Interictal spikes

Interictal spikes, also known as the interictal epileptiform discharge (IED), for a long time have been the most important interictal biomarker for epileptic tissue. The neuronal correlate of interictal spikes is the paroxysmal depolarization shift, which is a large amplitude depolarization potential originating in a pyramidal cell/neuron; the principal generator of almost all electrographic activity of the brain .

In scalp EEG IEDs are often described as sharply contoured events with a negative peak. In contrast IEDs in intracranial EEG can have both positive and negative peaks and occur in a large variability of shapes and patterns. The latter can be predictive of a type of epileptic lesion such as rhythmic spike runs and continuous spiking over focal cortical dysplasia (FCD). It is also important to note that some brain regions generate physiological spikes, that can be misleading to the unexperienced reader. This is for example the case in high-amplitude spikes occurring over the hippocampus .

It is well established that epileptic spikes are sensitive to identifying epileptic areas but not very specific. Per definition the area generating spikes, labeled the irritative zone is only partially overlapping with the EZ . It is well established that patients can be seizure-free even if not all spike-generating brain regions are removed . Thus, the persistence of IEDs after successful epilepsy surgery does not necessarily depict failure of treatment or persistence of seizures nor does their rate of occurrence correlate with the anticipated seizure frequency in the matter of a relapse . Additionally, it can be hard to separate between spike origin and propagation in clinical practice. Often spikes are seen over wide-spread areas as a propagation phenomenon. This is often seen when spikes and seizure onset originate from a deep source and bottom of the sulcus and can lead to mis-localization of epileptic areas .

Interictal electrographic patterns

Certain other, non-IED, electrographic patterns have been identified interictally as well as pre- or periictally which could be considered useful as biomarkers of epileptogenicity of a cortical region of interest. Interictal patterns that were commonly observed in intracranial EEG include low voltage fast activity, persistent focal electro-decrement, or the presence of delta brushes . However, the quality and the quantity of available evidence to guide the utilization of these biomarkers remain sparse, which in itself is a limitation.

Seizure onset zone and ictal onset patterns

The current gold standard for delineating margins for epilepsy surgery is the recording of spontaneous seizures and identification of the seizure onset zone (SOZ). The identification of the SOZ however is not without pitfalls. First, false localization due to fast propagation is common especially when deeper foci are involved. Additionally intracranial electrodes can only record activity from their direct vicinity. As a result, electrode placement influences the ability to identify SOZ areas correctly. Overall studies show varying success of surgery after complete removal of the SOZ .

Special emphasis has been placed on trying to identify ictal onset patterns that are specific to the SOZ and confirm that the first ictal activity observed in a patient is not the result of propagation phenomena. In one study, seven seizure-onset patterns were identified across 53 sampled seizures, namely, low-voltage fast activity (43%); low-frequency high-amplitude periodic spikes (21%); sharp activity at ≤13 Hz (15%); spike-and-wave activity (9%); burst of high-amplitude poly-spikes (6%); burst suppression (4%) and delta brush (4%) . Each pattern occurred across several pathologies, except for periodic spikes, which were only observed in association with mesial temporal sclerosis and the delta brush pattern which occurred exclusively in FCD . While these onset patterns are a key indicator of SOZ areas, no evidence systematically analyses that observation of any of these SOZ patterns in the region of the surgical resection correlates with postsurgical seizure outcome. Moreover, the utility of SOZ has lately been challenged by observations from long-term intracranial recording devices. These recordings suggest that SOZ areas in individual patients may change over time especially in patients with multifocal epileptic activity. In patients with bi-temporal epilepsy, for example, the predominant seizure onset may alternate between mesio-temporal structures over time .

