Neural mechanisms underlying the effects of physical fatigue on effort-based choice

The study addresses how physical fatigue alters the subjective valuation of effort and the neural mechanisms underlying effort-based decision-making. Prior work has identified ACC, bilateral anterior insula, and vmPFC in computing effort value in rested states, but has not examined how fatigue-induced bodily state changes influence prospective effort valuation and choices. The authors hypothesize that repeated exertion-induced fatigue increases the subjective cost (marginal disutility) of effort, manifesting as greater risk aversion for effort. Neurally, they predict that ACC and insula encode effort value and that insula, sensitive to interoceptive/proprioceptive states, will reflect fatigue-induced changes in effort valuation. They further propose that motor cortical state, particularly premotor cortex, provides information about bodily and motor capacity that modulates effort value computations during fatigue.

Prior neuroimaging work links ACC, anterior insula, and vmPFC to effort valuation and effort-based decisions, primarily in rested conditions. Studies of physical fatigue show decreased activity and excitability in motor and somatosensory cortices after fatiguing exertion (fMRI and TMS), suggesting reduced capacity for motor pathway recruitment. Theoretical accounts propose fatigue arises from mismatches between expected action consequences and actual sensorimotor outputs, implicating networks processing proprioceptive, exteroceptive, and interoceptive signals (including posterior and anterior insula). Meyniel et al. showed posterior insula encodes proprioceptive-like signals during exertion/rest cycles, while anterior insula encodes effort value, hinting at a role for bodily state in valuation, though prior designs did not examine prospective valuation under manipulated fatigue. Other work assessed impacts of physical or cognitive fatigue on cognitive control but not effort valuation. Motor control studies explored internal models of effort and decision-making but did not test fatigue effects or associated neural encoding. Together, literature points to valuation circuits and motor regions relevant to fatigue, but leaves a gap on how experimentally induced fatigue modulates prospective effort value and neural signals.

Participants: 30 healthy right-handed adults were recruited; after predefined behavioral exclusion criteria, 20 participants (mean age 24, 9 females) were included in the main fMRI analyses. All provided informed consent under IRB approval. Apparatus: fMRI at 3T (Philips Achieva); responses via left-hand button box; right-hand grip force with MRI-compatible dynamometer. Visual stimuli via projector/mirror. In Control Experiment 2, surface EMG recorded from right flexor digitorum superficialis. Task overview: Participants first completed association and recall phases linking numeric effort levels (0–100 scale; 100 = 80% MVC) to grip force. Maximum voluntary contraction (MVC) was measured at outset. Main fMRI experiment design: Two decision phases were scanned: (1) Baseline choice phase (rested): 170 trials of effort-based decisions. Each trial presented a choice between a Sure option (certainty of a smaller effort S) and a risky Flip option (50% chance of a larger effort G or 0 effort). Choices were prospective; gambles were not resolved during scanning. (2) Fatigue choice phase: Participants first performed an exertion block of repeated 4 s grip trials at target 80 units (success if within 80±5 for ≥2.67 s). Trials repeated until ≥75% within-block trials were failures (minimum 10 trials). Then participants alternated between choice blocks (10 effort choices sampled from baseline set) and exertion blocks (minimum 5 trials), for 17 alternating blocks, maintaining fatigue. Behavioral modeling: Subjective effort cost V(x) modeled as a power function representing costs as losses, with sensitivity parameter p (higher p = greater marginal cost; p>1 indicates risk aversion for effort). Choice stochasticity captured by softmax temperature r. Parameters (p, r) estimated separately for baseline and fatigue phases using maximum likelihood across 170 choices per phase. Model comparisons (including alternative utility forms) favored the power function V(x)=(-x)^p across phases. A model-free choice similarity metric compared acceptance of identical gambles across phases. Exertion performance quantified via mean exertion (effort units) and an exponential decay rate parameter over first 10 trials to index fatigue. fMRI acquisition and analysis: EPI parameters TR=2800 ms, TE=30 ms, 48 slices, 3x3x2 mm voxels; preprocessing in SPM12 (slice timing, motion correction, normalization, 8 mm FWHM smoothing). GLM regressors modeled trial onsets (from presentation to response) separately for baseline/fatigue and chosen/unchosen options with unorthogonalized parametric modulators for values of chosen and unchosen options; missed trials and exertion blocks included as nuisance regressors; motion parameters included. Primary contrasts: (i) chosen minus unchosen effort value across both phases; (ii) [fatigue − baseline] of chosen minus unchosen; (iii) main effect of [fatigue − baseline] at choice (motor state). Small-volume corrected ROI analyses focused on bilateral anterior insula and dorsal ACC for value; premotor cortex for fatigue-related motor signals, using a priori coordinates. Additional model included log RT to confirm effects were not driven by choice difficulty. Control Experiment 1 (behavioral): Separate N=9 performed two choice phases (baseline and a second phase after rest) without exertion blocks to test for mere exposure effects. Control Experiment 2 (behavioral+physiology): Separate N=17 alternated choices with exertion at two levels: low (10U) and high (60U), then rest (2 min), then low (10U) again. Success criterion: complete 5 successful exertions per block; experiment terminated if >20 failed exertions total. Self-reported fatigue ratings collected at start and end of exertion blocks. sEMG recorded; mean frequency of power spectrum used as a physiological fatigue index. Hierarchical Bayesian model estimated section-specific subjective effort parameters (p, r) across first low, high, and second low effort sections. Moderation analysis: Between-participant regression tested whether changes in premotor (PM) BOLD activity between phases moderated the relationship between exertion decay (performance) and changes in subjective effort parameter (Δp). Interaction terms were constructed after z-scoring predictors.

