Thalamic spindles and Up states coordinate cortical and hippocampal co-ripples in humans

Ripples are brief high-frequency local field potential oscillations. In rodent hippocampus during NREM sleep, sharpwave ripples mark replay critical for declarative memory consolidation. In humans, ~90 Hz cortical ripples occur in NREM and wake and couple to cortical and hippocampal ripples preceding recall, entraining replay of encoding-related spiking sequences. Co-rippling increases between language areas during reading and before correct semantic judgments, suggesting a general role in cortical integration. Spindles (10–16 Hz) during NREM are generated by thalamic circuits and broadly projected to cortex, often on the Down-to-Up transition, timing that supports consolidation. Down and Up states comprise the slow oscillation/K-complex in human NREM, with thalamo-cortical projections initiating, synchronizing, and terminating spindles and Down-to-Up states. Cortical ripples occur during spindles just prior to the Up peak and may mark replay; hippocampal ripples coordinate with cortical ripples, spindles, Down, and Up states. Cortical ripples co-occur and phase-lock across long distances without decay by separation, implying either dense intracortical coupling or a central subcortical synchronizer. The thalamus is a prime candidate, given direct connections with hippocampus and cortex and its ability to synchronize spindles and Down-to-Up states widely. Optogenetic thalamic stimulation in mice enhances consolidation by evoking cortical spindles co-occurring with Up states and hippocampal sharpwave ripples. Thus, the thalamus could either directly project ripples to cortex or indirectly coordinate co-ripples via spindles/Up states. Ripples have not been reported in the human thalamus previously. Using rare intracranial recordings from non-lesioned, non-epileptogenic anterior and posterior thalamus with simultaneous hippocampal and cortical recordings in epilepsy patients, we asked whether physiological thalamic ripples occur, characterized them, and tested if they co-occur or phase-lock with cortical/hippocampal ripples. We found ~90 Hz thalamic ripples with similar density and duration as cortical/hippocampal ripples that couple to local spindles and occur on the Down-to-Up transition, but thalamic ripples were only weakly related to cortical and hippocampal ripples. In contrast, thalamo-cortical spindles and Up states increased co-rippling between cortical sites, indicating that thalamus coordinates co-ripples via slower waves rather than by projecting ripples.

Prior work shows hippocampal sharpwave ripples in rodents mediate replay and consolidation; human cortical ~90 Hz ripples occur in NREM and waking, couple across cortex and hippocampus preceding recall, and entrain replayed spike sequences. Cortical co-rippling strengthens during language tasks, supporting distributed cortical integration by synchronous high-frequency oscillations. Spindles (10–16 Hz) are generated in thalamus and projected across cortex, often aligned to the Down-to-Up state transition, which is crucial for consolidation. Down states (neuronal silence) and Up states (near-waking firing) are core NREM slow oscillations and K-complexes; thalamo-cortical loops coordinate initiation, synchronization, and termination. Cortical ripples align with spindles just before Up peaks; hippocampal ripples coordinate with cortical slow waves and spindles. Cortical ripples co-occur and phase-lock over long distances without distance decay, implying either dense intracortical coupling or a common subcortical synchronizer. The thalamus, with direct hippocampo-cortical connectivity and demonstrated modulation of cortical spindles and slow waves, is a candidate; thalamic stimulation in mice enhances consolidation by co-occurring spindles/Up states and hippocampal ripples. Previous human studies documented thalamo-cortical spindle coordination and preferred anterior thalamus–anterior cortex coupling; however, thalamic ripples had not been studied in humans. The present study bridges this gap, testing whether thalamic ripples synchronize cortical/hippocampal ripples or whether thalamic spindles/Up states organize co-ripples.

