An fMRI Investigation of Alcohol Cue Reactivity: Test-retest Reliability and Link to Real-Life Drinking
Lauren Kuhns University of Amsterdam
Abstract
Cue reactivity paradigms are an integral methodological tool for studying the neural underpinnings of pathological drug and alcohol use. Functional neuroimaging research has demonstrated that alcohol cues elicit heightened activation in key reward-related areas in the brain, including the ventral striatum, anterior cingulate cortex, and ventromedial prefrontal cortex. Despite the prevalence of the paradigm, little is known about the test-retest reliability of neural alcohol cue reactivity. In addition, due to the use of self-report measures of alcohol consumption, which research suggests are subject to systematic biases, the relationship between neural alcohol cue reactivity and actual alcohol consumption remains unclear. The role of social processes in alcohol effects and conditioning have also been largely overlooked within cue reactivity research, despite decades of research suggesting that social motives are the most common reason for alcohol consumption and that social and alcohol effects interact and mutually contribute to each other. To address these gaps, we created a novel cue reactivity paradigm that incorporates social contexts and investigated the test-retest reliability of neural alcohol cue reactivity and its relationship to ad libitum alcohol consumption in a social drinking session. Results suggest that neural alcohol cue-reactivity elicited by the task was unstable over time in a
priori reward-related regions of interest—the ventral striatum, anterior cingulate cortex, and
ventromedial prefrontal cortex. In addition, neural alcohol cue reactivity in these regions did not predict alcohol consumption in a social drinking session. The implication of these results for our understanding of the test-retest reliability of neural alcohol cue reactivity, the relationship between cue reactivity and drinking behavior, and the role of social processes is discussed.
Introduction
Cue-reactivity paradigms have been an integral methodological approach to the study of addiction for decades, and a growing number of studies use neuroimaging techniques to
investigate the neural underpinnings of pathological alcohol use. Theoretically, cue-reactivity paradigms aim to measure the conditioned responses that alcohol-related stimuli come to elicit through continual pairings with alcohol effects over time (Robinson & Berridge, 2000). The incentive salience account of addiction argues that the repeated association between the intoxicating effects of alcohol and alcohol-related cues—ranging from physical materials
(alcohol bottles) to emotional states to situational contexts—leads to sensitization of the brain to these cues. The sensitized response to these conditioned cues is then believed to continue or increase the use of alcohol, even in spite of a person’s conscious goal or desire to abstain from or limit alcohol consumption. In line with this account, elevated responses to alcohol cues are associated with a variety of response patterns that are believed to contribute to or maintain alcohol use disorder (AUD) such as attentional bias (Völlstadt-Klein et al., 2012; Townshend & Duka, 2001), approach tendencies (Field, Mogg, & Bradley, 2005), and increased cravings (Monti et al., 1987).
On a neurobiological level, enhanced dopamine response to alcohol cues compared to other rewards (e.g. food) is believed to underlie the increased incentive salience of drug and alcohol cues (Nestler, 2005). In line with this, studies of cue reactivity have demonstrated increased brain activation to cues across a wide range of addictive substances in regions
constituting the mesocorticolimbic system (a key dopamine pathway) that is commonly referred to as the reward system of the brain (see Jasinska, Stein, Kaiser, Naumer, & Yalachov, 2014, for a review). According to a systematic review and quantitative meta-analysis of 28 alcohol cue
reactivity studies, alcohol cues elicit robust activations in the ventral striatum (VS), anterior cingulate cortex (ACC), and the ventromedial prefrontal cortex (vmPFC) in participants with AUD (Schacht, Anton, & Myrick, 2013). Among other processes, the VS is implicated in Pavlovian conditioning and predicting future rewards (Everitt & Robbins, 2005). The ACC plays an important role in coding of motivational value of events (Ridderinkhof, Ullsperger, Crone, & Nieuwenhuis, 2004). The vmPFC is implicated in the representation of hedonic properties of reward, especially in action selection processes that are influenced by emotion (Ridderinkhof, Van Den Wildenberg, Segalowitz, & Carter, 2004). Thus, elevated cue-reactivity to alcohol cues might be most robustly seen in brain regions involved in motivational and reward processing.
Cue-reactivity research has also been used to shed light on clinical outcomes of alcohol abuse and addiction. In a large study of neurobiological phenotypes associated with alcohol use disorders (AUD), alcohol use severity in heavy drinkers was positively correlated with cue-induced reactivity in reward areas of the brain, including the nucleus accumbens (a component of the VS), precuneus, insula, and dorsal striatum (Claus Ewing, Filbey, Sibbineni, & Hutchison, 2011). Furthermore, multiple studies report changes in alcohol cue reactivity following treatment for patients with AUD. In one such study, patients with AUD had decreased alcohol cue
reactivity in the amygdala, cerebellum, and hippocampus after a psychopharmacological treatment compared to before treatment (Schneider et al., 2001). In another study, patients exhibited decreased alcohol cue reactivity in the VS after an extinction treatment compared to a control group of non-treated patients (Vollstädt-Klein, 2011). These studies suggest that certain treatments for AUD reduce activation to alcohol cues in reward areas of the brain. Moving a step further, Grüsser and colleagues found that cue-induced reactivity in the ACC and medial
those who remained abstinent (2004). Further analyses showed that the amount of alcohol consumption during the relapse period was only associated with cue reactivity in these regions, and not self-reported levels of craving, past alcohol consumption history, or duration of
abstinence, suggesting that cue reactivity might provide valuable information about the success of treatment above and beyond any self-report measures of craving and alcohol use history.
Heightened responses to alcohol cues are not limited solely to clinical samples, however. Sub-clinical heavy drinkers also show an amplified response in emotional areas such as the insular cortex and the VS compared to light drinkers (Ihssen, Cox, Widget, Fadardi, & Linden, 2010). In addition, in a longitudinal study of American college students, BOLD responses to alcohol cues compared to neutral cues in the bilateral caudate, orbitofrontal cortex, ACC, and left insula were elevated in moderate drinkers who later escalated to heavy drinking (Dager et al., 2014). Furthermore, compared with other baseline risk factors such as impulsivity and family history of substance abuse, the BOLD response to alcohol cues was the best predictor of increased drinking in the future in currently moderate drinkers (Dager et al., 2014). These findings suggest that elevated alcohol cue-reactivity is not limited to clinical samples, and the relevance of the paradigm extends to moderate and heavy social drinkers as well.
