The effect of panic disorder versus anxiety sensitivity on event-related potentials during anticipation of threat

Abstract

Attention-related abnormalities are key components of the abnormal defensive responding observed in panic disorder (PD). Although behavioral studies have found aberrant attentional biases towards threat in PD, psychophysiological studies have been mixed. Predictability of threat, an important feature of threat processing, may have contributed to these mixed findings. Additionally, anxiety sensitivity, a dimensional trait associated with PD, may yield stronger associations with cognitive processes than categorical diagnoses of PD. In this study, 171 participants with PD and/or depression and healthy controls completed a task that differentiated anticipation of predictable vs. unpredictable shocks, while startle eyeblink and event-related potentials (ERPs [N100, P300]) were recorded. In all participants, relative to the control condition, probe N100 was enhanced to both predictable and unpredictable threat, whereas P300 suppression was unique to predictable threat. Probe N100, but not P300, was associated with startle eyeblink during both threatening conditions, and was strongest for unpredictable threat. PD was not associated with ERPs, but anxiety sensitivity (physical concerns) was positively associated with probe N100 (indicating reduced responding) in the unpredictable condition independent of PD diagnosis. Vulnerability to panic-related psychopathology may be characterized by aberrant early processing of threat, which may be especially evident during anticipation of unpredictable threats.

1. Introduction

Heightened defensive motivation is a key feature of many internalizing psychopathologies (i.e., anxiety disorders and depression; Lang, 1995; McTeague & Lang 2012). In panic disorder (PD), heightened defensive responding manifests as intense fear (i.e., panic attacks) and anxious apprehension of potential future panic attacks (Barlow, 2000). Learning-based conceptualizations of PD postulate that a key mechanism in PD is the unpredictability of panic attacks (Bouton, Mineka, & Barlow, 2001), which in turn leads to anticipatory anxiety about having subsequent panic attacks.

The importance of unpredictability is further highlighted by a growing literature demonstrating that predictable threats yield qualitatively different responses from unpredictable threats, with the former yielding fight-flight-freeze responses and the latter yielding more sustained states of preparedness for potential threat (Grillon, 2002; Shankman et al., 2013). The distinction between predictable and unpredictable threat has been validated by non-human animal (Davis, 1998, 2006; Gray & McNaughton, 2000), pharmacological challenge (Grillon et al., 2006; Grillon, Pine, et al., 2009; Moberg & Curtin, 2009), and neuroimaging (Alvarez, Chen, Bodurka, Kaplan, & Grillon, 2011) studies. To examine differences in defensive responding to predictable versus unpredictable threat, Grillon and colleagues developed the No-Predictable-Unpredictable (NPU) threat task (Schmitz & Grillon, 2012). There are three conditions in the NPU task: (1) no threat, (2) predictable threat (i.e., threat is signaled by a cue), and (3) unpredictable threat (i.e., threat is unsignaled and may be presented at any time). Defensive responding is typically operationalized as the magnitude of the startle eyeblink reflex to loud acoustic startle probes, and has been shown to be enhanced during predictable and unpredictable threat conditions relative to the no threat condition (Gorka, Lieberman, Shankman, & Phan, 2017; Grillon et al., 2009; Nelson et al., 2013; Shankman et al., 2013).

A few studies have examined the association between PD and the startle reflex during the NPU-threat task. Grillon and colleagues (2008) found that patients with PD, relative to healthy controls, exhibited greater startle potentiation to unpredictable threat, but did not differ on response to predictable threat. Shankman and colleagues (2013) sought to extend these findings by administering the NPU-threat task to individuals with PD without major depressive disorder (MDD), PD with MDD, and MDD without PD. Individuals with PD (irrespective of comorbid MDD), but not those with MDD only, exhibited heightened startle potentiation to unpredictable threat but unlike Grillon et al. also exhibited heightened startle potentiation to predictable threat. Furthermore, startle potentiation to unpredictable threat was uniquely associated with family history of PD, independent of participant diagnosis of PD (Nelson et al., 2013). These results suggest that response to threat, and perhaps only unpredictable threats, is a key feature of PD.

Another important component of defensive responding is attentional engagement with threat (Lang, 1995). Studies examining attentional biases to threat among individuals with PD have yielded mixed results, with some studies finding greater attentional bias toward threat- and/or or panic-related stimuli (Lundh, Wikström, Westerlund, & Öst, 1999; Reinecke, Cooper, Favaron, Massey-Chase, & Harmer, 2011) and others finding no differences from controls (De Cort, Hermans, Spruyt, Griez, & Schruers, 2008; Kampman, Keijsers, Verbraak, Naring, & Hoogduin, 2002).

