## INTRODUCTION

The opioid antagonist naltrexone (NTX), is one of only three drugs with US Food and Drug Administration (FDA) approval for the treatment of alcoholism, and is also used in the treatment of other addiction and impulse-control disorders. NTX reduces alcohol consumption, craving, and relapse in subjects with a history of alcohol abuse (O'Brien et al, 1996; Anton et al, 1999; Davidson et al, 1999; Hernandez-Avila et al, 2006; see Heidbreder, 2005 for a recent review). Moreover, NTX significantly reduces ethanol consumption in rats (Stromberg et al, 1998) and monkeys (Boyle et al, 1998). However, the mechanism underlying NTX's efficacy in reducing alcohol intake has yet to be determined. One possibility is that NTX decreases the rewarding effect of alcohol intake (Swift et al, 1994; Volpicelli et al, 1995; Sinclair, 2001). Another possibility is that NTX reduces drinking by generating aversive side effects, such as nausea, when ethanol is consumed (Davidson et al, 1999; de Wit et al, 1999; McCaul et al, 2000; Mitchell et al, 2005a). However, there is also evidence that NTX significantly reduces alcohol craving in alcoholics during abstinence (Monti et al, 1999; Rohsenow et al, 2000; O'Malley et al, 2002), and improves resistance to thoughts, urges, and behaviors associated with drinking (Anton et al, 1999). Thus, it is plausible that NTX acts, in part, via alteration of certain cognitive processes. Clinical and preclinical data converges on the hypothesis that NTX may reduce impulsive responding. This hypothesis is supported by animal studies that have shown NTX to be effective at decreasing impulsive choice bias. For example, in rats, NTX reverses a morphine induced increase in preference for small immediate rewards over larger delayed rewards (Kieres et al, 2004). Moreover, in humans, NTX attenuates a variety of impulse disorders, including pathological gambling (Kim et al, 2001), binge eating in bulimics (Marrazzi et al, 1995), compulsive sexual behavior (Raymond et al, 2002), self-injurious behavior (Symons et al, 2004), and kleptomania (Grant, 2005). Taken together, these results suggest that endogenous opioids promote impulsivity, and that NTX improves self-control by blocking endogenous opioid activity. However, despite this wealth of clinical and animal data, to date, no laboratory studies have explicitly investigated the impact of NTX on impulsive responding in human subjects.

To determine the effect of NTX on discrete elements of impulsive choice behavior in humans, we used a modified delay discounting (DD) task. This DD task allows the separate evaluation of both impulsive decision-making and motor impulsivity. We previously demonstrated that abstinent alcoholics (AA) select the smaller, sooner reward option significantly more frequently than control subjects (CS) in this task (Mitchell et al, 2005b), a tendency we have quantified as the ‘Impulsive Choice Ratio’ (ICR). If this tendency of AA to choose impulsively reflects an enhancement of endogenous opioid signaling in response to rewarding stimuli, then NTX should reduce the impulsive choice bias of the AA group. Using a double-blind placebo-controlled randomized crossover design, we tested whether discounting of delayed rewards was reduced by a single acute dose of NTX (50 mg), and whether this effect differed between groups. We predicted that NTX would specifically reduce impulsive decision-making, but would have no effect on inhibitory control (or motor mismatch (MM)).

