An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Noebels JL, Avoli M, Rogawski MA, et al., editors. Jasper's Basic Mechanisms of the Epilepsies. 5th edition. New York: Oxford University Press; 2024. doi: 10.1093/med/9780197549469.003.0016
This chapter reviews the history, technology, and scientific contributions of single-neuron recordings in humans, which have largely been conducted in epilepsy patients undergoing surgical procedures for clinical treatment purposes. Single-neuron recordings have been used by epileptologists to study interictal discharges and the onset, propagation, and termination of human focal seizures, and by neuroscientists and biomedical engineers to study normal cognition in areas such as visual and spatial memory and decision-making. While there are many practical and ethical limitations, this work has proven invaluable for both exploratory studies and for validating research from animal or computational models. Ongoing technological innovation is likely to expand the use and applicability of human single-neuron studies in the foreseeable future.
The first direct microelectrode recordings capable of capturing single-neuron activity in humans, over 60 years ago, were performed with glass micropipettes during surgical intervention in a patient with epilepsy, with an aim to locate the epileptogenic region (Ward and Thomas, 1955). While these recordings were deemed insufficient to compare normal and abnormal tissue, slowly over the following decades methods for these recordings—along with institutions performing them—have expanded greatly. Once constrained to acute recordings in the operating room, advancement in microfabrication techniques, together with simultaneous advances in computational processing, has enabled implantation of devices to record the activity of many single neurons over days to weeks during invasive electroencephalogram (EEG) monitoring as part of routine epilepsy surgery evaluation.
Given the invasive nature of these methods, a key question is what information they may provide that is unavailable to standard electrocorticography (ECoG) or stereo-EEG (sEEG) recordings. While the exact composition of the local field potential (LFP), as recorded with clinical scalp or intracranial electrodes, is still a matter of debate (Lęski et al., 2013), it is thought to be primarily a summation of nearby synaptic currents, with individual action potentials having little contribution to the overall signal at that location (Buzsàki et al., 2012). As a result, these traditional recording approaches give insight to a region’s input (i.e., the downstream result of distant neural activity) while the action potentials of the local neuronal population more accurately provide the region’s neural output. Recording only the input of a brain region thus leaves us with an incomplete picture of events. As we will describe in subsequent sections, the ability of human single-neuron recordings to fill in the missing information has had a substantial impact on our understanding of a range of normal and pathological brain function. In this chapter, we will explore the current technologies and methodologies, and provide an overview of the literature in this important and expanding field.
Since the advent of recording human single neurons (“single units”; see fourth section), tools have advanced from glass pipettes and oscilloscopes to microfabricated wires or silicon wafers and the capacity to store substantial data sets digitally in real time. During the improvements over these decades, multiple parallel tools have been designed (Fig. 16–1), each with their own specific merits and limitations.
Specialized microelectrode devices capable of single-neuron recordings in humans. A. Grid of fine wires (center) between macroelectrode grid contacts for recording from the pial surface at high resolution. B. The “Utah” array, comprised (more...)
The first devices were limited to cortical recordings, accessible during craniotomies (Rayport and Waller, 1967), and by the 1970s single-unit recordings in the mesial temporal lobe were being performed in humans, first with a platinum-iridium microelectrode with glass coating in the amygdala of awake, responsive patients (Verzeano et al., 1971), followed by the first recordings using microwires inserted through the lumen of clinical depth electrodes, which also showed stability of the extracellular traces of putative single neurons over multiple days (Babb et al., 1973). Subsequently, several microelectrode devices were developed, some of which have been made widely available and can be utilized in patients undergoing long-term (up to 30 days) invasive EEG monitoring studies.
Due in large part to the NIH Brain Initiative (Jorgenson et al., 2015), there has been an acceleration of interest in specialized human recordings and the development of new technologies. We therefore expect to see continued development of new devices, which will eventually expand the ability to conduct routine, multiscale human recordings and perhaps make it possible for such studies to be integrated into clinical practice (Chari et al., 2020). A recent example is the “Neurogrid”, a high density non-penetrating microelectrode array intended for use on exposed brain surfaces. This is an organic, polymer sheet with embedded microelectrodes. This microelectrode array can record action potentials intraoperatively from neurons in superficial neocortex and the alvear surface of hippocampus post-removal of overlying tissue (Khodagholy et al., 2015). At time of writing this chapter, a novel silicon array (“Neuropixels”) capable of recording from 384 individual microelectrodes on a single 10 mm shank has been tested intraoperatively in patients undergoing either DBS placement or craniotomies for surgical resections (Paulk et al., 2021).
While the choice of device for recording single units in epilepsy patients is necessarily dependent on clinical plans and ethical considerations (covered in the next section), it is worthwhile to consider some specific use cases. We describe here the usage in a clinical setting of two Food and Drug Administration (FDA)-approved microelectrode devices.
The most commonly used microelectrode device in epilepsy patients, and the first to be developed and approved by the FDA for human use in the United States, is the “Behnke-Fried” array (Fried et al., 1999; Fig. 16–1C). This device consists of a microwire bundle contained within a standard clinical depth array, thus allowing for microelectrode recording within the form factor of a clinically indicated array placement, and without the need to add additional devices to the implant plan. It is capable of sampling single-unit activity in structures such as the hippocampus, cingulate gyrus, and amygdala (Fried et al., 1999). An alternative “micro-macro” design was later developed with individual, nonpenetrating microelectrodes along the depth array shaft for simultaneous recording of single neurons in the vicinity of the clinical macro-contacts in deeper structures (Worrell et al., 2008). This device also received regulatory approval. Recently, another microwire depth array was developed with a novel design utilizing tetrodes with recording sites positioned between two macro-contacts, instead of wires emerging from the tip of the array shaft (Despouy et al., 2020). Tetrodes, which have long been used in animal work, allow the simultaneous recording of action potentials from the same neuron on multiple microelectrodes, thereby permitting the use of triangulation to better isolate individual neurons (O’Keefe and Recce, 1993; Gray et al., 1995; Harris et al., 2000).
Behnke-Fried bundles can be implanted stereotactically, as is commonly done in clinical depth electrode implants. The microwire bundle consists of nine wires of 40 µm diameter, which are trimmed such that they extend up to 5 mm beyond the tip of the macroelectrode after insertion. Eight of the microwires are coated in polyamide insulation until the tip, resulting in 100–300 kΩ impedance; the ninth wire is uninsulated, acting as reference (Fried et al., 1999; Misra et al., 2014). An advantage of this system is the ability to customize the microwire length, including cutting them at an angle, allowing for sampling across lamina (Bragin et al., 2002).
