Rationale: Evidence suggests that abnormalities in reward processing and impulsivity contribute to the pathophysiology of addiction. However, the relationship between the two is currently not well characterised. This study used functional magnetic resonance imaging during reward and inhibitory control tasks as well as subjective measures of impulsivity to investigate the association between reward and impulsivity.
Methods: Abstinent alcohol (AD, n=27), polydrug (PD, n=57) dependent and healthy control (HC, n=65) participants were recruited as part of the ICCAM platform study. Participants completed a battery of subjective and behavioural measures of impulsivity; Barratt Impulsiveness Scale, UPPS-P impulsive behaviour scale, the Kirby Delay Discounting task and the Stop Signal Task. Participants also completed a Monetary Incentive Delay and Go/No-go task during functional magnetic resonance imaging. An a-priori region of interest approach was used to image blood-oxygen level dependant response during the following contrasts: no-go>go, reward anticipation>neutral anticipation and reward outcome>neutral outcome. Additional whole brain analyses were also conducted.
Results: Abstinent alcohol and polydrug dependent individuals scored higher than healthy controls across self-report and behavioural measures of impulsivity. However, no group differences were found in BOLD response during the GNG task. No differences in impulsivity between addiction groups were found. BOLD response during reward anticipation and outcome was significantly blunted in prefrontal areas. Such blunting during reward anticipation was found to be associated with BOLD response during the GNG task, while blunted response during reward outcome correlated with subjective measures of impulsivity.
Conclusions: The overall findings of this study suggest that reward and impulsivity are associated and that this association can become dysregulated in addiction. The results infer that reward anticipation is associated with response inhibition and may be subject to change throughout the trajectory of addiction. Meanwhile, reward outcome was found to be better associated with trait impulsivity suggesting it may remain relatively stable. Subsequent analysis is needed to explore how such associations change as a function of abstinence and relapse.
Word count:
1. Introduction
Drug addiction is a chronic relapsing condition that affects around 10% of those who are exposed to drugs (Volkow and Morales, 2015). Hallmarks of the disorder include compulsive drug taking associated with an inability to control intake irrespective of the negative consequences to oneself or others (Koob and Volkow, 2010). It is estimated that around 3.6% of the UK population are alcohol dependent (Drummond et al., 2016) with a similar percentage estimated to suffer from drug dependence (Roberts et al., 2016), although there is likely a large degree of overlap between these groups with polydrug addiction being common. Only around one third of those affected are estimated to be in contact with treatment services and of these, less than 50% complete treatment dependence free (National Treatment Agency and Department of Health, 2017). Furthermore, even for those who achieve abstinence following treatment the risk of relapse remains high (Neto, Lambaz and Tavares 2007). It is therefore imperative that we better understand the neural mechanisms and biomarkers of the disorder and subsequent relapse in order to develop novel and effective treatment strategies.
1.1. The Reward System
In 1954 Olds and Milner found that rodents will self-administer electrical stimulation to certain areas of the brain but not others. They found that direct stimulation to these areas was more appealing to the rodents than eating, drinking or sex. The results of this study were the first compelling evidence to suggest that ‘pleasure centres’ exist within the brain (Olds and Milner, 1954). In light of this finding, subsequent research has identified structures, pathways and neurotransmitters that all contribute to what has since become known as the reward system. This system is responsible for goal-directed behaviour through reinforcement and is sensitive to stimuli related to survival such as food, sex and water. Activation of the reward system is associated with feelings of pleasure, enjoyment and arousal (Berridge and Kringelbach, 2008). The reward system is characterised neurochemically by the dopaminergic mesolimbic pathway. Originating in the ventral tegmental area (VTA) of the midbrain, dopaminergic projections extend through the mesolimbic pathway and innervate brain regions such as the nucleus accumbens (NAc), ventral striatum (VS), amygdala and prefrontal cortex (PFC). Following exposure to a rewarding stimulus the mesolimbic pathway becomes activated leading to an increase in the concentration of extracellular dopamine. This mechanism underlies the reinforcing properties of reward (Adinoff, 2004, Jentsch et al., 1999).
