Are rumination and neuroticism genetically or environmentally distinct risk factors for psychopathology?

Abstract

Neuroticism, a dispositional trait of heightened negative emotionality, is a vulnerability factor for psychopathology. Given neuroticism’s strong association with rumination, a repetitive thought pattern that intensifies and prolongs emotions, some question whether these constructs capture the same or unique information about vulnerability for psychopathology. The present study examined whether neuroticism is genetically and environmentally distinct from two clinically relevant ruminative subtypes—anger and depressive rumination—and whether genetic and environmental influences specific to rumination versus shared with neuroticism overlap with internalizing and externalizing psychopathology. These analyses were conducted on 439 same-sex twin pairs in the Colorado Longitudinal Twin study. Rumination and Neuroticism latent variables were created from multiple rumination questionnaires administered at age 23 and shortened Eysenck Personality Questionnaires administered at ages 17 and 21, respectively. Lifetime psychopathology symptoms, assessed by two structured clinical interviews, were used to create ordinal composite variables. Multivariate Cholesky decompositions indicated that Neuroticism, Anger Rumination, and Depressive Rumination have common genetic and nonshared environmental influences, but are differentiated by nonshared environmental influences specific to each ruminative subtype. Genetic influences common to Rumination and Neuroticism explained considerable variance in internalizing psychopathology, suggesting possible genetic mediation or common genetic causes. Genetic and environmental influences on externalizing psychopathology did not substantially overlap with those on Neuroticism and Rumination. These findings suggest that rumination and neuroticism share most genetic influences, yet are influenced by distinct environmental influences. Furthermore, our results indicate that a comprehensive understanding of transdiagnostic risk factors must include an examination of both genetic and environmental influences.

General Scientific Summary:

This study suggests that neuroticism and rumination have genetic overlap that is associated with lifetime internalizing symptoms and diagnoses. Neuroticism and rumination are differentiated by environmental influences that are specific to each twin within a twin pair (for example, life events or peer groups). Using a genetically-informed sample, this study improves understanding of the independent and additive effects of risk factors for psychopathology and encourages future research on environmental influences specific to neuroticism and rumination.

Neuroticism, a dispositional trait, and rumination, a cognitive coping style, are both transdiagnostic vulnerability factors for psychopathology (Aldao, Nolen-Hoeksema, & Schweizer, 2010; Kotov, Gamez, Schmidt, & Watson, 2010). Given that neuroticism is strongly associated with rumination (Nolan, Roberts, & Gotlib, 1998), some have questioned whether these constructs capture unique information about individual differences in vulnerability for psychopathology. Some studies suggest that rumination may be one process that mediates neuroticism’s relation to psychopathology symptoms (e.g., Roelofs, Huibers, Peeters, & Arntz, 2008), and other studies indicate that rumination may confer independent risk for psychopathology after statistically accounting for neuroticism (e.g., Nolan, Roberts, & Gotlib, 1998). The present study extends the literature on rumination and its relation to neuroticism by using a genetically informed design to address the following questions: 1) Why are rumination and neuroticism correlated, and what distinguishes them? 2) Does rumination increase genetic and environmental vulnerability for psychopathology above and beyond neuroticism? And 3) Does anger rumination—a less-studied subtype of rumination—also augment genetic or environmental risk for psychopathology above and beyond neuroticism and depressive rumination?

Neuroticism and Rumination

Genetically informed research on neuroticism suggests that it is moderately heritable (h² =.39–.58; Vukasović & Bratko, 2015) and shares genetic influences with psychopathology. For example, a study by Hettema, Neale, Myers, Prescott, & Kendler (2006) found neuroticism to be genetically correlated (rA=.58–82) with lifetime major depression, generalized anxiety, panic, and several kinds of phobia. Additional work by Hink and colleagues (2013) found overlapping genetic and nonshared environmental influences on neuroticism, internalizing (generalized anxiety, depression, social anxiety disorders), and externalizing psychopathology (oppositional defiance, conduct, and attention-deficit hyperactivity disorders) latent factors in adolescents. These findings support neuroticism as a phenotypic and genetic common feature of psychopathology that helps explain the comorbidity between internalizing and externalizing disorders.

Though individuals high on neuroticism often also report rumination, and both neuroticism and rumination are relatively stable individual differences, previous research has distinguished these constructs by defining neuroticism as a personality trait and rumination as a cognitive coping style. They are substantially correlated constructs (r=.57; Nolan et al., 1998) that increase risk for and maintain a range of psychopathological symptoms (Johnson et al., 2016; Ormel, Rosmalen, & Farmer, 2004). Prior studies suggest that rumination may partially mediate the association between neuroticism and psychopathology; however, cause-effect relations between neuroticism, rumination, and psychopathology remain unclear, as this work is largely cross-sectional in nature (e.g., Muris et al., 2005) or conducted over a short period of time (e.g., Nolan et al., 1998).

In contrast to the extensive literature examining genetic influences on neuroticism, few studies have investigated rumination using genetically informed designs, and none have examined overlapping and specific genetic and environmental influences on rumination and neuroticism. This literature has largely focused on one subtype of rumination, depressive rumination, suggesting that depressive rumination has low heritability in adolescents (h²=.21; Moore et al., 2013) and moderate heritability in adults (h²=.37–.43; Jang, Livesley, Vernon, & Jackson, 1996; Johnson, Whisman, Corley, Hewitt, & Friedman, 2014).¹ Research on the genetic correlation between rumination and specific disorders by Johnson and colleagues (2016) also suggests that rumination has overlapping genetic influences with major depression, generalized anxiety, eating disorder symptoms, and substance dependence vulnerability. However, no studies have examined whether the overlapping genetic influences between rumination and psychopathology may be accounted for by neuroticism.

