Altered hippocampal microstructure and function in children who experienced Hurricane Irma

Abstract Hurricane Irma was the most powerful Atlantic hurricane in recorded history, displacing 6 million and killing over 120 people in the state of Florida alone. Unpredictable disasters like Irma are associated with poor cognitive and health outcomes that can disproportionately impact children. This study examined the effects of Hurricane Irma on the hippocampus and memory processes previously related to unpredictable stress. We used an innovative application of an advanced diffusion‐weighted imaging technique, restriction spectrum imaging (RSI), to characterize hippocampal microstructure (i.e., cell density) in 9‐ to 10‐year‐old children who were exposed to Hurricane Irma relative to a non‐exposed control group (i.e., assessed the year before Hurricane Irma). We tested the hypotheses that the experience of Hurricane Irma would be associated with decreases in: (a) hippocampal cellularity (e.g., neurogenesis), based on known associations between unpredictable stress and hippocampal alterations; and (b) hippocampal‐related memory function as indexed by delayed recall. We show an association between decreased hippocampal cellularity and delayed recall memory in children who experienced Hurricane Irma relative to those who did not. These findings suggest an important role of RSI for assessing subtle microstructural changes related to functionally significant changes in the developing brain in response to environmental events.


| INTRODUC TI ON
As Hurricane Irma, the most powerful Atlantic hurricane in recorded history, moved toward the southern coast of Florida in September 2017, over 6 million Florida residents were evacuated from their homes leading to food, water, and fuel shortages throughout the state (National Weather Service, 2018). Irma made landfall as a Saffir-Simpson Category 4 hurricane, ripping down power lines, and leaving two-thirds of individuals and families in Florida without power as it tore roofs off their homes and flooded their streets. Ultimately, Irma left a death toll of over 120 in its wake in Florida alone (Issa et al., 2018).
Psychophysiological tools, such as scalp-recorded electroencephalogram (EEG) recordings in response to environmental cues of threat and error monitoring, have proven to be useful in identifying neural correlates of different forms of psychopathology across the life span (e.g., Shankman & Gorka, 2015). Over 50 studies have focused on the error-related negativity (ERN), an event-related potential (ERP) component typically measured at frontocentral electrodes 50-100 ms following commission of an error (Olvet & Hajcak, 2008;Weinberg, Riesel, & Hajcak, 2012). Source localization studies and investigations employing EEG and functional magnetic resonance imaging (fMRI;Debener et al., 2005;Fitzgerald et al., 2005;Mathalon, Whitfield, & Ford, 2003) have identified the anterior cingulate cortex (ACC) as the region of the brain that generates the ERN. The ACC is the primary brain mechanism involved in online monitoring for conflict between simultaneously active but incompatible streams of information (Shiels & Hawk, 2010). Greater ERN amplitudes are theorized to reflect processes of heightened conflict monitoring (Yeung, Botvinick, & Cohen, 2004), and sensitivity to threat (Weinberg et al., 2016) and punishment (Shackman et al., 2011;Zambrano-Vazquez, & Allen, 2014). Research has indicated that the ERN represents a trait-like neural response to errors, as it demonstrates strong test-retest reliability and rank-order stability. Furthermore, ERN amplitudes appear to be multiply determined by genetic and environmental influences (Weinberg, Klein, & Hajcak et al., 2012).
Research has shown that the ERN is a neural correlate of psychopathology, particularly anxiety disorders. Specifically, enhanced (i.e., more negative) ERN in anxious individuals is frequently observed, and has been consistently replicated in both adult (Hajcak, Klawohn, & Meyer, 2019;Weinberg, Dieterich, & Riesel, 2015) and pediatric samples (Ladouceur, Dahl, Birmaher, Axelson, & Ryan, 2006;Meyer, 2017). Enhanced ERN has been shown to prospectively predict the onset of anxiety psychopathology in school-age children (Meyer, Proudfit, Torpey-Newman, Kujawa, & Klein, 2015), adolescents (Meyer, Nelson specifically, time-marked, unpredictable events such as hurricanes and other natural disasters have been linked to alterations in brain and behavior (Chen et al., 2019;Ke et al., 2018;Kessel et al., 2018;Kopala-Sibley et al., 2016), with some evidence suggesting that children are disproportionately affected by natural disasters relative to adults (Satcher et al., 2007). Notably, prior investigations of neural mechanisms impacted by other forms of unpredictable stress provide important insight into how unpredictable events can lead to lasting or transitory alterations in brain and behavior.
The experience and expression of emotions related to environmental events is associated with the limbic system (Blum et al., 2000).
