We found that loss of white matter integrity was a significant cause of age-related cognitive slowing among otherwise cognitively normal older adults. Age-related reduction of cognitive processing speed was separate from other significant age-related processes that we observed in the same subjects, including gray and white matter atrophy, the accumulation of white matter lesions, and a decline in verbal episodic memory capacity. Thus slowing of cognitive processing speed with age emerged as a distinct phenomenon mediated by changes in DTI measures of white matter integrity.
Although we are not the first to show that FA and other DTI metrics decline with age () 
and correlate with reaction time 
, our study was large (N
131) and methodologically rigorous in the way that cognitive processing speed was tested. Specifically, we designed tasks that minimized the contribution of motor speed to the overall response latency interval. When measuring the time required to press a button after a simple perceptual task, or when timing performance on a task that requires handwriting or other complex motor activities, motor performance has a significant potential to influence reaction time. Our tasks required cognitively complex judgments but were motorically simple. As a further refinement, we averaged the performance on multiple tasks to extract a single metric of cognitive processing speed for each participant. Finally, because our tasks were computerized, they were not examiner or site dependent, and were thus amenable to easy reproducibility. On the other hand, it is important to note that we did not perform a direct comparison between our method and other common methods of processing speed evaluation (e.g., the pegboard test or Digit Symbol test), and we cannot conclude that those approaches are invalid or would give different results. It is not possible to separate, with absolute precision, cognitive processing speed from motor ability or other cognitive attributes. For instance, our tasks still required a button press. In addition, our tasks relied heavily on visuospatial processing, and slow reaction time in some participants could conceivably have resulted from subclinical dysfunction of the visual processing stream, such as very early neurodegeneration in the right parietal lobe, although we note that there was no evidence of gray matter atrophy there or elsewhere that correlated with task performance (). In sum, we observed a very tight correlation between performance on our computerized tasks and FA, but it is nevertheless unlikely that white matter integrity is the sole factor accounting for the observed variance in our data.
Using an unbiased, data-driven approach with TBSS, we found that a significant relationship between DTI metrics and cognitive processing speed emerged across a broad swath of cerebral white matter, including especially the genu and body of the corpus callosum and areas of the frontal lobes (, and ). Age-related changes to these same areas have been described by others 
. Our data thus support a general model in which the white matter tracts that are last to myelinate are among the earliest to deteriorate with age, a hypothesis recently tested in another TBSS study 
. Interestingly, cognitive processing speed exhibits a similar relationship with age, improving over the first two decades of life and declining from middle age onward 
. The convergent developmental timelines of white matter integrity and processing speed corroborate the important relationship between the two. The anatomical patterns of age-related and speed-related variability in FA were similar, except for the SLF and ILF (): Whereas many voxels within the SLF related speed to FA, the integrity of this ROI did not vary substantially with age. The significance of this finding is not evident from our study, and one direction for future work could be to investigate the specific determinants of SLF integrity among older adults. By contrast, the integrity of the ILF appeared to vary substantially with age but not with processing speed (), suggesting that our tasks were not dependent upon this particular ROI. Although individual tasks within our computerized processing speed battery contributed differentially to the relationship between response latency and FA, the overall distribution of involved white matter was not clearly distinct between tasks ().
How changes in DTI metrics and changes in cognitive processing speed correspond to histopathological features of white matter is not known. Some have used DR as a proxy for myelin integrity, as dysmyelination might be expected to enhance water diffusion across axons 
; using similar logic, DA has been used as a proxy for axonal integrity 
. In our study, reduced cognitive processing speed was associated with increased DR but not with any change in DA, suggesting dysmyelination as a possible contributing factor. This finding is in agreement with one past study 
, but contrasts with another recent report showing that DA but not other diffusion indices correlated with processing speed, and that this association was prominent in posterior brain regions but not in the frontal lobes 
. This latter study differed from ours in the method for measuring processing speed: Whereas we used spatial judgment tasks that did not rely upon memory, they used a computerized N-back test that relied on working memory, as well as a non-memory, non-computerized letter or pattern matching task. While we do not know the reason why their findings diverged from ours, we speculate that the inclusion of a memory-dependent test sampled more than simple cognitive processing speed. Early, preclinical Alzheimer’s disease, which is inevitably present in any large sample of older subjects, is an example of a process that could result in secondary axonal damage and poor memory.
Another prior report emphasized the importance of white matter lesion volume and white matter atrophy in causing apparent changes in FA with age 
. However, we observed no correlation between these factors and cognitive processing speed (), perhaps because of the very minimal extent of visible white matter lesions among our subjects. Others have suggested that in the process of age-related white matter deterioration, decreases in FA may precede and be more sensitive than volume loss 
, and so it is possible that subjects with lower FA values and normal-appearing white matter will go on to accumulate visible white matter lesions, a consideration that should factor into any longitudinal follow-up. We measured white matter lesion volume using a T1 sequence, which is less sensitive than FLAIR; we may therefore have underestimated the extent of subcortical ischemic disease. On the other hand, while we did not have FLAIR images on all of our participants, we did have this sequence on the individual who, of all our participants, had the greatest white matter lesion load; that FLAIR image is illustrated in and reveals only mild disease.
By including age as a confound regressor in voxel-wise regressions of FA with cognitive processing speed (), we show an age-independent effect of white matter integrity (see also 
). Therefore, other factors must contribute to differences in white matter integrity among healthy, older adults. It is likely that certain environmental factors, such as vascular risk factors, may contribute. Also likely – but currently unexplored – is a role for genetic factors, including polymorphisms in genes that affect central myelination.
Our study included some limitations: First, the number of subjects was not uniformly distributed across the age spectrum; many subjects were clumped in the late 60s, resulting in less information about individuals at the upper and lower ends of the age spectrum under consideration. Second, we used a very narrow definition of “normal” in order to achieve as homogeneous a sample population as possible, and our findings may not necessarily be applicable to more general populations of older adults. Third, despite the methodological advantages of voxel-wise cross-subject statistics afforded by TBSS, some information is inevitably lost during image coregistration. Future work may address these limitations, and also identify the factors responsible for age-related white matter changes. Hopefully these future insights will lead to interventions to stem the slowing of cognitive processing and its associated morbidity.