Abstract
Response inhibition is a primary executive control function that allows the withholding of inappropriate responses, and requires appropriate perception of the external environment to achieve a behavioral goal. It remains unclear, however, how response inhibition is achieved when goal-relevant information involves perceptual uncertainty. Twenty-six human participants of both sexes performed a go/no-go task where visually presented random-dot motion stimuli involved perceptual uncertainties. The right inferior frontal cortex (rIFC) was involved in response inhibition, and the middle temporal (MT) region showed greater activity when dot motions involved less uncertainty. A neocortical temporal region in the superior temporal sulcus (STS) specifically showed greater activity during response inhibition in more perceptually certain trials. In this STS region, activity was greater when response inhibition was successful than when it failed. Directional effective connectivity analysis revealed that, in more coherent trials, the MT and STS regions showed enhanced connectivity to the rIFC, whereas in less coherent trials, the signal direction was reversed. These results suggest that a reversible fronto-temporal functional network guides response inhibition and perceptual decision-making under perceptual uncertainty, and in this network, perceptual information in the MT is converted to control information in the rIFC via STS, enabling achievement of response inhibition.
SIGNIFICANCE STATEMENT Response inhibition refers to withholding inappropriate behavior and is important for achieving goals. Often, however, decision must be made based on limited environmental evidence. We showed that successful response inhibition is guided by a neocortical temporal region that plays a hub role in converting perceived information coded in a posterior temporal region to control information coded in the PFC. Interestingly, when a perceived stimulus becomes more uncertain, the PFC supplements stimulus encoding in the temporal regions. Our results highlight fronto-temporal mechanisms of response inhibition in which conversion of stimulus-control information is regulated based on the uncertainty of environmental evidence.
Introduction
Inhibition of inappropriate responses is a representative executive control function guiding adaptation to changing environments. Prior neuropsychological, electrophysiological, and neuroimaging studies have identified critical brain regions associated with response inhibition in the fronto-parietal cortex and subcortical nuclei (Konishi et al., 1996; Garavan et al., 1999; Bokura et al., 2001; Braver et al., 2001; Menon et al., 2001; Li et al., 2006; Xue et al., 2008; Chambers et al., 2009; Swann et al., 2009). Among these regions, the right inferior frontal cortex (rIFC) is thought to be responsible for a core mechanism underlying response inhibition (Aron et al., 2004; Chikazoe et al., 2007). The aforementioned studies assumed that task-relevant information is appropriately extracted by the perception of external environments, and used behavioral tasks involving perception of distinctive sensory stimuli (Fig. 1A). In our daily life, however, goal-relevant environmental information is not always available or distinctive. It is thus unclear how response inhibition is achieved when environmental information involves perceptual uncertainties.
A, Response inhibition and perceptual uncertainty. B, Three possible functional topological structures of response inhibition under perceptual uncertainty. C, Random dot motion stimulus. White dots were presented within a donut-shaped display circle as indicated by dotted lines. Each arrow indicates the direction of motion of its corresponding dot. D, Discrimination task and go/no-go tasks. Motion dot stimuli were cued by a red fixation cross followed by a white fixation cross.
Perceived sensory information from the external environment guides a course of action, a process referred to as perceptual decision-making. This function involves extraction of relevant sensory information and making suitable decisions. Studies of perceptual decision-making have used behavioral tasks that demand discrimination of presented sensory stimuli (Newsome and Paré, 1988; Corbetta et al., 1991; Erwin et al., 1992; Romo et al., 1998). In a common task, randomly moving dots are visually presented and motion strength (coherence) is manipulated by the proportion of dots moving coherently toward one direction. Behavioral agents are then required to judge the overall direction of movement of the dot stimuli. It has been shown that, when a low-coherence dot stimulus is presented, accuracy is reduced and reaction times (RTs) become longer (Shadlen et al., 1996; Mazurek et al., 2003; Palmer et al., 2005). Electrophysiological and neuroimaging studies have shown that perception of moving dots is associated with the middle temporal (MT or V5) cortex, with greater activity when the motion coherence is high (Newsome et al., 1989; Britten et al., 1992; Zohary et al., 1994; Kim and Shadlen, 1999; Braddick et al., 2001; Huk et al., 2002; Kayser et al., 2010). Importantly, MT activity is enhanced when behavioral agents actively attend to the movement (Beauchamp et al., 1997; Mante et al., 2013; J. Zhang et al., 2013; Kumano et al., 2016; Tsumura et al., 2021), suggesting that MT involvement reflects the perception of goal-relevant information rather than simple accumulation of stimulus evidence. As such, the MT plays an important role in perceptual decision-making regarding moving stimuli.
Given the characteristics of response inhibition and perceptual decision-making, these two functions may be closely related. Specifically, response inhibition depends on perceived information of the external environment, and this information should be relevant to achieving response inhibition. Then, we asked how response inhibition and perceptual decision-making interact when task-relevant information involves perceptual uncertainty.
Computational modeling studies have demonstrated that a standard two-choice model well explained decision-making under perceptual uncertainty (Palmer et al., 2005) and choice biases during response inhibition (Gomez et al., 2007; Ratcliff et al., 2018), suggesting that perceptual decision-making and response inhibition are guided by unitary processing. On the other hand, neuroimaging studies have reported decision-unrelated neural activity during response inhibition (Wessel, 2018; Raud et al., 2020), contrastively suggesting that response inhibition and perceptual decision-making involve distinct processing.
