Research has shown that impulsive personality traits (IPTs) confer transdiagnostic risk for psychopathology with an important role in disorders of the externalizing spectrum (e.g. substance use disorders, conduct disorder, antisocial personality disorder; Creswell, Wright, Flory, Skrzynski, & Manuck, Reference Creswell, Wright, Flory, Skrzynski and Manuck2019; Johnson, Carver, & Joormann, Reference Johnson, Carver and Joormann2013). Broadly, IPTs are characterized by lack of self-control and forethought of behavioral consequences in response to more temporally salient external stimuli or internal impulses (Whiteside & Lynam, Reference Whiteside and Lynam2001). Twin and genome-wide association studies (GWAS) have demonstrated that IPTs are heritable, though estimates vary across specific traits under study (Bezdjian, Baker, & Tuvblad, Reference Bezdjian, Baker and Tuvblad2011; Friedman et al., Reference Friedman, Hatoum, Gustavson, Corley, Hewitt and Young2020; Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019). Neuroimaging studies also suggest heterogeneity in the neural correlates of IPTs but support a general pattern of differential brain morphology with cortical regions involved in cognitive control and attention (orbitofrontal cortex), subcortical regions involved in reward- and emotion-processing (ventral striatum, amygdala), and connections between these regions relevant to both (mesocorticolimbic and frontostriatal pathways; Johnson, Elliott, & Carver, Reference Johnson, Elliott and Carver2020; Pan et al., Reference Pan, Wang, Zhao, Lai, Qin, Li and Gong2021).
In aggregate, the described studies support the hypothesis that IPTs may serve as useful endophenotypes for externalizing psychopathology (Cyders, Coskunpinar, & VanderVeen, Reference Cyders, Coskunpinar, VanderVeen, Zeigler-Hill and Marcus2016; Jonas & Markon, Reference Jonas and Markon2014). The endophenotype approach argues that studying genetic influences underlying intermediate constructs that confer risk for a manifest disorder may help identify shared neurobiological and genetic factors underlying that disorder (Hall & Smoller, Reference Hall and Smoller2010). Conceptualizations of IPTs, however, exhibit substantial heterogeneity, that while meaningful, contributes to a lack of clarity regarding relations between IPTs and clinical presentations (Strickland & Johnson, Reference Strickland and Johnson2021). Similarly, the indiscriminate use of multidimensional IPTs as endophenotypes in genetic studies may hamper the ability to identify causal loci. To address this, we argue that molecular genetic investigations of IPT models rooted in developmental and neurobiological frameworks, such as dual-systems models (Shulman et al., Reference Shulman, Smith, Silva, Icenogle, Duell, Chein and Steinberg2016b), can be used to develop, evaluate, and refine conceptualizations of IPTs as endophenotypes for psychopathology.
Dual-systems models posit that impulsive behaviors are the result of two complementary neurobiological systems associated with distinct neural substrates acting in dynamic tension to influence behavior: (1) a bottom-up system, involving activation of subcortical regions (ventral striatum, amygdala) involved in reward (e.g. sensation seeking) and/or emotion-based drive (e.g. urgency), and (2) a top-down system, involving activation of prefrontal cortical regions (PFC; orbitofrontal cortex) involved in effortful control and forethought (e.g. self-control; Carver & Johnson, Reference Carver and Johnson2018; Shulman et al., Reference Shulman, Smith, Silva, Icenogle, Duell, Chein and Steinberg2016b). Notably, dual-systems models align empirical neurocognitive observations with developmental theory (Steinberg et al., Reference Steinberg, Albert, Cauffman, Banich, Graham and Woolard2008). The transition from adolescence to adulthood is characterized by a developmental ‘spike’ in risky, impulsive behaviors driven by rapid increases in sensitivity to reward and affective salience (sensation seeking, urgency; Lopez-Vergara, Spillane, Merrill, & Jackson, Reference Lopez-Vergara, Spillane, Merrill and Jackson2016; Shulman, Harden, Chein, & Steinberg, Reference Shulman, Harden, Chein and Steinberg2016a) paired with relatively slower maturation of PFC regions that govern inhibition, planning, and self-control.
Post-GWAS investigations of dual-systems models can provide novel empirical support for these models and assessment of their constituent constructs and potentially aid in their refinement. Further, parsing genetic liability for increased bottom-up approach behaviors or lack of top-down cognitive control has the potential to identify unique risk pathways to psychopathology consistent with the endophenotype approach. Nonetheless, behavior genetic studies conducted to date suggest this may be difficult as prior twin research examining genetic influences for distinct top-down and bottom-up IPTs have indicated overlap between the two systems resulting from shared genetic factors (Ellingson et al., Reference Ellingson, Slutske, Vergés, Littlefield, Statham and Martin2018; Ellingson, Vergés, Littlefield, Martin, & Slutske, Reference Ellingson, Vergés, Littlefield, Martin and Slutske2013; Hur & Bouchard, Reference Hur and Bouchard1997). Though not explicitly testing dual-systems models, recent GWAS have shown that traits putatively characterizing top-down lack of self-control (lack of premeditation, non-planning impulsivity) are highly genetically correlated with each other and uncorrelated with putative bottom-up reward-based traits (sensation seeking; Gustavson et al., Reference Gustavson, Friedman, Fontanillas, Elson, Palmer and Sanchez-Roige2020; Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019). IPTs also demonstrate variability in their genetic correlations with other traits and disorders. For instance, emotion-based IPTs (urgency) show stronger genetic correlations with internalizing psychopathology (Gustavson et al., Reference Gustavson, Friedman, Fontanillas, Elson, Palmer and Sanchez-Roige2020), while lack of self-control traits show stronger genetic correlations with externalizing psychopathology (Linnér et al., Reference Linnér, Mallard, Barr, Sanchez-Roige, Madole, Driver and Dick2021; Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019).
