Emerging Evidence for Putative Neural Networks and Antecedents of Pediatric Anxiety in the Fetal, Neonatal, and Infant Periods

Abstract

Anxiety disorders are the most prevalent psychiatric disorders in youth and are associated with profound individual impairment and public health costs. Research shows that clinically significant anxiety symptoms manifest in preschool-aged children, and correlates of anxiety symptoms are observable in infancy. Yet, predicting who is at risk for developing anxiety remains an enduring challenge. Predictive biomarkers of anxiety are needed before school age when anxiety symptoms typically consolidate into diagnostic profiles. Increasing evidence indicates that early neural measures implicated in anxiety and anxious temperament may be incorporated with traditional measures of behavioral risk (i.e., behavioral inhibition) to provide more robust classification of pediatric anxiety problems. This review examines the phenomenology of anxiety disorders in early life, highlighting developmental research that interrogates the putative neurocircuitry of pediatric anxiety. First, we discuss enduring challenges in identifying and predicting risk for pediatric anxiety. Second, we summarize emerging evidence for putative neural antecedents and networks underlying risk for pediatric anxiety in the fetal, neonatal, and infant periods that represent novel potential avenues for risk identification and prediction. We focus on evidence examining the importance of early amygdala and extended amygdala circuitry development to the emergence of anxiety. Finally, we discuss the utility of integrating developmental psychopathology and neuroscience to facilitate future research and clinical work.

Anxiety disorders are the most prevalent psychiatric disorders in youth, affecting approximately a third of children and adolescents (1-3), and are increasingly recognized as neurodevelopmental in etiology (4). Anxiety problems emerge early. The median age of initial anxiety disorder diagnosis is 6 years (2); clinically significant anxiety symptoms manifest and can be impairing in preschool-aged children (1,3); and anxious features are observable in infancy, including negative affect, hypervigilance, and behavioral withdrawal (5). Childhood anxiety problems are often persistent (6), are associated with poor adult outcomes (7-10), and place a heavy burden on health care services and society (11,12). The public health significance of anxiety emphasizes the need to identify predictive biomarkers of anxiety symptoms or behavioral correlates, ideally before school age, when anxiety problems typically consolidate into diagnostic profiles (13,14). However, the etiology and expression of anxiety are complex and change across development at the behavioral and neural levels of analysis (15). Although pediatric anxiety risk has been widely studied, predicting who is at risk for developing anxiety remains challenging.

This review examines the phenomenology of anxiety disorders in early life, highlighting developmental research that interrogates the putative neurocircuitry involved in the etiology and maintenance of anxiety. First, we discuss barriers to identifying and predicting risk for pediatric anxiety using a traditional behavioral risk phenotype approach. Second, we review both normative neurodevelopment in the fetal, neonatal, and infant periods and putative neural antecedents and networks underlying risk for pediatric anxiety, with particular focus on the amygdala and extended amygdala circuitry. Finally, we address challenges to incorporating neural markers into pediatric anxiety risk assessment and the importance of integrating developmental psychopathology and neuroscience into future research and clinical work.

CHALLENGES TO IDENTIFYING AND PREDICTING RISK FOR PEDIATRIC ANXIETY

Identifying early risk factors for pediatric anxiety is complicated by the fact that anxiety results from the dynamic interplay between genes, environment, and development, with different presentations and trajectories in different individuals (16,17). The onset of symptoms may not align with onset of impairment; early indicators resolve without formal intervention for some; and a minority of at-risk individuals ultimately manifest persistent disorders, with presentations (and therefore diagnostic labels) sometimes varying across development. Longitudinal studies are needed for the following reasons: 1) to distinguish between risk markers and manifestations of transient or persistent impairment requires studies that often extend beyond normal funding periods, and 2) to characterize the phenotypic heterogeneity of impairing profiles over time requires creative study design strategies.

