Childhood socio-economic disadvantage predicts reduced myelin growth across adolescence and young adulthood

Socio-economic disadvantage (SED) increases exposure to life stressors. Animal research suggests early life stressors affect later neurodevelopment, including myelin developmental growth. To determine whether human childhood SED affects myelination in adolescence and early adulthood we measured the developmental increase of a sensitive myelin marker, magnetization transfer (MT), in a longitudinal study. Childhood SED was associated with globally reduced MT, as well as slower intra-cortical MT increase in widespread sensory-motor, cingulate, insular and prefrontal areas and subcortical areas. Parental education partially accounted for the SED effects on MT increase, while positive parenting provided a partial protection against the impact of SED. Thus, early socio-economic disadvantage, a vulnerability factor for a range of ill-health outcomes, is a risk factor for aberrant myelin growth during a critical developmental period that is associated with a high risk of psychiatric disorder.


55
Socio-economic disadvantage (SED) increases exposure to childhood adversity (Pascoe et al., 2016). 56 In turn, adversity is associated with specific patterns of impairments in human brain development 57 (Evans & Cassells, 2014;Johnson, Riis, & Noble, 2016;McDermott et al., 2019;Whittle et al., 2017). 58 The impact of early life socio-economic disadvantage (SED) on brain development is poorly 59 understood at the microstructural level. Here, we make use of a unique longitudinal sample of young 60 people and in-vivo quantitative MRI, a measure of macromolecular content sensitive to myelin 61 content, to examine the effects of SED on myelin development. 62 Animal studies, where early stressors are under experimental control, show a causal impact of 63 adversity on brain growth, even during adolescent development (Howell et al., 2013;Liu et al., 2012;64 Zhang, 2017). Some experimental manipulations, such as variable foraging demand (Coplan et al., 65 2016(Coplan et al., 65 , 2006, bear resemblance to the resource uncertainty encountered in human SED, but it is still 66 difficult to generalize from animal stress to human disadvantage. Thus, mindful of the dangers of 67 making causal implications (Wax, 2017), it is important to map longitudinally human microstructural 68 brain changes such as myelination in relation to SED. This can allow a more detailed comparison with 69 animal findings, and close a causal explanatory gap (Donahue et al., 2018).Cortical myelin likely 70 reflects local neuritic insulation and fibre density (Glasser et al., 2014). It enables myeloarchitectonic 71 parcellation (Glasser et al. 2016) and influences neuronal dynamics (Demirtas et al., 2019). 72 We recently mapped neurotypical myelin development during adolescence and young adulthood, 73 using myelin-sensitive magnetization transfer saturation (MT) (see also Turati et al., 2015) and 74 showed myelin growth is tied to mental health traits (Ziegler et al., 2018). Human neuroimaging 75 studies, for example in childhood as a function of cortisol reactivity (Sheikh et al., 2014) or in young 76 adulthood as a function of developmental stressors (Jensen et al., 2017), have linked adversity to 77 alterations in white matter myelin. However, these findings beg the question as to how myelination 78 unfolds in detail during adolescence, in both cortex and adjacent white matter, as a function of early 79 socio-economic disadvantage, rather than of specific stressors. 80 Here we seek to clarify whether patterns of longitudinal myelin growth during late development are 81 associated with early SED as well as establish what role parenting plays in mediating or moderating 82 any relationship (Hair, Hanson, Wolfe, & Pollak, 2015;Noble et al., 2015;Whittle et al., 2017). Under 83 the hypothesis that SED impacts brain development, we predicted that neighbourhood-level indices 84 of deprivation would be associated with both the mean level of myelination, and rate of myelin 85 growth, during late brain development. We found evidence in support of the latter but not the 86 former. Family factors, specifically lower parental education and poorer reported parenting, 87 explained important aspects of the observed relationship. 88

