SUPPLEMENTARY METHODS In order to assess whether our findings were of behavioral relevance, we investigated the relationship between diffusion metrics (FA and RD) and cognitive performance and how BDNF genotype modulated these. Neuropsychological assessment Cognitive test performance data was available for a subset of participants (Supplementary Table 1). As described in detail in (Dickinson et al, 2011), neuropsychological testing in the NIMH Clinical Brain Disorders Branch “Sibling Study” protocol is performed using a test battery including measures or variables from the Wechsler Adult Intelligence Scale (WAIS), Wechsler Memory Scale (WMS), California Verbal Learning Test (CVLT), Wisconsin Card Sorting Test (WCST), and Wide Range Achievement Test (WRAT), among others. Composite scores for six cognitive domains (verbal memory, visual memory, processing speed, n-back working memory, span working memory, and card sorting) and a general cognitive ability (“g”) composite were derived from the battery based on exploratory factor analyses. Factor-based composites were calculated as the average z-transformed score across contributing cognitive variables, and further transformed (e.g., squared or cubed transformations) to normalize the data distribution (see Supplementary Table 1 for demographic information). Data reduction of FA and RD data for exploratory analysis of relationships to cognition and BDNF genotype In an exploratory analysis, we examined the relationship between diffusion metrics derived from multiple tracts and all cognitive factors described above, with the exception of the visual memory factor where the number of available data points (n = 40) precluded a reasonable further analysis. In a first step, DTI estimates of brain white matter tracts were calculated and 1 reduced using factor analysis based on the full sample of 85 participants who had DTI data. Here, average FA and RD values were extracted for the ROIs provided by the JHU white matter atlas (Mori et al, 2008) as applied to the TBSS skeleton. AD was not explored given prior research indicating that its association with cognition is minor compared to FA and RD (Jacobs et al, 2011; Penke et al, 2010; Roberts et al, 2010). For each DTI metric (FA and RD), the number of variables was subsequently reduced by dropping ROIs belonging to the cerebellar peduncles and brain stem regions. The correlation matrix of all ROIs was then inspected and bilateral ROIs were averaged for most remaining tracts, since correlations between left and right hemisphere were r > 0.6. Only the ROIs for the left and right superior fronto-occipital (SFO) fascicles for FA (r = 0.4) were used as separate variables. This yielded 21 ROI values for FA and 20 for RD. All variables were normally distributed (Kolmogoroff-Smirnoff test: P > 0.05), with the sole exception of fornix estimates for RD. This variable was log transformed prior to entering it into further analysis. Factor analysis was subsequently run for FA and RD separately, and factors were extracted with the maximum likelihood method in Statistica 7 (Statsoft, Tulsa OK). The number of factors to be extracted were decided based on scree plots (Cattell, 1966), and factors accounting for less than 4% of the total variance were dropped. This procedure identified four factors for FA (see Supplementary Table 2 for details). After varimax normalized rotation, the identified factors could be interpreted as referring to 1) midline structures such as genu and body of the corpus callosum (11% explained variance), 2) subcortical tracts feeding into the frontal and occipital lobes (20% explained variance), 3) fronto-parietal and projecting fibers of the cortico-spinal tracts (18% explained variance), and 4) basal fronto-temporal tracts (4% explained variance). For RD, this procedure identified three factors (see Supplemental Table 2 for details) 2 including a regionally more unspecific, general factor (Factor 2, 34% explained variance), a posterior brain factor (Factor 1, 25% explained variance), and a medial structure factor (Factor 3, 10% explained variance). “Eigenvariate” values for RD and FA were calculated for each subject by weighing the data from each contributing ROI by the loadings for each factor. We then surveyed all the variables and dropped values that were more than 3 standard deviations below or above the mean in order to be less sensitive to outliers. This resulted in removal of one value for each of the following variables: card sorting, general cognitive ability, processing speed and for FA factor 3. The resulting individual