Abstract
Feelings of cognitive fatigue emerge through repeated mental exertion and are ubiquitous in our daily lives. However, there is a limited understanding of the neurobiological mechanisms underlying the influence of cognitive fatigue on decisions to exert. We use functional magnetic resonance imaging while participants (18 females, 10 males) chose to exert effort for reward, before and after bouts of fatiguing cognitive exertion. We found that when participants became cognitively fatigued, they were more likely to choose to forgo higher levels of reward for more effort. We describe a mechanism by which signals related to cognitive exertion in the dorsolateral prefrontal cortex influence effort value computations, instantiated by the insula, thereby influencing an individual's decisions to exert while fatigued. Our results suggest that cognitive fatigue plays a critical role in decisions to exert effort and provides a mechanistic link through which information about cognitive state shapes effort-based choice.
Significance Statement
Cognitive fatigue influences decisions to exert effort throughout our daily lives. However, it is unclear how the brain integrates information about cognitive fatigue to influence effort-based decision–making. We found that when individuals engage in repeated cognitive exertion, exertion-related signals in the dorsolateral prefrontal cortex influence effort value computations in the insular cortex, thereby influencing an individual's decisions to exert while fatigued. These results provide a mechanistic account of how information about cognitive state impacts choice.
Introduction
The workday is filled with tasks that require a great deal of cognitive effort, whether that be meeting with clients and coworkers, preparing for presentations, or simply replying to emails. As the day progresses, our engagement in these effortful tasks leads to fatigue, which impacts our choices by making us less willing to exert cognitive effort. Recent studies in cognitive neuroscience have begun to dissect the circuitry underpinning physical fatigue and its influence on effort-based decision–making (Hogan et al., 2020; Müller et al., 2021). While previous works have examined the effects of cognitive fatigue on task engagement and performance (Hopstaken et al., 2015, 2016; Matthews et al., 2023), there is a limited understanding of how the emergence of fatigue during repeated cognitive exertion influences effort-based decision–making at the level of brain and behavior.
Experiments examining the influence of physical fatigue on effort-based decision–making have shown that fatigued individuals increase their subjective costs of physical effort, and higher rewards are needed to incentivize exertion while in a fatigued state (Meyniel et al., 2013; Hogan et al., 2020; Müller et al., 2021). Physical effort costs and effort decisions fluctuate with individuals’ momentary fatigue (Meyniel et al., 2013, 2014). Neuroimaging during exertion and choice has revealed a network of brain activity that underlies effort-based choices and is impacted by physical fatigue (Croxson et al., 2009; Prévost et al., 2010; Hogan et al., 2019, 2020; Müller et al., 2021). This network includes the anterior cingulate cortex (ACC), bilateral anterior insula, and ventromedial prefrontal cortex (vmPFC), which are integral for computing the value of effortful options and making effort-based decisions. It has also been shown that during physical effort-based decision–making, the right anterior insula (rIns) is sensitive to fatigue-induced changes in prospective effort and is related to exertion signals in the premotor cortex (Hogan et al., 2020). A study that examined physical exertions directly interleaved with choice also showed that the medial and lateral prefrontal cortex tracked individuals’ states of fatigue and that a frontostriatal network integrated fatigue with effort value (Müller et al., 2021).
One important factor in signaling fatigue is the ability to sense one's internal state. Notably, the rIns is responsible for encoding proprioceptive signals, and the anterior insula encodes the effort value (Skvortsova et al., 2014; Chong et al., 2017; Pessiglione et al., 2018; Hogan et al., 2020). We recently showed that in the context of physical effort-based decision–making, feelings of fatigue mediate effort values encoded by the insula (Hogan et al., 2020). However, previous studies that examined neural signals at the time of exertion and rest were focused on physical effort and were not designed to examine prospective cognitive effort valuation and fatigue (Tanaka et al., 2014; Hopstaken et al., 2015, 2016; Wang et al., 2016; Müller et al., 2021). Furthermore, while recent studies have examined how physical and cognitive fatigue influence cognitive control during a temporal discounting task, they did not evaluate how fatigue influences cognitive effort valuation (Blain et al., 2016, 2019). Therefore, it is unclear how brain signals related to cognitive fatigue might influence effort valuation and whether similar functions of rIns are involved in signaling cognitive fatigue and cognitive effort-based choice. To our knowledge, there have been no studies that have directly tested how bouts of cognitive exertion influence brain signals related to the prospective valuation of cognitive effort and resulting decisions. As such, there is a limited neurobiological understanding of how cognitive fatigue influences effort valuation and decisions to engage in cognitive exertion.
