Neurocognitive variation in smoking behavior and withdrawal: genetic and affective moderators


  • D. E. Evans,

    Corresponding author
    1. Tobacco Research and Intervention Program
      *D. E. Evans, Department of Health Outcomes & Behavior, Tobacco Research and Intervention Program, Moffitt Cancer Center, 4115 East Fowler Avenue, Tampa, FL 33617, USA. E-mail:
    Search for more papers by this author
  • J. Y. Park,

    1. Department of Risk, Assessment, Detection, and Intervention, Moffitt Cancer Center, Tampa, FL
    2. Department of Oncologic Sciences
    Search for more papers by this author
  • N. Maxfield,

    1. Department of Communication Sciences and Disorders, University of South Florida, Tampa, FL, USA
    Search for more papers by this author
  • D. J. Drobes

    1. Tobacco Research and Intervention Program
    2. Department of Oncologic Sciences
    Search for more papers by this author

*D. E. Evans, Department of Health Outcomes & Behavior, Tobacco Research and Intervention Program, Moffitt Cancer Center, 4115 East Fowler Avenue, Tampa, FL 33617, USA. E-mail:


A burgeoning literature suggests that attentional factors are associated with smoking behavior (e.g. direct nicotine effects and smoking withdrawal). This study examined differences in attentional processing between nonsmokers, satiated smokers and overnight nicotine-deprived smokers by comparing the amplitude of the P300 (P3) component of the event-related brain potential (ERP) elicited during a go–nogo task. We also examined the moderating effects of a common dopamine receptor genotype and state negative affect (SNA) on this ERP index of attention. Nonsmokers relative to smokers had greater nogo P3 amplitude. Carrying the A1 allele at the dopamine receptor D2 (DRD2) Taq1A polymorphism site moderated the effects of withdrawal on nogo P3 amplitude, suggesting the A1 allele is a vulnerability marker for withdrawal-related attentional deficits. Increased SNA also predicted attenuated P3 amplitude among deprived smokers. These findings suggest that DRD2 status and SNA moderate the effects of smoking status and withdrawal on neurocognitive variation during attentional processing. This research contributes to a better understanding of the role of individual differences and attentional processing in smoking behavior.

Nicotine acutely enhances attentional cognitive processing (Kumari et al. 2003), and nicotine abstinence decreases attentional functioning among smokers (Heishman et al. 1994). Individuals with attentional deficits appear to derive greater cognitive benefit from smoking/nicotine and therefore might be at greater risk of becoming dependent (Newhouse et al. 2004). Nicotinic self-medication of cognition is consistent with elevated smoking rates and therapeutic effects of nicotine across several neuropsychiatric conditions, such as schizophrenia and attention deficit disorder (e.g. Gehricke et al. 2006).

The present study examined behavioral and brain [event-related potential (ERP)] measures of cognitive-attentional control in relation to smoking status and withdrawal. Specifically, we examined the P300 (P3) ERP component elicited by standard go and rare nogo stimuli (Roche et al. 2005) among satiated and nicotine-deprived smokers as well as nonsmokers. Nogo P3 amplitude at frontal and posterior scalp sites is lowered among individuals who report greater daily attention-related difficulties (Roche et al. 2005). Correct withholding of responses on this task involves multiple attention-related processes (e.g. action monitoring, working memory) and activates attention-related brain structures [e.g. anterior cingulate cortex (ACC)]. Amplitude of the P3 family of ERP components is associated with increased allocation of attention and working memory (Polich 2007). Kamarajan et al. (2005) found reduced nogo P3 amplitude among individuals at risk for alcohol dependence. Nogo P3 has not been examined in relation to smoking, yet traditional oddball P3 is reduced among smokers (Anokhin et al. 2000), increased by acute smoking (Polich & Criado 2006) and decreased by smoking abstinence (Daurignac et al. 1998).

We also examined the dopamine receptor D2 (DRD2) Taq1A polymorphism and state negative affect (SNA) as potential moderators of the association between smoking/nicotine and cognition. Although evidence regarding the DRD2 Taq1A polymorphism and smoking behavior is mixed (e.g. Munafòet al. 2004), it will be informative to determine if this genotype moderates smoking-related effects on nogo P3. Generally, more reliable gene-behavior relationships may be established when objective cognitive and/or biological phenotypes are employed. Indeed, being a DRD2 Taq1A A1 allele carrier predicts smoking status among individuals with reduced oddball P3 (Anokhin et al. 1999) and other nicotine-induced and withdrawal-related attentional effects (e.g. Gilbert et al. 2005). Further rationale for investigating the DRD2 Taq1A genotype in this study include opposing effects of smoking/nicotine intake and withdrawal on acute dopamine release and depletion, respectively, the association of the A1 allele with reduced dopamine activity (e.g. fewer D2 receptors and reduced receptor binding, Jonsson et al. 1999), the association between D2 receptors and executive attention and frontal brain structures (e.g. ACC, striatum, see Lumme et al. 2007) and findings that nogo P3 is maximal at anterior sites and associated with dopamine (Beste et al. 2008). Regarding SNA, although negative affect has been associated with reduced P3 amplitude (Moser et al. 2005), these relationships have not been investigated as a function of smoking status and nicotine withdrawal. The present study will examine SNA, including withdrawal-related SNA, as a moderator of smoking/nicotine effects on attention (Evans & Drobes 2009).



