Decreased functional segregation occurs with aging (Chan et al., 2014; Damoiseaux, 2017; Oschmann and Gawryluk, 2020). Functional segregation typically refers to neural processing in regions with similar functions, which usually manifests as sparse inter-network connectivity and dense intra-network connectivity (Sporns, 2013; Sporns and Betzel, 2016). In older adults, impaired functional segregation (as illustrated in Figure 1A) in the resting state occurs in the TPN and is generally characterized by decreased intra-network connectivity and increased inter-network connectivity (Damoiseaux, 2017; Oschmann and Gawryluk, 2020). For example, compared with young adults, older adults have reduced functional connectivity strength within the ventral and dorsal attentional networks (Archer et al., 2016; Zonneveld et al., 2019). Also, over a 4-year follow-up period, older adults showed a gradual decline in frontoparietal network connectivity (Oschmann and Gawryluk, 2020). In terms of inter-network connectivity, compared to young adults, older adults have increased functional connectivity between the frontoparietal and dorsal attention networks, between the frontoparietal and ventral attention networks, and between dorsal and ventral attention networks (Ferreira and Busatto, 2013; Ferreira et al., 2016; Zonneveld et al., 2019). Taken together, there is a clear association between decreased functional segregation in the TPN and aging.

Preliminary evidence has suggested that decreased functional segregation of the TPN is associated with a decline in cognitive control with age. For example, in older adults, lower intra-connectivity within frontoparietal and attention network indicates worse cognitive control, and enhanced inter-connectivity between the frontoparietal and dorsal attention networks is associated with cognitive control decline (Sala-Llonch et al., 2012; Hausman et al., 2020; Setton et al., 2021). Functional segregation reflects the extent to which densely connected regions perform specialized cognitive functions (Wig, 2017; Koen and Rugg, 2019) and often indicate a well-organized spontaneous network. Decreased functional segregation might reflect an impairment in spontaneous network organization, which in turn manifests as cognitive control decline in older adults.

Functional integration (also called functional coupling) reflects the integrated processing of information during tasks and usually manifests as enhanced functional connectivity (Fornito et al., 2012; Cocchi et al., 2013; Sporns, 2013). Enhanced coupling can occur within a specific network or across networks. For example, connectivity within the frontoparietal network was reorganized as control-demands increased, reflecting the engagement of rule representation, conflict monitoring, and resolution (Cocuzza et al., 2020; Ray et al., 2020; Nee, 2021). For inter-network connectivity, enhanced functional integration between frontoparietal and attention networks ensures that the information captured by attention can be processed according to the target rules (Cocchi et al., 2013; Cai et al., 2021; Menon and D’Esposito, 2022). In addition, functional integration between frontoparietal and sensorimotor networks ensures that subjects can inhibit dominant responses and behave correctly according to target rules (Cocchi et al., 2013; Cole et al., 2013; Cocuzza et al., 2020). In summary, functional integration is the core neural representations during tasks requiring cognitive control.

Functional integration among the frontoparietal network may be an important neural indicator of the cognitive control ability during aging. In older adults, functional connectivity between prefrontal and parietal cortex, i.e., within the frontoparietal network, positively predicts their performance during tasks requiring cognitive control (Madden et al., 2010). Further, in an emotional working memory task, older adults recruit the frontoparietal network to ignore irrelevant distractors and encode negative emotional items, whereas young adults only adopt the frontal regions to encode both positive and negative emotional items (Ziaei et al., 2017). Similar to young adults, functional integration among the frontoparietal network supports the control functions in healthy older adults. Specifically, frontal regions are thought to engage in rule representation and control execution, whereas parietal lobes are thought to support rule reconstruction (Brass et al., 2005a,b; Ghanavati et al., 2019; Nee, 2021). Thus, decreased functional integration within the frontoparietal network may indicate decreased rule-dependent control in CCA.