A great advantage of the Behnke Fried array is that its form factor resembles that of a standard depth array. That said, as with all penetrating microelectrodes, additional micro-injury can be expected (Misra et al., 2014), although there have been no reported clinical complications as a direct result of this research device (Cash and Hochberg, 2015). Instead, their complication rate is similar to that of standard clinical depth arrays (Carlson et al., 2018). A key benefit of this design is the ability to implant multiple bundles at once, allowing for simultaneous sampling of multiple regions in the same patient. However, Behnke-Fried microwires are not able to be placed in lateral neocortical structures, and they are generally limited to deep structures such as those of the limbic system. Similarly, as the microwires are designed to splay upon exit from the tip of the macroelectrode, the exact positioning of each wire is difficult to predict, and their diameter precludes reliable identification of individual wires in postimplantation imaging. Typically, standard co-registration techniques are employed to calculate the trajectory of the microwire bundle and to estimate their likely anatomical positioning (Papademetris et al., 2006; Ekstrom et al., 2008; Wu et al., 2014).
A second device that has been utilized in epilepsy surgery patients, although not as commonly as microelectrode depth arrays, is the “Utah” microelectrode grid (Fig. 16–1B). The “Utah” array was originally developed for a study of brain-machine interfaces in quadriplegic patients (Nordhausen et al., 1994, 1996), with devices implanted for periods of years (Hochberg et al., 2006, 2012; Masse et al., 2014). It consists of a square grid of 10 by 10 microelectrodes with inactive corner contacts (totaling 96 functional electrodes), fabricated from silicon. In contrast to microwire bundles, the “Utah” array requires a large burr hole or craniotomy for implantation, and it is placed onto an exposed neocortical structure. As a result, its use is restricted to cases with simultaneous implantation of grid or strip clinical contacts. The “Utah” array is also not magnetic resonance imaging (MRI)-compatible, which may further restrict its use. However, its design is perfectly suited for recordings of neocortical neurons, with configurations of 1.0- and 1.5-mm length electrodes available, typically targeting the cell-body-rich layer 4–5. With its fixed inter-electrode pitch of 400 µm, designed to approximate the scale of cortical macrocolumns, these are ideal for analyses of the propagation of epileptiform activity through the cortex (Smith et al., 2022), or for studying spatial interactions of single neurons across an extended 4 mm × 4 mm cortical region.
In contrast to microwire depth recordings, the Utah array is notable for its sensitivity to multi-unit and single-neuron activity, capable of distinguishing hundreds of single units simultaneously in a given recording (Truccolo et al., 2011; Peyrache et al., 2012; Merricks et al., 2015, 2021). Conversely, this largely precludes sampling multiple regions at once in the same patient, unlike their microwire depth array counterparts. Furthermore, as their implantation is more invasive, involving a pneumatic insertion, for ethical reasons their usage should strictly be limited to tissue that is confidently expected to be part of the subsequent clinical resection, although, similar to Behnke-Fried implants, no adverse clinical outcomes have been reported after Utah array implants (House et al., 2006; Waziri et al., 2009). However, in recent years, stereotactic depth array implants for invasive EEG monitoring (Baud et al., 2018) and implantable therapeutic stimulation devices have been rapidly displacing traditional subdural grid implants and brain resection as documented by the National Association of Epilepsy Centers (https://www.naec-epilepsy.org). Therefore, future use in epilepsy surgery patients of the Utah array device, or any device with similar placement requirements, is likely to be limited. Preservation and data sharing of existing patient recordings will be a priority for future studies, as many questions remain that can be studied with these valuable data.
While originally limited to intraoperative recordings, the majority of single-unit recording in humans is now performed extraoperatively (13.6% versus 86.4% of studies from a recent meta-analysis; Chari et al., 2020). This shift to recordings in the Epilepsy Monitoring Unit (EMU) during routine clinical observation over days to weeks, in combination with higher density of microelectrodes, has brought with it both technical and ethical complexities, as discussed below.
As with all direct-contact human research, studies involving single-unit recordings require institutional review board (IRB) approval and informed consent from the patient. Additionally, the research devices to be used are typically required to have appropriate regulatory approval or an investigational device exemption (IDE). Beyond this, however, there is an additional concern: all invasive research implants that have the potential to cause micro-damage carry added risk to the patient without any direct benefit, even though no clinical complications have been reported (House et al., 2006; Waziri et al., 2009; Misra et al., 2014; Carlson et al., 2018).
Acute or subchronic extraoperative recordings of human single units are almost exclusively performed in patients who are undergoing invasive EEG studies to localize seizure onset zones prior to potential epilepsy surgery. As discussed in the previous section, the type of research device available and its potential location is entirely dictated by the clinical plan, and the opportunity to participate in research is contingent on the agreement of the treating physicians and surgeons. The informed consent requires explanation of the device and its associated risks to the patient by the neurosurgical team, along with careful communication of the extent to which the research device adds to the risks beyond those of the clinical procedure, or how its presence may impact clinical procedures. If implantation proceeds, the researcher is ethically mandated to ensure that neither the data acquisition process nor associated research procedures interfere with clinical care, including patient-care team communications, medical procedures, and specifically the continuity and quality of the clinical EEG recording.
Human single-unit data are necessarily rare due to the aforementioned ethical and practical constraints on invasive implants conducted for research purposes, as well as the inherent technical and monetary requirements. As such, it is important to ensure the quality of recorded data and its safe storage. In the majority of settings, research recordings are limited only to the periods during which research tasks are being undertaken; however, a few centers worldwide maintain continuous research recording throughout the patient’s clinical assessment on the EMU (Schevon et al., 2008; Waziri et al., 2009; Truccolo et al., 2011; Bower et al., 2015; Niediek et al., 2016; Elahian et al., 2018; Misra et al., 2018; Diamond et al., 2021). Regardless of duration, a typical recording setup in the EMU consists of a separate research system from the clinical EEG. This allows for minimal impact to the clinical framework as well as the freedom to set recording parameters independently for the research device, which requires a far higher sampling rate to record single neurons (typically ≥ 20 kHz) than is necessary on the clinical system.
The parallel recording setup, however, brings a few potential issues, including the possibility of ground loops due to connecting the patient to two grounded systems, the need for hospital staff to disconnect the system quickly should an emergency arise, ensuring that additional connectors and cables do not impinge on patient comfort and mobility, and the need for frequent monitoring to maintain data quality and to ensure that the research system does not impact the clinical recording. Another consideration is the need for a consistent system that will permit aligning data from the clinical and research recording systems. Given these considerations, a few centers have opted to avoid these problems by using clinical acquisition systems that are also capable of handling microelectrode recordings.
In the case of recordings with many microelectrodes, the required high sampling rate quickly results in vast datasets. As an example, a recording from the 96-channel “Utah” array at 30 kHz, using 16-bit analog-to-digital resolution (i.e., 216 discrete voltage levels), results in just under 20 GB per hour of recording, translating to just under 1 TB of data for every 2 days of continuous recording. Carefully designed compression algorithms can reduce this size considerably (Brinkmann et al., 2009), but at the risk of making the data unrecoverable if a translation error occurs or if the encoding scheme is changed. Given the scarcity and value of these data, they must be safely and securely stored, ideally on a server cluster which is capable of continuous backups and meeting the institution’s IRB and Information Technology security requirements. Sharing data between research centers presents not only the opportunity to increase data sets for the individual researchers but also provides natural backups in the case of catastrophic data loss at any one location.
