It is a distinctive feature of many drugs of abuse that they produce subjective effects in humans. Subjective effects are sensory in nature and differ from many other drug effects in that they are accessible to conscious perception. A familiar example is the feelings of drunkenness and relaxation that can follow the ingestion of ethanol. Also known are the subjective effects of morphine and those of such drugs as cocaine, amphetamines, hallucinogens, such drugs as cocaine, amphetamines, hallucinogens, cannabinoids, barbiturates, benzodiazepines, and phencyclidine.

It is thought that the ability of opiates (Jasinski, 1977) and many other drugs to produce subjective effects relates intimately to their abuse; in as much as these effects can be euphoric or otherwise reinforcing, they may be involved in the initiation and maintenance of self-administration behavior. The single and, even more so, the repeated exposure to the drug may render the subject drug dependent. That is, the subject's behavioral and physiological homeostasis may come to depend on the intake of the drug in much the same manner as it depends on the intake of gaseous oxygen, fluids, and solid foods. Prominent and particularly powerful among the known drug dependencies is opiate dependence; opiates interfere with respiration (Eckenhoff & Oech, 1960) and with the ingestion of fluids and food (e.g., Holtzman, 1975; Locke, Brown, & Holtzman, 1982; Lowy, Starkey, & Yim, 1981; Ostrowski, Rowland, Foley, Nelson, & Reid, 1981).

Although the production of subjective effects, the occurrence of self-administration behavior, and the development of dependence may thus be relevant to drug abuse, preclinical research on drugs of abuse has been concerned largely with the ability of drugs to produce self-administration and dependence in laboratory animals. Until recently, it has not seemed feasible to study subjective drug efd feasible to study subjective drug effects in any species other than human.

It has been hypothesized (Colpaert, Lal, Niemegeers, & Janssen, 1975a; Colpaert, Niemegeers, Lal, & Janssen, 1975b), however, that the discriminative stimulus properties of opiates in the rat may relate to and can serve as an animal model of opiate-like subjective effects in man. It is not merely incidental that opiates were involved in this original proposal relating discriminative drug properties in animals to subjective drug effects in man. Opiates are a group of drugs of abuse for which formal methods have been developed (Fraser, 1968; Jasinski, 1977) to identify, quantify, and compare their subjective effects. These effects have made it possible that the subjective effects of a number of opiates have been documented and characterized extensively (e.g., Fraser, Van Horn, Martin, Wolbach, & Isbell, 1961; Haertzen, 1970, 1974; Jasinski, 1973; Jasinski & Mansky, 1972; Jasinski, Martin, & Haertzen, 1967; Jasinski, Martin, & Sapira, 1968). More so than with most other drugs of abuse, the clinical evidence on subjective effects of opiates thus offered a data base for comparison to animal data. The initial attempt (Colpaert et al., 1975a, 1975b) to substantiate the hypothesis did in fact consist of a comparison of discriminative effects of morphine, codeine, and diphenoxylate in rat to the results of a clinical study (Fraser & Isbell a clinical study (Fraser & Isbell, 1961) on their subjective effects in man. The evidence relating the discriminative effects of drugs in laboratory animals to their subjective effects in humans has been the subject of a recent review (Colpaert, 1984). Briefly, two lines of evidence suggest that discriminative drug effects in animals are homologous to subjective drug effects in humans. One similarity is methodological; in both cases the drug acts as a stimulus to behavior, and the formal relationship of the behavioral response to the drug stimulus is similar (Colpaert, 1978a; Schuster, Fischman, & Johanson, 1981). Second, it appears that the pharmacological characteristics of the discriminative effects of drugs in animals may closely match those of their subjective effects in humans. Evidence to this effect is relatively extensive and detailed with opiates (Colpaert, 1977a, 1978a, 1982; Herling & Woods, 1981; Holtzman, 1982a, 1982b) and is also available with such other prominent substances of abuse as CNS stimulants (Ho & Silverman, 1978; Silverman & Ho, 1977), hallucinogens (Appel, White, & Holohean, 1982; Glennon, Rosecrans, & Young, 1982), cannabinoids (Krimmer & Barry, 1977; Weissman, 1978), and ethanol (Barry & Krimmer, 1977).

The aim of the present chapter,¹ then, will be to describe some of the typical as well as the variable features of the drug discrimination experiment. We he drug discrimination experiment. We will also explore how drug discrimination can be used in analyzing the actions of drugs of abuse, and the utility it may have in the preclinical evaluation of drug abuse potential.
 

The Drug Discrimination Experiment

In a typical drug discrimination (DD) experiment, laboratory animals are trained to discriminate injections of a given dose (the training dose) of a particular drug (the training drug) from saline injections. Perhaps the most widely used of many possible procedures is pressing one of two levers according to a fixed ratio-10 schedule which results in food delivery to deprived rats (see Colpaert, Niemegeers, & Janssen, 1976a). That is, rats are trained to press the two levers for food, and the fixed ratio-10 schedule is gradually instituted prior to discrimination training. In any given session only one of the two levers is operative, and the operative lever is altevers is operative, and the operative lever is alternated between sessions.

Once the two responses—pressing the left and the right of the two levers—are established, discrimination training commences. That is, prior to every daily (5 days/week) 15-minute session, the animal is injected with either the training drug or saline. After the training drug injection only one of the two levers (the drug lever; DL) is operative, and pressing the alternative lever (the saline lever; SL) has no programmed consequences in drug sessions. After injection of saline SL responding is reinforced, and pressing the DL in saline sessions is inconsequential.

The following considerations apply to the sequence in which training drug (D) and saline (S) sessions are administered; (a) D and S sessions occur equally often; (b) simple alternation is avoided; (c) chains of at least two consecutive D and S sessions are thus allowed to appear; this accommodates the nonscheduled occurrence of such chains with later test sessions; finally, (d) the sequence must take into account that rats may learn to use olfactory cues that are left in the operant chamber by preceding animals (Extance & Goudie, 1981). Lever assignments can be: DL, left; SL, right; or DL, right; SL, left. These four conditions are implemented in the following two monthly alternating sequences—(1) D-S-S-D-S, S-D-D-S-S, S-D-S-D-D, D-S-D-S-D and (2) S-D-D-S-S, D-S-D-S-D, D-S-S-D-D,d (2) S-D-D-S-S, D-S-D-S-D, D-S-S-D-D, S-D-S-D-S. Rats with even serial numbers are initially run on sequence 1 and the others on sequence 2; the sequence assignments are reversed every 4 weeks (Colpaert, Niemegeers, & Janssen, 1977).

The implementation of these conditions will typically result in a change in behavior that is characteristic of the acquisition of a discrimination; the animals come to reliably press the DL in D sessions and the SL in S sessions. Once a given animal has reached some arbitrarily set criterion, tests of stimulus generalization can be conducted. For the purpose of such tests, the animal is treated with the pharmacological condition that is being studied, and its subsequent operant responding is observed. Stimulus generalization with the training drug is said to occur if the test treatment induces DL responding; generalization is said not to have occurred if the test treatment induces SL responding. Note that the wording "generalization with saline" is not ordinarily used when the test treatment makes the animal respond on the SL.

A built-in precaution against the possible development of state dependency in this procedure is that the two discriminative responses are established prior to the implementation of the pharmacological stimuli. It is also possible to verify whether state dependency occurs by determining if lever selection behavior early in training approaches what is then the chanceining approaches what is then the chance level (50%) in both D and S sessions (Colpaert, 1977b, 1983).

Typical Features

The DD method consists of a set of experimental conditions whereby one of two or more pharmacological stimuli predicts or signals (Colpaert, 1978b) which of several responses will be associated with reinforcement. Typically, the animals are trained to discriminate a training drug at a given training dose from the drug vehicle. The discrimination involves two responses. These responses are operant—instrumental in the occurrence of reinforcement. The two discriminative responses have a similar morphology (e.g., lever pressing), except for one particular feature such as location (e.g., pressing a left as opposed to a right lever). The responses are mutually exclusive; in the example the FR-10 schedule can be completed for the first time within a session on only one of the two levers.

The reinforcement of the two operant responses is the same; the same schedule of reinforcement is implemented for DL responses in D sessions and SL responses in S sessions. Since the reinforcement of the two responses is the same, it also follows that scheduled reinforcement is symmetrical in D and S sessions.

Another significant typical feature of the DD experiment is that it establishes a nominal (binary, quantal, categorical) relationship between the drug sgorical) relationship between the drug stimulus and the behavioral response (Colpaert, 1982a, 1983); SL responding is reinforced if, and only if, the animal was treated with saline, and DL responding is reinforced if, and only if, the treatment consisted of training drug. Specifically, the pharmacological treatment predicts only which of the two possible, mutually exclusive, locations of responding will yield reinforcement. The treatment is not predictive of any variation in reinforcement that may be consequential to any variation in behavior other than location.

Before considering the DD variables, it may be pertinent to note that DD constitutes an extremely simple and straightforward paradigm. Complexities may become apparent with the manipulation of independent variables and with the detailed numerical analysis of dependent variables. But these complexities should not obscure that the utility of the paradigm for pharmacological research owes much to the paradigm's simplicity (Colpaert, 1982a).

