Hypothalamic-pituitary-adrenal axis attenuation and obesity risk in sexually abused females

Abstract

Background:

Childhood sexual abuse (CSA) confers elevated risks for obesity in females. Mechanisms that explain this link remain unclear. This study tracked serum basal cortisol levels with body mass index (BMI) from childhood into adulthood to test whether hypothalamic-pituitary-adrenal (HPA) axis attenuation accounts for elevated obesity risks for sexually abused females.

Methods:

Data drew from six timepoints of a longitudinal study of the impact of CSA on development. Participants were females aged 6–16 years at time of study enrollment with substantiated CSA and demographically matched non-abused peers. Analyses included only participants who did not have obesity at study enrollment. Main outcomes were BMI growth trajectories across ages 6–27 (n=150; 66 abused, 84 comparisons) and early adulthood obesity status (ages 20–27; n=133; 62 abused, 71 comparison). HPA axis functioning indicators were intercept and linear slope parameters extracted from multilevel growth trajectories of serum basal cortisol levels across development. Racial-ethnic minority status, parity, steroid medication use, depression history and disordered eating history were covaried.

Results:

While controlling for covariates, multilevel modeling indicated that high initial serum basal cortisol levels in childhood and attenuated cortisol growth rate over time (i.e., HPA axis attenuation) were associated with accelerated BMI accumulation (p<.01). Attenuated cortisol growth rate mediated the effect of CSA on accelerated BMI accumulation and on elevated adulthood obesity rates (p<.05).

Conclusion:

This work establishes a mechanistic association between HPA axis attenuation and obesity, suggesting that trauma treatments for abuse survivors should include interventions that reduce health consequences associated with dysregulated stress physiology.

1. Introduction

Affecting 42.2% of adults (Hales, Carroll, Fryar, & Ogden, 2020) and 18.5% of youth (Hales, Fryar, Carroll, Freedman, & Ogden, 2018), obesity—body mass index (BMI) greater than or equal to 30 kg/m² (National Heart Lung and Blood Institute; National Institutes of Health, 1998)—is a global epidemic. Obesity engenders debilitating physical (Chu et al., 2018) and mental (Chu et al., 2019) health ailments and imposes heavy economic burdens (Chu et al., 2018). Women are disproportionately obese (Hales et al. 2020), which contributes to health disparities (National Heart Lung and Blood Institute; National Institutes of Health, 1998) and confers health risks onto offspring through pregnancy and birth complications (Kulie et al., 2011) and childhood obesity (Whitaker, Wright, Pepe, Seidel, & Dietz, 1997) and its comorbidities (O’Reilly & Reynolds, 2013). Targeted obesity prevention for females is needed; however, precision in such interventions requires a fortified understanding of the causal mechanisms involved in obesity development.

While lifestyle factors, including poor diet and physical inactivity, are targets of obesity interventions (Li et al., 2020; Nga et al., 2019), it is increasingly recognized that stressors, including sexual victimization, pose substantial risk (Van Tu et al., 2020; Felitti, 1991). A growing body of literature provides systematic evidence that childhood sexual abuse (CSA), in particular, confers inordinate risk for obesity in females (Danese & Tan, 2014; Noll, Zeller, Trickett, & Putnam, 2007), with obesity rates reported as high as 60% (Felitti, 1991). Accordingly, studies have linked CSA to obesity-related diseases including cardiorespiratory diseases (Irish, Kobayashi, & Delahanty, 2010) and metabolic syndrome (Lee, Tsenkova, & Carr et al. 2014). Given the roughly 63,000 new cases of substantiated CSA each year in the U.S. (Department of Health and Human Services, 2020) and that 1 in 5 adult females retrospectively report being sexually abused in childhood (Pereda, Guilera, Forns, & Gomez-Benito, 2009), targeted obesity prevention for CSA survivors may have a marked impact on overall obesity rates. Efforts to develop such targeted interventions have been hampered by the scarcity of empirical evidence for plausible mechanisms that explain the relationship between CSA and obesity.

CSA is arguably a highly stressful experience for a child, with the associated betrayal, powerlessness, body violations, and stigmatization further compounding the traumatic stress (Noll, 2021). The hypothalamic-pituitary-adrenal (HPA) axis is a primary stress-response system and thus a plausible mechanism linking CSA to obesity. During the acute phases of stress exposure, the HPA axis is activated through a series of hormonal signaling pathways that culminate in the increased adrenal secretion of glucocorticoids, particularly cortisol, which causes numerous physiological alterations that aid in responding to a stressor (Chrousos & Gold, 1992). Cortisol also functions as negative feedback inhibition on the hypothalamus, pituitary, and other brain structures to suppress HPA axis activity and restore cortisol concentration to basal levels (Ulrich-Lai & Herman 2009). Chronic and prolonged exposure to stressors can interfere with this negative feedback process such that the HPA axis is continually activated, resulting in chronically elevated levels of cortisol. This overactivation can eventually lead to dysregulation and pathophysiology (Pervanidou & Chrousos, 2012). In particular, HPA axis attenuation, characterized by initial hypersecretion followed by subsequent hyposecretion of cortisol over time, can result from glucocorticoid resistance caused by excessive cortisol exposure. HPA axis attenuation is considered a marker of the “biological embedding of stress,” a process by which chronic activation of the stress-response system alters multiple biological and developmental processes and promotes disease processes and illness (McEwen, 2012), and has been observed in conjunction with numerous physical and mental disorders (McEwen & Wingfield, 2003; McEwen, 1998). This precise HPA axis attenuation pattern has been observed in a prospective sample of female survivors of substantiated CSA (Trickett, Noll, Susman, Shenk, & Putnam, 2010). Given that obesity is among the illnesses tied to HPA axis attenuation (McEwen & Wingfield, 2003; McEwen, 1998), it is plausible that HPA axis attenuation plays a role in promoting obesity in CSA survivors.

