Adversity in early and mid-adolescence is associated with elevated startle responses to safety cues in late adolescence

Abstract

Elevated responding to safety cues in the context of threat is associated with anxiety disorder onset, but pathways underlying such responding remain unclear. This study examined whether childhood/adolescent adversity was associated with larger startle reflexes during safe phases of a fear potentiation startle paradigm (following delivery of an aversive stimulus) that predict anxiety disorders. Participants (N = 104) came from the Youth Emotion Project, a longitudinal study of risk factors for emotional disorders. Participants with no baseline psychopathology underwent a startle modulation protocol and were assessed for childhood and adolescent adversities using a validated interview. Adolescent adversity was associated with larger startle reflexes during the safe phases following an aversive stimulus. Neither child nor adolescent adversities were associated with responding during any other phase of the protocol. These findings suggest a pathway between adolescent adversity and a risk factor for anxiety disorders wherein adolescent adversity contributes to impaired responding to safety cues.

Several studies suggest that individuals with anxiety disorders show a failure to inhibit fear responding to cues that signal safety in a context of threat. For example, they typically show elevated startle eye blink responses to cues that are never paired with an unconditional aversive stimulus (i.e., CS-) within differential conditioning paradigms (see Lissek et al., 2005, and Craske, Rauch et al., 2009 for reviews). In addition, they show elevated responses throughout extinction and at extinction recall to cues that used to signal the aversive stimulus but now signal safety. More direct evidence for deficits in safety learning comes from conditional discrimination paradigms (AX+/BX-) (Jovanovic, Norrholm et al., 2009a), where individuals with anxiety (PTSD) show deficits in the acquisition and transfer of safety. In other words, when the danger cue (A, which was previously paired with the US) and safety signal (B, which was previously paired with no US) are paired together (AB), B does not inhibit responding to A in individuals with PTSD to the same extent as in those without PTSD (i.e., those with PTSD show slower fear extinction).

Our earlier work established that elevated fear responding to the CS+ and CS- during extinction and at extinction recall was not only characteristic of youths with anxiety disorders but also characteristic of youths at risk for anxiety disorders¹ (Craske, Waters et al., 2008). In addition, in a related fear potentiated startle protocol, we found that larger responses to safe phases within a series of alternating danger (i.e., potential for shock) and safe (i.e., no potential for shock) phases were associated with high levels of neuroticism (Craske, Waters et al., 2009), a risk factor for anxiety disorders (e.g., Krueger, Caspi, Moffitt, Silva, & McGee, 1996). Furthermore, over-responding to safe phases was a prospective predictor of the onset of anxiety disorders over three years of follow-up (Craske, Wolitzky-Taylor et al., 2011). These studies suggest that failure to attenuate fear responding to safe cues in a context of threat is a risk factor for anxiety disorders.

Elevated fear responding to safe cues may serve to enhance the generalization and persistence of fear responding following discrete negative life events, and thereby contribute to excessive and persistent anxiety. As an example, a child who is bullied on the playground and is prone to anxiety may then show excessive fear of other, friendly children on the playground. Other than neuroticism (Craske, Waters, et al., 2009), other factors that underlie elevated fear responding to safe cues have not yet been studied. Given that elevated responding during safe phases is a risk factor for anxiety disorders (Craske, Wolitzky-Taylor et al., 2011), understanding contributory factors to elevated fear responding to safety may help to define the pathways through which anxiety disorders develop.

One possibility is that early life adversities result in elevated fear responding to safe cues which in turn then increases risk for anxiety disorders. Indeed, various types of childhood adversity have been associated with several anxiety disorders (e.g., Kilpatrick et al., 2003), including physical abuse (e.g., Macmillian et al., 2010), emotional abuse and neglect (Gibb et al., 2006; Lochner et al., 2002), and sexual abuse (Cougle et al., 2010; Kendler et al., 2000). However, the mechanisms contributing to the development of these associations are not well understood. Conceivably, early adversity may be associated with increased risk of anxiety disorders because adversity elevates fear responding to safety in the context of threat. The viability of this model rests upon first demonstrating an association between childhood adversity and elevated fear responding to safety.

A small body of research shows an association between childhood trauma and larger eye blink reflexes to startling sounds (Jovanovic, Blanding et al., 2009b; Pole et al., 2007). However, few studies have examined the link between early adversity and responding to safety versus threat, and those that have do not indicate the expected link specifically between adversity and elevated responding during safe phases following an aversive stimulus. Instead, findings have been mixed. For example,an animal study found early life adversity to be associated with elevated startle reflexes across all phases of a fear conditioning paradigm (with the finding driven by startle during threat signals, or the CS+) (Nelson et al., 2009); and studies with humans have observed larger startle reflexes in a child abuse group compared to no child abuse group only during the baseline phase of a conditional discrimination paradigm in one study (Jovanovic, Blanding et al., 2009b) and during the safe as well as the danger phases of a fear potentiated startle paradigm in another study (Pole et al., 2007).²