In this study, we investigated the influence of cognitive fatigue on behavioral representations of the subjective effort value and the neural mechanisms by which fatigue influences the brain's decision-making circuitry. We hypothesize that fatigue, resulting from repeated cognitive exertion, will reduce individuals’ willingness to exert. In a fatigued state, compared with a rested state, when individuals face the option of exerting greater prospective cognitive effort for reward, they will be less willing to accept high-effort options. This hypothesis has its basis in previous studies of physical effort-based decision–making that found that individuals exhibit increased costs of physical effort when in a fatigued state (Hogan et al., 2020; Müller et al., 2021). We hypothesize that decisions about prospective cognitive effort exertion have their basis in a value signal encoded in the ACC and insula. This hypothesis is informed by neuroimaging studies, which consistently show that activity in regions of the ACC and insula is related to the subjective effort value (Chong et al., 2017; Arulpragasam et al., 2018; Hogan et al., 2019, 2020; Westbrook et al., 2019). Given our recent study of physical fatigue, which showed that rIns was sensitive to fatigue-induced changes in effort-based decision–making (Hogan et al., 2020), we predict that the insula is also sensitive to changes in cognitive effort value as a function of cognitive fatigue. We predict that brain regions related to cognitive exertion will be functionally coupled to rIns during choice, suggesting a network through which exertion and fatigue are translated into subsequent effort choices. These hypotheses form a neurobiological framework of cognitive fatigue, which recruits brain regions responsible for effort valuation and cognitive exertion to inform decisions about prospective effort while fatigued.
Materials and Methods
Participants
The Johns Hopkins School of Medicine Institutional Review Board approved this study. Participants were right-handed, had no self-reported history of psychiatric or neurological disorders, and had no self-reported previous adverse experiences undergoing a magnetic resonance imaging (MRI) exam. All participants provided informed consent before beginning the study.
Thirty-four individuals participated in this study. Two participants were excluded for poor performance on the n-back task (not identifying at least 50% of targets on average), two were excluded for making choices with no variability (i.e., accepting all offers), and two were excluded for having ratings of n-back difficulty that declined with increasing n (i.e., more difficult n-back tasks were not perceived as more demanding). This left a final cohort of n = 28 participants (mean age 25 years; standard deviation in age, 4 years; 18 females).
Experimental paradigm
Before the experiment, participants were informed that they would receive a $50 show-up fee and that the opportunity for additional monetary earnings depended on their choices and task performance.
This customized experimental procedure was built in MATLAB 2020b (The Mathworks Inc, 2020) using Psychophysics Toolbox Version 3 (Kleiner et al., 2007) for stimuli presentation and behavioral data collection.
The experiment began with an association phase, performed outside the MRI scanner, in which participants performed different cognitive effort levels of an n-back working memory task (Fig. 1A). In this paradigm, cognitive effort was operationalized as the level of the n-back task—higher n corresponded to higher cognitive effort. Each n-back trial consisted of 40 sequentially presented letters. Within this sequence were 10 target letters that a participant would have to identify as the same letter n letters prior. Participants had 2 s to identify whether the current letter on the screen was one of these targets. Once the participant entered their choice, the next letter was presented. Feedback was presented at the end of each task. If participants correctly identified 50% of the targets, they received a “Success!” message on the screen. If they failed to achieve 50% success, they received a message of “Please try harder!” and continued to the next n-back level. Six levels of n-back were presented (n = 1–6), a distinct color was pseudorandomly assigned to each level, and the stream of sequential letters was presented in this color. Participants were presented with varying levels of n-back in blocks of three trials. Each level was presented in a random order without replacement.
Following this association phase, participants were randomly presented with color cues that had previously been associated with n-back levels and were instructed to rate their perceived level of mental demand (Fig. 1B). Ratings were made on a continuous scale from “Very Low” to “Very High.” The association phase was also performed outside the MRI scanner.
To examine the effect of cognitive fatigue on behavioral and neural representations of effort valuation, we scanned participants’ brains with functional MRI (fMRI) while they made decisions about prospective cognitive effort and monetary reward, before and after they performed repetitive fatiguing cognitive exertions. Before being presented with the effort/reward decisions, participants were told that two of their decisions would be randomly selected (one from decisions before and one from after cognitive exertions) and played out at the end of the experiment. Since trials were extracted randomly, participants were instructed that they did not need to spread their exertions over all their trials and that they should treat each effort decision individually.
During the baseline choice phase, which was meant to elicit effort and reward preferences in a rested state, participants were presented with a series of effort/reward choices between performing a one-back task for $1 (default option) or a higher level of cognitive effort for reward (nondefault option). Participants made their choices by pressing one of two buttons on a handheld button box with their right hand (Cedrus Corporation, Cedrus RB-830). For the nondefault options, the effort levels were drawn from the six color cues introduced in the association phase, and rewards ranged from $1 to $8 in $1 increments. The effort/reward choices used for this experiment were specifically designed to span a range of potential effort and reward values, capturing behavioral extremes of choice acceptance and rejection centered on indifference. This design ensures that choice difficulty, the magnitude of the relative value of effort options, is orthogonal to the difference in value between the options (Shenhav et al., 2014). Effort/reward options were presented consecutively in a pseudorandom order, such that choices of approximately equal choice difficulty were shown in every block of 10 choices. This phase consisted of 40 unique choices. Offers were presented to participants twice: once with the nondefault effort presented first and once with the reward presented first. This allowed for the possibility of analyzing the separate valuation of effort and reward before the time of choice. Offers were presented in a pseudorandom order, such that choices of approximately equal difficulty were shown in every block of 10 choices to mitigate offer order presentation as a confound.