Identifying the importance of dopamine in the mesolimbic pathway, pre-clinical studies have shown that dysregulation of dopamine signalling or dopamine deficiency is sufficient enough to inhibit goal-directed behaviours such as feeding in rodents (Sotak et al., 2005, Cannon et al., 2004). However, rather than directly inducing the feeling of pleasure, recent work has found that mesolimbic dopamine not only increases following receipt of reward but also increases following prediction of a reward by a conditioned stimulus. It is through this prediction response that dopamine is capable of coding for prediction error. Reward prediction error evaluates the difference between predicted and received reward and is important in driving goal-directed behaviour. Briefly, in response to a greater reward than predicted, dopamine neurons are hyperactivated giving rise to a positive prediction error. The opposite is true for rewards that are less than predicted, in which case a negative prediction error is formed. By this mechanism the dopaminergic mesolimbic pathway is able to preferentially drive behaviour that is associated with a positive prediction error (Schultz, 2016). Whilst this system is crucial for survival it is also susceptible to pharmacological manipulation by drugs of abuse and has been identified as an important neural substrate for the manifestation of addiction.
1.1.1. Transition from Reward to Addiction
While drugs such as opiates do have non-dopaminergic reward pathways, dopamine is often considered the central neurotransmitter in drug addiction (Nutt et al., 2015). Drugs of abuse are able to ‘hijack’ the reward system by pharmacologically increasing the concentration of extracellular dopamine both directly and indirectly (Kandel, Schwartz and Jessell, 2000). Stimulants such as cocaine and amphetamines both act to directly increase dopamine concentration in the synapse. Cocaine blocks the presynaptic dopamine transporter (DAT) while amphetamines also increase release rate from synaptic vesicles. In contrast, other drugs such as opiates, alcohol and nicotine indirectly increase extra-synaptic dopamine by inhibiting gamma-aminobutyric acid (GABA) release onto VTA neurons. This results in an increase in the firing rate of dopaminergic neurons (Adinoff, 2004).
Through their pharmacological actions, drugs of abuse are able to increase dopamine to levels in excess of those induced by conventional rewards (Dawe, Gullo and Loxton, 2004). This increase occurs in the absence of a conventional reward prediction allowing drugs of abuse to always induce a positive reward prediction (Schultz, 2016). As drug taking transitions from experimental use to drug addiction, cues that are associated with the drug become conditioned stimuli, gaining salience and reinforcing drug-taking behaviour and craving (Beck et al., 2009, Koob and Volkow, 2016).
1.1.2. Reward Processing in Addiction
Since identifying the importance of the dopaminergic mesolimbic reward pathway in drug addiction, neuroimaging has allowed us to further advance our understanding of theunderlying mechanisms. Functional magnetic resonance imaging (fMRI) measures Blood Oxygen Level Dependant (BOLD) signal change and is often used in drug addiction research. Several fMRI paradigms exist to investigate reward processing; one of the most effective and commonly used is the monetary incentive delay (MID) task (Knutson et al., 2000). Designed to probe reward sensitivity, the task requires the individual to respond to a cue with a button press in order to win money. There are three types of trial; win, neutral or loss, with the individual being made aware of the trial type prior to the appearance of the cue. The task elicits two neuronal processes; anticipation and outcome, with anticipation during win trials being commonly explored in drug addiction research. A common finding in which is a blunted BOLD signal in the VS during MID reward anticipation (Murphy et al., 2017, Nestor et al., 2017, Rose et al., 2013, Beck et al., 2009). In response to drug related cues, increased BOLD signal is observed in the VS in drug addiction individuals compared with controls and is often associated with increased craving (Myrick et al., 2004, Chase et al., 2011, Wrase et al., 2002, Grüsser et al., 2004). Taken together, these findings suggest that in drug addiction the reward system attributes excess salience to drug related cues whilst reducing the response to conventional cues such as money.
However, there are inconsistencies in the literature. Nestor et al. (2017) reported that in drug addiction BOLD response during MID reward anticipation in the VS did not correlate with behavioural changes following pharmacological challenge with naltrexone. This finding highlights the importance of probing alternate brain regions which may contribute to reward processing. Reward outcome can also be analysed during the MID task. Interestingly, Jia et al. (2011) found increased BOLD signal in the VS in cocaine dependent individuals compared to healthy controls during reward outcome and that this was negatively associated with future abstinence. Inconsistencies in the literature and disparities between reward anticipation and reward outcome highlight a pressing need for further analysis of reward processing in drug addiction.
