High frequency oscillations and interictal discharges at 50 μm spatial resolution

Since the early days of EEG and epilepsy, clinicians have commonly viewed seizures and interictal discharges (IIDs) using macroelectrodes and relatively low-frequency bands (e.g., no higher than beta or gamma frequencies). While this view of seizures and IIDs continues to be an essential component of clinical practice, additional corroborating data is available at faster frequencies and smaller spatial scales. Indeed, over the last few decades, we have seen that as we open the door to more information, we continually obtain new insights. For example, higher sampling rates allow the acquisition of High Frequency Oscillations (HFOs), a promising biomarker of epileptogenic tissue. High temporal resolution data allowed the discovery of microseizures (Stead et al., 2010), providing new information about the various spatial scales relevant for seizure generation. However, many questions remain regarding the cellular and network activity that produce HFOs and microseizures and their relationship to epilepsy.

The work reported by Yang and colleagues in this issue of Clinical Neurophysiology (Yang et al., 2021) continues this pattern of novel data providing new information and insights. Interoperative recordings of several minute duration were acquired at 30 kHz (1 Hz to 7500 Hz bandpass filter) in 30 subjects using custom poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS) electrodes with 50 μm spatial resolution. While this is not the first study to investigate epileptic electrographic activity using microelectrodes, other studies typically had a much larger pitch between microelectrodes, e.g., 400 μm (Schevon et al., 2008). Using this high spatial and temporal resolution data, Yang et al. (2021) analyzed HFOs, IIDs, multiunit activity (MUA), and microseizures.

Previous data suggest that HFO generators are small. Micro-electrode recordings indicate that the generators may be less than 1 mm (Worrell et al., 2008). This is supported by macroelectrode data reporting that HFOs are often (but not always) observed on a single macroelectrode (Crépon et al., 2010). In contrast to HFOs, IIDs are generally believed to come from larger sources, supported by the fact that IIDs are easily recorded in scalp EEG.

Yang et al. (2021) found that IIDs and HFOs can be generated by regions with a radius as small as 50 μm. However, both IIDs and HFOs more commonly occurred across larger regions in their data. Thus, the generators of IIDs and HFOs have a wide range of possible sizes. This information can help guide future models regarding how these events are produced. We also note that the specifics of the recording technology also impact what is observed in any given dataset. For example, the data of Yang et al. (2021) do not disagree with the expectation that scalp EEG will preferentially record events (IIDs, HFOs) with larger generators and that microelectrodes may be needed to observe events with very small generators.

Yang et al. (2021) also considered the spatiotemporal dynamics of IIDs and HFOs. They observed multiple distinct patterns in the IID propagation and multiple unique HFO detection patterns. Given how little data is available at this spatial and temporal scale, reporting the existence of these patterns is an important first step.

However, these patterns perhaps raise more questions and hypotheses than they answer. The full importance and relevance of these patterns will be understood as these further hypotheses are tested. Future work is needed to quantify the clinical utility of these patterns and to understand how these patterns support or modify current ideas about how IIDs and HFOs are generated. In other words, the relevance of these patterns is not what they tell us today, but what they will lead to in the future.

The data in Yang et al. (2021) also address a long-standing question regarding the relationship between multiunit activity (MUA) and HFOs. Computational modeling (Fink et al., 2015; Shamas et al., 2018) and direct mathematics (Gliske et al., 2017) suggest that no oscillatory mechanism is needed to produce HFOs, conclusions supported by early work suggesting pathological HFOs are just action potentials from bursts of pyramidal cell firing (Bragin et al., 2002).

The detectors used by Yang et al. (2021) are designed well to address the relationship between MUA and HFOs. To avoid a priori correlation between MUA, HFOs, and IIDs due to the detection algorithm, the detector of MUA is based on clustering unitary events. This is in contrast with the detectors of IIDs and HFOs in this paper, which use the power within specific frequency ranges. Thus, each detector is viewing distinct data.

MUA detections were found to be coincident with HFOs and IIDs on occasion (Yang et al., 2021). However, the temporal correlation between MUA and HFOs was small but statistically significant, while the temporal correlation between MUA and IIDs was small and not statistically significant. Due to these small correlations, the authors thus conclude that the MUA, IID, and HFO detections in their data are sensitive to different physiological events.

While this is excellent, relevant data, it does not yet provide a clear conclusion about the nature of HFOs and MUA. If a strong correlation had been found between the detected MUA and HFO events, then one could infer they were the same event, which in turn would support the hypothesis that HFOs are caused by highly active (but not necessarily oscillatory) pyramidal cell firing. However, multiple scenarios could explain the observed lack of correlation between MUA and HFOs. For example, perhaps the single units of the cells which are generating the HFOs and/or IIDs were not recorded by the microelectrodes, possibly being too distant. If the single units were not recorded, the data then places no constraints on the firing patterns of these cells. It is also possible that the authors' choice to “filtered out errant, regular, rhythmic oscillatory events in the MUA detections” may have biased the correlation between MUA and HFOs.

Building off the foundation of the presented data, future work could consider the identification and tracking of specific single units (including separation of excitatory and inhibitory cells). Future research could also increase the number of implanted electrodes and the recording duration, improving the likelihood that single units are recorded from the cells causing the HFOs and IIDs. One could also analyze both ripples and fast ripples, as there is not a consensus in the community about whether ripples or fast ripples are more useful or more pathologic. An alternate approach would be to analyze single units during the HFO events rather than just temporal correlation between detections of MUA and HFOs.

One paper cannot answer all of the questions regarding HFOs, IIDs, and microseizures at these fast and ultra-short spatiotemporal scales. However, the work of Yang et al. (2021) clearly demonstrates that rich information is available beyond that observable at larger spatial scales. Parallel to this basic science research focused on increasing understanding, it will be another task to incorporate the acquisition, analysis, and interpretation of high temporal resolution, microscale data into clinical practice. While this road may be long, there is strong potential for these advances to lead to direct benefit for patients with epilepsy.