Content uploaded by Rebecca Price
Author content
All content in this area was uploaded by Rebecca Price on Oct 30, 2019
Content may be subject to copyright.
Connectivity Subtypes Predict Attentional Profiles in Anxiety
1
Word count: 8390
Tables: 1
Figures: 2
Supplement: 1
RUNNING HEAD: Connectivity Subtypes Predict Attentional Profiles in Anxiety
Neural Connectivity Subtypes Predict Discrete Attentional Bias Profiles Among
Heterogeneous Anxiety Patients
Rebecca B. Price, PhD1; Adriene M. Beltz, PhD2; Mary L. Woody, PhD1; Logan
Cummings, BS3; Danielle Gilchrist, BA1; Greg J. Siegle, PhD1
1Department of Psychiatry, University of Pittsburgh School of Medicine
2Department of Psychology, University of Michigan
3Department of Psychology, Florida International University
**Accepted manuscript: Clinical Psychological Science**
Author Note: Supported by a Career Development Award from NIMH (1K23MH100259; PI:
Price). The authors have no conflicts of interest.
Correspondence concerning this article should be addressed to Rebecca Price, Western
Psychiatric Institute and Clinic, 3811 O’Hara St., Pittsburgh, PA 15213, Phone: 412-648-6445,
Fax: 412-648-6451, email: rebecca.price@stanfordalumni.org. !
Connectivity Subtypes Predict Attentional Profiles in Anxiety
2
Abstract
On average, anxious patients show altered attention to threat—including early vigilance towards
threat and later avoidance of threat—accompanied by altered functional connectivity across brain
regions. However, substantial heterogeneity within clinical, neural, and attentional features of
anxiety is overlooked in typical group-level comparisons. We used a well-validated method for
data-driven parsing of neural connectivity to reveal connectivity-based subgroups among 60
adults with transdiagnostic anxiety. Subgroups were externally compared on attentional patterns
derived from independent behavioral measures. Two subgroups emerged. Subgroup A (68% of
patients) showed stronger executive network influences on sensory processing regions and a
paradigmatic “vigilance-avoidance” pattern on external behavioral measures. Subgroup B was
defined by a larger number of limbic influences on sensory regions and exhibited a more atypical
and inconsistent attentional profile. Neural connectivity-based categorization revealed an
atypical, limbic-driven pattern of connectivity in a subset of anxious patients that generalized to
atypical patterns of selective attention.
Keywords: fMRI, individual-level functional connectivity, community detection, attentional bias,
anxiety
!
Connectivity Subtypes Predict Attentional Profiles in Anxiety
3
Neural Connectivity Subtypes Predict Discrete Attentional Bias Profiles Among Heterogeneous
Anxiety Patients
Biological heterogeneity is common within psychological disorders. Anxiety disorders,
the most common form of psychiatric disorder (Kessler, Chiu, Demler, Merikangas, & Walters,
2005), are replete with heterogeneity at the level of symptom presentation (Grisanzio et al.,
2018), but little is known within this broad class of diagnoses regarding possible biobehavioral
subtypes, and treatment indications that might track with such subtypes. Though genetic and
neurocognitive studies suggest a broad taxonomy distinguishing anxiety disorders characterized
predominantly by chronic distress from those characterized by acute fear (Kendler, Prescott,
Myers, & Neale, 2003; Lahey, Van Hulle, Singh, Waldman, & Rathouz, 2011; McTeague &
Lang, 2012), such broad distinctions at the level of clusters of symptoms or symptom-based
diagnoses remain inherently heterogeneous with regard to specific biobehavioral mechanisms
that are likely to vary within the scope of a single diagnosis—and which may represent important
treatment targets.
In a complementary approach, biological heterogeneity across individual patients has
been parsed with data-driven methods to identify subgroups of patients within broad disorder
domains (Beltz, Moser, Zhu, Burt, & Klump, 2018; Clementz et al., 2016; Karalunas et al., 2014;
Yang et al., 2014), including affective disorders such as depression (Drysdale et al., 2017; Price,
Gates, Kraynak, Thase, & Siegle, 2017; Price, Lane, et al., 2017). Efforts to date suggest that
such biologically-based subtyping can have external relevance to clinically relevant features
across multiple levels of analysis, including gender, diagnosis (depressed vs. healthy; comorbid
anxiety), symptom severity, history of depression recurrence, and behavioral performance on
Connectivity Subtypes Predict Attentional Profiles in Anxiety
4
information processing tasks (Price, Gates, et al., 2017; Price, Lane, et al., 2017); and could
possibly inform clinical decision-making (Drysdale et al., 2017).
One well-established, clinically relevant, transdiagnostic dimension of anxiety disorders
involves altered attentional deployment towards threat (hereafter, “attentional bias”), which has
been described at both early and later stages of threat processing. Specifically, anxious
individuals may show a pattern of excessive vigilance towards threat during initial stages of
threat processing (Bar-Haim, Lamy, Pergamin, Bakermans-Kranenburg, & van IJzendoorn,
2007), but during later, more strategic stages of threat processing, may switch to an avoidant
pattern of attention (Mogg, Bradley, Miles, & Dixon, 2004)—consistent with the marked degree
of behavioral avoidance that is a core clinical feature of these conditions.
Both arms of this “vigilance-avoidance” pattern have been linked previously to altered
functional connectivity between regulatory regions of the prefrontal cortex (PFC) and
affective/limbic regions involved in stimulus-driven responses to threat-related stimuli (e.g.,
amygdala, hippocampus) (Bishop, 2007; Price, Allen, et al., 2016; Price, Eldreth, & Mohlman,
2011; Price et al., 2014; White et al., 2017). However, group-level observations of these
attentional phenomena in anxious samples mask considerable within-group heterogeneity, which
may contribute to notably mixed findings within the literature (Kruijt, Parsons, & Fox, 2018;
Mogg, Waters, & Bradley, 2017; Rosen, Price, & Silk, Submitted). Further contributing to
heterogeneity, attention is itself a multifaceted phenomenon, with divergent subcomponents
present not only as a function of time period (as predicted by vigilance-avoidance models), but
also with respect to overt vs. covert components (i.e., observable “overt” eye movements vs.
“covert” shifts in the ‘spotlight’ of attention within a stable visual field; (Posner, Snyder, &
Davidson, 1980; Weierich, Treat, & Hollingworth, 2008)), and engagement vs. disengagement
Connectivity Subtypes Predict Attentional Profiles in Anxiety
5
processes (i.e., selective initial capture of attention by threat vs. selective difficulties disengaging
from threat; (Grafton & MacLeod, 2014))—each with potentially unique neural substrates
(Petersen & Posner, 2012).
Individual differences in attentional patterns within anxious samples have been linked to
longitudinal and treatment outcomes (Amir, Taylor, & Donohue, 2011; Legerstee et al., 2009;
Price, Tone, & Anderson, 2011; Price, Rosen, et al., 2015; Price, Wallace, et al., 2016; Price,
Woody, Panny, & Siegle, In press), suggesting that attending to heterogeneity may help to
capture critical, clinically relevant information. Multiple treatment modalities (e.g.,
psychotherapy, pharmacology) produce acute reductions in attentional bias that may precede
mood effects (Browning, Holmes, & Harmer, 2010), suggesting that attentional bias
modification could be a final common pathway to symptom reduction. Recently, methods have
been developed to alter attentional bias directly through computer-based tasks that consistently
train initial attention towards neutral or positive cues. After establishing causal effects of the
experimental manipulation of attentional bias on emotional reactivity in healthy samples
(MacLeod, Rutherford, Campbell, Ebsworthy, & Holker, 2002), this approach was extended to
clinical populations (Heeren, Mogoase, Philippot, & McNally, 2015; Linetzky, Pergamin-Hight,
Pine, & Bar-Haim, 2015; Price, Wallace, et al., 2016). However, given widely-noted mixed
findings within this emerging mechanistic treatment literature (McNally, 2018), a better
understanding of the underlying neural contributors to heterogeneous attentional patterns within
anxious patients may offer key insights that are relevant to the refinement of mechanistic
treatments targeting attention and/or patient-treatment matching algorithms.
The neural substrates of both anxiety and attentional bias have typically been studied
through group comparisons (e.g., anxious patients vs. controls), but group-level summaries (e.g.,
Connectivity Subtypes Predict Attentional Profiles in Anxiety
6
brain maps) may not accurately represent even a single individual within the group (Beltz,
Wright, Sprague, & Molenaar, 2016; Gates & Molenaar, 2012; Miller et al., 2002; Molenaar &
Campbell, 2009). A more novel approach is to focus explicitly on heterogeneity (e.g.,
heterogeneity expressed within neural network connections) by analyzing data at the individual
participant level, searching for detectable biologically-derived subgroups, and then
characterizing these subgroups with respect to relevant observable characteristics and behaviors
(Drysdale et al., 2017; Price, Gates, et al., 2017; Price, Lane, et al., 2017).
In the present study, we aimed to robustly characterize the functional connectivity
patterns expressed by each individual within a transdiagnostic clinically anxious sample during
the presentation of threatening and neutral images. As in our previous studies, we applied a
connectivity method shown to reliably recover, for each individual, both the presence and the
direction of connectivity among regions [i.e., does A predict B after controlling all other
network-wide influences (including B’s influence on itself)? (Friston, 1994)]. Whereas concerns
have been raised about the ability of many connectivity methods to reliably recover brain
connections for individuals (Smith et al., 2011), validation tests suggest our selected approach,
Group Iterative Multiple Model Estimation with subgrouping [S-GIMME; (Gates, Lane,
Varangas, Giovanello, & Guiskewicz, 2017; Gates & Molenaar, 2012)], reliably recovers both
the presence and direction of paths within heterogeneous individuals when the number of
observations per person exceeds 120, as is the case in most neuroimaging data, even in relatively
small subsets of individuals (Gates & Molenaar, 2012; Lane, Gates, Pike, Beltz, & Wright, 2019;
Mumford & Ramsey, 2014; Nichols, Gates, Molenaar, & Wilson, 2014).
This approach thus allowed for neural network maps, across a network of regions
robustly modulated by the task, to be reliably constructed at the individual level, and with greater
Connectivity Subtypes Predict Attentional Profiles in Anxiety
7
specificity than is possible in non-directed (e.g., correlational) approaches, while incorporating
data-driven, subgroup categorization within these functional connectivity maps. In addition to
being able to recover models in best-case scenarios, the GIMME algorithm robustly recovers
individual-level directed paths in a number of conditions that emulate those seen in fMRI
studies, such as nonstationarity and additional noise in the region of interests selected (Gates &
Molenaar, 2012). Of particular relevance for subgrouping, GIMME has been strongly validated
at the individual subject level, robustly recovering directed influences in heterogeneous
individuals according to simulations (Gates & Molenaar, 2012).