In this context, neural mechanisms underlying response inhibition and perceptual decision-making may be coordinately involved in a brain-wide functional network. Then, motivated by the collective neuroimaging, neurophysiological, neuropsychological, and computational modeling studies, we hypothesized that there are three possibilities regarding the topological structure of the network: (1) a functionally merged region implements response inhibition under perceptual uncertainty (Fig. 1B, left) (Shackman et al., 2011); (2) distinct regions responsible for response inhibition (e.g., rIFC) and perceptual decision-making (e.g., MT) interact to guide response inhibition (Fig. 1B, middle) (Egner and Hirsch, 2005; Kayser et al., 2010; Waskom et al., 2014; Tsumura et al., 2022); and (3) a hub-like region links the regions involved in response inhibition and perceptual decision-making (Fig. 1B, right) (Cole et al., 2013; Nee and D'Esposito, 2016; Jiang et al., 2018). To test these possibilities, fMRI was performed while human participants performed a go/no-go task based on visually presented motion stimuli in which perceptual uncertainties were manipulated by the coherence of randomly moving dots (Fig. 1C,D). We first explored brain regions associated with response inhibition, motion strength, and their interaction. Then, we examined the directional effective connectivity between the task-related regions to examine their functional topological architecture.
Materials and Methods
Experimental design
Participants
Written informed consent was obtained from 26 healthy right-handed participants (10 females; age range: 18-22 years). Experimental procedures were approved by the institutional review boards of Keio University and Kochi University of Technology. Participants received 2000 yen for participation in each of two experimental sessions. Sample size was determined before the study based on behavioral piloting experiments.
Session procedures
The experiment consisted of two sessions conducted on separate days. The first day involved practice sessions, where participants performed a discrimination task for motion dot stimuli (Fig. 1C,D). On the second day, they first performed the same discrimination task during fMRI scanning. Then, they performed a go/no-go task based on motion dot stimuli identical to those used in the discrimination task (Fig. 1C,D).
Stimuli
All stimuli were generated with MATLAB version 2012a, using the Psychophysics Toolbox extension version 3.0.10 (Brainard, 1997; Pelli, 1997). The current stimuli were similar to those used in a previous study of perceptual decision-making (Chen et al., 2015) and task switching (Tsumura et al., 2021, 2022). Each motion stimulus was composed of 150 white dots moving inside a donut-shaped display patch, with a white cross in the center of the patch on a black background (Fig. 1C). The display patch was centered on the screen and extended from 6° to 12° of visual angle. Within the display patch, every dot moved at a speed of 10° of visual angle per second. Some dots moved coherently toward one direction while the others moved randomly. The percentage of coherently moving dots determined the “coherence,” which was set at one of three values (20%, 40%, and 80%). To remove local motion signals, dot presentation was controlled following a standard method for generating motion stimuli (Shadlen et al., 1996). Specifically, on stimulus onset, new dots were presented at new random locations in each of the first three frames. Each dot was relocated after two subsequent frames, so that the dots in frame 1 were repositioned in frame 4, and the dots in frame 2 were repositioned in frame 5, etc. When repositioned, each dot was either randomly presented at the new location or aligned with the predetermined motion direction (upward or downward), depending on the predetermined motion strength in that trial. Each stimulus was composed of 18 video frames at 60 Hz refresh rates (i.e., 300 ms presentation). Before dot presentation, the color of the central fixation cross was changed to red from white to cue dot stimulus presentation. The cue stimulus was presented for 0.75 s. The color of the fixation cross became white when dot stimuli disappeared and remained white for 1.75 s.
Experimental procedure
Direction discrimination task
At the beginning of each trial, a dot patch was presented, and participants were required to judge the direction of overall motion (up or down; Fig. 1D) and to press the corresponding button (left or right) with their right thumb as quickly and correctly as possible. In the discrimination task, participants were instructed to place their right thumb between the left and right buttons after each response. The response window was 1050 ms from the onset of a motion dot stimulus presentation that lasted 300 ms. The stimulus–response relationship was identical in the practice and scanning sessions and was counterbalanced across participants.
Each participant completed six runs, and each run involved 70 trials. For familiarization, the first five trials in each run used the highest coherence (80%) and were not included in data analysis. The last five trials in each run also used high coherence and the results were discarded. Thus, in each run, the remaining middle 60 trials (20 trials for each coherence level) were analyzed.
Response inhibition task
Participants performed a go/no-go task (Fig. 1D) using random dot stimuli. As in the discrimination task, a dot patch was presented in each trial. Participants had to press a button as quickly as possible when the motion of the dot stimuli was upward (or downward) (go trial), whereas in trials with the alternative direction, they had to withhold the response (no-go trial). Motion directions for the go and no-go trials were counterbalanced across participants.
They were told to use only one button for the go trials and not to use the other button that was used in the discrimination task. They were also instructed to place their right thumb on the button for the go trials and to avoid moving their thumb toward the other button. Thus, the errors in the no-go trials consisted of the commission errors only.
Each participant completed nine functional runs, each of which involved 70 trials. The first and last five trials of each run were high-coherence go trials, and their results were discarded. The remaining 60 trials involved 48 go trials and 12 no-go trials, and trials with each coherence level were equally distributed throughout one functional run (i.e., 20 trials for each coherence level).
Imaging procedure
MRI scanning was performed by a 3-T MRI scanner (Siemens Verio) with a 32-channel head coil. Functional images were acquired using a multiband acceleration gradient-echo EPI sequence (TR: 0.8 s; TE: 30 ms; flip angle: 45 degrees; 80 slices; slice thickness: 2 mm; in-plane resolution: 2 × 2 mm; multiband factor: 8). Each functional run involved 256 volume acquisitions. Data on the first 10 volumes were discarded to take into account the equilibrium of longitudinal magnetization. High-resolution anatomic images were acquired using an MP-RAGE T1-weighted sequence (TR: 2500 ms; TE = 4.32 ms; flip angle: 8 degrees; 192 slices; slice thickness: 1 mm; in-plane resolution: 0.9 × 0.9 mm2).
Statistical analysis
Behavioral analysis
Accuracy and RTs were calculated for each of the two tasks (discrimination and go/no-go tasks) and each coherence level (20%, 40%, and 80%), and then compared across tasks and coherence levels. Statistical tests were performed based on repeated-measures ANOVAs using SPSS 23 (IBM).
