Prenatal Substance Exposure, Early Life Adversity, and Parenting: Associations with Adolescent Stress Response

Abstract

We tested a conceptual model examining associations between prenatal substance exposure and adolescent cortisol reactivity profiles in response to an acute social evaluative stressor. We included cortisol reactivity in infancy, and direct and interactive effects of early life adversity and parenting behaviors (sensitivity, harshness) from infancy to early school age on adolescent cortisol reactivity profiles in model testing. Participants were 216 families (51% female children; 116 cocaine-exposed) recruited at birth, oversampled for prenatal substance exposure, and assessed from infancy to early adolescence (EA). Majority of participants self-identified as Black (72% mothers, 57.2% adolescents), and caregivers were primarily lower income (76%), single (86%), and had high school or below education (70%) at recruitment. Latent profile analyses identified three cortisol reactivity patterns including elevated (20.4%), moderate (63.1%) and blunted (16.5%) reactivity groups. Prenatal tobacco exposure was associated with higher likelihood of membership in the elevated reactivity compared to the moderate reactivity group. Higher caregiver sensitivity in early life was associated with lower likelihood of membership in the elevated reactivity group. Prenatal cocaine exposure was associated with higher maternal harshness. Interaction effects among early life adversity and parenting indicated that caregiver sensitivity buffered, and harshness exacerbated, the likelihood that high early adversity would be associated with the elevated and blunted reactivity groups. Results highlight the potential importance of prenatal alcohol and tobacco exposure for cortisol reactivity and the role of parenting as exacerbating or buffering the impact of early life adversity on adolescent stress response.

Introduction

Prenatal cocaine exposure (PCE) is often a marker of polysubstance exposure including alcohol and tobacco, continued adversities throughout childhood, and increased risk for poor caregiving experiences (Min et al., 2017). Prenatal substance exposure including cocaine, alcohol, and tobacco, constitutes an extreme form of prenatal stress that has the potential for lifelong alterations in hypothalamic-pituitary-adrenal (HPA) axis functioning (Horn et al., 2018; Lester & Padbury, 2009). Children experiencing high levels of prenatal polysubstance exposure often experience greater early life adversities that may include continued postnatal substance exposure, greater exposure to community and household violence, more chaotic and unstable caregiving environments, and caregivers who experience greater psychological distress (Eiden et al., 2014; Molnar et al., 2014). These adversities often co-occur and measures of cumulative exposure to these multiple adversities may be more predictive of developmental outcomes than any single risk factor alone (Conradt et al., 2014; Evans et al., 2013). Finally, prenatal substance exposure accounts for unique variance in parenting behaviors such as sensitivity and harshness even in the context of other risks in PCE samples (Eiden et al., 2011; Ettekal et al., 2020). Aspects of parenting behavior such as maternal sensitivity in early childhood have enduring effects on adult functioning (Raby et al., 2015), and both sensitivity and harshness may moderate the association between early adversity and cortisol reactivity in adolescence (Horn et al., 2018). Based on this literature, the goals of the present study were as follows: 1) to use person-centered analyses to identify profiles of HPA reactivity in adolescence; 2) to test a conceptual model predicting these profiles that included pathways from prenatal substance exposure (cocaine, alcohol, tobacco) to early life adversity and parenting behaviors (harshness, sensitivity), and 3) to examine interactive effects of early life adversity and caregiver sensitivity or harshness on adolescent profiles of HPA reactivity.

Prenatal Substance Exposure

Prenatal substance exposure presents a condition of toxic stress to the fetus (Shonkoff et al., 2012) and may alter HPA functioning through a number of different mechanisms (Horn et al., 2018; Lester & Padbury, 2009). These include vasoconstrictive effects of substances such as cocaine, alcohol, and tobacco, reducing blood flow to the developing fetus. This increases risk for fetal hypoxia and ischemia and changes in the neuroendocrine environment of the placenta (Hellemans et al., 2010; Horn et al., 2018; Knopik et al., 2012; Lester & Padbury, 2009) with increased risk for fetal exposure to high levels of glucocorticoids (Horn et al., 2018). The direct association between prenatal substance exposure and cortisol reactivity across development is also supported by fetal origins of health and disease models (Hellemans et al., 2010) and the adaptive calibration model (Del Guidice et al., 2011) that highlight the importance of the prenatal period as one of enhanced vulnerability to stressors with the potential to alter stress response systems.

The HPA system activates a slow cascade of signals that result in release of secretions from the adrenal gland such as cortisol (Granger et al., 2012; Kirschbaum & Hellhammer, 1989). Cortisol has a wide ranging impact on regulating metabolic and immune functions and responding to increased stress (Adam et al., 2017). Both over- (Eiden et al., 2009) and under-activation of the HPA system have been reported in cocaine-exposed compared to non-exposed children (Chaplin et al., 2010; Lester et al., 2010). For instance, there was depressed cortisol response to pain stimuli in a small sample of cocaine compared to non-cocaine exposed neonates (Magnano et al., 1992); higher urinary cortisol among preterm cocaine exposed neonates compared to non-exposed neonates (Scafidi et al., 1996); higher cortisol reactivity in response to frustration in infancy, especially among boys in the current sample (omitted citation, 2009); lower pre-stress but not post-stress cortisol in response to blood draw among 13-month old cocaine exposed compared to non-exposed infants (Jacobson et al., 1999), and no association between cumulative prenatal substance exposure and cortisol reactivity in middle childhood (Conradt et al., 2014). Similar mixed patterns of results have been reported with prenatal tobacco (Eiden et al., 2015, 2020; Ramsay, Bendersky, & Lewis, 1996; Schuetze et al., 2008) and alcohol exposure (Haley et al., 2006; Jacobson et al., 1999; Keiver et al., 2015; Oberlander et al., 2010; Ramsay et al., 1996). These differences in the pattern of findings in early childhood may be partly due to differences in the stressor (pain vs. frustration), measurement differences (e.g., urine vs. saliva), differences due to child age, or the nature of the comparison group. However, both over-and-under-activation of cortisol reactivity seem to reflect potentially “maladaptive” alterations in the stress response system, with studies indicating associations with accelerated biological aging, inflammation, and higher autonomic activity with over-activation (Revesz et al., 2014) and higher depression and anxiety associated with under-activation, particularly in the context of other risks (Badanes et al., 2011; Gunnar et al., 2020; Rudolph et al., 2018).

It is possible that there is a change from over- to under-activation as a function of increasing child age among children experiencing multiple pre-and-postnatal stressors, consistent with models of allostatic load (McEwen, 1998) and toxic stress (Shonkoff et al., 2012). Indeed, there is small but accumulating evidence that patterns of overactivation of the stress response system may follow periods of acute stress exposure but later in development, stress response systems may be hyporeactive (Joos et al., 2019; Trickett et al., 2010). There are only two published studies that have examined the association between PCE and adolescent stress response. In both these samples, PCE occurred in the context of polysubstance use. Chaplin et al. (2015) examined the association between PCE and stress response among adolescents ranging in age from 14–17 years. Results indicated lower cortisol response (using a cortisol difference score from baseline to peak response) among PCE adolescents. Similarly, Lester et al. (2010) divided adolescents into reactive and non-reactive groups based on increases or decrease/no changes from baseline to post-stressor. Results indicated that PCE adolescents had less of a cortisol response (flatter/blunted pattern), and these associations were stronger for more heavily exposed children compared to the non-cocaine-exposed group, indicating consistency across the two studies.

