Neural Mechanisms of Sensitivity to Peer Information in Young Adult Cannabis Users

Abstract

Though social influence is a critical factor in the initiation and maintenance of marijuana use, neural correlates of influence in those who use marijuana are unknown. In this study, marijuana-using young adults (MJ) (n = 20) and controls (CON) (n = 23) performed a decision-making task in which they made a perceptual choice after viewing the choices of unknown peers via photographs, while they underwent functional magnetic resonance imaging scans. MJ and CON did not show differences in the overall number of choices that agreed vs. opposed group influence, but only the MJ group showed reaction time slowing when deciding against group choices. Reaction time was associated with greater activation of frontal regions. MJ compared to CON showed significantly greater activation in the caudate when presented with peer information. Across groups, caudate activation was associated with self-reported susceptibility to influence. These findings indicate that young adults who use MJ may exhibit increased effort when confronted with opposing peer influence, as well as exhibit greater responsivity of the caudate to social information. These results not only better define the neural basis of social decisions, but also suggest that marijuana use is associated with exaggerated neural activity during decision-making that involves social information.

Marijuana (MJ) is the most widely used drug other than alcohol among young adults; according to the 2011 National Survey on Drug Use and Health, 38.7% of 18-20 year-old, and 32.9% of 21-25 year-old young adults reported using MJ in the past year, which is an increase in prevalence compared to earlier decades (Henry, Oldfield, & Kon, 2003). It is important to understand motivations for MJ use, as heavy use in adolescence/young adulthood has been robustly linked with harms such as future substance dependence (Berge et al., 2015; Green et al., 2015), worse psychosocial functioning (Lagerberg et al., 2015; Wilkinson, Stefanovics, & Rosenheck, 2015) and worse academic outcomes (Epstein et al., 2015; Meier, Hill, Small, & Luthar, 2015; Silins et al., 2015). Young adults may both initiate and continue to use MJ in part due to social factors. Peers play a pivotal role in introducing an individual to a drug (Clayton & Lacy, 1982; Khavari, 1993), and most drug use occurs in social and recreational settings (Terry-McElrath, O'Malley, & Johnston, 2009). Further, in a survey of 106 MJ users, almost half cited social pressure as a motive for using (Hartwell, Back, McRae-Clark, Shaftman, & Brady, 2012). Although the impact of social influence on behavior is greater in younger than in older adults broadly (Gardner & Steinberg, 2005), it remains poorly understood if susceptibility to peer influence is further exaggerated in young adults who use MJ and, if so, whether this is reflected in unique patterns of brain activity. The purpose of the current study is to establish an association between susceptibility to social influence and MJ use, as this relationship could indicate that susceptibility to influence is a pre-existing risk factor, and/or a neurodevelopmental outcome, of MJ exposure.

Scientists have used functional magnetic resonance imaging (fMRI) to establish that structures such as the nucleus accumbens (NAc), caudate, amygdala, cingulate, and anterior insula are activated when individuals make decisions in the context of peer influence. These regions are associated with social reward and learning, social cognition, and physiological arousal and anxiety, indicating that conforming or deviating from social norms may generate a variety of emotional responses (e.g. (Berns, Capra, Moore, & Noussair, 2010; Berns et al., 2005; Klucharev, Hytonen, Rijpkema, Smidts, & Fernandez, 2009). Neural sensitivity to peer influence appears to be especially salient in striatal regions. One of the first studies to use fMRI to study social influence demonstrated that the caudate was more active when participants saw ‘popular’ than ‘unpopular’ symbols (Mason, 2009). In a study in which participants rated attractiveness of faces before and after receiving information about how their peers rated each face, deactivation of the NAc during a conflict with group opinion predicted conformity in future trials (Klucharev et al., 2009). Another study using ratings of facial attractiveness found that agreeing with peers increased activity in the NAc, whereas disagreeing decreased NAc activity (Zaki, Schirmer, & Mitchell, 2011). These studies suggest that individuals process both the rewarding aspect of social stimuli and the congruence of consensus with other people using traditional reward circuitry such as the caudate and the NAc (Zaki et al., 2011).

Many studies support the idea that peer opinion can affect neural response, particularly during adolescence (e.g. (Guyer et al., 2014; Masten et al., 2009; Pfeifer et al., 2011; Welborn et al., 2016), and that caudate activity may be a marker of social reward in typically developing adolescents (Guyer et al., 2014). Chronic peer conflict was shown to be associated with greater risk-taking behavior and heightened activation in brain regions involved in affect and reward processing, such as the striatum and insula (Telzer, Fuligni, Lieberman, Miernicki, & Galvan, 2015). A series of studies showed that during risk tasks such as simulated driving experiments, as well as during non-risky decision-making tasks, being observed by peers elicited striatal activation in adolescents (Chein, Albert, O'Brien, Uckert, & Steinberg, 2011; Smith, Steinberg, Strang, & Chein, 2015). In addition, a behavioral study in which people rated risky behaviors before and after observing peer ratings of those behaviors found a steady decline in this social conformity from late childhood through adulthood (Knoll, Magis-Weinberg, Speekenbrink, & Blakemore, 2015). These together support the idea that the striatum is sensitive to social context and can be linked to individual differences in sensitivity to peer influence, particularly in adolescentsIn addition to literature on neural correlates of peer effects, there is also a growing literature regarding the effects of MJ on the brain. The main psychoactive component of cannabis, Δ-9-tetrahydrocannabinol (THC), acts as a direct agonist of cannabinoid CB1 receptors in the brain (Downer, Boland, Fogarty, & Campbell, 2001; Heath, Fitzjarrell, Fontana, & Garey, 1980; Lawston, Borella, Robinson, & Whitaker-Azmitia, 2000; Scallet et al., 1987), which are highly expressed in regions involved in the regulation of mood and cognition (amygdala, striatum, prefrontal cortex and hippocampus). There is now a well-established relationship between regular MJ use and a range of potentially adverse outcomes, including effects of learning and cognition (see (Solowij & Battisti, 2008; Solowij, Stephens, Roffman, & Babor, 2002) for review), effects on brain structure and function (see (Lorenzetti, Solowij, Fornito, Lubman, & Yucel, 2014) for review), and potential impacts on mental health (Hall & Degenhardt, 2009). Cannabis appears to affect both ventral (including the NAc) and dorsal (caudate and putamen) striatum function. The striatum has a central role in the mechanisms of reward; as such, its dysregulation is thought to relate to anhedonia and amotivation that is often observed with MJ use (Nestler & Carlezon, 2006). Activation of the ventral striatum/NAc is associated with acute MJ use, and in fact, data indicate an increase in dopamine concentrations in the human ventral striatum of 136% during cannabis administration (Bossong et al., 2009). THC-exposed rats showed structural changes in the NAc (e.g. increases in the length of the dendrites and number of dendritic spines) (Kolb, Gorny, Limebeer, & Parker, 2006), with similar increases in the NAc reported in human MJ users (Gilman et al., 2014) Functional MRI studies have also revealed greater activation of the caudate and putamen in heavy cannabis users versus nonusers during tasks that are dependent on the dorsal striatum memory system (e.g. (Ames et al., 2013; Bohbot, Del Balso, Conrad, Konishi, & Leyton, 2013)), and greater activation of the caudate during the Iowa gambling task (Acheson et al., 2015), indicating that all subregions of the striatum are likely impacted by MJ use.