factor diffusion values were then modeled as independent variables (together with BDNF genotype, sex, and age) in a general multiple regression backward stepwise analysis for association with six of the seven cognitive factor dimensions (described above) as dependent variables. The model was constrained as follows: all main effects were allowed into the model regardless of significance, and so were all two-way interactions with P < 0.1. No interactions beyond the second degree were entered into the model due to the complexity inherent in interpreting these interactions. If an independent variable was not significant with a p<0.1 either as a main effect or in an interaction, it was removed from the overall model, in order to obtain the simplest model possible, while explaining maximal variance. A corrected significance threshold of P < 0.008 for six comparisons was defined for the overall model. Univariate analyses of the models that survived this threshold were considered statistically significant when p<0.05. Hypothesis driven analyses 3 We addressed four hypotheses that have been put forth in the literature. 1) That working memory performance is predicted by the FA of the left superior longitudinal fasciculus (SLF) (Karlsgodt et al, 2008); 2) That processing speed is predicted by a principal component derived from the FA of eight white matter tracts (genu and splenium of the corpus callosum, bilateral dorsal cinguli, uncinates and SLFs) (Penke et al, 2010); 3) that verbal memory is predicted by the FA of the inferior longitudinal fasciculus (ILF), anterior and posterior cinguli (Kantarci et al, 2011); 4) that general cognitive ability is related to a principal component derived from the the FA of 12 tracts (genu and splenium, of the corpus callosum, bilateral dorsal cinguli, uncinates, anterior thalamic radiations, inferior longitudinal fasciculi and SLFs) (Penke et al, 2012). For all these hypotheses, the FA data was derived from the atlas-based ROIs discussed above. The initial models were backward stepwise multiple regressions with the cognitive factor of interest as the dependent variable and age, sex and the appropriate FA values as independent variables. After assessing these basic relationships, BDNF genotype was added to the model to assess whether it contributed significantly to prediction of the cognitive variables. The models were simplified in a second model building run by removing variables with p>0.1 as main effects or interactions. The significance for these models was set to 0.05 uncorrected because these were hypothesis-driven analyses with a prior basis in the literature. SUPPLEMENTARY RESULTS Exploratory Analysis of FA-cognition relationships and their moderation by BDNF Val/Met genotype Four factors were identified for FA (see Supplemental Table 2 for details). After varimax normalized rotation, the identified factors could be interpreted as referring to 1) midline structures such as genu and body of the corpus callosum (11% explained variance), 2) 4 subcortical tracts feeding into the frontal and occipital lobes (20% explained variance), 3) frontoparietal and projecting fibers of the cortico-spinal tracts (18% explained variance), and 4) basal fronto-temporal tracts (4% of explained variance: Supplemental Table 2). For RD, factor rotation identified three factors (Supplemental Table 3) including a regionally more unspecific, general factor (Factor 2, 34% explained variance), a posterior brain factor (Factor 1, 25% explained variance), and a medial structure factor (Factor 3, 10% explained variance). Multiple Regression Results for FA For FA, the results of multiple regression backward stepwise procedures are summarized in Supplementary Table 4a. For verbal memory, age, sex Factor1 and Factor 4 were dropped from the analysis. Univariate tests of significance are shown in Supplementary Table 4b, and the most statistically significant univariate result (the interaction of BDNF by FA Factor 2) is plotted in Supplementary Figure 1. For n-back working memory performance, age, sex and factor 4 were removed from the analysis. Results of the univariate analysis are shown in Table 4c and also in this case the most significant effect was the interaction of BDNF and factor 2, plotted in Supplementary Figure 2. This interaction remained significant after removing two possible outliers with the lowest working memory performance in the BDNF Val/Val group. For card sort performance, only age and factor 2 remained in the model, but the model was highly significant. Univariate results are shown in Supplementary Table 4d. The most significant effect was that of age, while factor 2 explained a smaller proportion of the variance (the relationship is shown in Supplementary Figure 3 prior to adjustment for age). 