Resolving the Micro-Macro Disconnect to Address Core Features of Seizure Networks

Summary

Current drug treatments for epilepsy attempt to broadly restrict excitability to mask a symptom, seizures, with little regard for the heterogeneous mechanisms that underlie disease manifestation across individuals. Here we discuss the need for a more complete view of epilepsy, outlining how key features at the cellular and microcircuit level can significantly impact disease mechanisms that are not captured by the most common methodology to study epilepsy, EEG. We highlight how major advances in neuroscience tool development now enable multi-scale investigation of fundamental questions to resolve the currently controversial understanding of seizure networks. These findings will provide essential insight into what has emerged as a disconnect between the different levels of investigation and identify new targets and treatment options.

In Brief

Farrell et al. discuss seizure mechanisms at the micro scale to highlight that such cellular-level information is necessary to explain epilepsy expression at the macro scale, where commonly used methodologies to locate seizure networks and guide interventions operate.

Introduction

Epilepsy affects 1 in 26 people in their lifetime (Hesdorffer et al., 2011a) and is characterized by the predisposition to generate seizures (Fisher et al., 2014). Despite continuous development of new antiseizure medications, seizures are inadequately controlled in 30-40% of cases, a situation that has seemingly not improved since the 1800’s (Löscher and Schmidt, 2011). This is troubling because epilepsy is associated with a 25 times greater risk of sudden death than the general population (Devinsky et al., 2016). Furthermore, epilepsy is often accompanied by cognitive, behavioral, and psychiatric comorbidities that reduce quality of life (Kanner, 2016). Seizure frequency plays a central role in the negative aspects of epilepsy and emphasizes the utmost importance for therapies that prevent seizures.

When possible, surgical intervention remains the best option for refractory epilepsy (Jette et al., 2014), though less invasive strategies should remain a goal (Krook-Magnuson and Soltesz, 2015). Successful surgical outcomes depend on proper characterization of the seizure-generating network. Accordingly, mapping and understanding seizure networks remains a major focus of contemporary epilepsy research at basic and clinical levels, but the current understanding is largely based on a macroscopic scale that is too far removed from the microscale cell-to-cell communication that generates and supports seizures.

It is sobering that “the rationale of temporal epilepsy surgery remains opaque or even contradictory” even in cases where there is a lesion or abnormality visible in brain imaging scans (Cleveland Clinic, 2017). This lack of confidence is exacerbated by the obscurity of linking electrographic recordings of seizures with the role of the local network in seizure activity. However, recent work highlighted in this review has begun to uncover the cellular ensembles that are responsible for the macroscopic electrographic properties of seizures and are refining our understanding of seizure networks. Here we discuss evidence that changes at the microscopic level, corresponding to the functional arrangement of specific cell types within microcircuits, are a critical source of differences in macroscopic epilepsy expression. Conversely, ostensibly similar epilepsy expression at the macroscopic scale can originate from a variety of mechanisms at the microscopic scale. Thus, a detailed understanding of seizure mechanisms at the microscale, encompassing cellular signaling and communication, is necessary to explain epilepsy expression at the macroscale, including seizure behavior, EEG patterns, and neuroimaging findings.

These microscale changes have a significant impact on the closely related, imperative outstanding question of how seizures generalize. In other words, what mechanisms underlie the ability of some seizures to spread beyond the brain region(s) of origin to cause behavioral convulsions (tonic-clonic activity)? Since generalized tonic-clonic seizures (GTCS) preceded all cases of witnessed sudden unexpected death in epilepsy (SUDEP) (Ryvlin et al., 2013) and frequency of GTCS is the greatest risk factor for SUDEP (Hesdorffer et al., 2011b), answering this question remains critical.

Finally, we highlight how modern neuroscience recording techniques that are genetically-targeted can be used to address basic science questions and validate clinical methodologies. Using these methods to resolve the underlying differences between macroscopically similar electrographic events is crucially informative to how seizures are conceptualized. Together, these insights will provide a more complete understanding of seizures to refine current therapies and develop novel treatments.

Heterogeneity Within Microcircuits: Possible Source of Individual Variation

While particular brain regions, such as the hippocampus, are often responsible for seizures, it is unclear how specific microcircuit elements, the non-uniform functional arrangement of cells within a brain region, are implicated across individuals with seizures originating from the same structure. Resolving these microscopic differences is critical to understanding and treating epilepsy (for insights from genetic epilepsies, see Box 1), especially when divergent epilepsies at the microscale are grossly classified as the same epilepsy (e.g. temporal lobe epilepsy; TLE) because they involve the same brain region. One prime example is the pyramidal cell population in the CA1 region of the hippocampus (Figure 1A). While CA1 has long been known to be the primary output node of the hippocampus, emerging data from the last decade has shown that there are robust differences among the principle cell population, which can be divided into deep or superficial groups along the radial axis of CA1 (reviewed by Soltesz and Losonczy, 2018). Differences between deep and superficial CA1 neurons have been described for gene expression, morphology, and the local and long-range connections coming to and from principal cells (Bannister and Larkman, 1995; Cembrowski et al., 2016; Jarsky et al., 2008; Lee et al., 2014). While the structure of CA1 cells are not classically thought to be arranged in a layered architecture as seen in cortex (except under genetic mutations that cause abnormal migration, see Jiang et al., 2016), these findings show that clear differences exist among the same cell types within a region. When also considering the division of pyramidal cell function along the dorsal-to-ventral and medial-to-lateral axis of the hippocampus (Igarashi et al., 2014; Cemborski et al., 2018), a macroscopic view of a cell layer is blind to these parallel pathways that are functionally distinct.

