Task-based functional connectivity identifies two segregated networks underlying intentional action

While much of motor behaviour is automatic, intentional action is necessary for the selection and initiation of controlled motor acts and is thus an essential part of goal-directed behaviour. Neuroimaging studies have shown that self-generated action implicates several dorsal and ventral frontoparietal areas. However, knowledge of the functional coupling between these brain regions during intentional action remains limited. We here studied brain activations and functional connectivity (FC) of thirty right-handed healthy participants performing a finger pressing task instructed to use a specific finger (externally-triggered action) or to select one of four fingers randomly (internally-generated action). Participants performed the task in alternating order either with their dominant right hand or the left hand. Consistent with previous studies, we observed stronger involvement of posterior parietal cortex and premotor regions when contrasting internally-generated with externally-triggered action. Interestingly, this contrast also revealed significant engagement of medial occipitotemporal regions including the left lingual and right fusiform gyrus. Task-based FC analysis identified increased functional coupling among frontoparietal regions as well as increased and decreased coupling between occipitotemporal regions, thus differentiating between two segregated networks. When comparing results of the dominant and nondominant hand we found less activation, but stronger connectivity for the former, suggesting increased neural efficiency when participants use their dominant hand. Taken together, our results reveal that two segregated networks that encompass the frontoparietal and occipitotemporal cortex contribute independently to intentional action.


Introduction
Intentional action is internally-triggered and self-generated motor behavior that is directed toward a specific goal and often requires selection among alternative movements ( Bonicalzi and Haggard, 2019 ;Jahanshahi and Frith, 1998 ;Passingham et al., 2010 ;Wolpert and Ghahramani, 2000 ). Paradigms used to examine the neural mechanisms of intentional action have often contrasted internallytriggered to externally-triggered conditions ( Brass and Haggard, 2010 ;Passingham et al., 2010 ;Si et al., 2021 ). Early studies on intentional action emphasized the importance of medial frontal regions, such as the pre-supplementary motor area (pre-SMA; Cunnington et al. 2002 ;Lau et al. 2004 ;Mueller et al. 2007 ;Nachev et al. 2007 ) and the anterior cingulate cortex (ACC; Cunnington et al. 2002 ;Deiber et al. 1999 ;Hoffstaedter et al. 2013 ) and the dorsolateral prefrontal cortex (dlPFC; Beudel and De Jong 2009 ;François-Brosseau et al. 2009 ; A prominent model of action planning proposes three groups of cognitive processes underlying intentional action, relating to what (action selection) , when (time of initiating the chosen action) and whether (mechanisms controlling start or interruption of the action) components ( Brass and Haggard, 2008 ). While frontoparietal engagement including the pre-SMA, dorsal premotor cortex, inferior and middle frontal gyrus, middle cingulum and supramarginal gyrus appear to be predominantly recruited for the what component ( Hoffstaedter et al., 2013 ;Zapparoli et al., 2018Zapparoli et al., , 2017, the SMA proper and insula are more involved in the when component of intentional action ( Zapparoli et al., 2018( Zapparoli et al., , 2017. Studies using intracranial or transcranial brain stimulation also support the implication of frontoparietal networks in intentional action. Desmurget et al. (2009) observed that electrical stimulation of the parietal cortex in awake surgery patients resulted in the desire to initiate a movement with the contralesional limb. When stimulation intensity increased, patients were even convinced of having executed a movement even though neither an observable action nor electromyographic activity was detected. Similarly, temporary inhibition of the left inferior parietal lobule with continuous theta burst stimulation in healthy volunteers led to impaired attribution of intention to action ( Patri et al., 2020 ). In contrast, transient disruption of the dorsal premotor cortex impaired selection and delayed the initiation of actions ( Mochizuki et al., 2005 ;Schluter et al., 1998 ) without affecting action execution ( O'Shea et al., 2007a ). Finally, direct electrical stimulation of the premotor cortex resulted in impairments of motor awareness ( Fornia et al., 2020 ). Taken together, brain stimulation data suggest that action intentions are generated in parietal regions, while action selection depends primarily on the premotor cortex.
These observations suggest a close interaction between frontal and parietal brain regions during the preparation of intentional action, in particular when it requires goal-directed hand movements ( Gallivan et al., 2013 ;Ptak et al., 2017 ). Unfortunately, while many previous studies examined functional activations during intentional action, only few focused on underlying interactions between functional networks. An early study by Rowe et al. (2005) found increased functional connectivity (FC) between prefrontal and motor cortices as well as frontoparietal regions when subjects selected a finger to press a button. Contrasting results were reported by Thimm et al. (2012) , who examined task effects on FC using fMRI. These authors found that free selection of actions was associated with increased coupling between the superior parietal lobules and occipitotemporal regions such as lingual and fusiform gyrus. However, a similar study by Welniarz et al. (2022) found increased connectivity between the PPC and cerebellum during the execution of freely chosen movements. Thus, the available evidence regarding functional interactions between brain regions -particularly the involvement of frontoparietal connections -during free generation of actions is not as well established as activation studies might suggest.
To fill in this gap, we aimed to identify FC between brain regions that were activated during a simple task requiring intentional actions. We first investigated signal increases during internally-triggered and externally-triggered finger movements, and then studied task-based FC between regions of interest (ROIs) defined by significant activation clusters . We hypothesized that the generation of intentional action would be accompanied by increased functional coupling between parietal and premotor cortices.

