Physiological Stress Response Reactivity Mediates the Link Between Emotional Abuse and Youth Internalizing Problems

Abstract

Youth who are raised in emotionally abusive families are more likely to have poor mental health outcomes such as depression and anxiety. However, the mechanisms of this association are unclear. The present study utilized a longitudinal sample of low-SES youth (N = 101, M_ageT1 = 10.24) to examine stress response reactivity (i.e., vagal withdrawal, sympathetic activation, and hypothalamic-pituitary-adrenal [HPA] axis activation) as mediators between emotional abuse and prospective youth internalizing symptoms. Results indicated that blunted HPA reactivity to a laboratory social stress task mediated the association between emotional abuse and youth internalizing symptoms. Emotional abuse was also associated with blunted parasympathetic nervous system activity (i.e., less vagal withdrawal than average). In sum, emotional abuse is a potent risk factor for youth internalizing symptoms, and this link may be mediated via dysregulation in physiological stress response systems. Primary prevention of childhood emotional abuse and secondary prevention programs that target self-regulation skills may reduce rates of youth internalizing symptoms and disorders.

Child maltreatment is an adverse and often traumatic experience that places healthy child development at risk. Adolescents who have been exposed to child maltreatment are more likely to exhibit mental health symptomology such as internalizing, externalizing, and substance use problems [1–4]. Empirical evidence suggests that emotional abuse, a type of child maltreatment that includes acts such as name-calling and failing to provide emotional support [5, 6], is a particularly salient risk factor for internalizing symptoms (i.e., depression, anxiety, and withdrawal) [7, 8]. Several mechanisms in the association between childhood emotional abuse and internalizing symptoms have been identified, including hopelessness, maladaptive cognitive schemas, and mother-adolescent conflict [8–10]. However, although biological mechanisms have been suggested generally in associations between childhood adversity and future maladaptive outcomes [11–13], the multilevel developmental mechanisms between childhood emotional abuse and adolescent internalizing symptomology remain unclear.

The developmental mechanisms that underlie the association between childhood emotional abuse and youth psychopathology can be explained with the developmental psychopathology perspective [14]. This framework suggests that child maltreatment leads to future psychopathology by way of disruptions in normative developmental processes, such as the attainment of self-regulation skills [15]. Corresponding evidence from psychophysiological research suggests that chronic stress caused by emotional abuse can disrupt the development of stress response systems and result in dysregulated physiological responses to acute stress [11]. These dysregulated physiological responses to stress are, in turn, thought to be a transdiagnostic mechanism for psychopathology [16–18]. Thus, it is theoretically plausible that the chronic stress induced by childhood emotional abuse can disrupt the normative development of self-regulation, including physiological stress regulation systems, thus cascading to psychopathology. However, to our knowledge, the role of bioregulatory stress systems in the association between childhood emotional abuse and adolescent internalizing symptoms has not been systematically investigated.

Emotional Abuse and Stress Response Reactivity

To address the aforementioned gaps in the literature, the present study investigates physiological stress reactivity in the autonomic nervous system (ANS) and the hypothalamic-pituitary-adrenal (HPA) axis in relation to emotional abuse and youth internalizing symptoms. The ANS and HPA axis are the two primary biological systems that are responsible for responding to threats and stressors in the environment [19]. The ANS consists of the sympathetic and the parasympathetic nervous systems (SNS and PNS, respectively). The sympathetic adreno-medullary (SAM) system is a component of the SNS that is primarily responsible for the fight-or-flight stress response [20], while the PNS plays a complementary role in the stress response by providing regulatory support and working to restore homeostasis. One of the crucial actions of the PNS during acute stress is the withdrawal of the vagal brake, a term referring to the consistent down-regulating effect on cardiac function performed by the vagal nerve. The present study defines PNS reactivity as vagal withdrawal, though it should be noted that vagal augmentation may also occur in non-threatening and engaging social situation [21]. Together, the actions of the SAM system and the PNS allow the individual to more effectively respond to external stressors [22]. The HPA axis works in conjunction with this ANS response by initiating and regulating the hormonal, or endocrine, response to stress. Cortisol is the primary glucocorticoid that the HPA system produces and circulates in response to stress [23]. Glucocorticoids have a variety of regulatory actions in the body during stress, including suppression of the immune and inflammatory response [24], regulation of an individual’s energy via glucose metabolism, and inhibition of short-term fight or flight responses [19]. The distribution of cortisol in the body typically peaks approximately twenty minutes after a stressor occurs, and its effects may last for days [20].

