Descriptive statistics and correlations
presents the number of individual twin participants, as well as the means and standard deviations for the raw (untransformed) variables. Complete information from both twins in a pair were available for RSA: 119 MZ male, 76 DZ male, 119 MZ female, 86 DZ female and 127 opposite-sex (DZOS) pairs, HR1: 121 MZ male, 76 DZ male, 119 MZ female, 86 DZ female and 127 DZOS pairs, HR2: 95 MZ male, 59 DZ male, 97 MZ female, 65 DZ female and 105 DZOS pairs, SCL: 125 MZ male, 78 DZ male, 130 MZ female, 91 DZ female and 137 DZOS pairs, NS-SCR: 115 MZ male, 70 DZ male, 126 MZ female, 83 DZ female and 132 DZOS pairs. The vast majority (>94%) of twins used in the genetic analyses were from complete twin pairs.
Means, standard deviations and number of participants (n) for respiratory sinus arrhythmia (RSA), heart rate (HR1 at rest 1, HR2 at rest 2), skin conductance level (SCL) and non-specific skin conductance response (NS-SCR), by sex and zygosity
No significant mean or variance differences were found between twin-1 and twin-2 (results available upon request). Moreover, there were no mean or variance differences between MZ and DZ twins of the same sex. For heart rate, mean sex differences were found, with males showing lower mean heart rate than females at both Rest 1 and Rest 2 [HR1: Mmale = 84.69 (SD: 10.74), Mfemale = 87.62 (SD: 10.27), χ2 = 23.78; df = 9; p<.001; HR2: Mmale = 82.49 (SD: 10.61), Mfemale = 86.06 (SD: 10.04), χ2 = 30.89; df = 9; p<.001].
shows phenotypic correlations among the measures. Heart rate was strongly correlated across the two rest periods for both boys and girls. Similarly, a significant, albeit somewhat more modest, association between SCL and NS-SCR was found for both sexes. There was a substantial negative association between RSA and heart rate, across Rest 1 and Rest 2 for both sexes, although the relationships between heart rate with SCL and NS-SCR were negligible or modest at best, and were significant only among girls.
Phenotypic correlations for respiratory sinus arrhythmia (RSA), heart rate (HR1 at rest 1, HR2 at rest 2), skin conductance level (SCL) and non-specific skin conductance response (NS-SCR), by sex
Twin and cross-twin cross-trait correlations are shown in . The consistently higher MZ as compared to DZ twin correlations suggest genetic influences on all five measures. For example, the twin correlations for RSA were r = .42 for MZ males and r = .31 for MZ females, and the corresponding values for DZ twins were lower: r = .27 and r = .06, respectively. The majority of cross-twin cross-trait correlations were consistently stronger for MZ twins compared to DZ twins, suggesting that genetic effects contribute to the covariation among RSA, HR1 and HR2, and among SCL, NS-SCR, HR1 and HR2. Also of note was the fact that the cross-twin, cross-trait correlations between HR1 and HR2 were nearly as strong as the cross-twin, within-trait correlations for HR1 or HR2, for both MZ and DZ twins, indicating the likely influence of shared measurement error across the two heart rate measures. Etiological patterns suggested by these twin correlations were next tested more formally using structural equation models.
Twin and cross-twin cross-trait correlations between for respiratory sinus arrhythmia (RSA), heart rate (HR1 at rest 1, HR2 at rest 2), skin conductance level (SCL) and non-specific skin conductance response (NS-SCR)
Univariate model fitting results
displays univariate model fitting results for each measure. Variance components were estimated from a full univariate ACE model, which compared favorably to the saturated model in all cases (specific results of model fitting are available from the first author). Genetic and environmental variances could be equated across sex for RSA (Δχ2 = 3.39, df = 3, p = 0.34), HR1 (Δχ2 = 3.66, df = 3, p = 0.30), HR2 (Δχ2 = 2.14, df = 3, p = 0.54), and NS-SCR (Δχ2 = 5.15, df = 3, p = 0.16). When sexes were combined, genetic influences accounted for between 30 and 48% of the variance in each of these measures, shared environmental influences were modest and non-significant, and the non-shared environment accounted for between 41 and 61% of the variance.
Univariate results for respiratory sinus arrhythmia (RSA), heart rate (HR1 at rest 1, HR2 at rest 2) skin conductance level (SCL) and non-specific skin conductance response (NS-SCR)
In contrast, estimates could not be constrained to be equal in males and females for SCL (Δχ2 = 14.35, df = 3, p<0.01). For males, genetic and shared environmental influences each accounted for a quarter of the variance, and the remaining variance was due to non-shared environmental effects. For females, a third of the variance was due to genetic influences, another third was due to shared environmental influences, and the remaining 35% of the variance was due to non-shared environmental influences.