Taken together, these studies demonstrate that early adversity is associated with elevated fear responding in general or to threat cues but not specifically to safe cues. However, these studies are limited in ways that may have mitigated their ability to detect the effects of life adversity upon fear responding specifically to safe cues. First, some either failed to take into account anxiety symptoms that may have already been present prior to the fear conditioning paradigm (Nelson et al., 2009) or included participants with particular anxiety disorders such as PTSD (Jovanovic et al, 2009b). Studies of non- disordered participants who experienced childhood adversity are needed to parse out the effects of the disorder itself. Second, some studies relied on a self-report questionnaire to assess for childhood abuse (Jovanovic et al, 2009b; Pole et al., 2007), which has lower reliability than interview measures of childhood abuse (Dohrenwend, 2006; Monroe, 2008). Third, only one study (Jovanovic et al., 2009b) assessed for different types of child abuse, and none considered the influence of the developmental period during the abuse (i.e., examining the effects of child and adolescent adversities separately). Finally, none of these studies examined eyeblink startle response in safe and danger phases separately for the phases before and after an aversive stimulus. This distinction may be important given our previous work showing that startle response specifically during the safe phase following an aversive stimulus predicted anxiety disorder onset (as opposed to safe phases in general, or safe phases before the aversive stimulus) (Craske et al., 2011).

Because our previous work found that elevated startle reflexes during safe phases after delivery of an aversive stimulus were uniquely associated with anxiety disorder onset (Craske, Wolitzky-Taylor et al., 2011), and because of evidence that childhood adversity is associated with anxiety disorder onset (e.g., Cougle et al., 2010), we aimed to examine whether childhood and adolescent adversity is associated with elevated fear responding to safe phases in our fear potentiation startle protocol. Participants in the current study were part of the Youth Emotion Project (YEP), a longitudinal study examining common and specific risk factors for emotional disorders. Participants with no Axis I psychiatric disorders at baseline underwent a fear potentiated startle protocol and completed a semi-structured interview assessing for the presence and severity of several types of adversity in childhood and adolescence.

Driven by our previous findings, we hypothesized that childhood and adolescent adversity would specifically be associated with elevated startle reflexes during the safe phases that predicted anxiety disorder onset relative to other phases of the startle protocol (i.e., baseline, context, and danger phases) that did not predict anxiety disorder onset. These findings would be consistent with a safety learning deficit hypothesis. A competing hypothesis, termed generalized defensive responding, derives from the small body of literature suggesting that those who experienced early life adversity may show elevated responding in general (i.e., baseline startle; Jovanovic et al., 2009b) or during threat/danger phases only (Nelson et al., 2009). Thus, we examined child and adolescent adversity in relation to each phase of our fear potentiation protocol in order to fully examine the specific associations between child adversity and elevated responding.

Methods

Participants

Our data derive from a larger prospective study, the YEP. Participants were high school juniors recruited over three years from one school in suburban Los Angeles, California and another in suburban Chicago, Illinois. Details of the selection procedure, which included over-sampling for high neuroticism, are provided elsewhere (Zinbarg et al., 2010). Of the invited participants, 627 completed the baseline assessment, which included the Structured Clinical Interview for DSM-IV, non-patient edition (SCID-I/NP; First et al. 2002) and several questionnaires.

Of these participants, 127 completed the startle protocol and the Childhood Trauma Interview. These 127 participants were part of the larger sample of participants whose data were included in two of our previous papers reporting findings on the startle protocol (Craske, Waters et al., 2009; Craske, Wolitzky-Taylor et al., 2011), and were part of the larger sample of participants whose data were included in a manuscript reporting findings from the Childhood Trauma Interview (Vrshek-Schallhorn et al., under review). The sample was approximately two thirds (65.4%) female. About half of participants (48.0%) identified themselves as Caucasian, 18.9% Latino, 7.1% African American, 5.5% Asian, 5.5% other race/ethnicity, 1.6% Pacific Islander, and 13.4% as having more than one race or ethnicity. At the time of baseline assessment, the sample ranged in age from 15 to 17 years (M = 16.39, SD = 0.51). At the time the startle protocol was completed, the sample ranged in age from 16 to 18 years (M = 16.99, SD = 0.43).

The startle modulation protocol was administered during the first 18 months of the study (1-12 months after the baseline diagnostic assessment). The Childhood Trauma Interview (CTI) was administered between four to six years after the baseline assessment. In order to examine the effects of childhood and adolescent adversity on startle reactivity without the confound of emotional disorders, only participants with no current, clinically significant Axis I psychiatric disorder at the baseline assessment were included in the analyses, thus excluding 23 of the 127 participants.³

Measures

Structured Clinical Interview for DSM-IV, non-patient edition (SCID-I/NP; First, Spitzer, Gibbon, & Williams, 2002)

Participants were assessed at baseline for DSM-IV psychiatric diagnoses using the SCID. Interviewers had at least a bachelor's degree and received extensive training and supervision. Each completed SCID was presented at a diagnostic consensus meeting led by a doctoral-level supervisor.