Few studies have specifically investigated peer influence within substance-using individuals, and none have specifically investigated this association among individuals who use MJ. A recent developmental neuroscience study that investigated links between resting-state activity of reward circuitry and substance use demonstrated that among adolescents who used alcohol, MJ, and cigarettes, the earlier they initiated substance use, the greater the strength of connectivity between the NAc and frontal regions (Weissman et al., 2015). This study fits nicely with the larger body of research showing that adolescent impulsivity and/or novelty seeking can be explained in part by maturational differences in frontal cortical and subcortical monoaminergic systems (see (Chambers, Taylor, & Potenza, 2003) for review). However, the lack of studies directly addressing peer influence, reward circuitry, and substance use, indicates a fundamental gap in the literature that the current study seeks to address.

In the current study, we adapted a task used in a series of classical social psychology experiments in the 1950s by psychologist Solomon Asch (Asch, 1951, 1952, 1956). Asch used a simple paper-and pencil line-judgment task to demonstrate that individuals were likely to agree with peers even at the expense of accuracy. We chose to use an objective task in which there are ‘right’ or ‘wrong’ answers (in contrast to subjective tasks such as rating attractiveness of faces or likability of music), in order to create a scenario in which there may be negative consequence of going along with the group if the group suggest an incorrect answer. This paradigm may be more analogous to social influence surrounding drug use, in which there is a cost to following substance-using peers. For the purpose of this study, social influence was defined in two ways: (1), sensitivity to peer information in general (compared to no peer information), and (2), likelihood of conforming to (compared to dissenting from) peer opinion. We hypothesized that sensitivity to peer information, would be reflected in greater caudate activation in the MJ compared to the CON group, and that likelihood of conforming to peer opinion, would be reflected in greater NAc activation in the MJ compared to the CON group.

Methods

Participants

Participants were 43 young adults, age 18-25 years; 20 who used MJ at least once a week (MJ), and 23 age and gender matched controls (CON). We chose to study light to moderate users, rather than heavy users, because evidence suggests that peer influence may have more of an impact in recreational users than it would in substance-dependent individuals, whose drug-taking may become less social and more habitual or compulsive (Wise, 1996). This sample size was calculated based on previous studies that used neuroimaging paradigms to investigate correlates of social influence (Berns et al., 2010; Berns et al., 2005; Klucharev et al., 2009). MJ were asked to refrain from using all substances on the day of the study. CON had used MJ on less than 5 lifetime occasions, and had not used MJ in the past 3 months. All participants were medically healthy, with no current psychiatric disorders (verified by the Structured Clinical Interview for DSM-IV Non-Patient Version (SCID; (First, 2002)) except for cannabis use disorders in the MJ group. Participants who met current or lifetime abuse or dependence criteria for any drug, including alcohol and nicotine, were excluded, though prior use of other drugs was not exclusionary (see Table 1). All participants gave written informed consent to a protocol approved by the Partners Human Research Committee Institutional Review Board.

Table 1. Participant Demographics.

		CON (n=23)	MJ (n=20)
Gender		11M/12F	9M/11F
Age	Yrs	21.6 (1.9)	20.6 (2.5)
Education	Yrs	15.1 (1.4)	14.2 (1.8)
TIPI	Extroversion	7.5 (2.8)	8.9 (3.5)
	Agreeableness	10.2 (1.9)	9.9 (2.1)
	Conscientiousness	11.5 (2.0)	9.9 (2.6)a
	Emotional Stability	11.5 (1.8)	10.6 (2.7)
	Openness	11.2 (2.5)	11.7 (1.9)
Substance Use
Alcohol	# Drinks/Week	2.4 (2.4)	2.8 (2.0)
Cigarettes	# Occasional smokers	0	7
Marijuana	# MJ Use Days/week	0	2.7 (1.2)
	# MJ Joints per week	0	6.6 (7.5)
	Age of Onset (years)	0	18.3 (2.0)
	Duration of Use (years)	0	2.3 (1.5)
Suggestibility	Consumer Suggestibility	18.0 (4.7)	20.3 (6.3)
	Persuadability	41.4 (6.5)	43.4 (8.4)
	Physiological Suggestibility	18.4 (3.3)	18.8 (4.6)
	Physiological Reactivity	41.4 (6.4)	43.8 (6.4)
	Peer Conformity	20.6 (7.0)	24.6 (6.8)b
	Mental Control	38.7 (8.2)	38.1 (9.6)
	Unpersuadability	53.7 (6.6)	53.0 (8.8)
	Total Suggestibility	139.8 (15.7)	150.8 (22.4)b

All values are expressed in means and standard deviations. There were no significant differences between groups on any measure other than conscientiousness.