Data from animal studies indicate that acute increases in cortisol activity associated with severe stress can increase appetite and food intake through stimulating the secretion of orexigenic peptides—e.g., neuropeptide Y and agouti-related proteins (Shimizu et al., 2008). Cortisol activity also stimulates insulin release, which promotes preference for higher-calorie foods (Dallman et al., 2003). Prolonged cortisol activity also increases leptin and insulin secretion and may eventually promote leptin and insulin resistance, both of which are known risk factors for obesity (Sominsky & Spencer, 2014). Additionally, cortisol enhances lipoprotein lipase activity in adipose tissues to promote fat storage (Sominsky & Spencer, 2014), particularly in the visceral region (Dallman, Pecoraro, & la Fleur, 2005). These metabolic disturbances are cumulative over time and can promote excessive weight gain.

The majority of extant studies linking CSA to obesity are cross-sectional and rely on retrospective reporting of CSA (e.g., Felitti et al., 1998), which hampers strong causal inference. Moreover, HPA axis dysregulation often co-occurs with other comorbid factors that may influence obesity risk in the CSA population. For instance, depression, a common sequela of CSA (Noll, 2021), is linked to increased odds of obesity (Stunkard, 2011), increased intake of high-calorie foods (Macht, 2008), physical inactivity (Paluska & Schwenk, 2000), sleep disruption, and endocrine abnormalities (Parker, Schatzberg, & Lyons, 2003). Similarly, disordered eating behaviors, especially binge and bulimic patterns, are common among CSA survivors (Li et al., 2018; Molendijk, Hoek, Brewerton, & Elzinga, 2017) and are tied to obesity (Stunkard, 2011) and HPA axis dysregulation (Gluck, 2006). Given the multiple pathways by which CSA may confer risk for obesity, comprehensive models will include plausible alternative explanations.

In a sample of CSA and demographically matched comparison females, the present study used an accelerated longitudinal cohort design to investigate the association between HPA axis and BMI growth trajectories across childhood into early adulthood. While accounting for potential confounds and alternative mechanisms, the following hypotheses were tested: (1) HPA axis attenuation will be associated with accelerated BMI growth trajectories, and (2) HPA axis attenuation will mediate the relationship between CSA and accelerated BMI growth trajectories and subsequent obesity status.

2. Materials and Method

Data for the present study drew from the Female Growth and Development Study (FGDS; R01 HD072468; PI: Noll) (Trickett, Noll, & Putnam, 2011). FGDS (1987–2019) is a multigenerational cohort study designed to investigate in a diverse female sample the long-term effects of CSA on a variety of developmental outcomes of multiple public health concerns. FGDS utilized an accelerated cross-sequential prospective cohort design (McArdle & Woodcock, 1997), in which cross-sections of development were followed longitudinally to generate an overall picture of development across the life course (Trickett et al., 2011). This powerful methodology allows for the depiction of a large range of the life course using relatively few time points.

2.1. Participants and procedure

Detailed descriptions of the sample, recruitment and screening procedures, and study design have previously been published elsewhere (e.g., Noll et al., 2007; Trickett et al., 2011; Trickett et al., 2010) and briefly recounted herein. Eighty-two females with substantiated sexual abuse were referred by Child Protective Services (CPS) agencies in Washington, D.C. between the years 1987 and 1989. To be eligible for study participation, individuals must 1) have been 6 to 16 years of age, 2) enroll within 6 months of disclosing the abuse to authorities, 3) have had substantiated sexual abuse, including genital contact and/or penetration, 4) have experienced abuse perpetration by a family member (e.g. parent, grandparent, older sibling, uncle), and 5) have a non-abusing caregiver (usually the biological mother) also participate in the study. CPS records indicated that the abuse characteristic of the sample (i.e. age at onset and duration and type of abuse) were similar to comparable information reported in the 1988 National Incidence Study (National Center of Child Abuse and Neglect, 1988; Trickett et al., 2011).

Eighty-four non-abused demographically matched comparison females were recruited via advertisements in newspapers and posters in welfare, childcare, and community facilities in the same neighborhoods as those in which the abused females resided. To be eligible for study participation, comparison individuals must have 1) had no previous contact with CPS agencies, and 2) been demographically similar with one or more same-aged, abused participant. Abused and comparison participants were recruited to be matched on age, race and ethnicity, residing zip codes, predisclosure socioeconomic status (SES), family constellation (1- or 2-parent households), and other nonsexual traumatic events as assessed using the Diagnostic Inventory for Children and Adolescents (Reich, 2000).