BOX 1. Genetic Epilepsies: Variability in Epilepsy Manifestation Across Scales Despite a Mutation Occurring in All Cells.

The advent of modern sequencing technologies such as whole-genome exome sequencing have increased the number of epilepsy patients undergoing genetic testing and added to the growing list of epilepsy-associated genes (Helbig et al., 2016). Despite such mutations being present in every cell at the macroscale, expression of mutations can vary across the scale of brain regions, cell types, and synapses to result in or modulate epilepsy manifestation, as illustrated by the prominent examples of epilepsy-related gene mutations below.

Brain Regions. Different epilepsy phenotypes are exhibited by knockout of either the Kv1.1 or 1.2 subunits, which belong to the same family of voltage gated potassium channels. While Kv1.1 knockouts display symptoms emblematic of limbic dysfunction, Kv1.2 knockouts display symptoms more associated with changes to brainstem and subcortical structures (Robbins and Tempel, 2012). It is thought that differences in the expression of these channels underlie the distinct epilepsy phenotypes, with Kv1.1 having greater expression in hippocampus than Kv1.2 (Prüss et al., 2010; Wang et al., 1994).

Cell Types. One of the strongest associations between a particular gene and epilepsy is Scn1a, encoding the sodium channel Nav1.1, and Dravet syndrome (Catterall et al., 2010; Oliva et al., 2012). About 70-90% of patients with Dravet have nonsense mutations in Scn1a, leading to a non-functional protein (Escayg and Goldin, 2010). Loss of function studies in mouse models have found a reduction in sodium currents specifically in GABAergic interneurons and not excitatory neurons early on, possibly due to higher expression of the channel in these cell types (Yu et al., 2006; Ogiwara et al., 2007).

Synapses. Mutations in genes for subunits of the GABA_A receptor (GABA_AR), a heteropentameric ion channel mediating the majority of inhibitory transmission in the brain (Sieghart, 2006; Simon et al., 2004), have been associated with a range of epilepsy syndromes (Oyrer et al., 2018). Receptors with particular subunit compositions can preferentially localize and function at specific synapses, showing how dysfunction in certain GABA_AR subunits can have circuit-specific effects. This is exemplified by the enrichment of α1 subunit containing GABA_ARs at synapses between parvalbumin (PV) positive perisomatic-targeting fast spiking interneurons and their granule cell targets in the dentate gyrus located close versus further away (Strüber et al., 2015). Similar synapse-specific targeting of GABA_ARs has also been found between connections from different interneuron types onto CA1 pyramidal cells, with synapses from PV-expressing inputs functionally relying more on β3 containing GABA_ARs compared to those from somatostatin (SST) expressing inputs (Nguyen and Nicoll, 2018).

These findings show how molecular components have a critical and specific impact on defining the circuit organization of the brain.

Figure 1. Differences in Underlying Microcircuit Activity Can Yield Similar Macroscopic Activity.

(A) Even within principal cell layers of a brain region, as depicted in the CA1 of hippocampus here, considerable heterogeneity exists in the organization and function of neuronal microcircuits (for review, see Soltesz and Losonczy, 2018). We represent two populations of CA1 pyramidal neurons (deep and superficial) that have divergent, inputs, outputs, and local inhibition (PV-expressing basket cell; PVBC).

(B) This biased microcircuit architecture brings about a heterogeneous network of activity during seizures both within and between patients (figure adapted from Muldoon et al., 2013). In the hypothetical example depicted here, while the EEG is macroscopically similar in both patients, the underlying firing pattern of distinct cell types (each row indicates the firing of a given cell, and different cell types are color coded as displayed in A) is different both within individual events and between different patients. Note that because of the divergent outputs, the resulting network disruption and behavioral manifestation will also be different between patients.

The segregation of function within hippocampal cell layers could have important implications for seizures and epilepsy (Figure 1). For example, it is unknown if seizures preferentially recruit deep or superficial cell layers. Given the divergent outputs of these cell types, this organization will undoubtedly influence where seizures spread and could vary within and between patients. Axonal sprouting between otherwise distinct cell groups could also disrupt the normally distinct parallel processing and contribute to excitability and interictal (between seizures) cognitive/behavioral comorbidities. Moreover, macroscopic cell loss of CA1 should be sub-characterized by the differential loss of deep versus superficial neurons (Towfighi et al., 2004), which could vary widely among patients and be a source of variation in epilepsy expression. The implications of heterogeneity within microcircuits extend to other brain areas where these characteristics are beginning to be understood, rather than being an exception of CA1 (Krook-Magnuson et al., 2012; Varga et al., 2010; Hunt et al., 2018). Since this connectivity bias is preserved in chronically epileptic animal models (Armstrong et al., 2016), determining how seizures and interictal abnormalities are expressed within heterogeneous microcircuits is a valid concern.

