The aim of this study was to provide the resting state MEG correlates of verbal fluency. Our results revealed a complex correlation pattern showing bilateral fronto-parietal clusters, right dominant in slow oscillations (delta to beta bands) and left dominant in highest oscillations frequency (gamma 2 and 3 bands). Anticorrelation clusters were also found in the left hemisphere temporal and occipital regions, in the right isthmus cingulate. Our results are in line with previous results report obtained during task-based approach of verbal fluency: similar activation was found in the left IFG/MFG (BA 6, 9, 44 & 45), the left precuneus (BA 7), and bilateral anterior cingulate gyrus (BA 24, 32). However, we did not find the right frontal lobe (BA 44, 47) nor the bilateral insula (BA 13) whereas we did find additional clusters in the bilateral motor cortex with paracentral lobule (BA 4), in the right premotor (BA 6, 8) and bilateral post central gyrus (BA 1, 2, 3) (Wagner et al., 2014). MEG fields are induced by synchronized neuronal currents, caused by synaptic transmission. Our results show that EEG/MEG power at rest may be correlated with performance during a test for different cortical sources and frequency bands.
We hypothesize that these regions are usually activated by cognitive processes similar to those solicited in the tests. These activities leave a trace related to the experience. The stronger the traces in the brain regions activated in the performance of a cognitive skill, the easier the activation when required by specific task, and the better the subject performs in that task. These traces can be detected by a difference in resting power in the MEG. The clusters of cortical sources where the resting MEG power correlates with the scores on a specific test thus reveal the clusters of traces in the brain regions associated with the competence assessed by the test. However, one should not expect that all regions actually activated in a task will yield resting state MEG clusters, and vice versa, because these two methods are based on very different approaches. Some of these regions may give rise to resting state MEG clusters but are not specific enough, activated in other processes, and subtracted from a closely matched control task in task-based brain imaging. Some tests would not reflect usual cognitive functions and requires ad hoc combinations of cognitive process which may not have a trace strong enough to be detected. In the resting-state MEG cluster approach, the relationship between inter-individual differences in test scores and in relative power at different cortical sources must fit a defined function (e.g., linear) for these clusters in the sample of subjects to be observed. These differences could account for discrepancies between resting states clusters and regions activated in tasks. Nevertheless, resting state clusters can complement the activity patterns and are relevant to provide a new perspective on the set of regions activated in VFL.
With conjunction 1, we aimed at identifying among the initial VFL clusters those accounting for a better performance in VFL as compared to VOC, thus for a relative advantage in executive over semantic abilities. A first group of clusters was found in the right premotor, motor and dorsolateral prefrontal cortex in the slow bands, most clearly in the alpha band. In early studies of executive function localization (Stuss, 2011; Stuss & Alexander, 2007), right lateral brain lesions, including the dorsolateral prefrontal cortex (BA9/46), thus overlapping some conjunction 1 clusters, had been linked to a deficit in monitoring ongoing performance in different executive tasks. This suggests that the relative advantage in executive abilities over semantic abilities is likely in monitoring verbal production. The second set of clusters was found in the dorsomedial prefrontal cortex, specifically in the bilateral rostral superior frontal and left rostral and caudal anterior cingulate, also in the slow frequency bands. This localization is consistent with lesions observed in patients whose performance deficits in various tasks have been described as an energization failure (Stuss, 2011; Stuss & Alexander, 2007). These patients showed lesions in superior medial cortex, primarily in BA areas 24, 32, 9, and 6. In the VFL task, these patients showed a marked decrease in the number of words in the last 45 seconds, compared with the first 15. This failure to energize is thus one of initiation and maintenance of performance. Moreover, these clusters are found in the alpha (8–13 Hz) and beta (13–30 Hz) oscillations, which are known to support inhibition mechanism (Klimesch, 2012; Klimesch, Sauseng, & Hanslmayr, 2007), which may be required to for maintenance of performance. These two clusters (right frontal lateral and dorsomedial frontal) correspond to the dual control network hypothesis (Dosenbach et al., 2007). The fronto-parietal network is optimized for rapid adaptive control and the other cingulo-opercular for stable set-maintenance. It also fits with the conjunction analyses across different executive functions (flexibility, inhibition and working memory), which also reveal activation in dorsolateral prefrontal (BAs 9, 46) and anterior cingulate (BA 32) cortex (Niendam et al., 2012). However, contrary to Wagner’s et al (2014) previous meta-analysis, we did not find any parietal clusters in this conjunction 1.