With regard to subsequent subgroup detection based on connectivity maps, the S-
GIMME approach has been validated in both simulated (Gates et al., 2017; Gates & Molenaar,
2012; Gates, Molenaar, Iyer, Nigg, & Fair, 2014; Lane et al., 2019) and empirical (Price, Gates,
et al., 2017; Price, Lane, et al., 2017) datasets. S-GIMME has been shown to reliably recover the
underlying subgroups across a range of conditions, such as varying number of subgroups,
various size and proportionality of subgroups, and differing sample sizes (Gates et al., 2017;
Lane et al., 2019), including much smaller samples (e.g., N=25 in the total sample, prior to
subgrouping) than the current sample (N=60). The stability and robustness of the subgroup
solution produced by S-GIMME has been established using simulated and empirical data (Gates
et al., 2014). Subgroups remain stable after randomly perturbing the similarity matrix of
connectivity weights to generate random fluctuations in individual data points (while holding
sample size/power constant). Additionally, an extensive Monte Carlo simulation study was
conducted to identify which clustering algorithms return the correct subgroups in the context of
GIMME-derived features (Gates et al., 2017). Walktrap (Pons & Latapy, 2006), the clustering
approach used in S-GIMME, outperformed all the others. Furthermore, Walktrap is an
Connectivity Subtypes Predict Attentional Profiles in Anxiety
8
unsupervised classification approach which does not rely on an a priori number of subgroups
specified by the researcher. Instead, it produces an optimal number of subgroups based solely on
shared patterns of connectivity across individuals, and, if there are no subgroups, Walktrap is
among the few methods that will return only one subgroup. This indicates to the researcher that,
if subgroups are obtained, they explain more variance than considering the sample as one group.
In summary, S-GIMME provides a robust and exceedingly well-validated data-driven approach
to parsing heterogeneity within the functional connectivity of a sample, even in the absence of
“big data” cohorts.
Given previous findings suggesting functional connectivity patterns across PFC and
limbic/affective are key substrates of attentional bias, we hypothesized that subgroups, derived
by S-GIMME based solely on functional connectivity maps, might differ on externally measured
behavioral indices of attention to threat. Attentional bias variables collected outside the scanner
(eyetracking, reaction times) were thus used to compare connectivity-based subgroups across
multiple subcomponents of attention, including varying time periods (initial vs. later stages of
threat processing) and discrete subcomponents of attention (overt vs. covert attention,
engagement vs. disengagement patterns), allowing us to further characterize the subgroups and
assess their external relevance to a putative treatment target in anxiety. Resulting connectivity-
based subgroup characteristics could ultimately suggest novel mechanistic targets for treatment
by revealing discrete attentional profiles that would be overlooked when averaging across
heterogeneous anxious individuals.
Methods
Participants were 60 individuals with clinically impairing anxiety recruited for a larger
treatment study [see (Price et al., 2018); Table and Supplement].
Connectivity Subtypes Predict Attentional Profiles in Anxiety
9
fMRI task. As described in detail previously (Price et al., 2018) and adapted from
Somerville and colleagues (Somerville et al., 2013), participants were shown a total of eight 78s
blocks that were either “Negative,” containing a series of 10 negative/threatening images,
presented for 3s each, from the International Affective Picture Set (IAPS; (Lang, Bradley, &
Cuthbert, 2008)), or “Neutral,” consisting of 10 neutral IAPS images. Images were jittered with a
pseudorandom number of numerals relating to either a numerical countdown (e.g., 3-2-1)
(“Predictable” blocks) or a random string of numbers (“Unpredictable” blocks). Participants
were alerted by a 2s text cue, presented at the start of each block, as to the nature of the
upcoming block (e.g., “Predictable Negative”). To encourage continued attention to the images,
participants completed an incidental task by responding via button press to indicate whether each
image depicted an indoor or an outdoor scene.
fMRI acquisition and preprocessing. T2*-weighted images depicting BOLD contrast
(TR=2000;TE=28;flip angle=73°;slices=38;FOV=200x200;3.125x3.125x3.2mm voxels) were
acquired on a 3T Siemens Trio. Standard preprocessing steps were applied using Analysis of
Functional Neuroimaging (AFNI; see Supplement).
11 functionally defined ROIs were selected because they showed robust group-level
activations or deactivations to negative/threatening images during the task (map-wise p<.05; see
details in Supplement), ensuring that connectivity was quantified within a network relevant to the
processing of visual threat cues. To improve interpretability of findings, these 11 regions were
then classified based on their known functions (e.g.,Kaiser, Andrews-Hanna, Wager, & Pizzagalli,
2015; Laird et al., 2011) as belonging to networks predominantly relevant to affective processing
(AN), sensory processing, and executive control (ExN). See Supplement and Figure 1A for
details of ROI definitions. Mean, preprocessed timeseries data were extracted per-participant for
Connectivity Subtypes Predict Attentional Profiles in Anxiety
10
each ROI. Of 69 participants who completed the task, 9 (13%) were excluded from analysis due
to excessive motion during the task (>30% of timepoints showed framewise motion >0.2mm or
>0.2°). To further protect against spurious connectivity patterns related to motion, we removed
individual timepoints with framewise motion >0.2mm or >0.2° from analysis [which are then
accounted for during full information maximum likelihood estimation during GIMME; see (Beltz
& Gates, 2017)] and verified that participants’ motion parameters were unrelated to any finding
(Supplement).
Directed connectivity and subgrouping. Directed paths (i.e., establishing which ROIs
statistically predict others) were derived for each individual in a data-driven fashion using S-
GIMME (Gates et al., 2017; Lane, Gates, & Molenaar, 2015), while controlling for
contemporaneous influences of the task on each ROI (i.e., threatening and neutral word blocks
convolved with a smoothed finite impulse response)(see (Gates, Molenaar, Hillary, &
Slobounov, 2011). Paths could be contemporaneous (marking prediction at the same TR) or
lagged (marking prediction from one TR to the next), or present for the full sample, only a
subgroup of the sample, or just for an individual. Large-scale simulations show that S-GIMME is
a valid and reliable connectivity mapping approach particularly when timeseries are long (as is
the case for fMRI data) and lagged autoregressive effects are modeled (as was done here) (Gates
et al., 2017; Lane et al., 2019)].
S-GIMME implements unified structural equation models (Kim, Zhu, Chang, Bentler, &
Ernst, 2007) and utilizes a Bayes net formulation. It first detects (only if they exist) lagged or
contemporaneous directed connections for the majority (defined here and in most simulations as
75%) of the sample. Next, it detects subgroups by using the individual-level estimates of these
group-level connections as well as anticipated estimates for candidate connections and
Connectivity Subtypes Predict Attentional Profiles in Anxiety
11
employing an ‘unsupervised’ community detection algorithm (i.e., Walktrap). These ‘shared’
patterns are defined based on their sign (positive/negative), significance (p<.05 vs. p>.05, after
Bonferroni correction for the number of subjects), direction of influence (e.g, region A->region
B), and temporal pattern (contemporaneous or lagged). If subgroups are identified, S-GIMME
then detects lagged or contemporaneous directed connections for the majority (defined here and
in most simulations as 50%) of a subgroup. Finally, S-GIMME iteratively detects (if they exist)
individual-level connections. The identification of all connections is based on LaGrange
Multiplier Equivalents, which indicate which connections (if added to a map) will maximally
increase the map’s explanatory power (i.e., model fit).
Characterization of subgroup connectivity. For each participant, S-GIMME generated
a connectivity map with group-level, subgroup-specific, and individual-level connections. To
understand the nature of the disparate connectivity patterns found across subgroups, subgroups
were characterized by the unique subgroup-level connections that were identified by the
algorithm, and by comparing the strength of each group-level path (excluding auto-regressive
paths) via independent t-tests (across subgroups) comparing individuals’ path beta weights, with
False Discovery Rate (FDR) correction.
External variables. Connectivity-based subgroups were compared across several
attentional bias indices described in detail previously [(Price, Brown, & Siegle, 2019; Price et al.,
In press); see details, missing data explanations, and resulting subgroup sample sizes for each
analysis in Supplement]. Briefly, a standard dot-probe task, consisting of threat-neutral word
pairs presented for short (500ms) and longer (1500ms) durations, was completed with concurrent
eyetracking. This task provided four indices for analysis, all of which demonstrated adequate-to-
strong psychometric reliability within this sample, as discussed in detail previously (Price et al.,
Connectivity Subtypes Predict Attentional Profiles in Anxiety
12
2019; Price et al., In press), which is critical given that attentional bias indices can be prone to
poor reliability (Price, Kuckertz, et al., 2015; Rodebaugh et al., 2016). Two were reaction time
indices that were derived following computational modeling [Drift Diffusion Modeling; (Ratcliff
& McKoon, 2008)] of trial-by-trial data from each individual dataset in order to separate the
attentional parameters of interest from incidental decision processes. Our previous publication in
the current sample (Price et al., 2019) illustrated marked improvement in split-half and test-retest
reliability for these attentional bias indices relative to conventional analysis methods for the dot-
probe; thus, we exclusively analyzed the DDM-derived dot-probe indices, which were quantified
separately for each of the two stimulus durations (short: 500ms and long: 1500ms), enabling us
to detect time-sensitive patterns consistent with the “vigilance-avoidance” hypothesis. Two
concurrently collected eyetracking indices reflected overt gaze patterns: initial fixation to threat
(as a % of all trials); and disengagement delay bias (see Supplement for further details). Both
overt/eyetracking indices demonstrated moderate split-half reliability in the present sample
[³.52; (Price et al., In press)]. Finally, a separate, widely used reaction time index of covert
attentional bias, the spatial cueing task (Bar-Haim, Morag, & Glickman, 2011; Price et al., In
press), was used to separately quantify biases in covert engagement and covert disengagement
from threat-related faces (see Supplement). The split-half reliability of the spatial cueing task
indices was strong in the present sample [³.86;(Price et al., In press)].
As a comparison, attentional bias variables were also compared across symptom-based
subgroups within the anxious participants (+/- generalized anxiety disorder, +/- comorbid
depression, primary distress- vs. fear-related disorder).
