The causes of epilepsy and its defining symptom – seizures – lie in cellular and architectural features of brain tissue. Though, for clinical diagnoses, electrophysiological measurements: e.g., stereo-electroencephalography (SEEG), are used to infer the causes of seizure activity. Therefore, understanding how brain tissue and its synaptic characteristics generate electrophysiological signals is essential for informed diagnosis. We addressed this issue using neural mass models (NMM) and dynamic causal modelling (DCM).

We asked how neuroreceptor density data from autoradiography studies are related to ‘healthy’ cortex SEEG signals from individuals with epilepsy. We first tested if a canonical microcircuit NMM replicates electrophysiological (SEEG) data using DCM. We then asked if receptor densities can be predicted by SEEG signals, and if regional receptor compositions (‘fingerprints’) can explain regional variation in SEEG spectra.

First, our DCM replicated ongoing awake cross spectral densities of intracranial EEG signals (1770 data series) highly accurately; with 40 exceptions (≅ 2.3%) DCM was able to explain key components of regional cortical signal variability.
Second, using both correlation between DCM parameters and receptor densities, and fitting DCM parameters with AMPA, NMDA and GABA regressors combined with Parametric Empirical Bayesian (PEB), we found that receptor densities are only predictable collectively but not individually.
Third, using PCA we captured regional receptor composition variability and showed that the principal components of receptor density fingerprints can explain regional variation in the generation of SEEG spectra, i.e., including receptor density fingerprints improves model evidence (free energy ≈ accuracy – complexity).

In summary, we show how tissue characteristics (i.e., receptor density) can be incorporated to improve biophysically grounded models and explain regional variations in electrophysiology. The results will be part of a toolbox (published on GitHub and EBRAINS) that enables integration of normative datasets as prior information to generate patient specific models of (pathophysiological) cortical dynamics.