Reduced discrimination between signals of danger and safety but not overgeneralization is linked to exposure to childhood adversity in healthy adults

Exposure to adverse childhood experiences (ACEs) is a strong predictor for developing behavioral, somatic and psychopathological conditions. Exposure to threat-related early adversity has been suggested to be specifically linked to altered emotional learning as well as changes in neural circuits involved in emotional responding and fear. Learning mechanisms are particularly interesting as they are central mechanisms through which environmental inputs shape emotional and cognitive processes and ultimately behavior. Multiple theories on the mechanisms underlying this association have been suggested which, however, differ in the operationalization of ACEs. 1,402 physically and mentally healthy participants underwent a fear conditioning paradigm including a fear acquisition and generalization phase while skin conductance responses (SCRs) and different subjective ratings were acquired. ACEs were retrospectively assessed through the childhood trauma questionnaire and participants were assigned to individuals exposed or unexposed to at least moderate adverse childhood experiences according to established cut-off criteria. In addition, we provide exploratory analyses aiming to shed light on different theoretical accounts on how ACE’s impact individual risk profiles (i.e., cumulative risk account, specificity model, dimensional model). During fear acquisition training and generalization, we observed reduced discrimination in SCRs between the CS+ and the CS-, primarily due to reduced CS+ responding in exposed individuals. During fear generalization, no differences in generalization gradients were observed between exposed and unexposed individuals but generally blunted physiological responses in exposed individuals. No differences between the groups were observed in ratings in any of the experimental phases. The lower CS discrimination in SCRs in exposed individuals was evident across operationalizations according to the cumulative risk account, specificity as well as dimensional model. However, none of these theories showed clear explanatory superiority. Our results stand in stark contrast to typical patterns observed in patients suffering from anxiety and stress-related disorders (i.e., reduced CS discrimination due to increased responses to safety signals). Thus, reduced CS+ responding in individuals exposed to ACEs, yet not showing signs of psychopathology, may represent a specific characteristic of this resilient subgroup that warrants further investigation with respect to its relation to risk and resilience. In addition, we conclude that theories linking ACEs to psychopathology need refinement.


Introduction
Exposure to adverse childhood experiences (ACEs) -particularly in early life -is a strong predictor for developing behavioral, somatic and psychopathological conditions (Anda et al., 2006;Danese, Pariante, Caspi, Taylor, & Poulton, 2007;Danese & Widom, 2023;Felitti, 2002;Gilbert et al., 2009;Green et al., 2010;Heim & Nemeroff, 2002;Klauke, Deckert, Reif, Pauli, & Domschke, 2010;McLaughlin et al., 2012;Moffitt et al., 2007;Teicher, Gordon, & Nemeroff, 2022) and hence causes substantial individual suffering as well as societal costs (Hughes et al., 2021). Exposure to ACEs is rather common with nearly two thirds of individuals experiencing one or more traumatic events prior to their 18th birthday (McLaughlin et al., 2013). Thus, it is central for the development of targeted intervention and prevention programs to understand the mechanisms through which ACEs become biologically embedded and contribute to the pathogenesis of stress-related somatic and mental disorders. As learning is a central mechanism through which environmental inputs shape emotional and cognitive processes and ultimately behavior, learning mechanisms are key candidates potentially underlying the biological embedding of exposure to ACEs and their impact on development and risk for psychopathology (McLaughlin & Sheridan, 2016a).
The fear conditioning paradigm is a prime translational paradigm for testing for potentially altered (threat) learning mechanisms following exposure to ACEs under laboratory conditions. The fear conditioning paradigm typically consists of different experimental phases (Lonsdorf et al., 2017). During fear acquisition training, a neutral cue is paired with an aversive event such as an electrotactile stimulation or a loud aversive human scream (unconditioned stimulus, US). Through these pairings, an association between both stimuli is formed and the previously neutral cue becomes a conditioned stimulus (CS+) that elicits conditioned responses.
In human differential conditioning experiments, typically a second neutral cue is never paired with the US and serves as a control or safety stimulus (i.e., CS-). During a subsequent fear extinction training phase, both the CS+ and the CS-are presented without the US which leads to a gradual waning of conditioned responding. A fear generalization phase includes additional stimuli (i.e., generalization stimuli; GSs) that are perceptually similar to the CS+ and CS-(e.g., generated through merging perceptual properties of the CS+ and CS-) which allows for the investigation to what degree conditioned responding generalizes to similar cues.
Fear acquisition as well as extinction are considered as experimental models of the development and exposure-based treatment of anxiety-and stress-related disorders. Fear generalization is in principle adaptive in ensuring survival ("better safe than sorry"), but broad overgeneralization can become burdensome for patients. Hence, aberrant fear acquisition, extinction and generalization processes may provide clear and potentially modifiable targets for intervention and prevention programs for stress-related psychopathology (McLaughlin & Sheridan, 2016a).

