Abstracts

Rationale: Cognitive impairment (CI) is a common comorbidity in epilepsy that is influenced by heterogeneous factors, including anti-seizure medication and disease activity. However, the neuronal mechanisms underlying cognitive variability in persons with epilepsy (PwE) are unclear. The brain criticality hypothesis postulates that brain dynamics are set at a phase transition where cognitive performance is optimized. Despite this, the experimental link between CI and criticality has not yet been established in humans. Long-range temporal correlations (TCs), measurable in intracranial EEG (iEEG), characterize information maintenance, are hallmarks of brain criticality, and offer an opportunity to investigate criticality and CI in PwE during presurgical evaluation. Here, we examine the relationship between TCs, CI profiles of PwE, and multiple factors known to affect cognition.


Methods: We simulated a neuronal network model to predict the effect of three cognition-related perturbations on TCs: Slow-wave-sleep (SWS), interictal epileptiform discharges (IEDs), and anti-seizure medication (ASM). Next, we tested these predictions on two independent datasets of multi-day iEEG recordings including a total of 104 persons with drug-resistant epilepsy: Dataset 1: 81 PwE (35 females, average age 32 ± 11 years), Dataset 2: 23 PwE (12 females, average age 29 ± 13 years). TCs were extracted from the high-γ power time series, SWS and IEDs were evaluated using established automated detectors. CI profiles were assessed using 13 cognitive tests and classified into four cognitive domains (language, attention, verbal memory, and working memory; Figure 1B). The relationship of TCs to CI profiles was tested across different recording days and brain regions (Figure 1B; Brunner-Munzel tests, Benjamini-Hochberg correction).


Results: Model simulations predicted a decline of TCs during SWS, with more IEDs and high ASM dosage. These predictions were confirmed experimentally (Figure 1A). TCs significantly declined during SWS compared to non-SWS epochs (Dataset 1: ΔTC = 8 ± 19%, p < 0.001; Dataset 2: ΔTC = 20 ± 28%, p < 0.05), with more IEDs (Dataset 1: ΔTC = 12 ± 33%, p < 0.001; Dataset 2: ΔTC = 13 ± 40%, p < 0.05), and high ASM dosage (Dataset 1: ΔTC = 29 ± 18%, p < 0.001; Dataset 2: ΔTC = 16 ± 35%, p < 0.001; Figure 1A). TCs predicted CI in language and attention domains across multiple brain regions (Figure 1B). No other studied features did, including EEG-derived spectral powers, surrogate TCs, high-γ power, IEDs, SWS, IEDs, and ASM.


Conclusions: Our results provide experimental evidence that signatures of criticality, i.e., TCs, are predictive of cognitive performance. Further, we show how sleep, epileptic activity, and medication dosage perturb the critical state. This suggests critical dynamics to be the setpoint to determine optimal network function, thus providing a framework for the heterogeneous mechanisms impacting cognition in conditions like epilepsy (Figure 2).


Funding: CM and PM are funded by NeuroCure Cluster of Excellence, EXC-2049-390688087.