Childhood adversity (CA) encompasses significant stressors that occur in childhood (e.g., physical illnesses, financial insecurity, death of loved ones, maltreatment). Exposure to CA results in developmental cascades that give rise to multifinal outcomes. One such consequence is disruption in interpersonal functioning (e.g., Huh, Kim, Yu, & Chae, Reference Huh, Kim, Yu and Chae2014; Johnson et al., Reference Johnson, Cohen, Gould, Kasen, Brown and Brook2002; Salwen, Hymowitz, Vivian, & O'Leary, Reference Salwen, Hymowitz, Vivian and O'Leary2014), which is in turn linked with sustained difficulties within relationships and increased risk for psychiatric disorders and related conditions (e.g., depressive disorders, suicide attempts; Hames, Hagan, & Joiner, Reference Hames, Hagan and Joiner2013; Johnson et al., Reference Johnson, Cohen, Gould, Kasen, Brown and Brook2002). However, given that not all individuals experience interpersonal difficulties following CA (e.g., Masten, Best, & Garmezy, Reference Masten, Best and Garmezy1990; Masten et al., Reference Masten, Hubbard, Gest, Tellegen, Garmezy and Ramirez1999), it is important to consider factors that may contribute to these differing trajectories, such as the interplay between genetic and environmental influences in shaping development. We propose that youth with high hypothalamic-pituitary-adrenal (HPA) axis-related genetic risk may be more likely to contribute to the occurrence of self-generated interpersonal stressors within their relationships (i.e., interpersonal stress generation) following exposure to interpersonal CA (ICA). In testing this notion, the current study sought to elucidate multilevel factors that may amplify the cascading effects of CA within the interpersonal realm.
Stress Generation
An abundance of literature supports the stress generation model (Hammen, Reference Hammen1991), which posits that individuals with vulnerabilities to depression, through personal characteristics and negative cognitive styles, are prone to precipitate or select into stressful experiences in their lives that have the potential to further increase their vulnerability to depression. These stressors are termed dependent events because their occurrence is, at least in part, due to the individual, and they stand in contrast with independent (i.e., fateful) events, which are not predicted by depression (Hammen, Reference Hammen1991). Importantly, these stress generation effects often occur within the interpersonal domain, and they are closely tied to disruptions in social relationships and interpersonal functioning (Hammen, Reference Hammen2006).
Environmental contributors
Several risk factors have been implicated in the generation of stress, including cognitive and interpersonal factors (e.g., rumination, insecure attachment) and personality traits (e.g., neuroticism; for reviews, see Hammen, Reference Hammen2006; Liu & Alloy, Reference Liu and Alloy2010). CA in particular has received attention as a risk factor for stress generation. Initial studies established that CA prospectively predicts increases in negative life events in adolescents and young adults (Hankin, Reference Hankin2005; Harkness, Lumley, & Truss, Reference Harkness, Lumley and Truss2008; Uhrlass & Gibb, Reference Uhrlass and Gibb2007). This link may partly be due to continuity in contextual factors (e.g., family dysfunction, financial instability) that contribute to the occurrence of CA and continued stress exposure beyond childhood (Hazel, Hammen, Brennan, & Najman, Reference Hazel, Hammen, Brennan and Najman2008; Uliaszek et al., Reference Uliaszek, Zinbarg, Mineka, Craske, Griffith, Sutton and Hammen2012).
However, youth with histories of CA may also be more vulnerable to stress generation. Recent research has supported the role of CA in stress generation, demonstrating that CA predicts increased dependent stress (but not independent stress) in samples of youth and emerging adults (Harkness et al., Reference Harkness, Bagby, Stewart, Larocque, Mazurka, Strauss and Kennedy2015; Kushner, Bagby, & Harkness, Reference Kushner, Bagby and Harkness2017; Liu, Choi, Boland, Mastin, & Alloy, Reference Liu, Choi, Boland, Mastin and Alloy2013). More recent evidence suggests that this effect may be specific to interpersonal stress generation, with Hernandez and colleagues (Reference Hernandez, Trout and Liu2016) showing that CA predicted higher levels of interpersonal dependent stress (but not noninterpersonal dependent or independent stress) in young adults. Notably, CA has been linked with greater stress reactivity, with CA predicting a stronger association between proximal stressors and depression (Kim et al., Reference Kim, Martins, Shmulewitz, Santaella, Wall, Keyes and Hasin2014; McLaughlin, Conron, Koenen, & Gilman, Reference McLaughlin, Conron, Koenen and Gilman2010; Shapero et al., Reference Shapero, Black, Liu, Klugman, Bender, Abramson and Alloy2014; Starr et al., Reference Starr, Dienes, Stroud, Shaw, Li, Mlawer and Huang2017; Starr, Hammen, Conway, Raposa, & Brennan, Reference Starr, Hammen, Conway, Raposa and Brennan2014). In addition to depression, changes in stress reactivity following CA may result in other negative outcomes, such as increased stress generation. Heightened stress reactivity may permeate interpersonal functioning, in turn leading to negative behaviors and interactions that contribute to the incidence of stressful life events within relationships.
Genetic contributors and gene–environment interactions
Research also indicates that exposure to stressful life events and stress generation tendencies may, in part, be linked to genetic vulnerabilities (Kendler, Karkowski, & Prescott, Reference Kendler, Karkowski and Prescott1999; Kendler & Karkowski-Shuman, Reference Kendler and Karkowski-Shuman1997; Plomin, DeFries, & Loehlin, Reference Plomin, DeFries and Loehlin1977; Scarr & McCartney, Reference Scarr and McCartney1983). For example, in several twin studies, Kendler and colleagues (Reference Kendler and Karkowski-Shuman1997, Reference Kendler, Karkowski and Prescott1999) found that individuals with histories of depression experience elevated rates of stressful life events, with genetic factors accounting for about one third of the association between the occurrence of stressors and depressive outcomes. Further, many of the risk factors associated with stress generation processes (e.g., attachment, neuroticism, negative cognitive styles) also appear to be genetically moderated (e.g., Lahey, Reference Lahey2009; Spangler, Johann, Ronai, & Zimmermann, Reference Spangler, Johann, Ronai and Zimmermann2009). Nonetheless, few studies have examined the contribution of genetic factors in interpersonal stress generation processes.