Unlike techniques available in animal research, such as patch-clamp recordings and calcium imaging, extracellular recordings from these devices cannot automatically isolate the action potentials of separate neurons without some degree of manual intervention. Therefore, to analyze the activity of single neurons, we must assign putative single neuronal identities to the recorded extracellular action potentials (commonly referred to as “spikes,” not to be confused with the clinical EEG epileptic spikes), a process known as spike sorting (Fig. 16–2). Due to their probabilistic nature, these clusters of waveforms are typically termed “single units,” as opposed to single-neuron recordings from confirmed methods such as cell-attached or intracellular recordings.
Spike sorting from raw data to single units. A. (i) An example of a raw, unfiltered signal from a microelectrode and (ii) the bandpass filter (300 Hz to 5 kHz) of the same signal, revealing the multi-unit activity (MUA). The red dashed line shows the (more...)
The synaptic activity in the LFP is dominated by larger amplitude oscillations, which are predominantly contained below 500 Hz (Buzsàki and Draguhn, 2004). The spikes arising from nearby action potentials, however, are less than 2 ms in duration, thus having minimal power below 500 Hz (Buzsàki et al., 2012). Additionally, an isolated action potential has a very small driving force compared to oscillations generated from population activity, so its ability to influence an EEG signal recorded from a site containing thousands of neurons is limited. Indeed, the excision of all isolated spikes from the LFP has been shown to have no significant impact on the power of the lower frequencies (Belluscio et al., 2012).
As a result, bandpass filtering the continuous data from each microelectrode, typically with a high pass of between 300 and 500 Hz and a low pass between 3 and 5 kHz (Quian Quiroga et al., 2004; Truccolo et al., 2011; Merricks et al., 2015, 2021; Elahian et al., 2018), results in signals primarily composed of multi-unit activity (MUA; Fig. 16–2A-B), with isolatable, individual deflections from neurons whose somata were within the immediate listening sphere, or sensing environment, of the microelectrode, typically within 50 to 100 µm of the electrode tip (Gerstein and Clark, 1964; Buzsàki, 2004). Multiple toolboxes, both open-source and proprietary, for sorting spikes after this stage are available—including many that aim to account for drift of spike shapes as a result of recording conditions or micromotion (Niediek et al., 2016; Steinmetz et al., 2021)—and the optimum strategy is still an open question (Carlson and Carin, 2019). However, the most common next step is to detect deflections in the MUA that have local maxima or minima beyond a chosen voltage threshold, and brief epochs around those detections are extracted to store the spike waveforms.
Assuming that a rigorous spike detection process has been employed, clustering of the waveshapes can then be performed in order to group spikes by their putative neuron of origin (Fig. 16–2C–E). While intracellularly recorded action potentials of specific cell types tend to show similar features from cell to cell, the extracellular record of these action potentials varies depending on the exact geometry between the microelectrode and the cell body, due to the flow of ions, the capacitive property of the cell membrane, and the low-pass filtering properties of the extracellular milieu (Gold et al., 2006). As such, since two neurons cannot inhabit the same space, these alterations to extracellular spike shape can be used to cluster spikes into groups of similar waveforms that likely arise from the same neuron. This process is typically semi-automated, requiring manual checking at the final stage (typically visual review, thus remaining to some degree a subjective process), and makes use of compression techniques on the spikes such as principal component analysis (e.g., Lewicki, 1998; Hill et al., 2011; Fig. 16–2C–E), or wavelet features (e.g., Quian Quiroga et al., 2004; Niediek et al., 2016).
Finally, insights into epileptiform activity are often derived from firing rates, cross- correlations, or other spike-rate-dependent analyses, and so it is important to conduct a quantitative assessment of both the false-positive and the false-negative rates of neuronal assignments of spikes. Any analysis of firing rate alterations in the presence of potential amplitude changes should be contextualized by a calculation through time of the percentage of waveforms that may be subthreshold and therefore not detected (Fig. 16–2F), even if its cluster appears well-isolated, in order to avoid false discovery of reduced—or even cessations of—firing rate. Overall, quality metrics for spike sorted data as per Hill et al. (2011) are an underutilized tool in the reporting of findings from human single-unit data.
Epilepsy is a disorder involving complex and dynamic interplay between inhibitory and excitatory activity (Nelson and Turrigiano, 1998; Isaacson and Scanziani, 2011; Pouille et al., 2013). Therefore, in using single-unit data to elucidate seizure mechanisms, treating all single units as the same cell type limits the conclusions that can be drawn and may even be misleading. For example, observed changes in firing rate during a given epoch of interest may include contrasting shifts in neural activity between pyramidal cells, which are relatively slow-firing, and fast-spiking inhibitory interneurons. Indeed, even among inhibitory interneurons, downstream effects of their activity are likely to be nuanced: somatostatin- and parvalbumin-containing interneurons primarily target the dendritic and somatic regions, respectively (Megías et al., 2001), thereby providing principally subtractive and divisive inhibition in turn (Wilson et al., 2012). Moreover, the activity of these two interneuron types, along with being each dissimilar to the excitatory pyramidal cell population, has been shown to differ during seizures in both in vitro and in vivo mouse models of epilepsy (Miri et al., 2018; Parrish et al., 2019). Testing these findings in humans, therefore, requires estimations of the cell types that are being recorded.
Putative subclassification of “fast-spiking” interneurons versus the “regular-spiking” population of cell types has been routine in animal work for some decades, with “thin spikes” being noted by Mountcastle et al. (1969), and later pinpointed as originating from a subset of GABAergic neurons (McCormick et al., 1985). These fast-spiking cells in murine recordings were ascertained to be primarily from parvalbumin-containing (PV+) interneurons, since their fast rebound was abolished by tetra-ethyl ammonium (TEA) but resistant to blockade of Kv1 and Ca2+-mediated channels (Erisir et al., 1999). As a result, putative PV+ interneurons have been subclassified in single-unit recordings by their spike durations routinely in both animal and human recordings, including finding expected differences in population activity of the cell types (Csicsvari et al., 1998; Barthó et al., 2004; Peyrache et al., 2012; Elahian et al., 2018).
That said, a few potential limitations to this approach must be acknowledged. First, computational modeling suggests the exact extracellular spike duration is dependent on the cell body’s distance from the electrode, even after amplitude normalization (Gold et al., 2006; Buzsàki et al., 2012), and so there should be evidence of a bimodal distribution in spike durations in order to achieve robust cell-type subclassification. Second, not all units with brief action potentials can be assumed to be inhibitory: recording from M1 in macaques and antidromically stimulating pyramidal tract neurons, Vigneswaran et al. (2011) found confirmed pyramidal cells with spike durations within the “fast-spiking” range; however, this may be a situation unique to M1, with its large Betz cells.