Heterogeneity in genetic correlations across IPTs can be difficult to interpret in the absence of strong theoretical models. Dual-systems models provide a potential framework for interpreting such results, but there is a lack of research attempting to validate these models and corresponding measures at the genetic level by examining their hypothesized relations to distinct neuroanatomical variation. Recent advances in modeling of GWAS summary statistics allow for theory-driven examinations of the interrelations among genetic influences on psychological traits as well as their genetic relations with hypothesized neural correlates to examine support for existing theories such as dual-systems models.
The aims of the present study were two-fold. First, the study aimed to leverage extant IPT GWAS and an advanced multivariate GWAS approach (GenomicSEM; Grotzinger et al., Reference Grotzinger, Rhemtulla, de Vlaming, Ritchie, Mallard, Hill and Tucker-Drob2019) to quantify and parse sources of unique and shared variance associated with dual-systems constructs. Separable factors of two distinct bottom-up constructs and a single top-down construct were hypothesized and empirically evaluated: (1) a reward-based bottom-up factor (sensation seeking), (2) an emotion-based bottom-up factor (urgency), and (3) a common top-down factor (lack of self-control). Second, the study aimed to distinguish between neurogenetic influences related to top-down and bottom-up constructs by examining genetic correlations with neuroimaging phenotypes. While these analyses were comparatively exploratory in nature, it was expected that dual-systems constructs would exhibit separable but overlapping genetic associations with brain regions implicated by previous phenotypic research and theory: lack of self-control with PFC regions, sensation seeking and urgency with subcortical regions involved in reward- and emotion-processing, respectively. Though previous GWAS research has examined differences in correlations between IPTs and other relevant behavioral phenotypes, including psychological disorders (Linnér et al., Reference Linnér, Biroli, Kong, Meddens, Wedow, Fontana and Beauchamp2019; Reference Linnér, Mallard, Barr, Sanchez-Roige, Madole, Driver and Dick2021; Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019), examinations of shared genetic architecture between IPTs and neuroanatomical features are conspicuously absent from both twin and GWAS literatures.
Methods
Table 1 contains descriptions of GWAS summary statistics for all phenotypes used. Summary statistics were restricted to individuals of European ancestry and common variants (minor allele frequency [MAF] > 0.01). See online Supplementary Methods for descriptions of genotyping, imputation, quality control, meta-analytic procedures, and additional measurement information for GWAS.
Table 1. Overview of GWAS used in study
Note: MAF, minor allele frequency; BIS-11, Barratt Impulsiveness Scale; UPPS-P, UPPS-P Impulsive Behavior Scale; TCI, Temperament and Character Inventory; NFBS, Northern Finland Birth Cohort; YFS, Cardiovascular Risk in Young Finns Study; HBCS, Helsinki Birth Cohort Study; QIMR, Australian Twin Registry; UKB, UK Biobank; CHARGE, Cohorts for Heart and Aging Research in Genomic Epidemiology; ENIGMA, Enhancing Neuro Imaging Genetics Through Meta-Analysis Consortium.
Publications: 1. Sanchez-Roige et al. (Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019), 2. Service et al. (Reference Service, Verweij, Lahti, Congdon, Ekelund, Hintsanen and Freimer2012), 3. Linnér et al. (Reference Linnér, Biroli, Kong, Meddens, Wedow, Fontana and Beauchamp2019), 4. Smith et al. (Reference Smith, Douaud, Chen, Hanayik, Alfaro-Almagro, Sharp and Elliott2021), 5. Grasby et al. (Reference Grasby, Jahanshad, Painter, Colodro-Conde, Bralten and Hibar2020), and 6. Satizabal et al. (Reference Satizabal, Adams, Hibar, White, Knol, Stein and Ikram2019).
Impulsive personality trait GWAS
GWAS IPT phenotypes were primarily measured using the brief version of the UPPS-P (Cyders, Littlefield, Coffey, & Karyadi, Reference Cyders, Littlefield, Coffey and Karyadi2014b), the BIS-11 (Patton, Stanford, & Barratt, Reference Patton, Stanford and Barratt1995), and Cloninger's Temperament and Character Inventory (TCI; Cloninger, Przybeck, Svrakic, and Wetzel, Reference Cloninger, Przybeck, Svrakic and Wetzel1994). Summary statistics were obtained from three primary sources: UK Biobank (UKB; risk-taking, Linnér et al., Reference Linnér, Biroli, Kong, Meddens, Wedow, Fontana and Beauchamp2019), direct-to-consumer genetics company 23andMe, Inc. (Sunnyvale, CA; BIS-11, UPPS-P, risk-taking, and adventurousness; Linnér et al., Reference Linnér, Biroli, Kong, Meddens, Wedow, Fontana and Beauchamp2019; Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019), and a meta-analytic sample comprised of four European-ancestry cohorts (TCI harm avoidance and novelty seeking; Service et al., Reference Service, Verweij, Lahti, Congdon, Ekelund, Hintsanen and Freimer2012).