One theory is that when drug taking transitions from experimental to compulsive use there is a shift in neural processing of reward from VS to dorsal striatal (DS) control (Everitt and Robins, 2005). In support of this hypothesis Vollstädt‐Klein et al. (2010) presented alcohol-related cues to both light non-dependent drinkers and heavy drinkers, of which two thirds were alcohol dependent. They revealed that increased BOLD signal was present in the VS of the light drinkers, although in the heavy drinkers BOLD signal was higher in the DS. Several studies analysing BOLD signal response to conventional, non-drug related rewards have identified blunted activation in the DS in drug addiction individuals (van Hell et al., 2010, Nestor et al., 2017, Wrase et al., 2007). These findings suggest that research into the dysregulation of reward processing in addiction may benefit from further exploring cue-response in the DS.
Despite a wealth of literature exploring the striatum, reward processing is not exclusively modulated by striatal activity. Areas of the frontal cortex such as the orbitofrontal cortex (OFC) and inferior frontal gyrus (IFG) are functionally connected with the striatum and form the fronto-striatal pathway (Jollans et al., 2016, Jarbo and Verstynen, 2015). Jollans et al. (2016) demonstrated in adolescent smokers that during the reward anticipation phase of the MID task, decreased functional connectivity was observed between the VS and IFG and this was associated with a blunted BOLD signal response in the VS. Prefrontal areas such as the OFC and IFG are also known to have a role in top-down executive control (Aron et al., 2003, Torregrossa, Quinn and Taylor, 2008). The medial OFC (mOFC) is responsible for predicting outcome expectancy and attributing salience to reward which in turn drives goal-directed behaviour (Volkow and Morales, 2015, O’Doherty, 2004). In response to an acute administration of heroin, activation in the OFC is positively associated with the subjective urge to use the drug and is thought to sustain goal directed behaviour (Sell et al., 2000, Goldstein and Volkow, 2011). It has also been shown in drug addiction individuals that activation in the OFC is associated with the expectancy of receiving a salient drug (Kufahl et al., 2008).
1.2. Modelling Addiction
The involvement of prefrontal areas such as the OFC and IFG in reward processing suggests a much larger network of modulation. Incorporating the multiple aspects of neural processing Koob proposed a comprehensive three-stage model identifying three recurring stages that characterise drug addiction (Koob and Volkow, 2010). The initial stage of the model is the binge/intoxication phase that is governed by reward processing and incentive salience in the striatum. This is followed by a withdrawal/negative affect stage associated with low mood and emotional stress in response to reward deficit and is mediated by dysregulation of the amygdala. The final stage of the model is the preoccupation/anticipation phase which manifests as craving due to deficits in prefrontal areas, responsible for inhibitory control. It is believed that in the early stages of drug abuse positive reinforcement of drug consumption concurrent with impulsivity drives this cycle. When the transition to addiction is more robust, drug taking becomes more compulsive and negative reinforcement becomes the driver of drug seeking and consumption (Koob and Volkow., 2010, Goldstein and Volkow, 2011). This model, a simplified version of which is illustrated in Figure 1, highlights the importance of considering addiction as a multi-faceted condition. Future work that aims to probe the interaction between these pathways in addiction may facilitate the development of novel, effective therapeutics.
1.3. Impulsivity
Impulsivity is a term used to describe a lack of inhibitory control characterised by reckless behaviour in the absence of pre-meditation. It has multiple domains including choice, trait and impulsive action (Mitchell and Potenza, 2014). Increased impulsivity has been associated with an increased risk of drug addiction in those who suffer from impulse control disorders such as attention deficit hyperactivity disorder (ADHD) (Biederman et al., 1997, Dalley, Everitt and Robbins, 2011). With some theories suggesting that a robust inhibitory control system may allow an individual to overcome drug addiction and successfully maintain abstinence, impulsivity has been identified as an important element of addiction (Adinoff, 2004). However, while dysregulation of the reward system in drug addiction is well characterised, the neural mechanisms that underpin impulsivity are not as well defined (Volkow, Koob and McLellan, 2016).
