The present study examined baseline and longitudinal associations between the environment, brain, and behaviour in adolescence. At baseline, we found that children with a better macro-and micro-environment showed larger brain volumes relative to their peers. Both better environments and higher brain centile scores were also associated with better mental health and cognitive outcomes. Importantly, in several cases, the relationship between environmental factors and cognitive and mental health outcomes was partially mediated via the brain. Longitudinally, we found that adolescents with a better childhood macro- and micro-environment moved higher up in their centile rankings across adolescence. Moreover, adolescents who moved upwards in their centile scores in targeted regions showed improvements in mental health and cognition. The childhood macro- and micro-environments were associated with changes in cognition, but not in mental health, across adolescence. The environment-brain-behaviour pathway therefore offered some explanatory power in predicting longitudinal changes in cognition across adolescence.

First, the findings provide confirmatory evidence for the associations between the environment and brain structure. At baseline, those with a better macro- and micro-environment showed higher brain volumes relative to their peers. Associations were observed across the brain and were particularly pronounced in multisensory and sensorimotor regions (e.g., middle temporal, lateral occipital, inferior temporal, precentral and postcentral gyri). Longitudinally, those with better environments showed increases in their relative centile rankings across adolescence. Associations were particularly pronounced in higher-order associative and multi-modal integrative regions (e.g., precuneus, insula, pericalcarine, lingual, inferior parietal, and middle temporal gyri). This trend aligns with known patterns of cortical maturation, where sensory regions develop first followed later by higher-order associative regions⁹. Environmental associations with brain structure therefore align with the general trajectory of cortical development.

Our findings further demonstrate that, at baseline, a better macro- and micro-environment was associated with better cognitive and mental health outcomes across all measured domains, consistent with a large body of previous literature^15,31,41. Longitudinally, a better childhood micro- and macro-environment was associated with increases on all the measured cognitive tasks across adolescence. However, the childhood environment did not predict changes in mental health traits. There are several potential explanations for these findings. Firstly, given that environmental factors are dynamic and may change at the individual-level, it is possible that the childhood environment is not a reliable predictor of mental health changes across adolescence, which may be more sensitive to the current environment. As such, measures of the environment that are more temporally aligned with mental health measures might result in more prominent associations. It is also possible that, while environment plays a significant role in shaping baseline effects, it may not play a further additive role in shaping longitudinal change across adolescence. Instead, these relationships may already become established pre-adolescence and, subsequently, the initial effects of the environment on the brain and mental health may become self-reinforcing in driving longitudinal change. Alternatively, it is also possible that the substantial reorganisation of the brain during adolescence may overshadow the impact of baseline environmental effects.

Indeed, as expected, brain structure significantly predicted mental health and cognitive scores, both at baseline and longitudinally. Higher baseline centile scores across multiple brain regions predicted lower internalising and externalising traits. Longitudinally, individuals showing decreases in their relative centile rankings across adolescence showed increased traits of poor mental health. Prominent regions showing associations with externalising included structures such as the insula and lateral orbitofrontal cortex, which have previously been implicated in processes such as self-awareness⁴², emotion regulation⁴³, impulse control⁴⁴, and reward processing⁴³. Structural variations within the insula, in particular, are implicated across psychiatric conditions, and more specifically with externalising psychopathologies⁴⁵. Associations were also observed across a wide range of regions, including the visual and associative regions (e.g., cuneus, pericalcarine, lateral occipital, and lingual cortex) as well as semantic and motor regions (e.g., middle temporal gyrus, pars triangularis, inferior parietal cortex, and precentral gyrus). Although these regions have been linked to a broad set of functions, some research suggests that various of them may also be involved externalising-relevant processes such as inhibitory control^45–48, attention control⁴⁹, and impulsivity⁴⁹. Indeed, structural variations in several of these regions have also previously been associated with externalising traits^53–55. For instance, largely consistent with the findings of the present study, it has previously been shown that externalising traits are linked to negative deviations in gray matter volumes from peers across structures such as the cuneus, inferior parietal lobe, lingual gyrus, and precentral gyrus in childhood and adolescence⁵⁰.

Several regions associated with externalising also overlapped with those associated with internalising, including the insula, cuneus, lingual gyrus, pars triangularis, and pericalcarine gyrus. This is consistent with prior evidence suggesting that internalising and externalising show both overlapping as well as distinct neural correlates^51,52. These regions have also been implicated in internalising-relevant traits and exhibit structural variations in internalising conditions such as depression^53–55. The insula, for instance, has been linked to interoception (potentially explaining somatic aspects of internalising) and emotion processing⁴², with smaller volumes and reduced surface areas showing associations with internalising^51,54. Regions specific to internalising included the fusiform gyrus and default mode regions such as the isthmus cingulate and parahippocampal gyrus – previously implicated in memory and self-referential cognition⁵³. Likewise, reduced volume, surface area, cortical thickness and gray matter density across these regions have been linked to depression and anxiety^55–57.

