Social support and mental health in late adolescence are correlated for genetic, as well as environmental, reasons

Late adolescence is a crucial, but underexplored, developmental stage with respect to the aetiology of social support. These individuals are experiencing many major life changes and social support can help them adjust to the associated environmental stressors of this time. Using 1,215 18-year-old twin pairs from the Twins Early Development Study, we collected measures of two indices of support: support quality and support quantity, as well as wellbeing and depression. Both support indices were moderately heritable (55% and 49%, respectively), an interesting finding given the many environmental changes that late adolescents are encountering that could be environmentally altering their social network structures. Finding a genetic influence on support suggests the presence of gene-environment correlation whereby individuals create and perceive their supportive environment based upon their genetic predispositions. Shared genetic influences mediated the moderate phenotypic correlation (mean r = 0.46) between support and mental health. Genetic correlations were higher between support quality and mental health (mean rA = 0.75), than between support quantity and mental health (mean rA = 0.54), reflecting the phenotypic pattern. This suggests that interventions should focus more on making late adolescents aware of the support quality around them than encouraging them to increase their social network size.


Supplementary Figures and
. Twin Correlations across zygosity groups split by sex Table showing within-trait cross-twin correlations obtained from the saturated univariate model through full information maximum likelihood across zygosity groups split by sex. These twin correlations were used to gauge the existence of qualitative and quantitative aetiological sex differences, both of which are suggested by our results

Supplementary Table S2. Means (Standard Deviation, SD) and ANOVA Results
Table showing means and standard deviations of the measures, unstandardized. Estimates were obtained from a subset consisting of one randomly selected twin from each twin pair, and were calculated for each measure across the whole sample, for monozygotic participants only, dizygotic participants only, males only and females only. Differences in measure scores between zygosity groups and between sex were investigated using ANOVA with eta-squared (proportion of variance attributed to an effect) used to look at the size of the effect of zygosity/sex on the measures.

Supplementary Table S3. Means and SD of individual traits across zygosity groups split by sex
Table showing means and standard deviations of the measures, unstandardized. Estimates were obtained from a subset consisting of one randomly selected twin from each twin pair, and were calculated for each measure within each zygosity pair split by sex.

Supplementary Table S4. Phenotypic correlations split across sex
Phenotypic correlations between social support measures and mental health measures for single sex male twins and single sex female twins to examine potential sex differences.

Supplementary Table 5. Fully saturated and constrained models to test twin assumptions
Table presenting the fit indices of each univariate saturated model and constrained submodel, obtained by structural equation model fitting, along with fit statistics comparing the fit of the constrained submodels (Sub1 and Sub2) to the fully saturated model (Sat). By testing for equality of variance across twin order and zygosity using our constrained submodels, we show that all our keep to the assumptions of equal variance. Table S6. Univariate Twin Analysis Results: Variance Components due to Additive Genetic (a 2 ), Non-Additive Genetic (d 2 ), Shared Environmental (c 2 ) and Non-Shared Environmental (e 2 ) effects.

Supplementary
Table presenting variance components due to additive genetic (a 2 ), non-additive genetic (d 2 ), shared environmental (c 2 ) and non-shared environmental (e 2 ) effects. Results were obtained from full univariate ACE/ADE models. ADE models are used when the DZ correlation is less than half of the MZ correlation. When DZ correlations are more than half of the MZ correlation, ACE models are used. Heritability estimates from these full models were similar to those from the reduced AE model (Table 2).

Supplementary Table S7. Univariate Twin Analysis Model Fit Statistics
Table presenting the fit statistics assessing the fit of each univariate model and submodel, obtained by structural equation model fitting. The best-fitting univariate model for each individual trait was selected based on AIC values, the fit of the model relative to the full/saturated model (as indicated by the likelihood ratio comparison) and parsimony.

Supplementary Table S8. Genetic and environmental correlations (95% confidence intervals)
Table showing estimates of genetic and environmental correlation between support and mental health measures. Estimates were obtained through bivariate AE model fitting. Most of the genetic correlations were moderate to high while the non-shared environmental correlations tended to be much lower.