Multivariate model fitting results
To investigate further the nature of the relationships among RSA, HR1 (Rest 1), HR2 (Rest 2), NS-SCR and SCL, we fit a series of multivariate models (Models #2–4 in ). Specifically, a fully saturated model (Model #1 in ) was used as a baseline to which a Cholesky decomposition (Model #2), a two-factor common pathway (Model #3), and a one-factor common pathway (Model #4) were compared. The two-factor common pathway model provided the best fit to the data based on BIC and AIC criteria, and did not significantly differ from the unconstrained fully saturated model (Δχ2 = 294.67; Δdf = 257; p = 0.053). In contrast, the one-factor model fit the data significantly worse than the fully saturated comparison model (Δχ2 = 738.30; Δdf = 271; p<0.001), and also resulted in more positive AIC and BIC values, indicating that one factor was insufficient to account for the genetic and environmental covariance across measures.
Multivariate model fit indices for respiratory sinus arrhythmia (RSA), heart rate (HR1 at rest 1, HR2 at rest 2) skin conductance level (SCL) and non-specific skin conductance response (NS-SCR)
Given the similarities across sexes in the univariate results, we also explored the extent to which the factor structure and underlying genetic and environmental etiology varied across sexes, using the full two-factor common pathway model (Model #3) as our comparison model. However, we could not equate the factor loadings (i.e., the paths stemming from the two latent ANS factors to the manifest variables—see ), common genetic and environmental influences (e.g., the A1, C1, E1 and A2, C2, E2 that explain variance in the two ANS factors), and measurespecific genetic and environmental effects (As, Cs, and Es in ) across sex without resulting in a significant deterioration in fit compared to the full two-factor common pathway model (: Model 3a, Δχ2 = 53.89; df = 29; p = 0.003). This suggests that there are sex differences in the factor structure and underlying genetic and environmental etiology of the five observed variables.
Fig. 1 Males and (Females). Standardized path estimates from the full two-factor common pathway model for RSA (respiratory sinus arrhythmia), HR1, HR2 (heart rate), SCL (skin conductance level) and NS-SCR (non-specific skin conductance) in 9–10 year (more ...)
Based on the magnitude of the factor loadings shown in , for both males and females, the first ANS factor appeared to reflect sympathetic activity, as it was indexed primarily by heart rate, SCL and NS-SCR. Furthermore, the second ANS factor appeared to reflect parasympathetic processes, indexed by strong (and inversely related) factor loadings for RSA and the two heart rate variables. However, loadings for SCL and NS-SCR were also significant for this factor, albeit only in males. Dropping RSA from the first ANS factor as well as SCL and NS-SCR from the second ANS factor for both boys and girls resulted in a significant deterioration in fit compared to the full two-factor common pathway model (: Model 3b, Δχ2 = 17.18; df = 6; p = 0.009). We consequently proceeded step-by-step, and first tested whether the factor structure could be simplified by dropping RSA from the first ANS factor in both males and females. This could be done without a significant reduction in fit (: Model 3c, Δχ2 = 2.40; df = 2; p = .302). We next tested whether SCL and NS-SCR could be dropped from the second ANS factor (as would be theoretically predicted) only in males. This resulted in a significant deterioration in fit compared to the full two-factor common pathway model (: Model 3d, Δχ2 = 17.03; df = 4; p<0.001). However, SCL and NS-SCR could be dropped from the second ANS factor in females without a reduction in fit (: Model 3e, Δχ2 = 3.41; df = 4; p = 0.493). Thus, the pattern in both sexes indicated that Factor 1 was defined by heart rate, SCL and NS-SCR, reflecting sympathetic activity. Factor 2 was mainly defined by the inverse relationship between heart rate and RSA in females, indicating parasympathetic processes, but the factor structure for Factor 2 could not be further simplified in males. Even though SCL and NS-SCR could not be dropped from the parasympathetic factor in males without a significant reduction in model fit, the factor loadings for these variables were rather small (.17 and .24). This indicates that the structure of the parasympathetic factor was largely similar in males and females.