A fatigue phase immediately followed the baseline phase. During this fatigue phase, participants were asked to rate their mental fatigue on an identical continuous scale from “Very Low” to “Very High” with seven tick marks on the number line to act as a reference. Participants then underwent a fatiguing trial, where they performed three 3-back tasks. They rated their mental fatigue again before going on to make 10 choices. This process was repeated a total of eight times, such that identical offers were given to the participant during both phases. Participants were made aware that they had to succeed on 50% of all fatiguing trials to receive the opportunity to earn additional rewards at the end of the experiment during the outcome phase.
Following imaging, the participant choices were realized. If the participant had met the fatigue phase success threshold, two offers from the baseline and fatigue phases were chosen randomly. The participant then completed their chosen option, exerting the effort required and receiving the associated reward. Participants had to achieve the effort task to receive the reward and were given three opportunities to succeed.
Control experiment
To test if changes in effort preferences between the first and second choice phases were the result of a mere exposure effect of the effort options, task order-related effects, or passage of time and not specifically related to fatiguing cognitive exertion, we performed a control experiment in which a new set of participants performed the same experimental phases described above without being exposed to fatiguing exertions. Instead, blocks of fatiguing cognitive exertions were replaced with rest blocks in which participants received a prompt to rest. The duration of each rest period block was randomly drawn from a uniform distribution of 189 (SD = 94) seconds to match the duration of cognitive exertion blocks in the main experiment. A group of 11 healthy participants, separate from those that performed the main experiment, took part in this experiment (mean age, 22 years; standard deviation, 3 years; six females).
MRI protocol
We conducted our MRI scans using a Philips dStream Achieva 3 T TX scanner equipped with a DirectDigital RF system. Participants lay supine on the MR exam table and had their head stabilized within a 32-channel SENSE head coil, whose signal was amplified by an 18 kW solid-state RF power amplifier.
High-resolution structural images were collected using an MPRAGE T1-weighted pulse sequence with an echo time (TE) of 3.57 ms and repetition time (TR) of 8.12 ms. The resulting whole-brain images had a 1 × 1 × 1 mm voxel resolution.
Functional images were collected at an angle of 30° from the anterior commissure–posterior commissure axis, which reduced signal dropout in the orbitofrontal cortex (Deichmann et al., 2003). Functional T2-weighted images included 48 slices in 1.875 × 1.875 × 3 mm resolution. We used an echoplanar imaging (FE EPI) pulse sequence (TR, 2,800 ms; TE, 30 ms; FOV, 240; flip angle, 70°).
Mixed-effect models of behavior
To evaluate participants’ ability to recall the cognitive effort associated with each color level, we employed a linear mixed-effect model as follows:
To evaluate how participants’ fatigue ratings increased with additional fatiguing blocks, we used the following model:
To analyze the aversive and appetitive nature of effort and reward, we evaluated the baseline portion of acceptance for each participant in choice trials across effort and reward levels:
Mixed-effect model of fatigue-induced difference in choice
To analyze the difference in behavior between the baseline and fatigue phases, we conducted a generalized linear mixed-effect model as follows:
Subjective value of effort
Participants’ choices were fitted to structural models using maximum likelihood estimation with a sigmoid function representing the likelihood of accepting the nondefault option (C = 1), where
τ is the stochasticity of the participants’ choice and
Image preprocessing and fMRI analysis
MRI images were preprocessed using the SPM12 software package (Wellcome Trust Centre for Neuroimaging, Institute of Neurology; Penny et al., 2011). Each functional image was registered to the mean of all images, smoothed with a Gaussian kernel of 5 mm FWHM and resliced, and interpolated with a fourth-degree B-spline. Slice timing correction was performed to correct for interleaved slices every seventh slice until the end of all 48 slices, using the first slice as a reference. Images were then coregistered to the MPRAGE anatomical image using a normalized mutual information objective function with a separation of 4 × 2 mm. Images were segmented with SPM12 Tissue Probability Maps. Finally, images were normalized and smoothed with an 8 × 8 × 8 mm FWHM Gaussian kernel. The exact preprocessing pipeline, with all specified settings, is available in the Open Science Framework repository (Steward and Chib, 2024).
