Cross-Frequency Coupling as a Neural Substrate for Prediction Error Evaluation

Ruffini, Lopez-Solà, Palma, Sanchez-Todo, Vohryzek, Castaldo, Friston — Preprint (bioRxiv), 2025.

Why I cared. Predictive coding says the brain is forever comparing predictions against inputs — but where, physically, does that comparison happen? I wanted a concrete neural substrate, not a metaphor.

What we did. We showed that our Laminar Neural Mass Model supports two kinds of cross-frequency coupling — slow rhythms modulating fast amplitudes (signal–envelope), and slow envelopes modulating fast envelopes (envelope–envelope) — and argued these build a hierarchical “Comparator” inside a cortical column.

What we found. Signal–envelope coupling can generate fast prediction-error signals, subtracting top-down predictions from bottom-up envelopes, while envelope–envelope coupling does slower gating — precision-weighting and routing. Perturbing it reproduces disease signatures: in Alzheimer’s, losing fast inhibition makes the Comparator over-amplify prediction errors early and flatten them late.

What it opened. If cross-frequency coupling is the comparator, can its signatures be read as a direct readout of prediction-error machinery in a living brain — and tracked as disease, or a drug, bends it?