One explanation for these discrepant findings is that the behavioral measures typically used to assess attentional biases, such as the dot probe paradigm (MacLeod, Mathews, & Tata, 1986), have poor reliability (Kappenman, Farrens, Luck, & Proudfit, 2014; Schmukle, 2005), potentially due to the temporal and neurophysiological separation between the construct being measured (i.e., attention) and the behavior indexing the construct (e.g., motoric response selection). In contrast, ERP indices of attention have better psychometric properties, (Levinson, Speed, Infantolino, & Hajcak, 2017; Segalowitz & Barnes, 1993), in part because of their closer temporal proximity to attentional processes. Although the startle eyeblink reflex is sensitive to individual differences in attention (Blumenthal et al., 2005), another advantage of ERP measures of attention-related processes is that they can distinguish different components of attention. Given evidence that anxiety specifically influences early, automatic attentional processes, as opposed to later, more elaborative stages of attentional processing (Weinberg & Hajcak, 2011; Weinberg, Perlman, Kotov, & Hajcak, 2016), it is particularly important to isolate early attentional components.

ERPs that index attention-related processes can be measured during the NPU-threat task by examining ERPs to startle probes during each condition. Probe-elicited ERPs differentiate several components of cognitive processing that likely relate to attention – most notably an N100 and a P300 (Cuthbert, Schupp, Bradley, McManis, & Lang, 1998; Schupp, Cuthbert, Bradley, Birmaumer, & Lang, 1997). The N100 is a negative-going deflection that likely indexes early attention and sensory processing (Cuthbert et al., 1998). The P300 is a positive-going deflection that likely reflects, in part, allocation of attention to salient stimuli, regardless of valence (Cuthbert et al., 1998). In prior threat-of-shock and affect modulation studies, the probe N100 is enhanced in threat or aversive relative to “safe” or positive conditions (Al-Abduljawad, Baqui, Langley, Bradshaw, & Szabadi, 2008; Cuthbert et al., 1998). In contrast, the probe P300 is attenuated; rather than attending to salient startle probes, attention is allocated to the more salient threatening context (Cuthbert et al., 1998; Shackman et al., 2011).

Our group previously investigated ERPs to startle probes during the NPU-threat paradigm in a sample of undergraduates (Nelson, Hajcak, & Shankman, 2015). Probe N100 was enhanced during the unpredictable (versus no threat) condition, but not during the predictable threat condition, suggesting that unpredictable threatening contexts may particularly increase this early sensory/attentional component. In contrast, probe P300 was attenuated during both predictable and unpredictable conditions relative to the no threat condition, suggesting that participants’ attention to salient contexts (i.e., threat of shock) may have increased to both threat conditions.

While several studies have examined the association between PD and the probe N100 and P300 (albeit with equivocal results; Clark, McFarlane, Weber, & Battersby, 2009; Di Giorgio, Velasques, Ribeiro, Nardi, & de Carvalho, 2015), no study has examined whether threat predictability impacts the association between PD and these ERP components. Given our aforementioned finding that familial vulnerability for PD was only related to startle eyeblink potentiation to unpredictable threat (Nelson et al., 2013), aberrant processing of threat in PD indexed by these ERPs may also be specific to unpredictable threats. The primary aim of the present study was therefore to extend the EMG startle findings of Shankman et al. (2013) by examining the association between ERP indices of attention-related processes and PD during predictable and unpredictable threatening contexts.

Studies of attentional deficits in PD may have yielded mixed results also because PD was defined categorically. Numerous studies of multiple psychopathologies have shown that a categorical conceptualization of psychopathology might not ‘carve nature at its joints’ and that a dimensional conceptualization likely has greater validity than categorical diagnoses (Helzer, Kraemer, & Krueger, 2006; Kendell & Jablensky, 2003). Given that PD and its mechanisms are heterogeneous, studies that define participants by individual differences on particular sensitivities rather than by categorical DSM diagnoses might have better predictive validity. Anxiety sensitivity (AS) is one such dimension that is principally relevant to PD (Taylor & Fedoroff, 1999). AS is a clinical trait that reflects sensitivity to physical sensations associated with threat responding that are perceived as harmful or having cognitive or social consequences, and has been shown to connote vulnerability for multiple psychopathologies (Epkins, Gardner, & Scanlon, 2013; Schmidt, Lerew, & Jackson, 1997; Taylor & Fedoroff, 1999).

Moreover, AS has been shown to be associated with attentional vigilance for physical threat-related words (e.g., breathless, harm; Keogh, Dillon, Georgiou, & Hunt, 2001; Teachman, Smith-Janik, & Saporito, 2007; but see Lang & Sarmiento, 2004 for null results). Consistent with these findings, Nelson, Hodges, Hajcak, & Shankman, (2015) reported that high levels of AS were associated with greater probe N100 enhancement during anticipation of unpredictable threat and greater probe P300 suppression in anticipation of both predictable and unpredictable threat. These findings suggest that AS is associated with altered patterns of attentional/cognitive processing of threat. Despite promising results, Nelson, Hodges et al. (2015) examined this question in a sample of college students (which tend to be healthier than clinical samples; Coyne, 1994) resulting in a restricted range of AS scores, preventing generalization to clinical populations. Thus, the second aim of the present study was to examine, in a clinical sample, the association between a continuous measure of anxiety (i.e., AS) and the probe N100 and P300 during the NPU-threat task.