Given the clinical relevance and the potentially additive information cue reactivity paradigms can add to our understanding of both the outcome of AUD and the drinking patterns of non-clinical social drinkers, it is critically important to have a foundational methodological understanding of the paradigm. An often-overlooked gap in the literature is the reliability of cue reactivity over time, which is especially important as the paradigm is used in test-retest protocols for AUD treatment. Without a basic understanding of the reliability of neural cue reactivity as a measure, we are fundamentally limited in our ability to interpret the findings of past and future
studies that use this paradigm. To our knowledge only one study has employed multiple fMRI scans over time to assess the test-retest reliability of cue reactivity. Schacht and colleagues looked at the stability of cue-elicited activation in the VS and DS of ten alcoholics who were not seeking treatment across two fMRI scans separated by two weeks, and found that cue activation was largely stable in these regions (2011). However, more research with larger sample sizes and varied populations are still needed for a more complete understanding of the reliability of the measure.
Furthermore, most of the studies examining the relationship between cue reactivity and alcohol use rely on retrospective self-report measures of alcohol consumption and patterns of use and abuse. These types of measurements are valuable tools in the effort to understand the
relationship between cue reactivity and alcohol consumption. However, several factors—such as social desirability, memory errors, and the use of cognitive heuristics—suggest that retrospective self-report measures of alcohol use are subject to systematic biases (Del Boca & Darkes, 2003; Davis, Thake, & Vilhena, 2010). For instance, people with higher levels of impression
management bias—defined as a tendency to self-attribute virtuous characteristics and deny socially deviant impulses or behavior—report 20 to 33% less consumption of alcohol and are 50% less likely to report risky drinking (Davis et al., 2010). Importantly, the rate of inaccurate reporting of alcohol consumption due to memory errors and behavioral estimation strategies is likely to vary depending on the frequency and intensity of drinking episodes (Del Boca and Darkes, 2003). These biases represent a significant threat to the validity of self-reported alcohol use, and therefore to our understanding of the relationship between the intensity of alcohol use and cue-reactivity.
The neurobiological focus of cue reactivity research has provided us with valuable insights into the neural correlates of alcohol rewards and the mechanisms of craving. However, parallel lines of research into the influence of social factors on alcohol consumption suggests that social contexts play an important role in the motivations to consume and rewarding aspects of alcohol. In a systematic review of research into drinking motives of young adults and
adolescents, Kuntsche and colleagues concluded that most young people report drinking due to social motives, and that these motives are associated with moderate levels of drinking (2008). Although social motives are often found to be less strongly associated with alcohol abuse and addiction than other drinking motives (e.g. coping or enhancement), social motives are powerful predictors of the frequency of alcohol consumption (Cooper, 1994) and the intensity of
consumption (Oostveen, Knibbe, & De Vries, 1996). Social motives are also associated with certain negative outcomes, such as driving while under the influence and campus violations in college students (Beck et al., 2008). These findings suggest that, in young people especially, social factors play a significant role in both the decision to drink and the outcomes of drinking behavior.
Not only does previous research indicate that social factors are important in drinking motivations in humans, animal studies suggest that the rewarding aspects of social situations and alcohol may interact and mutually contribute to each other. For instance, in rats, conditioned alcohol aversions can be attenuated via dyadic social interactions (Gauvin, 1994), social interaction can increase the responsiveness and sensitivity to alcohol (Varlinskaya, Spear, & Spear, 2001), and combining social interactions and alcohol increases alcohol intake (Tomie, Uveges, Burger, Patterson-Buckendahl, & Pohorecky, 2004). These findings suggest that the interaction of social and alcohol rewards may increase the reinforcement value of each, and
potentially lead to stronger conditioning effects. These findings are particularly important given that many formative experiences with alcohol in young adulthood and adolescence take place in social contexts, potentially leading to enhanced conditioning of alcohol-related cues.
Because of the demonstrated prevalence of social motivations for drinking and animal research suggesting that social contexts increase the rewarding effects of alcohol, incorporating social contexts in the cue reactivity paradigm may lead to more robust reward-area activation, especially in non-clinical social drinkers. Therefore, we created a novel multi-modal cue-reactivity task that includes visual and auditory social stimuli along with alcohol cues to begin investigating the aforementioned gaps in the literature surrounding the reliability and validity of cue reactivity as a measure. Thus, the main goals of the present study were, 1) to assess the test-retest reliability of alcohol cue reactivity over time, and 2) to investigate the relationship between cue reactivity and alcohol consumption in an ad libitum social drinking session. Alcohol cue reactivity was measured with two fMRI scans over the period of one week in a sample of
drinkers. In line with a recent systematic review of human neuroimaging studies of cue reactivity suggesting that multi-sensory substance-stimuli elicit more robust brain activation in reward-related areas compared to uni-sensory cues (Yalachov, Kaiser & Naumer, 2012), an audio-visual cue reactivity task with social contexts was used to elicit cue reactivity in the scanner. Alcohol consumption was assessed covertly in a social drinking session at the conclusion of the scanning protocol, during which participants could choose to consume alcohol ad libitum. Finally,
behavioral measures of impulsivity and inhibitory control were collected as potential covariates based on previous research suggesting that individual differences in these measures are related to alcohol cue reactivity (Papachristou, Nederkoorn, Havermans, van der Horst, & Jansen, 2012; Garland, Carter, Ropes, & Howard, 2012). We were a priori interested in the test-retest
reliability and association with alcohol consumption in the social drinking session of the VS, ACC, and vmPFC. We expected substantial test-retest reliability—based on Landis and Koch’s benchmarks of intraclass correlation coefficients (1977)—in all three regions of interest (ROI). We also expected brain activity in each ROI to correlate with alcohol consumption in the social drinking session.
Methods Participants
Thirty people were recruited from a pool of participants from a previous alcohol study as well as through advertisements on social media and student networks. Participants were required to be right-handed, have normal or corrected-to-normal vision, and no contraindications for MRI acquisition. In addition, participants had to have consumed alcohol at least once within the past six months (in order to exclude total abstainers) and report liking beer to ensure the alcohol cues were relevant for all participants. To avoid possible confounds with other substance use,
participants had to have no history of drug use within the past 30 days. Other exclusion criteria include any current major mental illnesses or learning disorders. Finally, participants were asked to refrain from consuming alcohol within 24 hours of each scan.
