The rat frontal orienting field dynamically encodes value for economic decisions under risk

The study addresses how frontal and parietal cortices causally contribute to economic decisions under risk, a behavior with major implications for public health and survival. Within expected utility theory, risk preferences arise from the curvature of the utility function that maps external rewards to subjective value; concavity implies risk aversion. While humans and nonhuman primates can be instructed about offers, animals must learn cue–outcome contingencies, complicating neural interpretations. Prior work has linked expected utility representations to reward/value areas as well as orienting-related regions (parietal and frontal cortices), but causal evidence is scarce, especially in rodents. Based on distinctions from perceptual decision-making—FOF often exerts strong effects and is argued to be post-decision, whereas PPC encodes momentary evidence—the authors hypothesized that FOF silencing would induce stimulus-independent biases and PPC silencing might cause stimulus-dependent biases. They developed a cued risky-choice task where rats choose between a surebet and a lottery with fixed probability and cue-indicated magnitude, enabling precise, model-based quantification of risk preference and bias.

• Expected utility theory explains risk aversion via utility curvature; concave utility yields preference for sure outcomes.
• Neural correlates of expected utility/value have been observed in human and primate reward systems and in orienting-related regions (parietal and frontal cortices), often in tasks requiring saccades.
• Causal data are limited: in monkeys, silencing supplementary eye field made choices less risk-seeking.
• In rodents, FOF and PPC are implicated in perceptual decisions: FOF perturbations reliably bias decisions, often interpreted as post-decision motor plan memory; PPC encodes sensory evidence, with less consistent causal effects. There are also suggestions of PPC’s role in priors and category re-mapping in auditory tasks. The current work extends these lines to economic decisions under risk, testing causal roles and neural codes in rat FOF and PPC.

Subjects and task: 26 rats (Sprague Dawley and Brown Norway; both sexes) were trained in custom chambers. In the risky choice task, rats fixated at a center port during an auditory cue indicating lottery magnitude, then chose between a lottery (fixed per-animal probability ~0.5–0.75; varying magnitudes across six cues) and a surebet (fixed magnitude per session). Rewards were delivered at a single reward port. Forced trials (surebet or lottery only) ensured exposure to outcomes. Reaction times and violations were recorded; violations excluded from analyses.

Perturbations: Muscimol infusions targeted bilateral/unilateral FOF (AP +2 mm, ML ±1.5 mm) and PPC (AP −3.8 mm, ML ±3.0 mm); doses included 0.075 and 0.3 µg per side (FOF) and 0.3–0.6 µg per side (PPC). In total, 7,456 choice trials from 127 infusion sessions in 8 rats. Optogenetic silencing employed bilateral AAV9-CaMKII-eNpHR3.0-EYFP in FOF with 532 nm laser (15–20 mW), 3-s constant pulses covering entire trials; unilateral and bilateral silencing interleaved (laser on 33% of trials). Controls were no-laser trials within the same sessions for opto; for muscimol, control sessions were typically the day prior to infusion.

Electrophysiology: Movable silicon probes implanted in FOF of six rats performing the task and in four naïve rats passively listening to lottery cues. Spiking was recorded at 30 kHz; units were sorted with Kilosort2 and curated in Phy; quality thresholds applied. Single-neuron analyses used mixed-effects linear models to assess encoding of lottery magnitude (ΔEV) and upcoming choice during late fixation (0.5–1 s post-cue).

Behavioral analyses: Generalized linear mixed-effects models (GLMMs) assessed effects of perturbations on P(choose lottery) as a function of ΔEV = EV_lottery − EV_surebet, and, for unilateral conditions, left/right biases as a function of EV_right − EV_left. Likelihood ratio (LR) tests compared full vs reduced models. Reaction times were analyzed by linear mixed-effects models on log(RT).

