Olga Boer

Substance use is commonly initiated during adolescence, a period that is also marked by extensive neurodevelopment. This temporal overlap has raised important questions about the nature of the relationship between early substance use and neurodevelopment. Specifically, it remains unclear whether observed associations reflect pre-existing vulnerabilities that predispose individuals to substance use, or whether substances itself have neuromodulatory effects on the developing brain, or whether this relationship is bidirectional. Furthermore, while long-term substance use has been linked to substance use disorders and alterations in brain structure and function, less is known about how these associations manifest in the general adolescent population. As effect sizes of brain-behavior associations are typically small, large samples are required to detect them reliably. In recent decades, large-scale initiatives focusing on brain morphology (using MRI) have emerged to meet this need and have advanced our understanding of structural brain correlates of adolescent substance use. However, comparable large-scale datasets in electroencephalography (EEG) research remain relatively uncommon.

The primary aim of this thesis was to address the following key questions:
1. Are there pre-existing brain differences in individuals who initiate substance use in adolescence?
2. How do brain structure and function relate to substance use initiated in adolescence?
3. What is the value of large EEG datasets in examining associations between brain and behavior?

These questions were addressed with a large epidemiological cohort study (The Generation R study) from the Netherlands, an adolescent sample (Brains and minds in transition; BRAINMINT) from Norway and two (systematic) reviews of the literature.

Pre-existing brain differences in adolescent substance use initiators
In recent decades, the field of neural correlates of substance use has gained a focus on what happens in the brain before substance use is initiated. This is made possible by longitudinal studies with a substance-naïve baseline brain measurement and follow-up visits in which substance use initiation is reported. In Chapter 2, we highlighted all studies examining this research question for alcohol, cannabis and tobacco use initiation. In these studies, pre-existing brain morphology variations were observed in future users: smaller anterior cingulate cortex (ACC) and superior frontal gyrus (SFG), and larger nucleus accumbens (NAcc) volume for future alcohol use; variations in orbitofrontal cortex (OFC) volume (both smaller and larger) were associated with future cannabis use; and smaller amygdala volume related to future tobacco use. However, the largest study on future alcohol use reported null findings. In Chapter 3, we explored this research question in the Generation R sample (N = 2200) and observed no pre-existing brain morphology differences in early (<13 years) alcohol or tobacco initiators. This might be due to our focus on very early initiation (<13 years) in our sample, whereas studies in the review primarily included later initiation and older adolescents. Recognizing pre-existing brain differences as potential risk factors for later substance use, in Chapter 4, we examined what might underlie these differences. We found that exposure to continued tobacco smoking during pregnancy was associated with lower global and regional brain volumes, smaller surface area, and reduced gyrification at age 10 years. As prenatal substance exposure has been linked to early substance use initiation, these findings help to connect the broader picture. The link between brain structure/function and substance use initiated in adolescence To explore the other direction of the association, in Chapter 5, we combined findings of over 100 studies that examined the long-term brain morphology correlates of alcohol, cannabis, tobacco, stimulant and opioid use initiated in adolescence. We reported that long-term alcohol and tobacco use were consistently associated with smaller frontal regions and altered white matter microstructure, particularly in the ACC for alcohol and the OFC for tobacco. Cannabis and stimulant use were linked to both increased (in adolescents) and reduced (in adults) hippocampal volumes, while opioid use was associated with smaller subcortical and insular volumes. Null findings were also common for all substances, particularly in cannabis use studies. Then, in Chapter 6 and 7, we explored this question for brain function; specifically, electrophysiological markers of error processing. In the Generation R cohort (N = 1525, Chapter 6), we observed an association between early initiation of alcohol use and more binge drinking, and altered neural markers of error processing. However, in the BRAINMINT cohort (N = 143, Chapter 7), we reported no link between neural markers of error processing and substance-related risks and problems, or externalizing problems in general. We argue this was partly due to lower substance use and related problems in the BRAINMINT sample, which limited variation and statistical power. Large EEG datasets in behavioral research In Chapter 8, we present data of one of the first population-based EEG studies, leveraging EEG data of almost 3000 participants. We used this data to inform future EEG-behavior research (both large- and small-scale). Specifically, we examined the minimum task duration to obtain reliable neural markers of error processing, and important confounding factors to adjust for in statistical models. For moderate reliability, we reported that a minimum of 9/10 errors was needed for the ERN, and 7 errors for the Pe. Fifteen No-Go trials were needed for moderate reliability of No-Go N2/P3. However, we emphasized that researchers should balance reliability and maintaining participants when applying such minima. Furthermore, we observed associations between higher IQ and larger Pe, No-Go P3, and FM-theta power, as well as similar patterns for maternal education and household income. Additionally, participants with a migration background showed smaller No-Go P3, and female participants showed larger ERN and No-Go N2 as compared to male participants. Discussion Finally, in Chapter 9, I integrated findings from all previous chapters, focusing primarily on the issue of directionality. I linked the findings in this thesis to recent studies, including studies with family-based designs (e.g., twin studies, family history of SUD), causal inference models, and genetic approaches such as Mendelian randomization and latent causal variable analysis. Together, this body of work supports a bidirectional model, suggesting both pre-existing brain vulnerabilities and neuromodulatory effects of substance use, although the expression of substance use (risk) in the brain seems to be heterogeneous. I discussed how confounding may contribute to inconsistencies in the brain-substance use literature, and emphasized the importance of using tools like directed acyclic graphs (DAGs) to identify causal pathways and confounders. I also highlighted situations where adjusting for confounders may be inappropriate for interpretation of findings, due to high co-occurrence of certain behaviors or problems in real-world settings. Population-based research offers valuable opportunities to address these challenges, including common issues like selection and attrition bias. However, I also highlighted that generalizability concerns can persist even in large samples, which affects interpretability of findings. Moreover, as there are numerous ways to measure substance use behavior, and the type of use is trend-sensitive and variable over time, this poses challenges for longitudinal population-based studies, which often require assessment that is brief (i.e., few items) and consistent over time. Looking ahead into future research possibilities, large-scale EEG shows strong potential for advancing our understanding of brain-substance use relationships. Furthermore, I discussed emerging multivariate and causal inference techniques that can help extract meaningful conclusions from observational data. Finally, I emphasized that the current findings do not seem appropriate for identifying at-risk youth using brain-based markers. However, this may change with advances in analytic methods, new approaches to conceptualizing substance use (e.g., focusing on underlying constructs or resilience), and the growing availability of large-scale EEG data.