Ethics: All patients provided informed consent; IRB approvals at University of Alabama at Birmingham (patients 1–10; anterior thalamus) and the French Institute of Health (patients 11–13; posterior thalamus). Participants: 13 adults (8 female; 39.5 ± 11.5 years) with pharmaco-resistant focal epilepsy undergoing stereo-EEG for 1–2 weeks to localize seizure onset zones. Thalamic implantation followed clinically accepted procedures, often advancing an insula operculum electrode deeper to thalamus; no thalamic bleeds occurred. Inclusion/exclusion: Patients included had normal background LFPs except infrequent epileptiform activity; 15 were excluded for frequent interictal spikes (>1/min) or artifacts (>1 mV/min). Recording: PMT electrodes (patients 1–10), Natus Quantum amplifier (2,048 Hz, 0.016–683 Hz bandpass); patient 11 at 512 Hz (160 Hz bandpass); DIXI electrodes (patients 12–13) at 1,024 Hz (0.16–340 Hz bandpass). Data downsampled to 1,000 Hz (anti-alias at 500 Hz). Notch filtering at 60 Hz (and harmonics; patients 1–10) or 50 Hz (and harmonics; patients 11–13). Channel selection: Bipolar derivations of adjacent contacts to ensure local LFP. Channels localized to non-lesioned, non-epileptogenic thalamus, cortex, and (in 9/13 patients) hippocampus. To ensure gray matter recordings (local activity, not volume conduction), candidate bipolars were chosen using co-registered pre-op MRI and post-op CT, then refined by physiological criteria: high delta (0.5–2 Hz) and high-gamma (70–190 Hz) analytic amplitudes during NREM, with significant correlation between low-pass delta waveform and high-gamma envelope. Thresholds: cortical delta peak >40 µV; hippocampal >15 µV; thalamic >5 µV. Visual inspection ensured absence of frequent interictal spikes/artifacts and normal broadband appearance. Among 2,145 bipolar channels, 336 were included (110 left-sided); among 39 thalamic bipolars, 30 were included. Channel localization: Cortical surfaces reconstructed via FreeSurfer; automated parcellation (Destrieux; aggregated to Desikan parcels). Anterior cortex parcels: orbitofrontal, prefrontal, cingulate; posterior cortex: remaining parcels. SEEG contacts localized by registering post-implant CT to pre-implant MRI in FreeSurfer space using 3D Slicer; contact centroids marked; parcels assigned via nearest white–gray surface vertex at bipolar midpoint. Thalamic nuclei segmentation via probabilistic atlas; nuclei included AnteroVentral, Ventral Anterior, Reticular, Ventral Lateral, Fasciculus, MedioDorsal, CentroMedian. Hippocampus anterior/posterior boundary defined at posterior limit of uncal head. Sleep staging: NREM (N2/N3) identified using Silber criteria; delta analytic amplitude increases during 8 PM–8 AM periods; visual confirmation of slow oscillations and spindles; epochs retained only if free of frequent spikes/artifacts. Time-frequency: EEGLAB used to compute ripple-triggered spectral power (1–500 Hz) using FFT with Hanning taper; normalized to −2 to −1.5 s baseline; significance via bootstrap (N=200 shuffles), FDR α=0.05. Ripple detection: Bandpass 70–100 Hz (sixth-order zero-phase Butterworth); Hilbert analytic amplitude peaks ≥3 SD above channel mean; onset/offset where z<0.75; events retained if at least 3 oscillation peaks in 120 Hz low-pass within any 40 ms window (stepped 5 ms) within ±50 ms of midpoint; adjacent events within 25 ms merged; ripple center at largest positive peak in 70–100 Hz bandpass. Artifact/epileptiform rejection: reject if 100 Hz highpass absolute z>7; within ±500 ms of putative interictal spikes (detected via combined high-frequency score and spike-template cross-covariance exceeding threshold 130; weights 13 and 25 respectively); reject if coincident with putative interictal spike on any channel; reject if largest peak-to-valley amplitude 2.5× greater than third largest (to exclude single prominent deflections). Oscillation frequency computed as f=N/(2×d), where N is number of 70–100 Hz zero crossings (half cycles, fractional allowed) and d is ripple duration. Down/Up state detection: Bandpass 0.1–4 Hz; top 10% amplitude peaks between consecutive zero crossings within 0.25–3 s identified; polarity inverted if needed so Down negative, Up positive, confirmed by increased high-gamma (70–190 Hz; or 110–160 Hz to avoid ripple band) around Up vs Down peaks; thalamic polarity confirmed by spindle-locked delta vs high-gamma envelope alignment (high-gamma increases during rising delta phase, i.e., Down-to-Up transition). Spindle detection: Bandpass 10–16 Hz; absolute value convolved with tapered 300 ms Tukey window; channel median subtracted; normalize by median absolute deviation; spindles when peak exceeds 1 for ≥400 ms; onsets/offsets when below 1; reject events with large 4–8 Hz or 18–25 Hz power to exclude broadband events and theta bursts. Ripple timing and co-occurrence: Event counts computed in 50 ms bins ±1,500 ms around ripple centers; histograms smoothed (50 ms Gaussian, σ=10 ms). Co-ripple defined as ≥25 ms overlap between ripples on two disjoint cortical bipolars. Co-ripple vs isolated cortical spindle vs thalamo-cortical spindle defined by thalamic spindle onset within 500 ms preceding cortical spindle onset on any thalamic channel. Probabilities computed as co-ripples divided by total time of cortical spindling (condition-specific). Similar method for Up states (cortical Up peaks ±500 ms; preceded by thalamic Up peak within 500 ms). Conditional probabilities: P(CX1–CX2 co-ripple | CX3 ripple) computed by instantaneous probability of CX1–CX2–CX3 co-ripple divided by instantaneous probability of CX3 ripple; similarly for P(CX1–CX2 co-ripple | TH ripple). Phase-locking analyses: PLV (phase consistency) computed for 70–100 Hz ripple or 10–16 Hz spindle phases; channel pairs required ≥40 co-events (ripples overlap ≥25 ms; spindles ≥200 ms). PLV time courses computed at 1 ms steps relative to co-event centers; null distribution via 500 random times −10 to −2 s preceding co-event centers; significance by comparing observed vs null in 5 ms (ripples, ±50 ms window) or 50 ms (spindles, ±250 ms window) bins; FDR-corrected; significant if ≥2 consecutive bins post-FDR p<0.05; plots Gaussian smoothed (10 ms window, σ=2 ms). White matter controls: Bipolar channels in adjacent white matter lateral to thalamic bipolars used to assess volume conduction; mean white matter LFP relative to thalamic ripple centers computed; ripple detection applied with same criteria/thresholds. Statistics: α=0.05; FDR correction across channels/pairs/bins; random shuffles N=200; paired and two-sample t-tests; peri-event significance via shuffled null comparisons; channel/pair significant modulation if ≥3 consecutive bins post-FDR p<0.05; Wilcoxon signed-rank tests for across-channel/pair modulation; leading/lagging via binomial test comparing 250 ms before vs after t=0; co-occurrence above chance via shuffling inter-ripple intervals; χ² tests for proportion comparisons; relationships categorized as weak (<25% increase from chance) vs strong, and infrequent (<25% significant channels) vs frequent.