To date, most work examining early predictors of risk for pediatric anxiety has focused on behavioral phenotypes of early risk, particularly inhibited temperamental dispositions, to identify children who might develop anxiety later in life. Multiple studies have examined extreme fearful temperament, most commonly characterized in infancy as behavioral inhibition (BI) (5) or dysregulated fear (DF) (18), as an early predictor of anxiety. However, prospective longitudinal studies and meta-analyses indicate that current characterizations of BI and DF are imperfect predictors of risk for clinically actionable anxiety problems. Specifically, less than 50% of behaviorally inhibited children develop anxiety disorders in childhood or adolescence (19). Research indicates that stable infant BI and DF have discriminant predictive validity for internalizing psychopathology (20), particularly for social anxiety disorder in childhood and adolescence (21-24). Yet, a meta-analysis suggests that only an estimated 40% of children who exhibit BI will go on to develop social anxiety disorder (24). Despite important work advancing our understanding of BI as one of the largest single risk factors for predicting risk for pediatric anxiety, the majority of children who exhibit BI do not meet criteria for social anxiety disorder in later life. This is partly due to heterogeneity in pediatric anxiety and lack of sensitivity and specificity of BI features. Not all fearful children show atypicality in the indices used to characterize BI (e.g., decreased vocalizations, freezing, measures of parasympathetic activity, attention bias to threat) (25). Moreover, research shows that for individuals who are captured by behavioral phenotypes such as BI and DF, these risk profiles may not be sufficient predictors on their own, with a wealth of studies investigating factors moderating the stability of BI and risk for anxiety problems among children who exhibit high BI [i.e., (26-28)]. Indeed, nontemperamental risk factors have already conceptually informed the development of novel, temperamentally based treatment approaches (29,30). Studies on BI moderators indicate that refined measurements of early risk at multiple levels [e.g., parenting (27,29,30) or attention (28)] are needed to specify differences not always observed through behavioral assessment and improve our ability to predict pediatric anxiety.

Given limitations to traditional approaches of predicting risk for pediatric anxiety using behavioral risk phenotypes, there is growing interest in the utility of neuroimaging-based predictive biomarkers (elsewhere called diagnostic biomarkers) to advance classification, prevention, and intervention efforts (31). Neural components implicated in the circuitry of anxiety and anxious temperament (AT), and therefore at the center of this interest, include the amygdala, cingulate cortex, prefrontal cortex (PFC), and autonomic nervous and endocrine systems (4,32). Increasing evidence suggests that neural measures associated with anxiety symptoms may help identify key mediators of phenotypic heterogeneity and may improve prediction of pediatric anxiety risk and outcomes when integrated with traditional behavioral approaches (33). Preliminary proof of concept can be found in recent work indicating that BI and DF are associated with partially distinct patterns of behavior, physiology, and neural and attentional processes (34-37). These results echo work in macaques delineating common biological substrates of inhibited and fearful temperament, as well as selective substrates uniquely associated with these two separate behavioral presentations (17). These findings in macaques and human children emphasize the possible utility of neural markers in identifying differences in risk not always captured by behavioral phenotypes, as well as the relative predictive utility of a single neural or behavioral risk marker. Recent research examining clinical predictors of response to treatment in an adult population demonstrates that neuroimaging predictors sometimes double (38) or triple (39) the amount of variance explained in symptom change following treatment beyond earlier symptom presentation. Identifying early brain-based biomarkers with sufficient sensitivity and specificity to classify or predict a given child’s phenotypic profile would dramatically enhance both the basic and clinical science of pediatric anxiety (40).

In sum, integrating neuroimaging data into multidimensional risk profiles may improve subgroup or individual-level prediction of risk by facilitating identification and characterization of key mediators that elucidate (and may presage) the heterogeneity of anxious risk phenotypes and individual trajectories. Moreover, parsing heterogeneity and characterizing key neural mediators is essential to refining our risk detection approach and promises to move the field toward the development of individualized risk prediction (41) that incorporates multiple levels of analysis, which is critical to effective prevention and treatment.