A general linear model (GLM) was used to estimate participant-specific (first-level), voxelwise, statistical parametric maps from the fMRI data. The GLM included categorical boxcar regressors beginning at the time of trial presentation and ending when a choice was indicated, for the baseline and fatigue choice phases. Each categorical regressor included orthogonalized parametric modulators of subjective chosen and unchosen value. Each participant's subjective value was calculated by transforming effort values using their estimated parabolic effort discounting model. Trials with missing choice responses were modeled as a separate nuisance regressor. Another categorical regressor was included to model the fatiguing n-back blocks. The n-back condition was modeled as a single block with duration beginning at the time of the n-back onset and ending at the completion of the last n-back trial. This condition included a parametric modulator associated with an n-back block number. Regressors modeling head motion as derived from the affine part of the realignment procedure were included in the model.
With these first-level models, we created group models (second level) to test brain areas that were sensitive to subjective cognitive effort/reward value. This was done by creating contrasts with the parametric modulators for the chosen value at the time of choice. To test for regions of the brain sensitive to subjective value, irrespective of fatigue state, we created a contrast that selected the subjective chosen value parametric modulators for both the baseline and fatigue choice conditions. We also tested for regions of the brain in which the chosen subjective value was sensitive to changes in bodily state induced by fatigue by taking the difference between the chosen value in fatigue and baseline choice phases. In addition, we tested for changes in brain activity that were sensitive to repeated fatiguing cognitive exertion by selecting the parametric modulator for the block number in the fatiguing n-back categorical regressor.
To conduct statistical inference, we extract signal change from independent regions of interest (ROIs) in the brain to conduct statistical inference using the MarsBaR (Brett et al., 2002) toolbox for SPM12 (Penny et al., 2011). ROIs during choice were independently defined by using the conjunction of predetermined functional regions (regions including the terms “vmPFC,” “dACC,” “NAc,” and “aIns” in its labeling) from a task-based fMRI atlas (James et al., 2016) and predictive maps from NeuroQuery (Dockès et al., 2020) associated with the search term “effort choice”. ROIs set for activity associated with increasing n-back were similarly determined using the intersection of regions of the a task-based fMRI atlas (James et al., 2016) labeled as the “dlPFC” (dorsolateral prefrontal cortex) and the predictive map from NeuroQuery (Dockès et al., 2020) for the search term “working memory task.” It is important to note that contrasts were used to illustrate brain activations, and formal statistical inferences were performed using the a priori defined ROIs.
For Psychophysiological Interaction (PPI) analysis, we used a simplified GLM that unified all activity during the time of choice by removing the chosen and unchosen value parametric modulators from the originally described GLM. A physiological seed was defined by a task-based fMRI atlas (James et al., 2016) that included the terms “aIns” in conjunction with “effort choice” predictive map from NeuroQuery (Dockès et al., 2020). Timeseries data were extracted using the GLMs explained above and contrasts associated with the difference between fatigue and baseline choice BOLD signal (fatigue choice > baseline choice). The same ROI used to evaluate activity during the n-back task was also used here to evaluate the activity connected to the rIns during fatigue at the time of choice. The exact analysis pipeline, with all specified settings, is available in the Open Science Framework repository (Steward and Chib, 2024).
Data availability
The datasets generated and analyzed during this study are available in the Open Science Framework repository (Steward and Chib, 2024). Included is an imaging dataset in accordance with the Brain Imaging Data Structure standard (Gorgolewski et al., 2016) and validated with the BIDS-Validator tool (Gorgolewski et al., 2020).
Code availability
The code used to collect and analyze data during this study is available in the Open Science Framework repository (Steward and Chib, 2024).
Results
There are many different cognitive tasks for which repeated engagement induces feelings of fatigue. In this study, we chose to focus on cognitive effort in the form of an n-back working memory task because it allows for clearly operationalized levels of exertion and well-mapped engagement of task-induced neural activity (Westbrook et al., 2013, 2019). During the working memory task, participants attended to a serial presentation of letters and attempted to identify when the letter presented matched the one n letters previously presented (Fig. 1A). Additionally, each effort level was assigned a unique color so that during choice trials, the cognitive effort level could be cued without explicitly presenting the numerical n-back level. This was meant to minimize the association of cognitive effort with the n-back number and instead increase reliance on the participants’ subjective feelings of cognitive effort.