The implications of parallel pathways within the same microcircuit are significant in the context of the differences between observed epilepsy mechanisms at the macro versus microscale. A basic assumption of epileptiform activity is that it is due to recurrent and hypersynchronous runaway excitation across multiple scales. This tenet arose from looking at neuronal activity at a population level (i.e. EEG). Closer examination, however, highlights that neuronal activity during seizures and epileptiform events are highly heterogeneous. Neuronal activation patterns observed from in vitro slice models of interictal spikes display spatially clustered arrangements of groups of neurons that vary between individual spikes, despite electrographically similar events (Muldoon et al., 2013; Sabolek et al., 2012). Similar analysis based on single unit recordings has also been performed on patients undergoing epilepsy monitoring and revealed heterogenous spiking outside of the seizure focus within a given seizure but consistent spiking patterns between seizures (Truccolo et al., 2011). These findings highlight the inadequacy of population-level measures, such as EEG, and the need to characterize microscale changes to potentially explain differences in epilepsy phenotypes, manifestations, and possibly treatment options (Figure 1B).

Optogenetic Control of Epilepsy is Achieved by Both Local and Long-Range Manipulations

The advent of molecular tools for functional analysis of specific circuits has enabled researchers to test the contribution of specific microcircuit elements to macroscopic seizure expression, most commonly EEG recordings and behavioral presentation. Optogenetics enables researchers to use light to control the activity of a genetically-defined cell population with spatiotemporal precision (Zemelman et al., 2002; Li et al., 2005; Boyden et al, 2005) and is a powerful tool for controlling seizures (recent reviews by Choy et al., 2017; Bui et al., 2017). One application of optogenetics is to increase or decrease the activity of a cell type of interest in a closed-loop manner such that control only occurs when seizures are detected. Ideally, this method could be used to identify critical nodes in a seizure network and to test the involvement of specific aspects of the microcircuit to enable more targeted control (Figure 2).

Figure 2. Three Strategies for Disrupting Seizures.

Here, we outline three conceptually distinct strategies that have demonstrated effective seizure control with optogenetics. It is possible to either disrupt the generation and maintenance of focal seizures (either at the focus, grey cells, or remotely, blue cells) or prevent their transition to more severe clinical manifestations (green cells). Clinically, it should be noted that remote electrical control of focal seizures via anterior thalamic stimulation may be a promising treatment option (SANTE trial, see Salanova et al., 2015) and the cerebellar influence on cortical and hippocampal seizures (see main text) may also take place indirectly through the thalamus.

For example, it has long been hypothesized that the dentate gyrus acts as a gate for temporal lobe seizures (Heinemann et al., 1992; Lothman et al., 1992). Optogenetic control of dentate granule cells was used to evaluate this theory in chronic epilepsy, and indeed, demonstrated their critical role in seizure expression, since inhibition almost immediately stopped seizures and excitation exacerbated seizures (Krook-Magnuson et al., 2015). Optogenetic inhibition in this study was highly successful when delivered on the ipsilateral side of the kainate-induced chronic epileptic focus but had no effect contralaterally. Since less than 5% of granule cells were estimated to have been inhibited by light, this result supports a critical role for a relatively small population of neurons. Moreover, similar results were obtained by stimulating inhibitory dentate interneurons during evoked seizures in hippocampus, but not entorhinal cortex, despite ongoing seizures in both locations (Lu et al., 2016). Results from these models suggest that seizures depend on relatively restricted networks.

Surprisingly, however, optogenetic manipulations well outside of the focus also have therapeutic efficacy (Paz and Huguenard, 2015). Though not classically considered part of the temporal lobe network, optogenetic modulation of cerebellar Purkinje cells reduced spontaneous seizure durations in chronic TLE in mice (Krook-Magnuson et al., 2014). This study also demonstrated that seizures increased firing rates of a proportion of Purkinje cells, indicating involvement of the cerebellum in temporal lobe seizures. Optogenetic activation of the cerebellum controlled corticothalamic seizures as well, which involve distinct networks from temporal lobe seizures (Kros et al., 2015) (see also Paz et al., 2013 and Chang et al., 2017 for thalamic optogenetic control of cortical seizures). Thus, while focal seizures may arise from spatially restricted networks, remote areas nevertheless are often able to exert robust effects on the primary pathological oscillatory circuit (Figure 2, blue cells), most likely through major network nodes within the thalamus.

Control of Seizure Generalization and Cognitive Comorbidities Through a Single Cell Type

Another advantage of the precise control afforded by optogenetics is to investigate the cell types that govern seizure generalization. Electrographic seizures can be controlled using closed loop optogenetic interventions targeted near the focus through either inhibition of the excitatory pyramidal cells or excitation of the GABAergic PV expressing interneurons (Krook-Magnuson et al., 2013) (Figure 2, grey cells). An additional study investigated the contribution of distinct cell types in the subiculum, another major output of the hippocampus, in the transition from focal to generalized seizures in chronic epilepsy (Wang et al., 2017). Activation of local GABAergic neurons exacerbated the duration and frequency of generalized seizures, and this paradoxical worsening of seizures by GABAergic activation was supported by evidence of a depolarizing effect of GABA on local pyramidal neurons resulting from chloride dysregulation. However, when the authors investigated cell type specific inhibition, the effects on seizure generalization were divergent. PV-expressing interneurons under these conditions exacerbated seizure generalization whereas SST-expressing interneurons inhibited seizure generalization. These results highlight a key role for distinct cell types within microcircuits in controlling seizure generalization.