The last set of clusters in conjunction 1 was found in the left inferior frontal gyrus (BA 44) in the gamma band. In phonemic and semantic verbal fluency tasks, activation in the left inferior frontal gyrus overlapped the same regions (BA 9, 45), but BA 47 appeared in the semantic condition, and BA 44 in the phonemic condition only (Wagner et al., 2014). This left inferior frontal gyrus was found in gamma band (90–120 Hz). Gamma band oscillations have been observed during word production and auditory perception. They reflect synchronized firing of neuronal assemblies and task-related cortical activation. Using electro-corticography (CoG), an increase in gamma range (70–120 Hz), activity was also observed continuously in the inferior frontal gyrus starting 500 msec prior to the onset of syllable-articulation and stopping at vocalization. The gamma-augmentation may thus be predominantly driven and/or monitored phonological processing (Brown et al., 2008).
We first discussed, the conjunction 1 clusters that correlated positively both with the VFL scores and with the individual factor loadings on the difference factor, and reflected an advantage in executive over semantic abilities. We also found clusters that were positively correlated with vocabulary scores (Figure S1) but negatively correlated with individual factor loadings of the difference factor. They reflected an advantage of semantic abilities over executive abilities, (Figure S3) and appeared to constitute a semantic knowledge network. These clusters were found for the low frequency bands in the left lateral temporo-parietal region: in the transverse, superior and middle temporal gyri, the inferior parietal and supramarginal gyri, the post-central gyrus as well as in the lateral occipital regions. According to the “embodied” view of semantic information, a word is associated with different representations of an object, for example how it looks like, or how it is used. The meaning of a word is then based on a network of visual, auditory, somato-motor representations. These widely distributed regions, and the various connections between them, constitute the semantic network (Barsalou, 1999, 2008; Martin, 2007; Pulvermuller, Hauk, Nikulin, & Ilmoniemi, 2005). Semantic representations can be based on different types of relations (Mirman, Landrigan, & Britt, 2017). For example, similarity can use shared features (taxonomic, e.g. coat for dog-bear), or contiguity which relies on the co-occurrence in events or scenarios (thematic, e.g., dog - leash). By performing voxel-based lesion-symptom mapping on taxonomic and thematic errors separately in individual with poststroke aphasia, thematic errors were located in the left temporoparietal junction, and taxonomic errors in the left anterior temporal lobe (Schwartz et al., 2011). The left temporo-parietal junction, where thematic knowledge has been proposed to be grounded, is the core of the F2 anticorrelation clusters we found. The temporo-parietal region (especially the posterior middle temporal cortex and the inferior parietal lobule) might be involved in mental simulation of events, or re-enactement of the subject’s own perceptual and motor experience. Using a picture matching task in which participants had to identify taxonomic and thematic relations between objects thematic processing specifically recruited a bilateral temporo-parietal network including the inferior parietal lobules and middle temporal gyri (Kalénine et al., 2009). The clusters we found also include postcentral and occipital regions, which is consistent with the role of visual and sensorimotor regions in visuo-motor processes supporting thematic representations. In sum, the stronger the traces left by language experience in this thematic semantic network, the better the individuals perform in the Vocabulary test. Relying on the thematic semantic network would optimize the score at Vocabulary as participants can define the given word by using synonym, by its use, a clear characteristic, some concrete example of action or causal relationship, and not necessarily by providing a general or more abstract category to which the word belongs.
Going back to the VFL correlation clusters, Conjunction 2 has two main differences from conjunction 1. The first difference was in slow frequencies, with bilateral clusters in the precuneus and paracentral lobule, and in the right superior parietal lobe. At the same time, clusters in the bilateral dorsomedial frontal cortex and in the right superior frontal gyrus disappeared. The second difference were additional gamma clusters in the left parietal, sensory-motor and premotor regions. These changes in resting state activity account for an advantage for VFL over TMT, as compared to VOC. VFL and TMT mainly assess two different executive functions, fluency and flexibility, and differ mainly on two aspects. First, VFL requires the production of a verbal sequence, whereas TMT requires an eye-hand coordinated sequence. Second, in VFL the subject must keep in memory the entire verbal sequence as it is spoken to check that it conforms to a set of rules (first letter of the word, no repetition, etc.), but without following a pre-established order. On the other hand, in the TMT, the subject must follow a pre-established sequence (the alphabetical and numerical orders), and switch between them, but without keeping in memory the entire sequence since the test sheet provides a visual support. We suggest that the changes in predominantly right parietal slow oscillations may reflect the fact that the subject must pay attention to items in episodic memory (a form of working memory) during the fluency task. On the other hand, larger left frontal gamma clusters may be attributed to the speech production itself.
