Hemispheric Asymmetry of Globus Pallidus Relates to Alpha Modulation in Reward-Related Attentional Tasks

Abstract

Whereas subcortical structures such as the basal ganglia have been widely explored in relation to motor control, recent evidence suggests that their mechanisms extend to the domain of attentional switching. We here investigated the subcortical involvement in reward related top-down control of visual alpha-band oscillations (8-13 Hz), which have been consistently linked to mechanisms supporting the allocation of visuospatial attention. Given that items associated with contextual saliency (e.g., monetary reward or loss) attract attention, it is not surprising that the acquired salience of visual items further modulates. The executive networks controlling such reward-dependent modulations of oscillatory brain activity have yet to be fully elucidated. Although such networks have been explored in terms of corticocortical interactions, subcortical regions are likely to be involved. To uncover this, we combined MRI and MEG data from 17 male and 11 female participants, investigating whether derived measures of subcortical structural asymmetries predict interhemispheric modulation of alpha power during a spatial attention task. We show that volumetric hemispheric lateralization of globus pallidus (GP) and thalamus (Th) explains individual hemispheric biases in the ability to modulate posterior alpha power. Importantly, for the GP, this effect became stronger when the value saliency parings in the task increased. Our findings suggest that the GP and Th in humans are part of a subcortical executive control network, differentially involved in modulating posterior alpha activity in the presence of saliency. Further investigation aimed at uncovering the interaction between subcortical and neocortical attentional networks would provide useful insight in future studies.SIGNIFICANCE STATEMENT Whereas the involvement of subcortical regions into higher level cognitive processing, such as attention and reward attribution, has been already indicated in previous studies, little is known about its relationship with the functional oscillatory underpinnings of said processes. In particular, interhemispheric modulation of alpha band (8-13 Hz) oscillations, as recorded with magnetoencephalography, has been previously shown to vary as a function of salience (i.e., monetary reward/loss) in a spatial attention task. We here provide novel insights into the link between subcortical and cortical control of visual attention. Using the same reward-related spatial attention paradigm, we show that the volumetric lateralization of subcortical structures (specifically globus pallidus and thalamus) explains individual biases in the modulation of visual alpha activity.

Keywords: attention; basal ganglia; magnetic resonance imaging; magnetoencephalography; reward.

Figure 1.

Illustration of selective attention task: stimuli and reward manipulation. A, Six Chinese symbols served as stimuli for the task and were associated with three values: two paired with reward, two with loss and two with no financial change (neutral). B, Representative trial of the learning phase. Symbols were displayed for 1000 ms, systematically paired with the corresponding (positive, negative or neutral) value, via visual and auditory feedback. Characters presentation was alternated with a 1000 ms fixation period. During the training phase, participants learned associations between the stimuli and their reward value. C, Representative trial of the testing phase. After a 1000 ms pretrial interval, participants were primed with a 500 ms preparatory cue signaling the upcoming stimuli. Two characters were then presented to the left and right hemifield, together with a spatial cue, instructing participants to covertly attend the symbol on the cued side (target) and ignore the other one (distractor). Participants' task was to report when the target stimulus changed contrast. Contrast change could either occur after 750 ms (13% of trials), 1450 ms (47% of trials), or 2350 ms (40% of the trials). In 95% of the trials, the target changed contrast (valid trials), whereas in 5% of the trials, the distractor changed contrast (invalid trials). (Figure adapted with permission from Marshall et al., 2018).

Figure 2.

Grand average MI and HLM distribution across participants. A, TFRs and topographical plot showing contrast between the 'attend right' and the 'attend left' trials trials. A clear modulation is visible at posterior sensors in the alpha band (8–13 Hz) in the −750 to 0 ms interval (this time window being considered for the computation of HLM(α) indices in B. Sensors included in the left and right ROIs are marked as dots. Trials are locked to the onset of the contrast change (t = 0). B, Side panels show the temporal evolution of modulation indices in the alpha range MI(α) averaged over sensors within left and right hemisphere ROIs. The magnitude (absolute value) of MI(α) progressively increased in the stimulus interval until the onset of the contrast change. Middle, Distribution of HLM(α) indices across participants, computed over the ROIs and 8 to 13 Hz frequency band (see Materials and Methods). A normal density function is superimposed, denoting no hemispheric bias in lateralized modulation values across participants (Shapiro–Wilk, W = 0.958, p = 0.392).

Figure 3.

Basal ganglia volumes resulting from semiautomated subcortical segmentation implemented. A, Orthogonal view and 3D rendering. Subcortical volumes are overlaid as meshes on the anatomical MRI of one of the participants (following defacing procedure in Freesurfer, where voxels outside the brain mask with identifiable facial features were excluded (Bischoff-Grethe et al., 2007). B, Histograms with superimposition of normal density function, showing the distribution of subcortical lateralization indices for each substructure. In our sample, Acb and Th volumes were left lateralized (p = 0.0001 and p = 0.0003, respectively), whereas CN showed a right lateralization (p = 0.0029).

Figure 4.

