevance is that different amphetamine doses produced optimal effects in monkeys, so if all animals had been tested with the same doses and mean data presented, the effects would not have been statistically significant. Studies of this type highlight the importance of individual subject variability in drug responses.

For studies involving drug choice, the primary dependent variable is percentage of trials the drug is chosen. There are two general variations of the choice procedure: drug versus drug choice and food versus drug choice. In one sense, if an investigator wanted to directly compare the reinforcing strength of a novel drug with a known drug of abuse, the drug–drug choice procedure is ideal (e.g., Johanson and Schuster, 1975). For these studies, animals are implanted with double-lumen catheters in which drug A is available through one lumen and drug B through the other. For example, Lile et al. (2002) compared the reinforcing strength of a novel DA transporter (DAT) blocker, PTT, with cocaine. When first studied under a PR schedule, the BP for PTT was significantly lower than that for cocaine (Lile et al., 2002). However, when monkeys were given the opportunity to choose between cocaine and PTT, at the highest dose of each, PTT and cocaine were chosen on 50% of the completed trials. Interestingly though, cocaine intake was reduced by nearly 90% relative to when choice was between cocaine and saline. That is, the monkeys did not complete many trials when both drugs were available (although half the trials resulted in cocaine and the other half PTT), suggesting that perhaps a long-acting DAT blocker would be an effective treatment for cocaine addiction in the context in which cocaine is still being used (see Nader et al., 2015 for additional discussion).

The second variation of drug choice involves comparing self-administration in the context of alternative nondrug reinforcers. However, the food–drug choice procedure is too labor intensive to use to directly compare novel drugs in terms of measures of reinforcing strength. That is, how different drugs dose–response curves appear in the context of a nondrug alternative are difficult studies to interpret. For example, Nader and Woolverton (1991) had different groups of monkeys, one choosing between cocaine and food the other between procaine and food. Under baseline conditions, the shapes of the dose–response curves for both drugs appeared similar. However, when the magnitude of the alternative was manipulated (i.e., increases in the number of food pellets available as an alternative to drug), the procaine dose–response curve became much flatter than the cocaine curve, suggesting that procaine had weaker reinforcing strength than cocaine.

When only one drug is studied (e.g., cocaine), investigators can utilize a food–drug choice procedure to compare different groups of subjects in terms of sensitivity to environmental context and alternative reinforcers. For example, when monkeys are placed in social groups, they form a linear hierarchy from most dominant to most subordinate (see Nader et al., 2012a) and the formation of these hierarchies results in changes in brain DA receptors and initial vulnerability to cocaine abuse (Morgan et al., 2002). However, if monkeys are permitted to continue self-administering cocaine under FR schedules, the differences between dominant and subordinate monkeys dissipate (Czoty et al., 2005). When the conditions are changed to a concurrent FR schedule of food and cocaine presentation, subordinate monkeys are more sensitive to cocaine reinforcement, choosing lower doses of cocaine relative to food compared to dominant monkeys (Czoty et al., 2005). Importantly, the ability of drugs to alter cocaine–food choice also varies depending on the social rank of the monkey (Czoty and Nader, 2013, 2015).

More frequently, food–drug choice studies are used to identify potential treatment drugs. As described by Banks and Negus (2012), if addiction is conceptualized as a choice (Heyman, 2009), then drug versus nondrug choice behavior may have the greatest face validity to the human condition (see Haney and Spealman, 2008; Hutsell et al., 2015). The primary objective of these studies is to examine novel treatment drugs on percent drug choice and a positive outcome would be represented by a shift in preference from drug to the food alternative. This reallocation of behavior would model the human condition in which the drug user chooses an alternative reinforcer (e.g., job) over continued drug use. Some recent examples are described by Nader and Banks (2014) and Banks et al. (2015).

3. CONCLUSIONS AND FUTURE DIRECTIONS

Models of drug addiction remain an integral preclinical research screen (see Table 1). There will always be a need to screen novel drugs for abuse liability and with the increases in recreational drug use (due in part to the legalization of marijuana) and the huge expense of dealing with drug addiction, the use of these animal models to evaluate novel treatments will be relevant for many years to come. There is a need to standardize research techniques and there is an important need to expand the study of THC abuse beyond the few laboratories that are currently examining this drug class. There is also a need to better understand individual differences in drug responses. Studies designed to investigate variable drug responses, rather than trying to control for them, will yield important new insights that will move the field of addiction treatment toward a personalized medicine approach. This chapter has highlighted the advantages of animal models for the understanding of brain changes accompanying drug addiction. It has been noted that there are certain behavioral phenotypes, as well as CNS markers (e.g., DA D2 receptor availability) that make certain individuals more vulnerable than others. How these phenotypic characteristics impact later treatment strategies remains to be determined. For example, when two individuals with long-term drug history are studied (in the animal or human lab or in the clinic), they may have very similar symptomatology. However, the treatment outcome may be different, and we have typically attributed these equivocal outcomes as a “lack of effect.” For example, recovery of DA receptor function varies among individuals and this may be related to some behavioral phenotype (Nader et al., 2006). Animal models of addiction can shed light on how these initial phenotypes impact later treatment efficacy—perhaps there is truly order in these “equivocal” outcomes, but only after understanding the long-term consequences of drug use can clinicians make individualized treatment decisions.

Table 1.

Summary of Animal Models

Model	Strengths	Limitations
Locomotor activity	No training	Not a measure of abuse liability
CPP	Easy training	No “drug-seeking” measures
	Evidence for most drugs of abuse
Drug discrimination	CNS-mediated effects	Because training dose and drug history can affect outcomes, care must be taken in designing studies
Self-administration	All routes of administration	Frequently requires surgery
	Predictive of human condition	Response rates are not measuring reinforcing strength