According to the attention-to-memory model, the dorsal parietal cortex is most active when top-down monitoring of memory content is maximal (Cabeza, 2008). The dorsal parietal cortex corresponds approximately to BA7 and includes the superior parietal lobule, but also the precuneus and part of the paracental lobule. The additional parietal clusters in conjunction 2 may therefore correspond to top-down attentional processes in accordance with internal goals as per VFL instructions, and thus account for an advantage in performance in VFL over TMT, as compared to VOC. The presence of a medial parietal structure (precuneus) in the slow oscillation clusters is likely related to the fact that the task requires the subject to monitor his or her own verbal output, given that the precuneus supports the recall of memories from a first-person perspective (Cavanna & Trimble, 2006; Freton et al., 2014). The dorsal parietal cortex is part of a dorsal frontoparietal pathway that includes the midlateral prefrontal cortex and enables the selection of internal goals and links them to appropriate responses (Nee et al., 2013). The absence of dorsomedial prefrontal clusters which were present in conjunction 1 may reflect the fact that the energization component of executive functioning is no longer an advantage for VFL when comparing with TMT over VOC, as initiation and maintenance of performance is greatly helped in TMT by the availability of the test sheet with the different letters and numbers to connect.
The rapid oscillatory cluster on the left corresponds with the word production sequence, from the left phonological store (i.e. left supramarginal gyrus) and then to pre-motor for the syllabification and ending in the pre-central gyrus for articulation (Indefrey & Levelt, 2004). Moreover, this cluster is found in the gamma bands, gamma 2 (60–90 Hz) and gamma 3 (90–120 Hz). As described previously, the gamma band were found in the left pre-central gyrus after onset of vocalization process (Brown et al., 2008) and reflects local synchronized firing of neuronal assemblies.
Finally, the conjunction 3 (Fig. 5) shows the overlap between conjunction 1 and conjunction 2, highlighting a set of anterior frontal clusters related to the executive monitoring process and a set of more posterior frontal clusters related to phonological sequence implementation. Conjunction 3 thus reflects regions that must coordinate activity in order to perform in the fluency task. This overlap is localized in the right hemisphere, in adjacent parts of the premotor, pre-central and post-central cortex in the mid-lateral regions. Previous lesions study shown these regions are implicated in monitoring and control of speech production (Stuss & Alexander, 2007).
To conclude, we first showed a complex set of correlation clusters, for the slow frequencies, in the right parietal and frontal regions, and in the bilateral medial frontal regions, bilateral paracentral and precuneus, as well as anticorrelation clusters in the left medial temporal lobe. By examining which of these clusters were related to a relative advantage in VFL as compared to two other tests (one verbal and one executive), we retained the core clusters of VFL. These core clusters clearly present verbal fluency as an executive test, related to word production and performance monitoring. Despite the verbal nature of verbal fluency, these clusters were located in the right hemisphere, which seems to reflect right-hemisphere dominance for executive control of attention (Spagna, Kim, Wu, & Fan, 2020). This study confirms the value of the cluster analysis method based on correlations between resting MEG and cognitive skills, such as working memory or verbal fluency.
These results and this method require replication and further internal and external validation. The sample was small and susceptible to sampling bias effects. As MEG temporal dynamics mainly include oscillatory and non-oscillatory (1/f noise-like) brain activities, the nature of the correlation needs to be elucidated. Although we have observed correlation clusters for various cognitive abilities (working memory, verbal fluency, and vocabulary), it is not known which other cognitive skills may generate such clusters. It is also not known whether differences in performance beyond the normal range can be detected by MEG clusters at rest, or whether other than linear functions should be used. This approach of cluster analysis is thus nascent, and requires much further research, but may have great potential.