Results
Social Network Size and Diversity
We first examined the associations of network indices with various demographic, social, affective and health factors (See Table 1). In general both diversity and network size were associated with similar measures, including age, income, education, PA and ISEL scores. Only network size was associated with current smoking status. There was no difference between men and women in social network size [t(154) = 0.029, P = 0.33] or diversity [t(154) = 0.963, P = 0.87]. Moreover, neither network measure was associated with measures of physical or cardiorespiratory health. In the analyses later, we control for central adiposity, age, sex and years of education because these factors are plausible confounds of white matter structure and inflammatory measures.
Inflammation and Social Network Structure
After controlling for central adiposity, age, sex and years of education, IL-6 was negatively associated with the diversity of a person's social network [r(128) = −0.194, P = 0.027], but uncorrelated with social network size [r(128) = −0.088, P = 0.319]. We did not find a relationship between network diversity and CRP [r(137) = 0.04, P = 0.638]. CRP was also not correlated with social network size [r(137) = 0.125, P = 0.144]. Finally, as expected CRP and IL-6 were moderately correlated with each other [r(120) = 0.415, P < 0.0001].
White matter and social network structure
We found that across all white matter voxels there was a predominant positive association between FA and diversity of a person's social network (mean = 0.012 ±0.039 s.d.; Figure 1A and B), after controlling for age, sex, education and central adiposity, which have all be found to be independently associated with measures of white matter integrity (Westlye et al., 2009; Gianaros et al., 2013; Verstynen et al., 2013). This means that in a numerical majority of voxels, individuals with more high-contact roles had greater microstructural white matter integrity. In the uncorrected statistical maps, we found several independent clusters with strong positive associations to diversity (Figure 1A), including near prefrontal boundaries of the dorsal medial prefrontal cortex, considered part of the 'social brain' (Lewis et al., 2011; Powell et al., 2012).
(Enlarge Image)
Figure 1.
(A) Uncorrected maps (P < 0.025) showing the voxels with strong positive (red) and negative (blue; none survive thresholding) associations with the SNI measure of diversity [SNI (Div)] and FA. (B) Probability distribution of regression coefficients between SNI (Div) and FA across all white matter voxels shown in panel A. The purple distribution shows the observed values, while the gray distribution shows the average distribution shape across a set of randomized permutation tests designed to model the null effect. (C) Location of the strongest cluster of voxels in the brain, with associations that survive multiple comparison corrections (FDR < 0.05) and clusterwise adjustments (k > 40).
Although there was a global positive association between network diversity and FA across a majority of voxels (Figure 1A and B), this effect was particularly strong in a cluster of voxels near the posterior section of the genu of the corpus callosum (Figure 1C). This large cluster of voxels surpassed both multiple comparison correction (FDR < 0.05) and cluster size thresholding (k > 40). No such patterns were observed for social network size, thus, no further exploratory analysis was done on this measure.
To ensure that head motion did not interfere with the associations between SNI and FA (Yendiki et al., 2013), we ran a linear regression between the two SNI variables and head motion parameters. We found no significant correlations between SNI and head motion (all r's < −0.137, all P's > 0.11); thus, spurious differences in head movement during the scans cannot account for the relationship between SNI diversity and FA that we detected.
In order to assess whether the association between SNI diversity and FA in this cluster is driven by particularly low diversity individuals, as would be predicted by the social isolation findings in rodents (Liu et al., 2012), we segmented the sample into three groups: low social diversity (<5 roles, n = 38), moderate diversity (between 5 and 7 roles, n = 65) and high diversity (8 or more roles, n = 42). The unadjusted FA values within the corpus callosum cluster were extracted and averaged across groups. After controlling for age, sex, education and central adiposity, FA increased consistently with each diversity group (Figure 2A), with a significant difference between low- and high-diversity individuals [t(79) = 2.79, P = 0.008] and marginal differences between low and medium diversity groups [t(106) = 1.48, P = 0.14] and medium and high-diversity groups [t(102) = 1.46, P = 0.14]. Thus, the FA associations within this cluster are not driven exclusively by low or high-diversity individuals.
(Enlarge Image)
Figure 2.
(A) Average cluster FA for each SNI (Div) group. Subjects were assigned to groups based on which tertile of the social network distribution they fell into. The same values for AD and RD with the cluster are shown in panels B and C, respectively. See text for statistical results. All error bars show the standard error of the mean.
Within the SNI-related cluster RD decreased as social network diversity increased, after age, sex, education and central adiposity were controlled for (Figure 2B; r = −0.15, P = 0.021). As with FA, we observed a significant difference between low and high-diversity subjects [t(79) = 2.04, P = 0.044], but no difference between low and medium diversity individuals [t(106) = 1.55, P = 0.12] or medium and high-diversity individuals [t(102) = 0.72, P = 0.47]. Unlike RD, we did not observe a significant group effect on AD within the cluster (Figure 2C; β = −6.55 × 10, P = 0.45, Spearman's r(145) = −0.05, P = 0.200). This selective association with the radial component of the diffusion signal is consistent with patterns seen in animal models of demyelination (Klawiter et al., 2011).