1.3.1. Impulsivity in addiction
Subjective measures of impulsivity such as the Barratt Impulsiveness Scale (BIS-11) and the Urgency, Premeditation, (lack of), Perseverance (lack of), Sensation Seeking, Positive Urgency, Impulsive Behaviour Scale (UPPS-P) are often used to measure trait impulsivity with drug addiction individuals scoring higher compared with healthy controls (Trifilieff and Martinez, 2014, Semple et al., 2005, Taylor et al., 2016, Xie et al., 2011). These results have been replicated using behavioural impulsivity measures such as the Kirby Delay Discounting task, used to measure choice impulsivity. Delay discounting tasks model impulsivity as a preference for smaller immediate rewards over larger delayed rewards (Mitchell and Potenza, 2014). Individuals with drug addiction have consistently been found to have increased delay discounting, meaning they will more often chose a smaller immediate reward compared with controls (Kirby et al., 1999, Coffey et al., 2003). Addiction related deficits in cortical and subcortical activation in response to delay discounting decision making have been located in the caudate, anterior cingulate cortex (ACC), OFC and dorsolateral prefrontal cortex (DLPFC) (Hoffman et al., 2008, McClure et al., 2004). Moreover, trait and choice impulsivity is thought to be suggestive of polydrug compared to single drug addiction (Semple et al., 2005, Taylor et al., 2016).
To probe impulsive action the Go/no-go (GNG) and Stop Signal Task (SST) are common paradigms in fMRI studies. Both tasks require participants to respond to a repetitive baseline of ‘go’ cues by pressing a button. Intermittent ‘no-go’ or ‘stop’ cues are presented to the participant in response to which they must inhibit their action. The GNG is designed to measure response inhibition prior to a response being initiated while the SST measures inhibition of an action following initiation (Dalley, Everitt and Robbins, 2011). In healthy volunteers, successful response inhibition in the GNG has been associated with activation in prefrontal regions including the ACC, OFC and DLPFC (Steele et al., 2013). GNG fMRI studies have also been conducted in drug addiction, successfully identifying potential neural substrates of impulsivity and relapse in addiction. Compared with healthy controls, abstinent cocaine addicts have been found to have increased BOLD signal in the right IFG (rIFG), superior frontal gyrus (SFG) and medial prefrontal gyrus (MFG) during successful inhibition in the GNG (Connolly et al., 2012). In contrast, active cocaine addicts have been found to experience hypoactivation in these ROIs associated with poorer accuracy (Hester et al., 2004). These finding suggest that prefrontal areas are recruited more in abstinent compared with active drug addiction. This may be a compensatory mechanism that aids successful abstinence. However, GNG fMRI outcomes in addiction are inconsistent with several studies reporting no addiction related differences in prefrontal activation (Taylor et al., 2016). Furthermore, a clear relationship between the different domains of impulsivity is yet to be established thus contributing to discrepancies (Dalley, Everitt and Robbins, 2011). More research is needed to determine the neural mechanisms of impulsivity and how such findings may contribute to the clinical presentation of addiction including risk of relapse.
1.4. Association Between Reward and Impulsivity
Associations between impulsivity and reward have been identified (Goldstein et al., 2002). Subjective impulsivity scores (BIS-11) are negatively associated with MID BOLD signal change in the VS during reward anticipation. (Beck et al., 2009, Murphy et al., 2017). However, to the best of our knowledge, there is an absence of evidence associating BOLD signal activation in the MID and GNG tasks. One explanation for this is that previous studies have attempted to correlate the activity of two different ROIs rather than investigating the activity of a region common to both tasks.
Several anatomical brain regions have previously been found to be functionally flexible across the MID and GNG tasks, suggesting related activity in reward and impulsivity (McGonigle et al., 2017). For example, activation in the OFC has been shown to play a role in conditioning and attributing salience (Hayashi et al., 2013). In addition OFC activation is often implicated in impulsivity (Goldstein and Volkow, 2011). Likewise, activation in the IFG is commonly associated with inhibitory control and has also been suggested to have a role in integrating sensory information with goal directed behaviour (Hampshire et al., 2010). Both the OFC and IFG are functionally connected to the striatum (Jollans et al., 2016, Jarbo and Verstynen, 2015). Combined with their common activation in both the MID and GNG tasks, these two regions are of interest when investigating the relationship between reward and impulsivity to better understand the neural mechanisms of addiction.
