In the early nineteenth century, Combe (1830) reported that ethanol could produce state-dependent learning (SDL). Specifically, he reported the case of an Irish deliveryman who had left a parcel at an incorrect address while drunk and could not remember where he had left it until several days later when he was once again intoxicated. No theoretical explanation for this phenomenon was advanced until 1882 when Ribot postulated that "organic sensations" were responsible for drug induced SDL and also for the temporary amnesias observed in cases of multiple personality, somnambulism, and fugue. According to Ribot’s theory, sensations reflecting the current physiological and pharmacological state of the body were powerful determinants of which memories could, and could not, be retrieved at any particular moment in time. Specifically, the memories that could be retrieved were those that had been encoded on previous occasions when the body was in the same state as that present at the time of retrieval (Ribot, 1882, p. 115; 1891, p. 34). Ribot’s theory was later revised by Semon (1904, 1909) who stated that the retrievability of memories was determined by the "energetic condition of the brain." In Semon’s model the influence on memory retrieval of the brain’s energetic condition was not assumed to be mediated by sensory events. Apparently, the idea that drugs could influence memory retrieval via an SDL-like process was generally accepted in nineteenth century England (Collins, 1868; Elliotson, 1840, p. 646; Macnish, 1835, p. 30; Siegel, 1982, 1983; Winslow, 1860, p. 338). No one has reviewed the subsequent literature to determine how widely this idea continued to be accepted during the first third of the twentieth century, although the inclusion of SDL in the plot of Charlie Chaplin’s movie City Lights indicates that the idea was not entirely lost (McDonald, Conway, & Ricci, 1965, pp. 191-196).

Girden and Culler (1937) reported the first experimental demonstration of SDL, which they called "dissociated" learning, in dogs virtually paralyzed with raw curare or erythroidine. In their dogs, conditioned leg flexion responses (muscle twitches) learned while the animals were drugged subsequently failed to appear during tests while the animals were undrugged and vice versa (Girden, 1942a, 1942b, 1942c, 1947). Girden did not refer to the earlier literature on SDL and did not interpret his observation as an instance of stimulus control. Instead, he proposed an alternate theory suggesting that drug rendered the cortex nonfunctional. Conditioning carried out in the normal undrugged state established associations only in the cortex which was rendered inoperative during tests when the animal was drugged. Conditioning in the drug state was postulated to establish associative connections in subcortical structures which were inhibited by the cortex during subsequent tests without drug. Theoretical models for SDL such as Girden’s, which did not postulate a causal role for drug-induced stimuli, were later labeled "neurological" models by Bliss (1974), and we will adopt that notation throughout this paper.

Conger (1951) first reported that ethanol could acquire discriminative control. He showed that in a telescope-alley task rats could learn to run down the alley when drugged and to withhold that response when undrugged or vice versa. He interpreted these results as showing that ethanol produced sensory stimuli that provided the basis for the rats’ discriminated responding. Subsequently, as in the nineteenth century, the notion became moderately widespread that drugs could produce SDL and that because they had "stimulus effects," performance of reion became moderately widespread that drugs could produce SDL and that because they had "stimulus effects," performance of responses learned in a drug condition might be impaired if the drug context was changed (e.g., Auld, 1951; Barry, Wagner, & Miller, 1962; Grossman & Miller, 1961; Holmgren, 1964; Sachs, Weingarten, & Klein, 1966; Shmavonian, 1956). Several investigators reported data which they interpreted as showing SDL effects of this type (Barry, Miller, & Tidd, 1962; Heistad, 1958; Heistad & Torres, 1959; Miller & Barry, 1960), and Heistad (1957) argued that new behaviors learned while a psychiatric patient was drugged might become unretrievable after drug treatment was discontinued. Of these reports Conger’s contained the most convincing data. The nature of the postulated drug-induced stimuli was not identified.

A few years later Overton (1961, 1964, 1966) and Stewart (1962) again trained rats to discriminate presence versus absence of drug and additionally conducted for the first time substitution tests. In these tests, after drug versus no-drug (D versus N) training with one drug (D1), the rats were tested under the influence of other drugs (D2, D3, D4 . . .) to determine under these novel drugs whether the rats would make the response that had been reinforced under the training drug or, alternately, the response that had been reinforced in the no-drug condition. Not surprisingly, the rats selected the response reinforced under the training drug during tests with drugs that were pharmacologically similar to the training drug. Only a few tests were conducted with drugs that differed pharmacologically from the training drug, and under those test conditions the rats selected the response that had been reinforced during no-drug training sessions.

These early reports of successful drug discrimination (DD) training raised two questions that were not immediately answered. First, which drugs would allow animals to easily learn DDs, and which drugs would make DD learning difficult or impossible? Second, what degree of "specificity" would be produced by D versus N training? In other words, after D versus N training, how wide a variety of drugs would produce D choices during test sessions, which drugs would produce a mixture of D and N choices, and which drugs would produce only N choices.

Overton (1971) reported on the relative discriminability of about three dozen centrally acting drugs. Also, on the basis of substitution test results primarily obtained with drugs in four pharmacological classes, he for the first time claimed that rats would make D choices during substitution tests if and only if they were tested under the influence of drugs that pharmacologically resembled the training drug. However, it took several more years before enough data accumulated to fully support this statement as a general property of DD-trained rats.

After 1970, the number of reported DD studies started to increase and by the late 1970s it was firmly established (1) that most centrally acting drugs could be discriminated, and (2) that after D versus N training, rats would usually make D choices during substitution tests with drugs that produced pharmacological actions similar to those of the training drug and not when tested with drugs that produced any of a variety of other pharmacological effects. Reviews of work conducted during this period can be found in several books and articles (Barry, 1974, 1978; Bliss, 1974; Colpaert & Rosecrans, 1978; Colpaert & Slangen, 1982; Erdmann, 1979; Ho, Richards, & Chute, 1978; Lal, 1977; Overton, 1968a, 1974, 1978d, 1984; Schuster & Balster, 1977; Stolerman, Baldy, & Shine, 1982; Thompson & Pickens, 1971; Winter, 1978a). The findings of these studies set the stage for the subsequent application of DD procedures as a method for the investigation of a variety of psychopharmacological and neuropharmacological questions, including studies of the effects of drugs that underlie drug abuse.

This section describes the behavioral procedures that have been developed for use in drug discrimination studies and the types of data that each procedure yields.

Drug Discrimination Training Procedures

Drug discriminations usually involve appetitively or aversively motivated instrumental behaviors. For example, a rat may be trained to turn left in a T-maze to escape shock when drugged (D) and to turn right in the same T-maze during sessions conducted without drug (N). Typically a series of training sessions are conducted in which the drug condition is varied in an alternating sequence (NDNDND . . .). Training sessions are usually conducted daily with drug elimination occurring overnight, thus allowing the D or N condition to be reinstated freshly during the next training session. Usually each daily session lasts 5 to 30 minutes, and responding during the first trial of each session (before any reinforcers are administered) is considered to reflect the degree to which the rat can select the state-appropriate response on the basis of the imposed drug condition.

The DD training procedure is analogous to procedures that are used to establish sensory discriminations (e.g., turn right when the cue light is on and turn left when it is off) except that the differential sensory conditions are replaced by drug conditions which are said to have stimulus effects. When the rat has learned to reliably perform the response appropriate to the current drug state, a drug discrimination is said to have been learned, and the drug is said to be discriminable. The term degree of discriminability is used to refer to the strength, amount, or salience of the drug’s stimulus effects, as reflected by the speed at which discriminative control is established or by the stability and accuracy of the resulting discrimination.

Many variations of this procedure are possible. Instead of alternating, the D and N conditions may be imposed in a pseudorandom or double alternation sequence. Rats can be trained to perform a response in one drug condition and to withhold it in a second condition (Ando, 1975; Conger, 1951; Harris & Balster, 1971; Winter, 1973, 1974, 1975, 1977). They can be trained to discriminate among three different drug conditions in a task where three different responses are possible (Overton, 1967; White & Holtzman, 1981, 1983a, 1983b). More generally, any of the following conditions may be discriminated.
 

The task may involve a series of discrete trials or continuous instrumental behavior as in lever-pressing operant paradigms. Reinforcement contingencies may be the same in each training condition (e.g., reinforce correct responses and extinguish incorrect responses) or may be different (e.g., use an FR-10 schedule of reinforcement of correct responses during D sessions and an FR-20 during N sessions).

Selection from among these procedures is based on empirical data showing the relative advantages of each procedure with respect to various criteria including the following: (1) Discriminations are more rapidly learned with some procedures than with others. (2) Higher asymptotic accuracy of discrimination is obtained with some procedures than with others. (3) Some procedures are more sensitive than others (i.e., rats can discriminate lower doses). (4) Some procedures yield a different degree of specificity than others. At this time the most widely used DD method employs appetitively motivated operant resp procedures are more sensitive than others (i.e., rats can discriminate lower doses). (4) Some procedures yield a different degree of specificity than others. At this time the most widely used DD method employs appetitively motivated operant responding in a compartment with two levers mounted side by side on one wall. Every tenth press on the correct lever is reinforced (FR-10 schedule), and presses on the incorrect lever are recorded but have no programmed consequence (Colpaert & Rosecrans, 1978; Lal, 1977; Overton, 1979b).

Performance during DD training sessions yields two major indices of interest: (1) speed of acquisition under each training condition, as reflected by sessions to criterion or number of errors during acquisition and (2) asymptotic accuracy of discrimination.

Substitution Test Procedures

After rats have learned to discriminate a particular training drug and dose versus the no-drug condition, test sessions can be scheduled with various novel drug conditions imposed. During these test sessions responses on either lever are reinforced (or sometimes extinguished) and any of the following drug conditions may be imposed.

Training Drug (TD) at a novel dose
TD with various time intervals after injection
TD plus putative antagonist
TD plus transmitter manipulator
TD plus synergistic drug
Metabolite of TD
Proposed precursor(s) of TD
One or more novel drugs

The two major types of data obtained from test sessions are (1) percentage of trials under each test condition when the D response is performed and (2) rate (or latency) of responding. The first index typically allows the investigator to subdivide the tested drug conditions into three categories: (1) drugs that mimic the effects of the training drug sufficiently so that D responses regularly occur, (2) drugs with which some intermediate percentage of D choices (25 to 75%) are observed, and (3) drugs which produce no D responses. The second index (rate) is usually considered to indicate the rate-suppressing effects of the test compound, although it is sometimes interpreted as reflecting the degree of novelty of the test condition (Barry, McGuire, & Krimmer, 1982).

As methods for studying the drug effects that underlie drug abuse, drug discriminations have both advantages and disadvantages. This section will review the advantages.

DDs Are Rapidly Learned

A priori one might think that DDs would be learned more slowly than sensory discriminations since specialized receptors, sensory pathways, and brain mechanisms have evolved to mediate the perception of sensory stimuli, whereas corresponding structures designed to mediate the perception of drug effects are not known to exist. Contrary to this prediction, however, the speed of acquisition of drug and of sensory discriminations is similar when comparable training procedures are employed (Duncan, Phillips, Reints, & Schechter, 1979; Kilbey, Harris, & Aigner, 1971, Overton, 1964, 1968b, 1971; Spear, Smith, Bryan, Gordon, Timmons, & Chiszar, 1980).