5 For processing speed (age and Factor 3 remained in the model), span (BDNF and Factor 2 dropped from the model) and G (only Factor 3 remained in the model) statistical significance of the overall model was not below the threshold established for multiple comparisons. Multiple Regression Results for RD For RD, the same procedure resulted in significant findings for n-back working memory and card sorting performance (Supplementary Table 5a). For working memory, all variables were retained in the model, but the role of BDNF genotype was modest and not significant in the univariate analysis. Factor 3 for RD appeared to negatively predict working memory performance (the higher the RD in the corpus callosum and columns of the fornix, the lower the working memory performance: Supplementary Figure 4). There was a significant interaction of factors 2 and 3, but this is difficult to interpret. For card sort performance (results for univariate analysis presented in Table 5c), BDNF genotype was dropped from the final model. The effects of RD factors were weakly significant in the univariate analysis. The main variable predicting performance was age. For verbal memory (BDNF, factors 1 and 3 in the final model), processing speed (age, factor1 and BDNF in the model), span (sex, Factors 1 and 2 in the model) and general cognitive ability (age, Factors 2 and 3 remaining in the model), results were below the threshold for statistical significance. Hypothesis based analyses 1) SLF FA predicts working memory performance The model using left hemisphere SLF FA alone as a predictor of n-back working memory was not significant, however, when we added age, sex, BDNF genotype and right hemisphere SLF to the model, statistical significance was present (multiple R2=0.26, F=2.43, p=0.015). Univariate results are presented in Supplementary Table 6. The most significant main effect was 6 that of sex (males>females). An interesting sex-by-right SLF FA interaction emerged, with reductions in FA predicting working memory in females, while the directionality of the association was the opposite in males (Supplementary Figure 5). BDNF genotype had a limited role in predicting working memory performance in this model. 2) The first principal component derived from the FA of 8 major tracts predicts processing speed The first principal component accounted for 54.5% of the variance in FA of the various tracts. Supplementary Table 7a shows the factor loadings for the first principal component. The overall model was not significant (multiple R2=0.047, F=1.32, p=0.27) and the addition of BDNF genotype in the model did not improve prediction of processing speed. 3) The FA of the ILF, anterior and posterior cinguli predict verbal memory The FA of the ILF (left and right hemisphere ROIs were entered into the model) did not predict verbal memory (multiple R2=0.016, F=0.31, p=0.87). The most parsimonious model included BDNF and the left ILF only, yielding a trend finding (multiple R2=0.085, F=2.47, p=0.068). Similarly, the FA of the ventral portion of the cingulum was not associated with verbal memory (multiple R2=0.053, F=1.11, p=0.36), but when BDNF was added to the model and the left cingulum dropped, trend significance was achieved (multiple R2=0.15, F=1.9, p=0.08). Also the dorsal cingulum did not achieve significance with (multiple R2=0.044, F=0.71, p=0.62) or without BDNF in the model (multiple R2=0.041, F=0.84, p=0.5). 4) The first component derived from the FA of 12 tracts predicts general intellectual ability 7 The factor loadings for the 1st principal component derived from the FA data of 12 tracts are shown in Table 7b. The first principal component derived from 12 tracts did not achieve significance, with or without BDNF in the model, in predicting general cognitive ability. SUPPLEMENTARY DISCUSSION The exploratory approach used here has the advantage of being relatively unbiased, since it includes samples of most forebrain tracts and it minimizes assumptions on which variables will enter the model by using stepwise regression in order to build the final model. Our analysis of the relationship between measures of white matter tract microstructure and cognition, which included the moderating effect of BDNF Val/Met, showed some specificity of FA factors and BDNF genotype in predicting different cognitive dimensions. For example, BDNF genotype had a role in