Time-limited stressors can inhibit the induction of long-term potentiation (LTP) in the hippocampus (Diamond et al., 1990;Foy et al., 1987), with predictability modulating the magnitude of LTP (Kavushansky et al., 2006;Shors et al., 1989Shors et al., , 1990. LTP is a reflection of synaptic plasticity and associated with dendritic arborization and the formation of new synapses (Bliss & Gardner-Medwin, 1973;Bliss & Lømo, 1973). Hippocampal neurogenesis is also linked to neural plasticity and together these processes are thought to support optimal exploration of novel events and environments (Glasper et al., 2012). Unpredictable stressful events are associated with hippocampal microstructure changes in adult animal models including reductions in hippocampal neurogenesis (Gould et al., 1997Tanapat et al., 2001), spine density in basal dendrites of CA1 (Diamond et al., 2006), and apical dendrites of CA3 (Chen et al., 2008(Chen et al., , 2010Magariños & McEwen, 1995;Stewart et al., 2005). Importantly, for the purposes of this study, alterations in the hippocampus to unpredictable events are observed in younger non-human animals (Hollis et al., 2013;Romeo, 2017;Simon et al., 2005;Tanapat et al., 1998). Specifically, pre-pubertal and adolescent non-human animals exhibit protracted stress-responses relative to adults (Lupien et al., 2009;Romeo, 2017). While these long lasting changes may be associated with elevated risk for disease (Kim et al., 2015), dendritic atrophy and decreases in hippocampal neurogenesis may facilitate adaptive responding and focus on a stressor by mitigating structural changes and increasing temporary vigilance in the face of uncertain environmental conditions (Cameron & Schoenfeld, 2018).
Together this work highlights the responsivity of the hippocampus to unpredictable experiences. However, less is known about microstructural changes that may underlie these associations in the developing human brain underscoring the importance of elucidating behaviorally significant changes in the hippocampus following natural disasters like Irma.
Recent advances in diffusion-weighted imaging (DWI) allow for non-invasively examining microstructural detail of the hippocampus.
For example, changes in hippocampal microstructure (e.g., neurite density and dispersion) measured with advanced DWI techniques are associated with age-related changes in memory performance in adults (Radhakrishnan et al., 2020;Zhang et al., 2012), demonstrating that probing the microstructure of gray matter may be important for characterizing functionally relevant changes. Other recent methodological advances using diffusion magnetic resonance imaging provide further opportunity to examine the microstructure of deep gray matter by providing estimates of cell and neurite density. Restriction spectrum imaging (RSI) provides an estimate of the cellularity of gray-matter tissue in specific areas of the brain by separating the diffusion-weighted MRI signal into restricted and hindered diffusion components, which represent intracellular and extracellular water signal, respectively F I G U R E 1 Restriction spectrum imaging schematic. Intracellular water diffusion within cells (neurons in orange; glial cells in gray) is restricted, while extracellular water diffusion (teal) is hindered. Stress-related alterations in the hippocampus (e.g., decreases in neurogenesis, dendritic atrophy) would be associated with a decrease in restricted (intracellular) diffusion (top panel) relative to control conditions (i.e., no acute stress; typical neurogenesis) (bottom panel) anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder (ODD;Kendall et al., 2001). It is well established that childhood comorbidity is associated with significantly worse short-and long-term psychosocial impairments (Fraire & Ollendick, 2013;Franco, Saavedra, & Silverman, 2007). Despite moderate rates of anxiety and externalizing psychopathology comorbidity, their neurobiological origins are not well understood, which is significant given increasing clinical research emphasis on understanding the underlying pathophysiology of common psychiatric disorders and syndromes, and identifying the extent to which disorders and subtypes are associated with similar or unique neurophysiological characteristics (Insel et al., 2010;Shankman & Gorka, 2015).
Psychophysiological tools, such as scalp-recorded electroencephalogram (EEG) recordings in response to environmental cues of threat and error monitoring, have proven to be useful in identifying neural correlates of different forms of psychopathology across the life span (e.g., Shankman & Gorka, 2015). Over 50 studies have focused on the error-related negativity (ERN), an event-related potential (ERP) component typically measured at frontocentral electrodes 50-100 ms following commission of an error (Olvet & Hajcak, 2008;Weinberg, Riesel, & Hajcak, 2012). Source localization studies and investigations employing EEG and functional magnetic resonance imaging (fMRI;Debener et al., 2005;Fitzgerald et al., 2005;Mathalon, Whitfield, & Ford, 2003) have identified the anterior cingulate cortex (ACC) as the region of the brain that generates the ERN. The ACC is the primary brain mechanism involved in online monitoring for conflict between simultaneously active but incompatible streams of information (Shiels & Hawk, 2010). Greater ERN amplitudes are theorized to reflect processes of heightened conflict monitoring (Yeung, Botvinick, & Cohen, 2004), and sensitivity to threat (Weinberg et al., 2016) andpunishment (Shackman et al., 2011;Zambrano-Vazquez, & Allen, 2014). Research has indicated that the ERN represents a trait-like neural response to errors, as it demonstrates strong test-retest reliability and rank-order stability. Furthermore, ERN amplitudes appear to be multiply determined by genetic and environmental influences (Weinberg, Klein, & Hajcak et al., 2012).
It is important to note that ERN findings in the anxiety and externalizing literatures have not been entirely consistent, such that symptom/disorder and ERN patterns do not always demonstrate the expected directional relationships, particularly when there is comorbid psychopathology. For example, some adult studies have found that ERN and anxiety disorder associations differ when individuals have comorbid depression (Weinberg, Klein, et al., 2012;Weinberg, Kotov, & Proudfit, 2015). In addition, Stieben et al. (2007) found that the association between externalizing problems and a blunted ERN was less apparent among children with comorbid internalizing problems.