Most Centrally Acting Drugs Are Discriminable

Most centrally acting drugs are measurably discriminable, and perhaps 75% of centrally acting drugs produce effects sufficiently discriminable so that DD training paradigms can establish stable response control. Table 1 indicates the relative degree of discriminability of various types of centrally acting drugs. Virtually all abused drugs are readily discriminated. The only known exceptions are the occasionally abused salicylates such as aspirin which are discriminated only with difficulty (Overton, 1982a; Weissman, 1976).
 

Drugs which act only outside the central nervous system have usually been found to be much less discriminable than centrally actDL. Classification based on data published by Overton, 1982a.  

Drugs which act only outside the central nervous system have usually been found to be much less discriminable than centrally acting drugs (Colpaert, Kuyps, Niemegeers, & Janssen, 1976; Colpaert, Niemegeers, & Janssen, 1975a; Downey, 1975; Hazell, Peterson, & Laverty, 1978; Ho, Richards, & Chute, 1978; Miksic, Shearman, & Lal, 1980; Overton, 1971; Schuster & Balster, 1977; Valentino, Herling, Woods, Medzihradsky, & Merz, 1981).

A few transmitter blockers such as naltrexone and mecamylamine are discriminable only at doses considerably higher than those required to produce the antagonistic actions characteristic of these drugs. Apparently we can conclude that the antagonistic actions per se of these drugs are not very discriminable (Carter & Leander, 1982; Overton, 1982a).

Sensitivity Is Relatively High

For several years DDs were a phenomenon most commonly produced by high doses of drugs (e.g., Overton, 1966). Subsequently, the two-lever fixed-ratio operant procedure (Colpaert, Niemegeers, & Janssen, 1975a) allowed the use of more moderate dosages of most drugs, and the dosages now used in DD procedures are comparable to those required by other preclinical behavioral techniques.

Several investigators have tried to determine the lowest dosage of drugs that could be discriminated by using a titration paradigm in which rats discriminated progressively lower dosages until a dosage was reached at which DDs no longer could be maintained (Colpaert, Niemegeger could be maintained (Colpaert, Niemegeers, & Janssen, 1980a; Overton, 1979a; Zenick & Goldsmith, 1981). These experiments showed that DDs could be maintained at dosages as low as one tenth those commonly employed in DD experiments. In other studies several groups of rats have been trained, each with a different dose of the training drug, to determine which doses could acquire control of D versus N DDs (Colpaert, Niemegeers, & Janssen, 1980b; Greenberg, Kuhn, & Appel, 1975; Overton, 1964, 1982a). The amount of training required to establish DDs increases as training dose is reduced, and the asymptotic accuracy of discrimination decreases. Therefore discriminations based on the lowest discriminable doses of a drug do not provide a very practical DD assay procedure.

Quantitative Measurement of Discriminable Effects Is Possible

We have referred several times to the degree of discriminability of drugs without describing how this property might be measured. Three indices of DD performance have been used to indicate degree of discriminability: (1) asymptotic accuracy, (2) speed of acquisition, and (3) ability to substitute for another discriminable drug.

Asymptotic accuracy refers to the relative frequency of correct (state appropriate) response selections after prolonged DD training. Investigators have generally categorized asymptotic accuracy as high (consistent selection of theacy as high (consistent selection of the correct response), medium (occasional lapses of discriminative control), or absent (discrimination never learned). Numerical indices computed from asymptotic accuracy have seldom been used to provide quantitative estimates of degree of discriminability, and such indices appear unlikely to be useful inasmuch as asymptotic accuracy is influenced by several factors other than degree of discriminability, including the schedule of reinforcement that is employed (Colpaert, et al., 1980b; Harris & Balster, 1971; Overton & Hayes, 1984; Schuster & Balster, 1977).

The speed of acquisition of a DD can be expressed in terms of the number of training sessions before some criterion level of accuracy is achieved or in terms of the average accuracy during early DD training sessions (i.e., the area under the learning curve). Figure 1 shows plots of sessions to criterion versus dosage for several drugs and indicates that as dosage is increased speed of acquisition also increases in a monotonic fashion (i.e., sessions to criterion decreases). Comparable data have been obtained with a variety of drugs (Colpaert et al., 1980b; Overton, 1974, 1982a), and a detailed comparison of the effectiveness of various indices of degree of discriminability has recently been reported (Overton, Leonard, & Merkle, 1985).

The substitution test procedure is quicker than either of the previously described techniqur of the previously described techniques but is only applicable in special cases. The center curve in the left half of Figure 2 shows the percentage of drug-lever choices during tests with various doses of phenobarbital in rats previously trained to discriminate phenobarbital (40 mg/kg) versus no drug. As the test dosage decreased, the percentage of drug-lever choices also decreased. There is no reason to expect a linear relationship between the percentage of presses on the drug lever and the degree of discriminability of the test dosage. However, such substitution test results can indicate the relative ability of various test drugs (or doses) to produce discriminable effects like those of the training drug (Overton, 1974). They are only useful for comparisons between drugs with discriminable effects that are essentially identical to those of the training drug. When this requirement is met, the substitution test technique will provide information about the relative discriminability of various test conditions more rapidly than either the speed of acquisition or the asymptotic accuracy method. For this reason among others, the substitution test method has been frequently employed in DD studies.
 

F height=366 width=521>

Figure 1: Plots showing the influence of dosage on the speed of acquisition of drug discriminations. In a T-Maze rats were trained to turn right to escape shock when undrugged, and to turn left when drugged. During daily 10-trial sessions, the drug condition alternated (NDNDND . . .) as did the reinforced choice. Criterion was correct choices on the first trial during 8 out of 10 consecutive sessions. Y-axis is the geometric mean number of training sessions before the beginning of criterion performance by three or more rats. Diazepam was suspended and the remaining drugs were dissolved before intraperitoneal injection. With each drug, 11 to 34 rats were trained.

DD methods compare favorably with self-administration procedures with regard to their ability to produce quantitative estimates of the strength of subjective drug effects that may underlie drug abuse. Recall that with self-administration procedures it has been difficult to provide more than a yes/no answer to the question of how frequently a drug might be abused. The influence of parameters such as unit dosage and time course of action has made it difficult to provide more quantitative estimates of abuse liability except via certain rather tedious techniques. By comparison, DD procedures can more easily yield data that allows a quantitative comparison ofield data that allows a quantitative comparison of the strength of the subjective effects of various drugs. Additionally, DD procedures may have the capability in certain instances to independently measure several different subjective effects of a drug, including some which promote drug abuse (euphorigenic) and others that deter drug abuse (nocioceptive).
 

Figure 2: Average percentage of drug-lever presses by three groups of rats during substitution tests conducted with various doses of two drugs. The rats were initially trained in a two-lever operant task with water reinforcement for correct responses (FR-10 schedule) and extinction of incorrect responses. In Groups 1 to 3, the imposed drug conditions during bar 1 and bar 2 sessions were as follows:

Group 1:  Group 2:  Group 3: (Phenobarbital) vs. (Ethanol or Diazepam or Ketamine or No Drug)  (Phenobarbital) vs. (No Drug)  (Phenobarbital or Ethanol or Ether or Diazepam) vs. (No Drug)

After the required discriminations had been learned, the plotted data were collected during test sessiodiscriminations had been learned, the plotted data were collected during test sessions in which responses on either lever were reinforced. Each test session continued until four reinforcements had been earned. Approximately four test sessions per drug at each dosage were conducted with each rat. N = 6 rats/group. Test results with phenobarbital show that the training procedures produced differing degrees of bias and/or quantitative specificity such that the dose-response curve was shifted to the right in Group 1 and to the left in Group 3. Test results with methaqualone showed larger differences in specificity; the Group 3 dose-response curve shifted further to the left, and substitution failed to occur at any tested dose in Group 1. Since the dose-response curves obtained with methaqualone are farther apart than those obtained with phenobarbital, the results suggest that the groups differed in qualitative specificity as well as in bias and/or quantitative specificity.

High Qualitative Specificity Produces Sharp Generalization Gradients

After D versus N DD training with a particular drug and dose, tests may be conducted using drugs that pharmacologically differ from the training drug. Such tests reveal very sharp generalization gradients. Typically the drug choice is not made during tests with any novel drugs except those producing pharmacological effects similar to those of the training drug. The effect is quite striking, and the results are ana quite striking, and the results are analogous to those which one might expect if animals had been trained using a visual discriminative stimulus and were then tested with the visual stimulus absent and with an auditory stimulus substituted for it. This type of experiment has been repeated many times using a different training drug in each successive group of animals; sharp generalization gradients have been obtained irrespective of the training drug employed (Overton, 1971, 1972, 1978a). The trained animals are said to exhibit high qualitative specificity and can be used as an assay to detect the presence or absence of the actions of the training drug and of its close pharmacological relatives. Indeed, the high qualitative specificity produced by D versus N training is the primary property which has made DD training a useful psychopharmacological assay procedure (Barry, 1974; Colpaert & Slangen, 1982; Lal, 1977). Since a high degree of qualitative specificity has been observed in a variety of DD training paradigms, it appears to reflect some intrinsic property of drug-induced stimuli or of their perception by trained animals rather than a characteristic induced by the particular training paradigm employed (Chance, Murfin, Krynock, & Rosecrans, 1977).

Specificity Can Be Controlled to Some Degree

Until recently, the degree of specificity produced by DD training could not be intentionally varied bng could not be intentionally varied by the investigator. However methods for controlling specificity are now under development. First, in many drug classes specificity can be decreased (or sometimes increased) by reducing the dose of the training drug that is discriminated. The mechanism by which training dosage influences specificity is not known (Colpaert et al., 1980a, 1980b; Colpaert & Janssen, 1982b; Koek & Slangen, 1982b; Shannon & Holtzman, 1979; Stolerman & D’Mello, 1981; Teal & Holtzman, 1980b; White & Appel, 1982). Second, specificity can be either increased or decreased by training with paradigms that require the rat concurrently to discriminate several drugs (Colpaert & Janssen, 1982c; Overton, 1982c). For example, a rat trained to discriminate chlordiazepoxide (lever 1) versus ethanol, pentobarbital, ketamine or no drug (lever 2) might provide a high specificity assay for new benzodiazepine drugs (see Figure 2). Third, response bias, and perhaps qualitative specificity, can be manipulated by using asymmetrical reinforcement contingencies of various types. With such contingencies the reinforcers applied for correct (or incorrect) presses are different during D than during N training sessions. For example, rats may be punished for incorrect responses in one drug condition and not in the other (Colpaert & Janssen, 1981).