Participants
Thirty-five healthy individuals participated in the experiment. All participants gave written informed consent prior to participation, and the study procedures were approved by the ethical commission of the Canton of Geneva (Switzerland). The data of four participants were excluded due to technical issues linked to MRI trigger prob-lems or excessive head movement. An additional participant was excluded due to inability to follow instructions. The analyzes were thus based on the data of thirty participants with an age range of 21-43 (mean = 27.4 ± 5.7; 13 females), who were all right-handed based on the Edinburgh Handedness Inventory (EHI; Oldfield 1971 ) (lateralization index: 86.67 ± 13.98). None of the participants had a history of neurological or psychiatric disorders.

Stimuli and procedure
Participants performed a finger pressing task with their right or left hand, similar to the one used previously by Rowe et al. (2010) . Stimuli consisted of the image of a left or right hand in white outline on black background and either a single white circle (cue) shown above the index, middle, ring, or little finger, or one circle above each of these fingers ( Fig. 1A ). At the start of each trial, only the outlined hand appeared on a completely black screen. After 750 ms the cue was added and remained on the screen for 1500 ms. For externally-triggered ( EXT ) trials, the cue was a single circle above the finger that had to be used for a button-press. For internally-triggered ( INT ) trials, the cue signalled that participants had to choose freely which finger to use. Participants had to provide a response within 1500 ms of cue onset, and finger presses registered after this period were excluded from the analysis of reaction times (RTs). After cue disappearance the hand outline remained on the screen for a further 950 ms before the next trial started. The experiment was created and presented using E-Prime 2.0 (Psychological Software Tools, Inc., Pittsburgh, PA).
Prior to the experiment participants practiced outside the scanner (18 trials with each hand). When installed in the scanner they were given earplugs to reduce scanner noise, and foam pads were placed around their heads to minimize head motion. Both hands were placed on the abdomen, and the responding hand held a button-box. During the experimental session, two different runs separated by a period of rest were performed with the left or right hand respectively, whereby hand order was counterbalanced across participants. Within each run five blocks of EXT and five blocks of INT conditions alternated, each block consisting of nine trials for a total of 45 trials for each condition and hand. The switch between conditions was separated by a short rest period (5760 ms) during which the hand outline remained on the screen without presenting any cues. Each run lasted approximately six minutes. At the end of the experimental task participants underwent a T1 structural MRI scan.