As devices with higher densities of microelectrodes potentially become available for use in humans, it will be more likely to be able to determine cell types in a subset of instances by their downstream effects through searching for putative monosynaptic connections through cell-wise cross correlations. Currently, this is likely to be rare among a population of simultaneous single-unit recordings, with Peyrache et al. (2012) finding putative connections in only 0.17% of all pairs in an overnight human Utah array recording, and with no connections separated by more than 2 microelectrodes (~800 µm). Note, also, that to identify these putative connections reliably requires recordings of considerable duration, typically multiple hours (Schwindel et al., 2014; Peyrache and Destexhe, 2019).
A wealth of mechanisms has been posited for the commonly unpredictable ictogenesis in unprovoked seizures, as discussed throughout this volume. There is little evidence for any consistent, distinctive preictal activity pattern in EEG (Mormann et al., 2007), and indeed the mechanism of ictogenesis may differ from patient to patient. However, once ictal activity is underway, animal models have routinely displayed a sharp transition, involving a period of tonic firing progressing to burst discharges comprising paroxysmal depolarizations—the intracellular hallmark of epilepsy in animal models (Kandel and Spencer, 1961a; Matsumoto and Ajmone Marsan, 1964; Traub and Wong, 1982).
Early recordings during the ictal period in humans showed, in contrast to animal data, that only a subset of neurons in the onset zone showed an increase in firing during the seizure itself (Babb and Crandall, 1976), with only 36% of combined single and multi-units increasing in focal unaware seizures, and even fewer in subclinical seizures and during auras (7% and 14%, respectively; Babb et al., 1987). Similarly moderate neuronal firing rate changes have been reported since in recordings of spontaneous seizures in humans both in subsets of neocortical tissue (Truccolo et al., 2011; Merricks et al., 2015) and deep, mesial structures (Wyler et al., 1982; Bower et al., 2012; Lambrecq et al., 2017). These latter studies found, as one would expect, synchronous firing at the onset of a seizure (Wyler et al., 1982) and increased neuronal synchrony following ictal onset (Bower et al., 2012). The surprising degree of heterogeneity of neuronal firing within the seizure-onset zone, however, led Lambrecq et al. to propose that human “ictogenic mechanisms operate in submillimeter-scale microdomains” (Lambrecq et al., 2017).
Similarly, Bragin et al. (2000) put forth the hypothesis that seizures coalesce from multiple interconnected neuronal clusters spread across an extended region, rather than being monolithic, organized events. There is, however, an alternative explanation for observations of heterogeneous ictal neuronal activity: the dual territory hypothesis. A long-standing observation in animal models of seizures is that epileptiform activity is surrounded by a region dominated by feedforward inhibition, restraining the propagation of the focal pathophysiology (Prince and Wilder, 1967; Dichter and Spencer, 1969; Wong and Prince, 1990; Schwartz and Bonhoeffer, 2001; Timofeev et al., 2002; Timofeev and Steriade, 2004). This inhibition is thought to be mediated by PV+ interneurons, which, by virtue of synapsing onto the somata and axon initial segments of the local pyramidal cells, are capable of vetoing the simultaneous, strong glutamatergic drive (Pouille and Scanziani, 2004; Cammarota et al., 2013), until a potential moment of failure. In studies of murine slices in vitro, pyramidal cells ahead of the propagating ictal activity have been shown to receive intense levels of inhibition, which appears to suppress the march of the ictal activity for some duration (Trevelyan et al., 2006, 2007). More recently, evidence for this mechanism of inhibitory restraint has been found also in humans using Utah array recordings (Schevon et al., 2012), demonstrating the classic tonic neuronal firing pattern followed by highly synchronous bursting across the population in some seizures, and in others the heterogeneous firing patterns corresponding to the “penumbral” region dominated by feedforward inhibition, albeit with large increases in firing rates in 10%–20% of neurons, in keeping with earlier human observations (Merricks et al., 2015; Smith et al., 2016; Tryba et al., 2019). These two patterns were subsequently identified simultaneously in a single patient (Merricks et al., 2021). Differentiating between the two regions without benefit of multi-unit or single-unit recordings appears nontrivial, since as discussed earlier, the lower frequencies in the range of visually interpreted EEG are dominated by the synaptic “input,” instead of the local neuronal firing “output” (Schevon et al., 2012).
An alternate hypothesis is that these contrasting ictal activity patterns represent two different types of seizures, rather than two coexisting, spatially defined states within each seizure. This would imply that animal models are faithful at capturing one seizure type but not another that arises in spontaneous human seizures: one defined by heterogeneous firing with no concomitant hypersynchrony (Truccolo et al., 2011; Bower et al., 2012). Indeed, single-unit firing patterns at the start of human seizures are dependent on the initial EEG pattern, with low voltage fast (LVF) and hypersynchronous onsets displaying different cell-type-specific firing patterns (Avoli et al., 2016; Weiss et al., 2019), as also predicted by computational modeling (Wang et al., 2017). Yet these, too, were posited to coexist in the same seizure, with hypersynchronous firing at the microelectrode level preceding the appearance of LVF at the macroelectrode (Weiss et al., 2016, 2019). While an exhaustive comparison of all seizure types/brain regions and the underlying neuronal firing has not been undertaken as of yet, in vitro rodent recordings show that LVF and hypersynchronous onsets involve increased inhibitory firing and increased excitatory activity, respectively, during the initiation of pharmacologically induced seizure-like events (Avoli et al., 2016).
Single neurons can be followed via action potential waveforms during the brief time windows required for most research (e.g., ~2 hours in Truccolo et al., 2011). There is also evidence that reliable tracking is possible for extended time periods when accounting for slow shifts in wave shapes, from over 12 hours (Niediek et al., 2016) up to at least 40 hours and across multiple seizures in one example (Merricks et al., 2015). In contrast, traditional spike sorting approaches have been shown to fail when applied to seizing brain once the local tissue has been “recruited” to the seizure, with recovery of spike sorting ability in the immediate postictal period (Merricks et al., 2015, 2021). This is an important concept when analyzing single-unit recordings of seizures, and it is worth addressing in some detail.