One way in which genetic risk may influence stress generation processes is by modifying the influence of environmental risk. Gene–environment interactions (G×E) have often been examined in the context of CA and psychopathological outcomes (for a review, see Manuck & McCaffery, Reference Manuck and McCaffery2014), but fewer studies have examined stress generation as an outcome. Given the potential role of stress reactivity in stress generation, genes related to stress reactivity may be an important starting point. At present, the existing literature on G×Es and stress generation processes has solely focused on a polymorphic region in the serotonin transporter gene (5-HTTLPR), a variant linked in many studies to stress reactivity (Caspi et al., Reference Caspi, Sugden, Moffitt, Taylor, Craig, Harrington and Poulton2003; Karg, Burmeister, Shedden, & Sen, Reference Karg, Burmeister, Shedden and Sen2011, although also see Culverhouse et al., Reference Culverhouse, Saccone, Horton, Ma, Anstey, Banaschewski and Bierut2018). Harkness and colleagues (Reference Harkness, Bagby, Stewart, Larocque, Mazurka, Strauss and Kennedy2015) examined the interaction of CA and 5-HTTLPR genotype in interpersonal stress generation processes in a sample of youth and young adults. Their results suggested that 5-HTTLPR risk allele status predicted greater levels of dependent interpersonal stress, but only for those who had experienced CA. Other studies have shown that constructs linked to CA (depression, relational security) also predict later interpersonal stress generation for those with high (but not low) genetic vulnerability (Starr, Hammen, Brennan, & Najman, Reference Starr, Hammen, Brennan and Najman2012, Reference Starr, Hammen, Brennan and Najman2013). These findings indicate that genetic risk may amplify the effects of CA on stress generation processes. However, the existing literature is limited in several ways. First, previous studies have exclusively considered a serotonergic genetic variant, and genetic variants from other biological systems involved in the stress response, such as the HPA axis, merit consideration. Furthermore, these studies have used a single-variant candidate gene approach, and recently developed polygenic approaches offer vastly improved statistical power.
HPA Axis and HPA Axis-Related Genetic Variation
The HPA axis facilitates the coordination of biological responses to stressors (for a review, see Gunnar & Quevedo, Reference Gunnar and Quevedo2007). HPA axis dysregulation has been linked to a wide range of negative outcomes (Anda et al., Reference Anda, Felitti, Bremner, Walker, Whitfield, Perry and Giles2006; Guerry & Hastings, Reference Guerry and Hastings2011; McEwen, Reference McEwen1998). Stressors that occur over the course of childhood have been shown to produce changes in HPA axis activity and cortisol levels, altering the typical course of HPA axis development (e.g., Kuras et al., Reference Kuras, Assaf, Thoma, Gianferante, Hanlin, Chen and Rohleder2017; Tarullo & Gunnar, Reference Tarullo and Gunnar2006). Indeed, many have pointed to disruptions in the development of the HPA axis and associated neural structures as key biological mechanisms for stress sensitization and increased depression risk following CA exposure (e.g., Cicchetti & Rogosch, Reference Cicchetti and Rogosch2012; Heim & Nemeroff, Reference Heim and Nemeroff2001; Starr et al., Reference Starr, Dienes, Stroud, Shaw, Li, Mlawer and Huang2017; Tarullo & Gunnar, Reference Tarullo and Gunnar2006). Further, HPA axis dysfunction (measured using cortisol responses) in response to a laboratory stressor has been found to predict stress generation among young adults (Morris, Kouros, Hellman, Rao, & Garber, Reference Morris, Kouros, Hellman, Rao and Garber2014).
HPA axis-linked genetic regions appear to predict both physiological and emotional stress reactivity, which may have implications for stress generation. For instance, research suggests that variation in genotype for the CRHR1 gene, which influences CRH receptors, affects cortisol responses following laboratory stressors in children (Sheikh, Kryski, Smith, Hayden, & Singh, Reference Sheikh, Kryski, Smith, Hayden and Singh2013) and adults (Mahon, Zandi, Potash, Nestadt, & Wand, Reference Mahon, Zandi, Potash, Nestadt and Wand2013). In individuals with a history of CA, the CRHR1 genotype is associated with greater cortisol dysregulation (Cicchetti, Rogosch, & Oshri, Reference Cicchetti, Rogosch and Oshri2011; Heim et al., Reference Heim, Bradley, Mletzko, Deveau, Musselman, Nemeroff and Binder2009). Genetic variation in the FKBP5 genotype is linked with glucocorticoid receptor regulation in response to stressors and is also associated with alterations in cortisol reactivity to laboratory stressors (Luijk et al., Reference Luijk, Velders, Tharner, van Ijzendoorn, Bakermans-Kranenburg, Jaddoe and Tiemeier2010; Zannas & Binder, Reference Zannas and Binder2014). Further, HPA axis dysregulation following stressful events has also been demonstrated in relation to variation in the NRC31 and NRC32 genes, which regulate mineralocorticoid receptors (for a review, see Derijk, Reference Derijk2009). HPA axis-related genotypes are also associated with greater cortisol dysregulation following CA (e.g., Buchmann et al., Reference Buchmann, Holz, Boecker, Blomeyer, Rietschel, Witt and Laucht2014; Cicchetti et al., Reference Cicchetti, Rogosch and Oshri2011; Gerritsen et al., Reference Gerritsen, Milaneschi, Vinkers, van Hemert, van Velzen, Schmaal and Penninx2017; Heim et al., Reference Heim, Bradley, Mletzko, Deveau, Musselman, Nemeroff and Binder2009; Sumner, McLaughlin, Walsh, Sheridan, & Koenen, Reference Sumner, McLaughlin, Walsh, Sheridan and Koenen2014) and have been shown to moderate the effects of CA on various negative outcomes, such as depression, suicide attempts, and posttraumatic stress disorder (e.g., Gerritsen et al., Reference Gerritsen, Milaneschi, Vinkers, van Hemert, van Velzen, Schmaal and Penninx2017; Laucht et al., Reference Laucht, Treutlein, Blomeyer, Buchmann, Schmidt, Esser and Banaschewski2013; Roy, Gorodetsky, Yuan, Goldman, & Enoch, Reference Roy, Gorodetsky, Yuan, Goldman and Enoch2010; Xie et al., Reference Xie, Kranzler, Poling, Stein, Anton, Farrer and Gelernter2010).
Most of these studies have applied single-candidate gene approaches to examine G×E effects; however, this method has recently come under fire following prominent nonreplications (de Vries, Roest, Franzen, Munafo, & Bastiaansen, Reference de Vries, Roest, Franzen, Munafo and Bastiaansen2016; Dick et al., Reference Dick, Agrawal, Keller, Adkins, Aliev, Monroe and Sher2015; Duncan & Keller, Reference Duncan and Keller2011), although the issue remains controversial (Caspi, Hariri, Holmes, Uher, & Moffitt, Reference Caspi, Hariri, Holmes, Uher and Moffitt2010; Karg et al., Reference Karg, Burmeister, Shedden and Sen2011; Vrshek-Schallhorn, Sapuram, & Avery, Reference Vrshek-Schallhorn, Sapuram and Avery2017). To address this issue, several research groups have developed multilocus genetic profile scores (MGPSs), additive indices of risk alleles from various single nucleotide polymorphisms (SNPs) that are selected due to their association with a given biological pathway (e.g., Nikolova, Ferrell, Manuck, & Hariri, Reference Nikolova, Ferrell, Manuck and Hariri2011; Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014; Vrshek-Schallhorn et al., Reference Vrshek-Schallhorn, Stroud, Mineka, Zinbarg, Adam, Redei and Craske2015); this theoretically driven approach of capturing cumulative, polygenic effects through the selection of specific SNPs linked to a specific biological system differs from atheoretical, genome-wide association study (GWAS)-derived polygenic risk scores (e.g., Musliner et al., Reference Musliner, Seifuddin, Judy, Pirooznia, Goes and Zandi2015). By capturing polygenic effects within specific biological systems, MGPSs appear to have greater predictive validity than does examining individual SNPs in isolation. Pagliaccio and colleagues (Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014) recently created an MGPS using 10 SNPs from HPA axis-related genes (CRHR1, NRC31, NRC32, FKBP5) that have been linked to HPA axis dysfunction and depression-related phenotypes. HPA axis-related MGPSs have been shown to predict cortisol reactivity in the context of laboratory stressors and interact with environmental stress (i.e., stressful life events, CA) to predict changes in emotional circuitry within the brain (i.e., amygdala reactivity; Di Iorio et al., Reference Di Iorio, Carey, Michalski, Corral-Frias, Conley, Hariri and Bogdan2017; Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014, Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2015), HPA axis dysregulation (i.e., diurnal cortisol regulation, Starr, Dienes, Li, & Shaw, Reference Starr, Dienes, Li and Shaw2019), and affective outcomes (i.e., depression, Feurer et al., Reference Feurer, McGeary, Knopik, Brick, Palmer and Gibb2017; Starr & Huang, Reference Starr and Huang2018). These effects may also extend to other negative outcomes following CA, such as interpersonal stress generation.