Results
Connectivity Subtypes Predict Attentional Profiles in Anxiety
13
Connectivity maps. Group-level: At the group level, connectivity paths depicted in
Figure 1A were present, in addition to lagged autoregressions at every ROI. As expected, ROIs
behaved as a strongly interconnected network. All contemporaneous network influences were in
the positive direction, while two lagged influences were negative, and redundant with positive
contemporaneous paths, suggesting possible negative feedback mechanisms. No group-level
paths were found for task vectors (i.e., negative and neutral blocks) predicting ROI timeseries in
the context of the observed ROI-to-ROI network influences. Subgroups. Based on unsupervised
search for the optimal number of subgroups, two subgroups emerged (see Supplement for
subgroup quality analyses). Subgroup A contained 68% (n=41) of participants; hence, Subgroup
B (32% of participants; n=19) was considered to exhibit ‘atypical’ connectivity patterns relative
to the majority of anxious patients. Subgroup was unrelated to motion and other data quality
measures (Supplement). In t-tests, Subgroup A exhibited stronger connectivity than Subgroup B
in two specific group-level paths (Figure 1A; FDR p<.05): a ExN->sensory path (DMPFC-
>thalamus) and an ipsilateral occipital lobe (sensory) path. Similarly, the paths unique to each
subgroup (Figures 1B-C) were also consistent with additional unique influences connecting ExN
and sensory regions in Subgroup A. By contrast, there were multiple unique directed influences
of AN regions on sensory regions in Subgroup B, including three AN paths converging on the
thalamus.
External variables. Dot-probe reaction times. In a repeated-measures ANOVA with
subgroup as a between-subjects factor and stimulus duration (short vs. long) as a within-subjects
factor, a group*duration interaction effect on attentional bias scores was found (F1,58=7.75,
p=.007; Figure 2). Specifically, subgroup A showed a “typical,” theoretically hypothesized
vigilant-avoidant pattern of attention, as reflected in vigilance towards threat during short trials
Connectivity Subtypes Predict Attentional Profiles in Anxiety
14
followed by avoidance of threat during long trials (t40=2.39; p=.02). Subgroup B displayed an
inverse pattern reflecting greater sustained vigilance to threat during long trials. Dot-probe
eyetracking. Subgroup B had a higher percentage of trials with initial fixation to the threat cue
than Subgroup A, suggesting greater overt engagement with threat (t55=2.32; p=.02). The groups
did not differ on overt disengagement bias (t53=1.27; p=.21). Spatial cueing task reaction times.
Patients in Subgroup A had a more vigilant pattern of engagement with threat relative to
Subgroup B (t53=2.58; p=.01). The groups did not differ with respect to disengagement from
threat (t53=0.54; p=.59).
In aggregate, analyses suggested connectivity-based subgroups had external behavioral
relevance across multiple subdomains of attention, particularly for measures reflecting
engagement with (rather than disengagement from) threat, and pointed to a more theoretically
paradigmatic pattern of attention in Subgroup A relative to Subgroup B.
Symptom-category-based subgroups. Attentional indices did not show significant
differences across anxious participants with (n=50) vs. without (n=10) a generalized anxiety
disorder diagnosis (the most prevalent diagnosis in the current sample), with (n=18) vs. without
(n=42) comorbid depression diagnosis, or those with a primary distress-related (n=51) vs. a
primary fear-related diagnosis (n=9), with only one specific exception: individuals with a
comorbid depression diagnosis displayed a more vigilant pattern towards threat during long
duration dot-probe trials relative to individuals without a comorbid depression diagnosis
(t58=2.68; p=.01).
Connectivity subgroups and clinical measures. Connectivity subgroups were unrelated
to diagnosis-based subgroups, as defined above (
𝜒#
’s<.76, p’s>.39). Furthermore, connectivity
subgroups also did not predict clinical symptom severity on any of the Mood and Anxiety
Connectivity Subtypes Predict Attentional Profiles in Anxiety
15
Symptoms Questionnaire [MASQ; (Watson et al., 1995)] subscales: anxious arousal, anhedonic
depression, and general distress (p’s>.41). Thus, clinical phenotypes were not strongly linked to
connectivity-based subtypes in the current sample.
Discussion
In the present study, a robust method was applied to characterize heterogeneous
individuals’ directed connectivity paths during threat and neutral image viewing (Gates et al.,
2017; Gates & Molenaar, 2012), revealing two brain-based subgroups of anxious individuals
(Figure 1). The majority of anxious patients in our sample (68%; Subgroup A) showed relatively
strong executive network influences on sensory processing regions (thalamus, occipital cortex),
relative to the smaller, “atypical” subgroup (Subgroup B). In contrast, Subgroup B was defined
by a larger number of directed influences from regions of the affective network onto sensory
regions—including three AN-driven paths converging on the thalamus, a sensory gating hub
(McCormick & Bal, 1994). Given the use of a robust and well-validated method (S-GIMME) for
subgroup identification, which is an ‘unsupervised’ approach that has been shown in simulated
data to arrive at accurate, complete, and stable subgroup solutions in total sample sizes as small
as N=25, the current study provides initial evidence that these two subtypes of neural network
patterns may well typify the larger population of anxious patients who share similar
characteristics (e.g., clinical, demographic, neurobiological) with the individuals in our sample.
The two unique neural network patterns exhibited by subgroups of anxious patients also
predicted unique attentional profiles at the behavioral level (Figure 2). Specifically, Subgroup A,
the more “executive-driven” subgroup, displayed a paradigmatic “vigilance-avoidance” pattern
on reaction time measures, which is consistent with influential theories of attentional processing
in anxiety (Mogg et al., 2004). Subgroup B exhibited an atypical and inconsistent pattern across
Connectivity Subtypes Predict Attentional Profiles in Anxiety
16
multiple measures of attentional bias, consistent with the multifaceted, neurobiologically
complex nature of attention (Peterson & Posner, 2012; Weierich et al., 2008), including greater
overt initial attention to threat (according to an eyetracking measure), accompanied by both
later/sustained vigilance to threat and decreased covert engagement with threat on a distinct
measure (Figure 2). Traditional diagnosis-based subgroups largely failed to predict distinct
attentional profiles in the current sample, and connectivity-based subgroups were also unrelated
to clinical variables (diagnostic subgroups, symptom scales), suggesting the connectivity-based
subgroups provided unique information that may have had stronger behavioral impact.
The field of attentional bias research has been fraught with mixed and inconsistent
findings, which may stem from a variety of sources including psychometric limitations of
widely-used behavioral indices (Price, Kuckertz, et al., 2015); inadequate dissection of the
relevant subcomponents of attention (Grafton & MacLeod, 2014; Price et al., In press); and
dominant theoretical and intervention models which presume anxious patients will exhibit fairly
uniform patterns of attentional bias relative to healthy individuals. Notably, in the present
analyses we utilized only attentional bias measures which we have previously demonstrated to
show adequate-to-strong reliability levels within this sample (Price et al., 2019; Price et al., In
press); thus, the divergent attentional profiles we uncovered are less likely to be heavily polluted
by measurement error, and more likely to reflect true heterogeneity that would be overlooked in
a conventional analysis averaging across anxious individuals.
The present findings suggest there is meaningful and consequential neural heterogeneity
expressed within this transdiagnostic sample with predominantly distress-related disorders (e.g.,
Generalized Anxiety Disorder; Table), and that these brain-based subgroups are predictive of
divergent attentional profiles that are complex and would be difficult to detect without first
Connectivity Subtypes Predict Attentional Profiles in Anxiety
17
parsing the sample on the basis of the neural patterns (or another impactful, independent
variable). In particular, divergent patterns exhibited by the “atypical” subgroup would dilute, and
thereby mask, the more paradigmatic patterns of attentional bias (e.g., vigilance-avoidance;
increased covert threat engagement) that were indeed displayed by the majority subset.
Conversely, while with increasing sample sizes the dominant attentional profile exhibited by the
majority would likely emerge as a detectable pattern at the full-group level (as in meta-analyses,
e.g. (Bar-Haim et al., 2007), this approach would mask the significant minority of anxious
patients who deviate from it systematically.
Our findings suggest multiple novel conclusions regarding the neural substrates of
anxiety and attentional bias that highlight future avenues for replication and extension. First, the
vigilance-avoidance model posits that bottom-up attentional capture by threat-related cues is
insurmountable in the early, relatively automatic stages of processing, but is then strategically
overcome and reversed (in favor of avoidance of threat) once stimuli are presented for a
sufficiently long duration (Mogg et al., 2004). The connectivity subgroup that was associated
with this pattern of attention in the current sample displayed additional directed influences of
executive control regions (DMPFC, parietal cortex) over sensory processing regions (thalamus,
occipital cortex; Figure 1B). Thus, the subgroup’s behavioral patterns may reflect that, when
anxious patients have enhanced prefrontal control of sensory processing at their disposal, they
may apply it in the service of threat avoidance at later stages of stimulus processing. Future
studies are needed to determine if this link between neural connectivity and attentional threat
avoidance is related to subsequent maintenance of anxiety in the long-term (e.g., Price, Allen, et
al., 2016).
Connectivity Subtypes Predict Attentional Profiles in Anxiety
18
Subgroup B, by contrast, exhibited numerous unique directed paths from AN regions to
sensory regions, with a particular preponderance of AN-driven paths converging on the
thalamus—a region with a central role in gating the sensory information that will be passed on
for further, elaborative processing (McCormick & Bal, 1994). Interestingly, this subgroup also
exhibited increased initial overt eye fixations to threat cues, relative to Subgroup A. Thus, AN-
driven influences on sensory processing regions—which included AN-driven paths to both
thalamus and occipital cortex—might be a corollary of increased overt eye saccades to threat
during early stages of attention. According to reaction time measures, this subset of individuals
also became increasingly attentive to threat cues over time (i.e., after longer presentation
intervals on the dot-probe task), suggesting that this subgroup’s unique AN influences on the
thalamus might reflect affectively driven ‘tagging’ of threat cues to be retained as foci of
attention during later, more elaborative stages of processing. Of note, no significant differences
across subgroups were observed on either of two [overt (eyetracking) and covert (spatial cueing)]
threat disengagement indices, suggesting difficulties with disengagement from threat were
relatively homogeneous within the current sample—as were the numerous directed influences
apparent across brain regions in group-level (i.e., homogenous) paths (Figure 1A).
Overall, findings suggest that, while numerous influences across brain regions were
robustly characteristic of the full sample of anxious patients (as reflected in group-level paths),
unique patterns of neural connectivity, which distinguished one anxious subgroup from the other,
were impactful in predicting attentional profiles measured on a distinct, external battery of tasks.
This novel, data-driven taxonomy of anxious neural subtypes, which is based on proximal neural
processes that support affective visual information processing, may have a tighter coupling with
behavioral information processing patterns relative to symptom-based categories (e.g.,
Connectivity Subtypes Predict Attentional Profiles in Anxiety
19
diagnoses—which were unrelated to both connectivity subgroups and nearly all attentional
indices), and might extend to other, unassessed domains of information processing with clinical
relevance. With regard to diagnostic subgroups, patients with comorbid depressive diagnoses
exhibited greater attentional bias towards threat during long-duration dot-probe trials—which is
consistent with an existing literature suggesting attentional biases characterize depressed patients
predominantly at later stages of processing (de Raedt & Koster, 2010; Gotlib & Joormann,
2010). However, no other link was found between diagnostic subgroups and any attentional
pattern examined. While this may lend preliminary support to the unique utility of connectivity-
based subtyping, larger and more heterogeneous transdiagnostic samples are necessary to
adequately test this hypothesis—particularly given previous reports of attentional patterns that
were moderated by specific anxiety diagnoses (Salum et al., 2012).