EMERGING EVIDENCE FOR PUTATIVE NEURAL NETWORKS THAT FUNCTION AS ANTECEDENTS OF PEDIATRIC ANXIETY IN THE FETAL, NEONATAL, AND INFANT PERIODS

Recent advances in neuroimaging methods have enabled investigations of normative brain development in the infant, neonatal, and even fetal periods, yielding insights into when and how neurodevelopment deviates to increase risk for psychopathology. Studies have demonstrated utility and value of early neural markers in predicting broad neurodevelopmental outcomes (42) and specific diagnoses [e.g., autism spectrum disorder (43)]. Although anxiety disorders are increasingly recognized as neurodevelopmental, the research and clinical potential for identification of putative neural markers of risk in the fetal, neonatal, and infant periods remains largely untapped. This is partly because the use of neuroimaging in developmental studies is still relatively new, and the majority of studies examining deviations in early neurodevelopment have not (yet) linked early risk markers to later diagnostic outcomes. While any review is constrained by this limitation, we examine emerging evidence of putative neural correlates of later pediatric anxiety that have been associated with risk for anxiety through a behavioral (primarily temperamental) phenotype or with neural deviations associated with anxiety symptoms in adolescent and adult populations. Although research is needed to investigate, replicate, and extend work on these early risk markers within prospective, longitudinal studies, we believe inclusion of neural measures of risk increases the utility and predictive value of a risk profile, thus leveraging multiple levels of analysis (44). Coupled with behavioral factors (such as BI), neural markers will improve our ability to predict specific individual-level risk for anxiety. We conclude this section by integrating what is known about early risk markers into a more established literature of neural markers in older samples and the role of developmental timing in specificity. The timing of risk marker measurement and identification is vital to identifying whether the measures we understand as markers solely represent risk for a future outcome or sensitivity (e.g., increased likelihood of developing anxiety) rather than evidence of a prior process or trajectory that had already affected development in some way, as indicated by this marker.

Key Principles of Normative Fetal, Neonatal, and Infant Brain Development

Early neurodevelopment unfolds in rapid, yet predictable ways. The second trimester is an important period of fetal brain development during which much gyrification occurs (45-48). Myelination occurs regionally beginning with the brainstem at 29 gestational weeks (49,50), and by 40 gestational weeks approximately 5% of total brain volume contains mature myelinated white matter (51). After birth the brain undergoes a period of dramatic growth, doubling in volume in the first postnatal year (52,53). Myelination rapidly increases in infancy (54,55) with protracted increases in volume and microstructural maturation in childhood (56). Structural connectivity and functional connectivity (FC) are apparent beginning in the fetal period, with subsequent regional and network developmental organization varying spatially and temporally. Across development, the maturation of large-scale brain networks is characterized by weakening of short-range FC and strengthening of long-range FC. These key principles illustrate that the developing brain is regulated by the dynamic processes of cell proliferation, overconnectivity, pruning, and apoptosis, which rewire and reorganize connectivity at the neuronal and network levels (57-59).

Key Principles of Amygdala Development

Converging evidence from nonhuman primate models and work in human infants suggests that the structural connectivity and FC of the amygdala undergo significant modifications during the first years of life (60-62). The amygdala begins to develop in early embryonic life (63). Broad nonlinear growth in the first postnatal year is followed by fine-tuning in the second year (61). Recent cross-sectional research suggests that subregions of the amygdala show distinct patterns of FC with sensory cortex areas by 3 months, with connectivity to associative and frontal cortical areas observable later in childhood and coordination between basolateral amygdala and portions of the PFC already resembling adult connectivity by 5 years (60). These results suggest that characterization of variability in extended amygdala structural and functional network connectivity may inform the search for early risk for a later emerging anxiety disorder. Accumulating evidence points to the importance of coordinated activity (64) of distributed neural circuits, with different neural mechanisms underlying heterogeneity in ATs or symptom presentations (17) and different presentations of preschool-age anxiety symptoms associated with different patterns of amygdala connectivity during emotional tasks at school age (65).

Emerging Evidence for Fetal, Neonatal, and Infant Neural Measures of Risk

As shown via experimental manipulations in nonhuman models (66,67), anxiolytic pharmacological agents (68,69), and neuroimaging methods (17,70), the heterogeneity observed in human pediatric anxiety presentations may likely be due, in part, to the complex and overlapping association between fear and anxiety and their neural correlates that may or may not be dissociable. Importantly, individual, narrowly focused neural markers alone are unlikely to underlie or predict the heterogeneity of pediatric anxiety risk phenotypes, symptoms, or diagnoses (70-72); rather, they are more likely to improve prediction of pediatric anxiety as part of a risk profile, potentially incorporating indices across early development.