High-frequency oscillations

HFOs are defined as spontaneous EEG or magneto-encephalographic (MEG) events in the frequency range between 80 and 500 Hz, consisting of at least four oscillations that clearly stand out from the background activity . Frequencies between 80 and 250 Hz are further subdivided as ripples while the frequencies that are between 250 and 500 Hz are identified as fast ripples and very-fast ripples would even exceed 500 Hz in frequency . HFOs are thought to be generated via multiple mechanisms . Physiological HFO that ubiquitously occur in cortical regions under natural circumstances reflect summated excitatory postsynaptic potentials. Pathologic HFO in the frequency range of either ripples or fast ripples reflect summated action potentials of synchronously bursting neurons or principal cell action potentials .

When it comes to the clinical utility, HFOs are probably some of the best studied biomarkers. First studies explore the relationship between HFOs and the SOZ. A strong argument for the link between fast ripples and epileptogenicity is the observation from the kainate model of epilepsy that fast ripples were found in the hippocampus of rats that developed chronic spontaneous seizures but were not found in rats that did not develop epilepsy . In patients with epilepsy this link naturally cannot be established because all patients investigated with intracranial electrodes have chronic seizures. In addition, intracranial EEG methods are not perfect when it comes to sampling only epileptogenic tissue and avoiding healthy brain tissue. For these reasons, only correlative evidence exists to establish a link between HFOs and the presumed epileptic tissue in patients. When it comes to proving HFOs are biomarkers of epileptic tissue, this link is maybe the weakest but also easiest to test. Most previous studies were retrospective group analyses, and current results show that in groups of patients the average rate of HFOs on electrodes recording from the SOZ is significantly higher than outside of it . However, the highest HFO rate does not occur in the SOZ for every patient. And, while these studies support the general finding of increased HFO rate in the SOZ, the results were not completely identical. Some studies found HFO superior to interictal EEG spikes while others did not . Moreover, some studies suggest a stronger link between ripples (80–250 Hz), rather than fast ripples, and the SOZ . Many factors contribute to this variability including different recording methods, different patient groups, and different epilepsy etiologies. Yet, despite this variability, the fact that HFOs are associated with the SOZ in groups of patients with epilepsy is reassuring rather than discouraging.

Studies clearly indicate HFOs exist beyond the SOZ , which again is unsurprising since it is generally accepted that the volume of brain tissue responsible for generating seizures extends beyond the SOZ and HFOs are present in normal brain tissue . To prove that HFOs are relevant to seizure generation in patients, one must establish a link between the removal of HFO-generating tissue and seizure outcome . Studies on seizure outcomes were designed in a very similar way to the SOZ studies . In general, these studies tested the hypothesis that patients would be seizure-free if most contacts with HFOs were removed, while patients would continue to have seizures if contacts with HFOs remained ( Fig. 3.1A ). Most seizure outcome studies were performed retrospectively using group analysis, but the first studies had single patient data, and showed in some surgical patients that the majority of HFOs were removed yet seizures continued . It was proposed that these findings do not discount HFOs as reliable markers of epileptic tissue; intracranial electrodes are always limited in their spatial coverage, and some HFOs may exist beyond the sampled brain areas . The fact that studies from different centers using different analysis methods essentially replicated earlier results was reassuring . However, metaanalysis of several larger studies showed the link between the removal of HFOs, and seizure freedom is weak , and none of the studies at this time would satisfy the criteria for a high-quality trial as, for example, measured by Cochrane .

(A)–(D) illustrate which network areas have to be removed to lead to a seizure free outcome after resection. The traditional hypothesis (1 A) assumed that all biomarker areas have to be removed to gain seizure freedom. Newer evidence suggests that thresholding the most active areas (B), being aware of physiological events (D) and considering propagation and network activity of events (C) are crucial for using biomarkers to predict seizure outcomes.

