Cognitive aging has been extensively investigated over the past two decades. One of the critical reasons for this research trend is due to the crisis of population aging which encourages researchers paying attention to evaluate the trajectories of cognitive function decline across chronical age to help people age gracefully. One pronounced diagram showing behavioral performance on measures of processing speed, working memory, long-term memory, and world knowledge was first reported by Park et al. (2002), in which while many cognitive functions, such as speed, spatial orientation, problem solving, numerical ability, verbal memory were significantly declined with age, whereas verbal ability (e.g., vocabulary) did not show decline with age. However, these trajectories of functional decline with age was based on the cross-sectional aging data (Light, 1991; Park, 2002; Park et al., 2002; Salthouse, 2004), it is still unclear whether similar results could also be seen in longitudinal (or follow-up) data. Literature has highlighted an important distinction between age effects and aging effects (see Rugg, 2017). An age effect refers to a scenario in which a dependent variable differs between groups of individuals with different mean ages (such as contrasting a younger group with an older group)—this is known as a cross-sectional design. Whereas an aging effect refers to a scenario in which we observe an individual’s performance changes over time—this is known as a longitudinal or follow-up design. Rugg (2017) has indicated several inferring limitations regarding age-related cognitive aging using a cross-sectional design. Some studies have also provided evidence in showing discrepant results coming from cross-sectional versus longitudinal data, hence highlighting the importance of longitudinal evidence to reveal actual aging effect in cognition (Hedden and Gabrieli, 2004; Hsieh and Yang, 2021).

Participants’ demographic information and DTI measures for TP1 and TP2 and their statistical tests are shown in Table 1. The results show that age, MoCA, MD, RD, and AxD, but not BDI-II and FA scores, are significantly different between TP1 and TP2. Please note, for each of the cognitive performance index and for each WM tract of the DTI measures between TP1 and TP2, the mean scores and respective paired t-tests results are shown in the Supplementary Tables 1, 2.

To address the issues set out in this study, we report the results as follows:

This study aimed to provide longitudinal data to compare with cross-sectional data, in which we were interested in examining if age-related differences (cross-sectional) in WM tracts and cognitive control functions can be also observed in age-related changes data with a follow-up interval of 2 years (longitudinal).

Regarding the first issue of the relationship between age and cognitive control function (including processing speed) cross-sectionally and longitudinally, we observed that while there were several significant age-related differences in processing speed, common (inhibition), shifting, and working memory updating, there were no significant correlations for age-related changes in cognitive control functions. Only changes in processing speed from TP1 to TP2 reflected on GPT_L were significantly related to age (at TP1). The results suggest that many significant correlations between age and cognitive control functions originally shown on cross-sectional data no longer exist on the longitudinal data, except that the relationship with age for processing speed was still retained even in the longitudinal data.

The current findings bring our attention in the fact that although we could observe age-related cognitive control deficits which are typically reported in most literature using cross-sectional approach (e.g., less working memory capacity – Salthouse, 1994; West, 1996; Grady and Craik, 2000; a deficit in inhibitory processing – Hasher and Zacks, 1988; Kramer et al., 1994; a lack of cognitive flexibility – Kramer et al., 1999; Mayr, 2001), when we followed up these individuals across the adult lifespan over an average of 2 years, we did not see many differences (except processing speed reflected on the left hand’s movement) in their changing trajectories. Literature has also indicated a similar finding in that age did not significantly correlate with working memory changes over a period of 2 years (Charlton et al., 2010; Lövdén et al., 2014). Therefore, we should not overlook the possibility that aging processing in cognitive control functions likewise occurs for younger adults. That is, the aging process in cognitive control functions may not be older-age specific, but more generally occur across lifespan.