Materials
Behavioral Measures. Alcohol use history and severity was assessed with the 14-Day Timeline Followback (TLFB; Sobell & Sobell, 1992) and the Alcohol Use Disorders
Identification test (AUDIT; Saunders et al., 2013). The TLFB is a reliable self-report measure of daily drinking. The AUDIT is a 10-item questionnaire that assesses harmful alcohol use and alcohol dependence, which is highly reliable and sufficiently sensitive and specific in
classification of alcohol use disorder (Babor, Higgens-Biddle, Saunders, & Montiero, 2001; Reinert & Allen, 2007).
Individual differences in impulsivity and inhibitory control were assessed with the Barrett Impulsivity Scale (BIS) and the Auditory Simon Task (AST; Patton et al., 1995; Xiong &
Proctor, 2016). The BIS is a reliable and valid measure of impulsivity and consists of 30 statements with a 4-point response scale ranging from “rarely or never” to “almost always or always,” with items such as “I do things without thinking” (Goudriann, Oosterlaan, De Beurs, & Van Den Brink, 2008; Stanford, Mathias, Dougherty, Lake, Anderson, & Patton, 2009). In the AST the speech sounds /a/ and /o/ are presented and the participant has to respond as quickly as possible depending on which sound is heard—the /a/ requires a left-hand response and the /o/ requires a right-hand response (counterbalanced across participants). Presentation of both sounds alternate between the right and the left ear on each trial, producing congruent and incongruent trials. On congruent trials, the sound is presented on the same side ear as the corresponding key press; on incongruent trials, the sounds is presented on the opposite side ear as the corresponding key press. In total, the task consisted of 16 practice trials, and three blocks with 100 trials in each. The measure of inhibitory control was then calculated using reaction times, subtracting congruent from incongruent trials.
In a social drinking session, alcohol consumption was measured by volume after
participants completed the session. The full session procedure will be described in the Procedure section.
fMRI Measures. Imaging was conducted using a Phillips 3T Achieva Scanner. During the cue-reactivity task, the blood oxygen dependent (BOLD) signal was measured with a T2* gradient-echo echo-planar imaging (EPI) sequence (TR 2 seconds, TE 27.63ms, 37 slices, slice
thickness 3mm, interslice gap 0.3 mm, FOV 240 x 240, in-plane resolution 80 x 80, flip angle 76.1°). A high-resolution T1-weighted structural scan was acquired after the functional scan in Session 1 for each participant (T1 turbo field echo, TR 8.28 seconds, TE 3.8ms, 220 slices, slice thickness 1mm, FOV 240 x 220, in-plane resolution 240 x 240, flip angle 8°).
To measure alcohol cue-induced neural activity, a multi-modal blocked design fMRI task was used. During the task, participants viewed images of either beer or water overlaid on images of three social drinking situations: a music festival, dinner party, and birthday party. Below the images, participants saw written offers of the beer or water (see Figure 1). During each block, participants also listened to an audio fragment corresponding to each of the three situations. These fragments are adapted versions of the Collegiate-Simulated Intoxication Digital Elicitation (C-SIDE), which are professionally scripted and recorded audio scenes of typical Dutch social drinking situations (Anderson, Duncan, Buras, Packard & Kennedy, 2013; Adapted Dutch version: Larsen, Salemink, Wiers, Andersen, submitted). The scenes consist of conversations between several speakers and background sounds relevant to each social context (e.g. music, clinking glasses, etc.) interspersed with offers for alcoholic and non-alcoholic beverages. Previous research in our lab has shown that willingness to accept alcohol offers within the C-SIDE simulation is correlated with retrospective and prospective alcohol use in students. For this task, the offers were removed from the simulations to remove response demands during each trial. Instead, participants were asked to imagine themselves in the simulation and consider the written offer on the screen.
In total, the task consisted of 6 blocked trials, which alternated between alcohol and water conditions. The condition that participants saw first during the task was randomized over
was a 16 second inter-trial-interval in which participants viewed a fixation cross. There were six possible orders for the trials given the alternation of condition and a requirement that no two scenes were shown back to back (e.g. alcohol-birthday party followed by water-birthday party), and the orders were randomized within the task. Immediately before and after the cue reactivity trials, participants rated their current level of craving for alcohol on a visual analogue scale from one to ten. The task and craving questions were presented using E-Prime 2.0 software
(Psychology Software Tools, Pittsburgh, PA).
Procedure
The present study used a test-retest reliability protocol in which participants were scanned twice and completed various behavioral tasks and questionnaires. The University of Amsterdam Faculty of Social and Behavioral Sciences ethics committee approved the study and all participants gave informed consent prior to participation.
Wil je een biertje?
Wil je water? Figure 1. Depiction of the alcohol cue reactivity scanning task. During the cue reactivity task, participants saw beer or water stimuli overlaid on three social drinking situations while listening to a corresponding audio fragment of each situation. Each trial lasted approximately two minutes.
The scanning sessions took place one week apart at the same time of day. The testing sessions took place in the evening to prevent social and practical concerns from inhibiting participants’ desire and willingness to consume alcohol. Furthermore, participants were scanned at the exact same time of day in each session. To prevent the content of questionnaires and behavioral tasks from influencing brain activity, scanning was conducted first in each session (see Figure 2 for depiction of experimental design). In Session 1, participants entered the scanner and were asked to report their current level of craving for alcohol on a scale of 1-10. Then, scanning began and the participants were presented with the cue reactivity task. Immediately following the task, participants were asked to report their current craving for alcohol again. After being removed from the scanner, participants were taken to a separate room and completed an online demographics questionnaire. This marked the completion of Session 1. In Session 2, the initial scanning procedure was exactly the same (including the pre- and post-task craving reports). After the scanning, they were again taken to another room where they first completed the auditory simon task on a computer. They then completed an online questionnaire containing the BIS, AUDIT, and TLFB.
