neuroimaging of connectome changes after working memory training

Posted comment on ´Dynamics of the Human Structural Connectome Underlying Working Memory Training` by K. Caeyenberghs, C. Metzler-Baddeley, S. Foley and D.K. Jones published in The Journal of Neuroscience, 6th April 2016, 36(14) p. 4056


Neuroimaging studies of the brain normally involve showing functional areas of the brain and that functioning responding to some change in condition. The work of Caeyenberghs and colleagues is no different except they have found that the results of neuroimaging studies relating to memory capability pre- and post-cognitive training can be dependent on the metrics used. Most of the previous research looking at this topic uses diffusion tensor MRI studies and the metric of fractional anisotropy (FA) looking at white matter and results obtained are inconsistent. In Caeyenberghs and colleagues study 40 healthy participants underwent either an adaptive training program for working memory (Cogmed – 45 sessions; 40 sessions in total with training for verbal and spatial memory) or non-adaptive training. The participants were assessed using MRI neuroimaging and computerized working memory and executive function tests.

The neuroimaging techniques used in the reported study combined well established (although nonspecific) diffusion tensor MRI metrics with both MRI relaxometry-based metrics (an indirect measure of myelin, but corrected for motion and distortion artifacts for example) and metrics derived from advanced models of diffusion tensor studies (gives estimates of axonal density and corrected for distortions induced by the diffusion-weighted gradients and motion of the head for example). Quantitative maps of axonal morphology were constructed using the CHARMED protocol and maps of myelin level using the mcDESPOT protocol. A total of eleven different kinds of networks were generated with a network being defined as a set of nodes denoting anatomical regions and interconnecting edges denoting undirected tractography-reconstructed fiber trajectories interconnecting the nodes. These were weighted using the mcDESPOT protocol. Network areas included the inferior and superior parietal cortex, supramarginal gyrus, caudal and rostral middle dorsolateral prefrontal cortex, superior frontal cortex, inferior ventrolateral prefrontal cortex, insula and anterior cingulate cortices and the subcortical regions of the basal ganglia i.e. caudate, putamen, globus pallidum and thalamus. The volumes of grey matter of 30 regions of interest were used to construct structural correlation networks. FreeSurfer was used for cortical reconstruction and volumetric segmentation reconstruction of the brain’s surface to compute cortical thickness. The authors then performed graph and theoretical analyses to obtain their imaging results.

In order to ascertain the possible relationship between various cognitive skills in their behavioural testing methods, Caeyenberghs and colleagues ran prior to training test exploratory principal component analysis and this showed three significant behavioral components that together amounted to 59% of the total variance. The first component, complex span working memory, accounted for 34% and related to all of the tasks in which information had to be actively maintained in short-term memory, eg. the automated symmetry span task, the spatial span task, and the odd one out task. The second component accounting for 13% of the variance was associated with tasks involving a verbal component and included the double trouble task, digit span tasks, and the grammatical reasoning task. Tasks requiring general reasoning including the Hampshire tree task and the self-ordered spatial span, related to the third component and this accounted for 12% of the variance. Statistical analyses of the results were carried out by combining the scores.

Caeyenberghs and colleagues found that the Cogmed training program carried out produced positive cognitive changes. The main effect of time (pre and post-training) and group (adaptive and non-adaptive training) produced significant results, but there was no significance between the three cognitive skills (ie. complex working memory, verbal or general reasoning). Both sets of results for time against group and time against cognitive skill were significant and a three-way interaction between factor, group, and time was also found to be significant. Varying levels of performance improvement were obtained with the highest levels for the complex working memory and the verbal component in the post-training session for the adaptive group compared to the non-adaptive group.