Findings have possible clinical implications for mechanistic treatments targeting
attentional patterns. By highlighting the divergent attentional profiles of subgroups, the study
suggests no one-size-fits-all approach to attentional remediation is likely to have a robust impact
on symptoms across all anxious patients. However, with further replication and extension of the
observed subgroups in larger validation samples, the present approach could eventually prove
useful in matching specific subgroups of patients to specific attention retraining batteries, e.g., to
automated interventions that explicitly target executive deficits vs. perceptually-driven
aberrations (Best, Milanovic, Iftene, & Bowie, 2019). With an eye towards clinical translation,
connectivity-based subtypes might also help to define a set of complex attentional profiles
(exhibited across a battery of simple behavioral attentional measures) that would best reflect
underlying neural subtypes, but may nevertheless be measurable without the need for cost-
prohibitive fMRI assessment. Refinements to attention modification procedures might be
Connectivity Subtypes Predict Attentional Profiles in Anxiety
20
informed by such future work. For instance, modification procedures might be made more
clinically impactful if they are designed to target such complex profiles by simultaneously
retraining multiple attentional subfeatures (e.g., vigilance and avoidance; overt and covert
attention). Furthermore, a nuanced understanding of connectivity-based subtypes could ideally
prove useful in designing synergistic treatment combinations that can leverage both neural (e.g.,
neuromodulation; neurofeedback) and behavioral (e.g., attention retraining) components of
information processing simultaneously to optimize behavioral and clinical effects (Wilkinson,
Holtzheimer, Gao, Kirwin, & Price, 2019).
The present sample did not include healthy comparison participants, and, though
participants were recruited transdiagnostically, the sample included a preponderance of distress-
related disorder (e.g., Generalized Anxiety Disorder) patients. Future work should explore the
degree to which connectivity-based subtypes overlap across healthy and disordered states, and
across a wider range of affective diagnoses. In our previous work applying S-GIMME to a
sample that included both depressed patients and healthy controls, patient status (healthy vs.
depressed) tracked statistically with data-driven connectivity-based subgroups, but did not
overlap perfectly with it (Price, Lane, et al., 2017). In particular, the connectivity-based subgroup
that contained the majority of depressed patients also contained 50% of rigorously screened
healthy controls. If a similar pattern were to emerge for anxious patients and controls when
adopting a focus on threat processing, such findings would suggest interesting novel avenues for
understanding potential latent risk factors and the nature of resilience or compensatory factors
among healthy controls who overlap neurobiologically with anxious patients. Alternatively, if
anxious and healthy subgroups were more cleanly divergent from one another when using data-
driven connectivity-based grouping, it would suggest connectivity-based subgroups may be
Connectivity Subtypes Predict Attentional Profiles in Anxiety
21
useful biomarkers that could potentially improve classification algorithms for clinical diagnosis
by explicitly accounting for patient heterogeneity, as previously suggested in depression research
(Drysdale et al., 2017).
Limitations. Connectivity subgroups were unrelated to available clinical measures
(diagnostic subtypes, continuous self-report measures) in the present dataset. Although the
ability to detect such relationships may have been constrained by the specific diagnostic
composition and/or small sample sizes of the present sample, these null relationships could also
limit clinical applications of the current findings if subtypes can only be identified via less
clinically available measures (i.e., fMRI and/or attention bias assessments). Nevertheless, our
general S-GIMME approach shows relevance in the current analysis to transdiagnostic
behavioral profiles in anxiety, and thus could be readily tested in future studies for its clinical
utility in predicting, for example, treatment response to conventional treatment options for
anxiety (e.g., psychotherapy, medications), particularly if more generalized forms of fMRI
connectivity (e.g, resting state)—rather than strictly the connectivity patterns relevant to threat
processing/attention measures per se—were incorporated as alternative (or additional) inputs to
S-GIMME. Our research group’s use of S-GIMME in patient samples to date suggests that
quantifying directed connectivity during both the resting state (Price, Gates, et al., 2017) and two
distinct task states [threat processing, in the current study; and positive mood induction (Price,
Lane, et al., 2017)] can yield subgroups that differ on external, clinically relevant variables.
Future work may benefit from incorporating connectivity patterns drawn from multiple fMRI
tasks/states into a unitary clustering algorithm, in an effort to consolidate the unique information
contained in each type of connectivity data into a single overarching classification scheme.
Connectivity Subtypes Predict Attentional Profiles in Anxiety
22
S-GIMME relies on accurate specification of a relevant network of ROIs. Results may
have differed with the inclusion of different regions in the connectivity models and/or with a
different fMRI task design, particularly given that the functional network of ROIs was identified
based on task activation patterns, which may have been sensitive to the overall sample size. Our
sample size of N=60 was well-powered to detect fMRI activations of medium-to-large, but not
small, effect size. The relatively small subgroup sample sizes (particularly Subgroup B) also
constrained power for comparisons on external variables and may increase risk of both Type I
and Type II error. The clinical utility of connectivity-based subgroups (e.g., patient sub-
classification for treatment assignment) is contingent upon establishing the robustness of these
subgroups in additional samples, and across time. In particular, although S-GIMME subgroup
solutions are robust and stable in simulated datasets as small as N=25, a direct empirical test of
subgroup stability and characteristics, both within and across anxious samples, is needed to build
further confidence in the current findings.
Conclusions. Our data-driven, brain-based categorization approach suggested that
transdiagnostic anxiety can be parsed into two unique subtypes, even among an anxious sample
with predominantly distress-related disorders. The two connectivity-based subgroups generalized
to external behavioral measures, displaying divergent patterns according to multiple indices of
attentional bias—a widely studied, transdiagnostic, clinically relevant behavioral marker of
anxiety. While a more executive-driven pattern of influences on sensory regions was predictive
of a relatively paradigmatic attentional profile [vigilance-avoidance; covert engagement bias;
(Mogg et al., 2004)], a sizable minority displayed a more affectively-driven pattern of sensory
processing, which was predictive of a divergent and atypical attentional profile (e.g.,
late/sustained vigilance in reaction times, coupled with greater initial orienting to threat in overt
Connectivity Subtypes Predict Attentional Profiles in Anxiety
23
eyetracking). DSM-based subgroups did not robustly predict attentional patterns or connectivity
subtypes, suggesting the connectivity-based subgroups provided unique, clinically relevant
information that may inform efforts to both characterize and mechanistically target attentional
patterns in anxiety. Connectivity-based subgrouping, a novel extension of biological subtyping,
could provide new insights into the diverse pathways that lead to anxiety—and, consequently,
the diverse intervention pathways that may lead back to adaptive functioning.
Authorship Statement. R.B.P. and G.J.S. developed the study concept and contributed to the
study design. Testing and data collection were performed by R.B.P., L.C., and D.G. R.B.P.,
A.M.B., and M.L.W. performed the data analysis and interpretation. R.B.P. drafted the paper. All
authors provided revisions and approved the final version of the paper for submission.
Acknowledgement. This research was supported by NIH Career Development Award
K23MH100259 (Price). We gratefully acknowledge Kathleen Gates, PhD for her assistance with
this work.
Financial Disclosures. The authors reported no biomedical financial interests or potential
conflicts of interest.
!
Connectivity Subtypes Predict Attentional Profiles in Anxiety
24
References
Amir, N., Taylor, C. T., & Donohue, M. C. (2011). Predictors of response to an attention
modification program in generalized social phobia. Journal of Consulting and Clinical
Psychology, 79, 533-541. doi:10.1037/a0023808
Bar-Haim, Y., Lamy, D., Pergamin, L., Bakermans-Kranenburg, M. J., & van IJzendoorn, M. H.
(2007). Threat-related attentional bias in anxious and nonanxious individuals: a meta-analytic
study. Psychological Bulletin, 133(1), 1-24. doi:2006-23058-001 [pii]
10.1037/0033-2909.133.1.1 [doi]
Bar-Haim, Y., Morag, I., & Glickman, S. (2011). Training anxious children to disengage
attention from threat: a randomized controlled trial. Journal of Child Psychology and
Psychiatry, and Allied Disciplines, 52, 861-869. doi:10.1111/j.1469-7610.2011.02368.x
Beltz, A. M., & Gates, K. M. (2017). Network Mapping with GIMME. Multivariate Behavioral
Research, 52(6), 789-804. doi:10.1080/00273171.2017.1373014
Beltz, A. M., Moser, J. S., Zhu, D. C., Burt, S. A., & Klump, K. L. (2018). Using person-specific
neural networks to characterize heterogeneity in eating disorders: Illustrative links between
emotional eating and ovarian hormones. International Journal of Eating Disorders, 51, 730-
740.
Beltz, A. M., Wright, A. G., Sprague, B. N., & Molenaar, P. C. (2016). Bridging the nomothetic
and idiographic approaches to the analysis of clinical data. Assessment, 23(4), 447-458.
doi:10.1177/1073191116648209
Best, M. W., Milanovic, M., Iftene, F., & Bowie, C. R. (2019). A Randomized Controlled Trial
of Executive Functioning Training Compared With Perceptual Training for Schizophrenia
Connectivity Subtypes Predict Attentional Profiles in Anxiety
25
Spectrum Disorders: Effects on Neurophysiology, Neurocognition, and Functioning.
American Journal of Psychiatry, 176(4), 297-306. doi:10.1176/appi.ajp.2018.18070849
Bishop, S. J. (2007). Neurocognitive mechanisms of anxiety: an integrative account. Trends in
Cognitive Sciences, 11(7), 307-316. doi:S1364-6613(07)00132-5 [pii]
10.1016/j.tics.2007.05.008 [doi]
Browning, M., Holmes, E. a., & Harmer, C. J. (2010). The modification of attentional bias to
emotional information: A review of the techniques, mechanisms, and relevance to emotional
disorders. Cognitive, Affective & Behavioral Neuroscience, 10, 8-20.
doi:10.3758/CABN.10.1.8
Clementz, B. A., Sweeney, J. A., Hamm, J. P., Ivleva, E. I., Ethridge, L. E., Pearlson, G. D., et
al. (2016). Identification of distinct psychosis biotypes using brain-based biomarkers.