predicting verbal and working memory performance by altering the relationship between Factor 2 and these cognitive domains, but we could not demonstrate a role in predicting performance in processing speed, card sort, span or general cognitive ability. Similarly, while Factor 2 (including the ventral portion of the cingulum, the fornix body/stria terminalis, a significant portion of posterior fibers and the anterior limb of the internal capsule [ALIC], and the right superior fronto-occipital fasciculi [SFOF]) had some role in predicting verbal memory, working memory and card sorting, only Factor 1 (midline structures such as the corpus callosum and the columns of the fornix) seemed to play a role in working memory. Factor 4 (external capsule and uncinate fasciculi) did not contribute to the determination of cognition alone or in combination with BDNF genotype. This somewhat regional pattern was also confirmed when we attempted to enter in the multiple regression analyses only one unrotated factor from the FA data, accounting for about 38% of the total variance. This analysis was not significant for any of the 8 cognitive variables, indicating that multiple rotated factors were necessary in order to show relationships to cognition in this sample. A finding that was not predicted was the significant interaction of BDNF and Factor 2 in predicting verbal and working memory performance. This is a difficult interaction to explain in biologic terms, but it suggests that the underlying basis of FA variance in Val/Val individuals is different than in Met carrier subjects. This may be indicative of BDNF genotype affecting the formation of white matter tracts and consequently memory performance during development, however, this inference is highly speculative given that this is just an association and we cannot prove the order of causality or if an unknown variable is driving this complex association. It should be noted that a similar finding with the same directionality was found by Chiang et al. (Chiang et al, 2011) for object assembly (a subtest that was not available to us) in the splenium of the corpus callosum and that further work by Chiang et al. found several SNPs that displayed this type of interaction in predicting performance intelligence quotient (Chiang et al, 2012). RD Factors were predictive of cognitive performance only in the case of Factor 3 and working memory. Also card sort performance was predicted by a model that included RD factors, but their statistical significance was limited. Also BDNF appeared to have a more limited role when combined with RD Factors than seen for the analyses of FA above. Although of interest, these associations suffer from multiple weaknesses that makes them difficult to interpret. Given our sample size, a reduction of the number of ROIs derived from the JHU atlas was necessary. We chose to drop regions associated with the brain stem and cerebellum and to average left and right ROIs with correlation coefficients above 0.6. Even after this data reduction, the ratio between the sample size and the number of variables entered in the 9 factor analysis (4:1, roughly) is not high enough to obtain stable solutions (Costello and Osborne, 2005). Moreover, the solutions found, even after varimax normalized rotation, contain several crossloadings (e.g. the genu of the corpus callosum loads on both Factor 1 and Factor 2 for FA), which complicate the biological interpretation of the data. The results are of difficult interpretation also due to the concomitant presence of main effects and interactions in the same model (e.g. verbal memory), which dilute the specificity of the findings. Finally, these analyses are extremely sensitive to outliers and to the particular group composition. A larger sample size and a completely independent replication sample are needed in order to support both the factor structure for the diffusion metrics and their associations with cognition. We also tried to approximate prior findings in the literature relating to prediction of cognitive variables based on FA. For example, Karlsgodt et al. (Karlsgodt et al, 2008) had reported in a sample of healthy volunteers (N=17) a positive correlation between the FA of the left superior longitudinal fasciculs (SLF) and verbal working memory performance. This was not true in our sample, although we used a transformed factor instead of simple accuracy during nback performance. Another difference from the work of Karlsgodt et al. (Karlsgodt et al, 2008) was that our n-back was a mixed numerical-spatial task rather than a verbal task. This