Meta-analyses suggest that patients suffering from anxiety-and stress-related disorders
show enhanced responding to the safe CS-during fear acquisition (Duits et al., 2015). During extinction, patients exhibit stronger defensive responses to the CS+ and a trend toward increased discrimination between the CS+ and CS-compared to controls, which may indicate delayed and/or reduced extinction (Duits et al., 2015). Furthermore, meta-analytic evidence also suggests stronger generalization to cues similar to the CS+ in patients and more linear generalization gradients (Cooper, van Dis, et al., 2022;Dymond, Dunsmoor, Vervliet, Roche, & Hermans, 2015;Fraunfelter, Gerdes, & Alpers, 2022).
In sharp contrast to these threat learning patterns observed in patient samples, a recent review provided converging evidence that exposure to ACEs is linked to reduced CS discrimination driven by blunted responding to the CS+ during experimental phases characterized through the presence of threat [i.e., acquisition training and generalization; Ruge et al. (2023)].
Of note, this pattern was observed in mixed samples (healthy, at risk, patients) and in pediatric samples and adults exposed to ACEs as children. The latter suggests that recency of exposure or developmental timing may not play a major role, even though there is some evidence pointing towards accelerated pubertal and neural (connectivity) development in exposed children (Machlin, Miller, Snyder, McLaughlin, & Sheridan, 2019;Silvers et al., 2016). There is, however, no evidence pointing towards differences in extinction learning or generalization gradients between individuals exposed and unexposed to ACEs (for a review, see Ruge et al., 2023). Ruge et al. (2023) also highlighted operationalization as a key challenge in the field hampering interpretation of findings across studies and consequently cumulative knowledge generation. Operationalization of exposure to ACEs, and hence translation of theoretical accounts of the role of ACEs into statistical tests, is an ongoing and current discussion in the field (McLaughlin, Sheridan, Humphreys, Belsky, & Ellis, 2021;. Historically, ACEs have been conceptualized rather broadly considering different adversity types lumped into a single category. This follows from the (implicit) assumption that any exposure to an adverse event will have similar and additive effects on the individual and its (neuro-biological) development . Accordingly, ACEs have often been considered as a cumulative measure ['cumulative risk approach'; Smith and Pollak (2021); McLaughlin et al. (2021)]. An alternative approach    2021)]. To this end, exposure to threat-related ACEs has been suggested to be specifically linked to altered emotional and fear learning ).
Yet, there is converging evidence from different fields of research suggesting that the effects of exposure to ACEs are cumulative, non-specific and rather unlikely to be tied to specific types of adverse events (Danese et al., 2009;D. A. Young et al., 2019)with few exceptions (Colich, Rosen, Williams, & McLaughlin, 2020;McLaughlin, Weissman, & Bitrán, 2019), which is also supported by a recent review on the association between threat and reward learning with exposure to ACEs (Ruge et al., 2023). Yet, the different theoretical accounts have not yet been directly compared in a single fear conditioning study. Here, we aim to fill this gap in an extraordinarily large sample of healthy adults (N=1402).
We operationalize ACE exposure through different approaches: Our main analyses employ the approach adopted by most publications in the field (see Ruge et al., 2023 for a review) -dichotomization of the sample into exposed vs. unexposed based on published cut-offs for the Childhood Trauma Questionnaire (CTQ). In addition, we provide exploratory analyses that attempt to translate dominant (verbal) theoretical accounts (McLaughlin et al., 2021; on the impact of exposure to ACEs into statistical tests while acknowledging that such a translation is not unambiguous and these exploratory analyses should be considered as showcasing a set of plausible solutions. With this, we aim to facilitate comparability, replicability and cumulative knowledge generation in the field as well as providing a solid base for hypothesis generation (Ruge et al., 2023) and refinement of theoretical accounts. More precisely, we attempted to exemplarily and exploratively translate a) the cumulative risk approach, b) the specificity model, and c) the dimensional model into statistical tests applied to our dataset, while also compiling challenges encountered when aiming to translate these verbal theories into statistical models in practice.
Based on the recently reviewed literature (Ruge et al., 2023), we expect less discrimination between signals of danger (CS+) and safety (CS-) in exposed individuals as compared to those unexposed to ACEs -primarily due to reduced responses to the CS+ -during both the fear acquisition and the generalization phase. Based on the literature (Ruge et al., 2023), we do not expect group differences in generalization gradients.