Participants were enrolled at mean age of 11 (range 6–16) years. At each assessment, caregivers provided informed consent and participants provided assent until they reached the age of 18, after which participants provided their own consent. A comprehensive assessment protocol that included biological, psychosocial, cognitive, and behavioral assessments with multiple informants and modalities was administered in hospitals, research centers, and participant homes during six key time points across development: childhood/early adolescence (Times 1, 2, 3; mean ages 11, 12 and 13, respectively), mid- to late-adolescence (Times 4, 5; mean ages 18 and 19, respectively), and early adulthood (Time 6; mean age 24).

After the Time 3 assessment, four CSA participants were found to have not met initial eligibility criteria (three experienced non-familial sexual abuse, one exceeded the age range at time of enrollment) and three comparison participants self-reported exposure to CSA. To fortify the sample, 21 new comparison females were enrolled at Time 4 and followed longitudinally. Subsequently, eight additional comparison participants self-reported CSA exposure. These fifteen females (CSA: 4, comparison: 11) were removed from statistical analyses in the current study to maintain adherence to the initial eligibility criteria and avoid potential contamination across the groups (Shenk et al., 2016). Twenty-two (CSA: 12, comparison: 10) participants at Time 1 were obese as indicated by a BMI score at or above the 95^th percentile on the age- and sex-specific population BMI percentiles (Barlow, 2007; Kuczmarski et al., 2002) and were also excluded from analyses in order to ascertain proper temporal sequencing among CSA exposure and obesity. The final sample was N=150 (CSA: 66 [82 – 4 - 12], comparison: 84 [84 – 3 + 21 – 8 - 10]). Power analyses conducted via G*Power 3.1.9.2 software (Faul, Erdfelder, Lang, & Buchner, 2007) demonstrated that, using the small (0.18) group effect on BMI trajectory previously published in this sample in Noll et al. (2007), the current analysis sample was sufficiently powered at β=0.86.

Table 1 contains the demographic characteristics for the analysis sample. The sample was 57% White/Caucasian, 41% Black/African American, 2% Hispanic/Latino, and 0.7% Asian/Pacific Islander, and of low to middle SES with a mean Hollingshead score of 37 (“working class”) (Hollingshead, 1975) at study enrollment.

Table 1.

Baseline characteristics for the analysis sample

	Total (N = 150)	CSA (n = 66)	Comparison (n = 84)	P < |t|
No. (%) racial-ethnic minoritya	65 (43)	23 (35)	42 (50)	.06
SES, mean (SD)b	37 (13)	35 (14)	38 (11)	.17
No. (%) parousc	53 (35)	29 (44)	24 (29)	.05
No. (%) steroid medication used	37 (25)	22 (33)	15 (18)	.03
No. (%) depressione	18 (12)	13 (20)	5 (6)	.01
No. (%) disordered eatingf	18 (12)	10 (15)	8 (10)	.29
HPA attenuation indicators
Initial cortisol level,g mean (SD), μg/dl	6.00 (0.99)	6.01 (1.30)	6.00 (0.67)	.94
Cortisol growth rate,h mean (SD), μg/dl per year	0.29 (0.07)	0.28 (0.08)	0.29 (0.05)	.26
No. with available adulthood (ages 20–27) obesity data	133	62	71
No. (%) with adulthood obesityi	33 (25)	22 (35)	11 (15)	.008

Note: CSA = childhood sexual abuse. SES = socioeconomic status. SD = standard deviation. No. = number. HPA = hypothalamic-pituitary-adrenal axis.

^aRacial-ethnic minority group included Black/African American (93.9%), Hispanic/Latino (4.6%), or Asian/Pacific Islander (1.5%). Group differences were tested via Chi-square difference test.

^bSocioeconomic status (SES) was defined using Hollingshead ratings (Hollingshead, 1975). Group differences were tested via independent samples t-test.

^cParity status was defined as ever having given birth across Times 1–6. Group differences were tested via Chi-square difference test.

^dSteroid medication use was defined as having reported using oral contraceptives or other steroids medications such as stimulants at Times 4, 5, or 6. Group differences were tested via Chi-square difference test.

^eDepression was defined as scoring above the clinical cutoff on the Child Depression Inventory (CDI; score: ≥ 25; Kovacs, 1981) prior to age 19 and on the Beck Depression Inventory (BDI; score: ≥ 20; Beck, Steer, & Brown, 1996) for age 19 and older and assessed as ever having had depression across Times 1–6. Group differences were tested via Chi-square difference test.

^fDisordered eating was defined as scoring above the recommended clinical cutoff of 20 on the Eating Attitudes Test (EAT)-26 (Garner et al., 1982) and assessed as ever having had disordered eating across Times 5 and 6. Group differences were tested via Chi-square difference test.

^gInitial cortisol level was derived from the intercept parameter estimate at age 6 of serum basal cortisol growth trajectories estimated via multilevel modeling with data arrayed by age. Group differences were tested via independent samples t-test.

^hCortisol growth rate was derived from the linear slope parameter estimates of serum basal cortisol growth trajectories estimated via multilevel modeling with data arrayed by age. Group differences were tested via independent samples t-test.

ⁱObesity was defined as having body mass index ≥ 30 kg/m² (National Heart Lung and Blood Institute; National Institutes of Health, 1998). Group differences were tested via Chi-square difference test.

The study received approval from the affiliated institutional review board and was awarded a federal certificate of confidentiality.