A priori hypotheses regarding appropriate phenotype structure for dual-systems models, given available GWAS data and prior phenotypic research, led to specification of two separate two-factor confirmatory genomic structural models. The first model, referred to as the sensation seeking-self-control (SSSC) model, consisted of a bottom-up ‘sensation seeking’ factor indexing genetic influences for reward-based drive, and a top-down ‘(lack of) self-control’ factor indexing genetic influences for low self-control, lack of planning, and lack of forethought (see Fig. 1a). The second model, referred to as the urgency-self-control (UGSC) model, consisted of a bottom-up ‘urgency’ factor indexing genetic influences for emotion-based rash action, and the same top-down ‘(lack of) self-control’ factor (see Fig. 1b). Notably, a model containing all three factors (correlated three-factor model) failed to converge. GWAS indicator selection and rationale are described below for each factor.
Figure 1. Final path diagrams of the SSSC (A) and UGSC (B) dual-systems models estimated using GenomicSEM. Presented parameters are standardized and SE are shown in paratheses. Variances and covariances are shown as dashed lines and factor loadings are shown as solid lines. See online Supplementary Tables S3 and S4 for model fit indices. BT, BIS-11 total score; PD, UPPS-P lack of premeditation; NS, TCI novelty seeking; AV, adventurousness; RT, risk-taking; SS, UPPS-P sensation seeking; NU, UPPS-P negative urgency; PU, UPPS-P positive urgency; HA, TCI harm avoidance.
Lack of self-control
BIS-11 total score, UPPS-P lack of premeditation, and TCI novelty seeking GWAS were specified as indicators of the top-down (lack of) self-control factor. The first two indicators were selected given prior research suggesting that BIS-11 subscales demonstrate inadequate psychometric properties (Morean et al., Reference Morean, DeMartini, Leeman, Pearlson, Anticevic, Krishnan-Sarin and O'Malley2014; Reise, Moore, Sabb, Brown, & London, Reference Reise, Moore, Sabb, Brown and London2013) and that BIS-11 total scores and UPPS-P lack of premeditation scores exhibit substantial genetic and phenotypic overlap related to lack of self-control and planning prior to action (Gustavson et al., Reference Gustavson, Friedman, Fontanillas, Elson, Palmer and Sanchez-Roige2020; Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019). Though some reviews have clustered novelty seeking with sensation seeking measures (Fischer, Smith, & Cyders, Reference Fischer, Smith and Cyders2008; Stautz & Cooper, Reference Stautz and Cooper2013), TCI novelty seeking was selected as an indicator of lack of self-control given empirical evidence across a number of samples, including the UPPS development sample (Whiteside & Lynam, Reference Whiteside and Lynam2001), suggesting this scale is more strongly associated with lack of premeditation than with sensation seeking (Evren, Durkaya, Evren, Dalbudak, & Cetin, Reference Evren, Durkaya, Evren, Dalbudak and Cetin2012; Savvidou et al., Reference Savvidou, Fagundo, Fernández-Aranda, Granero, Claes, Mallorquí-Baqué and Jiménez-Murcia2017; Vonmoos et al., Reference Vonmoos, Hulka, Preller, Jenni, Schulz, Baumgartner and Quednow2013). Relatedly, prior research suggests that TCI novelty seeking may not reflect a single construct but rather two: one reflecting characteristics more closely associated with sensation seeking and the other lack of self-control (Evren et al., Reference Evren, Durkaya, Evren, Dalbudak and Cetin2012; Herbst, Zonderman, McCrae, & Costa, Reference Herbst, Zonderman, McCrae and Costa2000; Jaksic et al., Reference Jaksic, Aukst-Margetic, Rózsa, Brajkovic, Jovanovic, Vuksan-Cusa and Jakovljevic2015; Vonmoos et al., Reference Vonmoos, Hulka, Preller, Jenni, Schulz, Baumgartner and Quednow2013). A series of sensitivity analyses examining models including novelty seeking as (1) an indicator of lack of self-control, (2) an indicator of sensation seeking, and (3) omitting novelty seeking from either factor further supported our proposed model (see online Supplementary Methods and Supplementary Tables S3–S5 and S14).
Urgency
UPPS-P negative and positive urgency and TCI harm avoidance GWAS were specified as indicators of the bottom-up urgency factor. Empirical studies have suggested that negative and positive urgency are highly correlated and together may represent a common transdiagnostic risk factor for psychopathology (Billieux et al., Reference Billieux, Heeren, Rochat, Maurage, Bayard, Bet and Baggio2021). This notion is substantiated by high phenotypic and genetic correlations between negative and positive urgency (r = 0.59; rg = 0.74) in the 23andMe sample (Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019). TCI harm avoidance is thought to reflect a tendency to respond intensely to aversive stimuli and negative affect with loss of control of behavioral responses (Cloninger, Reference Cloninger1987) and has been shown to be moderately correlated with negative urgency in clinical samples (r = 0.28–0.55; Jiménez-Murcia et al., Reference Jiménez-Murcia, Granero, Giménez, Del Pino-Gutiérrez, Mestre-Bach, Mena-Moreno and Fernández-Aranda2020; Savvidou et al., Reference Savvidou, Fagundo, Fernández-Aranda, Granero, Claes, Mallorquí-Baqué and Jiménez-Murcia2017).
