Descriptive statistics of the sample are presented in . The data were analyzed in a series of general linear models (GLMs). In each model, age (centered at the sample mean) served as a continuous independent variable; sex was a categorical independent variable, and regional FA and ADC were dependent variables with ROI as a within-subjects factor. To minimize rounding error and to simplify reporting, all ADC values (mm2/sec) were multiplied by 103. Full models including all the interactions were tested, but to maximize statistical power, all nonsignificant interactions with between-subject independent variables (p > .10) were removed from the models. Within-subjects interactions were adjusted by Huynh-Feldt factor to correct for violation of sphericity assumption.
Sample demographic information by sex, and hypertension status: mean ± SD
2.1 Age effects on white matter microstructure
Because all ROIs except corpus callosum were measured bilaterally, left and right hemisphere measures were compared. We observed minor, though significant laterality differences only in parietal FA (.44 on the left vs. 46 on the right, t(76) = 2.66, p = .01), prefrontal FA (.77 on the left and.78 right, t(76) = 2.72, p = .008), and temporal ADC (.82 left vs .81 right, t(76) = 7.65, p < .001). In light of virtual absence of lateral differences and lack of pertinent a priori hypotheses regarding asymmetry the measures were averaged across the hemispheres. Descriptive statistics for regional FA and ADC as well as their zero-order correlations with age are displayed in . As evident from that table, FA ranged from .45 in the frontal and parietal white matter regions (least directionally constrained) to .80 in the splenium of the corpus callosum (most directionally constrained). ADC had a narrower range of values from .73 in the posterior internal capsule (lesser diffusivity) to .83 in the occipital white matter (most diffusivity). FA displayed a significant negative correlation with age in all measured regions except the anterior limb and genu of the internal capsule. FA and ADC by region are compared graphically in .
Descriptive statistics for DTI coefficients by region of interest.
Mean diffusion properties by region. a) fractional anisotropy, b) apparent diffusion coefficient. Error bars represent standard error of the mean. Note the differences in regional variance are greater in FA than in ADC.
Fractional Anisotropy and Age
The Age × Sex interaction was nonsignificant and, therefore, removed from the model. In the reduced model, there was a significant main effect of Age, F(1, 74) = 38.15, p < .001, indicating age-related reduction in FA across the examined brain ROIs. However, the magnitude of the age effect on FA differed across the regions as shown by a significant ROI × Age interaction, F(8, 592) = 9.08, p < .001. Simple effects analyses through univariate regressions () revealed significant reduction of FA with age in all regions except the anterior and middle limbs of the internal capsule. The strongest age effects were in the occipital and frontal white matter and the splenium, where age alone accounted for 33%, 31%, and 27% of the variance in fractional anisotropy, respectively in those regions (see ). There was neither main effect of Sex (F < 1) nor ROI × Sex interaction (F = 1.49, ns). A significant main effect of ROI was noted: F(8, 592) = 728.11, p < .001 indicating that FA varied across the assessed regions.
Follow-up univariate analyses for the effects of age on regional FA
Apparent Diffusion Coefficient and Age
The effects of age on diffusivity were examined in the same manner as above with ADC ROI as a repeated measure. The main effect of Age on ADC was significant: F(1, 73) = 21.65, p < .001, but it was qualified by a significant ROI × Age interaction: F(8, 584) = 2.60, p < .05. The significant effects indicated that ADC increased with age but the magnitude of age differences therein varied across the ROIs. To decompose the interaction we performed univariate tests for each region (see for a summary). There was a significant increase in diffusivity with age in each measured location, except for a nonsignificant trend for the posterior limb of the internal capsule (p = .07). The strongest effect of age on diffusivity was observed in the splenium of the corpus callosum, where age accounted for 22% of the variance. See for slopes and R2 for ADC for each region. There was a significant main effect of ROI, F(8, 584) = 56.67, p < .001, indicating that diffusivity varied across the examined locations. Neither main effect of Sex (F < 1, ns), nor Sex × Age interaction, F(1, 73) = 2.79, p = .10, nor other within-subjects interactions with Sex (Fs < 1, ns) reached significance.
Follow-up univariate analyses for the effects of age on regional ADC
As expected, there was an inverse relation between FA and ADC in every region: Greater diffusivity was associated with reduced fractional anisotropy. Scatter plots of the relation between age and FA and age and ADC by region are displayed in .