The internal review board at the University of South Florida approved this study. Written informed consent was obtained from each participant. Eighty-eight participants (62 heavy smokers and 26 nonsmokers) were recruited from the Tampa Bay area, between 18 and 50 years. Inclusion criteria for the smokers included smoking an average of 15 or more cigarettes a day for at least the past year and not currently attempting to quit smoking. Criteria for the nonsmokers included having smoked at least 5 and no more than 100 lifetime cigarettes. The goal was to recruit nonsmokers who had experienced smoking, yet never became regular smokers. Participants were scheduled for a single 3- to 4-h experimental session that began between 0830 and 1100 h. Smokers agreed to participate before being randomly assigned to abstain from smoking overnight (deprived; minimum 8.5 h of nicotine deprivation; 10.5 h before the go–nogo task) or to continue smoking ad lib before the lab session. Smokers were informed that they would be given a breath test to verify compliance with these instructions. Carbon monoxide (CO) inclusion criteria were as follows: less than 5 parts per million (p.p.m.) for nonsmokers, greater than 8 p.p.m. for satiated smokers and 20 p.p.m. or less for deprived smokers. The CO upper limit for overnight deprivation was relatively inclusive, as baseline CO levels were not obtained and expected CO values because of half-life decay could not be computed on an individual basis. These criteria resulted in exclusion of one self-reported nonsmoker, one smoker assigned to the satiated group and two smokers assigned to the deprivation group. An urn randomization procedure (Muller et al. 2005) was used to balance the groups for gender, age and average number of cigarettes per day.


After arrival at the laboratory, participants provided informed consent. Next, participants completed the CO breath test to verify smoking status and to verify compliance with smoking instructions. An alcohol breathalyzer test was also conducted to verify no alcohol had been consumed that day. Next, buccal cells were collected for DNA extraction (e.g. Park et al. 1997). Satiated smokers smoked one of their own cigarettes before completing several questionnaires. All participants completed a demographic form (gender, ethnicity/race, age, medical history, current medication/drug use, education, income and occupation), and smokers completed the Fagerström Test for Nicotine Dependence (Heatherton et al. 1991), a widely used and reliable measure of smoking/nicotine addiction. Finally, smokers completed the anger, anxiety, sadness, concentration and craving subscales from the Wisconsin Smoking Withdrawal Scale (WSWS; Welsch et al. 1999). The WSWS full scale and subscales are highly reliable and valid in relation to the smoking withdrawal process (Welsch et al. 1999). Instructions were adapted to focus on withdrawal symptoms being experienced on the morning of the lab session. The anger, anxiety and sadness subscales were combined as an aggregate measure of SNA. Nonsmokers also completed the WSWS subscales, except for the craving subscale, to provide a basis for evaluating SNA as a moderator across all three groups. The WSWS sleep and appetite subscales were not administered, as they were not relevant to overnight deprivation. After completion of questionnaires, the 64-channel electroencephalogram (EEG) cap with silver chloride sintered electrodes was prepped for data collection.


Genomic DNA was extracted from buccal cells by proteinase K digestion and phenol extraction as described previously (Park et al. 1997). For genotyping of the DRD2 Taq1A polymorphism, previously described procedures (Lee et al. 2005) were used. Detailed information regarding this procedure is included in Appendix S1.

Go–nogo task

Participants completed a go–nogo task (Roche et al. 2005), which involved a series of ‘x’ or ‘y’ stimuli presented one at a time in the center of a computer monitor, each for 800 ms, with a 200-ms intertrial interval. Participants sat approximately 60 cm away from the monitor. Stimuli were approximately 2.5 cm in width presented in black font on a white background. Participants were instructed to use their dominant hand for pressing a response button each time the ‘x’ or ‘y’ was different from the letter presented on the preceding trial. Participants were instructed to respond before the letter was removed (i.e. 800 ms). Although the task requires relatively rapid responding (i.e. within 800 ms), accuracy was emphasized over speed, as responding too quickly increases the likelihood of perseverative responding to nogo trials. If the same stimulus was repeated across two trials then the participant was instructed to withhold responding (i.e. nogo, or lure, trials). Lures (repeated ‘X’ or ‘Y’) were placed between 4 and 28 trials apart, with an average of 11 trials apart. There were three blocks of trials, each having 300 items with a pseudorandomized order, resulting in a total of 900 trials that included 817 go trials and 83 nogo trials. Go and nogo trials were equally distributed across blocks. The task was approximately 25 min in duration. Participants completed three other cognitive tasks that are not reported here. E-prime version 1.0 experimental control software (Psychology Software Tools Inc., Pittsburgh, PA, USA) was used to program and administer the go–nogo task and was also used to trigger the continuous EEG recording for precise time locking of ERP waveform data to stimulus delivery.