Independent Variables

Pharmacological Variables

Pharmacological variables include the training drug, the training dose, and the route and time of drug administration prior to training sessions. Route and time of training drug administration are typically chosen in accordance with what is known about the drug'ccordance with what is known about the drug's kinetic properties and are the same for vehicle administration. The training dose is typically chosen to be within a known appropriate or relevant range of doses of the training drug. A wide variety of drugs can be discriminated from saline; suffice it here to indicate that this variety includes many substances of abuse (see Stolerman et al., 1982).

Procedural Variables

A survey of procedural variables is given in Table 1. Though a variety of species, responses, apparatus, reinforcers, and schedules have been used, numerous studies have revealed good agreement across these variables in the data that are obtained in generalization tests. The rat has been the most widely used animal species for obvious reasons. Some effort is involved in the initial DD training, and longevity may be a criterion in the choice of the animal species. The monkey and pigeon may excel the rat at this point, particularly in studies where number of animals is less of a consideration. However, little work has been done on the possible effects of prolonged exposure to the conditions of the DD experiment and of age. Effects of this nature may be significant, especially since the animals are exposed almost continuously either to deprivation schedules or to such stressful manipulations as electric shock. There is yet no evidence of any important change in generalization characteristics in conditions wheretion characteristics in conditions where the discriminandum remains unchanged for protracted periods of time (e.g., Colpaert, Niemegeers, & Janssen, 1976b). But caution must be exerted with the uncontrolled use of the same animals in experiments involving different discriminanda.

Procedures involving an operant chamber, the lever-press response, and food, water, or electric shock have been used more widely than mazes where locomotor responses are reinforced by the escape from or avoidance of electric shock or immersion in water. The operant chamber procedure can perhaps be considered as offering a set of conditions where sizable rates of responding can be maintained while implementing only limited stress. This apparently allows low training doses of drugs to be discriminable (Colpaert & Janssen, 1982b; Colpaert, Niemegeers, & Janssen, 1980b). Though no truly adequate comparisons have as yet been made, the shock- or water-escape maze procedures typically require much higher training doses for drugs to be discriminable (Overton, 1982). The differences between operant chamber and maze procedures may not be, however, simply quantitative. An (unpublished) experiment in our laboratory revealed that 10 mg/kg of pentobarbital is readily discriminable in both the operant chamber (Colpaert et al., 1976a) and a water-escape maze procedure (Colpaert, Niemegeers, & Janssen, 1978a). Cocaine, in contrast, was readily discriminable (atontrast, was readily discriminable (at 2.5 and 10 mg/kg) in the operant chamber procedure but essentially failed to acquire response control in the maze procedure at doses ranging from 0.63 to 40 mg/kg. It would seem possible from these data that severe stress may compromise drug discrimination and that training drugs which alleviate procedure-generated stress may be less vulnerable to this effect than those which do not. The issue deserves analysis but has not as yet been examined.
 

Fixed ratio (FR), variable interval (VI), differential reinforcement of low rate (DRL), and other schedules of reinforcement have been used in operant chamber procedures; maze procedures have typically utilized continuous reinforcement. Generalization test data have generally been consistent across different schedules. Schedules have often been chosen on the basis of (i) the amount of responding that they generate and (ii) their vulnerability to drug effects on rate. Specifically, schedules yielding modest rates of responding that are abolished by even low doses of drugs are considered undesirable. Overton (1979) has compared FR, VI, and DRL schedules in the two-lever appetitive procedure and concluded that the FR schedule yields relatively rapid acquisition and high asymptotic accuracy. These data add to the attractiveness of the two-lever, food-reinforced FR-10 procedure (Colpaert et al., 1976a); for practical pharmacological purposes, it is desirable to have available a rapid and accurate method which is sensitive to the discriminative effects of low doses and allows exploration of fairly wide ranges of training dose. The availability f fairly wide ranges of training dose. The availability of this procedure, however, should not act to discourage further methodological research. It would be of interest, for example, to examine and compare stimulus generalization with a given training dose of a given training drug in conditions where its discrimination is rapidly acquired and accurate and in conditions where it is slowly acquired and/or less accurate. In the context of such methodological research, the portrayal of "the ideal" DD procedure (Overton, 1979) obscures the important and interesting fact that the discriminative effects of drugs may vary markedly according to the conditions in which they are being analyzed (Colpaert & Janssen, 1981).

The procedures that have been used in tests of stimulus generalization differ mainly in terms of the delivery of reinforcement. When reinforcement is delivered in test sessions, either the selected response (e.g., Colpaert et al., 1976a) or both responses (e.g., Shannon & Holtzman, 1976) may be reinforced; reinforcing only the response which is alternative to the one selected by the animal is of course uncommon. The other possibility is to institute extinction. One reason for administering reinforcement in generalization tests is that the availability of reinforcement offers the opportunity to analyze the possible effects of test treatments on overall rate of responding (Colpaert et al., 1975a, 1976b; Shannon & Holtzman, 1976). , 1976b; Shannon & Holtzman, 1976). Such data on rate effects are obviously important. The rationale for testing in extinction has been largely intuitive; extinction is often thought to prevent new discrimination learning from taking place in test sessions. There is, however, no a priori reason to assume that less new learning takes place in extinction than with reinforcement; the learning, if any, may simply be opposite in sign. The new learning can be limited by implementing only brief episodes of extinction. But this maneuver does not preclude the possibility that test treatments become a signal for brief sessions of extinction; it may simply discourage responding in later test sessions. At any rate, the overall consistency of results that have been obtained across different test procedures renders it unlikely that much new discrimination learning occurs in test sessions. Consistent with this conclusion are the results of an (unpublished) experiment in which six rats were trained to discriminate 0.04 mg/kg of fentanyl from saline and then tested twice with 1.25 mg/kg of d-amphetamine. In the first test all animals selected the SL but were only reinforced for pressing the DL. The second test was administered one week later and revealed that all animals again selected the SL, despite this previous experience.

Different test procedures may nonetheless yield different results depending on the dependent variable that isng on the dependent variable that is being used to measure discrimination and generalization. Figure 1 summarizes earlier (Colpaert, 1977b) and new (unpublished) data on the effects of reinforcement on two different indices of generalization. Nine rats were trained to discriminate 0.04 mg/kg of fentanyl from saline in the two-lever, food-reinforced FR-10 procedure and were tested, eight times in all, with 0.04 mg/kg of fentanyl and saline in four different conditions: (1) Responses on the lever which the animal initially selected were reinforced, a standard practice in this laboratory; (2) responses on both levers were reinforced; (3) only responses on the lever opposite to the selected lever were reinforced; and (4) no reinforcement was delivered (i.e., the test consisted of an extinction episode). The session duration in Conditions 1, 2, and 3 was 15 minutes; the duration in Condition 4 was 60 minutes, but data were recorded after 0.5, 15, and 60 minutes. Two variates were derived from the responding that occurred in test sessions. One is the selected lever (i.e., the lever where the animal totaled 10 responses first); this variate can take one of only two values (i.e., SL and DL) and is nominal. The second variate consisted of the percentage of responding that was appropriate to the training drug (i.e., the percentage of DL responding); this variate is free to vary from 0 to 100% and is quantitative. The results of this experiment ative. The results of this experiment indicate that the nominal variate was entirely unresponsive to the manipulations of reinforcement in test sessions (see Figure 1). The reason for this is simple: The variate is based exclusively on responding which occurs prior to the time of the first possible delivery of reinforcement. It therefore is entirely independent from reinforcement. The percentage of drug-appropriate responses did vary as a function of the reinforcement procedure used in test sessions. The percentage was entirely treatment-appropriate when reinforcement was instituted on the selected lever. When either lever yielded reinforcement, one fentanyl-treated rat switched from the DL to the SL in the course of the session and continued to press the SL for some time. Behavior became particularly disorganized when pressing the nonselected lever was reinforced (Condition 3). The extinction procedure yielded responding that was almost entirely treatment-appropriate if only the first 30 seconds were considered. This duration corresponds roughly to the length of time it typically takes the animal to complete the first FR-10 schedule; the behavior in extinction at this point, therefore, is essentially identical to the behavior used as the basis for the nominal measure. However, the percentage of drug-appropriate responding rapidly deteriorated as the extinction episode grew longer, and by 60 minutes almost no discrimination was aminutes almost no discrimination was apparent anymore. In summary, the experiment described here utilized the training treatments in tests where the reinforcement procedure was varied. Obviously, adequate test conditions should yield results that are entirely appropriate to the training drug and saline treatments. This occurred at all times with the nominal variate; it also occurred with the quantitative variable when the selected or either lever was reinforced or when only a brief episode of extinction was instituted. The relevance of this data to the use of one or the other dependent variable will be discussed in a subsequent section.
 

Figure 1: Results of four different stimulus generalization tests with 0.04 mg/kg of fentanyl (filled circles) and saline (open circles) in nine rats trained to discriminate 0.04 mg/kg of fentanyl from saline. Two dependent variables are presented—the percentage of rats selecting the drug lever (upper panel) and the percentage of drug-appropriate responses (lower panel; mean ±1 SEM). Data points for condition 4 represent data obtained at different intervals of time (i.e., 0 to 0.5 minutes, 0.5 to 15 minutes, and 15 to 60 minutes). See text for detai 15 minutes, and 15 to 60 minutes). See text for details.