1.5. Aims of this Study
The Imperial College, Cambridge and Manchester (ICCAM) consortium initiated a multi-centre brain imaging platform study (Paterson et al, 2015). The primary aim of the ICCAM study is 1) to identify the neural circuitry and potential biomarkers of alcohol and polydrug addiction and 2) identify novel therapeutics for addiction. In this study we present the results of the baseline session. Using an a prioi ROI approach we aim to probe neural mechanisms of reward and impulsivity and the relationship between the two in alcohol and polydrug addiction. Based on previous evidence, this study also aims to determine if differences in reward and impulsivity exist between alcohol and polydrug addiction. The caudate, OFC and rIFG will be explored during the MID (Knutson et al., 2000) and GNG tasks (Garavan et al., 2002). The caudate was chosen based on the evidence that reward processing shifts from VS to DS when drug taking transitions from experimental to compulsive use. In addition, the current literature probing reward related activation in the VS is inconsistent. The OFC and rIFG were chosen based on their common activation in both the MID and GNG and their functional connectivity to the striatum.
We hypothesise that subjective measures of impulsivity will be higher in drug dependence compared with healthy controls with the greatest scores seen in polydrug dependence. Likewise, we hypothesise that activation during successful response inhibition in the GNG will be increased in drug dependence owing to an increased activation being protective for risk of relapse. We also expect to see blunted activation during the MID task in drug dependence. Finally, we expect to see a negative correlation between reward and impulsivity measures, with this association being strongest in the polydrug group owing to increased impulsivity and thus more inhibitory control effort needed to maintain successful abstinence.
2. Materials and Methods
2.1. Participants
Abstinent alcohol (AD, n=27), polydrug (PD, n=58) dependent and healthy control (HC, n=65) participants were recruited as part of the ICCAM platform study (Paterson et al., 2015). Participants who met the Diagnostic and statistical manual of mental disorders. 4th edition (DSM-IV) criteria (American Psychiatric Association, 2000) for at least one of the following dependencies; alcohol, cocaine or opiate and were currently abstinent and not intoxicated or in withdrawal were included. Participants were required to have a negative breath alcohol and urine drug test (including amphetamines, barbiturates, cocaine, opiates, cannabinoids and benzodiazepines). Healthy controls without history of dependence outlined by the criteria above were matched where possible for age, gender and smoking status. No participant had a current psychiatric diagnosis. Further inclusion and exclusion criteria are outlined in Paterson et al. (2015). This study was conducted in accordance with the Decleration of Helsinki. Ethical approval was obtained from West London & GTAC NRES committee (11/H0707/9). Relevant Research Governance and PIC (Participation Identification Centre) approvals were also obtained.
Three participants (two HC, one PD) were excluded from the MID task analysis. One was excluded for excessive motion (>1.00mm/s in at least one run, detected prior to this study using the Artifact Detection Toolbox (ART) (https://www.nitrc.org/projects/artifact_detect/)) and two for unusually fast reaction times, suggesting repetitive button pressing. Three participants (two HC, one AD) were excluded from the GNG task analysis for achieving a baseline accuracy of less than 80%. Therefore a total of 149 participants were included in the final analysis (65 HC, 27 AD, 57 PD)
2.2. fMRI Task Protocols
Participants performed GNG and MID tasks to measure neural impulsivity and reward processing, respectively. Participants familiarised themselves with these tasks prior to scanning in order to minimize learning effects. The tasks were presented to participants in the same order.
2.2.1. GNG task
The GNG task adapted from Garavan et al, (2002) uses an event-related design to measures suppression of a response that has not yet been initiated. Participants were shown a series of letters (X’s and Y’s) in an alternating fashion. Each letter was displayed for 900ms and was followed by an interval stimulus (blank screen) for 100ms. Participants were required to press a button upon seeing a letter that was different to its predecessor this was defined as a ‘go’ trial. A no-go trial occurred when the letter was the same as the previous letter in which participants were required to withhold their response. The trial design is outlined in Figure 2 (Taylor et al., 2016). No-go trials were pseudo-randomised to be unpredictable. Participants completed two runs, each of 4 minutes 22 seconds in length. Each run consisted of 250 trials; 220 go-trials and 30 no-go trials.