Upon completion of scanning, questionnaires, and behavioral tasks in the second session, participants engaged in an adapted social drinking session (Larsen, Overbeek, Granic, & Engels, 2010). The goal of the social drinking session protocol was to create a semi-naturalistic drinking environment and allow for ad libitum drinking behavior. The experimenter informed participants that there was a brief unexpected break in the study due to logistical concerns (i.e. the testing room for the final task was being used for another experiment). During this break, the
experimenter led the participant to a kitchen area with refreshments where they were to wait until the study could resume. In this area, a confederate—who was pretending to be another
participant who was also waiting to finish the study—was already seated with a glass of (non-alcoholic) wine in order to set a norm that drinking during this period was acceptable. The experimenter offered the participant refreshments—wine, beer, soda, juice, or water—as an apology for the delay. The experimenter then poured the refreshments for the participant, informed the participant and confederate that they were welcome to help themselves to more beverages and food, and then left the area for thirty minutes. After approximately 15 minutes, the confederate finished their first alcoholic beverage, announced they would have one more, and offered to get something for the participant. This ensured that it was clear to the participant that alcohol could be consumed without applying direct peer pressure. Because the experimenter or confederate always poured the drinks, the amount of alcohol in each glass was always
standardized in order to maintain a reliable measure of the volume of alcohol consumed by each participant.
The independent variables in the study are the conditions (alcohol, water, and fixation) in the cue reactivity task. The dependent variables are brain activity during the cue reactivity task, self-reported craving levels pre- and post-task, and volume of alcohol consumption in the social drinking session. Level of inhibitory control, impulsivity, and alcohol use patterns (as measured by the AUDIT and TLFB) are the predicted covariate variables.
Results fMRI Data Analysis
Data pre-processing and statistical analysis of fMRI data of the cue-reactivity task was conducted with FMRI Expert Analysis Tool (FEAT), part of FSL (FMRIB’s Software Library,
http://www.fmrib.ox.ac.uk/fsl; Smith et al., 2004). First, non-brain tissue and skull was removed
Session 1
7 daysSession 2
Figure 2. Experimental Design. Participants were always scanned first in each session, followed by behavioral tasks and questionnaires. At the end of the second session, participants engaged in the social drinking session.
Craving report Scanning: Alcohol Cue-Reactivity Task Craving report Behavioral Measures Simon Task Questionnaires -BIS -AUDIT -TLFB Craving report Scanning: Alcohol Cue-Reactivity Task Craving report
Social Drinking Session Behavioral Measures
Demographics questionnaire
for each participant with the Brain Extraction Tool (BET). Then, images were slice-time aligned, motion corrected (non-linear), temporally high-pass filtered (sigma=148 seconds), spatially smoothed with a 5mm full-with-half-maximum Gaussian kernel, and pre-whitened (Woolrich et al., 2001). Next, functional data was registered to participants’ structural T1 image and
transformed into MNI space (Montreal Neurological Institute) using FLIRT (FMRIB’s Linear Image Registration Tool).
A first-level analysis was run modeling brain responses to alcohol, water, and fixation conditions as three separate regressors. Each regressor was then convolved with a double gamma hemodynamic response function. The contrast of interest for further analyses was the subtraction contrast ‘alcohol>water.’ The three ROIs of interest—ACC, VS, and vmPFC- were created with binary masks in MNI space. The bilateral ACC masks were created using the Harvard-Oxford Cortical Atlas included in FSL. The bilateral VS masks—which encompass the nucleus
accumbens, ventral caudate, and ventral putamen (Tziortzi et al., 2011)—were created using the Oxford-GSK-Imanova structural striatal atlas included in FSL. The vmPFC is not a structurally well-defined region in the brain, and is not included in the anatomical atlases in FSL. In many neuroimaging studies, the vmPFC is functionally defined, leading to differing coordinates of the vmPFC in the literature. Because of this, we decided to use a vmPFC mask created using the “reverse inference map” on the meta-analytic platform neurosynth (Wager, 2016). This probabilistic mask is based on the location of the vmPFC ROI of hundreds of published fMRI studies. We then thresholded and binarized this probabilistic map for consistency with our other masks.
Mean activity level in each ROI was extracted using FSL’s Featquery tool. The ICC (3,1) of each region was then calculated in SPSS. The reported p-values refer to significance against
zero. Follow-up paired samples t-tests were calculated for each region to test for habituation to the task. To examine brain-behavior correlations, mean activity in each ROI was regressed on alcohol consumption, while controlling for significant covariates. When necessary, results are reported at Bonferroni corrected p-values for multiple comparisons.
An exploratory group analysis was conducted with FEAT to test for a main effect of alcohol cue reactivity using the Alcohol>Water contrast averaged across all participants in each session separately. A cluster-wise multiple comparison correction was used, with a Z-threshold of 2.3 and a cluster-p significance threshold of .05.
Behavioral Data Analysis
Exploratory analyses were conducted on self-reported measures to test whether the cue reactivity task induced craving and whether craving after the task predicted drinking behavior during the social drinking session. To test whether the cue reactivity task induced craving, a paired samples t-test was conducted on self-reported craving ratings, separately for each session. A paired samples t-test was then conducted on the craving difference scores for each session to test whether there was a significant difference in the effect of the task on craving scores across sessions. To test whether task-induced craving predicted alcohol consumption, we regressed craving rating after the task in Session 1 and two on volume of alcohol consumed in the social drinking session. We ran bivariate correlations with our predicted covariates—impulsivity, inhibitory control, and alcohol use history/severity—to test whether we needed to control for any of these variables within the regression. Those candidate covariates that were significantly associated with alcohol consumption were added to the regression analysis as covariates.
Sample Characteristics
All 30 (12 men; Mean Age=24.25; SD =2.48) participants reported drinking at least one alcohol beverage in the past six months, were right-handed, and had no diagnosed mental illness or learning disorder. Self-reported alcohol use within the sample was varied as measured by the TLFB, ranging from daily alcohol consumption to total abstention during the two weeks
assessed. Guidelines for the AUDIT indicate that scores of 20 and above indicates alcohol dependence. Three participants met this criterion for alcohol dependence. The average AUDIT score was 15.86±3.58, suggesting that the sample was composed of moderate to heavy drinkers. On average, participants reported consuming alcohol on 6.24±3.22 days in the previous two weeks.
Confirmatory Analyses
Three participants were excluded from the following fMRI analyses based on a priori exclusion criteria. One was excluded for excessive movement during one of the scans (>3mm). Two were excluded for exceeding ±3SD of the mean of alcohol cue reactivity in the ROIs.