Figure 3 Scatter plots and regression line of the association between age and regional fractional anisotropy (left panels) and apparent diffusion coefficient (right panels). The slopes of the regressions and the proportion variance explained for each panel is (more ...)
2.2 Nonlinearity of relation between white matter integrity and age
As evident in the regression plots in and confirmed by analyses summarized in , some regions exhibited a nonlinear trajectory of aging. Specifically, for most regions ADC evidenced significant age-related acceleration that was the strongest in the limbs of the internal capsule and the corpus callosum genu (p < .001; see for all regions). Inspection of the age-ADC scatter plots () indicates that age-related acceleration of diffusivity increase begins approximately at the end of the fifth decade.
Test of nonlinearity of effects of age on regional white matter ROIs across the age span for FA and ADC.
Fractional anisotropy, in contrast, evidences linear declines with age for all regions, except for the prefrontal white matter, which displayed a quadratic (decelerating) relation with age. Note, however, that a linear decline in anisotropy in that region was quite substantial, with the trend accounting for 31% of the age effect on FA. ADC in prefrontal and parietal white matter evidenced a trend toward nonlinearity that did not reach conventional levels of significance (F (1, 74) = 3.32 and 3.20, p = .07 and p = .08 respectively). Across the ROIs, the proportion of ADC variance explained by the combination of linear and quadratic age components ranged from 16% to 30%.
2.3 Effects of Hypertension on White Matter Microstructure
To examine the effect of hypertension on regional white matter, we analyzed a subsample of 64 adults above age 42 (the age of the youngest participant with hypertension). In the model, the nine FA ROIs served as a dependent repeated measures variable, and age, sex, hypertension status and their interactions as predictors. Main effects of age (F(1,56) = 11.64, p = .001) and ROI (F(8,448) = 407.83, p < .001) were observed. They were modified by a significant interaction: ROI × Age × Hypertension (F(8, 448) = 2.44, p =.029) and a trend for ROI × Age × Sex × Hypertension interaction (F(8, 448) = 2.09, p = .059). Univariate post-hoc analyses () revealed that Hypertension × Age interaction was significant only for the occipital and temporal white matter FA: F(1,56) = 6.87, p = .01 and F(1,56) = 5.60, p = .02, with a trend observed for the genu of the internal capsule: F(1,56) = 3.38, p = .07. The ROI × Age × Sex × Hypertension trend was due to stronger differential effects of age and hypertension on FA among women (F(8,264) = 3.18, p = .008) than men (F(8,264) = 1.53, p = .17).
Effects of hypertension on regional fractional anisotropy: Selective effects on posterior white matter (n = 64).
In the second model, regional ADC was the multivariate dependent variable. Again, we found main effects of age (F(1,57) = 24.50, p < .001), and ROI (F(8,456) = 37.43, p < .001), as well as Age × ROI interaction (F(8,456) = 2.78, p = .012). Neither main effect of hypertension (F(1,57) = 2.28, p = .14), nor ROI × hypertension interaction was found (F < 1). However, a significant Age × Sex × Hypertension interaction was noted: F(1,57) = 4.29, p = .043. Decomposition of that interaction revealed that it was due to the lack of association between ADC and age among hypertensive men (n = 9). Correlations of regional ADC with age ranged between r = −.02 for occipital white matter to r = .35 for the splenium, all ns. In contrast, correlations between ADC (averaged across ROIs) and age were r = .60, p = .015 for hypertensive women, r = .53, p = .014 for normotensive women, and r = .77, p < .001 for normotensive men.
Effects of Duration of Hypertension on FA and ADC
We examined the effects of duration of hypertension on regional FA and ADC in the subsample of 25 hypertensive adults (aged 42–84). Age and duration of hypertensive condition did not correlate (r =.09, ns), and thus both were used along with sex as predictors of regional FA or ADC. We found a trend for main effect of duration (F(1, 20) = 3.92, p =.06) on FA, and a significant age × duration interaction: F(1, 20) = 7.80, p = .001, with no differences among the ROIs. This finding indicates the age-related reduction in FA was exacerbated in persons with longer duration of hypertension. The results for regional ADC were similar: a significant age × duration interaction, F(1, 20) = 4.37, p < .05, suggesting that more prolonged hypertension was linked to a greater age-related increase in diffusivity of the white matter.