Anger and Depressive Rumination

Although the literature frequently uses the terms “rumination” and “depressive rumination” interchangeably, phenotypic research supports a multidimensional conceptualization of rumination. In the present study, we focus on anger and depressive rumination, as previous work has documented their transdiagnostic associations with psychopathology (du Pont, Rhee, Corley, Hewitt, & Friedman, 2018) and their separability (r=.57; Baer & Sauer, 2011). Anger and depressive rumination show unique associations with drinking behavior, aggression, and depressive symptoms (Ciesla, Dickson, Anderson, & Neal, 2011; Peled & Moretti, 2010). Even when both ruminative subtypes predict an outcome, the strengths of their associations differ. For example, both depressive and anger rumination correlate with internalizing psychopathology (i.e., depression and anxiety), though the magnitude of the association is larger for depressive rumination (r_men=.75; r_women=.61) than anger rumination (r_men=.33; r_women=.37; du Pont et al., 2018). However, the extent to which anger and depressive rumination have a shared or unique genetic or environmental etiology remains unclear.

The Present Study

Although significant progress has been made toward understanding the associations between neuroticism, rumination, and psychopathology, there are several limitations to this literature. First, there is no examination of the heritability of multiple ruminative subtypes. Second, it is unclear whether neuroticism and rumination, as well as depressive rumination and anger rumination, are genetically distinct as well as phenotypically distinct. Third, no study has examined whether neuroticism, anger rumination, and depressive rumination have independent genetic variance that overlaps with internalizing and externalizing psychopathology. The present study addresses these gaps in the literature by examining the overlap between genetic and environmental influences on: (1) anger and depressive rumination, (2) neuroticism and rumination, and (3) neuroticism, rumination, and psychopathology.

Addressing these gaps and evaluating the extent to which neuroticism and rumination predict unique phenotypic, genetic, and environmental variance in psychopathology has implications for several lines of research. First, previous phenotypic research suggests that rumination may be a potential mediator of the association between neuroticism and psychopathology (e.g. Nolan et al., 1998; Roelofs et al., 2008). Evidence that genetic and environmental influences shared by neuroticism and rumination also influence psychopathology will lend support to the mediation hypothesis, whereas evidence that genetic and environmental influences shared by anger and depressive rumination influence psychopathology after controlling for neuroticism will provide evidence consistent with additional, potentially direct, influences of rumination.

Second, both neuroticism and rumination have been suggested as heritable, transdiagnostic risk factors for psychopathology (Hettema et al., 2006; Johnson et al., 2016). Evidence of genetic influences shared by rumination and psychopathology after controlling for neuroticism would suggest that examining the genetic risk for both rumination and neuroticism may improve the prediction of genetic risk for psychopathology, whereas lack of such evidence would suggest that examining the genetic risk for neuroticism may be more efficient.

Third, some researchers have identified rumination as a target of psychological interventions (Watkins et al., 2011). Evidence of common nonshared environmental influences on rumination and psychopathology after controlling for those shared with neuroticism would be consistent with the suggestion that changes in rumination may also lead to changes in psychopathology symptoms, and support the clinical utility of targeting neuroticism and rumination as separate treatment goals.

Method

Participants

Participants were 877 individuals from 439 same-sex twin pairs. Of these participants, 469 individuals were from monozygotic (MZ) twin pairs (221 men; 248 women) and 408 individuals were from dizygotic (DZ) twin pairs (206 men; 202 women). The overall sample was 92% Caucasian, 5% multiracial, 2% other, and 1% not reported. Participants were recruited for the Colorado Longitudinal Twin Study, an ongoing study of the cognitive, emotional, and behavioral development of same-sex twins from infancy to adulthood. We report on 429 of the 483 twin pairs in the foundational sample who also participated in a multi-wave study conducted by the Center for Antisocial Drug Dependence (CADD; Rhea, Gross, Haberstick, & Corley, 2013; Rhea, Gross, Haberstick, & Corley, 2006), a study of sleep, and/or a study of executive functions and self-regulation. As participants were twin pairs, we accounted for the dependence in our data using Mplus’s TYPE=COMPLEX option for phenotypic analyses. All research protocols were approved by the University of Colorado’s Institutional Review Board.

Rumination and psychopathology measures were collected at age 22.28 (SD±1.29, range 21.00–27.00), whereas neuroticism measures were collected in prior assessments at age 17.26 (SD±0.64, range 16.13–20.08) and 21.09 (SD±2.05, range 18.00–28.03).

Zygosity determination.

Twin pairs were rated on 10 physical characteristics across time. Twin pairs were considered unambiguously MZ or DZ if 85% of the raters agreed on their zygosity. Zygosity determination by similarity rating was confirmed using 11 polymorphic microsatellite markers.

Measures

Neuroticism.

Neuroticism was assessed at age 17 and age 21 using revised versions of the neuroticism scale of the Eysenck Personality Questionnaire (EPQ; Eysenck & Eysenck, 1975). The neuroticism subscale measures one’s tendency to experience negative emotions (“Are you touchy about some things?”), and each item was dichotomous (yes or no).