ERPs: apparatus, recording and data reduction

The Neuroscan Synamps 2 system and its accompanying scan version 4.3.2 software system (Compumedics Neuroscan, Charlotte, NC, USA) were used to record the EEG data. The data were recorded at a sampling rate of 500 Hz from 64 electrodes that included the traditional 10–20 sites. Impedances were generally below 10 kilo-ohms. To monitor for eyeblink and other eye movement artifact, electrooculagram data were recorded from bipolar electrooculogram (EOG) electrodes placed above and below the left eye and lateral to each eye.

Band pass filtering at corner frequencies of 0.05 Hz (high pass) and 35 Hz (low pass) was applied offline. The continuous data were epoched into individual trials, ranging from a 100-ms prestimulus baseline to 800 ms following stimulus presentation. Epochs were baseline corrected offline. To minimize contamination of data from eyeblinks and other artifacts, epochs that included time-points with amplitudes greater than 90 microvolts at any of the EEG or EOG channels were rejected. Participants were excluded if they generated fewer than 20 artifact-free correct nogo trials. In addition to the four participants excluded for failure to meet CO criteria described earlier, this resulted in the exclusion of three nonsmokers, four satiated smokers and six deprived smokers for a final sample size of 71 participants. Mean artifact-free correct trials ranged from 44.6 to 45.7 for each of the groups and genotypes, with no significant differences as a function of smoking group, genotype or group × genotype (Ps > 0.58 for each effect). Artifact-free correct response epochs were averaged by stimulus type (i.e. correct go and correct nogo) and rereferenced from the vertex to the average of the two mastoid sites.

Statistical analyses


We used principal component analysis (PCA) as an alternative to traditional ERP measurement approaches, such as peak-picking and windowed amplitude measurements, for scoring ERP components. The rationale and a more detailed description of the specific PCA methodology are provided in Appendix S2. Briefly, PCA is a data reduction technique that takes into account covariance among variables to determine a more manageable smaller set of variables (i.e. principal components). We refer to components from the PCA as factors to avoid confusion with ERP components. Principal component analysis can be used to produce a smaller set of virtual electrodes (i.e. statistically derived electrodes weighted according to the covariation among actual electrode sites). Then, individual scores determined from this first step in the analysis can be used to follow-up with temporal PCA (Dien & Frishkoff 2005; Dien et al. 2004). A PCA toolbox program (Dien & Frishkoff 2005) was used to perform this type of spatiotemporal PCA on ERP data within the Matlab 7.0 program (Mathworks, Inc., Natick, MA, USA) environment. Rule N (Preisendorfer & Mobley 1988) was used to determine the number of factors to extract for both the initial spatial PCA and the subsequent temporal PCA. We first conducted a PCA of the covariance matrix with the 71 subject average waveforms for go and nogo trials computed separately at each of the 451 time-points (one data point per 2 ms including 100 ms prestimulus baseline) as rows and the 64 EEG scalp sites as columns/variables to identify spatial factors/virtual electrodes. This resulted in a PCA with 64 042 rows (71 subjects × 2 trial types × 451 time-points) and 64 variables/columns. Next, using all the factor scores from the first step as input, we computed a subsequent temporal PCA to determine ERP components and to generate spatiotemporal factor scores. This latter analysis involved analyzing the spatial factor scores from 71 participants on go and nogo trials at each of the spatial factors as 720 rows (i.e. 71 × 2 trial types × 5 spatial factors extracted, see Results) and the 451 time-points as columns/variables in the temporal PCA. The P3 temporal factor was then selected based on identifying the factor with highest loadings spanning the P3 time window observed in the raw waveform across electrodes. Spatiotemporal factor scores indicative of the P3 time window for each trial type at each substantive spatial factor (i.e. not noise factors) were then retained as the dependent variable.

As an ancillary approach, peak amplitude measurement at traditional midline (10–20 system) scalp sites was performed to verify that PCA-derived factor scores show expected moderate to high levels of correlations with traditional measures and to examine the consistency of smoking-related findings across methods. The most positive amplitude time-point occurring in the 250–800 ms time window was used to score P3 at Fz, Cz and Pz electrodes. A more detailed description of the traditional amplitude measurement is included in Appendix S3.

Behavioral data

Accuracy (go and nogo trials separately) and correct response reaction time data were analyzed among all participants included in the final ERP analyses. Correct go responses likely occur amid minor lapses in attention, as the rapid presentation format and infrequency of nogo trials result in robust prepotent responding that can become rhythmic and automatic. Nogo trials are more difficult to perform accurately, assuming one is responding accurately to go trials, so it was possible that smoking relevant interactions indicating greater nogo relative to go accuracy (i.e. trial type as a factor predicting accuracy) might be indicative of superior performance. Therefore, all responses to go trials occurring after presentation of the stimulus and before removal of the stimulus (800 ms) were treated as accurate.