Sensation seeking
A risk-taking GWAS meta-analysis including both UKB and 23andMe samples (see online Supplementary Methods for description and online Supplementary Fig. S1 for quantile–quantile [Q–Q] plot) was specified along with adventurousness and UPPS-P sensation seeking GWAS as indicators of the bottom-up sensation seeking factor.
Neuroimaging GWAS
Three sets of neuroimaging GWAS were utilized for genetic correlation analyses (see Table 1 and online Supplementary Methods). The first set included UKB GWAS of 62 (31 left/right hemisphere) cortical parcellation volumetric phenotypes obtained from the Oxford Brain Imaging Genetics web server (Smith et al., Reference Smith, Douaud, Chen, Hanayik, Alfaro-Almagro, Sharp and Elliott2021). The second set included GWAS of 34 cortical surface area and thickness parcellation phenotypes (Grasby et al., Reference Grasby, Jahanshad, Painter, Colodro-Conde, Bralten and Hibar2020). The third set included volumetric GWAS of seven subcortical structures (Satizabal et al., Reference Satizabal, Adams, Hibar, White, Knol, Stein and Ikram2019).
Data analysis
Genomic factor models
GenomicSEM (version 0.0.5; Grotzinger et al., Reference Grotzinger, Rhemtulla, de Vlaming, Ritchie, Mallard, Hill and Tucker-Drob2019) was employed using diagonally weighted least-squares estimation and unit variance identification to conduct genomic confirmatory factor analyses. Two primary models were tested: (1) the SSSC model reflecting shared genetic architecture between top-down lack of self-control and bottom-up sensation seeking, and (2) the UGSC model reflecting shared genetic architecture between top-down lack of self-control and bottom-up urgency. Model fit was assessed using χ2 tests, the comparative fit index (CFI), the standardized root mean square residual (SRMR), and the Akaike information criterion (AIC). Given that dual-systems models contained indicator GWAS from the same measures (UPPS-P, TCI) across correlated latent factors, follow-up models allowing within-measure cross-factor residuals to covary were fit with changes in model fit assessed using χ2 difference tests (Δχ2).
Single latent factor models were specified for each of the three dual-systems constructs in multivariate GWAS to: (1) minimize the effect of uneven sample sizes between traits as described in online Supplementary Methods, (2) increase the number of variants tested for each construct as variants are excluded using listwise deletion across indicators, and (3) limit any potential estimation bias introduced by residual covariance structures described above. Because these single latent factor models each had three indicator GWAS (fully saturated just-identified models, df = 0), model fit indices were unavailable, and fit was instead assessed by examining the significance of factor loadings and residual variances.
Multivariate GWAS
Multivariate GWAS of SNPs available across all indicator GWAS and present in the 1000 Genomes Project Phase 3 v5 reference panel with MAF⩾0.5% (The 1000 Genomes Project Consortium, 2015) were conducted in GenomicSEM to estimate SNP associations with each latent dual-systems genetic factor. Effective sample sizes for each latent factor ($\widehat{N}$
) were estimated (Mallard et al., Reference Mallard, Linnér, Grotzinger, Sanchez-Roige, Seidlitz, Okbay and Harden2022a). SNP-based heritability estimates ($h_g^2$
) of latent genetic factors derived using $\widehat{N}$
are more accurately referred to as genetic variances (Mallard et al., Reference Mallard, Linnér, Grotzinger, Sanchez-Roige, Seidlitz, Okbay and Harden2022a) and are subsequently denoted by ζg. To identify SNP effects not fully mediated by the specified latent factor (common pathway model), follow-up multivariate GWAS including unique pathways were conducted to calculate Q SNP tests of heterogeneity. SNPs with genome-wide significant (GWS; p < 5 × 10−8) Q SNP statistics exert effects on genetic indicators independent of the latent factor (Grotzinger et al., Reference Grotzinger, Rhemtulla, de Vlaming, Ritchie, Mallard, Hill and Tucker-Drob2019). Thus, these SNPs were removed from model-derived GWAS summary statistics to reduce heterogeneity in the latent genetic factors for downstream analyses.
Neuroimaging genetic correlation analyses
Linkage disequilibrium score regression (LDSC; Bulik-Sullivan et al., Reference Bulik-Sullivan, Finucane, Anttila, Gusev, Day, Loh and Neale2015) genetic correlation analyses were conducted to examine whether dual-systems constructs differed with respect to their genetic overlap with regional cortical volume, surface area, and thickness and subcortical structural volume. GenomicSEM was used to calculate genetic correlations between each genetic factor and each neuroimaging phenotype with a 5% FDR correction used to account for multiple testing within each imaging phenotype set for each latent construct separately. To determine whether correlations with each neuroimaging phenotype differed between paired top-down and bottom-up constructs, χ2 tests were used to evaluate the null hypothesis that each pair of genetic correlations could be constrained to equality (Demange et al., Reference Demange, Malanchini, Mallard, Biroli, Cox, Grotzinger and Nivard2021).