During clinically evident seizures, action potential waveforms may be obscured by an increase in the amplitude of background noise, due either to movement-related artifact or to the increase in population activity that occurs in the context of seizure invasion of the sampled brain sites. There is, however, a physiological explanation that is often underappreciated. It has been shown in animal and computational models that the paroxysmal depolarizing shift (PDS) causes action potential wave shapes to be reduced in amplitude and increased in duration (Kandel and Spencer, 1961b; Matsumoto and Ajmone Marsan, 1964; Traub and Wong, 1982). Taken together, these changes may lead to an erroneous impression of reduced firing rates or even cessation of firing of some neurons, and emergence of a “new” neuron firing during the seizure. To overcome these challenges, novel neuronal tracking techniques have been developed to increase reliability in the assessment of single units during seizures (Merricks et al., 2021) and across time periods of days (Niediek et al., 2016). Briefly, to automate tracking of the same units across many hours, Niediek et al. split the recording into blocks of time, with the duration set by the user, apply automatic clustering across all blocks in parallel, and then use template matching of the mean waveform of each unit to “stitch” together likely matches across blocks. Meanwhile, due to the instability of defined clusters during the seizure, Merricks et al. calculate convex hulls in principal component space around well-isolated single units from the peri-ictal time period, which are used to track spikes that are likely to have arisen from those units in a probabilistic manner.
Relating this single-neuron activity back to simultaneously recorded EEG or other physiological signals is critical for any investigations seeking to link cellular activity with clinically accessible data. Neuronal firing has been shown to correlate with high-frequency signals both under physiological conditions (Buzsàki et al., 2012) and during ictal discharges (Weiss et al., 2013; Smith et al., 2016; Liu and Parvizi, 2019). High-frequency oscillations (HFOs) have been posited as a potential biomarker for the seizure-onset zone (see Chapter 13 in this volume; Höller et al., 2015), and so the neuronal activity underlying these HFOs is of particular interest.
In limbic structures, Weiss et al. (2016) found that single-unit firing rates increased during incidents of increased power in the fast ripple band of HFOs (200–600 Hz) in specific subsets of mesial temporal lobe seizures. Similar findings were later reported in neocortical recordings, though with specific reference to whether the region had already been recruited to the seizure (after the passage of an ictal “wavefront” consisting of tonic neuronal firing) or while still in the penumbral region dominated by feedforward inhibition, finding that the two time points were defined by quantitatively dissimilar coupling of MUA to HFOs pre- and post-recruitment (Smith et al., 2020). This was hypothesized to derive from a “true” oscillation of firing existing only prior to recruitment as a result of intact inhibition, though was itself only indirect evidence of such cell-type-specific firing, which is explored in the next section.
Relating these neocortical findings both to seizure types and to mesial structures is complex, and, on their own, these do not discount the possibility of seizures arising from a pathological network of heterogeneous firing in other instances. Ultimately, to differentiate between the competing hypotheses of neuronal firing underlying spontaneous human seizures would require larger spatial coverage with microelectrodes in patients than is currently feasible. However, the co-occurrence of both synchronized and heterogeneous firing at different locations during a single seizure, while not definitive, suggests that these are coexistent regions in human seizures rather than distinct seizure types, especially given a single location’s fragility: unrecruited in one seizure; recruited 5.5 hours later (Fig. 16–3). To elucidate the different mechanisms ongoing in each of these regions, we must examine the cell-type-specific firing local to each. At its most basic level, for example, does inhibition dominate in one of these territories, but not the other?
Spatiotemporal ictal recruitment of single neurons during spontaneous human seizures. Two seizures (A and B respectively) 5.5 hours apart captured on a Utah array in the frontal lobe from a single patient. (i) Overview of spatiotemporal firing patterns (more...)
To treat all neuronal firing as equivalent during seizures would be naïve, with different types of inhibitory firing likely playing a significant and distinct role in ictal onset, propagation, and termination to that of excitatory activity. Animal models have suggested a discrepancy in the timing of firing increases by cell type, with inhibitory firing rates increasing prior to ictal onset, and preceding similar increases in pyramidal cell activity, in pharmacologically induced seizures in an ex vivo rodent model (de Curtis and Avoli, 2016) and during in vivo recordings with microelectrodes in rats (Grasse et al., 2013; Karunakaran et al., 2016). In a mouse pilocarpine model, Miri et al. (2018) further showed, optogenetically, that this preictal increase occurs in both PV+ and somatostatin-containing (SST+) interneurons, with PV+ cells alone undergoing a late increase immediately prior to the seizure. It is difficult to untangle whether this may represent a causative role of interneuronal firing in ictogenesis regardless of the methodological model, but similar increases in putative PV+ interneurons have been described in hippocampal recordings during spontaneous human seizures using Behnke-Fried microwires (Elahian et al., 2018) and in neocortical recordings with Utah arrays (Ahmed et al., 2014; Tryba et al., 2019).
The cellular mechanisms that govern the transition to seizure remain an open area of investigation, with active debate on key issues. Some groups have proposed that the increased inhibitory firing is part and parcel of the ictogenic mechanism, itself driving the seizure by increasing extracellular K+ concentration, sculpting the timing of pyramidal cells into highly synchronous bursts, and initiating the seizure via a “rebound” effect (Gnatkovsky et al., 2008; Avoli and de Curtis, 2011; Shiri et al., 2015; de Curtis and Avoli, 2016; also see Chapters 6, 8, and 9, this volume). Indeed, there is evidence in murine cortical slices that activation of local PV+ interneurons can both lower the threshold for ictogenesis and prolong ictal duration (Sessolo et al., 2015), and that optogenetically inhibiting PV+ and SST+ cells in turn reduces ictal duration (Khoshkhoo et al., 2017). Others have posited failure of this interneuronal firing as the key event via a depolarization block mechanism (Ahmed et al., 2014). Still others have described continued interneuron activity but with altered postsynaptic responses to GABA as a result of accumulation of intracellular chloride ions (Huberfeld et al., 2007; Pallud et al., 2014; Alfonsa et al., 2015). In an isolated-subiculum mouse model undergoing pharmacological blockade of KCC2 (a co-transporter that extrudes chloride ions from the neuron and has been found to be dysfunctional in temporal lobe epilepsy patients [Huberfeld et al., 2007]) but otherwise unperturbed, optogenetic activation of PV+ interneurons is sufficient to induce hypersynchronous pyramidal cell firing (Anstötz et al., 2021; see Chapter 6, this volume). In extreme cases chloride loading can lead to a reversal of EGABA, resulting in paradoxically depolarizing effects of GABA in the postsynaptic neuron (Dzhala et al., 2010; Burman et al., 2019).
In the case of seizure spread once ictal activity is underway, intense excitatory activity from the distant seizure triggers inhibitory postsynaptic potentials in surrounding tissue (Powell and Mountcastle, 1959; Schwartz and Bonhoeffer, 2001; Trevelyan et al., 2013). As a result, there is intense interneuronal firing immediately prior to local recruitment, but after the earliest ictal activity has already begun (Tryba et al., 2019). Of course, it is not unlikely that even if interneurons do not initiate some or all seizures, later chloride-loading—or other mechanism(s) of weakened inhibition—will have an impact on inhibitory efficacy once the seizure is underway, as evidenced by the fact that PV+ activation distant to a seizure focus in a mouse cortical slice model helped restrain the pathology, while activation within the already active focus promoted discharges (Sessolo et al., 2015). Moreover, in an in vivo mouse model, while optogenetic stimulation of PV+ interneurons in the preictal period showed the expected anti-ictal effect, a few seconds after seizure onset (>2 s) this shifted to a proictal response (Magloire et al., 2019). This shift was mitigated by overexpression of the chloride-cotransporter KCC2 in the postsynaptic pyramidal cells, insinuating chloride-loading as one mechanism of continued ictal activity once the seizure is underway.