^ap = 0.03,

^bp = 0.07

Measures

MJ participants completed a time-line follow-back (Sobell, Sobell, Klajner, Pavan, & Basian, 1986) asking them to indicate, for the past 90 days, the days that they used MJ. Cumulative number of joints (or joint equivalents), number of joints per week, and number of joints per day, were calculated for the 90 days prior to the study. MJ and CON also completed a time-line follow-back for alcohol use (Sobell et al., 1986), to detail their drinking behavior in the past 90 days, resulting in an estimate of past three-month standard drink consumption. We chose to include the TLFB for alcohol to ensure that alcohol use patterns were not different between the two groups. All participants also completed the Ten-Item Personality Inventory (TIPI; (Gosling, 2003)) to assess personality characteristics.

Participants also completed the Multidimensional Iowa Suggestibility Scale (MISS; (Kotov, 2004), which assesses self-reported susceptibility to influence in five social domains: consumer suggestibility (e.g., suggestibility to commercials, products), persuadability (e.g., changing one's mind based on other peoples' arguments), physiological suggestibility (e.g., feeling cold when someone else is shivering), physiological reactivity (e.g., feeling jumpy after watching a scary movie), and peer conformity (e.g., liking the same celebrities/fashion/music as friends). Scores across subscales were summed with higher total values indicating greater suggestibility. Because the aim of the current study was to understand the relationship between peer influence and brain activation, only the peer conformity subscale was regressed against activation in our predetermined regions of interest.

Social Influence Task Design

The social influence task was designed to measure an individual's likelihood of following group decisions or making independent choices in a visual discrimination task, and was based on a task previously developed by our group (Gilman et al., 2016). Each trial consisted of 5 events, shown in Figure 1A. There were 44 total trials, each consisting of 5 Events. The 44 trials were split into 2 runs of 22 trials each. Each trial took 16 seconds and the task took approximately 20 minutes to complete.

Fig 1.

Social Influence Task. This task consists of five events, described in “Methods,” which represent phases of decision-making. In Event 1 (Cue), participants were presented with two lines, and asked to judge which line was longer. In Event 2 (Influence), two photographs of fictitious “previous participants” were shown. The participant was either presented with “peer” responses (top), or not shown responses of peers; instead, they would only see “x” under each photograph (bottom).

In Event 3 (Choice), they were asked to make a choice, and could either agree or disagree with the peer responses in Event 2. In Event 4 (Confidence), they were asked to rate their confidence in their decision. In Event 5 (Feedback), they could earn 1 point for a correct response (or 0 for an incorrect response). All events were jittered with a fixation crosshair that was presented for a random non-integer interval between 1-5 seconds.

In Event 1 (Cue), the participant saw a cue, consisting of two lines, and was asked to judge whether the left or the right line was longer. In approximately 20% of the trials (8 trials), the task was “easy” (i.e. participants could easily tell which was longer, yielding < 5% error rate); and in the other 80% of the trials (36 trials), the task was “difficult” (yielding ≈ 50% error rate). We chose to make the majority of trials difficult because we found in piloting the task that participants rarely conformed to group influence on easy trials. In Event 2 (Influence), two photographs of fictitious ‘previous participants’ with responses were revealed to the participant. These photographs were taken from the Texas Center for Vital Longevity at University of Texas, Dallas (happy expressions; ages 18-29) (Minear & Park, 2004) and the Max Plank FACES database (happy expressions, young adults age 19-31)(Troje & Bulthoff, 1996). Participants were given a binder of 32 color photographs (16 of each gender) before beginning the task, and were allowed to select 8 peers; this was done to create an affiliation between the participant and the ‘participants’ in the photographs (Polonec, Major, & Atwood, 2006). In 28 of 44 trials, both peers showed the same response (both either chose ‘left’ or ‘right’), and the participant could either agree or disagree with these peers. In 6 trials, the peer responses were split, so that one peer chose ‘right’ and one chose ‘left’). The split condition was not analyzed, but was included only to make the task believable (i.e. participants would get suspicious if there were no trials in which the peers were split). In 10 trials, participants were not shown responses of peers; instead of responses, they would only see “x” under each photograph. In 50% of difficult trials (14 trials), the ‘group’ responses were correct (i.e. the line on the left was longer, and the ‘group’ recommended ‘left’), and in the other 50% (14 trials), the ‘group’ was incorrect (i.e. the line on the left was longer but that ‘group’ recommended ‘right’). These two conditions were combined, since group accuracy did not affect choice behavior, which was at chance level for difficult trials. The ‘group’ was always correct on easy trials in order to increase the believability of the task (Figure 1). Trial types were presented in a random order. In Event 3 (Choice), participants chose which line segment they judged to be longer (‘Left’ or ‘Right’). In Event 4 (Confidence), participants were asked to rate their confidence in that judgment on a 7-point Likert scale. In Event 5 (Points), participants were told if they were correct on that trial. If participants made the correct judgment, they received a point. A monetary reward was given at the end of the experiment based on the number of points received (over 80 points = $20, 70-79 points = $15, 60-69 points = $10, and 20-59 points = $5). This experiment was programmed in Python 2.7 using the package PyGame (http://www.pygame.org/) version 1.9.2.

Participants answered questions after the scan about how likely they were to go against group responses. Answers ranged, on a Likert scale, from “very unlikely” to “very likely.”

Behavioral Analysis

We conducted repeated-measures ANOVAs (SPSS, Version 16) to assess differences in behavior between groups. The independent variable was group (CON, MJ), and the dependent variable was choice (congruent with the ‘group’, incongruent with the ‘group’). We conducted ANOVAs to investigate whether reaction times (i.e. time from being presented with the choices “Left” or “Right in Event 3 to pressing a button to designate a response), and/or confidence ratings in Event 4 differed between groups or across conditions (i.e. when choice was congruent or incongruent with the group). Significant main effects and group by condition interactions were followed up with post-hoc Tukey tests. We also conducted linear regression models of MISS peer conformity score in and post-scan questionnaire scores and neural activation in a priori regions of interest across groups.

