Brain white matter structure and information processing speed in healthy older age

Cognitive ageing is associated with decline in average levels of some cognitive capabilities, such as information processing speed, abstract reasoning, executive function and aspects of memory, even in the absence of overt pathology (Salthouse 2010). Understanding the biological bases of cognitive ageing is therefore an increasingly important area of research for developed societies with ageing populations (Deary et al. 2007, 2009). Brain structural changes are also commonly observed in normal ageing and include atrophy of both grey and white matter regions (Bartzokis et al. 2001; Raz et al. 2005), ventricular enlargement, white matter hyperintensities (WMH) in deep and periventricular regions (Fazekas et al. 1987), enlarged perivascular spaces (Doubal et al. 2010) and mineral deposition (Bartzokis et al. 2007). The present study focuses on cerebral white matter, and hence the brain’s ability to pass information efficiently between different cortical regions, an important factor in terms of understanding cognitive functioning in older age (Gunning-Dixon et al. 2009; Malloy et al. 2007; Penke et al. 2012).

It has been argued that information processing speed is a fundamental characteristic of the brain’s cognitive efficiency. This is because it is relatively elementary; the tests which assess it require only simple and repetitive responses to questions which would not attract errors if given sufficient time. Performance on simple tests of processing speed is associated with scores on more complex cognitive tests, such as reasoning (Jensen 2006). Processing speed might therefore mediate how age-related brain changes influence the efficiency of ‘higher’ cognitive functions (Deary 2000; Salthouse 1996; Verhaeghen 2014; Penke et al. 2012).

A number of cognitive tests with different stimulus–response arrangements can provide measures of processing speed. These include paper-and-pencil psychometric tests, such as Digit-Symbol Substitution (Wechsler 1998), as well as psychophysical measurements, such as Choice Reaction Time (Deary et al. 2001; Jensen 1987) and Inspection Time (Deary 2000; Deary et al. 2010). Performance on processing speed tests is correlated; subjects who are more efficient on one speed test tend to be more efficient on the others. Therefore, data reduction techniques, which have the advantage of removing error and task-specific variance associated with each individual test, can be used to combine an individual’s performance on several tests into a general index of information processing speed (which we name ‘g _speed’). This is typically the first unrotated component (sometimes called a ‘factor’) from a principal component analysis (PCA) of scores from the individual tests which captures the largest proportion of the covariance between them (Luciano et al. 2009). Information processing speed, and hence g _speed, decreases with age and may therefore provide a useful indicator of cognitive decline in later life (Deary et al. 2010; Ritchie et al. 2014; Salthouse 1996; Verhaeghen 2014).

Brain white matter structure in older age can be investigated using imaging biomarkers derived from diffusion tensor MRI (DT-MRI), such as fractional anisotropy (FA) and mean diffusivity (MD) (Salat et al. 2005; Sullivan and Pfefferbaum 2006). Although ascribing these biomarkers directly to white matter ‘integrity’ is problematic (Jones et al. 2013), a number of authors have used them to investigate links between brain structure and processing speed in older age. For example, Kennedy and Raz (2009) examined relationships between FA measured in several white matter regions and cognitive tests, including processing speed, in a sample of 52 subjects aged between 19 and 81 years. After correcting for the effects of age, only working memory and processing speed remained significantly associated with FA in anterior brain regions. Haász et al. (2013) investigated links between general fluid intelligence (g _f) and white matter water diffusion parameters in a cohort of 100 healthy participants aged between 49 and 80 years. Using two measures of processing speed, a colour-word interference test and a visuo-spatial attention task, as well as several other tests to compute g _f, they found associations between g _f and MD, FA and radial diffusivity across the white matter skeleton with Tract-based Spatial Statistics (TBSS; Smith et al. 2006). However, of all g _f tests, processing speed scores contributed most towards associations with white matter structure.