American Journal of Psychiatry, 173(4), 373-384. doi:10.1176/appi.ajp.2015.14091200
de Raedt, R., & Koster, E. H. W. (2010). Understanding vulnerability for depression from a
cognitive neuroscience perspective: A reappraisal of attentional factors and a new conceptual
framework. Cognitive, Affective & Behavioral Neuroscience, 10, 50-70.
doi:10.3758/CABN.10.1.50
Drysdale, A. T., Grosenick, L., Downar, J., Dunlop, K., Mansouri, F., Meng, Y., et al. (2017).
Resting-state connectivity biomarkers define neurophysiological subtypes of depression.
Nature Medicine, 23(1), 28-38. doi:10.1038/nm.4246
Friston, K. J. (1994). Functional and effective connectivity in neuroimaging: a synthesis. Human
Brain Mapping, 2, 56-78.
Connectivity Subtypes Predict Attentional Profiles in Anxiety
26
Gates, K. M., Lane, S. T., Varangas, E., Giovanello, K., & Guiskewicz, K. (2017). Unsupervised
classification during time series model building. Multivariate Behavioral Research, 52(2),
129-148.
Gates, K. M., & Molenaar, P. C. (2012). Group search algorithm recovers effective connectivity
maps for individuals in homogeneous and heterogeneous samples. Neuroimage, 63(1), 310-
319. doi:10.1016/j.neuroimage.2012.06.026
Gates, K. M., Molenaar, P. C., Hillary, F. G., & Slobounov, S. (2011). Extended unified SEM
approach for modeling event-related fMRI data. Neuroimage, 54(2), 1151-1158.
doi:10.1016/j.neuroimage.2010.08.051
Gates, K. M., Molenaar, P. C., Iyer, S. P., Nigg, J. T., & Fair, D. A. (2014). Organizing
heterogeneous samples using community detection of GIMME-derived resting state
functional networks. PloS one, 9(3), e91322. doi:10.1371/journal.pone.0091322
Gotlib, I. H., & Joormann, J. (2010). Cognition and depression: current status and future
directions. Annual Review of Clinical Psychology, 6, 285-312.
doi:10.1146/annurev.clinpsy.121208.131305
Grafton, B., & MacLeod, C. (2014). Enhanced probing of attentional bias: the independence of
anxiety-linked selectivity in attentional engagement with and disengagement from negative
information. Cognition & Emotion, 28(7), 1287-1302. doi:10.1080/02699931.2014.881326
Grisanzio, K. A., Goldstein-Piekarski, A. N., Wang, M. Y., Rashed Ahmed, A. P., Samara, Z., &
Williams, L. M. (2018). Transdiagnostic Symptom Clusters and Associations With Brain,
Behavior, and Daily Function in Mood, Anxiety, and Trauma Disorders. JAMA Psychiatry,
75(2), 201-209. doi:10.1001/jamapsychiatry.2017.3951
Connectivity Subtypes Predict Attentional Profiles in Anxiety
27
Heeren, A., Mogoase, C., Philippot, P., & McNally, R. J. (2015). Attention bias modification for
social anxiety: A systematic review and meta-analysis. Clinical Psychology Review, 40, 76-
90. doi:10.1016/j.cpr.2015.06.001
Kaiser, R. H., Andrews-Hanna, J. R., Wager, T. D., & Pizzagalli, D. A. (2015). Large-Scale
Network Dysfunction in Major Depressive Disorder: A Meta-analysis of Resting-State
Functional Connectivity. JAMA Psychiatry, 72(6), 603-611.
doi:10.1001/jamapsychiatry.2015.0071
Karalunas, S. L., Fair, D., Musser, E. D., Aykes, K., Iyer, S. P., & Nigg, J. T. (2014). Subtyping
attention-deficit/hyperactivity disorder using temperament dimensions: toward biologically
based nosologic criteria. JAMA Psychiatry, 71(9), 1015-1024.
doi:10.1001/jamapsychiatry.2014.763
Kendler, K. S., Prescott, C. A., Myers, J., & Neale, M. C. (2003). The structure of genetic and
environmental risk factors for common psychiatric and substance use disorders in men and
women. Archives of General Psychiatry, 60(9), 929-937. doi:10.1001/archpsyc.60.9.929
Kessler, R. C., Chiu, W. T., Demler, O., Merikangas, K. R., & Walters, E. E. (2005). Prevalence,
severity, and comorbidity of 12-month DSM-IV disorders in the National Comorbidity
Survey Replication. Archives of General Psychiatry, 62(6), 617-627. doi:62/6/617 [pii]
10.1001/archpsyc.62.6.617 [doi]
Kim, J., Zhu, W., Chang, L., Bentler, P. M., & Ernst, T. (2007). Unified structural equation
modeling approach for the analysis of multisubject, multivariate functional MRI data. Human
Brain Mapping, 28(2), 85-93. doi:10.1002/hbm.20259
Connectivity Subtypes Predict Attentional Profiles in Anxiety
28
Kruijt, A. W., Parsons, S., & Fox, E. (2018). A meta-analysis of bias at baseline in RCTs of
attention bias modification: no evidence for dot-probe bias towards threat in clinical anxiety
and PTSD. PsyArXiv, https://doi.org/10.31234/osf.io/rfjup.
Lahey, B. B., Van Hulle, C. A., Singh, A. L., Waldman, I. D., & Rathouz, P. J. (2011). Higher-
order genetic and environmental structure of prevalent forms of child and adolescent
psychopathology. Archives of General Psychiatry, 68(2), 181-189.
doi:10.1001/archgenpsychiatry.2010.192
Laird, A. R., Fox, P. M., Eickhoff, S. B., Turner, J. a., Ray, K. L., McKay, D. R., et al. (2011).
Behavioral interpretations of intrinsic connectivity networks. Journal of Cognitive
Neuroscience, 23, 4022-4037. doi:10.1162/jocn_a_00077
Lane, S. T., Gates, K. M., & Molenaar, P. C. M. (2015). gimme: https://cran.r-
project.org/web/packages/gimme/index.html.
Lane, S. T., Gates, K. M., Pike, H. K., Beltz, A. M., & Wright, A. G. C. (2019). Uncovering
general, shared, and unique temporal patterns in ambulatory assessment data. Psychological
Methods, 24(1), 54-69. doi:10.1037/met0000192
Lang, P. J., Bradley, M. M., & Cuthbert, B. N. (2008). International affective picture system
(IAPS): Affective ratings of pictures and instruction manual. Gainesville, FL: University of
Florida.
Legerstee, J. S., Tulen, J. H. M., Kallen, V. L., Dieleman, G. C., Treffers, P. D. a., Verhulst, F.
C., et al. (2009). Threat-related selective attention predicts treatment success in childhood
anxiety disorders. Journal of the American Academy of Child and Adolescent Psychiatry, 48,
196-205. doi:10.1097/CHI.0b013e31819176e4
Connectivity Subtypes Predict Attentional Profiles in Anxiety
29
Linetzky, M., Pergamin-Hight, L., Pine, D. S., & Bar-Haim, Y. (2015). Quantitative evaluation
of the clinical efficacy of attention bias modification treatment for anxiety disorders.
Depression and Anxiety, 32(6), 383-391. doi:10.1002/da.22344
MacLeod, C., Rutherford, E., Campbell, L., Ebsworthy, G., & Holker, L. (2002). Selective
attention and emotional vulnerability: assessing the causal basis of their association through
the experimental manipulation of attentional bias. Journal of Abnormal Psychology, 111(1),
107-123.
McCormick, D. A., & Bal, T. (1994). Sensory gating mechanisms of the thalamus. Current
Opinion in Neurobiology, 4(4), 550-556.
McNally, R. J. (2018). Attentional bias for threat: Crisis or opportunity? Clinical Psychology
Review. doi:10.1016/j.cpr.2018.05.005
McTeague, L. M., & Lang, P. J. (2012). The anxiety spectrum and the reflex physiology of
defense: from circumscribed fear to broad distress. Depression and Anxiety, 29, 264-281.
doi:10.1002/da.21891
Miller, M. B., Van Horn, J. D., Wolford, G. L., Handy, T. C., Valsangkar-Smyth, M., Inati, S., et
al. (2002). Extensive individual differences in brain activations associated with episodic
retrieval are reliable over time. Journal of Cognitive Neuroscience, 14(8), 1200-1214.
doi:10.1162/089892902760807203
Mogg, K., Bradley, B. P., Miles, F., & Dixon, R. (2004). Time course of attentional bias for
threat scenes: Testing the vigilance-avoidance hypothesis. Cognition and Emotion, 18(5),
689-700.
Mogg, K., Waters, A. M., & Bradley, B. P. (2017). Attention Bias Modification (ABM): Review
of Effects of Multisession ABM Training on Anxiety and Threat-Related Attention in High-
Connectivity Subtypes Predict Attentional Profiles in Anxiety
30
Anxious Individuals. Clinical Psychological Science, 5(4), 698-717.
doi:10.1177/2167702617696359
Molenaar, P. C., & Campbell, C. G. (2009). The new person-specific paradigm in psychology.
Current Directions in Psychology, 18, 112-117.
Mumford, J. A., & Ramsey, J. D. (2014). Bayesian networks for fMRI: a primer. Neuroimage,
86, 573-582. doi:10.1016/j.neuroimage.2013.10.020
Nichols, T. T., Gates, K. M., Molenaar, P. C., & Wilson, S. J. (2014). Greater BOLD activity but
more efficient connectivity is associated with better cognitive performance within a sample
of nicotine-deprived smokers. Addiction Biology, 19(5), 931-940. doi:10.1111/adb.12060
Petersen, S. E., & Posner, M. I. (2012). The attention system of the human brain: 20 years after.
Annual Review of Neuroscience, 35, 73-89. doi:10.1146/annurev-neuro-062111-150525
Peterson, S., & Posner, M. (2012). The attention system of the human brain: 20 years after.
Annual Review of Neuroscience, 35, 73-89. doi:10.1146/annurev-neuro-062111-150525
Pons, P., & Latapy, M. (2006). Computing communities in large networks using random walks.
Journal of Graph Algorithms and Applications, 10(2), 191-218.
Posner, M. I., Snyder, C. R., & Davidson, B. J. (1980). Attention and the detection of signals.
Journal of Experimental Psychology, 109(2), 160-174.
Price, M., Tone, E. B., & Anderson, P. L. (2011). Vigilant and avoidant attention biases as
predictors of response to cognitive behavioral therapy for social phobia. Depression and
Anxiety, 28, 349-353. doi:10.1002/da.20791
Price, R., Cummings, L., Gilchrist, D., Graur, S., Banihashemi, L., Kuo, S., et al. (2018). Toward
personalized, brain-based behavioral intervention for transdiagnostic anxiety: Transient
Connectivity Subtypes Predict Attentional Profiles in Anxiety
31
neural responses to negative images predict outcomes following a targeted computer-based
intervention. Journal of Consulting and Clinical Psychology, 86, 1031-1045.