Given the high rates of comorbidity between anxiety and MDD (Shankman & Klein, 2003), it is important to isolate the effects of these conditions on cognitive processing (Miller & Chapman, 1985). Extant research indicates a strong relationship between anxiety and unpredictable threat responding, whereas findings regarding MDD are mixed (Grillon et al, 2013; Shankman et al., 2013). Additionally, because anxiety is characterized by deficits in early cognitive processes, whereas MDD is characterized by disruptions in later, elaborative processing (Sass et al., 2014; Weinberg et al., 2016), alterations in relatively early ERP components such as the N100 and P300 are likely to be unique to individuals with anxiety.

In sum, the present study seeks to extend the EMG startle findings from Shankman et al. (2013) to probe-elicited ERPs and test the following hypotheses:

• Hypothesis 1: All individuals would exhibit an enhanced probe N100 and suppressed probe P300 during threat anticipation relative to no threat conditions, but this effect would be more pronounced among those with PD compared to those without PD.

• Hypothesis 2: Given that AS is a transdiagnostic dimension of psychopathology that more specifically indexes sensitivity to threatening stimuli (Keogh et al., 2001), we hypothesized that AS would also be associated with enhanced N100 and suppressed P300 during threat anticipation relative to no threat conditions, and that these effects would be stronger than those of PD.

Based on EMG startle studies (Grillon et al., 2008; Nelson et al., 2013), these effects may be stronger for unpredictable than predictable threats. As the present study seeks to extend prior EMG startle findings, we will also examine concordance between EMG startle and probe-elicited ERPs.

2. Methods

2.1. Participants

Participants were a subset of 181 individuals from Shankman et al. (2013) (see also Gorka, Liu, Sarapas, & Shankman, 2015). Participants were recruited from the community and clinics and were included in the study if they met current diagnostic criteria for (1) panic disorder (PD; n = 27), (2) MDD (n = 37), (3) comorbid PD and MDD (PD+MDD; n = 56), or (4) no current psychopathology (healthy controls; n = 61) according to the Structured Clinical Interview for DSM-IV (SCID; First, Spitzer, Gibbon, & Williams, 1996). Individuals in the PD and PD+MDD groups could have other past or current anxiety disorders. Individuals in the MDD group were required to have a depression onset before age 18 to reduce heterogeneity of depression. Exclusion criteria were lifetime history of psychosis, bipolarity, or dementia; inability to read/write English; left-handedness; and history of head trauma. Seven participants were excluded due to unusable ERP data (more than 50% of trials contained artifacts¹) and three more were outliers (defined as ERPs below or above the first or third quartile, respectively, by at least 1.5 times the interquartile range), yielding the final N of 171. Demographic and clinical characteristics are in Table 1.

Table 1.

Sample demographic characteristics

	Control (n = 58)	PD-only (n = 26)	Dep-Only (n = 36)	PD+Dep (n = 51)	Total (n = 171)
Measure	M (SD)	M (SD)	M (SD)	M (SD)	M (SD)
Sex (% Female)	60.3	57.7	63.9	72.5	64.0
Age	32.21 (12.97)	33.33 (11.98)	30.85 (12.33)	36.81 (11.32)	33.47 (12.32)
Ethnicity (%)
White	39.7	42.3	47.2	47.1	43.4
Black	27.6	23.1	25.0	32.2	27.4
Asian	27.6	15.4	16.7	13.7	18.9
Hispanic	5.2	19.2	11.1	5.9	9.7
ASI-3 Total	9.29 (6.51)	27.56 (14.88)	21.06 (11.37)	29.67 (15.34)	20.78 (14.83)
PC	2.64 (2.44)	10.60 (5.42)	4.97 (3.28)	9.76 (6.43)	6.54 (5.68)
CC	1.36 (1.98)	6.48 (6.39)	6.62 (5.55)	7.71 (6.27)	5.20 (5.73)
SC	5.29 (3.55)	10.48 (5.45)	9.47 (5.09)	12.20 (5.51)	9.11 (5.58)
Data Quality (rejected trials)
No Threat	0.00 (0.00)	0.23 (1.18)	0.00 (0.00)	0.10 (0.71)	0.06 (0.60)
Predictable	0.05 (0.29)	0.42 (1.27)	0.11 (0.46)	0.48 (2.04)	0.25 (1.25) a
Unpredictable	0.05 (0.29)	0.50 (1.48)	0.14 (0.49)	0.44 (1.77)	0.25 (1.16) a

Note: ASI-3 = Anxiety Sensitivity Index 3; PC = Physical Concerns; CC = Cognitive Concerns; SC = Social Concerns

^aThe number of trials rejected due to artifacts in the NPU-threat task varied by Condition [F(2, 332) = 9.03, p = .002, η_p² = .052], such that fewer trials were rejected in the N condition than during both the P (p = .020) and U (p = .002) conditions, which did not differ from each other (p > .999). There were no effects of diagnostic group or a Group X Condition interaction (ps > .248).