The Discriminandum

The discriminandum specifies what pharmacological treatment is to be discriminated from what alternative treatment(s). This constitutes a third source of independent variability in the DD experiment. Table 2 gives a survey of some of the many possible discriminanda that can be used in drug discrimination (see also Jarbe & Swedberg, 1982) and indicates the additional sources of independent variability that are inherent in each of these discriminanda. The drugs used constitute a further, obvious source of variability but are not listed as such in the table. The DD discriminandum typically involves at least two discriminative responses, but it may also involve three or perhaps even more responses; the discriminanda listed in Table 2 are grouped according to the number of responses involved. Not included in this table is the single response procedure described by Winter (1975). In this procedure an operant response is reinforced for one group of subjects in the presence of drug but extinguished in its absence; reverse conditions apply for a second group. Data obtained with this procedure are, however, confounded by drug effects on rate of responding, and the procedure has been abandoned for this reason.

The drug vs. saline discrimination constitutes the single most widaline discrimination constitutes the single most widely used discriminandum, and its properties have been documented best. Generalization results obtained with this discriminandum are relatively simple to interpret, and the data serve many pharmacological purposes. More than any other single method of behavioral pharmacology, the single drug-saline discrimination offers an exquisitely specific method to characterize the agonist and antagonist activity of a broad range of drugs (Colpaert, 1982a); it is likely that this exceptionally powerful resolving capacity has made the DD paradigm so widely used in recent years. The drug-saline discrimination is simple also because it has only one independent variable (i.e., training dose). Fragmentary evidence on the effect of training dose originally suggested that it merely sets the sensitivity of the discrimination. Parametric analyses, however, have revealed far more fundamental effects (Colpaert, 1982a). Such analyses have been conducted in detail only on fentanyl (Colpaert, Niemegeers, & Janssen, 1980a; Colpaert et al., 1980b) and cocaine (Colpaert, Niemegeers, & Janssen, 1978c), and parametric data on the effects of training dose with other training drugs remain to be generated.
 

The dose₁ vs. dose₂ discrimination would seem to offer a method for the fine-grained analysis of the differences in discriminative effects of different doses of a given training drug. Two independent variations are inherent in this discriminandum; (i) the dose ratio (i.e., the ratio of the higher to the lower of the two training doses) and (ii) the absolute magnitude of these doses. A parametric analysis of the effects of dose ratio and of dose magnitude in dose₁-dose₂ discrimination has been carried out with fentanyl (Colpaert & Janssen, 1982a). The study revealed important effects of both factors on parameters of discrimination and generalization. A further study (Colpaert, 1982b) indicated the the dose₁-dose₂ discrimination may further improve the pharmacological specificity of the DD experiment relative to results obtained in the drug vs. saline discrimination. The effect is very impressive and challenges the current, intuitive view (Jarbe & Rollenhagen, 1982) that the dose₁-dose₂ discrimin1982) that the dose₁-dose₂ discriminandum represents a discrimination along a simple intensity continuum.

Only limited or no data are available with any of the other discriminanda. The drug_A AND drug_B vs. saline can perhaps demonstrate what can be referred to as residual discriminative effects of an agonist A in the presence of an antagonist B (e.g., Swedberg & Jarbe, 1982) where the combination elicits saline responding in drug vs. saline discrimination. The drug_A OR drug_B vs. saline discriminandum can perhaps be useful as a broadened screen for several types of discriminative effects (Colpaert & Janssen, 1982c). Drug_Avs. drug_B vs. saline may perhaps offer more resolving capacity than a single drug vs. saline discrimination does in otherwise similar conditions (White & Holtzman, 1981).

It will be apparent from this brief survey that the discriminandum presents an exceedingly large source of independent variation in the DD paradigm; discriminative effects of drugs can be examined in any of a great number of discriminanda, and generalization test data may vary accordingly. A potential danger with this large variability is that a given body of generalization data on any particular compound may become difficult to interpret. It has been this author's position, therefore, that it may be appropriate to examine thet it may be appropriate to examine the sources of independent variations in the simplest discriminanda (i.e., drug vs. saline; dose₁ vs. dose₂) through detailed parametric studies. The insights and concepts derived from such studies are likely to be a condition for comprehending the intricacies produced by the more complex discriminanda. In the meantime the existence of poorly understood generalization data from complex discriminanda may unduly compromise the face validity which DD research has rightfully acquired with the simple drug vs. saline discriminandum.

Subject Variables

A fourth source of independent variation are subject variables. DD research is typically conducted with what are referred to as normal animals, albeit that the subjects are invariably stressed by either deprivation of food or water, by shock, or by some other manipulation. Subjects whose conditions differ from normal may be able to discriminate otherwise nondiscriminable drugs or may show different generalization characteristics. Weissman (1976) has presented data that rats in pain, but not normal animals, may be able to discriminate aspirin. And opiate antagonists may become discriminable in opiate dependent subjects (Gellert & Holtzman, 1979). Subject condition thus has the potential of revealing drug effects that are not accessible otherwise and of establishing unique models of pathology. Dependent Variables

What is being observed in DD are discrete behavioral responses (e.g., depressions of two or more levers) which occur in the course of an interval of time which is termed a training or test session. What is being inferred from this responding are variates which ascribe some spatial or temporal organization to this responding. The dependent variables of the DD experiment, like those of many other paradigms, must therefore be considered as derived variables; the value which the variable takes is not directly or inherently apparent from what is being observed. Instead, the dependent variables are constructed by the experimenter and are designed to accommodate some (often largely implicit, ill defined) a priori concept or theory. The concepts underlying DD research have differed and so have the dependent variables. Importantly, the difference concerns the dependent variable used to measure discrimination and generalization.

Measures of Discrimination and Generalization

The two most important variables in current use to measure discrimination and generalization are (i) the percentage of drug-appropriate responses and (ii) response selection. These variables will be discussed here as they apply to a drug vs. saline discrimination in the two-lever, food-reinforced FR-10 procedure. The p food-reinforced FR-10 procedure. The percentage of drug-appropriate responses is most widely used, but an explicit justification of it is not available. Apparently, the use of this variable adheres to a tradition which can be traced to Skinnerian theory (Colpaert, 1983). Skinner (1938) argued that the acquisition of a discrimination represents two reflexes that draw apart in strength and that reflex strength is to be measured by means of response rate. The percentage of drug-appropriate responding occurring in a drug vs. saline discrimination has therefore been taken as a measure of the strength of the drug reflex relative to that of the saline reflex. The measure uses the percentage rather than the absolute number of drug-appropriate responses in an apparent attempt to accommodate possible drug effects on total response output. It is unclear, however, how well this maneuver achieves its goal; the effects of drugs on the rate of operant responding are rate-dependent (e.g., Dews, 1958), and it is conceivable that test drugs have effects on the rates of saline- and drug-appropriate responding which are not simply proportional to the often differing absolute rates.

Response selection, or lever selection as it applies here, has been proposed (Colpaert et al., 1976a) as a more appropriate measure on grounds that the dependent variable of the DD experiment is nominal in nature (for review, see Colpaert, 1983, 1984). Briefly, it isee Colpaert, 1983, 1984). Briefly, it is argued (i) that the only programmed variation in behavior which the DD paradigm relates with the pharmacological stimuli is DL as opposed to SL pressing (see above) and (ii) that, therefore, DL as opposed to SL pressing constitutes the only variation in behavior which can be taken to dependably reflect the discriminative effects of pharmacological treatments.

These concepts concerning the observable behavior in the DD experiment have thus given rise to the use of two different measures of discrimination and generalization. A paramount difference between the two measures concerns the level of measurement for scaling behavior: The percentage of drug-appropriate responses scales behavior along a quantitative scale, most often a ratio scale; the response-selection measure scales behavior according to a nominal or classificatory scale. Theory of measurement (Siegel, 1956; Stevens, 1946) specifies what theoretical differences exist between these scales. Some of the differences in data analysis and data interpretation emerging from this distinction will be discussed below.

Other Dependent Variables

In addition to measurements of discrimination and generalization, many DD procedures involve dependent variables that measure other characteristics of animal behavior. Total response output, or overall response rate, can be useful as an indication of possible drug effects an indication of possible drug effects on ongoing operant behavior. The training drug itself may have effects on this parameter, and its effects may change in the course of training or at some later point of the experiment. This parameter is often used in the choice of an appropriate dose or dose range of the training drug and of the test drugs. It may also assist in demonstrating the behavioral effectiveness of test drugs in cases where the test drug does not affect the main dependent variable (e.g., does not induce generalization). This parameter can further be used to monitor the subjects' general condition and may at times be instrumental in detecting apparatus failure.

Response latency is another rate measure; it differs from overall response rate by pertaining only to the initial segment of behavior rather than to all behavior occurring during the entire session. In procedures where the main dependent variable is nominal, response latency can be defined as the time which elapses before response choice or response selection occurs and can then perhaps be termed choice rate. Interestingly, it may appear that, relative to saline, a training drug condition decreasing overall rate nonetheless yields a smaller response latency (i.e., a higher choice rate; Colpaert, 1978b). These and other data (Colpaert et al., 1980b) suggest that response latency may perhaps in part reflect the time required for the decision process underlying a nor the decision process underlying a nominal response (e.g., Olton & Samuelson, 1974). In addition, however, response latency is also sensitive to what are referred to as nonspecific rate depressant drug effects.