2.2. Measures

CSA status was obtained at Time 1 via CPS records. A binary variable designated group membership according to CSA status, with score of “1” representing the CSA group and “0” representing the non-abused comparison group.

Outcome measures were BMI and adulthood obesity status. At each time point, height and weight measurements were obtained by trained study personnel using a calibrated upright Health-O-Meter balance beam scale (model 400GZD; Continental Scale Corp, Bridgeview, IL). Participants were measured once in street clothing without shoes, as outlined in Noll et al. (2007). BMI scores were calculated as weight in kilograms divided by height in meters squared. Obesity classification followed Centers for Disease Control (CDC) guidelines (Barlow, 2007; Kuczmarski et al., 2002), as having an age- and sex-specific BMI percentile of 95 or higher when participants were aged 19 and younger, and a BMI cutoff score of 30 kg/m² or greater when older than 19. Participants received a score of “1” for the adulthood obesity status variable if they had obesity at any of the assessment time points during which they were in the early adulthood period (ages 20–27) and “0” if they never had obesity during this period.

HPA axis functioning assessment involved the collection of morning nonstress (i.e., basal) serum and salivary cortisol levels, as previously documented in (Trickett et al., 2010). At Times 1–3 (years 1987–1992), unbound serum cortisol samples were assessed from blood samples collected using an indwelling catheter inserted into the forearm vein. Samples were collected immediately after the insertion and placement of the needle (i.e., zero minutes) and refrigerated until centrifugation (within 2 hours) and then frozen at −70 degrees Celsius (°C) until assayed by radioimmunoassay in triplicate by Hazleton Laboratories (Vienna, VA).

At Times 4–6, the less invasive salivary cortisol assessment method became more readily available, and the study protocol shifted to salivary assessments in order to minimize discomfort for participants. Participants were instructed to refrain from eating, drinking, brushing teeth, or using any substances such as cigarettes prior to providing samples. Stimulant-free salivary cortisol samples were collected and stored at −70 °C and assayed in duplicate by Salimetrics Laboratories (State College, PA) using a highly sensitive enzyme immunoassay (Trickett et al., 2010). For all participants, samples were collected in the morning on weekdays following a 30-minute resting period after the participant entered the testing area. Eighty-eight percent of the samples were collected before noon and 80% were collected before 11 AM.

In order to conduct longitudinal within-person analyses, salivary cortisol levels were mathematically converted to serum levels using the previously validated formula provided by Salimetrics Laboratories (State College, PA) (Salimetrics, 2003, June 10): y[serum level μg/dl] = 5.177 + 15.132 × [saliva level μg/dl]. This formula was supported by a Food and Drug Administration (FDA) Investigative New Drug (IND) application and has been previously been published in Trickett et al. (2010).

Due to potential variations in obesity risks, racial-ethnic minority (Ogden, Carroll, Lawman, & et al., 2016) (0=White; 1=racial-ethnic minority) and parity (Gunderson & Abrams, 2000) (0=have never given birth; 1=had given birth once or more times) statuses were covaried in analyses. Because steroid medications including stimulants and oral contraceptives have also been linked to obesity (San-Juan-Rodriguez, Bes-Rastrollo, Martinez-Gonzalez et al., 2020) and can influence cortisol levels (Hertel, König, Homuth et al., 2017), steroid medication use was assessed when the sample aged into adolescence (Time 4–6) via a self-reported medication checklist and a specific question regarding contraceptive use within the context of a menstrual history semi-structured interview. As such, a covariate was included in analyses which represented “0” for never used steroid medications or oral contraception, or “1” for have used steroid medications or oral contraception at any point (Time 4, 5 or 6). Due to their potentially confounding effect on the development of obesity, depression and disordered eating histories were also covaried. At each assessment, participants reported their depressive symptoms using the Child Depression Inventory (CDI) (Kovacs, 1981) (<19 years) and the Beck Depression Inventory (BDI) (Beck, Steer, & Brown, 1996) (≥19 years). Population-normed cutoff scores (≥25 for CDI; ≥20 for BDI) categorized depression history (0=never had depression; 1=had moderate to severe depression at any time point). At Times 5 and 6, participants reported on the presence of disordered eating symptoms using the Eating Attitudes Test (EAT)-26 (Garner, Olmsted, Bohr, & Garfinkel, 1982). A clinical cutoff score (total score ≥20) was used to classify disordered eating history (0=never had disordered eating; 1=had elevated levels of disordered eating symptoms at Times 5 or 6).

2.3. Analytic plan

Analyses were performed using SAS 9.4 (SAS Institute, Cary, NC) and SPSS Statistics Version 25.

2.3.1. Descriptive analysis.

Given the cross-sequential design of the FGDS, data were arrayed by age (rather than by time points) from 6 to 27, representing developmental trajectories from childhood into early adulthood. Multilevel modeling (MLM) (Singer & Willett, 2003) via SAS/PROC MIXED procedure estimated average growth trajectories across development for raw cortisol levels and BMI scores. Restricted information maximum likelihood estimation (REML) (Kenward & Roger, 1997) was used to obtain parameter and variance component estimates under assumption of ignorable missing data. This analytical technique is preferable for longitudinal data over other methods such as ordinary least squares regression that use listwise deletion of participants with missing data, which may produce biased parameter estimates where individuals only have available data for a limited portion of the growth trajectory.