Similar to the literature on PCE, there are a paucity of studies examining associations between prenatal tobacco or alcohol exposure and adolescent HPA functioning. Existing studies have found evidence of HPA dysregulation among adolescents with prenatal alcohol exposure (see Hellemans et al., 2010 for review) and higher levels of cortisol concentrations in hair samples of adolescents with prenatal tobacco exposure (Petimar et al., 2020). Finally, an additional study examining group differences in cortisol reactivity between polysubstance exposed (including cocaine and heroin) and non-exposed adolescents reported less reactivity in exposed compared to non-exposed group (Buckingham-Howes et al., 2016). Thus, the literature on prenatal substance exposure and adolescent cortisol reactivity is quite small but consistent, with most studies indicating lower cortisol reactivity among substance exposed compared to non-exposed adolescents.

Early Life Adversities

Prenatal cocaine and other substance exposure is often a marker of continued postnatal adversities (Conradt et al., 2014). Indeed, substance using women are at particularly high risk for greater psychological distress (Eiden et al., 2007), violence due to higher risk for relational victimization (Eiden et al., 1999), unstable relationships (Lynch & Cicchetti, 1998), and residence in high-risk neighborhoods (Osofsky et al., 1993). PCE children are also likely to experience frequent separations from the primary caregiver, changes in caregiving adults, frequent changes in the living situation, more unstable eating, sleeping, and bathing routines, and lower involvement with male caregivers (Brown et al., & Lynch, 2004). Furthermore, prenatal substance exposure is often a marker for continued substance use after delivery (Eiden et al., 2007).

There are robust associations between cumulative exposure to early life adversity and alterations in HPA functioning. Much of this work has been focused on diurnal cortisol patterns with several studies noting that diurnal patterns are flatter among preschool-school age (Bernard et al., 2015) and preadolescent (Martin, Bruce, & Fisher, 2012; Miles et al., 2018) children experiencing higher levels of adversity. A few studies of cortisol reactivity to a laboratory social-evaluative stressor (e.g., the Trier Social Stress Test) reported either increased reactivity (Mielock et al., 2017; Ouellet-Morin et al., 2018) or flatter patterns among adolescents experiencing higher violence exposure (Busso et al. 2017; Marsman et al., 2012; Peckins et al., 2020) and being institutionalized (see Wade et al., 2020), with some associations evident even after controlling for poverty (Busso et al., 2017). It has been postulated that a blunted pattern may be a marker of allostatic load following exposure to chronic stressors or adversities and reflect an adaptive pattern in the context of stress but less adaptive for overall health and functioning in more adaptive contexts (McEwen, 1998). Thus, early life adversity may be a significant risk factor for alterations in stress reactivity patterns from infancy to adolescence.

Role of Parenting

Maternal cocaine use also has implications for parenting behavior. Both animal and human studies indicate that cocaine use in pregnancy is associated with lower plasma oxytocin (sometimes called the “bonding hormone”) compared to non-substance using mothers (Johns et al., 1997; Light et al., 2004; McMurray et al., 2008). This may be one mechanism for cocaine effects on maternal caregiving or parenting behavior (Light et al., 2004; McMurray et al., 2008). Animal studies examining the effects of cocaine administration on caregiving behavior consistently report lower caregiving quality across multiple domains (Morrell et al., 2011). Similarly, human studies using experimental designs indicate associations between high dose cocaine use and higher aggressive behaviors (Licata et al., 1993). In the realm of parenting, studies report fewer positive reinforcements and more threats of physical discipline in the toddler/preschool period (Bauman & Dougherty, 1983); more harshness during different interactive paradigms at 2 years of age (Eiden et al., 2011); and more hostile and intrusive parenting in a structured teaching situation at 3 years of age (Johnson et al., 2002). It is possible that these associations between cocaine use and parenting in human studies may also be due to the role of childhood trauma or psychological distress that may be predictive of both substance use and parenting behaviors (Mayes & Truman, 2002).

Parenting is a significant predictor of stress reactivity in studies using animal models and human studies (Hostinar et al., 2014; review), although the literature on human studies is more mixed with mostly small associations (Hackman et al., 2018). Chronicity of caregiver harshness across childhood may present a condition of continued stress that significantly impacts children’s stress reactivity, while sensitive parenting may help regulate the stress response system (Hostinar et al., 2014). Indeed, in one randomized clinical trial of an intervention that enhances maternal sensitivity and nurturance, children from CPS-referred families participating in a foster care diversion program who received the intervention had more regulated diurnal cortisol compared to children in the control condition in toddler (Bernard et al., 2015a) and preschool ages (Bernard et al., 2015b), providing strong evidence for a causal linkage between caregiver sensitivity and children’s stress response system. Similarly, in a sample of children from rural, low-income households, higher maternal engagement in infancy was associated with higher cortisol reactivity in response to a social stressor in infancy, but prospectively associated with lower overall cortisol level at toddler age, indicating changes over time in the associations between parenting and patterns of cortisol responses (Blair et al., 2008). There were also no associations between cortisol responses in infancy and toddler age, perhaps indicating less continuity in stress response among children experiencing adversities related to being from low-income households (Blair et al., 2008). Relatedly, samples of preschool to school aged children who experienced harsh parenting practices characterized by higher levels of physical punishment had higher pre-task cortisol and higher cortisol over time (Hastings et al., 2011). However, less is known about the role of harsh or sensitive parenting across early life in predicting adolescent cortisol reactivity.

Interaction of Early Life Adversity and Parenting

Stress buffering (Cohen, 2004) theories highlight the importance of early life adversities and positive interactions with parents as interactive processes that predict stress reactivity and long-term mental health outcomes for children. Indeed, by the latter part of the first year, sensitive and responsive caregiving may largely buffer the HPA system from responding negatively, even under conditions that elicit marked distress and fearful behavior (Hostinar et al., 2014; Spangler & Schieche, 1998). Results of randomized clinical trials provide potent evidence that children in foster care who receive interventions promoting positive or sensitive caregiving do not exhibit a flattened diurnal cortisol response observed among those receiving standard foster care (Bernard et al., 2015a, 2015b; Fisher et al., 2007). Thus, sensitive parenting despite prenatal substance exposure and early life adversities may buffer the HPA system whereas harsh parenting may exacerbate these associations.

Person-Centered Approaches for Cortisol Reactivity

There are significant individual differences in cortisol reactivity (Conradt et al., 2014; Ji et al., 2015), particularly among children who experienced high levels of adversity (Gunnar et al., 2009; Joos et al., 2019). Person-centered approaches are well suited to examining individual differences in patterns of change over time and identify subgroups of individuals with a shared pattern of responses (Lanza & Cooper, 2016; Muthén & Muthén, 2020). There are a small number of studies using person-centered approaches to examine profiles of cortisol reactivity, but none in substance exposed samples. In a study using a toxic stress model to examine risk and protective factors predicting adolescent cortisol reactivity in an urban, low-income sample, there were three profiles of reactivity in early adolescence, an elevated reactive profile (11%), a moderate and non-reactive profile (26%), and a blunted and non-reactive profile (63%; Joos et al., 2019). Higher adolescent exposure to violence and lower levels of family support were associated with membership in the blunted profile. Similarly, in a study examining the role of maltreatment on adolescent cortisol reactivity (Peckins et al., 2015), latent profile analysis indicated three profiles of reactivity, elevated (11%), moderate (55%), and blunted reactivity (34%). History of maltreatment and non-white race was predictive of membership in the blunted profile. Both of these studies of adolescents experiencing high levels of adversity related to low-income and/or maltreatment yielded three profiles of cortisol reactivity. However, these studies did not examine the role of prenatal substance exposure or the role of early life adversity and parenting in substance exposed samples.