Despite the high prevalence of anxiety disorders (Beesdo et al., 2009;Copeland et al., 2014) and frequent co-occurrence of externalizing problems in childhood (Angold, Costello, & Erkanli, 1999;Kendall et al., 2010;Verduin & Kendall, 2003), how anxiety and externalizing comorbidity patterns relate to ERN profiles has yet to be tested. Research that examines these relations could have the potential to resolve some of the inconsistencies observed in the ERN pediatric literature, increase precision in our understanding of neural correlates of anxiety comorbidity subtypes, and elucidate primary sources of impairment that could have translational implications.
Accordingly, the primary aim of this preliminary study was to be one of the first to investigate how ERN responses could differentiate anxious subtypes as defined by those with and without comorbid externalizing problems in a pediatric sample. Healthy youth, defined as those without lifetime histories of psychiatric disorders, were also included in order to examine psychiatric symptom and ERN relations dimensionally. The overall sample consisted of anxious youth and healthy controls (ages 7-19 years old) who were recruited for a study of the treatment of pediatric anxiety disorders. Based on prior work on ERN in anxiety disorders (Meyer, 2017;Meyer et al., 2013), we hypothesized that youth with greater anxiety symptoms who did not have co-occurring externalizing problems would demonstrate enhanced ERN (i.e., greater sensitivity to errors). Drawing on preliminary studies suggesting anxiety and ERN relations could be qualified by psychiatric comorbidity (Stieben et al., 2007), we predicted that youth with co-occurring anxiety symptoms and externalizing | 3 CONLEY Et aL. (White, Leergaard, et al., 2013;White, McDonald, et al., 2013;White et al., 2014; Figure 1). Accordingly, a greater fraction of restricted component (i.e., a greater proportion of restricted motion of water molecules) is believed to indicate greater tissue cellularity, an interpretation which has been histologically confirmed (White, Leergaard, et al., 2013). This methodology provides a novel opportunity to investigate microstructural changes in the developing human brain following exposure to unpredictable and/or stressful events.
Here, we utilized RSI to test for microstructural differences in the hippocampus of children exposed to Hurricane Irma, a naturalistic, unpredictable event. More specifically, we evaluated whether Irma-exposed children, tested in the year prior to Hurricane Irma, show decreased hippocampal cellularity relative to a non-exposed control group, tested during the year before Hurricane Irma. To test for regional specificity of any hippocampal effects, we examined all subcortical brain regions. To evaluate functional consequences of Irma and changes in hippocampal cellularity between Irma-exposed and non-exposed groups, we tested whether Irma-exposure was associated with group differences in hippocampal-related memory (i.e., delayed recall). We hypothesized that the experience of Hurricane Irma would be associated with decreases in (a) hippocampal cell density and (b) delayed recall memory.

| Participants
Participants were 9-and 10-year-old children from the South  (Compton et al., 2019;Garavan et al., 2018). To assess whether Hurricane Irma exposure was related to hippocampal microstructure and memory, participants were divided into non-exposed and Irma-exposed groups based on scan date. Although Hurricane Irma made landfall in Florida on September 10, 2017, National Hurricane Center monitoring and media coverage began on August 26, 2017 (14 days prior; Blake, 2017). Because media coverage of disasters and stressful events has been associated with acute stress and later PTSD (Aber et al., 2004;Gil-Rivas et al., 2007;Pfefferbaum et al., 2001Pfefferbaum et al., , 2003Thompson et al., 2019), ABCD participants tested between August 26 and September 10 were excluded to avoid anticipatory effects of the hurricane resulting from news sources ( Figure 2).
Our primary hypotheses focused on differences between non-exposed and Irma-exposed groups from the same geographical location and social structure (i.e., South Florida) that were group matched on gender, pubertal status, age, and family income. However, to assess for specificity of the results to individuals proximal to Hurricane Irma, ancillary analyses considered differences between groups tested in the year prior to and following Hurricane Irma from a distal, non-exposed Northeastern ABCD site using a similar sample size and an identical scanner platform with the same DWI sequences. The operational definition of pre-and post-Irma groups was identical to the South Florida non-exposed and Irma-exposed groups (i.e., children scanned prior to August 26 and after September 10; Figure S2), however, geographic location, pubertal status, family income, parent education level, and racial and ethnic diversity varied significantly between these proximal and distal sites (see Figure S3 for distribution of demographic variables across sites).