89
Early disadvantage is associated with slower myelin growth 90 We obtained 497 repeated structural MRI scans on 288 (51.7% female) healthy participants between 91 14 and 25 years of age (see Methods). We used an observational accelerated-longitudinal design of 92 these community dwelling English young people and focused on longitudinal findings, a focus which 93 obviated biases seen in cross-sectional estimates with respect to development (Lindenberger,Von 94 Oertzen, Ghisletta, & Hertzog, 2011). We used MT rather than morphometric indices as a measure of 95 4 (myelin) development, following the priorities highlighted recently by neurodevelopmental 96 researchers (Walhovd, Fjell, Giedd, Dale, & Brown, 2017). 97 We analysed grey and white matter MT growth related to age, within as well as across participants, 98 using the efficient 'sandwich estimator' (Guillaume et al., 2014) Figure S1). 105 We examined putative explanatory variables both by entering them as covariates and moderating 106 factors in the imaging analyses (cf. methods). 107 Strikingly, we found that worse early life SED (i.e. living in a more deprived neighbourhood in before 108 age 12) correlated robustly with reduced rate of MT increase (MTr) in multiple brain areas (Figure 1  109  and Supplementary table 1). This supported an hypothesis that SED is associated with reduced 110 myelin growth (accounting for age and visits) during late adolescent development (Howell et al., 111 2013;Jensen et al., 2017). Moreover, a reduction in MTr was also observed globally (MTr within  112 whole-brain grey matter, t(321)=-2.87, p=.0022, one-sided, cf. methods). This contrasted with 113 current abode in a disadvantaged neighbourhood, which predicted reduced growth only weakly 114 (Supplementary Figure S2). Furthermore, there were no brain areas where SED was significantly 115 associated with increased MTr (or MTm), which might be expected if some changes reflected 116 adaptive or compensatory early growth (Ono et al., 2008;Ziegler et al., 2018) . 117 Early life SED correlated with reduced intra-cortical MTr across multiple brain regions including mid-118 and posterior cingulate, precuneus, operculum, insula (Figure 1a-b) and prefrontal areas, especially 119 on the right. Interestingly, MTr of juxta-cortical white matter and also subcortical regions was 120 similarly reduced (Figure 1b-c, Supplementary table 2). These rate reductions were most 121 pronounced in highly myelinated areas (e.g. M1, S1; for maps see e.g. Glasser et al., 2014) and in 122 regions showing significant developmental MT increase in the very same group of participants 123 (Ziegler et al., 2018). 124 Our hypothesis that the mean level of myelin, as reflected by the mean MT over visits would show a 125 similar reduction with SED was supported at the global brain level (MTm; accounting for age, visit, 126 sex, and confounds -see methods). Higher SED was associated with reduced global MTm (t=-2.15, 127 p=.016, one-sided, df=321). The local analysis however showed no significant associations between 128 early life SED and MTm (FDR corrected). In the light of the significant global result, greater statistical 129 power may be needed to resolve local MTm effects. Age-typical growth slows down with worsening early life SED in cortex, especially in bilateral precuneus/posterior cingulate, sensory-motor, premotor, sub-genual, and prefrontal areas (z-maps showing negative SED by time/visit interactions, p<.05 FDR corrected, N=328/185 scans/subjects in A-C, 45.7% female). B. MT growth is also reduced in insula, operculum (left) and the white matter adjacent to the affected cortex (right). C. Hippocampal and striatal grey matter, and core white matter regions also showed reduced growth of MT. D. SED-dependent rate of MT growth over visits in a region-ofinterest sphere encompassing the central operculum/posterior insula (radius 6mm, centre at x= -47, y=-22, z=13 mm, MNI). Plot shows subjects with higher SED (light yellow) compared to low SED subjects (dark red) express significantly less MT growth over visits (coloured lines in right panel indicate the interaction effect; yaxis: MT; x-axis: time of scan in years relative to each subject's mean age over visits). How early life SED influences brain myelination is likely to involve