Recent advances in understanding neural correlates of anxiety disorders in children and adolescents

Abstract

Purpose of review:

Anxiety disorders are some of the most common psychiatric diagnoses in children and adolescents, but attempts to improve outcome prediction and treatment have stalled. This review highlights recent findings on neural indices related to fear and anxiety that provide novel directions for attempts to create such improvements.

Recent findings:

Stimuli capable of provoking fear engage many brain regions, including the amygdala, medial prefrontal cortex, hippocampus, and bed nucleus of the stria terminalis. Studies in rodents suggest that sustained, low-level threats are particularly likely to engage the bed nucleus of the stria terminalis, which appears to malfunction in anxiety disorders. However, anxiety disorders, like most mental illnesses, appear less likely to arise from alterations in isolated brain regions than in distributed brain circuitry. Findings from large-scale studies of brain connectivity may reveal signs of such broadly distributed dysfunction, though available studies report small effect sizes. Finally we review novel approaches with promise for using such large-scale data to detect clinically relevant, broadly distributed circuitry dysfunction.

Summary:

Recent work maps neural circuitry related to fear and anxiety. This circuitry may malfunction in anxiety disorders. Integrating findings from animal studies, big datasets, and novel analytical approaches may generate clinically relevant insights based on this recent work.

Introduction

Anxiety disorders are among the most common psychiatric disorders with prevalence ranging from 10–14% during adulthood (1). The prevalence during childhood and adolescence might be even higher, with estimates ranging from 20–30% of at least one anxiety disorder during development (2,3). These data on prevalence across the lifespan resemble data for many mental disorders. Hence, like many mental illnesses, anxiety disorders represent developmental conditions, problems where persistent psychopathology in adults begins in children and adolescents (4).

This critical review unfolds in three steps. The review begins by describing evidence relating fear and anxiety to distinct neural mechanisms. This is followed by a description of two promising recent advances on translating findings from research in brain imaging to clinical practice. One advance involves the creation of big datasets, including longitudinal studies. The other involves the development of predictive models that aim at developing tools capable of translating research to clinically meaningful findings.

Classification of human responses to threat can adopt many approaches. One such approach is used in the RDoC (Research Domain Criteria) Negative Valence systems (5). As a framework for classification, RDoC attempts to bridge understanding of mental illness phenomenology with knowledge on brain-behavior relationships. The term “fear” has been used to refer to responses evoked in people by direct exposures to a threatening stimulus, whereas the term “anxiety” has been used to refer to responses evoked by the anticipation of such an exposure (4) (Figure 1). Following this distinction, acute threat (fear) and potential threat (anxiety) are considered two distinct constructs within the RDoC framework (5).

Figure 1 -. Schematic representation of brain regions involved in anxiety and fear.

A threat stimulus is represented by the spider. The brain circuit involved in fear and anxiety is represented by the hippocampus, the amygdala, the stria terminalis and the frontal cortex. These regions are involved in both fear response after the immediate presence of the threat and in the anxious response of an individual facing uncertainty. BNST = Bed Nucleus of the Stria Terminalis.

Beyond RDoC, other approaches focus more narrowly on varieties of clinical problems. Such clinical classification schemes categorize anxiety disorders as pathological entities into multiple diagnostic entities based on phenomenology. Of the anxiety disorders, selective mutism (6) and separation anxiety have a typical onset earlier in childhood; specific phobia peaks at late-childhood and early adolescence and the remaining anxiety disorders onset are more common in adolescence and young adulthood (2,3). Comorbidity among anxiety disorders are common as well as the progression from one disorder to another (7). This makes it especially difficult to isolate each diagnostic category in research.

