Children’s family income is associated with cognitive function and volume of anterior not posterior hippocampus

Children from lower income backgrounds tend to have poorer memory and language abilities than their wealthier peers. It has been proposed that these cognitive gaps reflect the effects of income-related stress on hippocampal structure, but the empirical evidence for this relationship has not been clear. Here, we examine how family income gaps in cognition relate to the anterior hippocampus, given its high sensitivity to stress, versus the posterior hippocampus. We find that anterior (but not posterior) hippocampal volumes positively correlate with family income up to an annual income of ~$75,000. Income-related differences in the anterior (but not posterior) hippocampus also predicted the strength of the gaps in memory and language. These findings add anatomical specificity to current theories by suggesting a stronger relationship between family income and anterior than posterior hippocampal volumes and offer a potential mechanism through which children from different income homes differ cognitively.

Exploratory analyses testing whether age moderated volume-cognition relationships. 144 We explored whether age moderated volume-cognition relationships by adding age as 145 an interaction term to models assessing how anterior and posterior volumes relate to 146 memory and vocabulary scores. There were no significant interactions between age and 147 anterior hippocampal volumes to predict memory or vocabulary scores (ps > 0.05, 148 Supplementary Tables 41 & 42). However, age moderated the relationship between 149 posterior hippocampal volumes and memory scores (b = −0.0007, SE = 0.0002,p = .003,r = 0.11) and vocabulary scores (b = −0.00007, SE = 151 0.00002,t(685) p = .005,r = 0.11,Supplementary Tables 43 152 & 44). 153 To further investigate this interaction, we divided our sample into separate age 154 bins (young children: 3-7, n=126; older children: 8-12, n=245; adolescents: 13-17, 155 n=170; young adults: 18-21, n=149) and ran a regression model to test which age bins 156 differed from each other. We found that adolescents had a more positive relationship 157 between the posterior hippocampus and memory scores than did older children 158 (interaction: b = 0.006, SE = 0.003, t(680) = 2.14, p = .032, r = 0.08). Examining the 159 relationship between posterior volumes and memory scores in these groups separately 160 revealed that the memory-volume relationships were not statistically significant in any  SE = 0.0003,p = .602,r = 0.04). We note that our results might differ from previous studies 170 examining how age moderates volume-cognition relationships 70 , however, owing to the 171 large sample size in our study, we are well powered to detect significant effects. 172 173 Supplementary Note 2 174 175 Exploratory analyses testing whether nonlinear age transformations altered our findings. 176 For models that included cognitive scores as the dependent variable, we controlled for 177 the linear effect of age, but did not include nonlinear age terms as covariates. However, 178 for completeness, we also tested whether nonlinear age transformations (quadratic, 179 cubic) explained additional variance in cognitive scores. We found that the linear, 180 quadratic and cubic age terms all explained significant variance in memory scores in the 181 full sample (ps < .05). In contrast, the quadratic (but not cubic) age term explained 182 additional significant variance in memory scores in the lower income sample (p < .05). 183 Similarly, the quadratic (but not cubic) age term explained significant variance in 184 vocabulary scores in the full sample, and in the lower income subsample (ps < .05). 185 Therefore, to ensure the pattern of results were maintained, we re-fit the models in 186 which cognitive scores were the dependent variable and included the relevant non-187 linear age transformations as covariates. We report the results from these models 188 below. 189 190 Adding non-linear age terms to income-cognitive score models. Including nonlinear age 191 transformations did not change the pattern of results for income-cognitive score models. Supplementary Tables  222  223  224  225 Supplementary participants). Age was mean centered, sex and income subsamples were effect-coded 367 (female = -1, male = 1; income ≤$75k = -1, income >$75k = 1) before being included in 368 the analysis. FDR corrected p value is reported in brackets for the critical effect of 369 interest.

371
Summary of findings presented in Supplementary Note: A linear regression was used for this analysis (n= 293 participants). Age was 519 mean centered, sex was effect-coded (female = -1, male = 1), income was log 520 transformed, and scanner was dummy coded prior to being included in the model. FDR 521 corrected p value is reported in brackets for the critical effect of interest.

523
Summary of findings presented in Supplementary was effect-coded (female = -1, male = 1), and age was mean centered in the model. 743 Cognitive scores were missing for 10 of the 689 participants who were included in the 744 education analyses. FDR corrected p value is reported in brackets for the critical effect 745 of interest.

747
Summary of findings presented in Supplementary was effect-coded (female = -1, male = 1), and age was mean centered in the model. 756 Cognitive scores were missing for 10 of the 689 participants who were included in the 757 education analyses. FDR corrected p value is reported in brackets for the critical effect 758 of interest.

760
Summary of findings presented in Supplementary Age was mean centered, sex was effect-coded (female = -1, male = 1), scanner was 770 dummy coded prior to being included in the model. FDR corrected p value is reported in 771 brackets for the critical effect of interest. 772 773 774 Summary of findings presented in Supplementary centered, sex was effect-coded (female = -1, male = 1), income was log transformed 796 prior to being included in the model. FDR corrected p value is reported in brackets for 797 the critical effect of interest. 798 799 800 Summary of findings presented in Supplementary Table 48: There was no significant  801 interaction between minority status and income, indicating that income-episodic memory 802 relationships were similar in minority status and non-minority status individuals. Notably, 803 income remained significant even after accounting for minority status, highlighting the 804 generalizability of income-episodic memory relationships. 805 806 807 808 Supplementary Volumes were modelled as a function of region, age, sex, and the interaction between age and region. Because regions were nested within participants, we modelled a random intercept for participants to account for the random effect of participant on regional volumes, i.e., lmer(Volumes ~ regions * age + sex + income + scanner + (1|participant)), n = 703. Prior to running the model, age was mean centered, and volumes were dummy coded (anterior = 0), posterior = 1). FDR corrected p value is reported in brackets for the critical effect of interest.