2.4. Effects of Pulse Pressure on White Matter Microstructure
In a sub-sample of normotensive individuals (N = 52), we examined the effect of a vascular risk factor derived from a normal measured blood pressure on regional white matter microstructure. Pulse pressure (the difference between systolic and diastolic pressure) served as a predictor of FA or ADC in general linear models that included age, sex, and the interactions among them. In this analysis, we observed a main effect of age (F(1,46) = 23.84, p < .001), and also a significant main effect of pulse pressure on FA: F(1,46) = 4.07, p = .05. Higher pulse pressure was associated with lower FA beyond the effects of calendar age. There was no pulse pressure by region interaction, indicating that this effect was global rather than region specific. As in the full-sample analyses, there was a main effect of ROI (F(8,368) = 331.84, p < .001) and a ROI × age interaction (F(8,368) = 5.19, p < .001).
A similar model with ADC as a dependent variable revealed a main effect of age, F(1,45) = 10.85, p = .002, and a sex × age interaction, F(1,45) = 7.33, p < .01. There was also a main effect of ROI, F(8,360) = 31.25, p < .001, and a ROI × pulse pressure interaction, F(8,360) = 2.48, p = .02. These effects, however, were qualified by a ROI × age × pulse pressure interaction, F(8,360) = 2.11, p < .05, which indicated that the effect of pulse pressure on regional ADC depended on age. To decompose this interaction, follow-up correlations between pulse pressure and regional ADC were computed for each age group (young ≤ 35, middle = 36-64, old ≥ 65 years). We found a significant effect of pulse pressure on the prefrontal white matter (r = −.50, p = .05, n = 24) in the older adults, on the anterior limb of the internal capsule for middle aged adults (r = .42, p = .04, n = 16), and no region reached significance for the young, although there was a trend for the splenium (r = −.54, p = .07, n = 12). Of all ROIs, regardless of age, higher pulse pressure was selectively associated with increased ADC in the genu of the corpus callosum: t(50) = 2.30, p = .025.
2.5. The influence of regional WMH burden on diffusion-based measures
In light of recent reports of the influence exerted by leukoaraiosis on diffusion-based indices of white matter integrity, we conducted a subsidiary analysis of WMH effect on FA and ADC. The manually measured regional volumes of WMH were available for the older part of the sample (n = 33, age 52–81, mean age 67.00 ± 7.15 years). In contrast to the full-range sample, in this sub-sample of a restricted age range, we found no age differences in FA in any region. No age differences in ADC were noted in frontal, parietal, or occipital subcortical white matter, or in the anterior limb of the internal capsule.
Lobar volumes of WMH were log-transformed to reduce the skew and were centered at their sample mean. There were also no age differences in WMH volumes in any of the examined lobes: F(1,29) = 1.88, p = .18 for the frontal, F(1,29) = 2.35, p = .14 for the occipital, and F < 1 for the parietal lobes. Temporal WMH were absent in almost half of the sub-sample (n = 15) and therefore were not analyzed.
The lobar WMH volumes were introduced into the models assessing the effects of age, sex, and hypertension on regional diffusion properties. Only WMH volumes from the lobes that corresponded to the ROIs were used in the analyses. The model for anterior regions (frontal subcortical white, anterior limb and genu of IC) contained frontal WMH as a covariate. The parietal WMH volume served a covariate in the models for parietal and posterior IC FA and ADC. Occipital WMH served as a covariate in the analyses of the occipital subcortical and posterior IC measures. No similar analyses were conducted for the corpus callosum as there were no WMH there.
No effects of age on FA were observed in the models that included WMH volumes along with sex and hypertension status. The only instance of a significant influence of WMH volume on FA was in the IC genu, where frontal WMH had a significant effect in the absence of age differences: F(1,28) = 7.77, p = .009. In contrast, several significant effects of WMH on ADC and age differences therein were noted. A significant age difference in ADC was observed in the IC genu (F(1,29) = 4.83, p = .04), but was attenuated to a trend (F(1,28) = 3.04, p = .09) after accounting for the frontal WMH volume. On the other hand, age-related increase in ADC in the posterior limb of IC was significant (F(1,29) = 11.33, p = .002), and was only slightly attenuated by covarying parietal (F(1,28) = 9.82, p = .004) or occipital (F(1,28) = 8.18, p = .008) WMH volumes. Independent effects of WMH burden on ADC were not widespread, although the effect of frontal WMH volume on the ADC of IC genu was marginally significant: F(1,28) = 4.02, p = .054, and higher occipital ADC was associated with larger occipital WMH volume: F(1,28) = 9.67, p = .004.