2.2. Measures

2.2.1 Anxiety Sensitivity Index 3 (ASI-3; Taylor et al., 2007)

AS was assessed with the ASI-3, an 18-item self-report measure consisting of three 6-item subscales that depict fear of specific anxiety symptoms and associated consequences - physical concerns (PC), cognitive concerns (CC), and social concerns (SC) related to anxiety. Items are rated on a Likert scale from 1 (“very little”) to 5 (“very much”). The ASI-3 subscales have demonstrated excellent psychometric properties (Olthuis, Watt, & Stewart, 2014; Wheaton, Deacon, McGrath, Berman, & Abramowitz, 2012). In the current sample, all scales were normally distributed, and internal consistency was excellent for the total score (α = .93) and acceptable to excellent on subscale scores (PC α = .88; CC α = .90; SC α = .78).

2.3. Procedure: NPU-Threat Task

The task is described in greater detail in Shankman et al. (2013; see Schmitz & Grillon, 2012). Stimuli were presented using PSYLAB (Contact Precision Instruments, London) hardware and software. Acoustic startle probes were 40-ms, 103dB bursts of white noise with near-instantaneous rise times presented bi-aurally through headphones. First, participants underwent a workup procedure to determine their idiographic (e.g, “highly annoying, but not painful”) shock level, followed by a habituation task consisting of 9 startle probes. Next, participants completed the NPU-threat task, which contained three conditions (no shock [N], predictable shock [P], unpredictable shock [U]) indicated via text at the bottom of the screen (i.e., “no shock,” “shock possible during square,” “shock possible at any time”) and presented in two blocks (condition order [counterbalanced]: PNUNPU or UPNUNP). Each 90s condition included four 8-second geometric shape “cue” presentations (N: blue circle, P: red square, U: green star to aid in differentiation of conditions) separated by inter-stimulus intervals (ISIs) that ranged from 7–17s (M = 12.4s). During the N condition, participants did not receive any shocks. During P, participants could only receive a shock when the cue was on the screen². During U, participants could receive a shock at any time (i.e., during cues or ISIs). Startle probes were presented during both cues and ISIs. Shocks were administered to participants’ left wrist for 400ms. Probes occurred at least 10s after shocks to minimize the shock’s impact on startle responding. Participants received 12 shocks (6 in each of the P and U conditions) and 72 startle probes (24 each in N, P, and U).

2.4. EMG startle recording and processing

EMG startle processing is described in prior studies with this sample (Nelson et al., 2013; Shankman et al., 2013). Briefly, EMG was recorded from the orbicularis oculi below the right eye and processed per published guidelines (e.g., filtered, baseline corrected, etc.). Average magnitude (non-response trials included in averages as “0”) for each condition (N, P, U; cues and ISIs) was calculated.

2.5. EEG Recording and Processing

Electroencephalography (EEG) was recorded using Neuroscan Synamp2 (Compumedics, Charlotte, NC, USA) and measured from Ag/AgCl electrodes in a 64-channel lycra cap. The ground electrode was AFz and the online reference was in between Cz and CPz. Electrodes at the right supra-and infra-orbital sites and the right and left outer canthi were used to monitor vertical and horizontal eye movements, respectively. Electrode impedances were under 5kΩ. EEG was recorded at a gain of 10K (5K for eye channels) with a band-pass of DC-200 Hz and digitized at 1000 Hz. Offline, EEG data were processed using Brain Vision Analyzer (BrainProducts, Munich). Data were re-referenced to the average of the left and right mastoid and band-pass filtered from 0.1 to 30 Hz. Eyeblink and ocular corrections were conducted using established standards (Gratton, Coles, & Donchin,1983; although see Plöchl, Ossandón, & König, 2012 for limitations of this method), which may be particularly useful when, as in the NPU-threat task, ERPs of interest occur temporally proximal to blinks (Gratton, 1998; Gratton et al., 1983).

EEG output was segmented by trial into 1200ms windows from 200ms before (i.e., baseline) to 1000ms after probe onset. A semiautomatic procedure was employed to detect and reject artifacts from individual channels on each trial. The criteria applied were a voltage step of more than 50μV between sample points, a voltage difference of 300μV within a trial, and a maximum voltage difference of less than 0.50μV within 100ms intervals. Remaining artifacts were rejected via visual inspection of data.

2.6. Principal Components Analysis

A principal components analysis (PCA) was conducted to isolate the probe-elicited N100 and P300 (see grand average waveforms in supplementary Figure 1 and 2). The baseline-corrected average for each condition (N_Cue, N_ISI, P_Cue, P_ISI, U_Cue, U_ISI), electrode location, and participant was entered into the data matrix. Using the MATLAB ERP PCA Toolbox–Version 2 (Dien, 2010b), a temporal PCA was performed in order to capture variance across time and to maximize the initial separation of ERP components (Dien & Frishkoff, 2005). Promax rotation was used to rotate to simple structure in the temporal domain (Dien, Khoe, & Mangun, 2007; Dien, 2010a). Following the first rotation, a parallel test (Horn, 1965) was conducted on the resulting Scree plot (Cattell, 1966), in which the Scree plot of the actual dataset is compared to that derived from a fully random dataset. The number of factors retained is based on the largest number of factors that account for a greater proportion of variance than the fully random dataset (see Dien, 2010b for more information). Based on this criterion, 40 temporal factors were extracted for rotation and the covariance matrix and Kaiser normalization were used for the PCA (Dien, Beal, & Berg, 2005). Next, a spatial PCA was performed on each temporal factor retained in the previous step to reduce the spatial dimensions of the datasets. Infomax rotation was used in the spatial domain (Dien et al., 2007; Dien, 2010a). Based on the results of the parallel test (Horn, 1965), four spatial factors were extracted from each temporal factor for Infomax rotation, yielding 160 temporospatial factor combinations that accounted for 86.7% of the variance. To directly assess timing and spatial voltage distributions, factors were translated back into voltages.