To examine whether perceptual sensitivity to motion stimulus differed between the discrimination and go/no-go tasks, receiver operating characteristic analysis (Luce et al., 1963) was performed. For each task and coherence level, differences in z scores for hit and false alarm rates were calculated as d′ values (Macmillian and Creelman, 1991), formulated as follows:
Where, ZFA and ZHit denote z-scored false alarm and hit rates, respectively.
Image preprocessing
MRI data were analyzed using SPM12 software (https://www.fil.ion.ucl.ac.uk/spm/). All functional images were first temporally realigned across volumes and runs, and the anatomic image was coregistered to the mean of the functional images. The functional images were then spatially normalized to a standard MNI template with normalization parameters estimated for the anatomic scans. The images were resampled into 2 mm isotropic voxels, and spatially smoothed with a 6 mm FWHM Gaussian kernel.
GLM analysis
Single-level
Correct trials in the go/no-go task
A GLM approach was applied to estimate parameter values for task events. The events of interest were correct no-go and go trials with the parametrical effect of coherence levels normalized (z-scored) across trials. For the go trials, normalized RTs were also added as a parametrical effect. Error trials in the go and no-go trials were separately coded in GLM as nuisance effects. These task events were time-locked to the onset of motion dot stimuli and then convolved with a canonical HRF implemented in SPM. Additionally, six-axis head movement parameters, white matter signals, and CSF signals were included in GLM as nuisance effects. Then, parameters were estimated for each voxel across the whole brain.
Corrects and errors in the no-go trials
To examine the effect of success in the no-go trials, a separate GLM analysis was performed in which the no-go trials were encoded with parametrical effects for success/error, coherence levels, and RTs (error trials only), normalized across trials. One issue to be considered when comparing success and error trials is that the errors occurred more often in low coherence trials than in high-coherence trials (Fig. 2A). Thus, if the correct and error trials were separately coded in GLM, the sets of these trials involved more high- and low-coherence trials, respectively. Then, the contrast between correct versus error would be contaminated by the coherence effect. To circumvent this potential confound, the correct and error trials were parametrically coded in one regressor. The parametrical effect of motion coherence was also used to reduce the effect of these unbalanced trial events. Correct and error go trials and nuisance effects were coded similarly to the analysis of the correct no-go and go trials.
Discrimination task trials
For the discrimination task, a separate GLM analysis was performed. Event coding was similar to that in the go/no-go task, with correct trials coded separately based on the direction used in the go/no-go task. Nuisance effects were coded similarly to the analysis of the correct trials in the go/no-go task.
Group-level analysis
Maps of parameter estimates were first contrasted for each participant. The contrast maps were collected from all participants and were subjected to group-mean tests based on nonparametric permutation methods (5000 permutations) implemented in randomize in the FSL suite (Eklund et al., 2016) (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki). Then voxel clusters were identified using a voxel-wise uncorrected threshold of p < 0.001. The voxel clusters were tested for significance with a threshold of p < 0.05 corrected by the family-wise error (FWE) rate. This procedure was empirically validated to appropriately control the false positive rate when correcting p values across the whole brain (Eklund et al., 2016). The peaks of significant clusters were then identified and listed in tables. If multiple peaks were identified within 12 mm, the most significant peak was kept.
To test whether brain regions involved in response inhibition and perceptual decision-making were overlapped spatially, a conjunction analysis was performed. We calculated logical AND activation maps showing significant activity in two contrasts: response inhibition (no-go vs go; Fig. 3B) and perceptual demand (negative coherence effect; Fig. 3A). Then, the centers of gravity of voxel clusters and their cluster sizes were listed in a table.
Dynamic causal modeling (DCM) analysis
To examine task-related functional connectivity during the go/no-go task under perceptual uncertainty, DCM (Friston et al., 2003) analysis was performed. DCM allows us to explore effective connectivity among brain regions under the premise of the brain as a deterministic dynamic system that is subject to environmental inputs and that produces outputs based on a space-state model. DCM constructs a nonlinear system involving intrinsic connectivity, task-induced connectivity, and extrinsic inputs. Parameters of the nonlinear system are estimated based on fMRI signals (system states) and task events.
The MT is well known to be a neocortical temporal region that is specifically associated with motion perception (Shadlen et al., 1996; Beauchamp et al., 1997; Corbetta and Shulman, 2002; Huk et al., 2002). The rIFC is thought to be a core region involved in implementing response inhibition (Garavan et al., 1999; Konishi et al., 1999; Bokura et al., 2001; Braver et al., 2001; Menon et al., 2001; Aron et al., 2004, 2007; Aron and Poldrack, 2006; Li et al., 2006; Chikazoe et al., 2007, 2009; Li et al., 2008; Xue et al., 2008; Chambers et al., 2009; Swann et al., 2009; White et al., 2014; Osada et al., 2019). The superior temporal sulcus (STS) showed prominent activity during high-coherent no-go trials in the current study (Fig. 4A; see Table 3). The pre-supplementary motor area (pre-SMA) is also thought to be an important region for response inhibition (Braver et al., 2001; Menon et al., 2001; Aron and Poldrack, 2006; Aron et al., 2007; Chikazoe et al., 2007, 2009; Li et al., 2008; Xue et al., 2008; Chambers et al., 2009). Given existing evidence regarding the MT, rIFC, and pre-SMA and the current results for the STS, the ROIs in the DCM analysis were defined as MT, STS, rIFC, and pre-SMA.
These ROIs were created as 6-mm-radius spheres centered on peak coordinates determined by the following procedures that avoided circular analysis (Kriegeskorte et al., 2009). The center coordinate of the MT ROI was defined based on the group-level statistical map for the coherence effect during the discrimination task (x, y, z) = (42, −66, 4) (Fig. 3A; Table 1), independently of our go/no-go data.