Researchers often administer a mildly aversive stressor in a laboratory setting to study the influence of childhood experiences such as emotional abuse on individual’s physiological responses to acute stress. A widely validated social stress-induction test is the mental arithmetic task, in which participants are asked to complete a series of mental math problems in front of a group of researchers who pretend to evaluate the participant’s performance [25]. Researchers commonly utilize pre-ejection period (PEP) and heart rate variability (HRV) to measure physiological reactivity in the ANS, and cortisol change to measure physiological reactivity in the HPA axis. PEP, obtained using an electrocardiogram (ECG), indexes the amount of epinephrine that is produced during a SAM reaction by measuring the amount of time it takes between a heartbeat and the ejection of blood into the aortic valve [26, 27]. Accordingly, PEP is often used to operationalize SNS activity. HRV is also obtained in research settings by using an ECG during a laboratory stress task and is a measure of the difference in inter-beat intervals between aspirations (i.e., breathing in) and expirations (i.e., breathing out). The high-frequency components of heart rate variability (HF HRV) are then isolated to operationalize PNS activity [22]. HF HRV indexes vagal tone, or the activity of the vagus nerve, wherein HF HRV reactivity to stress is synonymous with vagal withdrawal [28]. Thus, by obtaining PEP and HF HRV before and during a stress task, researchers can indirectly assess an individual’s active ANS response to acute social stress. Lastly, to measure physiological reactivity in the HPA axis, cortisol is typically collected via saliva before, during, and after a laboratory-induced stress task. Under normal conditions in response to acute stress, it is expected that cortisol levels will increase, and PEP and HF HRV will decrease (i.e., PEP shortening and vagal withdrawal), although extreme vagal withdrawal is also a form of dysregulation [19, 26].

Using the techniques mentioned above, investigators have linked adverse childhood experiences with stress response dysregulation in the ANS and the HPA axis [29, 30]. In the PNS, much of the evidence shows that early life stress is associated with a lack of vagal withdrawal (i.e., blunted PNS activity) during acute stress [29, 31]. Childhood adversity has also been linked to dysregulation to acute stressors in the SNS. For instance, youth who experienced severe early-life deprivation exhibited reduced PEP scores (i.e., higher sympathetic tone) during early adolescence [30]. In another study, hyper-reactivity in the SNS mediated the association between family conflict (a form of childhood adversity) and internalizing symptoms for boys [32]. However, there is mixed empirical evidence, with some studies also suggesting an association between maltreatment and blunted SNS responsivity to social stress during early adolescence [33]. The literature on childhood adversity and HPA axis reactivity is more consistent, showing that early life stress is associated with a blunted pattern of cortisol reactivity. Findings from the Bucharest Early Intervention Project suggest that institutionalized youth who endured severe and chronic deprivation were more likely to exhibit a blunted cortisol reaction to acute stress in comparison to youth who were placed with high-quality foster care families [29]. This finding is supported by a recent meta-analysis that indicated a moderate effect size between childhood adversity and blunted cortisol reactivity [34]. However, there are limited studies that have specifically investigated the association between childhood emotional abuse and physiological stress dysregulation in both the HPA axis and ANS.

Stress Response Reactivity and Youth Internalizing Symptoms

Investigating the link between emotional abuse and physiological stress reactivity is particularly important given the known associations between stress dysregulation and youth psychopathology. In particular, recent research has linked deficits in physiological stress regulation with elevated internalizing symptomology in youth [35]. There is strong evidence for the link between HPA axis dysregulation and internalizing psychopathology, specifically depression [35]. In fact, altered HPA axis responsivity to stress is one of the prominent neurobiological risk factors for depression in adult samples [36]. These findings have been replicated for children and adolescents with depression, with depressed adolescents more likely to exhibit hyper-reactivity to a psychosocial stressor [37]. However, the association between cortisol dysregulation and internalizing symptoms may vary due to a number of criteria, including timing and type of adversity, and sample characteristics [38]. This is demonstrated by recent research findings that hypo-cortisolism mediates the association between childhood adversity and youth internalizing symptoms [39].

A growing literature has also examined the associations between ANS regulation, particularly measured by HRV, and internalizing disorders, although much of this research has been conducted using adult samples. There is strong evidence that individuals with depression and anxiety exhibit lower vagal tone (i.e., resting HF HRV) [40]. Additionally, although HRV reactivity has been substantially less examined in relation to internalizing outcomes, some research shows that blunted vagal withdrawal is associated with adult major depressive disorder [31]. This work has been extended to youth samples with results similarly implicating blunted vagal withdrawal in depressive symptomology, although results with youth have been more mixed [31, 41]. Additionally, blunted SNS reactivity is associated with depressive symptoms in adults [42]. Evidence with youth samples tells a more complex story, showing for example that the association between SNS regulation and youth behavior problems is moderated by other factors including gender, type of laboratory stress task, and family environment [43, 44]. Ultimately, more work is needed to understand the associations between ANS reactivity and youth internalizing symptoms.

The Present Study

The present study aimed to investigate the indirect associations between childhood emotional abuse and internalizing psychopathology via cortisol reactivity and two biomarkers of ANS reactivity. We hypothesized that emotional abuse and youth internalizing psychopathology would be associated indirectly via blunted cortisol reactivity (H1); emotional abuse and youth internalizing psychopathology would be associated indirectly via blunted SNS stress reactivity (i.e., lower PEP-R) (H2); and that emotional abuse and internalizing psychopathology would be indirectly associated via lower levels of vagal withdrawal (i.e., lower HF HRV-R) (H3). Control variables included physically punitive parenting and neglect in order to isolate the independent influence of childhood emotional abuse. Additionally, we controlled for demographic and individual variables, including youths’ age, sex, race, pubertal stage, and body-mass index (BMI).