Interpersonal CA, HPA Axis-Related Genetic Risk, and Interpersonal Stress Generation
Interpersonal childhood adversities (ICAs) are comprised of significant stressors that occur over the course of childhood that are interpersonal in nature and/or in consequences (e.g., parental conflict or separation, deaths of loved ones). The effects of ICAs on subsequent interpersonal stress generation may be especially likely to be moderated by HPA axis-related genetic risk. ICAs have been identified as potent predictors of interpersonal stress generation (e.g., Chan, Doan, & Tompson, Reference Chan, Doan and Tompson2014; Hernandez et al., Reference Hernandez, Trout and Liu2016). Moreover, interpersonal stress may serve as a powerful “candidate environment,” with some studies suggesting G×E effects are limited to moderation of interpersonal stress, for both serotonergic and HPA axis-related genes (Feurer et al., Reference Feurer, McGeary, Knopik, Brick, Palmer and Gibb2017; Starr & Huang, Reference Starr and Huang2018; Vrshek-Schallhorn et al., Reference Vrshek-Schallhorn, Stroud, Mineka, Zinbarg, Adam, Redei and Craske2015). For example, Starr and Huang (Reference Starr and Huang2018) found that the effects of ICA (but not noninterpersonal CA) on depression were genetically moderated by HPA axis MGPS, suggesting that genetically vulnerable youth are specifically sensitive to interpersonal adversities. Altogether, these factors may put youth at greater risk for interpersonal stress generation following ICA.
Developmental Considerations
Adolescence is a developmental period marked by a confluence of changes relating to higher rates of psychiatric disorders (e.g., depression) associated with CA and stress generation, increases in stressful life events (especially interpersonal stressors), and alterations in HPA axis activity (Avenevoli, Swendsen, He, Burstein, & Merikangas, Reference Avenevoli, Swendsen, He, Burstein and Merikangas2015; Gunnar, Wewerka, Frenn, Long, & Griggs, Reference Gunnar, Wewerka, Frenn, Long and Griggs2009; Romeo, Reference Romeo2013; Rudolph, Reference Rudolph2002). During adolescence, basal HPA axis functioning shifts, resulting in greater release of related hormones and increased stress reactivity (Gunnar et al., Reference Gunnar, Wewerka, Frenn, Long and Griggs2009). These factors suggest that adolescence may be a sensitive period for stress and HPA axis functioning, so it may serve as an ideal period within which to examine our research questions.
The Current Study
We examined the moderating role of HPA axis-related genetic variation in the association between ICA and interpersonal stress generation in a sample of adolescents. We used an HPA axis-related MGPS based on previously established procedures (Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014) to examine genetic risk. We hypothesized that ICA would predict interpersonal stress generation (i.e., interpersonal dependent stress) but not independent stress or noninterpersonal dependent stress, in line with prior findings. Additionally, we predicted that this association would be specific to adolescents with high (and not low) HPA axis-related genetic vulnerability.
Method
Participants
The full study sample included 241 adolescents aged 14–17 years (130 female, 111 maleFootnote 1) who participated in a larger longitudinal study on adolescent experiences with their primary caregiver. Youth were recruited to participate from the community of a mid-sized metropolitan area. Families were recruited using a range of recruitment methods, including online and community advertisements (50.6% of families), a commercial mailing list of families with potentially age-eligible children (40.2%), and ResearchMatch (4.1%), an online clinical research registry (additional recruitment details are included in Starr et al., Reference Starr, Dienes, Stroud, Shaw, Li, Mlawer and Huang2017). Participants were excluded from study participation if they had a major physical, neurological, or pervasive developmental disorder, a prior diagnosis of any bipolar or psychotic disorders, English language difficulties, or previous participation of siblings or any other household member. Median parent-reported annual family income fell in the $80,000 to $89,999 range. Additionally, 24.1% of adolescents received free or reduced cost school lunches. Mothers comprised the majority of participating parents (87.6%).
As noted below, analyses conducted were largely specific to Caucasian adolescents in order to account for population stratification. The Caucasian sample included 192 youth (Mage = 15.89 years, SD = 1.08; 53.1% female). Parent-reported median annual family income was in the $90,000 to $99,000 range, with 16.7% of adolescents receiving free or reduced school lunches.
Procedure
Families completed a baseline in-lab session, during which youth and parents provided assent/consent, completed separate interviews, and participated in additional procedures unrelated to the present analyses. Saliva samples were also collected for DNA analysis during this visit. Participating families received $160 for completing baseline session procedures, and they were entered into raffles based on compliance. All procedures were approved by the Research Subjects Review Board of the University of Rochester.
Measures
Episodic stress
Trained interviewers administered the UCLA Life Stress Interview (LSI; Hammen, Reference Hammen1991), a semi-structured interview based on the contextual threat method of assessing life events (Brown & Harris, Reference Brown and Harris2012) that examines life events across multiple domains, to adolescents to measure youths’ episodic stress. During the LSI, interviewers collected information about life events that occurred within the previous 12 months across six domains (romantic relationships, peer relations, close friendships, family relationships, academic experiences, and behavioral functioning). Interviewers also obtained information about the nature, timing, duration, and context for each event and integrated details from both respondents if both discussed the same event. On average, youth reported 2.95 episodic events. An independent team of trained coders consensus-rated each event based on contextual factors and provided an objective rating of negative impact on a scale from 1 (no negative impact) to 5 (extremely severe impact). Events were also rated on level of independence, which was dichotomized as dependent or independent, and coded on interpersonal status (interpersonal or not). Inter-rater reliability based on independent raters recoding negative impact for a subset of episodic events yielded an interclass correlation of .87. Negative impact scores were summed (excluding “nonevents” rated as “1”) to obtain indices of total independent stress, interpersonal dependent stress, and noninterpersonal dependent stress.
CA
A modified version of the Youth Life Stress Interview (Rudolph et al., Reference Rudolph, Hammen, Burge, Lindberg, Herzberg and Daley2000) was completed with parents to assess the adolescents’ experience of CA. Information was collected solely from parents due to time constraints and their potential better recall of events from the youth's early childhood. Interview probes related to youths’ potential experiences with negative events and circumstances (e.g., parental conflict/divorce, separation from parents, death of close others, financial difficulties) that had occurred from birth through a year before study participation. Interviewers elicited information about the context of each event, including duration and impact. Parents reported an average of 4.56 CA events. A coding team provided an objective rating of the negative impact for each event using the same rating scale as the LSI. Further, each event was also coded as interpersonal or noninterpersonal. A second independent team of coders rated a subset of episodic stress interviews with excellent reliability, achieving an interclass correlation of .87. Impact scores were summed to assess overall noninterpersonal CA and ICA, excluding nonevents (for frequencies of reported ICAs, see Table 1).