Behavioral fatigue manipulation and performance: - Exertion performance declined with fatigue: mean repetitions to reach failure threshold decreased between first and second exertion blocks by 9.26 trials (paired t18=4.69, p=0.0002). Mean exertion within the first exertion block declined by 10.14 effort units from start to end (paired t19=3.36, p=0.003), and remained reduced across subsequent blocks. Subjective valuation changes with fatigue: - Baseline rested state: most participants had p>1 (mean p_baseline=1.39, SD=0.56; t17=3.17 vs 1, p=0.005), indicating risk aversion for effort. - Fatigue increased marginal cost of effort: Δp = p_fatigue − p_baseline = 0.49 (SD=0.86), paired t19=2.54, p=0.02; effect remained significant when holding r constant (paired t=3.22, p=0.005). Model-free analysis showed decreased acceptance of risky effort under fatigue. - Individual variability: 5 participants showed decreased p under fatigue (greater risk tolerance to avoid certain effort), but group trend was increased risk aversion for effort. - Choice stochasticity did not change significantly (Δr = −0.08, SD=0.25; t19=−1.42, p=0.17). - Greater exertion decay predicted larger increases in p (standardized robust regression coefficient=0.87, p=0.0001), linking motor performance decline to valuation changes. Neural encoding of effort value: - Across phases, chosen-minus-unchosen effort value correlated with BOLD in bilateral anterior insula (peaks ~[34,20,2] and [−32,24,2], SVC p<0.05) and dorsal ACC (peak [−8,24,42]). - Fatigue-specific increase in value encoding observed in right anterior insula (rIns; peak [34,18,2], SVC p<0.05); rIns was insensitive to chosen/unchosen values at baseline. rIns effects did not correlate with individual changes in risk preferences (r=0.03, p=0.91), supporting an effort value interpretation rather than risk per se. Motor cortical state under fatigue and its relation to valuation: - At choice time, fatigue vs baseline showed reduced activity in left primary motor cortex and premotor cortex (PM) (e.g., peaks near [−36,−18,−54]; SVC p<0.05), consistent with central fatigue effects. - Individuals with larger increases in p showed less PM deactivation between baseline and fatigue (PM peak [−32,−14,−58], SVC p<0.05), suggesting that reduced adjustment (miscalibration) of PM state relates to inflated subjective effort costs under fatigue. - Moderation: PM deactivation moderated the relationship between exertion performance decay and Δp; less PM deactivation strengthened the link between motor performance decline and increased subjective effort cost. No significant direct correlation between PM deactivation and performance decay (r=−0.20, p=0.40), precluding mediation. Control experiments: - Control 1 (no exertion): No significant change in choice behavior across two phases, indicating observed effects are not due to mere exposure. - Control 2 (varying exertion with self-report and EMG): Self-reported fatigue was low during initial low-effort section (mean 1.81; t167=7.48 vs 3, p<0.001), increased during high-effort section (mean within-block increase 1.14; t166=6.58, p<0.001), and remained elevated after rest and during subsequent low-effort section (t16=5.69, p<0.001). EMG mean frequency decreased between first and second low-effort sections by −4.39 Hz (paired t16=−2.25, p=0.04), consistent with muscle fatigue. Hierarchical Bayesian estimates showed increased p during high-effort vs first low-effort section (95% HDI for p_60U−p_10U,1 > 0) and persistence into the second low-effort section (95% HDI for p_10U,2−p_10U,1 > 0). A model-free comparison across groups equating number of exertions found that high-effort (80U) group became more risk-averse for effort whereas low-effort (10U) group did not (between-group unpaired t34=2.60, p=0.013; main group one-sample t18=2.13, p=0.047; control group unpaired t16=1.61, p=0.126).