Fetal Heart Rate.

Research has repeatedly linked fetal vagal activity and behavioral risk markers for pediatric anxiety, with more recent evidence for associated neural antecedents. A modest link has been shown between high reactivity to novelty at 4 months and fetal cardiac measures in the third trimester, including faster resting fetal heart rate (FHR) (73) and greater FHR reactivity to a maternal stress challenge (74). Posner et al. (75) extended this work by examining neural correlates of FHR in early infancy. These authors showed that fetuses exposed to greater levels of prenatal mood symptoms exhibited greater FHR reactivity to a maternal stress challenge at 34 to 37 gestational weeks, which was mediated by altered amygdala-PFC connectivity at approximately 5 weeks after birth. Specifically, greater FHR reactivity was associated with decreased structural connectivity between the right amygdala and the right ventral PFC, a pattern associated with greater trait anxiety in adults (76). Nonhuman models demonstrate that the amygdala modulates cardiovascular responses to affective stimuli via projections to the brainstem (77). This mechanistic account provides biological justification for FHR as a potential early risk marker for deviant neurocircuitry underlying risk for pediatric anxiety, as supported by these recent data in humans.

Amygdala Volume and Circuitry.

A growing body of evidence shows that amygdala features or amygdala network connectivity in early life are consistent with findings from children, adolescents (78-80), and adults with anxiety disorders (81-83). This work implies that amygdala hyperreactivity is not only a symptom but also potentially a risk factor for pediatric anxiety (84) or interacts bidirectionally with other anxiety risk factors throughout the life span. Postulating a role for the amygdala in anxiety disorders is not new (85), but work advancing a developmental conceptualization of amygdala function that includes interrogation of amygdala network connectivity early in development has increased in recent years.

Several studies have reported associations between neonatal resting-state FC (rs-FC) of the amygdala and later measures of fear in infancy (86,87), broader internalizing behaviors (88), or BI (33), all in the first 2 years of life. Thomas et al. (87) demonstrated that these patterns of connectivity (e.g., amygdala-insula) are specific to early fear and not internalizing behaviors broadly. Rogers et al. (88) also provided evidence that different phenotypes of anxious presentations at 2 years of age may already demonstrate specificity in regional amygdala rs-FC, with BI and general anxiety subscales of the Infant-Toddler Social and Emotional Assessment (89) uniquely associated with amygdala-dorsal anterior cingulate and medial PFC, respectively. Importantly, these connectivity patterns are consistent with work in older children and adults, suggesting that neural differences underlying behavioral phenotypes emerge early and could be indicative of both precocious or canalized fear learning and development and a neurophenotype representing decreased plasticity or sensitivity to environmental factors (87). Collectively, these studies provide important evidence of associations between neural connectivity and behavioral outcomes or symptoms well before most children receive an anxiety diagnosis and highlight the utility of developmental neuroscience in identifying early indicators of later anxiety symptoms, behaviors, and phenotypes.

Additionally, emerging research suggests that putative alterations to offspring amygdala during fetal development may mediate effects of the in utero environment on offspring risk for pediatric anxiety (85). Because the amygdala contains a high concentration of glucocorticoid receptors (90,91), it is hypothesized that prenatal stress (variously operationalized as hypothalamic-pituitary-adrenal axis functioning, psychological distress, frank psychiatric symptoms, negative life events, etc.) can influence in utero glucocorticoid concentrations and subsequently initiate a deviant amygdala neurodevelopmental trajectory. Prenatal cortisol has been associated with increased amygdala volume in later childhood (92). However, relatively few studies have explicitly examined the role of the amygdala in the link between prenatal stress and early life development. Within this emerging literature, Graham et al. (93,94) have shown that prenatal maternal inflammation (93) and cortisol concentrations (94) are associated with amygdala connectivity at birth and later internalizing behaviors in toddlerhood. Additionally, Buss et al. (92) have linked prenatal cortisol concentrations to greater right amygdala volumes in 7-year-olds and shown that amygdala volume partially mediated an association between prenatal cortisol and child internalizing symptoms.