Generally, prospectively designed studies have less potential for bias (e.g., investigator bias, loss to follow-up) and confounding effects (e.g., type of epilepsy or seizure type, location of SOZ) than retrospective studies. In the context of HFOs and epilepsy surgery, prospective studies can be performed using a group or individual patient approach. Most prospective HFO studies conducted thus far were observational, meaning that HFOs were not used to tailor a resection. Two studies performed at the group level show it is possible to use HFOs to predict 1-year seizure outcomes . This information, however, is less relevant to a patient who would want an individual prediction of his or her surgical outcome. In this regard, the results of HFO studies are not convincing. Two complementary scenarios were observed: (1) either all sites with HFO were removed and the patient continued to have seizures or (2) brain regions with frequent HFOs remained and the patient was seizure-free. The results from two prospective studies show that 6 of 37 (16%) patients did not become seizure-free despite removing most of the HFO-generating tissue. By contrast, 14 of 35 (40%) patients became seizure-free after incomplete removal of HFO . These results challenge the reliability of HFOs as a biomarker for epilepsy. We provide below some explanations for these results (Sections 2.3 and 2.4) and hope for future studies to improve HFO analysis and enhance clinical utility (Sections 2.5, 3, and 4).

Most recently a prospective randomized controlled trial was completed using intraoperative ECOG to identify HFO and tailor the surgical resection according to the HFO distribution . This trial was a noninferiority trial aiming to prove that HFO performs similarly to using interictal spikes in regard to identifying epileptic areas. This trial could not show that HFO are as good as epileptic spikes to tailor the resection when looking at all patients. Pathologies associated with poor prognosis were found to be a confounder in the HFO-tailored resection group. After correcting for confounders, HFO were noninferior to epileptic spikes in patients with neocortical epilepsy onset. Even if this trial did not show the expected superiority of HFO compared to classical EEG markers, it is a milestone in neurophysiology research and evaluations of presurgical diagnostics as it is at this point the only randomized controlled evidence in the field.

The discussed results challenge the reliability of HFOs as a biomarker for epilepsy. In the following we provide some possible explanations for these results, which are not unique to HFO and should be considered whenever establishing a biomarker for epileptic tissue.

A first challenge to consider is the fact that HFOs are found not only in epileptic tissue; oscillations in the same frequency band are generated during physiological activity . Ripples in the mesio-temporal structures are clearly linked to memory consolidation , and fast ripple-frequency oscillations have been linked to sensory function in the central and occipital brain areas . Thus, some of the remaining HFOs in seizure-free patients might not be pathological at all ( Fig. 3.1D ). No methods exist to differentiate normal from pathological HFOs within the same brain regions. Compounding this issue is evidence for normal and pathological HFOs in the same tissue , and recent data from the normal brain stereo-EEG brain atlas suggesting that normal HFOs are much more abundant than previously thought . Attempts to identify normal HFOs by establishing a link with other physiological events, such as sleep slow waves or spindles, are promising but have not been tested in a prospective surgical setting . An alternative approach is to use only HFOs with interictal spikes to identify pathological HFOs. This limits the HFO analysis to brain regions with interictal spikes however, in some types of epilepsy, this could be an advantage rather than a limitation.

The second challenge is finding the right threshold when interpreting HFO data ( Fig. 3.1B ). In most patients, there are some brain areas with very frequent HFOs, many areas with moderate HFOs, and fewer areas with little or no HFO activity. This leaves epileptologists with the same challenge associated with interpreting interictal spikes to delineate the irritative zone: how to threshold the events. How much HFO activity is required to identify pathological brain regions, which areas with HFOs can be ignored, or should any area with HFOs be ignored? In retrospective studies, some groups addressed this challenge by using thresholding methods like the “upper fence” , which help to individually determine the area of highest HFO activity. It is important to note that these threshold measures are not based on any mechanistic understanding but are instead motivated by statistical notions. In addition, HFO rates change—for example, with sleep stage, medications, lesion type, or extended recordings (see Section 4)—which may cause any type of thresholding to fail.