The two-lever, FR-10 procedure and some similar procedures also define two further measures. FRF (first reinforcement responding) represents the total number of responses occurring on either lever prior to the time the animal totals 10 responses on either the appropriate lever (in training sessions) or the selected lever (in test sessions). With an FR-10 schedule median values are typically 10, but individual FRF measures may show variations (Colpaert, 1978b; Colpaert, Niemegeers, & Janssen, 1978b) which suggest that FRF, like choice latency, may at times reflect the varying degrees of certainty for choosing among nominal responses. However, variations of both response latency and FRF often are limited or erratic, and the two measures have as yet found little use in DD research.

Finally, the percentage of responses on the selected lever is a variate which may reflect response control by the reinforcer in test sessions if responding on the selected lever is being reinforced (see Figure 1). Some of the features of this parameter will be demonstrated (see below).

The methods used in the analysis of drug discrimination and druhe analysis of drug discrimination and drug generalization data vary considerably according to whether the main dependent variable is taken to be either quantitative or nominal. Where applicable, the two analytical approaches will be considered below in parallel.

Analysis of Discrimination

The purpose for analyzing a given drug discrimination can be one of at least two different types. One may wish to determine the degree a given discriminandum controls the discriminative responding in one or different sets of experimental conditions, and methods are available to measure response control. Given a particular set of experimental conditions, one may also wish to analyze the discriminability of one or several training drugs, and additional methods are available to measure drug discriminability.

Measurement of Response Control

DD training is typically² continued until the animal reaches an arbitrarily chosen criterion of performance in S and D sessions. The conventional method of measuring response control consists of determining the number of sessions required to reach this criterion (sessions-to-criterion: STC; e.g., Overton, 1982). This method may have some practical utility, but it acts to conceal a number of interesting variations in response control. The control of discriminative responding which drugs may exert in the DD paradigm is a somewhat complex, dynamic process whica somewhat complex, dynamic process which can be divided into two phases—an acquisition phase and an asymptotic phase.
 

As indicated above, acquisition can be characterized by the number of (S and D) sessions required to reach some arbitrarily set criterion. When the main dependent variable is taken to be quantitative, the criterion can be the percentage of treatment-appropriate responding (e.g., 70, 80, 90%) or some other value a in 2 or N consecutive S and D sessions. This measurement can be written: STC_{a; N; S,D}. Note that DD studies have differed in terms of the values of a and N that were implemented. When a nominal variable is used, the STC can be determined for a criterion requiring that the response was correct (i.e., appropriate to treatment) in n out of N consecutive S and D sessions (STC_{n/N; S,D}). Note that measurements of acquisition using this formula may differ according to the length of the criterion run n/N (Colpaert et al., 1980b). STC has also been analyzed for S and D sessions separately (Colpaert, 1978b); such an analysis may be of interest since it may appear that the STC to respond S in S sesit may appear that the STC to respond S in S sessions (STC_{n/N; S}) is dissimilar to the STC to respond D in D sessions (STC_{n/N; D}). Cocaine HCl at 10 mg/kg, for example, is peculiar among many other training drug conditions because rats acquire the D response in cocaine sessions much faster that the S response in saline sessions (Colpaert, 1978b); the asymmetry may vary with training dose and, again, with n/N (Colpaert, 1978b).

Measures of asymptotic performance offer an alternative to measures of acquisition in characterizing response control (Colpaert et al., 1980b). Unlike measures of acquisition, measures of asymptotic performance are independent from the unavoidably arbitrary criterion that must be chosen to define acquisition. This may be important, since speed of acquisition as measured by STC may not strictly covary with asymptotic accuracy; for example, STC is longer with 0.04 mg/kg fentanyl (vs. saline) than with 10 mg/kg of cocaine (vs. saline), but asymptotic accuracy seems to be higher with fentanyl than with cocaine (Colpaert, 1978b). However, it may take a large number of sessions before asymptotic performance is reached (e.g., Colpaert et al., 1980b), and a sizable number of sessions at asymptote is required for measures to be accurate and reliable. The principal measure of asymptotic performance is the percentage of sessions in which the response was treatment appropriate (with the nominal vatment appropriate (with the nominal variable) or some measure of central tendency of the percentage of treatment-appropriate responding over sessions (with the quantitative variable). If applied to nominal data, this measure (% AP_S,D) of asymptotic performance is similar to the measure d' of discriminability in signal detection theory (SDT; Colpaert, 1978b). Further refinement can be achieved by deriving the percentage of errors from saline (% AE_S) and drug sessions (% AE_D) separately. If applied to nominal data, the % AE_D to % AE_S ratio offers a measure of symmetry in performance which is similar to the measure of response bias in SDT (Colpaert, 1978b). The determination of symmetry in asymptotic performance may be important because generalization data may covary with changes in response bias (Colpaert & Janssen, 1981).

SDT requires³ the response to be nominal in nature (Green & Swets, 1974), and the DD response may fulfill this requirement (Colpaert, 1978b). This raises the question whether the attractive analytical framework developed for SDT can be applied to DD data. Apart from some practical considerations such as number of observations, a consideration arguing against the application of SDT analyses to DD data is that the threshold dose for generalization varies (Colpaert et al., 1978c; Colpaert, Niemegeers, & Janssen, 1978d). This has led t & Janssen, 1978d). This has led to the argument that the criterion underlying the decision is not stable (Colpaert, Maroli, & Meert, 1982). It is a fundamental assumption of SDT that the criterion be stable (Green & Swets, 1974) and a classical SDT analysis strictly requires that this assumption be satisfied for the analysis to be permissible (Pastore & Scheier, 1974; Treisman, 1976). The discrimination data thus may not be amenable to SDT analysis (Colpaert et al., 1982). It is of interest to note here that the SDT assessment of pain has recently been criticized (Coppola & Gracely, 1983) on similar grounds.
 

The statistical analysis of the measures of response control discussed above is fairly straightforward. The statistical analysis differs, however, depending on the use of a nominal or quantitative variable, and appropriate nonparametric tests are available (Siegel, 1956) for the comparison of both related and independent samples. Adequate data on the distribution characteristics of the quantitative variable are not available, and it is prudent to use nonparametric statistics as opposed to statistics that make any important assumption about the distribution of the dependent variable.bout the distribution of the dependent variable.

Measurement of Drug Discriminability

To measure drug discriminability requires (i) that some criterion of response control be defined and (ii) that it be determined to what extent the criterion level of response control is reached at several doses covering the training drug's full dynamic range.

The above has made it apparent that a multitude of criteria of response control can be defined, and the outcome of an analysis of drug discriminability probably varies depending on the criterion implemented. An example of a criterion is that responding be appropriate on 10 out of 10 consecutive sessions and that it be reached within a cut-off number of 100 D and S sessions.

Two general methods are then available to explore the training drug's dynamic range. One consists of using different groups of animals at different training doses and of determining the percentage of animals reaching the criterion of response control (e.g., Overton, 1982). It typically appears that this percentage increases orderly as a function of training dose, and the statistical methods of Litchfield & Wilcoxon (1949) or Finney (1971) are often used to derive ED₅₀ values, confidence limits, and slope; these methods can be used further to compare drugs in terms of potency and of slope. A second method consists of initially training a single set of subjects on a relatively high training dose jects on a relatively high training dose and of determining the percentage of animals reaching criterion. The same animals are then retrained on a progressively lowered training dose until a dose is found where none of the animals reach criterion (Colpaert et al., 1980a). The percentage of animals reaching criterion can be used to obtain ED₅₀ values, confidence limits, and slope. Fentanyl is the only drug whose discriminability has been examined with both methods, and the two studies (Colpaert et al., 1980a, 1980b) yielded similar ED₅₀ values. The method of lowering training dose progressively has the unique feature of establishing the lowest discriminable dose (LDD) in individual animals; it also has revealed characteristics of drug discrimination and generalization which are not accessible by any other method (Colpaert et al., 1980a).

Drug discriminability is often viewed, albeit implicitly, as an immutable drug property (e.g., Overton, 1982), and some caution may be appropriate regarding this view. There is evidence that drug discriminability may vary with the subjects' condition (Weissman, 1976), and it may also depend on other independent variables, including procedural variables (Richards, 1978).

Analysis of Generalization

It is with the analysis and subsequent interpretation of generalization data that the difference in using a quantitative or nominal dependent variable ntitative or nominal dependent variable is perhaps most critical; the two variables are for that reason discussed separately here.

Using a Quantitative Measure

The analysis of the percentage of drug-appropriate responding in test sessions is essentially an analysis of group data. In any given test the percentage typically varies anywhere between 0 and 100% among subjects, and some measure of central tendency is taken to represent the group data. The mean (±1 SEM) is used most commonly for this purpose but may be misleading. This is because the distribution of the quantitative variate often is not Gaussian, nor even symmetrical (Colpaert, 1977b; Stolerman & D'Mello, 1981); its typical shape is in fact bimodal, with one peak occurring around 10% (a value similar to what appears in saline sessions) and with a second peak occurring around 90% (a value similar to what is obtained in training drug sessions). Alternative measures of central tendency of the quantitative variate are the median (and 95% confidence limits) and modal values.