• Human thalamic ripples occur during NREM sleep in both anterior and posterior thalamus, centered near ~90 Hz, with features resembling cortical and hippocampal ripples, and are temporally embedded within local spindles on the Down-to-Up transition. • Ripple characteristics (mean ± SD across channels): Anterior thalamus: density 19.8 ± 4.6 min⁻¹; peak 70–100 Hz analytic amplitude 1.79 ± 0.91 µV; oscillation frequency 92.4 ± 0.6 Hz; duration 62.6 ± 5.9 ms. Posterior thalamus: density 11.3 ± 2.2 min⁻¹; amplitude 1.95 ± 0.34 µV; frequency 90.2 ± 0.4 Hz; duration 93.3 ± 23.0 ms. Cortex: density 21.9 ± 2.5 min⁻¹; amplitude 4.17 ± 1.69 µV; frequency 90.1 ± 0.5 Hz; duration 65.3 ± 3.6 ms. Hippocampus: density 21.3 ± 4.7 min⁻¹; amplitude 11.39 ± 7.23 µV; frequency 88.5 ± 1.3 Hz; duration 74.8 ± 7.9 ms. • Thalamic ripple timing relative to slow waves/spindles: Anterior thalamic ripples occur ~225 ms after thalamic Down peaks and ~125 ms before thalamic Up peaks; they tend to occur shortly after spindle onset (~50 ms). Posterior thalamic ripples occur ~275 ms after thalamic Down peaks, ~275 ms before thalamic Up peaks, and ~350 ms after spindle onset. Multiple channels showed significant order preferences with Down states preceding ripples and ripples preceding Up peaks. • Thalamic ripples infrequently and weakly couple/co-occur with cortical or hippocampal ripples compared to robust cortico-cortical and hippocampo-cortical coupling: Among anterior thalamo-cortical pairs, 14% (88/649) significantly coupled and 6% (40/649) had significant co-occurrences; posterior thalamo-cortical pairs: 27% (7/26) coupled and 31% (8/26) co-occurred; anterior thalamo-hippocampal pairs: 6% (5/81) coupled and 2% (2/81) co-occurred. Cortico-cortical pairs: 52% (4,063/7,796) coupled and 47% (1,828/3,898) co-occurred. Hippocampo-cortical pairs: 51% (437/865) coupled and 15% (126/865) co-occurred. • Conditional co-occurrence probabilities: The probability of cortico-cortical co-rippling given rippling on another cortical channel was significantly higher than given rippling on a thalamic channel (e.g., 0.18 ± 0.003% vs 0.09 ± 0.001% for anterior thalamus; p=1×10⁻²³³; and 0.67 ± 0.09% vs 0.17 ± 0.03% for posterior thalamus; p=6×10⁻⁷). Larger-amplitude ripples co-occurred more often, but cortico-cortical co-ripple probabilities remained 5–8× (vs anterior thalamo-cortical) and 13–20× (vs posterior thalamo-cortical) higher. • Phase-locking: Thalamic ripples rarely phase-lock with cortical ripples (1/451 significant aTH–CX pairs; 0/20 pTH–CX), and never with hippocampal ripples (0/47 thalamo-hippocampal, 0/490 hippocampo-cortical). In contrast, cortico-cortical pairs showed significant ripple phase-locking (52/2,833; 7/20 in separate dataset) (χ²=16.2, p=0.001 across pair types). • Thalamo-cortical spindle phase-locking: Robust 10–16 Hz PLV between anterior thalamus and anterior cortex (188/369 significant pairs) and, to a lesser extent, anterior thalamus with posterior cortex (71/280). Posterior thalamus did not phase-lock with anterior cortex (0/4) but did with posterior cortex (9/22). Mean PLV modulation within ±250 ms was greater for anterior vs posterior cortex (p=5×10⁻¹⁵). Thalamo-hippocampal co-spindles were significantly phase-locked in 31/81 (38%) pairs. • Thalamic spindles and Up states enhance co-rippling: Cortico-cortical co-ripple probability during cortical spindles increased by 53% when an anterior thalamic spindle co-occurred (p≈3×10⁻¹⁰⁹) and by 68% when cortical Up states were preceded by thalamic Up states (p≈3×10⁻¹⁰⁵). Posterior thalamic spindles increased cortico-cortical co-rippling by 77% (p=4×10⁻⁶); no significant increase with posterior thalamic Up states. Hippocampo-cortical co-ripples increased by 31% during anterior thalamo-cortical vs isolated cortical spindles (p=3×10⁻⁵) and by 11% when cortical Up states were preceded by anterior thalamic Up states (p=0.003). • Co-rippling is further enhanced when at least one cortical site has phase-locked spindles with thalamus (p=4×10⁻⁴), supporting a model where thalamo-cortical spindles/Up states modulate cortical excitability to promote co-rippling, while precise ripple phase-locking arises from intracortical mechanisms.

