Anxiety and threat-related attentional biases in adolescents with fragile X syndrome

Abstract

Background:

Fragile X syndrome (FXS) is a single-gene disorder highly associated with anxiety; however, measuring anxiety symptoms in FXS and other neurogenetic syndromes is challenged by common limitations in language, self-awareness, and cognitive skills required for many traditional assessment tasks. Prior studies have documented group-level differences in threat-related attentional biases, assessed via eye-tracking, in FXS and non-FXS groups. The present study built on this work to test whether attentional biases correspond to clinical features of anxiety among adolescents and young adults with FXS.

Methods:

Participants included 21 males with FXS ages 15–20 who completed an adapted eye tracking task that measured attentional bias toward fearful faces of varied emotional intensity.

Results:

Among participants without anxiety disorders, attentional bias toward fear increased across age, similar to non-FXS pediatric anxiety samples. In contrast, participants with anxiety disorders exhibited greater stability in fear-related attentional biases across age. Across analyses, subtle fear stimuli were more sensitive to within-group anxiety variability than full-intensity stimuli.

Conclusions:

Our results provide novel evidence that although threat-related attentional biases may correspond with anxiety outcomes in FXS, these associations are complex and vary across developmental and task factors. Future studies are needed to characterize these associations in more robust longitudinal samples, informing whether and how eye tracking tasks might be optimized to reliably predict and track anxiety in FXS.

Introduction

Anxiety disorders constitute one of the most common and impairing comorbidities associated with fragile X syndrome (FXS), with the majority of males meeting diagnostic criteria (Cordeiro et al. 2011; Ezell et al. 2018). Although elevated anxiety-related features are well-documented in FXS as early as infancy (Tonnsen et al. 2013, 2017; Tonnsen and Roberts 2016) and intensify across development (Cordeiro et al. 2011), measuring features of anxiety in FXS and other neurogenetic syndromes is inherently challenging due to common limitations in language, self-awareness, and cognitive skills required for traditional assessment tasks. Recently, eye tracking has been highlighted as a valid, non-invasive, and objective alternative for measuring anxiety-related risks in FXS (Crawford et al. 2016; Burris et al. 2017). In particular, threat-related attentional biases have been shown to correlate with anxiety symptoms in typical development (Bar-Haim et al. 2007) and present atypically in young children with FXS relative to non-FXS controls (Burris et al. 2017). Although these data suggest clinical promise, it remains unclear whether atypical threat-related attentional biases actually index anxiety features within FXS, as prior work has primarily focused on group-level comparisons between FXS and non-FXS samples. As such, the present study aimed to test associations between attentional biases and clinical anxiety symptoms in this population, a critical next step to informing whether such measures could be eventually leveraged to predict, diagnose, or monitor anxiety symptoms in FXS.

Methods

Participants

Participants included 21 males with FXS assessed between ages 15–20 years as part of a multi-site study on language development in adolescence. Procedures were approved by the Institutional Review Board. Fragile X status was confirmed via polymerase chain reaction analysis (CGG repeats >200). Six additional participants were excluded due to non-compliance (n=3) and <32 valid trials (n=3; one trial block), and are not discussed further. The sample was predominantly White (95%; 5% Black), 15% reported incomes <$30,000, and 85% of mothers reported post-high school education.

Methods and Procedures

Clinical procedures, task development, apparatus, and procedures are detailed in supplementary material and are briefly outlined here.

Clinical Measures.

Anxiety diagnoses were measured using the Children’s Interview for Psychiatric Symptoms-Parent (P-ChIPS) (Weller et al. 1999). We collapsed participants as any anxiety disorder (n=10) versus no disorder (n=11). Diagnoses were specific phobia (n=7, 33%), social phobia (n=1, 5%), separation anxiety (n=1, 5%), generalised anxiety disorder (n=4, 19%), and obsessive compulsive disorder (n=2, 10%). Five participants met criteria for one disorder, and five met for two disorders. We measured cognitive ability using the Leiter International Performance Scales-Revised Brief IQ composite growth score (Roid et al. 1997).

Apparatus and Procedure.

Our task was adapted from a previous study of attentional bias in anxiety (Mogg et al. 2007). For each trial, two face stimuli from a single actor were presented side-by-side, with one face depicting a neutral expression and the other face depicting either 25% fear (“partial-intensity”) or 100% fear (“full-intensity”). Two pairings (partial/neutral, full/neutral) were presented from 8 actors (4 male, 4 female) and counterbalanced across two locations (left/right), producing 32 configurations randomized across 192 trials.