The next step in the analysis of quantitative generalization data consists of implementing a (preferably nonparametric) statistical test such as the Wilcoxon test (Siegel, 1956) for related samples to compare the test data with control data; the latter are obtained in saline and training sessions administered prior to or after the test session but in close temporal proximity session but in close temporal proximity to it. The test data are thus analyzed by means of two comparisons; the four most common outcomes⁴ of the analysis are given in Table 3.
 

Test data which are similar to saline control and significantly lower than training drug control are simply taken to conclude that generalization did not occur (see Table 3, outcome 1). Test data which are similar to training drug control and higher than saline control are taken to conclude that generalization did occur (see Table 3, outcome 4). Outcome 2 likely indicates that the number of observations and/or the magnitude of response control were inadequate for any statistical analysis of the data to be conclusive. Outcome 3 may be obtained when the central tendency of drug-appropriate responding is around 50%. Interpretations of this outcome have been that it represents random responding, that it may be due to behavioral drug toxicity, or that it represents some lesser form of stimulus generalization (e.g., Koek & Slangen, 1982). None of these interpretations are entirely satisfactory (Colpaert, 1984), and several authors note that (Colpaert, 1984), and several authors note that any interpretation of this outcome is troublesome (Holtzman, 1982a; Winter, 1978).
 

It may be useful to note here that, in addition to the percentage of drug-appropriate responses, some other quantitative measures of discrimination and generalization have also been proposed. One example is the progressive ratio procedure which assesses the breaking point of a ratio schedule (Winter, 1981). Another procedure is termed extended schedule transfer and assesses response perseverance (Schechter, 1981).

Using a Nominal Measure

A test of stimulus generalization using the nominal variable determines whether a given animal selected the training drug-appropriate response following administration of the test treatment; the analysis of generalizationl selected the training drug-appropriate response following administration of the test treatment; the analysis of generalization with this variable therefore is essentially an analysis of individual data. Generalization testing is, of course, conducted in a number of animals, and the results can be reported concisely by the percentage⁵ of animals selecting the drug lever (Colpaert et al., 1975a).
 

The statistical analysis of nominal generalization data is extremely simple. The analysis often is so straightforward that it is not even reported.that it is not even reported. Depending on the criterion used for training the animal at the time of acquisition, the actual total (S and D sessions combined) error rate can often be reduced to about 10% to 5%, or even less; as a result, response selection data can be accepted within the 0.10, 0.05, or < 0.05 level of statistical significance. For most practical purposes, therefore, a DL response occurring in a test session is said to indicate, quite simply, that generalization occurred in that particular animal; the occurrence of a SL response indicates that the animal did not generalize the test treatment.

In cases where error rates in S and D sessions differ (Colpaert, 1978b), it may be useful to refine the analysis. For example, rats discriminating, rats discriminating 10 mg/kg of cocaine from saline may show an error rate of 3.7% in saline sessions and of only 0.6% in training drug sessions (Colpaert et al., 1978c). Thus, in this study, the conclusion that SL responding in a test session indicates no generalization could be drawn with 6.2 times more confidence than the conclusion that DL selection indicates generalization. This refinement is often felt to be redundant, however, especially when neither the S nor the D error rate exceeds the universally accepted 5% level. Thus the outcome of a generalization test using the nominal variable is simply that generalization occurred in from 0 to 100% of the animals and that no generalization occurred in the other subjects (see Table 4).ects (see Table 4).
 

It is clear, then, that although the actual data may be very similar in appearance the conclusion reached from a generalization test differs markedly depending on the use of either a quantitative or nominal variable. The quantitative approach leads to the nominal conclusion that the test treatment did or did not produce generalization and finds no satisfactory interpretation for outcomes 2 and 3 in Table 3.⁶ The nominal approach leads to the conclusion that generalization occurred in a specified percentage or quantity of animals, and that it did not occur in the remaining, complementary quantity of subjects. It will become apparent below that this difference in conclusions presents little problem withs little problem with unanimous (0 or 100%) data but does become extremely important with all intermediary levels of responding.
 

Generalization Gradients

When different doses are tested of a treatment that produces stimulus generalization, then the effects of the test drug can be of two general types. The type being considered in this section is that the generalization proceeds orderly from 0 to 100% as a function of test dose (percentage generalization refers to the mean percentage of drug-appropriate responding in the quantitative approach and to percentage of animals selecting the D response in the nominal approach). The percentage-generalization data points are often plotted graphically in log-linear coordinates. The generalizationinates. The generalization gradient or dose-response curve can then be obtained either by connecting the data points or by fitting them by the function of best fit. The function of best fit may be curvilinear, but it appears that a rectilinear function is often adequate in log-linear coordinates. An important advantage of a rectilinear function is that it allows the description of the data by two simple numericals (i.e., ED₅₀ and slope b; the ED₅₀ and slope are obtained from the regression equation y = a + bx; Colpaert, 1982a).

The methods of Finney (1971) and of Litchfield and Wilcoxon (1946) are usually used in the statistical evaluation of ED₅₀ and slope. The use of these methods is not entirely appropriate, however, since both, however, since both analyses require that the data samples taken at different doses be independent. The problem can, of course, be avoided by obtaining independent samples, but the solution is often impractical since it requires much larger numbers of trained animals. Truly appropriate statistical methods for evaluating ED₅₀ doses and slope for related data samples do not seem to be available. The somewhat inappropriate use of these methods, however, is not likely to yield exceedingly misleading results since there is little reason to assume that related generalization data would differ greatly from independent data. It is hoped that alternative statistical methods will be developed to rectify this problem.

The application of the Finney (1971) and Litchfield & Wilcoxon (1946)& Wilcoxon (1946) analyses to quantitative generalization data presents an additional, more important problem. Both analyses have been developed for nominal data, and the analyses' definitions of such terms as ED₅₀ values, confidence limits of the ED₅₀, and slope do not apply when data are quantitative.

Partial Generalization

Partial generalization can be said to occur (Colpaert & Janssen, 1984) when test treatments produce levels of discriminative responding which are intermediate between those produced by the treatments which the animals are trained to discriminate. Partial generalization can occur either along a dose-response curve progressing orderly from the 0 to the 100%rom the 0 to the 100% level of effect, or it may represent the ceiling level of a drug's effect.

A previous section has made it apparent that the analysis of partial generalization is simple with the use of a nominal variate. The use of the nominal variate has revealed (Colpaert, 1984) that (a) the lowest generalized dose (LGD) may differ among rats, (b) the slope of the gradient reflects between-subject variation in terms of LGD, and (c) the occurrence of partial generalization along a curve proceeding from 0 to 100% indicates that the test dose is higher than or equal to LGD in some subjects and lower than LGD in the other subjects (case 1 in Table 5). Where partial generalization represents the ceiling level of a drug's effect, the use of the nominal use of the nominal variate reveals that this can result from one of two sources (Colpaert, 1984; Colpaert, Niemegeers, & Janssen, 1979). First, generalization may occur in only some but not all animals, and generalization can then be specified as partial in terms of subjects (case 2 in Table 5). Secondly, it may occur that all animals generalize the test treatment, but the generalization does not persist at higher doses. In this case the ceiling level of drug effect fails to reach 100% because the doses showing generalization do not coincide among all subjects; this generalization is considered as partial in terms of doses (case 3 in Table 5).

It is difficult, and for many practical purposes impossible, to identifyssible, to identify and characterize partial generalization with the quantitative variable. This is because the quantitative variable is responsive both to the discriminative stimulus and to the primary reinforcer (e.g., Figure 1) so that variations in the percentage of drug-appropriate responding cannot simply and directly be attributed to discriminative control alone. The previous section delineates some of the difficulties in data analysis and subsequent interpretation. The examples given below will reveal that converting the quantitative variable into a nominal variate to overcome the problem may be neither adequate nor permissible. It would thus seem that the quantitative variable cannot serve to analyze partial generalization because it is co-determinedit is co-determined by variables other than the discriminative effects of the treatment being tested.
 

Other Variables

Overall response rate, response latency, first reinforcement responses, and the percentage of responding on the treatment-appropriate (training sessions) or selected lever (test sessions) are analyzed by conventional methods of statistical analysis. The distributions of these variables may be of a number of different types, and it is prudent to use nonparametric statiatistical analysis. The distributions of these variables may be of a number of different types, and it is prudent to use nonparametric statistics throughout. Latency and first reinforcement responding data have at times been analyzed for subjects individually (Colpaert, 1978b), but it is common practice to only consider group data.

Examples

The foregoing discussions have made it apparent that two dependent variables are currently being used in DD research. The two variables differ greatly in terms of their conceptual origin and in terms of the formal analyses and statistical treatment which they can be given. It has been argued, however, that quantitative and nominal outcomes of generalization tests correlate (Koek & Slangen, 1982; Stolerman & D'Mello, 1981) and that this correlation is reason to suggest that the quantitative and the nominal variable constitute equally adequate measures of discrimination and generalization. The examples given below show that the two variables effectively correlate in some but not in other cases and demonstrate some of the mechanisms for the occurrence of similarities and dissimilarities.

