Fig. 6.

Nutrient-sensitive reinforcement learning, economic choice models, and candidate neuronal mechanisms. (A) A nutrient-sensitive reinforcement learning model updates values of choice options based on the nutrient content of reward outcomes. Depending on an agent’s nutrient–value function, high-nutrient rewards ($R^H>1$) elicit stronger value-updating than low-nutrient rewards ($R^L=1$). By comparison, a standard reinforcement model updates value irrespective of nutrient content $(R^H=R^L=1$). (B) Nutrient-sensitivity parameter η increments value for high-nutrient options (e.g., for η = 0.2, value increases from 1.0 to 1.25). (C) In a simulated reversal-learning task with high- and low-nutrient rewards ($\alpha =0.2$, $\beta =5$,$R^H=1.25R^L$,$N=100$ trials), a nutrient-sensitive model learns faster to choose the high-nutrient option when it is associated with higher reward probability (P(H) = 0.6); following probability reversal (P(H) = 0.4), the nutrient-sensitive model switches more slowly to choosing low-nutrient reward than the standard model. (Inset) This learning pattern results in similar total reward outcomes (R) as the standard model but achieves higher nutrient intake (N). (D) Implications for economic choice theory. (Top) Indifference maps between combinations of rewards A and B describe an individual’s preferences between two reward options. (Middle) If choices depend on nutrient preferences, establishing indifference maps in nutrient space could predict preferences for untested rewards (C and D) of known nutrient composition (Bottom). (E–I) Neuronal hypotheses. (E) Nutrient-sensitive attractor-based decision circuit. Decision-making is implemented by competition through mutual inhibition between choice-coding neurons, biased by value inputs. Nutrient-sensitivity is implemented by fat, sugar, and texture inputs onto value neurons. (F) Neurons sensitive to dietary fat may receive separate viscosity and friction inputs. (G) Parallel circuits implementing decision-making based on sugar and fat content. (H) Decision circuit for nutrient prioritization. (I) Energy-tracking neurons integrate information across macronutrients.