Mixed modeling

Mixed modeling, including hierarchical linear modeling (HLM; see Tabachnick & Fidell 2007), was used to examine study predictions. This approach allowed us to model SNA as a continuous predictor, as well as potential interactions between SNA and other independent variables, including variables at other hierarchical levels. State negative affect was centered (i.e. converted to deviation scores) to prevent unnecessary multicollinearity resulting from the same continuous predictor being repeated in different effects (Tabachnick & Fidell 2007). Dependent variables included behavioral accuracy and reaction time, spatiotemporal ERP factor scores indicative of P3 and traditional peak amplitude measures as an ancillary measure. All dependent scores were age and sex adjusted. This was accomplished using regression models, with age predicting each dependent variable followed by retaining the unstandardized residuals, thereby eliminating shared variance between the dependent variables and the age. These residuals were obtained through separate regressions for men and women. The accuracy testing included go and nogo trials as two levels of the same factor as well as separate testing of go and nogo accuracy as dependent variables. Level 2 predictors for response accuracy included smoking group (nonsmoker, satiated smoker and deprived smoker) and genotype (A2A2 vs. A1 carrier) as categorical predictors and SNA as a continuous predictor. Trial (go vs. nogo) was nested within subjects as a level 1 predictor. The additional testing for accuracy with separate testing for go and nogo trials included the same predictors as the level 2 predictors from the accuracy model, but trial type was no longer a predictor, so only included one level. This was also the case for reaction time analysis, as accurate reaction times pertained to go trials only. For ERP analysis, the level 3 (subjects) predictors were smoking group and genotype as categorical predictors and SNA as a continuous predictor. Scalp location was nested within subjects as a level 2 predictor and trial type nested within location as a level 1 predictor.

We were interested in smoking group × trial fixed effects to test smoking and withdrawal status on accuracy and in smoking group ×DRD2 genotype × trial and smoking group × negative affect × trial interactions to examine moderators. The same effects were of interest in predicting reaction time except that there was no trial type predictor (i.e. only go trials). For ERP analyses, we were first interested in trial (go vs. nogo), location (virtual electrode) and trial × location effects as manipulation checks regarding the experimental paradigm and P3 amplitude. Then, the smoking-related effects of interest included (1) smoking group × trial (and group × trial × location) to determine if there were P3 differences among groups, (2) smoking group × genotype × trial (and group × genotype × trial × location) to examine the moderating effects of the DRD2 Taq1a polymorphism and (3) smoking group × SNA × trial (and group × SNA × trial × location) to examine the moderating effects of SNA.


Sample characteristics and group comparisons

Table 1 displays demographic and smoking characteristics across the three groups for the final sample. There were no significant differences across groups for age, race, gender or level of education attained (all > 0.05). There were also no differences in smoking characteristics between the two smoker groups, with the exception of CO level and withdrawal, consistent with the deprivation manipulation. Mean CO levels upon arrival to the lab were 2.2 (SD = 0.8) p.p.m. for nonsmokers, 24.5 (SD = 10.8) p.p.m. for satiated smokers and 10.5 (SD = 5.2) p.p.m. for deprived smokers. Seventy per cent of deprived participants had CO of 12 p.p.m. or less, and 92% of satiated participants had CO > 12 p.p.m. There were significant CO differences between nonsmokers and satiated smokers [t(25.3) = 10.46, < 0.001; equal variances not assumed] and between satiated and deprived smokers [t(47) = 5.64, < 0.001]. During the 2 h elapsed between CO measures and performance of the nogo task, deprived smokers did not smoke, but satiated smokers smoked two cigarettes, including one 15 min before performing the nogo task. As expected, the deprived group scored higher on the WSWS overall withdrawal score [t(47) = 1.74, < 0.05, one-tailed test] as well as the WSWS craving subscale [t(47) = 3.66, = 0.001].

Table 1.  Demographic and smoking-related participant characteristics
  • Values in parentheses indicate SD.

  • FTND, Fagerström Test for Nicotine Dependence.

  • *

    < 0.05 for all possible group comparisons.