Severity ratings and reported incidents of each type of interpersonal childhood adversities
Note: Caucasian subsample n = 192, Full sample n = 241. ICA = Interpersonal childhood adversity. Data do not incorporate childhood adversities that were coded as noninterpersonal. Mean event severity ratings were calculated in the full sample. Percentage of sample endorsing reflects participants reporting at least one event in each category.
Depressive symptoms (covariate)
Youth were interviewed using the Schedule for Affective Disorders and Schizophrenia for School-Aged Children—Present and Lifetime (KSADS-PL; Kaufman et al., Reference Kaufman, Birmaher, Brent, Rao, Flynn, Moreci and Ryan1997) to assess both past and current symptoms of major depressive disorder (MDD) and dysthymia. For past depression, the worst episode of depressive symptoms was coded. Consistent with prior work (e.g., Rao, Daley, & Hammen, Reference Rao, Daley and Hammen2000; Steinberg & Davila, Reference Steinberg and Davila2008), disorder-level and subsyndromal symptoms were rated following a dimensional rating scale: 0 (no symptoms), 1 (mild symptoms), 2 (moderate, subthreshold symptoms), 3 (meets DSM-IV criteria), 4 (meets DSM-IV criteria with high severity/impairment). Maximum scores between current MDD and dysthymia were used to capture depressive symptoms (consistent with prior GxE studies; e.g., Conway, Hammen, Brennan, Lind, & Najman, Reference Conway, Hammen, Brennan, Lind and Najman2010). For current depression, 3.6% of adolescents met criteria for a depression diagnosis, whereas 20.8% met criteria for past depression. Independent coders re-rated 20% of completed interviews with 100% reliability.
Pubertal Development (covariate)
Participants completed the Pubertal Development Scale (PDS; Petersen, Crockett, Richards, & Boxer, Reference Petersen, Crockett, Richards and Boxer1988). This measure consists of three questions about physical maturation for both sexes (e.g., skin complexion, growth spurts, and body hair) and two additional sex-specific items (girls: breast development, menarche; boys: facial hair, deepening voice). Items were scored on a 4-point scale from 1 (has not yet begun) to 4 (growth or development is complete), with a dichotomized menarche item (1 = no, 4 = yes). Item responses were averaged to create an index of pubertal development.
Genotyping and MGPS Computation
Genotyping
Youth submitted saliva samples using Oragene (DNA Genotek, Ontario, Canada) collection kits. The DNA samples were analyzed by the University of Wisconsin-Madison Biotechnology Center. DNA concentration was detected and quantitated using the Quant-iT PicoGreen dsDNA kit (Life Technologies, Grand Island, NY). Standard salting-out procedure was used for DNA extraction. Genotyping was carried out using KBiosciences’ competitive allele specific PCR SNP genotyping assay based on dual FRET (KASPar). KASPar assays were amplified with the Eppendorf Mastercycler pro384 thermal cycler using allele specific primers. End point fluorescence signals were analyzed by the Synergy 2 (BioTek) plate reader and Gen5 software program.
Following Pagliaccio's (Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014) established MGPS procedures, genotypes for 10 SNPs from four HPA axis-related genes, CRHR1 (rs4792887 T allele, rs110402 G allele, rs242941 T allele, rs242939 G allele, rs1876828 G allele), NR3C1 (rs41423247 G allele, rs10482605 T allele, rs10052957 A allele), NR3C2 (rs5522 G allele), and FKB5 (rs1360780 T allele) were included in the MGPS. Pagliaccio and colleagues (Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014) developed this MGPS from a large list of HPA axis-related SNPs, pruning them down to the current 10 SNP profile. These specific SNPs were selected from genes involved in the coding of HPA axis proteins and had been found to be associated with altered stress responsivity (e.g., increased cortisol reactivity), vulnerability to depression, and associated phenotypes (for further detail regarding the development of this MGPS, see Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014). Individual SNPs were coded based on the presence of at-risk genotypes and summed. Higher MGPS indicated greater HPA axis-related genetic risk. Distributions of genotype frequencies are available upon request. All genotype distributions were in Hardy-Weinberg equilibrium, χ2 (1) ≤ 2.82, ps > .05), except rs1876828, χ2 (1) = 4.12, p = .041. Excluding this SNP in analyses had no influence on the results.
Data Analytic Approach
Analyses were conducted using the SPSS PROCESS macro (Hayes, Reference Hayes2017). Prior to analysis, the data were inspected for univariate outliers (greater than three times the interquartile range away from the 25th or 75th percentiles, consistent with previous work; e.g., Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2015). Skew and kurtosis were in the normal range for all major variables (George & Mallery, Reference George and Mallery2010). Predictor variables were mean-centered prior to analysis of interaction models. In testing GxE models, ICA, MGPS, and their interaction were entered, with episodic stress as the outcome. In all analyses, gender, pubertal stage, and age were included as covariates as both main and interaction effects (e.g., gender × MGPS, gender × E), following guidelines by Dick et al. (Reference Dick, Agrawal, Keller, Adkins, Aliev, Monroe and Sher2015). We also conducted Cook's distance tests for the identification of potential multivariate outliers within these models (using a 1.0 threshold for Cook's D; no issues were identified). Significant interactions were examined using simple slope tests (M ± 1 SD).
Further, the Johnson-Neyman technique was applied to examine levels of ICA at which MGPS predicted stress outcomes, and alternately, MGPS values at which ICA predicted these outcomes. Finally, we conducted a set of sensitivity tests to assess the effects of individual SNPs in driving potential MGPS interaction effects. First, we examined individual SNP GxE effects in predicting interpersonal stress generation, our hypothesized outcome (given the large number of exploratory tests, we applied False Discovery Rate corrections to reduce the risk of Type I error for these analyses; Benjamini & Hochberg, Reference Benjamini and Hochberg1995; results should still be interpreted with caution). Second, we conducted “n – 1” analyses by re-running models after removing single SNPs from the MGPS profile one variant at a time (creating 10 nine-SNP profiles) to test whether MGPS interactions were robust to the removal of individual SNPs (see Vrshek-Schallhorn et al., Reference Vrshek-Schallhorn, Stroud, Mineka, Zinbarg, Adam, Redei and Craske2015).
Power analyses
Because no prior studies have examined the prediction of stress generation using an MGPS-based, GxE approach, we derived estimated parameters for power analyses from a variety of sources. On the lower end, we included an estimate of R 2 = .0015, based on a recent genome-wide by environment interaction study predicting depression (e.g., Arnau-Soler et al., Reference Arnau-Soler, Macdonald-Dunlop, Adams, Clarke, MacIntyre, Milburn and Thomson2019), which suggested power of 8–9% for the White and full samples, respectively. On the highly optimistic end, we estimated R 2 at .05, based on recent analyses within the current sample of this MGPS interacting with interpersonal childhood adversity to predict depression (i.e., the same model in the same sample, but with a different dependent variable; see Starr & Huang, Reference Starr and Huang2018); this suggested power of 89% to 94%. At this time, it is unclear whether these widely differing estimates are the result of very different methodological approaches (atheoretically versus theoretically selected SNPs, self-report versus interview-based phenotyping) or whether the higher effect size is an outlying effect derived from a much smaller sample. However, readers should be aware that, according to genome-wide-study-based estimates of variance captured by GxEs, our study may be underpowered to detect effects, and the results should be interpreted with a priori caution and an eye towards a need for future replication.