F I G U R E 2 Experimental design. Participants from the same geographic location were divided into two groups based on when Adolescent Brain Cognitive Development baseline assessments and magnetic resonance imaging scans were acquired. The Irma non-exposed group was scanned during the year prior to Hurricane Irma. The Irma-exposed group was scanned during the year following Hurricane Irma anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder (ODD;Kendall et al., 2001). It is well established that childhood comorbidity is associated with significantly worse short-and long-term psychosocial impairments (Fraire & Ollendick, 2013;Franco, Saavedra, & Silverman, 2007). Despite moderate rates of anxiety and externalizing psychopathology comorbidity, their neurobiological origins are not well understood, which is sig- Research has shown that the ERN is a neural correlate of psychopathology, particularly anxiety disorders. Specifically, enhanced (i.e., more negative) ERN in anxious individuals is frequently observed, and has been consistently replicated in both adult (Hajcak, Klawohn, & Meyer, 2019;Weinberg, Dieterich, & Riesel, 2015) and pediatric samples (Ladouceur, Dahl, Birmaher, Axelson, & Ryan, 2006;Meyer, 2017). Enhanced ERN has been shown to prospectively predict the onset of anxiety psychopathology in school-age children (Meyer, Proudfit, Torpey-Newman, Kujawa, & Klein, 2015), adolescents (Meyer, Nelson

| Image preprocessing, RSI, and volumetric data
Diffusion magnetic resonance imaging and structural magnetic resonance imaging (sMRI) data were processed by the ABCD Study Data Analysis, Informatics and Resources Center using methods previously detailed in Hagler et al. (2019) . Restricted normalized isotropic (N0) metrics were calculated for subcortical gray matter using a linear estimation approach (White, Leergaard, et al., 2013;White, McDonald, et al., 2013;White et al., 2014) with atlas-based segmentation (Fischl et al., 2002). Hippocampal volume differences were tested using data computed from the pre-processed T1 images using FreeSurfer v5.3 and labeled using an atlas-based volumetric segmentation procedure (Fischl et al., 2002;Hagler et al., 2019).

| Hippocampal-related behavioral function
The RAVLT is a widely used and robust measure of auditory learning, memory, and recall (Lezak et al., 2004;Luciana et al., 2018). The test involves five learning trials of 15 unrelated words (list A). After each trial, participants are asked to recall as many words as possible. After the initial five learning trials, participants are presented with a distractor list of 15 new words (list B) and are then asked to recall as many words as possible from the new list (list B). Next, an immediate recall trial is assessed for words from the initial list (list A). After a 30-min delay (where participants complete other non-verbal tasks or rest), a final delayed recall trial is assessed for words from the initial list (list A). Previous work has established the RAVLT as a reliable measure of hippocampal integrity (Saury & Emanuelson, 2017) and hippocampaldependent memory (Stevenson et al., 2018), linking the delayed recall trial in particular to hippocampal function (Wolk et al., 2011). Here, we assessed a behavioral correlate of hippocampal function using performance (total correct) on the delayed recall trial (i.e., RAVLT Trial VII).

| Analytic approach
Analyses were performed in R version 3.6.3 (R Core Team, 2020) using the gamm4 package (Wood & Scheip, 2020). Mixed-effect models were used to evaluate RSI measures in subcortical regions as well as verbal memory between non-exposed and Irma-exposed groups. For all models, covariates included fixed effects for gender, interview age, race/ethnicity, parental education, and household income, and a random effect for family ID. In addition, RSI models also included intracranial volume, motion, pubertal development, and the interaction of gender and pubertal development as fixed covariates.
Supplemental analyses restricting the sample to only pre-and earlypubertal participants were conducted (Supplemental Analyses 1) in addition to supplemental analyses including trauma history, threat exposure, and history of anxiety disorders and PTSD as other fixed covariates (Supplemental Analyses 2). All analyses were Bonferroni corrected for multiple comparisons. Non-parametric significance was assessed using permutation testing by randomly shuffling data 10,000 times. Non-parametric p-values were computed by dividing the number of times the randomly permuted t-statistic was greater than the observed t-statistic by the number of tests performed (i.e., # observations > |t|/10,000 + 1). Additionally, because there were no a priori hypotheses that effects would be lateralized, RSI was averaged between hemispheres for each subcortical region. Ancillary analyses applied identical models to the distal, Northeastern sample to evaluate if any of the findings detected in the South Florida sample could be attributed to other background cohort characteristics (e.g., sampling protocols, other overlooked events occurring in September 2017).
Following the initial region of interest (ROI) analyses, a post hoc analysis of voxel-wise data was performed to further examine the spatial specificity of our imaging results. Post hoc analyses were performed by applying the same model from the ROI analyses to every voxel within the subcortex utilizing the cifti (Muschelli, n.d.) and lme4 (Bates et al., 2015) packages. Given previous work showing that the effects of stress can impact different subfields of the hippocampus McEwen et al., 2015), the distribution of voxel-wise RSI (i.e., restricted diffusion) values was plotted as a function of the anterior/posterior coordinate axis (i.e., y-axis coordinates) to further evaluate spatial specificity within the hippocampus.

