THEORY-097: Long-range coordination in the human forebrain is carried by brief ripples that co-occur at distant sites: during them neurons co-fire more whatever the distance between them, and at retrieval they carry stimulus-specific reinstatement of encoding co-firing

Source

• Verzhbinsky, Daume, Cheng, Rutishauser & Halgren (2026), LIT-633, read in full in NOTE-494: Figs. 1–6, Extended Data Figs. 3, 9 and 10, the Methods on ripple, co-ripple and co-firing definitions, and the Discussion and Limitations.

What was actually shown

The data. Verzhbinsky et al. (LIT-633) reanalyse an open dataset: 35 patients with refractory epilepsy, 43 sessions, Behnke–Fried microwires placed for clinical reasons in hippocampus, amygdala, vmPFC, ACC and preSMA of both hemispheres, during a Sternberg working-memory task with load 1 or load 3. A ripple is a 70–100 Hz burst of at least three cycles; a co-ripple is ripples on two bundles overlapping by at least 25 ms; co-firing is spikes from two units within 25 ms.

Distance does not matter. Ripples occur at about 0.5 per second in every region, last about 70 ms and peak near 91 Hz. Between bundles they co-occur at a median 5% (13% within a bundle), and the rate barely changes with population-average tract length from 71 to 223 mm, ipsilateral or contralateral. Across 31,489 unit pairs, co-firing is a median 34% higher during co-ripples than when neither site ripples; the abstract’s “~30%” is the task-stage figure. The gain is about as large for contralateral cortical pairs (44%) as for ipsilateral ones (42%), and rises slightly, not falls, with distance (r = 0.04).

It is not just more spikes. Against an independent-firing null, co-firing is 0.059 Hz in co-ripples against 0.038 Hz outside them, and the rate-corrected spike time tiling coefficient is 117% higher (Extended Data Fig. 3). This is the result that could most easily have come out the other way: if ripples only raised excitability, the rate-corrected measures would not move.

It tracks load. Load 3 raises co-ripple co-firing over load 1 by 13% in maintenance and 19% at retrieval, against about zero outside ripples. Ripple-band co-occurrence is load-modulated in 12 of 15 region pairs at retrieval, low gamma in 2 and non-oscillatory very high gamma in none.

It carries content. A cell pair that co-fired when a stimulus was encoded co-fires again when that stimulus is retrieved in 0.29% of cell-pair trials during co-ripples, against 0.14% in duration-matched no-ripple periods, and in 0.36% against 0.23% for fast and slow load 3 responses (Fig. 6). A stimulus-label shuffle inside co-ripple periods controls for excitability.

What this does not say

• It does not say ripples cause the coordination. Nothing was perturbed. The authors’ own Methods say ripple rates explain only about 5% of firing-rate variance in maintenance and retrieval, and that ripples “modulate but do not completely determine” firing. “Carried by” in the title is the paper’s reading of a correlation, and is what the promote_when asks to be tested.
• It does not say ripples are a general cognitive mechanism. The evidence is one task with two load levels, and the authors themselves ask for other paradigms and a wider range of difficulty before calling it general.
• It does not say most of the time is co-rippling. Co-occurrence is a few percent; co-ripple time per recording is about 45 s. The reinstatement effect is a fraction of a percent of cell-pair trials, significant because the counts are very large.
• It does not cover the whole forebrain. Five regions, no posterior parietal or sensory cortex, in patients with epilepsy, and distances are atlas averages, not measured per patient.
• It does not establish the coupled-oscillator mechanism. That distant sites lock at zero lag despite conduction delays because many pathways join them is argued in the Discussion from cited models and not tested.

Connections

• Integration and the workspace. The authors say that the “general lack of evidence for integration of neuronal firing over wide expanses of association cortex” has challenged workspace-type theories, and offer this as evidence for “an essential component of that process”. That is a claim about cognitive integration in working memory, not about experience.
• THEORY-063. That account takes phenomenal unity to be the integration of one world-model into a single functional cluster, with the neural level borrowed from the dynamic-core hypothesis. Co-ripples are one measured mechanism by which distant neurons could form such a cluster, but nothing here measures whether they go with experience being unified, and a cluster of maximal causal density is not a 5% co-occurrence of brief bursts. No relation is declared: this account neither extends nor rivals that one, and gives it no support.
• LIT-385. Damasio’s convergence-zone proposal, that recall is the time-locked re-enactment across regions of activity laid down at encoding, has the shape of the reinstatement result. The paper does not cite it and the record has not read it, so it is an analogy only.
• Propagation against synchrony. Xu et al.’s spirals (LIT-638, THEORY-088) and Muller et al.’s traveling-wave review (LIT-647, unread) put large-scale coordination in phase that moves across cortex. This account puts it in co-occurrence with no measured direction of travel, at a timescale about three orders faster. No paper in the record sets the two against each other.