## METHODS

AA subjects (n=9) were recruited through flyers posted by local alcoholism recovery organizations and through referral by addiction treatment professionals. CS (n=9) were recruited by flyers posted in the local community. To minimize systematic differences in physiological state between subject groups, we elected not to recruit active alcoholics, as our protocol would have required those subjects to be in a state of acute withdrawal. AA subjects were recruited on the basis of a minimum of 2 weeks of self-reported abstinence from alcohol. Subjects were screened by telephone before participation. All subjects were between 19 and 38 years of age, and were screened for neurological disease, current treatment for other psychological disorders, and addiction to substances other than alcohol. CS were also screened for a history of alcohol abuse. Owing to the high incidence of chronic tobacco use among alcoholics, nicotine addiction was not considered grounds for exclusion for either group. Subjects who met our inclusion criteria in the phone screening were invited to participate in the experiment, which took place in a laboratory setting on the University of California, Berkeley (UCB) campus. Subjects provided written, informed consent, as approved by the UCB Committee for the Protection of Human Subjects. Subjects were compensated approximately $150 for their participation. Subjects participated on 2 days, separated by at least 48 h, to allow complete elimination of NTX from the system. The DD task itself lasted approximately 1 h. In addition to the DD task, during session one, subjects also completed a battery of standard questionnaires (see ‘Behavioral Inventories’). ### NTX Administration After being screened for contraindications for NTX, including a urine pregnancy test for women, subjects were administered either a 50 mg NTX capsule or an identical placebo capsule. Experimenter and subject were both blind to capsule content; capsule order was counter balanced across subjects. During session one, subjects were asked to fill out a series of questionnaires, and were then free to relax during the remaining time. Administration of the DD task began approximately 3 h following capsule ingestion. This interval was chosen to minimize any acute physiological effects of NTX during the DD task while still achieving significant opioid receptor antagonism (Atkinson, 1984; Swift et al, 1994; King et al, 1997). ### Behavioral Inventories We quantified alcohol abuse severity via the AUDIT (Saunders et al, 1993), based on its sensitivity and strong correlation to the DSM-III-R diagnosis for alcoholism (eg, Bradley et al, 1998), and its demonstrated cross-cultural validity (Cherpitel, 1998). Domain I of the Drug Use Screening Inventory (DUSI-I; Tarter, 1990) provided additional information regarding the severity of each subject's alcohol abuse behaviors. DUSI-I scores are reported in terms of the percent of affirmative answers from Domain I, part B. For the AUDIT & DUSI-I questionnaires only, subjects were instructed to answer based on the year before achieving abstinence if they had ceased consuming alcohol. To determine whether group differences in other affective and/or behavioral factors could impact NTX effects on choice behavior, we also collected responses from each subject for the Beck Depression Inventory (BDI; Beck and Steer, 1987), the Depression Anxiety and Stress Scales (DASS; Lovibond and Lovibond, 1993), the Barratt Impulsivity Scale-11 (BIS; Barratt, 1994), Rotter's Locus of Control Scale (LOC; Rotter, 1966), the South Oaks Gambling Screen (SOGS; Lesieur and Blume, 1987), the Future Time Perspective Inventory (FTPI; Wallace, 1956), and the Schizotypal Personality Questionnaire (SPQ; Raine, 1991). We also collected information about familial alcohol abuse using the Family Tree Questionnaire (FTQ; Mann et al, 1985). Occupation and education information were collected to calculate the Hollingshead Socioeconomic Status (SES) score (Hollingshead, 1975). ### DD Task The details of this procedure have been reported elsewhere (Mitchell et al, 2005b). In brief, during each session subjects completed eight blocks of 47 or 48 trials, with rest periods between blocks, as needed (total duration 1 h). Subjects were given a brief practice session (4 min) before beginning the DD task. For each