family-related factors, indexed 144 both by demographic characteristics but also how parents look after their children (Whittle et al., 145 2017). We examined whether parental education, parental occupation (a proxy for family income), 146 and self-report measures of parenting quality accounted for the effect of SED on myelin growth 147 trajectories (Ronfani et al., 2015;Sarsour et al., 2011). Parental educational qualifications were 148 translated to years-of-education to derive a continuous measure. This variable partially accounted 149 for our MT findings, broadly replicating but also expanding upon existing findings (Noble et al., 150 2015). Specifically, while peak clusters remained significant, their extent was much reduced, 151 especially in the medial surface of the brain. For example, the left subgenual, right medial motor and 152 right posterior cingulate clusters were largely abolished (cf. Figure 2A vs. Figure 1A). Therefore 153 parental education appears to index influences overlapping with neighbourhood-level SED, but 154 further important influences operate within SED. Against our hypothesis, controlling for poorly paid 155 parental occupation ("Standard Occupational Classification: SOC2000 | HESA," n.d.) without other 156 family covariates had little impact on the relationship between SED and myelination ( Figure 2B). 157 We next examined family factors proximal to the experience of participants, specifically subjectively 158 reported parenting quality. Here, we used overall parenting quality as an independent variable, 159 estimated by subtracting scores of negative parenting (e.g. harsh parenting or neglect) from those of 160 positive ones (e.g., parental warmth or praise; see Methods). Component negative and positive 161 scores were derived from three self-report questionnaires (Kiddle et al., 2017). We found that 162 parenting quality was not associated with MTr and thus could not mediate the effect of SED on MTr 163 ( Figure 2C). 164 By contrast to the absence of mediation effects, we found a significant moderating effect of 165 parenting on SED, such that better parenting significantly reduced the detrimental effect of early life 166 SED on adolescent MTr. This expands upon existing studies (Sheikh et al., 2014;Whittle et al., 2017). 167 Topographically, this moderating effect was largely confined to lateral prefrontal cortical MT ( Figure  168 3B) and subcortical MT (Supplementary table 3). Thus SED and parental education index overlapping 169 psychobiological influences, while improved parenting quality indexes a separate influence whose 170 presence might have a protective effect in more adverse environments. 171 7 173

Figure 2. Slower growth of myelin-sensitive MT as early life SED increases is partially explained by parental education but not other factors.
We present z-maps showing negative SED by time/visit interactions, p<.05 FDR corrected, N=328/185 scans/subjects, 45.7% female, when additionally controlling for multiple covariates and their respective time/visit interactions in A-C). A. Controlling for parental education reduces the impact of SED, in medial motor and premotor areas more than right lateral prefrontal ones (cf. Figure  1A) B. In contrast, controlling for parental occupation has minimal impact. C. Overall parenting quality has small impact (cf. see Figure 3). D. Controlling for time-varying IQ raw scores (here, WASI matrix) has negligible effect on the interaction (similar results for vocabulary, not shown). Controlling for baseline (or mean) IQ over the study period had a very similar effect. Colour scale is identical for A-D. The effect of parenting quality on MT was significantly steeper in males compared to females (see also Supplementary table 4). Z-maps show negative sex by parenting quality interactions, p<.05 FDR corrected, N=328/185 scans/subjects in A-C, 45.7% female, accounting for age, visit/time, sex, interactions and confounds. The right panel plots MT in superior temporal gyrus (6mm sphere around peak voxel) over parenting quality (x-axis, z-scored) and with adjusted data (grey/black) and model predictions (red/orange, effects of interest: intercept, parenting, sex by parenting). Higher parenting quality only showed a trend towards a positive main effect on cortical MT (p<.001, unc., not shown).

