Dose-Response Analysis in RiskAssessment: Evaluation ofBehavioral SpecificityJohn R. GlowaLaboratory of Medicinal Chemistry, National Institute of Diabetes andDigestive and Kidney Diseases, National Institutes of Health, Bethesda,Maryland

Several methods of quantitative risk assessment that have been described recently areparticularly applicable to neurotoxic end points. These methods can be broadly divided into twotypes of approaches based on their treatment of dose-response data to estimate risks.Benchmark approaches estimate risks using variability in response to a fixed dose level incomparison with background control variability. Probabilistic approaches estimate risks using thevariability in the dose to produce a small effect in the sample population. The current report seeksto extend the development of probabilistic approaches for neurotoxic end points. Becausebehavioral data are often used to assess therapeutic efficacy as well as toxicity (unwanted effects),this analysis focused on the relative risks of producing these effects with the same agent. Thetherapeutic potential of GBR 12909 was determined by its ability to decrease cocaine-maintainedresponding in monkeys. The effects of this agent were also assessed in the same monkeys usingfood-maintained responding to provide an indication of behavioral toxicity. GBR 12909 decreasedboth behaviors, with complete decreases on drug-seeking behavior occurring at doses that hadminimal effects on food-maintained responding. The difference in the estimates of doses todecrease drug-seeking and food-maintained behavior suggested that specific therapeutic effectscould be obtained in the absence of unwanted side effects for a definable proportion of thepopulation. These results also suggest that multiple behavioral end points can be useful foridentifying specific effects of chemicals for the purposes of risk assessment. Environ HealthPerspect 104(Suppl 21:391-396 (1996)

Key words: risk assessment, drug abuse, rhesus monkeys, GBR 12909, schedule-controlledbehavior, neurotoxicology

IntroductionRisk assessment is the attempt to estimate agent that produce adverse effects inthe chance of obtaining an adverse effect of humans, risk assessments are typically basedexposure to an agent. Although it is possible on effects obtained over a range of non-to predict virtually safe levels if sufficient toxic-to-toxic doses determined in animals.information is known about doses of an Such methods were developed initially to

This paper was prepared as background for the Workshop on Risk Assessment Methodology forNeurobehavioral Toxicity convened by the Scientific Group on Methodologies for the Safety Evaluation ofChemicals (SGOMSEC) held 12-17 June 1994 in Rochester, New York. Manuscript received 1 February 1995;manuscript accepted 17 December 1995.

This work was supported, in part, by a NIDA Interagency Agreement #RA-ND-94-24 (JR Glowa, principalinvestigator). Portions of the data were presented at the College for Problems of Drug Dependence, Toronto,Canada, 1993. All animal procedures conformed to the Guide for Care and Use of Laboratory Animals endorsedby the National Institutes of Health. The author gratefully acknowledges the counsel and support of PB Dewsand RC MacPhail. Their continued discussions of risk assessment methodology, as well as comments on ear-lier versions of this manuscript, are greatly appreciated.

Address correspondence to Dr. John R. Glowa, Chief, Behavioral Pharmacology Unit, LMC/NIDDK/NIH,Building 14D, Room 311, Bethesda, MD 20892. Telephone: (301) 496-5059. Fax: (301) 402-0378. E-mail:G7W@cu.nih.gov

Abbreviations used: ADI, acceptable daily intake; RfD, reference dose; NOAEL, no observable adverse effectlevel; LOAEL, lowest observable adverse effect level; DA, dopamine; TLVs, threshold limit values; FRF, fixed-ratio 3G-response food presentation; FRC, fixed-ratio 30-response cocaine injection; ANOVA, analysis of vari-ance; LD50, median lethal dose; ED5, median effective dose.

predict the chance of producing carcino-genic and mutagenic effects with exposureto ionizing radiation. Higher doses wereoften associated with an increased inci-dence of cancers, so the focus of risk assess-ment shifted toward the attempt to predicteffects produced by very low doses. Twoproblems developed. First, the spontaneousrate of cancer production called into doubtthe ability to measure an increase in risksassociated with very low doses of an agent.Second, several low-dose extrapolationapproaches were developed based on theassumption that any increment in dosewould increase the probability of an effect.These issues have questioned the relevanceof the low-dose extrapolation approach forother types of adverse effect. For example,few neurotoxic events occur spontaneously,although disease, age, or other processesmay exacerbate their effects. Likewise,mechanisms such as cellular repair, plastic-ity, system redundancy, or tolerance maycontribute to the lack of measurable effectfollowing exposure to low doses of neuro-toxic agents. Alternatively, some agents havebeen simply found to lack effects at lowdoses. As a result, risk assessment methodol-ogy for noncancer end points shifted awayfrom the estimation of effects of very lowdoses toward the attempt to estimate levelsof an agent that might be considered safe.

