Valuation system connectivity is correlated with poly-drug use in young adults

Main Text

The pursuit of a desired internal state is known to drive drug-seeking and -taking behaviour (Hellberg et al., 2019). For example, an individual in pain may consume sedating substances such as marijuana or sedatives for relief (Choi et al., 2017; Shurman et al., 2010), while individuals looking to experience enhanced levels of euphoria may choose stimulants or cocaine (Breiter et al., 1997; Klega and Keehbauch, 2018). Poly-drug users mix-and-match different combinations in an attempt to increase the efficacy (Scott et al., 2007) or attenuate drawbacks (Fisk et al., 2011) of a particular substance. This tendency to co-use multiple drugs can cause a fatal overdose (Su et al., 2019). Many rehabilitating patients having a history of co-using more than one drug (Su et al., 2019; Wang et al., 2018) thereby requiring different treatment strategies (Hiebler-Ragger and Unterrainer, 2019; Weigl et al., 2017) due to higher levels of risky behaviour and psychopathology, decreased cognitive functioning as well as poorer treatment outcome and engagement (Wang et al., 2017).

Addictive drugs are thought to disrupt the brains’ reward system (Koob and Le Moal, 2008). These substances are consumed to achieve desired internal states (e.g. euphoria or pain relief), with regions of the Valuation and Executive Control System responsible for weighing drug value in relation to these states (Loganathan and Ho, 2021). Given that valuation of drugs changes based on internal and external factors as well as the increased complexity of the addiction phenotype among polyu-drug users, this study will focus on the interaction between Valuation and Executive Control System connectivity and poly-drug use in young adults. Value, the currency of choice, fluctuates based on internal states (e.g. satiety, deprivation) as well as external circumstances (e.g. socioeconomic factors, health status) (Compton et al., 2007; Davis and Johnson, 2008; Helie et al., 2017). The process of assigning value to different options may be regulated by the valuation and executive control systems of the brain (van den Bos and McClure, 2013). The valuation system (VS, comprised of the ventromedial prefrontal cortex, posterior cingulate cortex, amygdala and nucleus accumbens) may be involved in estimating the incentive value of different options (van den Bos and McClure, 2013) and is hypothesized to interface between pleasure and pain (Kringelbach, 2005). The executive control system (ECS, consisting of the lateral prefrontal cortex, dorsal anterior cingulate cortex and posterior parietal cortex), guides behaviour in pursuit of long-term goals (van den Bos and McClure, 2013). The ECS may assist in refining VS-encoded values by incorporating elements of future-thinking and careful deliberation (van den Bos and McClure, 2013). Addiction is thought to be associated with an imbalance in valuation, as drug-related stimuli are attributed greater salience compared to non-drug stimuli (Helie et al., 2017). The VS and ECS are linked with substance use behaviour, both as a network (Zhai et al., 2015) and as individual regions (Bedi et al., 2015). Normally, the interaction between VS and ECS is balanced, modulating reward processing by assigning values to various reward and non-reward options to facilitate optimal decision-making (Dalley et al., 2011; Xie et al., 2014; Zhai et al., 2015). During drug dependence the valuation system activation appears not to be counteracted by the control system which has since surrendered its regulatory abilities, tilting this once-mutually-reinforcing system towards the instrumental pursuit of drugs (Xie et al., 2014; Zhai et al., 2015).

Addictive drugs operate via different pathways in the brain (Leurquin-Sterk et al., 2018; Solinas et al., 2019). Taken together with the tendency to exclude poly-drug users from individual substance use neuroimaging studies (Zou et al., 2015), these limitations have restricted the development of a unifying concept underpinning drug use, further complicating efforts to develop focused treatment plans targeted at poly-drug users (for a review, see Wang et al (2017)). Given that poly-drug use contributes to more than 50% of fatal overdoses (Coffin et al., 2003), interrogating the interactions between different drugs is a necessary first step in unravelling the complex psychopathology underpinning drug addiction and what is required to treat it. This study aims to provide preliminary evidence linking functional connectivity of the VS and ECS with poly-substance use to infer the common combinations of substance co-use and associated neural patterns. This is done by correlating the connectivity of the VS, ECS and the valuation-control complex (VCC, a hybrid network comprised of VS and ECS regions) with poly-drug use in the Human Connectome Project Healthy Young Adult dataset using multivariate generalized linear models. We hypothesize that functional connectivity within the VS, ECS and VCC would all be correlated with alcohol, tobacco, opiates, cocaine, stimulants, sedatives and marijuana use. We further hypothesize that in some cases, connectivity may be correlated with substance use via a paired interaction with another drug (eg VS connectivity paired with sedative use may be correlated with cocaine use).

