Enhanced structural connectivity within a brain sub-network supporting working memory and engagement processes after cognitive training

Highlights

• Here we study brain changes after cognitive training within the connectome framework.

• Network-based statistics (NBS) and graph theoretical analyses were applied for studying the interaction groups × times.

• The training group show enhanced connectivity, strength and global efficiency.

• The identified network supports cognitive processes required by the training.

Abstract

The structural connectome provides relevant information about experience and training-related changes in the brain. Here, we used network-based statistics (NBS) and graph theoretical analyses to study structural changes in the brain as a function of cognitive training. Fifty-six young women were divided in two groups (experimental and control). We assessed their cognitive function before and after completing a working memory intervention using a comprehensive battery that included fluid and crystallized abilities, working memory and attention control, and we also obtained structural MRI images. We acquired and analyzed diffusion-weighted images to reconstruct the anatomical connectome and we computed standardized changes in connectivity as well as group differences across time using NBS. We also compared group differences relying on a variety of graph-theory indices (clustering, characteristic path length, global and local efficiency and strength) for the whole network as well as for the sub-network derived from NBS analyses. Finally, we calculated correlations between these graph indices and training performance as well as the behavioral changes in cognitive function. Our results revealed enhanced connectivity for the training group within one specific network comprised of nodes/regions supporting cognitive processes required by the training (working memory, interference resolution, inhibition, and task engagement). Significant group differences were also observed for strength and global efficiency indices in the sub-network detected by NBS. Therefore, the connectome approach is a valuable method for tracking the effects of cognitive training interventions across specific sub-networks. Moreover, this approach allows for the computation of graph theoretical network metrics to quantify the topological architecture of the brain network detected. The observed structural brain changes support the behavioral results reported earlier (see Colom, Román, et al., 2013)

Introduction

There is a growing interest in the study of changes evoked by cognitive training programs both at the behavioral (Au et al., 2015, Karbach and Verhaeghen, 2014, Schwaighofer et al., 2015, Weicker et al., 2016) as well as at the biological level (Buschkuehl, Jaeggi, & Jonides, 2012). Here we study changes in structural connectivity (white matter connection paths) after completing an adaptive cognitive intervention based on the n-back task (Colom, Román, et al., 2013). This approach allowed us to investigate training-induced changes in the connectome (Iturria-Medina, 2013, Rubinov and Sporns, 2010), which underlie brain plasticity (Tymofiyeva, Hess, Xu, & Barkovich, 2014), experience-dependent learning, and structural changes related to cognitive training outcome.

Taya, Sun, Babiloni, Thakor, and Bezerianos (2015) argue that studying changes in the connectome leads to important insights because (a) cognitive training might affect spatially distributed and functionally relevant sub-networks, (b) brain regions might be differentially impacted by training depending on their topological properties, and (c) the study of changes in the connectome allows for the analysis of changes in network efficiency. For example, Langer, von Bastian, Wirz, Oberauer, and Jäncke (2013) investigated changes in the topological architecture of the functional brain network via EEG after working memory training. Their main conclusion was that participants with the highest working memory training performance showed an increase in the small-worldness within a distributed fronto-parietal network (see also, Kundu, Sutterer, Emrich, & Postle, 2013), suggesting an increase in network efficiency.

Others have investigated changes in functional brain networks after cognitive training. For instance, Hempel et al. (2004) observed changes in brain activity (increases after 2 weeks of training, and decreases after 4 weeks) as a function of visuospatial working training. Olesen, Westerberg, and Klingberg (2004) reported an increase in activity within brain areas underlying working memory after 5 weeks of practice. Buschkuehl, Hernandez-Garcia, Jaeggi, Bernard, and Jonides (2014) used the n-back task. They found differences in brain activity comparing two difficulty levels (4-back vs.1-back), and furthermore, the observed brain activity was correlated with task performance. Similar effects have been observed in older adult populations. For example, Chapman et al. (2015) found training-related increases in global and regional cerebral blood flow, as well as greater connectivity in the default mode and the central executive network. Furthermore, there have been reports of longitudinal effects that were observed as long as one year after cognitive training (see Cao et al., 2016, Katz et al., in press).

Here, we focused on the effects of cognitive training on structural brain connectivity (SC). A topic that has rarely been addressed to date although there are some exceptions. Takeuchi et al. (2010) observed that after cognitive training, there was an increase in a general index of white matter integrity (fractional anisotropy, FA) in parietal and frontal cortices. In another study, training on working memory updating, mental set shifting, episodic memory, and processing speed tasks produced an increase in the white matter microstructure of the anterior section of the corpus callosum (Lövdén et al., 2010). Colom et al. (2012) compared a training group that played Professor Layton—a cognitively complex commercial videogame—with a control group, observing training-related increases in white matter integrity in the right hippocampal cingulum bundle and in the left inferior longitudinal fasciculus. Tang, Lu, et al. (2010) observed increased FA in the anterior cingulate cortex after 11 h of integrative body–mind training. Finally, Mackey, Whitaker, and Bunge (2012) reported decreased mean diffusivity (MD) within frontal and parietal cortices in participants who completed a training course for the Law School Admissions Test.

In the present study, we quantified anatomical connections among gray matter regions using a graph-based global tractography algorithm to compute individuals’ connectivity matrices (Iturria-Medina et al., 2007). We also applied network-based statistics (NBS) to analyze changes at the sub-network level (Zalesky, Fornito, & Bullmore, 2010). To date, NBS has been mainly applied to compare healthy individuals and patients (Bai et al., 2012, Verstraete et al., 2011, Zhang et al., 2011), and researchers have only recently used NBS within healthy populations (e.g. comparing bilinguals and monolinguals) (García-Pentón, Fernández, Iturria-Medina, Gillon-Dowens, & Carreiras, 2014). Furthermore, we computed graph-theory indices (characteristic path length, mean clustering coefficient, global efficiency, local efficiency and strength) to describe the overall network, and to reveal potential training-related changes in sub-networks. Finally, we calculated the relationships between changes at the network level and changes in the training task itself, as well as in four cognitive domains assessed before and after training: fluid intelligence, crystallized intelligence, working memory capacity and attention control.

We predicted that cognitive training would lead to increased connectivity within brain regions underlying fluid intelligence and working memory, given that the cognitive training paradigm was designed to engage these processes (Colom, Román, et al., 2013). Specifically, we expected frontal and parietal involvement in the sub-network as detected by NBS (Barbey et al., 2014, Burgess et al., 2011, Colom, Burgaleta, et al., 2013, Jung and Haier, 2007). In addition, we predicted training-related changes within the temporal cortex (especially middle temporal areas), given that we observed gray matter changes after training in this area before (Colom, Martínez, et al., 2016, Colom, Hua, et al., 2016, Román et al., 2016; see also Basten, Hilger, & Fiebach, 2015). Based on previous work, we also expected increased global efficiency (Eg), local efficiency (El) and strength (S) for the sub-network sensitive to cognitive training as a result of increased inter-regional communication related to training (Lövdén et al., 2010, Tang, Lu, et al., 2010). Additionally, we assumed that this sub-network would demonstrate more pronounced small worldness properties after cognitive training, especially in the case of training-related improvements in the efficiency of the sub-network. Therefore, we predicted increased clustering, (C) and decreased characteristic path length (L) (Sporns & Zwi, 2004). We did not have specific hypotheses regarding the correlations between changes at the psychological level and graph-theory indices, especially since previous findings have been contradictory (Lövdén et al., 2010, Takeuchi et al., 2010).