Thresholding is only one of many methodological problems when analyzing HFOs in a prospective fashion. Different automatic detection approaches, electrode types, or recording methods may differ in reliability when analyzing HFOs . A multicenter trial led by some of the authors aimed to address these concerns. However, neither thresholding, different recording methods, different detector types, nor different statistical approaches generated a reliable prediction of seizure outcomes using HFOs .

Last but not least HFO like other epileptic activity might not be isolated localized events, but also be connected via networks ( Fig. 3.1C ). New thinking has emerged from a study using pre- and postoperative acute electrocorticography that might help to explain results from prior HFO studies . The authors showed that the distribution of HFOs is variable before and after a surgical resection. This observation suggests that HFOs are organized in networks, and the removal of a core generator could leave remote sites silent. This concept cannot be tested in the same way in chronic stereo-EEG recordings as for ethical reasons patients are not recorded with intracranial EEG after surgery. As illustrated in Fig. 3.2 the network concept is very similar to what we know from interictal spikes, that is, some areas with spikes might be due to propagated activity. Currently no measure has been established to differentiate between the core and the propagating sites of HFOs. Van’t Klooster and coworkers suggest that the combined analysis of pre- and post-operative HFO distributions helps to provide a better predictor for outcome on an individual basis . Meanwhile more studies suggest that HFOs can form networks of variable sizes and should not be interpreted as standalone events . These results are consistent with the notion of complex epileptic networks, and fundamentally differ from the existing approach to HFO analysis. Historically HFOs were considered to have very focal generators, local impact, and no propagation. These concepts should be reconsidered in light of recent results, and future HFO analysis should focus on interactions between HFOs in separate brain regions as well as their variable impact on seizure generation.

illustrates the interaction between interictal epileptiform discharge (IED) (epileptic spikes) and high frequency oscillations. (A)–(C) are data from a simultaneous macro–mirco electrode study in the hippocampus which showed significantly different multiunit firing between IED with and without high-frequency oscillation (HFO) (C). The graph in (D) shows how combining the two biomarkers to look at IED with HFO can be the best compromise between sensitivity and specificity to identify the seizure onset zone.

In summary, HFOs are promising biomarkers of epileptic activity, but recent studies which focus on higher-level evidence data, like prospective studies and randomized trials have dampened the enthusiasm for HFO being a simple ubiquitously used new biomarker. Rather we have learned that a deeper understanding and better methodology is necessary prior to widespread application.

Ictal infra-slow activity

ISA has been traditionally invisible in EEG, as a lower filter setting was set to eliminate activity below 0.5 Hz. More recently lower frequency bands in the subdelta range (<0.5 Hz) have started to gain popularity and electrophysiologist’s interest . In contrast to the work in HFO, ISA has been mostly analyzed in the preictal and ictal periods. This is mostly related to the fact that ISA is a very slow and long-term change that occurs regularly over different brain regions and then is nonspecific to epilepsy. To use the method specific to epileptic activity analysis around seizures is necessary to identify a time point around which epilepsy specific changes in ISA might occur. Very few studies looked at interictal very low activity and its clinical relevance is currently unclear. Underlying mechanisms are likely different from ISA .

In feline models with experimental tonic-clonic seizures, depolarization of the pyramidal neurons caused a negative direct current (DC) shift when recorded using conventional DC amplifiers . This gradually declined and changed into a positive shift during the postictal period. Further studies in feline models showed that these DC shifts had a horizontal dipole with a maximum amplitude at the focus and a field extending to a perimeter of 10 mm . In human epilepsy, ictal DC shift was first described by Ikeda et al. with the use of subdural electrodes in three patients with focal refractory epilepsy .

A slow-rising negative potential (DC shift) was observed in a localized area, 1–10 seconds prior to the observed conventional ictal discharges . The terms slow DC shift (slow DC shift), ISA as well as ictal baseline shifts (IBSs) have been used interchangeably to describe this electrographic phenomenon which is in the range of subdelta (<0.5 Hz) frequency range.