Functional MRI data from the Human Connectome Project (HCP, Smith et al., 2013) was sourced for healthy adults of both genders (n = 992, 463 males, 529 females; age range = 22-37). Recruitment procedures, inclusion/exclusion criteria and pre-processing steps are described elsewhere (Glasser et al., 2013; Smith et al., 2013a). Additional steps such as global signal regression are deemed too aggressive for use with HCP imaging data and are not implemented as it has been known to introduce artifactual changes in correlations patterns, fundamentally altering interregional correlations. These changes are thought to be dependent on the underlying true interregional correlation pattern (Anderson et al., 2011; Glasser et al., 2013; Saad et al., 2012; Smith et al., 2013b). The functional scans used in this study have already been cleaned of structured noise via joint use of independent component analysis (ICA) and FMRIB’s ICA-based X-noisifier (FIX) to remove non-neural spatio-temporal components from each highpass-filtered fMRI scan. Initial head motion correction is performed by registering fMRI data to a distorted gradient echo EPI single-band reference image. For each subject, rigid body head motion was estimated using FSL’s MCFLIRT routine to derive a motion transformation matrix for each time point. Each transform is described by 6 motion parameters consisting of three translations and three rotations, which is then condensed to form a vector of framewise displacement by summing the absolute values of the differentials of the 6 parameters (Power et al., 2012). As a part of the artefactual cleanup process, 24-counfound timeseries derived from the motion estimation (comprised of the 6 rigid body-parameter timeseries, their backwards-looking temporal derivatives, plus all 12 resulting regressors squared) had temporal highpass filtering applied to them and then regressed out of the data aggressively (WU-Minn, 2017). The HCP pipeline ultimately resulted in the removal of spatial and temporal artefacts without removing potentially useful data, particularly relevant for resting-state fMRI data that is particularly sensitive to artefactual corruption across voxels (Smith et al., 2013a).

Only subjects with all four repeated resting-state fMRI sessions (first and second scan sessions with left-right and right-left phase encoding directions), were used for this study. Preparation of masks used in this study has been detailed elsewhere (Loganathan et al., 2020). Briefly, the VS, ECS and VCC were delineated using the binary masks that combined regions of interest (ROIs) from both the Desikan-Killiany (Desikan et al., 2006) and Destrieux (Destrieux et al., 2010) parcellations. All anatomical labels were extracted and merged using the FMRIB Software Library (Smith et al., 2004, https://fsl.fmrib.ox.ac.uk/fsl/). The complete list of all anatomical ROIs used to delineate the VS and ECS are shown in Table Supplementary 1, along with their central coordinates given in Montreal Neurological Institute (MNI) space.

The following substance use measures were obtained from the Human Connectome Project Restricted Access Data (Table 1). These substance use and frequency measures were collected using two approaches: a detailed questionnaire prepared specifically for the Human Connectome Project (for short-term alcohol and tobacco use), followed by the Semi-Structured Assessment for the Genetics of Alcoholism (SSAGA) for long-term alcohol, tobacco and illicit drug use. Additionally, parental use history, age, gender and framewise displacement (FD) measures were included as demographic variable.