To obtain indicators of HPA axis functioning, an unconditional MLM model was constructed to estimate the sample average growth trajectory of raw cortisol scores and individual deviations from the sample trajectory. Random individual intercept and the linear slope parameter estimates, representing serum basal cortisol level in childhood (age 6; “initial cortisol level”) and cortisol growth rate over time (“cortisol growth rate”), respectively, were extracted for use as predictors in subsequent analyses. Higher values of initial cortisol level and lower values of cortisol growth rate indicated greater extent of HPA axis attenuation (i.e., dysregulation) (Susman, 2006).

To model BMI accumulation across development, conditional MLM estimated sample average growth trajectory of raw BMI with random intercept and linear slope and fixed quadratic slope, with covariates included. The extent to which HPA axis functioning indicators accounted for individual variation in the parameter estimates of the BMI trajectory was evaluated by using initial cortisol level or cortisol growth rate as a time-invariant predictor of the BMI intercept and BMI linear slope effects. All predictor and covariate variables were centered on their respective mean values (“mean-centered”) in order to facilitate the interpretation of parameter estimates as coefficients for individuals scoring at “average levels” for the variables.

2.3.2. Mediation analysis.

Mediation was tested using two methods. First, the extent to which HPA axis attenuation accounted for the effect of CSA on the accelerated BMI growth trajectory was estimated by constructing a conditional MLM growth model with CSA status as a Level 2 time-invariant predictor along with covariates. Then, initial cortisol level or cortisol growth rate was added as an additional Level 2 time-invariant predictor to examine whether accounting for HPA axis functioning indicators reduced the CSA effect to statistical insignificance (Baron & Kenny, 1986). Percentage reduction in the CSA effect on BMI growth rate was also calculated.

Second, the extent to which HPA axis attenuation mediated the effect of CSA on adulthood obesity status was tested using indirect effects analysis (Preacher & Hayes, 2008) via Bootstrap Sobel Test. SPSS INDIRECT macro (Preacher & Hayes, 2008) was used to test the relationships (i.e., “paths”) among CSA, cortisol growth rate, and adulthood BMI and obesity status, while controlling for initial cortisol level and covariates. Total indirect effect of CSA on adulthood obesity status through cortisol growth rate, as well as the bias-corrected bootstrap 95% confidence interval for the total indirect effect, were obtained.

3. Results

Table 1 presents the descriptive statistics for the analysis variables. CSA group had significantly higher rates of steroid medication use (33% vs. 18%, χ²=4.76, p=.03), depression (20 vs. 6%, χ²=6.61, p=.01), and adulthood obesity (35% vs. 15%, χ²=7.09, p=.008) relative to the comparison group. The proportion of racial-ethnic minorities was slightly smaller (35% vs. 50%, χ²=3.46, p=.06) and the proportion of having had one or more children was slightly larger (44 vs 29%, χ²=3.82, p=.05) in the CSA group relative to the comparison group. Table 2 presents the correlations among the analysis variables.

Table 2.

Correlations and p-values among variables used in analyses, N=150

	1	2	3	4	5	6	7	8	9
1. Childhood sexual abuse (CSA)	1	--	--	--
2. Initial cortisol levela	0.007 (.94)	1	--	--
3. Cortisol growth rateb	−0.10 (.24)	−0.73 (<.0001)	1	--
4. Adulthood obesityc	0.18 (.04)	0.11 (.21)	−0.30 (.0004)	1
5. Racial-ethnic minorityd	0.15 (.06)	−0.04 (.60)	0.11 (.19)	−0.07 (.43)	1
6. Paritye	0.16 (.05)	−0.04 (.60)	−0.09 (.25)	−0.01 (.89)	−0.20 (.02)	1
7. Steroid medication usef	0.18 (.03)	−.06 (.49)	0.07 (.40)	0.03 (.72)	−0.09 (.26)	−0.03 (.67)	1
8. Depressiong	0.21 (.01)	−.09 (.26)	0.13 (.11)	0.07 (.42)	0.07 (.36)	0.07 (.39)	−0.02 (.80)	1
9. Disordered eatingh	0.09 (.30)	−0.10 (.20)	0.08 (.30)	0.03 (.69)	0.03  (0.69)	−0.02 (.85)	−0.07 (.40)	0.24  (.003)	1

Notes: Results are presented as correlation coefficient r (p−value).

^aInitial cortisol level was derived from the intercept parameter estimate at age 6 of serum basal cortisol growth trajectories estimated via multilevel modeling with data arrayed by age.

^bCortisol growth rate was derived from the linear slope parameter estimates of serum basal cortisol growth trajectories estimated via multilevel modeling with data arrayed by age.

^cProportion of sample with obesity in early adulthood (ages 20–27). N = 133.

^dRacial-ethnic minority group included Black/African American (93.9%), Hispanic/Latino (4.6%), or Asian/Pacific Islander (1.5%).

^eParity status was defined as ever having given birth across Times 1–6.

^fSteroid medication use was defined as having reported using oral contraceptives or other steroids medications such as stimulants at Times 4, 5, or 6.