2.2.2. MID task
The MID task, adapted from Knutson et al. (2001) uses a mini-block design to probe reward anticipation and outcome. The task consists of win, neutral and loss trials. A cue is used to make the participant aware of the type of trial. Following this cue is an anticipation period. A target stimulus is then presented to the participant in response to which they must press a button as quick as possible. The outcome of the trial is then presented on the screen. The trial design is outline in Figure 3. In the win trials, participants were able to win £0.50 if they pressed the button quick enough. In the loss trials they could lose £0.50 if they did not respond quick enough. Money was neither won nor lost during neutral trials, but participants were still required to respond to the stimulus. The task was adaptive to achieve a target accuracy of 66% for win trials and maximum winnings of approximately £10. The starting duration of the target stimulus for a win and loss trial was 280ms and 240ms, respectively. If a participant responded successfully then the target stimulus duration was reduced by 10ms while a missed response resulted in an addition of 10ms to the stimulus presentation time. The floor and ceiling durations were 150ms and 300ms respectively. Participants completed two runs of the task with each run being 7 minutes and 12 seconds in duration. Each run contained 18 win, 18 neutral and 6 loss trials.
2.1. Structural and functional acquisition
All data were collected prior to this study, complete procedures are described elsewhere (McGonigle et al., 2017).
Briefly, all images were acquired using a 3 tesla (T) MRI machine. London and Cambridge centres used identical 3T Siemens Tim Trio systems while Manchester used a 3T Phillips Achieva. In London and Cambridge T1-weighted volumes were acquired for registration using a magnetization-prepared rapid gradient echo (MPRAGE) sequence (TR=2300 ms, TE=2.98 ms, TI=900 ms, flip angle =9°, field of view =256 mm, image matrix =240×256) with a resolution of 1 mm isotropic. Functional imaging was acquired using a multi-echo gradient echoplanar imaging (EPI) sequence (TR=2000 ms, TE=13 ms and 31 ms, flip angle =80°, field of view =225 mm, image matrix =64×64) with an in-plane resolution of 3.516×3.516 mm and a slice thickness of 3.000 mm. For each volume, 36 ascending, abutting oblique axial slices were collected at an angle of around 30° to the anterior (AC) and posterior commissure (PC). In Manchester T1-weighted volumes were acquired using MPRAGE sequence (TR=6.8 ms, TE=3.1 ms, TI=900 ms, flip angle =9°, field of view =270 mm, image matrix =256×256) with an in-plane resolution of 1.055×1.055 mm and a slice thickness of 1.200 mm. Functional imaging was acquired using the same EPI sequence as described above. However, for each volume, 24 rather than 36 slices were collected. In all three centres due to the functional imaging parameters, whole brain coverage was not achieved with the most superior 9mm being outside of the range for most participants.
2.2. Pre-processing
Pre-processing occurred prior to this study and details can be found elsewhere (McGonigle et al., 2017).
Briefly, FMRIB Software Library’s (FSL) (version 5.0.6) FMRI Expert Analysis Tool (FEAT) (version 6.00) was used to pre-process both runs of the MID and GNG separately. The functional image obtained during the first run was then used to realign the second functional image. The mean functional realigned image was then co-registered to the T1 weighted structural MRI image aligning them in the same space. The structural image then underwent segmentation and normalisation using the unified segmentation approach to ensure it was in standard stereotactic space. Functional images were then transposed into standard space. An isotropic Gaussian smoothing kernel was used to perform smoothing. A FWMH of 8mm and 7mm were used for the GNG and MID tasks respectively. The reason for this was that during reward processing in the MID task, subcortical structures are known to play an important role and thus a smaller smoothing kernel is recommended (Sacchet and Knutson, 2013).
2.3. fMRI task modelling
All task modelling was conducted prior to this study, full details can be found elsewhere (McGonigle et al., 2017).