Individual risk factors 179
A number of factors measured at the level of the individual might form either important causal 180 contributors, outcomes or markers for the association between SED and MTr. We predicted that 181 lower IQ would be associated with altered patterns of myelination consequent upon SED for three 182 reasons. First, prior evidence suggests that IQ might be related to individual differences of 183 myelination (Dunst, Benedek, Koschutnig, Jauk, & Neubauer, 2014). Second, there is evidence that 184 IQ and socio-economic status share similar genetic determinants (Trzaskowski et al., 2014). Thus,185 genes that directly contribute to parental socio-economic success may also directly contribute to 186 differences of brain structure, as indexed by IQ. In other words, SED and brain growth could be 187 associated through horizontal genetic pleiotropy (Supplemental Figure S3). Third, the correlation 188 9 between socio-economic indices and IQ might be explained by morphometric brain measures 189 (McDermott et al., 2019). To test the hypothesis that IQ would index biological processes largely 190 overlapping with those present in SED, we controlled for verbal and matrix IQ scores ( Figure 2D), 191 both separately and as a total score. However, accounting for IQ in any of these ways left the 192 relationship between SED and slower MTr unchanged. 193 We also examined the effects of self-reported ethnicity and alcohol drinking, as these are thought to 194 be reflected in brain structure and connectivity (Noble et al., 2015;Smith et al., 2015). These failed to 195 account for our key findings (Supplementary Figure S4A-B). 196 197 We were interested whether determinants of poor physical condition, such as sedentary habits and 198 poor quality nutrition, at least as indexed by body mass index (BMI), explained why more deprived 199 children had lower MTr during adolescence. Thus, we tested a prediction that an association of MT 200 trajectories with early life SED would be partially accounted by an increased BMI. This is important as 201 both SED and poor parenting (Sleddens, Gerards, Thijs, Vries, & Kremers, 2011)) increase the risk of 202 being overweight (Salmasi & Celidoni, 2017), which is in turn associated with deviant white matter 203 development (Kullmann, Schweizer, Veit, Fritsche, & Preissl, 2015). 204 Mean BMI was positively associated with early SED and negatively to MTm but not MTr, and did it 205 account for the relationship between SED and MTr. As expected BMI increased with age during 206 adolescence, but, importantly, it increased faster the greater the degree of early SED ( Figure 4A). 207 Correcting for age, and variables of no interest (see Methods), greater BMI was associated with lower 208 MTm in anterior insula, anterior cingulate and other areas ( Figure 4B and 4C). The relationship was, 209 significantly more pronounced in males. Early life SED is associated with faster gain of body mass index (BMI) during youth. Linear-mixed modelling revealed positive age effects on BMI (t=4.6, p=3.9e-6, two-tailed, df=559), positive main effects of early life SED on BMI (t=2.05, p=.0407, two-tailed, df=559, N=568/384 observations/subjects) and a steeper age-related increase with higher SED (t=2.2, p=.0265, df=559, two-tailed), accounting for age, visits, sex, and interactions. B. Greater BMI is associated with lower cortical myelin-sensitive MT in the anterior cingulate, superior temporal, anterior insula cortex (z-maps showing BMI effects, p<.05 FDR corrected, N=277/155 scans/subjects, 47.5% female, accounting for age, visit/time, sex, early life SED, interactions and confounds). C. Right panel shows the plot of MT in insula (6mm sphere around peak voxel) over BMI (x-axis, centred) and with adjusted data (grey/black) and model predictions for sexes (red/orange, effects of interest: intercept, BMI, sex by BMI). The decline of MT with higher BMI is steeper in males than females. See also Supplementary Table 5. 217 218

220
We studied the relationship of early socio-economic disadvantage to myelin growth during 221 adolescence, de facto a relationship between economic disadvantage and deviant neurodevelopment 222 (Johnson et al., 2016). Early disadvantage correlated with a globally reduced myelin and reductions in 223 myelin growth longitudinally, as indexed by the sensitive marker magnetization transfer saturation, 224 MT. Reminiscent of other developmental findings (Sheikh et al., 2014;Whittle et al., 2017), parental 225 education, but not other predictors, accounted statistically for much of this effect. Better parenting 226 moderated the relationship, lessening the effect of economic disadvantage. As hypothesized, 227 increased