Acquisition and Analysis of Neuroimaging Data

Participants were scanned using a 3T Skyra Siemens (Erlangen, Germany) scanner with a 32 channel head coil at the MGH Martinos Center for Biomedical Imaging. Whole-brain T1-weighted 1 mm isotropic structural scans were collected using a 3D multiecho MPRAGE sequence (176 sagittal slices, 256 mm FoV, TR 2530 ms, TI 1200 ms, 2× GRAPPA acceleration, TE 1.64/3.5/5.36/7.22 ms, BW 651 Hz/px, T_acq 6:03 mins) (van der Kouwe, Benner, Salat, & Fischl, 2008). Functional scans were collected using a 2D gradient echo EPI sequence (31 slices, 3 mm thick, 0.6 mm gap, 216 mm FoV, 3×3 mm² in-plane resolution, TR 2 s, TE 30 ms, BW 2240 Hz/px). All acquisitions were automatically positioned using AutoAlign (van der Kouwe et al., 2005).

Data processing was carried out using FEAT (fMRI Expert Analysis Tool) Version 5.98, part of the FSL fMRI processing stream (fMRIB's Software Library, www.fmrib.ox.ac.uk/fsl). Each subject's functional and structural scans were registered using FSL's linear registration tool (FLIRT), and then these scans were registered to high resolution structural and standard space images using both FLIRT and FSL's nonlinear registration tool (FNIRT) (Jenkinson & Smith, 2001) so that each subject's brain was registered to the ICBM152 T1 template (Chau & McIntosh, 2005). In addition, the following pre-processing was applied; non-brain removal using FSL's brain extraction tool (S. M. Smith, 2002); spatial smoothing using a Gaussian kernel of FWHM 5 mm; grand-mean intensity normalization of the entire 4D dataset; highpass temporal filtering (Gaussian-weighted least-squares straight line fitting, with sigma=50.0 s). Analysis was performed in two steps. First, each subject's time series data was fit using a linear signal model with 11 regressors of interest, and 6 movement regressors of no interest. Regressors of interest included the following explanatory variables (EVs): (1) Cue presentation, (2) Influence (e.g. both peers showing the same response), (3) No Influence (e.g. x's displayed below the peers), (4) Split Influence (e.g. one peer chose ‘right’ and one chose ‘left’), (5) Choice after Influence when the participant followed the group, (6) Choice after Influence when the participant dissented from the group, (7) Choice after No Influence, (8) Choice after Split Influence, (9) Confidence, (10) Positive feedback (Point-based Gains), (11) Negative Feedback (No Gains). We modeled our regressors using FSL's 3 column format, in which each event was programmed with a triplets of numbers, consisting of the onset of the event (in seconds), the duration (in seconds) of the event, and weight of the input (all were assigned a weight of “1”). Events were averaged and convolved using a gamma waveform. The experiment was jittered with the inclusion of a fixation point presented between each task event. This fixation point was presented for a random time interval between 1-5 seconds (at a non-integer value), which allowed us to separate each event of a trial for analysis.

Next, higher-level group analysis was carried out using FLAME (FMRIB's Local Analysis of Mixed Effects) stage 1 and stage 2 (Woolrich et al., 2009). To correct for multiple comparisons, Z (Gaussianised T/F) statistic images were thresholded using clusters determined by Z>2.3 and a (corrected) whole-brain cluster significance threshold of p<0.05 (Worsley, 2001). Our analysis consisted of a whole-brain, random effects analysis of two primary contrasts of interest, which reflected our hypotheses regarding social influence: (1) response to influence vs. no influence, which isolated areas of the brain that responded to group information in general (e.g. EV 2 vs 3), and (2) congruent vs. incongruent choices, which isolated brain areas that showed different responsivity to following vs. dissenting from group information (e.g. EV 5 vs 6). Participants with too few responses (5 or less) were removed from the second analysis, as these participants had too few responses to accurately estimate neural response to these events. The other task events were necessary in order to model the task, but were not included in the primary analysis.

In parallel with whole-brain analyses, individual parameter estimates for a region-of-interest (ROI) analysis were extracted using the FSL program featquery (http://fsl.fmrib.ox.ac.uk/fsl/fsl4.0/feat5/featquery.html). Because we had a priori, anatomically-focused hypotheses on the caudate and the NAc, we chose an ROI approach in order to determine whether there was evidence of task-related changes in brain activity within these regions and whether this activation correlated with behavioral outputs. We chose to use an ROI approach to (1) reduce the number of voxels to a single mean data point, yielding measurements that are less variable than the voxel-level measurements, and (2) to mitigate the multiple comparisons issue, which provided increased sensitivity to detect task-related differences in brain activity (Bowman, Guo, & Derado, 2007). We extracted data from each participant from the NAc, and the caudate (see fig. 4A which illustrates the anatomical masks that were used). Activation signal was extracted from using the following steps: (1) the signal at each voxel was converted to a (percentage) deviation from the mean for that voxel across the entire time series, (2) the signal was averaged by stimulus type and spatially translated into MNI space, and (3) anatomical masks were designated consisting of the volume of interest through which each individual participant's data was extracted. All masks were parcellated from the ICBM152 T1 brain at the MGH Center for Morphometric Analysis (CMA) (Gasic et al., 2009; Perlis et al., 2008), using validated anatomical landmarks for the NAc and caudate.

Fig 4.

Association between self reported conformity and brain activation. A. Depiction of anatomical masks used in the ROI analysis. B. Peer conformity scores on the MISS correlated with activation in the right caudate in response to viewing peer information (top). Self-reported likeliness to go along with group influence correlated with deactivation of the right NAc in response to making incongruent choices (bottom). C. MJ showed significantly greater activation compared to CON in the right caudate when viewing group information (top); there was no difference between groups in activation of the NAc (bottom).