Experimental design. A, Participants first associated different effort levels of n-back task with distinct colors. During the n-back cognitive effort task, participants identified when a letter presented on the screen matched the one presented n letters previously. B, Following this association phase, participants rated the difficulty of each color level on a continuous scale from “Very Low” to “Very High” mental demand. C, During choice trials, participants made decisions between a default effort/reward (one-back for $1) option or an option with varying levels of effort for different reward amounts. Each trial began with the presentation of one piece of information regarding the nondefault option (either the effort level or reward amount), followed by presentation of the second piece of information. Participants were instructed to make their choice after the second piece of information was presented. D, Participants rated their state of mental fatigue on a continuous scale from “Very Low” to “Very High.” This rating was repeated several times during the fatigue phase. E, Experiment schedule. To study the effects of fatigue on effort-based decision–making, blocks of cognitive exertion trials were interspersed with blocks of effort-based choice trials. The experiment was divided into baseline and fatigue choice phases, which were both scanned with fMRI. The baseline choice phase consisted of 80 effort-based choices presented in a pseudorandomized order. Following the baseline choice phase, participants performed the fatigue choice phase of the experiment, in which they underwent repeated exertion trials (indicated in black) to bring them into a fatigued state. The fatigue choice phase was comprised of alternating cognitive exertion and choice blocks. Cognitive exertion trials involved performing a three-back working memory task three times. Each choice block consisted of 10 effort-based choices randomly sampled from the same set used in the baseline choice phase. At the beginning and end of each exertion block, participants rated their cognitive fatigue. Completion of the fatigue choice phase consisted of eight back-to-back exertion and choice blocks.
To develop an association between cognitive effort levels and color cues, we randomly presented the participants with n-back tasks from Levels 1 through 6, with their associated colors (Fig. 1A). To ensure that participants had developed an association between the extent of cognitive effort associated with each color cue, participants rated the mental challenge of each n-back level based on the color cue alone, following the assessment trials (Fig. 1B).
To study how decisions about prospective cognitive effort are influenced by cognitive fatigue, we scanned participants’ brains with fMRI while they made choices to exert different levels of cognitive effort for varying amounts of money, interspersed with bouts of fatiguing cognitive exertion (Fig. 1C). First, during a baseline choice phase, participants made choices to either accept a default option of a one-back task for the reward of $1 or accept a variable higher-level task for a greater reward (nondefault option). These choices were meant to characterize a person's decision-making before fatigue. This baseline phase was immediately followed by a fatigue choice phase, during which participants completed alternating blocks of fatiguing working memory exertions and the same effort-based choices. Participants rated their mental fatigue at the beginning and end of each exertion block (Fig. 1D). All choices were for prospective effort and reward, and at the end of the experiment, two choices were randomly selected from the baseline and fatigue choice phases and played out.
Fatigue-induced changes in effort-based choice
To confirm that participants understood the association between cognitive effort levels and color cues, we examined the relationship between participants’ ratings of mental demand and the n-back–associated color cues. We found that participants’ mental demand ratings increased with cues associated with higher n-back levels (linear mixed-effect model; t(166) = 12.78; p = 1.91 × 10−26; Fig. 2A). This correlation between ratings of mental demand and cognitive effort levels indicates that participants understood the mapping between feelings of cognitive exertion and effort color cues. As a result, participants could make meaningful choices about prospective effort options that utilized these color cues during the choice phases.
Effects of fatigue on choice. A, Participants ratings of working memory task difficulty as a function of the n-back level, during trials polling mental demand. B, Ratings of fatigue during the fatigue phase, as a function of exertion block. Probability of accepting the nondefault effort/reward option as a function of cognitive effort level (C) and reward (D). Error bars indicate SEM. E, Proportion of nondefault option choices accepted, across all participants, as function of the effort level and reward at different time points in the experiment. Participants were more likely to choose options with higher reward and lower effort. The main difference in behavior between the baseline and fatigue phases was at the highest levels of offered reward and effort. F, Contrast of nondefault acceptance proportion between the baseline and the figure phase, where higher values represent offers more likely to be accepted in the fatigue phase than baseline and lower values vice versa. G, To better understand the effect of fatigue on behavior, we used a generalized linear mixed-effect model and found a significant effect of the baseline/fatigue condition on choice. When participants became fatigued, they were more likely to take the default option and forgo greater amounts of reward for increased effort. Error bars indicate SE. ***p < 0.001.
To verify that participants became cognitively fatigued through repeated exertion during the fatigue phase, we examined how their mental fatigue ratings changed as a function of cognitive exertion blocks. We found that as participants progressed through the fatigue blocks, their rating of mental fatigue increased (linear mixed-effect model; t(222) = 6.95; p = 3.94 × 10−11; Fig. 2B). We also examined the relationship between participants’ performance and exertion block. We did not find evidence that participants’ task performance declined with additional fatigue. Instead, there was a significant increase in performance with additional fatigue blocks (linear mixed-effect model; t(222) = 6.95; p = 3.90 × 10−11; Extended Data Fig. 1-1). These results illustrate that repeated cognitive exertion of the working memory task leads to increased feelings of fatigue without resulting in decrements in performance.