% Caucasian959291
Gender (% male)415839
Age25.0 (7.2)30.2 (10.6)29.6 (9.0)
CO p.p.m.*2.2 (0.8)24.5 (10.8)10.5 (5.2)
Cigarettes per day20.1 (5.3)22.3 (5.8)
Years smoking11.2 (9.9)12.9 (9.6)
FTND4.3 (1.8)5.3 (2.0)
WSWS*1.4 (0.8)1.8 (0.8)

Genotypic/allelic characteristics

There were 40 A2A2, 28 A1A2 and 3 A1A1 genotypes. A binary code for scoring the genotypes as either A1+ (A1A1, A1A2) or A1− (A2A2) was used, as is typical in the literature. The genotype distribution of the DRD2 Taq1A polymorphism among the study population was consistent with the Hardy–Weinberg equilibrium (= 0.88). Among nonsmokers, there were 11 (4 male and 7 female) A1+ and 11 (9 female and 2 male) A1− genotypes; among satiated smokers, there were 13 (6 male and 7 female) A1+ and 13 (9 male and 4 female) A1− genotypes and among deprived smokers, there were 7 (5 male and 2 female) A1+ and 16 (4 male and 12 female) A1− genotypes. Chi-square testing did not find any significant difference in genotype across the smoking groups. Being an A1 carrier has sometimes been associated with smoking status (Munafòet al. 2004), but we were substantially underpowered to test this hypothesis. In addition, genotype and genotype × group effects did not approach significance with respect to predicting SNA or withdrawal.

Go–nogo performance

There was one extreme outlier on accuracy to go trials. This participant responded accurately to 656 (80% correct) go trials, substantially less than the next lowest performer at 769 (94% correct). This participant was therefore removed from accuracy analysis but was retained for reaction time and ERP, as only accurate trials were included in these latter analyses and the participant clearly performed above chance. Go and nogo accuracy and reaction time to go trials were normally distributed. Accuracy to go trials ranged from 94.1% to 100%, with a mean of 98.2% (SD = 1.5%). Accuracy to nogo trials ranged from 32.5% to 96.4%, with a mean 68.6% (SD = 14.8%). Individual mean reaction time to go trials ranged from 240 to 449 ms, with an overall mean of 331 ms (SD = 44 ms). There were no significant smoking-related effects involving SNA, genotype, group or trial type in predicting reaction time or accuracy.

Go–nogo ERP

The grand averaged ERP waveforms for go and nogo trials at midline electrode sites (Fz, Cz and Pz) as well as sites to the left and right of these midline sites (i.e. F3, F4, C3, C4, P3 and P4) are presented in Fig. 1. [A smoothing function (i.e. filter) was implemented to the graphs in Fig. 1. This did not change the morphology of the waveform except for removing 60 Hz noise.] For the PCA analysis, the above mentioned rule N criterion for factor extraction resulted in five spatial factors, followed by temporal PCA of the complete set of spatial factor scores, which resulted in 11 temporal factors. Three spatial factors/virtual electrodes with marked frontal, central and posterior scalp topographies were of greatest interest and are presented in Fig. 2. The other two spatial factors were indicative of a mastoid reference effect and a noise factor.

Figure 1.

ERP waveforms for averaged correct go and nogo trials across the 0- to 800-ms epoch at nine electrode sites. Go trials are shown as thin lines and nogo trials as bold lines. Consistent with the standard 10–20 electrode montage, F3, left frontal; Fz, medial frontal; F4, right frontal; C3, left central; Cz, medial central; C4, right central; P3, left parietal; Pz, medial parietal; and P4, right parietal scalp sites.

Figure 2.

Frontal (factor 1), posterior (factor 2) and central (factor 3) spatial factor unstandardized loadings derived from a covariance-based PCA of the 64 electrode sites across all subjects. The legend to the left of each head map indicates factor loadings.

We targeted a single temporal factor, that is the factor indicative of the P3 time window. Temporal factor 2 from the subsequent temporal PCA is clearly indicative of a large positive going wave occurring within the P3 time window. Figure 3 shows the temporal factor loadings for the P3 component and shows the 250–800 ms window as having higher loadings (peak amplitude at 428 ms). The other temporal factors represent earlier occurring ERP components as well as a number of noise factors. We were interested in three sets of spatiotemporal factor scores: the frontal (spatial factor 1), central (spatial factor 3) and posterior (spatial factor 2) spatial factors associated with the P3 (i.e. temporal factor 2). We will, henceforth, refer to factor scores as indicative of frontal P3, central P3 or posterior P3 amplitudes, rather than by order of factor extraction. Spatiotemporal factor scores were sex and age adjusted as described above. Factor scores indicative of go P3 were subtracted from nogo scores indicative of P3 at each spatial location to obtain correlations among locations. Factor scores indicative of frontal P3 were marginally correlated with central P3 (= 0.16, < 0.10, one-tailed test), as was the case with frontal and posterior P3 scores (= 0.17, < 0.10, one-tailed test). Central and posterior scores were positively correlated (= 0.55, < 0.0001). Nogo minus go traditional peak amplitude P3 measures were more highly intercorrelated (Fz and Cz, = 0.78; Fz and Pz, = 0.58 and Cz and Pz, = 0.89). Previous research using this go–nogo task did not use a PCA approach to scoring the P3 (Roche et al. 2005). However, there was substantial overlap between our virtual electrodes and prior analyses of single electrode sites. To verify that our approach shows reasonable level of overlap with traditional amplitude measures, we also correlated the nogo minus go traditional peak amplitude measures with factor scores. Frontal spatiotemporal P3 factor scores correlated strongly with Fz (= 0.71), central scores with Cz (= 0.72) and posterior with both Pz (= 0.62) and Cz (= 0.65).