Childhood Trauma Interview (CTI; Fink, 1995)

The CTI is a semi-structured interview that positively compares positively to similar measures (Roy & Perry, 2004), and that was used here to retrospectively assess childhood and adolescent adversity. Interviews were completed over the phone. Interviewers queried participants about six domains of concrete and behaviorally defined adversity through the age of 16 years old including separations from or loss of a caregiver, caregiver neglect, emotional abuse, physical abuse, witnessing violence, and sexual abuse and assault. For each adversity endorsed, interviewers rated severity based on more than 260 coding examples in an interview manual, ranging from 1 (minimal or mild) to 6 (very extreme, sadistic). Thus, the CTI is designed to elicit reports of a full range of severity of adversities, not only trauma, as the name might suggest. In previous reports, the CTI had high inter-rater reliability as well as better convergent and discriminant validity than questionnaires about early life adversity.It also had good predictive validity in its associations with symptoms of dissociation, depression, and PTSD (Brown et al., 2005; Fink, 1996; Simeon et al., 2001). Its practice of eliciting concrete behavioral information rather than judgment or interpretation has been associated with reasonable validity of retrospective reports; positive reports are generally accurate, although underreporting of severe abuse is known to occur with retrospective methods (Hardt & Rutter, 2004).

Due to a significant psychometric flaw in a previously-applied scoring system (i.e., multiplying ordinal scales), a novel scoring system was applied in which the sum of adversity severity scores was calculated for each domain of adversity (using CTI data from the YEP; Vrshek-Schallhorn et al., under review). For example, if a participant experienced three physical abuse events with severity scores of two, three, and four, respectively, the score for physical abuse for that individual would be nine. This yielded dimensional measures, which tend to provide greater power in comparison to dichotomous measures (e.g., whether a participant did or did not experience a particular type or severity of adversity). Additionally, in order to permit separate analyses for adversities reported to have occurred relatively early versus later in development, adversities were classified as occurring during early to middle childhood (age 0 to 9 years) or pre-adolescence to adolescence (age 9 to 16 years). These ages were chosen because there is evidence of prepubertal gonadal hormone changes beginning at age nine, which might influence sensitivity to adversity (e.g., Romeo, 2010). Within each domain, patterns of adversity were scored separately if they differed in several defining characteristics—perpetrator, duration, frequency or severity. Taken together, in addition to yielding summed severity scores in both age ranges for each of the six domains of adversity (e.g., witnessing violence in childhood), a total summed score across all six domains was calculated for each age range (i.e., adversities during childhood and adversities during adolescence).

Startle Modulation Protocol

The startle protocol included an initial baseline condition, application of shock electrode condition (i.e., context), and a series of alternating safe and danger phases, during which an aversive electrical biceps muscle contraction was threatened to occur during the final 15 sec of danger phases only. Baseline and context conditions were then repeated. Auditory startle stimuli were presented throughout. Contextual modulation was defined as the startle response during the electrode application phase. Explicit threat cue modulation was assessed during danger phases relative to safe phases.

Baseline and Context Conditions

For the first baseline condition, eight startle stimuli were delivered with a mean inter-stimulus interval (ISI) of 22 s, while participants focused on a white fixation cross. The first context condition followed next, during which participants were fitted with two muscle contraction electrodes to the biceps muscle, and told that instructions would be given when contractions would happen. Eight startle stimuli were presented while Ps focused on the fixation cross.

Fear Potentiation Condition

Before initiating the safe-danger phases, participants were told that no contractions would be delivered while the words ‘Safe: no contraction will be given’ were on the green screen, and that they might receive a contraction when the words ‘Danger: contraction might be given’ were on the red screen. There were eight safe and eight danger phases in alternating order, always commencing with a safe phase. Participants were also told that the progressing bar on each screen would count down from 0 to 55 sec, and if a contraction were to occur in the danger phase, it would happen in the last 15 sec as the bar turned from pink to red in color. Finally, they were told they might receive a contraction up to three times, of increasing intensity each time. All participants received only one contraction in the final 15 sec of the fourth danger phase. Sixteen startle probes were presented in each phase, with two trials per phase, at 5 and 35 or 15 and 45 sec, resulting in 32 startle probes during the safe-threat phases.

After the fear potentiation condition, the muscle contraction electrodes were removed for the second baseline condition. They were reattached for the second context condition. For more methodological detail of the startle paradigm, see Craske et al. (2009) and Craske et al. (2011), which both report findings using the YEP sample.