Participants
In total, 1678 healthy participants (ageM = 25.26 years, ageSD = 5.58 years, female = 60.10%, male = 39.30%) were recruited in a multi-centric study at the Universities of Münster, and Hamburg, Germany (SFB TRR58;for Würzburg 2013for Würzburg -2020. The study was approved by the local ethic committees of the three Universities and was conducted in agreement Running head: CHILDHOOD ADVERSITY AND FEAR CONDITIONING with the Declaration of Helsinki. Current and/or lifetime diagnosis of DSM-IV mental Axis-I disorders as assessed by the German version of the Mini International Psychiatric Interview (Sheehan et al., 1998) led to exclusion from the study (see supplementary material for additional exclusion criteria). All participants provided written informed consent and received 50 € as compensation. Data from subsamples of this dataset have been published previously (see supplementary material) on research questions unrelated to the ones investigated here.
A reduced number of 1402 participants (ageM = 25.38 years, ageSD = 5.76 years, female = 60.30%, male = 39.70%) were included in the statistical analyses because 276 participants were excluded due to missing data (for CTQ: n = 21, for ratings: n = 78, for skin conductance responses [SCRs]: n = 182), for technical reasons and due to deviating study protocol. Five participants had missing CTQ and missing SCR data, so the sum of exclusions in specific outcome measures does not add up to the total number of exclusions. We did not exclude physiological SCR non-responders or non-learners, as this procedure has been shown to induce bias through predominantly excluding specific subpopulations (e.g., high trait anxiety), which may be particularly prevalent in individuals exposed to ACEs . See Table 1 and supplementary material for additional sample information including trait anxiety and depression scores (see Supplementary Figure 1) as well as information on socioeconomic status (see Supplementary Figure 3). Table 1: Descriptives of the subsamples being exposed or not exposed to ACE's with exposure being defined as at least one CTQ subscale exceeding the moderate cut off (Bernstein & Fink, 1998;Häuser et al., 2011)

Procedure
Participants underwent a fear conditioning and generalization paradigm which was adapted from Lau et al. (2008) and described previously in detail (Herzog et al., 2021;Schiele, Reinhard, et al., 2016;Stegmann et al., 2019). Details are also provided in brief in the supplementary material.

Ratings
At the end of each experimental phase (habituation, acquisition training and generalization) as well as after half of the total acquisition and generalization trials, participants provided ratings of the faces with regards to valence, arousal (9-point Likert-scales; from 1 = very unpleasant/very calm to 9 = very pleasant/very arousing) and US contingencies (11-point Likert-scale; from 0 to 100% in 10% increments). As the US did not occur during the habituation Fear conditioning and generalization paradigm.
phase, contingency ratings were not provided after this phase. For reasons of comparability, valence ratings were inverted.

Physiological data recordings and processing
Skin conductance was recorded continuously using Brainproducts V-Amp-16 and Vision Recorder software (Brainproducts, Gilching, Germany) at a sampling rate of 1000 Hz from the non-dominant hand (thenar and hypothenar eminences) using two Ag/AgCl electrodes. Data were analyzed offline using BrainVision Analyzer 2 software (Brainproducts, Gilching, Germany).
The signal was filtered offline with a high cut-off filter of 1 Hz and a notch filter of 50 Hz.
Amplitudes of SCRs were quantified by using the Trough-to-peak (TTP) approach. According to published guidelines (Boucsein et al., 2012), response onset was defined between 900-4000 ms after stimulus onset and the peak between 2000-6000 ms after stimulus onset. A minimum response criterion of 0.02 mS was applied, with lower índividual responses scored as zero (i.e., non-response). Note that previous work using this sample (Schiele, Reinhard, et al., 2016;Stegmann et al., 2019) had used square-root transformations but we decided to employ a logtransformation in order to approximate a normal distribution of the data and a range correction to control for individual variability (i.e., dividing each SCR by the maximum SCR per participant) due to the study focus (Lykken, 1972;Lykken & Venables, 1971).