Recent studies in experimental mouse models of chronic TLE also revealed that it is possible to control seizure generalization to motor, clinical seizures, even if the intervention has no or only minor effect on the primary focus (Figure 2, green cells). The latter possibility is especially relevant in light of the challenges in the clinic in trying, and unfortunately too many times failing, to identify the actual seizure focus in order to neurosurgically remove it. Specifically, mossy cells of the dentate gyrus were demonstrated to play a crucial role in seizure generalization while exerting little to no effect on the electrographic seizures recorded from near the primary focus in the CA1 (Bui et al., 2018). Mossy cells are excitatory, glutamatergic cells that reside in the hilus, and their cardinal feature is that they are very richly and widely connected: their axons form the commissural-associational pathway of the dentate gyrus, innervating up to 30,000 granule cells each, in addition to interneurons, and mossy cells receive highly convergent strong excitatory inputs from the dentate granule cells. Mossy cells are upstream of the primary hippocampal output nodes (CA1 and subiculum) and their partial loss in chronic epilepsy has long been identified as a key pathological hallmark in both animal models and human patient samples (Blümcke et al., 2000). Using a closed-loop approach, it was shown that activation of the glutamatergic mossy cells suppressed seizure generalization and behavioural seizures, whereas inhibition promoted seizure generalization (Bui et al., 2018). These results suggest that the loss of mossy cells in chronic epilepsy plays a mechanistic role in disease progression and worsening of clinical seizures. But could their loss also be related to the cognitive comorbidities in epilepsy, given their emerging role in cognitive functions of the hippocampus (Senzai and Buzsáki, 2017; Danielson et al., 2017; GoodSmith et al., 2017)? Remarkably, impaired spatial memory could be recapitulated in non-epileptic mice by optogenetically suppressing mossy cells. These results highlight a central role for a single cell type in two core issues in TLE – seizure generalization and cognitive comorbidities.

While optogenetics is clearly a valuable tool to determine which cellular elements of neural networks are sufficient to supress or exacerbate seizures, necessity of a cell type is more difficult to unequivocally determine. One must consider the indirect consequences of silencing a specific cell type to the function of the microcircuit. In other words, are seizures disrupted because the intended cell population is inhibited or because a more critical synaptically connected cell type has had its activity altered indirectly? This uncertainty, in part, can be resolved by genetically-targeted recording technologies, which are covered in a later section. Monitoring the spatiotemporal involvement of specific cell types in seizures provides an additional level of confidence and could be used to generate hypotheses for future optogenetic studies. These tools can also be harnessed to clarify the currently controversial definition of a seizure network.

Do Focal Seizures Rely on Focal or Widespread Networks?

While evidence from genetic, microcircuit, and optogenetic studies highlight a potential role for focally-restricted seizure generation, an increasingly prevalent and seemingly alternative view of temporal lobe seizures is that they rely on widespread networks throughout the brain (Bernhardt et al., 2015; Kramer and Cash, 2013; Stam, 2014). Specifically, the pathophysiology of focal epilepsies, such as TLE, occur on a systemic scale and are not localized to the temporal lobe per se (Bernhardt et al., 2015). This paradigm-shifting idea is attractive because it offers an explanation for why surgical resection of temporal lobe structures can fail to control seizures in a sizable proportion of the cases. This widespread network model is, in part, supported by data from structural neuroimaging (e.g. cortical thickness and white matter tractography) and resting-state functional characterization, typically performed when patients are not having seizures (e.g. electroencephalography: EEG and functional MRI: fMRI). However, it is currently unclear if broad structural changes or functional changes assessed in the interictal period are truly representative of the seizure-generating network.

A widespread network model for focal TLE is potentially more directly supported by extensive electrographic seizure activity apparent on EEG obtained from scalp, subdural, and targeted depth recordings. Combining unique patient structural data and electrographic seizure recordings may even allow researchers to build a personalized model of the epileptic network (Jirsa et al., 2017). Moreover, researchers could perform virtual tissue resections to test hypotheses before irreversibly ablating tissue clinically (Khambhati et al., 2016). Considering the heterogeneity within epilepsy, it is easy to envision the potential benefits of this personalized approach.

While electrographic seizures impact broad networks, the evidence to support that the entire network is “core” to the disease should be closely examined (critically reviewed by Smith and Schevon, 2016). It should be considered that surgery can fail to control seizures because intervention often lags many years behind diagnosis (Berg, 2004) and increases the likelihood of generating other foci (Janszky et al., 2005). Seizures arising from independent foci is a fundamentally different concept from the notion that seizures initiate from widely distributed networks. Furthermore, as we have highlighted with microcircuit dynamics, macroscale electrographic events provide at best incomplete, and at worst misleading insights on their underlying origins. We suggest that a basic understanding of the cellular contributions to seizure activity should guide definitions of seizure networks.

Conceptually Defining Excitatory vs. Electrographic Seizure Networks

The current definition of an epileptic seizure is “a transient occurrence of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity in the brain” (Fisher et al., 2005). Therefore, seizures are defined by the behavioral signs and symptoms they produce (e.g. loss of awareness, motor activity, etc.) (Fisher et al., 2017), which are a reflection of the networks that are engaged. As we will examine below, both local excitatory and inhibitory activity can result in electrographic discharges, but would be expected to have divergent effects on seizure behavior. It is therefore insufficient to characterize the involvement of a brain region in a seizure network based on electrographic activity alone.