Brain regions showing a significant motion coherence effect in the discrimination taska
The center coordinates of the rIFC and pre-SMA ROIs were defined based on a prior study using a go/no-go task [rIFC: (48, 14, 28); pre-SMA: (4, 14, 50)] (R. Zhang et al., 2017), independently of our go/no-go data. For reproducibility, another set of rIFC and pre-SMA ROIs was defined based on meta-analysis of response inhibition [rIFC: (46, 10, 28), pre-SMA: (2, 12, 48)] (Yarkoni et al., 2011) (https://www.neurosynth.org/analyses/terms/response%20inhibition/), and we confirmed that fundamental results were preserved.
For the STS, the center coordinate of the ROI was defined based on the coherence effect of the no-go trials relative to go trials using a leave-one-subject-out procedure. More specifically, the STS ROI was defined on an individual basis: for each participant, the coordinate was determined based on the group-level statistical map from which the participant was excluded when the group-level statistics were calculated (Tsumura et al., 2021, 2022). The mean center coordinate was (53.4, −32.6, −2.4) ± (1.7, 1,5, 1.7) (mean ± SD).
The signal time courses of four ROIs and the regressors in events of interest were then extracted from first-level GLMs. Time courses are concatenated across functional runs, and nuisance effects of head motion, white matter signals, ventricle signals, whole-brain signals, functional runs, and contrast were subtracted out for ROI time courses.
For the no-go trials, causal models were defined as those that differed among ROIs in terms of external inputs and modulatory effects. In the analyses of three ROIs (MT, STS, and rIFC, Fig. 5A; MT, STS, and pre-SMA, Fig. 5B), the tested models comprised 512 types (i.e., 23 inputs and 26 connection effects). In the analysis of four ROIs (MT, STS, rIFC, and pre-SMA), the tested models comprised 65,536 types (i.e., 24 inputs and 212 connection effects). Then, connectivity matrices reflecting (1) first-order connectivity, (2) effective changes in coupling induced by the inputs, and (3) extrinsic inputs to ROIs were estimated using SPM12 for each of possible models (i.e., 512 models in three-ROI analysis and 65,536 models in four-ROI analysis) based on DCM analysis. A parametric regressor for motion coherence level in the no-go trials was used as the extrinsic effect for effective connectivity between ROIs and ROI inputs.
To estimate the effective connectivity strength, a Bayesian model reduction method (Friston et al., 2016) was used. This reduction method enables the calculation of posterior densities of all possible reduced models, and these posterior densities were then inverted to a fully connected model. In the current study, following the default setting of SPM12, models with posterior probability of 0 were excluded. Then, estimation of model parameters was achieved using 442 of 512 models for the MT, STS, and rIFC ROIs (Fig. 5A), and 448 of 512 models for the MT, STS, and pre-SMA ROIs (Fig. 5B). In the analysis of the MT, STS, rIFC, and pre-SMA ROIs, 65,536 (i.e., 24 inputs and 212 connection effects) models were possible; and of these, 45,730 models were retained to estimate model parameters. The reduced models were supplemented with the second-level parametric empirical Bayes method (Friston et al., 2016) to apply empirical priors that remove subject variability from each model.
Subsequently, parameters of these models were estimated based on Bayesian model averaging (Penny et al., 2010) to estimate connectivity strength. Because the current analysis aimed to identify the average effective connectivity observed across participants, we used a fixed effect estimation assuming that every participant uses the same model, rather than a random effect estimation assuming that different participants use different models; this latter estimation method has been often used to examine group differences in effective connectivity (Penny et al., 2010). The significance of connectivity was then tested by a posterior probability density and corrected for FWE rates across task-related connectivity patterns.
In our previous studies (Tsumura et al., 2021, 2022), we examined Bayesian model averaging-based DCM without model reduction or an empirical prior, and confirmed that the overall results of the previous studies were consistent across these estimation methods.
The numbers of the correct no-go trials analyzed in the current DCM analysis were 30.6 ± 1.0, 26.0 ± 1.1, and 19.5 ± 1.4 (mean ± SD) trials in the high-coherence (80%), middle-coherence, and low-coherence trials, respectively, which were comparable to our previous studies (Tsumura et al., 2021, 2022). We also note that, in the current and previous studies, DCM analyses were performed using similar procedures, and neuroimaging data were collected using the identical MRI scanner and high temporal resolution scanning parameters.
Results
Behavioral results
In the discrimination task, accuracy was lower in the low-coherence (i.e., more uncertain) trials (F(1,25) = 111.5; p < 0.001, planned contrast of a linear effect; Fig. 2A), replicating previous studies of perceptual decision-making (Shadlen et al., 1996; Kim and Shadlen, 1999; Mazurek et al., 2003; Palmer et al., 2005; Kayser et al., 2010).
Behavioral results. A, Accuracy in the discrimination and go/no-go tasks as a function of motion coherence. The vertical and horizontal axes indicate accuracy and motion coherence, respectively. Blue and orange lines indicate the discrimination and go/no-go tasks, respectively. Solid and dotted lines indicate trials with no-go and go directions, respectively. B, RTs for the discrimination and go/no-go tasks. The vertical and horizontal axes indicate RTs and motion coherence, respectively. C, Sensitivity to motion-dot stimuli during the discrimination and go/no-go tasks. The vertical and horizontal axes indicate d′ values and motion coherence, respectively.
In the go/no-go task, the commission error rate in the no-go trials (i.e., 1 – accuracy) was higher in low-coherence trials (F(1,25) = 47.2; p < 0.001, planned contrast of a linear effect; Fig. 2A), and the omission error rate in the go trials (i.e., 1 – accuracy) was higher in low-coherence trials (F(1,25) = 38.0; p < 0.001, planned contrast of a linear effect), suggesting that the commission and omission errors occurred because of the failure of motion direction perception. The commission error rate was higher than the omission error rate (F(1,25) = 42.0; p < 0.001; Fig. 2A). This suggests that the commission errors occurred because of the prepotency of go response, consistent with previous studies of go/no-go tasks (Bokura et al., 2001; Chikazoe et al., 2009). The commission error rate was higher than the error rate in the direction-matched trials in the discrimination task (F(1,25) = 62.7; p < 0.001; Fig. 2A).