Present Study

One major goal was to identify subgroups of children with similar patterns of cortisol reactivity over time. Accordingly, we first conducted latent profile analyses to identify profiles of cortisol reactivity in low-income, prenatally substance exposed adolescents residing in disadvantaged urban communities. We conducted exploratory analyses to examine associations between cortisol reactivity profiles and adolescent perceptions of their own stress/anxiety at multiple time points during the stress paradigm. Finally, we tested a conceptual model (Figure 1) for the association between prenatal substance exposure and adolescent cortisol response to a social evaluative stressor that included direct, indirect, and interactive pathways. This model reflected the following hypotheses: 1) prenatal substance exposure would be associated with higher likelihood of a blunted cortisol reactivity profile during adolescence; 2) the association between prenatal substance exposure and adolescent cortisol reactivity may be indirect via greater exposure to early life adversity and lower caregiver sensitivity/harshness; 3) there would be an interaction of early life adversity and parenting, such that in the context of high early life adversity, low caregiver sensitivity or high harshness may be associated with greater likelihood of membership in a blunted reactivity group, but high sensitivity may have a buffering effect.

Figure 1.

The Conceptual Model

Method

Participants

Participants consisted of 216 mother-child dyads recruited into an ongoing longitudinal study. One-hundred and sixteen dyads were cocaine-exposed and 100 were non-cocaine-exposed (NCE) controls matched on maternal education, maternal race/ethnicity, and infant sex. Biological mothers ranged in age from 18 to 42 (M = 30.52, SD = 5.45). The majority of the mothers were Black (72%), were receiving Temporary Assistance for Needy Families (TANF, 76%) at the time of their first laboratory visit (years 2001–2004), had high school or below education (70%), and were single (86%). In early adolescence (EA), the youth self-identified their race/ethnicity as 57.2% Black, 17.8% Mixed Race, 8.6% Hispanic, 6.6% White, 2.0% American Indian, 0.7% Asian, and 7.1% other. Of the 216 children, 106 (49%) were male. The study received approval from the institutional review boards of the hospitals as well as the primary institutions with which the investigators were affiliated. In addition, informed written consent was obtained from all recruited participants and HIPAA authorization was obtained from all participants after April 2003.

Mothers who participated were more likely to be between 18 and 25 years of age at recruitment (p < .001), and were more likely to have a high school or below high school education (p < .001), compared to those who were eligible but not enrolled. Mothers who participated were also more likely to be in the PCE group (with a participation rate of 91% among PCE group eligible) compared to those who were eligible but not enrolled. Mothers in the PCE group who were eligible but not enrolled in the study were more likely to have children who were placed in non-maternal care. There were no other differences on any demographic variables between those who participated and those who were eligible but not enrolled or between mothers in the PCE group who participated compared to those who did not.

Procedure

Mothers were approached by study staff at two local hospitals serving predominantly low-income families and were invited to participate in a study of maternal health and infant development. Interested and eligible mothers were given detailed information about the study and asked to sign consent forms. Once a family was recruited into the PCE group, the closest matching NCE family was recruited. However, a significantly higher proportion of mothers in the NCE group declined participation or withdrew before formal enrollment, resulting in a smaller number of families in the control group. Of the 4,800 women screened at delivery, 340 were eligible for participation in either group. Of these 340 women, 35% either declined participation or were not enrolled in the study because they expressed initial interest but later withdrew, resulting in a sample of 220 mother-infant dyads. Of these 220 mother-infant dyads, 4 were excluded from analyses (two infants were later diagnosed with fetal alcohol syndrome, one was later diagnosed with shaken baby syndrome, and one infant was severely delayed), resulting in the final sample of 216 dyads.

About 2 weeks after delivery, mother–infant dyads were contacted and scheduled for their first laboratory visit, which took place at the time that their infant was approximately 4–8 weeks old. Follow-up assessments were conducted every 6 months until and including early school age (ESA; M = 5.52 years, SD = .36) and in EA (M = 13.26 years, SD = .82). Laboratory visits, which consisted of a combination of maternal interviews, observations of mother–child interactions, and child assessments occurred at child age 1, 7, 13, 24, 36, 48 months, at least 3 months after child entry into kindergarten (early school age, ESA), middle childhood, and EA. Questionnaire and phone interviews were conducted with mothers at child age 18, 30, 42, and 54 months. Data collected from all assessments except middle childhood were used in the current analyses. By EA, 64 (37%) children had been or currently were in non-biological parental care (foster care or kinship care). All assessments were conducted with the primary caregiver of the child at that time who was identified as the adult who had legal guardianship of the child. Biological mothers were interviewed at the 4- to 8-week assessment in addition to primary caregivers in order to obtain accurate information about prenatal substance use. Participants were compensated for their time in the form of gift certificates, checks, and small gifts at each assessment, with the amount increasing over time. Participant compensation ranged from $35 to caregivers and $10 toys for the infants at the 4–8 week assessment to $50 for caregivers and $50 for adolescents at the EA assessment.

Measures

Identification of Substance Use

Cocaine status was determined by a combination of maternal report and chart review that included urine toxicologies and maternal hair analysis. Urine toxicology results were available for 90% of the families in the NCE group and 92% of the families in the PCE group, and hair samples were collected from all mothers. Mothers who were positive for cocaine based on any of these methods were recruited into the PCE group. Urine toxicologies consisted of standard urine screening for drug level or metabolites of cocaine, opiates, benzodiazepines, and tetrahydrocannabinol. Urine was rated as positive if the quantity of drug or metabolite was greater than 300 g/ml. Hair samples obtained after delivery were screened for cocaine, followed by a gas chromatography/mass spectrometry (GC/MS) confirmation for positive cocaine screens.

The Timeline Follow-Back Interview (TLFB; Sobell et al. 1986) was used to assess maternal report of substance use during pregnancy and postnatally. Participants were provided with a pregnancy and past month calendar at the first laboratory assessment and over the time period since the last assessment for the other postnatal assessments and asked to identify events of personal interest (e.g., holidays, birthdays, vacations) as anchor points to aid recall. This has been established as a reliable and valid method of obtaining longitudinal data on substance use patterns, has good test–retest reliability, and is highly correlated with other intensive self-report measures (Brown et al., 1998). The TLFB yielded data about the average number of days of cocaine use per week, average number of joints smoked per week, average number of cigarettes smoked per week, and average number of standard drinks consumed per week per trimester and during the entire pregnancy. Average number of days per week of cocaine use during pregnancy based on the TLFB ranged from 0 to 6.63 days. Approximately 63% of families in the cocaine group were positive for cocaine or metabolites in hair, 42% were positive based on maternal or infant urine assays, 85% reported cocaine use in pregnancy, and 19% of families in the cocaine group were positive on all four indicators (self-reported use in screening interview; self-reported use in the Timeline Follow-Back interview; maternal or infant urine is positive; maternal hair is positive for cocaine or metabolites), highlighting the importance of multi-method assessments. Self-reported cocaine use decreased substantially over the duration of pregnancy from approximately once per week during the first trimester to once per month during the third trimester, although only 5 women reported use only during the first trimester. The remainder of women in the PCE group reported use throughout pregnancy. The majority of women in the CE group used crack cocaine. Postnatal substance use was computed by taking the average of number of days of cocaine use, number of cigarettes smoked per week, number of standard drinks consumed per week, and number of joints smoked per week at each assessment from 7 months to ESA. The final variables used in model testing included PCE group status (dummy coded, based on maternal reports, maternal or infant urine, or maternal hair assays), the average number of cigarettes per week, average standard drinks per week, and average joints per week.

Early Life Adversity

A composite early life adversity score was created by computing a count variable that included maternal psychopathology, caregiving instability, maternal substance use, and maternal exposure to violence at early infancy (4–8 weeks), late infancy (average of scores from 7 and 13 months), toddler age (24 months), early preschool age (36 months), late-preschool age (48 months), and at least 3 months after entry into kindergarten at ESA. Scores for cumulative early environmental risk could range from 0 to 7 within each time point. Maternal psychopathology was measured using the Brief Symptom Inventory (BSI; Derogatis, 1993), consisting of 53 items rated on a 5-point scale. A positive symptom distress index was created by summing all items and dividing by the number of items endorsed with a positive response with higher scores reflecting higher maternal psychopathology. Scores in the upper quartile were included in the risk composite as having high BSI risk following previous studies (Evans et al., 2013).