The rsfMRI signal was averaged over all voxels comprising each ROI (node) of the VS (12 nodes), ECS (20 nodes) and VCC (32 nodes), yielding a regionally averaged signal for each node. The Pearson correlation coefficient in the regionally averaged signals were then computed between all pairs of nodes, resulting in a 12 x 12 (VS), 20 x 20 (ECS) and 32 x 32 (VCC) connectivity matrix for each of the four scan runs (Rest 1-LR, Rest 1-RL, Rest 2-LR, Rest 2-RL). This computation was repeated for each subject. The four connectivity matrices computed from the four rsfMRI sessions for each subject were then averaged to improve the signal-to-noise ratio, yielding a single connectivity matrix for each of the three systems for every subject (see Supplementary Results). These connectivity matrices were then averaged per subject, such that for each of the VS, ECS and VCC, each subject’s connectivity would be represented by a single value. Generalized linear models (GLM) were then computed to correlate the average connectivity of VS, ECS and VCC with each subject’s substance use profile. All dependent variables, independent variables and connectivity measures are standardized to z-scores with mean of 0 and standard deviation of 1. Regression coefficients are therefore representative of SD change in outcome per unit SD change in predictor and may be interpreted as partial correlation coefficients (Bring, 1994). Throughout this study, a total of 288 tests were performed (16 independent variables X 3 brain networks X 6 dependent variables). Each dependent variable was tested against 15 independent variables for all 3 networks (VS, ECS, VCC) producing a single model per dependent variable per network. For the purposes of illustration, let us assume one wishes to model Cocaine use (dependent variable) and its correlation with Valuation System average connectivity (brain network). The 16 independent variables would therefore comprise of Total Drinks 7 Days, Drink Frequency 7 Days, Drinks/Day 12 Months, Drink Frequency 12 Months, Total Tobacco Use 7 Days, Tobacco Use Frequency 7 Days, Hallucinogen, Opiate, Sedatives, Stimulants, Marijuana, Age, Gender, Fathers’ Use History, Mothers’ Use History, Average Framewise Displacement). Multiple comparisons across all models were controlled using the False Discovery Rate using the p-values attributed to each component of each model. p-values < 0.05 post-correction are deemed significant.

Correlation between substance use measures is shown in Table 2. The most highly correlated pairing of substances are opiates and sedatives (r² = 0.725), followed by cocaine and hallucinogens (r² = 0.614). Marijuana use was not strongly correlated with other substances, except with hallucinogens (r² = 0.559). Alcohol (Total Drinks 7 Days, Drink Frequency 7 Days, Drinks/Day 12 Months and Drink Frequency 12 Months) and tobacco measures (Total Tobacco 7 Days, Tobacco Frequency 7 Days) did not show strong correlation with illicit drug use measures. Stronger correlations were reported between alcohol and Marijuana measures as well as tobacco and Marijuana measures, but even these r² values were no higher than 0.326 and 0.336 respectively. Correlations between resting-state functional connectivity and substance use measures are shown in Table 3. Network connectivity of all three systems is weakly correlated with substance use measures, with no r² value higher than 0.1.