The Fentanyl Gradient

Figure 2 summarizes generalization test data with 0.0025 to 0.04 mg/kg fentanyl obtained (Colpaert et al., 1975a) from 12 rats trained to discriminate 0.04 mg/kg of fentanyl from saline in the two-lever, FR-10, food-reinforced DD procedure. In test sessions responses on the selected lever were reinforced, and the data are expressed botforced, and the data are expressed both nominally (percentage of animals selecting the DL) and quantitatively (percentage of responses on the DL; mean ±1 SEM). It appears that in this set of data the two variables correlate almost perfectly (see Figure 2). The mechanism of covariation is evident: Whatever the lever that individual animals selected, all or almost all subsequent responding occurred on the selected lever (for detailed individual data, see Table 5 in Colpaert et al., 1975a). Apparently, the pharmacological stimulus controlled lever selection, while the reinforcement of subsequent responding on the selected lever controlled much of the behavior occurring later in the session (see also Figure 1).

Haloperidol and Fentanyl

The following (unpublished) experiment was conducted in seven rats trained to discriminate 0.04 mg/kg of fentanyl from saline, and it is an extension of an earlier study (Colpaert et al., 1977). Before test sessions, the animals were trained with 0.04 mg/kg of fentanyl (subcutaneously, 30 minutes prior to testing) and pretreated (subcutaneously, 60 minutes prior to testing) with a haloperidol dose ranging from 0.0025 to 0.63 mg/kg. All rats pretreated with any dose of haloperidol selected the DL following the injection of fentanyl. The percentage of drug-appropriate responding, however, decreased to reach a level of about 50% at the 0.63 mg/kg dose (see Figure 3). In these data, t dose (see Figure 3). In these data, therefore, there is no correlation between the nominal and the quantitative variable. The lower part of Figure 3 further indicates that the percentage of responses on the selected lever was decreased by haloperidol; this decrease covaried perfectly with the decrease of the quantitative variable and is responsible for it.
 

Figure 2: Fentanyl generalization gradient in rats trained to discriminate 0.04 mg/kg of fentanyl from saline. Two dependent variables are presented in the upper graph—the nominal (left ordinate) and the quantitative(left ordinate) and the quantitative variable (right ordinate). The lower graph presents the percentage of responding on the lever (i.e., SL or DL) which the animals selected.

The source of this discrepancy is that the quantitative, but not the nominal, variable is responsive to the effects of the primary reinforcer (see Figure 1). In the data in Figure 2, the effects of the primary reinforcer happened to be isodirectional with those of the pharmacological stimulus so that the quantitative variable covaried with the nominal variable. The treatment tested in Figure 3 interfered with the effects of the primary reinforcer and thus caused the quantitative variable to diverge from the nominal variable. These and similar data (Colpaert et al., 1978b) demonstrate the particularly powerful effects that haloperidol and other antipsychotic agents may exert on control of responding by a primary reinforcer.
 

Figure 3: Effects of haloperidol in rats trained to discriminate 0.04 mg/kg of fentanyl from saline. The symbols S and F are adjacent to data points representing performance on saline and 0.04 mg/kg fentanyl control sessions, respectively. See also legend to Figure 2.

Haloperidol and Cocaine

The data presented in Figure 3 highlight the difficulties which occur with interpreting intermediate quantitative results. The problem originates from the susceptibility of the quantitative variable to the effects of the primary reinforcer. In addition, it is not readily possible to control for the effects of test treatments on the primary reinforcer; in one study (Colpaert et al., 1977) 0.08 mg/kg of haloperidol and 0.04 mg/kg of fentanyl had only limited effects on the percentage of responding on the selected lever, but the percentage dropped very markedly (i.e., to 50%) when the two treatments were combined. To administer haloperidol and fentanyl aloneedly (i.e., to 50%) when the two treatments were combined. To administer haloperidol and fentanyl alone thus would not have served as an appropriate control for the effects of the haloperidol-fentanyl combination on primary reinforcement. The haloperidol-fentanyl combination also reduced overall response rate to about 10% of saline control values (Colpaert et al., 1977). In our experience there is no simple correlation between drug effects on rate and on the percentage of responses on the selected lever. But it often appears that treatments which severely reduce overall response rate also block the effects of the primary reinforcer on responding in at least some animals. The vulnerability of the quantitative variable to such effects renders its use in such conditions particularly adventurous. The nominal variable, of course, is also reinforced; the requiremented; the requirement for the nominal response to occur is that it be reinforced by secondary reinforcement. At sufficiently large doses almost any pharmacological treatment reduces responding altogether and thus blocks the effects of secondary reinforcement; at this point it is no longer possible to measure discrimination and generalization.⁷
 

The interference of the effects of reinforcement with quantitative generalizationith quantitative generalization data has generally been a problem and has been particularly confusing in studies on the effects of neuroleptic agents on discriminative effects of cocaine and other stimulants. Specifically, some authors (Ho & Silverman, 1978) conclude that haloperidol blocks cocaine discrimination, whereas others (Cunningham & Appel, 1982) find any possible haloperidol effect on cocaine discrimination to be of little pharmacological relevance. The confusion is apparent from a set of data (Colpaert et al., 1978b) presented in Figure 4. Haloperidol reduced the value taken by the quantitative variable (see Figure 4) much like it did in the haloperidol-fentanyl experiment (see Figure 3). This time, however, haloperidol also made some animalso made some animals select the saline lever, indicating that cocaine antagonism occurred. Unlike what appeared in Figure 2, however, the decrease in the quantitative variable in Figure 4 cannot simply be attributed to pharmacological antagonism, because haloperidol also reduced the percentage of responding on the selected lever. The data in Figure 4 thus demonstrate just how confusing and misleading the use of the quantitative variable can be. They also demonstrate, however, the validity of the nominal variable in what are admitted to be difficult experimental conditions. Specifically, that haloperidol does in effect antagonize cocaine could be confirmed in an additional (unpublished) study. Nine rats were trained to discriminate 10 mg/kg of cocaine from saline,ocaine from saline, and the cocaine gradient (doses 0.63 to 10 mg/kg; administered 30 minutes prior to testing) was determined following pretreatment (60 minutes prior to testing) with saline and 0.04 or 0.16 mg/kg of haloperidol. It appeared that haloperidol made the cocaine dose-response curve shift to the right (see Figure 5). The magnitude of this shift was proportional to haloperidol dose and leaves no doubt that the neuroleptic antagonizes the discriminative effects of cocaine.
 

Figure 4: Effects of pretreatment (60 minutes prior to test)etreatment (60 minutes prior to test) with saline or 0.04 to 0.31 mg/kg of haloperidol on responding in the DD task after injection with 10 mg/kg of cocaine (30 minutes prior to test). The data were obtained from seven rats trained to discriminate 10 mg/kg of cocaine from saline (Colpaert, Niemegeers, & Janssen, 1978).

Figure 5: Dose-response curve of cocaine following pretreatment with either saline (open circles) or 0.04 (filled circles)i>open circles) or 0.04 (filled circles) and 0.16 mg/kg (X) of haloperidol. Data were obtained from nine rats that were trained to discriminate 10 mg/kg of cocaine from saline. See text for details.

As indicated in the introduction, many classes of drugs of abuse produce pharmacologically specific discriminative effects in laboratory animals. The preclinical evaluation of drug abuse potential with the DD method essentially consists of determining whether the novel drug produces discriminative effects similar to those of any one or several of the known classes of drugs of abuse. The use of drug discrimination in the preclinical evaluationination in the preclinical evaluation of drug abuse potential is suggested to proceed along the following phases.

The first phase is determining whether the new drug which is to be evaluated produces stimulus generalization with a training drug (i) that has some significant pharmacological property in common with the new drug and (ii) that belongs to one of the classes of drugs of abuse. For example, a new drug exerting naloxone-reversible inhibition of gastrointestinal motility will be tested for generalization in rats trained to discriminate morphine or another opiate agonist from saline (Colpaert et al., 1975b). The observation that the new drug generalizes with morphine is taken to predict that it may produce opiate-like subjective effects in humans. The alternative outcomealternative outcome is taken to predict that the new drug will not produce such subjective effects but does not guarantee that it is devoid of stimulus properties similar to that of drugs of abuse other than opiates. The latter can be determined in phase two, where the drug is tested for generalization in different sets of rats trained, for example, on cocaine, chlordiazepoxide, D⁹-THC, LSD, nicotine, and phencyclidine. Phase two testing may require a sizable effort and is required only when there are reasonable grounds to suggest that the new drug shares pertinent pharmacological or structural properties with any of these substances of abuse. In both phase one and phase two, negative outcomes have great face validity, especially if training doseslly if training doses and other conditions have been chosen to yield inclusive generalizations (see Colpaert, 1982a). The significance of positive outcomes can be examined more closely; a new drug producing generalization with a 0.005 mg/kg training dose of fentanyl may produce subjective effects resembling those of cyclazocine (Colpaert et al., 1980b). But the drug is unlikely to produce the more threatening subjective effects of prominent opiates of abuse such as heroin if it fails to generalize with a 0.04-mg/kg training dose of fentanyl (Colpaert, 1984; Colpaert & Janssen, 1984; see also Holtzman, 1982a, 1982b). Much work is currently being undertaken to further develop the drug discrimination methodology such that stimulus properties of drugs can be characterizedan be characterized with extreme refinement in both quantitative and qualitative terms. A third step can be considered if the new drug yields negative outcomes in phases one and two. The third phase is determining whether animals can be trained to discriminate the drug from saline. In as much as it cannot be ascribed to drug impairment of learning at relevant doses, a negative outcome here is strongly suggestive of the drug producing no significant subjective effects. Generalization tests with prototypical agents of all important classes of drugs of abuse are pertinent if training does succeed. Positive outcomes of these tests have a relevance that is similar, but perhaps not identical, to those that may occur in phases one and two. An entirely negativen entirely negative outcome of these tests is essentially inconclusive; it merely indicates the drug to possess discriminative effects that are unlike those of known substances of abuse. Whether these unprecedented stimulus effects are likely to generate an unprecedented drug abuse remains for clinical data to decide.