Twenty temporospatial factor combinations each accounted for more than 1% of the variance and in total accounted for 61.5% of the variance. Five of these factors differentiated the conditions using a Holm-Bonferroni (Holm, 1979) correction (ps ≤ .003) and two resembled the temporal and spatial characteristics of the N100 and P300 (see Figures 1 and 2). TF1SF2 resembled the N100, in that it was maximal at FCz approximately 100 ms after the onset of the startle probe, and accounted for 4.0% of the variance. TF6SF1 resembled the P300, in that it was maximal at Cz approximately 300 ms after the onset of the startle probe, and accounted for 2.9% of the variance. Therefore, TF1SF2 and TF6SF1 were used as the PCA-derived N100 and P300, respectively, for subsequent analyses.

Figure 1.

PCA-derived ERP grand-average waveforms at FCz, separated by Cue (left) versus ISI (right). The probe N100 was increased during both P and U conditions compared to the N condition. The head map shows the scalp distribution of the probe N100 enhancement in the P and U conditions relative to the N condition, separated by Cue (left) versus ISI (right).

Figure 2.

PCA-derived ERP grand-average waveforms at Cz, separated by Cue (left) and ISI (right) phases of each condition. The head map shows the scalp distribution of the probe P300 suppression in the P condition relative to the N (left) and U (right) conditions.

2.7. Data analysis

Three sets of models were conducted to test the study’s aims. First, a 3 (Condition: N, P, U) X 2 (Cue: Cue, ISI) repeated-measures ANOVA examined the basic effect of the NPU-threat task, with Condition and Cue as within-subjects factors. Second, a 3 (Condition: N, P, U) X 2 (Cue: Cue, ISI) X 2 (PD Status: PD, no PD) X 2 (MDD Status: MDD, no MDD) mixed model ANOVA examined the moderating effect of diagnosis on task effect, with Condition and Cue as within-subjects factors and both PD status and MDD status as between-subjects factors. Third, a 3 (Condition: N, P, U) X 2 (Cue: Cue, ISI) X AS mixed model ANOVA examined the moderating effect of AS on the task effect, with Condition and Cue as within-subjects factors and AS as a continuous predictor. Bonferroni corrections were utilized in follow up analyses for variables with more than two levels. A Huynh-Feldt correction was used for repeated-measures analyses involving factors with more than two levels (e.g., Condition), and for these analyses, Huynh-Feldt epsilons are included.

3. Results

3.1. Demographic and Clinical Characteristics of Diagnostic Groups

There were no between-groups differences in age, sex, or ethnicity (p > .07). A 2 (PD Status: PD, no PD) X 2 (MDD Status: MDD, no MDD) multivariate analysis of variance (ANOVA) was run on ASI-3 subscale scores. As expected, there was a main effect of both PD Status [F(3, 159) = 26.45, p < .001, η_p² = .333] and MDD Status [F(3, 159) = 7.50, p < .001, η_p² = .124], but no PD Status X MDD Status interaction [F(3, 159) = 2.13 p=.099, η_p² = .039]. Follow-up analyses indicated that individuals with PD had higher AS on all three subscales than those who did not. Individuals with MDD had higher SC and CC scores than those who did not (ps < .001), but similar PC scores (p = .324).

3.2. NPU-Threat Overall Task Effect

A 3 (Condition) X 2 (Cue) repeated-measures ANOVA was run separately on the PCA-derived probe N100 and P300 to examine attentional response to threat. Analyses adjusted for age, sex, and psychotropic medications yielded nearly identical patterns of results, therefore analyses are reported without covariates in the models.

3.2.1. Probe N100

For the probe N100, there were main effects of Condition [F(2, 340) = 56.66, p < .001, ε = .88, η_p² = .250] and Cue [F(1, 174) = 12.03, p = .001, η_p² = .066], which were qualified by a Condition X Cue interaction [F(2, 340) = 3.83, p = .024, ε = .94, η_p² = .022]. Paired samples t-tests indicated that the probe N100 did not differ between the Cue and ISI phases in either the N [t(170) = −0.33, p = .740] or U [t(170) = −1.58, p = .115] conditions. In contrast, during the P condition, the probe N100 was enhanced during the Cue relative to ISI [t(170) = −3.24, p = .001]. As a follow-up, potentiation scores were calculated for the threat conditions by subtracting out activity in the corresponding N condition (i.e., P_Cue - N_Cue, and average of U_Cue, and U_ISI minus average of N_Cue and N_ISI), and one-sample t-tests were run on these potentiation scores. Both potentiation scores differed from zero [P_Cue t(170) = −7.33, p < .001; U_Cue/ISI t(170) = −8.40, p < .001], but a paired samples t-test indicated that these potentiation scores did not differ from each other [t(170) = −0.88, p = .378], suggesting that the probe N100 was enhanced (i.e., more negative) equally across threatening contexts (P and U; see Figure 1).