Reliability analyses. The reliability of activity in the right ACC for the contrast
Alcohol>Water across the two sessions was fair , while the reliability in the left ACC was poor (Right ACC: ICC(3, 1) = .388 (95% CI: .008 to .670), p=.006; Left ACC: ICC(3, 1) =.417(95% CI: .043 to .688), p=.015). The reliability of activity in the right and left VS was poor (Right VS: ICC(3, 1) =.171 (95% CI: -.224 to .518), p=.197; Left VS: ICC(3, 1) = .093 (95% CI: -.299 to .457), p=.323). The reliability of activity in the right and left vmPFC was also poor (Right vmPFC: ICC(3, 1)=.175(95% CI: -.221 to .521), p=.191; Left vmPFC: ICC(3, 1)=.158(95% CI: -.238 to .508), p=.216). To summarize, based on Cicchetti’s reliability benchmarks, the only ROIs that achieves fair reliability is the left ACC, with all other ROIs demonstrating poor reliability (See Figure 3;
Cicchetti, 1994). Paired samples t-tests were then conducted to test for habituation. None of the six ROIs showed a significant reduction in activity in Session 2 compared to Session 1.
Figure 3. Results of the reliability analysis for the ROIs. For each ROI, each participant’s activity
in the two sessions is plotted. The orange line in each scatter plot represents absolute consistency of activity across sessions. Individual data points closer to the line are more consistent.
ICC(3, 1) = .388 ICC(3, 1) =.417 ICC(3, 1) =.171 ICC(3, 1) = .093 ICC(3, 1) =.175 ICC(3, 1) =.158
Drinking prediction analyses. Linear regressions were run separately for each of the six ROIs to predict the amount of alcohol consumed in the social drinking session. Regressions were first run with only the respective ROI activity as a predictor, and then again with AUDIT, BIS, and Simon Effect scores as covariates. However, controlling for covariates did not affect the results, so only the uncontrolled results are reported in Table 1. None of ROIs significantly independently predicted drinking behavior during the social drinking session.
Table 1
Regression results for predicting drinking behavior from alcohol cue-reactivity
Predictor Variable B SEB β*
Right ACC .013 .013 .200 Right VS .017 .018 .190 Right vmPFC .005 .007 .145 Left ACC .013 .014 .183 Left VS .021 .018 .225 Left vmPFC .005 .006 .156
Note: All predictors were regressed independently on alcohol consumption due to low power
*No β values were significantly different than zero
Exploratory Analyses
fMRI results. To check whether we see significant alcohol cue reactivity in the brain during the task, we conducted an exploratory group analysis for each session separately using FSL’s FEAT. We found no significant activation to the alcohol cues in either Session 1 or Session 2 when controlling for multiple comparisons. Contrary to our behavioral results, which showed increased craving after the task, we do not see increased activation to alcohol cues compared to control cues on a neural level during the task.
We conducted follow-up exploratory analyses in an attempt to examine whether the variability in brain activity across sessions was related to behavioral measures including impulsivity, inhibitory control, and alcohol use history. To do so, we first calculated difference scores across sessions for each ROI, and then ran exploratory Pearson correlations with sum scores of the AUDIT and BIS, Simon effect, and composite scores of the TLFB (total number of drinks, frequency of drinking sessions, and number of drinks per session in two weeks), and difference in baseline craving before the task. However, no correlation reached a Bonferroni-corrected level of significance.
Behavioral results. One participant was excluded from the exploratory behavioral analyses for failing to provide answers to the questionnaires. Furthermore, for analyses that include a measure of alcohol consumption during the social drinking session, two additional participants were excluded for failing to complete the social drinking session due to logistical issues during the experiment.
To examine whether the cue reactivity task increased craving, we conducted paired samples t-tests of self-reported craving before and after the task for each session separately. In Session 1, self-reported alcohol cravings were significantly higher after the task (M=5.59, SD=1.9) compared to before the task (M=4.76, SD=2.047), t(28)=3.147, p=.004. However, in Session 2, self-reported alcohol cravings were only marginally higher after the task (M=5.14, SD=2.083) compared to before (M=4.69, SD=2.156) t(28)=1.436, p=.081. To determine whether there was a significant difference in craving induction as a result of the cue reactivity task across sessions, we conducted a follow-up paired samples t-test on individual difference scores for craving before and after the task in Session 1 and two. The cue reactivity task did not induce
increase craving more in Session 1 (M=.8276, SD=1.416) than in Session 2 (M=.4483, SD=1.68), t(28)=t=.799, p=.216.
To examine whether self-reported craving after the cue reactivity task in Session 2 predicted drinking behavior during the social drinking session (also conducted in Session 2) we ran a regression analyses. Post-task craving significantly predicted alcohol consumption in the social drinking session, β =.536, t(26)=3.172, p=.004 before controlling for covariates. To test for covariates, we ran bivariate Pearson correlations between alcohol consumption and our candidate covariates. Only impulsivity was (marginally) significantly correlated with a alcohol consumption in the social drinking task, r=.302, p=.067. Therefore, we reran our original regression analysis with impulsivity as a covariate. After controlling for impulsivity, post-task craving still significantly predicted alcohol consumption, β=.499, t(26)=2.642, p=.014.
Furthermore, post-task craving ratings in Session 1 also significantly predicted alcohol
consumption one week later in Session 2 with and without controlling for impulsivity (without
control: β =.441, t(26)=2.459, p=.021; with control: β =.407, t(26)=2.279, p=.032). Interestingly
pre-task craving in neither Session 1 nor Session 2 significantly predicted alcohol consumption after controlling for impulsivity (Session 1: β =.309, t(26)=1.636, p=115.; Session 2: β =.358, t(26)=1.953, p=.063), although pre-task craving in Session 2 was marginally significantly predictive. This suggests that cue-induced craving is particularly predictive of alcohol
consumption, and not just baseline craving levels in each session. To further examine this, we conducted two further regression analyses controlling for both baseline craving and impulsivity in Session 1 and two. The results can be seen in Table 2 and Table 3. Although the beta values of post-craving ratings fall below significance in both sessions (Session 1: t=1.474, p=.154; Session
control for baseline craving. Given our small sample size for this analysis (N=27), lack of power is likely to partly explain the lack of significance when adding baseline craving as an additional covariate.