may be consistent with the fact that while they found that the left SLF was predictive of performance, only the right SLF contributed as a main effect in our models. The effect of sex (males>female) seen here is consistent with males outperforming females in spatial tasks and with the data collected by our group in a much larger sample deriving from the sibling study (data available on request). Sex had not been included as a variable of interest in the analysis by Karlsgodt et al. It should be noted, however, that these models are highly dependent on the variables included. For example, a main effect of sex on working memory was not seen when we entered Factors 1-3 10 and BDNF in the model (Supplementary Table 4c) during our exploratory analyses of working memory performance. Another cause for skepticism in interpreting these results is that the model became statistically significant only when BDNF genotype was added into the model, however this variable was not significant at the p<0.05 level in the univariate analysis as a main effect or in interactions. The proportion of the variance explained in working memory by using the bilateral SLF values rather than the FA Factors (exploratory analysis) was similar (multiple R2=0.26 for SLF vs. 0.25 for Factors 1-3), an indication that it is possible to simplify the model and pick single variables or combinations of un-weighted variables in order to achieve better interpretability of the findings. Picking which regions of white matter should enter the prediction scheme, however, remains problematic due to the high interconnectedness of the brain. Future studies confirming the factor structure of the diffusion metrics in different age cohorts are necessary. Another issue we tried to address in this sample, is the reported relationship of processing speed with the first principal component derived from the FA of eight white matter tracts (Penke et al, 2010). The same group (Penke et al, 2012) also reported that the first principal component derived from the FA of 12 tracts, predicted general cognitive ability. Both of these studies (Penke et al, 2012; Penke et al, 2010) were conducted in healthy elderly subjects in their seventies. Most of the literature on relationships between diffusion metrics and cognition with large Ns has been centered on samples that include participants over 60 years of age. The first principal component obtained in our investigation accounted for 54.5% of the variance when derived from eight tracts and for 49% when using 12, which is roughly in accord with Penke et al. (Penke et al, 2012; Penke et al, 2010). Supplementary Tables 7a and 7b show the factor loadings for the first principal component derived from 8 or 12 tracts. Several differences emerge 11 from the findings of (Penke et al, 2010) (see their Table 1 in comparison with Table 7a). Our factor loadings all have negative signs, while theirs are all positive, and the ranking of the ROIs is different. The splenium of the corpus callosum had the lowest loading in (Penke et al, 2010), while in our study the uncinate fasciculi have the lowest loadings. Using these principal components, we were not able to predict either general cognitive ability or processing speed. This difference is most likely due to the difference in age between the two samples (~33 years old in our sample vs. ~72 in (Penke et al, 2012; Penke et al, 2010), however there are significant differences in the way the FA values were obtained between our study (atlas derived ROIs in the template space of the TBSS skeleton) and (Penke et al, 2012; Penke et al, 2010) (tractography based results). A final hypothesis that we tried to test with our data derived from a study by Kantarci et al. (Kantarci et al, 2011) who found significant relationships between the FA or the ILF, dorsal and ventral cinguli with memory performance in a large sample (N=220, 71 of which were cognitively intact) of elderly individuals (~79 years old). Our data do not support that the FA of these tracts predicts verbal memory, however their results (Kantarci et al, 2011) were based on a larger sample with greater variance in memory performance due to the inclusion of a large group of individuals with mild cognitive impairment. It is likely that the physiology of the aging brain is substantially different from that of middle aged individuals. The overall reduction in anisotropy that occurs with age could make white matter diffusion metrics much more relevant to directly predicting cognitive function in a fairly diffuse manner, which may not be the case earlier