Caregiving instability was assessed using the Structured Clinical Interview (SCI; Platzman et al., 2001). The individual items from the SCI were summed into a cluster called caregiving instability and included: separations from the primary caregiver for more than 48 hours, there was no eating, sleeping, or bathing routine, child was fed, bathed less frequently than average, there were custody changes, there was no male caregiver (resident or non-resident) who was involved in caring for the child given previous studies indicating the importance of fathers or other male caregivers involved in caregiving (Cabrera et al., 2014). The caregiving instability scores ranged from 0 to 4 in the current study, with the majority of children in the sample having a score of 0 or 1. Thus, caregiving instability may be viewed as a risk index as opposed to a scale with items that are highly correlated and reflect an underlying construct.

The TLFB (Sobell et al., 1986) was used to assess postnatal maternal substance use as noted above. The following cut-offs were used to dummy code these variables at each time point: any cocaine use, postnatal marijuana use of 7 or more joints per week, postnatal alcohol use of 28 standard drinks or more per week (4 or more drinks per day), and postnatal cigarette use of 70 or more per week (10 cigarettes/day).

The TLFB (Sobell et al., 1986) was also used to measure maternal exposure to any mild, moderate, or severe violence. Although the original interview was a calendar-based method to assess substance use, it has also been used to measure episodes of violence in various studies (e.g., Mignone et al., 2009). At each time, mothers were asked about their witnessing, experiencing, or perpetrating violence using a daily calendar at each assessment point. The total number of days reported were summed within each time point. Given the apparent bimodal distributions at each time point, exposure to violence was dummy-coded (i.e., 0, 1) at each time point. The final composite score for cumulative risk was computed by taking the average of the within time count variables with a theoretical range of 0 to 7 (sample range was 0 – 4.29).

Caregiver Sensitivity and Harshness

Caregiver sensitivity and harshness were assessed using behavioral observation at child ages of 7, 13, 24, 36, and 48 months, and ESA. For all assessments except ESA, caregivers were asked to interact with their children as they normally would at home for 10 minutes in a lab room filled with toys. At the ESA assessment, dyads decorated a picture frame together for 20 minutes (Kochanska & Murray, 2000). These interactions were video recorded and coded using the Parent Child Early Relational Assessment (PCERA), a collection of global 5-point rating scales (Clark et al., 1980; Clark, 1999), with higher scores indicating higher sensitivity and harshness. The scale for caregiver sensitivity included items such as warm, kind tone of voice, expressed enthusiasm, positive affect, contingent responsiveness to child initiations, high flexibility, and low intrusiveness. The scale for caregiver harshness consisted of items such as angry and hostile tone of voice, expressed harshness, angry and hostile mood, and displeasure or disapproval or criticism. Two sets of coders blind to group status rated maternal harshness at different ages. Both coders were trained on the PCERA by the first author until the inter-rater reliability criterion was reached (agreement of 90% or above). Interrater reliability for 11% to 14% of the interactions at each age ranged from intraclass correlation coefficients of .82 to .98.

Infant Cortisol Reactivity

At 7 months, infant cortisol reactivity was assessed in response to frustration using the arm restraint paradigm taken from the Laboratory Temperament Assessment Battery (Lab-TAB; Goldsmith & Rothbart, 1988). Assessments were scheduled at least a month after vaccinations during a time when infants were normally awake and alert, and families were rescheduled if infants were or had been sick during the past 24 hours. The order of the procedures was as follows: the Time 1 or pretask saliva sample was collected after the infant arrived at the laboratory with all appointments scheduled after 11am; the infant was then seated in a high chair, was hooked up to electrodes (for measurement of heart rate used in the larger project), and watched a Baby Einstein (emotionally neutral) video for 3 minutes; a puppet show was presented for the next 3 minutes, followed by another 3 minutes of the neutral video. The Time 2 saliva sample was collected at the end of the second 3 minutes of the neutral video. The arm restraint procedure was presented next (consisting of two trials), followed by another 3 minutes of the neutral video. The infant was then unhooked from the heart rate monitor and presented with some blocks for a 3-minute play episode, followed by the presentation of four masks lasting a total of 60 seconds (a fear paradigm). The mask paradigm was taken from the Lab-TAB and presented one at a time for 10 seconds each (through an opaque screen, held up by the researcher), with a 5 second break between each presentation. Next, the mother–infant dyad was moved to the floor with a variety of toys for 8 minutes of interaction. This was followed by the Time 3 saliva sample. The infant was measured and weighed, and the mother was interviewed. The Time 4 saliva sample was collected 20 minutes after Time 3. Saliva samples were collected by placing an absorbent dental cotton roll in the mouth of infants. The saliva in the cotton roll was expressed into a storage vial using a 10 cc needleless syringe. Samples were immediately placed in a −80° C freezer and were shipped on dry ice to Salimetrics Laboratories (State College, PA) for assay. On the day of testing, all samples were centrifuged at 1500 × g (3000 RPM) for 15 minutes to remove mucins. All samples were assayed for salivary cortisol using a highly sensitive enzyme immunoassay the U.S. Food and Drug Administration (510 k) has cleared for use as an in vitro diagnostic measure of adrenal function (Salimetrics, State College, PA). The test used 25 μL of saliva, had a lower limit of sensitivity of .007 lg /dL, a range of sensitivity from .007 to 3.0 lg /dL. All samples were assayed in duplicate, average intra- and inter-assay coefficients of variation were less than 5% and 10%, and averaged duplicate scores were used in all statistical analyses. Time 1 cortisol values were used to examine group differences at pre-task. Although this value may not be a true baseline and may reflect some reaction to stresses associated with preparing to leave the house and the cab ride to the laboratory, it does reflect the initial cortisol level before major affective stress. Following previous studies (e.g., Ramsay & Lewis, 2003), cortisol reactivity was indexed by the difference between the peak post-stressor cortisol value (Time 3 or 4) and the pre-task value. Higher values indicated higher cortisol reactivity.

Early Adolescent Stress Response

Overall Procedure.

In EA, a modified version of the Trier Social Stress Test for Children (TSST-C; Buske-Kirschbaum et al., 1997) was used as the acute social stressor. The TSST-C is a behavioral paradigm designed to induce social evaluative stress within a controlled laboratory setting. To account for diurnal cortisol patterns, these procedures were completed in the late afternoon. A questionnaire was administered that included questions about medication use and general health, and adolescents were asked to refrain from brushing teeth, eating or having drinks with dairy or citrus for at least an hour before the appointment. The order of the procedures was as follows: After arriving at the laboratory, adolescents were hooked up to electrodes for collection of autonomic data (for purposes of the larger project) and were instructed to sit in a comfortable chair and relax for 15 minutes. While relaxing, adolescents were directed to a screen showing nature scenes and calming music through the Calm.com website. Adolescents then rated their current level of stress/anxiety using a visual analog scale (T1 stress ratings). The time 1 (T1) saliva sample was collected following these procedures. Afterward, adolescents were escorted into a separate area and the TSST-C was administered. First, adolescents completed the story-telling portion of the task and then rated their stress/anxiety for a second time (T2 stress ratings). Next, adolescents completed the math portion of the task and subsequently rated their stress/anxiety for a third time (T3 stress ratings). Time 2 (T2) saliva sample was collected after the completion of both TSST-C tasks. After relaxing for 20 minutes after the end of the stress paradigm using the Calm.com website, adolescents rated their stress/anxiety for a fourth time (T4 stress ratings) and provided a Time 3 (T3) saliva sample – theoretically indicating reactivity to the stressor. The final saliva sample (T4) and stress ratings (T5) were collected about 20 minutes after the T3 saliva sample – theoretically indicating recovery from the stressor. After the collection of the final sample, adolescents were debriefed and allowed to ask questions (see Table 1). Adolescents were informed that the tasks were exceedingly difficult, that they were not really being evaluated, and that they had performed well.