Figure 3.

Temporal factor 2 unstandardized loadings. These loadings are indicative of the P3 component shown across the 0- to 800-ms epoch.


The dependent variable generated from the spatiotemporal PCA included six scores for each participant (i.e. two trial types × three spatiotemporal factor scores associated with the nonnoise spatial factors). The omnibus HLM model resulted in a significant effect for trial, with nogo trials producing greater spatiotemporal factor scores indicative of P3 amplitude, F(1,213) = 419.56, < 0.001. As expected, the location × trial interaction was also significant, F(2,213) = 7.86, = 0.001, with the frontal location showing greatest and the posterior the lowest differentiation between go and nogo trials. These location × trial effects are consistent with previous research involving this go–nogo paradigm (Roche et al. 2005).

Using the same omnibus model, there were significant smoking group × trial [F(2,213) = 5.27, = 0.006], smoking group ×DRD2 genotype × trial [F(2,213) = 6.40, = 0.002] and smoking group × SNA × trial [F(2,213) = 4.18, = 0.02] effects. More specifically, smoking group predicted nogo relative to go P3 amplitude as well as showing DRD2 genotype and SNA as moderators of this effect. Although there were no significant higher order interactions that included location, different scalp sites typically generate P3-type waveforms from relatively independent brain sources, which reflect different, although related types of cognitive processing (e.g. Dien et al. 2004; Polich 2007). We therefore elected to follow-up the significant interactions from the omnibus model at each virtual electrode to establish which spatial P3 components are sensitive to subject characteristics. However, because there was no interaction with location and smoking-related effects, it is important to note that the simple effects analyses we report as significant at specific locations were also significant without differentiating locations, thereby verifying that the effects in our follow-up analyses at each location are the same as those found in the omnibus model, except that the effects were significant only at select locations in the analyses included differentiating location. Therefore, as more elaborately addressed in the discussion, the pattern of findings at each specific location may also apply to a more general P3 that is not broken down by location. At the frontal virtual electrode, the interactions involving smoking group, DRD2 genotype and SNA did not approach statistical significance.

Central virtual electrode

At the central virtual electrode, the smoking group × trial interaction was significant [F(2,71) = 4.45, = 0.01]. Figure 4 indicates that nonsmokers have greater differentiation between nogo and go trials in comparison to both smoker groups. Indeed, follow-up t tests confirmed that nonsmokers had greater P3 difference scores (nogo minus go age- and sex-adjusted factor scores) than both satiated [t(46) = 2.32, = 0.02] and deprived [t(43) = 2.21, = 0.03] smokers.

Figure 4.

Nonsmokers show age- and sex-adjusted spatiotemporal factor scores indicative of greater central nogo P3 (relative to go) amplitude than both satiated and deprived smokers. Black bars indicate go trials and gray bars indicate nogo trials. Whiskers above bars indicate 1 SE.

In addition, the smoking group ×DRD2 genotype × trial interaction was significant at the central virtual electrode [F(2,71) = 3.36., = 0.04]. Figure 5 displays the DRD2 genotype by trial interaction for each group. Deprived A1+ smokers exhibited less differentiation between nogo and go trials compared with deprived A1− smokers [t(21) = 2.42, = 0.03]. Although the deprived A1+ cell was the smallest (= 7), six of the seven deprived A1+ central P3 factor scores were below the median for deprived group, suggesting that this effect was not unduly weighted by a smaller subset of scores. A trend in the opposite direction emerged among satiated smokers [t(24) = 1.50, = 0.15]. Nonsmokers did not differ as a function of DRD2 genotype.

Figure 5.

Smoking group × genotype × trial type predicts age- and sex-adjusted spatiotemporal factor scores indicative of central P3. Panel A, nonsmokers; panel B, satiated smokers and panel C, deprived smokers. Left bars indicate factor scores for A1+ and right bars indicate factors scores for the A1−. Black bars indicate go trials and gray bars indicate nogo trials. Whiskers above bar indicate 1 SE.

Posterior virtual electrode

At the posterior virtual electrode, the smoking group ×DRD2 genotype × trial interaction was significant [F(2,71) = 3.37, = 0.04]. As shown in Fig. 6, the nature of this interaction was similar to that found at the central virtual electrode. Satiated A1+ smokers showed greater differentiation in P3 amplitude between nogo and go trials compared with satiated A1− smokers [t(24) = 2.36, = 0.03]. In contrast, deprived A1+ smokers showed a trend for less nogo relative to go P3 amplitude compared with deprived A1− smokers [t(21) = 1.66, = 0.11]. Although the A1+ cell was the smallest, all seven of these A1+ participants scored below the median for deprived smokers (< 0.01). Again, nonsmokers did not differ according to A1 status.