At least two mechanisms can be postulated by which a training procedurestulated by which a training procedure might lead to increased specificity. (1) The subjects might discriminate a smaller subset of the discriminable effects of the training drug. (2) The subjects might insist on a closer match between the effects of the training drug and those of a test drug as a condition for making the D response during a test session. At present it is not known which of these mechanisms is operative in the procedures for varying specificity described above. In this context we should mention the breaking point substitution test procedure described by Winter (1981a) in which the subject is subjected to progressively increasing ratios of reinforcement until its exclusive preference for the D lever gives way to mixed responding. Apparently this procedure yields higher specificity than do conventional substitution test paradigms, even though the training paradigm is unaltered. Hence, the second mechanism appears to be operative.

In the past specificity was generally viewed as fixed, and the results of substitution tests were often interpreted as directly reflecting the degree of stimulus overlap between the training drug and a test drug. For example, 100% D choices were often interpreted to mean that the test drug was essentially identical with the training drug, 0% D choices were inferred to mean no overlap in actions, and 50% D choices were interpreted as indicating partial (perhaps 50%) overlap. Unfortunatartial (perhaps 50%) overlap. Unfortunately, no two investigators exactly agreed on the interpretation of test results. The recent discovery that specificity can vary depending on the training paradigm employed should appreciably alter interpretation of test results. We now know that a particular test drug can produce 0, 50, or 100% D choices, depending on the degree of specificity induced by the training procedure (Overton, 1982c). Apparently, the percent D choices observed during a substitution test has no fixed meaning and can only be interpreted with reference to the degree of specificity induced by the training paradigm employed.

Masking Apparently Does Not Occur

During substitution tests after D versus N training, suppose that drug X substitutes for the training drug and that substitution does not occur during tests when drugs X and Y are concurrently administered. One possible interpretation is that Y pharmacologically antagonizes or blocks X so that the actions of X are truly absent during the test. Another possibility, at least in principle, is that the discriminable effects of Y somehow mask (occlude, prevent the animal from perceiving) the effects of X even though they are present. The term mask is used here to convey the same meaning that it denotes in sensory psychophysics—a situation in which one stimulus prevents the perceptin which one stimulus prevents the perception of a second concurrently presented stimulus.

After D versus N DD training, rats frequently have been tested while simultaneously drugged with the training drug and with another drug; such tests usually have been performed to identify antagonists of the training drug. In most data thus far reported, the only instances when a second drug prevents the rat from performing the drug response are those in which the second drug pharmacologically antagonizes the effects of the training drug so that its discriminable effects are not present (Altshuler, Applebaum, & Shippenberg, 1981; Browne, 1981; Browne & Ho, 1975a; Browne & Weissman, 1981; Hernandez, Holohean, & Appel, 1978; Hirschhorn & Rosecrans, 1974a, 1974b; Jarbe, 1977; Jarbe & Ohlin, 1977; Romano, Goldstein, & Jewell, 1981; Rosecrans, 1979; Shearman & Lal, 1981; Silverman & Ho, 1980). Thus the results of these experiments suggest that masking seldom if ever occurs. Indeed, in interpreting their data most DD investigators have assumed that masking did not occur, and absence of masking has been an implicitly assumed property of DDs which, surprisingly, has almost never been explicitly discussed or demonstrated (except see Colpaert, 1977a; Witkin, Carter, & Dykstra, 1980).

However, the conclusion that masking does not occur between drug stimuli can hardly be regarded as firmly established. Firstle regarded as firmly established. Firstly, experimental designs can be envisaged that should be considerably more sensitive to masking than any of the accidental tests for its occurrence presently in the literature (see below). Secondly, there are a few reported data which suggest that masking may have occurred (e.g., Browne, 1981; Browne & Weissman, 1981; Witkin et al., 1980). Possibly these reports indicate instances of masking; if so, they constitute a first step towards enumerating the conditions under which masking will in fact occur. However, notwithstanding these occasional instances of possible masking, it appears that most drugs do not mask the effects of most other drugs in DD preparations (Overton, 1983, 1984).

DDs Have Some Face Validity for the Study of Drug Abuse

For the study of most psychopharmacological issues, DDs have little or no face validity (see below). However, DDs do appear to provide an animal model which has some face validity for the study of (1) subjective side effects of drugs and (2) subjective or hedonic drug actions which may underlie drug abuse.

With respect to subjective side effects of drugs, it has not been demonstrated that DDs provide accurate preclinical data on such effects. However, it makes sense that if DDs are based on the subjective (perceived) sensory consequences of drug effects in animals, then DDs may be able to provide information about analogoble to provide information about analogous drug effects in humans.

The application of DDs to the investigation of drug abuse has less face validity. Drug abuse is probably based, at least in part, on reinforcing stimulus effects of drugs (e.g., euphorigenic effects), and such reinforcing effects are probably one of the types of subjective effects capable of supporting DDs. Hence, DDs apparently provide a method for measuring and analyzing the sensory effects that underlie drug abuse. However, this application is complicated by the fact that DDs presumably can also be based on noxious subjective effects that would deter drug abuse and on affectively neutral subjective effects that are irrelevant to drug abuse. Also, drug abuse is often ascribed to nonsensory effects of drugs such as reduction of anxiety or selective amnesia for the dysphoric effects of drugs, and the DD preparation has no intrinsic face validity for the investigation of such drug actions. This paper primarily discusses the use of DDs to study the reinforcing stimulus effects of drugs.

Additionally, the DD preparation has been used to investigate the subjective effects of withdrawal in drug-tolerant animals and appears to have face validity for this purpose (Gellert & Holtzman, 1979; Miksic, Sherman, & Lal, 1981; Valentino, Herling, & Woods, 1983; Valentino, Smith, & Woods, 1981).

This review would be incomplete if it did not describe the most prominent disadvantages and limitations of the DD procedure.

DD Training Procedures Are Cumbersome and Slow

We indicated earlier that DDs are learned as rapidly as sensory discriminations, but even sensory discriminations are learned slowly when animals receive only one training trial per day, as is required in all DD training paradigms to allow drug elimination between sessions. Hence, the process of shaping and training an animal to reliably discriminate a drug typically requires 1 to 3 months. Even longer training is required to establish DDs involving several manipulanda, several drugs, or low doses (Overton & Hayes, 1980; White & Holtzman, 1981). Such training is expensive. Parametric improvements in DD training paradigms have somewhat reduced the duration of training and increased the accuracy of the resulting discriminations, and some further improvement along these lines can be expected (e.g., Overton, 1978c, 1979b; Overton & Hayes, 1984). Nonetheless, the time required to establish DDs will probably continue to be an expensive aspect of the procedure. Apparently, this is only a practical difficulty which does not significantly compromise interpretation of DD results once they are obtained.

Each Substitution Test Yields Only a=+1>Each Substitution Test Yields Only a Bit of Data

With most DD procedures another practical difficulty is posed by the small yield of data produced by each individual substitution test. In the majority of published DD experiments, one substitution test was conducted per day, and each substitution test yielded one binary bit of data per rat as the animal made either a D or an N choice. Hence to measure performance under a particular test condition with 4-bit accuracy, the test procedure had to be repeated four times. Since training sessions must be interspersed between tests and since each test drug is typically evaluated at several doses, more than a month of work is commonly required to generate a dose-response curve showing the ability of a single test compound to substitute for a single training drug. Obviously, such procedures do not always fit conveniently into the life spans of rodents or of investigators.

Fortunately, new substitution test procedures have recently been developed that use repeated tests with cumulatively increasing doses within a single test session. These allow a complete dose-effect curve for a novel compound to be obtained during a single protracted test session. Although these procedures only yield binary data at each individual dose and are not yet fully validated or universally employed, it appears that they will reduce the amount of time required to conduct DD experiments (Herling, Hamptonnduct DD experiments (Herling, Hampton, Bertalmio, Winger, & Woods, 1980; Jarbe, Swedberg, & Mechoulam, 1981; McMillan, 1982; McMillan, Cole-Fullenwider, Hardwick, & Wenger, 1982; Shannon & Holtzman, 1976a, 1976b). A priori, we might expect such procedures to have three disadvantages. First, after a series of injections at various intervals prior to a particular test trial, the blood level of drug is not entirely predictable and is obviously influenced by a variety of factors. Second, the result obtained during a particular test trial within a test session may not be independent of the result obtained during the preceding test trial, and a different percentage of drug choices may be obtained at a given blood level depending on whether the level was higher or lower during the preceding test trial. Finally, the repeated series of test trials that is employed during such cumulative dosing procedures appears more likely to disrupt accurate response control under the training drug conditions than would be the case if only a single test trial was administered. Apparently this effect is most likely to be troublesome in instances where the training conditions are only weakly discriminable, and in such cases it may be necessary to schedule a higher percentage of training sessions than would otherwise be required. In spite of these difficulties, experience with the method thus far suggests that none of its potential problgests that none of its potential problems are terribly serious and that the cumulative dosing procedure does increase the data yield obtained in DD studies.

Other procedures that yield more than one bit of data per substitution test have also been developed. However, these procedures have not been widely adopted, perhaps because they require relatively difficult training and/or test procedures (McMillan et al., 1982; McMillan & Wenger, 1983a; Schechter, 1981a; Shannon & Holtzman, 1976a, 1976b, 1979; Winter, 1981a).

DD Procedure Is Easiest With High Doses

Although DDs can be controlled by relatively low drug doses, DDs based on high doses are more stable and are less likely to be disrupted by substitution tests. Hence, in practice most DD studies use as high a training dose as is possible without disrupting behavioral performance in the DD task. With current DD preparations these doses are usually not outlandish, and they are lower than the doses used in preparations where disruption of ongoing behavior is used to indicate the presence of drug effects. Nonetheless, because high doses make DD experiments easier to conduct, one can recognize in the DD literature a tendency to use relatively high doses, even though this is not absolutely necessary in most cases.

Dose-Response Curves Are Somewhat Unstable

Discriminating a high training dose from no drug is easy be training dose from no drug is easy because the training dose and the no-drug condition are several JNDs (just noticeable differences) apart (Colpaert & Janssen, 1982a; Overton, 1968b, 1977a). However, it appears that the generalization gradient obtained during tests with intermediate doses is not entirely determined by such D versus N training; hence, the ED50s may vary considerably from rat to rat and within a single rat at different points in time (Colpaert, Niemegeers, & Janssen, 1978a, 1978c; Overton, 1984). Usually these shifts in quantitative specificity are not obvious in the average generalization curves for a group of subjects, as shifts in the ED50s for individual animals tend to cancel one another (Colpaert et al., 1980a). Nonetheless, the preparation would provide more quantitatively reliable results if some method could be found to reduce the variability of the dose response curves both across and within subjects. Colpaert and Janssen (1982a) showed that high dose versus low dose DD training would produce steeper dose response curves and presumably smaller ED50 differences within and between animals, but this technique has not yet been used in enough studies so that its utility can be evaluated.

Specificity Varies in Different Drug Classes

When discussing specificity it is convenient to refer to pharmacological classes of drugs. In this paper the term pharmacological class will be used in its conventional sense to indicate a group of drugs that are generally agreed to produce one or more of the same effects (e.g., depressants, antimuscarinics, tricyclic antidepressants). Obviously, the definitions of these classes change as our knowledge of drug actions improves.