Finally, findings from the GUSTO (Growing Up in Singapore Towards Healthy Outcomes) cohort also provide provisional evidence that deviant neurodevelopmental trajectories begin in early life and can be observed through alterations to neonatal or infant amygdala volume or connectivity. GUSTO results have linked prenatal maternal depression to lower fractional anisotropy (FA) (but not volume) in the right amygdala of neonates approximately 2 weeks old (95). No association was found between maternal prenatal anxiety and neonatal amygdala structures, although prenatal anxiety was linked to lower FA in the right insular cortex (96), an area implicated in anxiety disorders in adults (83). Later in early childhood, GUSTO findings show that prenatal maternal depression is associated with lower functional organization of the cortico-striatoamygdala circuitry and greater right amygdala volumes at 4.5 years old for girls but not boys (97,98). Additional groups have shown differing associations between prenatal maternal depression and anxiety symptoms in boys and girls, with one group reporting lower FA in girls and higher FA in boys exposed to higher levels of prenatal maternal depression and anxiety (99). These findings may underlie differences in anxiety prevalence rates for boys and girls. Further longitudinal research is needed to understand if these deviant early amygdala features are predictive of child behaviors, symptoms, or diagnoses related to pediatric anxiety.

Bed Nucleus of the Stria Terminalis.

The bed nucleus of the stria terminalis (BNST) is part of the extended amygdala. To date, developmental investigation of the BNST in humans is limited by its extremely small size, which makes it difficult to isolate from surrounding neuroanatomy using current 3T neuroimaging methods. The limited human and nonhuman data we have significantly implicate alterations in BNST structure, function, or connectivity in anxiety, making the BNST a novel, untapped target for future work. For example, alterations in increased BNST activity are linked to individual differences in normative anxiety as well as chronic trait anxiety in adolescents (100). Neuroimaging studies show that the anticipation of unpredictable threat engages the BNST; in contrast, predictable threat tasks engage the amygdala, with parallel findings in rodent models (66,101). Recent research from Fox et al. (70) using a nonhuman primate model of early AT showed that FC between the central nucleus of the amygdala and the BNST was correlated and coinherited with AT and accounted for approximately 4% of the variance in AT. Additional results showed that BNST metabolism and amygdala-BNST connectivity are unrelated and account for unique variance in AT that together accounted for approximately 8% of variance in AT. These results suggest that the two imaging markers reflect dissociable mechanisms underlying risk for anxiety. Although these results must be replicated in human children, the broader implications suggest that a single neural risk marker is unlikely to fully explain or predict heterogeneity seen in pediatric anxiety, while using more than one neural marker may improve prediction.

Thinking Developmentally: Incorporating Neural Measures Into Risk Assessment of Pediatric Anxiety

As reviewed above, neural structures and networks implicated in later anxiety develop early, with the amygdala demonstrating adult-like connectivity by early childhood (60). Deviations in amygdala FC observable in infancy may persist into later life, including to areas such as the PFC, which often dampens amygdala responses, a process that can be abnormal in anxiety (102). Research findings in early neurodevelopment align with older samples, finding associations between aberrant amygdala FC and risk for later or concurrent anxiety. This suggests that differences in neural circuits associated with an increased vulnerability for anxiety may emerge in infant and fetal periods and reflect neurodevelopmental deviation just as early in development.

Researchers have recently begun to examine early neurodevelopmental milestones within the context of parameters that reflect expected maturation to understand whether deviations from typical development may indicate accelerated, normative, or decreased neurodevelopmental trajectories and if deviant trajectories may be a marker of risk. For example, longitudinal modeling suggests a fetal brain growth spurt (47,103) or a potential spline point at 28 to 32 weeks of gestational age (45-48,104-108), representing a possible important transitional point in fetal development. Research also implicates unique or interactive effects of various sensitive periods within early development, such as findings that show the early postnatal period (i.e., 0–3 mo) is a possible period of emergence for white matter markers, as research suggests that white matter development may be particularly sensitive to the environment during this period compared with the prenatal period (109). These results are consistent with our understanding of the timing of typical white matter development. In 1-month-old infants, prenatal maternal anxiety and depression was associated with less organized white matter microstructure (99). Additionally, deviation in rs-FC between the PFC and various other neural regions at birth has been associated with behavioral differences at 2 years (33,87), suggesting that neural differences in the same regions implicated in task-based studies with anxious adolescents and adults (110) and rs-FC in adults formerly labeled as BI (111) are present very early in neurodevelopment.