Tying together these observations from many modalities and experimental setups in the context of spontaneous seizures in humans is an ongoing area of investigation. Human single-unit recordings can provide valuable confirmation of hypotheses arising from mechanistic animal model studies. To that end, when it comes to interpreting these single-unit recordings by cell-type-specific activity, it is important to consider our field of view. Microelectrode access in patients is spatially limited and is likely to miss a true focal onset, thus initially seeing a downstream response to ictal activity elsewhere that was not captured on the recording. The region of seizure onset is likely to be highly spatially confined, limited to microdomains on the order of millimeters and possibly submillimeter in extent (Goldensohn, 1975; Schevon et al., 2008; Stead et al., 2010; Paz and Huguenard, 2015).
For example, the view that interneuronal firing may by itself “jump-start” seizures (de Curtis and Avoli, 2016) is in part based on observation of interneuronal activity recorded at the putative seizure onset site. But what if the seizure originated at a nearby, unrecorded location? By way of analogy, imagine an observer with a view of the central building in a row of town houses, and with the adjacent homes obscured by other structures. After a while, this observer sees firefighters rush into view and activate their hoses, spraying water toward the next-door home, itself out of view. Moments later, the building in view catches alight, while the firefighters continue their efforts. The observer, with prior knowledge, naturally realizes the fire spread from out of view, and the firefighters’ arrival was an initial attempt to stop the spread. But without said prior knowledge and no clear view of the adjacent buildings, the observer instead may have interpreted the arrival of firefighters, and their spraying water, as a pre-fire activity, perhaps even being causative of the subsequent blaze. That is, of course, not to say that interneurons cannot be an important driving force in the ictogenesis process—for example, what if the firefighters were unwittingly spraying a flammable liquid, as might be the case if GABAergic efficacy is altered—but that any evidence that they are must be considered carefully within this context of not knowing what is occurring “out of view” of our electrodes.
Examining cell-type-specific activity in human recordings has to date been limited to assessment of PV+ interneurons versus all other neuronal cell types. While this population is of particular interest given their role in surround inhibition, animal studies have highlighted a complex interplay between interneuron subclasses during ictogenesis, including PV+, SST+, and vasoactive intestinal peptide-containing (VIP+) cells (Sloviter, 1987). In an in vitro model utilizing cell-type-specific Ca2+ imaging paired to electrophysiological recordings, both PV+ and SST+ interneurons were intensely activated by, and entrained to, an oncoming glutamatergic drive, with PV+ surpassing and preceding SST+ activation, and depolarization block occurring in a subset of PV+ but not SST+ interneurons (Parrish et al., 2019). In chemoconvulsant-induced seizures in an in vivo mouse model, optogenetic stimulation of hippocampal PV+ and SST+ cells—independently to one another—showed that while PV+ had multiple stages of firing rate changes during early ictal stages unlike SST+, only SST+ displayed evidence of an altered input-output relationship (Miri et al., 2018). Lastly, inhibition among interneurons is itself complex, with VIP+ cells primarily synapsing onto SST+ and, to a lesser degree, PV+ interneurons, making them highly plausible culprits in reducing inhibition onto local pyramidal cells (Karnani et al., 2014).
As a result, analyses of firing patterns of one type of interneuron without concurrent recording of the others can make it difficult to disentangle the underlying mechanism; for example, does a reduction in PV+ interneuron firing occur because of depolarization block, reduced excitatory input, or active inhibition from other interneuron classes? While tracking the trajectory of single-unit action potential amplitudes through time—with reference to the detection threshold—can help differentiate depolarization block from the latter mechanisms, this ultimately may prove a limitation of human single-unit recordings. As discussed in the previous section, current methods of subclassifying extracellularly recorded neurons into cell types primarily relies on features of PV+ action potentials that are (predominantly) dissimilar to the rest of the population, and while recent approaches have begun to include cell-intrinsic firing to putatively isolate regular-spiking interneurons from pyramidal cells, differentiating between SST+ and VIP+ based on electrophysiological recordings does not appear promising (Burkhalter, 2008; Peyrache and Destexhe, 2019), and in fact even morphological subclassification appears inconsistent among groups (DeFelipe et al., 2013).
Although the relationship of interictal findings to seizures is much more difficult to interpret, human single-unit analyses during this time period have the advantage that available data are far more common, compared to the relatively rare instances of captured seizures. To date, however, research into epilepsy utilizing human single-unit recordings during the interictal period has been limited, likely due to uncertainties and corresponding lack of basic science studies regarding the mechanisms and spatial relationship of interictal activity to ictogenesis. Nevertheless, recent studies have provided insight into the temporal and spatial neuronal underpinnings of interictal epileptiform discharges (IEDs; Smith et al., 2022; Diamond et al., 2023), of high-frequency oscillations (HFOs) that are under active investigation as an epileptic biomarker, and of ongoing fluctuations in firing patterns independent to these events.
Recordings with microwires in the mesial temporal lobe from 11 patients found that cell “burstiness” (as measured by the mean autocorrelation lag) across 459 single units was higher during the interictal period in excitatory cells in the ipsilateral hemisphere to the seizure onset than the contralateral (Gast et al., 2016). This burstiness in ipsilateral principal cells was hypothesized by the authors to be a protective mechanism—ensuring high-frequency activity was not directly propagated through subsequent synapses—because there was no equivalent significant difference between hemispheres in the preictal period. However, comparing burstiness between interictal and preictal epochs of the same population (rather than the aforementioned comparing of p-values from each epoch) showed no difference, and so evidence for seizures arising from a reduction in this burst firing was inconclusive.
An alternative hypothesis is that the higher burstiness in ipsilateral versus contralateral excitatory cells results from ongoing IEDs or HFOs, both from their underlying neuronal mechanisms and their impact on local firing (Jiruska et al., 2017). Equivalent recordings from mesial temporal lobe and neocortex, with specific reference to IEDs, have found around 30% of units increased firing during IEDs, with a post-IED suppression of firing in 40%–50% consistent with the afterhyperpolarization from PDS (Keller et al., 2010; Alvarado-Rojas et al., 2013). Units that were modulated by IEDs displayed significantly higher burstiness (Keller et al., 2010). Notably, the same studies found that between 19% and 30% of neurons exhibited firing rate changes up to 300 milliseconds prior to the IED, and that in a subset of patients this corresponded to oscillations in the lower range of HFOs (Alvarado-Rojas et al., 2013). Both the timing and frequencies observed here partially overlap with interneuronal increases in the lead up to the peak in power of physiological ripples in human hippocampal recordings, prior to the maximal firing of the pyramidal cells (Le Van Quyen et al., 2008).