Psychometric assessment
Participants completed a computerized battery of questionnaires (for a full list, see  Hautzinger and Bailer (1993)]. The CTQ contains 28 items for the retrospective assessment of childhood maltreatment across five subscales (emotional, physical, and sexual abuse, as well as emotional and physical neglect; for internal consistency, see supplementary material) and a control scale, the STAI-T consists of 20 items addressing general negative affect (Laux & Spielberger, 1981;Spielberger, 1983) and the ADS-K includes 15 items assessing depressiveness during the past 7 days.
We implemented different approaches to operationalize exposure to ACEs (see Table 2): Operationalization of "exposure".

Main analyses
Moderate exposure based on CTQ (exposed vs. unexposed) Short description: At least one subscale met the published cut-off for at least moderate exposure (Bernstein & Fink,1998;Häuser et al., 2011) Procedure: emotional abuse >= 13, physical abuse >= 10, sexual abuse >= 8, emotional neglect >= 15, physical neglect >= 10), a cut-off employed in previous work by our team  and in the literature (Ruge et al., 2023) Statistical test: See "Methods: Statistical analyses" • Not based on an existing theory but on what is commonly used in the literature (Ruge et al., 2023) • Different cut-offs published (for a discussion, see Ruge et al., 2023) • (Statistical) Challenges linked to dichotomization of an inherently continuous variable

Exploratory analyses
Cumulative risk model Evans et al., 2013;McEwen 2003) Short description: Based on the assumed key role of cumulative exposure (exposure intensity and frequency) Procedure a): classification into the four severity groups (no, low, moderate, severe exposure) based on cut-offs published by Bernstein & Fink (1998) Statistical test a): comparison of conditioned responding of the four severity groups by using one-way ANOVAs • Problem with CTQ sum score: it assigns the same "value" to all CM types (see also "General operationalizational challenges" below) • Number of subscales exceeding cut-off: calculate ANOVA or regression? • Cumulative risk scores are based on the implicit assumption that different types of adverse events affect the same mechanisms and are of equal impact Procedure b): number of subscales exceeding an at least moderate cut-off based on Bernstein & Fink (1998) and Häuser et al.,(2011) Statistical test b): number of sub-scales exceeding an at least moderate cut-off as predictor and conditioned responding as criterion in simple linear regression models Procedure: summing up the CTQ subscales emotional abuse, physical abuse and sexual abuse yielding a composite score for exposure to "abuse" and summing up the subscales emotional neglect and physical neglect to yield a composite score for "neglect" (or threat vs. deprivation as done by Sheridan et al., 2017) Statistical test: the abuse and neglect composite score is tested for associations with conditioned responding in separate regression models In our sample n = 52 and n = 96 individuals were exposed to abuse only and neglect only, respectively, while n = 55 reported to have experienced both abuse and neglect. We included all participants in all analyses as done previously (Sheridan et al., 2017) • What qualifies as a specific exposure type (i.e., subscales or composite scales for neglect vs. abuse?) • Which exposure subcategories are "too specific" or "too broad"?
(A heterogeneous category may obscure potentially relevant discrete associations) • Include only participants who experienced only one specific type but not any other types despite this being rather artificial due to high co-occurrences of different exposure types and requiring extremely large samples? Which cut-off should be used then to define exposure? We decided to include all participants in the analyses as done in previous studies (Sheridan et al., 2017) • Lack of specificity of exposure subtypes (e.g., sexual abuse also has an emotional component)

Dimensional model
Short description: consideration of specific exposure types (i.e., abuse and neglect) that are assumed to co-occur and be controlled for the • Non-comparability of dimensional and categorical approaches: CTQ sum score assumes an equal contribution of all items which contradicts different thresholds for being considered as exposed for different subscales (e.g., lower cut-off for sexual abuse as compared to emotional neglect) • Associations in a full sample may differ from associations in the group of exposed individuals only which is a challenge for interpretation of data • Multiple cut-offs published (Bernstein et al. 1997;Bernstein & Fink, 1998) • Specific challenges relating to abuse and neglect: They ○ often co-occur ○ are not the only relevant dimensions (e.g., unpredictability, loss) ○ are not strongly supported as distinct dimensions in the literature (Carozza et al., 2022; • Heterogeneity in the assessment of childhood adversity across studies -both with respect to the assessment tools (e.g., questionnaires, interview) as well as with respect to the operationalization of adversity (i.e., definition) • Different response formats (yes/no vs. specification of duration and frequency) and the number of trauma types/events included in assessment tools impact on prevalence rates and potentially also associations between the number of adverse experiences and symptom severity [e.g., Contractor et al., 2018) • Distinction between stressful events and trauma is often unclear (Richter-Levin & Sandi, 2021)