Functional integration between frontoparietal network regions (e.g., the dlPFC) and task-specific regions is associated with CCA (as illustrated in Figure 1B). By adding face-based distractors to a delayed-recognition task, Clapp et al. (2011) explored task-based connectivity during working memory performance and observed decreased connectivity between the right dlPFC (called the middle frontal gyrus in their report) and parahippocampus in older adults. Also, during the switch-based task, the functional connectivity between the dlPFC and task-specific regions (e.g., cerebellum, thalamus) was lower in older adults than young people (Madden et al., 2010). In contrast, connectivity preservation between the dlPFC and task-specific regions (e.g., inferotemporal, left premotor cortex) was associated with better cognitive control performance in older adults (Nagel et al., 2011; Hakun et al., 2015). The above evidence suggests that decreased functional integration between frontoparietal network and task-specific brain regions is associated with the decline in cognitive control with age, but the pattern is influenced by tasks or specific subcomponents of cognitive control.

Finally, functional integration, as measured by network efficiency properties, declines with aging. Network integrative processes can be viewed as the communication efficiency measured by graph-theoretic properties (Sporns, 2013). In addition to the strength of functional connectivity, graph-theoretic coefficients such as rich clubs and hub integrity can be used to reflect the functional integration. The assortativity, a graph coefficient that reflects how regions connect to other regions with a similar degree, was lower in older adults during control tasks than at rest, indicating an aging-related decline in functional integration (Grady et al., 2016). In contrast, preserved functional integration was associated with better cognitive control performance. Specifically, both higher hub integrity of the dlPFC and anterior cingulate cortex (Pa et al., 2014), and greater integration of the dorsal attention network predicted better inhibitory control ability in older adults (Rieck et al., 2021). The graph-theoretic analysis offers an additional approach to elucidate CCA-related changes in functional integration.

Cognitive control relies on functional antagonism between large-scale networks. Functional antagonism differs from functional segregation. Functional segregation usually refers to positive functional connectivity, where blood-oxygen dynamics between regions are predominantly positively correlated, whereas functional antagonism refers mainly to negative connectivity, where the blood oxygen level-dependent signals between paired regions are predominantly negatively correlated (Fornito et al., 2012; Li et al., 2021; Demertzi et al., 2022). Functional antagonism can exist in both resting and task states and is mainly present between the TPN and DMN. In the resting state, the TPN is usually only slightly activated, while the DMN is highly activated, and there is a stable negative connection between the TPN (e.g., attention and frontoparietal network) and DMN (Raichle, 2015; Buckner and DiNicola, 2019). In a cognitive control task, the activation level of the DMN is decreased, while that of the TPN is increased; thus, the functional antagonism between these networks is maintained. Functional antagonism between the TPN and DMN positively predicts cognitive control performance in healthy adults (Kelly et al., 2008; Xin and Lei, 2015).

Declines in functional antagonism are important neurological indicators of the decreased cognitive control in older adults (as illustrated in Figure 1C). There is ample evidence that functional antagonism between the TPN and DMN decreases with aging, as evidenced by the decreased anti-correlation between activity in the DMN and in the ventral and dorsal attention network (Geerligs et al., 2015; Ferreira et al., 2016; Spreng et al., 2016; Zonneveld et al., 2019). Furthermore, even in healthy older adults, a large number of connections shift from negative to positive, implying impaired functional antagonism in the aging brain (Ferreira et al., 2016). Putcha et al. (2016) found that decreased functional antagonism between attention networks and the DMN was associated with cognitive control decline in older adults. The TPN antagonism supports the appropriate suppression of the DMN during a range of cognitive control tasks, which ensures that the external-oriented cognitive process is not affected by internal processes from the DMN (Raichle, 2015; Buckner and DiNicola, 2019). Thus, decreased DMN–TPN antagonism putatively indicates that the decrease in CCA is associated with enhanced internal inference from the DMN.