3.2.2. Probe P300

As with the N100, there was a Condition X Cue interaction [F(2, 348) = 8.16, p < .001, η_p² = .045], but there were no main effects of Condition or Cue (ps > .425). In the U condition, the probe P300 did not differ in the Cue and ISI phases [t(170) = 0.29, p = .771], but was enhanced in N_CUE versus N_ISI [t(170) = 2.60, p = .010] and blunted in P_Cue relative to P_ISI [t(170) = −2.94, p = .004]. Potentiation scores were calculated in the same manner as for the probe N100. One-sample t-tests indicated that only P_Cue potentiation was significantly different from zero [t(170) = −3.12, p = .002], but U_Cue/ISI potentiation did not differ from zero [t(170) = −0.17, p = .869]. Additionally, a paired-samples t-test indicated that P_Cue potentiation was significantly greater than U_Cue/ISI potentiation [t(170) = −1.79, p < .001], suggesting that probe P300 suppression was unique to predictable threat (see Figure 2).³

3.3. Moderating Effect of PD

A Condition X Cue X PD Status X MDD Status mixed-model ANOVA was run separately for the probe N100 and P300 to examine the moderating effect of categorical diagnosis on these components. There were no main effects or interactions involving either PD Status or MDD Status for either the probe N100 or the P300 (ps > .173). See Figure 3.

Figure 3.

Residualized N100 (top) and P300 (bottom) during P and U threat conditions (i.e., P_Cue adjusted for N_Cue; average of U_Cue and U_ISI adjusted for the average of N_Cue and N_ISI) among individuals with versus without PD.

3.4. Moderating Effect of AS

To examine the moderating effect of AS, a series of Condition X Cue mixed-models were run on the probe N100 and the probe P300, with all three ASI-3 subscale scores (PC, SC, and CC) included as continuous predictors⁴. Results indicated a Condition X ASI-3-PC interaction (F[2, 320] = 4.15, p = .020, η_p² = .025, ε = .90). There were no main or interactive effects of CC (ps > .562) or SC (ps > .254) independent of the other two subscales.

To follow up the Condition X ASI-3-PC interaction, partial correlations were conducted to examine the unique associations between the ASI-3-PC subscale and probe N100 to the two threat conditions after adjusting for ASI-3-CC and ASI-3-SC. Per recommendations by Meyer et al., (2017), residualized scores were calculated by regressing the probe N100 during the P and U conditions separately on the probe N100 during the N condition (i.e., P_Cue adjusting for N_Cue; average of U_Cue and U_ISI adjusting for the average of N_Cue and N_ISI). Greater ASI-3-PC was associated with reduced (i.e., more positive) residualized probe N100 to the U condition (r = 0.17, p = .028; see Figure 4), but was unrelated to residualized probe N100 to the P condition (r = 0.01, p = .878). Interestingly, there was no association between the probe N100 and ASI-3-PC during either the N (r = −0.01, p = .861) or U (r = 0.07, p = .370) conditions, suggesting that ASI-3-PC was associated with probe N100 potentiation to U, not each condition individually.

Figure 4.

Scatterplot of the association between ASI-3-PC residual and reactivity during the U_CD/ISI condition adjusting for reactivity during the N condition (i.e., the average of N_CD and N_ISI). Probe N100 during U (adjusted for probe N100 during N) was attenuated (i.e., less negative values) with greater self-reported ASI-3-PC independent of CC and SC.

For the probe P300, there were no main effects or interactions involving any of the ASI-3 subscales (ps > .211).

3.5. Moderating Effect of AS Independent of PD

To examine whether the relationship between the probe N100 and AS was independent of diagnostic status, we ran a Condition X Cue X PD Status X MDD Status with the subscales of the ASI-3 entered as continuous predictors. The Condition X ASI-3-PC interaction remained significant [F(2, 316) = 6.42, p = .003, η_p² = .039, ε = .91]. Partial correlations examining the unique relationship between the probe N100 and ASI-3-PC independent of the other two ASI-3 subscales, and additionally controlling for diagnostic status, indicated that greater ASI-3-PC continued to be associated with reduced (i.e., more positive) probe N100 in U adjusted for N (r = .21, p = .006), but not probe N100 in P adjusted for N (r = .06, p = .491).

3.6. Concordance between EMG and ERP Threat Responding

Pearson correlations were run to evaluate the relationship between ERP and EMG startle potentation (i.e., using the above approach of regressing out responses during the N condition). The probe N100 was correlated with startle eyeblink responding during both P and U threat, such that a more enhanced (i.e., more negative) probe N100 was associated with greater eyeblink startle; the probe N100 and EMG residualized potentiation scores were also correlated during both P and U, but the association was significantly stronger during U than P (see Table 2). There was no relationship between probe P300 and the magnitude of the EMG startle response.