Results
Preliminary Analyses
Population stratification
As a preliminary step, we tested for population stratification effects (i.e., confounding effects that occur when race is correlated with outcomes of interest and specific genotypes). As reported elsewhere (for full details, see Starr et al., Reference Starr, Dienes, Stroud, Shaw, Li, Mlawer and Huang2017), in the current sample, non-Caucasian youth had higher MGPS than did Caucasian youth, t (239) = 2.10, p = .036, and race moderated the association between MGPS and depressive symptoms. Race was also marginally associated with ICA, t (239) = 1.68, p = .094; scores were higher for non-White adolescents. As a conservative measure to address potential population stratification issues, primary analyses were restricted to the Caucasian sample (n = 192) and then conducted in the full sample.
Main effects and gene–environment correlations
Descriptive data and bivariate correlations are presented in Table 2. Results show that HPA axis-related MGPS was not significantly correlated with any of the variables, and no gene–environment correlations were found between MGPS and any stress outcomes.
Bivariate main effect and gene-environment correlations and descriptive data
Note: MGPS = Multilocus genetic profile score, ICA = Interpersonal childhood adversity, CA = Childhood adversity. Data are from the Caucasian sample (n= 192). *p< .05 **p< .01.
We further examined environmental main effects using linear regression analyses. As hypothesized, there was evidence for an environmental main effect, with ICA and predicting interpersonal dependent stress, ß = .18, p = .012. These effects were largely exclusive to the relationship between ICA and interpersonal dependent stress. ICA marginally significantly predicted independent stress, ß = .14, p = .060, but it did not significantly predict noninterpersonal dependent stress,Footnote 2 ß = .10, p = .152. Additionally, noninterpersonal CA did not predict interpersonal dependent stress, ß = .09, p = .223. We also re-ran these analyses controlling for depressive symptoms, given past research indicating that depression predicts stress generation (Hammen, Reference Hammen1991; Rudolph et al., Reference Rudolph, Hammen, Burge, Lindberg, Herzberg and Daley2000). Following control of both past and current depressive symptoms, ICA marginally significantly predicted interpersonal dependent stress (ß = .11, p = .09). Finally, ICA did not significantly predict independent stress (p = .102) or noninterpersonal dependent stress (p = .155), nor did noninterpersonal CA predict interpersonal dependent stress (p = .267).
Tests of Gene–Environment Interactions
Supporting hypotheses, we found a significant G×E interaction between ICA and MGPS on adolescents’ interpersonal dependent stress, interaction term ß = .18, p = .016 (see Table 3); ΔR 2 was .03, suggesting that this G×E interaction explains approximately 3% of the variance in interpersonal dependent stress. Simple slopes analyses (see Figure 1) indicate that ICA did not significantly predict interpersonal dependent stress at low MGPS (M - 1 SD), b = .03, SE = .22, p = .893; however, at high MGPS (M + 1 SD), ICA significantly predicted higher interpersonal dependent stress, b = .72, SE = .19, p = .0002 (see Figure 1). Johnson-Neyman analyses indicated an association between ICA and interpersonal dependent stress at MGPS ≥ 4.34 (49th percentile). Higher genetic risk scores were associated with greater adolescent interpersonal dependent stress at high levels of ICA (Johnson-Neyman significance region beginning at 83rd percentile of ICA). At very low levels of ICA, MGPS predicted marginally less interpersonal dependent stress (p = .060; lowest 5th percentile, which encompassed only adolescents who reported zero ICA).
HPA axis MGPS at low moderate, and high level of interpersonal childhood adversity (ICA) predicting (a) interpersonal dependent stress and (b) independent stress. Data presented are from the Caucasian subsample (n = 192).
Model examining the interaction of HPA axis multilocus genetic profile scores and interpersonal childhood adversity in predicting interpersonal dependent stress
Note: MGPS = Multilocus genetic profile score; ICA = Interpersonal childhood adversity; PDS = Pubertal Development Scale. Presented analyses are from Caucasian subsample (n = 192).
We re-ran the model including both current and past depressive symptoms as covariates; the G×E interaction remained significant (interaction term ß = .17, p = .027), and simple slope patterns were unchanged. Further, to test that the findings were specific to stress generation results, analyses were repeated with independent (i.e., fateful) stress as the outcome. In line with hypotheses and the stress generation model, MGPS did not significantly moderate the association between ICA and independent stress (see Figure 1), interaction term ß = .02, p = .836; results were similar in a robust model including all interactive covariates, ß = .03, p = .733.
We next confirmed whether the results were unique to interpersonal stress, both as a predictor variable (ICA versus noninterpersonal CA) and outcome (interpersonal versus noninterpersonal dependent stress). Aligning with predictions, MGPS did not significantly interact with noninterpersonal CA to predict adolescent interpersonal dependent stress in the initial model, ß = −.02, p = .816. Furthermore, as hypothesized, there was no significant interaction between MGPS and ICA in predicting noninterpersonal dependent stress, ß = −.02, p = .821.
Exploratory tests of gender moderation
Although not initially hypothesized, we also conducted exploratory analyses involving gender moderation. Although biological sex was used in covariate analyses, for these tests, we focused on gender, given research suggesting greater risk for depression and stress generation in girls (e.g., Hammen, Reference Hammen1991; Rudolph et al., Reference Rudolph, Hammen, Burge, Lindberg, Herzberg and Daley2000). Thus, we excluded adolescents endorsing nonbinary gender (n = 3, 2 in the White sample) from our analyses. We tested gender as a moderator of the association between ICA and dependent interpersonal stress, MGPS and dependent interpersonal stress, and a 3-way interaction with ICA, MGPS, and gender predicting dependent interpersonal stress. Models including the MGPS variable were conducted in the White sample, and models with no genetic variable were conducted in the full sample. All models were nonsignificant, suggesting no support for gender moderation.
Sensitivity tests
We conducted exploratory analyses to test whether our significant findings were largely driven by single SNPs within the genetic profile. For these sensitivity analyses, we first tested G×E effects for each MGPS SNP in interaction with ICA to predict dependent interpersonal stress. We controlled for the main effects of covariates (i.e., gender, age, pubertal development) and tested covariate interactions for cases in which the single SNP interactions reached significance. We then further assessed the effects of individual SNPs by conducting n – 1 analyses, removing individual SNPs from the original 10 SNP genetic profile one at a time and re-running our original model with these revised 9-SNP MGPSs (Vrshek-Schallhorn et al., Reference Vrshek-Schallhorn, Stroud, Mineka, Zinbarg, Adam, Redei and Craske2015). Results (presented in Table 4) suggested that several SNPs reached nominal significance in their interaction with ICA (CRHR1 rs242939, CRHR1 rs242941, CRHR1 rs110402, NR3C1 rs41423247, and NR3C2 rs5522). However, after FDR corrections, only two SNPs remained significant (NR3C2 rs5522 and CRHR1 rs243939; Benjamini-Hochberg adjusted p-values = .025). These ICA interactions with NR3C2 rs5522 and CRHR1 rs243939 remained significant after controlling for covariate interactions, p = .002 and p = .033, respectively. In the n - 1 analyses, re-running analyses with 9-SNP MGPS profiles after removing each SNP individually, all G×E effects remained significant (ps ≤ .024), suggesting that it is the cumulative influence of these SNPs within the profile, rather than an individual SNP with large effects.