Electrophysiological materials and equipment

Auditory startle stimuli (105 dB, zero rise time, 50 ms white noise bursts) were presented binaurally through stereophonic headphones (Sony, Model MDRV700). The amplitude and latency of startle responses (SRs) were measured by electromyogram (EMG) activity of the orbicularis oculi. EMG was recorded from electrodes placed beneath the right eye approximately 10 mm apart, edge to edge, and 9 to 11 mm below the lower lid margin. The lateral electrode was placed 5 mm medial to the outer canthus. A vertical electro-oculogram was recorded from electrodes above and below the left eye to facilitate recognition of spontaneous blinks and eye movements. The impedance level of electrodes was 15 KOhm or less. EMG was amplified by 10,000 with low and high frequency cut-off values of 30 Hz and 1000 Hz, digitized at 1000 Hz, full-wave rectified after analogue to digital conversion, and smoothed with a 2 ms moving average filter.

EMG magnitudes were expressed as the difference between the mean amplitude of the 200ms of EMG preceding the startle stimulus and the peak response (between response onset and 104 ms following the startle stimulus). Response onset was defined as the first EMG increment between 20 and 80 ms (response onset window) following startle stimulus onset that exceeded 2 standard deviations above the mean of baseline and did not drop below that level for more than 10 ms. Given the highly skewed nature of startle EMG (Yamada, et al., 1980), analyses were performed on natural log (ln) transformed eyeblink data.

The muscle contraction, delivered by a Digital 807 Electrical Muscle Stimulation Device (Everyway Medical Instruments), was a 20.4 mA peak current (i.e., equating to 50 V peak) for .5 sec. The experience of the contraction is one of a very rapid onset, uncomfortable muscle contraction across the biceps for .75 sec. For more details, see Craske, Waters et al., 2009 and Craske, Wolitzky-Taylor et al., 2011.

Procedure

After signing informed consent, participants completed a baseline diagnostic assessment followed by the startle protocol (1-12 months after study enrollment). The startle protocol was completed by 168 participants, although only a subset of those completed the CTI as well (N=127; N=74 from UCLA and N=53 from NU). The two laboratories used identical hardware, software, manualized procedures, and technician training procedures. Participants were seated upright in a sound attenuated room adjacent to the experimental room, interconnected via intercom and closed-circuit cameras from two angles (UCLA) or one-way mirrors (NU). Participants were instructed to sit quietly and as still as possible throughout the protocol. Three to six years (depending on cohort) after baseline assessment (i.e., in the sixth through seventh year of YEP data collection), 127 of the participants who had undergone the startle protocol completed a CTI interview over the phone.

Statistical Analysis

The analyses included participants who completed the startle protocol and the CTI and who did not meet the diagnostic criteria for any DSM-IV-TR (APA, 2004) Axis I psychiatric disorder at the baseline assessment (N = 104). A series of hierarchical linear regressions were conducted in SPSS. Each regression examined whether the sum of early life adversity severity scores across all domains for a given developmental period (i.e., childhood or adolescence) predicted startle response indices.

Analyses of the safe and danger phase variables included the mean EMG startle response (SR) during the eight trials preceding the safe-danger sequences as a covariate in the first block (i.e., context SR) to account for individual differences in responding before the fear potentiation portion of the startle protocol. Predictors in the second block included the sum of severity scores across domains of adversity during childhood and adolescence (childhood and adolescence scores entered in two separate analyses). The dependent variables included (a) safe phase SR before the muscle contraction, (b) safe phase SR after the muscle contraction, (c) danger phase SR before the muscle contraction, and (d) danger phase SR after the muscle contraction. Startle response for each of these phases was identical to those examined as predictors of anxiety disorder onset in this sample (Craske et al., 2011). For analyses predicting SR during the safe and danger phases after the administration of the muscle contraction, the SR during those phases before the contraction were also entered into the first block as covariates (e.g., when predicting SR during the safe phase post-contraction, SR during the safe phase pre-contraction was entered as a covariate). Adjusted R² is presented as an index of effect size when the adversity predictor variable of interest was entered alone in the model (e.g., childhood adversity predicting SR during baseline pre), and R² Δ is reported as an index of the effect size when the adversity predictor variable of interest is added after covarying for other variables (e.g., childhood adversity predicting baseline post, since baseline pre is entered into the first block of the model).

As a secondary analysis, we separately examined child and adolescent adversities as predictors of the two baseline phases (before and after the safe/danger phases) and the two context phases (before and after the safe/danger phases). When predicting the baseline and context phase SR after the safe/danger phases, the respective baseline and context phase SRs prior to the safe/danger phases were entered as covariates. For example, in testing whether childhood adversities predicted responding during the second context phase (that occurred after the safe/danger phases), the SR during the context phase prior to the safe/danger phases was entered in the first block as a covariate.