Supplementary Table S9. Proportion of the phenotypic correlation explained by genetic and non-shared environmental overlaps (95% confidence intervals)
Table showing the proportion of the phenotypic correlation between our measures that is accounted for by common genetic or environmental influences. Estimates were obtained through bivariate AE model fitting. Genetic influences explained a larger proportion of the phenotypic correlations between all of the measures of mental health and support

Supplementary Table S10. Skew of measures
Table showing skew of measures. We transformed any measure which had a skew greater than 1 or less than -1.

Supplementary Figure S11: Comparing univariate results (with 95% confidence intervals) of transformed versus untransformed scales
For measures with an absolute skew value of over 1, we repeated our analyses on a reflected then logtransformed version of those scales with negative skew and directly log-transformed version of those scales with positive skew. This graph shows the a 2 and e 2 components obtained from our univariate analysis for the untransformed and transformed versions of the scales. Estimates were very similar, all with overlapping confidence intervals.  Figures S13-16 show only genetic correlations between support quality and mental health. Each bar represents the genetic correlation between a mental health measure and support quality (total score/significant other subscale/family subscale/friend subscale). The support quality total score and subscales all had a negative skew less than -1 meaning that we reflected and then log-transformed the scale/subscales. We then obtained the genetic correlation between the untransformed and transformed versions of these (sub)scales with untransformed and (where absolute skew was greater than 1) transformed versions of individual mental health measures. Estimates were very similar for untransformed and transformed measures, with overlapping confidence intervals.

Supplementary Figure S16. Genetic correlations between transformed and untransformed measures of mental health and support quantity with 95% confidence intervals
Figure showing only genetic correlations between support quantity and mental health. Support quantity had a skew of between 0 and -1 so was not transformed. This untransformed score was correlated with untransformed and transformed versions of mental health measures which had an absolute skew value greater than 1. For those mental health measures with an absolute skew value of less than 1, results are already presented in Figure 1. Estimates were very similar for untransformed and transformed measures, with overlapping confidence intervals.  Figuring showing the bivariate Cholesky decomposition which allows us to decompose the covariance between two traits, indicating the degree of genetic and environmental overlap between our measures.

Supplementary Figure S25: Illustration of the Correlated Factors Solution
Figure showing the correlated factors solution, which is mathematically equivalent to the bivariate Cholesky decomposition. It does not impose an order on the included variables and allows us to estimate genetic and environmental correlations between our measures. We can also calculate the proportion of the phenotypic correlation between our measures of mental health and support that is accounted for by overlapping genetic or environmental influences. Observing the patterns of twin correlations between MZ male and DZ males compared to MZ females and DZ females, we find evidence of potential quantitative sex differences, where the same genetic and environmental factors influence both males and females but to varying extents. DZ opposite sex twin correlations are generally lower than DZ same sex twin correlations. Therefore, it seems like there may be qualitative sex differences, where there are different genetic and environmental influences on a trait for females and males. However, many of the confidence intervals overlap, and equally, we cannot make any inferences from our point estimates due to our small sample sizes. The small sample sizes within each zygosity group in this study means that we are underpowered to run a sex-limitation model.

Supplementary Table S2. Means (Standard Deviation, SD) and ANOVA Results
Note. Table showing means and standard deviations of the measures, unstandardized. Estimates were obtained from a subset consisting of one randomly selected twin from each twin pair, and were calculated for each measure across the whole sample, for monozygotic participants only, dizygotic participants only, males only and females only. Differences in measure scores between zygosity groups and between sex were investigated using ANOVA with eta-squared (proportion of variance attributed to an effect) used to look at the size of the effect of zygosity/sex on the measures. *p<0.05, **p<0.01, ***p<0.001. MZ= monozygotic twins; DZ = dizygostic twins (same and opposite sex) Note.  13 across sex. In our saturated model, we account for the effects of covariates (age and sex) from our traits using a Means Model. This model specifies one Means Model across twins and zygosity group. The saturated model includes a covariance within twin pairs for each zygosity (between MZ twin 1 and MZ twin 2 and between DZ twin 1 and DZ twin 2), and variances for each person-category (MZ twin 1, MZ twin 2, DZ twin 1, DZ twin 2). The constrained submodels allows us to check the assumptions of the twin method. We can test equality of variance across twin order within zygosity (sub1) and equality of variance across twin order and across zygosity (sub2) −2LL = negative 2 log likelihood; df = degrees of freedom; AIC = Akaike's Information Criterion; ∆χ2 = difference in chi square; ∆df = difference in degrees of freedom; p = pvalue. Bonferroni corrected alpha level, correcting for multiple testing (30 tests) = 0.001667. -2LL provides a relative measure of model fit as differences in -2LL between models are distributed as χ2. Therefore the significance of ∆χ2 provides an indication of the constrained sub model being a significantly worse fit to the data than the fully saturated model. Lower AIC values reflect better model fit.
Equating variances across twin pair and zygosity did not significantly worsen fit for any of our measures, using a Bonferroni corrected p threshold (lowered for multiple testing). Note. Table presenting variance components due to additive genetic (a 2 ), non-additive genetic (d 2 ), shared environmental (c 2 ) and non-shared environmental (e 2 ) effects.