MRI acquisition
Functional and structural images were acquired on a 3T Trio MRI scanner (Siemens Medical Solutions, Erlangen, Germany) in one session using a 64-channel array coil. Structural images consisted of a highresolution T1-weighted MPRAGE sequence (TR: 2300 ms; TE: 1.96 ms; number of slices: 176; FA: 9°; voxel size = 1.0 mm isometric). A fast Echoplanar Imaging sequence was used for the functional images (TR: 720 ms; TE: 30 ms; number of slices = 56 axial; voxel size = 2.5 mm isometric; flip angle (FA) = 50°). Using a short TR allowed to increase statistical power of the analysis by collecting more data while reducing acquisition time to prevent boredom for the monotonous task. We also acquired individual field maps (with the same field of view as functional images) to take possible inhomogeneities of the magnetic gradient into account. The MR session consisted of 472 volumes for each experiment run.

fMRI data processing
Functional imaging data were preprocessed and analyzed using Statistical Parametric Mapping (SPM12; Wellcome Trust Center for Neuroimaging, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/ ). Preprocessing steps included slice-timing correction, realignment and un- warping by using the acquired individual field maps, coregistration to the individual T1, spatial normalization to the Montreal Neurological Institute (MNI) template and spatial smoothing with a 8 mm Gaussian kernel.
To better control for motion inside the MR scanner, scan-nulling regressors ( Lemieux et al., 2007 ) were added to the standard 6 realignment parameters derived by the SPM realignment and their Volterra expansion ( Friston et al., 1998 ). Further nuisance regressors included the average time course from the white matter as well as from the cerebrospinal fluid (CSF), which were extracted and estimated using the MarsBar toolbox ( Brett et al., 2002 ). For each subject, we computed a two-factor general linear model (to account for left and right hands) including the EXT and the INT conditions as regressors of interest, convolved by the hemodynamic response function. The main contrast of interest reflecting intentional action was computed as ( . Moreover, to examine the effect of condition as a function of hand, we considered the contrasts ( > ) for the left hand and ( > ) for the right hand. Group level results were analyzed with t-tests, with an FWE -corrected threshold of p < 0.05 and a cluster extent threshold of k > 15 voxels. Activation results were visualized using the BrainNet Viewer toolbox ( Xia et al., 2013 ).

Functional connectivity analysis
The FC analysis was performed using the CONN toolbox ( Whitfield-Gabrieli and Nieto-Castanon, 2012 ). Sixteen ROIs were defined in MNI space using the MarsBar toolbox ( Brett et al., 2002 ) as 10 mm-diameter spheres centered at the most significant activation peak of each brain region that was identified by the ( Supplementary Table 1). To maintain a symmetrical distribution of ROIs we added contralateral ROIs when the contrast only identified unilateral engagement of a brain region.
Physiological noise was handled with the CompCor method ( Behzadi et al., 2007 ) implemented in the CONN toolbox, which regresses out timeseries from white matter and cerebrospinal fluid for each participant. Moreover, subject motion and main task effects were added as additional confounds, the latter to prevent misinterpreting task-related coactivations as connectivity measures. A temporal bandpass filter of 0.01-infinite was applied to eliminate high-frequency noise. A weighted general linear model (wGLM) was used to estimate pairwise ROI-to-ROI connectivity across experimental conditions by computing bivariate correlation across all time points at the first-level analysis. Blocks that were defined by each condition were convolved with the hemodynamic response function to define the condition-specific weights. Average time series were extracted across all voxels within each ROI, and each pair of source and target ROIs was computed separately, producing correlation coefficients which were then Fisher-transformed. This approach assesses task-related functional connectivity changes in a similar manner as the psychophysiological interaction (PPI) approach does ( Poletti et al., 2018 ). While PPI estimates connectivity measures at different time points, wGLM allows comparing connectivity across different runs in long block designs. This approach was chosen because our subjects performed the task in two different runs (one hand at a time) with relatively long blocks. One-sample t-tests were performed for the second-level ROI-to-ROI functional connectivity with a threshold connection of p < 0.05. All second-level ROI-to-ROI FC results were corrected for multiple comparisons using a false discovery rate (FDR) of p < 0.05 at cluster level.