Using graph theoretical network analysis of the working memory training effects, the authors found that there were significant changes for the interaction of group against time using the mcDESPOT protocol. This demonstrated that there had been an increase in global efficiency between the nodes of the network in the adaptive group from pre- to post-training. Marginal significant interaction effects were observed for the global efficiency of the graphs weighted by different diffusion-derived parameters, including FA, 1/mean diffusivity (MD), axial diffusivity, tissue volume fraction, 1/radial diffusivity, and a number of streamlines. The parameter that best captured the effect was the relaxation rate. No significant interaction effects were observed for the graph weighted by the total restricted fraction derived from the CHARMED protocol, or in the graph derived from the covariance of gray matter volumes.

Analysis of the neuroimaging results obtained for different regions led to the identification of the nodes responsible for the effects of the working memory training. The authors found that right anterior rostral cingulate gyrus, an area associated with attentional control and mental effort, showed a significant group against time interaction. Using a different method, significant results for a group against time interaction effect was also achieved for the right inferior ventrolateral prefrontal cortex which is associated with attentional orienting processes. These observations were supported by post hoc two-sided t test results.

Caeyenberghs and authors also found that correlation analyses between the changes in global efficiency from prior training periods to training and afterwards and the composite scores of the behavioral parameters showed little direct association between changes in structural network metrics and the improved performance on the cognitive tests. However,  using an exploratory uncorrected threshold of p < 0.05, correlations were observed between the changes in Cogmed tasks performance and changes in global efficiency of the R1-weighted networks. This was associated with better working memory performance on the Cogmed pairing with higher efficiency of information transfer (i.e. more global integration). None of the correlations were significant when the necessary correction for multiple comparisons was carried out.

The scores of global efficiency of the networks constructed with different metrics as ´connection strengths` were found to be highly intercorrelated at the baseline and therefore the authors interpreted this as representing non-independent observations. For example, the global efficiency of the network whose connections strengths were defined by the quantitative relaxation rate R1 (1/T1) derived from the mcDESPOT protocol correlated strongly with global efficiency of the network weighted by the TRF derived from the CHARMED protocol. However, the difference scores in global efficiency of the networks weighted by R1 were found not to correlate significantly with difference scores in the efficiency of the network weighted by the TRF metric or MWF-weighted network. Therefore, Caeyenberghs and authors concluded that although all metrics correlated at the baseline and that the reduction in correlation post-training suggested that the white matter network underwent changes during training, these changes are better detected with R1 than the other metrics tested. Since the different metrics relate to different aspects of the white matter microstructure and because relaxation times T1 and T2 are affected by changes in water, lipid, and protein content, T2 by iron within the oligodendrocytes and MWF by lipid myelin content, the changes observed with training in Caeyenberghs and colleagues study are suggested as being linked to alterations in these cell components.

Therefore, Caeyenberghs and authors showed in their neuroimaging study that changes occur in the structural connectome as a result of adaptive cognitive training. These changes relate to improved performance of working memory and verbal tasks and less so the far-transfer tasks involving general reasoning. The positive performance changes relate to increased global efficiency and there are likely to be white matter changes as shown by relaxation-rated network increased neuroimaging sensitivity. Since the authors discovered that some MRI metrics are not ideal for this type of neuroimaging of these particular cognitive skills (eg. techniques normally used for observing global efficiency changes rely on FA or MD), Caeyenberghs and authors concluded their paper by emphasizing the need for using specific microstructural markers for this type of experimentation.


What makes this paper and others like it interesting is to see the shift over the years of emphasis in neuroscience from laboratory test-tube experiments of long-term physiological changes in single brain area samples to the current fields of neuroimaging of real-time neuronal firing and functional networks. This gives another dimension to cognitive research and this paper takes advantage of these modern real-time techniques to demonstrate how training can affect neuronal cell firing and neural networks. However, every technique has its problems and drawbacks and neural imaging is no exception with the authors here demonstrating that not all metrics are ideal for every cognitive situation and that some can mask effects that would normally be seen or give results different to those obtained by other means. Coupled with these experimental problems are others relating to the fact that living subjects are being used and so with all such experiments of this nature stress, anxiety, timing, previous medication etc. can all affect the imaging results obtained. Even providing enough participants to produce significant results can be a problem.