2.3.1. MID
Nine exploratory variables were used to model the MID task. These included the type of trial; win, neutral or loss and the phases; anticipation, successful outcome or successful loss. Loss trials were included to increase the salience of the win trials and as such the task was not powered to examine the loss trials. Therefore, for the purpose of this study we analysed the following four explanatory variables; reward and neutral anticipation and successful outcome. The anticipation phase was modelled as a block beginning at cue onset which lasted 1 second and ending upon presentation of the target stimulus 2-4 seconds later. The successful outcome phase was modelled as a block beginning at presentation of the target stimulus (immediately after the end of the anticipation phase) and lasting for two seconds. Successful outcome was modelled so as to include the presentation of the outcome of the trial. A high-pass filter cut-off of periods above 50s was applied to remove unwanted signal. Blocks were then convolved with the haemodynamic response function (HRF) and averaged across both runs. Realignment parameters six and movement parameters were added to the model as nuisance regressors. The contrasts of interests used in this analysis were ‘reward anticipation’ compared with ‘neutral anticipation’ (‘reward anticipation>neutral anticipation’) and ‘reward successful outcome’ compared with ‘neutral successful outcome’ (‘reward outcome>neutral outcome’).
2.3.2. GNG
Two explanatory variables were used for modelling the GNG task; successful no-go and unsuccessful no-go each lasting 0.1seconds. Both were modelled against an implicit baseline of ‘go’ trials and convolved with the HRF. If a stop was preceded by a go trial to which the participant had not responded, then this was considered a false inhibition due to potential lack of attention and these were removed from the model. A high-pass filter cut off of periods greater than 120 seconds was applied. Realignment parameters and movement outliers were added to the model as nuisance regressors. For the purpose of this analysis we used the ‘successful no-go’ compared with ‘go’ trials (‘no-go>go’).
2.4. ROI Analysis
2.4.1. Masks
Guided by previous literature we chose ROIs that are activated by both the MID and GNG tasks; the caudate, OFC and rIFG. Anatomical ROIs were defined using the Harvard-Oxford cortical and subcortical structural atlas. We made masks of the bilateral caudate and bilateral OFC but due to a lack of evidence for the role of the left IFG in both tasks, we decided to lateralise the IFG anatomical mask to the rIFG. Using fslmaths the bilateral OFC and rIFG masks were thresholded at 10% and binarized. The caudate mask was thresholded at 40% to avoid overlap with any other region of the striatum.
To produce the functional masks we overlaid the anatomical masks with whole brain functional activation in the MID and GNG tasks in a pilot cohort. The pilot cohort consisted of 17 healthy participants none of whom participated in the main study. MID and GNG tasks were carried out in the same manner as in the main study. Guided by the pilot data we produced functional masks for activation in the reward anticipation and successful no-go phases and deactivation for the reward outcome phase. ROIs used in this analysis are shown in Figure 4.
2.4.2. 2nd Level ROI Analysis
Contrast estimates averaged across both runs of each task were extracted from each ROI using FSL Featquery and converted into a percentage BOLD signal change. To determine between group (HC, AD, PD) differences in BOLD signal, extracted ROI data for each contrast were entered into a one-way analysis of variance (ANOVA) or Kruskal-Wallis H test as appropriate in IBM’s Statistical Package for the Social Sciences (SPSS) (version 25.0). When significant between group differences were found, pos- hoc t-tests and Mann-Whitney U tests were used to determine what was driving this effect. Small sample size gave rise to a power issue in this study and thus all analyses were also conducted between HC and a collapsed drug addiction group (DD). Between group differences were then analysed using one sample t-tests or Mann-Whitney U tests where appropriate. Holm-Bonferroni correction for multiple comparisons was used.