| Decreased hippocampal cellularity in Irmaexposed group
Although a substantial literature across the fields of neuroscience and psychology describes relationships between unpredictable anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder (ODD;Kendall et al., 2001). It is well established that childhood comorbidity is associated with significantly worse short-and long-term psychosocial impairments (Fraire & Ollendick, 2013;Franco, Saavedra, & Silverman, 2007). Despite moderate rates of anxiety and externalizing psychopathology comorbidity, their neurobiological origins are not well understood, which is sig- Accordingly, the primary aim of this preliminary study was to be one of the first to investigate how ERN responses could differentiate anxious subtypes as defined by those with and without comorbid externalizing problems in a pediatric sample. Healthy youth, defined as those without lifetime histories of psychiatric disorders, were also included in order to examine psychiatric symptom and ERN relations dimensionally. The overall sample consisted of anxious youth and healthy controls (ages 7-19 years old) who were recruited for a study of the treatment of pediatric anxiety disorders. Based on prior work on ERN in anxiety disorders (Meyer, 2017;Meyer et al., 2013), we hypothesized that youth with greater anxiety symptoms who did not have co-occurring externalizing problems would demonstrate enhanced ERN (i.e., greater sensitivity to errors). Drawing on preliminary studies suggesting anxiety and ERN relations could be qualified by psychiatric comorbidity (Stieben et al., 2007), we predicted that youth with co-occurring anxiety symptoms and externalizing | 5 CONLEY Et aL. events and macroscale changes in hippocampal structure and function, less is known about how these associations emerge in the developing human brain. Thus, the primary aim of the current study was to determine if RSI could be used to detect subtle microstructural differences in the hippocampus of children exposed to a naturalistic unpredictable event, Hurricane Irma (hypothesis 1).
Consistent with hypothesis 1, and with previous literature linking unpredictable stress to dendritic atrophy and decreased neurogenesis in the hippocampus, lower restricted diffusion signal, consistent with lower cellularity, was observed in the hippocampus of the South Florida Irma-exposed group relative to the South Florida non-exposed control group (β = −3.90 × 10 -3 (SE = 1.35 × 10 -3 ), t = −2.875, p = .004, r 2 (adj) = .05, Δr 2 (adj) = .02; Figure 3a). 3,4 Control analyses demonstrated that these findings are robust to the impact of puberty and various adverse life events or experiences on hippocampal microstructure and are described in supplemental materials. Next, we used permutation testing to evaluate the specificity of this effect to the hippocampus by testing all subcortical regions. Across F I G U R E 3 Differences in restriction spectrum imaging-based hippocampal cellularity among South Florida Irma-exposed relative to non-exposed children. (a) Irma exposure was related to decreased hippocampal cell density; (b) post hoc voxel-wise analysis demonstrated spatial specificity of findings to the hippocampus in right and left hemispheres; (c) results of permutation testing (10,000 + 1 iterations) demonstrated specificity of Irma exposure on hippocampal cellularity and not to other subcortical regions. The histograms show the null distribution of t-statistics with observed effects indicated with vertical lines (solid lines represent non-parametric significance and dashed lines represent non-parametric non-significance) anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder (ODD;Kendall et al., 2001). It is well established that childhood comorbidity is associated with significantly worse short-and long-term psychosocial impairments (Fraire & Ollendick, 2013;Franco, Saavedra, & Silverman, 2007). Despite moderate rates of anxiety and externalizing psychopathology comorbidity, their neurobiological origins are not well understood, which is sig-  all subcortical areas, only the hippocampus showed significant differences in cellularity between the Irma-exposed and non-exposed control group (Figure 3c).
A post hoc voxel-wise analysis of deep gray matter further revealed the spatial specificity of these findings. Consistent with the ROI results above, voxel-wise models revealed a robust relationship between cellularity measured with RSI and Irma exposure in the hippocampus (Figure 3b). Further evaluation of the spatial distribution of differences in hippocampal cellularity between the Irma-exposed and non-exposed groups revealed that this effect was more concentrated in anterior hippocampus (Figure 4). There was no significant difference in hippocampal volume between Irma-exposed and non-exposed groups (β = 12.5 (SE = 32.2), t = 0.389, p = .70, r 2 (adj) = .38, Δr 2 (adj) < .001). Together these results are consistent with the hypothesis that decreased hippocampal cellularity is associated with exposure to Hurricane Irma.

| Distal comparison groups show no differences in hippocampal cellularity or delayed recall
To assess for specificity of results to individuals proximal to Hurricane Irma, ancillary analyses considered differences between individuals tested in the year prior to and following Hurricane Irma from a distal, non-exposed Northeastern ABCD site. This analysis revealed no significant difference in RSI-based hippocampal cellularity between groups scanned prior to and following Irma from the distal, Irma non-exposed Northeastern site (β = −1.67 × 10 −3 (SE = 1.31 × 10 −3 ), t = 1.28, p = .20, r 2 (adj) = .03, Δr 2 (adj) < .001; Figure 6a). Additionally, this comparison analysis revealed no significant difference in delayed recall between groups tested prior to and following Irma from the distal, non-exposed Northeastern site (β = 0.06 (SE = 0.35), t = 0.187, p = .85, r 2 (adj) = .05, Δr 2 (adj) < .001; Figure 6b).