Results for separate regression models predicting interpersonal dependent stress from the interactions between individual SNPs and interpersonal childhood adversity
Note: Analyses conducted in Caucasian subsample (n = 192). B-H p = Benjamini-Hochberg corrected p value (Benjamini & Hochberg, Reference Benjamini and Hochberg1995). Covariates included gender, age, and pubertal development.
Analyses with Full Sample
Analyses were re-run in the full sample, which included adolescents from all racial groups. Given that the majority of the sample reported European heritage (80%), these results should be interpreted more cautiously; note that there was insufficient power to examine any other racial group individually. These findings paralleled those from the White youth sample. MGPS moderated the association between ICA and interpersonal dependent stress, interaction term ß = .13, p = .045. Results were near-identical in the robust model, ß = .14, p = .040. Simple slope patterns were consistent with those from the White sample. As in the White sample, MGPS did not significantly moderate the relationship between ICA and independent stress or noninterpersonal dependent stress. No moderation effects were found for the association between noninterpersonal CA and interpersonal dependent stress.
Discussion
The current study examined the interaction between ICA and HPA axis-related genetic variation in predicting interpersonal stress generation. Our findings supported study hypotheses; first, aligning with previous findings (e.g., Chan et al., Reference Chan, Doan and Tompson2014; Hernandez et al., Reference Hernandez, Trout and Liu2016), we found that ICA significantly predicted greater adolescent interpersonal dependent stress (but not noninterpersonal dependent) stress. The association between ICA and independent stress was marginally significant, which suggests that some of the effects may be attributable to continuity in high-stress environments, rather than stress generation alone. However, we found that the association between ICA and interpersonal dependent stress was qualified by its significant moderation by HPA axis-related genetic variation. Importantly, HPA axis-related MGPS G×E did not predict independent stress, which suggests that these effects contribute to generation of stress and not overall stress exposure. These results are consistent with the stress generation model (Hammen, Reference Hammen2006) and prior findings supporting the role of genetic factors and CA in stress generation (Harkness et al., Reference Harkness, Bagby, Stewart, Larocque, Mazurka, Strauss and Kennedy2015; Starr et al., Reference Starr, Hammen, Brennan and Najman2012, Reference Starr, Hammen, Brennan and Najman2013). Further, results were exclusive to interpersonal stress: MGPS did not moderate the association between noninterpersonal CA and interpersonal dependent stress, nor was there an interaction between ICA and MGPS in predicting noninterpersonal dependent stress.
While previous studies have shown that genetic risk intensifies interpersonal stress generation (Harkness et al., Reference Harkness, Bagby, Stewart, Larocque, Mazurka, Strauss and Kennedy2015; Starr et al., Reference Starr, Hammen, Brennan and Najman2012, Reference Starr, Hammen, Brennan and Najman2013), the present study extends previous findings in several key ways. This is the first study to apply a multilocus genetic risk approach in examining stress generation. Previous research focused on G×Es involving a specific serotonergic genetic variant (5-HTTLPR) linked with stress reactivity; the MGPS approach considers the cumulative, polygenic effect of several genes linked to a specific biological system (Aliev, Latendresse, Bacanu, Neale, & Dick, Reference Aliev, Latendresse, Bacanu, Neale and Dick2014; Caspi et al., Reference Caspi, Hariri, Holmes, Uher and Moffitt2010; Dick et al., Reference Dick, Agrawal, Keller, Adkins, Aliev, Monroe and Sher2015). Further, HPA axis-related genetic risk has never been examined in relation to stress generation. Recent work has shown that HPA axis dysregulation following laboratory stress predicts stress generation (Morris et al., Reference Morris, Kouros, Hellman, Rao and Garber2014). Our findings indicate that genetic risk linked to HPA axis functioning may interact with environmental stress, namely ICA, to promote stress generation. Given that few studies have examined HPA axis functioning in relation to stress generation, an important future research direction would be to examine physiological mechanisms that might underpin this process. For example, CA has been shown to alter specific indices of diurnal HPA axis regulation (e.g., latent trait cortisol, cortisol awakening response; Chen, Stroud, Vrshek-Schallhorn, Doane, & Granger, Reference Chen, Stroud, Vrshek-Schallhorn, Doane and Granger2017; Starr et al., Reference Starr, Dienes, Stroud, Shaw, Li, Mlawer and Huang2017; Stroud, Chen, Doane, & Granger, Reference Stroud, Chen, Doane and Granger2016); it would be interesting to examine how these are, in turn, related to daily interpersonal behaviors which may culminate in stress generation.
In addition, our results provide further evidence of ICA as a specific environmental risk factor predicting interpersonal stress generation, as moderated by MGPS. Findings align with prior research that identified interpersonal stress, including ICA, as a particularly powerful candidate environment for the prediction of depression among genetically vulnerable youth (Feurer et al., Reference Feurer, McGeary, Knopik, Brick, Palmer and Gibb2017; Starr & Huang, Reference Starr and Huang2018; Vrshek-Schallhorn et al., Reference Vrshek-Schallhorn, Stroud, Mineka, Zinbarg, Adam, Redei and Craske2015). Several previous studies have examined specific types of adversities that fall into the interpersonal domain (e.g., parent–child conflict, childhood emotional abuse; Chan, Doan, & Tompson, Reference Chan, Doan and Tompson2014; Hernandez et al., Reference Hernandez, Trout and Liu2016) as predictors of interpersonal stress generation, but this study was the first to explicitly classify CA by its interpersonal nature in the context of stress generation. Importantly, many of the previous studies examining the link between CA and stress generation focus on maltreatment, often occurring specifically in early childhood (e.g., Harkness et al., Reference Harkness, Bagby, Stewart, Larocque, Mazurka, Strauss and Kennedy2015; Hernandez et al., Reference Hernandez, Trout and Liu2016; Liu et al., Reference Liu, Choi, Boland, Mastin and Alloy2013). In contrast, the CA reported within this study largely reflect more commonplace stressors (e.g., parental divorce or separation, family conflicts) that occur over the course of childhood. These results suggest that genetic moderation of stress generation does not exclusively occur following very severe CA, and they are in line with previous research suggesting that the downstream effects of CA are not specific to childhood maltreatment (e.g., Felitti et al., Reference Felitti, Anda, Nordenberg, Williamson, Spitz, Edwards and Marks1998; Green et al., Reference Green, McLaughlin, Berglund, Gruber, Sampson, Zaslavsky and Kessler2010; Hazel et al., Reference Hazel, Hammen, Brennan and Najman2008), potentially indicating that results are directly relevant to a broader portion of the population.
