Plasticity in brain activity dynamics after task-shifting training in older adults

Highlights

• Investigation of neural changes after task-shifting training in old age.

• Methodically applying an hybrid fMRI design to capture brain activity dynamics.

• Training promoted changes in transient mechanisms in fronto-parietal networks.

• Training promoted overall reductions of sustained activation in striatal circuits.

• Spatio-temporal interactions may underlie training-related brain activity changes.

Abstract

Cognitive control is supported by a dynamic interplay of transient (i.e., trial-related) brain activation across fronto-parietal networks and sustained (i.e., block-related) activation across fronto-striatal networks. Older adults show disturbances in this dynamic functional recruitment. There is evidence suggesting that cognitive-control training may enable older adults to redistribute their brain activation across cortical and subcortical networks, which in turn can limit behavioral impairments. However, previous studies have only focused on spatial rather than on temporal aspects of changes in brain activation. In the present study, we examined training-related functional plasticity in old age by applying a hybrid fMRI design that sensitively tracks the spatio-temporal interactions underlying brain-activation changes. Fifty healthy seniors were assigned to a task-shifting training or an active-control group and their pretest/posttest activation-change maps were compared against 25 untrained younger adults. After training, older adults showed the same performance as untrained young adults. Compared to the control group, task-shifting training promoted proactive (i.e., early, cue-related) changes in transient mechanisms supporting the maintenance and top-down biasing of task-set representations in a specific prefrontal circuitry; reactive (i.e., late, probe-related) changes in transient mechanisms supporting response-selection processes in dissociable fronto-parietal networks; overall reductions of sustained activation in striatal circuits. Results highlight the importance of spatio-temporal interactions in training-induced neural changes in age.

Introduction

A hallmark of human cognition is the ability to adapt thoughts and actions according to rapidly changing environmental contexts. To deal with such contextual changes, we rely on a set of cognitive-control processes, such as representing and maintaining task rules in working memory (WM) and flexibly shifting among them according to changing rule priorities (Friedman and Miyake, 2017; Miyake and Friedman, 2012). It is a well replicated finding that older adults show impairments in tasks that put demands on such cognitive-control processes (for a review, see Kray and Ferdinand, 2014). Meanwhile, it is also known that targeted cognitive training can reduce deficits in cognitive control even into very old age (for a recent meta-analysis, see Karbach and Verhaeghen, 2014). However, only a few studies so far have investigated whether improved cognitive control processes after training are associated with neural changes in brain activation in older adults, and their outcomes are controversial (for a review, see Simons et al., 2016). Contributing to an ongoing debate, the general aim of the present study was to specify the neural mechanisms that support training-induced neural plasticity in cognitive control in older adults.

There is ample empirical evidence for age differences in cognitive control processes that comes, for instance, from the task-shifting literature (for a review, see Kray and Ferdinand, 2014; for a meta-analysis, see Wasylyshyn et al., 2011). In the task-shifting paradigm, participants are instructed to continuously switch between two or more simple cognitive tasks (for an example, see Fig. 2). In addition to such mixed-task blocks, some researchers also measure the performance of each task in isolation in so called single-task blocks (e.g., Kray and Lindenberger, 2000). This allows calculating two types of costs. Mixing costs (also termed global or general shifting costs) are defined as differences in performance between mixed-task and single-task blocks. Mixing costs are assumed to arise if participants have to maintain and alternate between two overarching task-rule sets, such as in mixed-task blocks, as compared to performing only one task in isolation, such as in single-task blocks. Shifting costs (also termed local or specific shifting costs) are defined as differences in performance between shift and repeat trials within mixed-task blocks. These costs are assumed to reflect the ability to disengage from a previous task rule and to shift the priority to another task rule. Hence, mixing and shifting costs refer to different cognitive control processes; one is operating on an abstract set of multiple task rules, and the other is changing the currently relevant rule priority.

Previous studies have shown that for most variants of the task-shifting paradigm, older adults reliably show larger mixing costs and sometimes larger shifting costs than younger adults (Kray and Ferdinand, 2014; Kray and Lindenberger, 2000). However, behavioral cost differences have not always been consistent. For instance, Kray et al. (2002) applied a cued-variant of the task-shifting paradigm in combination with a response deadline and found age differences in shifting costs but not mixing costs. In the present study, we also applied a cued task-shifting variant to measure age differences in cognitive control.

We know that older adults are sensitive to training demands on cognitive control as they showed a larger reduction of performance costs after task-shifting training than after single-task training, and this effect was larger for older than for younger adults (Karbach and Kray, 2009; Kray and Fehér, 2017). Hence, task-shifting training can reduce age differences in task-shifting costs and leads to a compensation of cognitive control deficits. Similarly, in the present study, we expected that task-shifting training should enable larger performance improvements (i.e., larger reductions of behavioral mixing and shifting costs) than single-task training in older adults.