Generally, cortical structures are the most common sources of epilepsy and recurrent glutamatergic neurotransmission is thought to be necessary to support seizures in these structures. Thus, recruitment of glutamatergic neurons is a fundamental substrate for seizures (Figure 3A,B), notwithstanding the potential involvement of depolarizing GABA responses in seizure events (see above). Beyond cortical structures, the excitatory seizure network should be extended to include neurons of brain regions whose neurotransmitter system(s) cause excitation of their synaptic targets (e.g. basal forebrain cholinergic neurons, midbrain dopamine neurons, excitatory neuropeptidereleasing neurons), as these neurons would also be fundamental to seizures. While widespread recruitment of inhibitory networks is also important to seizure semiology (Blumenfeld et al., 2004), these areas should be dissociated from the networks that drive and support seizures. Throughout the rest of the review, we refer to brain areas that drive and support seizure activity as belonging to the excitatory seizure network, which is a subdivision of the broader electrographic seizure network where electrographic seizures are observed. Finally, before addressing the mechanisms that drive this distinction, it should be noted that these definitions vary with respect to time: seizures are spatiotemporally dynamic, the effect of neurotransmitters can change with seizure-driven changes in ionic gradients, and changes in protein expression and synaptic connectivity can alter how cells communicate.

Figure 3. Conceptual Framework for Defining Seizure Networks: Excitatory vs. Electrographic.

A seizure onset zone is depicted with an adjacent brain region. Hypothetical intracellular traces from excitatory (depicted as glutamatergic: Glu) and inhibitory (depicted as GABAergic: GABA) neurons and EEG recordings are portrayed for each brain region. Example mechanisms illustrate cases where electrographic seizures are the result of local excitatory activity (A and B) or other mechanisms (C-D), the latter giving the impression of a larger seizure network. D and E highlight the role of inhibition in driving rhythmic IPSPs onto excitatory cells to yield large, synchronous LFP oscillations. Clarifying the local excitatory contribution to seizure activity provides a more precise way to describe seizure networks rather than electrographic activity alone.

Clarifying the Micro-Macro Distinction: How Electrographic Seizures Can Occur Independently of Local Excitatory Seizure Activity

As mentioned previously, electrographic seizures may not reflect the underlying activity of local excitatory neurons. One mechanism by which seizure activity can be recorded without recruitment of any local network component, excitatory or inhibitory, is through volume conduction (reviewed by Buzsáki et al., 2012) (Figure 3C). The extracellular space of the brain is composed of a continuous medium of water and salts and, not surprisingly, is conductive. In fact, scalp EEG primarily relies on volume conduction to record brain activity from outside of the brain. Moreover, EEG potentials in cats can be recorded over a resected hemisphere (Cobb and Sears, 1960), clearly highlighting a role for this phenomenon in generating EEG. As a rule, low frequency oscillations travel better than higher frequencies, which tend to be filtered out (Bazhenov et al., 2011). Indeed, both physiological and electrographic seizures travel effectively through the brain without lag and are a major component of the observed EEG (Jeffreys, 1995; Christodoulakis et al., 2013). This mechanism could also apply to depth electrodes in brain regions containing passing white matter tracts that are actively seizing without recruitment of local grey matter. However, it should be noted volume conduction is not necessarily an epiphenomenon, since volume-conducted changes in electrical potential can influence the physiology at the sites they travel to (termed ephaptic coupling) (Qui et al., 2015; Anastassiou and Koch, 2015). Thus, volume conduction can be both a confound for seizure localization and a mechanism for seizures to spread non-synaptically (Qui et al., 2015).

Secondly, electrographic seizures can be driven by distant brain regions through feed-forward inhibition, independent of local excitatory neurons at the recording site (Figure 3D). Feed-forward inhibition is a common feature of neural circuits where afferent glutamatergic activity drives local inhibitory neurons, which in turn, inhibit the local excitatory population (Buzsáki, 1984; Pouille and Scanziani, 2001). This network feature serves to drive behaviorally-relevant oscillatory field potentials at remote brain regions (Zemankovics et al., 2013; Isaacson and Scanziani, 2011; Varga et al., 2014). Thus, it is conceivable that anatomically remote seizures acting through feed-forward inhibition could drive the appearance of electrographic seizures on EEG recordings without recruiting ictal firing of local glutamatergic neurons. This is evident in cortex where seizure spread is thought to be often opposed by “surround” feed-forward inhibition, in which glutamatergic neurons of seizing cortex strongly recruit inhibitory neurotransmission and restrict glutamatergic-driven seizure spread (Prince and Wilder, 1967; Trevelyan et al., 2007). Using high-density micro-electrode arrays to simultaneously record local field potential (LFP) and neuronal firing activity in patients undergoing epilepsy monitoring, Schevon et al (2012) observed rapid spread of electrographic seizures across all electrodes, but more restricted local firing. Surprisingly, different electrodes could record indistinguishable LFP (or EEG) but drastically different local firing rates. Detailed follow-up work in a mouse model demonstrated that focally-generated seizures engage a widespread inhibitory network outside of the focus that is responsible for electrographic seizures in the absence of excitatory recruitment (Liou et al., 2018). These results reiterate the concepts introduced in the microcircuit heterogeneity section that the macro view provided by EEG is of limited utility for drawing conclusions about the underlying circuitry.