Further, our results suggest that the cascading effects of ICA may have particular relevance to the interpersonal domain, especially for those with HPA axis-related genetic risk, resulting in negative outcomes such as interpersonal stress generation. It may be that ICAs serve as acute threats that powerfully influence HPA axis functioning. Stress response systems are exquisitely sensitive to social threat, with interpersonal stressors reliably predicting HPA axis activation (Miller, Chen, & Zhou, Reference Miller, Chen and Zhou2007; Stroud et al., Reference Stroud, Chen, Doane and Granger2016). Our results build upon this past research, demonstrating that youth with high HPA axis-related genetic risk may be especially sensitive to ICA. Given that social relationships in childhood are particularly important (McLaughlin, Reference McLaughlin2016; Sroufe, Reference Sroufe2000), ICAs that occur during this stage may lead to repeated or sustained HPA axis activation and consequent long-term alterations in HPA axis activity, especially for adolescents who are at genetic risk for HPA axis dysfunction. Moreover, ICAs and associated stress sensitization may be particularly relevant to social development. For example, attachment theory (Bowlby, Reference Bowlby1982) suggests that early relationships and interpersonal interactions serve as models for future relationships and interpersonal patterns. Internalized expectations from prior negative interpersonal experiences have been linked to greater interpersonal conflict within relationships (Downey, Freitas, Michaelis, & Khouri, Reference Downey, Freitas, Michaelis and Khouri1998). Furthermore, ICAs have been shown to predict chronic interpersonal difficulties (e.g., Salwen et al., Reference Salwen, Hymowitz, Vivian and O'Leary2014). Thus, social disruptions due to ICA may contribute to impairments in interpersonal functioning, such as interpersonal stress generation. Further, within our ICA construct, there may be specific dimensions of interpersonal adversities that may be more potent predictors of social disruptions and later stress generation. For example, research indicates that maladaptive family functioning (e.g., parental mental illness, neglect, physical or sexual abuse) strongly predicts later psychopathology (Green et al., Reference Green, McLaughlin, Berglund, Gruber, Sampson, Zaslavsky and Kessler2010). It may be that stressful or highly conflictual family environments in particular, especially during key developmental periods, may increase vulnerability, posing more potent threats to attachment bonds and cognitions and expectations about interpersonal behavior (e.g., Dodge, Bates, & Pettit, Reference Dodge, Bates and Pettit1990; Styron & Janoff-Bulman, Reference Styron and Janoff-Bulman1997). Future studies on HPA axis-related biological mechanisms should investigate whether effects are unique to the interpersonal realm and examine whether specific dimensions of ICA increase vulnerability for interpersonal stress generation.
In addition, these effects may more specifically put youth at greater risk of generating interpersonal (but not noninterpersonal) stress. Notably, there were relatively few noninterpersonal dependent events in our sample. Although the lack of prediction of noninterpersonal stress may be a consequence of restricted range, it may also imply that the fusion of ICA and genetic risk specifically leads to interpersonal dysfunction. While CA may lead to interpersonal stress generation through a number of mechanisms, one of particular relevance may be stress sensitization. According to the stress sensitization hypothesis (Post, Reference Post1992), CA is linked with a lower threshold for depressive onset following future stressful events, such that depressed youth with CA histories report lower levels of recent stress than do depressed youth with no CA history (Hammen, Henry, & Daley, Reference Hammen, Henry and Daley2000; La Rocque, Harkness, & Bagby, Reference La Rocque, Harkness and Bagby2014; Monroe & Harkness, Reference Monroe and Harkness2005; Shih, Eberhart, Hammen, & Brennan, Reference Shih, Eberhart, Hammen and Brennan2006). Likewise, CA amplifies the relationship between proximal stressors and depression, again suggesting greater stress reactivity (Kim et al., Reference Kim, Martins, Shmulewitz, Santaella, Wall, Keyes and Hasin2014; McLaughlin et al., Reference McLaughlin, Conron, Koenen and Gilman2010; Shapero et al., Reference Shapero, Black, Liu, Klugman, Bender, Abramson and Alloy2014; Starr et al., Reference Starr, Dienes, Stroud, Shaw, Li, Mlawer and Huang2017). Further, HPA axis-related genetic risk has also been shown to moderate stress sensitivity following CA, suggesting that genetic vulnerabilities may also increase risk for stress sensitization (Starr et al., Reference Starr, Hammen, Conway, Raposa and Brennan2014). Research has also found that HPA axis-related genetic variation interacts with CA to predict alterations in threat-related amygdala function, which has been implicated in reactivity to stress and linked to greater psychological vulnerability to subsequent life stress (Di Iorio et al., Reference Di Iorio, Carey, Michalski, Corral-Frias, Conley, Hariri and Bogdan2017; Swartz, Knodt, Radtke, & Hariri, Reference Swartz, Knodt, Radtke and Hariri2015). HPA axis MGPS has also been shown to interact with environmental stress to predict differences in diurnal cortisol regulation, an alteration in HPA axis functioning that may also contribute to stress reactivity (Starr et al., Reference Starr, Dienes, Li and Shaw2019). Increases in stress reactivity, along with potential alterations in neural circuitry and other indices relating to HPA axis functioning, may bias perceptions of social threats in these situations and affect stress responses, resulting in further stress experienced and contributing to interpersonal stress generation. For instance, for someone who is hyper-responsive to stress, a perceived slight may rapidly escalate into a conflict and subsequent friendship dissolution, culminating more readily into a significant stressor. Supporting this model, several variables related to stress reactivity directly predict stress generation, including neuroticism and rejection sensitivity (Hernandez et al., Reference Hernandez, Trout and Liu2016; Uliaszek et al., Reference Uliaszek, Zinbarg, Mineka, Craske, Griffith, Sutton and Hammen2012). For example, one recent study found that CA predicted greater rejection sensitivity, which, in turn, led to greater interpersonal stress generation (Hernandez et al., Reference Hernandez, Trout and Liu2016). These results suggest that CA produces increased reactivity within interpersonal relationships, which contributes to greater self-generated interpersonal stress. Thus, heightened sensitivity and associated negative interpersonal processes following CA may increase the likelihood of interpersonal stress generation.
Further research is necessary to understand potential interpersonal mechanisms involved. For example, CA is related to a range of interpersonal risk processes, such as insecure attachment, ineffective interpersonal stress responses (e.g., involuntary engagement/ disengagement with stressors), and excessive reassurance seeking (e.g., Massing-Schaffer, Liu, Kraines, Choi, & Alloy, Reference Massing-Schaffer, Liu, Kraines, Choi and Alloy2015; Mickelson, Kessler, & Shaver, Reference Mickelson, Kessler and Shaver1997; Shih, Abela, & Starrs, Reference Shih, Abela and Starrs2009; Troop-Gordon, Sugimura, & Rudolph, Reference Troop-Gordon, Sugimura and Rudolph2017). These interpersonal risk processes have also been linked to interpersonal stress generation (Flynn & Rudolph, Reference Flynn and Rudolph2011; Shih et al., Reference Shih, Abela and Starrs2009; Starr et al., Reference Starr, Hammen, Brennan and Najman2013). Future research should examine how the interaction of ICA and HPA axis-related genetic risk might create a marked vulnerability for these negative interpersonal processes and later interpersonal stress generation.