trial, two options were presented, each consisting of a dollar amount at a point in time. On every trial, one option was one of six ‘full’ amounts, ($1, $2,$5, $10,$20, or $100) at one of five future delays (1 week, 2 weeks, 1 month, 3 months, or 6 months). The other option was a lesser amount available at a sooner point in time. The discount rate randomly varied among the following four percentages: 70, 85, 90, or 95%. We refer to these two alternatives as the ‘earlier’ vs the ‘later’ option. Right–left position of the earlier and later options was randomized. Subjects indicated their choice by pressing one of two buttons on a keypad. Trial type was indicated by an instruction cue, which indicated how the subject was to select one of the two options that subsequently appeared (see Figure 1). Trials were one of four instruction conditions: WANT (W), DON’T WANT (DW), SOONER, and LARGER, which are considered together as ‘CONTROL’ (CON). In the W condition, subjects chose the option they preferred. In the DW condition, subjects were asked to make the same evaluation, but to press the button corresponding to the opposite choice. On CON trials, subjects selected the side with the sooner time point or larger amount of money. This verified that subjects comprehended the task and were compliant with task instructions. Trial types were randomly intermixed, with weighted ratios of 1/2 for the W condition and 1/6 each for each of the other conditions. The order of trial types did not vary across subjects; however, the delayed amount, delay time, and discount rate were randomly selected on each trial. The decision to use hypothetical rewards was based on results from numerous studies comparing choices for real vs hypothetical monetary rewards in discounting paradigms (Critchfield and Kollins, 2001; Johnson and Bickel (2002); Madden et al, 2003; Madden et al, 2004; Lagorio and Madden, 2005). ### Data Analysis We used two indices of discounting, the proportion of earlier choices, which we have termed the ICR, and the cumulative gain ratio (CGR). The latter metric is a ratio of the total dollar amount chosen over the course of the experiment, divided by the maximum total dollar amount available. These values were calculated across all W trials, as well as separated according to delay time and delayed amount. As described previously (Mitchell et al, 2005b), our design does not allow robust determination of a hyperbolic discount rate (k), thus k was not used as a dependent measure. We also calculated the criterion interest rate acceptance threshold for each subject. To do so, we first calculated the simple interest rate for each trial according to the following equation: where delay is equal to the Later timeSooner time interval. We then plotted the percentage of trials in which the subject accepted the Later option against the interest rate. We fitted the data with a logistic regression of the following form: Given the two alternative forced-choice structure of our task design, we defined the interest rate criterion acceptance threshold (from the logistic fit) as the interest rate for which the subject chose the Later option 75% of the time. We were unable to successfully fit a logistic regression curve to the interest rate choice data for one subject, so these data were excluded from further interest rate analyses. Reward preference in DW trials was inferred as the non-selected option. On the basis of these responses, we calculated an inferred ICR (iICR) for each delay time. The absolute difference between the ICR and iICR for each delay time was calculated, and the sum of the ICR–iICR across all delay times was used as an index of motor control. We refer to this sum value as ‘MM’. For single factor statistical comparisons between subject groups, we used unpaired two-tailed t-tests for continuous measures and χ2 tests for categorical measures. For multi-factorial comparisons, we used mixed repeated measures ANOVAs as implemented in SPSS, with group as a between subjects factor. Where sphericity assumptions were violated, a Greenhouse-Geisser non-sphericity correction was applied. Post hoc paired comparisons were performed where indicated using two-tailed t-tests. To ensure the validity of parametric statistical tests, when data were not normally distributed, appropriate arcsine-root transformation was