Figure 6.

Smoking group × genotype × trial type predicts age- and sex-adjusted spatiotemporal factor scores indicative of posterior P3. Panel A, nonsmokers; panel B, satiated smokers and panel C, deprived smokers. Left bars indicate factors scores for A1+ and right bars indicate factors scores for the A1−. Black bars indicate go trials and gray bars indicate nogo trials. Whiskers above bar indicate 1 SE.

The smoking group × SNA × trial interaction was significant at the posterior virtual electrode [F(2,71) = 3.39, = 0.04]. For ease of interpretation, Fig. 7 shows this interaction based on a median split for the continuous SNA variable. This effect was driven largely by the deprived group, with less differentiation according to trial type among deprived smokers with higher levels of SNA. Using mixed modeling, the slope of the association between negative affect and P3 factor score difference (nogo minus go) scores was significantly more negative in the deprived group relative to both the satiated [t(71) = 2.06, = 0.04] and the nonsmoker [t(71) = 2.43, = 0.02] groups.

Figure 7.

Smoking group × SNA × trial type predicts age- and sex-adjusted spatiotemporal factor scores indicative of posterior P3. Panel A, satiated smokers and panel B, deprived smokers. Left bars indicate factor scores among lower SNA smokers and right bars indicate factor scores among higher SNA smokers. Black bars indicate go trials and gray bars indicate nogo trials. Whiskers above bar indicate 1 SE.

Traditional peak amplitude

The traditional amplitude approach resulted in the same effects as the PCA generated dependent measure in the omnibus model, with one notable exception. The group × trial type × genotype interaction did not approach significance, although the pattern of means was in the same direction as with the PCA analyses and subsequent simple effect tests. A more detailed description is provided in Appendix S3.


The primary goal of the study was to compare nonsmokers, satiated smokers and overnight-deprived smokers on behavioral and brain (ERP) measures of cognitive-attentional processing. We also examined a dopamine genotype and SNA as potential moderators of smoking/nicotine-related effects on cognition. We did not find behavioral differences between nonsmokers, satiated smokers and deprived smokers. Studies using similar attention-related tasks have found differences in performance as a function of nicotine abstinence (e.g. Hendricks et al. 2006). Nevertheless, it is common to find ERP group differences evoked by basic cognitive-attention tasks that do not show up at the behavioral level (e.g. Roche et al. 2005). Equivalent behavioral performance on simple experimental tasks may involve compensatory mechanisms (e.g. greater effort) that mask cognitive deficits that are evident using physiological indices of neural processing (Evans & Drobes 2009).

Null location effects were found for all smoking-related effects. It is possible that the more posterior and central sites were driving the P3 effect and that the study was underpowered to detect more complex interactions that included location. Alternatively, null location effects may indicate that a substrate common to the various P3 types is driving the effect. Indeed, each of the simpler effects found at specific virtual electrodes were also present without differentiating by location. Correlations among spatiotemporal factor scores were positive and modest (ranging = 0.16–0.55), indicating a common substrate across locations. These correlations were substantially less than the correlations among traditional amplitude measures at midline sites (range = 0.58–0.89).

Event-related potential brain source localization studies have found both unique and shared neural substrates among the various P3 components, suggesting that either global P3 and/or more specific P3-type effects may be related to smoking behavior. For example, the ACC, a region heavily implicated in overall attentional control, has been identified as a source for P3 components that emphasize both posterior (e.g. Iwaki et al. 2007) and anterior (Fallgatter et al. 2002) scalp distributions. Polich (2007; Polich & Criado 2006) suggests direct neural linkages between frontal and posterior–central P3 activity and that the broader function of P3 may generally involve inhibitory modulation of extraneous stimuli, thereby increasing the probability of performing well on tasks requiring attentional control and concomitant working memory. Thus, it may be that our null effects regarding location may reflect common features across the P3 components. Nicotine affects a diverse range of attentional control processes (Evans & Drobes 2009), so detection of a more global effect seems plausible.

P3 differences

The finding that nonsmokers relative to smokers had greater nogo P3 amplitude is consistent with Anokhin et al.’s (2000) oddball P3/P3b findings and extends findings regarding an association between risk for alcohol dependence and reduced nogo P3 (Kamarajan et al. 2005). Anokhin et al. (2000) suggested that smokers who have reduced P3 may find it more difficult to quit, thereby resulting in lower P3 amplitude among the pool of smokers who have not successfully quit. Lowered P3 is associated with a variety of externalizing/impulsivity-related behavior (Iacono et al. 2003), which indeed may impede successful cessation.