Table 1.

Timeframe of Adolescent Stress Ratings and Cortisol Collection

Stress Rating	Cortisol Collection
Baseline Relaxation (15 minutes)
T1	T1
Story Telling Task (TSST-C)
T2
Math Task (TSST-C)
T3	T2
Relaxation (20 minutes from the end of TSST-C)
T4	T3
Recovery (20 minutes after Relaxation)
T5	T4

Note. TSST-C: Trier Social Stress Test-Child; T1-T5 refer to Time 1 to Time 5 for saliva collection and/or stress ratings.

Perceived Stress/Anxiety.

Adolescent perceptions of stress/anxiety were measured using a visual analog scale represented by a bipolar line or ruler used to measure a characteristic across a continuum (e.g., Bond & Lader, 1974; Hellhammer & Schubert, 2012). Participants were asked to point to a spot on the ruler ranging from 1 (“Very calm and relaxed”) to 10 (“Very nervous”) with a score of 5 indicating “Neutral” feelings. By marking a spot on the line, subjects indicated their subjective appraisal of their own perceived stress at the current time with higher scores indicating greater perceived stress/anxiety.

Early Adolescent Cortisol Reactivity.

Saliva samples were collected using passive drool method into a cryovial (product # 5004.01-06) and a saliva collection aid (product # 5016.02) purchased from Salimetrics. Samples were placed in – 80 °C freezer and sent to the Institute for Interdisciplinary Salivary Bioscience Research at the University of California, Irvine. Samples were assayed for cortisol using a highly sensitive enzyme immunoassay cleared by the US Food and Drug Administration (510 k) for use as an in vitro diagnostic measure of adrenal function. The test uses 25 μL of saliva and has a lower limit of sensitivity < 0.007 μg/dL and assay range of 0.012–3.00 μg/dL. All samples were assayed in duplicate and averaged scores were used in analyses.

Control Variables

Puberty was measured during EA using the widely used Pubertal Development Scale (PDS; Petersen et al., 1988), consisting of 6 items rated on Likert scales with low scores indicating that pubertal changes had not yet begun and high scores indicating that they were well underway. For both boys and girls, items addressed growth spurts, growth of body hair, skin changes, and their own perception of their overall development. For boys, two additional questions addressed the growth of facial hair and voice deepening. For girls, two additional items addressed breast growth and menstruation. An index was created by summing all items with scores ranging from 1 to 4 (M = 2.50, SD = .59) and higher scores reflecting more pubertal development. The measure has high reliability and is highly correlated with physician ratings of pubertal development (Brooks-Gunn et al., 1987). Mothers also reported their highest level of education which ranged from 8 to 18 years with an average of 11.80 years (SD = 1.91; i.e., 11th grade). Additional potential control variables included biological mother’s age, parity, birth weight, birth length, head circumference, gestational age (extracted from medical records), adolescent medication use (dummy coded yes/no based on adolescent report), hours of sleep the night before laboratory visit (adolescent report), body mass index (from measurement of height and weight during the laboratory assessment), being placed in non-biological (foster or kinship care) parental care between birth through EA, and child chronological age at the EA assessment.

Data transformations

Outliers (± 3 SD from the mean; Kline, 2016) on maternal harshness at 7 months of child age (2 outliers), infant cortisol reactivity at 7 months (4 outliers), and early adolescent stress ratings (for stress ratings 1 and 4, 2 outliers for each) were winsorized to 3 SD with rank order maintained. Early adolescent cortisol data were screened for biologically improbable values above 4 ug/dL (Jacobson et al., 1999) and/or outliers (+ 3 SD from the mean; Gunnar et al., 1996). There were no cases with values above 4 μg/dL. One outlier was detected for Time 1 cortisol, two for Time 2 cortisol, four for Time 3 cortisol, and two for Time 4 cortisol. Outlying scores were winsorized to 3 SD with rank order maintained. The cortisol data were positively skewed and were thus transformed (log 10) before further analyses.

Missing Data and Analytic Strategy

Of the 216 mother-infant dyads, 166 completed assessments for infancy/toddlerhood (7–36 months), 166 completed assessments for early childhood (48–60 months), and 138 had complete data on early adolescent measures. There were no associations between missingness and demographic variables assessed at recruitment or constructs used in this paper. The analyses examining profiles of cortisol reactivity in early adolescence included participants with available EA data (N= 138). Subsequent analyses examining the conceptual model incorporated the full sample (N = 216) and Full Information Maximum Likelihood (FIML) estimation was used to account for missing data.

In the first step of data analysis, latent profiles of cortisol reactivity were estimated using latent profile analysis (LPA) in Mplus (version 8.0; Muthén & Muthén, 1998–2017). Models with varying numbers of classes (i.e., one through six classes) were specified based on the log-transformed cortisol data collected across four measurements which were conducted before and after participants completed the social evaluative stressor task (TSST-C; see Table 1). Maximum likelihood estimation with robust standard errors was used, an estimation method that is robust to skewness and nonnormality (Yuan & Bentler, 2000). To determine the optimal number of classes, fit indices including the Akaike information criterion (AIC), Bayesian information criterion (BIC), sample-size adjusted BIC (aBIC), Lo–Mendell–Rubin adjusted likelihood ratio test (LMR-LRT), Bootstrap likelihood ratio test (BLRT), and entropy were assessed (Collins & Lanza, 2010; Nylund et al., 2007). Models with smaller AIC, BIC and aBIC values indicate better fitting solutions. Significant p values on the LMR-LRT and BLRT indicate that a k class solution fits the data better than a k −1 class solution within a model. Entropy values range from 0 to 1 with values closer to 1 indicating greater classification precision. Finally, the qualitative nature of the classes was assessed to ascertain whether they were interpretable in consideration of extant empirical findings and theory.

After the most optimal solution from LPA was selected, we examined associations between profile membership and adolescent stress ratings. More specifically, adolescent stress ratings (across five measurements; see Table 1) were estimated as distal outcomes and class-specific means were assessed in order to examine mean differences across the cortisol reactivity classes that were identified. To assess mean differences, the Model Constraint command in Mplus was used and pairwise comparisons were made among the cortisol reactivity classes. This model relied on the use of the manual ML 3-step approach. This approach better accounts for classification errors compared to more traditional approaches (e.g., the classify-analyze method) thereby reducing potential biases of the estimated associations between class membership and the auxiliary variables (e.g., predictors or distal outcomes) in the model (see Asparouhov & Muthén, 2014; Lanza et al., 2013; Nylund et al., 2014, 2019; Vermunt, 2010).

The manual 3-step approach was also used to test the conceptual model (see Figure 1). Path analyses were used to estimate direct, indirect and interaction effects on the cortisol reactivity classes that were previously identified, and multinomial logistic regression was used to examine the associations among the classes. To facilitate the interpretations of these effects, all of the continuous predictors in the model were standardized, and the estimated log odds were exponentiated into odds ratios (ORs). Furthermore, to probe for significant interaction effects, we estimated log odds and their corresponding 95% confidence intervals using the Johnson-Neyman (J-N) technique (Bauer & Curran, 2005; Johnson & Neyman, 1936). The J-N technique allows for examination of the magnitude and significance of the simple slope for a range of different values of the moderator (e.g., caregiver sensitivity) instead of examining the simple slope at fixed values (e.g., one standard deviation above or below the mean). Thus, instead of simple slopes examined at values chosen on an arbitrary basis, the J-N technique provides estimations of the simple slope and its corresponding confidence intervals across the range of possible values of a given moderator. A statistically significant interaction effect is indicated by confidence intervals which do not include zero.