Average connectivity of VS, ECS and VCC is correlated, either independently or via a two-way interaction with substance use (Table 4). VS connectivity is correlated with Sedative use (Estimate=0.04, t=2.3, p=0.03) while ECS connectivity is correlated with Hallucinogen (Estimate=0.08, t=2.4, p=0.025) and Opiate (Estimate=0.05, t=2.3, p=0.03) use. Cocaine use is correlated with VS, ECS and VCC connectivity via two-way interactions with Total Tobacco 7 Days, Hallucinogen use, Opiate use, Sedative use and Fathers’ Use History. Opiate use is correlated with VS (Estimate=0.07, t=3.3, p=0.002), ECS (Estimate=0.08, t=3.6, p=0.0008) and VCC (Estimate=0.1, t=4.4, p<0.00001) connectivity via two-way interaction with Total Tobacco 7 Days. Sedative use is correlated with VS (Estimate=0.04, t=2.6, p=0.017), ECS (Estimate=0.04, t=2.2, p=0.04) and VCC (Estimate=0.04, t=2.3, p=0.03) connectivity via two-way interaction with Drinks/Day 12 Months. Stimulant use is correlated with VS (Estimate=-0.07, t=-2.9 p=0.007), ECS (Estimate=-0.25, t=-7.2, p<0.00001) and VCC (Estimate=-0.16, t=-4.8 p<0.00001) connectivity via two-way interaction with Opiate use. Hallucinogen use is correlated with VS (Estimate=-0.06, t=-2.6, p=0.015) and VCC (Estimate=-0.05, t=7.8, p=0.04) connectivity via two-way interaction with Total Tobacco 7 Days. Opiate use is correlated with ECS (Estimate=0.19, t=5.8, p<0.00001) and VCC (Estimate=0.09, t=3.6, p=0.0009) connectivity via two-way interaction with Hallucinogen use. Sedative use is correlated with ECS (Estimate=0.05, t=2.45, p=0.02) and VCC (Estimate=0.08, t=3.6, p=0.0006) connectivity via two-way interaction with Tobacco Frequency 7 Days. Sedative use is also correlated with ECS (Estimate=0.19, t=7.96 p<0.00001) and VCC (Estimate=0.18, t=8.3, p<0.00001) connectivity via two-way interaction with Opiate use. Sedative use is correlated with VS (Estimate=0.09, t=3.17 p=0.003) and ECS (Estimate=-0.08, t=-2.86, p=0.008) use via two-way interaction with Cocaine use.

Research suggests that substance use may stem in part from a disrupted decision-making system stemming from aberrant valuation of drugs (Zhai et al., 2015). The value of drugs varies between individuals, driven by hedonic sensations in some or pain relief in others (Hogarth, 2020). These underlying motivations increase the value of drugs and bias choice towards their use (Hogarth and Field, 2020). Additionally, connections both within and between the Valuation and Executive Control System act as dependency risk and resilience markers (Ersche et al., 2020). Here, we correlated the average network connectivity of the Valuation and Executive Control Systems, as well as the Valuation-Control Complex with drug use in the brains of healthy young adults. Common co-use patterns and their network correlations can thus be inferred based on trends observed in this study. Understanding the needs of one who takes addictive substances with respect to a desired internal state (e.g., euphoria or numbness) represents the first step to improving existing forms of interventions, with greater focus on individual needs that are being met via substance use.

Correlations are observed between combinations of addictive substances, for example Sedatives and Cocaine, Opiates and Sedatives. Highly correlated pairs of drugs reflect popular combinations used to achieve a particular state (euphoria, numbness etc) (Floyd et al., 2010; Laursen et al., 2016). Thus, two-way interactions involving substance use measures and average network connectivity of choice networks may signify the high value placed on these combinations to achieve said state. Stimulating and sedating substances are commonly co-used due to their synergistic interactions. Stimulant-type substances (cocaine, stimulants like amphetamine) energize the user, producing longer-lasting euphoria while off-setting sedating effects (Hernández-López et al., 2002). Sedative-type substances (alcohol, tobacco, marijuana, pain-killing sedatives, heroin) calm the user, ameliorating unpleasant side-effects of stimulant use such as anxiety and aggression (Fisk et al., 2011).