To compare accuracy in the go trials and direction-matched trials in the discrimination task, a repeated-measures ANOVA with the trial (go and direction-matched discrimination task trials) and the coherence levels (high, middle, and low) as factors was performed. Significant main effect of trials (F(1,25) = 7.6; p < 0.05; Fig. 2A) and a significant interaction effect (F(1,25) = 9.4; p < 0.01, planned contrast of a linear effect) were observed, indicating that, in the go trials, accuracy was higher and coherence effect was weaker relative to those in the direction-matched discrimination task trials. This result suggests that prepotent tendency for go response was enhanced in the go/no-go task.
Then, we compared the coherence effect between the no-go and go trials. A repeated-measures ANOVA with the trial (go and no-go) and the coherence levels as factors revealed a significant interaction effect (F(1,25) = 6.8; p < 0.05; linear coherence effect; Fig. 2A), indicating that the motion coherence effect was greater in the no-go trials, compared with the go trials. This suggests that the commission errors occurred because of the low motion coherence and the response prepotency conjointly. We then compared the coherence effect between the no-go and motion direction-matched discrimination task trials. A repeated-measures ANOVA with the trial (no-go and direction-matched discrimination task trials) and coherence levels as factors revealed a significant interaction effect (F(1,25) = 5.1; p < 0.05; linear coherence effect), again greater motion coherence effect in the no-go trials than in the discrimination task trials.
To test whether the greater coherence effect in the no-go trials is attributable to the change in perceptual sensitivity to motion stimuli or task demands, sensitivity to motion stimulus was examined based on d′ indices for each task and coherence level (see Materials and Methods). In both the discrimination and go/no-go tasks, d′ indices were greater in the high-coherence trials, consistent with the accuracy results (Fig. 2C). If the change in sensitivity to motion stimuli explains the greater coherence effect on accuracy in the no-go trials (i.e., lower accuracy in low coherence trials), the slope of d′ along the coherence level would be steeper in the go/no-go task than in the discrimination task. However, the d′ indices did not support this possibility because the d′ slope was steeper in the discrimination task (Fig. 2B). These results suggest that the greater coherence effect in the no-go trials is attributable to differential task demands rather than the change in the sensitivity to motion between the go/no-go and discrimination tasks.
RTs were shorter in high-coherence trials in both the discrimination task (F(1,25) = 82.5; p < 0.001; linear coherence effect; Fig. 2B) and the go trials in the go/no-go task (F(1,25) = 32.0; p < 0.001; linear coherence effect). RTs were also shorter in the go trials than the trials in the discrimination task (F(1,25) = 120.1; p < 0.001). A repeated-measures ANOVA with trials (go trials and direction-matched discrimination trials) and coherence levels as factors revealed a significant interaction effect (F(1,25) = 6.9; p < 0.05; linear coherence effect).
These collective behavioral results suggest that the current behavioral task successfully manipulated response inhibition and perceptual decision-making; and more importantly, the motion coherence affected both perceptual decision-making and response inhibition.
Imaging results
We first explored brain regions associated with motion coherence. As shown in Figure 3A and Table 1, prominent activation was observed in the MT cortex in high-coherence trials, consistent with prior studies (Shadlen et al., 1996; Beauchamp et al., 1997; Corbetta and Shulman, 2002; Huk et al., 2002; Tsumura et al., 2021, 2022). On the other hand, in low-coherence trials, activation was greater in multiple frontal regions, including IFC and pre-SMA, also consistent with previous studies (Zatorre et al., 1994; Gevins et al., 1997; Duncan and Owen, 2000; Paulus et al., 2001; Lavie, 2005; Vickery and Jiang, 2009; Graves et al., 2010; Sheth et al., 2012; Tsumura et al., 2021, 2022).
A, Statistical maps of the parametrical effect of motion coherence in the discrimination task. Maps are overlaid onto a 3D surface of the brain. Hot and cool colors represent positive and negative effects, respectively. White arrows indicate the MT, IFC, and pre-SMA regions with MNI coordinates. B, Statistical activation maps for signal increases and decreases in the no-go trials relative to go trials. Hot and cool colors represent signal increases and decreases, respectively, in the no-go trials. Formats are similar to those in A. White arrowheads indicate the IFC, MT, pre-SMA, anterior insular (aINS), dorsolateral PFC (DLPFC), inferior frontal junction (IFJ), posterior parietal cortex (PPC), temporoparietal junction (TPJ), and ACC regions with MNI coordinates. C, Brain regions showing a conjunction effect of perceptual demand (negative effect in A) and response inhibition (B).
We next explored brain regions associated with response inhibition. In the no-go trials, relative to the go trials, robust activity was observed in fronto-parietal regions, including the IFC, inferior frontal junction, dorsolateral PFC, anterior insula, ACC, pre-SMA, posterior parietal cortex, and temporo-parietal junction, predominantly in the right hemisphere (Fig. 3B; Table 2). These results are consistent with prior studies of response inhibition (Garavan et al., 1999; Rubia et al., 2003; Aron et al., 2004; Li et al., 2006; Chambers et al., 2009; Chikazoe et al., 2009).
Brain regions showing a significant signal increase in the no-go trials relative to the go trials in the go/no-go taska
The negative coherence effect (Fig. 3A) and response inhibition (Fig. 3B) both involved fronto-parietal regions. We then performed a conjunction analysis to test whether these regions were spatially overlapped. As shown in Figure 3C and Table 3, multiple fronto-parietal regions, including the rIFC and pre-SMA, showed the conjunction effect, suggesting that these regions are associated with both response inhibition and greater decision demands in low-coherence trials. This result is consistent with a previous study (Tsumura et al., 2021) and agrees well with the fronto-parietal multiple-demand system (Camilleri et al., 2018).