^gDepression was defined as scoring above the clinical cutoff on the Child Depression Inventory (CDI; score: ≥ 25; Kovacs, 1981) prior to age 19 and on the Beck Depression Inventory (BDI; score: ≥ 20; Beck, Steer, & Brown, 1996) for age 19 and older, assessed as ever having had depression across Times 1–6.

^hDisordered eating was defined as scoring above the recommended clinical cutoff of 20 on the Eating Attitudes Test (EAT)-26 (Garner et al., 1982), assessed as ever having had disordered eating across Times 5 and 6.

3.1. Descriptive analysis

Descriptive analysis tested the associations between HPA axis functioning indicators and BMI growth trajectories across development, controlling for covariates. Table 3 presents the conditional MLM effects of the HPA axis functioning indicators on BMI trajectories. Controlling for covariates, initial cortisol level × BMI linear slope effect was statistically significant (β=0.10, t_1,114=2.46, p=.02). Initial cortisol level × BMI intercept effect was not significantly different from zero. Results indicate that, although serum basal cortisol level was not associated with BMI in childhood, higher initial cortisol level in childhood was associated with an accelerated rate of BMI accumulation across development.

Table 3.

Conditional multilevel modeling results for the effect of hypothalamic-pituitary-adrenal axis functioning indicators on body mass index growth trajectories across development, N=150

Parameter	Initial cortisol level (cortisol intercept)	Cortisol growth rate (cortisol slope)
Fixed effects
Intercepta (γ00)	15.22****	15.29****
(se)	0.72	0.74
HPA (γ01)	−0.42	1.55
(se)	0.31	4.68
Linear Slope of Timeb
Intercept (γ10)	1.08****	1.10****
(se)	0.12	0.13
Time × HPA (γ11)	0.10*	−1.77**
(se)	0.04	0.56
Quadratic Slope of Timec
Intercept (γ20)	−0.02****	−0.02****
(se)	0.004	0.004
Random Effects
Level 1 effect
Residuals (rij)	3.59****	3.58****
Level 2 effects
Intercept (μ0j)	8.57****	8.95****
Covar.(μ0j)(μ1j)	−0.61**	−0.67***
Linear slope (μ1j)	0.17****	0.17****
Fit Statistics
Deviance (-2LL)	2834.8	2816.1
AIC	2842.8	2824.1
BIC	2854.8	2836.2

Note

^****p<.0001

^***p<.001

^**p<.01

^*p<.05

^†p<.10.

HPA = hypothalamic-pituitary-adrenal axis functioning indicator. se = standard error. All models covaried minority status, parity status, steroid medication use, depression history, and disordered eating history. Results for covariate variables are not presented. All independent variables were mean-centered for continuous variables and centered on 0.50 for dichotomous variables.

^aRandom intercept centered on childhood (i.e., age 6).

^bRandom linear slope of time across development, ages 6–27 years.

^cFixed quadratic slope of time across development, ages 6–27 years.

Controlling for covariates, cortisol growth rate × BMI linear slope effect was statistically significant (β=−1.77, t_1,115=−3.17, p=.002). Cortisol growth rate × BMI intercept was not significant. Results indicate that, although cortisol growth rate was not associated with BMI in childhood, attenuated cortisol growth rate (i.e., HPA axis attenuation) was associated with an accelerated rate of BMI accumulation across development.

3.2. Mediation analysis

Table 4 presents the conditional MLM results showing the reduction in CSA effect on the linear growth rate (i.e., linear slope effect) of BMI due to accounting for HPA axis functioning indicators. Controlling for covariates, the original model containing solely CSA as predictor showed a significant CSA × linear slope effect (β=0.20, t_1,126=2.28, p=.02), indicating that CSA participants showed an accelerated rate of BMI accumulation across development.

Table 4.

Conditional multilevel modeling results for the effect of accounting for hypothalamic-pituitaryadrenal axis indicators on the effect of childhood sexual abuse on linear growth rate of body mass index across development, N=150

Parameter	CSA only	CSA and initial cortisol level	CSA and cortisol growth rate
Fixed effects
Intercepta (γ00)	15.34****	15.20****	15.28****
(se)	0.86	0.86	0.90
CSA (γ02)	−0.04	−0.003	−0.01
(se)	0.73	0.72	0.74
HPA (γ01)		−0.43	1.69
(se)		0.31	4.78
Linear Slope of Timeb
Intercept (γ10)	0.95****	0.97****	1.01****
(se)	0.13	0.13	0.14
Time × CSA (γ12)	0.20*	0.19*	0.16†
(se)	0.09	0.09	0.09
Time × HPA (γ11)		0.09*	−1.61**
(se)		0.04	.56
Quadratic Slope of Timec
Intercept (γ20)	−0.02****	−0.02****	−0.02****
(se)	0.004	0.004	0.004
Random Effects
Level 1 effect
Residuals (rij)	3.58****	3.59****	3.58****
Level 2 effects
Intercept (μ0j)	8.97****	8.72****	9.10****
Covar.(μ0j)(μ1j)	−0.68**	−0.62**	−0.68**
Linear slope (μ1j)	0.18****	0.17****	0.16****
Fit Statistics
Deviance (−2LL)	2829.1	2829.1	2812.5
AIC	2837.1	2837.1	2820.5
BIC	2849.2	2849.1	2832.5

Note

^****p<.0001

^***p<.001

^**p<.01

^*p<.05

^†p<.10.