Behavioral results
We first examined RTs, which were defined as the time taken to give a response from the onset of the cue until its disappearance. Even though our task was slow and self-paced, trials with an RT below 150 ms were discarded as anticipations. Median RTs computed for each hand and condition were submitted to a repeated-measures ANOVA with the factors hand (left, right) and condition ( EXT, INT ). This analysis revealed a main effect of condition ( F 1 , 29 = 34.94; p < 0.0001), reflecting faster RTs in the INT compared to the EXT condition. There was no effect of hand ( F 1 , 29 = 1.48; p = 0.23) and no significant interaction between hand and condition ( F 1 , 29 = 0.09; p = 0.77).
We next analyzed whether participants conformed to the instruction to select the responding finger randomly. We calculated the approximate entropy (ApEn; Pincus, 1991 ) as well as sample entropy (SampEn; Richman and Moorman, 2000 ) using the TSEntropies package in RStudio (RStudio Team, 2022), where m (dimension of subseries length) was set to 2 and r (similarity tolerance) was set to 0.2 times the standard deviation of each participant's series of choice, by default. This resulted in an average ApEn of 0.89 ± 0.07 and a SampEn of 1.08 ± 0.14. These high values indicate a low level of repeated patterns ( Delgado-Bonal and Marshak, 2019 ). This finding shows that participants did not select specific finger sequences in the INT condition, indicating that they attempted to act randomly. Table 1 shows the activation coordinates and peaks identified with the contrast ( + ) > ( + ), and Fig. 2 shows activations projected on a template brain. Compared to externally-triggered finger movement, internally-triggered movement showed increased neural signals in bilateral inferior parietal lobules (IPL) and the right superior parietal lobule (SPL), as well as the opercular part of the left and right inferior frontal gyrus (IFG operculum). Bilateral premotor cortex (PMd) was activated in the lateral and medial part of the superior frontal gyrus (SFG) and left pre-SMA. Engagement of the right PMd extended into the SMA, while the pre-SMA cluster extended into the midcingulate cortex (MCC). Moreover, parts of the dlPFC were involved, showing involvement of the middle frontal gyrus (MFG) bilaterally. Surprisingly, how-  ( Fig. 4A ). By contrast, using the right hand produced less spread and weaker signals in bilateral IPL, left SPL, and bilateral lingual/fusiform gyrus ( Fig. 4B ).

Functional connectivity (FC) results
Contrasting the INT vs. EXT condition across hands revealed significantly increased ROI-to-ROI connectivity between SPL and premotor regions ( Fig. 3 ). While both left and right SPL showed a significant positive connectivity with SFG bilaterally, left SPL was additionally connected with left IFG. Interestingly, none of these dorsal ROIs exhibited significant FC with the occipitotemporal ROIs in lingual and fusiform gyri ( Fig. 3 ). Instead, the latter showed positive FC within each hemisphere and negative FC across the hemispheres during task execution ( Fig. 3C ). The ROI-to-ROI analysis thus revealed that the dorsal and ventral ROIs function as two relatively isolated networks, which do not interact with each other.
Examining INT vs. EXT connectivity for the right hand ( Fig. 4 B) yielded positive FC between the left SPL, bilateral SFG and left IFG . Among the occipitotemporal regions, left lingual gyrus showed a positive connectivity pattern with left fusiform gyrus. In contrast, a comparable analysis of FC when participants responded with their left hand did not identify any significant connectivity ( Fig. 4 A).

Correlation analysis of cortical activation and behavior
To assess if any of the significant clusters from the contrast of interest predicts behavior, we computed Spearman's coefficients between mean and maximum activation peaks of each subject and RT differences between INT and EXT conditions. These analyses failed to identify any significant correlations (all p > 0.09).