The earliest practice used to predict safelevels was the acceptable daily intakeapproach (ADI), which attempted topredict a dose (within an order of magni-tude of uncertainty) that could be toleratedover the lifetime without producing harm.However, while some argue that it was vir-tually impossible to determine an entirelysafe dose, others pointed to an apparentinappropriateness of using the word "accept-able" in the context of a poison. This led toa revised terminology called the referencedose (RfD) approach. This approach firstfinds a dose level with no effect (i.e., noobservable adverse effect level, or NOAEL)or a minimal effect (i.e., lowest observableadverse effect level, or LOAEL) in animalsand then divides this dose by a series ofuncertainty factors (invariably 10-fold each)to calculate a RfD for humans. The result-ing RfD could be several orders of magni-tude smaller than the original NOAEL orLOAEL (1). There have been a number ofcriticisms of the RfD approach (2-6),including a) its dependence on the actualdose spacing used, b) its failure to incorpo-rate the slope of the dose-response function,

Environmental Health Perspectives * Vol 104, Supplement 2 - April 1996 391

J.R. GLOWA

c) a lack of biological justification foruncertainty factors, d) its questionableapplicability for some types of data, e) thefact that the RfD is not a risk figure butrather a point estimate of an exposure atwhich risk is considered negligible, andf) its perverse ability to reward biased riskassessors with less conservative risk figuresfor using small sample sizes. In response, anew trend in risk assessment that uses morequantitative approaches has emerged.These approaches are of interest becausethey circumvent many of the problems asso-ciated with the RfD and low-dose extrapola-tion approaches and thus provide truequantitative risk estimates. Given the recentincrease in concerns over other end pointssuch as neurotoxicity (7,8), and the equivo-cal performance of the older approaches toestimate risks (9,10), the time is ripe toscrutinize these quantitative methods.

Quantitative approaches generallydescribe dose-effect functions in mathe-matical terms and specifically incorporatethe variability in different parameters of thisfunction to assess risks (6,11). One subfam-ily of these approaches can be characterizedby its focus on variability in an effect at afixed-exposure level (effect-tolerance orbenchmark approaches). One advantage ofthis approach is that statistical techniquesto fit functions and characterize variabilityare readily available. Another subfamily ofapproaches is characterized by a focus onvariability in exposure levels that can beshown to produce a fixed effect (dose-toler-ance or probabilistic approaches). Theadvantage of the dose-tolerance approachesis that they characterize individual differ-ences, thus providing direct information asto the potential adequacy of the use of anuncertainty factor for individual variability(6). The current report extends the use ofthe probabilistic (dose-tolerance) approachto multiple behavioral end points in thesame organisms to address the specificity ofthe effect of the agent.

The agent used for these analyses is adrug that may have therapeutic merit.Though drugs are not commonly thoughtof as neurotoxicants, any pharmacologist ortoxicologist will acknowledge that highdoses of therapeutic agents will have adverseeffects. It has been a common practice overthe years to compare doses expected to pro-duce a therapeutic effect with thoseexpected to produce a toxic effect. The cur-rent report departs from this approach byattempting to compare the risk of producingthese effects. Another small departure fromstandard methods is the use of a different

end point in the current report. Behavioraltoxicology has been primarily concernedwith the direct effects of toxic agents onlearned or acquired behaviors. Most often,these behaviors are experimentally devel-oped in the laboratory, typically by train-ing animals to respond under schedules offood reinforcement. Stable rates and pat-terns of behavior provide a baseline fromwhich a dose-effect function can be deter-mined. While almost all of these studieshave employed behaviors that lead to thedelivery of food as a reinforcer, some othershave shown that the event used to main-tain behavior can determine differenteffects for some agents (12). It has beenknown for some time that drugs can beused to maintain responding (e.g., self-administration studies). Interestingly, sometoxicants have been shown to serve as rein-forcers. For example, toluene has beenshown to support responding in monkeys(13). Little is known of solvent abuse;however, growing knowledge suggests that,if left unchecked, drug abuse results inchanges in behavior that are sufficient tomeet the criteria of a psychiatric disorder(14). Neurotoxic changes are also seenwith prolonged exposure to drugs of abuse(15). Thus, drug abuse may be the mostprevalent form of behavioral toxicologyobservable in humans today. One questionis whether a therapeutic agent could bedeveloped that would decrease drug abusebut not alter other normal behavior.