Lastly, the presence of long-range-projecting GABAergic neurons in seizure-generating brain regions could be a source of remotely-driven electrographic seizures (Figure 3E). Classically, the axonal arbors of interneurons were thought to occupy discrete regions within a given subdomain of a network, however, several exceptions to this rule have been recently characterized. In the dentate gyrus, for example, as many as 15-20% of GABAeric inputs originate from the CA regions of the hippocampus, highlighting prevalent non-local inhibitory networks (Szabo et al., 2017). Long-range axons of brain region spanning interneurons have been identified in the septum, hippocampus, entorhinal cortex, intracortical networks, and other brain areas (Tomioka et al., 2005; Jinno et al., 2007; Melzer et al., 2012; Basu et al., 2016) and represent an important and underappreciated conduit for brain regions to influence each other. Since these projections influence synchrony, plasticity, and the behavioral outputs of their projections (Melzer et al., 2012; Basu et al., 2016), future work should aim to characterize their role in epilepsy. Since long-range inhibitory neurons primarily target GABAergic neurons in temporal lobe networks and potentiate LFP oscillations (Melzer et al., 2012), these GABA-GABA synapses could drive electrographic seizure patterns or even disinhibit the local network and render it more sensitive to afferent excitatory seizure activity. Thus, using targeted tools to study the neural networks underlying seizures will clarify the currently opaque understanding of seizures.

A New Era for Basic Epilepsy Research: Genetically-Defined Cellular Recordings at Several Scales

At the basic science level, the development of cell type-targeted tools permit researchers to span the gap between micro and macro networks while retaining specificity. Adoption of these tools will dramatically enhance our understanding of seizures with major implications for future epilepsy research. As suggested earlier, genetically-defined recordings could be used to generate hypotheses for future optogenetic interventions by first identifying which neural populations are recruited. Covered below, these tools can also be used to validate the potential cell type-specificity of commonly used clinical neuroimaging methodologies by identifying the cellular sources of neuroimaging signals. Lastly, cell type-specific recordings can clarify the poorly understood boundaries of seizure networks. Here we highlight some of these tools and emphasize recent developments.

High density microelectrode arrays offer a way to separate and track the activity of neurons to generate valuable insights of the local spiking activity (Buzsáki, 2004; Figure 4A). A recent paper using this approach to record hippocampal and parietal cortical neurons outside of the seizure focus in chronic TLE models in rats showed that seizures predominantly engage presumed interneurons (fast-spiking units), which demonstrate conserved ictal spiking sequences that were coordinated with the LFP (Neumann et al., 2018). This is consistent with the idea that seizures can engage widespread inhibitory synaptic activity to influence LFP, but with limited recruitment of excitatory neurons, as discussed above. Though major recent advances have been made in microelectrode designs (Chen et al., 2017) and optogenetic “tagging” of units is possible for cell type specific recordings (Kvitsiani et al., 2013; Roux et al., 2014), a major problem with microelectrode recordings, particularly in the seizure focus, is that it is often rather challenging to separate overlapping spike waveforms during seizures (Merricks et al., 2015). Fortunately, other techniques described below are more suitable for this task.

Figure 4. Probing for Cell Type-Specific Neural Activity Across Multiple Scales.

(A) Microelectrode recordings are commonly used to track the spiking activity of local neurons that can be separated into fast-spiking (presumed inhibitory) and non-fast-spiking (presumed excitatory) units. These recordings are more suited for recordings outside of the seizure focus due to current limitations.

(B) Genetically-encoded calcium (GECIs) and voltage sensors (GEVIs) provide a readout of cell type specific neural activity and can be targeted to excitatory neurons, for example. A simple tool to measure activity of a given brain region is through an optical fiber which acts as a conduit to excite and collect emitted photons from an activity sensor (fiber photometry).

(C) Imaging calcium activity (also voltage with additional challenges) at the cellular level is possible in both head-fixed (e.g. 2-photon) and freely-behaving animals (miniscopes). The principal advantage of 2-photon microscopy is that illumination is restricted to a thin z-plane, which limits off-axis sampling and minimizes photobleaching. Miniscopes have greater out of focus sampling and photobleaching, but preserve a more complete behavioral repertoire by enabling free movement. Options for gaining optical access will depend on the brain region of interest. Here we demonstrate a few examples of how coverslips, cannulae, GRadient INdex refractive (GRIN) lenses, or prism probes can be utilized to access near-surface, immediately subcortical, deep, and orthogonal structures, respectively. See Supplemental Table 1 for software to extract signal from calcium movies.

(D) Wide-spread cortical activity can be measured with cell type specificity (e.g. VGLUT1-GECI transgenic mice for glutamatergic recordings), for example, using simple, low-cost macroscopes.

(E) Fiber photometry in (B) can be adapted to fiber-bundles allowing access from several discrete brain regions at depth.

