We scanned 66 healthy subjects aged between 23 and 81 years and found evidence for a widespread linear negative association of GM volumes and a more localised negative association of WM volumes with age. Furthermore, we found that FA, a putative marker of integrity of WM microstructure, was negatively associated with age in many fibre tracts, mainly due to a positive association of diffusivity perpendicular to the main axis of the fibre tracts. MD, which reflects overall diffusivity, showed a positive association with age in many WM regions, and this was less extensive in the occipital lobe. Each of these patterns of change, and the putative mechanisms underlying them, are discussed in detail below.
Grey matter volume
In our study most cortical and deep GM regions showed a linear negative association between volume and age. These findings are broadly in line with previous studies of similar cohorts (Good et al., 2001; Jernigan et al., 2001; Pfefferbaum et al., 1994; Raz et al., 1997; Resnick et al., 2000; Salat et al., 2004; Sowell et al., 2003; Sullivan et al., 1995, 2004; Walhovd et al., 2005
), although the volume reductions we detected are even more widespread than some previous reports that have found preservation of volume in specific structures such as amygdala and hippocampus (Good et al., 2001; Sullivan et al., 1995
There are a number of potential cellular processes that could explain these volume reductions in GM. In early adulthood, elimination of neurons and synapses (Ge et al., 2002; Webb et al., 2001
) can contribute, whereas in middle and late adulthood volume reductions could be due to shrinkage of large neurons (Peters et al., 1998; Terry et al., 1987
), or rarefaction in the GM microvasculature, resulting in a loss of neurons (Farkas and Luiten, 2001; Riddle et al., 2003
). In addition, GM volume reductions inferred from T1-weighted images may be apparent, arising not only from genuine decreases in neuronal volume but also from changes in myelination, affecting the intensity of the MR signal (Sowell et al., 2003
). This may be a plausible mechanism in early adulthood because of ongoing processes of myelination (Yakovlev and Lecours, 1967
) but is less likely to play a role in later adulthood. This last hypothesis may explain why different ageing trajectories have been previously found in different cortical areas, appearing linear in visual, auditory and limbic cortices, which complete myelination early in life, and nonlinear or more complex in associative frontal and parietal cortices and posterior temporal cortex, which continue myelination into adulthood (Sowell et al., 2003
White matter volume
The total WM volume follows a nonlinear inverted “U-shaped” relationship with age, slightly increasing during the early adulthood, consistent with the notion of an ongoing maturation of the WM beyond adolescence, and then reaching the peak in the fourth decade (Ge et al., 2002; Sowell et al., 2003; Walhovd et al., 2005
Using a VBM-style analysis, we found linear negative association of WM volume with age especially in the external capsule bilaterally, and in the right anterior limb of the internal capsule, anterior thalamic radiations, cerebral peduncle and cerebellum. Significant nonlinear (quadratic) relationships between WM volume and age were present in the superior corona radiata bilaterally and in the left superior longitudinal fascicle. These findings seem to contrast with those of a recent study on 84 normal subjects aged 13-70 years (Pagani et al., 2008
), where linear negative associations between volume an age were found in the fronto-parietal lobes (corona radiata, anterior cingulum, fornix) and left cerebellar peduncle and nonlinear relationships with age in the genu of the corpus callosum and in the right deep temporal association fibres. The presence of different areas where WM volume is related to age, found by the two studies, may be due to the fact that Pagani et al. (2008)
included younger subjects and used a DTI-based technique, and not VBM, to assess white matter fibre volumes.