We used a 9-item version of the EPQ neuroticism subscale at age 17 (EPQ-17; Floderus-Myrhed, Pedersen, & Rasmuson, 1980) and a 12-item version of the EPQ neuroticism subscale at age 21 (EPQ-21; Francis, Brown, & Philipchalk, 1992). Items within each time point were averaged to create an age 17 and age 21 neuroticism score for each person. The two neuroticism scores were used as indicators for the Neuroticism latent variable described in the Results section. The EPQ questionnaires administered at age 17 and 21, as well as the rumination measures administered at age 23, were completed on a computer in the laboratory along with a series of other questionnaires. Subscale scores were calculated for participants who answered at least 80% of the subscale items.

Depressive rumination.

Depressive rumination was assessed at age 23 with the Rumination-Reflection Questionnaire (RRQ; Trapnell & Campbell, 1999), and a revised version of the Ruminative Responses Scale (RRS; Treynor, Gonzalez, & Nolen-Hoeksema, 2003). The RRQ is a 24-item questionnaire consisting of two subscales: rumination (RRQ-Ru) and reflection (RRQ-Re). The RRQ-Ru measures negative self-focused thought (“I spend a great deal of time thinking back over my embarrassing or disappointing moments”). The RRQ-Re measures a general tendency to think introspectively (“I’m very self-inquisitive by nature”). As general reflection is considered distinct from depressive rumination (Trapnell & Campbell, 1999), the RRQ-Re subscale was excluded from the present study.

The revised RRS excludes items from the original scale that overlap with depression inventories. The RRS asks participants how they typically respond when they are “down, sad, or depressed” and has two subscales: brooding (RRS-B) and reflection (RRS-R). The RRS-B subscale measures negative, self-focused thoughts (“I think ‘What am I doing to deserve this?’”), and the RRS-R subscale measures the tendency to reflect on sadness (“I analyze recent events to understand why I am depressed”). The RRS-R scale was included in the current study because it measures reflection on sadness rather than general introspection.

Anger rumination.

Anger rumination was assessed at age 23 using the 19-item Anger Rumination Scale (ARS; Sukhodolsky et al., 2001). The ARS has 4 subscales: angry afterthoughts (AA), thoughts of revenge (TR), angry memories (AM), and understanding causes (UC). The AA and TR subscales measure cognitive rehearsal after an anger episode (“I re-enact the anger episode in my mind after it has happened”) and desired retribution (“I have day dreams and fantasies of a violent nature”), respectively. The AM subscale captures thoughts of past anger episodes (“I ponder about the injustices that have been done to me”) and the UC subscale measures thoughts about why the anger episode occurred (“I think about the reasons people treat me badly”).

Psychopathology.

Internalizing and externalizing psychopathology were measured at age 23 using the Diagnostic Interview Schedule–IV (DIS-IV; Robins et al., 2000) and the Composite International Diagnostic Interview–Substance Abuse Module (CIDI-SAM; Cottler, Robins, & Helzer, 1989). The DIS-IV is a structured interview that diagnoses major psychiatric disorders. The CIDI-SAM is a structured interview used to diagnose abuse and dependence for tobacco, alcohol, cannabis, and illicit drugs. The psychopathology interviews were conducted over the phone, typically within a month from the in-person assessment of rumination (mean age difference=0.05 years, SD=0.2).

Internalizing psychopathology was measured using lifetime symptom endorsement from the MDD and GAD modules of the DIS-IV and coded as a composite variable.² As symptom count measures are highly skewed, we created an ordinal variable that enables estimation of the underlying liability based on the frequencies within each category and decreases the risk of biased parameter estimates (Derks, Dolan, & Boomsma, 2004). The internalizing psychopathology composite variable was coded as each person’s maximum level of symptoms on MDD and GAD: A score of 0 indicated no MDD or GAD symptoms, a score of 1 indicated symptoms of MDD and/or GAD but no diagnosis, and a score of 2 indicated a diagnosis of MDD and/or GAD.

Externalizing psychopathology was measured using lifetime symptom endorsement from the DIS-IV antisocial personality disorder (ASPD) module and the CIDI substance abuse module. Externalizing was coded as a composite of ASPD and substance use disorders; a score of 1 indicated symptoms of ASPD and/or substance use disorder(s), and a score of 2 indicated a diagnosis of ASPD and/or a substance use disorder. The low prevalence of abuse and dependence symptoms for illicit drugs in our sample led us to use the illicit drug class with the highest number of symptoms for each person who endorsed illicit drug use. Problematic illicit drug use was coded as 0 for no symptoms and 1 for 1 or more symptoms. Symptom frequencies for the ordinal symptom and composite variables are presented in supplemental Table S1.

Statistical Analysis

Structural equation models (SEM) were estimated with Mplus 8.1 (Muthén & Muthén, 1998–2015). We used the means and variance adjusted weighted least squares (WLSMV) estimation method, which uses pairwise deletion for missing data, for models including the categorical psychopathology variables. Maximum likelihood (ML) estimation, which uses full-information maximum likelihood (FIML) for missing data, was used for all other models.

Model fit was assessed using the χ² statistic, the comparative fit index (CFI), and the root mean square error of approximation (RMSEA). As recommended by Hu and Bentler (1998), we used a CFI>.95 and RMSEA<.06 as indications of good fit. For phenotypic models, we used an alpha level of .05 for the z-test formed by the ratio of the parameter to its standard error to assess significance of the parameters. For genetic models, we used chi-square difference (Δχ²) tests (Satorra & Bentler, 2001) and an alpha level of .05 to assess the significance of all parameters (Neale, Heath, Hewitt, Eaves, & Fulker, 1989). Phenotypic analyses, which were used to examine the independent effects of neuroticism, depressive rumination, and anger rumination on psychopathology, were conducted using Mplus’s TYPE=COMPLEX option to account for the data being nested within families. This option provides scaled chi-square values and standard errors that are robust to nonindependence.