applied before making statistical comparison. Simple regression analysis and macro-based analyses of covariance (ANCOVA) were performed using Excel. To estimate which continuous variables had the greatest predictive value for the NTX effects on ICR and MM, linear multiple regression analyses were carried out using SPSS. For each multiple regression analysis, we entered variables stepwise, divided into four blocks. The blocks were as follows: block 1—group (1 or 2), age, years of education, HH-SES; block 2—AUDIT, DUSI—1B, FTQ density, time abstinent; block 3—BDI, DASS, SOGS, LOC, FTPI-I-max extension, FTPI-I-mean extension, BIS, SPQ; block 4—WANT trial RT change, DW trial RT change, CON trial RT change. ## RESULTS ### Demographic and Psychometric Data The AA and CS groups differed somewhat in terms of demographic variables. Specifically, on average, the CS had a higher level of education, and were more likely to be Caucasian and female; age and SES did not differ between groups (Table 1). However, we have shown previously that gender, ethnicity, and education level do not have a significant impact on performance in this DD task (Mitchell et al, 2005b), a result we have confirmed in the present study. Consistent with our expectations, the AA subjects reported significantly higher levels of alcohol abuse and alcohol-related problems (Table 1). Members of the AA group also reported a significantly higher incidence of familial alcohol abuse (Table 1). To determine whether any additional individual or group differences in personality traits predicted NTX response, we collected additional psychometric data from subjects. The psychometric data are shown in Table 1. We found that the AA and CS groups did not differ on measures of emotional distress (DASS) or future orientation (FTPI). However, the AA group scored significantly higher on measures of depression (BDI), impulsivity (BIS-11), and gambling (SOGS). Furthermore, the AA group had a significantly less internal LOC. The impact of these differences on NTX's effects are considered in the multiple regression analyses below. ## TASK REACTION TIMES Reaction time (RT) data from the DD task are shown in Table 2. A mixed model ANOVA indicated no significant main effect of acute NTX administration (F1,16=2.35, p=0.145) or subject group (F1,16=0.01, p=0.915) on RT. However, consistent with our previous findings, we did detect a significant effect of trial type (F1.18,18.9=48.18, p<0.001). Post hoc paired-comparisons determined that the trial type effect on RTs was due to significantly longer RTs for W trials than for CON trials (1959 vs 1478 ms, respectively, F1,16=34.46, p<0.001), and significantly longer RTs for DW trials than for W trials (2174 vs 1959 ms, respectively, F1,16=48, p<0.001). We found no significant interaction between drug and group (F1,16=0.25, p=0.622) or drug and trial type (F2,32=0.77, p=0.47). Additionally, NTX had no effect on CON trial accuracy (F1,16=1.16, p=0.298). ### Acute NTX does not Reduce Impulsive Choice in Either Group Consistent with previous findings (Vuchinich and Simpson, 1998; Petry, 2001; Mitchell et al, 2005b), the AA group made impulsive choices significantly more often than the CS group (F1,16=5.846, p=0.028). Also, consistent with our previous findings (Mitchell et al, 2005b), ICR was sensitive to both the later reward delay (Figure 2c; F1.74,27.83=12.35; p<0.001) and the later reward amount (Figure 2d; F1.86,29.67=17.49, p<0.001). However, as shown in Figure 2a, NTX did not significantly alter ICR for either subject group (F1,16 0.60, p=0.451). We also failed to detect a significant drug × group interaction (F1,16=0.02, p=0.899). We did, however, find a significant drug × later reward amount × group interaction (F5,80=4.51, p=0.001; see Figure 2d). Post hoc tests determined that this interaction was due to a group difference in the relative effect of NTX on the probability of discounting$5 or $10 (F1,16=5.0, p=0.04). In the AA group, the mean ratio of sooner choices given a$10 later reward increased on NTX to a greater extent (0.72 to 0.79) than for a later reward of $5 (0.82 to 0.84; Figure 2d, left panel). In the CS group, the opposite relationship was true: NTX increased discounting of a$5 later reward to a greater degree (0.48 to 0.58) than a \$10 later reward (0.37 to 0.39; Figure 2d, right panel).