When child or adolescent adversity was associated with a startle variable, an additional analysis was conducted including all six types of adversity (i.e., separation/loss, neglect, emotional abuse, witnessing violence, physical abuse, and sexual abuse) entered simultaneously into the model as predictors in order to examine which, if any, types of adversity were uniquely associated with that startle variable. To be conservative, all analyses were run a second time covarying for gender and site in order to account for variance that may have been explained by these variables. We report these findings only when they differed from the main analyses.

Results

Table 1a presents the means and standard deviations (SDs) for SR in each of the phases of the startle protocol, and Table 1b reports the means and SDs for each domain of adversity separated by developmental period.

Table 1a.

Means (SDs) of SR magnitudes (ln-transformed μV) for all phases

Phase	M (SD)
Pre Baseline	4.49 (0.79)
Post Baseline	3.53 (1.01)
Pre Context	4.25 (0.93)
Post Context	3.75 (1.00)
Safe Pre Muscle Contraction	4.06 (0.85)
Safe Post Muscle Contraction	3.76 (0.93)
Danger Pre Muscle Contraction	4.29 (0.81)
Danger Post Muscle Contraction	4.12 (0.85)

Table 1b.

Mean and SD sum of severity scores across domains of adversity

Adversity Domain	Mean (SD)
	Child	Adolescent
Separation/loss	2.19 (3.40)	2.65 (3.10)
Neglect	1.15 (2.87)	6.11 (4.49)
Emotional abuse	1.87 (1.92)	4.16 (4.03)
Witnessing violence	1.02 (2.29)	1.25 (2.08)
Physical abuse	2.22 (3.14)	1.96 (2.89)
Sexual abuse	0.25 (0.99)	0.26 (1.40)
Across domains	8.69 (9.24)	16.38 (10.83)

Manipulation Check

In this sub-sample, startle response (SR) was significantly higher in the danger phases compared to the safe phases, t (206) = −2.71, p < .01. SR did not significantly differ between the baseline and context phases, t (206) = −0.02, p = .98, or between SR during the fear potentiation paradigm and the other (baseline and context) phases, t (206) = 0.44, p = .66. For more details with the larger sample, see Craske, Wolitzky-Taylor et al., 2011.

Analyses of Associations between Adversity and Startle Variables

Tables 2 and 3 report the statistics for each of the final models predicting safe and danger phase SR variables with childhood adversity (Table 2) and adolescent adversity (Table 3) as the predictors of interest. Tables 4 and 5 report the statistics for the final models predicting baseline and context phase SR variables with childhood adversity (Table 4) and adolescent adversity (Table 5) as the predictors of interest.

Table 2.

Child adversity predicting safe and danger phases (final models)

	B	SE	β	t
Model 1: DV Safe pre contraction
Pre context	.81	.04	.89	19.92***
Child adversity (sum severity)	−.004	.004	−.05	−1.06
Model 2: DV Safe post contraction
Pre context	.27	.12	.27	2.34*
Safe pre contraction	.67	.13	.61	5.22***
Child adversity (sum severity)	.01	.01	.05	1.02
Model 3: DV: Danger pre contraction
Pre context	.77	.04	.88	18.77***
Child adversity (sum severity)	−.002	.004	−.02	−0.43
Model 4: DV: Danger post contraction
Pre context	.23	.10	.25	2.34*
Danger pre contraction	.67	.11	.64	5.94***
Child adversity (sum severity)	.003	.05	.03	0.67

^*p < .05

^***p < .001

b= unstandardized beta; β= standardized beta

Table 3.

Adolescent adversity predicting safe and danger phases (final models)

	B	SE	β	t
Model 1: DV Safe pre contraction
Pre context	.82	.04	.90	19.94***
Adolescent adversity (sum severity)	.000	.004	.001	0.03
Model 2: DV Safe post contraction
Pre context	.27	.11	.27	2.38*
Safe pre contraction	.66	.13	.60	5.25***
Adolescent adversity (sum severity)	.01	.004	.12	2.36*
Model 3: DV Danger pre contraction
Pre context	.77	.04	.88	18.77***
Adolescent adversity	.003	.004	.04	0.78
Model 4: DV: Danger post contraction
Pre context	.23	.10	.25	2.31*
Danger pre contraction	.67	.11	.64	5.95***
Adolescent adversity	−.003	.004	−.04	−0.72

^*p < .05

^***p < .001

b=unstandardized beta; β = standardized beta

Table 4.

Child adversity predicting baseline and context phases (final models)

	B	SE	β	t
Model 1: DV Pre baseline
Child adversity (sum severity)	−.01	.01	−.17	−1.67
Model 2: DV Post baseline
Pre baseline	1.05	.08	.81	13.59***
Child adversity (sum severity)	.004	.04		0.59
Model 3: DV: Pre context
Child adversity	−.01	.01	−.09	−0.91
Model 4: DV: Post context
Pre context	.91	.06	.85	15.79***
Child adversity	.000	.01	−.003	−0.07

^***p < .001

b=unstandardized beta; β = standardized beta

Table 5.