Twin models were used to examine genetic and environmental influences on neuroticism, rumination, and psychopathology. For these models, twins were randomized to be twin 1 or 2 using the same random assignments as prior work (Friedman et al., 2016). Twin models estimate additive (A) and dominant (D) genetic influences, and shared (C) and (E) nonshared environmental influences on the trait of interest. An MZ correlation (rMZ) that is greater than the DZ correlation (rDZ) indicates genetic effects, because MZ’s and DZ’s genomes are, on average, 100% and 50% identical by descent, respectively. An rMZ that is more than twice the rDZ indicates dominant genetic effects in addition to additive genetic effects. Shared environmental influences, which make twins more similar (e.g., family socioeconomic status), are indicated if the rMZ is less than twice the rDZ. As MZ twins share their genes and their shared environments, differences between them (i.e., rMZ<1.0) support the influence of nonshared environmental influences (e.g., individual-specific life events). The E estimate also captures measurement error for manifest variables. The correlation patterns are used to determine whether an ACE or ADE model is most appropriate for each measure, as classical twin models cannot estimate D and C influences simultaneously. However, as described in the results section, the D and C paths accounted for little variance, and so we used a more parsimonious AE model for all final genetic analyses.

We used AE Cholesky decompositions to examine the genetic and environmental relations among multiple variables. The Cholesky decomposition partitions the covariance matrix with a set of orthogonal AE components. For example, in a trivariate Cholesky decomposition, the first A and E components represent all of the genetic and nonshared environmental influences on the first latent variable, which are allowed to predict the second and third latent variables. The second set of A and E components represent the remaining genetic influences on the second latent variable (i.e., unrelated to the first latent variable), and these paths are allowed to predict the third latent variable. The last A and E represent all of the genetic and nonshared environmental influences that are unique to the final latent variable. To improve the interpretability of the results from the Cholesky decompositions, we report the percent of the variance explained by genetic and environmental influences that are unique versus overlapping with another variable in addition to the standardized path estimates (which are included in the figures). This value is calculated by squaring the standardized path estimate and multiplying it by 100.

Results

Factor Structure and Measurement Invariance

Descriptive statistics and zero-order correlations are presented in supplement Table S2 and S3, respectively. As described earlier, we used composite internalizing and externalizing psychopathology variables rather than latent variables given the complexity of the genetic models (Table S1).

We created latent variables of Neuroticism and both ruminative subtypes, and tested each measurement model for invariance across sex (see supplement section S6 for additional details about invariance testing procedures and model fit statistics). The scores from the 9- and 12-item versions of the EPQ neuroticism subscale (administered at ages 17 and 21, respectively) were used as two separate indicators of the Neuroticism latent variable. A factorially invariant model of Neuroticism with constrained factor variances across sex fit the data adequately. As in previous work (du Pont et al., 2018) we used a two-factor model of Anger and Depressive Rumination. Noninvariance of the Anger and Depressive Rumination indicators was driven by the ARS Thoughts of Revenge subscale, and the RRS Brooding subscale. Scale-specific sex differences in genetic and environmental influences on rumination similarly revealed noninvariance that was confined to the Thoughts of Revenge and Brooding subscales (see supplement Table S4). The fact that these sex differences were not consistent across measures, and that sex noninvariance was confined to two indicators, led us to examine the combined sample and use sex residualized variables for all models. Model fit for the correlated Neuroticism and Rumination measurement models was adequate, χ²(22)=86.72, p<.001, RMSEA=.059 [.046, .072], CFI=.978.

Phenotypic Analyses

The correlational model for Neuroticism, Anger Rumination, and Depressive Rumination is presented in Figure 1. Collapsing the Neuroticism and Depressive Rumination latent variables into one factor significantly hurt model fit, Δχ²(2)=33.94, p<.001. Similarly, we were unable to collapse the Neuroticism and Anger Rumination latent variables into one factor, Δχ²(2)=67.60, p<.001. As described in du Pont et al. (2018), collapsing Anger and Depressive rumination into a single factor (without Neuroticism in the model) also significantly hurt model fit, Δχ²(2)=172.12, p<.001.

Figure 1.

Correlation model of Neuroticism, Anger Rumination, and Depressive Rumination. Standardized coefficients are presented, and standard errors are in parentheses. Ellipses and rectangles represent latent and manifest variables, respectively. RRS-B, Ruminative Responses Scale-Brooding subscale; RRS-R, Ruminative Responses Scale-Reflection subscale; RRQ-Ru, Rumination-Reflection Questionnaire-Rumination subscale; AM, Anger Rumination Scale-Angry Memories subscale; TR, Anger Rumination Scale-Thoughts of Revenge subscale; AA, Angry Afterthoughts subscale; UC, Anger Rumination Scale-Understanding Causes, EPQ-21 and EPQ-17, Eysenck Personality Questionnaire neuroticism scale at age 21 and 17. All paths were statistically significant, *p<.05.

Independent associations with psychopathology.