Table 2
Regression results for predicting drinking behavior in Session 1
Predictor Variable B SEB β*
Impulsivity .017 .013 .238
Baseline Craving .003 .081 .009
Post-task Craving .126 .086 .400
*No β values were significantly different than zero
Table 3
Regression results for predicting drinking behavior in Session 2
Predictor Variable B SEB β*
Impulsivity .007 .014 .101
Baseline Craving .020 .070 .070
Post-task Craving .128 .078 .445
*No β values were significantly different than zero
Discussion
One of the main goals of this fMRI study was to assess the test-retest reliability of a novel block-designed multi-sensory cue-reactivity paradigm. To our knowledge, only one other study—solely focused on the VS—has investigated the reliability of alcohol cue reactivity as a measure (Schacht et al., 2011). We focused our analyses on three a priori determined ROIs that are known to be involved in the reward-circuit of addiction: the ACC, VS, and vmPFC (Schacht, Anton, & Myrick, 2013). Contrary to our hypothesis, our results suggest that brain activation to alcohol cues compared to control cues in the task may not be reliable within individuals over time. Activity in the left and right VS and vmPFC was highly variable within individuals with only a week in between scanning sessions, with neither of the lateralized ROIs exceeding an ICC
of .2. Activity in the left and right ACC demonstrated less variability, with the left ACC reaching the minimal cutoff for the fair reliability benchmark (Cichetti, 1994). The low reliability does not appear to be a result of habituation to the task, as alcohol cue reactivity in Sessions 2 was not significantly lower than Session 1 in any ROI.
We conducted exploratory analyses to examine whether state-dependent differences between sessions (e.g. change in baseline craving) or individual differences between subjects might explain the high variability in cue reactivity in these regions. We did not find evidence to suggest that individual differences in alcohol use and severity, impulsivity, or inhibitory control are associated with variability in neural alcohol cue reactivity over time. In addition, neither baseline craving nor differences in the strength of self-reported cue-induced craving between sessions were associated with variability in cue reactivity. However, given the relatively small sample size, we were underpowered to detect small to medium correlations. Therefore, it is possible that variability in cue reactivity between sessions is associated with individual differences or state-dependent factors, and further research should explore this possibility.
The poor reliability of alcohol cue reactivity in the VS in the current study diverges from Schacht and colleagues’ (2011) findings of an ICC of .77 for this region. However, in another study, Fliessbach and colleagues assessed the test-retest reliability of the VS during reward processes using an event-related design and found poor to fair reliability (2010). Furthermore, across a wide-variety of paradigms (working memory, emotion processing, auditory signal detection, etc.) the test-retest reliabilities of BOLD fMRI signals have been heterogeneous, ranging from below zero to very high (>.9; Specht, Wilmes, Shah, & Jäncke, 2003; Caceres, Hall, Zelaya, Williams, & Mehta, 2009; Johnstone et al., 2005). A variety of factors contribute to error variance in the measurement of BOLD signals including scanner field strengths, scanning
parameters, strength of the head coil, quality of the data, and experimental design (Fleissbach et al., 2010). Although we cannot explicitly determine how many of these factors contribute to our measurement variability over time, speculatively, certain aspects of our experimental design could contribute to the low reliability we found.
Specifically, because we used audio simulations of drinking situations along with visual stimuli in the task, each trial was approximately 2-minutes. Although Schacht and colleagues (2011) also used a blocked-design in their study of VS alcohol cue reactivity reliability, the blocks only lasted for 24 seconds, with five visual cues displayed for 4.8s each during each block. In our task, participants were instructed to imagine themselves in the situation and consider the alcohol/water offer for the entirety of the trial. However, the lack of response demands in the task prevented us from knowing whether participants were engaged in the task for the entirety of each trial, or whether they were imagining very similar situations each time even if engaged in the task. In addition, in order to compare reactivity to alcohol versus control cues, participants heard the same audio simulation twice in each session. Although blocked-designs are generally assumed to yield more robust results in fMRI research (Friston, Zarahn, Josephs, Henson, & Dale, 1999), the length and repetition of the trials make it likely that mind wandering and off-task thought processes might have occurred during the task, thereby reducing our signal of interest. Because we cannot parse the role that the task-related factors played in variability across sessions, we cannot conclusively state based on our results that neural alcohol cue reactivity is an inherently momentary assessment that fluctuates over time. Instead, we conclude that cue-induced activity in our novel task is unstable over time. In light of this, reliability of neural alcohol cue reactivity tasks should not be assumed when interpreting the results of studies using this paradigm. Future research should aim to examine the test-retest
reliability of other variations of the cue reactivity paradigm—such as those with shorter blocks, event-related designs, and response-demands to get a clearer picture of the reliability of alcohol cue reactivity in general.
The second goal of the current study was to examine whether neural alcohol cue reactivity could predict ad libitum alcohol consumption in a social drinking session. To our knowledge, most research investigating the association between alcohol cue reactivity and drinking behavior relies on report questionnaires; however, research suggests that self-reports of alcohol consumption are subject to systematic biases (Del Boca & Darkes, 2003; Davis, Thake, & Vilhena, 2010). Alcohol cue reactivity in our a priori ROIs did not significantly predict the amount of alcohol consumption in the drinking session. Neural reactivity to alcohol cues in the ROIs was also not significantly associated with any of the self-reported drinking measures. To check whether the task elicited the anticipated enhanced neural activation to alcohol cues, we conducted exploratory group-level analyses of the main effect of alcohol cues compared to control cues in each session. We did not find significantly increased activation in any regions of the brain for either session. It remains unclear whether the lack of this main effect is a result of the sample characteristics (e.g. a range of social drinkers) or due to the task-related issues previously discussed (e.g. block length). Regardless, given that our task did not elicit alcohol cue reactivity, it is not entirely surprising that we do not see an association between reactivity and behavior. We believe that future research should continue to investigate the relationship between neural cue reactivity and actual alcohol consumption with an altered task that has been previously shown to elicit alcohol cue-induced activation.
When considering future research possibilities, it is important to note that only 25% of the sample consumed alcohol during the social drinking session. Following our experimental
protocol, experimenters informed participants that there was an unexpected break in the study during which they could help themselves to our refreshments to pass the time with a confederate who was already drinking. However, participants may have been suspicious of the premise of the drinking sessions and may have refrained from drinking when they otherwise would have. Future studies attempting to create a naturalistic drinking session should consider whether it is
necessary for participants to be blind to the purpose of the drinking session. Participants might actually be more comfortable accepting alcohol if they knew it was a standard part of the study.