Methods

Sample

Participants (N = 101) were youth aged 9–12 years old (M = 10.28) and one of their caregivers who were recruited into a larger study that examined the family and community context of youth decision making. The sample was equally divided by gender (51% female). The racial makeup of the youth sample included 75% African American, 11% Caucasian, 9% Hispanic or Latinx, 1% Native American, and 4% Other. Families were recruited from a non-metropolitan region of the Southeastern United States in an area immediately surrounding the university community. In order to be eligible for the study, youth must have been (a) between the ages of 9–12, (b) English speakers, and (c) able to read and answer questions at an elementary reading level. Additionally, families who participated must have had a household income at or below 200% of the federal poverty level, which at the time (2017) was indicated by an annual income of $48,600 for a family of four. The average income for families in the study was $21,740 (SD = 12,800). Caregivers reported having on average 13.33 years of education (SD = 2.76). Among the caregivers, 1.82% completed less than 8 years of education, 18.18% completed 9–12 years with no diploma, 25.45% received a high school diploma or GED, 34.54% completed some college, 16.36% obtained a college degree, and 3.65% earned a post-graduate degree. Participants were ineligible for the study if (a) the parent was pregnant, or (b) the youth had type II diabetes or significant developmental disabilities. Of the full sample, there were approximately 9% (n = 8) and 16% (n = 16) of families who had an open or closed case with child protective services, respectively. The majority of primary caregivers recruited into the study were the youth’s biological mother (n = 91).

Procedures

The institutional review board for ethical conduct in research approved all study procedures. The larger study consisted of two waves of data collection. During the first wave, youth and their primary caregiver completed research procedures at a university-affiliated clinical unit. Trained research staff and licensed pediatric nurses implemented all study procedures. Before any study procedures took place, parents provided their written informed consent, and youth provided their informed assent.

After informed consent took place, researchers collected saliva via the passive drool method to obtain measures of cortisol. Then, ANS measures were collected from participants using mobile electrocardiogram (ECG) units. The ECG was attached using seven dermal ECG electrodes, which were attached on the clavicle (both sides) and lower rib cage (both sides), on the sternum, and the upper and lower spine. After electrodes were attached to the youth, a baseline measure of ANS activity was obtained via the ECG. Researchers instructed participants to close their eyes and listen to a five-minute video of nature sounds (e.g., waterfalls or rainforest sounds). This baseline procedure was informed by recommendations for using HRV in behavioral research [45]. After baseline was established, researchers implemented a mental arithmetic social stress task to obtain measures of youth ANS reactivity. The procedures were modified from the commonly administered Trier Social Stress Test [46]. Youth were instructed to answer a series of arithmetic problems aloud in front of a group of researchers and their parent. The difficulty of subtraction problems increased as the task went on, and the difficulty was adjusted accordingly to the participant’s ability. Researchers abstained from giving any feedback to the participants and maintained a neutral facial expression throughout the task. The complete task took five minutes to complete. We chose to utilize this particular task for the present study as the mental arithmetic task is a well-validated experimental social stressor that has been shown to reliably elicit a physiological stress response [47–49]. Individual differences in physiological stress responses elicited by the Trier Social Stress Test have been linked to self-reported emotion regulation [50] and psychopathology [51].

After the mental arithmetic task, families completed a discussion task intended to induce stress. Youth and their caregiver received index cards that contained common topics of disagreement (e.g., completing homework, chores) and were instructed to discuss three topics they had the most disagreement on. The full task spanned ten minutes. The task was previously utilized as part of the Early Head Start 5^th Grade Follow-up Study [52]. The second collection of saliva took place approximately 20-minutes after the first stress task began to assess cortisol reactivity. All ANS parameters were collected during the first stress task (mental arithmetic). Following the mental math and discussion tasks, youth and parents completed a battery of survey measures. Youth completed their surveys with the assistance of a trained research assistant who read each item and the corresponding answer choices to the youth. Parents completed surveys independently on a laptop computer in a room apart from the child.

The second wave of data collection took place approximately one year after the first wave. There were 71 families who completed the follow-up assessment. Two trained research assistants visited families at their homes to conduct the follow-up appointment. Parents and youth provided their informed consent and assent and then completed assessments separately using a handheld device. The follow-up visit took approximately one hour to complete.

Measures

Child maltreatment.

Child maltreatment was measured at Wave 1, utilizing subscales from the Parent-Child Conflict Tactics Scale (CTS-PC) [53]. Parents were asked to indicate the frequency of specific behaviors towards their child in the past year, and answer choices ranged from “0” (this has not happened in the past year) to “6” (more than 20 times in the past year). Emotional abuse was measured using five items from the Psychological Aggression subscale (α = .72). An example item was “swore or cursed at him/her.” Physically punitive parenting was measured using six items from the Corporal Punishment subscale (α = .83). An example item was “slapped your child on the hand, arm, or leg.” Neglect was measured using four items from the Neglect subscale (α = .70). An example item was “had to leave your child home alone, even when you thought some other adult should be with him/her.”

Stress response reactivity.

All physiological measures of stress reactivity were collected during the first wave of data collection.

Cortisol reactivity.

Peak cortisol reactivity was measured via salivary cortisol that was collected immediately before and 20-minutes after the first stress task protocol. Saliva samples were collected via the passive drool method for collecting whole saliva with SalivaBio’s 2 mL cryovials and the Saliva Collection Aid (exclusively from Salimetrics, State College, PA). Immediately after collection, samples were frozen at −80°C. Samples were assayed twice using enzyme-linked immunosorbent assay (ELISA), and the average value in μg/dL was utilized for the present analysis. Peak reactivity was calculated by subtracting the baseline level of cortisol (M = .16, SD = .13) from the 20-minutes post-stress cortisol level¹. Thus, higher peak reactivity scores (i.e., larger difference score from baseline to peak, or larger Δ cortisol) represent a greater increase in cortisol levels from baseline to the stress task (i.e., more HPA activation).