However, the results obtained from neuroimaging can expand the neuroscientific knowledge front and what was observed here and in other studies of this nature are the functioning changes in neuronal firing and neural networking relating to cognitive training. The authors here found a positive training effect suggested as a result of a positive effect on axonal connectivity within certain areas and between certain areas. Caeyenberghs and authors identified 11 different kinds of networks and found a significant result for the group containing the right anterior rostral cingulate gyrus, an area normally associated with attentional control and mental effort. Using a different method significant results were also obtained for a group with the right inferior ventrolateral prefrontal cortex, an area normally associated with attentional orienting processes and decision-making. These results supported observations by Dreher and Grafman who investigated task switching and dual-task performance using fMRI. They showed that performing two tasks successively or simultaneously activated a common prefrontal-parietal neural network relative to performing each task separately. Performing two tasks simultaneously brought about activation in the rostral anterior cingulate cortex whereas switching between two tasks activated the left lateral prefrontal cortex and the bilateral intra-parietal sulcus region. The results were interpreted as indicating that the rostral anterior cingulate cortex serves to resolve conflicts between stimulus–response associations when performing two tasks simultaneously (attentional control and mental effort), whilst the lateral prefrontal cortex dynamically selects the neural pathways needed to perform a given task during task switching (attentional orienting processes).

The involvement of more brain areas during specific tasks and the effect of training has also been observed by other researchers, too. In August 2015, a post was published on this blog about the Cogmed program and a meta-analysis of the results which had been performed by Spencer-Smith and Klingberg, the authors of the paper. In their paper, they reported an average improvement of 16% after participation in the Cogmed training program independent of participant health status. The repetitive nature of the training program with participants learning through adaptation during the program time span, applying routine, use of memory chunking, improving attention and concentration, and taking note of feedback all possibly providing reasons why the training program led to cognitive improvement. The authors also noted that there was a greater effect on visuospatial memory than verbal suggesting that the improvement could be associated to increased working memory. The improvements obtained were also sustained for 2-8 months after training had finished.

From a neuroscientific point of view, studies have shown that training increases connectivity in frontoparietal and parietal occipital networks (Kunden). We also know that working memory requires acetylcholine and glutamate and the involvement of many brain areas. For example, neurons in the prefrontal cortex are associated with multi-tasking, working memory and attention (Messinger) and visual working memory requires activity in the inferotemporal cortex, V4, medial temporal cortex, prefrontal cortex and globus pallidus, lateral infero parietal cortex (guided eye movements in attention), post-parietal cortex (Koenigs – manipulation of information in working memory), as well as fronto-hippo connectivity (Cordesa-Cruiz). Working memory activity sees changes in prefrontal oscillations with  theta oscillations increasing during temporal order maintenance and alpha oscillations increasing over the posterior parietal and lateral occipital regions for item maintenance (Hsieh) with these alpha oscillations being used to maintain the relevant memory contents rather than suppressing unwanted or no longer relevant memory traces (Manza). In spatial working memory, theta oscillations occur in the medial prefrontal cortex with the ventral hippocampus playing a role in synchronization (O`Neill).

Cogmed training has also been linked to increased attention as well as working memory. In fact, it is likely that Cogmed decreases inattention and increases how much verbal and visuospatial information a subject can temporarily work with (Slezak). What part is effected by training must be ascertained and it could be an increase in efficiency of Posner`s control networks of alerting, orienting or central executive or to the components of orienting (Posner and Petersen) with disengagement (responsibility of the parietal area), shifting and reengagement of focus (responsibility of the What-Where pathway, cortical medial temporal cortex or pulvinar nuclei of thalamus). Another area that exhibits higher efficiency after training is the selection process that retrieves the relevant items from memory (activation in the rostral superior frontal sulcus and posterior cingulate cortex) or the updating process that changes the focus of attention on it (caudal superior frontal sulcus and post parietal cortex). Training could also shift the top-down, bottom-up balance of the control systems (control – stimulus driven and goal directed – Asphland, or top-down and bottom up system – Corbetta and Shulman) with training leading to better control of the top-down attentional systems or by improving working memory biases of attention by initiating the novel parieo-medial-temporal pathway proposed by Soto. The interconnectivity of all these areas can be observed by neuroimaging.