Results

Table 2 consists of correlations among study variables and descriptive statistics. The bivariate correlations indicated significant positive associations between average number of drinks/week in pregnancy and higher adolescent stress ratings after relaxation (Time 1) and after TSST (Time 4). Higher average cigarettes/week during pregnancy was also associated with higher stress ratings post-TSST (Time 4). There were no bivariate associations between infant cortisol and EA cortisol. Early life adversity risk index and caregiver harshness were not associated with EA cortisol at any time point. Caregiver sensitivity was negatively correlated with Time 1 cortisol.

Table 2.

Descriptive Statistics and Zero-Order Correlations for Study Variables

Variable	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19
1. Prenatal cocaine exposure (1 = yes)
2. Prenatal average cigarettes/week	.33**
3. Prenatal average drinks/week	.21**	.34**
4. Maternal education	−.11	−.23**	−.21**
5. Early Life Adversity (1M to ESA)	.15*	.24**	.10	−.12
6. Infant cortisol reactivity (7M)	.11	−.04	.09	.03	−.01
7. Pubertal development	.06	−.09	−.09	.10	−.06	.07
8. Child sex	.02	−.09	−.15*	.02	−.06	.01	.55**
9. Caregiver harshness (7M to ESA)	.18*	.08	.09	−.27**	.18**	.04	−.04	−.08
10. Caregiver sensitivity (7m to ESA)	−.11	.01	−.14*	.16*	−.02	.02	.02	.06	−.57**
11. Time 1 cortisol (EA)	.00	−.08	−.15	.07	−.03	−.03	.19*	.13	−.02	−.19*
12. Time 2 cortisol (EA)	.00	−.03	−.05	.09	−.12	.04	.24**	.12	−.15	−.12	.81**
13. Time 3 cortisol (EA)	−.03	.06	−.07	.09	−.07	.11	.21*	.08	−.07	−.05	.57**	.79**
14. Time 4 cortisol (EA)	−.08	−.02	−.10	.03	−.02	.04	.21*	.16	.04	−.09	.64**	.75**	.84**
15. Time 1 stress ratings (EA)	.06	.04	.25**	−.24**	−.01	−.01	.02	−.05	.05	.04	.04	.02	−.08	−.05
16. Time 2 stress ratings (EA)	.07	.16	.10	−.05	−.04	.05	.24**	.21*	.10	.04	.05	.15	.19*	.18*	.24**
17. Time 3 stress ratings (EA)	.02	.13	−.04	−.05	.03	−.05	.11	.13	−.04	.03	−.04	−.04	.03	.02	.07	.60**
18. Time 4 stress ratings (EA)	.02	.21*	.33**	−.18*	−.06	.13	−.02	−.13	.16	.03	−.06	−.02	.06	.02	.27**	.34**	.22*
19. Time 5 stress ratings (EA)	−.06	−.02	.07	.04	.06	.04	.00	−.01	.02	.08	−.02	−.02	.02	.04	.33**	.27**	.02	.19*
M	0.54	26.30	2.09	11.80	0.00	0.10	2.50	1.51	2.49	3.93	0.13	0.15	0.18	0.15	2.05	6.85	6.91	1.97	1.93
SD	0.50	38.38	8.45	1.91	1.01	0.29	0.59	0.50	0.56	0.67	0.11	0.14	0.16	0.12	1.59	2.74	2.89	1.39	1.29

Note. Prenatal cocaine exposure: cocaine group =1, control = 0. Prenatal tobacco exposure = Average cigarettes/week during pregnancy. Prenatal alcohol exposure = Average standard drinks/week during pregnancy. M = Month. ESA = early school age. EA = Early adolescence. ELA = Early life adversity.

^*p < .05

^**p < .01

LPA Examining Cortisol Reactivity in Early Adolescence

LPA models were estimated to examine models with varying numbers of classes (one through six classes), and for each model, multiple fit indices were evaluated (see Table 3). In scenarios in which the information criteria (AIC, BIC, aBIC) continually decrease as the number of classes increase, thus favoring the solution with the most classes, methodologists have recommended the use of scree plots to aid in the selection of more parsimonious solutions (see Collins & Lanza, 2010). Consequently, scree plots were created to assess changes in the information criteria among models with varying numbers of classes, and in order to identify potential elbow points (i.e., solutions in which the changes in the information criteria become relatively smaller). Using this approach, it appeared that the changes in the information criteria were relatively negligible for models with five or more classes (compared to the 4-class model). Consistent with this pattern, the LMR-LRT was also statistically significant for the 3- and 4-class models. Thus, the class-specific means were plotted for the 3- and 4-class models to better ascertain model interpretability. The 3-class model identified three clearly distinct and interpretable groups which were also consistent with findings from previous studies (Joos et al., 2019; Peckins et al., 2015). In the 4-class model, three of the identified classes were very similar in form to the classes identified in the 3-class model; however, the fourth class consisted of a small sample size (n = 7). Consequently, the 3-class model was selected as the optimal solution.

Table 3.

Fit Indices for LPA Models with One to Six Classes

Number of classes	Log Likelihood	AIC	BIC	aBIC	LMR-LRT (p)	BLRT (p)
1	−63.90	143.80	167.21	141.90
2	45.26	−64.51	−26.46	−67.59	0.13	0.00
3	94.15	−152.30	−99.61	−156.56	0.02	0.00
4	132.92	−219.84	−152.52	−225.28	0.03	0.00
5	148.52	−241.05	−159.08	−247.67	0.25	0.00
6	160.87	−255.74	−159.14	−263.54	0.30	0.00

Note. AIC: Akaike information criterion, BIC: Bayesian information criterion, aBIC: Sample-size adjusted Bayesian information criterion, LMR-LRT: Lo-Mendell-Rubin likelihood ratio test, BLRT: Bootstrap likelihood ratio test. Model in bold was selected as the optimal solution.

The classes identified in the 3-class model (see Figure 2) consisted of 16.0% of children (n = 22) with high baseline levels of cortisol which increased during the TSST-C and subsequently declined afterwards (labelled as the elevated reactivity class). About 65.2% of children (n = 90) had moderate baseline levels of cortisol which increased slightly during the TSST-C and subsequently declined afterwards (labelled as the moderate reactivity class). About 18.8% (n = 26) had low baseline levels of cortisol which remained relatively flat (unchanging) during and after the TSST-C (labelled as the blunted, non-reactive class).

Figure 2.

Latent Profile Analysis Results for Three Class Solution

Note: The TSST was administered in approximately ten minutes, corresponding with Time (T) 2 in the figure above (see Table 2 for additional details on the timeframe of the TSST and the adolescent stress ratings and cortisol collection)

Associations between Cortisol Profiles and Stress Ratings

The associations between cortisol profiles and perceived stress/anxiety ratings were examined next and are displayed in Figure 3. Adolescents in the moderate reactivity group had the highest perceived stress/anxiety ratings, with significantly higher stress/anxiety ratings compared to the low reactivity group at all time points before, during, and after the TSST (p = < .001 at all time points). Adolescents in the moderate reactivity group also reported significantly higher stress/anxiety compared to the high reactivity group at all time points except time 2 (post-story telling task of the TSST) with p values ranging from p = .00 to p = .04. There were no significant differences between adolescents in the high reactivity and low reactivity classes in their perceptions of their own stress/anxiety except for time 2 (post-story telling task of the TSST).

Figure 3.