Previous studies have shown that oper-ant responding can be maintained bycocaine delivery (self-administration), whichprovides an animal model of drug abuse(16). Left unrestrained, cocaine self-admin-istration can occur to the point of produc-ing serious harm in animals (17). Suchexcessive behavior may result in neurotoxiceffects in central mesolimbic dopamine(DA) pathways associated with reinforce-ment (18). In contrast to some forms ofdrug abuse, there is no currently acceptedstrategy to treat cocaine abuse. However,clues from successful treatment approachessuggest that targeting mechanisms associ-ated with the abuse potential of a drug maylead to a treatment drug. Recently, aspecific neuronal site (the DA reuptake ortransporter) has been identified as a recep-tor that is strongly associated with the rein-forcing effects of cocaine (19). Studieshave shown that drugs that bind to this siteneed not exhibit reinforcing effects inhumans (20) or in animals (21). Theseobservations have led to the suggestion thatlong-acting DA reuptake inhibitors may

attenuate cocaine-seeking behavior in amanner similar to the ability of methadoneto decrease opiate-seeking behavior ornicotine patches to decrease smoking (22).However, there is no clear consensus as tothe type of effect a medication shouldexhibit on cocaine self-administration inanimals that would predict its efficacy todecrease cocaine-seeking behavior inhumans. The current research is meant toaddress this issue from a therapeutic pointof view; however, it is obvious that someconcern should be directed at the potentialtoxicity of such an agent. Thus, the currentstudies are designed to compare the relativetherapeutic actions of an agent being devel-oped to treat cocaine abuse with otherunwanted effects. The current studies focuson one compound, GBR 12909, which haspreviously been shown to exhibit a numberof effects similar to those of cocaine andhas a high affinity for the DA reuptake site.In contrast, because GBR 12909 is longacting and has a chemical structure unlikecocaine, it may block the effects of cocaine.Interestingly, this drug has been reportedto exhibit a nonstimulant profile of actionin normal human volunteers following oraladministration (23). Thus, this agent couldserve as a prototype from which even moretherapeutic and less toxic analogs could bedeveloped. The primary focus of the cur-rent work was to separate those effects.For these purposes, the effects of GBR12909 on food-maintained behavior weretaken to represent a behaviorally toxiceffect, and its effects on cocaine self-administration were used as an indicationof its therapeutic potential.

To accomplish the risk assessmentanalysis of these data, methods previouslydeveloped to assess the risks of exposure toagents using a single behavioral end pointwere used (24). This approach was origi-nally described by Dews (3,4). Data fromindividual subjects are first described bymeans of a mathematical function, and thedose resulting in a small but measurableeffect is estimated from each function.Through replications in different subjects,a distribution of individual point estimatesis obtained. The variability in these esti-mates is used to predict population toler-ances, from which the proportion of thepopulation that would be expected to beaffected at lower levels can be determined.This method has been used extensively toassess the risks associated with exposure to avariety of organic solvents (25-27). Theseestimates of a small decrement in neurobe-havioral functioning in relatively small

Environmental Health Perspectives * Vol 104. Supplement 2 * April 1996392

NEUROBEHAVIORAL RISK ASSESSMENTS FOR THERAPEUTICS

sample sizes (n= 10-12) agreed well withestablished threshold limit values (TLVs). Asubsequent report (28) compared estimatesfrom small (n= 10) and large (n= 40) samplesizes to illustrate the effectiveness of these rel-atively small experiments, as well as to com-pare estimates obtained with this approachto those of others. This approach has alsobeen compared with other approaches interms of its use of the dose-effect data(6,10), applicability to cross-species extrapo-lation (29), and ability to characterize theneurotoxic effects of the pesticide carbaryl(6). The potential advantage of using twoend points in the present studies is that thenonspecific (toxic) effects of the agent canbe directly compared to its intendedtherapeutic effects, thereby producing athzerapeutic ratio.

MethodsThis report uses methods that were previ-ously published (30,31). Briefly, rhesusmonkeys were surgically implanted with asubcutaneous port/catheter system (32)and trained to respond under a multiplefixed-ratio 30-response food presentation(FRF), fixed-ratio 30-response cocaineinjection (FRC) schedule to establish abaseline where approximately equal rates ofresponding were maintained by food andcocaine. The acute effects of GBR 12909were then determined on these perfor-mances using eight monkeys. Individualcontrol rates of responding for both perfor-mances were obtained for several noninjec-tion control sessions and served as a baselinefor each monkey. The effects of pretreat-ment with GBR 12909 were assessed onthis baseline by comparing the mean over-all rates of responding during drug ses-sions with rates during control sessions.Two advantages of these data are that theeffects of the agent were well characterized(doses were assessed 2-3 times per monkey)and the individual differences in effectoccurring among the monkeys could serveto model individual differences in thehuman population.