Using multiple regression, we examined whether Neuroticism, Depressive Rumination, and Anger Rumination have independent associations with internalizing and externalizing psychopathology (Figure 2). Neuroticism (β=.46, p<.001) and Depressive Rumination (β=.34, p<.001) had positive independent associations with internalizing psychopathology. Anger Rumination had a negative association with internalizing psychopathology (β=−.24, p=.001) after controlling for Neuroticism and Depressive Rumination. Neuroticism (β=.24, p=.012) and Anger Rumination (β=.15, p=.029) had positive associations with externalizing psychopathology above and beyond the other predictors. Depressive Rumination was not significantly associated with externalizing after controlling for Neuroticism and Anger Rumination (β=–.06, p=.486)³.

Figure 2.

Multiple regression model of Neuroticism, Depressive Rumination, Anger Rumination, and psychopathology. Ellipses represent latent variables and rectangles represent the composite psychopathology variables. For simplicity, the manifest variables that load onto the latent variables are not depicted. *p<.05. Dashed lines indicate p>.05.

Genetic Analyses

The pattern of twin correlations (Table 1) indicated that modeling dominant (D) genetic influences would be more appropriate than common environmental (C) influences; however, univariate AE models, in which the D was constrained to zero, hurt model fit for only 2 of the 9 indicators.⁴ This does not mean that there are no non-additive genetic influences on rumination and neuroticism, but rather reflects the low power of the twin design to differentiate between additive and dominant genetic influences (Martin, Eaves, Kearsey, & Davies, 1978; Posthuma & Boomsma, 2000). Thus, we conducted AE models for all analyses. Heritability estimates for Neuroticism, Depressive Rumination, and Anger Rumination were 48%, 36%, and 47% in the present sample.

Table 1.

Within-trait and Cross-trait Twin Correlations (MZ/DZ)

	Depressive Rumination	Anger Rumination	Neuroticism	Internalizing Psychopathology	Externalizing Psychopathology
Depressive Rumination	.47*/.08
Anger Rumination	.71*.45*/.06	.52*/.09
Neuroticism	.69*.46*/.02	.62*.45*/.14+	.59*/.08
Internalizing	.51*.24*/.07	.27*.21*/.09	.59*.58*/.09	.44*/.40*
Externalizing	.22*.12+/−.03	.28*.24*/.05	.41*.26*/.03	.33*.31*/.10	.61*/.42*

Note: All variables were regressed on sex. MZ/DZ cross-twin within-trait correlations are presented on the diagonal. Within-twin phenotypic correlations are presented above the MZ/DZ cross-twin cross-trait correlations in the off-diagonal. These correlations came from a model in which the cross-twin cross-trait correlations were constrained to be equal across twins, and the within-twin cross-trait (i.e., phenotypic) correlations were constrained to be equal across twins and zygosity, χ²(499)=511.51, p=.340, RMSEA=.011[.000, .026], CFI=.993. *p<.05, ⁺ p<.10.

Genetic and environmental influences on anger and depressive rumination.

We used a bivariate Cholesky decomposition to determine the extent to which Anger Rumination is influenced by genetic and environmental influences that also influence Depressive Rumination, χ²(195)=260.98, p=.001, RMSEA=.042[.027, .055], CFI=.979 (Figure 3a). A substantial proportion of the variance (41%) in Anger Rumination was attributable to genetic influences that also influence Depressive Rumination (A1, Δχ²(1)=41.38, p<.001). Nonshared environmental influences on Depressive Rumination explained 16% of the variance in Anger Rumination (E1, Δχ²(1)=73.28, p<.001). Thirty-seven percent of the variance in Anger Rumination was attributable to nonshared environmental influences specific to Anger Rumination (E2, Δχ²(1)=107.62, p<.001), whereas Anger Rumination-specific genetic influences only explained 7% of the variance in Anger Rumination, Δχ²(1)=0.81, p=.367.

Figure 3.

Cholesky decomposition of (a) Anger Rumination and Depressive Rumination, and (b) Neuroticism, Depressive Rumination, and Anger Rumination. Ellipses represent latent variables. For simplicity, the manifest variables that load onto the latent variables are not depicted. A1-A3 indicate additive genetic influences and E1-E3 indicate nonshared environment. *p<.05, based on chi-square difference tests. Dashed lines indicate p>.05.

Genetic and environmental influences on rumination and neuroticism.

Next, a multivariate Cholesky decomposition was used to determine the extent to which neuroticism and rumination are genetically distinct, χ²(320)=441.37, p<.001, RMSEA=.042[.032, .051], CFI=.966 (Figure 3b). Results indicated that 24% of the variance in Depressive Rumination, Δχ²(1)=26.58, p<.001, and 28% of the variance in Anger Rumination, Δχ²(1)=31.31, p<.001, was explained by the genetic influences on Neuroticism (A1). Although not statistically significant, 11% of the variance of Depressive Rumination was explained by genetic influences not shared with Neuroticism (A2; Δχ²(2)=2.96, p=.228)⁵; these same influences also explained 13% of the variance in Anger Rumination, Δχ²(1)=1.96, p=.162. Only 6% of the variance in Anger Rumination was attributable to Anger Rumination-specific genetic influences (A3; Δχ²(1)=0.63, p=.419), controlling for Neuroticism and Depressive Rumination.