Brain regions showing a conjunction effect of response inhibition (no-go vs go) and negative coherence effectsa
Then, we explored brain regions showing significant interaction effect of motion coherence and response inhibition during the go/no-go task by contrasting parametric modulatory effect of motion coherence between the no-go and go trials. A robust difference in coherence effect was observed in the right STS (Fig. 4A; Table 4). Notably, this area showed greater activation in the high-coherence no-go trials relative to low-coherence no-go trials (Fig. 4B, left; Table 5) and in the no-go relative to go trials (Figs. 3B and 4B, right; Table 2), collectively indicating that the STS activity was enhanced when motion coherence was higher in the no-go trials.
A, Statistical maps for differential parametrical effects of motion coherence between the no-go versus go trials. Hot and cool colors represent greater positive coherence effects in the no-go and go trials, respectively. The area enclosed in the black rectangle (left) represents the area magnified on the middle. Green solid line on the surface indicates the axial section on the right. Formats are similar to those in Figure 3A. White arrowheads indicate the STS region. B, Statistical maps for parametrical effects of motion coherence for the no-go trials (left) and for activation during the no-go trials relative to go trials (right; Fig. 3B). Formats are similar to those in Figure 3A (bottom). C, Statistical activation map for signal increases and decreases indicating success and error in the no-go trials. Formats are similar to those in Figure 3A. White arrowheads indicate the STS region. D, The STS region shows greater activation in successful no-go trials (green), and coherence effects in the no-go trials are mapped onto a 2D parasagittal slice of the standard brain. The section is indicated by the X coordinate on the bottom right. OTS, Occipitotemporal sulcus.
Brain regions showing a significant difference in the coherence effect between the no-go and go trialsa
The STS involvement in high-coherence no-go trials may point to an important role of this region in both response inhibition and perceptual decision-making. In particular, STS activity during response inhibition is elevated in situations where perceptual decision-making is more facilitated, suggesting that the STS is specifically affected by perceptual decision-making during response inhibition. We then hypothesized that the success of response inhibition depends at least in part on the STS playing a hub role between perceptual decision-making and response inhibition. To test this hypothesis, we compared brain activity between success and error trials in the no-go trials, and whole-brain exploratory analysis revealed robust activation in the right STS (Fig. 4C; Table 6). Of note, this STS region is located very close to the STS region that showed a strong coherence effect in the no-go trials (Fig. 4D), with the two areas only 12.2 mm apart (Tables 2 and 6). Together, our results highlight important roles of the STS in both response inhibition and perceptual decision-making.
Brain regions showing a significant coherence effect in the no-go trialsa
Directional effective connectivity
The whole-brain exploratory analyses identified three types of brain regions: (1) the MT, associated with perceptual decision-making (Fig. 3A); (2) the rIFC and pre-SMA, both associated with response inhibition (Fig. 3B); and (3) the STS, associated with both of them (Fig. 4A,B). In order to examine the functional relationships between these regions during response inhibition under perceptual uncertainty, we performed interregional effective connectivity analysis of coherence effects in the no-go trials based on DCM (Friston et al., 2019). DCM allowed us to examine the directionality of task-related functional connectivity based on state-space models.
Brain regions showing a significant activity increase in the correct no-go trials relative to the error no-go trialsa
We first performed DCM analysis for a fronto-temporal network comprising the MT, STS, and rIFC. Figure 5A shows the task-related parametrical modulatory effect of motion coherence in the no-go trials. In the high-coherence no-go trials, connectivity was enhanced from the MT toward the rIFC via the STS (p values < 0.05, FWE-corrected). In contrast, connectivity was enhanced from the rIFC to the MT and STS in low-coherence no-go trials (p values < 0.05, FWE-corrected; Fig. 2). These results suggest a reversal of the connectivity between rIFC, STS, and MT regions depending on motion coherence: bottom-up signaling from the MT to the rIFC via the STS under low perceptual uncertainty, which reversed to top-down signaling from the rIFC to the STS and MT regions under high perceptual uncertainty.
Effective connectivity analysis. A, A model comprising MT, STS, and IFC regions. B, A model comprising MT, STS, and pre-SMA regions. C, A model comprising MT, STS, IFC, and pre-SMA regions. Red and blue arrows indicate enhancements of effective connectivity in high- and low-coherence trials, respectively, as reflected by positive and negative values of the estimated connectivity strength. *p < 0.05 (FWE-corrected).
We next performed DCM analysis for another fronto-temporal network comprising the MT, STS, and pre-SMA, but not the rIFC. In the high-coherence no-go trials, connectivity was enhanced from the MT toward the pre-SMA via the STS (p values < 0.05, FWE-corrected; Fig. 5B). In contrast, connectivity was enhanced from the pre-SMA to the STS in the low-coherence no-go trials (p values < 0.05, FWE-corrected; Fig. 5B). Interestingly, the connectivity results were similar to those of the analysis involving the rIFC instead of the pre-SMA (Fig. 5A), although top-down signaling from the pre-SMA to MT failed to reach statistical significance. Together, the connectivity results in Figure 5A, B suggest that the pre-SMA implements a function similar to that of the rIFC regarding alternated top-down and bottom-up signaling between fronto-temporal networks.
Finally, we performed a DCM analysis comprising the four ROIs. In the high-coherence no-go trials, connectivity was enhanced from the MT toward the rIFC and pre-SMA via the STS (p values < 0.05, FWE-corrected; Fig. 5C), fully replicating the bottom-up signaling in the three-ROI analyses (Fig. 5A,B). In contrast, the top-down signaling was partially replicated: in the low-coherence trials, connectivity was enhanced from the rIFC to the STS (p values < 0.05, FWE-corrected). However, the top-down signaling from the rIFC to the MT (Fig. 5A) and from the pre-SMA to STS (Fig. 5B) failed to replicate in this four-ROI analysis. These connectivity results may not be robust against the number of ROIs because of the high degree of freedom of possible models in the four-ROI analysis (65,536 models) relative to the three-ROI analysis (512 models).