With respect to cognition, at baseline, children whose brain volumes were larger relative to age and sex-matched peers showed higher performance across the range of tested cognitive tasks measured. Longitudinally, more region-specific and task-dependent associations were observed with mixed directionality. For instance, increases on the Flanker Inhibitory Control and Attention Task were associated with downward shifts in the lateral orbitofrontal cortex, implicated in inhibitory control⁴⁴, as well as the precuneus, implicated in self-referential processing⁵⁸ and attention-shifting⁵⁹. Increases on the Oral Reading Recognition task were likewise associated with downward shifts in the precuneus and the pars ocupalaris, both of which have been linked to aspects of language and semantic processing^60,61. On the other hand, increases on the Picture Memory Task, which assesses episodic memory and sequencing, were associated with upward shifts in centile scores across the brain. Increases on the Picture Vocabulary Task were also associated with upward shifts in the isthmus of the cingulate cortex, while performance on the Pattern Comparison Processing Speed Task was associated with downward shifts in the superior temporal sulcus, implicated in various broad aspects of cognition (e.g., sensory integration, attention control)⁶². Although seemingly counterintuitive, it is possible that these downward shifts, in fact, reflect more advanced maturation, since overall reductions in gray matter volumes are typical during adolescent development⁶³. These effects may be region and task-dependent, with certain tasks requiring increasing centiles in specific regions for better performance while others requiring decreased centiles. It is also important to note that since these analyses are based primarily on structural measures, any functional interpretations are largely speculative and should be interpreted with caution.

Finally, a key aim of the present study was to investigate the environment-brain-behaviour mediation pathway. At baseline, several such mediation effects were identified - for instance, an improved macro-environment was associated with higher centile scores, which, in turn, predicted lower internalising and externalising. Centile scores also mediated the association between the macro-environment and scores on the Flanker Inhibitory Control and Attention task, as well as the association between both the micro- and macro-environment and scores on the Oral Reading Recognition Task. These results suggest that the associations between the environment and behaviour are mediated, at least partially, via the brain.

Longitudinally, changes in regions such as the cuneus and lateral occipital gyrus mediated the relationship between environmental measures and changes on the Picture Memory Task. Similarly, changes in the cingulate cortex mediated the relationship between the micro-environment and scores on the Picture Sequence Memory Task. These findings suggest that the childhood environment predicts aspects of cognitive development across adolescence partly through modulating changes in brain structure. Besides these associations, no other longitudinal mediation effects were observed for the other outcomes examined. This can largely be attributed to the fact that the baseline environment was not predictive of longitudinal changes across outcome such as internalising and externalising, which precluded further mediation analyses.

There are a number of limitations to consider when interpreting these findings. First, mediation analyses typically require high statistical power. In the present study, the longitudinal sample, although still large, was smaller than the baseline cohort. As a result, some subtle mediation effects may not have been detected in the longitudinal analyses due to comparatively lower power. Secondly, while brain and behavioural changes were measured longitudinally, environmental factors were measured only at baseline. It is possible that changes in the environment may be more informative in predicting changes in the brain, mental health, and cognition across adolescence. This is particularly important considering that the COVID-19 pandemic occurred during data collection and resulted in various substantial changes to the health, social, economic, and educational environment. Third, macro-environmental variables were derived from geospatial data, which hold limited spatial and temporal resolution. Fourth, the ABCD sample is concentrated in US metropolitan areas, which may limit the global generalisability of these findings. Fifth, because several environmental factors are highly collinear and interrelated, we used summary-level scores as a method to parsimoniously obtain a broad overview of the impact of the environment on brain development and behaviour. However, as a result, this approach limited inference on the specific contributions of individual environmental predictors. Additionally, some other important aspects of the adolescent micro-environment, such as cyberbullying or discrimination, were not considered as these measures were not fully available for all timepoints in the ABCD dataset. Finally, several environmental factors that were investigated are not fully separable from genetic factors. For instance, micro-environmental factors such as parental education can be highly heritable ⁶⁴ and are also linked with macro-environmental factors such as neighbourhood characteristics. It will therefore be important for future work to assess the extent to which the observed associations persist after accounting for genetic factors.

In light of these limitations, future studies may consider disentangling the roles of individual environmental predictors on outcomes of interest. Future research should also consider incorporating both repeated measures of the environment as well as additional brain-based measures (e.g., white matter microstructure, functional connectivity). Collectively, these measures will produce more explanatory mediation pathways. Studies should also consider integrating genetic data into longitudinal environment-brain-behaviour frameworks to clarify how genetic predispositions may moderate sensitivity to environmental effects on the brain and behaviour. Finally, since adolescence is also a critical period for sex differences in the brain and behaviour⁶⁵, sex differences in the environment-brain-behaviour mediation pathway should also be explored further.

In summary, the present study identified both baseline and longitudinal associations between the environment, brain, and behaviour during adolescence. Key strengths include its large sample size, longitudinal design, and use of brain metrics benchmarked against population-based norms. The findings represent an important step towards the identification of brain markers and enviro-types that shape cognition and mental health during this critical period of development. In turn, this may support the development of both neurobehavioral interventions and policy initiatives aimed at improving the environment to promote positive outcomes for adolescents.