The nonshared environmental influences on Neuroticism (E1) explained 19% of the variance in Depressive Rumination, Δχ²(1)=22.30, p< .001, and 10% of the variance in Anger Rumination, Δχ²(1)=18.72, p< .001. After controlling for Neuroticism, 44% of the variance in Depressive Rumination was explained by nonshared environmental influences (E2). These influences only explained 8% of the variance in Anger Rumination, Δχ²(1)=7.64, p<.001. Anger Rumination-specific nonshared environmental influences (E3) explained 35% of the variance in Anger Rumination, Δχ²(1)=11.81, p<.001. Together, these results indicate that Depressive and Anger Rumination predominately share genetic influences with Neuroticism and are distinguished from Neuroticism and each other primarily by nonshared environmental influences.

Genetic and environmental influences on neuroticism, rumination, and psychopathology.

Internalizing psychopathology.

The Cholesky decomposition examining internalizing psychopathology, χ²(432)=451.05, p=.254, RMSEA=.028[.000, .028], CFI=.988, is depicted in Figure 4a. Genetic influences on Neuroticism explained the most variance in internalizing psychopathology (A1; 40%; Δχ²(1)=24.88, p<.001). After controlling for Neuroticism, genetic influences on Depressive Rumination (A2, Δχ²(1)=2.76, p=.097) explained 5% of the internalizing psychopathology variance, and genetic influences specific to Anger Rumination also explained 5% of the variance in internalizing psychopathology (A3, Δχ²(2)=1.96, p=.560).⁶ There were no genetic influences specific to internalizing psychopathology (A4).

Figure 4.

Cholesky decomposition of Neuroticism, Anger Rumination, Depressive Rumination, and (a) a composite measure of internalizing psychopathology, or (b) externalizing psychopathology. Ellipses represent latent variables. For simplicity, the manifest variables that load onto the latent variables are not depicted. A1-A4 indicate additive genetic influences and E1-E4 indicate nonshared environment. *p<.05, based on chi-square difference tests. Dashed lines indicate p>.05.

In contrast, only 3% of the internalizing variance was attributable to nonshared environmental influences on Neuroticism (E1; Δχ²(2)=3.21, p=.201⁶). Twelve percent of the variance in internalizing psychopathology was attributable to nonshared environmental influences on Depressive Rumination controlling for Neuroticism, Δχ²(1)=7.69, p=.006. Anger Rumination-specific nonshared environmental influences explained 2% of the variance in internalizing, Δχ²(1)=3.80, p=.051. The largest percentage of variance explained by nonshared environmental influences was attributable to internalizing-specific influences (33%).

Externalizing psychopathology.

The Cholesky decomposition with externalizing psychopathology, χ²(432)=463.37, p=.1434, RMSEA=.014[.000, .031], CFI=.980, is depicted in Figure 4b. Genetic influences on Neuroticism (A1) explained 9% of the externalizing variance, Δχ²(1)=7.84, p=.006, and genetic influences on Depressive Rumination controlling for Neuroticism (A2) explained 3% of the externalizing variance, Δχ²(1)=1.02, p=.313. Most of the genetic variance in externalizing was unique (A4; 39%), but constraining this path to zero did not significantly hurt model fit, Δχ²(1)=1.28, p=.258. Closer examination of the path estimates revealed that when the A4 path was constrained to zero, the A variance was redistributed across A2 and A3. Dropping the path from A2 to externalizing did not hurt model fit, Δχ²(1)=.1.02, p=.313, nor did dropping the path from A3 to externalizing, Δχ²(1)=0.71, p=.401. A two degree of freedom test, in which the paths from A3 and A4 to externalizing were dropped, similarly did not hurt model fit, Δχ²(2)=1.13, p=.567. However, dropping the A2, A3, and A4 paths associated with externalizing psychopathology did significantly hurt model fit, Δχ²(3)=49.79, p<.001. This is indicative of significant common genetic influences on externalizing psychopathology that is uncorrelated with Neuroticism, but we are unable to determine whether this variance is overlapping with depressive and anger rumination or specific to externalizing.

Only 2% of the variance in externalizing psychopathology was attributable to nonshared environmental influences on Neuroticism (E1; Δχ²(1)=2.21, p=.137). After controlling for Neuroticism, nonshared environmental influences on Depressive Rumination (E2; Δχ²(1)=1.29, p=.255) explained 1% of the variance in externalizing, and nonshared environmental influences specific to Anger Rumination (E3; Δχ²(1)=.081, p=.776) explained less than 1% of the externalizing variance. Nonshared environmental influences on externalizing were largely specific to externalizing psychopathology, and explained 33% of the variance in externalizing.

Taken together, these results suggest that the genetic variance, but not the environmental variance, that rumination shares with Neuroticism is associated with psychopathology and that the environmental variance that distinguishes rumination from Neuroticism also predicts internalizing psychopathology.

Discussion

The present study examined genetic and environmental influences on Neuroticism, Depressive Rumination, and Anger Rumination in a sample of young adults. Results indicated that Neuroticism, Depressive Rumination, and Anger Rumination are highly genetically related. Furthermore, the common genetic influences on Neuroticism and rumination also influence internalizing and externalizing psychopathology. The separability of Neuroticism and Rumination is largely due to unique nonshared environmental influences. Nonshared environmental influences specific to rumination were also related to internalizing psychopathology, indicating that these environmental influences are likely responsible for the independent phenotypic association between rumination and internalizing psychopathology after controlling for Neuroticism. Examination of the two ruminative subtypes indicated a similar pattern: Anger and Depressive Rumination have common genetic influences but are differentiated by nonshared environmental influences. We elaborate on these findings and their implications in the following sections.