BMI also indexed deficits in myelin development. However, although BMI was robustly 228 related to SED, BMI had an effect on myelination independently of SED (and not explaining it). These 229 findings have important implications for further, translational research and are relevant to protecting 230 brain development during youth as we shall see. 231 As our sample was healthy, slower myelination constituted neurodiversity rather than 232 neuropathology. Diversity can be seen as biologically encoding prior beliefs (Dayan, 2012;Moutoussis 233 et al., 2018) conferring adaptation, maladaptation (Jensen et al., 2017) or both, depending on 234 functions yet to be studied. On the one hand, the fact that we observed only reductions in MTr, that 235 these were very extensive, and that they were related to early but not current SED, suggest 236 maladaptation, at least with respect to current circumstances. 237 Parental education substantially mediated the effects of SED (Figure 2), suggesting these two 238 processes index overlapping biological pathways, reducing MTr. Remarkably, according to our results 239 this is unlikely to be due to IQ-related genes, to parenting quality as perceived by the young person, 240 over-weight, alcohol drinking or ethnicity. It is thus important to further understand what (non-IQ) 241 genes or proximal environmental drivers underpin different myelination trajectories. These may be 242 diverse and could include different parental behaviours not captured by our measure of parenting, 243 the provision of enriched primary schooling, other family environment factors such as parental 244 conflict, or factors associated with early peer processes. 245 Early poverty is associated with lower life achievement and IQ (Evans & Cassells, 2014)  would disproportionally affect the genetically vulnerable (Gage, Smith, & Munafo, 2016). Vertical 254 pleiotropy is further explained in Figure S3. 255 Primate experiments show that early stressors (Howell et al., 2013) cause long-term problems, 256 including an impact on brain myelination. By analogy, early developmental stressors are likely to be 257 commoner among disadvantaged children. However here, unlike in animal studies, diffuse low-impact 258 mechanisms rather than focal, high-impact ones are more likely to account for the relationship 259 between SED and myelination. This is because the sample was healthy, and severe adversity was 260 under-represented ( Figure S1). If broad low-impact influences operate, future research should also 261 prioritize broad influences and interventions over searches for focal, high-risk subgroups. 262 In primate variable foraging demand (VFD) experiments, it is resource insecurity (Coplan et al., 2006), 263 rather than the average resource level, that affects infant neurodevelopment, suggesting that 264 neighbourhood SED may index stressful economic insecurity for all, not just the poor, consistent with 265 parental occupation not accounting for the effect of SED. That VFD effects may be mediated by 266 maternal preoccupation by insecurity (Coplan et al., 2006) would be consistent with positive parenting 267 modestly mitigating the effects of SED (Figure 3). 268 We did not replicate a number of findings in the literature connecting SED to macroscopic measures 269 such as grey matter volume or surface area. Our smaller sample, though a limitation, is likely to mean 270 that the myelination effects are more prominent, so easier to detect. Results are also consistent with 271 myelination not being straightforwardly reflected in macroscopic measures (Grydeland,Walhovd,272 Tamnes, Westlye, & Fjell, 2013 and/or improve parental education and parenting in richer countries should prospectively examine 288 their impact on myelination. 289 In conclusion, neighbourhood deprivation during development was associated with slower myelin 290 growth markers during adolescence and young adulthood. This was independent of baseline IQ, 291 ethnicity or parental occupation, while parental education statistically explained much of the effect 292 and may give clues about causal mechanisms. Causation, functional consequences of myelination and 293 policy implications provide a fertile context for future investigations. 294 13

Recruitment, demographic and psychological measures 296