GBR 12909 (1-1.7 mg/kg) decreasedresponding maintained by cocaine almostcompletely without affecting food-main-tained responding. This type of effectshowed that cocaine-maintained respond-ing could be decreased by low doses of thisdrug. With higher doses of GBR 12909,food-maintained responding decreased.This effect may have been associated witha drug-induced disruption of behavior orbecause some aspect of food-maintainedresponding provided a greater resistance

to the drug effect than that seen withcocaine-maintained responding. A meandose-effect function for each of these behav-ioral end points was then constructed usingthe individual data from the eight animals.While there was no overlap of these curves(i.e., GBR 12909 always decreased cocaine-maintained responding at lower doses thanfood-maintained responding), some animalswere more sensitive to the effects of GBR12909 than others. However, traditionalstatistical approaches confirmed that GBR12909 had different effects on these twobehaviors. For example, single factorrepeated measures analysis of variance(ANOVA) indicated a significant maineffect of maintaining event (F= 18.428;df = 1, 14; p< 0.0007), dose (p < 0.)0001),and an interaction.

Figure 1 shows the essential elements ofthe method of risk assessment applied tothese data. First, the effect of GBR 12909on each type of response was determinedrelative to each monkey's baseline rate foreach response by dividing rates during drugsessions by rates during control sessions foreach component. This represents a normaldose-effect function for a single animal,only two different end points could bedescribed. Next, a straight line was fit to thelinear portion (only those doses that pro-duced greater than a 20% decrement butless than an 80% decrement in responding)of each curve using linear regression. Effect

100 -

80 -

60 -

40 -

20 -

2 -1 0 1

In dose GBR 12909, mg/kg

Figure 1. Transformed dose-effect summaries of therate-decreasing effects of GBR 12909 on respondingmaintained under multiple FR30 schedules of cocainedelivery (solid line) and food-pellet delivery (dottedline). Effects are the mean of double-determined obser-vations of the rate-decreasing effects (see text for adescription of the restriction of the range) in eightmonkeys, expressed as a percentage of the individualbaseline rate of responding. Doses were transformedto the natural logarithm (In). Variability (1 z-score) inthe dose to produce 90 and 10% effects, based on dif-ferences in the individual curves, is indicated by thehorizontal bars.

(percent of control) was plotted against thenatural logarithm (ln) of the dose. Thefigure shows only the mean circumscribeddescending limb, from 90 to 10% of con-trol. The variability (SD) in the 90 and 10%of control was determined from the individ-ual functions (horizontal lines). For exam-ple, the mean ln dose of GBR 12909 toproduce a 10% decrement in cocaine-main-tained responding was -1, with a SD ofabout 0.5. As is apparent from Figure 1, thedoses of GBR 12909 required to decreasefood-maintained responding were high. Infact, for three animals complete decreases infood-maintained responding were not pro-duced by the highest dose of GBR 12909studied. Since previous studies suggestedthat a higher dose (5.6 mg/kg) would com-pletely abolish responding, it was conserva-tively assumed that similar effects would beobtained in these animals and that it wouldbe in the best interest of the animals to nottest higher doses.

The estimates of risk were thenobtained by calculating the successive prob-abilities (i.e., p=0.1, p=0.01, p=0.001) ofobtaining a 10% decrement in this group ofanimals using a z-score approach. In thepast, only estimates from the low end ofthese distributions have been assessedbecause of the specific interest in the effectsof low doses. In this study, the overlap ofthe effects of similar doses of GBR 12909on both cocaine- and food-maintainedresponding could be directly compared.Transformation then allowed determina-tion of doses (in milligram per kilogram)expected to decrease responding by 10% inthe specified proportion (i.e., 1 out of 10,1 out of 100, etc.) of the population (24).Figure 1 characterizes the descending limbsof the dose-effect curve for each end point.(A mean curve was never used for theanalysis; it simply represents the mean slopejoining the mean 90 and 10% point esti-mates.) The mean parameters (a=intercept;b=slope) for the individual curves, the 90%effect (in In dose and in milligram per kilo-gram), and the SD of In dose for cocaine-maintained responding were

FRC, mg/kga b 90% (In) 90%

Mean 18.103 -74.52 -1.0538(SD) (0.4226)