CSA = childhood sexual abuse. HPA = hypothalamic-pituitary-adrenal axis functioning indicator. se = standard error. All models covaried minority status, parity status, steroid medication use, depression history, and disordered eating history. Results for covariate variables are not presented. All independent variables except for CSA were mean-centered for continuous variables and centered on 0.50 for dichotomous variables.

^aRandom intercept centered on childhood (i.e., age 6).

^bRandom linear slope of time across development, ages 6–27 years.

^cFixed quadratic slope of time across development, ages 6–27 years.

When added as an additional predictor to CSA status, both initial cortisol level (β=0.09, t_1,113=2.41, p=.02) and cortisol growth rate (β=−1.61, t_1,115=−2.89, p=.005), each in separate models, showed a statistically significant BMI linear slope effect. However, only the addition of cortisol growth rate had reduced the CSA × BMI linear slope coefficient to statistical insignificance (β=0.16, t_1,124=1.89, p=.06) with a 16% reduction in the effect size r of CSA, indicating that HPA attenuation completely mediated the effect of CSA on the accelerated rate of BMI accumulation across development.

The indirect effects analysis was conducted with 133 participants (abused: 62, comparison: 71), who had obesity status data available during early adulthood (age 20–27 years). Figure 1 presents the mediation model diagrams showing paths and corresponding coefficient estimates for CSA through cortisol growth rate to adulthood obesity status. There were statistically significant effects across all the relationship paths: total effect of CSA on obesity status (path c) β=1.24, SE=0.46, Wald χ²=7.21, p=.007; effect of CSA on cortisol growth rate (path a) β=−0.02, t_1,122=−2.13, p=.04; and effect of cortisol growth rate on obesity status (path b) β=−19.28, Wald χ²=10.27, p=.001. Bootstrap Sobel Test revealed an indirect effect (path ab) of CSA on adulthood obesity status through cortisol growth rate that was statistically significant (bootstrap indirect effect ab=0.38, bias-corrected [BC] 95% confidence interval [CI]: 0.02–1.06), indicating that HPA axis attenuation significantly accounted for the effect of CSA on elevated adulthood obesity rates. Even after accounting for cortisol growth rate, the direct effect of CSA on adulthood obesity remained statistically significant (total effect [path c’] β=1.07, Wald χ²=4.97, p=.03), indicating that HPA attenuation partially mediated the effects of CSA on adulthood obesity status.

Figure 1.

Mediational model for the effect of childhood sexual abuse (CSA) on adulthood (defined as ages 20–27) obesity status through cortisol growth rate across development (N=133, CSA: 62, comparison: 71). There were significant effects of the dichotomous variable childhood sexual abuse (0=non-abused comparison, 1=CSA) on the dichotomous variable adulthood obesity status (body mass index [BMI] greater than or equal to 30 kg/m² as per Centers for Disease Control and Prevention guidelines; 0=no obesity, 1=has obesity; c path); of childhood sexual abuse on the continuous variable cortisol growth rate (a path); and of cortisol growth rate on adulthood obesity status (b path). The indirect effect of childhood sexual abuse on adulthood obesity status was significantly different from zero (bootstrap indirect effect ab path=0.38, bias corrected 95% confidence interval: 0.02–1.06), indicating that cortisol growth rate significantly mediated the effect of childhood sexual abuse on adulthood obesity status. Models covaried initial cortisol level in childhood (age 6), racial-ethnic minority status, parity status, steroid medication use, depression history, and eating disorder history. Participants who had preexisting obesity at the time of study entry were excluded. **p<.01; *p<.05. ^†p<.10.

4. Discussion

The present study provides prospective evidence for the association between HPA axis dysregulation—specifically, attenuation—and increased risk for subsequent obesity in female CSA survivors. The associations remained significant after accounting for several possible confounders— racial-ethnic minority status, parity, steroid medication use—as well as viable accounting for alternative explanations for the development of obesity in the CSA population—namely, a history of depression and disordered eating behaviors. The robustness of the association provides further evidence for the unique role of HPA axis dysregulation in the development of obesity. The prospective, longitudinal data on BMI and serum basal cortisol levels across the first half of the life course in a well-retained sample of females with substantiated CSA and a matched non-abused counterfactual comparison group strengthen the evidence for the hypothesis that HPA axis dysregulation is an important mechanism in the development of obesity in the CSA population.

The association between hypercortisolism (i.e., the precursor of HPA axis attenuation) in childhood and accelerated rate of BMI accumulation across development suggests that HPA axis functioning could be used as a screening tool for the risks for obesity development in stress-exposed populations. Early detection of risks can aid in the timely delivery of interventions that can help to curtail excessive weight gain and promote healthy development. Although much of the hypothesized processes through which stress may promote obesity have focused primarily on the effects of hypercortisolism, the present study further showed that attenuated cortisol growth rate over time (i.e., hypocortisolism) had a stronger effect than did high initial cortisol levels in explaining the accelerated rate of BMI accumulation across development. The Attenuation Hypothesis (Susman, 2006) posits that HPA axis functioning adapts to prolonged periods of cortisol hypersecretion by downregulating basal cortisol secretion to protect the body from cortisol-induced damages. Therefore, the downregulated HPA axis may be an indication of the earlier presence of the upregulated HPA axis and suggests that toxic stress had been sufficiently “embedded” into the physiological system (McEwen, 2012). In conjunction, the metabolic disturbances associated with HPA axis dysregulation would then have accumulated over time, ultimately leading to excessive weight gain and obesity.