Nonetheless, the results of the four-ROI analysis indicate distinctive roles of the rIFC and pre-SMA. In particular, in the low-coherence trials, connectivity was enhanced from the rIFC to the pre-SMA (p values < 0.05, FWE-corrected), suggesting that the rIFC may supplement executive control demands toward the pre-SMA when stimulus evidence of the motion dot stimulus is limited.
Discussion
The current study provides new insights regarding the relationships between perceptual decision-making and response inhibition. Our results highlighted the important role of the STS in high-coherence no-go trials and in the success of response inhibition (Fig. 6), and support Hypothesis 3: “a hub-like region links the regions involved in response inhibition and perceptual decision-making” (Fig. 1B, right; see also Introduction). The STS serves as a hub-like region that links the rIFC, which is involved in response inhibition, and the MT, which plays a role in perceptual decision-making. Specifically, in high-coherence trials, the STS received task-related signals from the MT that was associated with motion perception, and signaled the rIFC regarding response inhibition. The signal direction was reversed in low-coherence trials, such that the rIFC signaled the MT and STS. These results demonstrate that bottom-up and top-down signals are reversible, and their status depends on the uncertainty of perceptual information (Fig. 6). As such, the fronto-temporal network may convert perceptual information to executive control information via the STS to achieve response inhibition.
Schematic diagrams of neural mechanisms among the STS, IFC, and MT regions during response inhibition with perceptual uncertainty. Arrows indicate signal directions. The size of disks reflects activity magnitude.
The STS links perceptual and executive information
It is known that the MT receives motion-related information from early visual cortices and is more activated when motion coherence is higher (Desimone and Ungerleider, 1989; Ungerleider and Haxby, 1994; Desimone and Duncan, 1995; Tootell et al., 1995; Ungerleider et al., 1998). It is also known that the rIFC and pre-SMA are activated during response inhibition and signaled to basal ganglia (Rubia et al., 2003; Duann et al., 2009; Cai and Leung, 2011; Jahfari et al., 2011). In the current study, the STS was activated during successful high-coherence no-go trials; it may receive perceptual information from the MT, and send signals to the rIFC, where response inhibition is encoded. Thus, the STS may encode both perceptual decision-making and response inhibition and serve as a hub role between them (Fig. 6).
Prior studies have suggested that top-down and bottom-up signaling between neocortical temporal regions and prefrontal regions is implicated in executive control (Kawamura and Naito, 1984; Wilson et al., 1993; Desimone and Duncan, 1995; Tomita et al., 1999; Moore and Armstrong, 2003). Our task-related effective connectivity analysis demonstrated that which of these two types of signaling was active depended on perceptual uncertainty, as demonstrated in our recent study (Tsumura et al., 2022). Additionally, the STS relayed signals from the MT to rIFC, suggesting a functional route from the temporal cortex to the frontal cortex. We note that our connectivity analysis used a fixed effect model approach, which did not consider the variability of participants, limiting the generalizability of our results (compare Penny et al., 2010).
It is well known that visual information coded in the temporal cortex becomes more abstract along the posterior-to-anterior axis of the cortical areas (Desimone and Ungerleider, 1989; Ungerleider and Haxby, 1994; Desimone and Duncan, 1995; Ungerleider et al., 1998). For example, when a face stimulus is perceived, posterior temporal regions represent visual features of the face, whereas anterior regions represent the identity of the face (Freiwald and Tsao, 2010). This hierarchical information stream in the temporal cortex is consistent with the current findings. In particular, the MT, which is located in the posterior end of the inferior sulcus, represented visual perceptual information (Huk et al., 2002), and the STS, which is located anteriorly, represented a mixture of visual and control information that was converted into executive control information in the rIFC.
Many studies of visual motion perception have consistently reported the involvement of the MT (Newsome and Paré, 1988; Britten et al., 1992, 1993; Zohary et al., 1994; Shadlen et al., 1996; Beauchamp et al., 1997; Britten and Newsome, 1998; Kim and Shadlen, 1999; Braddick et al., 2001; Corbetta and Shulman, 2002; Huk et al., 2002; Mazurek et al., 2003), but some neuroimaging studies have also reported that the STS is active during the perception of visual motion stimuli (Braddick et al., 2001; Noguchi et al., 2005; Kayser et al., 2010; Tsumura et al., 2021, 2022). This STS involvement may partially reflect executive control processing, as is shown in the present study. Indeed, previous studies of response inhibition have reported involvement of the lateral temporal areas, including the temporo-parietal junction (Aron and Poldrack, 2006; Xue et al., 2008; Chikazoe et al., 2009) and distributed regions along the STS (Aron et al., 2007; Chikazoe et al., 2009). It is possible that these temporal regions convert visual information into executive information along the bottom-up stream in the neocortical temporal areas (Osada et al., 2021).
Perceptual decision-making and executive control
Computational analyses based on the drift diffusion model (DDM) have revealed that decision processes in the go/no-go task are explained by a traditional two-choice perceptual decision model. In particular, the faster RTs in the go trials and larger number of errors in the no-go trials can be modeled by a response bias toward the go trial (Gomez et al., 2007; Ratcliff et al., 2018). In the current study, this decision bias may be produced by the asymmetric stimulus proportion of the go and no-go trials (80% for go and 20% for no-go) and corresponds to the prepotent response tendency in the context of response inhibition (Wessel, 2018).
The DDM account of binary decision processes involved in the go and no-go trials is consistent with an ERP study demonstrating that decision processes in the no-go and go trials are comparable, but postdecision processing differed between these trials (Raud et al., 2020). On the other hand, in stop-signal tasks, where the response to a go signal is canceled after its initiation, the initiated processes trigger EMG activity (Jana et al., 2020; Raud et al., 2020). Notably, such EMG activity was observed in successful no-go trials (Raud et al., 2020). Additionally, when the no-go trials were presented more infrequently, task-related activity was enhanced in the successful no-go trials (Wessel, 2018). Such processing in the no-go trials can be explained by a nondecision (postdecision) component in the DDM. Together, our go/no-go task required response inhibition involved in high-order processing beyond simple two-choice decisions.
