Contribution
Earlier work showed that human cortical ripples co-occur across large distances in the LFP, and that they raise co-firing between nearby neurons. This paper puts single units and LFP from five bilateral forebrain regions together during a working-memory task. It shows that the co-firing boost reaches distant and cross-hemisphere pairs with no loss over distance, that it scales with memory load, and that co-ripples carry stimulus-specific co-firing from encoding into retrieval.
Key insight
Long-range coordination need not be a signal sent along a path. If distant sites tend to burst at about 90 Hz at the same moments, those moments become shared windows in which spikes at both sites are more likely to coincide. The windows open more when the task is harder, and what coincides in them at retrieval is more often the same pairing that coincided when the item was seen. Because the effect does not fall off with distance, the authors read it as a property of the network as a whole rather than of point-to-point transmission.
Assumptions
- Dataset: the open Daume et al. dataset (DANDI 000673), 35 of its patients (one excluded for lack of clean LFP), 43 sessions, Behnke–Fried microwires in vmPFC, ACC, preSMA, amygdala and hippocampus, bilaterally, placed for clinical reasons in patients with refractory epilepsy.
- Task: modified Sternberg, 140 trials per session, load 1 or load 3 images, 2.5–2.8 s maintenance, then a probe. Mean accuracy 93%, so error trials were too few to analyse.
- Ripple definition (Methods): 70–100 Hz band, at least three cycles, peak z above 2.5, events merged within 25 ms; rejected near large deflections and epileptiform activity; channel averages and co-ripples visually reviewed. Spike waveforms were subtracted from the raw signal before LFP processing.
- Co-ripple: at least 25 ms overlap between ripples on two bundles. Co-firing: spikes of two units within 25 ms. Co-firing analyses keep pairs with baseline co-firing of at least 0.025 Hz (12,335 of 16,274).
- Distance is population-average tractography streamline length between atlas parcels assigned to each region, not measured per patient.
- Statistics: permutation tests of medians, linear mixed effects with patient as random effect (Rate ∼ Load + (1|Patient)), FDR correction, χ² tests for repetition rates.
Key results
- Ripples everywhere. Per-region density about 0.51–0.55 per second, duration about 69–75 ms, frequency about 91 Hz (Fig. 1f–i). Ripple rates rise over baseline in all regions in every task phase, by 13% in encoding and 10% in maintenance on average, and by 36% in preSMA at retrieval (Fig. 2a).
- No decline with distance. Co-occurrence is a median 13% within a bundle and 5% (IQR 4–6%) between bundles, and cross-hemisphere rates differ from within-hemisphere ones by 0.1 percentage points (Fig. 1j). The co-ripple co-firing gain rises slightly with distance (r = 0.04).
- Co-firing. Across 31,489 pairs, co-firing is a median 34% higher in co-ripples than in no-ripple periods; median gains by pathway are 49% (amygdala–cortex), 21% (hippocampus–cortex), 43% (amygdala–hippocampus), 42% (cortex ipsilateral) and 44% (cortex contralateral) (Fig. 1k).
- Not rate alone. Co-firing above an independent-firing null (rate_A × rate_B × 2Δt) is 0.059 Hz in co-ripples against 0.038 Hz outside them; STTC is 0.023 against 0.011, 117% higher (n = 26,005 pairs; Extended Data Fig. 3).
- Task stages. Against baseline, co-ripple co-firing rises 28%, 29% and 24% in encoding, maintenance and retrieval; no-ripple co-firing changes by +1%, −3% and −4% (Fig. 4).
- Load. Co-ripple co-firing is 13% (maintenance) and 19% (retrieval) higher at load 3; no-ripple co-firing changes by −1% and −0.2% (Fig. 5). Ripple-band co-occurrence shows load modulation in 12 of 15 pairs at retrieval, low gamma in 2 and very high gamma in 0. Amplitude envelope correlations show no robust load effect in any band (Extended Data Fig. 9).
- Speed. Hippocampal and amygdala ripple rates after the probe are higher on fast than slow load 3 trials, and hippocampal ripples lead amygdala ripples (379 versus 473 ms).