Adolescent adversity predicting baseline and context phases (final models)

	B	SE	β	t
Model 1: DV Pre baseline
Adolescent adversity (sum severity)	.004	.01	.06	.57
Model 2: DV Post baseline
Pre baseline	1.04	.08	.81	13.63***
Adolescent adversity (sum severity)	.004	.05	.05	0.76
Model 3: DV: Pre context
Adolescent adversity	.01	.01	.06	.63
Model 4: DV: Post context
Pre context	.91	.06	.85	15.84***
Adolescent adversity	.00	.01	.01	0.09

^***p < .001

b=unstandardized beta; β = standardized beta

Associations between childhood and adolescent adversity and SR during safe phases

In order to test our safety learning deficit hypothesis, we first examined whether childhood and adolescent adversities were associated with SR during the safe phases of the startle protocol. The sum of severity scores across domains of childhood adversity was not significantly associated with SR during the safe phases before or after the muscle contraction (R²Δs = .02 and .03 for safe phases pre- and post-muscle contraction, respectively, ps > .17). The sum of severity scores across domains of adversity during adolescence was not associated with SR during the safe phase before the muscle contraction (R²Δ = .00, p = .72), but was significantly associated with SR during the safe phase after the muscle contraction, b = .01 (standardized β = .12), t (101) = 2.36, p = .02, Adjusted R² = .74, R²Δ = .014, F (1, 98) = 5.57, p = .02, such that higher adolescent life adversity severity scores were associated with higher SR during this phase. Consistent with hypothesis, adolescent adversity was significantly associated with elevated SR during the safe phase after the aversive stimulus, the variable that predicted anxiety disorder onset in this sample (Craske, Wolitzky-Taylor et al., 2011). However, inconsistent with hypothesis, childhood adversities were not associated with SR during the safe phase after the aversive stimulus.

A series of additional analyses were conducted to examine whether this effect was unique to the safe phase following the muscle contraction and unique to adolescent adversity. First, an analysis was conducted covarying for SR during the danger phases before and after the muscle contraction in order to examine whether the effect remained significant after accounting for the variance explained by the SR during the danger phases. Adolescent adversity remained significantly associated with SR during the safe phase following the muscle contraction, b = .01 (standardized β = .11), t (101) = 2.22, p = .03, Adjusted R² = .75, R²Δ = .012, F (1, 98) = 4.92, p = .03. Next, an analysis was conducted to examine whether the effect was specific to adolescent adversity by covarying for childhood adversity. Adolescent adversity remained significantly associated with the SR during the safe phase following the muscle contraction after accounting for the variance explained by childhood adversity, b = .01 (standardized β = .14), t (101) = 2.17, p = .03, Adjusted R² = .74, R²Δ = .012, F (1, 97) = 4.69, p = .03. Finally, an additional analysis covaried for site and gender in order to account for these individual differences. The effect of adolescent adversity on the safe phase SR following the muscle contraction was similar in magnitude to the previous analyses, but did not attain statistical significance, b = .01 (standardized β = .13), t (101) = 1.94, p = .056, Adjusted R² = .75, R²Δ = .01, F (1, 95) = 3.74, p = .056. Because the effect was virtually identical in magnitude to the others in this series of analyses, we presumed that the addition of two covariates may have resulted in insufficient power to detect a statistically significant effect. Taken together, these findings indicate that adolescent adversity is uniquely associated with safe phase SR after the delivery of the muscle contraction.

Because adolescent adversity in general was associated with elevated SR during the safe phase following the aversive stimulus, an additional analysis was conducted including all of the domains of adversity entered simultaneously as predictors in order to examine whether any specific domains uniquely contributed to the model. Only neglect uniquely contributed to the model, [b = .03 (standardized β = .14), t (101) = 2.57, p = .012, Adjusted R² = .75; Final model including all domains: R²Δ = .03, F (6, 93) = 2.11, p = .059, semi-partial correlation for neglect = .13], with higher sum of severity scores for the neglect domain associated with higher SR during the safe phase of the startle protocol that followed the muscle contraction. The finding remained significant after accounting for childhood adversity (across domains) [b = .01 (standardized β = .11), t (101) = 2.22, p = .03, Adjusted R² = .74; Final model including all domains: R²Δ = .03, F (6, 92) = 1.91, p = .088, semi-partial correlation for neglect = .13]. The finding also remained significant after accounting for SR during the threat phases before the muscle contraction [b = .03 (standardized β = .13), t (101) = 2.57, p = .012, Adjusted R² = .75, Final model including all domains: R²Δ = .03, F (6, 92) = 1.91, p = .087, semi-partial correlation for neglect = .13]. This finding remained nearly identical when covarying for gender and site, although the effect of neglect only approached statistical significance [b = .02 (standardized β = .11), t (101) = 1.97, p = .052, Adjusted R² = .75, R²Δ = .03, F (6, 91) = 1.80, p = .108, semi-partial correlation for neglect = .10]. The effect sizes stayed consistent across these analyses, suggesting the analysis covarying for additional individual difference variables (i.e., site and gender) may have resulted in insufficient power to detect a statistically significant difference. Taken together, these findings suggest that neglect may be partly driving the association between adolescent adversity and deficits in responding to safety cues. Importantly, neglect accounted for approximately 56% of the variance due to the predictors on this step, indicating that a substantial amount of the variance in safe phase SR following the muscle contraction was accounted for by shared features across the abuse and neglect variables entered on that step (44%).