The rIFC showed brain activity during both response inhibition and high perceptual demand (Fig. 3C), which is consistent with our previous study demonstrating a conjunction of executive control and perceptual decision-making (Tsumura et al., 2021), and agrees well with the functionality of a fronto-parietal multiple-demand system (Camilleri et al., 2018). The rIFC did not show an interaction effect between them, however (Fig. 4A). The imaging results, together with computational modeling, suggest the rIFC involvement in the low-coherence no-go trials concurrent response inhibition and perceptual demand, but these processes may be independent (Fig. 6, right).
Behavioral analysis revealed that, in the go trials, accuracy was higher and the coherence effect was weaker relative to those in the direction-matched discrimination task trials. Our previous study using identical dot motion stimuli examined performance improvement in the identical motion discrimination task and found that accuracy did not improve through 6 runs within one experiment (Sarabi et al., 2018). Thus, a practice effect was not responsible for the higher accuracy in the go trials.
Our recent studies examined the relations between perceptual decision-making and task switching, another representative executive control function (Tsumura et al., 2021, 2022). In these studies, to examine task switching under perceptual uncertainty, we manipulated the motion coherence of similar random dot stimuli, as in the current study, while participants alternated discrimination tasks.
In one study (Tsumura et al., 2021), the left lateral PFC was implicated in task switching, and stimulus-modality-dependent occipitotemporal regions received complementary signals from the right lateral PFC involved in perceptual decision-making, supporting Hypothesis 2: “distinct regions responsible for response inhibition and perceptual decision-making interact to guide task switching” (Fig. 1B). In another study (Tsumura et al., 2022), top-down signaling from the PFC supplemented task-relevant information in the occipitotemporal regions, also supporting Hypothesis 2.
In contrast, the current results support Hypothesis 3: “a hub-like region links the regions involved in response inhibition and perceptual decision-making.” One helpful way to explain the variability in the possible topological architecture of the functional network is to examine whether the executive control functioning is affected by the perceptual decision-making. This can be statistically tested by interaction effects of motion coherence and executive control. Behaviorally, interaction effects between motion coherence and task switching were absent in our previous studies, indicating that the task-switching effect itself was not affected by the motion coherence level. Consistent with the behavioral results, neuroimaging analysis did not reveal significant interaction effects of task switching and motion coherence.
In the current study, by contrast, accuracy showed an interaction effect between response inhibition (no-go vs go) and motion coherence, with decreased accuracy in lower-coherence trials, particularly the no-go trials (Fig. 2A). The STS region showed a consistent interaction effect, with greater activity in high-coherence no-go trials. These behavioral and neural interaction effects between perceptual decision-making and response inhibition may reflect the hub role of the STS. As such, the STS may relay task-related information from the MT to the rIFC by converting perceptual information in the MT into response inhibition information in the rIFC, thus successfully exerting executive control to achieve a behavioral goal.
One important functional characteristic of the STS is that it is activated during response inhibition, and its activity is enhanced in high-coherence trials (Fig. 4A,B), which corresponds to the statistical interaction effect of behavioral performance (Fig. 2A). On the other hand, the interaction effect of executive control and perceptual uncertainty was absent both behaviorally and neurally in our previous studies (Tsumura et al., 2021, 2022). If an executive control function shows a statistical interaction effect of behavior with perceptual uncertainty, the function may involve the STS. This issue should be addressed in the future. For example, stop-signal tasks require the cancelation of an already initiated response, whereas go/no-go tasks require withholding of prepotency of a response, and these two tasks involve differential cognitive control processing (Raud et al., 2020). Thus, introducing a random dot motion stimulus in a stop-signal task is an interesting and important extension of the current study.
Prefrontal involvements in high decision demands
In low-coherence trials, the MT was less activated than in high-coherence trials, reflecting the slower accumulation of perceptual information (Newsome and Paré, 1988; Britten et al., 1992, 1993; Zohary et al., 1994; Shadlen et al., 1996; Beauchamp et al., 1997; Britten and Newsome, 1998; Kim and Shadlen, 1999; Braddick et al., 2001; Corbetta and Shulman, 2002; Huk et al., 2002; Mazurek et al., 2003; Kayser et al., 2010). On the other hand, fronto-parietal regions, such as the IFC, inferior frontal junction, and pre-SMA, and posterior parietal cortex, were activated in low-coherence trials in the current discrimination task (Fig. 3A) and in our previous studies (Tsumura et al., 2021, 2022). This strong prefrontal involvement may reflect greater decision demands for low-coherence motion stimuli, which is consistent with increased frontal activation in high cognitive load situations (Zatorre et al., 1994; Gevins et al., 1997; Duncan and Owen, 2000; Paulus et al., 2001; Lavie, 2005; Vickery and Jiang, 2009; Enriquez-Geppert et al., 2010; Graves et al., 2010; Sheth et al., 2012). Our results suggest that such strong prefrontal involvement in low-coherence trials supplements the perceptual information in the MT, consistent with our previous studies (Tsumura et al., 2021, 2022).
Footnotes
This work was supported by Japan Society for the Promotion of Science Kakenhi 24800023, 19K12121, 20K07727, and 21H05060 to K.J., 20H00521 and 21K18267 to M.T., 17H00891 to K.N., and 17H06033 to J.C.; Uehara Memorial Foundation Grant to K.J.; and Takeda Science Foundation Grants to K.J. and M.T. We thank Drs. Russell A. Poldrack, Alexander C. Huk, and Corey N. White for scientific advice; and Maoko Yamanaka for administrative assistance.
The authors declare no competing financial interests.
- Correspondence should be addressed to Koji Jimura at koji.jimura{at}gmail.com