SNS reactivity.

PEP reactivity was utilized to measure the function of the SNS in response to an acute stressor. Cardiac impedance was measured by utilizing ECG data using the BioNex system from MindWare Technologies (Gahanna, OH), and the MindWare IMP 3.1.4 Software module. Impedance cardiography isolates the sympathetic influence on the heart by measuring the opposition to a small constant electrical current, which is generated through four electrodes that are placed on the front and back of the participant’s thorax. This current is modulated, or opposed, by the amount of blood in the chest. The pre-ejection period (PEP) indicates the time interval between the initial electrical stimulation of the heart (onset of the R peak) and the opening of the aortic valve (B point of the dZ/dT wave) [54]. Longer PEP intervals indicate less SNS activity, while PEP shortening indicates more SNS activity. Impedance data were ensemble-averaged in 30-s epochs in combination with R waves that were obtained from the electrocardiogram. We used the MAD/MED algorithm to detect physically improbable IBIs, and peak detection algorithm was set to detect dynamic R-peaks. The Q point was set using the minimum K-R value method, and the B point was detected using the max slope change method. Then, trained research assistants cross inspected the data to correct abnormal R-R intervals, inadvertent cardiac fluctuations, and ectopic beats due to physical movement or breathing. Research assistants (RAs) were trained as a group and then had one-on-one trainings with the same researcher. The RAs also viewed written guides and videotaped materials to supplement their training. During these initial sessions, all RAs were trained to analyze ECG data from the same three participants. After individual trainings concluded, RAs were assigned an additional three participants’ data and asked to analyze the ECG data independently. The training was considered complete if the data cleaned by RAs matched the original data cleaned by the researcher. Additionally, during independent data cleaning, RAs were asked to take notes on details of their operations (i.e., segment and time when a revision was made, the specific operation, the change of RSA before and after the revision). We frequently checked these data cleaning notes for the purpose of quality control.

Mean values of PEP across the 30-s epochs were calculated for the baseline and stress conflict task, respectively. The mean baseline PEP level for the sample was 77.42 (SD = 24.75). To measure PEP reactivity (PEP-R), a residualized change score was created using the mean level of PEP during the rest period and during the stress task, with a higher change score representing less SNS activity:

$$\Delta PEP=\frac{PE{P}_{Stress}-PE{P}_{Baseline}}{SD\left(PE{P}_{Baseline}\right)\times \sqrt{1-r\left(PE{P}_{Stress},PE{P}_{Baseline}\right)}}$$

Vagal withdrawal.

High-frequency heart rate variability (HF HRV) reactivity was utilized to measure the function of the PNS in response to an acute stressor (i.e., vagal withdrawal) All procedures were in accordance with current standards for measuring HF HRV in psychophysiological research [55]. We utilized the BioNex system from MindWare Technologies (Gahanna, OH) and the MindWare HRV 3.1.4 Software module to obtain and digitize HRV data. Respiration was derived using spectral analysis of thoracic impedance [56]. First, ECG data was filtered using a .5–45 Hz bandpass, in order to remove noise related to movement and baseline drift. Then, inter-beat intervals (IBIs) were converted into 120s segments using an interpolation algorithm provided by the MindWare HRV software. Then, the Fast Fourier Transformation was used to convert the time-series domain to frequency-domain. Finally, high-frequency components of HRV were captured via power spectrum analysis to isolate PNS activity. According to previous research, high-frequency components of HRV can be used to measure PNS activity [22]. The high-frequency bandpass was set at .27 – .50 Hz for youth aged 9 years old, .25 – .50 Hz for youth aged 10 years old, and .23 – .50 for youth aged 11– 12 years old [57], with a sample rate set at 1000 Hz. In order to detect and remove noise from the heart rate data that was not previously filtered by the autonomic bandpass filter, trained researchers inspected the ECG data for artifacts (e.g., extra beats, double beats), and subsequently corrected the time series. Data cleaning and training procedures were consistent with those outlined above for PEP reactivity.

HRV was calculated as the natural log of the high-frequency power. To calculate HF HRV reactivity (HF HRV-R), a residualized difference score was calculated using the youth’s mean HF HRV during the rest period and the stress task. This type of calculation allows for the adjustment of the typical variation in baseline HF HRV [55]. The average baseline logarithmic HF HRV in the sample was 6.34 (SD = 1.33). Lower HRV-R residualized change scores indicate a decrease in PNS influence from baseline to the stress task (i.e., vagal withdrawal). This is typically indicative of more self-regulation, though it should be noted that extreme withdrawal can also be a form of dysregulation [58]. Alternatively, higher ΔHF HRV scores represent a lack of vagal withdrawal during stress. The equation to calculate ΔHF HRV is below:

$$\Delta HFHRV=\frac{HR{V}_{Stress}-HR{V}_{Baseline}}{SD\left(HR{V}_{Baseline}\right)\times \sqrt{1-r\left(HR{V}_{Stress},HR{V}_{Baseline}\right)}}$$

Internalizing symptoms.