Results
Genomic factor models
Preliminary univariate and bivariate LDSC estimates for all indicator GWAS are shown in online Supplementary Tables S1, S2 and Fig. S2. SNP-based heritability estimates were all significant at p < 0.05 ($h_g^2$
= 0.040–0.362). Ratio values (LDSC intercept − 1)/(mean χ2 − 1) were not significantly different from zero for most traits, suggesting negligible inflation of test statistics from sources other than true genetic effects (e.g. uncontrolled population stratification). Of note, the novelty seeking and harm avoidance GWAS were likely underpowered, as evidenced by λ GC, mean χ2, and LDSC intercept values below 1, suggesting $h_g^2$
estimates (0.305–0.362) are likely inflated. Nevertheless, genetic correlations between these traits and other constituent indicator GWAS demonstrated appreciable clustering among indicator GWAS for each dual-systems latent genetic factor (online Supplementary Fig. S2). Further, these GWAS contributed to the polygenic signal and ζg of subsequent latent genetic factors as described in the Multivariate GWAS section below.
GenomicSEM analyses showed that the correlated factors dual-systems models provided good fit to genetic covariance matrices. The SSSC model exhibited good fit (χ2 = 10.67, df = 8, p = 0.22, AIC = 36.67, CFI = 1.00, SRMR = 0.09) and was not improved with the inclusion of within-measure cross-factor residual covariation between UPPS-P sensation seeking and lack of premeditation (χ2 = 10.69, df = 7, p = 0.15, AIC = 38.69, CFI = 1.00, SRMR = 0.08; Δχ2 = −0.02, df = 1, p = 0.887). The bottom-up sensation seeking factor and the top-down (lack of) self-control factor in the SSSC model were significantly correlated (rg = 0.60, s.e. = 0.12, p = 2.15 × 10−7; Fig. 1a; online Supplementary Table S3).
The initial UGSC model exhibited poor fit (χ2 = 56.15, df = 8, p = 2.63 × 10−9, AIC = 82.15, CFI = 0.56, SRMR = 0.16), but fit was drastically improved with the inclusion of within-measure cross-factor residual covariances between UPPS-P negative and positive urgency and lack of premeditation and between TCI harm avoidance and novelty seeking (χ2 = 10.67, df = 5, p = 0.058, AIC = 42.67, CFI = 0.95, SRMR = 0.09; Δχ2 = −45.49, df = 3, p = 7.29 × 10−10). The bottom-up urgency factor and the top-down (lack of) self-control factor in the UGSC model exhibited a moderate, but non-significant, correlation (rg = 0.42, s.e. = 0.23, p = 0.063; Figure 1B; online Supplementary Table S4).
For the single factor models, loadings were acceptable to large (λ = 0.38–1.00) and significant at p < 0.05 apart from novelty seeking (λ = 0.38, s.e. = 0.22, p = 0.083). Residual variances were generally small and non-significant apart from novelty seeking, harm avoidance, and risk-taking (ɛ NS = 0.86, s.e. = 0.19, p = 6.83 × 10−5; ɛ HA = 0.78, s.e. = 0.24, p = 0.001; ɛ RT = 0.32, s.e. = 0.05, p = 1.52 × 10−9, respectively). See online Supplementary Table S5 and Fig. S3, respectively, for model parameters and path diagrams.
Multivariate GWAS
The multivariate sensation seeking GWAS ($\widehat{N}$
= 710 971) identified 1092 independent GWS variants (online Supplementary Table S6). LDSC analysis indicated that results reflect the extensive polygenicity of this trait (ζg=0.087, s.e. = 0.003; mean χ2 = 2.26; online Supplementary Table S7 and Fig. S4 for Q-Q plot), and were not due to uncontrolled inflation, bias, or stratification (ratio value = 0.01, s.e. = 0.01). For a more detailed description of these results see Miller and Gizer (Reference Miller and Gizer2023).
In contrast, no variants in the (lack of) self-control ($\widehat{N}$
= 27 656) or urgency GWAS ($\widehat{N}$
= 28 316) reached GWS (online Supplementary Tables S8, S9). LDSC analyses suggested these traits displayed significant genetic variance ([lack of] self-control ζg = 0.072, s.e. = 0.019; urgency ζg = 0.093, s.e. = 0.022; online Supplementary Table S7), but examination of Q–Q plots (online Supplementary Figs S5, S6) and mean χ2 values (1.03–1.06) implied that sample sizes for these traits lack the power necessary to identify meaningful variant-level associations.
Q SNP analyses identified no significant heterogeneity in individual SNP effects for the sensation seeking factor or the (lack of) self-control factor, but 323 GWS Q SNPs were identified for the urgency factor. These were removed from urgency summary statistics prior to $\widehat{N}$
calculation and downstream analyses.