Modeling: A hierarchical Bayesian three-agent mixture model captured choices via: (1) a rational expected-utility agent with power-law utility U = V^p and Gaussian decision noise σ; (2) a lottery-biased agent; (3) a surebet-biased agent. Choices are a weighted mixture with mixing weights ω summing to 1. The model estimated subject- and condition-level parameters jointly (control and perturbation) using Hamiltonian Monte Carlo (Stan via rstan/brms). Parameter shifts due to perturbations (Δρ for log p, Δα for log σ, Δω terms) had priors centered at zero to favor no effect; model validation included synthetic data recovery.

Dynamical modeling: A six-node rate network with random all-to-all connectivity represented an action value network receiving inputs proportional to lottery magnitude. The FOF was modeled as one node; silencing it assessed how the population transform from reward magnitude to firing rates (utils) scaled and affected the effective utility exponent p and behavior. Alternative two-node models (hemispheric competition or post-decision inputs) were also tested.

Additional behavioral paradigms: A model-based surebet magnitude shift procedure (re-categorization) altered the decision boundary without changing cue–lottery mapping to test PPC’s role in updating category boundaries; predictions from the three-agent model were compared to observed shifts. Free trials (both ports rewarding equally) were added post hoc to test for classic PPC unilateral inactivation effects on ipsilateral bias as a positive control.

• Baseline behavior: Rats increased lottery choices with increasing EV and showed few violations of dominance; most were risk-averse by psychometric indifference points.
• PPC inactivation effects on risky choice: Bilateral PPC muscimol did not significantly affect risk preference (LR test P≈0.21). Unilateral PPC also showed no significant effect on risk preference (P≈0.16) and no contralateral neglect (P≈0.28), and no reliable RT effect. However, analyses restricted to early trials revealed short-lived effects: within the first ~45–60 trials, PPC silencing increased risk aversion via stimulus-independent bias (e.g., with 45-trial cutoff, P=0.017), supported by three-agent model fits showing shifts in mixing weights.
• FOF silencing robustly reduced lottery choice: Bilateral FOF muscimol decreased lottery choices (P=2.7×10^-4), shifting the mean indifference point from 50.9 ± 11.6 to 154.4 ± 23.5 µl at 0.3 µg (t = −3.95, P < 0.001), equivalent to adding ~100 µl to the surebet. No consistent RT change (P=0.764). Bilateral FOF optogenetic silencing replicated smaller but significant shifts away from lottery (P=3.3×10^-3) without RT effects (β_opto = 0.026 ± 0.018, P=0.068).
• Unilateral FOF inactivation: Muscimol produced a small but significant reduction in lottery choices (P=1.2×10^-3) without reliable RT change (P=0.06). Unilateral optogenetic silencing similarly reduced lottery choices (P=2.4×10^-6). Contralateral neglect effects were weak but detectable: muscimol P=0.010, opto P=0.032.
• Three-agent model inference: Across four FOF silencing experiments, the utility exponent p decreased (Δρ < 0), indicating increased concavity/risk aversion; 3/4 experiments showed significant decreases by 97.5% credible interval. Bilateral muscimol showed bimodal posterior (either Δp<0 or shift in ω increasing surebet agent). There was also a tendency for increased surebet mixing weight (Δω_surebet>0), significant in unilateral silencing. Control behavior was dominated by the rational agent (ω_rational ~0.84 CI 0.79–0.88), with σ ~0.05 and decelerating utilities in most animals (p<1).
• Dynamical model: Silencing the FOF node scaled down network encoding of lottery value (e.g., for largest lottery, mean firing ~45 Hz to ~22 Hz), changing the effective power-law transform from p≈0.76 (control) to p≈0.60 (FOF silenced) and producing stimulus-dependent rightward shifts in psychometric curves consistent with data. Alternative hemispheric competition and post-decision models failed to reproduce bilateral shifts and predicted primarily noise increases or large unilateral biases.
• Neural coding in FOF: During fixation, FOF neurons encoded lottery value and upcoming choice. Of 1,690 neurons, 423 (25.0%) correlated with ΔEV (63.6% positively) controlling for choice; 702 (41.5%) predicted upcoming choice controlling for ΔEV; 309 (18.3%) encoded both. Passive listening in naïve rats showed no above-chance tuning to the cues (6/105 cells, χ²=0.051, p=0.82). Pseudopopulation decoding of lottery magnitude from FOF activity was above chance with as few as 32 neurons and reached r>0.8 with >200 neurons; large lotteries were compressed (consistent with concave utility).
• PPC and re-categorization/prior: When the surebet magnitude was changed, behavior shifted toward the new boundary; predicted vs actual shift correlated strongly (r=0.905, P=1.6×10^-11). Bilateral PPC inactivation did not impair updating; instead, it reduced contraction bias (actual shifts were closer to predicted; β_predicted shift×PPC = 0.251 ± 0.091, P=0.011).
• PPC free trials control: Unilateral PPC muscimol induced a significant ipsilateral bias on free trials (52 ± 16%, t(5)=3.09, P=0.027) without affecting performance or inducing bias on interleaved choice trials (15 ± 8%, P=0.11), confirming effective inactivation.