Despite the lack of association between cue reactivity and alcohol consumption in a social drinking context, we believe future research should continue to incorporate social context within the cue-reactivity paradigm. The role of social factors in the motivational and reward processes of alcohol has yet to be fully examined on a neural level. Understanding how motivational processes and social rewards interact with alcohol in context is particularly important given the role that social contexts play in people’s decisions to drink. Neuroimaging research has the potential to elucidate the way in which alcohol and social rewards interact, providing insight into the role that social contexts may play in alcohol-related decision-making.
In conclusion, we found poor to fair retest reliability of alcohol cue reactivity in three reward-related regions—the VS, ACC, and vmPFC—using a novel blocked-design cue reactivity task. Although specific characteristics of the task may have contributed to the low reliability, these results suggest that we should interpret the results of alcohol cue reactivity studies using fMRI with caution—especially those using test-retest protocols (such as treatment studies)— until we have a better understanding of the retest reliability of the measure. Furthermore, we did not find the expected association between neural alcohol cue reactivity and alcohol consumption in the social drinking session. Future studies examining the association between alcohol cue
reactivity and ad libitum alcohol consumption should attempt to create a more naturalistic setting and a less suspicious cover story such that participants feel more comfortable accepting alcohol offers.
References
Anderson, K. G., Duncan, K., Buras, M., Packard, C. D., & Kennedy, C. (2013). C-SIDE: Drinking simulation for college students. Journal of studies on alcohol and drugs, 74(1), 94-103.
Babor, T. F., Higgins-Biddle, J. C., Saunders, J. B., & Monteiro, M. G. (2001). The alcohol use disorders identification test. Guidelines for use in primary care, 2.
Beck, K. H., Arria, A. M., Caldeira, K. M., Vincent, K. B., O'Grady, K. E., & Wish, E. D. (2008). Social context of drinking and alcohol problems among college
students. American journal of health behavior, 32(4), 420-430.
Caceres, A., Hall, D. L., Zelaya, F. O., Williams, S. C., & Mehta, M. A. (2009). Measuring fMRI reliability with the intra-class correlation coefficient.Neuroimage, 45(3), 758-768.
Cicchetti, D. V. (1994). Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology.Psychological assessment, 6(4), 284. Claus, E. D., Ewing, S. W. F., Filbey, F. M., Sabbineni, A., & Hutchison, K. E. (2011).
Identifying neurobiological phenotypes associated with alcohol use disorder severity. Neuropsychopharmacology, 36(10), 2086-2096.
Cooper, M. L. (1994). Motivations for alcohol use among adolescents: Development and validation of a four-factor model. Psychological assessment, 6(2), 117.
Dager, A. D., Anderson, B. M., Rosen, R., Khadka, S., Sawyer, B., Jiantonio‐Kelly, R. E., ... & Fallahi, C. R. (2014). Functional magnetic resonance imaging (fMRI) response to alcohol pictures predicts subsequent transition to heavy drinking in college
Davis, C. G., Thake, J., & Vilhena, N. (2010). Social desirability biases in self-reported alcohol consumption and harms. Addictive behaviors, 35(4), 302-311.
Del Boca, F. K., & Darkes, J. (2003). The validity of self‐reports of alcohol consumption: state of the science and challenges for research. Addiction,98(s2), 1-12.
Everitt, B. J., & Robbins, T. W. (2005). Neural systems of reinforcement for drug addiction: from actions to habits to compulsion. Nature neuroscience,8(11), 1481-1489.
Field, M., Mogg, K., & Bradley, B. P. (2005). Craving and cognitive biases for alcohol cues in social drinkers. Alcohol and Alcoholism, 40(6), 504-510.
Fliessbach, K., Rohe, T., Linder, N. S., Trautner, P., Elger, C. E., & Weber, B. (2010). Retest reliability of reward-related BOLD signals. Neuroimage, 50(3), 1168-1176.
Friston, K. J., Zarahn, E. O. R. N. A., Josephs, O., Henson, R. N. A., & Dale, A. M. (1999). Stochastic designs in event-related fMRI. Neuroimage, 10(5), 607-619.
Garland, E. L., Carter, K., Ropes, K., & Howard, M. O. (2012). Thought suppression, impaired regulation of urges, and Addiction-Stroop predict affect-modulated cue-reactivity among alcohol dependent adults. Biological psychology, 89(1), 87-93.
Gauvin, D. V., Briscoe, R. J., Goulden, K. L., & Holloway, F. A. (1994). Aversive attributes of ethanol can be attenuated by dyadic social interaction in the rat. Alcohol, 11(3), 247-251. Goudriaan, A. E., Oosterlaan, J. A. A. P., De Beurs, E. D. W. I., & Van Den Brink, W. (2008).
The role of self-reported impulsivity and reward sensitivity versus neurocognitive
measures of disinhibition and decision-making in the prediction of relapse in pathological gamblers. Psychological medicine,38(01), 41-50.
Grüsser, S. M., Wrase, J., Klein, S., Hermann, D., Smolka, M. N., Ruf, M., ... & Heinz, A. (2004). Cue-induced activation of the striatum and medial prefrontal cortex is associated with subsequent relapse in abstinent alcoholics. Psychopharmacology, 175(3), 296-302. Ihssen, N., Cox, W. M., Wiggett, A., Fadardi, J. S., & Linden, D. E. (2010). Differentiating
heavy from light drinkers by neural responses to visual alcohol cues and other motivational stimuli. Cerebral cortex, 21(6), 1408-1415.
Jasinska, A. J., Stein, E. A., Kaiser, J., Naumer, M. J., & Yalachkov, Y. (2014). Factors
modulating neural reactivity to drug cues in addiction: a survey of human neuroimaging studies. Neuroscience & Biobehavioral Reviews, 38, 1-16.
Johnstone, T., Somerville, L. H., Alexander, A. L., Oakes, T. R., Davidson, R. J., Kalin, N. H., & Whalen, P. J. (2005). Stability of amygdala BOLD response to fearful faces over multiple scan sessions. Neuroimage, 25(4), 1112-1123.
Kuntsche, E., Stewart, S. H., & Cooper, M. L. (2008). How stable is the motive–alcohol use link? A cross-national validation of the drinking motives questionnaire revised among adolescents from Switzerland, Canada, and the United States. Journal of Studies on Alcohol and Drugs, 69(3), 388-396.
Landis, J. R., & Koch, G. G. (1977). The measurement of observer agreement for categorical data. biometrics, 159-174.