Utilizing polymer microelectrode arrays to record intraoperatively from the cortical surface, Yang et al. (2021) analyzed MUA during IEDs and HFOs in superficial neocortex, including over eloquent cortex, from 30 patients. While these recordings were necessarily brief and performed under anesthesia, raising the possibility of anesthesia-induced burst discharges confounding results (Liou et al., 2019), IEDs and HFOs were detected in a majority of patients, with similar temporal findings to the aforementioned mesial temporal recordings: HFOs occurring significantly frequently in the moments prior to IEDs. In contrast, the MUA was found to be significantly aligned to HFOs, but no significant covariance was found to IEDs. Though again, this is not evidence that MUA does not covary with IEDs—absence of evidence is not evidence of absence. That said, a study assessing the impact of IEDs on single-unit activity and its correlation to memory performance (discussed further in the next section) found similarly low rates, with only 6.9% of 728 neurons being significantly modulated by hippocampal IEDs, though with putative interneurons showing significantly greater alteration in the form of increased firing rate (Reed et al., 2020).
One interpretation of these findings might be that the varied responses to IEDs is suggestive of multiple irritative regions with limited local firing that nonetheless coalesces into network disruption, thereby contributing to seizure susceptibility. However, analysis of tetrode recordings of single units from both neocortical and mesial structures with reference to both local and distributed HFOs—all of which were associated with an IED—has revealed that the local neuronal response to the HFO is increased firing while there is a distributed inhibitory response (Curot et al., 2021). These data in combination with the prior observations of inhibitory-specific increased firing and varied population responses suggest that the dual-territory structure previously described for seizures also applies to interictal discharges, consistent with early in vitro studies of surround inhibition (Prince and Wilder, 1967). There is some evidence that the excitatory core of interictal discharges could be submillimeter in extent (Schevon et al., 2008; Stead et al., 2010; Yang et al., 2021). Indeed, assessment of the single-neuron activity underlying IEDs with versus without concomitant HFOs has revealed dissimilar firing patterns: IEDs with HFOs showed significant firing rate increases in many neurons during discharge onset with almost no decreases in firing, while IEDs without HFOs were more varied (Guth et al., 2021). Concurrently, in in vitro recordings of resected human tissue, depolarizing effects of GABA have been found during IEDs in the subiculum (Cohen et al., 2002) while the hyperpolarizing nature of GABAergic activity was maintained in equivalent neocortical slices (Köhling et al., 1998), along with evidence for nonsynaptic transmission of HFOs (Roopun et al., 2010). This is indicative that differing mechanisms coexist, either temporally or spatially, during IEDs and more research in this nascent area is necessary to disentangle the interplay between them, along with their pathological nature and potential for localizing ability.
Neuroscientific insights from single-unit recordings in epilepsy patients are not limited to studies of pathology. Such patients present a rare and exceedingly valuable opportunity to record from human single neurons during cognitive tasks or stages of relatively unperturbed sleep. Of course, these studies are not irrelevant to epilepsy, as cognitive deficits are a common comorbidity (Elger et al., 2004) and brain states such as sleep stages can play a significant role in seizure generation and management (Ewell et al., 2015). Moreover, it must be kept in mind that all such studies are necessarily from potentially pathological tissue, and that most epilepsy patients have neuropsychiatric deficits related to their condition.
The opportunity to probe cognition in this patient cohort was utilized as early as the 1970s (Halgren et al., 1978), and since then it has expanded manifold to include studies of many areas that are largely out of reach in animal studies, such as language (Ojemann et al., 1988; Creutzfeldt et al., 1989), music perception (Creutzfeldt and Ojemann, 1989), recognition of concepts (e.g., studies linking hippocampal neuronal firing to recall of well-known faces or images from popular culture; Quian Quiroga et al., 2005, 2009), food preferences (Mormann et al., 2019), and the substrates of consciousness (Navajas et al., 2014; Wenzel et al., 2019). Furthermore, the access to single neurons in these patients has been beneficial in not just confirming that groundbreaking observations from animal studies have homologous mechanisms in human processing (e.g., place, direction, and grid cells; Ekstrom et al., 2003; Jacobs et al., 2010, 2013) but also enables an exploration one step further, finding neuronal firing pattern is repeated during recall of previous navigation (Miller et al., 2013). Similarly, a subset of neurons’ firing encodes the location of remote targets rather than one’s own position (Tsitsiklis et al., 2020). These results build, in part, on findings from functional MRI studies (Bellmund et al., 2016; Horner et al., 2016), highlighting the value of the unique spatiotemporal resolution in these electrophysiological recordings to validate and expand on discoveries from other modalities. Human single-unit recordings even enable transitioning these findings into the realm of theoretical, rather than physical, space: firing patterns that encode “position” within goal states (Qasim et al., 2021) or in time during a memory task (Reddy et al., 2021).
Neocortical single-unit recordings in epilepsy patients, meanwhile, have elucidated the neuronal mechanism of memory retrieval, with the temporal order of spiking that encoded the memory reoccurring during recall (Vaz et al., 2020). In addition, the interplay between single-neuron firing in the entorhinal cortex and hippocampus has been shown to underlie context-specific recollection, with the hippocampal activity coordinating the retrieval of the memory from cortex (Staresina et al., 2019). Working memory, too, has been shown to be encoded in both the firing rate (Kamiński et al., 2017) and phase-relationship to low-frequency oscillations of single units in humans (Kamiński et al., 2020). These findings are nuanced, with neurons in specific regions being more likely to be associated with specific roles, for example, item retrieval during memory in parahippocampal cortex (Derner et al., 2020) and encoding methods in units differing across brain regions, each playing distinct roles in memory and decision making (Minxha et al., 2020). Indeed, neuronal firing in prefrontal cortex plays a specific role in decision-making in the presence of competing or conflicting options, with spike-phase coupling in the dorsal anterior cingulate cortex (dACC) being modulated in the presence of conflict during decision-making, in turn altering activity in the dorsolateral prefrontal cortex (Smith et al., 2019). Meanwhile, the reduction in cognitive performance after sleep deprivation has been shown to involve altered single neuronal firing in the mesial temporal lobe in recordings from epilepsy patients (Nir et al., 2017). Relatedly, there is evidence that learning involves the replay of neural firing sequences during subsequent rest, from similar recordings in patients with chronic implants as part of a brain-computer interface clinical trial (Eichenlaub et al., 2020).