Associations between childhood and adolescent adversity and SR during danger, baseline, and context phases

In order to test the generalized defensive responding hypothesis, we examined whether childhood and adolescent adversities were associated with the other phases of the startle protocol. The sum of severity scores across domains of childhood adversity was not significantly associated with SR during the danger phases before or after the muscle contraction (R²Δs = .00 and .001 for danger pre- and danger post-muscle contraction, respectively, ps > .54). Similarly, the sum of severity scores across domains of adolescent adversity was not significantly associated with SR during the danger phases before or after the muscle contraction (R²Δs = .001 for both danger pre- and danger post-muscle contraction, ps > .40).

The sum of severity scores across domains of adversity during childhood was neither associated with the baseline SR prior to the fear potentiation portion of the paradigm nor after the fear potentiation protocol (Adjusted R²s = .02 for baseline pre and R²Δ = .001 for baseline post, ps > .32). Nor were childhood adversity scores associated with SR during either of the context phases (prior to, or after the fear potentiation protocol; Adjusted R²s = −.002 for context pre and R²Δ = .00 for context post all ps > .36). The sum of severity scores across domains of adversity during adolescence was neither associated with the baseline SR prior to the fear potentiation portion of the paradigm nor after the fear potentiation protocol (Adjusted R²s = −.01 for baseline pre and R²Δ = .002 for baseline post ps > .45). Nor were adolescent adversity scores associated with SR during either of the context phases (prior to, or after the fear potentiation protocol; Adjusted R²s = -.01 for context pre and R²Δ = .00 for context post, all ps > .53). Thus, these data did not support the generalized defensive responding hypothesis that adversities would be associated with elevated SR in general or in phases other than the safe phases.

Discussion

The safety learning deficit hypothesis was partially supported: adolescent adversity was significantly associated with larger startle reflexes during safe phases following the aversive stimulus and not during other phases. The results did not extend to childhood adversity. The current study indicates that relatively recent life stressors (i.e., adolescent adversities, which could have occurred up to two years before the startle protocol, but as long as eight years before the startle protocol) were associated with elevated responding during safe phases following the aversive stimulus, whereas these data did not yield a significant association with more earlier childhood adverse experiences. Further, the effect of adolescent adversity on safe phase startle responding remained statistically significant after accounting for the variance explained by childhood adversity, providing support for the specificity of adolescent adversity as a predictor of startle response during the safe phase following the aversive stimulus. Conceivably, with the passage of time since earlier childhood adversity, and in the absence of subsequent adversity, other protective factors may intervene and buffer the effects of that earlier adversity. Childhood adversity might thus be associated with elevated startle response during safe phases if testing was conducted closer to the timing of the adversity (e.g., during early to middle adolescence).

It is important to note that previous studies examining the relationship between childhood trauma and fear potentiated startle responding (e.g., Jovanovic et al, 2009b) or anxiety disorder onset (e.g., Cougle et al., 2010) often used questionnaires and interviews that included adolescent experiences but categorized all adversities before age 18 under the “childhood trauma or adversity” umbrella. Thus, it is possible that previous findings were driven by adolescent adversities that were not examined separately. It is also possible that advancing pubertal status might differentially moderate the relationship between childhood versus adolescent adverse events and subsequent startle reactivity. Indeed, there is some evidence that advancing pubertal status has effects on the psychophysiology of defensive and appetitive responding (Quevedo et al., 2009), as well as interactions with increasing stress and emotional symptoms (Patton et al., 2008; Conley & Rudolph, 2009). Another possibility is that low base rates (resulting in data skew) for several childhood adversities (see Table 1b) may have resulted in insufficient power to detect effects of childhood adversity on responding to safety cues. Although this is a problem likely to occur in any examination of childhood adversity in a normal, non-disordered sample, it nonetheless should be considered as a viable explanation for the null findings.