As the hub of many different intersecting biopsychosocial systems, the HPA axis may exert its effect on obesity development in several ways. In addition to direct metabolic consequences, prolonged HPA axis upregulation can lead to glucocorticoid resistance, which results in the disinhibition of the proinflammatory pathway and thus elevated levels of inflammation (Danese & Baldwin, 2017; Silverman & Sternberg, 2012). Inflammation can damage leptin and insulin receptors and lead to leptin and insulin resistance akin to that observed in more severe and persistent cases of obesity (Sominsky & Spencer, 2014). Another indirect influence is the unique sequela of CSA—traumatic sexualization—wherein the sexual violation of a child can fundamentally disrupt sexual development in ways that result in distorted and dysfunctional attitudes and behaviors around sex (Noll, 2021) in the form of disrupted sexual schemas and “sexual ambivalence”—the concurrence of sexual preoccupation coupled with sexual avoidance and aversion (Noll, Trickett, & Putnam, 2003). Such disruptions may result in the experience of heightened stress in response to sexual stimuli (i.e., receiving unwanted sexual attention) and the attempt to avoid these stimuli through building a physical barrier of obesity. Clinical literature has documented reports of CSA survivors purposefully gaining excessive weight to render themselves “unattractive,” and thus, deter unwanted sexual attention and advances (Ross, 2009). These possible indirect pathways toward obesity should be further explored in future studies.

Several limitations must be considered when interpreting the findings from this study. Because the FGDS sample included only CPS-substantiated, relatively severe, intrafamilial CSA, the findings may not generalize to other sexual abuse experiences (e.g., self-reported, extra-familial perpetrators, non-contact). Further, because the FGDS sample included only females, findings cannot be extrapolated to males. The historical nature of the data limited the precision or availability of certain indicators. For instance, only one cortisol sample per participant per day had been assessed and diurnal variations could not be examined. Efforts to minimize diurnal variation were made by ensuring that cortisol assessments at each time point were within the same time frame in the mornings during typical weekdays for every participant (Trickett et al., 2010). The change of cortisol collection methods from serum to saliva also presented a methodological challenge; however, the validated formula supported by the FDA IND application was used to convert the salivary levels to serum levels to facilitate the modeling of intraindividual change. The solid evidence base (e.g., Hellhammer, Wust & Kudielka, 2009) showing high correlation between serum and salivary levels (ranging from r=0.62 to 0.96 across the literature) provides further support for the viability of this conversion approach. Since FGDS was not originally designed to focus on obesity, precise information on lifestyle behaviors such as diet, physical activity, and sleep was not systematically collected. Proxies for these lifestyle factors were included as statistical controls in analyses via measures of depression (which includes sleep disturbances, lethargia, and physical inactivity) and disordered eating (which includes poor dietary intake). However, it is recognized that these assessments have limited precision with respect to the assessment of lifestyle factors which could impact the development of obesity over time. Future research should include direct assessments of lifestyle factors in order to test more comprehensive models of the impact of stress on obesity.

With its strong evidence for the role of HPA axis dysregulation in obesity development, the present study provides an important first step toward a mechanistic understanding of the link between CSA and obesity.

5. Conclusions

Through contributing to the knowledge regarding the role of HPA axis in obesity development, the present study provides a basis for additional testing for potential protective processes that could reduce risks and promote health and resilience (Masten, 2001) in CSA survivors. It is worth investigating whether the mitigation of HPA axis dysregulation in CSA survivors might be effective in curtailing excessive weight gain. Potential protective factors that may help individuals to avoid HPA axis dysregulation and subsequent obesity can also be fruitful areas of investigation. These factors may include social support, adaptive coping strategies, and support for developing healthy diet and physical activity patterns. Future research should include clinical trials that test trauma resolution and stress reduction methods aimed at re-regulating HPA axis functioning. Improved understanding of effective treatments that promote healthy HPA axis functioning may improve not only body weight outcomes but also numerous mental and physical health outcomes for CSA survivors as well as other individuals exposed to toxic stress and adversity. Obesity prevention efforts targeting survivors of early-life trauma may hold promise for reducing the overall population rates of obesity.

Highlights.

• Childhood sexual abuse is associated with increased risk for obesity but the mechanisms by which sexual abuse survivors develop obesity is uncertain.

• Hypothalamic-pituitary-adrenal (HPA) axis dysregulation has been observed in survivors but the link between HPA functioning and obesity development has yet to be established.

• HPA axis attenuation, marked by initial hypersecretion followed by subsequent hyposecretion of cortisol, is mechanistic of accelerated BMI accumulation observed across the first half of the lifespan and of elevated obesity rates in early adulthood for females with substantiated childhood sexual abuse.

Abbreviations:

FGDS

Female Growth and Development Study

CSA

Childhood sexual abuse

HPA

Hypothalamic-pituitary-adrenal

CPS

Child Protective Services

BMI

Body mass index

SES

Socioeconomic status

CDC

Centers for Disease Control and Prevention

MLM

Multilevel Modeling

FDA

Food and Drug Administration

IND

Investigative New Drug