The preparation of this paper was supported in part by a grant from the I.W.O.N.L.

Appel, J. B., White, F. J., & Holohean, A. M. (1982). Analyzing mechanism(s) of hallucinogenic drug action with drug discrimination procedures. Neuroscience & Biobehavioral Reviews, 6, 529-536.Reviews, 6, 529-536.

Barry, H., III, & Krimmer, E. C. (1977). Discriminable stimuli produced by alcohol and other CNS depressants. In H. Lal (Ed.), Discriminative stimulus properties of drugs (pp. 73-92). New York: Plenum Press.

Colpaert, F. C. (1977a). Narcotic cue and narcotic state. Life Sciences, 20, 1097-1108.

Colpaert, F. C. (1977b). Drug-produced cues and states: Some theoretical and methodological inferences. In H. Lal (Ed.), Discriminative stimulus properties of drugs (pp.5-21). New York: Plenum Press.

Colpaert, F. C. (1978a). Discriminative stimulus properties of narcotic analgesic drugs. Pharmacology Biochemistry & Behavior, 9, 863-887.

Colpaert, F. C. (1978b). Some properties of drugs as physiological signals:as physiological signals: The FR procedure and signal detection theory. In F. C. Colpaert & J. L. Slangen (Eds.), Stimulus properties of drugs: Ten years of progress (pp. 217-242). Amsterdam: Elsevier/North Holland Biomedical Press.

Colpaert, F. C. (1982a). The pharmacological specificity of opiate drug discrimination. In F. C. Colpaert & J. L. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 3-16). Amsterdam: Elsevier/North Holland Biomedical Press.

Colpaert, F. C. (1982b). Increased naloxone reversibility in fentanyl dose-dose discrimination. European Journal of Pharmacology, 84, 229-231.

Colpaert, F. C. (1983). Drug discrimination and the behavioral analysise behavioral analysis of drug action. In C. F. Lowe, M. Richelle, D. E. Blackman, & C. M. Bradshaw (Eds.), Behaviour analysis and contemporary psychology. Hillsdale, NJ: Lawrence Erlbaum & Associates.

Colpaert, F. C. (1984). Drug discrimination: Behavioral, pharmacological, and molecular mechanisms of discriminative drug effects. In S. R. Goldberg & I. P. Stolerman (Eds.), Behavioral analysis of drug dependence. New York: Academic Press.

Colpaert, F. C., & Janssen, P. A. J. (1981). Factors regulating drug cue sensitivity: The effect of frustrative non-reward in fentanyl-saline discrimination. Archives Internationales de Pharmacodynamie et de Therapie, 254, 241-251.

Colpaert, F. C., & Janssen, P. A. J. (1982a). Factors regulating982a). Factors regulating drug cue sensitivity: The effects of dose ratio and absolute dose level in the case of fentanyl dose-dose discrimination. Archives Internationales de Pharmacodynamie et de Therapie, 258, 283-299.

Colpaert, F. C., & Janssen, P. A. J. (1982b). Factors regulating drug cue sensitivity: Limits of discriminability and the role of a progressively decreasing training dose in cocaine-saline discrimination. Neuropharmacology, 21, 1187-1194.

Colpaert, F. C., & Janssen, P. A. J. (1982c). OR discrimination: A new drug discrimination method. European Journal of Pharmacology, 78, 141-144.

Colpaert, F. C., & Janssen, P. A. J. (1984). Agonist and antagonist). Agonist and antagonist effects of prototype opiate drugs in rats discriminating fentanyl from saline: Characteristics of partial generalization. Journal of Pharmacology and Experimental Therapeutics, 230, 193-199.

Colpaert, F. C., Lal, H., Niemegeers, C. J. E., & Janssen, P. A. J. (1975a). Investigations on drug produced and subjectively experienced discriminative stimuli. 1. The fentanyl cue, a tool to investigate subjectively experienced narcotic drug actions. Life Sciences, 16, 705-716.

Colpaert, F. C., Maroli, A. N., & Meert, T. (1982). Parametric effects in the discrimination of intracranial stimulation: Some methodological and analytical issues. Physiology & Behavior, 28, 1047-1058.>28, 1047-1058.

Colpaert, F. C., & Rosecrans, J. A. (Eds.). (1978). Stimulus properties of drugs: Ten years of progress. Amsterdam: Elsevier/North Holland Biomedical Press.

Colpaert, F. C., & Slangen, J. L. (Eds.). (1982). Drug discrimination: Applications in CNS pharmacology. Amsterdam: Elsevier/North Holland Biomedical Press.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1976a). Theoretical and methodological considerations on drug discrimination learning. Psychopharmacologia, 46, 169-177.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1976b). Discriminative stimulus properties of fentanyl and morphine: Tolerance and dependence. Pharmacology Biochemistry & Behavior, 5,mp; Behavior, 5, 401-408.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1977). Differential haloperidol effect on two indices of fentanyl-saline discrimination. Psychopharmacology, 53, 169-173.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1978a). Drug-cue conditioning to external stimulus conditions. European Journal of Pharmacology, 49, 185-188.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1978b). Neuroleptic interference with the cocaine cue: Internal stimulus control of behavior and psychosis. Psychopharmacology, 58, 247-255.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1978c). Factors regulating drug cue sensitivity. A long term study on the cocaineg term study on the cocaine cue. In F. C. Colpaert & J. A. Rosecrans (Eds.), Stimulus properties of drugs: Ten years of progress (pp. 281-299). Amsterdam: Elsevier/North Holland Biomedical Press.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1978d). Changes of sensitivity to the cueing properties of narcotic drugs as evidenced by generalization and cross-generalization experiments. Psychopharmacology, 58, 257-262.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1979). Discriminative stimulus properties of cocaine: Neuropharmacological characteristics as derived from stimulus generalization experiments. Pharmacology Biochemistry & Behavior, 10, 535-546.i>, 10, 535-546.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1980a). Factors regulating drug cue sensitivity: Limits of discriminability and the role of a progressively decreasing training dose in fentanyl-saline discrimination. Journal of Pharmacology and Experimental Therapeutics, 212, 474-480.

Colpaert, F. C., Niemegeers, C. J. E., & Janssen, P. A. J. (1980b). Factors regulating drug cue sensitivity: The effect of training dose in fentanyl-saline discrimination. Neuropharmacology, 19, 705-713.

Colpaert, F. C., Niemegeers, C. J. E., Lal, H., & Janssen, P. A. J. (1975b). Investigations on drug produced and subjectively experienced discriminative stimuli. 2. Loperamide, an antidiarrheal devoid of narcoticheal devoid of narcotic cue producing actions. Life Sciences, 16, 717-728.

Coppola, R., & Gracely, R. H. (1983). Where is the noise in SDT pain assessment? Pain, 17, 257-266.

Cunningham, K. A., & Appel, J. B. (1982). Discriminative stimulus properties of cocaine and phencyclidine: Similarities in the mechanism of action. In F. C. Colpaert & J. L. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 181-192). Amsterdam: Elsevier/North Holland Biomedical Press.

Dews, P. B. (1958). Stimulant actions of methamphetamine. Journal of Pharmacology and Experimental Therapeutics, 122, 137-147.

Eckenhoff, J. E., & Oech, S. R. (1960). The effects of narcotics and antagonists upon respiration and circulation in man. Clinical Pharmacology and Therapeutics, 1, 483-524.

Extance, K., & Goudie, A. J. (1981). Inter-animal olfactory cues in operant drug discrimination procedures in rats. Psychopharmacology, 73, 363-371.

Finney, D. J. (1971). Statistical methods in biological assay (2nd ed.). London: Griffin Press.

Fraser, H. F. (1968). Methods for assessing the addiction liability of opioids and opioid antagonists in man. In A. Wikler (Ed.), The addictive states (pp. 176-187). Baltimore: Williams and Wilkins Co.

Fraser, H. F., & Isbell, H. (1961). Human pharmacology and addictiveness of ethyl 1-(3-cyano-3,3-phenylpropyl)iveness of ethyl 1-(3-cyano-3,3-phenylpropyl)-4-phenyl-4-piperidine carboxylate hydrochloride (R-1132, diphenoxylate). Bulletin on Narcotics, 13, 29-43.

Fraser, H. F., Van Horn, G. D., Martin, W. R., Wolbach, A. B., & Isbell, H. (1961). Methods for evaluating addiction liability. (A) "Attitude" of opiate addicts toward opiate-like drugs, (B) a short-term "direct" addiction test. Journal of Pharmacology and Experimental Therapeutics, 133, 371-387.

Gellert, V. F., & Holtzman, S. G. (1979). Discriminative stimulus effects of naltrexone in the morphine-dependent rat. Journal of Pharmacology and Experimental Therapeutics, 211, 596-605.