Youth self-reported on their depressive symptoms using an 11-item version of the Center for Epidemiological Studies Depression Scale (CES-D) [59, 60] at Wave 1 and Wave 2. A total sum score was created from ten items. One item, “I felt everything I did was an effort”, was dropped from the measure at Wave 1 and Wave 2 in order to improve internal consistency. All items were on a Likert scale that ranged from “0” (not at all) to “3” (nearly every day), and youth reported on how often they experienced symptoms in the past two weeks. An example item is “I felt sad”. The scale exhibited adequate internal consistency in the present sample at Wave 1 (α = .72) and Wave 2 (α = .73).

At Wave 2, parents reported on youths’ internalizing symptoms with the Child Behavior Checklist (CBCL) [61]. Parents were instructed to report whether items described their child’s behavior in the last six months. All items were assessed on a Likert scale that ranged from “1” (not true, as far as you know) to “2” (very true or often true). The Withdrawn and Anxious/Depressed subscales were utilized for the present study. The Withdrawn subscale consisted of nine items, including “Would rather be alone” and “Refuses to talk”. The Anxious/Depressed subscale consisted of 14 items, including “Cries a lot” and “Too fearful or anxious”. The internal consistencies for the Withdrawn and Anxious/Depressed subscales were α = .74 and α = .86, respectively.

Control variables.

Control variables included youths’ age, sex, race, BMI, and pubertal stage. Primary caregivers reported on child’s race and ethnicity (coded as 1 = African-American and 0 = Other). Youth pubertal stage was measured via three self-report questions on the Pubertal Development Scale [62]. Time 1 reports of youth internalizing were also included in the model as a control variable.

Analytic Plan

Structural equation modeling (SEM) was used to test study hypotheses using Mplus version 7.4 with maximum likelihood estimation [63]. A series of SEMs were constructed that evaluated the indirect association between emotional abuse and internalizing symptomology via youths’ stress response reactivity. Mediators were tested separately, as opposed to a parallel mediator model, due to sample size and power considerations. Models 1–3 tested the indirect effect between emotional abuse and internalizing psychopathology through physiological stress reactivity (i.e., cortisol reactivity, PEP reactivity, and vagal withdrawal). For all models, control variables included physically punitive parenting, neglect, sex, age, pubertal stage, race, and BMI. Youths’ depressive symptoms at Wave 1 was also included in the model as a control variable. Nonsignificant covariates were trimmed from the final models to improve model parsimony, in accordance with recommendations in the SEM literature [64]. Models were evaluated based on absolute and relative fit indices. According to published standards, model fit was determined to be adequate if CFI and TLI values were at or above .95, RMSEA was at or below .06, and SRMR was at or below .08 [65]. Indirect effects were tested using the distribution-of-the-product method via RMediation [66, 67].

At Wave 1, the percentage of missing data ranged from zero to 4.95%. Missing data at Wave 1 were due to participant non-response and errors in obtaining physiological data. Specifically, PEP reactivity and HF HRV reactivity were each missing five cases due to errors with the electrocardiogram equipment, the presence of too many artifacts in the ECG data, or administrative error (e.g., data not saved correctly). There were two cases missing cortisol reactivity due to errors in collection (i.e., not enough saliva collected). At Wave 2, there were missing data for 30 participants (29.7%). All missing data at Wave 2 were due to participant attrition. An attrition analysis found that participant dropout was significantly related to age, such that youth who dropped out of the study were younger than youth who completed both waves, t (98) = 2.49, p < .05. Data missingness was also tested using Little’s Missing Completely at Random (MCAR) test, χ² (113) = 80.25, p = .99. Although the Little’s test indicated MCAR, data were treated as missing at random (MAR) due to results from the attrition analysis. Thus, subsequent hypothesis testing utilized FIML, which utilizes all cases when estimating parameters [68].

Results

Descriptive Statistics and Bivariate Correlations

Variables were first inspected for normality. Corporal punishment and neglect at Wave 1 were positively skewed at 3.39 and 4.14, and subsequently were transformed by calculating their logarithmic value. Cortisol peak reactivity was negatively skewed (−3.78) and subsequently log transformed (log[1-x]) as recommended for negative skewness [69]. This variable was then multiplied by −1 to retain directions of effect. We inspected the data for outliers and determined that there were no outliers due to data coding error. There were three variables with data points more than 3 SD from the mean (PEP-R, HRV-R, and CBCL Anxious/Depressed subscale). We conducted analyses both with and without these datapoints recoded (±3 SD from the mean). Results did not differ, and thus we present results using the full data. Descriptive statistics and bivariate correlations were then calculated for all study variables (See Table 1).

Table 1.

