Collaborative roles of Temporoparietal Junction and Dorsolateral Prefrontal Cortex in Different Types of Behavioural Flexibility

Behavioural flexibility is essential for everyday life. This involves shifting attention between different perspectives. Previous studies suggest that flexibility is mainly subserved by the dorsolateral prefrontal cortex (DLPFC). However, although rarely emphasized, the temporoparietal junction (TPJ) is frequently recruited during flexible behaviour. A crucial question is whether TPJ plays a role in different types of flexibility, compared to its limited role in perceptual flexibility. We hypothesized that TPJ activity during diverse flexibility tasks plays a common role in stimulus-driven attention-shifting, thereby contributing to different types of flexibility, and thus the collaboration between DLPFC and TPJ might serve as a more appropriate mechanism than DLPFC alone. We used fMRI to measure DLPFC/TPJ activity recruited during moral flexibility, and examined its effect on other domains of flexibility (economic/perceptual). Here, we show the additional, yet crucial role of TPJ: a combined DLPFC/TPJ activity predicted flexibility, regardless of domain. Different types of flexibility might rely on more basic attention-shifting, which highlights the behavioural significance of alternatives.


Subjects
Among the twenty-four students participating in this study, twenty-two were righthanded and two were ambidextrous. Exclusion criteria included a history of neurological injury or disease, serious medical/surgical illness, and substance abuse.
Further, all subjects underwent structural MRI scanning to rule out significant cerebral anatomic abnormalities. This study was approved by the Committee on Medical Ethics of Kyoto University (permission number: E1924) and carried out in accordance with The Code of Ethics of the World Medical Association. In this study, we adopted onetailed significance tests, which reflect our a priori unidirectional hypotheses.

Ultimatum Game (UG)
UG task was designed based on a previous study consisting of 20 trials per subject 1 .
Following the conventional implementation of this study 1 , we also established five types of offers. That is, the proposer offered the responder 50, 40, 30, 20 or 10 yen.
Moreover, we kept the same offering ratio with this study: offers of 50 and 40 yen (50% and 40% of 100 yen) were regarded as fair offers, and those of 30, 20, and 10 yen (30%, 20%, and 10% of 100 yen) were regarded as unfair offers. Accordingly, fair offers (40% of total offers: 8 offers) and unfair offers (60% of total offers: 12 offers) were presented randomly.

Wisconsin Card Sorting Test (WCST)
During WCST, four stimulus cards were displayed on the computer screen. These cards varied according to three categories: the number, colour, and shape of their elements.
The subjects were instructed to select one of the four cards, but were not told which perceptual category to use in order to choose the card. After each selection, there was a feedback ('Right' or 'Wrong') on the screen. The 'Right' category shifted among the three categories (number, colour, and shape) during the test, which continued until the subjects had selected all 48 cards.

Psychophysiological interaction analysis
To investigate whether R-TPJ and R-DLPFC interact to adjust attention allocation during flexible decision-making, we conducted Psychophysiological interaction (PPI) analysis. We used the generalized PPI (gPPI) toolbox, which has increased sensitivity and flexibility of statistical modeling (http://www.nitrc.org/projects/gppi/). In gPPI, deconvolved activity of the seed region is assigned to separate regressors dependent on the psychological conditions and reconvolved with the hemodynamic response function.
Following a conventional study, deconvolved time series of R-TPJ activity (x, y, z = 48, -52, 18) were obtained for each subject as a seed region 2 . Based on the manual, we included 1) C/B; 2) R/W; and 3) Y/N conditions in the analysis. Contrast images associated with the PPI regressor were then entered into a regression analysis. The PPI contrast compared the C/B condition*R-TPJ (contrast weight: +1) with the R/W condition*R-TPJ (contrast weight: -1). Based on previous studies 3,4 , the statistical threshold was set at corrected cluster-level p < 0.01 (individual voxel p < 0.05, cluster size ≥ 193 voxels) to reduce the risk of false negatives 5 . This threshold was obtained by Monte Carlo simulations using Gaussian random field theory multiple comparison correction that applied 10,000 iterations within the R-DLPFC [REST (http://restfmri. net/forum/rest)]. Specifically, based on the manual, Gaussian filter widths were determined as 7.681, 8.757, 11.481 (FWHMx, FWHMy, FWHMz, in mm), and the cluster connectivity criterion was set at 3 (rmm). The risk of getting an inappropriate clustering threshold may increase by the amount of smoothing of the observed fMRI signals (i.e., degree of false positives). Namely, immoderate smoothing enhances spatial correlation of the fMRI noise across voxels, which is characterized by a spatial autocorrelation function that assumes a Gaussian shape across the whole brain. In effect, this would result in a slight overestimation of the clustering threshold 6,7 , in which the degree of overestimation may vary marginally depending on each dataset 8 .
In this analysis, we expected either positive or negative connectivity between TPJ and DLPFC in the conflict resolution contrast (C/B > R/W), which represented greater flexibility. This is because both positive and negative connectivity between two brain regions may potentially be accompanied by co-activation in GLM group-level activation analyses (i.e., increased hemodynamic activity). In other words, positive connectivity between two regions (accompanied by co-activation) can be due to these regions working in concert 9 . Meanwhile, negative connectivity (accompanied by co-activation) can be due to reciprocal modulation 9 . More specifically, in an activation study, a prediction of activity in one brain region by greater activity in another region indicates functional integration, whereas negative connectivity between these regions may further suggest a dynamic interplay via reciprocal modulation 10 . Accordingly, the results of the PPI analyses represent the strength of regression of activity in one region on another (i.e., the level of activity in two regions correlates over time). That is, activity in one region may be driven/explained by activity in the other, and that hemodynamic activity in these two regions can increase and decrease in synchronization/coupling 9 . In this respect, either positive or negative connectivity of R-TPJ and R-DLPFC, accompanied by co-activation in activation analyses, may represent behavioural flexibility via collaboration or reciprocal modulation of these brain regions.

Supplementary Results and Discussion
The results of PPI analysis showed a significant decrease in TPJ-DLPFC functional connectivity in the C/B compared to the R/W condition in the conflict resolution contrast (C/B > R/W; p < 0.01, cluster-level corrected; Supplementary Table S3). This negative functional connectivity was in agreement with findings from the GLM grouplevel activation analysis, which showed greater simultaneous TPJ and DLPFC activity (i.e., co-activation) in this contrast. These results suggest a functional integration between these regions. Specifically, our findings of negative connectivity, accompanied by co-activation of these regions (observed in the activation analysis) might imply that TPJ and DLPFC modulate each other in a reciprocal manner via bi-directional inhibitory processing 11 . In effect, we speculate that this might facilitate thinking about conflicting perspectives simultaneously, which involves both maintaining and shifting between decision-rules [12][13][14] . However, to consider this issue, the threshold in the PPI analysis is relatively liberal for arguing robustly about the interaction of the link between TPJ and DLPFC. Furthermore, statistical procedure of this PPI at the grouplevel is incompatible with that of the activation analysis. To argue about the role of the TPJ-DLPFC link in flexible behaviour, additional experiments that are especially focused on the interaction between TPJ and DLPFC should be conducted in future studies.

Supplementary Tables
Supplementary Table S1 There was negative condition-dependent connectivity between R-TPJ and R-DLPFC in C/B compared to R/W condition.