Participant Characteristics
A total of 88 subjects (Mage=9.36, SDage=2.61; 44 male) were included in analysis (Table 1). The flow of participants through the study are shown in Fig. 1. There were 25 participants in the NF1 group (Mage=9.20, SDage=2.27; 15 male); 31 in the NS group (Mage=10.0, SDage=3.22; 12 male); and 32 in the TD group (Mage=8.88, SDage=2.13; 15 male). Details on participant mutations are in the Supplement. The groups did not differ significantly by age (p = .090), sex (p = .259), or pubertal stage (p > .05) (Supplementary Table 1). Significantly lower scores were found for clinical groups compared to TD on all IQ measures (p < .010). Results of the sensitivity analysis for T2-hyperintensities are shown in Supplementary Table 2.
Table 1
Demographics and descriptive statistics of included participants.
| All (n = 88) | TD (n = 32) | NF1 (n = 25) | NS (n = 31) |
Age | 9.36 (2.61) | 8.88 (2.13) | 9.20 (2.27) | 10.0 (3.22) |
Sex (M/F) | 44/44 | 17/15 | 15/10 | 12/19 |
FSIQ | 104 (15.4) | 115 (12.9) | 96.5 (13.3) | 98.5 (13.2) |
VIQ | 106 (16.5) | 116 (13.1) | 101 (17.5) | 102 (15.0) |
PIQ | 101 (18.8) | 110 (17.0) | 92.5 (11.5) | 98.7 (21.6) |
Tanner 1- Stage 1/2/3/4 | 43/19/14/6 | 15/11/4/1 | 9/6/5/3 | 19/2/5/2 |
Tanner 2- Stage 1/2/3/4 | 58/10/11/4 | 20/7/3/1 | 15/3/5/0 | 23/0/3/3 |
Stimulant | 15 | 0 | 9 | 6 |
Growth hormone | 8 | 0 | 0 | 8 |
Data presented as counts or mean (SD). Tanner 1 refers to Tanner Pubic Hair Scale. Tanner 2 refers to Female Breast Development/Male External Genitalia Scale. Tanner Scales exclude 1 TD, 2 NF1, 2 NS. |
ANOVA = analysis of variance; FSIQ = Full-Scale Intelligence Quotient; NF1 = neurofibromatosis type 1; NS = Noonan syndrome; PIQ = Performance Intelligence Quotient; TD = typical developing; VIQ = Verbal Intelligence Quotient. |
ROI analysis of NDI, ODI, and MK
In subcortical regions, the NDI results showed a trend of NF1 < NS < TD (Table 2). Lower NDI values were found in NF1 and in NS compared to TD (Fig. 2, Supplementary Fig. 3). NDI was lower in NF1 compared to NS in the amygdala, hippocampus, pallidum, and thalamus (all p < .001), with the largest effect observed in the thalamus (estimate − 0.044 [95% CI: -0.053, -0.034], d=-2.36). Overall, NDI detected the greatest number of differences between groups and therefore showed the most sensitivity of the 3 diffusion measures investigated.
Table 2
Confidence intervals from analysis of subcortical regions.
| | TD – NF1 | | TD – NS | | NF1 – NS | |
| ROI | Estimate (95% CI) | p | d | Estimate (95% CI) | p | d | Estimate (95% CI) | p | d |
NDI | Amygdala | 0.040 (0.028, 0.051) | < .001 | 1.97 | 0.009 (-0.001, 0.020) | .105 | 0.45 | -0.030 (-0.042, -0.019) | < .001 | -1.66 |
Caudate | 0.025 (0.009, 0.040) | .001 | 1.00 | 0.012 (-0.003, 0.026) | .134 | 0.22 | -0.013 (-0.028, 0.003) | .134 | -0.61 |
Hippocampus | 0.061 (0.050, 0.072) | < .001 | 3.34 | 0.012 (0.002, 0.023) | .015 | 0.59 | -0.048 (-0.060, -0.037) | < .001 | -2.34 |
Pallidum | 0.098 (0.064, 0.132) | < .001 | 1.41 | 0.034 (0.002, 0.066) | .035 | 0.27 | -0.064 (-0.098, -0.030) | < .001 | -0.98 |
Putamen | 0.007 (-0.004, 0.018) | .315 | 0.37 | 0.014 (0.003, 0.025) | .007 | 0.50 | 0.007 (-0.005, 0.018) | .322 | 0.08 |
Thalamus | 0.060 (0.050, 0.070) | < .001 | 3.38 | 0.016 (0.007, 0.025) | < .001 | 1.07 | -0.044 (-0.053, -0.034) | < .001 | -2.36 |
ODI | Amygdala | -0.000 (-0.021, 0.021) | .999 | 0.51 | 0.005 (-0.014, 0.024) | .812 | 0.15 | 0.005 (-0.016, 0.026) | .825 | 0.09 |
Caudate | 0.010 (-0.009, 0.029) | .409 | 0.49 | -0.010 (-0.028, 0.007) | .345 | -0.45 | -0.021 (-0.040, -0.002) | .030 | -0.77 |
Hippocampus | 0.012 (-0.007, 0.031) | .287 | 0.66 | 0.001 (-0.017, 0.018) | .997 | 0.03 | -0.011 (-0.031, 0.008) | .326 | -0.43 |
Pallidum | 0.004 (-0.022, 0.031) | .917 | 0.15 | 0.008 (-0.017, 0.033) | .746 | 0.14 | 0.003 (-0.024, 0.030) | .955 | 0.02 |
Putamen | 0.001 (-0.015, 0.016) | .996 | 0.30 | -0.015 (-0.029, -0.000) | .043 | -0.51 | -0.015 (-0.031, 0.000) | .051 | -0.68 |
Thalamus | 0.012 (0.004, 0.021) | .002 | 1.07 | -0.006 (-0.014, 0.002) | .193 | -0.42 | -0.018 (-0.026, -0.010) | < .001 | -1.39 |
MK | Amygdala | 0.014 (-0.012, 0.041) | .400 | 0.45 | -0.008 (-0.033, 0.017) | .703 | -0.30 | -0.023 (-0.050, 0.004) | .108 | -0.71 |
Caudate | -0.008 (-0.061 0.046) | .938 | -0.05 | 0.005 (-0.045, 0.055) | .971 | 0.09 | 0.012 (-0.041, 0.066) | .845 | 0.13 |
Hippocampus | 0.065 (0.044, 0.086) | < .001 | 1.87 | 0.013 (-0.006, 0.033) | .244 | 0.22 | -0.051 (-0.072, -0.030) | < .001 | -1.37 |
Pallidum | 0.136 (0.075, 0.198) | < .001 | 1.07 | 0.041 (-0.017, 0.100) | .209 | 0.05 | -0.095 (-0.157, -0.033) | .001 | -0.84 |
Putamen | 0.037 (-0.002, 0.076) | .064 | 0.57 | -0.003 (-0.040, 0.033) | .971 | -0.21 | -0.041 (-0.080, -0.017) | .039 | -0.80 |