Recent reports suggest that the degree of specificity produced by D versus N training depends on the type of drug used for training (Overton, 1978b). Three cases may be distinguished. (1) With several types of training drugs, specificity approximates the width of the pharmacological class in which the training drug is found. This is the case for antimuscarinics, narcotic agonists, tetrahydrocannabinols, and antihistamines among others (Barry & Kubena, 1972; Colpaert, 1978; Colpaert, Niemegeers, & Janssen, 1975b; Overton, 1977a, 1978a; Woods, Young, & Herling, 1982). (2) In other cases, specificity appears to be higher, and not all drugs in the training drug’s pharmacological class will mimic the training drug. This is true for cholinergics, antipsychotics, and narcotic mixed agonist/antagonists among others (Goas & Boston, 1978; Hirschhorn, 1977; Overton & Batta, 1979; Romano et al., 1981; Rosecrans, Spencer, Krynock, & Chance, 1978; Schechter & Rosecrans, 1972a). (3) In still other cases, specificity appears to be lower, as generalization is observed to drugs outside the training drug’s pharmacological class. For example, anesthetics, benzodiazepines, muscle relaxants, and some anticonvulsants all tend to generalize to one another, perhaps because they share depressant effects (Barry, 1974; Barry & Krimmer, 1979; Barry & Kubena, 1972; Colpaert, 1977a; Colpaert, Desmedt, & Janssen, 1976; Haug & Gotestam, 1982; Herling, Valentino, & Winger, 1980; Jarbe & McMillan, 1983; Lal & Fielding, 1979; Overton, 1966, 1976, 1977b; Takada, 1982). Other instances of generalization to drugs outside the training drug’s class also have been reported (Herling, Coale, Hein, Winger, & Woods, 1981; Herling & Woods, 1981; Holtzman, 1980; Overton, 1974; Shannon, 1981; Teal & Holtzman, 1980a).

In the preceding paragraph we used the breadth of pre-existing drug categories as a standard against which to compare specificity. Alternately, however, it may be the case that a particular D versus N training paradigm produces a fixed degree of specificity in all classes and that some pharmacological classes include a more diverse array of drugs than others, with regard to their discriminable effects, thus allowing cross-generalization between all members in some classes but not in others. At this time we have no way to measure breadth of generalization except by counting the number of drugs to which generalization occurs and by noting their apparent degree of pharmacological similarity to the training drug. Comparison to pharmacological categories established by other psychopharmacological procedures has been very useful in this enterprise and has led to the conclusion that generalization occurs to a greater or smaller portion of the drugs in the training drug’s class and sometimes to drugs in other classes depending on the type of drug used for training. Whether this result should be interpreted as showing that some classes are larger than others or as showing that specificity is higher after training in some classes than in others cannot be determined at this time.

The varying degree of specificity produced by D versus N training with different types of drugs is basically a disadvantage of the procedure, since a substantial amount of effort must be devoted to determining the degree of specificity obtained in a drug class before the results of substitution tests can be readily interpreted.

Methods for Controlling Specificity Are Poorly Developed

The methods now available for controlling the degree of specificity produced by DD training have serious limitations and/or have not been tested sufficiently to establish their limitations or capabilities.

Firstly, the method which involves a reduction in training dosage only alters specificity with some types of training drugs (Shannon & Herling, 1983). Additionally, in some cases a low training dosage may produce qualitatively different discriminable effects than a high dosage, thus changing the basis of the discrimination as well as the degree of specificity achieved (Colpaert, et al., 1976; Emmett-Oglesby, Wurst, & Lal, 1983). Finally, this method usually increases the duration of training and decreases the asymptotic accuracy and stability of DDs, thus making substitution test data more difficult to obtain and to interpret.

Secondly, the method which involves the use of several additional training drugs to manipulate specificity also poses some problems. In order to substantially increase specificity, ancillary training drugs rather similar to the primary training drug must be used. This prolongs training considerably, is only possible in classes where such drugs exist and where the relationship of their discriminable effects to those of the primary training drug is known, reduces the accuracy and stability of asymptotic discrimination, and probably can only be done in drug classes which produce readily discriminable actions (Colpaert & Janssen, 1982c; Overton, 1982c).

Thirdly, the method which uses asymmetrical reinforcement contingencies biases the animals to press either the D or N lever and thereby shifts the ED50 that is obtained during substitution tests with various doses of the training drug as well as with other drugs (Colpaert & Jato press either the D or N lever and thereby shifts the ED50 that is obtained during substitution tests with various doses of the training drug as well as with other drugs (Colpaert & Janssen, 1981; Koek & Slangen, 1982a; McMillan & Wenger, 1983b). However in unpublished studies we have found that such procedures do not alter qualitative specificity (the amount of substitution for the training drug obtained with dissimilar drugs) to a degree any greater than the extent to which they modify quantitative specificity (the amount of substitution observed with reduced doses of the training drug). If our results are correct and applicable to all varieties of asymmetrical reinforcement, this will indicate that such procedures are not useful for varying the qualitative specificity of the DD preparation. Further development of methods for controlling specificity will be required before such methods can be employed with confidence.

DD Procedures May Induce Tolerance

In some studies regimens of drug administration have apparently failed to induce tolerance to the discriminable effects of a drug. However, all these studies used a design that was flawed, since DD training continued while tolerance was induced thereby allowing the rats to learn to discriminate the reduced actions produced by drug in the now-tolerant animals (e.g., Bueno & Carlini, 1972; Colpaert, Kuyps, Niemegeers, & Janssen, 1976; Colpaert, Niemegeers, & Janssen, 1978b; Hirschhorn & Rosecrans, 1974c). In other studies which did not suffer from this unfortunate experimentuffer from this unfortunate experimental design, tolerance to discriminable drug effects has been observed (Barrett & Leith, 1981; Jarbe & Henriksson, 1973; McKenna & Ho, 1977; Miksic & Lal, 1977; Overton & Batta, 1979; Schechter & Rosecrans, 1972d; Shannon & Holtzman, 1976b; Witkin, Dykstra, & Carter, 1982; York & Winter, 1975). Most of these studies induced tolerance by administration of doses higher than the DD training dosage, and only Miksic & Lal (1977) have shown that the DD training regimen itself induced tolerance.

It is likely that all DD training procedures induce some degree of tolerance to the training drug and that all DD results are thus obtained from "abnormal" rats. Additionally, it appears that whenever the DD training paradigm is interrupted (e.g., over weekends) tolerance may begin to dissipate, thus shifting the pharmacological responsiveness of the subjects. Apparently tolerance may be reduced by the administration of test sessions with reduced doses of the training drug or by tests with drugs that differ from the training drug as well as by more prolonged vacations from training. An additional complexity is posed by the possibility that tolerance to various actions of the training drug may develop (and dissipate) to different degrees and at different rates. Such effects probably do not unduly complicate the interpretation of many DD results. However, Modrow et al. (1981) esults. However, Modrow et al. (1981) found different dosage generalization curves during test sessions preceded by D and by N training sessions—an effect possibly produced by short term (24 hour) tolerance induced by the training drug—and other effects of changes in tolerance may go unnoticed in many DD studies. It appears that the role of tolerance is often disregarded in DD experiments, and it would be reassuring if more investigators conducted data analyses designed to identify the effects of shifts in tolerance, if any, in their DD preparations.

Discriminated Effects May Not Be Related to Abuse

As mentioned earlier, not all sensory consequences of drug actions are conducive to drug abuse, and a priori it appears that we can distinguish three categories of subjective drug effects: (1) effects such as euphorigenic actions which promote drug abuse; (2) aversive (nocioceptive) effects which deter drug abuse; and (3) neutral subjective effects which can provide a basis for discriminative control but which neither increase nor decrease a compound’s abuse liability.

It would be convenient to speculate that effects in the first category are more frequently produced or are more readily discriminable than those in categories two and three but this apparently is not true. Some time ago this writer performed a rather extensive study designed to determine whether drugs with a high liability for abusher drugs with a high liability for abuse would always be readily discriminated and, conversely, whether drugs with low abuse liability would be relatively nondiscriminable. A high correlation between degree of discriminability and liability for abuse would have suggested that effects in category one were primarily responsible for most drug discriminations. However, the data showed a different result, as only a modest correlation was observed between degree of discriminability and abuse liability (Overton & Batta, 1977). The result clearly suggests that subjective effects in categories two and three form the basis for many, if not most, DDs.

Apparently, the face validity of DDs as a method for studying drug abuse is of a very limited nature. Even though DDs apparently can detect, quantify, and categorize many of the subjective effects that underlie drug abuse, they also detect many other types of subjective effects, and care must be taken to ascertain what type of subjective effect underlies each DD. Thus far, relatively little effort has been directed towards determining which DDs are based on effects related to drug abuse and which are not.

Theoretical Explanations for DDs Are Inadequate

In the next section we will discuss the adequacy of extant theories for DDs. Here, under the topic of disadvantages of the DD method, let us simply note that the primitive nature of DD theories makes it more diffture of DD theories makes it more difficult to interpret certain experimental results and extremely difficult to discover improved DD procedures except via a trial and error approach.

In spite of about 500 published studies elucidating the empirical properties of SDL and DDs, there is still no generally accepted and adequately detailed theoretical formulation about their causes that satisfactorily predicts all of the observed properties of SDL and DDs. With respect to this issue, it is useful to recall that SDL and DDs are probably based on the same drug effects (Overton, 1964) and that a complete theoretical formulation must predict the properties of both phenomena.

Symmetrical SDL: D Cues Replace N Cues

All early investigators who attributed SDL to drug-induced "stimuli" expected SDL to be symmetrical (i.e., they expected response decrements to be produced by both D—>N and N—>D transitions). This expectation was based on the assumption that "normal" or no-drug interoceptive cues would be replaced by "drug" cues when drug was administered, and that the converse would also be true (Auld, 1951; Barry, Miller, & Tidd, 1962; Barry et al., 1962; Heistad & Torres, 1959; Miller, 1957; Shmavonian, 1956). Model 1 in Table 2 shows this tan, 1956). Model 1 in Table 2 shows this traditional sensory formulation for SDL. In the N—>N and D—>D groups, the same environmental contextual cues (C) and interoceptive cues (N or D cues) are present during training and during the test for retention; hence, test session retrieval and performance are normal. In the N—>D group, learning takes place in the presence of C and N cues; during the retrieval test, N cues are replaced by D cues and hence memory retrieval is impaired. In the D—>N group, training takes place in the presence of C and D cues; during the test, D cues are replaced by N cues and so memory retrieval is impaired here also. The model appears straightforward and has survived for at least twenty-five years—a full century if we count Ribot’s initial presentation of the theory. A number of studies have reported data congruent with the theory (e.g., Bustamante, Jordan, Vila, Gonzalez, & Insua, 1970; Cole & Gay, 1976; Overton, 1964).
 