Turning from studies of normal cognition, the impact of epileptiform pathology and transient pathophysiological activity such as IEDs and pathological HFOs deserves to be explored, given the well-known cognitive comorbidities suffered by epilepsy patients. Indeed, cognitive impairment as a result of interictal events is well-documented (Holmes and Lenck-Santini, 2006) and interictal spikes disrupt cognition in in vivo animal models (Khan et al., 2010; Kleen et al., 2010). Hippocampal IEDs in humans impact the ability to retrieve declarative memories, and the degree of firing rate increase in a subset of single neurons during IEDs—particularly putative FS interneurons—correlates with participants’ uncertainty in their memory (Reed et al., 2020). Of note, these IEDs and associated single-unit firing did not significantly alter memory formation, as opposed to retrieval, and only IEDs at least 2 seconds prior to stimulus onset interfered with memory retrieval.
Combined with the fact that different cognitive processes appear to be disrupted to different degrees within the seizure onset zone (memory-selective vs. visually selective; Lee et al., 2021), it is likely that differing cognitive task modalities and recording locations will be unequally impacted by interictal events. For example, there is evidence for decreased memory retrieval function in the right hippocampus (Reed et al., 2020; Lee et al., 2021). Of course, these findings are comparing cognitive performance either during the presence or absence of IEDs, or within or outside the focal onset. While cognitive deficits may be at their worst in these locations and periods, one cannot assume that outside these regions and times is equivalent to control, “healthy” tissue.
Many neuroscientific insights have been gleaned to date from human single-neuron recordings in epilepsy patients, related to both epilepsy-specific questions and general neurocognitive investigations. An obvious question, however, is whether there is, or may be, any direct clinical use of these recordings. Despite the single neuron’s quintessence in physiological processing and its role underlying many disorders of the brain, it has long been omitted from the world of the practicing neurologist (Cash and Hochberg, 2015). This, of course, is partially due to inaccessibility. Since microelectrode recordings are now available and the possibility of capturing single-neuron activity is within clinical reach, it may now be time to explore this question (see, for review, Chari et al., 2020). At the same time, the question of the ethics of adding invasive electrodes—however small—to an invasive EEG study, without any direct benefit to the patient, remains a concern. It is possible that the era of human microelectrode recordings may be a relatively brief one, focusing on research aimed at contextualizing data available from less invasive recordings. Once we learn to meet clinical goals such as pinpointing the true ictogenic focus and to define a therapeutic plan, surgical or otherwise, that maximizes probability of seizure freedom—while minimizing damage to healthy cortex—from less invasive tools such as EEG, MEG, or fMRI, it may be argued that it will then be wholly unethical to continue microelectrode implantations.
As such, while these recordings continue, it is important to maximize the benefit that can be achieved through them. This may be more complex than is immediately apparent: they are necessarily rare, and it is hard to control for variability in these data since equivalent control recordings in humans are, with limited exception, unavailable. Similarly, in contrast to animal studies in which environment can be controlled and breeding allows for homogeneous populations, individual patient recordings that are similar but not necessarily identical must be pooled to gain scientific insight. However, while etiologies and ictogenic processes may be varied, mechanisms of ictal propagation and termination are likely more similar across patients (Kramer et al., 2012), especially given the limited repertoire of seizure-onset EEG patterns (Perucca et al., 2014; Salami et al., 2015). As a result, single-neuron research across a seemingly disparate cohort of patients can provide considerable insights and benefits into specific seizure types and their spread. Nonetheless, this highlights a necessity to pool datasets between the relatively limited research centers undertaking these recordings, to best reach consensus on such mechanisms.
To the same point, it is statistically unlikely in the case of focal onset seizures that microelectrodes of any type will be implanted at the precise location of seizure origin, instead only ever capturing spreading ictal activity. However, for the same reason, this may be counterintuitively an advantage: many potential mechanisms have been posited for ictogenesis, making it difficult to draw meaningful conclusions from a given dataset with mixed etiologies and ictogenesis pathways. Conversely, the mechanisms of seizure propagation are likely to be far less variable. Recordings from regions of early ictal spread will provide key insight into mechanisms of propagation, which could evolve into therapeutic tools to contain ictal spread in a large cohort of focal onset seizures regardless of onset mechanism, thereby minimizing their clinical impact to the patient. To return to a firefighter analogy—there are many ways to start a fire: matches, magnifying lenses, flints, lighters, and so on. But once the fire is underway from a point source, it spreads in a similar way regardless, and so the optimum first lesson is how to contain that spread, perhaps more so than targeted techniques to stop diverse fire-starting tools.
Of course, microelectrode studies of single units have their limitations, human or otherwise. These recordings are inherently an under-sampling, recording from only a very small fraction of the total neurons in the region—typically tens to hundreds in regions of many thousands. Utilizing MUA simultaneously can ameliorate this to a certain degree, to get a handle on what the general population of cells are doing while simultaneously sampling a few individuals as single units. In fact, MUA and LFP alone may be enough in some instances: complex movement can be predicted accurately from MUA alone in premotor cortex recordings in macaques (Stark and Abeles, 2007). Since recruitment happens at the level of single neurons in spontaneous human seizures (Merricks et al., 2021), however, it is important not to discard spike sorting in favor of solely MUA analyses yet. Similarly, the majority of microelectrodes are incapable of recording from the considerably more abundant glial cells, which have been suggested to contribute to ictogenesis through astrocytic control of NMDA receptors (Tian et al., 2005; Clasadonte et al., 2013), and so the optimum therapeutic target may be upstream from the neurons recorded in these studies.
In the case of human-specific single-unit studies, it can prove difficult to capture seizures, requiring continuous recording over many days. A pivot to include more studies of single-unit firing during IEDs and the interictal period in general would bypass some of these issues, although the seizure itself will remain the gold standard. Another underutilized avenue in these recordings is analysis of single-unit firing in response to routine stimulation mapping done for clinical purposes: assessing aberrant firing by location with respect to stimulated region may provide insight into inherently pathological tissue, or potential routes of propagation during seizures.
Ultimately, the first single-neuron recordings in humans were driven by Dr. Ward’s concern that experimental models in animals bore resemblance to spontaneous human seizures, focusing specifically on electrophysiological similarities of the human neuron to those in these models (Ojemann, 2014). Some 60 years on from these groundbreaking recordings, a lot of insight has been gained, but similar concerns persist. Human single-neuron studies in epilepsy patients continue to be ideally suited to answering these questions—validating certain components of experimental models, perhaps invalidating others—and in doing so honing better treatments for patients with epilepsy.
The authors declare no relevant conflicts.
This is an open access publication, available online and distributed under the terms of a Creative Commons Attribution-Non Commercial-No Derivatives 4.0 International licence (CC BY-NC-ND 4.0), a copy of which is available at https://creativecommons.org/licenses/by-nc-nd/4.0/. Subject to this license, all rights are reserved.
Your browsing activity is empty.
Activity recording is turned off.
See more...