Genetic correlations with neuroimaging phenotypes
Key findings from neuroimaging genetic correlation analyses included the following significant associations (p FDR < 0.05): (1) sensation seeking exhibited small positive correlations with cortical surface area across several regions (0.07<|rg|<0.10); (2) (lack of) self-control exhibited moderate negative correlations with cortical thickness across the majority of tested regions (0.22<|rg|<0.43); and (3) urgency exhibited a negative correlation with cortical thickness in the rostral middle frontal gyrus (rg = −0.39, p FDR = 0.031). Dual-systems factors were not associated with regional cortical brain volumes nor subcortical structural volumes following FDR correction (online Supplementary Tables S10, S11).
Differences in genetic correlations with cortical surface area and thickness were generally robust across factors (Fig. 2; Supplementary Tables S12, S13). Broadly, genetic correlations between dual-systems factors and cortical surface area were in the positive direction while genetic correlations with cortical thickness were negative. However, (lack of) self-control and urgency were only nominally associated with cortical surface area in two and one regions, respectively (p = 0.012–0.025). Sensation seeking, in contrast, was associated with cortical surface area following FDR correction across more than 25% of regions tested (p FDR < 0.05) with equal representation in the frontal, parietal, and temporal lobes. Notably, χ2 tests constraining the magnitude of these correlations to equality across traits were generally non-significant (p diff > 0.05), suggesting that these weaker associations were less specific to sensation seeking. Conversely, (lack of) self-control was negatively correlated with cortical thickness across more than 60% of regions tested (p FDR < 0.05) with the greatest representation in the PFC (max-rg = −0.41, pars orbitalis) and parietal lobe (max-rg = −0.38, precuneus), where correlations were significantly larger than those between sensation seeking and cortical thickness (p diff < 0.05) which were not significant.
Figure 2. Genetic correlations between dual-systems factors and regional cortical brain volume, cortical surface area, and cortical thickness. Cortical patterning of genetic correlations plotted as z statistics (blue = positive correlation, red = negative correlation) across IPT dual-systems factors for cortical regional volume (top) according to the Desikan–Killiany–Tourville atlas (Klein & Tourville, Reference Klein and Tourville2012), and cortical regional surface area (middle) and thickness (bottom) according to the Desikan–Killiany atlas (Desikan et al., Reference Desikan, Ségonne, Fischl, Quinn, Dickerson, Blacker and Killiany2006). Plots were constructed using the ggseg package in R (Mowinckel & Vidal-Piñeiro, Reference Mowinckel and Vidal-Piñeiro2020).
Discussion
The current study represents the first investigation of the latent genetic structure of dual-systems models of IPTs and their neuroanatomical correlates using GWAS data. Genomic factor analyses supported the distinct but related hypothesized genetic components of the tested dual-systems models. The SSSC model was strongly supported with fit indices demonstrating that the putative bottom-up sensation seeking and top-down (lack of) self-control factors represented separable, though correlated (rg = 0.60), constructs. These results are consistent with prior twin studies, which reported similar support for these constructs and a similar genetic correlation between them (Ellingson et al., Reference Ellingson, Vergés, Littlefield, Martin and Slutske2013; Hur & Bouchard, Reference Hur and Bouchard1997). The UGSC model was also supported, showing satisfactory model fit with a modest, non-significant correlation between the putative bottom-up urgency and top-down (lack of) self-control factors (rg = 0.42). Though non-significant, this correlation was within the range of previous estimates for similar traits (e.g. rg = 0.26–0.64; Gustavson et al., Reference Gustavson, Franz, Kremen, Carver, Corley, Hewitt and Friedman2019, Reference Gustavson, Friedman, Fontanillas, Elson, Palmer and Sanchez-Roige2020).
To further evaluate the validity of the modeled dual-systems factors, hypotheses regarding the neural underpinnings of top-down and bottom-up constructs driven by dual-systems model theory and prior research (Shulman et al., Reference Shulman, Smith, Silva, Icenogle, Duell, Chein and Steinberg2016b) were tested by estimating genetic correlations between the dual-systems factors and relevant neuroimaging phenotypes. Consistent with prior research and theory, cortical thickness of PFC regions was negatively correlated in the present study with (lack of) self-control reflecting overlap in genetic variation associated with thinner PFC regions and diminished self-control (max-rg = −0.41). This finding mirrors results from previous neuroimaging studies suggesting negative associations between frontocortical thickness and lack of self-control IPTs (Holmes, Hollinshead, Roffman, Smoller, & Buckner, Reference Holmes, Hollinshead, Roffman, Smoller and Buckner2016; Kaag et al., Reference Kaag, Crunelle, van Wingen, Homberg, van den Brink and Reneman2014; Kubera et al., Reference Kubera, Schmitgen, Maier-Hein, Thomann, Hirjak and Wolf2018; Schilling et al., Reference Schilling, Kühn, Romanowski, Schubert, Kathmann and Gallinat2012) and complements prior research reporting a positive correlation between cortical thickness in the precentral gyrus and general cognitive functioning using these GWAS data (Grasby et al., Reference Grasby, Jahanshad, Painter, Colodro-Conde, Bralten and Hibar2020). Together, these lines of evidence imply that associations between reduced frontocortical thickness and diminished self-control may be partially explained by a common underlying genetic basis, thus lending further support to the interpretation of the modeled latent (lack of) self-control factor as a top-down construct.