Asymmetrical SDL

Surprisingly, in many experiments impaired retrieval has been observed only after D—>N state changes and not after N—>D state changes. In other studies retrieval deficits have been observed after both D—>N and N—>D state changes, but they have been larger after D—>N transitions. The term asymmetrical SDL has been used to describe such results in which retrieval deficits were observed primarily in the D—>N group (Overton, 1968a).erved primarily in the D—>N group (Overton, 1968a).

Probably because of our theoretical presuppositions, asymmetrical SDL has been a neglected phenomenon. The original stimulus model for SDL predicted symmetrical SDL, and most neurological models for SDL have also been developed to explain symmetrical SDL (Bliss, 1974; Overton, 1973, 1978d). The statistical methods that are most commonly used to test the significance of SDL effects pool results in the D—>N and N—>D groups and do not indicate whether both N—>D and D—>N transitions produce similarly large retrieval deficits (Swanson & Kinsbourne, 1979).

In this context, the frequent reports of data indicating asymmetrical SDL have been problematic, since they appeared to reflect a phenomenon contrary to theoretical expectations. However, the data clearly do suggestta clearly do suggest that both symmetrical and asymmetrical SDL exist as bona fide phenomena. In the majority of reported SDL studies, some degree of asymmetry is present; in many reports impairment of retrieval is entirely lacking in the N—>D group (Avis & Pert, 1974; Barnhart & Abbott, 1967; Berger & Stein, 1969a, 1969b; Deutsch & Roll, 1973; Eich, Weingartner, Stillman, & Gillin, 1975; Evans & Patton, 1968; File 1974; Goldberg, Hefner, Robichaud, & Dubinsky, 1973; Henriksson & Jarbe, 1971; Holloway, 1972; Overton, 1974; Pappas & Gray, 1971; Peters & McGee, 1982). We will argue here that the sensory model for symmetrical SDL that has been accepted by most investigators since the mid-1950s is inconsistent50s is inconsistent with contemporary theories regarding the operation of contextual cues and that both symmetrical and asymmetrical SDL are entirely consistent with contemporary theories.

"No-Drug" Cues

First, let us ask what N cues are. Ribot (1882) asserted that they were sensations reflecting the normal functioning of the organs of the body and normal levels of activity in various regions of the brain. Today we might specify that, at a particular instant in time, N cues reflect the current degree of arousal, hunger, thirst, depression, distention of the stomach, vertigo, ringing in the ears (if any), degree of fatigue, et cetera. Next consider a question which is crucial; when a moderate dose of a psychoactiveose of a psychoactive drug is administered, how many N cues are substantially changed? To this writer it appears that the answer is either "some" or "none," depending on the drug and dosage administered. For example, pentobarbital may increase thirst and decrease arousal but may not alter the other interoceptive sensations mentioned. An antihistamine may decrease motion sickness and dry one’s mucosa but will probably have no other effects. Most N cues will still be present in the D state. In addition, drugs may create new sensory cues (e.g., hallucinations, euphoria). Such drug-induced cues (a different set of cues for each type of drug) will be superimposed on the still-present N cues and appear to match what contemporary DD investigators apparentlytigators apparently have in mind when they refer to drug-induced discriminative stimuli or D cues.

Consequences if D Cues Are Superimposed on N Cues

Model 2 in Table 2 recasts the traditional sensory interpretation of SDL in a more contemporary vein by assuming that drug injection simply adds D cues while leaving N cues unchanged. The N—>N and D—>D groups experience no change in contextual stimuli between training and testing and hence show unimpaired retrieval. So does the N—>D group because, although D cues have been added to its internal environment during the test session, both the C and N cues that were present during training are still present during test sessions and allow efficient retrieval. Only the D—>N group shows the D—>N group shows a retrieval deficit which occurs because D cues were present during training, became associated with the learned response, and are absent during the test for retrieval. It appears that if drug states simply create D cues without abolishing N cues, then a sensory model for SDL actually predicts asymmetrical SDL—not symmetrical SDL. This idea, incidentally, does not originate with the present writer. It has been repeatedly mentioned in the literature in recent years but has not been generally accepted as yet (Barry, 1978; Boyd & Caul, 1979; File, 1974; Hinderliter, 1978; Hinderliter, Webster, & Riccio, 1975; Mactutus, McCutcheon, & Riccio, 1980; Richardson, Riccio, & Jonke, 1983).

Finally, Model 3 in Table 2 portrays an intermediate case in which somete case in which some N cues are modified by the drug (N1 becomes D1) and other N cues are unaffected (N2 and N3 remain), and additionally the drug creates some new D cues (D2). The effectiveness of memory retrieval during the test session is considered to be proportional to the number of cues that were present during training and are still present during the test session; all cues are assumed to be equally strong and effects are assumed to be additive. In the N—>N and D—>D groups, no cues change and retrieval is normal. In the N—>D group, the C, N2, and N3 cues are still present during testing, but N1 is missing, yielding somewhat impaired retrieval; the addition of cues D1 and D2 has no effect. In the D—>N group, the C, N2, and N3 cues are present at the are present at the time of retrieval, but both D1 and D2 are missing, yielding a larger impairment in retrieval than in the N—>D group. The result is a partially asymmetrical SDL effect. Obviously, we can manipulate the predicted result by varying the assumed relative strength and number of C, N, and D cues and by varying the number of N cues that we assume to be modified by the drug. Depending on the assumptions that we make (the drug and dosage that we use?), the predicted SDL effect may be symmetrical, partially asymmetrical, or entirely asymmetrical.

The theoretical issues raised in the preceding paragraphs have not been empirically addressed in SDL studies. However the vast majority of published SDL results are at least partially asymmetrical and thus inconsistent withhus inconsistent with the traditional sensory model for SDL (Model 1 in Table 2).

Sensory Theories for DDs

A parallel theoretical problem has existed in the interpretation of the results of DD studies. Early DD investigators often argued that their animals discriminated drug cues versus no-drug cues (Brown, Feldman, & Moore, 1968; Browne & Ho, 1975a; Jones, Grant, & Vospalek, 1976; Schechter, 1973). However, most contemporary investigators apparently believe that animals discriminate presence versus absence of specific drug stimuli; N cues are never mentioned. This formulation is congruent with the common observation that during tests with drugs that are pharmacologically dissimilarogically dissimilar from the training drug, the animals usually perform the N response; such results are interpreted as showing that since D cues like those of the training drug are absent, the N response is performed (Colpaert, 1978; Overton, Merkle, & Hayes, 1983).

The change in the nomenclature used to describe DD results has taken place quietly during the past twenty years without any theoretical discussion in the literature and appears to have taken place so that DD investigators’ theoretical formulations would not be obviously inconsistent with the data that they obtained. However, a parallel change in theoretical conceptions has not occurred among SDL investigators, most of whom still accept a theoryill accept a theory which predicts symmetrical SDL, use statistical tests that can only properly evaluate the occurrence of SDL if it is symmetrical, and regularly obtain results that are asymmetrical. In this paper we are clearly taking the position that SDL theory is in need of revision. However, we might also note that few DD publications have been theoretically oriented and that even DD theory is not very well developed.

One interpretation of DDs likens the occurrence of N choices during tests with a novel drug after D versus N training to results obtained after sensory discrimination training using a discriminative stimulus (Sd) in one modality (e.g., visual Sd) when tests are conducted with the Sd replaced by a stimulus in a different sensory modality (e.g., auditory Sd). This., auditory Sd). This interpretation of DDs implies that each type of drug produces stimuli in a different modality. But what are these modalities, and how can there be so many of them (about 20 have been demonstrated)? By what brain mechanisms do these sensory "modalities" get processed? Why have we been unable to identify sensory events that can mimic the drug stimuli which presumably underlie DDs?

Note that the assumption that each type of drug produces cues in a different modality is counterintuitive (see previous paragraph), as it implies the existence of too many modalities. As an alternative model, we might postulate that all drugs produce stimulus effects in a rather small number of sensory modalities and that each type of drug produces a different pattern of sensory events in these modalities. This alternate hypothesis implies that we should, in time, discover more interactions between the stimulus effects of pharmacologically dissimilar drugs than have thus far been reported. Instances of masking, generalization to pharmacologically dissimilar drugs, and expected generalizations to drug mixtures should all occur, at least occasionally, and patterns of these drug interactions should eventually be discernible.

We face an unpleasant choice between the first model, which claims more distinct modalities of interoceptive experience than appear reasonable, and the second, which fails to predict the high qualitative specificity that D versus N training is usually reported to produce. Neither modelsually reported to produce. Neither model appears very satisfactory. It appears that neither DD nor SDL investigators are presently in a very good position to predict or to explain their results on the basis of available theoretical models. Since available theories match available data so loosely, they provide a weak foundation for building a scientific understanding of the subjective drug effects that cause drug abuse, and they do not allow us to precisely interpret the results of many DD experiments or to rationally develop new DD paradigms that have better properties than the ones presently in use. To remedy this problem, we can only hope that future DD research will incorporate substantial efforts directed toward an improved understanding of the effects of behavioral variables in the DD paradigm.

The most important disadvantage of the DD procedure is the fact that DDs lack face validity as a method for investigating most psychopharmacological questions. Because this issue is important, we will discuss it at length, beginning with a general discussion of the face validity of DDs as a neuropharmacological research method and then proceeding to special issues involved in the application of DDs to investigate issues related to drug abuse.

Face Validity of DDs as a General Preclinical Assay

DDs have little face validity ft>

DDs have little face validity for the preclinical investigation of many psychopharmacological issues! There is no a priori reason to believe that all antianxiety drugs will produce similar subjective effects in rats that are not fearful or that all phenothiazines will produce shared sensory consequences irrespective of the strength of their extrapyramidal side effects. It is not intuitively obvious that the clinically relevant actions of all of these drugs should be the primary cues used by rats when DDs are learned, nor is it obvious why peripheral side effects should be disregarded by the animals. DDs do, in fact, appear to be based on clinically (or scientifically) important drug actions in many instances. However, this was hardly predictable a priori and is not very well understood even when it occurs.

To clarify this point, Table 3 presents a number of fictitious examples illustrating the varied relationships that may exist between discriminable drug effects and clinically important drug effects. The table assumes that drugs can produce only ten possible actions on the central nervous system, five discriminable and five nondiscriminable; the five discriminable drug effects are assumed to be equally discriminable. The 10 drug effects are classified as clinically relevant (C), referring to important actions that are prototypic of the drug in question, and as side effects (S), referring to actions produced by (S), referring to actions produced by the drug that are not required in order to achieve its desired clinical actions. Obviously, in some cases the term scientifically important could be substituted for clinically relevant (e.g., competitive blockade of serotonin would be the scientifically important action of cyproheptadine, and this effect might or might not have significant clinical utility).