Vernooij et al. (2009) investigated relationships between brain white matter water diffusion parameters and cognitive ability in 860 people aged 60 and older. They found that elevated mean, axial and radial diffusivities and decreased FA were associated with worse performance on processing speed and general cognitive ability tasks for both normal-appearing white matter and that located within WMH. In their study of 554 adults aged 70–82 years, van den Heuvel et al. (2006) found that periventricular WMH were significantly associated with decreased processing speed performance as measured by the Letter-Digit Coding test.

Although these studies provide valuable insights into links between brain white matter structure and cognition in older age, they tend to include participants with a wide range of ages. This is problematic since chronological age is a major confounder of cognition and white matter structure, and age-heterogeneous cohorts may mask its true effect (Hofer and Sliwinski 2001); for example, white matter FA and MD are known to decrease and increase, respectively, with advancing age requiring mediation analyses to account for such effects in brain-cognition studies of older adults (Madden et al. 2012). In the present study, we investigate relationships between white matter structure and information processing speed using voxel-based methods in the Lothian Birth Cohort 1936 (LBC1936), a large group of generally healthy subjects in their seventies with a narrow age range. These attributes, coupled with the availability of a validated measure of cognitive ability from childhood and a diverse range of processing speed tests performed in older age, make this cohort valuable for studying associations between brain structure and cognition in older age.

Previous studies conducted in the LBC1936 have assessed associations between general factors of information processing speed (g _speed), general intelligence (g) and brain white matter DT-MRI biomarkers measured in tracts segmented using quantitative tractography. Penke et al. (2010) reported associations between tract-average white matter FA, MD, axial and radial diffusivity, and g _speed in a pilot sample of the LBC1936 (n = 132). They found that g _speed had a significant association with tract-average FA and MD in all eight major tracts investigated (genu and splenium of corpus callosum, and bilateral cingulum cingulate gyri, uncinate and arcuate fasciculi). This suggests that g _speed is likely to be a property related to global brain connectivity rather than to specific tracts. Penke et al. (2010) also demonstrated that, like g, a general brain white matter FA factor (explaining 40 % of the individual differences across all eight tracts) could be extracted from the tractography data and this factor was associated with g _speed (r = −0.24, p = 0.007). In a further analysis conducted when all LBC1936 participants’ brain imaging data were available, Penke et al. (2012) studied the relationships between tract-average FA, longitudinal relaxation time and magnetisation transfer ratio with g and g _speed in 12 major white matter tracts in 420 subjects (the eight tracts measured in Penke et al. (2010) plus bilateral inferior longitudinal fasciculi and anterior thalamic radiations). General factors of all three tract-averaged biomarkers (explaining 40–70 % of the individual differences across all 12 tracts) were independently associated with g (together explaining 10 % of the variance) and, importantly, this effect was completely mediated by g _speed. This is amongst the strongest evidence to date that processing speed is a mediator between brain white matter structure and general cognitive functioning. However, these previous studies have measured imaging biomarkers averaged across specific tracts, and have not investigated whether, and where, individual tests might be correlated with white matter structure using methods that allow greater spatial localisation of effects.

The present paper reports new analyses on the LBC1936, testing the extent to which white matter structure is related to differences in processing speed measures. For these analyses, we used TBSS rather than quantitative tractography methods. TBSS allows analysis of DT-MRI data to be conducted on a voxel-wise basis across the brain, therefore revealing more detail on the spatial distribution of associations between white matter structure and cognitive ability than is possible with quantitative tractography. We first tested white matter FA associations with both g _speed and the individual speed tests contributing to it. We also assessed whether specific aspects of the individual processing speed tests were associated with white matter structure beyond that accounted for by g _speed (Booth et al. 2013). Finally, since our participants had measures of their childhood intelligence available, we tested whether later life associations between white matter FA and processing speed were still present after the speed tests had been adjusted for earlier cognitive ability.