Discussion
Our study examined the implication of frontoparietal regions in intentional action, which we operationalized as the contrast between internally and externally triggered finger presses. Similar to previous studies this comparison elicited activations in bilateral IFG, IPL, SFG and MFG, and additional activations in left pre-SMA, right SPL, and left insula ( Beudel and De Jong, 2009 ;François-Brosseau et al., 2009 ). Surprisingly, the strongest neural signals were detected in the lingual and fusiform gyri, which may provide an additional role in intentional hand action, as will be discussed in detail below.
The main difference between the INT and EXT condition was the requirement to select the responding finger, which mainly reflects the what component of intentional action ( Brass and Haggard, 2008 ). Participants reacted significantly faster during the internally-triggered compared to the externally-triggered condition, which appears to be somewhat counterintuitive given that accelerated RTs with externally cued movements were reported in some previous studies ( Rowe et al., 2010 ). However, our finding is consistent with other reports ( Beudel and De Jong, 2009 ;Doty, 2018 ;François-Brosseau et al., 2009 ;Gerardin et al., 2004 ;Mueller et al., 2007 ), and is likely due to the predictability of the task. Our analyses of entropy (which evaluates the degree of 'randomness' in a sequence of responses) indicate that participants acted at least as randomly in the INT as in the EXT condition. However, given the presentation of several INT trials in succession, participants could prepare themselves for the upcoming trial and select the responding finger in advance. This contrasted to the reactive mode in the EXT condition, where they had to wait for the cue display before planning their response. It is thus possible that participants selected the responding finger in advance and then waited for the signal to execute the movement. In addition, anticipated response planning in the INT condition explains well the absence of a correlation between RTs and peak activations.