Previous observations in rodents (Hermes et al., 2006; Karelina et al., 2009) predict that inflammatory factors should be moderately associated with white matter structure. Within the SNI-related cluster we found that, after controlling for age, sex, education and central adiposity, FA was negatively correlated with circulating levels of IL-6 (Figure 3; Spearman's r(126) = −0.14, P = 0.017), and trending, but not significant, when correlated against circulating CRP (Spearman's r(135) = −0.12, P = 0.084). However, no such correlation was observed between IL-6 and either RD [r(126) = 0.07, P = 0.165] or AD [r(126) = −0.09, P = 0.096]. This negative association between IL-6 and FA generally suggests that inflammation may be playing a role in white matter variation. However, this trend disappears after controlling for age and sex (all P's > 0.1), thus negating IL-6 as a mediating variable within this sample.
(Enlarge Image)
Figure 3.
Scatterplot showing the negative correlation between clusterwise FA and IL-6 (natural log transformed).
Finally, we expanded our analysis to include a full set of psychosocial and health factors against the cluster-wise FA values. These results are shown in Table 2. Of these factors only smoking status correlated with FA within the target cluster. This association is consistent with a possible inflammation link to the FA variation in this region. However, given the lack of association with socioeconomic and general health factors that correlated with social network measures, it is unlikely that this SNI and white matter association is the mediated by socioeconomic (Gianaros et al., 2013) or physical health (Verstynen et al., 2013) pathways previously reported in this sample.
Network Connectivity Through SNI-related Voxels
To characterize the pathways running through this cluster, we performed deterministic tractography on a template of 60 neurologically healthy volunteers who were scanned using a high angular resolution form of diffusion imaging that is optimal for tractography in MNI space (see Methods). Using the SNI-related cluster as a region of interest, we found that a majority of the cluster covers interhemispheric connections between the superior and middle frontal gyrus (Figure 4). The dorsal aspects of the cluster also include projections from the left cingulum tract while the ventral aspects intersect with portions of the superior infundibulum. Thus, the voxels having particularly strong associations with social network diversity predominantly reflect pathways connecting the left and right dorsolateral prefrontal cortex, with some carry over to frontoparietal and limbic pathways.
(Enlarge Image)
Figure 4.
Coronal, A and sagittal, B, views showing the tracked fiber streamlines that run through the SNI (Div) related cluster (yellow region). Tractography was performed on the CMU-60 template brain.
To determine the extent that variation in the integrity of the white matter cluster associates with the functional properties of this connected circuit, we selected the cortical clusters of the tractography endpoints that had more than 50 contiguous voxels (ROIs; Figure 5A) and used resting-state BOLD time series to evaluate the functional connections between all ROIs for each subject. After correcting for multiple comparisons (FDR < 0.05) and controlling for sex, age and central adiposity, one cluster pair was found to be negatively associated with FA within the SNI-related cluster (β = −1.09, P = 0.0146). This region reflected functional connections between an area on the superior frontal sulcus [SFG; center of mass (COM) in MNI-space = 22, 45, 30] and the rostral striatum near the nucleus accumbens (NAcc; COM = 9, 7, 13), both in the right hemisphere. The diffusion component that explained the most variance between subjects' functional connectivity was the RD component (β = 652.30, P = 0.0016). Although the AD component was also associated with changes in functional connectivity (β = 351.95, P = 0.0286), this significance does not survive the threshold for multiple comparisons (Bonferroni adjusted P = 0.0167). Thus, within this corticostriatal pathway that runs through the SNI-associated cluster, we observed that individual differences in FA predict variance in functional connectivity, particularly with radial component of the DTI signal.
(Enlarge Image)
Figure 5.
(A) Sagittal slices showing the eight ROIs masks generated from cortical endpoints of tracked fiber pathways from analysis shown in Figure 4. These ROIs were used as masks for resting state functional connectivity analysis on the subset of the sample (N = 110) that had viable resting state fMRI data. Variation in FA within the SNI (Div) related cluster correlated with variation in functional connectivity between a region on the right medial wall of the SFG and a region near the right NAcc (lower row). ACC, anterior cingulate cortex; PCC, posterior cingulate cortex; NAcc, nucleus accumbens; vmPFC, ventromedial prefrontal cortex; SFG, superior frontal gyrus; R, right; L, left. (B) Mediation analysis showing how FA and RD served as indirect pathways linking SNI (Div) to functional connectivity of the significant corticostriatal pathway identified in A. Size of the lines represents the magnitude of the indirect (a*b) pathway. Dashed lines indicate non-significant results after correcting for multiple comparisons. Bracketed numbers indicate the upper and lower bound of the bias corrected and accelerated 95% CI for the indirect pathway.
Although white matter integrity correlated with functional connectivity of the corticostriatal pathway, social network diversity did not have a direct association with the functional connectivity between these regions (β = −0.0027, P = 0.36). However, because social network diversity is associated with white matter integrity within this cluster, it is possible that white matter serves as an indirect pathway linking SNI diversity with corticostriatal functional connectivity. Using a statistical mediation analysis (Preacher and Hayes, 2008), we found that FA (a*b = −0.0049, P = 0.0005) and RD (a*b = −0.0053, P = 0.0002) served as significant indirect pathways linking SNI diversity with corticostriatal functional connectivity (Figure 5B). This indirect pathway was marginally significant with AD (a*b = −0.0013, P = 0.046), but this does not survive multiple comparison correction and the 95% confidence interval (CI) includes zero. Therefore, only FA and RD are indirect mediators linking social network diversity to corticostriatal functional connectivity.