0.37

The resulting risk figures for estimation ofa 90% effect in successively smaller propor-tions (1/10 [p= 0.1], 1/100 [p= 0.0I], and1/1000 [p= 0.001]) of the population basedon cocaine-maintained responding were

Environmental Health Perspectives * Vol 104, Supplement 2 - April 996

tm.sa

0CLU)9)-6r

0L)

0

a)PCDCL

393

J.R. GLOWA

In doseDose, mg/kg

The mean par

Probabllity In addition, the effects of GBR 12909 on0.1 0.01 0.001 cocaine-maintained responding may also-0.652 -2.321 -3.076 be similar to those of delivering free food0.19 0.10 0.05 on behavior maintained by food delivery.ameters for food-maintained However, in contrast to the decrease in

responding wereFRF

cocaine-maintained behavior produced byGBR 12909, the decrease in food-main-

a b 90% (in) 90%, mg/kg tained behavior by this drug was consid-170.45 -134.6 0.5579 2.02 ered adverse (i.e., an effect that might

(0.6258) interfere with the normal behavioral func-tioning). The variability in the likelihoodof producing each effect in individual ani-

tulting risk figures for estimation of mals was then used to specify the probabil-effect in successively smaller propor- ity of seeing that effects at doses that would/10 [p= 0.1], 1/100 [p= 0.01], and be expected to affect certain proportions ofD [p = 0.001O]) of the population the population.in food-maintained responding were Further analysis of the current data was

Probability used to explore potential therapeutic0.1 0.01 0.001 indices for GBR 12909. Traditional thera-

peutic ratios have been constructed usingglkg 0.72 0.27 0.09 comparable levels of effect on different endpoints (e.g., LD50/ED50). The current

approach could extend these methods tomparisons of the distributions of compare the extent to which similar levelso decrease each type of responding of effect on both behaviors overlapped. Fory t-tests indicated (p <0.001) the example, the low end of the distribution ofral specificity of the effects of GBR effects for food-maintained responding(i.e., the two behaviors were affected suggests that 1 out of 10 individuals wouldrent doses). exhibit a 10% decrease in responding when

lusions given 0.72 mg/kg GBR 12909. Likewise, 1out of 100 individuals would be expected

irrent report is not intended to to exhibit this effect when given 0.27tively explore the relationships mg/kg. For cocaine-maintained responding,a different effects of an agent on dif- the high end of the distribution suggests,nd points but rather to extend the that 9 out of 10 individuals would exhibitment of a risk assessment technique at least a 10% decrease in drug-seekingwell-defined data set. The current behavior at 0.45 mg/kg. Thus, a dose could

Is are of interest because they direcdy be conceived that diminished drug-seekinge sources of variability that have been in at least 90% of the population and with-d in other risk assessment processes. out producing adverse effects in 90 to 99%Lmple, the use of the sample variance of the population.nate risks provides data directly rele- In contrast, it may be of interest toone source of uncertainty in risk compare the overlap of a highly therapeutic

lent (individual differences). In the dose (i.e., the ED90 for cocaine-maintainediproach this uncertainty factor is typi- responding) with a minimally toxic dosesumed to be an order of magnitude. (i.e., EDIO for food-maintained respond-irrent studies also suggested that, ing) in order to assess the probabilities of:here was a clear difference in the obtaining the most benefit at the least cost.Df GBR 12909 on these two types of Alternatively, it might be argued that onlyDr, the variability in effects on each a single dose ofGBR 12909 could be givened relevant data from which risks at one time, focusing the emphasis of suchoe assessed. analyses on distributions of two different! current methods established dose- levels of effect (e.g., a dose of 1.7 mg/kgelationships for GBR 12909 on two decreased drug-maintained respondingit behaviors. These dose-effect func- almost completely while having little effect,uggested that each behavior was on food-maintained responding). Obviously,i differently by this drug, as expected. the extent to which there is no apparent2909 decreased drug-seeking behav- overlap between the mean desired andese effects may be similar effects to undesired effects suggests the drug may beofmethadone to treat heroin or nico- relatively safe, although a 1/10 chance of antches to treat tobacco abuse (28,29). adverse effect may warrant concern. The

present data suggest that there is someoverlap of effects with GBR 12909,although it is by no means complete. It maybe constructive to compare the degree ofoverlap with that of other therapeutic agents(e.g., diazepam) when assessed on relevantbehaviors (e.g., punished and nonpunishedresponding). In the present case, some of thisoverlap may have resulted from the treat-ment of the data. The slope of the functionfor food-maintained responding was steeperthan that for cocaine-maintained respond-ing, suggesting that the artificial decreasesimposed for monkeys that did not exhibitcomplete decreases in food-maintainedresponding may have inflated variability.