In child and adolescent anxiety, MRI studies have either relied on case-control design or correlational approaches, relating symptoms or diagnosis to brain-based measures. These two approaches are mutually informative. Dimensional approaches are useful because individual differences in the general tendency to experience fear and anxiety are continuously distributed in the population. Thus, anxiety disorders can be seen as lying on one end of the dimension, and subclinical symptoms among healthy subjects at another end. This dimensional approach is especially relevant to children, since fearful thoughts represent one component of normal development. Despite generating knowledge about the potential neurological mechanisms involved in fear and anxiety, there is a gap to fill that can relate research findings to differential diagnosis and clinical outcomes. This review focuses on novel MRI studies that investigate the underlying circuits related to childhood anxiety and recent advances in generating useful clinical application of neuroimaging biomarkers to patients. While the field remains years away from such application, recent findings provide a path for pursuing this long-term goal.

Neural circuits of Fear and Anxiety

Research in basic neuroscience can help focus the lens of imaging techniques on particular aspects of brain structure or function. Such basic research connects many brain regions to the mammalian response to threats, which are engaged when the organism detects stimuli capable of producing harm. Many of these brain regions are depicted in Figure 1. These regions include amygdala, stria terminalis (particularly the bed nucleus), prefrontal cortex, anterior cingulate cortex and hippocampus (8). Fear and anxiety, viewed as the physiological response to danger detection, is readily observed across species (9). The amygdala has been implicated in threat conditioning (10) and response (11), and more recently in valence and salience (12). The amygdala is a complex structure, with multiple subregions that regulate distinct phenomena. The basolateral amygdala integrates sensory information, and excites the central nucleus of the amygdala, which in turn projects to other regions triggering fear response. There is evidence that the bed nucleus of the stria terminalis (BNST) is another key region, being associated with sustained threat response, thus being central to responses in humans that can be characterized as provoking “anxiety” (13). Such recent work has pointed towards the existence of precisely functioning microcircuits. These circuits regulate distinct responses, such as approach or avoidance, to one or another stimulus from circuitry components that lie very close to one another.

Key sectors of the human neocortex possess no homologue with other species and, for those regions, the translation of animal findings to humans is difficult. Other structures do exhibit such homology, such as portions of the insula, including mid and posterior sections, which are implicated in the monitoring of internal stimuli and the regulation of response to aversive stimuli (14). In humans, portions of the medial PFC (15) and anterior cingulate cortices (16) function as part of the default mode network (DMN) (17), which relates to threat responding and anxiety. These regions have been implicated in cognitive processes such as episodic memory and self-representations. Animal models of anxiety in rodents associate portions of the prefrontal cortex, involved in working memory function, with afferent and efferent connections to the amygdala (9). The dorsolateral PFC (dlPFC) is thought to exert a fundamental role in maintenance of goal representation and motor plans for achieving such goals. These functions enable working memory in humans, and there is evidence of altered dlPFC activation in anxiety patients (18). The higher cognitive functions associated with frontal cortical regions are targeted by current therapeutic approaches for the treatment of anxiety disorders (for review see (19)).

Basic science research suggests that parallels are likely to exist in aspects of fear and anxiety among children, adolescents and adults. However, in humans, a robust body of evidence shows that the brain changes during development (20). Structural MRI shows that the cortex volume increases during early childhood, decreasing during late childhood and adolescence. Cortical thickness results point to a monotonic decline from childhood through adolescence (21). The hippocampus and amygdala volume appear to peak early in life, followed by a relatively stable period during adolescence and young adulthood (22). There is at least some evidence of difference in the slope (rate of change) of the right ventromedial PFC in youths with any anxiety compared to healthy volunteers (23).

Likewise, brain connectivity patterns appear to change during childhood and adolescence (24). Overall findings using different methodology show an increase in integration of different brain regions during this period (24,25). There is data suggesting that the within-subject connectivity is relatively stable over time in healthy subjects, including across adolescence (26). Interestingly, the reliability of connectome metrics does not appear to be stable over time, being less reliable in infancy and old age, than in adulthood (24). It is still unknown how subtle changes detectable at group levels are related to changes in subject specific patterns of connectivity over development (27). This opens a line of research for the use of brain connectivity in predictive work (see below).