- Reinstatement. Encoding-to-retrieval repetition of a pair’s stimulus-specific co-firing: 0.29% of cell-pair trials in co-ripples versus 0.14% in no-ripple periods (χ² = 213.9); 0.36% versus 0.23% for fast versus slow load 3 match trials (χ² = 57.5). Matching co-firing clusters around co-ripple centres more than around single-site ripples or shuffled spikes (Fig. 6c,d). A stimulus-label shuffle within co-ripple periods is the control for excitability (Methods).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Ripple co-occurrence between human forebrain sites does not decline with distance beyond the local bundle, up to about 220 mm and across hemispheres | strong for these five regions | Fig. 1j; tract lengths are atlas averages |
| C2 | Co-ripples raise cross-region co-firing beyond what the concurrent rise in firing rate predicts | strong | Extended Data Fig. 3 (two rate-corrected measures) |
| C3 | Co-ripple co-firing scales with working-memory load, more than co-firing outside ripples or co-occurrence in other high-frequency bands | moderate: one task, two load levels | Figs. 3, 5 |
| C4 | Co-ripples hold more stimulus-specific reinstatement of encoding co-firing at retrieval, more on fast trials | moderate: rates are small (fractions of a percent), effects significant by χ² on very large counts | Fig. 6, Extended Data Fig. 10 |
| C5 | Co-ripples are a general mechanism for integrating distributed representations in human cognition | weak: an interpretation | Discussion; one task, clinically placed electrodes, no causal manipulation |
| C6 | Distance invariance indicates an emergent network of coupled oscillators rather than directed transmission | weak: argued from cited models | Discussion |
Concepts
- ripple: a 70–100 Hz oscillatory burst of at least three cycles, detected per microwire (Methods).
- co-ripple: ripples on two bundles overlapping by at least 25 ms.
- co-firing: spikes from two units within 25 ms of each other.
- STTC (spike time tiling coefficient): a pairwise synchrony measure corrected for firing rate, bounded between −1 and 1.
- co-firing repetition: a cell pair that co-fired during encoding of a stimulus co-firing again during retrieval of the same stimulus.
Connections
- Xu et al. 2023 (LIT-638) and Muller et al. 2026 (LIT-647). Both place large-scale coordination in propagating waves, phase gradients across cortex. This paper places it in brief synchronous bursts at distant sites with no measured direction of travel. Neither study tests the other’s mechanism, and they work at very different timescales (0.01–0.1 Hz fMRI against 90 Hz LFP).
- Damasio 1989 (LIT-385). Recall as time-locked multiregional re-activation of encoding activity is the shape of C4. The paper does not cite it, and LIT-385 is unread; the analogy is mine.
- Kelso 2021 (LIT-601). Zero-lag locking of coupled oscillators is the regime HKB and its Kuramoto extension describe for relative phase. The paper’s coupled-oscillator account (C6) cites other models.
Bearing on the record
- THEORY-063 takes phenomenal unity to be one global functional cluster. This reading shows a single-neuron-level integrating mechanism across hemispheres in working memory, not unity of experience; it neither supports nor tests the theory.
- It supplies one source for a candidate theory: that long-range coordination in the human forebrain is carried by transient synchronous high-frequency events whose effect does not decay with distance. That theory is proposed, not filed.
- No ML instruction. Nothing here belongs in the anthology.
Limitations
- Correlational. Ripples are detected, never 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.
- One dataset, one task, two loads. The authors call the load range narrow and ask for other paradigms before calling ripples a general cognitive mechanism.
- Clinical sampling. Five regions, no posterior parietal cortex, in patients with epilepsy; epileptiform periods and seizure-onset contacts were excluded.
- Small absolute effects. Repetition rates are below half a percent of cell-pair trials; significance comes from very large counts.
- Distance is not measured per patient, but taken from population tractography between parcels.
- Partial overlap of the rate controls. STTC is computed only for pairs with co-firing in both conditions, because co-ripple time per recording is only about 45 s.
Open questions
- Does disrupting co-ripples (for example by closed-loop stimulation) remove the load effect and the reinstatement advantage?
- How do co-ripples relate to the theta–gamma coupling the same dataset shows in hippocampal neurons?
- Is the distance invariance kept in posterior and sensory cortex, which this sample omits?
Corrections
- none to a seeded skim (there was no seed)
- Abstract against Results. The abstract’s “~30%” co-firing increase matches the task-stage figures (24–29% over baseline); the all-pairs co-ripple versus no-ripple median is 34%.