Thalamus | 0.075 (0.051, 0.098) | < .001 | 1.86 | 0.026 (0.004, 0.047) | .016 | 0.53 | -0.049 (-0.072, -0.025) | < .001 | -1.39 |
CI = Tukey-Kramer corrected and least squares means adjusted confidence intervals; d = Cohen’s d effect size; L = left; MK = mean kurtosis; NDI = neurite density index; ODI = orientation dispersion index; p = Tukey-Kramer corrected p-value from pairwise comparison between two groups; R = right; ROI = region-of-interest; SE = standard error. |
ODI demonstrated the least sensitivity to changes of the three measures investigated, and the results did not show a clear pattern across the groups. We found evidence of decreased ODI in NF1 compared to NS in the caudate (estimate − 0.021 [95% CI: -0.040, -0.002], p = .030) and thalamus (estimate − 0.018 [95% CI: -0.026, -0.010], d=-1.39) p < .001).
The pattern of MK results was like NDI where NF1 < NS < TD. Lower MK was found in NF1 compared to TD in the hippocampus, pallidum, and thalamus (all p < .001). When comparing TD and NS, lower MK was found in NS compared to TD in the thalamus only (p = .016). MK was lower in NF1 compared to NS in the hippocampus (p < .001), pallidum (p = .001), putamen (p = .039), and thalamus (p < .001).
Tract-based analysis of NDI, ODI, and MK
The number of participants included in analysis for each tract are shown in Supplementary Table 3. In each of the 39 white matter tracts, we found lower NDI values in NF1 and NS compared to TD (all p < .001) (Supplementary Figs. 4 and 5). However, only two tracts showed evidence of differences between the clinical groups. Lower NDI was found in NF1 compared to NS in the parietal body of the corpus callosum (estimate − 0.018 [95% CI: -0.036, -0.000], p = .046) and in the middle cerebellar peduncle (estimate − 0.038 [95% CI: -0.056, -0.021], p < .001). Lower ODI was found in NF1 compared to TD in nine of the tracts; in TD compared to NS in three tracts; and in NF1 compared to NS in 18 tracts. Like NDI, MK was lower in NS compared to TD in all 39 tracts (p < .01). MK was also lower in NF1 compared to TD in all tracts except the rostrum of the corpus callosum and the left extreme capsule. The only difference between the clinical groups was found in the middle cerebellar peduncle where MK was lower in NF1 compared to NS (estimate − 0.057 [95% CI: -0.089, -0.026], p < .001).
Given the similar findings in NDI and MK, we tested the associations between these parameters using Pearson correlations. The results suggest the parameters are highly positively correlated, particularly in the white matter tracts. For example, MK and NDI were positively correlated in the middle cerebellar peduncle (R = 0.81, p < .001) (Supplementary Fig. 6).
Multivariate analysis
To complement the univariate results, values for each white matter tract were entered into a principal components analysis. The first component alone represented 49.0% of the variance with similar contributions from each tract (Fig. 3A-D). The second component represented 11.6% of the variance (cumulative variance of 60.6%) again with similar contributions from each tract. The eigenvalues for the first 10 principal components are detailed in Supplementary Table 4. A biplot representation of the loadings in PC1 and PC2 suggest some discrimination between TD and the clinical groups (NF1, NS). Subjects in the clinical groups generally have lower values on PC1 compared to subjects in the TD group. Given that NDI and MK values contribute the most to PC1, we can infer that lower values of NDI and MK are generally found in clinical groups compared to TD. Upon visual inspection of the biplot with respect to PC2, subjects in the clinical and TD groups appear to have similar loadings. ODI values contributed the most to PC2, suggesting the groups may have similar ODI values. These inferences consolidate our pattern of univariate results.
As the univariate results showed several differences between clinical groups in the subcortical regions, we tested the ability of the subcortical diffusion values to classify the subjects into groups using linear discriminant analysis. The discriminant functions classified participants with NF1, TD, and NS with 98%, 85% and 84% accuracy respectively (Supplementary Table 5). Two linear discriminant functions were generated, LD1 and LD2 (Supplementary Table 6). For LD1, NDI in the thalamus contributed the most with a weighting of -1.92. Following the leave-one-out cross-validation, the accuracy of the model in classifying NF1 remained high at 97.2% though there were drops in accuracy for TD and NS classifications to 71% and 70% respectively. Figure 4A-D shows that the NF1 observations can be separated from TD and NS observations, however TD and NS observations show overlap.