Glennon, R. A., Rosecrans, J. A., & Young, R. (1982). The use of R. (1982). The use of the drug discrimination paradigm for studying hallucinogenic agents. A review. In F. C. Colpaert & J. L. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 69-98). Amsterdam: Elsevier/North Holland Biomedical Press.

Green, D. M., & Swets, J. A. (1974). Signal detection theory and psychophysics. Huntington: Robert E. Krieger Publishing Co.

Haertzen, C. A. (1970). Subjective effects of the narcotic antagonists cyclazocine and nalorphine in the Addiction Research Center Inventory (ARCI). Psychopharmacologia, 18, 366-377.

Haertzen, C. A. (1974). Subjective effects of narcotic antagonists. In M. C. Braude, L. S. Harris, E. L. May, J. P. Smith, & J. E. Villarrealh, & J. E. Villarreal (Eds.), Narcotic antagonists (pp. 383-398). New York: Raven Press.

Hayes, R. L. (1978). Experimental design and data analysis in studies of drug discrimination: Some general considerations. In B. T. Ho, D. W. Richards, III, & D. L. Chute (Eds.), Drug discrimination and state dependent learning (pp. 193-201). New York: Academic Press.

Herling, S., & Woods, J. H. (1981). Discriminative stimulus effects of narcotics: Evidence for multiple receptor-mediated actions. Life Sciences, 28, 1571-1584.

Ho, B. T., & Silverman, P. B. (1978). Stimulants as discriminative stimuli. In F. C. Colpaert & J. A. Rosecrans (Eds.), Stimulus properties of drugs: Ten years of progress (pp. 53-68). Amsterdam: Elsevier/Northerdam: Elsevier/North Holland Biomedical Press.

Holtzman, S. G. (1975). Effects of narcotic antagonists on fluid intake in the rat. Life Sciences, 16, 1465-1470.

Holtzman, S. G. (1982a). Stimulus properties of opioids with mixed agonist and antagonist activity. Federation Proceedings, 41, 2328-2332.

Holtzman, S. G. (1982b). Discriminative stimulus properties of opioids in the rat and squirrel monkey. In F. C. Colpaert & J. L. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 17-36). Amsterdam: Elsevier/North Holland Biomedical Press.

Jarbe, T. U. C., & Swedberg, M. D. B. (1982). A conceptualization of drug discrimination learning. In F. C. Colpaert & J. L. Slangenert & J. L. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 327-341). Amsterdam: Elsevier/North Holland Biomedical Press.

Jasinski, D. R. (1973). Effects in man of partial morphine agonists. In H. W. Kosterlitz, H. O. J. Collier, & J. E. Villarreal (Eds.), Agonist and antagonist actions of narcotic analgesic drugs (pp. 94-103). Baltimore: University Park Press.

Jasinski, D. R. (1977). Assessment of the abuse potentiality of morphine-like drugs (methods used in man). In W. R. Martin (Ed.), Drug Addiction I (Handbook of Experimental Pharmacology, Vol. 45, pp. 197-258). Berlin: Springer-Verlag.

Jasinski, D. R., & Mansky, P. A. (1972). Evaluation of nalbuphineluation of nalbuphine for abuse potential. Clinical Pharmacology and Therapeutics, 13, 78-90.

Jasinski, D. R., Martin, W. R., & Haertzen, C. A. (1967). The human pharmacology and abuse potential of n-allylnoroxymorphone (naloxone). Journal of Pharmacology and Experimental Therapeutics, 157, 420-426.

Jasinski, D. R., Martin, W. R., & Sapira, J. D. (1968). Antagonism of the subjective, behavioral, pupillary, and respiratory depressant effects of cyclazocine by naloxone. Clinical Pharmacology and Therapeutics, 9, 215-222.

Koek, W., & Slangen, J. L. (1982). The role of fentanyl training dose and of the alternative stimulus condition in drug generalization. Psychopharmacology, 76, 149-156.

Krimmer, E. C., & Barry, H., III. (1977). Discriminable stimuli produced by marihuana constituents. In H. Lal (Ed.), Discriminative stimulus properties of drugs (pp. 121-136). New York: Plenum Press.

Litchfield, J. T., & Wilcoxon, F. (1949). A simplified method of evaluating dose-effect experiments. Journal of Pharmacology and Experimental Therapeutics, 96, 99-113.

Locke, K. W., Brown, D. R., & Holtzman, S. G. (1982). Effects of opiate antagonists and putative mu- and kappa agonists on milk intake in rat and squirrel monkey. Pharmacology Biochemistry & Behavior, 17, 1275-1279.

Lowy, M., Starkey, C., & Yim, G. (1981). Stereoselective effectsey, C., & Yim, G. (1981). Stereoselective effects of opiate agonists and antagonists on ingestive behavior in rats. Pharmacology Biochemistry & Behavior, 15, 591-596.

Olton, D. S., & Samuelson, R. (1974). Decision making in the rat. Journal of Comparative and Physiological Psychology, 87, 1134-1147.

Ostrowski, N., Rowland, N., Foley, T., Nelson, J., & Reid, L. (1981). Morphine antagonists and consummatory behaviors. Pharmacology Biochemistry & Behavior, 14, 549-559.

Overton, D. A. (1979). Influence of shaping procedures and schedules of reinforcement on performance in the two-bar drug discrimination task: A methodological report. Psychopharmacology, 65, 291-298.

Overton, D. A. (1982). Comparison of the degree of discriminabilityree of discriminability of various drugs using the T-maze drug discrimination paradigm. Psychopharmacology, 76, 385-395.

Pastore, R. E., & Scheier, C. J. (1974). Signal detection theory: Considerations for general applications. Psychological Bulletin, 81, 945-958.

Richards, D. W., III. (1978). A functional analysis of the discriminative stimulus properties of amphetamine and pentobarbital. In B. T. Ho, D. W. Richards, III, & D. L. Chute (Eds.), Drug discrimination and state dependent learning (pp. 227-247). New York: Academic Press.

Schechter, M. D. (1981). Extended schedule transfer of ethanol discrimination. Pharmacology Biochemistry & Behavior, 14, 23-25.vior, 14, 23-25.

Schuster, C. R., Fischman, M. W., & Johanson, C. E. (1981). Internal stimulus control and subjective effects of drugs. In T. Thompson & C. E. Johanson (Eds.), Behavioral pharmacology of human drug dependence (National Institute on Drug Abuse Research Monograph 37, pp. 116-129). Washington, DC: U.S. Government Printing Office.

Shannon, H. E., & Holtzman, S. G. (1976). Evaluation of the discriminative effects of morphine in rats. Journal of Pharmacology and Experimental Therapeutics, 198, 54-65.

Siegel, S. (1956). Nonparametric statistics. New York: McGraw-Hill.

Silverman, P. B., & Ho, B. T. (1977). Characterization of discriminative response control by psychomotor stimulants. In H. Lal (Ed.), Discriminative.), Discriminative stimulus properties of drugs (pp. 107-119). New York: Plenum Press.

Skinner, B. F. (1938). The behavior of organisms: An experimental analysis. New York: Appleton-Century-Crofts.

Stevens, S. S. (1946). On the theory of scales of measurement. Science, 103, 677-680.

Stolerman, I. P., Baldy, R. E., & Shine, P. J. (1982). Drug discrimination procedure: A bibliography. In F. C. Colpaert & J. L. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 401-441). Amsterdam: Elsevier/North Holland Biomedical Press.

Stolerman, I. P., & D'Mello, G. D. (1981). Role of training conditions in discrimination of central nervous system stimulants by rats. Psychopharmacology,i>Psychopharmacology, 73, 295-303.

Swedberg, M. D. B., & Jarbe, T. U. C. (1982). Morphine cue saliency: Limits of discriminability and third state perception by pigeons. In F. C. Colpaert & J. C. Slangen (Eds.), Drug discrimination: Applications in CNS pharmacology (pp. 147-164). Amsterdam: Elsevier/North Holland Biomedical Press.

Treisman, M. (1976). On the use and misuse of psychophysical terms. Psychological Review, 83, 246-256.

Weissman, A. (1976). The discriminability of aspirin in arthritic and non-arthritic rats. Pharmacology Biochemistry & Behavior, 5, 583-586.

Weissman, A. (1978). Generalization of the discriminative stimulus properties of D ⁹-tetrahydrocannabinol to cannabinoidsdrocannabinol to cannabinoids with therapeutic potential. In F. C. Colpaert & J. A. Rosecrans (Eds.), Stimulus properties of drugs: Ten years of progress (pp. 99-124). Amsterdam: Elsevier/North Holland Biomedical Press.

White, J. M., & Holtzman, S. G. (1981). Three-choice drug discrimination in the rat: Morphine, cyclazocine and saline. Journal of Pharmacology and Experimental Therapeutics, 217, 254-262.

Winter, J. C. (1975). The stimulus properties of morphine and ethanol. Psychopharmacologia, 44, 209-214.

Winter, J. C. (1978). Drug-induced stimulus control. In D. E. Blackman & D. J. Sanger (Eds.), Contemporary research in behavioral pharmacology (pp. 209-237). New York: Plenum Press.7). New York: Plenum Press.

Winter, J. C. (1981). Drug-induced stimulus control and the concept of breaking point: LSD and quipazine. Psychopharmacology, 72, 217-218.