Intentional action engages frontoparietal and occipitotemporal brain regions
Consistent with several studies that examined intentional action with paradigms involving finger movements ( Cunnington et al., 2002 ;Gerardin et al., 2004 ;Hoffstaedter et al., 2013 ;Jenkins et al., 2000 ) we found increased neural signals in the pre-SMA, extending into the MCC. These regions are strongly recruited when movement is initiated voluntarily and in the absence of external cues ( Lau et al., 2006 ;Mueller et al., 2007 ). They appear to be essential nodes of a movement selection network ( Hoffstaedter et al., 2013 ) and thus strongly contribute to the what component of intentional action. Additionally, lesion studies implicate the pre-SMA in response selection ( Nachev et al., 2007 ) and the MCC in cognitive control under conflict ( Tolomeo et al., 2016 ). The INT condition required increased cognitive control, since the necessity to randomly select fingers forced participants to monitor previously selected responses in order to avoid repeated sequences. Engagement of the pre-SMA and MCC are therefore in line with the role of these regions in motor control and action monitoring ( Hoffstaedter et al., 2013 ;Misra and Coombes, 2015 ;Nachev et al., 2007 ;Tolomeo et al., 2016 ). In addition to midline premotor cortex associated with the contrast INT vs. EXT , we also found involvement of right dorsal premotor cortex (PMd: SFG), bilateral ventral premotor cortex (PMv: IFG) and right dlPFC. There were also bilateral IPL and right SPL activations. All these regions have been identified in previous studies examining neural mechanisms of intentional action ( Ariani et al., 2015 ;Beudel and De Jong, 2009 ;François-Brosseau et al., 2009 ;Gerardin et al., 2004 ;Rowe et al., 2010 ). The PMd and PMv play a crucial role in motor control and the storage of abstract motor representations ( Kantak et al., 2012 ). While PMv is known to encode movement precision, the planning of goal-directed movements and action observation, imitation or even imagery ( Burianová et al., 2013 ;Caspers et al., 2010 ;Johnson-Frey et al., 2003 ;Liakakis et al., 2011 ;Tunik et al., 2008 ), PMd appears to be directly involved in action selec-tion ( Cisek and Kalaska, 2005 ;Rushworth et al., 2003 ). Interestingly, in our study the right PMd was engaged when subjects used the contralateral hand, whereas bilateral activations were present when responses were made independently of hand. This is in line with previous studies suggesting a preference of right PMd for the left hand, while left PMd is implicated in selecting actions with either hand ( Johansen-Berg et al., 2002 ). Verstraelen et al. (2021) suggest that effector selection, action definition and planning of movement kinematics rely on the left PMd, while the right PMd is responsible for the fine-tuning of ongoing movements. This hypothesis could partly explain why the right PMd was more recruited when participants responded with their slightly clumsier, nondominant hand.
Regarding the parietal lobe, Desmurget and Sirigu (2012) argued that the IPL specifically generates movement intentions, as patients with damage to this region may execute movements in the absence of intentions. Frontoparietal engagement is related to the sense of agency ( Renes et al., 2015 ;Tsakiris, 2010 ), which is more pronounced in selfgenerated actions ( Borhani et al., 2017 ;Villa et al., 2021 ). Bilateral parietal areas also contribute significantly to motor imagery ( McInnes et al., 2016 ), suggesting a major role in the generation of action plans and intentions ( Tumati et al., 2019 ). Finally, involvement of the SPL can be linked to the specification of spatial parameters for movement execution, sensorimotor integration and motor attention ( Striemer et al., 2011 ).
To summarize, internally-triggered action recruited premotor and parietal cortex more than externally-triggered action. This finding is consistent with a stronger reliance of self-generated action on processes related to goal and effector selection ( Ariani et al., 2015 ). The frontoparietal action network thus generates abstract motor representations, translates them into a motor action, and monitors their execution ( Löffler et al., 2020 ;Si et al., 2021 ). While these findings are consistent with previous observations, we surprisingly found the strongest contrast between INT vs. EXT conditions in bilateral occipitotemporal cortex. This was unexpected, as these regions are implicated in processing various visual categories, such as letters, tools or faces ( Iidaka, 2014 ;Kanwisher and Yovel, 2006 ;Kuriki et al., 1998 ;Ptak et al., 2022 ). It is unlikely that these activations were only related to differences in visual stimulation between conditions, since the engagement of occipitotemporal areas was strongly modulated by the hand effect (see Fig. 4 ) although visual stimulation was equivalent. Some recent studies implicated occipitotemporal cortex in biological movement processing ( Nocchi et al., 2012 ) and internally-directed attention ( Benedek et al., 2016 ). Also, several studies incorporating execution or observation of hand movements observed engagement of these areas ( Crotti et al., 2022 ;Uhlmann et al., 2020 ;Zapparoli et al., 2016 ). Recently, Okamoto et al. (2021) found increased activation of the right fusiform and bilateral lingual gyri when a hand was presented in a first-person compared to a third-person perspective. Furthermore, the occipitotemporal cortex is also implicated in visual imagery of objects and motor imagery of hands during voluntary control of action ( Helmich et al., 2007 ;Zapparoli et al., 2016 ). Interestingly, in a motor tracking task the lingual gyrus was particularly engaged when participants received no visual feedback about their hand, suggesting that lingual activation is more linked to imagery than the vision of the hand ( Beets et al., 2015 ). Thus, the activations of lingual and fusiform gyrus in our study may be related to an internal representation of one's own hand, since lying inside the scanner made it challenging to execute the task without visual bodily feedback. While this interpretation remains hypothetical, it is at least partly supported by previous literature.

Functional coupling during intentional action identifies segregated networks
Previous and the current fMRI findings suggest that a widespread network of regions contributes to intentional action. However, since only a modest number of studies examined functional connectivity, it is unknown to what extent the activated areas functionally interact during the planning and execution of such actions. For example, while one study identified FC between the primary motor cortex (M1) with prefrontal and parietal areas during freely selected finger presses ( Rowe et al., 2005 ), others observed increased connectivity between the M1 and premotor regions ( Koch et al., 2006 ;O'Shea et al., 2007b ), or even between parietal areas and the cerebellum ( Welniarz et al., 2022 ). By considering all activated areas as possible network nodes we were able to distinguish between two functionally segregated networks contributing to intentional action: a dorsal network involving bilateral SPL, bilateral SFG and left PMv, and a ventral network interconnecting the fusiform and lingual gyri. Functional segregation is often interpreted as reflecting the specialization to distinct types of mental representations or distinct cognitive processes. We propose that even a simple finger press task relies on distinct representations, where frontoparietal areas are important for representations related to motor plans and action execution, while the medial occipitotemporal cortex is more involved in generating bodily representations that accompany the action. We also hypothesize, that the two types of representations are relatively independent, and that action triggered by external cues may operate independently from occipitotemporal activity. An important difference between the dorsal and ventral network was that while the former exclusively showed positive connectivity (suggesting functional 'collaboration'), the latter exhibited negative FC across the hemispheres. Though it is difficult to deduce a specific functional meaning from this pattern it suggests that there is a greater need for coordination and integration in the dorsal network. A possible reason might be that the planning of a specific motor response requires selecting the appropriate motor plan and effector, while avoiding interference from alternative action plans (e.g., through activation of another, or even multiple fingers). A smooth selection and preparation of the motor output is thus easier in an integrated functional system.