in life. 12 In summary, although we were not generally able to confirm some prior findings in the literature, our exploratory analysis indicates that white matter microstructure as measured with DTI is predictive of cognitive performance at least for card sorting, verbal and working memory domains. More than one FA derived factor (i.e. subregions of the brain) is likely necessary to uncover this relationship in middle age, contrary to old age, where one factor has been shown to be sufficient (Penke et al, 2012; Penke et al, 2010) to predict processing speed and general cognitive performance. Our data do not allow to definitively address the details of how different white matter ROIs should be combined in order to obtain maximal predictability of cognitive performance. Larger sample sizes and replications are needed to inform this choice. FA seems to be more predictive of cognition than RD in our hands. BDNF genotype does appear to play a role in predicting verbal and working memory by interacting with white matter microstructure. This role is complex and could not have been predicted from prior studies, although a similar finding with the same directionality was found by Chiang et al. (Chiang et al, 2011) for object assembly (a subtest that was not available to us) in the splenium of the corpus callosum. 13 SUPPLEMENTARY REFERENCES Cattell RB (1966). The scree test for the number of factors. Multivar Behav Res 1: 245-276. Chiang MC, Barysheva M, McMahon KL, de Zubicaray GI, Johnson K, Montgomery GW, et al (2012). Gene Network Effects on Brain Microstructure and Intellectual Performance Identified in 472 Twins. J Neurosci 32(25): 8732-8745. Chiang MC, Barysheva M, Toga AW, Medland SE, Hansell NK, James MR, et al (2011). BDNF gene effects on brain circuitry replicated in 455 twins. NeuroImage 55(2): 448-454. Costello AB, Osborne JW (2005). Best Practices in Exploratory Factor Analysis: Four Recommendations for Getting the Most From Your Analysis. Practical Assessment, Research and Evaluation 10(7): 1-9. Dickinson D, Goldberg TE, Gold JM, Elvevag B, Weinberger DR (2011). Cognitive factor structure and invariance in people with schizophrenia, their unaffected siblings, and controls. Schizophr Bull 37(6): 1157-1167. Jacobs HI, Leritz EC, Williams VJ, Van Boxtel MP, Elst WV, Jolles J, et al (2011). Association between white matter microstructure, executive functions, and processing speed in older adults: The impact of vascular health. Hum Brain Mapp. Kantarci K, Senjem ML, Avula R, Zhang B, Samikoglu AR, Weigand SD, et al (2011). Diffusion tensor imaging and cognitive function in older adults with no dementia. Neurology 77(1): 26-34. Karlsgodt KH, van Erp TG, Poldrack RA, Bearden CE, Nuechterlein KH, Cannon TD (2008). Diffusion tensor imaging of the superior longitudinal fasciculus and working memory in recentonset schizophrenia. Biol Psychiatry 63(5): 512-518. Mori S, Oishi K, Jiang H, Jiang L, Li X, Akhter K, et al (2008). Stereotaxic white matter atlas based on diffusion tensor imaging in an ICBM template. NeuroImage 40(2): 570-582. Penke L, Maniega SM, Bastin ME, Valdes Hernandez MC, Murray C, Royle NA, et al (2012). Brain white matter tract integrity as a neural foundation for general intelligence. Mol Psychiatry. 14 Penke L, Munoz Maniega S, Murray C, Gow AJ, Hernandez MC, Clayden JD, et al (2010). A general factor of brain white matter integrity predicts information processing speed in healthy older people. J Neurosci 30(22): 7569-7574. Roberts RE, Anderson EJ, Husain M (2010). Expert cognitive control and individual differences associated with frontal and parietal white matter microstructure. J Neurosci 30(50): 1706317067. 15 SUPPLEMENTAL TABLE 1: Demographics, performance and genotype composition for the available neurospcyhological domains VERBAL MEMORY (N=84) VISUAL MEMORY (N=40) PROCESSING SPEED (N=84) N-BACK WORKING MEMORY (N=82) SPAN (N=76) CARD SORTING (N=76) GENERAL COGNITIVE ABILITY (N=82) BDNF genotype Val/Val : Met carriers sex M:F age (mean +/- SD) years of education (mean +/- SD) performance (z score mean +/- SD) 50 : 34 45 : 39 33.7 +/- 9.5 16.5 +/- 2.5 0.059 +/- 0.87 50 : 34 45 : 39 33.7 +/- 9.5 16.5 +/- 2.5 0.015 +/- 0.88 48 : 34 43 : 39 33.6 +/- 9.5 16.5 +/- 2.4 0.13 +/- 0.84 45 : 31 40 : 36 33.7 +/- 9.6 16.5 +/- 2.5 0.008 +/- 0.79 43 : 33 39 : 37 33.9 +/- 9.6 16.5 +/- 2.4 0.089 +/- 0.69 48 : 34 43 : 39 33.3 +/- 9.3 16.5 +/- 2.4 0.073 +/- 0.48 omitted from analysis due to low N 16 SUPPLEMENTARY TABLE 2: Factor loadings of tract estimates for FA after factor analysis (contributing factor loadings are