Our findings suggest that both heavily myelinated fibres (e.g., corticospinal tracts, located in the posterior limb of the internal capsule) and thinly myelinated fibres (e.g., association fibres of the frontal lobe) are affected in the same way by the pathogenic mechanisms underlying the ageing process. The involvement of the motor system during ageing is confirmed in the current study by the fact that volume in precentral gyrus, where that system is expected to originate, showed a significant inverse correlation with age and by the findings of previous histological studies, where an age-related decrease in the number of pyramidal fibres in the spinal cord and a thinning of precentral gyrus were reported (Salat et al., 2004; Terao et al., 1994
WM atrophy could be a result of the ageing process per se
or a consequence of the effects of WM lesions, which may be present in the neurologically asymptomatic elderly population (Enzinger et al., 2005; Pagani et al., 2008; Rovaris et al., 2003
). However, none of our subjects had the presence of WM lesions and, further, a previous study found no overlap between regions showing age-related atrophy of WM fibres and the presence of WM lesions (Pagani et al., 2008
Overall, we demonstrated here that the loss of WM integrity over adulthood and ageing might be explained in the corticospinal tract by WM atrophy.
Diffusion metrics in white matter
When comparing the different age subgroups, both middle-aged (41-59 years) and older (60-81 years) adults showed a significant widespread FA decrease compared to young adults (23-40 years) whereas middle-aged and older adults were not significantly different from each other, suggesting that the loss of directionality in WM fibres becomes apparent during the transition from early to middle adulthood. Again, this is in line with the study by Salat et al. (Salat et al., 2005a
), where FA differences between young adults (21-37 years) and older adults (65-76 years) were apparent by middle adulthood (42-59 years). In the current study, the presence of widespread FA reductions in the transition from young to middle adulthood was demonstrated earlier than the volume reductions shown by our WM VBM-style analysis and earlier than the existing volumetric literature on WM (Bartzokis et al., 2001; Ge et al., 2002; Sowell et al., 2003; Walhovd et al., 2005
). This regional decline in FA with early ageing occurs in the absence of significant regional increase in FA elsewhere, suggesting that middle adulthood may be the time when the transition from development to ageing in white matter fibre coherence takes place.
The findings of our study challenge the idea that early myelinating posterior regions are less susceptible to age-related change than late myelinating regions of frontal lobe white matter. Although previous DTI studies have provided some support for this idea, i.e. greater age-related FA decline in frontal compared to posterior WM regions (Abe et al., 2002; Ota et al., 2006; Pfefferbaum et al., 2005; Salat et al., 2005a,b; Sullivan et al., 2006
), our results are more closely in line with studies where a decrease in FA was found not only in regions of the frontal lobe but also in more posterior regions such as the splenium of the corpus callosum and posterior limb of the internal capsule (Ardekani et al., 2007; Furutani et al., 2005; Salat et al., 2005a
Our results also demonstrate that the two major tracts connecting limbic structures in Papez circuit (i.e., fornix and cingulum) are both affected by ageing. The involvement of the fornix in our cohort of subjects is also confirmed by the fact that volume of hippocampus, a GM structure connected by the fornix, showed a significant inverse correlation with age. However, these results appear to conflict with a recent study in subjects aged 18 to 88 years, where cingulum but not fornix appeared resistant to ageing (Stadlbauer et al., 2008
In our study, MD showed a spatial pattern of associations with age broadly similar to FA, although it seemed that posterior regions of the WM were less extensively affected. The comparison across different age subgroups showed that, in contrast to FA changes, changes in WM MD appear as a late ageing phenomenon, consistent with a previous study which used slightly different age ranges for the two groups (Ardekani et al., 2007
The assessment of parallel and perpendicular diffusivity may enhance the specificity of DTI findings and shed light on the mechanisms underlying FA changes. Changes in parallel and perpendicular diffusivity have been related to axon and myelin damage, respectively, in mouse models of multiple sclerosis (Budde et al., 2007a,b; Song et al., 2005; Sun et al., 2007; Wu et al., 2007
), dysmyelination (Song et al., 2002
) and retinal ischemia in the optic nerve (Song et al., 2003
). Moreover, perpendicular diffusivity, together with FA and MD, was shown to be a robust predictor of myelin content in post-mortem
human brain, prior to and after fixation (Schmierer et al., 2008
). However, the degree to which these relationships hold in the healthy living human brain remains to be determined. Nevertheless, interrogating how the different diffusivities contribute to observed FA changes provides complementary information. In the current study, we found that negative correlation of FA with age was mainly driven by positive correlation of perpendicular diffusivity, whereas parallel diffusivity only showed a slight non-significant positive correlation with age. These findings are in line with other recent studies testing the tensor diffusivities during ageing (Bastin et al., 2009; Bhagat and Beaulieu, 2004; Ota et al., 2006; Salat et al., 2010; Stadlbauer et al., 2008; Sullivan et al., 2010; Vernooij et al., 2008
) and may be explained by the presence of degenerative changes in the myelin sheaths or, alternatively, of a reduced fibre organisation or “packing” (i.e., a decrease in the number of axons) within WM tracts.