In addition to the ICR, we also examined whether NTX affected CGR. We found that NTX failed to alter the total reward amount that subjects accumulated (F1,16=1.22, p=0.285). We also found no significant drug × group interaction for CGR (F1,16=0.18, p=0.675). Finally, we tested whether NTX impacted subjects' criterion interest rate acceptance threshold. We found a significant main effect of group on the interest rate criterion threshold (F1,15=6.49, p=0.022), with the CS group being willing to accept significantly lower interest rates (placebo criterion rates, AA: 24±7%, CS: 6±2%). However, as for the other choice measures (ICR and CGR), we found neither a significant NTX effect (F1,15=0.73, p=0.406), nor a significant drug × group interaction (F1,15=0.11, p=0.748). Taken together, these results indicate that NTX does not have a generalized effect on decision-making behavior under normal circumstances in either AA or CS.

### Acute NTX Reduces MM in CS, but not AA

In contrast to the lack of consistent NTX effects on explicit decision-making, we detected a robust NTX effect on MM. NTX reduced the absolute difference between each subject's ICR from the W condition, and their iICR from the DW condition (Figure 3; F1,16=5.77; p 0.029). A mixed-design repeated measures ANOVA found no main effect of group on MM (F1,16=2.05; p=0.172). Although there was a trend towards greater mismatch reduction in the CS group (Figure 3), we found no significant group × drug interaction (F1,16=1.58; p=0.228). As shown in Figure 3b, within the AA group, we observed considerable inter-subject variability in the effect of NTX on MM (t7=0.54, p=0.645), whereas the effect of NTX on MM was very consistent in the CS group, as a paired t-test demonstrated (t7=4.34, p=0.003).

### Factors Predicting NTX Effect on Impulsive Choice

We next tested the value of the collected demographic and psychometric measures in predicting the effect of NTX on ICR. We found that only one variable, the LOC score held significant predictive value (r=−0.51, t17=−2.4, p=0.03; Table 3; Figure 4a). Lower LOC scores, reflecting a more internal attribution style, correlated with a NTX-induced increase in impulsive choice, whereas more external attribution style correlated with reduced impulsive choice on NTX. We found that no other factor added into the model added significantly more predictive power than LOC alone for the NTX effect on ICR. As this correlation appeared to be driven by the AA group, we repeated the multiple linear regression procedure with data from each group independently. Within the AA group, we again found a significant correlation between LOC scores and NTX's effect on ICR (r=−0.80, t8=−3.47, p=0.01), however, the FTPI (part I) mean event extension time provided significant added predictive power to the LOC alone (Table 3). In contrast, for the CS group, we found that the FTPI part I mean extension score alone demonstrated significant correlation with the NTX effect on ICR (r=0.73, t8=2.83, p=0.03; Table 3). Importantly, this correlation is in the opposite direction relative to the correlation seen in the AA group, which likely explains why the FTPI mean extension did not emerge as a significant predictive factor in the whole group multiple regression analysis (Figure 4b). Among the CS group, greater future orientation correlated with increased ICR on NTX, whereas among the AA group, greater future orientation was associated with decreased ICR on NTX.

Given the small sample sizes of the present study, it is important to consider whether we have adequate statistical power to rule out correlations between other factors and NTX's effect on ICR. However, power analyses indicated that we had greater statistical power to detect a significant relationship with every other variable, excepting FTQ density, than for LOC. It remains possible that we failed to detect additional, weaker correlations, so negative findings should be considered with some caution.

### Factors Predicting NTX Effect on MM

We also used multiple linear regression to identify any demographic or psychometric measures that had significant predictive value in terms of the NTX effect on MM. We found that a single variable had significant predictive power: the NTX effect on RT in the DW condition. The NTX effect on RT in the DW condition was positively correlated with the effect on MM (r=0.48, t17=2.19, p=0.04; Table 4; Figure 5a). That is, responding more quickly in the DW condition on NTX was associated with less MM on NTX. No other factor added into the model added significantly more predictive power than the DW RT change alone for the NTX effect on MM. Repeating the multiple regression procedure within the AA group, we found that no factor was significantly correlated with the NTX effect on MM, however, we found a trend for a similar correlation between change in MM and change in DW RT (r=0.59, t8=1.93, p=0.09). Carrying out the multiple regression procedure within the CS group, we found that a single factor significantly correlated with the NTX effect on MM: LOC score. We found that CS with more external attribution styles showed less MM on NTX (r=−0.67, t8=−2.40, p<0.05; Table 4; Figure 5b). A similar correlation was seen with LOC scores and the effect of NTX on inferred ICR (r=−0.88, t8=−4.92, p=0.002), consistent with the hypothesis that NTX is reducing MM by reducing errors in the DW condition.