Major advances in optical methods to probe for neural activity can be applied to address questions at several scales with varying spatiotemporal resolution. Genetically-encoded calcium sensors display increased fluorescence upon increased intracellular calcium, which provide a good readout of neural activity, since intracellular calcium transients (lasting ~100ms) are coupled to action potential firing (Helmchen et al., 1996; Broussard et al., 2014). Calcium sensors come in several varieties that allow flexibility depending on the research question (e.g. red-shifted to allow optogenetic interventions, fast decay, brighter signal, ratiometric etc.) (Zhao et al., 2011b; Akerboom et al., 2012; Chen et al., 2013; Thestrup et al., 2014). Since these tools are endogenously synthesized within the cell (i.e. genetically-encoded), genetic tools can be used to restrict expression to cell types of interest through transgenic mouse lines, viral induction, or a combination of the two (see Kim et al., 2017 for a review outlining these strategies). For example, expression can be limited to neurons that rely on promoter sequences specific to glutamatergic neurons (e.g. VGLUT1, a synaptic glutamate transporter) to determine local excitatory neuron activation during seizures. Calcium signal from these neurons can be assessed at the local population level (average activity) with a chronically implanted fiber optic probe (fiber photometry) at a single site (Gunaydin et al., 2014; Figure 4B) or up to 48 locations throughout the brain (Kim et al., 2016a; Sych et al., 2018; Figure 4E). Similarly, the average calcium activity can be visualized across the entire cortical surface with a wide-field macroscope to visualize seizure spread through the excitatory network (Rossi et al., 2017; Liou et al., 2018; Figure 4D). Comparing these data to concurrent LFP recordings allows researchers to determine if the local excitatory population is indeed part of the seizure network. Furthermore, these methods could examine the contribution of distinct cell types to the types of electrographic activity seen on clinical stereoEEG (stereotaxically positioned depth electrode recording of EEG signals) that are thought to underlie different parts of the seizure network (Grinenko et al., 2018).

As we have reiterated, understanding which neurons are recruited during seizures and when is crucial information. For instance, in the CA1, where we have stressed that the local pyramidal cells are actually a mixed population, it would be meaningful to resolve which subpopulation is more implicated in seizures, which is not possible with an averaged signal from CA1. Two-photon microscopy (Kaifosh et al., 2014) and head-mounted miniature epifluorescent microendoscopes (miniscopes) (Ghosh et al., 2011; Cai et al., 2016; Jacobs et al., 2018; see Resendez et al., 2016 for protocol) are commonly used for tracking calcium transients from individual neurons in head-fixed and freely-behaving mice, respectively (Figure 4C). Recordings at several scales are possible, from small deep brain nuclei using thin GRadient INdex refractive lenses up to imaging the entire cortical surface by replacing the dorsal cranium with custom glass (crystal skull technique; Kim et al., 2016b). Given the unique ensemble dynamics at the cellular level but homogeneous macroscopic EEG effects of interictal spikes (Muldoon et al., 2013; Sabolek et al., 2012), an outstanding question that could be addressed with these in vivo imaging techniques is whether these principles are conserved during seizures.

Finally, the development of fluorescent voltage sensors has been rapidly progressing and provide a glimpse into the future. Unlike calcium transients, which are inherently slow (Helmchen et al., 1996), voltage fluctuations are sub-millisecond, making it possible to record fast subthreshold changes and action potential firing (Lin and Schnitzer et al., 2016; Hochbaum et al., 2014; Gong et al., 2015; Piatkevish et al., 2018). Using a modified fiber photometry approach, a technique named “transmembrane electrical measurements performed optically” (TEMPO) enables LFP-like recordings in a cell type-specific manner (Marshall et al., 2016). A key difference between TEMPO and LFP is that TEMPO is a genetically-targeted intracellular recording, which is therefore impervious to uninfluential changes in extracellular field potential. Therefore, TEMPO or cell type-specific voltage imaging would be valuable tools to resolve the differences between conventional LFP and that of specific classes of neurons during seizures, but without the time-lag of calcium sensors for a better understanding of temporal kinetics. We have provided links and references to open source options of the tools mentioned in Supplemental Table 1.

Towards Utilizing Cell Type-Specificity in Cerebral Blood Flow Control to Delineate Seizure Networks

The rapid development of cell type-specific recording tools for basic science is exciting, but is unlikely to be adopted clinically in the near future due to their need for genetic access and invasiveness. However, perhaps nature has offered a way to probe for local excitatory neuron recruitment, deemed critical for seizures in the sections above, across the entire brain of people with epilepsy. This is possible because (1) seizures and epileptiform activity drive dynamic changes in blood flow and (2) control of cerebral blood flow is cell type-specific. Therefore, commonly used tools such as ictal single photon emission computed tomography (SPECT) and fMRI can be used to detect hotspots during or after seizures. These tools are commonly included in a pre-surgical investigation and are appreciated to be more localizing than EEG (LeVan et al., 2010; Duncan, 2010), but understanding the cellular basis for the more restricted activation patterns would be exceedingly informative of the underlying circuitry.