Bivariate Correlations (N = 101)

	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
1. W1 Corp. Pun.	-
2. W1 Neglect	.45***	-
3. W1 EA	.43***	.23*	-
4. W1 CESD	−.04	.02	−.02	-
5. W1 INT	.03	.10	.43***	.05	-
6. W1 ΔCortisol	−.09	−.07	−.23*	−.07	−.12	-
7. W1 ΔPEP	−.17	−.08	.05	−.07	.08	−.03	-
8. W1 ΔHF HRV	.21*	.03	.22*	.06	.07	−.18	−.03	-
9. W2 CESD	−.05	.31**	.16	.28*	.04	−.29*	−.10	.08	-
10. W2 Anx/Dep	−.07	.23	.21	.25*	.28*	−.22	.15	.19	.39**	-
11. W2 With.	−.08	.17	.42***	.07	.48***	−.36**	.12	.02	.47***	.43***	-
12. Age	.06	.03	.24*	−.16	.30**	.08	−.01	.16	.25*	.15	.17	-
13. Sex	.06	−.03	.02	−.04	−.02	.07	−.04	.10	.08	.05	−.09	.14	-
14. Race	−.02	−.23*	−.00	.06	−.04	.14	−.01	−.09	−.27*	−.21	−.10	−.20	−.05	-
15. Puberty Status	.09	.00	.20*	−.07	.13	.10	−.07	.12	.12	.05	−.01	.52***	.39***	−.13	-
16. BMI	.04	−.02	.14	.16	.25*	.03	−.03	−.03	.14	−.15	.04	.30**	.04	−.05	.19	-
Mean	8.16	3.66	21.74	4.77	37.21	−.03	−.06	−1.35	3.90	1.72	1.84	10.28	.52	.75	6.89	21.96
SD	14.79	10.43	22.39	4.73	6.28	.13	1.39	1.42	4.01	2.74	2.34	1.19	.50	.43	1.95	6.41

Note. W1 = Wave 1; W2 = Wave 2; EA = Emotional abuse; CESD = Center for Epidemiological Studies Depression Inventory for Children; INT = CBCL Internalizing subscale; ΔPEP = pre-ejection period reactivity change score; ΔHF HRV = heart rate variability reactivity change score; Anx/Dep = CBCL Anxious/Depressed subscale; With= CBCL withdrawn subscale; BMI = Body mass index. Point biserial and phi correlations are included for nominal variables (sex and race).

^*p < .05,

^**p < .01,

^***p < .001

Confirmatory Factor Analysis of Internalizing Symptoms

A confirmatory factor analysis supported a latent variable of internalizing symptomology at Wave 2 with three observed factors (youth-reported depressive symptoms, CBCL depressive and anxiety symptom subscale, CBCL withdrawn symptom subscale). The model was just-identified, CFI = 1.00, TLI = 1.00, SRMR = .00, RMSEA = .00. Factor loadings ranged from λ = .59 to λ = .74 and all were significant at p < .001.

Indirect Effect Models

After examining descriptive statistics and bivariate correlations, indirect effect hypotheses were tested using a SEM framework. In Model 1 (see Figure 1), cortisol peak reactivity was tested as a mediator in the pathway between child maltreatment and adolescent internalizing symptoms. The model exhibited excellent fit, χ² (11) = 12.18, p = .35, CFI = .98, TLI = .96, SRMR = .06, RMSEA = .03. Emotional abuse was significantly associated with blunted cortisol reactivity, β = −.28, p < 01, 95% CI [−.47, −.09], and blunted cortisol reactivity in turn was associated with elevated internalizing symptoms, β = −.25, p < .05, 95% CI [−.49, −.02]. There was also a significant direct association between emotional abuse at W1 and W2 internalizing symptoms, β = .30, p < .05, 95% CI [.04, .56]. W1 depressive symptoms and age were both significantly associated with W2 internalizing symptoms, respectively: β = .37, p < .01, 95% CI [.11, .63]; β = .35, p < .01, 95% CI [.12, .59]. Race was also associated with internalizing symptoms such that African American youth were less likely to exhibit internalizing, β = −.24, p < .05, 95% CI [−.47, −.01]. Lastly, age and race were both associated with cortisol reactivity such that youth who were younger and White were more likely to have lower peak cortisol reactivity, respectively: β = .23, p < .05, 95% CI [.04, .43]; β = .19, p = .06, 95% CI [−.01, .38]. There was a significant indirect effect between emotional abuse and youth internalizing symptoms via peak cortisol reactivity, α*β = .07, SE = .01, p < .05, 95% CI [.001, .034]. The model explained 53% of the variance in internalizing symptomology.

Figure 1.

Structural equation model testing cortisol reactivity as a mediator between emotional abuse and youth internalizing symptoms.

Note. Bolded lines represent paths testing study hypotheses. Standardized parameters are indicated. The model exhibited excellent fit, χ² (11) = 12.18, p = .35, CFI = .98, TLI = .96, SRMR = .06, RMSEA = .03.

*p < .05, **p < .01, ***p < .001

In Model 2, the PNS stress response was tested as a mediator in the pathway between child maltreatment and youth internalizing symptoms. The model exhibited excellent fit: χ² (10) = 9.70, p = .47, CFI = 1.00, TLI = 1.00, SRMR = .05, RMSEA = .00. Emotional abuse at W1 significantly predicted the latent factor for internalizing symptoms at W2, β = .63, p < .001, 95% CI [.38, .88]. Unexpectedly, corporal punishment also predicted internalizing symptomology such that higher rates of corporal punishment predicted lower levels of internalizing symptomology, β = −.40, p < .01, 95% CI [−.66, −.15]. However, we conducted additional analysis to understand this association and concluded that this inverse association was due to a suppression effect, and thus should not be interpreted directly (see footnote for more information²). Emotional abuse at W1 was significantly associated with higher ΔHF HRV (i.e., less vagal withdrawal), β = .22, p < .05, 95% CI [.03, .41], but ΔHF HRV did not predict internalizing symptomology at W2, β = .12, p = .35, 95% CI [−.13, .35]. There were no significant indirect effects found in Model 2. The model explained 41% of the variance in internalizing symptomology.