Participants were recruited from the Neuroscience in Psychiatry network participant pool (Kiddle et 297 al., 2017). From this pool, also known as the '2K sample', 300 participants were recruited for the 298 present scanning study. We aimed to exclude all but the most minor psychiatric and neurological 299 symptomatology, and therefore screened by self-report participants to not have current or previous 300 relevant medical histories. We finally analysed 497 available brain scans from 288 healthy (149 female) 301 individuals that passed quality control. In particular, data from 100, 167, and 21 subjects with one, 302 two or three visits per person were available, with mean (standard deviation) follow-up interval of 1.3 303 (0.32) years between first and last visit. 304 305 Self-defined ethnic group was asked about shortly after recruitment and in terms of the following 306 broad classes: White (1858 or 80% of declared ethnicity), Black (3.7%), Asian (8.5%) Mixed (6.0%), 307 Other (2.1%), 'Prefer not to say'. On the day of scanning, participants also completed the Wechsler 308 Abbreviated Scale Intelligence of (WASI) (Kiddle et al., 2017). As ours was a developmental study, we 309 used the raw subscale scores for vocabulary and matrix IQ and explicitly analysed, and their 310 dependence of age. Unless otherwise stated, IQ measurements refer to the time of the first, 'baseline' 311 scan. 312 The measure of overall parenting quality that we used here was a composite of the Positive Parenting 313 Questionnaire (PPQ), Alabama Parenting Questionnaire (APQ) and Measure of Parenting Style 314 (MOPS). All three were obtained within about a month of the first scan (Kiddle et al., 2017). We took 315 the reversed positive parenting total score from the PPQ and the similarly reversed positive parenting 316 scales from the APQ, the negative parenting scales from the APQ (inconsistent discipline, poor 317 supervision, and corporal punishment) and the negative parenting scores for the MOPS (abuse, control 318 and neglect), which were standardized and summed to make a composite negative parenting scale. 319 The internal consistency of the resulting total score was alpha = .96. 320 As far as parental education is concerned, the young people in our study reported the highest 321 qualification and occupational level of their parents. This data was obtained for the mother, the father, 322 and if applicable the mother's partner and the father's partner. These were converted to an ordinal 323 scale, according to a categorization of educational achievement in England -that is, none, primary 324 school, secondary school -GCSE's, sixth form -A levels, skills-based trainings, undergraduate 325 education, postgraduate education or higher professional training. We then took as starting score the 326 education level of the female parent (usually biological mother) and compared it with the primary 327 male parent (mother's partner or biological father, in that order of priority). It was unusual for these 328 to differ by more than one on this ordinal scale. Therefore, if only one parent score was available, we 329 used that for 'parental education'; otherwise we averaged the two prioritized scores. 330 331

Measures of Socio-Economic Disadvantage 332
We used the following measures of socio-economic disadvantage. As our central measure, we used 333 the neighbourhood proportion of households below the official poverty income around the 334 14 participant's residence ("Small area model-based households in poverty estimates, England and Wales 335 -Office for National Statistics," n.d.) at the time of first scan. We also used an index of parental 336 education (IPE); and the mother's and father's SOC2000 occupational class ("Standard Occupational 337 Classification: SOC2000 | HESA," n.d.). 338 339 MRI data acquisition and longitudinal preprocessing 340 Brain scans were acquired using the quantitative MPM protocol (Weiskopf et al., 2013)  the tools described in more detail below and in the methods section of the preceding paper that 347 focussed on effects of demographics in the same sample (Ziegler et al., 2018). 348

349
To assess macromolecular growth during development, we used a longitudinal Voxel-Based 350 Quantification (VBQ) pipeline that follows the following steps (for more details and illustration see 351 Ziegler et al., 2018). First, images were serially registered. Each baseline -follow-up mid-point image 352 was then segmented into grey matter (GM), white matter (WM) and cerebrospinal fluid (using the 353 CAT12 toolbox of SPM). MT maps from all time-points were then normalized to MNI space, manually 354 inspected and checked for outliers. A during scan motion proxy was used to discard 10% of images 355 with strongest motion-induced artefacts and as a confounding covariate during all analyses. We 356 constructed masks for both grey and adjacent white matter using SPM neuromorphometrics atlas for 357 tissue-specific analysis of MT parameters. Finally, normalized MT maps were processed with a tissue-358 weighted smoothing procedure (7 mm FWHM). 359 360