Training-induced changes in older adults were even larger for more demanding training conditions, such as training variability (cf. Karbach and Kray, 2009) or high ambiguity between the training tasks (Kray and Fehér, 2017). Therefore, we applied variable task-shifting training coupled with ambiguous task settings in the present training study.

Of most interest in the present study were training-induced neural changes in cognitive control networks in older adults. Cognitive-control processes are assumed to be supported by distributed fronto-striatal, cingulo-opercular and fronto-parietal networks (Dosenbach et al., 2008; Dosenbach et al., 2006, 2007; Gazes et al., 2012; Gold et al., 2010; Kim et al., 2012; Nee et al., 2013). A key region among these networks is the prefrontal cortex (PFC¹) due to its unique ability to integrate multiple pieces of internal and external information into coherent representations that guide context-appropriate behavior.

The integrative function of the PFC is enabled by a dynamic interplay of many sub-regions serving distinct functions along a rostral-caudal abstraction gradient (Badre & D'Esposito, 2009; Bunge and Zelazo, 2006; O'Reilly, 2010): Anterior PFC portions, such as the frontal pole (FP), encompassing Brodmann Area (BA) 10, and the anterior part of the cingulate cortex (ACC), including BA 6, 8, and 32, have been associated with the representation and maintenance of the global task context, such as required in a mixed-task situation during task shifting (Braver et al., 2003; Jimura and Braver, 2009; Nee and Brown, 2013). In contrast, shifting between currently relevant tasks has been linked to more posterior PFC portions (Montojo and Courtney, 2008; Nee and Brown, 2013); that is, to the lateral PFC, which stretches along the dorsal (dlPFC, BA 9_posterior, 46) and ventral parts (mid-vlPFC, BA 44, 45) of the inferior frontal sulcus, and to the inferior frontal junction (IFJ), a small insulated area in the transition zone between prefrontal and premotor cortices (at the intersection of BA 6, 8, 9, and 44; Derrfuss et al., 2005; Muhle-Karbe et al., 2015).

Importantly, the seminal work of Nee and Brown (2013) indicated that this hierarchical organization within the PFC is differentially modulated by the basal ganglia (BG), including the putamen and the caudate head, and the posterior parietal cortex (PPC), that can be subdivided into its superior (SPL; BA 5, 7) and inferior lobules (IPL; BA 39, 40). More specifically, the authors revealed dissociable functional roles of a PFC-BG and a PFC-PPC loop. While the PFC-BG is associated with the representation and maintenance of the global task context, the PFC-PPC loop is involved in the shifting between relevant task rules.

This is of particular interest for the present study as we already know that the fronto-striatal and the fronto-parietal circuits are vulnerable to early and disproportionate age-related atrophy (Hedden and Gabrieli, 2004). Moreover, both show altered task-related activations (e.g., Grady, 2008; Schneider-Garces et al., 2010) as well as reduced functional connectivity between the involved network hubs in older age (Geerligs et al., 2014; Klostermann, Braskie, Landau, O'Neil and Jagust, 2012). In the present study, we will therefore focus on training-induced neural changes after task-shifting training within these two age-sensitive circuitries by applying a region-of-interest (ROI) based approach.

A variety of studies applying high-demand training of cognitive control have revealed functional plasticity in younger adults (Erickson et al., 2007b; Hempel et al., 2004; Jolles et al., 2012), while empirical evidence, to date, for training-induced functional brain changes in older adults is scarce. However, one common finding, also in older adults, is a change in the spatial distribution of brain activation across canonical control networks (Erickson et al., 2007a; Heinzel et al., 2016), such as hemispheric asymmetry reductions in neural processing (HAROLD account; Cabeza et al., 2002) or posterior-anterior brain activation shifts (PASA account; Davis et al., 2008) in high-functioning (or well-practiced) older adults (for a recent review, see Martins et al., 2015). Here, we were specifically interested in training-induced redistribution of brain activity after task-shifting training in older adults within the PFC-BG and the PFC-PPC circuitries.

The aforementioned HAROLD or PASA accounts focus exclusively on a spatial redistribution of brain activation to compensate for age-related losses. A recent theory called the temporal hypothesis for compensation (THC; alternatively termed as ‘early to late shift in aging’ (ELSA); see Dew et al., 2012) instead emphasized not only “WHICH regions of the brain show increased activation, but WHEN (…) [or HOW LONG]” these regions are increasingly activated (Martins et al., 2015, p. 10–11). In this line, it is assumed that hierarchical cognitive control functioning is also defined temporally by a distinction between transient and sustained time scales (Koechlin et al., 2003). Transient timescales underlie processes that span a short-lasting temporal episode, such as a task trial, while sustained timescales underlie temporally extended processes, enduring over a longer time interval, such as a task block. Only hybrid epoch-/event-related fMRI designs are suited to dissociate such temporal dynamics (Visscher et al., 2003). We therefore employed a hybrid fMRI design in the present study to capture training-induced changes in the spatio-temporal dynamics of cognitive control in the aging brain.