Regarding the second issue of the relationship between age and DTI measures (FA, MD, RD, AxD), we observed that FA indexes in many WM tracts at TP1 were sensitive to age effect (cross-sectional data: all tracts except CG_R, CH_L, CH_R, and CST_R). In particular, Fmin has the strongest age effect reflected on FA, which is in line with the findings in the literature either using a longitudinal or cross-sectional approach (e.g., longitudinal approach: Barrick et al., 2010; cross-sectional approach: Salat et al., 1997; Vernooij et al., 2008; Sullivan et al., 2010; Jolly et al., 2016; Zavaliangos-Petropulu et al., 2019; Hsieh et al., 2020). Conversely, the FA index’s changes from TP1 to TP2 were much less correlated to the aging effect (i.e., follow-up data). The result suggests that FA might not be a robust index when examining the aging effect. On the other hand, both MD and RD at TP1 and their changes from TP1 to TP2 showed relatively similar relationships with age (except CST_L/R, Fmaj, and Fmin tracts)² Although AxD showed a different pattern from other DTI measures in relation to age on the cross-sectional data (e.g., they were more non-linearly correlated with age compared to other DTI metrics), its changes somehow showed similar patterns as MD and RD changes in the follow-up results, in which several WM tracts were significantly correlated with aging (except Fmaj and Fmin tracts). Although previously based on the cross-sectional data, even our lab also claimed that FA might play a more important role than MD, RD, or AxD in reflecting the age effect (Hsieh et al., 2020), the current results based on both cross-sectional and longitudinal data conversely suggest that MD, RD, and AxD indexes in WM tracts might be more representative than FA when examining both age and aging effects with DTI (note: AxD yielded more non-linear correlations with age cross-sectionally). FA is a ratio value reflecting the directional consistency of diffusion, with higher values indicating that diffusion within a voxel is primarily restricted to one direction (O’Donnell and Westin, 2011), conversely, MD refers to the average amount of diffusion occurring within an image voxel, RD is calculated as the amount of diffusion perpendicular to the main directional axis of fibers, and AxD refers to the magnitude of diffusivity parallel to fiber tracts. Early studies of optic nerve fiber damage (Song et al., 2003, 2005) indicated that increases in RD values are associated with axonal demyelination (Wheeler-Kingshott and Cercignani, 2009), whereas lower AxD may reflect axonal injury, reduced axonal caliber, or less coherent orientation axons. Literature has shown a negative correlation between age and FA, suggesting that in older individuals the diffusion tensor is less fractionally anisotropic than in relatively younger individuals (e.g., a cross-sectional study by Boekel and Hsieh, 2018). In addition, literature using a cross-sectional approach also often report a positive correlation between MD, RD, or AxD and age (e.g., Sullivan et al., 2001; Pfefferbaum and Sullivan, 2003; Abe et al., 2008; Westlye et al., 2009; Burzynska et al., 2010; Lebel et al., 2012). Since the current longitudinal data showed that MD, RD, and AxD changes are more closely related than FA changes to the aging process, we would suggest that axonal demyelination and/or axonal degeneration is a more robust phenomenon for observing intra-individual aging processes.

When we further compared age-related declines in different WM tracts, we found association fibers of CG_R and CH_L/R were less correlated to both age and aging effects when compared to other WM tracts. In addition, ATR_L/R, and Fmin tracts across FA/MD/RD/AxD appeared to be most correlated to the age effect (cross-sectional data), in which the findings are consistent with the suggestion of frontal and callosal areas being most affected by the age effect (e.g., cross-sectional studies: Zavaliangos-Petropulu et al., 2019; Hsieh et al., 2020; Behler et al., 2021).

However, the current longitudinal data showed that the commissural fiber of Fmin no longer yielded an aging effect, despite its strong relationship with the age effect cross-sectionally. On the other hand, although the frontal association fibers of ATR_L/R remain to be correlated with aging, they were only present in MD/RD/AxD, but no in FA. Therefore, combining the cross-sectional and longitudinal results in the current study, we suggest that ATR tracts in MD/RD/AxD are the most representative when examining both age and aging effects.

Turning to the third issue of the association between cognitive control functions (including processing speed) and DTI measures, we observed that processing speed and cognitive control functions (including common/inhibition, shifting, and working memory/updating) were related to DTI measures cross-sectionally, with processing speed showing the most correlations with WM tracts in FA/MD/RD. Literature using a cross-sectional approach has also shown that age-related WM decline may contribute to age-related cognitive control declines. For example, some studies have explored age-dependent relationships between white matter integrity and composite measures of cognitive control function (Kennedy and Raz, 2009; Vernooij et al., 2009; Zahr et al., 2009). Furthermore, some researchers have found that frontoparietal WM differences are linked to age-related differences in task-switching performance (Madden et al., 2009; Gold et al., 2010). Perry et al. (2009) using a cross-sectional approach also observed that advancing age was associated with declines in task-set shifting performance and with decreased FA in corpus callosum and in association tracts that connect the frontal cortex to more posterior brain regions. A more recent DTI study using a cross-sectional approach led by Karayanidis also demonstrated the role of WM microstructure in age-related cognitive decline (Jolly et al., 2013). Bucur et al. (2008) also using a cross-sectional approach observed that declines in FA in the pericallosal frontal region and in the genu of the corpus callosum, but not in other regions, mediated the relationship between perceptual speed and episodic retrieval reaction time. Hence, they suggested that WM integrity in prefrontal regions is one mechanism underlying the relationship between individual differences in perceptual speed and episodic retrieval (Bucur et al., 2008). Nevertheless, most of these studies cited above are cross-sectional results.