Next, we analyzed choice trials across both the baseline and fatigue phases. We found that individuals’ willingness to accept the nondefault effort/reward option decreased as cognitive effort levels increased (linear mixed-effect model; t(138) = −9.18; p = 5.65 × 10−16; Fig. 2C) and increased with additional reward (linear mixed-effect model; t(222) = 8.96; p = 1.36 × 10−16; Fig. 2D). These results align with several previous studies of cognitive and physical effort-based decision–making, which found that increasing effort and decreasing reward have inverse effects on effort-based choice (Meyniel et al., 2013; Klein-Flügge et al., 2015, 2016; Chong et al., 2017; Westbrook et al., 2019).
To evaluate the effect of fatigue on effort-based choice, we analyzed the difference between individuals’ propensity to accept the nondefault effort option in the baseline and fatigue phases. We found that individuals tended to accept fewer nondefault option trials in the fatigue phase compared with baseline and that this propensity appeared to increase later in the fatigue phase (Fig. 2E). As in previous studies of effort-based decision–making (Chong et al., 2017), we used a parameter estimation procedure to fit models of subjective value for each participant and found that a parabolic effort discounting function best described participants’ choice (Extended Data Fig. 2-1; Extended Data Table 1-1). To formally test the effect of fatigue on effort-based choice, we examined the relationship between participants’ propensity to accept the nondefault subjective effort/reward options as a function of effort and reward magnitude and experimental phase (baseline/fatigue). We found a decreased acceptance of the nondefault effort/reward option in the fatigue phase relative to the baseline (generalized linear mixed-effect model; βFatigue = −0.349; SE = 0.097; t(4,434) = −3.60; p = 3.24 × 10−4; Fig. 2F) in a model that performed better than the null model with no effect of phase (LR = 12.773 > χ2(1) = 3.841). When fatigued, participants preferred the low-effort, low-reward default option compared with when in a rested state.
To examine how repeated exertions during the fatigue phase influenced effort-based decision–making, we evaluated how choices for the nondefault effort/reward option varied as a function of the option's subjective value, trial number, and fatigue ratings. Neither the trial number (linear mixed-effect model; t(2,218) = 0.003; p = 0.44) nor fatigue rating (linear mixed-effect model; t(2,218) = −1.63; p = 0.10) significantly influenced choice in the fatigue phase. We likely did not find an effect of the trial number or rating because our paradigm was designed to fatigue participants before choice and hold them at relatively constant fatigue levels during the fatigue choice phase. Our experimental paradigm was optimized to examine gross differences in choice behavior between the baseline and fatigued states and not trial-to-trial functions in the fatigue state.
To exclude the possibility that changes in choice behavior between the choice phases were the result of order-related effects, choice task repetition, or passage of time and not an effect of exertion-induced fatigue, we performed a control experiment in which a separate group of participants performed two phases of effort choices where fatiguing exertion blocks were replaced with blocks of rest (see Materials and Methods, Control experiment). We found that participants’ fatigue ratings did not increase in the control experiment and fatigue ratings were significantly higher in the main experiment than in the control experiment (t(37) = 6.136; p = 5.03 × 10−8; Extended Data Fig. 3-1). We also found that, unlike the main experiment, there were no significant changes in effort choice preferences between the baseline and second phase of choices in the control experiment (generalized linear mixed-effect model; βFatigue = −0.03; SE = 0.17; t(1,726) = −0.19; p = 0.85; Extended Data Fig. 4-1). Together, these results suggest that the fatigue-induced changes in fatigue ratings and choice observed in the main experiment were the results of repeated cognitive exertion and not simply the results of mere exposure to the task or task order effects.
Neural encoding of the cognitive effort value
We found that the BOLD signal in a network of brain regions, including the dorsal ACC (dACC), nucleus accumbens (NAc), vmPFC, and rIns, was modulated by the subjective chosen value of the effort/reward option across both the baseline and fatigue phases (Fig. 3A,B). This finding is consistent with several studies of effort-based decision–making for cognitive and physical effort that implicated these brain regions as part of an effort valuation network (Chong et al., 2017; Arulpragasam et al., 2018; Westbrook et al., 2019).
Neural representations of the chosen value. A, Overall chosen value encoding. Whole-brain activity thresholded at voxelwise p < 0.005 with an extent threshold of 10 voxels. B, Activity in the dACC (ROI analysis, peak, [−8, −2, 54]; t(27) = 3.22; p = 0.007), NAc (ROI analysis, peak, [18, 8, −2]; t(27) = 3.55; p = 0.003), rIns (peak, [36, 26, −12]; t(27) = 3.94; p = 0.001), and vmPFC (ROI analysis, peak, [−8, 42, −2]; t(27) = 3.38; p = 0.004) encoded the chosen subjective value (computed from subjective effort value and reward) across both the baseline and fatigue phases. Error bars indicate SEM. **p < 0.01. C, Whole-brain analysis contrasting chosen subjective value encoding between fatigue and baseline phases thresholded at voxelwise p < 0.005 with extent threshold of 10 voxels. D, Activity in rIns (rIns ROI analysis, peak, [34, 16, −12]; t(27) = −2.14; p = 0.042) reflects the difference in the chosen and unchosen subjective value, between the baseline and fatigue phase. Contrasts were used to illustrate brain activations, and formal statistical inference was performed using a priori defined ROIs. Error bars indicate SEM. *p < 0.05.