Results

Participant Characteristics

CON and MJ were not different in gender, age, or years of education (Table 1). MJ reported a trend for greater peer conformity (p = 0.07) and greater suggestibility (p = 0.07) than CON, and CON reported greater conscientiousness than MJ (p = 0.03). Personality measures were not related to task behavior, self-reported task behavior, or MISS scores.

Behavioral Results

There was a main effect of choice type (F (1,41) = 36.27, p < 0.0001, ηp² = 0.469) (Fig 2A), such that both CON and MJ were more likely to make congruent than incongruent choices. CON participants had an average of 12.0 congruent choices (SD = 2.7, range = 8-18) and 7.9 incongruent responses (SD = 2.6, range = 2-13). MJ participants had an average of 12.2 congruent choices (SD = 2.6, range = 8-16) and 7.0 incongruent responses (SD = 2.4, range = 2-11). There was no main effect of group or group by choice type interaction, indicating that choice behavior did not differ between groups, but that both groups preferred to follow group responses.

Fig 2.

A. Choice Behavior in Social Influence Task. All participants were more likely to follow group responses than to go against group responses; there were no significant differences between groups. Each point represents an individual subject; the black line represents the mean, and the red/blue lines represent the 95% confidence interval. B. Confidence and Reaction time measures. Controls (top left) did not show modulation of reaction time across conditions. MJ (top right) showed a significant effect of choice type on reaction time, responding more slowly for choices that were incongruent with group responses. For confidence ratings, both controls (bottom left) and MJ (bottom right) reported greater confidence when agreeing than when disagreeing with the group responses.

There was, however, a group by choice type (congruent, incongruent, ‘split’, or ‘none’) interaction for reaction time (F (1,41) = 4.86, p = 0.03, ηp²= 0.10). Post-hoc tests indicated that MJ had longer reaction times when making choices incongruent to peer responses compared to choices congruent with peer responses, whereas CON showed no difference in reaction time between congruent and incongruent responses (Fig 2B). Additionally, there was a main effect of choice type on confidence ratings (F (3,39) = 14.83, p < 0.001, ηp²= 0.503), indicating that both groups were more confident when agreeing with than when disagreeing with group responses (Fig 2B). There was no main effect of group or group by confidence interactions. There was no group difference in points earned and money received (p = 0.72). See Table S1 for descriptive statistics on task behavior.

Neuroimaging Results

Activation to Social Influence vs No Influence (Whole-Brain Analysis)

The following brain imaging results refer to changes in activity during Event 2, Influence, in which participants were or were not shown group responses (Table 2). When participants saw information from faces with peer responses, compared to no peer information, voxel-based analyses showed a main effect of social influence across a broad network of brain regions associated with social cognition and decision-making, predominantly frontal structures such as the rostral cingulate zone (RCZ), dorsolateral prefrontal cortex (DLPFC), and bilateral caudate (Fig 3, far left). When we examined each group separately, we observed that the MJ group was driving this main effect, with significant activation in the caudate, RCZ, inferior frontal gyrus (IFG), and DLPFC, while CON showed significant activation only in occipital structures. When directly comparing MJ and CON, MJ showed significantly greater activation than CON in the right caudate (Fig 3, far right). To test whether the caudate difference was modulated by other variables, we ran an ANOVA with caudate activation as the dependent variable, group membership (MJ, CON) as the independent variable, and covariates of interest including alcohol use, age, gender, and personality. We found that the significant predictors of caudate activation included group (F = 15.10, p < 0.001, η² = 0.31) and age (F = 7.10, p < 0.012, η² = 0.18). Alcohol use and personality measures did not predict caudate activation.

Table 2. Activation to Social Influence.

Area	HEM	Region	x	y	z	z-stat	VOL
Activation in CONTROLS
Occipital	R	Lateral occipital cortex	28	-76	28	3.76	61
	R	Occipital cortex	30	-86	-2	3.74	118
	L	Lateral occipital cortex	-26	-90	-2	4.42	204
Activation in MJ Participants
Frontal	L/R	Rostral cingulate zone	6	48	32	4.60	853
	L/R	Medial frontal cortex	4	44	-14	3.74	52
	R	Orbitofrontal cortex	46	30	-8	3.95	143
	R	Inferior frontal gyrus	52	22	14	4.03	142
	L	Inferior frontal gyrus	-48	10	20	4.07	395
	L	DLPFC	-46	0	36	3.92	123
Parietal	L/R	Precuneus	4	-56	24	4.75	582
Occipital	L/R	Lingual Gyrus	-4	-76	-2	3.81	138
	L	Occipital Pole	-30	-94	-4	3.82	115
Subcortical	R	Caudate	12	10	8	4.53	208
	L	Caudate	-14	12	8	3.79	125
	L	Insula	-34	20	0	3.87	100
MJ > CONTROLS
Subcortical	R	Caudate	12	10	6	3.35	95
	R	Caudate	6	2	2	3.10	16

Z (Gaussianised T/F) statistic images were thresholded using clusters determined by Z>2.3 and a (corrected) whole-brain cluster significance threshold of p<0.05. HEM represents hemisphere. Coordinates are in MNI space. VOL = volume, in number of voxels (2 × 2 × 2 mm³).

Fig 3.

Regions Associated with Social Influence in CON and MJ. The contrast of social influence compared no influence activated a wide network of structures in both groups (see left-most figure for main effect of influence across groups). Notably, the caudate only activated in the MJ group. The difference image (right-most figure) demonstrates that MJ showed significantly greater activation than CON to social influence in the right caudate. All clusters shown were thresholded using clusters determined by Z>2.3 and a (corrected) whole-brain cluster significance threshold of p<0.05.