However, when we examined the longitudinal data in the current study, we observed that changes in processing speed as compared to changes in other cognitive control performance were more correlated to the changes in all DTI indexes. Therefore, processing speed appeared to be the function that was mostly related to WM integrity both cross-sectionally and longitudinally. Processing speed is a basic cognitive or brain process that subserves many other higher-order cognitive domains. One possibility is that over a short period of follow up, changes in processing speed would be more apparent than other cognitive control functions, thus would be easier to be observed to be related to changes in WM integrity. Future research with a longer period of follow up is needed to clarify the speculation.

Finally, to provide a more direct evidence in showing if changes in DTI mediated age-related changes in processing speed, we tested a series of mediation models, and the result showed that four models yielded good model fits, in which ΔRD, ΔMD, and ΔAxD mediated age-related changes in processing speed. Therefore, converging the regression results and mediation models, we could conclude that MD, RD, and AxD appear to be the most representative DTI measures to reveal age-related changes in processing speed.

Some final remarks are worth mentioning before closing. First, in the current study, only a period of 1–2 years follow-up interval was performed, it is likely that the period is too short to be sensitive enough to reveal the aging process. For example, in the literature, Charlton et al. (2010) and Lövdén et al. (2014) using a follow-up approach likewise did not find an aging effect on cognitive changes, but Vonk et al. (2020) found a linear aging effect in working memory and cognitive speed. The discrepant findings among these studies are likely due to their differences in the length of the follow-up interval. The follow-up interval between the two timepoints in Vonk et al.’s (2020) research was 4∼8 years, whereas in the other two research (i.e., Charlton et al., 2010; Lövdén et al., 2014), the age interval was about 2 years which is similar to the current study. Therefore, future studies with a longer follow-up interval or more times of follow-up are warranted. Second, while we examined the aging effect in relation to the changes in DTI from TP1 to TP2, we observed very few and even paradoxically positive correlations for three WM tracts (i.e., CH_L/R, CST_L) in FA. The finding was the opposite to the cross-sectional results reported by the prior and current research. One possibility is that the reduction rate for FA with aging was not as prominent as one could expect from other DTI indicators, and because this study only followed 1.77 years on average, thus very few significant changes could be observed for FA. However, why these three tracts (CH_L/R, CST_L) showed FA increases rather than decreases with aging remain puzzled. We suspected that individual differences might contribute to the paradoxical findings. One longitudinal study with 1-year follow up comparing healthy control and Alzheimer’s disease also found that healthy controls did not demonstrate FA changes in the hippocampal cingulum (i.e., CH) as were observed in those with Alzheimer’s disease (Mayo et al., 2017). The authors thus concluded that changes in microstructural integrity for the hippocampal cingulum over short time intervals (i.e., 1 year) may more specifically reflect ongoing degenerative processes due to Alzheimer’s disease. Since the participants in the current study were all healthy controls, some of them might be even more reserved than middle-aged or young adults. Future research with a longer follow-up period and considering individual differences is needed to clarify the issue. Third, the participants’ age range in this study covered a wide range of 20 to 80 years, it may be worth considering sub-groups of different ages rather than treat all participants as a single group (e.g., see Yap et al., 2013 and Lebel et al., 2019 for reviews of DTI findings related to development). Future studies with more numbers of participants are encouraged to split into subgroups with smaller age ranges (e.g., every 10 years old per group). Fourth, this study did not analyze myelin water fraction (MWF) which should not be overlooked and should be analyzed complementary to the conventional DTI metrics. A recent study (Kiely et al., 2022) has indicated that although DTI metrics (e.g., RD and FA – indices of myelin content) might be related to the age effect, they could not serve as specific metrics to myelin, so that further studies using more specific myelin measure, such as MWF relaxometry, are required. Fifth, in this study, we used the TBSS processing pipeline in the FSL, which was developed to reduce the effects of local mis-registrations by projecting all FA voxels onto the nearest location on a “skeleton” approximating WM tract centers (Smith et al., 2006). Some studies however pointed out that TBSS was designed to compensate for local registration errors, which might in return cause many more limitations due to over compromises (see Zalesky, 2011; Schwarz et al., 2014, for more examples). Despite the limitations outlined, TBSS remains a popular method for DTI analyses.

To conclude, the current results warrant the importance of longitudinal research for aging studies to elucidate actual aging processes on cognitive control function. More critically, the current results provide new insights to which indicator of WM integrity and which type of cognitive changes are most representative (i.e., potentially to be effective neuroimaging biomarkers) to reflect intra-individual cognitive aging processes.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by Research Ethics Committee of the National Cheng Kung University, Tainan, Taiwan (Contract No. 104–004). The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

SH contributed to the grant resources, the design of the experiments, data processing, and preparation of the manuscript. M-HY contributed to the data collection and analysis, and preparation of figures and tables. Both authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.