Asymmetric pattern of activation and connectivity for the dominant and nondominant hand
The INT vs. EXT contrast yielded different activation and FC patterns for the dominant and nondominant hand. Performing the task with the dominant right hand resulted in less, and more focalized neural signals that were composed of small clusters in the medial occipitotemporal and parietal regions. The nondominant hand produced more widespread differences, including parietal, premotor and medial occipitotemporal regions. In contrast, we found a strong connectivity pattern between ROIs for the dominant hand, but nearly no significant coupling for the nondominant hand. This inverse pattern of less activation -stronger connectivity may reflect increased neural efficiency when participants use their dominant hand, since similar observations were made when comparing easy tasks to more complex tasks ( Meister et al., 2005 ) or automatized (i.e., efficient) vs. non-automatized movements ( Wu et al., 2008 ). Together, these findings suggest a complementary relationship between fMRI signal strength (suggesting neural recruitment or 'effort') and integration (suggesting efficiency of information processing) during intentional action as a function of limb dominance ( Godde and Voelcker-Rehage, 2010 ;Mochizuki et al., 2009 ).

Limitations
Our study used a block design to achieve a high signal-to-noise ratio and high statistical power, the drawback of this being that participants could anticipate their responses in the INT condition while being in a reactive mode in the EXT condition. Anticipation may have rendered the INT condition easier, while simultaneously introducing the need to withhold the response until the go-signal appeared. A more important point is that we used a slightly more complex visual display for the INT condition compared to the EXT condition. We argued in Section 4.1 that occipitotemporal engagement may reflect bodily representations rather than only visual stimulation, but the precise role of occipitotemporal cortex (in particular, the lingual gyrus) remains to be determined. Engagement of occipito-temporal areas may have been partly due to increased visual stimulation in the INT condition, though its modulation by handedness makes a purely visual explanation unlikely. A possibility to disentangle visual responses from activation due to imagery is to perform a similar task that uses auditory cues. Several studies have observed engagement of lingual and fusiform gyri in tasks that require participants to establish a motor representation of the hand. One such study found a strong lingual activation when contrasting motor imagery of hands and mental rotation of hands, even though the two tasks used the same visual stimuli ( Hamada et al., 2018 ). Lingual and fusiform gyri are also involved in tasks requiring self-focused processing, such as during self-criticism ( Kim et al., 2020 ). Based on these findings, we hypothesize that lingual and fusiform gyri are not exclusively visual regions, but also establish self-related bodily representations during the generation of motor plans ( Crotti et al., 2022 ;Okamoto et al., 2021 ).

Conclusion
To summarize, our findings indicate that intentional action recruits several midline and lateral frontoparietal brain areas, as well as medial occipitotemporal cortex. By studying task-based connectivity we observed that these brain regions represent two segregated networks. Our findings suggest that even a very simple motor task recruits widespread parietal, premotor, prefrontal and even associative occipitotemporal areas, suggesting that a complex interaction of functional components contributes to the generation of action intentions.

Declaration of Competing Interest
None.

Data Availability
The datasets generated during this study are available from the corresponding author.
Data will be made available on request.