Aging-related changes in the functional connectivity of large-scale networks are important contributors to CCA. Specifically, decreased functional segregation, integration, and antagonism are associated with worse behavioral performance measured by control-based tasks in older adults (Madden et al., 2010; Putcha et al., 2016; Setton et al., 2021). Despite the debilitating effects of disrupted network connectivity, there is also evidence that functional integration is positively correlated with task performance in older adults (Nagel et al., 2011; Hakun et al., 2015). Thus, connectivity-related compensation, especially in terms of functional integration, could be examined in future studies. Functional segregation in the resting state is a hallmark of the well-organized spontaneous network organization, functional integration in the task state reflects collaborative processing of information, and functional antagonism ensures resistance to endogenous interference during cognitive control (Cocchi et al., 2013; Sporns, 2013; Wig, 2017; Buckner and DiNicola, 2019). Aging is accompanied by abnormalities in all three types of network interaction patterns, suggesting that decreased CCA is associated with disrupted spontaneous network organization, impaired information coordination, and increased endogenous interference. In both resting and task states, changes in the intra- and inter-connectivity of the frontoparietal network states play a central role in decreasing CCA (Madden et al., 2010; Sala-Llonch et al., 2012; Setton et al., 2021). Finally, it is known that maintenance of functional connectivity depends on structural connections, consisting of white matter fibers. How does the white matter structural network change with age? How do functional and structural networks interact with and participate in CCA? In the next section, we introduce structural evidence to elucidate the network connectivity mechanisms related to CCA further.

Age-related structural changes in the white matter involve both decreases in white matter volume and impairment of structural connections (Damoiseaux, 2017; Farokhian et al., 2017). Voxel-based morphometry analysis of T1-weighted images revealed a 26% decrease in the total white matter volume from the age of 30 years to the age of 90 years (Jernigan et al., 2001). Compared to the occipital and temporal lobe regions, the frontal white matter in the anterior hemisphere is the most vulnerable to age-related loss due to aging (Gunning-Dixon et al., 2009). In addition, several studies have analyzed diffusion tensor imaging and diffusion-weighted imaging using white matter fiber-tracking techniques to examine the relationship between structural connectivity and aging. These results showed that the structural connectivity of the cingulate gyrus, cuneus, precuneus, superior frontal gyrus, and parietal lobe gradually decreased with age, suggesting brain-wide structural connectivity impairment (Madden et al., 2012; Hirsiger et al., 2016; Damoiseaux, 2017; Liu et al., 2017).

Both white matter loss and structural disconnection suggest a decrease in neural signaling capacity and are associated with decreased cognitive control (Charlton et al., 2010; Ystad et al., 2011; Madden et al., 2012). In older adults, volume loss in the frontal, inferior frontal, and parietal white matter predicts declines in multiple subcomponents of cognitive control, including working memory, inhibitory control, and cognitive flexibility (Charlton et al., 2010; Madden et al., 2012). In addition, lesions of the superior longitudinal fasciculus can predict cognitive flexibility decline in older adults (Gold et al., 2010). Notably, white matter lesions associated with decreased CCA are mainly concentrated in the frontal and parietal lobes (Charlton et al., 2010; Madden et al., 2012), which overlap spatially with the frontoparietal network. In addition, structural connectivity between subcortical regions (i.e., the thalamus and nucleus accumbens) and regions within the DMN and dorsal attention network positively predicts cognitive control in older adults (Ystad et al., 2011). The structural connectivity between the thalamus and pre-supplementary motor region is also positively correlated with inhibitory control in older adults (Coxon et al., 2012). In summary, both aging-related regional and connectivity damage to the white matter play a role in the decline of cognitive control with age.