While we focus on interpersonal stress generation as an outcome, it is also important to consider downstream effects that may follow. A wealth of literature links interpersonal dysfunction (including interpersonal stress generation) to later depression (Hames et al., Reference Hames, Hagan and Joiner2013; Joiner & Timmons, Reference Joiner and Timmons2002; Liu & Alloy, Reference Liu and Alloy2010; Rudolph et al., Reference Rudolph, Hammen, Burge, Lindberg, Herzberg and Daley2000). Thus, another avenue for future research may be to examine whether processes highlighted in our study are a probable factor in propagating depression, such that increased interpersonal stress generation predicts later increases in depressive symptoms. Moreover, there is evidence to suggest that HPA axis-related genetic risk may moderate this relationship, as HPA axis MGPS has been shown to increase reactivity to acute interpersonal stressors, predicting stronger associations between interpersonal stress and depressive symptoms in youth (Feurer et al., Reference Feurer, McGeary, Knopik, Brick, Palmer and Gibb2017). As such, it would be interesting to examine how the increased self-generated interpersonal stress might bridge ICA with later emotional outcomes among youth at high genetic risk. Longitudinal research examining these questions may be an important future direction.
Furthermore, our results suggested that at very low levels of ICA (i.e., no ICA reported), high HPA axis-related MGPS was marginally associated with lower interpersonal dependent stress. This pattern is consistent with differential susceptibility models, which propose that genetic factors that leave adolescents more susceptible to negative outcomes in negative, stressful environments may actually also predispose them to thrive in positive, supportive environments (Belsky, Bakermans-Kranenburg, & van IJzendoorn, Reference Belsky, Bakermans-Kranenburg and van IJzendoorn2007; Belsky & Pluess, Reference Belsky and Pluess2009). Prior studies (e.g., Feurer et al., Reference Feurer, McGeary, Knopik, Brick, Palmer and Gibb2017; Starr & Huang, Reference Starr and Huang2018) have found evidence for the differential susceptibility model using HPA axis-related MGPS in predicting depressive outcomes. Given this converging set of findings, further work examining potential differential susceptibility effects for interpersonal stress generation is needed. For example, our ability to detect this pattern may potentially have been limited by the environmental measures in our study, as low ICA may not necessarily reflect a true positive context; rather, it may better capture the absence of a negative environment. It could be the case that a differential susceptibility pattern would be more evident using a measure that encapsulates the extent to which the environment was warm, nurturing, and supportive.
Adolescence provided an ideal developmental period for our research questions, given the changes in reactivity to stress, increased rates of disorders (e.g., depression) associated with stress generation and CA, shifts in biological functioning, and pronounced increases in stressful life events during this time (Avenevoli et al., Reference Avenevoli, Swendsen, He, Burstein and Merikangas2015; Gunnar et al., Reference Gunnar, Wewerka, Frenn, Long and Griggs2009; Romeo, Reference Romeo2013; Rudolph, Reference Rudolph2002). An important remaining question is whether effects replicate across other age groups. Stress generation processes occur across development and are not solely relevant to adolescence (Alloy, Liu, & Bender, Reference Alloy, Liu and Bender2010). Research suggests that there are changes in HPA axis activity across development, including increases in basal cortisol from childhood into adulthood (Gunnar & Vazquez, Reference Gunnar and Vazquez2006). These shifts may have implications for stress reactivity and stress generation processes, particularly for those who are more genetically vulnerable to stress system dysregulation. Furthermore, twin study research indicates that G×E effects may also change with age, as the relative influence of environmental versus genetic factors vary over time (e.g., Rice, Harold, & Thapar, Reference Rice, Harold and Thapar2002; Tully, Iacono, & McGue, Reference Tully, Iacono and McGue2010). Thus, the extent to which HPA axis-related genetic variation modulates the effects of CA on stress generation processes may vary across development. Longitudinal studies examining whether these HPA axis-related G×E effects on stress generation and related processes shift with age are needed.
This study featured several important limitations. First, the sample size was small by typical G×E standards, which has been associated with concerns over statistical power and robustness. Of note, our effect of R 2 of .04 is somewhat larger than those typically associated with predictions of complex behavioral phenotypes and could be anomalous to this sample (Bogdan, Baranger, & Agrawal, Reference Bogdan, Baranger and Agrawal2018); replication is paramount. We used the MGPS approach, which appears to yield greater predictive validity than single polymorphisms examined in isolation and allow for theoretically-driven hypotheses about intermediate phenotypes (e.g., Pagliaccio et al., Reference Pagliaccio, Luby, Bogdan, Agrawal, Gaffrey, Belden and Barch2014; Vrshek-Schallhorn et al., Reference Vrshek-Schallhorn, Stroud, Mineka, Zinbarg, Adam, Redei and Craske2015), but GWAS-based approaches that map genetic risk using the entire human genome offer additional computational benefits (Bogdan et al., Reference Bogdan, Baranger and Agrawal2018). Our study findings were preserved after conservative control of covariates and covariate interactions and sample racial stratification, which we believe supports that the sample was adequately powered for these analyses. Of note, we controlled for past and current depressive symptoms, given their established association with stress generation; however, given that computation of the CA variables involved aggregation across all relevant CA events, we were prevented from testing whether depressive symptoms temporally preceded these events. Given the smaller sample size and the novelty of these findings, however, replication is important. Further, because this study did not use GWAS-based genotyping, we were unable to conduct competitive significance testing in order to assess the performance of this MGPS in comparison with MGPSs derived from randomly drawn SNPs (Di Iorio et al., Reference Di Iorio, Carey, Michalski, Corral-Frias, Conley, Hariri and Bogdan2017), which will be an interesting next step for future studies. Additionally, in our analyses featuring noninterpersonal events (both for CA and dependent stress), reported interpersonal events outnumbered noninterpersonal ones; these differences in frequency may have limited power for these specific analyses.
The study also featured a cross-sectional sample. One problem this raises is bias that occurs with retrospective reports, especially in the assessment of adversities occurring many years prior to the interview. Further, information about CA was obtained from parents. This means that reported events may have been described from the parent's perspective, and any experiences that occurred outside of the parent's awareness may not have been included in their report. There may also have been instances in which parents were reluctant to report specific events they had perpetrated or were at fault for in some way. Future research would benefit from longitudinal tracking of CA across childhood and incorporating multisource data (e.g., cross-checking with multiple reporters, child protection services records).
These limitations notwithstanding, this study highlights the interaction between ICA and HPA axis-related genetic risk in adolescent interpersonal stress generation and sheds light on the complex processes involved in stress generation, contributing to the growing literature on G×Es using HPA axis MGPSs. Future studies should consider their use in multilevel analyses to help further elucidate the many pathways that contribute to this and other stress-related outcomes.
Acknowledgments
The authors thank Y. Irina Li, Zoey A. Shaw, and Fanny Mlawer for data collection management, as well as the participating families for generously volunteering their time. We are also grateful to Sheree Toth for her comments on an earlier draft of this manuscript. The authors acknowledge the University of Wisconsin Biotechnology Center DNA Sequencing Facility for providing genotyping facilities and services.
Financial Support
Study funding was provided by the University of Rochester.