Price, R. B., Allen, K. B., Silk, J. S., Ladouceur, C. D., Ryan, N. D., Dahl, R. E., et al. (2016).
Vigilance in the laboratory predicts avoidance in the real world: A dimensional analysis of
neural, behavioral, and ecological momentary data in anxious youth. Developmental
Cognitive Neuroscience, 19, 128-136. doi:10.1016/j.dcn.2016.03.001
Price, R. B., Brown, V., & Siegle, G. J. (2019). Computational Modeling Applied to the Dot-
Probe Task Yields Improved Reliability and Mechanistic Insights. Biological Psychiatry,
85(7), 606-612. doi:10.1016/j.biopsych.2018.09.022
Price, R. B., Eldreth, D. A., & Mohlman, J. (2011). Deficient prefrontal attentional control in
late-life generalized anxiety disorder: an fMRI investigation. Translational Psychiatry, 1,
e46. doi:10.1038/tp.2011.46
Price, R. B., Gates, K., Kraynak, T. E., Thase, M. E., & Siegle, G. J. (2017). Data-driven
subgroups in depression derived from directed functional connectivity paths at rest.
Neuropsychopharmacology, 42(13), 2623-2632. doi:10.1038/npp.2017.97
Price, R. B., Kuckertz, J. M., Siegle, G. J., Ladouceur, C. D., Silk, J. S., Ryan, N. D., et al.
(2015). Empirical recommendations for improving the stability of the dot-probe task in
clinical research. Psychological Assessment, 27(2), 365-376.
Price, R. B., Lane, S., Gates, K., Kraynak, T. E., Horner, M. S., Thase, M. E., et al. (2017).
Parsing heterogeneity in the brain connectivity of depressed and healthy adults during
positive mood. Biological Psychiatry, 81(4), 347-357. doi:10.1016/j.biopsych.2016.06.023
Price, R. B., Rosen, D., Siegle, G. J., Ladouceur, C. D., Tang, K., Allen, K. B., et al. (2015).
From anxious youth to depressed adolescents: Prospective prediction of 2-year depression
Connectivity Subtypes Predict Attentional Profiles in Anxiety
32
symptoms via attentional bias measures. Journal of Abnormal Psychology, 125(2), 267-278.
doi:10.1037/abn0000127
Price, R. B., Siegle, G. J., Silk, J. S., Ladouceur, C. D., McFarland, A., Dahl, R. E., et al. (2014).
Looking under the hood of the dot-probe task: An fMRI study in anxious youth. Depression
and Anxiety, 31, 178-187.
Price, R. B., Wallace, M., Kuckertz, J. M., Amir, N., Graur, S., Cummings, L., et al. (2016).
Pooled patient-level metaanalysis of children and adults completing a computer-based
anxiety intervention targeting attentional bias. Clinical Psychology Review, 50, 37-49.
Price, R. B., Woody, M. L., Panny, B. P., & Siegle, G. J. (In press). Pinpointing mechanisms of a
mechanistic treatment: Dissociable roles for overt and covert attentional processes in acute
and long-term outcomes following Attention Bias Modification. Clinical Psychological
Science.
Ratcliff, R., & McKoon, G. (2008). The diffusion decision model: Theory and data for two-
choice decision tasks. Neural Computation, 20, 873-922.
Rodebaugh, T. L., Scullin, R. B., Langer, J. K., Dixon, D. J., Huppert, J. D., Bernstein, A., et al.
(2016). Unreliability as a threat to understanding psychopathology: The cautionary tale of
attentional bias. Journal of Abnormal Psychology, 125(6), 840-851. doi:10.1037/abn0000184
Rosen, D., Price, R. B., & Silk, J. S. (Submitted). The Vigilance-Avoidance Model in Pediatric
Anxiety Disorders: Are We Looking in the Wrong Place? An Integrative Review.
Salum, G. a., Mogg, K., Bradley, B. P., Gadelha, a., Pan, P., Tamanaha, a. C., et al. (2012).
Threat bias in attention orienting: evidence of specificity in a large community-based study.
Psychological Medicine, 1-13. doi:10.1017/S0033291712001651
Connectivity Subtypes Predict Attentional Profiles in Anxiety
33
Smith, S. M., Miller, K. L., Salimi-Khorshidi, G., Webster, M., Beckmann, C. F., Nichols, T. E.,
et al. (2011). Network modelling methods for FMRI. Neuroimage, 54(2), 875-891.
doi:10.1016/j.neuroimage.2010.08.063
Somerville, L. H., Wagner, D. D., Wig, G. S., Moran, J. M., Whalen, P. J., & Kelley, W. M.
(2013). Interactions between transient and sustained neural signals support the generation and
regulation of anxious emotion. Cerebral cortex, 23, 49-60. doi:10.1093/cercor/bhr373
Watson, D., Weber, K., Assenheimer, J. S., Clark, L. a., Strauss, M. E., & McCormick, R. a.
(1995). Testing a tripartite model: I. Evaluating the convergent and discriminant validity of
anxiety and depression symptom scales. Journal of Abnormal Psychology, 104, 3-14.
Weierich, M. R., Treat, T. A., & Hollingworth, A. (2008). Theories and measurement of visual
attentional processing in anxiety. Cognition and Emotion, 22(6), 985-1018.
White, L. K., Sequeira, S., Britton, J. C., Brotman, M. A., Gold, A. L., Berman, E., et al. (2017).
Complementary Features of Attention Bias Modification Therapy and Cognitive-Behavioral
Therapy in Pediatric Anxiety Disorders. American Journal of Psychiatry, 174(8), 775-784.
doi:10.1176/appi.ajp.2017.16070847
Wilkinson, S., Holtzheimer, P. E., Gao, S., Kirwin, D., & Price, R. (2019). Leveraging
neuroplasticity to enhance adaptive learning: The potential for synergistic somatic-behavioral
treatment combinations to improve clinical outcomes in depression. Biological Psychiatry,
85, 454-465.
Yang, Z., Xu, Y., Xu, T., Hoy, C. W., Handwerker, D. A., Chen, G., et al. (2014). Brain network
informed subject community detection in early-onset schizophrenia. Scientific Reports, 4,
5549. doi:10.1038/srep05549
Connectivity Subtypes Predict Attentional Profiles in Anxiety
34
Table. Demographic and Clinical Characteristics of the Transdiagnostic Anxiety Sample
Caucasian, n (%)
31 (63%)
Female, n (%)
38 (78%)
Age
30.07 (9.70)
Anxiety diagnoses met, n (%):
GAD
SAD
Panic/agoraphobia
PTSD
Specific Phobia
OCD
50 (83%)
22 (37%)
12 (20%)
7 (12%)
6 (10%)
4 (7%)
Comorbid depressive disorder, n (%)
18 (30%)
Primary distress-related anxiety diagnosis, n (%)
51 (85%)
Number of anxiety diagnoses
1.72 (1.04)
Self-report symptom indices
MASQ: Anxious Arousal
32.65 (10.82)
MASQ: General Distress
58.77 (16.26)
MASQ: Anhedonic Depression
74.30 (11.52)
PSWQ
65.95 (9.34)
Note: Data presented as mean (SD) unless otherwise noted. PSWQ=Penn State Worry Questionnaire; MASQ=Mood
and Anxiety Symptoms Questionnaire; GAD=Generalized Anxiety Disorder; SAD=Social Anxiety Disorder;
PTSD=Posttraumatic Stress Disorder; OCD=Obsessive-Compulsive Disorder; NOS=Not Otherwise Specified.
Figure 1: (A) Functional regions of interest represented as nodes in rough anatomical space. Nodes of the affective network (AN) are
presented in blue; sensory processing regions in green; and executive network (EN) in purple. Group-level directed
connectivity paths between regions are depicted with arrows. Black, solid arrows represent positive, contemporaneous paths;
red, dashed arrows represent negative, lagged paths. Brown, solid arrows represent positive, contemporaneous paths present in
the whole group, but significantly stronger in Subgroup B (corrected for multiple comparisons). Not shown: positive, lagged
autoregressive paths were also present for every region.
(B) Directed connectivity paths unique to subgroup A (green/solid=positive, contemporaneous path; red/dashed=negative, lagged
path), superimposed on group-level connectivity map (in grey).
(C) Directed connectivity paths unique to subgroup B (in green), superimposed on group-level connectivity map (in grey).
Figure 1
1A: Group-level paths
Anterior
Posterior
Ventral
Dorsal
pgACC
DMPFC
Thalamus
R Insula
L
Amygdala
R
Amygdala
L Parietal
R Parietal
R Occipital
L Insula
L Occipital
1B: Subgroup A paths
Anterior
Posterior
Ventral
Dorsal
pgACC
DMPFC
Thalamus
R Insula
L
Amygdala
R
Amygdala
L Parietal
R Parietal
R Occipital
L Insula
L Occipital
pgACC
DMPFC
Thalamus
R Insula
L
Amygdala
R
Amygdala
L Parietal
R Parietal
R Occipital
L Insula
L Occipital
1C: Subgroup B paths
Anterior
Posterior
Ventral
Dorsal
Figure 2: Behavioral attention bias indices as a function of connectivity-based subgroups (Subgroup A: blue gradient; Subgroup B:
solid orange). All attention bias measures are quantified on a continuum such that higher/more positive scores indicate
increasing vigilance to threat cues, while lower/more negative scores indicate relative avoidance of threat cues. Top left panel:
dot-probe reaction time bias scores after applying drift-diffusion modeling (in seconds); top right panel: spatial cueing task
reaction time bias scores (in milliseconds); bottom left panel: % of trials with initial overt eye fixation made to threat cue
during dot-probe task (as % of all trials); bottom right panel: bias in the delay to disengage overt eye gaze from threat vs.
neutral cues during dot-probe task (units = eyetracking samples; 1 unit=16.7ms). P-values displayed reflect all significant
(p<.05) and trend-level (p<.10) effects according to t-tests.
!
Figure 2
-0.025
-0.02
-0.015
-0.01
-0.005
0
0.005
0.01
0.015
0.02
0.025
Dot-probe RT bias: short trials Dot-probe RT bias: long trials
Subgroup A Subgroup B
p=.02
p=.06
-10
-8
-6
-4
-2
0
2
4
6
8
10
Spatial cueing: engagement bias Spatial cueing: disengagement bias
p=.01
0.36
0.37
0.38
0.39
0.4
0.41
0.42
0.43
0.44
0.45
Dot-probe eyetracking: initial fixation
p=.02
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Dot-probe eyetracking: disengagement bias
Avoidant Vigilant
l
Avoidant Vigilant
l
Avoidant Vigilant
l
More Vigilant
l
p=.02
Price et al. Supplement
1
Neural Connectivity Subtypes Predict Discrete Attentional Bias Profiles
Among Heterogeneous Anxiety Patients
Supplemental information
Supplemental Methods & Materials ....................................................................................... 2!