Amongst the 730 participants who underwent MRI, 581 subjects (311 males) fulfilled the inclusion criteria of providing complete DT-MRI, childhood intelligence and processing speed data. The mean age at time of scanning was 72.7 ± 0.7 (range 71.0–74.2) years. Table 1 shows participant health characteristics for the whole sample, and male and female subgroups. In general, the male and female subgroups had fairly similar health characteristics, although males had significantly lower Mini-Mental State Examination scores, and significantly higher alcohol consumption, and self-reported incidence of diabetes and cardiovascular disease; conversely females had significantly higher incidence of arthritis and blood circulation problems. The majority of subjects had WMH loads rated on the two lowest Fazekas scores for both periventricular (67.5 %) and deep (78.8 %) white matter; females had a significantly higher Fazekas score in deep white matter compared with males.

Mean and SD values for age 11 IQ, and the processing speed tests and their residuals are shown in Table 2. Males and females performed similarly on these tests, although females had significantly higher age 11 IQ and provided a significantly larger number of correctly decoded symbols in the Digit-Symbol Substitution test than males, while males produced a significantly greater number of correct responses in the Inspection Time task than females. As indicated in Table 3, Pearson’s product-moment correlations between the individual processing speed tests were moderate to large suggesting that a common factor could be extracted. As described above, PCA was then run on these data which produced a strong common factor, g _speed, which explained 65 % of the total test variance; all cognitive tests had a large correlation with g _speed, except Simple Reaction Time which had a moderate correlation.

The present study used voxel-based methods to extend previous analyses that have employed quantitative tractography to investigate links between cognitive ability and white matter structure in the LBC1936 (Penke et al. 2010, 2012). The results showed that, in older age, there were skeleton-wide associations between white matter FA and measures of processing speed which supports the hypothesis that the latter is not localised to specific tracts (Haász et al. 2013; Penke et al. 2010, 2012). This was the case regardless of whether individual tests or g _speed was used to index processing speed. There were, however, some specific areas of the brain that were associated with variance in some tasks that were not part of general processing speed. Finally, intelligence assessed at age 11 years was not associated with white matter FA in most models suggesting that older age associations between white matter structure and processing speed are generally not accounted for by cognitive ability measured in youth. Our findings highlight the crucial role white matter plays in providing a neuroanatomical structure for individual differences in the information transfer underpinning fundamental aspects of human intelligence. They also provide support for viewing the brain as a highly connected distributed network with the slower processing speed found in older age being part of a process of cortico-cortical disconnection (Bennett and Madden 2014; Felleman and Van Essen 1991; Sporns 2013).

Beyond g _speed, we investigated whether the individual processing speed tests had specific associations with white matter FA. This was achieved by removing the shared variance with all the other tests from each individual test by residualising for g _speed. The residuals of Symbol Search, Simple Reaction Time and Inspection Time were no longer associated with white matter FA, indicating that variance specific to these tests was not related to white matter structure beyond their shared variance with g _speed. However, residual Four-Choice Reaction Time was significantly associated with FA in the genu of corpus callosum, and residual Digit-Symbol Substitution was associated with FA in frontal white matter. This may be due to these tasks’ overlap with cognitive abilities other than information processing speed. For example, there is evidence that the Digit-Symbol Substitution task also involves use of memory which is associated with frontal fibres (Charlton et al. 2010; Piccinin and Rabbitt 1999). Similarly, faster responses on Four-Choice Reaction Time require inter-hemispheric coordination and selection of manual responses, which is associated with the corpus callosum (Doron and Gazzaniga 2008).