| D ISCUSS I ON
The current study examined whether the unpredictable and uncontrollable events of Hurricane Irma were associated with neural and behavioral signatures in children. Given that unpredictable stressful events have been associated with cellular alterations in the hippocampus, we employed a novel application of RSI to characterize hippocampal microstructural changes (i.e., decreases in cellularity) in response to a naturalistic, unpredictable, and stressful event. The results showed an association between exposure to Hurricane Irma and decreased RSIbased hippocampal cellularity in 9-and 10-year-old children. Analyses across all subcortical regions revealed specificity of this finding to the F I G U R E 4 Group differences in hippocampal cellularity are predominantly concentrated in anterior hippocampus. The distribution of restriction spectrum imaging values (i.e., restricted diffusion) derived from a post hoc voxel-wise analysis based on the Fischl atlas (Fischl et al., 2002) revealed decreases in left and right hippocampal cellularity in the Irma-exposed relative to non-exposed group are more concentrated in anterior hippocampus anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder (ODD;Kendall et al., 2001). It is well established that childhood comorbidity is associated with significantly worse short-and long-term psychosocial impairments (Fraire & Ollendick, 2013;Franco, Saavedra, & Silverman, 2007) hippocampus and not to other subcortical regions, demonstrating the sensitivity of RSI to detect changes in hippocampal microstructure as a function of environmental experiences. The behavioral significance of these findings was supported by significantly poorer delayed recall performance in children exposed to Irma compared to non-exposed children, which was associated with less hippocampal cellularity. These  & Romeo, 2013;Magariños & McEwen, 1995;Stewart et al., 2005) and neurogenesis reductions (Gould et al., 1997Tanapat et al., 2001) in the hippocampus following exposure to unpredictable stress. This work suggests that hippocampal microstructural differences between Irma-exposed and non-exposed children may be related to these previously established cellular alterations following stress. Additionally, the spatial distribution of cellularity differences associated with Hurricane Irma demonstrated specificity in the anterior hippocampus relative to the posterior hippocampus.
This finding is consistent with previous animal work demonstrating decreases in ventral (analogous to anterior in primates; Fanselow F I G U R E 5 Differences in hippocampal function in Irma-exposed and non-exposed children. (a) Irma exposure was associated with poorer delayed recall; (b) less RSI-based hippocampal cellularity was associated with poorer delayed recall memory. RAVLT: Rey Auditory Verbal Learning Test; RSI, restriction spectrum imaging F I G U R E 6 The distal, non-exposed Northeastern site revealed no group differences in (a) hippocampal cellularity; or (b) delayed recall between Pre-and Post-Irma groups. RAVLT, Rey Auditory Verbal Learning Test anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder  & Dong, 2010) hippocampal neurogenesis following unpredictable stress Maggio & Segal, 2007). Although much remains to be learned about the role of neurogenesis and dendritic atrophy in neurobiological and behavioral responses to unpredictable events, previous studies posit that these stress-related alterations may serve to temporarily increase cautiousness and focus on stressors (Cameron & Schoenfeld, 2018;Schoenfeld et al., 2017). For example, memory impairment following stress (Chen et al., 2010;de Quervain et al., 1998;Logue et al., 2018) may serve to promote discernment between stress-relevant and -irrelevant information (Oitzl et al., 2010). In the face of a hurricane, for example, a heightened focus on novel, relevant information may facilitate quick responding in preparation for evacuation. Previous animal work shows that the magnitude of memory deficit following brief, acute stress correlates with reduced density of apical dendritic spines (Chen et al., 2010), suggesting that alterations in hippocampal microstructure may be involved in adaptive responding to environmental stressors. We found that Irma exposure was related to lower delayed recall performance, and that hippocampal cellularity was significantly associated with delayed recall in children, which raises the possibility that changes in hippocampal microstructure detected with RSI may relate to less processing of irrelevant information when adapting to a real-world stressful event (i.e., Hurricane Irma). Alternatively, these changes may be associated with long-term poor outcomes in children following a natural disaster as suggested by Satcher et al. (2007). Further research is needed to understand the temporal dynamics of changes in hippocampal microstructure and how they relate to outcomes, and whether these changes are stable over time.