In general, the table is self-explanatory. Case 2 describes a simple situation in which drugs C and D both produce the same clinically important action (drug action #8) which is discriminable, and where these drugs have no other effects on the brain either discriminable or nondiscriminable. In this instance DDs can be used very effectively to investigate the properties and mechanisms of action of drugs C and D. Case 4 involves two drugs that produce the same clinically relevant effect (drug action #1) which is not discriminable. Additionally, drugs G and H both produce other actions which are discriminable and which differ for the two drugs. In this case DDs will be learned with both G and H but will provide no information relevant to the clinically important actions of these drugs (Rats trained to discriminate G versus N could, however, be used to assist in the development of new drugs that did not produce side effect #6.). Case 6 depicts a situation where DDs will provide some correct and some misleading results. Drugs K annd some misleading results. Drugs K and L will be correctly identified as producing effects different than those of drugs M and N, but the differences between the actions of drugs M and N will not be identified by the DD procedure. Finally, in Case 7 drugs O and R will mimic P, but substitution in the reverse direction is not likely as P will produce only one third of the discriminated effects of O or R. Incidentally, this is the only point in this review where we allude to the phenomenon of one-way substitution which is sometimes observed in DD studies (Bueno, Carlini, Finkelfarb, & Suzuki, 1972; Colpaert, Niemegeers, & Janssen, 1976; D’Mello, 1982; Overton, 1966).
 

The table suggests that the a priori likelihood is low that DDs can provide a useful research paradigm with any particular class of drugs. Hence, it is hardly surprising that widespread adoption of the DD method did not occur until the results of DDs procedures were shown to be useful in several drug classes. Evidence is now available suggesting that DDs are based on clinically or scientifically important drug effects in several pharmacological classes, and from this fact we draw two inferences. First, a relatively high proportion of drug effects on the centrirst, a relatively high proportion of drug effects on the central nervous system appear to have discriminable consequences. Second, the prototypic or desired effects of drugs in these classes appear to be more discriminable than are the undesired side effects of the same drugs.

Face Validity of DDs for Study of Drug Abuse

DDs may have face validity for investigation of the drug effects that underlie drug abuse if the following three conditions are met.

1. Sensory effects underlie and cause drug abuse.
2. Sensory events underlie DDs.
3. The sensory events underlying DDs are the same as those that underlie drug abuse.

Are SDL and DDs Based on Sensory Effects of Drugs?

In view of the current popularity of stimulus interpretations of both SDL and DDs, it is probably wise to state explicitly that there are alternativeere are alternative mechanisms that could be responsible for the phenomena; Bliss (1974) referred to these as neurological models. Although the first theoretical explanation for SDL was a sensory theory (Ribot, 1891), the second was a neurological model (Semon, 1904), and Girden (1942a, 1942b, 1942c, 1947) in his pioneering experimental reports on SDL never even mentioned sensory drug effects. The controversy between sensory and neurological models continued throughout the 1970s with most review papers on theories for SDL devoting more space to neurological theories than to sensory theories (Bliss, 1974; Overton, 1978d). Even at this point in time, although a sensory interpretation enjoys wide popularity, the evidence supporting such an interpretation is lesserpretation is less than definitive. Hence the decision to devote a substantial portion of this paper to a consideration of this question.

Most of the evidence regarding this issue falls into one of the following three categories: (1) Evidence which is compatible with a sensory interpretation but is equally compatible with some or all of the alternate neurological models. (2) Evidence which is compatible with a sensory interpretation and which is incompatible with many or all of the neurological models. (3) Evidence which is somewhat embarrassing to a sensory interpretation of DDs, although not necessarily strongly supportive of a neurological model. We will devote one section of this paper to each of these categories.f these categories.

This section will enumerate properties of DDs and SDL that are compatible with a sensory interpretation of these phenomena and are equally compatible with most neurological models for SDL and DDs.

Appearance of DD Behavior

The behavior that animals exhibit during the performance of DDs resembles in many respects that seen during the performance of sensory discriminations. In maze tasks vicarious trial and error behavior can be seen. Learning curves showing the acquisition of discriminations are indistinguishables are indistinguishable from those obtained when sensory stimuli are discriminated.

Relative Strength of Response Control by Drugs and by Stimuli

The strength of SDL and DD effects is markedly dependent on the drug and dosage employed (Mayse & DeVietti, 1971; Overton, 1982a). Hence an explicit comparison of the degree of response control produced by drug states and by sensory stimuli is difficult. A few SDL and DD studies have attempted this comparison and, depending on the parameters employed, have observed that response control by drug stimuli was either stronger or weaker than response control by sensory stimuli (Balster, 1970; Connelly & Connelly, 1978; Connelly, Connelly, & Epps, 1973; Duncan et al., 1979; Duncan et al., 1979; Jarbe, Laaksonen, & Svensson, 1983; Kilbey et al., 1971; Overton, 1964, 1968b, 1971; Spear et al., 1980). In general it appears that the strength of drug SDL effects can be considered comparable to the effects of contextual stimuli on memory retrieval.

Broad Generalization Gradients Across Dosage

After D versus N training using a particular drug and dosage, substitution tests can be conducted with various doses of the training drug. Usually the drug response generalizes to doses significantly higher and lower than the training dose with an appreciable percentage of drug responses occurring with doses down to about 30% of the training dose (i.e., rather broad generalizationroad generalization gradients are obtained along the intensity [dosage] continuum; Barry, 1978; Colpaert & Slangen, 1982; Overton, 1966). In a few instances sharp gradients have been obtained, and the particular factors that cause such steep gradients have not yet been determined. To some degree the broad gradients may be artifactual, since levels of drug in the brain vary somewhat during successive training sessions as the result of drug excretion, redistribution in body tissues, and other factors. However, even if this source of variation were controlled, it appears likely that rather broad dosage generalization gradients still would be obtained. Such generalization gradients obtained during test sessions with reduced doses of the training drug are reminiscentrug are reminiscent of those seen during analogous generalization testing along the quantitative (intensity) dimension of sensory discriminative stimuli.

Peak Shift Is Usually Not Seen

After high-dose versus low-dose DD training, peak shift has been reported during test sessions with various doses of the training drug (Akins, Gouvier, & Lyons, 1980). However, after D versus N training, peak shift is usually not seen (e.g., White & Appel, 1982). This may be an artifact caused by a "ceiling effect" in the vicinity of the training dosage. Additionally, it might be necessary to test with doses as high as twice the training dosage in order to see peak shift, and this is seldom done (but see Emmett-Oglesby see Emmett-Oglesby et al., 1983). Finally, note that in sensory discrimination paradigms, training to discriminate the presence versus absence of a discriminative stimulus is not the paradigm that most readily allows observation of peak shift. In any case, since some neurological models for DDs predict peak shift, its occurrence versus nonoccurrence is not critical to acceptance of a sensory model.

Overtraining Abolishes SDL

Several studies have shown that a well-learned response will generalize successfully across changes in drug state that would disrupt performance of a response that was not overtrained. In these studies the effect of overtraining on SDL was demonstrated by using several groups of subjectsal groups of subjects which received different amounts of training before the effects of a change in drug state were evaluated (Bliss, 1973; Eich & Birnbaum, 1982; Iwahara & Noguchi, 1972, 1974; Modrow, Salm, & Bliss, 1982). In most other 2x2 studies where SDL was reported, subjects were in fact only trained to a rather weak criterion. Indeed, in the entire literature reporting SDL effects, all positive findings involve recently mastered responses. Hence, the literature suggests that a change in drug state cannot prevent retrieval and performance of a well-trained, overly practiced response. SDL appears to be demonstrable only with responses that are not overtrained.

SDL Effects Summate with Stimulus Context Effects Effects

Duncan (1979) reported that a simultaneous change in drug state and in stimulus context could impair memory retrieval even though neither change by itself would impair retrieval. Other studies have shown that the addition of sensory retrieval cues or, in human subjects, category cues can prevent SDL effects that would otherwise occur (Connelly et al., 1973; Connelly, Connelly, & Nevitt, 1977; Connelly, Connelly, & Phifer, 1975; Eich, 1977, 1980; Eich et al., 1975; Petersen, 1977). It appears that sensory events and drug state manipulations conjointly determine the efficiency of memory retrieval.

Stimulus Blocking Occurs

Recently, Jarbe showed that initial discriminative training with ethanoltraining with ethanol would block the subsequent acquisition of discriminative control by the illumination conditions light versus dark. Conversely, discriminative light versus dark training would partially block the subsequent acquisition of control by the conditions pentobarbital versus no drug (Jarbe, Svensson, & Laaksomen, 1983). Although this type of interaction between exteroceptive stimulus control and DDs certainly suggests a sensory interpretation of SDL and DDs, it is compatible with some neurological models for SDL.

Learning Set Phenomena Occur in DD and SDL Experiments

In a T-maze DD task, Overton (1971) demonstrated that light versus dark discriminations could be more rapidly learned if rats had previously masteredpreviously mastered an ethanol versus N discrimination in the same task. This suggested that a learning set established by DD training could facilitate subsequent acquisition of sensory discriminations. In a different SDL task Bliss (1974) required monkeys to learn a series of visual discriminations while drugged. Each individual discrimination failed to generalize to the no-drug state, but after several had been learned the resulting learning set did generalize to the no-drug condition. This result appears somewhat counterintuitive since it is usually argued that simple types of learning should generalize more easily across changes in drug state than complex types of learning. However Bliss’s data suggest the opposite—the learning set generalizeding set generalized (perhaps because it was overtrained?) even though the individual discriminations failed to generalize. Modrow, Salm, and Bliss (1982) obtained similar results. These learning set phenomena appear to be compatible with both neurological and sensory models for SDL and DDs.

All of the preceding similarities between drug and stimulus control of behavior are compatible with a sensory interpretation of SDL and DDs. However, most of the neurological models for SDL and DDs (reviewed by Bliss, 1974, and by Overton, 1973, 1978d) also predict the same phenomena.

This section describes several types of evidence that do support the acceptance of a sensory interpretation of SDL and DDs because the data are incompatible with most or all of the proposed neurological models for these phenomena.

No-Drug Responses Are the Default Choice During Substitution Tests

We previously described the fact that during substitution tests conducted after D versus N training, D choices are only observed during tests conducted with drugs that pharmacologically resemble the training drug. In other words, the majority of centrally acting drugs do not produce D choices. We suggested that this result is analogous to what might be expected aftermight be expected after training using a discriminative stimulus in one modality (e.g., auditory) if tests were then conducted with the discriminative stimulus replaced by a stimulus from a different sensory modality (e.g., visual). In the present context we should add that most neurological models for SDL and DDs apparently do not predict this outcome but instead predict mixed, random, or confused choice behavior during tests with drugs that pharmacologically differ from the training drug. Hence, the high qualitative specificity produced by D versus N training appears to support a sensory interpretation of DDs. Parenthetically, a few neurological models may predict N responses as the default choice during such substitution tests.substitution tests.

Threshold Dosages for SDL Are Higher Than for DDs

It appears that a determination of the lowest dose that is capable of producing SDL and DDs, respectively, can test whether the mechanism underlying these effects is sensory in nature or involves one of the postulated neurological mechanisms. The argument is as follows.