Findings show that physical fatigue increases the subjective marginal cost of effort, leading to more risk-averse effort choices. Neural data reveal that bilateral anterior insula and ACC encode effort decision values, with rIns specifically amplifying chosen effort value encoding under fatigue. This supports the role of interoceptive/proprioceptive representations in modulating valuation when bodily state is altered. Fatigue also reduces activity in motor/premotor cortices at decision time, indicating central fatigue signals present even during prospective choices. Critically, individuals exhibiting greater increases in subjective effort cost showed less fatigue-induced PM deactivation, suggesting that miscalibration of motor cortical state (failure to downregulate PM activity in line with reduced motor capacity) contributes to inflated effort valuation under fatigue. Moderation analyses indicate PM state shapes how declines in motor performance translate into changes in subjective effort valuation. Control experiments confirm that effects are specific to high-intensity fatigue, accompanied by subjective and physiological signatures (EMG), and not merely repeated exposure or low-level exertion. Together, results provide a neurobiological mechanism linking motor cortical state to insula-based effort valuation to guide decisions under fatigue.

The study integrates computational modeling, behavioral measures, and fMRI to demonstrate that fatigue inflates the subjective cost of effort, shifts risk attitudes for effort, and modifies value signals in right anterior insula. It identifies premotor cortex state as a key modulator of how fatigue affects effort valuation, consistent with accounts where dyshomeostatic sensorimotor representations bias perceived effort costs. Contributions include: (1) direct experimental manipulation of fatigue to reveal changes in prospective effort valuation; (2) identification of rIns value encoding specific to fatigue; (3) linkage between PM deactivation and behavioral valuation changes. Future work should combine central neural measures with peripheral muscle/motor unit recordings to dissociate central vs peripheral fatigue contributions, employ designs that directly probe interoceptive and proprioceptive sensations during choice, and test generalizability with whole-brain statistical corrections and larger samples. Investigating broader effort contexts (e.g., varying reward, cognitive effort) and interventions to recalibrate motor cortical state may inform strategies to mitigate fatigue-related decision biases.

- Main fMRI experiment used a single high effort level (80U) for exertion and did not collect subjective fatigue ratings or EMG, limiting direct linkage between central signals and perceived/physiological fatigue in that dataset (addressed in Control Experiment 2 but not concurrently with fMRI). - Imaging inferences relied on small-volume corrected ROI analyses rather than whole-brain family-wise corrections; results require confirmation with whole-brain corrections and independent datasets. - Sample size after exclusions was modest (N=20), and individual variability in responses was present. - Design did not directly measure interoceptive/proprioceptive sensations during choice, so mechanisms by which bodily signals modulate valuation are inferred rather than directly tested. - The study focuses on central mechanisms; peripheral fatigue contributions were not measured in the main imaging experiment, limiting conclusions about their interplay. - Moderation analysis is correlational and assumes causal ordering based on task timing; experimental manipulations of PM state (e.g., TMS) would be needed to establish causality.