Larsen, H., Overbeek, G., Granic, I., & Engels, R. C. (2010). Imitation of alcohol consumption in same-sex and other-sex dyads. Alcohol and alcoholism, 45(6), 557-562.
Monti, P. M., Binkoff, J. A., Abrams, D. B., Zwick, W. R., Nirenberg, T. D., & Liepman, M. R. (1987). Reactivity of alcoholics and nonalcoholics to drinking cues. Journal of abnormal psychology, 96(2), 122.
Nestler, E. J. (2005). Is there a common molecular pathway for addiction?.Nature
neuroscience, 8(11), 1445-1449.
Oostveen, T., Knibbe, R., & De Vries, H. (1996). Social influences on young adults' alcohol consumption: norms, modeling, pressure, socializing, and conformity. Addictive Behaviors, 21(2), 187-197.
Papachristou, H., Nederkoorn, C., Havermans, R., van der Horst, M., & Jansen, A. (2012). Can’t stop the craving: the effect of impulsivity on cue-elicited craving for alcohol in heavy and light social drinkers.Psychopharmacology, 219(2), 511-518.
Patton, J.H., Stanford, M.S., Barratt, E.S. (1995): Factor Structure of the Barratt Impulsiveness Scale. J Clin Psychol, 51,768-74.
Psychology Software Tools, Inc. [E-Prime 2.0]. (2012)
Reinert, D. F., & Allen, J. P. (2007). The alcohol use disorders identification test: an update of research findings. Alcoholism: Clinical and Experimental Research, 31(2), 185-199. Ridderinkhof, K. R., Ullsperger, M., Crone, E. A., & Nieuwenhuis, S. (2004). The role of the
medial frontal cortex in cognitive control. science, 306(5695), 443-447.
Ridderinkhof, K. R., Van Den Wildenberg, W. P., Segalowitz, S. J., & Carter, C. S. (2004). Neurocognitive mechanisms of cognitive control: the role of prefrontal cortex in action selection, response inhibition, performance monitoring, and reward-based learning. Brain and cognition, 56(2), 129-140.
Robinson, T. E., & Berridge, K. C. (2000). The psychology and neurobiology of addiction: an incentive–sensitization view. Addiction, 95(8s2), 91-117.
Saunders, J. B., Aasland, O. G., Babor, T. F., De la Fuente, J. R., & Grant, M. (1993). Development of the alcohol use disorders identification test (AUDIT): WHO
collaborative project on early detection of persons with harmful alcohol consumption‐ II. Addiction, 88(6), 791-804.
Schacht, J. P., Anton, R. F., & Myrick, H. (2013). Functional neuroimaging studies of alcohol cue reactivity: a quantitative meta‐analysis and systematic review. Addiction
biology, 18(1), 121-133.
Schacht, J. P., Anton, R. F., Randall, P. K., Li, X., Henderson, S., & Myrick, H. (2011). Stability of fMRI striatal response to alcohol cues: a hierarchical linear modeling
approach. Neuroimage, 56(1), 61-68.
Schneider, F., Habel, U., Wagner, M., Franke, P., Salloum, J. B., Shah, N. J., ... & Gaebel, W. (2001). Subcortical correlates of craving in recently abstinent alcoholic
patients. American Journal of Psychiatry, 158(7), 1075-1083.
Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E., Johansen-Berg, H., ... & Niazy, R. K. (2004). Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage, 23, S208-S219.
Sobell, L. C., & Sobell, M. B. (1992). Timeline follow-back. In Measuring alcohol consumption (pp. 41-72). Humana Press.
Specht, K., Willmes, K., Shah, N. J., & Jäncke, L. (2003). Assessment of reliability in functional imaging studies. Journal of Magnetic Resonance Imaging, 17(4), 463-471.
Stanford, M. S., Mathias, C. W., Dougherty, D. M., Lake, S. L., Anderson, N. E., & Patton, J. H. (2009). Fifty years of the Barratt Impulsiveness Scale: An update and review. Personality and individual differences, 47(5), 385-395.
Tomie, A., Uveges, J. M., Burger, K. M., Patterson-Buckendahl, P., & Pohorecky, L. A. (2004). Effects of ethanol sipper and social opportunity on ethanol drinking in rats. Alcohol and Alcoholism, 39(3), 197-202.
Townshend, J., & Duka, T. (2001). Attentional bias associated with alcohol cues: differences between heavy and occasional social drinkers.Psychopharmacology, 157(1), 67-74. Tziortzi, A. C., Searle, G. E., Tzimopoulou, S., Salinas, C., Beaver, J. D., Jenkinson, M., ... &
Gunn, R. N. (2011). Imaging dopamine receptors in humans with [11 C]-(+)-PHNO: dissection of D3 signal and anatomy.Neuroimage, 54(1), 264-277.
Varlinskaya, E. I., Spear, L. P., & Spear, N. E. (2001). Acute effects of ethanol on behavior of adolescent rats: role of social context. Alcoholism: Clinical and Experimental
Research, 25(3), 377-385.
Vollstädt-Klein, S., Loeber, S., Kirsch, M., Bach, P., Richter, A., Bühler, M., ... & Kiefer, F. (2011). Effects of cue-exposure treatment on neural cue reactivity in alcohol dependence: a randomized trial. Biological Psychiatry, 69(11), 1060-1066.
Vollstädt‐Klein, S., Loeber, S., Richter, A., Kirsch, M., Bach, P., von der Goltz, C., ... & Kiefer, F. (2012). Validating incentive salience with functional magnetic resonance imaging: association between mesolimbic cue reactivity and attentional bias in alcohol‐dependent patients. Addiction biology, 17(4), 807-816.
Wager, T. (2016, August 16). Brain masks. Retrieved August 23, 2017, from https://canlabweb.colorado.edu/wiki/doku.php/help/core/brain_masks
Woolrich MW, Ripley BD, Brady M, Smith SM (2001) Temporal autocorrelation in univariate linear modeling of FMRI data. Neuroimage 14:1370–1386.
Xiong, A., & Proctor, R. W. (2016). Decreasing auditory Simon effects across reaction time distributions. Journal of experimental psychology: human perception and
performance, 42(1), 23.
Yalachkov, Y., Kaiser, J., & Naumer, M. J. (2012). Functional neuroimaging studies in addiction: multisensory drug stimuli and neural cue reactivity.Neuroscience & Biobehavioral Reviews, 36(2), 825-835.