We are beginning to understand that aberrant regional brain growth disturbances likely begin in utero [i.e., (75)] and that prenatal and early postnatal measures are associated with distal behavioral and neural markers of risk [i.e., (92)]. However, it is not yet clear if neural risk markers may be expressed consistently or change across development. Emerging evidence has consistently identified the same marker of risk in multiple cohorts/studies (e.g., right amygdala structure and connectivity) at various points in fetal, neonatal, and infant development. These cross-sectional data are intriguing and must be replicated within individual longitudinal data to support individual-level predictive ability. Prenatal stress is just one area of research to incorporate into neural markers and developmental trajectories as well as measurements of early anxiety risk. It is only with data spanning early development and associated longitudinal diagnostic outcomes that we can best use neural markers together with behavioral and other risk factors.

FUTURE DIRECTIONS AND CHALLENGES

While progress has been made in advancing behavioral temperamental phenotypes (e.g., BI) as risk markers for later emerging anxiety, incorporating neural markers into our conceptualization of early anxiety risk provides a more complete and precise measurement of precursors and correlates of pediatric anxiety. Deviations at a behavioral level are reflective of, and bidirectionally interact with, differences in neural structure, function, and connectivity, conjointly increasing risk for later anxiety. Additionally, accounting for individual variation (e.g., multifinality and equifinality in risk and outcome) is improved only by the inclusion of additional levels of measurement, better accounting for the full developmental context and improving prediction for any given individual.

Several problems hinder our ability to more effectively identify individuals at risk for anxiety, as follows. First, how anxiety is expressed across the life span changes. Second, not all individuals who will develop later anxiety are identifiable via behavioral markers in early life (e.g., lack of sensitivity). Third, early anxious features observed in infancy can reflect the broad presentation of risk for psychopathology seen in early life (e.g., lack of specificity to anxiety). Fourth, many children will transiently express anxious temperamental features and anxiety symptoms in early life and subsequently develop into healthy adults (8,102,112).

Individual, neural-based prediction has shown promise for other diagnoses, such as autism spectrum disorder (43) and Alzheimer’s disease (113), as well as behaviors such as risk-taking (114). Cross-sectional work has used amygdala connectivity to predict individual differences in childhood anxiety (115), and similar neuroimaging markers have accurately differentiated adults with social anxiety disorder from healthy control subjects (116,117), sometimes outperforming demographic, clinical, and genetic markers in individual patient prediction analyses (116). Additional work is required to advance our understanding of deviant early trajectories of anxiety and fear neurobiology and circuitry [e.g., (118,119)] and demonstrate the full utility of neuroimaging-based predictive biomarkers for application to anxiety (120), especially in pediatric samples. Sensitive and specific biomarkers identified before the consolidation of a diagnostic profile promise to catalyze new approaches to early identification and, perhaps, preemptive interventions.

As with any method, neuroimaging faces challenges and limitations. For example, some functional magnetic resonance imaging tasks previously used to measure various aspects of anxiety in older populations have demonstrated poor reliability (121). While imaging of younger individuals has been methodologically challenging in the past, advances in protocols and technology have already improved success rates (122). As we continue to study typical and atypical neural development in early life, methodological advances in imaging young populations will allow for the collection of higher-quality data and may increase the specificity, robustness, and utility of neural markers. For example, recent latent factoring of neural indices suggests that anxiety symptoms can be separated from other behavioral indices such as irritability in childhood samples (123).

While the relative cost of neuroimaging makes it prohibitive as a singular level 1 screening method for identifying risk for childhood anxiety, further research will clarify when specific imaging can provide essential clarification of an individual’s risk for later anxiety, such as known genetic risk [e.g., (124)] or early physiological measures such as neonatal cardiac activity [e.g., (125)]. Further research into neural risk markers can elucidate mechanisms and moderators of developmental trajectories of anxiety, highlighting opportunities for prevention and intervention at an individual level. In the future, an integration of neural, physiological, and behavioral indicators may provide precise, individual-level predictions of anxiety risk before clinically significant symptoms emerge or consolidate.