However, negative correlations between the (lack of) self-control factor and cortical thickness extended to other cortical regions not hypothesized by the dual-systems model (e.g. occipital lobe), and other findings also ran contrary to the hypothesized neurobiology of the dual-systems model. For example, the observed positive genetic correlations between sensation seeking and cortical surface area across a number of regions were unexpected as dual-systems theory contends that sensation seeking, as a bottom-up construct, is primarily localized to subcortical reward structures (Steinberg et al., Reference Steinberg, Albert, Cauffman, Banich, Graham and Woolard2008). Similarly, previous research would partially situate urgency in the morphology of subcortical structures involved in emotion-processing (amygdala, basal ganglia; Chester et al., Reference Chester, Lynam, Milich, Powell, Andersen and DeWall2016; Cyders et al., Reference Cyders, Dzemidzic, Eiler, Coskunpinar, Karyadi and Kareken2015; Halcomb, Argyriou, & Cyders, Reference Halcomb, Argyriou and Cyders2019). Contrarily, urgency was significantly associated with a single neuroimaging phenotype following FDR correction: rostral middle frontal cortical thickness (rg = −0.39), though this replicates prior phenotypic studies of neuroimaging correlates of urgency (Cyders et al., Reference Cyders, Dzemidzic, Eiler, Coskunpinar, Karyadi and Kareken2014a; Cyders et al., Reference Cyders, Dzemidzic, Eiler, Coskunpinar, Karyadi and Kareken2015; Muhlert & Lawrence, Reference Muhlert and Lawrence2015). In the current study, all dual-systems constructs were uncorrelated with subcortical structural volume.
The reported results have two primary implications. First, they provide support at the genetic level for our modeled constructs ([lack of] self-control, urgency, and sensation seeking) as an organizational framework for understanding IPTs as putative endophenotypes for psychopathology but suggest that some further refinement of the hypothesized neurobiological underpinnings of these constructs as suggested by dual-systems models may be needed. As described, the top-down self-control factor was partially supported, demonstrating significant genetic correlations with its hypothesized neural correlates, but also with reduced thickness more broadly across the cortex. In contrast, neurogenetic evidence supporting sensation seeking and urgency as bottom-up constructs was more limited. While findings generally support the latent genetic structure of each, there was a lack of evidence relating these traits genetically to their hypothesized subcortical neural correlates, though previously described relations to cortical regions (rostral middle frontal gyrus) were replicated.
Criticisms of the dual-systems model as overly simplified have noted that, while dual-systems constructs are theoretically and empirically separable (Duckworth & Steinberg, Reference Duckworth and Steinberg2015; Shulman et al., Reference Shulman, Harden, Chein and Steinberg2016a), the underlying neurobiology is likely dynamic and multifaceted (Casey, Galván, & Somerville, Reference Casey, Galván and Somerville2016; Pfeifer & Allen, Reference Pfeifer and Allen2012). Moreover, given that these constructs tend to be highly correlated (Steinberg et al., Reference Steinberg, Albert, Cauffman, Banich, Graham and Woolard2008) and were strongly genetically correlated in the case of lack of self-control and sensation seeking in the current study, investigations of genetic correlations between one of these constructs and any neuroimaging analogue in isolation will be contaminated by the contribution of the unmeasured construct (Shulman et al., Reference Shulman, Smith, Silva, Icenogle, Duell, Chein and Steinberg2016b). As such, the present study provides further evidence supporting this criticism consistent with structural neuroimaging studies demonstrating that relations between bottom-up constructs and subcortical structural volume have been mixed and of generally small effect (Holmes et al., Reference Holmes, Hollinshead, Roffman, Smoller and Buckner2016; Owens et al., Reference Owens, Hyatt, Gray, Miller, Lynam, Hahn and Garavan2020; Reference Owens, Hyatt, Xu, Thompson, Miller, Lynam and Gray2023). Notably, functional neuroimaging studies of bottom-up constructs have reported unique patterns of connectivity across frontostriatal pathways (Burnette et al., Reference Burnette, Grodin, Lim, MacKillop, Karno and Ray2019; Demidenko, Huntley, Weigard, Keating, & Beltz, Reference Demidenko, Huntley, Weigard, Keating and Beltz2022; Hawes et al., Reference Hawes, Chahal, Hallquist, Paulsen, Geier and Luna2017; Um, Hummer, & Cyders, Reference Um, Hummer and Cyders2020; Zhu, Cortes, Mathur, Tomasi, & Momenan, Reference Zhu, Cortes, Mathur, Tomasi and Momenan2017), suggesting that assessment of coordinated cortical-subcortical activity may help with further refinement of dual-systems models and their measurement. In aggregate, study findings suggest that further refinement and validation of dual-systems constructs from neuroimaging and genetic perspectives is needed, though the consistency of some findings reported here with prior research highlight the promise of this approach.