The current findings advance our current understanding of microstructural changes related to unpredictable events and hippocampal-related memory function; however, limitations of observational approaches must be considered. First, although a substantial literature links hurricane exposure to stress (Garrison et al., 1995;Neria et al., 2008), the baseline ABCD study data do not include a validated measure of self-perceived stress. In addition, psychopathology, social support, and past trauma exposure or other hardships such as financial burden or limited access to resources (Layne et al., 2014) can moderate the effects of unpredictable events on perceived stress (Furr et al., 2010). It is also likely that participants in the South Florida non-exposed control group experienced varying levels of adversity at other times. However, Supplemental Analyses 2 showed that our findings were robust even when controlling for parent-reported trauma history, threat exposure, and clinical disorders such as anxiety and PTSD. Future work that incorporates measures of self-perceived stress and other important moderators and mediators such as family relationships or parental care (Callaghan & Tottenham, 2016) is needed to further inform the specificity of the effect of unpredictable stress on hippocampal microstructure and function. Although the current study suggests an association between hippocampal cellularity and long-term memory, we are unable to relate these changes in memory to specific details about Irma because the ABCD study, on which this study is based, does not evaluate memory details for Irma-related events. Second, while Hurricane Irma provided an opportunity to evaluate the impact of a naturalistic unpredictable event on subcortical microstructure, the South Florida subsample of the ABCD cohort does not reflect the geographic, demographic, or socioeconomic distribution of children across the United States or world (Compton et al., 2019;Garavan et al., 2018) and it is unclear whether our findings would generalize to other populations. Moreover, all participants were 9-and 10-yearolds and previous work suggests that changes in neurodevelopment with age and puberty can interact with the effects of stress (Gee and Casey, 2015;Gunnar & Quevedo, 2007;Lupien et al., 2009;Romeo, 2017). For example, the hippocampus continues to develop throughout adolescence (Anderson & Teicher, 2004;Benes et al., 1994;Giedd et al., 1996;Giedd & Rapoport, 2010;Meyer & Ferres-Torres, 1978) when the brain may be more sensitive to glucocorticoids (Lee et al., 2003) and when gender differences in stress-response and stress-related outcomes such as depression and anxiety emerge (Leussis & Andersen, 2008). Given previous work indicating that the effects of stress during adolescence may be longer-lasting relative to adults (McCormick et al., 2010;Romeo, 2018), it will be important to assess potential sensitive periods of greater risk or resilience throughout development. Similarly, our analyses are cross-sectional and cannot address causality. It is possible that differences in hippocampal microstructure between Irma-exposed and non-exposed children could be attributed to other background factors. Our ancillary analyses of a distal, non-exposed Northeastern site that showed no differences between children assessed the year prior to and following Hurricane Irma provide some confidence of the specificity of the results to Irma exposure. However, future longitudinal analyses and studies with optimized measures of stressor timing, type of stress, and cumulative stress will be needed for understanding how changes in subcortical microstructure emerge and relate to previously established interactions between gender, age, pubertal development, and stress on outcomes (Ho et al., 2012;Siddiqui & Romeo, 2019). It should also be noted that we did not observe significant differences in hippocampal macrostructure (i.e., MRI-based volumetric measures) between Irma-exposed and non-exposed group. While some developmental human imaging studies of stress have shown null volumetric alterations in the hippocampus (De Bellis et al., 1999; or even increases in hippocampal volume (Tupler & De Bellis, 2006;Woon & Hedges, 2008), others have demonstrated that traumatic events are associated with reductions in hippocampal volume (Lambert et al., 2017;Lee et al., 2018;McLaughlin et al., 2016McLaughlin et al., , 2019Weissman et al., 2020). That said, our current findings suggest that RSI may be a more sensitive measure than macrostructure measures of the hippocampus in the human brain. Future work is needed to further evaluate the relationship between RSI and other imaging modalities (e.g., sMRI, fMRI) to better understand how changes in RSI-based cellularity estimates relate to changes in brain structure and function. Lastly, there is speculation that the proportion of restricted diffusion estimated with RSI may also detect long cylindrical glial processes (White, Leergaard, et al., 2013). Given previous work showing microglial sensitivity to changes anywhere from 15% to 23% of anxious youth meet criteria for attention-deficit hyperactivity disorder (ADHD; Angold, Costello, & Erkanli, 1999;Kendall, Brady, & Verduin, 2001), and about 9% of anxious youth meet criteria for oppositional defiant disorder (ODD;Kendall et al., 2001). It is well established that childhood comorbidity is associated with significantly worse short-and long-term psychosocial impairments (Fraire & Ollendick, 2013;Franco, Saavedra, & Silverman, 2007) Research has shown that the ERN is a neural correlate of psychopathology, particularly anxiety disorders. Specifically, enhanced (i.e., more negative) ERN in anxious individuals is frequently observed, and has been consistently replicated in both adult (Hajcak, Klawohn, & Meyer, 2019;Weinberg, Dieterich, & Riesel, 2015) and pediatric samples (Ladouceur, Dahl, Birmaher, Axelson, & Ryan, 2006;Meyer, 2017). Enhanced ERN has been shown to prospectively predict the onset of anxiety psychopathology in school-age children (Meyer, Proudfit, Torpey-Newman, Kujawa, & Klein, 2015), adolescents (Meyer, Nelson It is important to note that ERN findings in the anxiety and externalizing literatures have not been entirely consistent, such that symptom/disorder and ERN patterns do not always demonstrate the expected directional relationships, particularly when there is comorbid psychopathology. For example, some adult studies have found that ERN and anxiety disorder associations differ when individuals have comorbid depression (Weinberg, Klein, et al., 2012;Weinberg, Kotov, & Proudfit, 2015). In addition, Stieben et al. (2007) found that the association between externalizing problems and a blunted ERN was less apparent among children with comorbid internalizing problems.
Despite the high prevalence of anxiety disorders (Beesdo et al., 2009;Copeland et al., 2014) and frequent co-occurrence of externalizing problems in childhood (Angold, Costello, & Erkanli, 1999;Kendall et al., 2010;Verduin & Kendall, 2003), how anxiety and externalizing comorbidity patterns relate to ERN profiles has yet to be tested. Research that examines these relations could have the potential to resolve some of the inconsistencies observed in the ERN pediatric literature, increase precision in our understanding of neural correlates of anxiety comorbidity subtypes, and elucidate primary sources of impairment that could have translational implications.