Eye movements were recorded using an SR Research Eyelink 1000 Plus eye tracker operating in remote desktop mode. As depicted in Figure 1, each trial began with a central fixation “X” at which gaze had to be detected for 100ms before the trial would continue. Next, a face pair was presented until a fixation was detected for 100ms, or until 5000ms had elapsed. Inter-trial interval was randomly set to 750–1250ms. Methods and adaptations for encouraging on-task behavior are detailed in supplementary material and included presenting colorful non-face reward stimuli between trials and, to encourage on-task behavior, asking the participant to verbally indicate if they saw a car. For each participant, we calculated the proportion of trials in which the participant initially looked toward the fearful face (“fear bias”) for each condition (partial, full) as [n trials oriented toward fear / n trials].

Figure 1.

Task sequence

Prior to analyses and bias calculations, we excluded trials with reaction times (RTs) >1250 ms or two standard deviations above the individual participant’s mean RT (9% of trials) per prior studies (Mogg et al. 2007). An additional 17% of trials were excluded for missing data.

Data Analytic Strategy

Analyses were conducted using SAS 9.4 with an alpha criterion of .05. We verified model assumptions and centered age at the mean for regression models. We tested task integrity using split-half Spearman correlations between odd versus even trials, as well as by testing whether biases differed from chance (.5) at each intensity level using one-sample median tests. Sensitivity power analysis conducted using G*Power 3.1.9.4 (Faul et al. 2007) indicated our one-sample median tests were powered to detect medium-sized effects (d=.56).

To test our hypothesis that higher anxiety correlates with enhanced fear bias, particularly at older ages, we used linear regression models that predicted fear bias from anxiety, age, and the interaction of anxiety x age. Separate models were constructed for full and partial intensity trials. Significant anxiety x age interactions were probed using the Johnson-Neyman technique (Johnson 1950) via MODPROBE (Hayes and Matthes 2009). Sensitivity power analysis indicated our regression model with three predictors (main effects + interaction) was powered to detect large effects only (f²=.42). As such, we also conducted follow-up, simplified Spearman correlations between fear bias and age for both anxiety and non-anxiety groups.

Results

Preliminary Analyses

Table 1 includes descriptive data, and Table 2 includes models. Anxiety groups did not differ in age (Wilcoxon Z=0.18, two-tailed p=.860), administered trials (Z=.66, p=.506) or completed trials (Z=.49, p=.621). Trial data did not correlate with demographic features or task performance, and cognitive abilities were not associated with task performance or outcomes.

Table 1.

Descriptive Statistics and Spearman Correlations (σ) for Continuous Variables

	1	2	3	4	5	6
1. Age
2. Leiter Growth Score	.09
3. Full Intensity Bias	.29	.02
4. Partial Intensity Bias	.26	.04	.41b
5. n Administered Trials	.09	.26	−.02	.28
6. n Usable Trials	−.10	.37b	−.23	.23	.84a
M (SD)	17.20 (1.54)	38.33 (4.31)	56.31 (8.10)	51.57 (4.81)	107.71 (29.25)	83.52 (25.34)

Note:

^ap < .05

^bp < .10.

Table 2.

Regression Analyses for Fear Bias Predicted by Anxiety, Age, and their Interaction

	B	SE	T	p
Full Intensity Fear Bias (R2=.07)
Intercept	56.71	2.68	21.14	<.001
Anxiety	−0.79	3.71	−0.21	.833
Age	0.60	1.91	0.32	.756
Anxiety*Age	1.08	2.49	0.43	.670
Partial Intensity Fear Bias (R2=.37)
Intercept	51.69	1.31	39.53	<.001
Anxiety	−.33	1.81	−0.18	.856
Age	−1.41	0.93	−1.51	.149
Anxiety*Age	3.58	1.22	2.94	.009

We also probed potential association of autism symptoms with both clinical status and task performance. Anxiety status did not vary according to whether participants exceeded the clinical threshold on the Autism Diagnostic Observation Schedule, 2^nd Edition (ADOS-2) (Lord et al. 2012; Fisher’s two-tailed p=.149), and attentional bias did not correlate with ADOS-2 severity scores (full intensity Spearman ρ=−.05, p=.845; partial intensity ρ=.04, p=.853). Thus, we did not covary autism symptoms in primary models. These decisions are consistent with previous work documenting independence of anxiety diagnoses from autism and cognitive features in an overlapping sample (Ezell et al., 2018).

Task integrity analyses indicated consistent individual task performance across odd versus even trials (ρ=.77, p<.001). Participants also exhibited a significant bias toward the fearful face on full-intensity trials (S=84.5, p=.001) but not on partial-intensity trials (S=42.5, p=.116), replicating previous work (Mogg et al. 2007).