In this line, theoretical frameworks, such as the context-processing account of Braver and Barch (2002), emphasized that it is the interaction between spatial and temporal dynamics of brain activations that may enable successful cognitive control (also in older adults). The researchers found activation peaks in the sustained processing stream (a) in the anterior PFC that has been attributed to the stable maintenance of overarching context representations; (b) in the ACC associated with supervisory monitoring of PFC functions and the detection of ongoing conflict (see also Botvinick et al., 2001); and (c) in the BG, where an interplay between sustained and transient dopamine (DA) signals² may serve as a gate regulating the entrance of task-relevant information into the PFC (Gruber et al., 2006). According to their recent dual-mechanisms of control (DMC) theory (Braver, 2012), this sustained activation in more anterior regions corresponds to early top-down modulations of ongoing processes (proactive control mode). In contrast, activation peaks in the PFC-PPC loop (Braver et al., 2003; Jimura and Braver, 2009), especially in the lateral PFC and the SPL (Richter and Yeung, 2014), correspond to transient processes. While in the DMC theory sustained activation strictly serves a proactive control mode, transient activation can be flexibly utilized in both a pro- but also a reactive manner, depending on whether it is invested right after cue onset or only after probe onset. The latter reactive control mode refers to a late correction mechanism that depends on the bottom-up information from encountered stimuli (Braver, 2012).

Older adults show reduced amplitudes of sustained brain activation in the FP and ACC during task shifting (Jimura and Braver, 2009) as well as generally reduced sustained DA in the striatum (Braver et al., 2005), even in the absence of age differences in behavioral performance costs. Such alterations have been attributed to losses in sustained activation resources with increasing age (Dennis et al., 2007). In contrast, transient activation mechanisms, such as within the PFC-PPC circuitry, appear to be largely preserved in old age, or are only partly affected on their cue-related (i.e., proactive) component (Paxton et al., 2006). Hence, although a proactive control mode has been speculated to be more effective in most cognitive tasks for younger adults, shifts towards more transient and reactive control have been observed in older adults (Braver et al., 2009; Braver and West, 2008). Again, this idea fits nicely with empirical evidence from the task-shifting literature showing that older adults that rely more heavily on the recruitment of preserved late and transient activation in the PFC-PPC network achieved similar task-shifting performance as younger adults (Jimura and Braver, 2009; Madden et al., 2010).

In this line, in the present study, we were specifically interested in whether task-shifting training would lead not only to a change in brain activation levels across canonical networks but also to a change of temporal dynamics in a way of promoting a transient at the expense of a sustained control mode (i.e., a sustained-to-transient crossover shift in brain-activation dynamics). We assumed less specific changes of sustained brain activation but a dynamic sustained-to-transient crossover shift, such that we expect to see a reduction of sustained activation paralleled by an enhancement of transient activation, that should be more pronounced after task-shifting training than after single-task training.

To summarize, our main goal was to investigate the spatio-temporal interactions underlying training-induced changes after task-shifting training compared to single-task training in older adults. Therefore, we combined a hybrid epoch-/event-related fMRI design (temporal sensitivity) with an ROI-based analysis approach (spatial sensitivity).

Derived from the above-mentioned literature, our research questions were as follows:

• Does task-shifting training enable greater performance improvements than single task training in older adults?

• Is brain activation associated with transient mixing and/or shifting costs within the PFC-BG and/or the PFC-PPC circuitries more strongly modulated by task-shifting than single-task training?

• Is sustained brain activation within the PFC-BG and/or the PFC-PPC circuitries more strongly modulated by task-shifting than single-task training?

On the behavioral level, we predicted that task-shifting training should enable larger performance improvements than single-task training in older adults (i.e., larger reductions of behavioral mixing and shifting costs).

On the neural level, we predicted, first, that the transient mixing- and/or shifting-costs activation within the PFC-BG and/or the PFC-PPC circuitries would be more strongly modulated by task-shifting than by single-task training (Braver et al., 2009; but see Paxton et al., 2006). Second, we predicted less specific changes of sustained brain activation but a dynamic sustained-to-transient crossover shift, suggesting temporal changes, that should be again more pronounced after task-shifting training than after single-task training (Jimura and Braver, 2009; Madden et al., 2010).

In two complementary analyses, we aimed to explore the nature of any brain-behavior relationships and to compare the post-intervention neural patterns of older adults with those of younger adults to capture potential effects of age-related compensation.