Activation to Choice During Congruent vs Incongruent Decisions (Whole-Brain Analysis)

These brain imaging results refer to changes in activity during Event 3, Choice, in which participants could choose to either follow the group's recommendation (congruent) or go against the group response (incongruent). Four participants in the CON group and 4 participants in the MJ group were removed from this analysis because they had too few incongruent responses (e.g. 5 or less). Across groups, participants showed greater activation when making congruent choices compared to incongruent choices in numerous brain regions, including the bilateral anterior and posterior cingulate, frontal medial cortex, as well as in occipital regions. Across groups, participants showed greater responses to incongruent than to congruent choices in frontal regions (right middle frontal gyrus and right frontal pole), and bilateral lateral occipital cortex. (Table S2). There were no differences between CON and MJ in activation to congruent or to incongruent decisions.

Correlations Between Self-reported Peer Conformity and Brain Activation

To investigate whether self-reported peer conformity correlated with brain activation, data were extracted from the ROIs in the caudate and the NAc of each participant (Fig. 4A), and regressed against scores on the MISS, as well as self-reported task behavior. Self reported tendency toward peer conformity, as assessed with the MISS, correlated with activation in the caudate in Event 2, Influence (minus no influence), across groups on both the left (r² = 0.17, p = 0.006) and right (r² = 0.25, p < 0.001) (Fig 4B, top). There were no differences in slopes between MJ and CON, indicating that both groups showed similar effects, thought there was a group difference in magnitude of caudate activation (t = 2.89, p = 0.006) (Fig 4C, top). Self-reported peer conformity scores did not correlate with NAc activation. To explore whether other brain regions also correlated with self-reported peer conformity, we ran an additional analysis in which peer conformity scores were regressed into the whole-brain random effects analysis. The correlation between caudate activation and peer conformity revealed several interesting clusters, including the bilateral caudate and the bilateral NAc but these clusters were generally small and did not survive whole-brain correction (see Tables S3 and Fig S1).

Post-scan questionnaire responses correlated with actual task behavior, as well as activation in the NAc in response to choice behavior. There was a strong relationship between self-reported likelihood of going along with the group on the task, and actual number of congruent responses on the task (r² = 0.45, p = 0.002). There was a significant correlation across groups so that those who reported that they were most likely to go along with group influence showed greatest NAc deactivation when making incongruent choices, particularly on the right (r² = 0.14, p = 0.015) (Fig 4B, bottom). This correlation indicated that those who reported they were most likely to go along with the group had greatest NAc deactivation when making the choice to go against the group. These questionnaire ratings did not correlate with caudate activation. An ROI-based comparison between groups did not show significant differences in NAc activation (Fig 4C, bottom). An additional whole-brain analysis in which questionnaire scores were regressed into the whole-brain random effects analysis did not reveal any significant or sub-threshold clusters. There were no significant group differences in NAc activation.

Correlations Between Reaction Time and Brain Activation to “Choice” Event During Incongruent Choices

In order to test the hypothesis that increased reaction time in the MJ group was related to increased effortful processing, we ran an additional analysis in which reaction time was regressed into the whole-brain model for the condition of incongruent choices (Table 3). We found that both groups showed significant correlations between reaction time and activation, particularly in frontal regions such as the bilateral DLPFC (in CON) and the IFG (in MJ), and well as in parietal and occipital regions (Fig 5). In a direct comparison in which we tested whether the linear relationship between Event 3 (Incongruent Choice) and reaction time differed between groups, we found a significant interaction in the left IFG, as well as in other frontal, parietal, and occipital regions. This interaction indicated that the slope of the correlation between activation and reaction time within the MJ group was steeper than that in the CON group. Reaction time did not correlate with either self-reported peer conformity, or with brain activation in the caudate.

Table 3. Brain Activation to Event 3, Choice (Incongruent Responses) Correlated with Reaction Time.

Area	HEM	Region	x	y	z	z-stat	VOL
Activation in CONTROLS
Frontal	L	DLPFC	-50	2	44	3.98	1718
	R	DLPFC	48	6	36	3.4	513
Parietal	L	Superior Partietal Lobule	-30	-44	48	4.1	2341
Occipital	R	Occipital Cortex	28	-76	30	3.38	1200
Activation in MJ Participants
Frontal	L/R	Superior Frontal Gyrus	4	50	36	3.61	1496
	L	Inferior Frontal Gyrus	-38	22	18	3.56	631
	R	Inferior Frontal Gyrus	52	22	14	3.42	393
	R	Orbitofrontal cortex	46	30	-8	3.45	457
Parietal	L/R	Precuneus	4	-58	24	4.04	618
MJ > CONTROLS
Frontal	L	Inferior Frontal Gyrus	-42	28	14	3.37	532
	L	Superior Frontal Gyrus	-14	-4	76	3.43	399
Parietal	L	Supramarginal Gyrus	-36	-46	42	3.57	1271
Occipital	L	Lateral Occipital Cortex	-50	-68	16	3.49	541
Subcortical	R	Insula	32	8	12	3.31	518

Z (Gaussianised T/F) statistic images were thresholded using clusters determined by Z>2.3 and a (corrected) whole-brain cluster significance threshold of p<0.05. HEM represents hemisphere. Coordinates are in MNI space. VOL = volume, in number of voxels (2 × 2 × 2 mm³).

Fig 5.

Both groups showed significant correlations between reaction time and activation, particularly in frontal regions such as the bilateral DLPFC (in CON) and the IFG (in MJ), and well as in parietal and occipital regions. In brain regions such as the left IFG, the correlation between activation and reaction time within the MJ group was stronger than that in the CON group. All clusters shown were thresholded using clusters determined by Z>2.3 and a (corrected) whole-brain cluster significance threshold of p<0.05.