Association Between Cortisol Reactivity Profiles and Perceived Stress/Anxiety Ratings

Note: The TSST was administered in approximately ten minutes, corresponding with Time (T) 2 and 3 in the figure above (see Table 2 for additional details on the timeframe of the TSST and the adolescent stress ratings and cortisol collection)

Testing Conceptual Model

Puberty was significantly correlated with cortisol at all four time points (rs ranged from .19 to .24, ps < .05). Maternal education correlated positively with caregiver sensitivity (r = .16, p < .05), and negatively with maternal harshness (r = −.27, p < .01) and early adolescent stress ratings at time 1 (r = −.24, p < .01) and time 4 (r = −.18, p < .05). Finally, child biological sex was used as an additional covariate given the potential for sex differences in cortisol reactivity. Thus, these variables were included as control variables in subsequent model testing.

Two path models were estimated to examine the direct, indirect, and interactive associations specified in the conceptual model (Figure 1), one with caregiver sensitivity and the second with caregiver harshness. Caregiver sensitivity and harshness associations were estimated in two separate models given the conceptual distinction between them, the potential of sensitivity to buffer the impact of early life adversity on cortisol profiles, and the potential of caregiver harshness to exacerbate these associations. The moderate group was chosen as the referent because moderate levels of cortisol reactivity would be an expected response to a stressor and as noted in the introduction, both elevated and a blunted response pattern may reflect over or under activation of the HPA system. Results from the path model (Figure 4) including caregiver sensitivity indicated a significant direct association between prenatal tobacco exposure (average cigarettes/week in pregnancy) and greater than double the odds (OR = 2.39) of membership in the elevated reactivity class. Higher prenatal tobacco exposure was also associated with higher early life adversity (b = .20, p < .05). Higher caregiver sensitivity was associated with a significantly lower likelihood of membership in the elevated reactivity class in early adolescence (b = −.94, p < .05, OR = .39). Greater number of drinks per day was associated with higher cortisol reactivity in infancy (b = .12, p < .01) and higher cortisol reactivity in infancy was associated with lower odds of membership in the blunted, non-reactive class in early adolescence (b = −.69, p < .05, OR = .50). As hypothesized, there was a significant interaction between caregiver sensitivity and early life adversity indicating that the effect of adversity on cortisol reactivity classes in early adolescence differed by levels of caregiver sensitivity in childhood (see Figure 5). We used the Johnson-Neyman (J-N) technique to further examine the conditional relation between early life adversity and caregiver sensitivity on the cortisol reactivity classes (Bauer & Curran, 2005; Johnson & Neyman, 1936). The simple slope for early life adversity across varying levels of sensitivity on the cortisol reactivity classes in early adolescence are depicted in Figure 5 (see Figure 5a for elevated reactivity class and Figure 5b for blunted, low reactivity class). The estimated log odds were exponentiated to ORs in these figures to facilitate their interpretation. The results indicated that early life adversity increased the likelihood of elevated reactivity when maternal sensitivity was low (i.e., the region of significance consisted of standardized scores ranging from −.70 to −2.70; ORs = 2.5 to 24.0), and similar results were found for the blunted, low reactivity profile (i.e., standardized scores ranging from −.30 to −2.70; ORs = 1.9 to 49.0).

Figure 4.

Estimated Path Model Including Caregiver Sensitivity

Note. Only significant (p < .05) paths are shown. Sex and puberty were included as covariates, but all paths were not significant so they are not depicted in the figure. Within time covariances were included but not depicted for ease of presentation. The LPA model included 3 classes, and the moderate reactivity class (not shown above) was used as the reference group.

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

Figure 5.

Johnson-Neyman Plots Displaying the Interaction Effects of Early Life Adversity on Cortisol Reactivity Moderated by Caregiver Sensitivity/Harshness

Results from the path model with caregiver harshness (Figure 6) was supportive of a causal pathway from lower maternal education and prenatal cocaine exposure to higher caregiver harshness, but there was no main effect of harshness on the cortisol reactivity profiles in early adolescence. Instead, there was an interactive association between early life adversity and caregiver harshness. As shown in Figure 5c, the interaction effect of early life adversity and caregiver harshness was not significantly associated with the high reactivity profile. However, early life adversity was significantly associated with a greater likelihood of blunted, low reactivity at high levels of caregiver harshness (see Figure 5d; the region of significance consisted of standardized scores ranging from 2.20 to 3.60; ORs = 6.7 to 18.9).

Figure 6.

Estimated Path Model with Caregiver Harshness

Note. Only significant (p < .05) paths are shown. Sex and puberty were included as covariates, but all paths were not significant so they are not shown in the figure. Within time covariances were included in the model but not depicted for ease of presentation. The LPA model included 3 classes, and the moderate reactivity class (not shown above) was used as the reference group.

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

Discussion

The purpose of this study was to examine direct and indirect associations between prenatal substance exposure and early adolescent cortisol reactivity via early life adversity and parenting. Our results indicated both direct and indirect associations via caregiver sensitivity on profiles of early adolescent stress reactivity and an interaction of early life adversity and parenting on membership in risk profiles. Results expanded models of toxic stress (Shonkoff et al., 2012) to include the role of prenatal substance exposure and continued postnatal adversity and highlighted the role of parenting as a significant moderator of these associations.

Our results supported three distinct profiles of cortisol reactivity: an elevated reactivity profile, moderate reactivity profile, and a blunted, non-reactive profile. This is consistent with the limited number of prior studies using latent profile analysis to identify profiles of cortisol reactivity in adolescence, with both studies including adolescents who experienced high levels of relational and environmental adversity (Joos et al., 2019; Peckins et al., 2015). In both prior studies, authors identified three latent profiles that were similar to those obtained in the current study, although the distribution of adolescents within each profile differed. The number of adolescents in each profile was similar to those from a study of mostly Black, Hispanic, or Mixed Race adolescents (89%) with a history of maltreatment, with over 50% in the moderate reactivity profile and fewer in the blunted or elevated cortisol profiles. In contrast, although there were similar profiles (elevated, moderate, blunted) in a study of urban, low-income early adolescents (10–12 years of age, Joos et al., 2019), 63% of adolescents were in the blunted, non-reactive profile compared to the other two (elevated and moderate) profiles. In addition, although the range of salivary cortisol levels in the blunted and moderate profiles are similar, the range of cortisol levels over time in the elevated profile is higher in the current sample (.22 to .42 μg/dL) compared to those reported by Joos et al. (2019; .16 to .25 μg/dL). It is possible that these differences may be due to differences in procedures, with adolescents in the Joos et al. (2019) study engaging in other study tasks in the period before the first cortisol assessment resulting in less of an elevation but could also be due to differences in prenatal substance exposure levels. In addition, in all three studies (including the current study), three profiles were identified, which is not consistent with the four profiles (sensitive, buffered, vigilant, unemotional) proposed by the Adaptive Calibration Model (Del Guidice et al., 2011) in the context of high adversity. Perhaps extremely large samples are needed for identification of profiles that have lower base rates even among adolescents facing greater disadvantage. Finally, it is worth noting that although the moderate reactivity class was used as the referent group given the expectation that too high or lack of reactivity may both be problematic, the level of reactivity in the moderate group was fairly small compared to other studies using the TSST where the moderate group had higher reactivity (Helminen et al., 2021). On the one hand, this may reflect the chronic stressors experienced by the children in this sample. On the other hand, children in this group also reported the highest levels of anxiety/stress in response to the TSST. Thus, this raises the possibility that the moderately elevated group may not have exhibited an optimal stress response or be interpreted as being the most healthy group within this sample.