To test for brain regions sensitive to changes in chosen subjective value induced by cognitive fatigue, we contrasted the difference between chosen subjective values in the baseline and fatigue phases. We found that a region of the rIns had increased sensitivity to the chosen subjective value during the fatigue phase, compared with the baseline (Fig. 3C,D). This suggests that activity in rIns is sensitive to changes in subjective effort value resulting from cognitive fatigue, which aligns with previous studies of effortful exertion that have suggested that this brain region encodes representations of individuals’ internal state as a function of physical exertion and fatigue (Hogan et al., 2020).
Cognitive fatigue-induced changes in working memory neural activity
We examined how cognitive fatigue influenced brain activity during the fatiguing working memory task. As fatiguing n-back blocks increased, the signal in the bilateral dlPFC also increased (Fig. 4A,B). Activity in this region has been shown to increase with increasing cognitive demand (i.e., increasing n-back level; Braver et al., 1997; Lamichhane et al., 2020). Increasing blocks of cognitive exertion in the context of our fatigue paradigm may increase the neural resources associated with increased cognitive demand.
Neural signatures of cognitive exertion-induced fatigue. A, n-back exertion-induced changes in brain activity. Activity in bilateral dlPFC increased as a function of exertion block, at the time of fatiguing exertions. We estimated a GLM with a parametric modulator for each fatiguing block completed. The resulting contrast showed a significant increase in activation after the first fatigue block in areas previously associated with working memory function (ROI analysis; L dlPFC peak, [−38, 4, 32]; R dlPFC peak, [4, 8, 28]; t(27) = 3.80; p = 7.5 × 10−4). B, Illustration of the effect size in bilateral dlPFC ROI increasing with additional fatiguing blocks. The contrasts were used to illustrate brain activations, and formal statistical inference was performed using a priori defined ROIs. Error bars indicate SEM.
Functional connectivity between rIns and dlPFC
Finally, given our hypothesis that information about one's cognitive state is integral to assessing cognitive capacity and associated effort value, we tested the idea that the neural circuit modulating cognitive effort value representations in rIns might be influenced by computations about cognitive exertion instantiated in dlPFC during choice. To test this hypothesis, we conducted a PPI analysis between rIns (seed) and dlPFC (target) with fatigue state as a psychological variable (Fig. 5A). This analysis revealed a robust modulation of connectivity between the dlPFC and rIns as a function of fatigue state at the time of choice (Fig. 5B). Connectivity was increased in the fatigued state compared with the baseline, and in the fatigued state, activity in the rIns at the time of choice was associated with increased connectivity to the dlPFC. These results provide support for the hypothesis that activity in dlPFC and rIns are functionally related during effort-based decision–making and suggest that interactions between these brain regions could facilitate the transfer of information about cognitive state that is used to inform choices about prospective cognitive effort when cognitively fatigued.
Functional connectivity between rIns and dlPFC. A, Illustration of the PPI analysis. We computed a PPI between rIns and bilateral dlPFC with the psychological variable of baseline/fatigue state, at the time of choice. Activity in bilateral dlPFC shows an increased functional coupling with rIns during the fatigue choice phase, compared with the baseline choice phase (ROI analysis; L dlPFC peak, [−44, 0, 30]; R dlPFC peak, [34, −2, 42]; t(27) = 3.92; p = 5.52 × 10−4). Statistical inference was determined via an ROI analysis of the contrast of the two phases. B, Effect size plots by phase of rIns dlPFC functional coupling. Error bars indicate SEM. ***p < 0.001.
Discussion
In this study, we show that cognitive fatigue decreases an individual's willingness to exert higher cognitive effort, that decisions to exert effort while fatigued are related to value signals in the rIns, and that signals related to cognitive exertions in dlPFC are functionally coupled with signals in rIns at the time of choice. Our neuroimaging findings are consistent with previous studies showing that effort value signals are represented in a network of brain regions, including the rIns, and that activity in rIns follows the time course of feelings associated with exertion and rest (Westbrook et al., 2019; Hogan et al., 2020). However, previous studies primarily focused on physical effort and fatigue and did not examine how cognitive fatigue is related to decisions to exert effort. Our results build on these previous studies by illustrating a mechanism by which signals related to the cognitive fatigue-modulated effort value are shaped by brain regions responsible for generating fatiguing cognitive exertion.