2.5. Trait, Choice and Impulsive Action
Questionnaires were used to measure trait impulsivity. The UPPS-P (Lynam et al., 2006), and BIS-11 (Patton et al., 1995) are questionnaires designed to measure subjective impulsiveness. The BIS-11 is commonly used in addiction and impulsiveness studies and is split into three subsections; attentional impulsiveness, motor impulsiveness and non-planning impulsiveness. For our results to be comparable with previous literature, we measured total BIS-11 score encompassing all three subsections. The UPPS-P is split into five subsections designed to probe impulsive personality traits; negative urgency, lack of premeditation, lack of perseverance, sensation seeking and positive urgency. In order to ensure we were able to capture complete data on trait impulsivity we used total UPPS-P score for this analysis. Between group differences were analysed using a one-way ANOVA. When significant between groups differences were found a Tukey’s honestly significant difference (HSD) post-hoc test was used.
Choice impulsivity was probed using the Kirby Delay Discounting task which measures the extent to which an individual prefers a smaller more immediate reward over a larger more delayed reward (Kirby and Maraković, 1996). Hypothetical immediate rewards of £11-80 and delayed rewards of £25-85 were used. Delays ranged from 7-186 days. A discount parameter score (k) was generated for each participant representing the proportion of preference for immediate rewards over delayed. Higher k values are associated with increased impulsivity (Kirby et al, 1999). Between group differences were calculated using Kruskal-Wallis H tests. In the event of significant between group differences a Mann-Whitney U post-hoc test was used.
Impulsive action was measured with the SST task, taken from the CANTAB® (Cambridge Cognition, 2018) neurophysiological test battery (http://www.cambridgecognition.com/cantab/cognitive-tests/executive-function/stop-signal-task-sst/). It is designed to measure response inhibition following initiation of an action (Dalley, Everitt and Robbins, 2011). The participant is required to respond to a baseline of directional arrows by pressing a button with the corresponding hand. When an auditory tone is heard, the participants must withhold their response. The task employs a staircase design in order to adapt to the participant’s ability creating an average success rate of 50%. In this study we measured the reaction time of the SST (SSRT) of the last 20 blocks of the study. The reaction time represents the amount of time a participant requires to inhibit their response. We excluded the first block as this is defined as a ‘practice block’ in the CANTAB eclipse version 3.2 Test Administration Guide. Between group differences were calculated using Kruskal-Wallis H tests followed by Mann-Whitney U post-hoc tests.
Multiple comparisons were corrected for using the Holm-Bonferonni method.
2.6. Correlational analysis
Partial correlations covarying for centre were performed to explore the relationship between ROI BOLD signal activation and subjective measures of impulsivity. For each ROI a partial correlation or partial rank correlation as appropriate, controlling for centre, was performed between each task contrast and the following measure of impulsivity; BIS-11, UPPS-P, Kirby Delay Discounting and the SSRT. Correlations were performed across the whole cohort and where a significant association was found post-hoc between group correlations were then conducted. Fisher’s r to z transformation was then used to determine whether group correlations were significantly different to each other. Using the same method, investigations were carried out to explore whether there was an association in BOLD signal response in the MID and GNG tasks for each of our ROIs. The two addiction groups were collapsed and partial rank correlation controlling for centre was used to investigate the effect of length of abstinence on BOLD signal response in the ROIs.
2.7. Whole Brain Analysis
A mixed effects whole brain group analysis was performed using FSL’s FMRIB's Local Analysis of Mixed Effects (FLAME) 1 for activation during reward anticipation>neutral anticipation and reward outcome>neutral outcome for the MID task and successful no-go>go for the GNG task. For the purpose of the whole brain analysis, addiction groups were collapsed. Therefore, independent-samples t-tests were conducted between HC and DD controlling for centre. A z threshold of 3.1 was applied to the cluster based statistical images based on guidance from Woo et al., (2014) with a voxelwise threshold of P<0.05. If no differences were seen at 3.1, the threshold was reduced to 2.3 to ensure no potential differences were missed from this analysis. The locations of clusters were identified using local maxima co-ordinates and the Harvard-Oxford cortical and subcortical structural atlases. Contrast of parameter estimates (COPEs) were then extracted from activated clusters using FSL Featquery. Activation clusters in the occipital cortex were not extracted as these were probably associated with the visual aspect of the task and therefore not of interest to this analysis. This was also applied to any regions that were probably associated with motor actions. Extracted COPE estimates were then statistically analysed in SPSS. Exploratory partial correlation and partial rank correlations covarying for centre were conducted between extracted clusters and subjective measures of impulsivity.