The literature on contextual control suggests that moderately large sensory changes are required to produce retrieval failures caused by a change in stimulus context. In contrast, after discrimination training it is reasonable to expect that an animal will be able to discriminate much smaller differences between stimuli because prolonged training willonged training will allow the animal to gradually learn to attend to the relevant attributes of the discriminated stimuli. Although the literature actually provides little evidence regarding the relative magnitude of the sensory changes that are required to produce (1) retrieval deficits based on changes in sensory context and (2) an adequate basis for discriminative control, some evidence suggests that the threshold for discriminative control is considerably lower than for contextual retrieval effects—perhaps an order of magnitude lower (Riccio, Urda, & Thomas, 1966). Analogously, a sensory interpretation of SDL and DDs appears to predict that the threshold for discriminative control will be considerably lower than the threshold required to producerequired to produce SDL decrements.

In contrast, all neurological models for SDL predict that the thresholds for DDs and for SDL will be equal. According to these theories, DDs are based on weak (but measurable) SDL effects. During DD training the animal is assumed to learn two responses that have approximately equal habit strengths. During each DD trial both habits are believed to be retrieved from memory. However, the habit that was learned in the currently imposed drug state is retrieved somewhat more efficiently and dominates the overt behavior of the animal (Spear et al., 1980). Clearly, in such a situation where two responses are competing for expression, if the state-appropriate response is consistently performed, then its engram must be appreciablym must be appreciably (measurably) more retrievable than that of the other response. Hence, the lowest drug dosage that is capable of controlling discriminative responding should be no lower than the lowest dosage that will produce a measurable SDL effect after D—>N or N—>D state changes.

The differing predictions of the sensory and neurological theories have, to some degree, been experimentally tested. Suppose that one starts with a task, drug, and dosage so selected that both SDL and DDs can be obtained. If SDL and DD experiments are repeated using lower and lower dosages, the SDL effects will disappear first. Then, at lower doses discriminative control will also be lost. Although only a few explicit comparisons of the thresholdns of the threshold dose adequate to produce SDL and DDs have been reported (Overton, 1979a, 1982b; Zenick & Goldsmith, 1981), the total literature on these phenomena clearly support the generalization just stated. SDL is only produced by some drugs at the highest doses compatible with sustained behavioral responding (Eich, 1980). In contrast, DDs can be maintained by intermediate doses of the same drugs and by a variety of other drugs which are unable to produce measurable SDL effects at any usable dose. This difference in the threshold dosage for SDL and for DDs is discordant with the predictions of all neurological models for SDL and DDs and supports a sensory interpretation of both phenomena.

Several types of evidence deter us from an entirely uncritical acceptance of a sensory interpretation for SDL and DDs.

Lack of Direct Supporting Evidence

A substantial embarrassment for sensory theories of SDL and DDs is provided by the fact that no one has been able to identify a sensory stimulus that would mimic or substitute for a drug stimulus or vice versa. If drugs truly achieve contextual and discriminative control via the mechanism of drug-induced stimuli, then it might be possible to duplicate the stimulus conditions produced by drugs via appropriate manipulations of the internal or external milieu, and such a finding would be important for two reasons. First, such results would help identify the specific sensory actions responsible for SDL and DDs produced by the specific drug in question, thus providing a more rational basis than is presently available for evaluating the results of DD experiments conducted with that drug. More importantly, successful identification of the sensory mediators of SDL and DDs, even for a single drug, would more firmly establish the general principle that such effects could be mediated by sensory drug effects—a conclusion presently supported only by indirect evidence.

There are at least three different types of sensory effects which drugs apparently can produce, any or all of which might cause SDL and DDs. (1) Some drugs directly induce changes in peripheral organs, and these changes produce altered sensory input returning to the brain via the classical afferent pathways. For example, antimuscarinics reduce the flow of saliva and produce sensations of "dry mouth." Sedative drugs such as ethanol produce ataxia and the associated altered proprioceptive feedback. (2) Other drugs can modify the processing and perception of interoceptive or exteroceptive stimuli. Blurred or double vision is one example of such a drug-induced modification in sensory processing. Analgesia produced by narcotics provides a second example. (3) Finally, some drugs may directly induce central "sensory" effects by altering the organism’s emotions, drive states, arousal level, et cetera.

The first type of mediating mechanism for DDs—drug-produced changes in peripheral organs that cause altered sensory feedback—has been investigated on several occasions with negative results. For example, antimuscarinic drugs produce SDL and DDs and produce altered functioning in a variety of organs innervated by the autonomic nervous system. Tertiary antimuscarinic drugs act at both central and peripheral sites whereas quaternary antimuscarinics cross the blood-brain barrier less easily and thus act only at peripheral sites. After D versus N DD training with scopolamine hydrobromide (which produces both central and peripheral actions), animals fail to generalize the D response to quaternary scopolamine compounds which produce only peripheral actions. Thus it appears that the peripheral actions of scopolamine are not responsible for its discriminable actions (Overton, 1977a). Similarly, after D versus N DD training with pentobarbital, the D response fails to appear during tests with gallamine, a curare-type drug which produces muscular weakness and a lack of coordination vaguely reminiscent of the ataxic actions of pentobarbital (Overton, 1964). Other tests of this type have also been reported, but in no case has a peripheral manipulation been identified that could mimic the discriminable effects of a drug (Downey, 1975; Hazell et al., 1978). Additionally, it has generally been found that drugs which act only on peripheral organs do not produce SDL and are discriminated only with difficulty whereas centrally acting drugs are more likely to produce SDL and are more readily discriminated (Miksic, Shearman, & Lal, 1980; Overton, 1971). This has discouraged further attempts to identify specific peripheral sensory stimuli which might mediate SDL and DDs.

The second possibility—that drug-induced alterations in central sensory processing might mediate SDL and DDs—has also been investigated with negative results. For example, Overton (1968b) hypothesized that blurred vision might mediate the discriminable effects of pentobarbital. To test this possibility, he first blinded rats and then required them to learn a D versus N discrimination in a T-maze. This discrimination was learned as rapidly by blind as by sighted rats, indicating that drug-induced alterations in visual stimuli were not a prerequisite for the establishment of the discrimination. In another experiment sighted rats were required to discriminate pentobarbital versus N; these rats were then blinded, and training was continued. Only a transient disruption in discriminative control was noted at the time of blinding, suggesting that even in sighted rats alterations in visual perception do not mediate discriminative control. In a similar vein Overton hypothesized that pentobarbital versus N discriminations in a shock-escape T-maze task might be mediated by drug-induced analgesia or at least by a drug-induced insensitivity to some of the consequences of electric shock. However, two pieces of evidence contradicted this hypothesis. First, after D versus N training with high shock levels, undrugged rats could not be induced to make D choices by the application of low (less painful) shock intensities (Overton, 1968b). Secondly, after D versus N training with pentobarbital, the D response does not occur during tests with morphine or other narcotic analgesics.

The third type of possible sensory mediator for SDL and DDs—drug induced central sensory effects—implies, as a corollary, that altered states of the central nervous system induced by other manipulations besides drug injections might also produce contextual and discriminative effects analogous to those of drugs. Two questions follow. (1) Which altered CNS states, if any, produce contextual and/or discriminative effects? (2) Do any of the altered CNS states that can be induced by nonpharmacological manipulations produce sensory effects equivalent to those induced by drugs?

Regarding the first question, electroconvulsive shock (ECS) produces SDL, and ECS versus no ECS discriminations are robust (McIntyre & Reichert, 1971; Overton, Ercole, & Dutta, 1976). Alterations in hunger or thirst are discriminable and produce SDL (Bolles, 1958; Nahinsky, 1960; Peck & Ader, 1974). Electrical brain stimulation is discriminable (Colpaert, 1977b; Hirschhorn, Hayes, & Rosecrans, 1975; Stutz & Maroli, 1978). Temporary ablation of certain brain structures can produce SDL (Duncan & Copeland, 1975; Greenwood & Singer, 1974; Langford, Freedman, & Whitman, 1971; Pianka, 1976; Reed & Trowill, 1969; Schneider, 1966, 1967, 1973). REM-sleep deprivation can produce SDL (Joy & Prinz, 1969), and learning that occurs during REM sleep is state dependent (Evans, 1972; Evans, Gustafson, O’Connell, Orne, & Shor, 1966, 1969, 1970). These studies show that a variety of alterations in CNS activity can be discriminated and/o that occurs during REM sleep is state dependent (Evans, 1972; Evans, Gustafson, O’Connell, Orne, & Shor, 1966, 1969, 1970). These studies show that a variety of alterations in CNS activity can be discriminated and/or produce SDL.

With regard to the second issue, there are very few studies that we can cite. Huang (1973) trained rats to press one lever after normal sleep and the second lever after REM-deprivation. This discrimination was learned, and subsequent tests showed that amphetamine would cause some REM-deprivation responses in rats that had slept normally and conversely that pentobarbital would cause some responses on the normal-sleep lever in REM-deprived rats. Although the drugs did not completely antagonize (or mimic) the sleep manipulations, the results suggested that the effects of REM deprivation and of normal sleep were to some degree overlapped by the effects of pentobarbital and/or amphetamine. In a related study Schechter (1981b) trained rats to discriminate pentobarbital versus amphetamine and then tested them with saline at various times of day. During midafternoon tests saline caused responding predominantly on the pentobarbital lever, but during saline tests at 2:00 a.m., the rats showed only 50% responses on that lever. Both studies suggest that the drug injections and the time of day (or REM deprivation) manipulations may have moved the animals along some shared "sensory" dimension (arousal level?) which provided at least part of the basis for discriminative control. Finally, Gardner, Glick, and Jarvik (1972) reported some similarities between the SDL effects of ECS, physostigmine, and scopolamine. However, Overton, Ercole, and Dutta (1976) were unable to obtain evidence for analogous effects using a DD paradigm.

In summary, attempts to identify sensory or physiological manipulations that would mimic the postulated sensory effects of drugs have failed, almost without exception, to find such manipulations. This is not entirely surprising because due to the high specificity produced by D versus N training, it might be necessary to rather exactly mimic the sensory effects of a drug in order to observe substitution. Nonetheless, one single explicit demonstration of a DD mediated by identifiable sensory events would make it much easier to accept a sensory interpretation of all DDs.

Masking Has Not Been Reported

We mentioned earlier the less than conclusive evidence suggesting that masking does not occur in the DD preparation. This property is helpful in many DD experiments. However, if DDs are based on sensory events, it appears that masking really should occur. Hence the failure to observe masking can be counted as evidence against a sensory interpretation of DDs.

One should note that attempts to detect masking have not used as yet the drug stimulus conditions that apparently would be most likely to produce masking. For example, it appears that masking will be most likely if a rapidly discriminated drug (with strong sensory effects) is used to mask a training drug with weak, slowly discriminated effects. Additionally, if the sensory literature is a useful guide, it may be the case that drugs producing similar sensory effects (effects in the same modality?) can mask one another more effectively than can drugs that produce markedly dissimilar effects. Even though we know almost nothing about the topography or dimensionality of the sensory modalities in which drugs produce stimuli, it might be feasible to test this possibility by selecting drugs that were sufficiently similar to substitute for one another in a low specificity DD paradigm and by then testing whether these drugs would mask one another in a high specificity paradigm. Tests using such combinations of drugs have not been reported. If future studies detect the occurrence of masking and define the conditions for its occurrence, this will remove an impediment to acceptance of a sensory interpretation of DDs.