Participants ........................................................................................................................... 2!
fMRI Preprocessing ............................................................................................................. 3!
Region of Interest Definitions .............................................................................................. 3!
Handling and Impact of Motion During Scanning ............................................................. 6!
Further Methodological Details on Indices of Overt and Covert Attention Bias ............. 7!
Sources of Missing Data ................................................................................................... 10!
Supporting Findings ........................................................................................................... 10!
Analyses of Model Fits ...................................................................................................... 10!
Supplemental References ..................................................................................................... 12!
Price et al. Supplement
2
Supplemental Methods & Materials
Participants
Participants were recruited and enrolled from 03/2013-11/2016, stopping when the target
randomization sample of 70 participants was obtained. Inclusion criteria specified that
participants: 1) be between the ages of 18 and 55 years; 2) score >=45 on the Spielberger State-
Trait Anxiety Inventory—trait form, a clinically relevant cut-point that statistically maximizes
discrimination between clinical and non-clinical anxiety (1) and eliminates the bottom ~84% of a
typical healthy control distribution (2); and 3) score >=75th percentile on the World Health
Organization Disability Assessment Schedule 2.0—clinician-rated version, a percentile that is
characteristic of individuals with one or more mental disorders (3).
Exclusion criteria included the following: 1) current medication or Cognitive-Behavioral
Therapy for anxiety or depression; 2) failure to meet standard Magnetic Resonance Imaging (MRI)
safety criteria; 3) pregnancy, determined by pregnancy tests on females; 4) currently suicidal or
at risk for harm to self or others; 5) visual disturbance (<20/40 as per the Snellen test, corrective
lenses allowed); 6) <6th grade reading level as per the Wide Range Achievement Test; 7)
presence of bipolar, psychotic, autism spectrum, substance dependence, or primary depressive
disorder; 8) positive urine drug test.
These criteria produced a study sample in which 93% of randomized participants had one
or more diagnosed anxiety disorder at baseline (mean=2.10 DSM-IV-TR disorders), while the
remaining 7% (n=5) captured ‘diagnostic orphans’ with clinically significant anxiety and associated
impairment but who did not meet full criteria for any DSM-IV-TR anxiety disorder. Diagnoses were
established by experienced master’s-level (or higher) clinicians using the MINI International
Neuropsychiatric Interview.
The study was approved by the Internal Review Board of the University of Pittsburgh and
was pre-registered at clinicaltrials.gov (NCT02303691).
Price et al. Supplement
3
fMRI Preprocessing
The following preprocessing steps were applied, as described in more detail previously
(4): slice time correction, cross-registration of functional data to a high-resolution structural scan
acquired in the same fMRI session (axial MPRAGE: TR=2100; TE=3.31; 176 slices; flip
angle=8°; FOV=256x208; 1mm isotropic voxels), 6-parameter motion correction, linear
detrending to correct drift, conversion to percent change, 32-parameter nonlinear warping to the
Montreal Neurological Institute Colin-27 brain data set, spatial smoothing [6-mm full width half
maximum].
Region of Interest Definitions
Network nodes were defined based on task activation/deactivation maps for the Negative
image block (sustained) and event (transient) regressors across the full sample. The mixed
block/event-related task design and regression analyses were optimized in previous research (5)
to allow estimation of transient responses to each image type embedded within the sustained
response to each overarching block type. At the single-subject level, transient and sustained
responses were modeled simultaneously using box-car regressors for each block type (Negative
Predictable, Negative Unpredictable, Neutral Predictable, Neutral Unpredictable) and finite
impulse response functions over 20sec to flexibly model responses to each transient image
(separate Negative-Transient and Neutral-Transient regressors), along with nuisance covariates
(e.g., motion parameters). As in prior work (5), an area under the curve (AUC) was computed for
each transient regressor to capture the magnitude of transient response from 2-6 TRs (4-12secs)
after each image presentation as the sum of finite impulse response function beta weights derived
at each of the relevant timepoints (4-12s post-stimulus onset) in response to transient images of
each type (negative, neutral). Beta-weights for sustained block-type regressors and AUCs for
transient event-related regressors were then averaged for each emotional image type (negative
Price et al. Supplement
4
and neutral) and subjected to one-sample t-tests (t-value vs. 0) at each voxel, applying multiple
comparisons correction to hold map-wise p<.05 [voxel-wise p<.001, map-wise multiple
comparisons correction via AFNI’s 3dclustsim utilizing the spatial autocorrelation function (‘-acf’)
option; (6)]. The resulting clusters of activation (or deactivation) to negative task inputs that were
significant at the group level were converted to region-of-interest spheres centered on the peak
voxel from the activated cluster, using sphere radii of 8-12mm, selected according to the
maximum radius that approached but did not significantly transgress the boundaries of the
functional cluster. This approach ensured the ROIs closely approximated the size, shape, and
location of each functional cluster, while simultaneously ensuring each ROI was spatially and
anatomically distinct and non-contiguous with all others in the model. In the case of regions that
were significantly modulated by both block (sustained) and event (transient) regressors, the peak
voxel locations from the block-type analysis were used.
Timeseries data were then extracted from each spherical ROI for each participant, and
along with vectors specifying Negative and Neutral task-related timepoints (convolved with a finite
impulse response), subjected to connectivity analyses with S-GIMME (as described in main text).
These connectivity analyses are based on the covariance structure across regions for each
individual, computed across the full timeseries, and thus are distinct from the task-based, group-
mean analyses used to define ROIs functionally; the inclusion of the task-based vectors in the S-
GIMME analyses confirmed that task-related findings were not driving results. Applying S-GIMME
to regions of interest that were robustly functionally modulated by the negative task inputs allowed
us to characterize the connectivity within a network of regions that was most relevant to visual
threat processing within this sample, while retaining our focus on the identification of individual
differences in the covariance (connectivity) patterns across this network of interacting regions that
were stable across all portions of the task and reflected region-to-region influences present above
and beyond the influence of specific external (task-related) inputs. See Table S1 and Figure S1
below for ROI details.
Price et al. Supplement
5
Table S1. Analysis of regional response to sustained blocks of negative images and/or event-
related negative image presentations across full sample
Note: Coordinates for each cluster’s peak voxel are presented in Talairach space. Findings are
from unrestricted whole brain analysis with voxel-wise error rate p <.001; map-wise corrected
error rate p < .05 via AFNI’s 3dClustSim, utilizing the spatial autocorrelation function (‘-acf’)
option. This approach provides accurate Type I error control (6). DMPFC=dorsomedial
prefrontal cortex; pgACC=perigenual anterior cingulate cortex.
!
Region
Location of peak voxel
x
y
z
Sphere
radius
Network label (in Figure
1A, main text)
Activated regions:
negative blocks > 0
1: L occipital
L middle occipital gyrus
-36
-80
0
10
Sensory Processing
2: R occipital
R middle occipital gyrus
39
-80
3
10
Sensory Processing
3: DMPFC
L medial frontal gyrus
-4
3
54
12
Executive Network (ExN)
4: R insula
R insula
44
11
15
12
Affective Network (AN)
5: L insula
L insula
-40
21
3
12
Affective Network (AN)
6: Thalamus
thalamus
0
-17
13
12
Sensory Processing
7: L Parietal
L inferior parietal lobule
-50
-51
38
10
Executive Network (ExN)
8: R Parietal
R inferior parietal lobule
53
-52
40
10
Executive Network (ExN)
(Non-redundant)
activated regions:
negative events > 0:
9: L amygdala
L amygdala
-22
-4
-11
8
Affective Network (AN)
10: R amygdala
R amygdala
26
-7
-12
8
Affective Network (AN)
Deactivated regions:
negative events < 0:
11: pgACC
L anterior cingulate
-1
44
3
12
Affective Network (AN)
Price et al. Supplement
6
Figure S1. Anatomical placement of spherical ROIs (as numbered in Table S1 above)
Handling and Impact of Motion During Scanning
Of 69 participants who completed the task, 9 (13%) were excluded from analysis due to
excessive motion during the task (defined as >30% of timepoints showed framewise motion
>0.2mm or >0.2°). Participants with vs. without usable imaging data from the task did not differ
on any demographic variable presented in the Table (main text; p’s>.38), but those who were
excluded due to excessive motion reported significantly less depression and general distress (per
MASQ subscales) and less worry (PSWQ) (all p’s<.05), suggesting the final analyzed sample was
more severely affected by mood and anxiety symptoms relative to those who were excluded for
motion. Thus, our findings may generalize more readily to the more severe end of the spectrum
of affective dysfunction.
To further protect against spurious connectivity patterns related to motion, we removed
individual timepoints with framewise motion >0.2mm or >0.2° from analysis. These timepoints
(6.7% of data) were marked as missing data and accounted for using full information maximum
likelihood estimation during the S-GIMME procedure to further safeguard against spurious
2
1
3
4
5
6
2
8
7
10
9
11
11
5
z=45
z=11
z=-6
Price et al. Supplement
7
(motion-induced) connectivity. While similar to ‘scrubbing’ (i.e. deleting timepoints), this approach
maintains the temporal ordering of scans.
Although this approach mitigates the influence of motion on connectivity patterns (7), we
further verified that motion did not influence findings. Across 6 motion parameters calculated for
each participant (maximum change from baseline across each of 6 movement planes: roll, pitch,
yaw, right-left, front-back, up-down), movement did not differ significantly as a function of
connectivity subgroup according to a MANOVA omnibus test (p=.11). Furthermore, no finding of
subgroup differences on attentional bias indices reported in the main text were altered after
covarying the average degree of motion across the 6 parameters and/or the percentage of
timepoints that were scrubbed for micromovement for each participant.
Further Methodological Details on Indices of Overt and Covert Attention Bias
Dot-probe task and eyetracking. Eyetracking indices were collected during an assessment
dot-probe task. Ten idiographic threat words capturing the primary foci of anxiety were selected
collaboratively by the participant and a clinical interviewer, and were subsequently matched
ideographically, on both subjective familiarity ratings and word length, to 10 neutral words drawn
from a larger normative corpus used previously in ABM research (e.g., 18). To supplement this
idiographic list with words pertinent to a broader and more generalized range of threat-related
content, these idiographic words lists were supplemented by 20 threat words and 20 neutral words
from the normative corpus. During each of 300 trials, word pairs (80% threat-neutral; 20% neutral-
neutral) were presented vertically for either 500ms (150 trials) or 1500ms (150 trials; in random
order), followed by a probe (‘E’ or ‘F’) in either the upper or lower word location. The probe
replaced either the threat or neutral word with equal likelihood. Participants responded via button
press to indicate the probe letter displayed. All text (words and probes) was presented in 14pt
font, with the distance between the upper and lower screen position transcending a visual angle
of 2°.