Lateralization of individual subcortical structures in relation to HLM in the task. A, Bar plot showing beta coefficients associated with a GLM where LV values were defined as explanatory variables for HLM(α). Error bars indicate SEM. Asterisks denote statistical significance; **p < 0.01. B, Partial regression plot showing the association between LV_GP and HLM(α) while controlling for the other regressors in the model in A. C, Partial regression plot showing the association between LV_Th and HLM(α) while controlling for the other regressors in the model in A. Given Equations 1 and 2 (see Materials and Methods), positive HLM(α) values indicate stronger modulation of alpha power in the right compared with the left hemisphere, and vice versa; similarly, positive (or negative) LV_s indices denote greater right(or left) volume for a given substructure s. The dotted curves in B and C indicate 95% confidence bounds for the regression line fitted on the plot in black.

Figure 5.

TFR of regression coefficient t-statistics on the linear relationship between low-frequency power modulation MI(f) and LV_GP (A) and LV_Th (B) indices averaged over ROIs (Fig. 2A). A black outline is used to highlight the significant time–frequency cluster found. For the LV_GP, the analysis revealed a clear α-band-limited association between the variables across the full time window of interest (see Materials and Methods) extending up to 1 s before the response.

Figure 6.

Alpha modulation indices for left and right hemispheres associated with two subgroups of the sample. A, Topographical plot of MI(α) values for the two participants groups, clustered according to directionality of GP lateralization (right vs left lateralized GP). Left and right sensors of interest are marked as dots and correspond to the same ROIs as in Figure 2. B, Individual data points superimposed on bar graph showing individual scores and MI(α) averaged over ROIs in the two subgroups. As indicated in the cluster-based permutation results, a difference is particularly observable for right hemisphere alpha modulation between the two groups, being higher in participants exhibiting a right lateralized GP. C, Individual data points showing HLM(α) scores for all participants. The horizontal blue line superimposed on the data indicates average HLM(α) index for each subgroup.

Figure 7.

Linear association between GP volumetric asymmetry and alpha modulation asymmetry as a function of value saliency occurrences in the task. A, Correlation between GP volume lateralization and HLM(α), grouped accordingly to the number of value-salient stimuli in the trials (see Materials and Methods). From left to right, respectively, two, one and zero value saliency occurrences are displayed. GP asymmetry significantly explained HLM(α) only when value-salient stimuli featured as both target and distractors, regardless of their valence (r = 0.68, significant at the p < 0.001 level after Bonferroni correction for three comparisons). B, The association between HLM(α) and GP volume lateralization increased as a function of value saliency in the task: the linear relationship was stronger when two value-salient stimuli were presented, when compared with conditions characterized by either one or value salience pairings, with a 95% CI of [0.106, 0.672] and [0.125, 0.897], respectively for the two comparisons. This suggests that, when both target and distractor were associated with a salient value, participants exhibiting bigger GP volume in the left hemisphere than in the right hemisphere were also better at modulating alpha oscillations in the left compared with the right hemisphere. Asterisks denote statistical significance; **p < 0.01. C, Raincloud plot (Allen et al., 2019) showing the bootstrap distribution of the difference in pairwise correlation coefficients examined.

Figure 8.

Linear association between Th volumetric asymmetry and alpha modulation asymmetry as a function of value saliency occurrences in the task. A, Correlation between TH volume lateralization and HLM(α), grouped accordingly to the number of value-salient stimuli in the trials (see Materials and Methods). From left to right, respectively, two, one and zero value saliency occurrences are displayed. When considering individual correlations between Th asymmetry and HLM(α), no significant linear relationship was found. B, Association between the two measures also did not significantly differ as a function of saliency in the trials. C, Raincloud plot showing the bootstrap distribution of the difference in pairwise correlation coefficients examined.

Figure 9.

Alpha modulation indices for left and right hemispheres associated with fast versus slow trials, neutral condition only. A, Topographical plot of MI(α) values for the two trial groups, clustered according to median split of reaction times (fast versus slow trials). Left and right sensors of interest are marked as dots and correspond to the same ROIs as in Figure 2. B, Individual data points superimposed on bar graph showing individual scores and MI(α) averaged over ROIs in the two subgroups. C, Individual data points showing LI(α) scores for all participants (difference in MI(α) values between right and left ROIs above). The horizontal blue line superimposed on the data indicates average LI(α) index for each subgroup.

Figure 10.

Mean and lateralized RTs and accuracy values across the three value saliency occurrences in the task. Mean RT (A) and accuracy (C) values averaged across participants in the three value-salient occurrences conditions in the task. No significant difference was found between groups by means of one-way repeated-measures ANOVA, indicating that different levels of value saliency pairings did not influence behavioral performance. No significant difference emerged also when comparing average lateralized values of RT (B) and accuracy (D) across the same conditions, and by means of same statistical analysis, indicating that the behavioral spatial bias was not affected by the different levels of value saliency pairings. Respective individual scores are superimposed on bars in all plots.

Figure 11.

GLM displaying combined lateralized subcortical volumes and hemispheric lateralized modulation as multiple regressors for the prediction of spatial behavioral bias in RT (A) and accuracy (B). No significant regression was found that could account for either the lateralized accuracy or RTs (p = 0.429 and p = 0.570, respectively).