Importantly, the prediction of startle response was unique to the phase of the paradigm that has been shown to predict anxiety disorder onset (Craske et al., 2011). This finding provides a direction for future research in identifying mediational pathways between the experience of adversity and anxiety disorders. Further, the association between adolescent adversity and startle response during the safe phases following the aversive stimulus remained significant after covarying for startle response during the danger phases. The competing generalized defensive responding hypothesis that elevated fear responding in general or to threat cues in particular would relate to early life adversity was not supported. Thus, the current results were not in accord with previous work examining the associations between startle reflex responses and childhood trauma (i.e., Jovanovic et al, 2009b; Nelson et al., 2009; Pole et al., 2007).

More work with consistent methodology across studies is needed to clarify discrepancies in this small body of literature. For example, separating indices of startle response before and after an aversive stimulus proved to be an important distinction that was not made in previous work. The current study attempted to improve upon previous methodology by (a) using a validated, semi-structured interview that assessed for a number of different types of adversities (summed scores represented experiences across six domains of adversity) and a continuous severity rating scale, (b) including only participants without any Axis I psychopathology to avoid potential confounds, and (c) examining two developmental periods separately. The current findings indicate that adverse experiences are specifically associated with elevated fear responding to safe cues following an aversive stimulus. Thus, future research should explore mediational models to examine whether this pattern of elevated responding to safety after an aversive stimulus mediates the association between adolescent adversity and anxiety disorder onset.

The current data indicate that neglect may uniquely contribute to elevated fear responding to safety cues. On the other hand, the number of predictors in the model for a sample of this size may have resulted in an underpowered test, particularly for adversity domains that were rarely endorsed, such as sexual abuse, physical abuse, and witnessing violence. Nonetheless, neglect is an understudied area of child and adolescent adversity relative to other types of adversity such as physical and sexual abuse and the findings point to the value of further investigation. At present, these data suggest that neglect experiences are associated with elevated fear responding to safety cues that later contribute to anxiety disorders, but why this finding emerged for neglect in particular, above and beyond other types of adversity, is unclear.

It is also worth reiterating that a substantial proportion of the variance in the model including all six domains of adversity was shared across the predictors, suggesting that common features that cut across adversities explain deficits in responding to safety cues. One possible area for exploration is the role of the perpetrator. Specifically, future research might examine whether the familiarity of the perpetrator (e.g., caregiver v. noncaregiver) moderates these effects. Adverse events perpetrated by those with whom there is an expectation of safety may be perceived by the child or adolescent as more uncontrollable and more unpredictable, both of which have been theorized to play a role in the development of anxiety and fear (Foa, Zinbarg & Rothbaum, 1992; Mineka & Kelly, 1989). If impairments in safety responding are most prominent in individuals whose adverse experiences were perpetrated by caregivers, this would lend support to the idea that violation of safety expectations play an especially significant role in subsequent deficits in responding to safety cues. This theory is consistent with previous work showing that close relationships between the perpetrator and victim are more often associated with PTSD following childhood sexual abuse (e.g., McLeer, Deblinger, Atkins, Foa, & Ralphe, 1988; Ullman, 2007).

The present study indicates several future directions for research on the role of early life adversity in anxiety. Whether elevated fear responding to safe cues mediates the relationship between adolescent adversity and the onset of anxiety disorders is a key test, but one the present sample is unfortunately underpowered to address. Clearly, larger sample sizes are needed to evaluate this purported mechanistic relationship. One such mechanistic study could include a diagnostic assessment, startle protocol, and administration of the Childhood Trauma Interview at baseline (thereby reducing the potential biases involved in retrospective recall), followed by subsequent diagnostic assessments over several years. The analysis would focus on whether the relation between adolescent adversity and anxiety disorder onset is mediated by elevated responding during the safe phases of the startle protocol that follow delivery of the aversive stimulus.

Another critical study that draws upon this and our previous work is the development of a prevention intervention to modify responding to safety cues among those who experienced adversity in adolescence, particularly if the study described above does in fact demonstrate that deficits in safety learning mediate the relationship between adolescent adversity and anxiety disorder onset. To identify the target population, a prevention program could first assess adolescents for adverse experiences, and then use a startle protocol or fear conditioning paradigm to identify which of those individuals who experienced adversity in adolescence also show deficits in safety learning or over- responding to safety cues. Participants could be randomized to an intervention that aims to train these adolescents to respond differently to safety cues (i.e., training to inhibit fear responding to these cues, or an attentional bias modification paradigm to train individuals to attend to safety cues) or a control group. The design of the study would provide a strong test of whether fear responding to safety cues plays a causal role in the development of anxiety pathology; and if so, the intervention may prevent the onset of anxiety disorders.

Finally, this study uncovered one risk factor for deficits in safety responding, but more comprehensive models are needed to understand how other risk factors (such as genetic or other biomarkers) in combination influence deficits in safety responding. Models that include several putative risk factors as main effects as well as their interactions are needed to elucidate which risk factors contribute uniquely to deficits in safety responding. Taken together, research that examines the mechanisms by which early life adversity elevates fear responding to safe cues could inform strategies and targets for prevention interventions.