Price et al. Supplement
8
The dot-probe task was administered individually in a quiet, moderately lit room on a
monitor at ~68cm from the participant using Eprime software running on a PC. Eyetracking data
were collected during the task using a table-mounted RK-768 eye-tracker, consisting of a video
camera and infrared light source pointed at a participant’s eye and a device that tracked the
location and size of the pupil and corneal reflection at 60 Hz (every 16.7 ms). As previously
described (19,20), data cleaning and preprocessing were applied in Matlab. Blinks were
automatically identified and corrected using interpolation. Prior to the task, the eyetracker was
calibrated using a 9-point sequential fixation task. Gaze position during the dot-probe task was
scaled and offset based on each individual’s calibration parameters. Hand quality-checking was
performed offline to ensure accurate registration of each participant’s gaze position to the two
relevant screen positions (upper/lower) at the time that accurate responses to the probe in each
of these positions were made. Participants (baseline: n=2; post-treatment: n=2; 1-month follow-
up: n=2) whose data quality was deemed poor were excluded from all subsequent eyetracking
analysis.
Eye fixations were defined as eye positions stable within 1° of visual angle for at least
100ms. Trials comprised of less than 25% fixation or with incorrect probe responses were
removed from analysis. Fixations were then used to identify, for each individual: 1) trials in which
the first (initial) fixation that was made following word pair onset fell within a region-of-interest
defined by the threat word’s screen position (an index of engagement); and 2) trials in which the
participant’s fixation, at the time of probe onset (i.e., just when the word pair was replaced by a
probe), fell in the opposite word location from where the probe appeared (thus requiring
disengagement from one location and an overt eye movement in order to respond to the probe
accurately). These trials were separately identified as requiring disengagement from the threat
word location (disengage-threat) or disengagement from the neutral word location (disengage-
neutral). A minimum of 5 usable trials of each type (disengage-threat and disengage-neutral) was
required in order for eyetracking data to be deemed usable at a given assessment point. Among
Price et al. Supplement
9
included participants/sessions, an average of 38 trials per participant were available to generate
overt disengagement bias scores.
Spatial cueing task. The emotional spatial cueing task was developed as a modification of
covert attentional probes widely used in basic cognitive research to distinguish covert from overt
attentional processes (21). It was identical to the task used in a prior study (22) and similar to
emotional spatial cueing tasks used widely in attention research. In brief, participants were
instructed to rest their eyes on a central fixation cross which remains on-screen throughout the
task, to discourage overt eye movements. Consistent with a large literature utilizing this task to
assess covert attention, compliance with this instruction was not explicitly assessed, but task
features (e.g., placement of large, unambiguous spatial probes well within the field of peripheral
vision) encourage compliance. On each trial, a single face was presented either left or right of the
fixation cross located at the center of the screen. Face stimuli consisted of 12 different actors
(50% male) displaying neutral or angry expressions from the NimStim stimulus set (23). After a
500ms pause, two black rectangle frames appeared to the left and right of the cross, one of which
framed either a neutral or an angry face. After 500ms, the face was removed and a target (a star)
appeared for 200ms at the center of one of the two rectangles. 75% of the trials were “valid” trials,
meaning the target appeared at the location of the face cue, while the remaining 25% of the trial
targets were “invalid”, meaning the target appeared on the opposite side of the screen.
Participants pressed one of two buttons on a keyboard to indicate the location of the star. The
next trial began 1,800ms after target offset. A beep tone was played to indicate an incorrect
response or no response within the response window.
The task commenced with 8 practice trials comprised of 2 trials per condition. The
participant could elect to repeat the practice as needed. 192 trials (96 of each emotion type) were
then presented across two blocks. 25% of the trials of each emotion type were invalid cues and
75% were valid cues.
Price et al. Supplement
10
Sources of Missing Data
Of 70 participants in the study, 69 completed the MRI task, and 60 of these met our criteria
for usable data within motion limits (see main text). Of these 60 participants included in the
connectivity-based subgrouping, the following numbers of participants had unusable data on
specific attentional bias indices and were excluded from those specific analyses only: n=3 had
unusable eyetracking data due to equipment failure (n=1) or due to poor eyetracking quality
according to hand quality-checking procedures performed offline (n=2); an additional n=2 had
unusable overt disengagement bias scores because they did not meet the minimum of 5 usable
trials of each type (disengage-threat and disengage-neutral) necessary for calculating this
disengagement score from eyetracking data; n=5 did not complete the spatial cueing task due to
time constraints. The total subgroup sample sizes with available data on each external variable
(attentional bias) measure are detailed below.
Table S2.
Outcome measure Subgroup A Subgroup B
Dot-probe RTs n=41 n=19
Dot-probe eyetracking: initial fixation n=38 n=19
Dot-probe eyetracking: disengagement n=36 n=19
Spatial cueing RTs n=38 n=17
Supporting Findings
Analyses of Model Fits
S-GIMME results were assessed for robustness (e.g., potential over-fitting; see (24)) by
evaluating the covariance matrices of the error terms and standard errors of the estimated
Price et al. Supplement
11
connection parameters. A small proportion of the maps (i.e., 17%, or 10/60) had a high error
variance associated with a single ROI. However, these maps were not more likely to belong to
participants in Subgroup A (19.5%) or B (10.5%; c2=0.76, p=.39), and removal of participants from
the inferential analyses (i.e., comparisons of subgroups on attentional bias indices, etc.) did not
alter the pattern or statistical significance of any result presented in the main text. These
conservative sensitivity analyses suggest that the reported findings were robust to possible
variations in model fitting, which are common in time series and covariance structural modeling
(see (25)).
Price et al. Supplement
12
Supplemental References
1. Fisher PL, Durham RC (1999): Recovery rates in generalized anxiety disorder following
psychological therapy: an analysis of clinically significant change in the STAI-T across outcome
studies since 1990. Psychol Med. 29:1425-1434.
2. Spielberger CD, Gorsuch RL, Lushene R, Vagg PR, Jacobs GA (1983): Manual for the
State-trait Anxiety Inventory (Form Y Self-evaluation Questionnaire). Palo Alto, CA: Consulting
Psychologists Press.
3. Andrews G, Kemp A, Sunderland M, Von Korff M, Ustun TB (2009): Normative data for
the 12 item WHO Disability Assessment Schedule 2.0. PloS one. 4:e8343.
4. Price R, Cummings L, Gilchrist D, Graur S, Banihashemi L, Kuo S, et al. (2018): Toward
personalized, brain-based behavioral intervention for transdiagnostic anxiety: Transient neural
responses to negative images predict outcomes following a targeted computer-based intervention.
J Consult Clin Psychol. 86:1031-1045.
5. Somerville LH, Wagner DD, Wig GS, Moran JM, Whalen PJ, Kelley WM (2013):
Interactions between transient and sustained neural signals support the generation and regulation
of anxious emotion. Cereb Cortex. 23:49-60.
6. Cox RW, Reynolds RC, Taylor PA (2016): AFNI and clustering: false positive rates redux.
BioRXiv. doi: http://dx.doi.org/10.1101/065862.
7. Power JD, Barnes Ka, Snyder AZ, Schlaggar BL, Petersen SE (2011): Spurious but
systematic correlations in functional connectivity MRI networks arise from subject motion.
Neuroimage. 59:2142-2154.
8. Gates KM, Molenaar PC (2012): Group search algorithm recovers effective connectivity
maps for individuals in homogeneous and heterogeneous samples. Neuroimage. 63:310-319.
9. Gates KM, Molenaar PC, Iyer SP, Nigg JT, Fair DA (2014): Organizing heterogeneous
samples using community detection of GIMME-derived resting state functional networks. PloS
one. 9:e91322.
10. Gates KM, Lane ST, Varangas E, Giovanello K, Guiskewicz K (2017): Unsupervised
classification during time series model building. Multivar Behav Res. 52:129-148.
11. Lane ST, Gates KM, Pike HK, Beltz AM, Wright AGC (2019): Uncovering general,
shared, and unique temporal patterns in ambulatory assessment data. Psychol Methods. 24:54-69.
12. Nichols TT, Gates KM, Molenaar PC, Wilson SJ (2014): Greater BOLD activity but more
efficient connectivity is associated with better cognitive performance within a sample of nicotine-
deprived smokers. Addict Biol. 19:931-940.
13. Price RB, Gates K, Kraynak TE, Thase ME, Siegle GJ (2017): Data-driven subgroups in
depression derived from directed functional connectivity paths at rest. Neuropsychopharmacology.
42:2623-2632.
14. Price RB, Lane S, Gates K, Kraynak TE, Horner MS, Thase ME, et al. (2017): Parsing
heterogeneity in the brain connectivity of depressed and healthy adults during positive mood. Biol
Psychiatry. 81:347-357.
15. Smith SM, Miller KL, Salimi-Khorshidi G, Webster M, Beckmann CF, Nichols TE, et al.
(2011): Network modelling methods for FMRI. Neuroimage. 54:875-891.
16. Mumford JA, Ramsey JD (2014): Bayesian networks for fMRI: a primer. Neuroimage.
86:573-582.
17. Henry T, Gates K (2016): Causal search procedures for fMRI: review and suggestions.
Behaviormetrika.1-33.
Price et al. Supplement
13
18. Amir N, Beard C, Burns M, Bomyea J (2009): Attention modification program in
individuals with generalized anxiety disorder. J Abnorm Psychol. 118:28-33.
19. Silk JS, Stroud LR, Siegle GJ, Dahl RE, Lee KH, Nelson EE (2012): Peer acceptance and
rejection through the eyes of youth: pupillary, eyetracking and ecological data from the Chatroom
Interact task. Soc Cogn Affect Neurosci. 7:93-105.
20. Price RB, Siegle G, Silk JS, Ladouceur CD, McFarland A, Dahl RE, et al. (2013):
Sustained neural alterations in anxious youth performing an attentional bias task: A pupilometry
study. Depress Anxiety. 30:22-30.
21. Posner MI, Snyder CR, Davidson BJ (1980): Attention and the detection of signals. J Exp
Psychol. 109:160-174.
22. Bar-Haim Y, Morag I, Glickman S (2011): Training anxious children to disengage
attention from threat: a randomized controlled trial. J Child Psychol Psychiatry. 52:861-869.
23. Tottenham N, Tanaka JW, Leon AC, McCarry T, Nurse M, Hare TA, et al. (2009): The
NimStim set of facial expressions: judgments from untrained research participants. . Psychiatry
Res. 168:242-249.
24. Beltz AM, Gates KM (2017): Network Mapping with GIMME. Multivariate Behav Res.
52:789-804.
25. Lütkepohl H (2005): New introduction to multiple time series analysis. Berlin, Germany:
Springer.