Discussion

Though decades of research support the importance of social influence in the initiation and maintenance of drug use, the literature on the neural correlates of social influence in the context of substance use is sparse. Here we report results from a novel experimental approach to evaluating the effect of social influence on both behavioral susceptibility to influence, and neural activation to influence. Both MJ users and controls were more likely to follow than oppose group recommendations, at the expense of accuracy, but only MJ showed choice reaction time slowing when going against the group, reflected in increased IFG activation correlating with reaction time, suggesting more effortful/contemplative processing. The MJ group also exhibited greater caudate activation than controls when presented with peer information/social influence. The caudate, which is a region that is particularly affected by MJ use because of its high density of cannabinoid receptors (see (Goodman & Packard, 2015) for review), is involved in a variety of functions, from social reward to goal-directed behavior, the selection of correct actions, and behavioral control. This hyperactivation in the MJ group could indicate that the reward regions of the brain may be more responsive to social information in MJ users than among non-users, or alternatively, that the MJ users are cognitively compromised such that greater recruitment of the caudate is needed to inform the selection of the correct action during cognitively demanding social influence trials. Across groups, caudate activation was associated with self-reported peer conformity, providing evidence that caudate activation is a core component of the neural circuitry underlying social influence, and that this circuitry may be hyper-activated in late adolescent/young adult MJ users.

Though we hypothesized that MJ users would be more likely than controls to follow group information, all participants were more likely to follow group opinions, with no between-group differences in decision-making. However, only the MJ group had slower reaction time when making choices that were incongruent with social influence. Reaction time for decision-making can be seen as the time from deliberation to choice selection, and deliberation usually takes a longer time when a decision is harder (Lo & Wang, 2006). Previous work using a similar social decision-making task found that reaction times were slowest for trials in which the social influence was “incongruent” with the expected choice and fastest for “congruent” trials (Gilman, Treadway, Curran, Calderon, & Evins, 2015). Critically, in this previously published work, individuals who showed a greater bias towards conformity also showed larger reaction time slowing effects during incongruent trials.

There are several potential explanations for the longer reaction times in the MJ group during incongruent trials. The differences in reaction time with an absence of overt choice differences may indicate that although the MJ group was capable of coming to the same decisions as the control group in a laboratory environment, it may be more effortful for young adult MJ users to oppose group influence. Indeed, prior studies in rodents have demonstrated that various drugs of abuse can impact the ability to exercise cognitive control when making decisions (e.g., (Cocker, Hosking, Benoit, & Winstanley, 2012)). This is the first study to our knowledge to investigate this effect in humans with social influence and in the context of MJ use. Decision speed is modulated by cognitive variables such as attention and working memory (Ester, Ho, Brown, & Serences, 2014; Nunez, Srinivasan, & Vandekerckhove, 2015; Whitney, Rinehart, & Hinson, 2008) and it has been widely demonstrated that MJ use is associated with these domain-specific deficits (Colizzi et al., 2015; Jacobus et al., 2015; Price et al., 2015). An alternative explanation is that longer reaction time during the incongruent trials could indicate that the MJ users experience impaired cognition associated with identifying and/or resolving information conflicts. As all participants were less confident during incongruent trials, it is likely that all participants were engaging in cognitive processing of identifying error detection/conflicting information during these trials, skills that rely largely on the function of the caudate (Davidson et al., 2004; Delgado, Miller, Inati, & Phelps, 2005; Haruno & Kawato, 2006; O'Doherty et al., 2004; O'Doherty, 2004). It is therefore possible that the MJ users had greater difficulty coming to a decision during these conflict trials than did control participants. In real-life situations, this “extra” effort required to make decisions that go against influence could potentially confer a greater risk of suboptimal decision-making regarding. Future studies will need to examine whether the tendency to slower processing when making decisions that go against group influence confers vulnerability to initiate and continue using MJ as well as to engage in other risk behaviors. The neuroimaging results showing increased correlation between reaction time and activation of the IFG in the MJ group compared to the CON group supports that hypothesis that resisting peer influence may be more effortful in the MJ group (Swick et al., 2008).

In addition to reaction time differences, we also found group differences in brain activation during presentation of social influence. MJ users but not controls showed activation bilaterally in the caudate. The dorsal striatum, including the caudate and the putamen, has a prominent role in decision-making. While the ventral striatum has been implicated in passive receipt of reward (e.g. drug reward (Gilman, Ramchandani, Crouss, & Hommer, 2012; Gilman, Ramchandani, Davis, Bjork, & Hommer, 2008), and monetary reward (O'Doherty et al., 2004), the dorsal striatum is thought to mediate important aspects of decision-making related to goal-directed action, and underlie the selection of actions on the basis of their expected reward value. Studies suggest that the caudate nucleus in particular is involved in coding reward-prediction errors during goal-directed behavior (Davidson et al., 2004; Delgado et al., 2005; Haruno & Kawato, 2006; O'Doherty et al., 2004; O'Doherty, 2004), and even more specifically, behavior regarding complex social issues such as social cooperation (Rilling et al., 2002), revenge (de Quervain et al., 2004), and the acquisition of social reputations (King-Casas et al., 2005). In the current study, the hyperactivation of the caudate during the receipt of social information could indicate that perhaps the MJ group was particularly responsive to social information.

Data from this study support the notion that the caudate is important for social decision-making and may be sensitive to individual differences. Across groups, activation of the caudate positively correlated with self-reported peer conformity on the MISS, indicating that activation of the caudate is associated with social reward, particularly in those who value group conformity. Prior studies have suggested that valuation of social stimuli relies on mesocorticolimbic circuitry that largely overlaps with that of other reinforcers (Jones et al., 2011; Klucharev et al., 2009). Social conformity may serve as an additional value signal that is incorporated with other standard reinforcement parameters (e.g., effort, reward magnitude, probability) to derive a single subjective value for a given option (Kable & Glimcher, 2009; Sescousse, 2014). Greater activation in the MJ-using group may indicate that the neural circuitry within this group is particularly sensitive to social information.