Statistical analyses
Manipulation checks were performed to test for successful fear acquisition and generalization (for more details, see supplementary material). Following previous studies (Imholze et al., 2023;Stegmann et al., 2019), we calculated three different outcomes for each participant for SCRs and ratings: CS discrimination (for acquisition training and the generalization phase), the linear deviation score (LDS; only for the generalization phase) as an index of the linearity of the generalization gradient (Kaczkurkin et al., 2017) and the general reactivity (across all phases including habituation, acquisition training and the generalization phase). CS discrimination was calculated by averaging responses to CS+ and CS-across trials (except the first acquisition trial) and subtracting averaged CS-responses from averaged CS+ responses. The first acquisition trial was excluded as no learning could possibly have taken place due to the delay conditioning paradigm. The linear deviation score (LDS) was calculated by subtracting the mean responses to all GSs from the mean responses to both CSs during the generalization phase.
To calculate the general reactivity in SCRs and ratings, trials were averaged across all stimuli (CSs and GSs) and phases (i.e., habituation, acquisition training and generalization phase). Note that raw SCRs were used for analyses of general physiological reactivity.
CS discrimination during acquisition training and the generalization phase, LDS and general reactivity were compared between participants who were exposed and unexposed to ACEs by using two-tailed independent t-tests. For CS discrimination in SCRs, a two-way mixed ANOVA was conducted to examine the effect of ACE exposure on responses to the CS+ and CSby including CS type and ACE exposure as independent variables. As the interaction between CS type and ACE exposure was statistically significant, post-hoc two-tailed paired t-tests were used to compare SCRs to CS+ and CS-between exposed and unexposed participants.
In addition, exploratory analyses used different operationalizations of ACE exposure (see Table 2 for details). Note that no correction for alpha inflation was applied in these analyses, In statistical procedures where the assumption of homogeneity of variance was not met, Welch's tests, robust trimmed means ANOVAs (Mair & Wilcox, 2020a) and regressions with robust standard errors using the HC3 estimator (Hayes & Cai, 2007) were calculated instead of ttests, ANOVAs and regressions, respectively. Note that for robust mixed ANOVAs, the WRS2 package in R (Mair & Wilcox, 2020a) does not provide an effect size. Post-hoc t-test or Welch's tests were corrected for multiple comparisons by using the Holm correction. As post-hoc tests for robust ANOVAs, Yuen independent samples t-test for trimmed means were calculated including the explanatory measure of effect size [Mair and Wilcox (2020a); Values of 0.10, 0.30, and 0.50 represent small, medium, and large effect sizes, respectively. Even though such rules of thumb have to be evaluated with a critical view, we provide these benchmarks here as this effect size might be somewhat unknown.].
Following previous calls for a stronger focus on measurement reliability (Cooper, Dunsmoor, Koval, Pino, & Steinman, 2022;Klingelhöfer-Jens, Ehlers, Kuhn, Keyaniyan & Lonsdorf, 2022), we also provide information on split-half reliability for SCRs as well as Cronbach's alpha for the CTQ in the supplementary material. For all statistical analyses described above, the a priori significance level was set to α = 0.05. For data analysis and visualizations as well as for the creation of the manuscript, we used R (Version 4.1.3; R Core Team, 2022b) and the R-packages apa (Aust & Barth, 2020 Wilcox, 2020b), and zoo (Version 1.8.8; Zeileis & Grothendieck, 2005).

Results
As expected, participants exposed to ACEs reported significantly higher trait anxiety and depression levels than unexposed participants (all p's < 0.001; see Table 1 and Supplementary   Figure 1). Exposed and unexposed participants were equally distributed across data recording sites ( 2 (3) = 3.72, p = .293).

Main Effect of Task
In brief, and as reported previously, fear acquisition was successful in SCRs as well as ratings in the full sample (all p's < 0.001; see supplementary material for details). During fear generalization, the expected generalization gradient was observed with a gradual increase in SCRs and ratings with increasing similarity to the CS+ (all p's < 0.01 except for the comparisons of SCRs to CS-vs. GS4 as well as GS1 vs. GS2 which were non-significant; see supplementary material).