Accordingly, the primary aim of this preliminary study was to be one of the first to investigate how ERN responses could differentiate anxious subtypes as defined by those with and without comorbid externalizing problems in a pediatric sample. Healthy youth, defined as those without lifetime histories of psychiatric disorders, were also included in order to examine psychiatric symptom and ERN relations dimensionally. The overall sample consisted of anxious youth and healthy controls (ages 7-19 years old) who were recruited for a study of the treatment of pediatric anxiety disorders. Based on prior work on ERN in anxiety disorders (Meyer, 2017;Meyer et al., 2013), we hypothesized that youth with greater anxiety symptoms who did not have co-occurring externalizing problems would demonstrate enhanced ERN (i.e., greater sensitivity to errors). Drawing on preliminary studies suggesting anxiety and ERN relations could be qualified by psychiatric comorbidity (Stieben et al., 2007), we predicted that youth with co-occurring anxiety symptoms and externalizing | 9 CONLEY Et aL.
in environment (Walker et al., 2013), it is possible that the changes in hippocampal microstructure observed in the current study may also relate to changes in glial processes. However, current research on the response of glial cells to unpredictable stress is mixed (Delpech et al., 2016;Jauregui-Huerta et al., 2010;Paolicelli & Ferretti, 2017;Pearson-Leary et al., 2016) and more research is needed.
Overall, the present study provides novel evidence suggesting the potential utility of RSI in detecting behaviorally significant, yet subtle microstructure changes in the hippocampus in response to naturalistic stressful events. Although further research is needed, our results suggest that advanced diffusion MRI methodology may provide novel opportunities to elucidate how cumulative stressors and dimensions of stressors (e.g., threat vs. deprivation; McLaughlin et al., 2014), such as those experienced during the coronavirus disease 19 pandemic, impact youth. Given changing weather patterns and predicted increases in hurricanes and other natural disasters (Bender et al., 2010), our results provide important information that has relevance for researchers, clinicians, families, and policy makers for identifying and ultimately minimizing the impact of these events on young people.

ACK N OWLED G EM ENTS
This work was inspired by the work of the late Bruce McEwen and his novel insights on the impact of stress on brain structure and function.

CO N FLI C T O F I NTE R E S T
The authors declare no competing financial or other interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data used in the preparation of this article were obtained from the ABCD Study (https://abcds tudy.org), held in the NIMH Data Archive (NDA). This is a multisite, longitudinal study designed to recruit more than 10,000 children ages 9-10 and follow them over A full list of supporters is available at https://abcds tudy.org/feder alpartn ers.html. A listing of participating sites and a complete listing of the study investigators can be found at https://abcds tudy.org/ conso rtium_membe rs/. ABCD consortium investigators designed and implemented the study and/or provided data but did not necessarily participate in analysis or writing of this report. This manuscript reflects the views of the authors and may not reflect the opinions or views of the NIH or ABCD consortium investigators. The ABCD data repository grows and changes over time. However, we focus on the sample with 25 participants randomly excluded in our main text to eliminate significant differences in age between the Irma-exposed and non-exposed groups.
2 Given previous work showing that development and hormonal changes during puberty can affect neuronal proliferation within the hippocampus (Allen et al., 2014; Hueston et al., 2017) and stress response (Viau, 2002), supplementary analyses using data from only pre-or early-pubertal participants was conducted and is provided in supplementary materials (Supplemental Analyses 1).
Despite the high prevalence of anxiety disorders (Beesdo et al., 2009;Copeland et al., 2014) and frequent co-occurrence of externalizing problems in childhood (Angold, Costello, & Erkanli, 1999;Kendall et al., 2010;Verduin & Kendall, 2003), how anxiety and externalizing comorbidity patterns relate to ERN profiles has yet to be tested. Research that examines these relations could have the potential to resolve some of the inconsistencies observed in the ERN pediatric literature, increase precision in our understanding of neural correlates of anxiety comorbidity subtypes, and elucidate primary sources of impairment that could have translational implications.
Accordingly, the primary aim of this preliminary study was to be one of the first to investigate how ERN responses could differentiate anxious subtypes as defined by those with and without comorbid externalizing problems in a pediatric sample. Healthy youth, defined as those without lifetime histories of psychiatric disorders, were also included in order to examine psychiatric symptom and ERN relations dimensionally. The overall sample consisted of anxious youth and healthy controls (ages 7-19 years old) who were recruited for a study of the treatment of pediatric anxiety disorders. Based on prior work on ERN in anxiety disorders (Meyer, 2017;Meyer et al., 2013), we hypothesized that youth with greater anxiety symptoms who did not have co-occurring externalizing problems would demonstrate enhanced ERN (i.e., greater sensitivity to errors). Drawing on preliminary studies suggesting anxiety and ERN relations could be qualified by psychiatric comorbidity (Stieben et al., 2007), we predicted that youth with co-occurring anxiety symptoms and externalizing | 13 CONLEY Et aL.