A number of cognitive processes might be associated with the nogo P3 evoked by this task, including response inhibition, conflict detection and working memory. However, these processes require attention. Response inhibition on the nogo task requires sustained attentional control to consciously override and thereby inhibit prepotent go response tendencies. The present nogo task can also be conceptualized as a one-back working memory task and P3 is also associated with updating of working memory. Contemporary conceptualizations of working memory emphasize the importance of executive attention (Baddeley 2003). Because the family of P3 components is associated with various facets of attention that contribute to overall executive attentional control, it may be that smokers’ reduced P3 reflects attentional deficits and may serve as a marker for smoking to self-medicate attentional dysfunction (Newhouse et al. 2004). One substrate of this effect may be ACC function, an area with complex reciprocal connectivity with multiple brain areas. Nicotine increases ACC activity during executive attention tasks (e.g. Kumari et al. 2003), and as noted, this area has been identified as a source across P3-type components.

Genetic moderation of P3

Similar to effects of smoking status and withdrawal discussed above, nogo P3 was moderated by the DRD2 genotype at central and posterior sites but not at frontal sites. In particular, satiated A1+ smokers had greater nogo P3 amplitude relative to satiated A1− smokers at the posterior site, with a trend for this effect at the central site. In contrast, deprived A1+ smokers had decreased nogo P3 relative to A1− smokers at the central site, with a trend for this effect at the posterior site. The P3b (traditional oddball P3) is maximal at posterior and central sites. Thus, the observation of smoking related and genetic influences on nogo P3 at relatively posterior scalp sites bears resemblance to Anokhin et al.’s (1999) results linking the traditional oddball P3b (at parietal electrode) to smoking status and DRD2 genotype. These authors found that being a DRD2 Taq1A A1 allele carrier predicted smoking status among individuals with reduced oddball P3.

Lower P3b amplitude has been previously associated with smoking deprivation (Daurignac et al. 1998). Deprivation-induced nogo P3 deficit at the posterior–central sites among A1+ is also consistent with findings concerning the DRD2 Taq1A polymorphism and the P3. First, the posterior–central P3b component has been associated with dopamine activity and the DRD2 Taq1A polymorphism (Hill et al. 1998). Second, the A1 allele may be associated with lower dopamine activity through fewer D2 receptors and reduced receptor binding in frontal structures, including the ACC (Jonsson et al. 1999). In contrast, smoking is associated with dopamine release, particularly among individuals with genes that may indicate lower dopamine activity (Brody et al. 2006). Thus, smoking might normalize neurocognitive indices of attention among A1+, which is consistent with the proposition that nicotine might be especially reinforcing among individuals with attentional deficits (Newhouse et al. 2004). In addition to the ACC as a common source across P3 components, this area is dopamine rich and plays a major role in attentional control. Importantly, executive attention is modulated by D2 receptors in the ACC region (Lumme et al. 2007).

Among satiated smokers, nogo P3 was maximized for A1+ at posterior and central sites. It is interesting to speculate that either dopamine deficits (deprived A1+) or excesses (satiated A1−) may be associated with cognitive-attentional dysfunction. Indeed, a D2 dopamine antagonist has been found to increase P3 amplitude among individuals with reduced P3 and was found to decrease amplitude among individuals with greater P3 (Takeshita & Ogura 1994), suggesting that optimization of dopamine activity may play a role in P3 amplitude. The present findings are consistent with an optimization of dopamine level among A1 carriers under conditions of nicotine satiation or those without an A1 allele who are nicotine deprived.

SNA moderation of P3

Among deprived smokers, the nogo P3 effect was reduced at the posterior site as a function of increased SNA, whereas SNA did not influence nogo P3 for satiated smokers or nonsmokers. As noted above, P3 at the posterior virtual electrode may reflect P3b-related activity. Consistent with findings that show a negative association between temperamental negative affect and attentional control (Evans & Rothbart 2007), this finding suggests that P3-related attention is reduced in the presence of withdrawal-related SNA. Smoking as self-medication for either SNA or cognitive deficits has been the target of a great deal of research, but the relationship between affective and cognitive domains has not received much attention in the context of smoking research (Evans & Drobes 2009).

Limitations and future directions

The present study included a number of design limitations. P3 scores were corrected for age and sex, but other confounds such as psychopathology were not controlled. Future research might provide more detailed matching of smokers with nonsmokers. Further, it would be optimal to lengthen the required deprivation period to at least 12 h. To better differentiate P3 effects, future research might include a go–nogo task with 50% go and 50% nogo trials along with a separate oddball task to evoke both P3a (frontal) and P3b (posterior), thereby enabling finer-grained testing of specific P3 contributions to smoking behavior. An additional improvement would be to increase the number of overnight-deprived A1+ participants.

It is our hope that the current findings contribute to a better understanding of important individual differences that moderate links between cognition and smoking. Improved detection of people who display greater cognitive improvements from smoking/nicotine (or greater cognitive disruptions during withdrawal) could lead to the development of programs tailored to treat smokers who find smoking more reinforcing for cognitive reasons.


We thank Heather Jim for providing helpful statistical advice. We also thank Illia Deleon and Jane Weigl for assistance with collecting and processing data. This research was funded by the Moffitt Cancer Center.