bolded) Tracts of interest Corpus Callosum (genu) Corpus Callosum (body) Corpus Callosum (splenium) Fornix (columns) Superior fronto-occipital fasciculus (right) Factor 1 Factor 2 Factor 3 Factor 4 0.474031 0.866927 0.204244 0.720810 0.437264 0.145686 0.546239 0.377518 0.398595 0.389299 0.497539 -0.175613 0.108273 0.155405 0.162366 -0.095802 0.220221 0.420087 0.050687 0.242593 Superior fronto-occipital fasciculus (left) 0.070015 0.135888 0.433217 0.132936 Internal capsule (anterior limb) Internal capsule (posterior limb) Internal capsule (retrolenticular) Anterior corona radiata Superior corona radiata Posterior corona radiata Posterior thalamic radiation Sagittal stratum-ILF-IFO External Capsule Dorsal cingulum Ventral cingulum Fornix body-Stria Terminalis Superior longitudinal fasciculus Uncinate fasciculus Tapetum 0.165516 -0.201717 -0.098229 0.431961 0.198768 0.055535 0.251759 0.229731 0.325302 0.370354 0.067500 0.370052 -0.067404 0.010986 0.138341 0.546144 0.043377 0.617341 0.562459 0.147000 0.188532 0.682061 0.779479 0.508495 0.354900 0.459553 0.555840 0.547625 0.166893 0.069813 0.189477 0.692114 0.597171 0.244418 0.775027 0.823254 0.340950 0.281091 0.275765 0.451497 0.086868 -0.049104 0.570402 0.130676 0.367790 0.358171 0.413112 0.100227 0.215644 0.189879 0.015706 -0.075663 -0.030196 0.566059 0.340426 0.155263 0.225376 0.242221 0.752925 -0.059283 ILF = Inferior Longitudinal Fasciculus; IFO = Inferior Fronto-Occipital fasciculus 17 SUPPLEMENTAL TABLE 3: Factor loadings of tract estimates for RD after factor analysis (contributing factor loadings are bolded) Tracts of interest Factor 1 Factor 2 Factor 3 Corpus Callosum (genu) Corpus Callosum (body) Corpus Callosum (splenium) Fornix (columns) Internal capsule (anterior limb) 0.468993 0.494637 0.608369 0.058225 0.213091 0.535018 0.304524 0.493367 -0.050485 0.820091 0.363298 0.532952 0.088669 0.849035 -0.036011 Internal capsule (posterior limb) 0.331198 0.615010 -0.448344 Internal capsule (retrolenticular) Anterior corona radiata Superior corona radiata Posterior corona radiata Posterior thalamic radiation Sagittal Stratum-ILF-IFO External capsule Dorsal cingulum Ventral cingulum Fornix body-Stria Terminalis Superior longitudinal fasciculus Superior fronto-occipital fasciculus Uncinate fasciculus Tapetum 0.591401 0.511806 0.681966 0.917795 0.710428 0.613059 0.332801 0.390412 0.214099 0.160231 0.556249 0.652848 0.651463 0.589345 0.309371 0.421355 0.606843 0.808338 0.723173 0.640468 0.648441 0.674727 -0.175010 0.359966 0.051041 -0.027603 0.231541 0.202342 0.185866 0.246983 -0.016744 0.266644 -0.077087 0.432631 0.581726 0.328856 0.100376 0.695552 0.729273 0.008273 -0.088306 0.101563 18 SUPPLEMENTARY TABLE 4a: Results of overall regression predicting cognitive performance based on BDNF genotype, age sex and four factors based on FA measures. FA verbal memory processing speed working memory span card sort G Multiple R 0.45 0.30 0.49 0.42 0.75 0.25 Multiple R² 0.20 0.09 0.25 0.18 0.56 0.06 Adjusted R² 0.15 0.07 0.19 0.09 0.36 0.05 df Model 5 2 5 7 14 1 df Res F p 77 79 75 67 31 78 3.82 3.97 4.87 2.09 2.80 5.15 0.0038 0.0227 0.0007 0.0565 0.0083 0.0260 19 SUPPLEMENTARY TABLE 4b: Results for univariate tests of significance (backward stepwise solution) for verbal memory performance and factors derived from FA. Verbal Memory Intercept BDNF Genotype Factor2 Factor3 BDNF Genotype*Factor2 Factor2*Factor3 Error Df 1 1 1 1 1 1 77 F 882.1058 0.0089 0.0824 4.2632 13.0424 3.2566 p 0.000000 0.924927 0.774803 0.042314 0.000540 0.075045 SUPPLEMENTARY TABLE 4c: Results for univariate tests of significance (backward stepwise solution) for working memory performance and factors derived from FA. Working Memory Intercept BDNF Genotype Factor1 Factor2 Factor3 BDNF Genotype*Factor2 Error Df 1 1 1 1 1 1 75 F 1119.806 0.459 7.012 0.205 4.488 14.477 p 0.000000 0.500315 0.009866 0.652367 0.037438 0.000287 SUPPLEMENTARY TABLE 4d: Results for univariate tests of significance (backward stepwise solution) for card sort performance and factors derived from FA. Card Sorting Intercept AGE Factor2 Error Df 1 1 1 73 F 125.6947 12.6488 4.9246 p 0.000000 0.000664 0.029584 20 SUPPLEMENTARY TABLE 5a: Results of overall regression predicting cognitive performance as a function of BDNF genotype, age sex and three factors based on RD measures. RD verbal memory processing speed working memory span card sort G Multiple R 0.38 0.36 0.51 0.36 0.55 0.17 Multiple R² 0.15 0.13 0.26 0.13 0.30 0.03 Adjusted R² 0.09 0.08 0.17 0.08 0.23 -0.01 df Model 5 4 9 4 7 3 df Res 78 79 72 71 68 78 F 2.68 2.86 2.88 2.57 4.11 0.77 p 0.0274 