Childhood intelligence was not significantly associated with older age white matter FA in the model with age and sex as the only covariates. Similarly, there were no significant associations between childhood intelligence and white matter FA when g _speed, or any of its individual processing speed tests, were also included in the model. However, when the same analysis was repeated with residual processing speed scores, childhood intelligence became significantly associated with white matter FA in the splenium and posterior body of corpus callosum for all residual scores except Digit-Symbol Substitution. These data therefore suggest that although early life intelligence does not strongly predict later life global white matter FA, it does predict the specific relationship between splenium FA and better scores on the g _speed-independent components of these individual cognitive tests. A possible explanation for this finding is that the residual test scores reflect other orthogonal cognitive domains to processing speed, such as memory, which are known to relate to measures of childhood cognitive ability (Deary 2014). These results should be considered in the context of other analyses performed in the LBC1936, for example regarding WMH burden and brain volume, which show that early life cognitive function has a significant effect on white matter structure in older age. Specifically, Valdés Hernández et al. (2013) found in approximately 600 LBC1936 participants that childhood intelligence was negatively correlated with WMH volume (r = −0.08) suggesting that lower cognitive ability in youth leads to increasing white matter damage with age, while Royle et al. (2013) reported that childhood intelligence was a predictor of estimated prior (r = 0.12) and measured current brain volume (r = 0.11) in older age.¹ However, Penke et al. (2012) found that white matter FA measured in 12 major tracts using quantitative tractography was more associated with intelligence in later rather than early life, as reported here. It is therefore possible that DT-MRI biomarkers, even measured skeleton-wide in voxel-based analyses, reflect the effects of cognitive ageing more than they reflect lifelong-stable differences in intelligence. Clearly, the availability of measures of childhood cognitive ability is an uncommon feature of the current study which makes generalising these findings more difficult; it would be tempting to conclude that early life cognition need not be a key feature of DT-MRI studies of normal ageing. Further work, in both cross-sectional and longitudinal analyses, in cohorts with such early life cognitive data, for example the Aberdeen Birth Cohorts (http://​www.​abdn.​ac.​uk/​birth-cohorts), is required to clarify this finding.

Our results showed some subtle sex differences: females had higher white matter FA in isolated parts of the genu and splenium of corpus callosum, whereas males had higher white matter FA in small mid-brain segments of the superior corona radiata. This may reflect differential brain ageing according to sex, although males and females had fairly similar healthy characteristics in this cohort. These sex differences in white matter structure will require replication in other large, later life cohorts.

This study has several strengths compared to previous studies of information processing speed and white matter structure in older age. The LBC1936 is a large cohort with a narrow age range and generally healthy participants. Subjects took five processing speed tasks drawn from different traditions, namely psychometric, experimental psychological and psychophysical tests. Together, they created a reliable measure of g _speed. Furthermore, the availability of childhood intelligence data from age 11 allows prior cognitive ability to be controlled for and thus permits changes that are specifically due to ageing to be identified more efficiently. This is an important issue since we have shown previously that all of the processing speed tests are strongly associated with general cognitive ability from youth, about 60 years earlier (Deary et al. 2010). Analysis of the DT-MRI data using TBSS allows relations between processing speed and white matter structure to be investigated voxel-by-voxel across the brain, rather than in a few major tracts, in an automated and user-independent fashion. However, this method only allows comparison of individual differences in white matter FA across those areas which show good spatial concordance with the white matter skeleton. Thus, local gyral white matter and other areas, which show a high degree of morphological variability, were not included in these analyses.

The narrow age range of this sample can, conversely, also be viewed as a limitation because our findings cannot be generalised to younger populations and are only representative of older subjects. Similarly, because the LBC1936 subjects are generally healthy, our finding cannot be directly applied to ageing populations with decreased processing speed due to neurodegenerative diseases, such as dementia. In order to keep the number of models manageable, we did not investigate associations between processing speed test scores and other DT-MRI biomarkers, such as mean, radial and axial diffusivity; these variables may be more prone to partial volume CSF-contamination errors than FA (Metzler-Baddeley et al. 2012). Finally, we did not take into account WMH; how these features might affect the results reported here is currently being investigated in the subgroup of LBC1936 participants who have significant lesion burden.

In conclusion, we found using voxel-based analyses that g _speed, as well as five varied tests of information processing speed, had significant skeleton-wide associations with white matter FA. These results indicate that quicker and more efficient information processing requires global connectivity in older age. Conversely, they can also be interpreted to imply that there is an age-related global reduction in connectivity that affects all networks, including those associated with the administered processing speed tests. Childhood intelligence was not associated with white matter FA, except for several residual (g _speed-independent) test score models, suggesting that associations between white matter structure and information processing speed measured in older age are not generally accounted for by cognitive ability measured in youth.