Association between different outcomes and exposure to childhood adversity
During both the acquisition training and generalization phase, CS discrimination in SCRs was significantly lower in individuals exposed to ACEs as compared to unexposed individuals (see Table 3 and Figure 1 Table 3).
No significant effect of exposure to ACEs in either SCRs or ratings was observed for generalization gradients (see Table 3 and Figure 3). It is, however, also evident from the generalization gradients that both groups differ specifically in reactivity to the CS+ (see above and Figure 3).
In addition, general physiological reactivity in SCRs (i.e., raw amplitudes) was significantly lower in participants exposed to ACEs compared to unexposed participants (see Table 3 and Figure 4) while there were no differences between both groups in general rating response levels (see Table 3).  [Insert Figure 4 about here]

Exploratory analyses
The cumulative risk model operationalized through the different cut-offs for no, low, moderate and severe exposure (Bernstein & Fink, 1998) did not yield any significant results for any outcome measure and experimental phase (see Supplementary Table 2). However, on a descriptive level (see Figure 5), it seems that indeed exposure to at least a moderate cut-off level may induce behavioral and physiological changes [see main analysis; Bernstein and Fink (1998)] suggesting that what is commonly applied in the literature (see Ruge et al., 2023) may indeed represent a reasonable approach supported by our data.
For an illustration, how the different adversity types (i.e., subscales) are distributed among the different numbers of subscales, see Supplementary Figure 5.
The operationalization of ACEs in the context of the specificity model tests the association between exposure to abuse and neglect experiences on conditioned responding statistically independently, while the dimensional model controls for each other's impact (see Table 2 for details and Figure 7 for an exemplary illustration of SCRs during acquisition training). Despite these conceptual and operationalizational differences, results are converging. More precisely, no significant effect of exposure to abuse was observed on CS discrimination, generalization (i.e., LDS) or general reactivity in any of the outcome measures and in any experimental phase (see Supplementary Table 4   . However, upon inspecting the data to assess collinearity assumptions, we found that multicollinearity was not a concern in our study (abuse: VIF = 8.64; neglect: VIF = 7.93).
Furthermore, the statistical analyses of the specificity model additionally revealed that greater exposure to neglect significantly predicted a generally lower SCR reactivity (F = [1, Hence, we found no evidence of differential explanatory strengths among theories.
[Insert Figure 5 about here] [Insert Figure 6 about here] [Insert Figure 7 about here] Running head: CHILDHOOD ADVERSITY AND FEAR CONDITIONING