One interesting feature of the currentapproach is the variability of an effect (andhence, risks estimates) that was almost con-stant over the range of the dose-effectfunction. For example, examination of thevariability at different effect levels (i.e., 90,50, 10%) on each curve demonstrated aremarkable similarity at different points forfood-maintained responding (from 0.639 at90% to 0.626 at 10%). Variability increasedslightly with the increasing effect of GBR12909 on cocaine-maintained responding(from 0.423 at 90% to 0.724 at 10%).These observations illustrate that dose-toler-ance and benchmark approaches character-ize variability in a completely differentmanner. Confidence intervals invariably flareat the high and low ends of the dose-effectfunction, even if the observed variability isgreatest at the intermediate doses. One fea-ture of these hyperbolidike functions is that,with sufficient variability in the dose-effectdata, this flare may preclude the determina-tion of a benchmark dose because the entirelower confidence interval may never inter-cept the 10% effect level.A feature of quantitative approaches that

appears to be gaining acceptance is the useof logarithmic conversions of dose. Whilelog-normal distributions have been used andaccepted in pharmacology for many years(33) primarily because receptor interactionscan span orders or magnitude, their use inrisk assessment has received less attention.Previous studies have pointed to the advan-tage of preventing negative numbers in riskassessment (3,4) and to the validity of log-normal distributions of individual effectswhen sufficient numbers of subjects areused (27). Standard benchmark approachesmay also convert doses to log doses in unex-posed algorithms (2). In the current datathis conversion had an unusual effect on theupper end of the ED estimates. When thenormally distributed ln point estimates were

Environmental Health Perspectives * Vol 104, Supplement 2 * April 1996

Mean(SD)

The resa 90% (

tions (11/100(based o

In doseDose, mj

Cordoses t410% bbehavio12909by diffe

ConcThe ctexhausbetweerferent e

developusing amethodmeasurtassume(

For exato estinvant toassessmRID apcally asThe cuwhile teffects (behavicprovidcould b

Thceffect ridifferentions s

affectedGBR 1ior. Ththe usetine pai

394

NEUROBEHAVIORAL RISK ASSESSMENTS FOR THERAPEUTICS

converted back to real numbers, the con-verted distribution dispersed at a logarith-mic rate of change. The result should bethat observed effects rarely exceed thosepredicted at either end of the distribution,although in practice predicted effects at thelow end seem to occur with more regular-ity than at the high end. This could dimin-ish some concern of the predicted overlapof these two effects described above.

Another feature of the current methodthat deserves consideration is the use oflinear functions to fit individual dose-effectdata. There may be several issues includingthe likelihood that other types of functionsmay fit the data better. However, previousauthors have noted that the type of thefunction used to fit the linear region usuallyhas little effect on the risk estimate (24,34).In fact, traditional statistical techniques(ANOVA, etc.) may bias the slope of thefunction by incorporating excessive determi-nations of 0 and 100% effects. Eliminatingexcessive determination of doses withouteffect seems desirable, but it requires anarbitrary rule for data exclusion. The use offunctions that can effectively use all the datato more accurately fix the linear portion(e.g., logistic curves) should be explored.Linear functions are preferred, of course,because they strongly suggest that a directrelationship exists between dose and effect.While it is obvious that agents may havemany effects, the most parsimoniousconclusion upon observation of a linearfunction is that the one that is being mea-sured is directly affected by the agent.Unfortunately, not all dose-effect data canbe accurately characterized by a linear func-tion. For example, there is often an increasein responding with some agents whenassessed on behavioral measures. These

increases inevitably give way to decreases,creating an inverted U-shaped function.Wood and Cox (35) recently reported riskassessments using data in which low dosesof toluene increased rates and higher dosesdecreased rates. They fit a linear function tothe ascending limb (rate-increasing effects)alone. While this approach clearly avoidsthe issues of the interpretation of a complexfunction and uses the most sensitive effectof toluene, it questions the issue of whetherany effect of a toxicant is adverse. In thecurrent studies, effects of GBR 12909 ondrug-seeking behavior were the most sensi-tive but were not considered toxic.