## DISCUSSION

### NTX Effects on Impulsive Choice

In testing the effect of acute NTX on decision-making behavior, we found that NTX did not reliably reduce impulsive choice in the AA group as a whole. However, NTX's effect on choice bias across individuals was found to be robustly predictable by a single factor. Scores on Rotter's LOC scale reliably and robustly predicted NTX's effect on choice bias. This relationship was noted across the entire group of subjects tested, but was particularly strong in the group of AA. Owing to the widely reported variability in the therapeutic efficacy of NTX in treating alcoholism (O'Malley et al, 1992, 1996; Volpicelli et al, 1992, 1997; Anton et al, 1999; Chick et al, 2000; Johnson and Ait-Daoud, 2000; Heinala et al, 2001; Krystal et al, 2001; Morris et al, 2001; Guardia et al, 2002), identifying factors that predict therapeutic response to NTX is a critical goal of alcoholism research. Although the cognitive effects of acute and chronic NTX may differ, recent data supports equivalent or greater efficacy of acute NTX dosing, relative to daily maintenance, in reducing excessive alcohol intake (Hernandez-Avila et al, 2006). Thus, determining the cognitive effects of acute NTX doses is warranted. qOur data suggest that LOC scores index a biological difference that impacts NTX's effect on impulsive choice. Strengthening this conclusion is the fact that we have recently replicated the LOC finding in a separate, cohort of 40 subjects (Altamirano et al, 2006). Thus, investigating the predictive value of LOC scores in NTX clinical outcomes is warranted.

LOC is a personality trait referring to an individual's perception of control over events and consequences (Rotter, 1966), which has been extensively studied. Much of this work has led to the hypothesis that LOC scores reflect individual differences in tonic frontal dopamine (DA) transmission (Declerck et al, 2006). Evidence includes the correlation of LOC scores with serum DA metabolite levels (De Brabander and deClerck, 2004), as well as with frontal lobe function (Amrhein et al, 1999; Stevens et al, 1999). By this theory, NTX's effect on impulsive choice in this study depended on frontal dopaminergic tone, with low DA subjects becoming less impulsive and high DA subjects more impulsive.

Given that impulsive choice is reduced by acute elevation of DA levels; (de Wit et al, 2002; Wade et al, 2000), a simple explanation of our findings is that NTX exerts frontal DA tone-dependent effects on tonic DA release. This is consistent with studies demonstrating opioid regulation of DA neurons (eg, Ostrowski et al, 1982; Sesack and Pickel, 1992; Oswald and Wand, 2004; Berthele et al, 2005), including those projecting to prefrontal cortex (Margolis et al, 2006). Given that NTX acts at both μ- and κ-opioid receptors, which exert opposing effects on forebrain DA release (Spanagel et al, 1992; Herz and Spangel, 1995; Margolis et al, 2006), differential effects of NTX on DA levels could be due to differences in relative μ- and κ-mediated effects of NTX, as depicted in Figure 6. Relative μ-receptor to κ-receptor blockade effects would be expected to differ in subjects with low circulating levels of endogenous μ-receptor ligands, as is the case with alcoholics and their offspring (Govoni et al, 1983; Vescovi et al, 1992; del Arbol et al, 1995; Dai et al, 2005), or in subjects with low levels of μ-receptor expression, as is found in subjects with low frontal DA levels (Berthele et al, 2005) owing to their catechol-O-methyltransferase genotype (Meyer-Lindenberg et al, 2005), and in subjects with the A118G polymorphism of the μ-receptor (Zhang et al, 2005). This latter polymorphism has also shown a functional differentiation in the therapeutic response to NTX (Oslin et al, 2003), and has distinguished alcoholics from non-alcoholics in some populations (Bart et al, 2005). Based on this line of reasoning, we would expect to find a group difference in NTX on decision-making, however given the small sample size and large variance within groups, this study lacked sufficient statistical power to draw such a conclusion.

### NTX Effects on MM

With respect to the effects of NTX on MM, our data is consistent with one of two possibilities. The first possibility is that NTX improves inhibitory motor control by altering a pathway with differential baseline functioning between the CS and AA groups. A candidate pathway is baseline μ-receptor signaling. As noted above, low μ-receptor ligand tone has been demonstrated in AA. However, an alternative explanation for the increased NTX effect in the CS group is that endogenous opioids are elevated due to more recent alcohol intake rather than a constitutive difference. To our knowledge, however, there is no evidence to date supporting the idea that NTX improves motor control.