Discussion
The objective of this study was to examine the relationship between adverse childhood experiences (ACEs) and conditioned responding as a potential mechanistic route linking experience-dependent plasticity in the nervous system and behavior related to risk of and resilience to psychopathology using a large community sample of healthy participants. In additional exploratory analyses, we examined these associations through different approaches by translating key theories in the literature into statistical tests. In line with the conclusion of a recent systematic literature review (Ruge et al., 2023), individuals exposed to (an at least moderate level of) ACEs exhibited reduced CS discrimination in SCRs during both acquisition training and the generalization phase compared to those classified as unexposed (i.e., no or low exposure).
Generalization gradients themselves were, however, comparable between exposed and unexposed individuals.
The systematic literature search by Ruge et al. (2023) revealed that the pattern of decreased CS discrimination, driven primarily by reduced CS+ responding, was observed despite substantial heterogeneity in ACE assessment and operationalization, and despite differences in the experimental paradigms. Although both individuals without mental disorders exposed to ACEs and patients suffering from anxiety-and stress-related disorders (e.g., Duits et al. (2015)) show reduced CS discrimination, it is striking that the response pattern of individuals exposed to ACEs (i.e., reduced responding to the CS+) is remarkably different from what is typically observed in patients (i.e., enhanced responding to the CS-). It should be noted, however, that ACE exposure status was not considered in this meta-analysis. As exposure to ACEs represents a particularly strong risk factor for the development of later psychopathology, these seemingly contrary findings warrant an explanation. In this context, it is important to note that all individuals included in the present study were mentally healthy -at least up to the assessment.
Hence, it may be an obvious explanation that reduced CS discrimination driven by decreased CS+ responding may represent a resilience rather than a risk factor because individuals exposed to ACEs in our sample are mentally healthy despite being exposed to a strong risk factor.
In fact, there is substantial heterogeneity in individual trajectories and profiles in the aftermath of such exposures in humans and rodents (Russo, Murrough, Han, Charney, & Nestler, 2012). While some individuals remain resilient despite exposure, others develop psychopathological conditions (Galea, Nandi, & Vlahov, 2005). Consequently, the sample of the present study may represent a specific subsample of exposed individuals who are developing along a resilient trajectory. Thus, it can be speculated that reduced physiological reactivity to a signal of threat (e.g., CS+) may protect the individual from overwhelming physiological and/or emotional responses to potentially recurrent threats (for a discussion, see Ruge et al. (2023)).
Similar concepts have been proposed as "emotional numbing" in post-traumatic stress disorders (for a review, see e.g., Litz & Gray, 2002).
While this seems a plausible theoretical explanation, decreased CS discrimination driven by reduced CS+ responding is observed rather consistently and most importantly irrespective of whether the investigated samples were healthy, at risk or included patients (Ruge et al., 2023).
Thus, this response pattern which was also observed in our study might be a specific characterization of ACE exposure distinct from the response pattern generally observed in patients suffering from anxiety-and stress-related disorders (i.e., increased responding to the CS-) -even though individuals exposed to ACEs in this sample indeed showed significantly higher anxiety and depression despite being free of any categorical diagnoses (see supplementary material) which was also previously reported by Kuhn, Mertens, & Lonsdorf (2016). Hence, future work should investigate potential links between reduced CS discrimination in individuals exposed to ACEs and the developmental trajectories of psychopathological symptoms inoptimally -prospective studies. In fact, there is first evidence that exposure to stressful life events were longitudinally associated with increases in anxiety sensitivity (McLaughlin & Hatzenbuehler, 2009).
In addition to reduced CS discrimination in SCRs, a generally blunted electrodermal responding was observed, which may, however, be mainly driven by substantially reduced CS+ responses. Yet, it is noteworthy that reduced skin conductance in children exposed to ACEs was also observed during other tasks such as attention regulation during interpersonal conflict (Pollak, Vardi, Putzer Bechner, & Curtin, 2005) or passively viewing slides with emotional or cognitive content (Carrey, Butter, Persinger, & Bialik, 1995), whereas other studies did not find such an association (Ben-Amitay, Kimchi, Wolmer, & Toren, 2016). Moreover, in various threat-related studies, also enhanced responding or no significant differences were observed across outcome measures (Estrada, Richards, Gee, & Baskin-Sommers, 2020;Huskey, Taylor, & Friedman, 2022;Jovanovic et al., 2009;Jovanovic et al., 2022;Kreutzer & Gorka, 2021;Lis et al., 2020;Pole et al., 2007;Rowland et al., 2022;Thome et al., 2018;E. S. Young et al., 2018). While generally blunted responding might be particularly related to decreased CS+ responding in the present study, differences in general reactivity need to be taken into account for data analyses and interpretation -in particular as the exclusion of physiological non-responders or so called "nonlearners" (i.e., individuals not showing a minimum discrimination score between SCRs to the CS+ and CS-) has been common in the field until recently [for a critical discussion, see Lonsdorf et al. (2019)]. Future work should also investigate reactivity to the unconditioned stimulus, which was not implemented here and may shed light on potential differences in general reactivity unaffected by associative learning processes (see e.g., Harnett et al., 2019;Machlin et al., 2019).
Contrary to the association between CS discrimination as well as general (electrodermal) reactivity and exposure to ACEs, no such relationship was found for generalization gradients. In a subsample of this study, it was previously observed that fear generalization phenotypes explained less variance as compared to CS discrimination and general reactivity (Stegmann et al., 2019), and CS discrimination as well as general reactivity but not fear generalization predicted anxiety and depression scores after the COVID-19 pandemic (Imholze et al., 2023). The lack of associations with fear generalization measures (i.e., LDS) may be specific to the paradigm and sample used in these studies, but it may also be an interesting lead for future work to disentangle the relationship between CS discrimination, general reactivity, and generalization gradients, as they have been suggested to be interrelated (Imholze et al., 2023;Stegmann et al., 2019).