The current report applies a quantita-tive approach to assess the risks of obtainingtwo different effects with the same agent.Drug-seeking behavior was targeted forselective elimination by doses of GBR12909 that were not expected to producetoxic effects. It should be clear from theseresults that a decrease (or increase) in onebehavior alone may not be the best indica-tion of behavioral toxicity. Likewise, selec-tive decreases in food-seeking behaviorproduced by other agents [e.g., appetitesuppressants (36)] may be an intendedeffect. Literature is incomplete on the dif-ferent behavioral effects that agents canhave on behaviors maintained by differentevents. In the absence of specific knowledgeof the effect of an agent on behaviors main-tained by certain events, complete decreasesin responding may warrant concern. Whena single behavior is studied, it must be con-sidered that doses which decrease respond-ing completely might decrease all types ofresponding completely, a clear reason forconcern. However, when one behavior ismore affected than another, it could bebecause the susceptible behavior is less well

maintained. This possibility points to theimportance of reinforcement processes inthe assessment of behavioral toxicology. Inthe current case, a decrease in the relativelywell-maintained food-presentation behavioris clear reason for concern.

The dose-tolerance method was easilyadapted to demonstrate a selective effect ofGBR 12909 on two different behaviors, aswell as to estimate the risks of obtainingthose effects in a population. This methodshould be considered positively over otherapproaches since it directly employs indi-vidual differences in sensitivity to an agentas a metric of risk. However, not all types ofend points can be studied using single-sub-ject, repeated-exposure designs. Recently,we have developed a variant of the dose-tol-erance approach for use with groups-designsingle-exposure data (Bogdan et al., unpub-lished data). This approach uses an iteratingline-fitting program to calculate all possibledose-effect combinations that could be cre-ated from individual effects when differentgroups of animals are treated with differentdoses (e.g., three dose groups with six ani-mals each would produce a 6 x 6 x 6 matrixyielding 216 possible dose-effect functions).This approach was very effective whenapplied to a data set addressing the effects ofcarbaryl on motor activity of rats (6). Therisk figures it produced were less conserva-tive than those of the current (within-sub-ject) approach but were more conservativethan those produced by the maximal likeli-hood estimation/confidence-intervalapproach. Future studies will continue tocompare the results of these approaches toprovide the practitioner a range of optionsfor quantitative risk assessment.

REFERENCES

1. Barnes G, Dourson M. Reference dose (RfD): description and usein health risk assessments. Reg Toxicol Pharmacol 8:471-486(1988).

2. Crump KS. A new method for determining allowable dailyintakes. Fundam Appl Toxicol 4:854-871 (1984).

3. Dews PB. On the assessment of risk. In: DevelopmentalBehavioral Pharmacology (Krasnegor N, Gray J, Thompson T,eds). Hillsdale, NJ:Lawrence Erlbaum, 1986;53-65.

4. Dews PB. Some general problems in behavioral toxicology. In:Neurobehavioraf Toxicology (Annau Z, ed). Baltimore:TheJohns Hopkins University Press, 1986;424-434.

5. Dourson ML. New approaches in the derivation of acceptabledaily intake (ADI). Comments Toxicol 1:35-48 (1986).

6. Glowa JR, MacPhail RC. Dose-response models for quantitativerisk assessment in neurotoxicology. In: Handbook ofNeurotoxicology, Vol 3: Approaches and Methodologies (ChangL, Tilson H, eds). New York:Raven Press, 1995; 777-787.

7. McMillan DE. Risk assessment for neurobehavioral toxicity.Environ Health Perspect 76:155-161 (1987).

8. Raffaele KC, Rees DC. Neurotoxicology dose/response assess-ment for several cholinesterase inhibitors: use of uncertaintyfactors. Neurotoxicology 11:237-256 (1990).

9. Weiss B. Neurobehavioral toxicity as a basis for risk assessment.Trends Pharmacol Sci 91:59-62 (1988).

10. National Research Council. Environmental Neurotoxicology.Washington:National Academy Press (1992).

11. Cote I, Rees DC, Glowa JR. An introduction to the principlesof risk assessment. In: The Vulnerable Brain and EnvironmentalRisks, Vol 3: Toxins in Air and Water (Isaacson R, Jensen K,eds). New York:Plenum Press, 1994;183-207.

12. Barrett JE. Effects of alcohol, chlordiazepoxide, cocaine andpentobarbital on responding maintained under fixed-intervalschedules of food or shock presentation. J Pharmacol Exp Ther196:605-615 (1976).

Environmental Health Perspectives * Vol 104, Supplement 2 - April 1996 395

J.R. GLOWA

13. Wood RW. Stimulus properties of inhaled substances: anupdate. In: Nervous System Toxicology (Mitchell CL, ed).New York:Raven Press 1982; 199-212.