Functional networks may serve as mediators of the cognitive control decline triggered by structural network damage. The impairment perspective proposes that damage to structural connections can weaken functional connectivity. Studies have shown that structural connections support functional connectivity in healthy older adults. For example, Andrews-Hanna et al. (2007) found that fractional anisotropy (a measurement of white matter integrity) of the anterior–posterior fiber was positively correlated with the functional connectivity between hub regions within the DMN, including the medial prefrontal cortex and PCC. Other studies have also shown that the integrity of the fornix and cingulum bundle can positively predict functional connectivity (e.g., cortex–subcortical connectivity) (Kehoe et al., 2015; Fjell et al., 2016). Not surprisingly, damage to structural connections results in reduced functional connectivity. A longitudinal study suggested that age-related impairments in the structural connectivity of the cingulum bundle trigger decreased cortical–subcortical functional connectivity (Fjell et al., 2016). Since functional connectivity supports the collaborative processing of information in control tasks, the pathway “structural connectivity damage → functional connectivity damage → cognitive control decline” may be a valid concept in terms of CCA (Figure 2A, damaged pathway).

However, the damage perspective cannot explain some of the existing empirical results and may lead to a view of functional connectivity compensation (Figure 2A, functional compensation). Specifically, while lesions in structural connectivity can trigger cognitive control decline in older adults, increased functional connectivity can attenuate cognitive decline associated with structural damage. Benson et al. (2018) reported that increased functional connectivity of the frontoparietal and attention networks counteracted cognitive control declines associated with white matter lesions. Therefore, white matter lesions do not necessarily trigger functional connectivity impairment. In older adults, the TPN can maintain cognitive function as much as possible by enhancing functional connectivity. Why do structural connectivity lesions in the aging brain not necessarily weaken functional connectivity? What is the structural basis of compensatory enhancement of functional connectivity? The possible explanation of above questions is discussed in the following paragraph.

A relatively intact structural network may be a prerequisite for functional connectivity compensation, while severe lesions in the structural network can trigger reduced functional connectivity. When considering a network as a whole, a marked similarity between functional and structural networks can be observed (Figure 2B; Wang et al., 2015; Suo et al., 2021). Structure–function coupling reflects the support and restraint of structural networks on functional networks (Suárez et al., 2020; Bazinet et al., 2021; Suo et al., 2021). However, structural and functional connections do not strictly correspond to each other, which reflects the adaptive adjustments of specific connections in the human brain (Suárez et al., 2020). On the other word, functional connectivity can be built on third-party structural pathways. For older adults, when the structural network is relatively well preserved, lesions of specific structural connections do not directly result in damage to functional connectivity, and the preserved third-party structural pathways ensures the keep of functional connectivity. However, severe structural network damage makes it difficult to rely on third-party structural pathways to maintain the initial functional connectivity and results in cognitive control decline in older adults. For example, Burianova et al. (2015) selected hyperactivated regions in working memory tasks to define the compensatory network and found that integrity of the bilateral frontoparietal tracts positively predicted the strength of functional connectivity of the compensatory network as well as task performance. Notably, although functional connectivity can be maintained when the structural network is relatively intact, the functional connectivity built on a third-party structural pathway may have reduced communication efficiency due to the increased length of the structural route. This can be explained using a metaphor: when the direct road to the destination is blocked, one can take a detour to get to the destination, but the total distance is longer. The aging brain may recruit more neural synchronization to offset the reduced communication efficiency and manifest as increased functional connectivity.

Structural and functional networks show complex interactions affecting CCA. Because structural connections support and constrain functional connections, it could be concluded that structural connection damage triggers cognitive control decline by weakening functional connectivity. However, when the structural network is locally damaged but globally preserved, the effects on functional connectivity may be compensated through third-party structural pathways and may counteract the cognitive control decline associated with the structural damage. Functional compensation determined by preservation of the structural network should be considered in future explorations of network mechanisms underlying CCA.

Common and unique network mechanisms underlie cognitive control. Cognitive control can be divided into three core subcomponents: working memory, inhibitory control, and cognitive flexibility (Diamond, 2013; Zink et al., 2021). All three core subcomponents are based on top–down control but have unique cognitive mechanisms. For example, working memory is related to the representation and extraction of information, whereas inhibitory control is mainly associated with reducing interference and highlighting the target (Diamond, 2013). Imaging evidence suggests that the three subcomponents rely on the frontoparietal network to implement conflict monitoring, resolution, and goal-directed processes (Cocchi et al., 2013; Zanto and Gazzaley, 2013; Qiao et al., 2020). However, the frontoparietal network dynamically rewires functional connections with regions involved in other networks according to task rules, forming component-specific network mechanisms (Cole et al., 2013; Zanto and Gazzaley, 2013). For example, working memory involves increased functional connectivity between the frontoparietal network and memory-related brain regions (Yamashita et al., 2018; Cai et al., 2021), while responsive control is associated with enhanced functional connectivity between the frontoparietal and sensorimotor networks (Cocchi et al., 2013).