A second possibility is that NTX reduces MM by reducing response conflict. This possibility can also explain the differential effects seen in the AA and CS groups. In the CS group, attraction to the larger reward creates response conflict in the DW condition. There, the subjects must override the pull of responding to the more attractive large reward to respond to the small reward. On placebo, CS chose the large reward in DW trials more often than was predicted by their W condition choices, indicating response errors in the DW condition, as confirmed by post-experiment debriefing. In the CS group, NTX significantly reduced such errors. This result is consistent with NTX reducing response conflict by reducing attentional bias toward desirable stimuli or the tendency to select a preferred reward. In fact, NTX selectively reduces consumption of preferred palatable liquids in rats (Taha et al, 2006). In contrast, the DW error rate of the AA group was much smaller. This difference in errors could be explained by the fact that for the AA group, the correct response in the DW condition is more often the larger reward. Thus, for AA subjects, attentional bias toward the larger reward reduces, rather than increases, response conflict in the DW condition, resulting in fewer errors (ie, less MM). NTX would then be predicted to have little or no effect on MM in the AA group (or perhaps to slightly increase MM by reducing bias toward the larger reward), which is consistent with our results. If NTX is indeed reducing response conflict owing to choosing against the larger reward, one would predict that RT on placebo in the W condition would positively correlate with NTX's effect on ICR, which is indeed the case (r=0.48, t=2.17, p<0.05). In other words, if slower RTs in the W condition indicate greater response conflict owing to attentional bias toward the large reward, and if opioids contribute to such bias, NTX should reduce it. Our results are consistent with this interpretation of NTX's effect.

Finally, the fact that speeding of RT in the DW condition on NTX correlated with the magnitude of MM reduction lends further support to the notion that NTX is reducing response conflict. Although a strictly motor effect of NTX is possible, the alternative that NTX decreases attentional bias toward desirable rewards holds greater appeal, as it is consistent with the results of a number of related studies. For example, NTX selectively reduces consumption of preferred foods (Apfelbaum and Mandenoff, 1981). Moreover, decreased attentional bias toward attractive stimuli by NTX is consistent with data showing that NTX reduces cue-elicited alcohol craving in alcoholics (Monti et al, 1999; Rohsenow et al, 2000; O'Malley et al, 2002). Thus, effects on attentional bias could be one means by which NTX exerts its therapeutic effects in alcoholics. As previous work has correlated attentional bias toward alcohol-related cues with alcohol abuse severity (Weinstein and Cox, 2006), further exploration of this possibility is needed.

### Relevance to Alcoholism

An important consideration is the relevance of our findings to the ability of NTX to reduce alcohol consumption. As outlined in the Introduction, several possible mechanisms for this effect of NTX on a complex behavior have been proposed, to which we now add the potential shift in preference from immediate to delayed rewards. Our model raises the possibility that NTX reduces impulsive choice in proportion to the degree of κ-receptor mediated effect relative to μ-receptor mediated effect. This notion is supported by data showing that people with the A118G μ-receptor polymorphism show reduced μ-receptor expression (Zhang et al, 2005), and more effective NTX-mediated abstinence from alcohol (Oslin et al, 2003). However, our small sample sizes make it difficult to conclusively attribute the impulsive choice differences we find between the AA and CS groups to propensity to alcoholism per se, or other related factors, such as depression.

### Summary

In conclusion, we find that the effect of an acute NTX dose on impulsive choice in humans is predicted by the individual's LOC score. We find that a lower LOC, reflecting a more internal attribution style, correlates with an increase in impulsive choice on NTX, whereas a more external attribution style correlates with reduced impulsive choice on NTX. This relationship was particularly strong in the AA group. Future orientation, as measured by the FTPI was also predictive of NTX's effect on impulsive choice, although the direction of the correlation differed between groups. We also found that NTX reduced motor errors in a conflict condition. Our results suggest that this effect was due to a reduction of response conflict, a hypothesis that bears further investigation. Together, the results reported here provide new insights into possible mechanisms for NTX's ability to reduce drinking, and suggest several novel lines of research in this area.