Table 2.

Correlation between EMG and ERP residualized threat potentiation scores.

Measure	N100 and EMG	P300 and EMG	Fisher’s Z (N100 vs. P300)
PCue	−.234**	−.078	−1.46
UCue	−.275**	−.006	−2.52*
UISI	−.295**	−.015	−2.63**
P Potentiation	−.153*	−.023	−1.20
U Potentiation	−.302***	.016	−2.99**

Note: P Potentiation = residualized P_Cue adjusting for N_Cue; U Potentiation = the average of U_Cue and U_ISI residualized adjusting for the average of N_Cue and N_ISI.

^*p < .05,

^**p < .01,

^***p < .001

4. Discussion

The present study examined ERP indicators of attention-related processes (probe N100 and P300) during the NPU-threat task, and their association with categorical and continuous measures of panic-related phenomenology in a clinical sample. Results indicated that the probe N100 was enhanced (i.e., more negative) during both the predictable and unpredictable threat (compared to no threat) conditions, and the probe P300 was only attenuated during the predictable threat condition. Furthermore, dimensional AS, and not categorical PD diagnosis, was associated with a smaller (i.e., more positive) probe N100 during the unpredictable threat condition.

It is noteworthy that AS, but not PD, was associated with probe N100 in the U condition, and this effect remained independent of PD. AS has historically been considered a trait-like vulnerability for PD, although recent literature has shown that it relates to other internalizing psychopathologies (Schmidt, Zvolensky, & Maner, 2006; Zvielli, Bernstein, & Berenz, 2012). Indeed, AS is a heterogenous construct, and its components differentially predict different psychopathologies (Brown, Meiser-Stedman, Woods, & Lester, 2016; Taylor et al., 2007). Given that the present task utilized electric shocks as the aversive stimulus, it is interesting that the physical-concerns component of AS was the only dimension related to task performance. However, the direction of the relationship between the N100 and AS was opposite of the hypotheses, in that higher AS was associated with reduced (i.e., more positive) rather than enhanced N100 in the unpredictable threat condition. As the probe elicited N100 reduces when individuals deviate their attention (Cuthbert et al., 1998), individuals with high physical-concerns may have been attending to sensations associated with the sustained anxiety elicited by the unpredictable threat condition rather than to the threat of shock itself, thereby reducing N100 reactivity. This interpretation would be consistent with findings that individuals high in AS evidence heightened reactivity in anticipation of interoceptive (e.g., hyperventilation) but not exteroceptive (e.g., shock) threats (Melzig, Michalowski, Holtz, & Hamm, 2008).

In a prior behavioral study, AS-physical concerns was linked to a reaction time measure of attention bias to threat and the findings were specific to physical threat versus other forms of threat (Keogh et al., 2001). The present study extended these results to a neurophysiological measure of cognitive processing and further suggests that this attention-related bias is specific to unpredictable physical threats. Additionally, as the findings were specific to probe N100, it is possible that the reaction time deficits reported by Keogh et al. correspond to deficits in early attentional components subsumed by the probe N100 (i.e., early vigilance), but the more active participation required by such attentional assessments precluded participants from focusing on their internal experience during presentation of the aversive stimuli. Alternatively, ERP indices during anticipation of exteroceptive threat may involve different processes than interoceptive threat.

Our prior study in undergraduates (Nelson, Hodges, et al., 2015) also examined the impact of AS subscales but found slightly different results. Both studies found that ASI-3-PC was associated with altered threat responding, but the prior study found that it related to probe P300 enhancement while the present study found it was related to probe N100 suppression. Our prior study also found effects for cognitive concerns that were not evidenced here. These differences may have been due to the fact that the present sample was a clinical sample while our previous study was not. Indeed, the means and variances for the AS subscales were greater in the present study, perhaps increasing the reliability of effects in the present study.

The present study is consistent with the mission of NIMH’s Research Domain Criteria (RDoC; Cuthbert, 2014; Insel et al., 2010) as it utilized multiple measures of threat responding (EMG startle, probe-elicited ERPs) during the NPU-threat task, the recommended task for eliciting the ‘potential threat’ RDoC construct (NIMH, 2016). It is interesting that individual differences in EMG startle and the N100 (but not P300) were correlated. Since the probe N100 reflects early components of attention/vigilance (Cuthbert et al., 1998), the EMG startle response may be influenced by early aspects of attention and not more elaborative processing that is captured by the probe P300; in other words, both the probe N100 and EMG startle may represent relatively early deployment of defensive responding. Furthermore, as potentiation of EMG startle and probe N100 were more strongly associated for responses to unpredictable threat, and sensitivity to unpredictable threat is central to vulnerability to certain anxiety disorders (Gorka et al., 2017; Grupe & Nitschke, 2013; Nelson et al., 2013), utilizing ERPs and startle together during the NPU-threat task could help identify multiple important aspects of those at risk for anxiety.