Findings demonstrate a causal and computational role for rat FOF in encoding the value of actions for economic choices under risk. Contrary to a strictly post-decision interpretation, FOF silencing produced stimulus-dependent shifts consistent with changes in utility curvature, and FOF neurons encoded both lottery value and impending choice during planning. A simple dynamical network model in which FOF is a node in a distributed action value circuit explained the behavioral and neural effects via a scaling of the value transform (lower effective p), leading to reduced lottery preference.

PPC, despite correlational associations with decision variables in other species and tasks, was not necessary for ongoing risky choices in this paradigm; its effects were minimal or short-lived and best explained by transient biases rather than changes in utility computation. PPC did not impair re-categorization of stimuli when the decision boundary shifted; instead, silencing reduced contraction bias, suggesting PPC may contribute to representing priors or expectations about surebet value. Classic ipsilateral biases on free trials confirmed PPC inactivation efficacy.

Together, results support a division in the frontal–parietal network: FOF contributes directly to representing expected utility of options for action selection, while PPC may modulate biases and priors depending on task demands and timescale. The work bridges rodent models to human/primate neuroeconomics by isolating utility curvature as a key determinant of risk aversion under expected uncertainty and identifying its cortical substrate.

This work establishes the rat frontal orienting field as a critical node for dynamically representing action values during risky economic decisions. Causal manipulations show that FOF silencing increases risk aversion by reducing the utility exponent p, and electrophysiology reveals FOF neurons encode lottery value and upcoming choice. A dynamical network model parsimoniously links neural scaling to behavioral utility curvature. PPC is not required for ongoing choice under risk but may encode priors, influencing contraction bias.

Future directions include simultaneous recordings and perturbations across FOF, PPC, and downstream structures (e.g., superior colliculus) to dissect network interactions and timescales; testing modality-specific PPC contributions; and dissociating attention from utility valuation. Extending these paradigms to varying forms of uncertainty will help separate utility curvature effects from learning and history-dependent strategies.

• PPC effects on risky choice were short-lived and identified via analyses restricted to early trials; the definition of “early” was explored post hoc, warranting caution.
• Unilateral FOF muscimol experiments were not fully counterbalanced (7/8 rats had lottery on the right), potentially confounding lateralized bias estimates.
• Bilateral FOF muscimol had the least data and yielded bimodal posterior interpretations (Δp vs ω shifts), limiting certainty.
• Some slowing in RT under bilateral FOF muscimol could reflect spillover to adjacent M1.
• PPC recordings/optogenetics were not pursued initially due to nominal muscimol effects; later discovery of short-lived effects suggests further targeted studies are needed.
• Modality specificity for PPC remains unresolved; results may depend on auditory vs visual stimuli.