Mouse dmPFC Encodes Motivation Along Three Orthogonal Axes: Value, Valence, and Salience
> Nature 2026 | Nanci Winke lab > > dmPFC is not an undifferentiated soup of "excitation" and "inhibition." It spreads value, valence, and salience across three mutually orthogonal coordinate sheets.
The Problem: Three Entangled Variables
Animals constantly judge whether to approach or avoid a cue. Behind this decision lie at least three dimensions:
- Valence: good or bad? Reward or punishment?
- Salience: how attention-grabbing is the event?
- Value: quantitatively, how rewarding or how painful?
- Sometimes shuttling earned water reward (CSw)
- Sometimes it avoided shock (CSs)
- A third cue (CSr) signaled larger reward or weaker punishment
- Freely moving + miniscope calcium imaging: mice shuttled spontaneously while dmPFC calcium signals were recorded head-mounted. Electrophysiology follow-ups confirmed the sustained representations seen in calcium signals reflect genuine neuronal activity, not indicator kinetics artifacts.
- Linear classifier decoding: if dmPFC coded salience, equally motivated reward and punishment should show similar patterns; if valence, orthogonal patterns; if value, intensity-matched patterns. Results ruled out pure-salience and pure-valence models, supporting value-dominant coding with orthogonal valence/salience subspaces.
- Orthogonal subspace constraints across motivational dimensions
- Decoupled value assessment: separating "how important" from "how good"
- Flexible recomposition via interpolation in an orthogonal basis
- Winke N, et al. (2026). *Prefrontal neural geometry of learned cues guides motivated behaviours*. Nature, 651(8104):164-173. DOI: 10.1038/s41586-025-09902-2
- Winke N, et al. (2023). Preprint on bioRxiv: 10.1101/2023.03.03.530871
- Jercog D, et al. (2021). Sustained threat representations in dmPFC. *(Related prior work)*
- Laubach M. (2018). dmPFC anatomy definitions in mice.
- Tye KM. (2018). Neural circuit motifs in valence processing. *Nature Neuroscience*.
- Dalley JW, et al. (2004). Prefrontal cortical control of executive functions.
In nature these are almost always confounded—a piece of meat for a starved rat is positive in valence, high in salience, and high in value. The Winke team's trick: keep the physical action identical and vary only the consequence, so neural coding of outcome could be isolated.
The 7-Second Shuttle Task: Action Locked, Consequences Split
Mice in a two-chamber box had to shuttle from room A to B within 7 seconds of a light cue:
Crucially, the motor action was identical in all cases. When CSs and CSw were tuned to equal motivational strength (same shuttling probability), neither salience nor value could explain differential neural responses—leaving valence as the only explanation for differing activity patterns.
Findings: Orthogonal Axes in dmPFC
Using calcium imaging plus machine-learning decoding, the team found:
1. Population level encodes Value — dmPFC ensembles most accurately decode quantified reward/punishment magnitude, not a vague good/bad judgment. 2. Subpopulations encode Valence and Salience separately — one ensemble codes positive vs. negative outcomes (appetitive vs. aversive), another codes salience, and these coding spaces are approximately orthogonal. 3. Orthogonality enables recomposition — a single dimension can be adjusted independently. A mouse can learn "this tone meant water before, now it means shock"—a valence flip with salience unchanged.
Technical Notes
Implications for AI: Orthogonal, Disentangled Representations
The finding offers a biological template for a core ML problem—entangled representations where reward, risk, and curiosity signals interfere in one vector space. dmPFC suggests:
Work like β-VAE and FactorVAE pursues disentanglement, but dmPFC provides biological validation that brains use orthogonal geometry for motivation—and that evolution preserved this scheme.
Core Conclusions
1. dmPFC populations encode value (quantified appetitive/aversive magnitude), not diffuse "emotion." 2. Subpopulations orthogonally separate valence and salience. 3. Orthogonal structure allows flexible recomposition of motivational judgments. 4. The 7-second shuttle paradigm is a methodological breakthrough: behavior locked, motivation disentangled, decoded via calcium imaging + machine learning.