Lastly, Model 3 tested pre-ejection period reactivity, a measure of the SNS response to stress, as a mediator in the pathway from child maltreatment to youth internalizing symptoms. The model exhibited excellent fit with the data: χ² (10) = 9.68, p = .47, CFI = 1.00, TLI = 1.00, SRMR = .06, RMSEA = .00. Consistent with other models, emotional abuse at W1 predicted heightened internalizing symptomology at W2, β = .63, p < .001, 95% CI [.38, .87]. However, emotional abuse was not associated with SNS reactivity (PEP-R), β = −.02, p = .85, 95% CI [−.22, .18], nor was PEP-R associated with internalizing symptomology, β = .04, p = .75, 95% CI [−.23, .31]. Corporal punishment also predicted internalizing symptomology such that higher rates of corporal punishment predicted lower levels of internalizing symptomology. However, as noted above, this association is likely due to a suppressor effect and should be interpreted with caution. There were no significant indirect effects in the model, and the model explained 38% of the variance in internalizing symptomology at W2.

Post Hoc Analyses

We conducted a post hoc power analysis using a Monte Carlo power simulation via Mplus. Using a significance level set at .05, the power estimates for hypothesized structural paths ranged from .84 to .99, and for the indirect effect was .78. Thus, the model had adequate power to detect our hypothesized pathways.

Further, we tested the robustness of our structural models by adding W1 parental reports of internalizing symptoms (via the CBCL) as an additional covariate. The magnitude and direction of effects in our structural models remained unchanged. However, the model fit for all models with this additional covariate was poor (CFI range = .74 – .81; SRMR = .09). Thus, we report models without this additional covariate.

Discussion

Emotional abuse has continuously been found to be a risk factor for affective psychopathology and other socioemotional problems across the life span [9, 70]. The rate of internalizing problems increase during the transition to adolescence, posing a significant developmental point of vulnerability. One proposed mechanism in the link between childhood adversity, such as emotional abuse, and the development of adolescent affective psychopathology is disruptions to systems of self-regulation. The present study extended upon this body of research by studying the indirect association between emotional abuse and youth internalizing symptoms via physiological stress reactivity (measured in the SNS, PNS, and HPA axis). We found evidence for an indirect association via physiological stress reactivity in the HPA axis, in which childhood emotional abuse and youth internalizing symptoms were associated via blunted levels of cortisol reactivity. Alternatively, we did not find support for the indirect role of physiological stress reactivity in the SNS and PNS, as measured by PEP and HRV reactivity. There was some support, however, for a direct association between emotional abuse and lower levels of vagal withdrawal, although vagal withdrawal was not related to youth internalizing symptomology.

These results highlight the harmful impact of parental emotional abuse on youth development. Our findings corroborate previous research on the damaging consequences of exposure to emotional abuse [9, 70]. Moreover, these findings support an emerging body of research on the deleterious effects of emotional abuse compared to other maltreatment types [71, 72]. Investigators found that, compared with other types of abuse and neglect by parents and peers, parental emotional abuse during early adolescence was the most salient predictor of depression for adolescent boys [72]. Similarly, Cecil and colleagues found that emotional abuse was a robust predictor for psychopathology in a sample of at-risk youth, after adjusting for other types of maltreatment [71]. These previous findings and those from the present study are concerning given that emotional abuse is often under-reported and under-substantiated [73], leaving youth vulnerable to the harmful psychological impacts of the abuse.

Our data provided evidence for the indirect role of blunted physiological stress reactivity in the HPA axis, or hypo-cortisolism, in the association between childhood emotional abuse and youth internalizing symptoms. This finding is not surprising, as the HPA axis has long been considered one of the primary pathways in which chronic stress “gets under the skin” to affect health outcomes [74]. Further, the direction of our finding is consistent with the overall body of literature that shows an association between childhood adversity and blunted cortisol reactivity [34]. Chronic exposure to stress early in life leads to sustained overactivity of physiological stress-response systems, and this overactivation can result in modifications in the development of neuroendocrine regulatory systems, including the down-regulation of cortisol [75]. This is demonstrated by a meta-analysis that found an inverse association between stressor timing and cortisol, such that cortisol output was initially elevated at the time of current or recent stressors but then decreased to concentrations below normal as more time passed since the stressor was experienced [74]. Additionally, we found support for a direct association between emotional abuse and lower levels of vagal withdrawal. This is consistent with prior research, including a study that established a causal link between childhood deprivation via institutionalization and blunted vagal withdrawal during early adolescence [29]. Together, these findings support evolutionary developmental models on the effect of early adversity on youth adjustment [76–78]. For example, the adaptive calibration model suggests that stress responsivity is biologically programmed to respond optimally to environmental demands. In stressful environments, including those in which emotionally abusive caretakers are present, stress response systems may become blunted as an adaptation to the environment.