In sum, the current results converge with the literature in identifying reduced CS discrimination and decreased CS+ responding as key characteristics in individuals exposed to ACEs. As highlighted recently (see Koppold, Kastrinogiannis, Kuhn, & Lonsdorf, 2023;Ruge et al., 2023), the various operationalizations of ACEs as well as general trauma (Karstoft & Armour, 2023) represent a challenge for integrating the results in the existing literature. Hence, future studies should focus on in-depth phenotyping (Ruge et al., 2023), an elaborate classification of adversity subtypes  and methodological considerations (Ruge et al., 2023).
Running head: CHILDHOOD ADVERSITY AND FEAR CONDITIONING Several proposed (verbal) theories describe the association between (specific) ACEs and behavioral as well as physiological consequences differently and there is currently a heated debate rather than consensus on this issue (McLaughlin et al., 2021;. In the field, most often a dichotomization in exposed vs. unexposed individuals is used [for a review, see Ruge et al. (2023)]. We adopted this typical approach of an at least moderate exposure cut-off from the literature for our main analyses, despite the well-known statistical disadvantages of artificially dichotomizing variables that are (presumably) dimensional in nature (Cohen, 1983). It is noteworthy, however, that this cut-off appears to map rather well onto psychophysiological response patterns observed here (see Figure 5). More precisely, our exploratory results of applying different exposure cut-offs (low, moderate, severe, no exposure) seem to indicate that indeed a moderate exposure level is "required" for the manifestation of physiological differences, suggesting that ACE exposure may not have a linear or cumulative effect.
Of note, comparing individuals exposed vs. unexposed to an at least moderate level of ACEs is not derived from any of the existing theories, but rather from practices in the literature (see Ruge et al., 2023). For this reason, we aimed at an exploratory translation of key (verbal) theories into statistical tests (see Table 2). Several important topical and methodological take home messages can be drawn from this endeavor: First, translation of these verbal theories into precise statistical tests proved to be a rather challenging task paved by operationalizational ambiguity. We have collected some key challenges in Some limitations of this work are worth noting: First, despite our observation of significant associations between exposure to ACEs and fear conditioning performance in a large sample, it should be noted that effect sizes were small. Second, we cannot provide a comparison of potential group differences in unconditioned responding to the US. This is, however, important as this comparison may explain group differences in conditioned responding -a mechanism that remains unexplored to date (Ruge et al., 2023). Third, the use of the CTQ, which is the most commonly used questionnaire in the field (see Ruge et al., 2023), comes with a number of disadvantages, most prominently, an exclusive focus on the presence or absence of exposure without consideration of individual and exposure characteristics that have been shown to be of crucial relevance (Danese & Widom, 2023;see Smith & Pollak, 2021), such as controllability, burdening, exposure severity, duration and developmental timing. These characteristics are embedded in the framework of the topological approach , another important model linking ACE exposure to negative outcomes, which, however, was not evaluated in the present work. Testing this model requires an extremely large dataset including in-depth phenotyping, which was not available here, but may be an important avenue for future work.
Fourth, across all theories, significant effects of ACEs have been shown primarily on physiological reactivity (i.e., SCR). Whether these findings are specific to SCRs or might generalize to other physiological outcome measures such as fear potentiated startle, heart rate or local changes in neural activation, remains an open question for future studies.
In sum, when ultimately aiming to understand the impact of exposure to adversity on the development of psychopathological symptoms (Anda et al., 2006;Felitti, 2002;Gilbert et al., 2009;Green et al., 2010;Heim & Nemeroff, 2001;McLaughlin et al., 2012;Moffitt et al., 2007;Teicher et al., 2022), it is crucial to understand the biological mechanisms through which exposure to adversity "gets under the skin". To achieve this, emotional-associative learning can serve as a prime translational model for fear and anxiety disorders: One plausible mechanism is the ability to distinguish threat from safety, which is key to an individual's ability to dynamically adapt to changing environmental demands (Craske et al., 2012;Vervliet, Craske, & Hermans, 2013) -an ability that appears to be impaired in individuals with a history of ACEs. This mechanism is of particular relevance to the development of stress-and anxiety-related psychopathology, as the identification of risk but also resilience factors following exposure to childhood adversity is essential for the development of effective intervention and prevention programs.
Running head: CHILDHOOD ADVERSITY AND FEAR CONDITIONING

Data and Code Availability Statement
The data will be made available to editors and reviewers only, as open access to the data was not indicated in the ethical approval. R Markdown files that include all analyses and generate this manuscript are openly available at zenodo (DOI: 10.5281/zenodo.8190359). generalization phase (C-D) for individuals unexposed (gray) and exposed (pink) to ACE's.
Barplots (A and C) with error bars represent means and standard errors of the means (SEMs).
Distributions of the data are illustrated in the raincloud plots (B and D). Points next to the densities represent CS discrimination of each participant averaged across phases. Boxes of boxplots represent the interquartile range (IQR) crossed by the median as a bold line, ends of whiskers represent the minimum/maximum value in the data within the range of 25th/75th percentiles ±1.5 IQR For trial-by-trial SCRs across all phases, see Supplementary Figure 4). log = log-transformed, rc = range corrected.  well as generalization (i.e., the LDS) and the general reactivity are lower in all groups exposed to ACEs at an at least moderate level as compared to those with no or low exposure -which corresponds to the main analyses (see above). log = log-transformed, rc = range corrected. Note that the different ranges of CTQ composite scores result from summing up two and three subscales for the neglect and abuse composite scores, respectively (see also Table 2 for more details). log = log-transformed, rc = range corrected, R squ. = R squared.