Tobacco and alcohol (Drink Frequency, Total Drinks and Drinks/Day) use measures form two-way interactions with VS, ECS and VCC average network connectivity for all six substances tested as dependent variables (Cocaine, Hallucinogens, Opiates, Sedatives, Stimulants, Marijuana). Tobacco and alcohol both possess sedating qualities (Fultz et al., 2021; Kishore, 2014; Walker, 1980) and are considered gateway substances that pave the way towards the development of illicit drug consumption (Kandel and Kandel, 2015), underpinning poly-substance use (Anderson et al., 2018; Jongenelis et al., 2019; Moss et al., 2014). Stimulants complement alcohol to produce longer-lasting euphoria while off-setting the sedating qualities of alcohol (Hernández-López et al., 2002), reducing the anxiety and agitation (Fisk et al., 2011; Knackstedt and Ettenberg, 2005). Alcohol and sedating substances (opiates, sedatives, marijuana) are often co-used for pain management (Downey et al., 2013; Hood et al., 2020; Ronen et al., 2010; Wilson et al., 2020), albeit at the expense of cognitive control over reward-driven behaviour (Crean et al., 2011; Müller-Oehring et al., 2013). Similarly, tobacco use sensitizes the reward centres of the brain making one susceptible to other drugs (Kandel and Kandel, 2015). Tobacco is thought to encourage cocaine and opiate use (Cross et al., 2017; Mello et al., 1980; Stark and Campbell, 1993) by stimulating the reward centres of the brain (Cross et al., 2017; Nuechterlein et al., 2016). The findings of the current study support existing evidence regarding the involvement of alcohol and tobacco in poly-drug use, as these two substances form two-way interactions with average connectivity of VS, ECS and VCC for all illicit drugs (Cocaine, Hallucinogens, Opiates, Sedatives, Stimulants, Marijuana). Given the “gateway” nature of these two substances (Kandel and Kandel, 2015), particularly among young adults (Laursen et al., 2016; McKetin et al., 2014), it is theorized that these participants may have started using alcohol and/or tobacco to achieve an internal state of euphoria or numbness, but began to incorporate illicit drugs to boost the strength of the desired sensation.

Two-way interactions between all three networks and Opiates in Cocaine use could be indicative of “speedball”, a cocaine-heroin mixture that induces enhanced feelings of euphoria (Lacy et al., 2014), while reducing the aggression and anxiety (Fisk et al., 2011). A similar effect may underpin other two-way interactions between all three networks and sedating substances (Opiates, Sedatives) with Stimulants and vice versa. Studies revealed ECS involvement in regulation of attentional processing of external cues (Sarter et al., 2001) including auditory (Napoli et al., 2021; Plakke et al., 2013) and visual stimuli (Geliebter et al., 2016). Both stimulant and cocaine use are capable of enhancing the senses (Devonshire et al., 2007; Oken et al., 2006) and inducing euphoria (Bellone et al., 2020; Dlugos et al., 2011) by increasing activation within the ECS (Jan et al., 2014) and VS (Crane et al., 2018) respectively.

Co-use of sedating substances are commonly used for pain management (Peele, 2016) especially when a single substance is found to be inadequate for the purpose (Wilson et al., 2020). The VS acts as the middle ground for processing both pleasure and pain (Leknes and Tracey, 2008). Connections between the VS and ECS regions is observed in patients experiencing chronic pain (Jiang et al., 2016). In the current study, VS, ECS and VCC each form a two-way interaction with alcohol in Sedatives use, suggesting a pain-killing combination. Two-way interactions involving VS, ECS and VCC connectivity and sedating substances (Marijuana, Sedatives) when paired with other sedating substances (such as alcohol) could indicate these combinations of substances are being used for pain management.

This study is not without limitations. There are various types of stimulants, sedatives and hallucinogens, some of which have been manufactured to have pharmacological properties of other abusable substances (Klega and Keehbauch, 2018). Given that each drug has its own mechanism of action that targets specific receptors or pathways, information regarding the specific types of drugs used may help shed additional light on the relationships between network connectivity and poly-drug use measures. Additionally, global signal regression was not performed in this study. The fMRI scans used in this study have been minimally pre-processed by the HCP team to ensure minimal information loss (Glasser et al., 2013). Global signal regression is deemed too aggressive for use with HCP imaging data and never implemented. This method may introduce artifactual changes in correlations patterns, fundamentally altering interregional correlations. These changes are thought to be dependent on the underlying true interregional correlation pattern (Anderson et al., 2011; Saad et al., 2012).