The fact that there was an effect for AS but no effect of categorical diagnosis is also consistent with the RDoC framework, which advocates for a more dimensional approach towards characterizing psychopathology. Specifically, variance in underlying mechanisms of psychopathology (such as attentional engagement with threat) may be better captured by transdiagnostic characteristics (such as AS) than DSM disorders. This is also consistent with prior studies that found that symptom dimensions are more strongly associated with neurophysiological variables than categorical diagnoses (MacNamara, Kotov, & Hajcak, 2016; Weinberg, Kotov, & Proudfit, 2015).

The present findings largely replicate the basic effect of the NPU-threat task on ERP indices of cognitive processing found in our prior work in undergraduates (Nelson, Hajcak, et al., 2015; Nelson, Hodges, et al., 2015). Both studies found a similar PCA-derived probe N100 component during the NPU-threat task that was increased during threat. This suggests that the early cognitive processes that are captured by the probe N100 are heightened while under threat (Cuthbert et al., 1998). However, the probe N100 was enhanced to both P and U in the present study but only U in our prior studies (Nelson, Hajcak, et al., 2015; Nelson, Hodges, et al., 2015). This is likely due to the fact that the present study employed a slightly different form of the NPU-threat task than our other studies. Specifically, in the present study, the cue in the P condition was more predictive of threat than the U condition, but participants did not receive a shock during every shape cue; additionally, the cue was on the screen in the P condition for 8 seconds and thus participants did not know exactly when the shock would occur while the shape was on the screen. In contrast, the P condition of the NPU-threat task used by Nelson et al. was fully predictable as (a) participants were shocked during every P trial and (b) the task utilized countdowns rather than shape cues, thereby providing specific information as to when the shock would occur. Thus, the probe N100 enhancement observed during the P condition in the present study might have been due to the fact that there was still some uncertainty in the P condition, rendering it more similar to the U condition.

In contrast, in the present study, probe P300 suppression was specific to the P condition. Threat of shock is particularly salient and requires that cognitive resources are diverted in order to anticipate threats. Although both our present and prior studies (Nelson, Hajcak et al., 2015) found a similar PCA-derived probe P300 component, our prior study found probe P300 suppression in both P and U conditions and not just the P condition. It is unclear why the results of the probe P300 analyses did not fully parallel those of Nelson and colleagues. It is possible that the use of a clinical sample in the present study led to slighly different results. However, we did not observe any effects for PD, MDD, or AS on the probe P300, so further studies are needed to examine the impact of unpredictable and predictable threat on probe P300 responses.

Despite several notable strengths (examining the effects of PD independent of MDD in a clinical sample; using PCA-derived ERP components, etc), these results should be interpreted in light of several limitations. First, the number of participants in the PD-only group was small, which may have precluded finding PD X MDD interactive effects. Second, as previously discussed, the predictable threat condition in this version of the NPU-threat task was more predictable than was the unpredictable threat condition, but was not fully predictable. Additionally, characteristics of the shape stimuli signaling the conditions may have influenced ERP responding (Luck, 2014). Replication of these findings using different versions of the NPU-threat task (e.g., Nelson, Hodges, et al., 2015) would strengthen the conclusions drawn from this study. Third, the fact that individuals with PD were allowed to have other anxiety disorders may have reduced the likelihood of finding effects for PD (although it increased the generalizability of the results; Craske et al., 2009). However, the question remains whether these effects might generalize to individuals with non-PD anxiety pathology. Finally, because the NPU-threat task did not include an attention manipulation (e.g., attend versus distract), it is difficult to draw definitive conclusions regarding whether the N100 specifically reflects attention to threat. However, given that the N100 has been found in other studies to be modulated by attentional manipulations (e.g., Cuthbert et al., 1998), this appears to be a plausible interpretation.

In conclusion, this study found that the physical concerns component of AS was associated with reduced probe N100 to unpredictable threat independent of response to no threat and PD (and MDD) diagnoses. Additionally, threat responding indexed by the probe N100 was associated with threat responding indexed by EMG startle, particularly during the unpredictable threat condition. These results support the use of the NPU-threat task in examining neural indicators of threat responding (particularly to assess early cognitive processing of threat) and highlight the utility of dimensional, transdiagnostic constructs in the investigation of mechanisms integral to PD-related psychopathology.

Highlights.

• Panic disorder entails heightened sensitivity and responding to unpredictable threat

• Event-related potentials (e.g., probe N100 and P300) can index attention to threat

• Examined in predictable and unpredictable threat in panic and anxiety sensitivity

• N100 was enhanced during threat and P300 was suppressed during predictable threat

• N100 was associated with anxiety sensitivity, not panic, as well as with startle

Abbreviations

PD

panic disorder

ERP

event-related potential

MDD

major depressive disorder

NPU

No-Predictable-Unpredictable

AS

anxiety sensitivity

PC

physical concerns

CC

cognitive concerns

SC

social concerns

PCA

principal components analysis

EMG

electromyography

EEG

electroencephalography

ANOVA

analysis of variance

ANCOVA

analysis of covariance