It is also possible that training programs change the balance of task relevant to task irrelevant (or attended to unattended) information, hence improving overall performance this way. It is known that working memory performance is dependent on effectively filtering out irrelevant information through neural suppression (Zonto). The dorsal parietal cortex exhibits influence on top-down attention and ventral parietal cortex on bottom up (Curicella) with prefrontal cortex playing a role (Nieuwenhaus – 5HT and dopamine amplify task relevant information rather than inhibiting distraction), and cingulate cortex (Egner – increase in task relevant information rather than inhibiting task irrelevant). This balance in task relevant information and task irrelevant can be affected by various factors. For example, there appears to be a decreased inhibition of task irrelevant information with age (Blair; Heshier; Redrick); anxiety appears to decrease information with relevance (Weltman), but on a positive note effects can be mitigated by training (Matzell) and computer games (Gofper; Green).

In order to see how the training altered the activation levels at the physiological level, the authors of this paper used different neuroimaging metrics. They found that working memory training brings about effects in water, lipid, and protein content, T2 by iron within the oligodendrocytes and MWF by lipid myelin content and associated the changes observed with training to alterations in these cell components. Other researchers have found that working memory training is associated with variability in white matter (Golestani) and more specifically, increased myelination in white matter neurons in the intraparietal sulcus and anterior body of corpus callosum (Takeichi). Chapman and colleagues also reported changes in blood flow. They found in their MRI studies (arterial spin labeling MRI, functional connectivity, and diffusion tensor imaging) healthy seniors tested pre, mid, and post training (12 weeks) had significant training-related changes in the brain in the resting state. Specifically there were increases in global and regional cerebral blood flow and connectivity particularly in the Default Mode Network and the central executive network and that there was increased white matter integrity in the left brain areas connecting parts of the limbic system in the temporal lobe (eg. hippocampus, amygdala) with parts of the frontal lobes such as the orbitofrontal cortex. They suggested that cognitive training enhanced resting-state neural activity and connectivity and increased the blood flow to certain brain areas, an idea supported by the results given in Caeyenberghs and colleagues paper.

Therefore, it can be summarized that training can induce positive effects on certain types of memory and processing and that these effects are likely to be associated with improved task-relevant information levels, increased attention and increased connectivity between brain areas responsible for attentional systems, visual systems and working memory. However, there is a word of caution in that training of this type does not give an unlimited positive effect, ie. the more you train the more you improve. Even after 8 weeks of training which is a long time and requires a high level of commitment to the training program, the improvement in cognitive performance lies only at less than 25% for healthy individuals. The advantage appears to come when cognitive problems are evident pre-training. It would be interesting to see how the neuroimaging results reported here correlate to training programs undertaken under these conditions.

Since we`re talking about the topic…….

….can we assume that the neuroimaging techniques applied here can be used to demonstrate neural functioning and neural connectivity in other types of memory which have a time element eg. conditioning?  Should connectivity between the areas relating to emotions and decision-making also appear strengthened?

…using this neuroimaging technique would provide interesting comparisons of functioning and neural connectivity if knock-out mice, or transgenic mice, are compared to controls.

….would the consumption of caffeine which increases alertness and consolidation of memory, or other stimulants shortly before each participation in the training program have a noticeable effect on the neuroimaging results as well as performance post-training?

This entry was posted in neuroimaging, neuronal networks, training, Uncategorized, working memory and tagged , , , . Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s