Supplementary
Results were obtained from full univariate ACE/ADE models. ADE models are used when the DZ correlation is less than half of the MZ correlation. When DZ correlations are more than half of the MZ correlation, ACE models are used. Heritability estimates from these full models were similar to those from the reduced AE model (Table 2).     Note. Table showing skew of measures. We transformed any measure which had a skew greater than 1 or less than -1.

Supplementary Figure S12. Genetic correlations between transformed and untransformed measures of mental health and support quality (total) with 95% confidence intervals
Note. Scales were transformed only when they had an absolute skew value greater than 1. The support quality total score had a negative skew less than -1 meaning that we reflected and then log-transformed this scale. We then obtained the genetic correlation between the untransformed and transformed versions of this scale with untransformed and (where absolute skew was greater than 1) transformed versions of individual mental health measures. For comparative purposes, genetic correlations with negative affect and depression shown here are absolute, as both measures are negatively correlated with support. Pa(i) = Positive Affect (untransformed) and Support Quality total (untransformed), PA(ii) = Positive Affect (untransformed) and Support Quality total (transformed), NA(i) = Negative Affect (untransformed) and Support Quality total (untransformed), Na(ii) = Negative Affect (transformed) and Support Quality total (transformed), SHS(i) = Subjective Happiness (untransformed) and Support Quality total (untransformed), SHS(ii) = Subjective Happiness (untransformed) and Support Quality total (transformed), LS(i) = Life Satisfaction (untransformed) and Support Quality total (untransformed), LS(ii) = Life Satisfaction (transformed) and Support Quality total (transformed), Grat(i) = Supplementary Figure S13. Genetic correlations between transformed and untransformed measures of mental health and support quality (significant other subscale) with 95% confidence intervals Note. Scales were transformed only when they had an absolute skew value greater than 1. The support quality significant other subscale had a negative skew less than -1 meaning that we reflected and then log-transformed this scale. We then obtained the genetic correlation between the untransformed and transformed versions of this subscale with untransformed and (where absolute skew was greater than 1) transformed versions of individual mental health measures. For comparative purposes, genetic correlations with negative affect and depression shown here are absolute, as both measures are negatively correlated with support. Pa(i) = Positive Affect (untransformed) and Support Quality significant other (untransformed), PA(ii) = Positive Affect (untransformed) and Support Quality significant other (transformed), NA(i) = Negative Affect (untransformed) and Support Quality significant other (untransformed), Na(ii) = Negative Affect (transformed) and Support Quality significant other (transformed), SHS(i) = Subjective Happiness (untransformed) and Support Quality significant other (untransformed), SHS(ii) = Subjective Happiness (untransformed) and Support Quality significant other (transformed), LS(i) = Life Satisfaction (untransformed) and Support Quality significant other (untransformed), LS(ii) = 29 Life Satisfaction (transformed) and Support Quality significant other (transformed), Grat(i) = Gratitude (untransformed) and Support Quality significant other (untransformed), Grat(ii) = Gratitude (transformed) and Support Quality significant other (transformed), MiL(i) = Meaning in Life (untransformed) and Support Quality significant other (untransformed), MiL(ii) = Meaning in Life (untransformed) and Support Quality significant other (transformed), Auto(i) = Autonomy (untransformed) and Support Quality significant other (untransformed), Auto(ii) = Autonomy (untransformed) and Support Quality significant other (transformed), Comp(i) = Competence (untransformed) and Support Quality significant other (untransformed), Comp(ii) = Competence (untransformed) and Support Quality significant other (transformed), Rel(i) = Relatedness (untransformed) and Support Quality significant other (untransformed), Rel(ii) = Relatedness (untransformed) and Support Quality significant other (transformed), Dep(i) = Depression (untransformed) and Support Quality significant other (untransformed), Dep(ii) = Depression (transformed) and Support Quality significant other (transformed) Supplementary Figure S14. Genetic correlations between transformed and untransformed measures of mental health and support quality (family subscale) with 95% confidence intervals Note. Scales were transformed only when they had an absolute skew value greater than 1. The support quality family subscale had a negative skew less than -1 meaning that we reflected and then log-transformed this scale. We then obtained the genetic correlation between the untransformed and transformed versions of this subscale with untransformed and (where absolute skew was greater than 1) transformed versions of individual mental health measures. For comparative purposes, genetic correlations with negative affect and depression shown here are absolute, as both measures are negatively correlated with support. Pa(i) = Positive Affect (untransformed) and Support Quality family (untransformed), PA(ii) = Positive Affect (untransformed) and Support Quality family (transformed), NA(i) = Negative Affect (untransformed) and Support Quality family (untransformed), Na(ii) = Negative Affect (transformed) and Support Quality family (transformed), SHS(i) = Subjective Happiness (untransformed) and Support Quality family (untransformed), SHS(ii) = Subjective Happiness (untransformed) and Support Quality family (transformed), LS(i) = Life Satisfaction (untransformed) and Support Quality family (untransformed), LS(ii) = Life Satisfaction (transformed) and Support Quality family Supplementary Figure S16. Genetic correlations between transformed and untransformed measures of mental health and support quantity with 95% confidence intervals Note. Support quantity had a skew of between 0 and -1 so was not transformed. We obtained the genetic correlation between this untransformed scale with untransformed and (where absolute skew was greater than 1) transformed versions of individual mental health measures. For comparative purposes, genetic correlations with negative affect and depression shown here are absolute, as both measures are negatively correlated with support. NA(i) = Negative Affect (untransformed) and Support Quantity (untransformed), Na(ii) = Negative Affect (transformed) and Support Quantity (transformed), LS(i) = Life Satisfaction (untransformed) and Support Quantity (untransformed), LS(ii) = Life Satisfaction (transformed) and Support Quantity (transformed), Grat(i) = Gratitude (untransformed) and Support Quantity (untransformed), Grat(ii) = Gratitude (transformed) and Support Quantity (transformed), Dep(i) = Depression (untransformed) and Support Quantity (untransformed), Dep(ii) = Depression (transformed) and Support Quantity ( Note. This figure shows the genetic and environmental relationships within twins for one trait and forms the basis of all twin analyses. The total observed phenotypic variance of a trait, shown by the rectangular boxes, is the sum of additive genetic (A), shared environmental (C), and non-shared environmental (E) influences, shown as latent factors in the circles. The path coefficients of these latent variables are represented by a, c and e respectively. We know that MZ twins are 100% genetically identical while DZ twins are on average 50% genetically identical. We also assume that both reared together MZ and DZ twins share 100% of their shared environment. Using this information of genetic and environmental similarity, we can predict concordances of a trait between MZ and DZ twin pairs and compare this to actual observed concordances to enable us to decompose the phenotypic variance of a trait using simultaneous equations (Falconer's equations). When non-additive genetic (D) are used instead of C, rMZ = 1.0 and rDZ=0.25. Note. A bivariate Cholesky decomposition allows us to decompose the covariance between two traits, indicating the degree of genetic and environmental overlap between our measures. The first genetic factor (A1) represents genetic influences on life satisfaction. The extent to which these same genes also influence support quality is estimated (displayed by the diagonal pathway). The second genetic factor (A2) represents the genetic influence on support quality, which is independent of genetic influences shared with life satisfaction. The same decomposition is done for shared and non-shared environmental influences.