In the test of the conceptual model, there were direct associations between higher prenatal tobacco exposure and more than double the odds of membership in the elevated reactivity (instead of moderate reactivity) profile. These results are generally consistent with higher hair cortisol among adolescents exposed to prenatal tobacco exposure (Petimar et al., 2020) and higher salivary cortisol concentrations among 6-year-old children exposed to both tobacco and cannabis (Cajachugua-Torres et al., 2021). However, these results are not consistent with a previous study indicating no association between prenatal tobacco exposure and slopes of salivary cortisol reactivity or with a more blunted cortisol reactivity pattern in a low-income sample of children exposed to both tobacco and cannabis at early school age (Eiden et al., 2020). Perhaps these differences across studies may be due to the measure of tobacco exposure, with the current study using a continuous measure of exposure instead of group status (exposure vs. not). Our results indicate a dose-response association with cortisol reactivity profiles in early adolescence, extending the existing literature on prenatal tobacco exposure effects on stress response. Although there were no direct associations between prenatal cocaine exposure and adolescent cortisol profiles, prenatal cocaine exposure was associated with higher caregiver harshness consistent with previous studies indicating the potential of maternal cocaine use to interfere with caregiving behaviors in both animal (Heyser et al., 1992; McMurray et al., 2008) and human studies (Johnson et al., 2002; Lagasse et al., 2003; Tronick et al., 2005; Ukeje et al., 2001). The lack of direct associations between PCE and cortisol profiles are contrary to at least two previous studies (Chaplin et al., 2015; Lester et al., 2010) indicating significant associations between PCE and adolescent cortisol responses. Some possible explanations for the lack of associations with PCE in the current sample compared to those previously reported may be child age, sample size, and differences in analytic method. The sample of adolescents in the Chaplin et al. (2015) study were older (14–17 years of age) and it is possible that stress response systems for these children become increasingly hypo-reactive with increasing age in response to accumulating stressors. In contrast, the study by Lester et al. (2010) had a large sample (n = 743) of 11-year-olds and examined this association by dividing children into groups based on their cortisol response (i.e., children who exhibited cortisol increase vs. decrease). However, our results are similar to the results obtained in the same sample (Lester et al., 2010) when examining prenatal cocaine and other substance exposure using the number of different substances as a predictor of individual differences in cortisol reactivity at 11 years of age and indicating no direct associations (Conradt et al., 2014). Thus, when potential effects of PCE were examined in the context of other substances, there were no significant associations with cortisol reactivity, consistent with current results of PCE not accounting for unique variance in cortisol profiles.

Higher caregiver sensitivity was predictive of lower odds of being in the elevated vs. moderate reactivity profile, suggesting a buffering effect of caregiver sensitivity on elevated cortisol response to stress. These results are generally consistent with previous studies highlighting the role of parenting as a proximal predictor of stress reactivity (Hostinar et al., 2014, review), although a recent meta-analysis indicated that this association is fairly small across studies (Hackman et al., 2018). Importantly, caregiver sensitivity and harshness moderated the association between early life adversity and cortisol profiles in theoretically expected ways. These interaction effects are also supported by reviews of the literature on pre and perinatal stress (including substance exposure) with neuroendocrine functioning, highlighting the role of parenting as a moderator of adversity effects on children’s stress response systems (Horn et al., 2018). Results are also consistent with findings from clinical trials indicating that enhancements in maternal sensitivity in early childhood result in more regulated diurnal cortisol at toddler (Bernard et al., 2015a) and preschool ages (Bernard et al., 2015b). The critical role of caregiving quality in the current sample is also similar to those reported from longitudinal studies of children who experienced early deprivation in orphanages and exhibited more blunted stress response systems (Gunnar, 2001). Similarly, Jaffee et al. (2015) reported an interaction of harsh parenting and recent traumatic events on cortisol reactivity in middle childhood. Children who had experienced higher levels of harsh parenting and more traumatic events in the past year had the lowest levels of cortisol reactivity. Finally, results add to a gap in the literature on the role of early social interactions having a long-term effect on the stress response system from infancy to early adolescence. Future studies may examine the social buffering framework posited by Hostinar and colleagues (2014) and examine changes in stress response systems from infancy to adolescence via hormonal and neural mechanisms.

Contrary to previous studies and our hypothesis, our results were not supportive of any significant association between early life adversity and cortisol profiles in adolescence. One explanation may be that the sample as a whole experienced higher levels of adversity given the cumulative, co-occurring risks (Eiden et al., 2007). A second explanation may be that with a few exceptions (e.g., Conradt et al., 2014), many studies of early life adversity and adolescent stress response have not included prenatal substance exposure effects. Finally, it is possible that using a cumulative risk approach obscured more fine-grained associations between specific aspects of adversity and cortisol profiles. Recent theories suggest that childhood experiences that examine differential associations with threat, deprivation, and unpredictability are important to consider (McLaughlin et al., 2014). Although our conceptualization of early life adversity reflected these dimensions, we did not differentiate among them. Results from longitudinal studies using hair cortisol as a marker of overall cortisol output across longer periods of time indicate that results vary by type of stress – threat versus deprivation (Doom et al., 2020), and may vary by unpredictability in the caregiving context (e.g., Gunnar, 2020; Tarullo et al., 2020). This may be an important direction for future studies when examining etiological pathways to stress reactivity among samples at high risk due to prenatal substance exposure. In addition, although using a person-centered approach to examine profiles or reactivity addresses the important question of variability in the stress response instead of examining mean changes (as in an ANOVA framework) and addresses a different question than examinations of overall output (as in Area Under the Curve measures), another method that examines variations in timing or latency of slopes and peak responses has been proposed (Lopez-Duran et al., 2014). Future studies may use this method of growth curve modeling with landmark registration to examine associations between adversity and stress response.

Finally, our exploratory analyses of the association between cortisol profiles and early adolescent perceptions of their own stress/anxiety yielded some unexpected findings. Although the blunted, non-reactive profile was consistent with adolescent ratings of lower anxiety/stress across time compared to the other two profiles, adolescents in the moderate reactivity profile rated themselves as experiencing greater stress or anxiety than did adolescents in the elevated cortisol reactivity profile. In one of the few studies with prenatal substance exposure samples including adolescent stress ratings, Chaplin et al. (2015) reported no significant associations between cortisol response and adolescent ratings of sadness, anxiety, or anger. Perhaps the different time scales of the adolescents’ own perceptions of their stress or anxiety in the moment compared to the slower responding HPA system may account for the lack of correspondence between the moderate/high profiles and their perceptions of their stress/anxiety.

Limitations and Future Research Directions

Although our study had several major strengths (i.e., prospective longitudinal design with high retention among an at-risk sample, multi-method assessment of prenatal exposure, objective observational assessments of parenting, examination of individual differences in stress response, addition of adolescent stress ratings), there were also significant limitations. Although adolescent self-reports validated that there were task related increases in perceived stress, it is unclear the extent to which a laboratory stress task would be reflective of adolescents’ stress response in real-life daily interactions. Similarly, measurements of acute stress responses in the laboratory may not reflect diurnal patterns and these results may not generalize to stress response system functioning in daily life. Although the laboratory paradigm in adolescence was designed to allow enough time for relaxation so that the time 1 cortisol was more reflective of baseline levels, the context of measurement was different from daily experiences of adolescents and the relaxation paradigm may not have been equally effective for all adolescents. In addition, although we had observational assessments of caregiving quality and adversity across multiple time points, peer networks and romantic partners may be increasingly important social buffers of adversity with increasing age (Hostinar et al., 2014). Future studies of adolescent stress responses may include these potentially protective social relationships. Future studies may also examine the role of individual differences in adolescent stress responses in later adolescent outcomes. While LPA is a robust person-centered approach, the “solution” or profiles may not generalize beyond this sample and there is a need for replication with both substance exposed and non-substance exposed samples to understand generalizability. Despite these limitations, the current study adds to a growing literature on the role of prenatal stressors on adolescent stress responses and highlights the important role of parenting in moderating the effects of early life adversities on adolescent stress response.

Data Sharing

The data that support the findings of this study are available from the corresponding author upon reasonable request.