In conclusion, we correlated the average network connectivity of the Valuation and Executive Control Systems, as well as the Valuation-Control Complex with drug use in the brains of healthy young adults. Common co-use patterns and their network correlations can thus be inferred based on trends observed in this study. Two-way interactions involving substance use measures and average network connectivity of choice networks may signify the high value placed on these combinations to achieve said state. Particularly, tobacco and alcohol use measures form interactions with VS, ECS and VCC average network connectivity across all forms of illicit substances (Cocaine, Hallucinogens, Opiates, Sedatives, Stimulants and Marijuana), suggesting that participants may have started using alcohol and/or tobacco initially but began to incorporate illicit drugs to enhance their experience. Additionally, two-way interactions involving sedating substances (e.g. Opiates) with connectivity of all 3 networks is observed in stimulant-type substances (e.g. Cocaine) and vice-versa, suggesting a combination aimed at increasing feelings of euphoria while dampening anxiety and aggression. Finally, connectivity of all three networks each form a two-way interaction with alcohol among sedating substances, suggesting a combination for pain management. These results provide early evidence linking functional connectivity of the VS and ECS with poly-substance use, enabling us to infer common combinations of substance co-use and associated neural patterns.

Funding

Kavinash Loganathan is grateful to IBRO-APRC and Yayasan Universiti Teknologi PETRONAS (YUTP) for funding to undertake this study. Eric Tatt Wei Ho gratefully acknowledges the Ministry of Higher Education, Malaysia for funding through the Higher Institution Center of Excellence (HI-CoE) program awarded to the Center for Intelligent Signal & Imaging Research, Universiti Teknologi PETRONAS and the YUTP-Fundamental Research Grant award. Vanessa Cropley was supported by an Australian National Health and Medical Research Council (NHMRC) Investigator Grant (1177370).

Supplementary Methods

The following table shows the list of regions of interest (ROIs) and their center of gravity (in mm) used in the construction of the valuation, executive control and valuation-control complex masks.

Supplementary Results

A multicollinearity test was performed among all substance use measures (Supplementary Table 3). Results do not indicate the presence of multicollinearity among these variables. Among the substance use measures, no condition index scored more than 15, while no variance inflation factor (VIF) readings are above 10.

The following figures show average VS, ECS and VCC connectivity in the brains of Human Connectome Project Healthy Young Adult participants used in this study. Connectivity within the VS was higher between the regions of the vmPFC and PCC (Supplementary Figure 1A). Connectivity within the ECS was higher between regions of the PPC, DLPFC and VLPFC (Supplementary Figure 1B). In the VCC, connectivity between the ECS and VS regions was higher between the PPC-DLPFC and the vmPFC-PCC as well as between the lateral OFC and VLPFC (Supplementary Figure 2).

• Download figure
• Open in new tab

Supplementary Figure 1:

Valuation system component regions (upper) and resting-state functional connectivity matrix (lower) in the human brain (A). Control System component regions (upper) and resting-state functional connectivity matrix (lower) in the human brain (B).

• Download figure
• Open in new tab

Supplementary Figure 2:

Valuation system (yellow-orange) and control system (green) component regions in the human brain. The cortical regions (vmPFC, PCC) of the valuation system are labelled yellow, while the subcortical regions (NA, amygdala) are labelled orange). The valuation-control complex is a combination of the valuation and control systems (A). Resting-state connectivity matrix of the valuation-control complex. Connections within the valuation system are shown in yellow, the control system in green. PPC-DLFPC connections with vmPFC-PCC are highlighted in red, VLPFC-lateral OFC connections are highlighted in orange (B).

Given that head motion both affects functional connectivity measures (Ciric et al., 2018) and is affected by substance use (Givens, 2016), average framewise displacement values were correlated with both average functional connectivity of VS, ECS and VCC (Supplementary Table 4) as well as substance use measures (Supplementary Table 5). Strong correlations were not observed between average framewise displacement with either average connectivity or substance use measures with only tobacco measures (Total Tobacco 7 Days and Tobacco Frequency 7 Days) registering r values higher than 0.1.

Valuation System

Cocaine

Hallucinogens

Opiates

Sedatives

Stimulants

Marijuana

Executive Control System

Cocaine

Hallucinogens

Opiates

Sedatives

Stimulants

Marijuana

Valuation-Control Complex

Cocaine

Hallucinogens

Opiates

Sedatives

Stimulants

Marijuana