How different stages of brain development relate to the emergence of symptoms is still unclear. There are, however, a few studies directly comparing childhood and adolescent anxiety to adult anxiety. In a study with 200 participants aged 8 to 50, Gold et. al. (28) showed group differences between anxious and healthy adults during threat appraisal in the vmPFC but not in youth. In the inferior temporal gyrus youths with anxiety showed greater activation during memory tasks, but not appraisal, while the opposite was seen in adults.

ENIGMA and other large scale initiatives

Research on genetics shows that understandings of mental illnesses benefit from large-scale research combining data across multiple research groups. This suggests the promise of creating similar approaches with imaging. The ENIGMA (Enhancing NeuroImaging Genetics through Meta Analysis) Consortium supports multi-group efforts that are generating valuable insights (29). These insights concern the nature of altered brain structure in several psychiatric disorders. The ENIGMA-Anxiety working group includes subgroups dedicated to specific disorders (Generalized Anxiety Disorder, Social Anxiety Disorder, Panic Disorder, Specific Phobia) (30). Unlike traditional meta-analyses, ENIGMA conducts preprocessing and analytical steps simultaneously across samples (31), to reduce bias arising from different preprocessing and quality assurance methods. Many ENIGMA working groups have further implemented mega-analysis methodology, whereby the individual participants’ data (IPD) are shared within the group. This allows for even further standardization of processing methods, as all data can be assessed by the team that is leading the analysis (31). The sharing of IPD helps improve the consistency of inclusion criteria, treatment of confounds and handling of missing data (32).

Recently finalized analyses by ENIGMA-Anxiety (Harrewjin, submitted) show no evidence of structural alteration in patients with GAD when compared with healthy subjects. The interaction between age by GAD was also non-significant. This is in line with previous work that showed no evidence of structural differences in anxiety disorders, dimming enthusiasm for attempts to find diagnosis-specific structural findings.

There is work in progress to conduct a similar mega-analysis with functional data. Resting-state fMRI (rs-fMRI) data from participant centers is being centralized for processing and analysis. The advantage of resting-state data as opposed to task fMRI is that it is collected somewhat similarly across all centers, with the research subject being asked to remain still and look at a fixation cross, or close their eyes during scanning. Rs-fMRI measures have been shown to be relatively stable over time and conditions (26). Although limited by relatively small sample sizes, previous work shows that anxiety patients exhibit alterations in within and between network connectivity in using rs-fMRI when compared with controls. A recent meta-analysis showed alterations in multiple networks: affective , salience , default mode and executive control (33).

Another promising avenue involves studies of prospective cohorts such as the Adolescent Brain Cognitive Development (ABCD) (34), Generation R (35) and the Brazilian High-Risk Cohort Study (BHRCS) (36). These studies aim to collect data from many subjects that includes neuroimaging measures and behavioral data from childhood through young adulthood. One difficulty in past work is that both brain measures and measures of psychopathology change during this period (Figure 2). Results from prospective studies will help identify how different symptom trajectories relate to differences in neuroimaging measures. A difficulty with large studies is that they require streamlined data acquisition that is feasible for many subjects across multiple centers. This limits the depth of coverage for particular dimensions (e.g.: anxiety).

Figure 2 -. Schematic representation of correlation vs prediction.

Cross-sectional studies are limited to correlating neuroimaging findings - in this case represented by brain connections - to concurrent symptoms. Psychopathology can present multiple dimensions - represented by the blue lines - with fluctuating time course. Predicting outcomes based on brain measures is still not possible. Cross-sectional designs do not allow for the full comprehension of the timing of brain differences found in anxiety. It is unclear if brain changes remain after symptom remission or worsening of symptoms and how brain functioning reflects shifts in diagnosis or the development of comorbidities; or even how underlying traits that confer risk to later development of anxiety disorders relates to the brain.