Prior Exposure to Drug Does Not Slow the Acquisition of DDs

In several conditioning paradigms prior exposure to a stimulus has been observed to impede the subsequent acquisition of conditioned responses or discriminative control, presumably by habituating the animal to the to-be-conditioned stimulus. However, this effect has not been observed in connection with drug stimuli. In several such studies investigators have repeatedly exposed animals to drugs before commencing D versus N training. Such prior exposure to drug has neither facilitated nor impaired the subsequent acquisition of DDs (Hinderliter, 1978; Jarbe & Henriksson, 1973; Jarbe & Holmgren, 1977; Kilbey et al., 1971; McKim, 1976; Overton, 1972).

Poor Correlation with Human Subjective Reports

One final problem for sensory DD theories is posed by some inconsistencies between data obtained from animal and from human subjects. Recreational, clinical, and experimental drug use in humans have yielded anecdotal and experimental reports on the strength and nature of the subjective effects of drugs including hallucinations, changes in affect, and assorted unpleasant side effects. Somewhat analogous data on the discriminable effects of drugs have been obtained from DD studies in animal subjects (e.g., Overton 1982a). In many instances the two types of data correspond. For example, all morphine-like drugs produce similar subjective effects in humans and similar discriminable effects in rats. However, there are several instances in which data from human and animal subjects are disparate. For example, in rats ethanol is very readily discriminated and chlorpromazine is discriminated only with difficulty whereas in humans both drugs produce more or less equally reportable subjective effects. As another example nicotine and marihuana are both readily discriminated in rats, with nicotine perhaps being more rapidly discriminated. In humans the subjective effects of marihuana appear to be more noticeable than those of nicotine. Instances such as these suggest that subjective drug effects in humans and discriminable drug effects in animals are not entirely isomorphic.

Conclusions Regarding Sensory Mediation of SDL and DDs

Several pages ago we listed the following three prerequisites that had to be met if DDs were to be capable of providing information regarding the causes of drug abuse.

1. Sensory effects must underlie and cause drug abuse.
2. Sensory events must underlie DDs.
3. The same sensory events must underlie both DDs and drug abuse.

This section enumerates the most obvious ways in which DDs can be used to improve ohe causes of drug abuse.

This section enumerates the most obvious ways in which DDs can be used to improve our understanding of drug abuse.

Definition of Categories of Drugs

DD investigators have made considerable progress toward developing a new categorization system for psychoactive drugs based entirely on their discriminable properties. In many instances the categories match those already established by other techniques, and one can view these studies as a simple replication of previous work in psychopharmacology (e.g., Barry, 1974; Browne, 1981; Cameron & Appel, 1973; Colpaert, 1977a, 1978; Colpaert & Rosecrans, 1978; Goas & Boston, 1978; Goudie, 1977, 1982; Lal, 1977; Overton, 1977a, 1978a). In other instances DD studies have allowed a subdivision of pre-existing drug classes, thereby defining differences between the actions of drugs that were previously considered to be similar. These results are likely to have relevance to drug abuse since drugs with similar clinical actions but differing subjective effects are likely to have differing liabilities for abuse. Finally, some genuinely new "classes" of compounds have been defined by DD procedures. For example, the class of phencyclidine-like compounds now includes not only phencyclidine (PCP) analogs but also certain narcotic mixed antagonists (dextrorphan, cyclazocine, SK10047) which appear to share some of PCP’s discriminable effects (Hein, Young, Herling, & Woods, 1981; Herling et al., 1981). This work provides a good example of the application of DD procedures to obtain information about drugs that is likely to have considerable relevance to their liability for abuse.

Identification of Neuropharmacological Mechanisms of Drug Action

DDs have been used in numerous studies designed to identify the neurotransmitters which mediate the effects of drugs. A typical procedure is to establish a D versus N DD based on the drug in question and then to manipulate various transmitter systems using available agonists, antagonists, depleters, and blockers. The utility of the DD method for such studies derives, as usual, from its relatively high specificity, and a number of rather ingenious and productive studies of this type have been reported (Barrett, Blackshear, & Sanders-Bush, 1982; Barrett & Steranka, 1983; Bennett & Lal, 1982; Browne & Ho, 1975a; Chipkin, Stewart, & Channabasavaiah, 1980; Ho & Huang, 1975; McKenna & Ho, 1980; Schechter, 1977, 1980; Schechter & Cook, 1975; Schechter & Rosecrans, 1972b). Among the most sophisticated of these studies have been those investigating the role of various neurotransmitters in mediating the actions of hallucinogenic drugs (Appel, White, & Holohean, 1982; Glennon, Rosecrans, Young, & Gaines, 1979; White & Appel, 1981, 1982; White, Appel, & Kuhn, 1979; Winter, 1978b).

Some investigators have used DDs to identify the neuroanatomical location where drugs act to produce their subjective effects. However, such studies have been difficult in rats because a long time is required to establish DDs and a relatively small number of tests with intracerebral drugs can be conducted per rat (e.g., Krynock & Rosecrans, 1979; Meltzer & Rosecrans, 1981; Rosecrans & Glennon, 1979).

Identification of Antagonists

Drugs which block or antagonize the actions of the training drug can be easily identified with the DD procedure. Although early reports dealt mainly with well-known antagonists and blockers (e.g., Hirschhorn & Rosecrans, 1974b; Overton, 1966, 1969; Romano et al., 1981; Schechter & Rosecrans, 1972c), more recent studies have identified novel antagonists (Bennett, Geyer, Dutta, Brugger, Fielding, & Lal, 1982; Browne, 1981; Colpaert, 1977a; Dantzer & Perio, 1982; Herling & Shannon, 1982; Shearman & Lal, 1979; Winter, 1981b). Excellent examples are the reports provided by Colpaert, Niemegeers, and Janssen (1982) who identified pirenperone as an LSD antagonist and by Browne who identified compounds which reduced the effects of phencyclidine (Browne & Welch, 1982; Browne, Welch, Kozlowski, & Duthu, 1983). In general, the DD method appears well suited to identify new antagonists and competitive blockers.

Identification of Active Metabolites

The DD method can apparently identify the metabolites of a drug which produce actions similar to those of the parent compound (e.g., Barry, Steenberg, Manian, & Buckley, 1974; Beford, Nail, Borne, & Wilson, 1981; Brady & Balster, 1981; Browne, Harris, & Ho, 1974; Browne & Ho, 1975b; Holtzman, 1979).

Establish Structure-Activity Relationships

DDs can obviously be used to test a series of related chemicals to determine which produce discriminable effects like those of a parent compound (Chance, Kallman, Rosecrans, & Spencer, 1978; Huang & Ho, 1974; Katz, Woods, Winger, & Jacobson, 1982; Meltzer, Rosecrans, Aceto, & Harris, 1980; Shannon, McQuinn, Vaupel, & Cone, 1983; Solomon, Herling, Domino, & Woods, 1982). Industrial DD laboratories have tended not to use DDs as a primary assay, due to the expense of the DD procedure, and have instead used DDs to ask specific questions about compounds already known to have CNS activity. However, Glennon and his associates have used the DD preparation as a primary screen to identify compounds producing hallucinogenic subjective effects (Glennon & Rosecrans, 1982; Glennon, Young, & Jacyno, 1983; Glennon, Young, Jacyno, Slusher, & Rosecrans, 1983; see also Glennon & Young, this volume). Like many other assays, the DD procedure used in this way can only identify compounds that produce effects similar to those of pre-existing drugs. With presently available procedures, it is not feasible to use each new compound as a training drug in order to determine whether it produces any discriminable actions (which might differ from those of pre-existing drugs).

Screen for New Compounds Lacking Abuse Liability

One very promising application of DDs which has been proposed (by J. Woods, personal communication) but not yet employed is screening new compounds to identify those which lack effects already shown to produce abuse. For example, Woods proposes that new analgesics could first be identified in a standard test for analgesic efficacy (e.g., tail flick). Compounds shown to have analgesic efficacy could then be tested in rats that had been trained to discriminate morphine versus no drug. Compounds that substituted for morphine would be rejected on the assumption that they shared with morphine its euphorigenic actions. Compounds that produced analgesia but failed to mimic morphine would remain as candidates for further testing. Note that compounds that survived the dual test procedure might still have liability for abuse due to hallucinogenic or other abuse-producing subjective effects. However they would probably not produce the morphine-like subjective effects which are responsible for a great proportion of the abuse of currently available analgesic drugs.

Characterization of the Sensory Effects That Cause Drug Abuse

It seems likely that a variety of different sensory effects of drugs can cause drug abuse. Investigators using the self-administration method have tended to lump these effects together, referring to all as "reinforcing" stimulus effects, and self-administration procedures have been relatively ineffective in differentiating the various types of drug stimulus effects that could underlie drug abuse. In contrast, DD procedures apparently have the capability to differentiate between the various sensory effects that can underlie drug abuse and have already shown that abused drugs produce at least half a dozen clearly distinct discriminable effects. This research effort ility to differentiate between the various sensory effects that can underlie drug abuse and have already shown that abused drugs produce at least half a dozen clearly distinct discriminable effects. This research effort is continuing. When attempting to analyze the sensory effects that underlie drug abuse, we appear to face a difficult choice. Self-administration procedures have greater face validity but currently lack the ability to make fine discriminations between different types of sensory effects, some or all of which may underlie self-administration. DD procedures yield data not necessarily based on drug effects that underlie drug abuse, but these procedures have a much greater ability to differentiate between dissimilar drug stimuli. At the very least it appears that DD procedures can be a valuable adjunct to self-administration procedures when attempting to analyze the sensory causes of drug abuse. Possibly DD procedures will become the more important analytic technique with self-administration experiments used as a sort of post hoc test to determine which DD results relate to sensory effects that will promote self-administration and which do not. It seems likely that via a not too well defined series of intermeshed DD and self-administration experiments we could identify and differentiate between the various sensory abuse-inducing effects of drugs. This is not an established use of DDs. Rather, it is an optimistic prediction that a judicious mixture of DD and self-administration experiments might provide much more precise information about the sensory causes of drug abuse than could be produced by either method alone.

It should be obvious that several of the research goals outlined above (e.g., categorization of compounds, identification of neurochemical substrates) can also be accomplished by techniques other than the DD procedure. However, in the context of drug abuse, the DD procedure has the advantage of providing experimental results that are based on the sensory consequences of drug action. Hence, insofar as DD results differ from those obtained by more molecular methods, the DD results are more likely to have relevance to abuse liability.

Drug discrimination procedures have less face validity than self-administration procedures for investigating the sensory drug effects that presumably underlie drug abuse. However, currently available DD procedures can provide more detailed information comparing and contrasting the subjective effects of drugs than can be produced by self-administration procedures thus far developed. Hence, DDs appear to provide a valuable method for improving our unde