0.0288 0.0059 0.0448 0.0008 0.5167 21 SUPPLEMENTARY TABLE 5b: Results for univariate tests of significance (backward stepwise solution) for working memory performance and factors derived from RD. N-Back Working Memory Intercept SEX BDNF Genotype AGE Factor1 Factor2 Factor3 BDNF Genotype*AGE SEX*Factor2 Factor1*Factor3 Error Df F p 1 1 1 1 1 1 1 1 1 1 72 57.91 1.38 2.82 0.00 0.00 0.56 6.74 3.57 4.38 8.33 0.000 0.244 0.097 0.972 0.946 0.456 0.011 0.063 0.040 0.005 SUPPLEMENTARY TABLE 5c: Results for univariate tests of significance (backward stepwise solution) for card sort performance and factors derived from RD. Card Sort Intercept SEX AGE Factor1 Factor2 Factor3 SEX*Factor2 Factor1*Factor3 Error Df 1 1 1 1 1 1 1 1 68 F 104.22 0.60 10.43 0.90 3.04 0.70 3.48 4.29 p 0.0000 0.4424 0.0019 0.3452 0.0858 0.4072 0.0666 0.0420 22 Supplementary Table 6. Prediction of working memory performance by FA of the bilateral SLF and BDNF. Working Memory Intercept SEX BDNF Genotype AGE SLF R FA SLF L FA BDNF Genotype*AGE SEX*SLF R FA AGE*SLF R FA AGE*SLF L FA SLF R FA*SLF L FA Error Df 1 1 1 1 1 1 1 1 1 1 1 71 F 5.40 8.71 2.03 0.09 6.65 2.73 3.28 8.95 5.39 3.34 4.73 p 0.023 0.004 0.159 0.759 0.012 0.103 0.075 0.004 0.023 0.072 0.033 23 Supplementary Table 7a. Factor loadings for the first principal component based on the average FA of eight white matter tracts. ROI Genu CC FA Splenium CC FA Dors. Cing. R FA Dors. Cing. L FA SLF R FA SLF L FA UF R FA UF L FA factor loadings -0.721110 -0.796593 -0.813736 -0.854284 -0.834219 -0.815361 -0.483780 -0.466819 Supplementary Table 7b. Factor loadings for the first principal component based on the average FA of twelve white matter tracts. ROI Factor loadings Genu CC FA Splenium CC FA ALIC R FA ALIC L FA SS-ILF-IFO R FA SS-ILF-IFO L FA Dors. Cing. R FA Dors. Cing. L FA SLF R FA SLF L FA UF R FA UF L FA -0.305799 -0.331982 -0.268846 -0.282537 -0.284854 -0.277110 -0.309260 -0.334474 -0.326140 -0.328967 -0.179487 -0.178663 Abbreviations: ROI=region of interest, FA=fractional anisotropy, CC=corpus callosum, Dors.= dorsal, Cing.=cingulum, SLF=superior longitudinal fasciculus, UF=uncinate fasciculus, ALIC= Anterior limb of the internal capsule; SS-ILF-IFO=sagittal stratum-inferior longitudinal fasciculus-inferior fronto-occipital fasciculus. 24 Supplementary Figure 1. Interaction of BDNF genotype by FA factor 2 in predicting verbal memory performance. BDNF Genotype::Trans_Factor1_VerbalMemory: 1 Trans_Factor1_VerbalMemoryr 2==26.8988+2.7714*x BDNF Genotype: 1 FACTOR2 0.0943; r = 0.3071, p = 0.0301 BDNF Genotype: 2 Trans_Factor1_VerbalMemoryr 2==25.6753-3.3015*x BDNF Genotype: 2 FACTOR2 :Trans_Factor1_VerbalMemory: 0.1363; r = -0.3691, p = 0.0317 45 40 Verbal Memory Performance 35 30 25 20 15 10 5 0 -3 -2 -1 0 1 2 3 -3 -2 BDNF Val/Val -1 0 1 2 3 BDNF Met carriers FA Factor2 BDNF Genotype 1= Val/Val, Genotype 2 = Met carriers 25 Supplementary Figure 2. Interaction of BDNF genotype by FA factor 2 in predicting nback working memory performance. BDNF1 Genotype: Trans_Factor2_Nback =r226.7881+3.3461*x BDNF Genotype: FACTOR21:Trans_Factor2_Nback: = 0.1481; r = 0.3848, p = 0.0069 BDNF Genotype: 2 Trans_Factor2_Nback =r227.6547-2.8962*x BDNF Genotype: 2 FACTOR2 :Trans_Factor2_Nback: = 0.1331; r = -0.3648, p = 0.0339 45 40 Working Memory Performance 35 30 25 20 15 10 5 0 -3 -2 -1 0 1 2 3 -3 BDNF Val/Val -2 -1 0 1 2 3 BDNF Met Carriers FA Factor 2 BDNF Genotype 1= Val/Val, Genotype 2 = Met carriers 26 Supplementary Figure 3. FA Factor 2 predicts Card Sort performance Trans_Factor5_CardSorting =r2138.4656+15.0822*x = 0.0971; r = 0.3117, p = 0.0061 FACTOR2 :Trans_Factor5_CardSorting: 240 220 200 Card Sort Performance 180 160 140 120 100 80 60 40 20 0 -3 -2 -1 0 1 2 3 FA Factor 2 27 Supplementary Figure 4. Relationship of RD-based Factor 3 and n-back working memory performance. Trans_Factor2_Nback = 27.032-1.985*x FACTOR3 :Trans_Factor2_Nback: r2 = 0.0535; r = -0.2314, p = 0.0365 45 Working Memory Performance 40 35 30 25 20 15 10 5 0 -3 -2 -1 0 1 2 3 4 RD Factor 3 28 Supplementary Figure 5. Interaction of right SLF FA by sex in predicting n-back working memory performance. 1 Trans_Factor2_Nback SEX: 1 SLF R SEX: FA:Trans_Factor2_Nback: r2 == 4.4523+47.5289*x 0.0136; r = 0.1166, p = 0.4566 SEX: 2 Trans_Factor2_Nback SEX: 2 SLF R FA:Trans_Factor2_Nback: r2==73.5037-97.1013*x 0.1426; r = -0.3777, p = 0.0178 45 Working Memory Performance 40 35 30 25 20 15 10 5 0 0.42 0.44 0.46 0.48 0.50 0.52 0.54 0.56 0.42 0.44 0.46 0.48 0.50 0.52 0.54 0.56 Males Females SLF R FA SEX: 1= Males; SEX: 2= Females 29
© Copyright 2026 Paperzz