In our study, FA changes overlap with regions showing evidence for WM atrophy (i.e., reductions in WM volume using VBM-style analysis) in the forceps minor, internal and external capsules, cerebral peduncle and temporal association fibers, consistent with previous reports showing similar changes in FA and VBM measures of WM (Vernooij et al., 2008
). At the level of the superior corona radiata, however, out data suggest that this relationship seems to be present only after 50 years of age. Further, in all others areas loss of WM microstructure was detected in the absence of significant WM changes on VBM-style analysis. This suggests that DTI can provide more sensitive information on ageing-related WM degeneration compared to volume metrics.
Our study benefited from some methodological advances compared to previous DTI studies investigating ageing-related FA changes. First, data were acquired using 60 diffusion directions, thus providing increased power and higher angular resolution compared to some previous investigations. Second, we used a technique for sensitive and robust voxelwise analysis of DTI data (Smith et al., 2006
) as well as an improved approach to statistical thresholding which enhances all cluster-like structures present in an image avoiding the need for the arbitrary initial cluster-forming threshold (Smith and Nichols, 2009
In studies on the ageing brain, the quality of image registration is an important issue, especially as it applies to small structures such as the hippocampus. Indeed, an age-related volume decline of such structures may represent an artifact of distorted shape from adjacent, expanding fluid-filled spaces rather than a real atrophy process. However, as registration was accurately checked, we do not believe that results of this study are due to misregistration.
However, there are some limitations to this study. First, we had fewer subjects in the older age groups, and this might have underpowered the subgroup comparison analyses. Second, the study has a cross-sectional design and therefore, because it contains different cohort of subjects in different age groups, it may be less sensitive to brain changes across the lifespan than a longitudinal study. Third, we recognise that age clustering in this study is arbitrary. However, as we mention in the Methods section, grouping selection was broadly in line with previous studies. In addition, we deliberately included complementary analyses (i.e., testing for correlations across the whole age range) that were not dependent on this grouping. Finally, after performing residual analysis of all plotted measures in the different regression analyses, we found no clear evidence for heteroscedasticity and no outliers in the VBM or FA analyses and only one outlier in the MD analysis (data not shown). However, it should be noted that regression analysis is relatively robust to homoscedasticity and small-to-moderate violations of homoscedasticity have only minor impact on regression estimates (Fox, 2005
In conclusion, we assessed a number of measures of brain structure to characterise the spatial and temporal pattern of changes that occur in both GM and WM with normal ageing. Reductions in GM volume appear in middle age and then become more widespread into older adulthood. Widespread age-related decline in WM microstructure occurs earlier and can be detected more sensitively using DTI-based measures of microstructure than using markers of WM volume derived from conventional T1-weighted imaging. The observed patterns and dynamics of normal age-related changes may have important implications for future studies on chronic neurological conditions that show an impact of age on disease onset, course and progression, such as Alzheimer's disease, Parkinson's disease and multiple sclerosis (Baquer et al., 2009; Reeve et al., 2008; Trojano et al., 2002
). The application of these novel MRI techniques may be useful for furthering our understanding of pathologic substrates underlying such neurological conditions, for charting their evolution and monitoring response to treatment. Finally, the study of age-related brain changes may potentially benefit from the use of high field scanner MRI acquisition, which provides increased signal-to-noise ratio and so opens possibilities for greater spatial resolution, although issues related to field inhomogeneities and increased chemical shift (Pagani et al., 2007
) deserve further investigation.