The cellular mechanisms that mediate functional hyperemia/neurovascular coupling (increased blood flow upon increased neural activation) have been extensively investigated in rodent physiology and involve specific neuronal cell types. Generally, glutamatergic cellular activity plays a crucial role in generating vasodilation and associated fMRI response (Lee et al., 2010; Mukamel et al., 2005) and blocking glutamate receptors prevents functional hyperemia in the cortex (Rungta et al., 2018). Moreover, a major molecular contributor to functional hyperemia, cyclooxygenase-2 (COX-2), is generally restricted to excitatory neurons (Niwa et al., 2000; Lacroix et al., 2015) and engaged during seizures (Yoshikawa et al., 2006; Serrano et al., 2011), though other mechanisms with less glutamatergic-specific expression also play a role (reviewed by Nippert et al., 2018; Iadecola, 2017). These results support that ictal hyperperfusion is an indication of local excitatory activity. Conversely, ictal hypoperfusion may occur at areas with excessive inhibition since inhibitory neurons can drive vasoconstriction (Devor et al., 2007; Uhlirova et al., 2016). Hypoperfusion is evident in the surrounding cortex in acute focal seizure models, highlighting a potential role for feed-forward inhibition in the inhibitory surround to reduce blood flow (Zhao et al., 2011a). Similarly, decreased ictal perfusion is observed well outside of seizure foci in humans (Blumenfeld et al., 2004). While these results provide insight into the cell type-specific mechanisms that drive changes in blood flow during seizures, combining neuroimaging with the cell type-specific recordings above could provide a more complete understanding of clinical neuroimaging signals and ultimately lead to more precise seizure network mapping.

Postictal vascular changes have recently been mechanistically characterized and represent a new opportunity to identify regions activated by seizures. Severe, local hypoperfusion begins after seizure termination, lasts over an hour, and is almost completely blocked by genetic-disruption or inhibition of COX-2 (Farrell et al., 2016; see Farrell et al., 2017 for the potential consequences of stroke-like events in epilepsy). Like ictal hyperperfusion, the expression of COX-2 in excitatory neurons could be informative of the underlying excitatory networks recruited by seizures. Use of non-invasive arterial spin labelling (ASL) MRI to map blood flow 45-60 minutes after patients’ seizures demonstrated that this method had similar or superior localization value to other commonly used imaging modalities (Gaxiola-Valdez et al., 2017). One major advantage to consider is that this approach provides a time-window of approximately an hour to detect postictal hypoperfusion, which is sufficient time to transfer a patient to the imaging facility to capture a scan. This obviates the need to record during the seizure, which is unpredictable and costly. As with ictal perfusion changes, cell type-specific recordings at the basic science level would be valuable to more fully understand the cellular basis for macroscopic postictal neuroimaging signals in the clinic.

Concluding Remarks

In this modern era, the complex disorder of epilepsy continues to be treated with broad-acting drugs and irreversible surgical resection, with persistent challenges posed by frequent undesirable side effects, a variety of comorbidities, and even outright failure in treatment outcomes in a substantial proportion of patients. Here we have described how heterogeneous mechanisms at the microscale level, with particular emphasis on functional microcircuit connectivity, provide essential information about critical features of disease manifestation that are generally missing with macroscale recordings, such as EEG. Differences in the organization of microcircuits and the manner by which they are recruited during seizures can lead to differences in the underlying mechanisms of seizure generation and spread. From these insights, we argue that the conceptual understanding of seizure networks should be refined to draw attention to the cellular networks that are capable of driving and/or supporting seizures (excitatory seizure network) versus the widespread electrographic seizure network. Since seizure network mapping is used to inform treatments and irreversible surgeries, it is imperative that this distinction is resolved. Application of basic science tools to reveal the cell type-specificity of commonly used clinical neuroimaging techniques (Lee et al., 2010) could provide a more complete understanding of what ictal and postictal neurovascular changes mean at the cellular level. Providing microscopic context to macroscopic recordings of seizures is necessary to explain and understand the unique mechanisms of epilepsy across individuals and should be used to develop more accurate clinical models of epileptic networks.

The advent of new tools and techniques to study epilepsy will yield a more comprehensive understanding of seizure dynamics. Optogenetics remains an important tool to dissect the specific cellular mechanisms that govern seizure initiation, seizure generalization, extra-focal seizure control, and interictal comorbidities. Combining cell type-specific recording techniques with optogenetics will be extremely valuable to observe endogenous network properties, generate hypotheses, and subsequently intervene to test these hypotheses, all with high temporal and spatial precision. These results will clarify the currently opaque understanding of seizures. Furthermore, these results can be applied in silico to biologically-realistic models of hippocampal networks (Bezaire et al., 2016) to observe how changes at the microcircuit level influence and contribute to seizure properties. Together, these efforts will help to explain the fundamental disconnect between microscopic and macroscopic observations in epilepsy.

With a detailed understanding of the crucial cellular elements of seizure networks, better targeted clinical therapies are possible. More restricted ablation of epileptogenic tissue is one obvious approach to refine the currently large tissue resections performed clinically. Pharmacological treatments could also be tailored to inhibit the identified cell populations most important for seizure generation and spread. For example, seizure generalization could be potentially prevented by drugs that target receptors enriched on mossy cells. Spatiotemporally precise optogenetic control of epilepsy may also be clinically translated as obstacles continue to be overcome (reviewed by Kim et al., 2018). Specific interventions to target specific problems is logical, however, as highlighted in the optogenetics section, less targeted manipulations well outside the seizure focus are also capable of seizure control. Stimulation of a relay structure, such as the thalamus (e.g. SANTE trial, see Salanova et al., 2015), or even sensory inputs to the thalamus (e.g. tactile stimulation in Parkinson’s disease, see Syrkin-Nikolau et al., 2018) could potentially control many epilepsies. Specific and non-specific therapies both have merit and it will be interesting to understand how advancements in these domains will ultimately lead to more effective treatments for epilepsy.