Previous studies of effort cost have mainly focused on the trade-off between prospective physical effort (e.g., grip exertion, button pressing) and reward (Arulpragasam et al., 2018; Hogan et al., 2019, 2020; Müller et al., 2021) and have implicated a network of brain regions, including the vmPFC, ACC, and insular cortex. Studies examining trade-offs between cognitive effort (e.g., working memory tasks, arithmetic) and reward have identified a network that largely overlaps with physical effort, suggesting a general effort valuation network that subserves decisions to exert (Vassena et al., 2014; Chong et al., 2017; Westbrook et al., 2019). Our results are consistent with these findings, which identify regions in the ACC and rIns responsible for effort/reward trade-offs in the context of cognitive effort decisions in both rested and fatigued states. Neuroimaging studies of the influence of fatigue on decisions to exert have focused on the domain of physical effort and shown that regions of the prefrontal cortex, rIns, and motor cortex are sensitive to fatigue state and influence decisions about effort exertion. Here, we show that a region of rIns previously shown to be sensitive to physical chosen effort value when physically fatigued is also sensitive to cognitive chosen effort value when cognitively fatigued (Hogan et al., 2020). Moreover, these regions are functionally coupled with working memory cognitive exertion-related signals in dlPFC at the time of choice, which suggests that information about cognitive state could influence these choices. These results point to a general role of rIns in being sensitive to feelings of fatigue while experiencing both cognitive and physical fatigue.
We found that dlPFC signals related to fatiguing exertion were inversely related to the extent to which participants rated fatigue—those individuals who became more fatigued through repeated exertion exhibited smaller changes in fatigue-induced dlPFC activity. These results are consistent with the idea that cognitive fatigue may be associated with dyshomeostatic representations of cognitive exertion in dlPFC (Hogan et al., 2020). Those individuals who do not tune their dlPFC activity following cognitively fatiguing exertion find prospective effort to be more costly. The cognitive effort might feel particularly costly to these individuals because they continue to recruit the same levels of neural activity as in a baseline state, although they are physiologically unable to recruit higher levels of neural resources to efficiently achieve the target levels of cognitive performance. This idea aligns with recent theoretical accounts of fatigue that suggest discrepancies between perceptions of ability and actual sensorimotor capacity may give rise to feelings of fatigue (Stephan et al., 2016).
Recent studies have shown that neurotransmitters such as gamma-aminobutyric acid (GABA) and glutamate play a key role in judgments of physical effort and are related to cognitively fatiguing exertion (Strasser et al., 2020; Hu et al., 2022). For physical effort, GABA inhibition has been related to changes in descending motor drive and variability in motor output, neural and behavioral signatures that impact feelings of exertion (Hu et al., 2022). However, the neurochemical underpinnings of cognitive effort are less straightforward. Working memory tasks that result in greater cognitive fatigue and require greater cognitive control have been associated with higher glutamate concentration and glutamate/glutamine diffusion in task-related regions (Wiehler et al., 2022). It has been proposed that these neurochemicals could be associated with recycling potentially toxic substances accumulated in the brain during sustained cognitive exertion. The neural system's ability to monitor these state changes could be integral in signaling and updating feelings of fatigue.
A recent behavioral study of cognitive effort-based decision–making found that reductions in cognitive task performance resulting from fatiguing cognitive exertion were associated with momentary increases in feelings of fatigue (Matthews et al., 2023). Those results align with previous work on physical effort, which found that larger motor errors are also associated with increased feelings of effort and result in judgments of physical effort as more costly (Hu et al., 2022; Padmanabhan et al., 2023). Our present study could not dissociate the influence of task performance on feelings of fatigue, as we found that successful performance increased with repeated fatiguing exertions in the context of our experiment. We may not have observed a performance effect on fatigue because our cognitive task may have been easier than that used in the previous study, which showed such a performance effect on feelings of fatigue. More work is needed to distinguish the possibility that cognitive task difficulty could mediate between the effects of performance on judgments of fatigue—as a cognitive task becomes more difficult, poor performance on that task may inflate the subjective cost of effort.
In conclusion, our study provides evidence of how the brain integrates information about cognitive fatigue into effort valuation and decisions to exert. Our data implicate a mechanism by which cognitive exertion-related signals in dlPFC influence effort valuation signals in the rIns, which underlie decisions to exert while in a fatigued state. The present work begins to bridge the gap in understanding how the brain represents cognitive fatigue and the influence of cognitive fatigue on decisions to exert, which to date has received limited study. Moreover, from a clinical standpoint, the neurobiology of cognitive fatigue is likely to prove important in understanding feelings of amotivation that are prevalent across neurological and psychiatric conditions. From this perspective, the behavioral and neural underpinnings of fatigue provide a foundation for the development of interventions aimed at optimizing effortful exertion.
Footnotes
This work was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number R01HD097619 and the National Institutes of Mental Health R01MH119086.
The authors declare no competing financial interests.
This paper contains supplemental material available at: https://doi.org/10.1523/JNEUROSCI.1612-24.2025
- Correspondence should be addressed to Vikram S. Chib at vchib{at}jhu.edu.
