4. Discussion
Reduced WM performance in old age is associated with changes in both local neural dysfunction and functional connectivity. In the current study, alterations in task-evoked activity and task-based functional connectivity were independently examined by applying the background connectivity approach, and the major and novel findings are that: (1) a decline in the magnitudes of task-evoked activity was accompanied by increased seed-based background functional connectivity with posterior parietal regions, and both were tightly correlated with WM performance; and (2) while task-evoked activity mediated age-related WM impairment, task-based connectivity, on the contrary, has contributed to better WM performance in old age.
Standard univariate analyses on task-evoked activity revealed an age-related decrease in activity of the right IPL and an increase in activity of the right ACC. Although alterations in the IPL activities of older individuals have been repeatedly identified in previous studies (Cabeza et al., 2004; Carp et al., 2010), the involvement of the ACC has been relatively rarely reported in WM ageing. Some studies have identified special changes in the ACC related to WM performance (Bush et al., 2000; Lenartowicz and McIntosh, 2005). There was evidence from developmental studies that neural responses (negative activity during task) in the ventral ACC could predict between-subject differences in WM (Huang et al., 2016; Vogan et al., 2016), and several training programs on WM also highlighted the importance of modulations in ACC activities (Olesen et al., 2004; Brehmer et al., 2011). All these reports linked a more deactivated ACC (especially ventral) to better WM performance, possibly due to the successful inhibition of irrelevant information during tasks (Jonides et al., 1998; Vogt, 2009). It should also be noted that as previous investigations into WM ageing mainly focused on regions with task-evoked positive activity (as detailed in recent meta-analyses (Wager and Smith, 2003; Rottschy et al., 2012)), our findings of ACC with negative activity during task, along with a few recent investigations into task-negative areas (Sambataro et al., 2010; Anticevic et al., 2012a), thus implied that those negatively activated regions were as important as the positively activated regions in supporting human cognitions, especially for externally oriented cognitive functions such as working memory and executive function. Moreover, our studies failed to find any pattern of increased activity in frontal and parietal regions, and it was possible that the age-related regression analysis method based on a single group of older individuals, rather than a typical comparison between young and older individuals, was the reason that no increases were identified in activated regions.
Along with the alterations in local cortical processing, subsequent seed-based functional connectivity analyses on the background activity derived from the WM task also showed that, regions with increased and decreased task-evoked activity were more functionally connected to other regions, as the ACC was connected to the bilateral lateral parietal area and the IPL was coupled with the posterior cingulate gyrus, and these patterns of a more connected brain were linked to better performance on the N-back task. The associations between functional connectivity and WM performance have been well-characterized in previous studies. Consistent evidence from resting state neuroimaging data has identified that regions activated in the WM task, such as DLPFC, PreCG, and IPL, were highly connected (Fox et al., 2005; Power et al., 2011) and that connections among task-deactivated regions (such as the medial PFC and PCC) also made considerable contributions to WM (Hampson et al., 2006; Hampson et al., 2010; Sambataro et al., 2010). Furthermore, task fMRI studies, which modelled task-induced BOLD signals using methods such as PPI and DCM, provided further evidence on how information flow was modulated during WM tasks, depicting more details on frontoparietal interactions in different conditions (Ma et al., 2012; Heinzel et al., 2017; Jung et al., 2018).
In addition to the well-documented relationships between local task-evoked brain activity and WM, we found seed-based connectivity closely associated with task performance after regressing out the local activity of seed regions. These results fairly agreed with the basic logic of the background connectivity approach, that the task-induced BOLD signals contained two sources of variance (i.e., stimuli-related and state-related), with distinct and complementary information on cognitive processes (Otten et al., 2002; Rissman et al., 2004). Classic FC estimated by resting-state fMRI data has been best suited to studying the latent functional architecture of the brain at rest, however, it still lacks the ability to well depict the relationship of brain interactions to specific cognitive tasks (Friston, 2011; Turk-Browne, 2013; Misic and Sporns, 2016). In contrast, there have been several studies showing that during cognitive tasks such as visual attention and memory, background connectivity could selectively respond to specific stimulus types or cognitive processes (Summerfield et al., 2006; Al-Aidroos et al., 2012; Norman-Haignere et al., 2012; Wimmer and Shohamy, 2012; Duncan et al., 2014), and changes in connectivity could be detected even when cognitive demands were modulated by diseases (Lou et al., 2015) or drugs (Anticevic et al., 2012b). Thus, our results provided more evidence that brain activity and connectivity played independent but complementary roles in the process of WM ageing, and since age-related changes in WM made critical contributions to the ageing of other higher cognition such as memory and decision-making, this evidence could also be one possible neural mechanism underlying the ageing of broader cognitive functions.
Meanwhile, based on results from mediation modelling, that is, while a decline in activity was linked to the age-related decrease in AR and increase in the RT, the background connectivity was enhanced to support higher AR and shorter RT, we proposed that along with local processing impairment in advanced age, more regions were involved in the current WM task as a result of functional compensation (Reuter-Lorenz and Cappell, 2008; Park and Reuter-Lorenz, 2009). Specifically, although we could not clarify which alteration happened early, the compensational mechanisms might be the situation in which the age-related decline in activity led to the decline in behavioural performance, followed by the enhancement of background functional connectivity, to reduce the impact and to help older individuals maintain behavioural performance. In addition, our findings also revealed patterns that positively activated regions (IPL) during tasks were more connected to regions that were negatively activated (PCC), and connections were enhanced between regions that were negatively activated (ACC) and those that were positively activated (IPL and PreCG). Since these regions were the core regions in the frontoparietal network (FPN) and default mode network (DMN), the pattern might also catch a glimpse of network reconfiguration among brain networks (i.e., FPN and DMN) in ageing to support higher cognitive functions (Reuter-Lorenz and Lustig, 2005).
There are several points that could be explored in future studies. First, the block design of the current experiment could not fully consider correct and failed trials during the WM task, which made the impacts of regional activity and connectivity on successful WM difficult to assess. However, based on the decent accuracy rate of WM tasks (90.91 ± 6.04%) in all subjects, it could be inferred that they were focused on the tasks, which agreed with the hypothesis of maintaining cognitive status in background connectivity analyses. Second, the computation of functional connectivity was based on seed regions derived from regional activity analyses, which could have been computed at the whole brain level to examine the network reconfiguration or reorganization. Due to the primary proposal of the present study to jointly explore and compare the roles of brain activation and connectivity in WM ageing, the seed-based analyses could provide more direct evidence on their relationships and contributions to WM ageing. Finally, we assessed functional alterations within a cross-sectional sample. Although a group of elderly subjects avoided the possible biases of strategies or patterns in WM tasks, a longitudinal design would give an improved and more detailed depiction of the regional activity and connectivity alterations as well as their contributions to the ageing of WM.