The discovery of ~90 Hz ripples in human anterior and posterior thalamus, occurring on the local Down-to-Up transition and coupling to thalamic spindles, expands the repertoire of sleep-related high-frequency events beyond cortex and hippocampus. However, the rarity of thalamo-cortical and thalamo-hippocampal ripple co-occurrence and the near absence of ripple phase-locking with cortex or hippocampus argue strongly against a direct wave-by-wave thalamic driving of widespread cortical or hippocampal ripples. Instead, thalamo-cortical spindles and Up states—waves the thalamus is known to initiate and synchronize—substantially increase the likelihood of cortico-cortical and hippocampo-cortical co-rippling, particularly when thalamo-cortical spindles are phase-locked and aligned to the Down-to-Up transition. This supports a modulatory mechanism wherein thalamic slow oscillations and spindles transiently depolarize cortical networks (~50–100 ms), triggering local ripple-generating circuits; intracortical connectivity then establishes precise ripple phase-locking over long distances, consistent with high-bandwidth information binding models. The findings align with prior human and animal work showing thalamo-cortical coordination of spindles and slow waves and the role of hippocampal ripples in information transfer. Preferential phase-locking of anterior thalamus with anterior cortex and posterior thalamus with posterior cortex reflects anatomical connectivity and further supports targeted thalamic modulation. Overall, the thalamus appears to orchestrate the timing and conditions for distributed ripple co-occurrence via spindles and Up states, while intracortical networks determine the fine-scale phase synchrony that may underlie memory consolidation and cross-cortical binding.

This study provides the first evidence of physiological ~90 Hz ripples in human anterior and posterior thalamus with characteristics similar to cortical and hippocampal ripples and occurring on the Down-to-Up transition during spindles. Despite this, thalamic ripples rarely co-occur or phase-lock with cortical or hippocampal ripples, indicating they do not directly synchronize forebrain ripple networks. In contrast, thalamo-cortical spindles and Up states—especially when phase-locked—significantly increase cortico-cortical and hippocampo-cortical co-rippling, implicating a thalamic modulatory role that sets the stage for intracortical mechanisms to establish precise ripple synchrony. These insights advance our understanding of how sleep rhythms coordinate widespread neural replay for memory consolidation in humans. Future work should increase sampling across thalamic nuclei with finer spatial resolution, assess causal interactions (e.g., targeted thalamic stimulation), delineate state-dependent dynamics across NREM stages, and probe how these mechanisms impact specific memory processes and clinical interventions.

• The cohort comprised patients with pharmaco-resistant focal epilepsy; although stringent exclusion of epileptiform activity and artifacts was applied, it is impossible to definitively rule out residual influences of epilepsy on the findings. • Thalamic sampling was limited, with relatively wide spacing and large recording contacts; assignments to specific thalamic nuclei are approximate due to registration uncertainties and indistinct nuclear boundaries. • Granular distinctions beyond anterior vs posterior thalamus were not feasible, constraining anatomical specificity of conclusions. • High-gamma magnitude in thalamus is small, complicating polarity determinations and slow-wave alignment; multiple validation approaches were used but residual uncertainty may remain. • Hippocampal ripple detection did not restrict to sharpwave-associated events to maximize co-ripple detection, potentially mixing ripple subtypes. • Sleep in the epilepsy monitoring unit may be disrupted; despite careful NREM epoch selection, residual variability could affect sleep-wave dynamics.