Associations between Fear Bias, Anxiety, and Age

Table 2 includes results of linear regression models, which are depicted in Figure 2. Anxiety was not associated with fear biases in response to full-intensity stimuli. However, the association between anxiety and fear bias on partial-intensity trials varied significantly across age, with the combined predictors accounting for 37% of variance in the model. Simple slopes indicated that participants with anxiety disorders’ fear bias marginally decreased an average of 1.41 percentage points per year [F (1,8)=3.51, p=.098, R²⁼.31], whereas participants without anxiety disorders’ fear bias increased 2.17 percentage points per year [F (1,9)=5.89, p=.038, R²⁼.40]. These patterns are consistent with post-hoc analyses showing higher fear bias at younger ages (region of significance ≤ 15.57 years) but lower fear bias at older ages (region of significance ≥ 18.64 years). These findings were maintained with number of trials covaried.

Figure 2.

Differential association between fear bias and anxiety across age, separated by partial- and full-intensity trial types.

Follow-up Spearman correlation models paralleled these results. On full-intensity trials, neither group exhibited significant associations between fear biases and age (non-anxiety ρ=.35, p=.285; anxiety ρ=.28 p=.434). However on partial-intensity trials, the anxiety group exhibited marginal decreases in fear biases with age (ρ=.−.59, p=.073), and the non-anxiety group exhibited increased fear bias with age (ρ=.65, p=.030).

Discussion

The present study offers novel evidence that threat-related attentional biases may relate to anxiety disorders in adolescents with FXS, however, patterns are complex and vary across developmental and task-related factors. These findings suggest that although eye tracking may be a useful tool for measuring anxiety-related features in FXS in the future, substantial work is needed to distinguish anxiety-specific correlates from the broader FXS phenotype, probe effects of task properties on this association, and establish psychometric rigor of this approach.

Similar to prior studies (Burris et al. 2017), individuals with FXS generally exhibited a bias toward threat-related stimuli. However, the association between attentional biases and clinical symptoms was complex and varied across age and task factors. For example, biases became more blunted over time in the subgroup of individuals with anxiety. Although this blunting seemingly conflicts with the general notion that threat-related biases are enhanced in children with anxiety (Dudeney et al. 2015), this finding is consistent with emerging evidence that in children, the direction of threat-related biases varies according to anxiety subtype: whereas children with “distress” disorders, such as generalised anxiety disorder (GAD), exhibit attentional biases toward threat, children with “fear” disorders (e.g., social phobia, specific phobia, and separation anxiety disorder) exhibit bias away from threat (Waters et al. 2014). Similarly, children who display dysregulated fear – characterised by persistently high fear despite low threat – later exhibit biases away from threat (Morales et al. 1998). In our sample, nine participants met criteria for fear disorders, whereas only 2 exclusively met criteria for distress disorders (4 total met criteria for GAD), supporting the notion that blunted biases may reflect the emergence of fear-related disorders in this sample. Another possible explanation is that our sample was not yet at the age when fear disorders would be fully expressed, as initial evidence suggests some fear disorders such as social phobia may be more common in adults than children with FXS (Cordeiro et al. 2011), and only one participant in our adolescent sample (ages 15–20 years) was diagnosed with social phobia relative to over 30% of participants in Cordeiro et al. (ages 5–33; 61% with adapted diagnostic criteria).

We also found that associations between threat-related attentional biases and anxiety were exclusively present on partial-intensity trials. This finding was particularly striking given individuals’ overall fear biases only significantly differed from chance across the sample on full-intensity trials. These distinct patterns may indicate that more subtle stimuli provide a context for anxiety-related signals to emerge, consistent with previous studies in which adults with anxiety disorders respond to sub-threshold threat cues with enhanced attentional biases toward (Mogg 2002) and differentiate emotions at lower expressional intensities than controls (Bui et al. 2017). Future work is needed to clarify the developmental and task-related factors that affect the translation of research related to attentional biases to clinical outcomes.

The complexity of our findings highlight the challenges of translating biological and behavioral tasks to patient care: despite the promise of biomarkers – broadly defined – in improving assessment procedures in special populations, actual translation of these approaches will be a complex process and currently remanis premature in many cases (Sahid et al. 2018). For example, additional work is needed to map expected within-subject changes over time, ensure that biomarkers are generalizable and reproducable, and develop standareds for rigorous quality control, particularly in the context of clinical trials (Sahid et al.). Specific to FXS, follow-up studies are also needed to expand our understanding of the presentation, development, genetic substrates of anxiety-related biases in FXS, and to extend this work to females and other age groups. Together, these studies may inform whether threat related attentional biases could offer malleable indices of anxiety in FXS and other neurogenetic groups, similar to studies in non-FXS groups (Hakamata et al. 2010), informing next steps for noninvasively monitoring behavioural or pharmacological treatment response in FXS and other neurodevelopmental populations.

Notably, this study was limited by a small sample, as is common in rare disorder research. In addition to compromising power, small samples can be vulnerable to unstable or inflated statistical estimates, particularly in regression models (Lombardo et al., 2019). As such, our findings should be interpreted with caution and replicated, per best practice.