Aging-related network mechanisms underlying the different subcomponents of cognitive control may be both shared and unique. Declines in functional connectivity within the frontoparietal network are associated with decreased CCA (Sala-Llonch et al., 2012; Setton et al., 2021); however, compensatory enhancement of functional connectivity within the network counteracts the adverse effects of aging (Benson et al., 2018). In addition, white matter lesions within the frontoparietal network predict declines in multiple aspects of cognitive control (Charlton et al., 2010; Madden et al., 2012). Multimodal imaging evidence points to common mechanisms underlying CCA in the frontoparietal network. Moreover, aging-related changes in inter-network connectivity show task-specific patterns, suggesting the existence of unique network mechanisms related to CCA. Notably, previous studies used different MRI data pre-processing protocols, and network construction and functional connectivity analysis methods (Stanley et al., 2015; Tsvetanov et al., 2018); therefore, it is impossible to compare results across studies directly. In future, datasets from the same cohort of older adults, addressing multiple cognitive control components, will be needed to explore this issue.

Multimodal network mechanisms for CCA are mostly theoretical, and empirical studies are rare. The impairment perspective suggests that CCA follows the pathway of “structural connectivity impairment → functional connectivity impairment → cognitive control decline.” However, the compensation view suggests that if the structural network is mostly preserved, the compensation of functional connectivity can be occurred through third-party structural pathways, as manifested by increased functional connectivity and cognitive control maintenance. When the structural network is severely disrupted, the functional network will be damaged and result in declines in cognitive control. Thus, structural and functional networks act together in a complex way in CCA, and the degree of structural network impairment may be an important a priori variable. In future, researchers should systematically collect multimodal imaging data, select targeted connections or networks, and construct mediating and moderating multivariate models to clarify how structural and functional networks interact and influence CCA.

Notably, communication models and multilayer network analyses that emerge from network science would offer approaches to explore the structure–function coupling of brain networks (Bassett and Sporns, 2017). By formulating models of multimodal connectivity, these methods hinder the non-independence of multilevel connectivity measures and assess the extent to which the biologically realistic model conforms to the properties of the functional network (Kivela et al., 2014; Muldoon and Bassett, 2016; Bassett and Sporns, 2017; Suárez et al., 2020). These methods can characterize the similarity of structural and functional networks at the macro level and will allow an integrated assessment of the supporting and constraining effects of structural networks on functional networks, thereby offering a new approach for exploring multimodal network mechanisms underlying CCA.

Whether changes in dynamic functional connectivity are involved in decreased CCA should be examined. Dynamic functional connectivity refers to the time-varying fluctuations of functional networks (Hutchison et al., 2013; Allen et al., 2014; Calhoun et al., 2014). Cognitive control has been found to be associated with dynamic functional connectivity (Hutchison and Morton, 2016; Nomi et al., 2017). For example, Nomi et al. (2017) found that the frequency of an asynchronous brain state was associated with control-based tasks that require flexible cognition. Besides, aging-related changes in dynamic functional connectivity have been verified (Cabral et al., 2017; Viviano et al., 2017). Nevertheless, investigations of CCA have largely taken into account the connectivity properties from a static perspective. Based on the available studies, it is difficult to form specific conclusions about whether and how aging-related changes in dynamic functional connectivity are involved in decreased CCA. Hence, dynamic approaches of functional connectivity, such as sliding-window methods (Hutchison et al., 2013), should be adopted in future studies.