Next, we investigated whether variation in the two latent factors was significantly influenced by both genetic and shared environmental factors by dropping common genetic influences (Model #3f) and common shared environmental influences (Model #3g) from the simplified two-factor common pathway model (Model #3e). While neither model was statistically significant, submodels dropping both common A and common C from Factor 1 (Model #3h) or Factor 2 (Model #3i) did result in a significant decrease in fit. Because the AIC and BIC values were similar for Models #3f and #3g, it was impossible to determine whether genetic or shared environmental influences were more important for variation in the underlying latent factors. Therefore, both common genetic and common shared environmental estimates were kept in the model. Finally, we ran a series of submodels investigating the significance of the genetic and shared environmental factors that were specific to each of the fives measures (results from all submodels available from the first author). The greatest simplification of the model was ultimately achieved by dropping both specific genetic and specific shared environmental influences on HR1, HR2 and NS-SCR, and specific shared environmental influences on RSA (: Model 3j, Δχ2 = 8.50; df = 14; p = .862). and display standardized parameter estimates from this reduced two-factor common pathway model for males and females, respectively.
Fig. 2 Final model estimates (males). Standardized path estimates with confidence intervals from the reduced two-factor common pathway model for RSA (respiratory sinus arrhythmia), HR1, HR2 (heart rate), SCL (skin conductance level) and NS-SCR (non-specific (more ...)
Final model estimates (females). See figure caption
Squaring the standardized parameter estimates presented in (males) and 3 (females) provides the relative contributions to the phenotypic variance. In addition, squaring the loadings for the common A, C, and E influences on the latent factors gives the overall heritability of the underlying latent sympathetic and parasympathetic factors. For the first latent ANS factor (labelled sympathetic activity) in males, 27% of the variance was due to genetic factors, 28% (p<.05) due to shared environmental factors, and 45% (p<.05) due to non-shared environmental factors. Variance in the second latent ANS factor (labelled parasympathetic activity) was decomposed into respective values of 27, 23, and 50% (p<.05). For females, 31% of the variance in sympathetic activity was due to genetic factors, 41% (p<.05) due to shared environmental factors, and 28% (p<.05) due to non-shared environmental factors. The respective proportions of variance for parasympathetic activity were 35, 18, and 47% (p<.05). For both boys and girls, there were significant variable-specific genetic influences on RSA. In addition, there were variable-specific genetic and shared environmental influences on SCL. While neither estimate was statistically significant, our detailed series of analyses investigating the significance of the variable-specific factors indicated that the specific genetic and specific shared environmental influences on SCL could not be dropped simultaneously from the model without a significant reduction in fit (Δχ2 = 19.82; df = 4; p<.001); thus, both estimates were left in the model.
shows the variance components for each of the five measures, based on the factor loadings presented in and . The variance components are divided into influences due to genetic (A), shared environmental (C), and non-shared environmental factors (E), and are further differentiated by common variance due to the sympathetic factor (Factor 1), common variance due to the parasympathetic factor (Factor 2), and variable-specific variance. Comparing variance due to Factor 1 versus Factor 2, it can be seen that a larger proportion of variance in RSA, HR1, and HR2 was due to the parasympathetic factor, whereas a greater proportion of the variance in SCL and NS-SCR was due to variation in the sympathetic factor. Despite the fact that we could not drop SCL and NS-SCR from the parasympathetic factor in males, this factor accounted for only a minority of the overall variance in these two variables (i.e., 4–6% of the total phenotypic variance). While genetic variance from the two latent factors accounted for all of the heritabilities of HR1, HR2, and NS-SCR, a large proportion of the heritabilities of RSA and SCL were due to genetic factors that were specific to each variable. Finally, overall, the estimates shown in are consistent with the results from the univariate genetic analyses (). A possible exception is that the non-shared environment accounts for a greater proportion of variance in HR2 in the multivariate analyses (.61 for both males and females) compared to estimates obtained in the univariate analyses (.41 for males and .40 for females). Over half of this nonshared environmental variance comes from non-shared environmental influences on the parasympathetic factor. We noted earlier that the cross-twin, cross-trait correlations for HR1 and HR2 were similar in magnitude to the cross-twin, within-trait correlations for either HR1 or HR2, for both MZ and DZ twins (). Thus, the higher estimate of non-shared environmental influence on HR2 is likely due to correlated errors of measurement across HR1 and HR2.
Variance Components from Best-Fitting Reduced 2-Factor Model for respiratory sinus arrhythmia (RSA), heart rate (HR1 at rest 1, HR2 at rest 2) skin conductance level (SCL) and non-specific skin conductance response (NS-SCR)