Though the right caudate was the only region that showed significantly greater activation in the MJ than CON to peer influence after whole-brain correction, it is important to point out that several other brain regions were more highly activated in the MJ group, including the IFG, orbitofrontal cortex, and anterior cingulate. These regions, particularly the IFG, are important in cognitive control of decision-making, specifically in inhibitory control and task-switching, (Aron, Robbins, & Poldrack, 2004), domains that may be impacted by MJ use (e.g. (Gruber & Yurgelun-Todd, 2005). Longer reaction times in the MJ group compared to controls, as well as the observation that the MJ group demonstrated a stronger correlation between reaction time and activation of the IFG, could indicate that the MJ users experiencing more difficulty with error monitoring/cognitive control during the task. Though differences in activation of these regions did not survive whole-brain correction in the current study, they nonetheless could be followed up in future studies with larger sample sizes.

We assessed participants' self-reported propensity to follow group decisions and found a correlation such that those who reported high susceptibility to influence (e.g. that were likely to follow group decisions) had greater deactivation of the NAc in response to incongruent choices. These results are interesting in light of results from previous studies (Klucharev et al., 2009; Zaki et al., 2011) showing that agreeing with peers increased activity in the NAc, whereas disagreeing decreased NAc activity. In a landmark study of social influence in which participants were asked to rate attractiveness of faces before and after learning about how peers had rated the faces, deactivation of the NAc during a conflict with group opinion predicted conformity in future trials (Klucharev et al., 2009). We found this decreased NAc activity to disagreeing with the group was strongest among those who reported that they were likely to follow the group. Though there are differences in the temporal sequence of decision-making between the Klucharev paradigm, in which NAc deactivation occurred in response to finding out that peer opinion conflicted with one's already submitted choice, and the current study, in which participants saw peer judgments before making their choices, the current finding replicates research showing that activation of the NAc is an important signature of conformity vs independence, and may be sensitive to individual differences in peer susceptibility among both MJ users and non-users. This study also suggests that the NAc may be sensitive to making/implementing choices to conform, whereas the caudate may be more sensitive to the receipt of differentially salient peer information. The implication of this distinction may be that while peer information itself can activate social goals and relevant behavior, actual choice may be more related to hedonic value.

There are several caveats to this study. First, though MJ users and CON showed differences in activation of the caudate, this brain difference occurred in the absence of a behavioral difference in susceptibility to influence during the social influence task. There are several explanations for why brain-based differences may be observed despite lack of differences in task performance (for commentary, see (Gilman, Bjork, & Wilens, 2015), but most notably, it is important to acknowledge that scanner-based experimental psychology tasks represent grossly simplified traces of complex real-life behaviors. Our task investigates social influences on decision-making, during a low-arousal task where the outcome (comparing the relative length of lines) is not particularly meaningful to the participants. Social influences that occur in the context of drug use are likely to be high-arousal, personally meaningful, and require complex balancing of costs and benefits to following or dissenting from peer opinion. The difference in brain signal could indicate that in the artificial scanner environment, MJ users are able to exhibit normative decision-making patterns, but may be more challenged relative to healthy controls in real-life settings involving peer influence. Future studies can employ tasks more relevant to drug-seeking and drug-using behavior, such as risk-taking and motivation tasks, to further probe whether more effortful tasks would reveal behavioral differences as well better understand the predictive utility of slowed decision making in understanding real-world functional outcomes.

A further limitation of this study was that because of the large number of experimental conditions of the task in order to create a balanced design, the number of trials per condition was relatively low (44 trials total, with 28 active “influence” trials), and this may not have been optimal for revealing statistically robust effects. Future studies can include either more task runs, or fewer conditions, in order to maximize statistical effects and minimize the chance of a Type II error.

An additional limitation is that the criteria for eligibility into the MJ group required only weekly or more use. Though weekly or more use is indicative of a regular pattern of use, studies of heavier users may better capture brain-based correlates of MJ use. Furthermore, though the focus of the current study is on MJ use specifically, young adults using other substances such as alcohol or nicotine may show similar brain-based hyperactivity to peer influence. Though alcohol use did not contribute to the results reported in the current study, all participants were fairly light drinkers. Future studies could recruit heavy drinkers/cigarette smokers to test whether these individuals also show hypersensitivity in the caudate to peer information, or whether this is an effect of MJ specifically. An additional avenue for research is to examine the role of educational attainment on susceptibility to influence; in the current study, most participants were currently college students or recent college graduates, but it is possible that in addition to age, level of education would influence susceptibility. Future studies could also examine how different motivations for drug use (e.g. coping, social engagement, etc.) contribute to different patterns of brain activity underlying social influence and drug use.

Finally, this study cannot detect whether increased neural activation to influence is a cause or an effect of drug use. There are several possible explanations of the relationship between MJ use and susceptibility to social influence; 1) Increased susceptibility to social influence is a trait that reflects a pre-existing risk for initiation of MJ use, 2) susceptibility to social influence developed as a result of MJ exposure due to disruption to neurodevelopmental processes underlying social processing, 3) a third variable is responsible for giving rise to both MJ use and susceptibility to social influence, such as striatal reactivity, or 4) some combination of the explanations above is responsible for of the relationship between MJ use and susceptibility to social influence. Although the current study cannot differentiate among these explanations, this study may help researchers to develop hypotheses for future work addressing causality.

Identifying brain regions that differ in response to social influence may influence treatments such as lifestyle management, which could encourage MJ users to find alternative positive reinforcement in the natural environment. Many studies have found that social network support is a critical factor in promoting healthier lifestyle behaviors in addition to ongoing abstinence (e.g. (Bond, Kaskutas, & Weisner, 2003; Kaskutas, Bond, & Humphreys, 2002)). If marijuana users are indeed more susceptible to peer influence, future studies can investigate whether positive peer influences may be particularly effective as an intervention in this group.

Supplementary Material

13415_2016_421_Fig6_ESM

Fig S1. Correlation between brain activation and peer conformity. When peer conformity scores were regressed into whole-brain analysis, uncorrected maps revealed clusters in the bilateral caudate and the bilateral NAc. Clusters shown were thresholded using clusters determined by Z>2.3 and an uncorrected voxel significance threshold of p<0.01.