Rumination and Neuroticism have a Shared Genetic Etiology

Our results suggest that Rumination and Neuroticism have considerable genetic overlap. In models that included Neuroticism and the two ruminative subtypes, there were no significant genetic influences on rumination after statistically accounting for Neuroticism. This finding suggests that genome-wide studies of neuroticism may also provide information about rumination, which has never been examined at the genome-wide level. For example, Luciano and colleagues (2018) found that genes influencing neuroticism are implicated in biological pathways involved in antidepressant action. They suggested that neuroticism may be a clinically useful tool to identify people who may respond well to antidepressants, and our results indicate that rumination may be of similar clinical utility.

Although the majority of the genetic influences on rumination overlapped with genetic influences on Neuroticism, rumination-specific genetic influences were statistically significant when we included additional information in the model (i.e., a measure of internalizing psychopathology). This suggests that there are likely some genes that directly influence rumination. Rumination-specific genetic influences did not explain a large proportion of the variance in Depressive or Anger Rumination, and did not have a statistically significant overlap with variance in psychopathology. However, it is possible that a small genetic relation between rumination and psychopathology after controlling for neuroticism may be statistically significant in a larger sample. It is also possible that different ruminative subtypes have a shared genetic etiology with some psychiatric disorders that is independent from neuroticism, though theoretical and empirical research on the genetic relations between ruminative subtypes and psychopathology is lacking.

Neuroticism and rumination were primarily differentiated by nonshared environmental influences. Our use of latent variables separated random error from nonshared environmental influences, suggesting that these nonshared environmental influences are not random measurement error. Prior research has identified several potential nonshared environmental factors that are related to neuroticism and rumination, including childhood trauma (e.g., Kim, Jin, Jung, Hahn, & Lee, 2017) and stressful life events (e.g., Jeronimus, Ormel, Aleman, Penninx, & Riese, 2013). However, little is known regarding the environmental influences that are specific to rumination versus neuroticism. Understanding environmental factors that are specific to neuroticism and rumination may facilitate assessment of risk and protective factors that contribute to or buffer a person from neuroticism and rumination.

Common Genetic Influences on Neuroticism and Rumination, but Rumination-Specific Environmental Influences, Predict Internalizing Psychopathology

We found a substantial genetic overlap of rumination and internalizing psychopathology with Neuroticism (rA =.90), which is consistent with prior studies (rA=.58–.82; Hettema et al., 2006). Furthermore, prior work in the present sample has reported strong genetic correlations between rumination and depression symptoms and diagnoses (rA=.40–.82; Johnson et al., 2014), which is consistent with our results that indicate genetic overlap between rumination and internalizing. Our finding that Neuroticism, rumination, and internalizing psychopathology have some shared genetic etiology is consistent with two existing theories: the vulnerability model, which hypothesizes that processes like rumination mediate the causal association between neuroticism and internalizing psychopathology, and the common cause model, which suggests that neuroticism and internalizing psychopathology share a genetic and environmental etiology (Ormel et al., 2013).

Common genetic influences on neuroticism, rumination, and psychopathology also indicate that rumination may be a potential candidate for inclusion in a multi-trait analysis of genome-wide association studies (MTAG; Turley et al., 2018). This method analyzes genome-wide association study (GWAS) summary statistics of genetically correlated traits to improve statistical power to detect genetic correlations with each of the examined traits. MTAG can be used to create polygenic risk scores that explain more variance than traditional single-trait methods; thus, examination of neuroticism, rumination, and psychopathology using this approach could improve the prediction of genetic risk for psychopathology (see Turley et al., 2018 for an example using depression, neuroticism, and subjective well-being).

A different pattern of results emerged for environmental influences. Environmental influences on Neuroticism did not significantly influence internalizing psychopathology, indicating that the neuroticism–internalizing association is primarily genetic. There was evidence of rumination-specific nonshared environmental influences that also influenced internalizing psychopathology. The overlapping nonshared environmental influences on rumination and internalizing suggests that changes in rumination may also lead to changes in symptoms of internalizing psychopathology, supporting the clinical utility of interventions that target rumination (e.g., rumination-focused cognitive behavioral therapy; Watkins et al., 2011).

With respect to externalizing psychopathology, we found statistically significant additive genetic influences on rumination and externalizing that overlapped with genetic influences on Neuroticism. However, the genetic correlation between Neuroticism and externalizing was .38, explaining only 9% of the variance in externalizing psychopathology. This finding is consistent with the weak-to-moderate genetic correlation between neuroticism and externalizing found in prior work (e.g., Kendler, Prescott, Myers, & Neale, 2003). There were no statistically significant nonshared environmental influences on externalizing psychopathology that were shared with rumination and Neuroticism, indicating that the majority of genetic and nonshared environmental influences on externalizing psychopathology were externalizing-specific. Our results also converge with prior phenotypic work indicating that anger and depressive rumination together explain less of the variance in externalizing (7–15%) than internalizing psychopathology (38–70%), and that depressive rumination is more strongly associated with internalizing than externalizing psychopathology (du Pont et al., 2018).

Although the majority of the variance in externalizing was not explained by neuroticism or rumination, we found that the correlation between internalizing and externalizing psychopathology decreased from .33 to .14 when internalizing and externalizing psychopathology were regressed on neuroticism, anger rumination, and depressive rumination. This finding suggests that neuroticism and rumination explains a portion of the covariance between internalizing and externalizing or, as suggested by Ormel et al. (2013), capture general risk for psychopathology.