Longitudinal MT image analyses 361
In order to quantify myelin development we took advantage of the observational accelerated 362 longitudinal design. We focused on how the brains change over study time/visit. We compared and 363 contrasted this to how brain structure varied with scan-midpoint age across participants in the study. 364 To do so, we used the Sandwich Estimator (SwE) method for voxel-based longitudinal image analysis 365 (Guillaume et al., 2014, http://www.nisox.org/Software/SwE/). This so-called marginal model 366 describes expected variability as a function of predictors in a design matrix, while additionally 367 accounting for correlations due to repeated measurements and unexplained variations across 368 individuals as an enriched error term. 369 In our analyses, we focused on factors time/visits and mean age of the individual (over all visits) on 370 MT across the whole brain. To investigate how exposure to poverty was related to brain trajectories 371 and altered growth, we enriched the models by adding a main effect (SED as measured by the NPI, as 372 a predictor of mid-point MT) as well as the interaction of SED with individual MT change over scan 373 sessions (visits or within-subject study time). The latter metric allowed us to assess how myelin growth 374 is associated with SED (e.g. lower myelin growth upon exposure to high SED), whereas the former 375 15 indicates how SED relates to overall myelination differences across individuals accounting for other 376 covariates, such as visit, mean age, and sex. We a-priori hypothesized reduced levels of myelin and 377 impaired myelin growth with higher SED. The effects of visit, age, sex, and non-linearities (e.g. in terms 378 of age by age and age by time interactions) of age-related trajectories, and for first order interactions 379 among all demographic variables are presented elsewhere (Ziegler et al., 2018). All analyses were 380 carried out with scanning site, total intracranial volume and motion regressors as confounds. More 381 mathematical details on SwE and longitudinal design specification can be found in supplementary  382 information of Ziegler et al., 2018. 383 We then tested whether the observed associations of SED might be explained by further covariates, 384 by including the latter on the same footing as SED in analyses. All models were tested for indications 385 of effects of sex, IQ, parental education, parental occupation and self-reported ethnicity. We further 386 conducted moderation analysis in terms of indications whether above family or individual factors 387 show a significant interaction with SED either on mean level MT or MT growth over visits. Thus, two-388 way interaction with SED (e.g. parenting quality by SED) and three-way interaction terms (e.g. 389 parenting quality by SED by time/visit) were included in addition to all main effects, time, age, sex, 390 and their interactions in SwE models of local MT. We controlled for the False Discovery Rate (FDR,391 p<.05) during corrections for multiple comparisons in all image analyses. 392 393

Linear mixed effects modelling of global MT and BMI 394
To assess the effects of SED on global MT and on BMI, we used linear mixed-effects modelling (LME,395 cf. supplementary information Ziegler et al., 2018). We specified corresponding fixed effects design 396 matrix including time, age, sex, SED, and first order interactions while accounting for confounds. 397 Random-effect intercepts were included and proved optimally suited using likelihood ratio tests. T-398 values of fixed effects coefficients and corresponding (one-sided) p-values were calculated to test for 399 detrimental main effects of SED and time/visit or age interactions. More mathematical notes on LME 400 and longitudinal design specification can be found in supplementary information of Ziegler et al., 2018. 401 402 403

Macrostructural measures 404
Finally, to complement the main focus of this study in assessing SED-related correlates of novel, 405 quantitative, myelin-sensitive MT (using VBQ), we also tested for previously reported relationships of 406 SED with conventional metrics, i.e. Voxel-Based and global Surface-based Morphometry (VBM & 407 SBM). For this purpose we used non-linear registration to obtain normalized (grey and white matter) 408 tissue segment maps using both within-and between-subjects modulation. This was followed by 409 Gaussian smoothing (6mm). Moreover, cortical surface reconstructions of all participants' midpoint 410 was obtained (using CAT Toolbox), and cortical thickness, surface area, gyrification index, and sulcal 411 depth was assessed in native space, resampled to a surface template and smoothed with 12mm. 412 413