Translating findings to clinical practice: prediction

Large datasets support applications of prediction algorithms in neuroimaging research (37,38). Most knowledge comes from studies deploying classical statistical inference when quantifying associations between variables. However, the effect sizes in applications of this approach to imaging are frequently small, as noted in the material appearing above on structural findings (39). Associations between behavior and fMRI measures may reflect aggregate effects from small, diffuse effects as opposed to large localized effects (40). Small sample sizes can lead to underpowered studies and spurious associations. Likewise small effects can become detectable in large datasets, rendering statistically significant findings clinically irrelevant, or leading to the detection of unrelated noise as signal (41,42). Complicating this even further is the fact that many psychiatric disorders and symptoms have overlapping imaging findings.

Work that uses a predictive approach estimates an outcome using data from one or more imaging modality. Figure 2 illustrates some of the considerations that inform such predictive approaches. The available data is usually split between a training and a test dataset, with the training dataset being used to create (or train) the model, which is then used to make estimates of the variable of interest in the test dataset (which can be constituted of a single test subject). An additional dataset can be used to further assess model validity (43). This is particularly important when a cross-validation method is used. Many predictive methods have been applied in neuroimaging with no superiority of one approach over another (e.g., deep learning, support vector machines, support vector regression, random forests, clustering, linear discriminant analysis); some methods seek to combine multiple algorithms (44). Some of these methods provide models that produce difficult-to-interpret results. Neural networks and deep learning may rely on a series of non-linear relations that have been called ‘black box’ models. The interpretation of the effect of each parameter in the prediction model, and thus the biological meaning, might be more difficult to obtain (45,46). There are other approaches, however, that favor interpretability. One of such approaches, called connectome predictive modeling (CPM) uses a connectivity matrix (the correlation between brain regions one with another), which is then correlated with a measure of interest. The resulting correlations are then used to select the connections of the correlation matrix. The sum of the selected connections is then used in the model for prediction (47). CPM has been used in healthy volunteers to predict trait anxiety with promising results (48).

Much work is still needed before neuroimaging predictive models make the leap to become clinically significant. To date, work in the field has focused on predicting the category of diagnosis or a symptom or neurocognitive score. There has been insufficient work in clinically significant topics, such as treatment outcome. There is an inherent difficulty when trying to build predictive models for specific anxiety diagnosis in children. Anxiety diagnoses are frequently comorbid and it is unclear how to accommodate that in predictive models. Another difficulty is that prediction algorithms might not perform well on an imbalanced dataset (i.e.: many more healthy volunteers than patients) (49). This might make it particularly difficult to apply predictive models on infrequent diagnosis or symptoms, even with large datasets becoming available. One question that remains is if predictive algorithms will be able to work across different age ranges, ethnicities, culture and across all the slight phenotypic variations seen in clinical settings.

Conclusion

Recent work has been instrumental in helping unveil mechanisms associated with fear, anxiety and anxiety disorders. Ongoing collaborative research efforts such as ENIGMA and multi-site studies are underway and will help make it clearer how these disorders develop during childhood and adolescence. Hopefully the rapid growing field will help develop clinically useful markers of anxiety.

KEY POINTS:

1. Fear follows from immediate encounters with a threat, whereas anxiety arises when anticipating a potential forthcoming encounter.

2. Anxiety disorder symptoms are believed to relate to function in neural circuits involved in threat processing and response.

3. Rodent, non-human primate, and human studies implicate similar structures in the response to danger: amygdala, hippocampus, medial prefrontal cortex, and the bed nucleus of the stria terminalis.

4. Neuroimaging studies have shown small effect sizes for when comparing structure and function in particular brain regions among patients with anxiety and healthy volunteers.

5. Large datasets are becoming available and will allow for novel multivariate approaches for analyzing neuroimaging data that may detect clinically-relevant findings.