14. American Psychiatric Association. Diagnostic and StatisticalManual of Mental Disorders, 3rd ed rev. Washington:APAPress, 1987.

15. Nestler EJ. Molecular mechanisms of drug addiction. JNeurosci 12:2439-2450 (1992) .

16. Balster RL. Drug abuse potential evaluation in animals. Br JAddict 86:1549-1558 (1991).

17. Aigner T, Balster RL. Choice behavior in rhesus monkeys:cocaine versus food. Science 201:534-535 (1978).

18. Koob GF, Bloom FE. Cellular and molecular mechanisms ofdrug dependence. Science 242:715-723 (1988).

19. Ritz MC, Lamb RJ, Goldberg SR, Kuhar MJ. Cocaine recep-tors on dopamine transporters are related to self-administrationof cocaine. Science 237:1219-1223 (1987).

20. Peck AW, Hamilton M. Psychopharmacology of bupropion innormal volunteers. J Clin Psychiat 44:202-205 (1983).

21. Bergman J, Madras BK, Johnson SE, Spealman RD. Effects ofcocaine and related drugs in nonhuman primates: III. Self-administration by squirrel monkeys. J Pharmacol Exp Ther251:150-155 (1989).

22. Rothman RB. High affinity dopamine reuptake blockers aspotential cocaine antagonists: a strategy for drug development.Life Sci 46:PL17-21 (1990).

23. S0gaard U, Michalow J, Butler B, Lund Laursen A, IngersenSH, Skrumsager BK, Rafaelsen 0. A tolerance study of singleand multiple dosing of the selective dopamine uptake inhibitorGBR 12909 in healthy subjects. J Int Clin Psychopharmacol5:237-251 (1990).

24. Glowa JR, DeWeese J, Natale ME, Holland JJ, Dews PB.Behavioral toxicology of volatile organic solvents: I. Methods:acute effects of to1uene. J Environ Pathol Toxicol Oncol6:153-168 (1986).

25. Glowa JR. Behavioral toxicity of n-octane. In: Advances inModern Environmental Toxicology, Vol 6 (Mehlman MA, ed).Princeton, NJ:Princeton Scientific Publishers, 1984;245-253.

26. Glowa JR, Dews PB. Behavioral toxicology of volatile organicsolvents: IV. Comparisons of the behavioral effects of acetone,methyl ethyl ketone, ethyl acetate, carbon disulfide, and tolueneof the responding of mice. J Am Coll Toxicol 6:461-469 (1987).

27. Glowa JR. Behavioral toxicology of volatile organic solvents: V.Comparisons of the behavioral and neuroendocrine effectsamong n-alkanes. J Am Coll Toxicol 10:639-646 (1992).

28. Glowa JR. A comparison of dose-effect based risk assessmentmethods. Neurosci Biobehav Rev 15:153-158 (1991).

29. Rees DC, Glowa JR. Extrapolations to humans for neurotoxi-cants. In: The Vulnerable Brain and Environmental Risks, Vol3: Toxins in Air and Water (Isaacson R, Jensen K, eds). NewYork:Plenum Press, 1994;207-230.

30. Glowa JR, Wojnicki FHE, Matecka D, Bacher J, Mansbach RS,Balster RL, Rice KC. Effects of dopamine reuptake inhibitors onfood and cocaine maintained responding: I. Dependence on unitdose of cocaine. Exp Clin Psychopharmacol 3:219-231 (1995).

31. Glowa JR, Wojnicki FH, Matecka D, Rice K, Rothman RB.Effects of dopamine reuptake inhibitors on food and cocainemaintained responding: II. Comparisons of effects of psychomo-tor stimulants and subchronic efects. Exp Clin Psychopharmacol3:231-239 (1995).

32. Wojnicki FHE, Bacher JD, Glowa JR. The use of subcuta-neous vascular access port in rhesus monkeys. Lab Anim Med44:491-494 (1994).

33. Gaddum JH. Lognornal distributions. Nature 156:463-466(1945).

34. Crump KS, Allen BC, Faustman EM. The Use of theBenchmark Dose Approach in Health Risk Assessment. Finalreport. Washington:U.S. Environmental Protection Agency,1992.

35. Wood RW, Cox CC. A repeated-measures approach to thedetection of the minimal acute effects of toluene. Toxicology6:221 (1986).

36. Corwin R, Schuster CR. Anorectic specificity as measured in achoice paradigm in rhesus monkeys. Pharmacol BiochemBehav45:131-141 (1993).

396 Environmental Health Perspectives - Vol 104, Supplement 2 - April 1996