Anger and Depressive Rumination have Specific Environmental Influences

Our results indicate that Anger and Depressive Rumination have common genetic and nonshared environmental influences, and are differentiated from each other primarily by nonshared environmental influences. Future work identifying environmental factors and experiences that are specific to anger versus depressive rumination may provide important information about risk factors for rumination and psychopathology as well as potential targets for prevention initiatives and interventions.

As noted previously, prior work has examined environmental factors associated with depressive rumination. Less is known about environmental influences that are associated with anger rumination. As anger is often elicited in response to unfair behavior or one’s goals being blocked (Berkowitz & Harmon-Jones, 2004), nonshared environmental influences related to perceived injustice, structural discrimination, or rejection may be specific to anger rumination. For example, research by Borders & Hennebry (2015) found that anger rumination was associated with self-reported discrimination among ethnic minority participants. Given the nascent literature on anger rumination, further identification of potential environmental factors that precipitate heightened and habitual patterns of rumination and direct examination of environmental influences that facilitate the development of distinct ruminative subtypes is needed.

Limitations and Future Directions

Our findings should be interpreted in the context of the following limitations. First, our measures of neuroticism came from earlier time points than our rumination and psychopathology measures. These data were collected as a part of a larger longitudinal study, but the measures of rumination were administered only at age 23 and the neuroticism measure was administered up until the age 21. This limited our analysis of rumination to age 23, and prevented us from using a concurrent measure of neuroticism. By using earlier measures of neuroticism, it is possible that nonshared environmental influences on rumination that overlap with internalizing psychopathology may in part be explained by age-specific variance. Prior work suggests that genetic influences on neuroticism are largely stable from adolescence into adulthood (Nivard, Middeldorp, Dolan, & Boomsma, 2015), and we used latent variables with indicators from age 17 and 21 to minimize age-specific measurement error. However, future work should replicate our results in a sample with concurrent neuroticism and rumination measures. As genetic and environmental influences can change across the lifespan, future work should also explore the extent to which the genetic and environmental influences that are common and specific to rumination and neuroticism vary across the lifespan.

Second, the cross-sectional twin design of this study can provide evidence that is consistent with genetic or environmental mediation, but does not rule out non-causal relationships. Examining these questions using genetically-informed longitudinal designs (e.g., longitudinal multivariate or discordant twin designs; Pingault et al., 2018; Rutter, 2007) is an essential next step to understanding the ways in which neuroticism and rumination contribute to psychopathology.

An additional challenge of working with neuroticism, rumination, and internalizing disorders is the inherent overlap between these constructs. Our internalizing composite variable included GAD, a disorder characterized by excessive and uncontrollable worry. Prior studies suggest strong correlations between worry and rumination (e.g., rs =.49-.61; de Jong-Meyer et al., 2009; Gustavon et al., 2018), which may have led rumination to explain more of the variance in internalizing than it would if other internalizing disorders were included. Inflated relations between overlapping constructs can be a result of criterion contamination (Sullivan, 2009), which can occur when measures of predictors and outcomes share similar item content. Methodological overlap can be remediated sometimes by removing specific items (see footnote 3). However, as noted by Riese, Ormel, Aleman, Servaas, & Jeronimus (2016), depressive and anxious traits are two of the facets of neuroticism. The problem of criterion contamination is an important and ongoing discussion in the field because it affects the interpretation of results (e.g., is there evidence of mediation or simply redundancy in the measures?) and may hinder theoretical advances in this area.

Future research should also examine the extent to which genetic and environmental influences specific to rumination predict psychopathology in a clinically-ascertained sample. The sample used in the present study was a population sample of young adults, which is consistent with National Institute for Mental Health’s dimensional Research Domain Criteria (RDoC; Insel et al., 2010) approach and prevents attenuated correlations due to a restricting the range of values in our sample (Sackett & Yang, 2000). However, as expected, psychopathology endorsement was lower in the current sample than would be expected in a clinical sample, which led us to use lifetime rather than current symptom endorsement. Prior work indicates that the structure of rumination may vary across depression symptom severity (Whitmer & Gotlib, 2011), so it may be important to examine the extent to which rumination-specific genetic and environmental influences are associated with current psychopathology symptoms in a clinical sample.

Lastly, although the present study was well powered to answer the primary research questions, it was not adequately powered to explore additional questions about sex differences. Previous research indicates that mean levels of internalizing disorders are higher in women, whereas mean levels of externalizing disorders are higher in men (Eaton et al., 2012). However, prior research examining sex differences in depressive rumination and depression in the present sample found small sex differences that were largely measure-specific, and no evidence of sex as a moderator of the rumination–depression association (Johnson et al., 2014).

Conclusions

Our results indicate that rumination has genetic and environmental overlap with neuroticism, and is distinguished from neuroticism by nonshared environmental influences. Rumination’s genetic overlap with neuroticism, rather than genetic influences specific to rumination, are related to both internalizing and externalizing psychopathology. In contrast, rumination’s environmental overlap with neuroticism was unrelated to internalizing and externalizing psychopathology and rumination-specific nonshared environmental influences were associated with internalizing psychopathology. That is, the environmental influences that distinguish rumination from neuroticism explain the independent association between rumination and internalizing psychopathology after controlling for neuroticism. Our findings highlight the genetic overlap between neuroticism and rumination and the importance of understanding and integrating environmental factors that augment risk for rumination into models of psychopathology.