Our findings on the association between blunted cortisol reactivity and internalizing symptomology are less consistent with the literature, as many studies have shown that depressive symptoms are related to hyper-activity in the HPA axis [35]. This finding may have been due to specific sample characteristics, namely the early adolescent age of our sample. It is possible that the associations between physiological stress reactivity and internalizing symptomology differ based on development stage. Adolescence, in particular, is a time in which the HPA axis undergoes major changes, including an increase in the secretion of cortisol [79]. Thus, relations among child maltreatment, cortisol reactivity, and depression may differ by individuals’ developmental stage. Additionally, most of the studies linking hyper-cortisolism and depression have been conducted with adult samples. Alternatively, in a sample of 12-year old youth, blunted cortisol reactivity was linked to increased risk for socioemotional and behavioral problems, particularly among youth who had been exposed to early life abuse or bullying [80]. Further, Harkness and colleagues found that adolescents with severe depressive symptoms showed a blunted cortisol reaction to social stress, but youth with mild to moderate symptomology showed cortisol hyper-reactivity to social stress [81]. Our finding of an indirect effect between emotional abuse and internalizing symptoms via blunted cortisol reactivity may also be understood in the context of developmental psychopathology, and more specifically, attachment theory [82]. Youth who are emotionally abused may not form secure attachment relationships through which children learn to regulate emotions via coregulation with their caregiver [83]. In turn, an inability to self-regulate when faced with social stressors can lead to psychopathology including internalizing symptoms.

The results of the present study should be considered in light of several limitations. First, our sample was recruited from a non-urban setting in the southeast United States and consisted of families who were at or below 200% of the poverty level. The majority of the sample (75%) identified as African-American. Due to these unique sample characteristics, the results from the present study may not extend to the general population of youth in the United States or other specific youth demographics. However, the sample composition of the present study may also be a strength, given that African American and low-SES families are an understudied population. Second, parent report on the Conflict Tactics Scale was used to assess childhood emotional abuse. This measurement was likely biased due to participants’ hesitation to report adverse parenting practices. However, despite this likelihood of underreporting, the present study did reveal associations between emotional abuse and youth outcomes. These associations would likely be stronger if parents fully reported on emotionally abusive parenting practices. Third, several measures exhibited only fair internal consistency, which may have impacted the study findings by limiting the ability to detect associations between variables. Fourth, due to limited statistical power, we were unable to test a model with all three mediators included. Future research that simultaneously models stress response reactivity in the HPA, SNS, and PNS has the potential to uncover information about the unique contributions of each physiological stress response mediator, and about the interactions between these systems in the etiology of psychopathology. Additionally, our model showed a direct association between corporal punishment and lower levels of internalizing symptoms; however, we found that this association was due to a suppressor effect and thus should not be interpreted directly. We are also limited in our ability to make causal assumptions about the associations between emotional abuse, blunted physiological reactivity, and adolescent internalizing symptomology, as child maltreatment and physiological reactivity were collected at the same time point. Last, although the mental arithmetic task is a well validated social stressor for laboratory settings [25], it does not have perfect ecological validity as it relates to the stress caused by emotional abuse. However, to address this limitation, we invited parents of youth to be part of the audience evaluating the youth during the social stress task.

In spite of these limitations, the present study has several notable implications. Results suggested that emotional abuse is associated with disruptions to physiological stress response systems during adolescence, resulting in a dysregulated response (i.e., under-responsiveness) to acute psychosocial stressors. Thus, clinicians who work with maltreated youth may wish to work with their clients to develop more adaptive responses to acute stress. Mindfulness relaxation [84] and other psychosocial interventions [for a review, see 85] have been shown to produce changes in stress-response systems and therefore, might be incorporated into therapeutic services for maltreated youth in order facilitate more adaptive responses to stress. However, more research is needed to assess whether these interventions are helpful specifically for youth who exhibit hyporeactivity to stress. Additionally, therapeutic techniques that enhance emotion regulation and coping may be particularly beneficial for maltreated youth to improve their ability to regulate when facing acute psychosocial stressors. In particular, there are several interventions that have been found to enhance stress regulation processes (e.g., cortisol production) in samples of maltreated youth, including Attachment and Biobehavioral Catch-Up (ABC) [86] and the Multidimensional Treatment Foster Care for Preschoolers (MTFC-P) intervention [87]. Last, these findings indicate that the primary prevention of emotional abuse will likely have upstream benefits for preventing psychopathology during adolescence. Parent-child psychotherapy is one candidate intervention that may prevent emotional abuse by improving the parent-child relationship [88].

Summary

In conclusion, results of the present study provide further evidence that emotional abuse can negatively affect youth in terms of both physiological and socioemotional outcomes. Specifically, findings shed light on a possible etiological pathway between emotional abuse and youth psychopathology via physiological reactivity to an acute psychosocial stressor. Results provide support for the indirect role of blunted HPA reactivity to a laboratory social stress task in the association between emotional abuse and youth internalizing symptoms one year later. These findings enhance the present literature on the association between early life adversity, stress reactivity, and youth psychopathology by examining three types of stress reactivity (e.g., vagal withdrawal, PEP reactivity, and cortisol reactivity) in a sample of non-urban low-SES adolescents. Ultimately, knowledge on the mechanisms that underlie the association between child maltreatment and adolescent psychopathology can inform clinical work and prevention programs to reduce the rates for psychopathology and increase mental wellness among at-risk adolescents.