Second, the current study provides an important demonstration of how post-GWAS approaches can complement studies using other methodologies to refine our models of psychopathology and endophenotypic measures based on these models. As noted, the present study is the first to investigate the latent genetic architecture of dual-systems models using GWAS data, and the first of any type to investigate the genetic architecture of an urgency and (lack of) self-control dual-systems model. While prior studies have examined genetic correlations amongst IPTs using GWAS approaches (Sanchez-Roige et al., Reference Sanchez-Roige, Fontanillas, Elson, Gray, De Wit, MacKillop and Palmer2019, Reference Sanchez-Roige, Jennings, Thorpe, Mallari, van der Werf, Bianchi and Palmer2023), focal investigations of the latent genetic architecture underlying dual-systems models are limited to a small number of twin studies (Ellingson et al., Reference Ellingson, Vergés, Littlefield, Martin and Slutske2013, Reference Ellingson, Slutske, Vergés, Littlefield, Statham and Martin2018; Harden et al., Reference Harden, Kretsch, Mann, Herzhoff, Tackett, Steinberg and Tucker-Drob2017). As a result, findings from the present study demonstrate how post-GWAS approaches can be used to critically evaluate theoretical models of psychological traits across multiple levels of analysis from the genetic level to that of a manifest disorder, representing an important and novel extension of this prior work. Despite deviations from dual-systems theory regarding neurobiological bases of these constructs, findings here emphasize unique neurogenetic components of putative top-down and bottom-up constructs which may have distinct etiological influences on psychopathology development. Therefore, the present study may serve as a benchmark for future studies assessing evidence for neurogenetic influences underlying dual-systems models and shared associations with related psychopathology.
Limitations
The current study is not without limitations. First, smaller sample sizes likely hindered analyses of the (lack of) self-control and urgency traits, demonstrating the need for larger IPT GWAS (Sanchez-Roige et al., Reference Sanchez-Roige, Jennings, Thorpe, Mallari, van der Werf, Bianchi and Palmer2023). Second, and equally important, is the exclusion of non-European ancestry samples from the present study. Our European ancestry-specific findings may only generalize to European ancestry populations and thus contribute to the disparity in applicability of research findings to non-European populations (Martin et al., Reference Martin, Kanai, Kamatani, Okada, Neale and Daly2019). As non-European ancestry groups are extremely underrepresented in extant GWAS studies (Mills & Rahal, Reference Mills and Rahal2019, Reference Mills and Rahal2020), addressing this limitation is of dire importance for leveling health disparities across groups. Third, IPT GWAS were available only at scale-level, rather than item-level, resulting in a small number of appropriate indicators for each genomic factor. Relatedly, all GWAS included sex and age as covariates obviating examination of sex- or age-specific associations relevant to both dual-systems models and brain development (Casey, Getz, & Galvan, Reference Casey, Getz and Galvan2008; Shulman, Harden, Chein, & Steinberg, Reference Shulman, Harden, Chein and Steinberg2015). Future investigations using item-level data (Mallard et al., Reference Mallard, Savage, Johnson, Huang, Edwards, Hottenga and Sanchez-Roige2022b) in combination with sex-specific (Silveira, Pokhvisneva, Howard, & Meaney, Reference Silveira, Pokhvisneva, Howard and Meaney2023) and developmentally relevant (Couto Alves et al., Reference Couto Alves, De Silva, Karhunen, Sovio, Das and Taal2019) analytic approaches will continue to improve our understanding of these complex pathways.
Conclusion
Results of the current study suggest dual-systems models of the genetic architecture of IPTs are generally well-validated through genomic structural equation modeling, though factors derived from this model were not consistently associated with theoretically relevant neuroimaging phenotypes. As such, this study serves as an important first step in defining the shared and unique genomic and neurobiological correlates of dual-systems constructs and underscores the importance of using imaging genetics to further elucidate neurobiological substrates underlying genetic overlap between complex traits (Bogdan et al., Reference Bogdan, Salmeron, Carey, Agrawal, Calhoun, Garavan and Goldman2017).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291723003367
Data availability
The full GWAS summary statistics for the 23andMe discovery data set will be made available through 23andMe to qualified researchers under an agreement with 23andMe that protects the privacy of the 23andMe participants. Please visit https://research.23andme.com/collaborate/#dataset-access/ for more information and to apply to access the data.
Acknowledgments
This study made use of summary statistics data from a number of sources which we wish to acknowledge. First, we thank Susan Service, MSc, and Nelson Freimer, MD (University of California, Los Angeles), for providing TCI summary statistics from Service et al. (Reference Service, Verweij, Lahti, Congdon, Ekelund, Hintsanen and Freimer2012) for which compensation was not received. Second, this study made use of GWAS summary statistics data from 23andMe, Inc. (Sunnyvale, CA). We thank the 23andMe research participants and employees for making this work possible. Second, this research used summary data from UKB, a population-based sample of participants whose contributions we gratefully acknowledge. Finally, this study also made use of data generated by the UK10K Consortium, derived from samples from UK10K_COHORT_IMPUTATION REL-2012-06-02 (EGAD00001000776). A full list of the investigators who contributed to the generation of the data is available from www.UK10K.org. Funding for UK10K was provided by the Wellcome Trust under award WT091310. All secondary data analysis of GWAS summary statistics and reference panels were considered exempt by the Institutional Review Board at the University of Missouri. The computations for all analyses were performed on the high-performance computing infrastructure provided by Research Computing Support Services and in part by the National Science Foundation under grant number CNS-1429294 at the University of Missouri, Columbia, MO. DOI: https://doi.org/10.32469/10355/69802
Funding statement
Investigator effort was supported by the National Institutes of Health (APM, F31AA027957, T32DA015035).
Competing interests
None.
Open access