The term “DC shift” stems from the use of a DC amplifier based on the fact that in the earlier studies slow baseline shifts could not be appreciated on conventional EEG recordings obtained with alternative current (AC) amplifiers. This was because of the short time constant used in low-frequency filters. However, it was demonstrated that with the use of AC amplifiers coupled with a low frequency filter harnessing a longer time constant, ictal DC shifts could be recorded in humans, in intracranial as well as on scalp EEG recordings . Therefore, either ISA or IBS are better-suited terms in describing this particular electroencephalographic phenomenon.

The mechanism by which the ISA originated has been widely debated. Experimental feline epilepsy models as well as nonhuman primate (Baboons) models have demonstrated a rise in extracellular potassium associated with the ictal DC shifts secondary to the failure of glial mechanisms which buffer this system .

As mentioned above, ISAs are ubiquitous throughout the brain and not specific to epileptic areas. Ictal ISAs however were found to be either preceding, co-occurring with or following the conventional EEG ictal onset, and had a duration ranging between 1.9 and 2.9 seconds .

Extensive intracranial coverage shows that ictal ISA can be spatially distributed over a wide surface area . However, its true spatial extent is generally smaller than the area defined by conventional frequency activity in both neocortical and temporal lobe epilepsy patients .

It is therefore a challenge of the method to identify spatially restricted ictal ISA that are linked to epilepsy and delineate the SOZ. Currently no prospective trials exist that assess the sensitivity or specificity of ISA to improve outcome prognostication. Evidence from smaller studies suggests that ISA can improve the precision for identifying epileptic areas. Most of these studies as retrospective and focus on identifying the SOZ . In a prospective study of six patients, full resection of areas with ictal ISA and partial resection of the SOZ defined by conventional EEG frequencies, led to Engel class I outcome in 2/6 and Engel class II in 4/6 . 5 out of 6 patients with eventual class I outcomes had the DC contacts that were resected in a separate study .

Overall, the evidence for ISA in the presurgical setting is lower than for other biomarkers such as HFO. On reason is the methodology used for most studies. DC coupled amplifiers have been used historically in the past to record slow EEG baseline shifts. However, most researchers studying slower EEG fluctuations nowadays use AC amplifiers instead, which are more stable compared to DC amplifiers . Pure, unaltered DC recordings can only be achieved using nonpolarizing silver/silver chloride electrodes which are potentially toxic to human brain parenchyma . For a successful recording, apart from using an AC amplifier with a longer time constant, using platinum electrodes spread over a larger surface area (e.g., subdural as opposed to depth electrodes) and using a higher input impedance of the amplifier (>50 MΩ) can overcome some of the technical difficulties . Recent studies suggest that different time constants for ISA exist which would allow use of regular amplifiers in some circumstances .

Epileptogenicity index

The EI was introduced as a new quantitative measure of characterizing the epileptogenic potential of the involved anatomical brain structures and is a measure of both epileptogenicity and connectivity. It is based on the assumption that there is an increase in rapid discharges and activity during the transition of an interictal pattern into an ictal pattern on intracranial recordings . Often there is an increase in beta/gamma activity accompanied by a voltage attenuation. EI takes into account, both the spectral and temporal properties of the fast brain activity as well as the occurrence and interaction of this activity between brain regions covered by electrodes and involved in seizure generation. It characterizes the ability of a brain structure to generate fast oscillations during a seizure and the time delay in producing these in comparison with the time of seizure onset (t0). EI functions semiautomatically, ranking each of the intracranial electrodes based on the tonicity of the observed rapid discharges at ictal onset and the time delay in involving the particular electrode. The quantification process gives a value between 0 and 1. The developers of EI suggest that any electrode area with an EI of more than 0.3 can be identified as epileptogenic.

EI has been designed to allow a more automatized identification of brain areas that are involved in seizure generation. Its use can avoid bias in visual review. Studies that compared retrospectively EI values with visually identified SOZ areas showed a high correlation between both regions. In a group of 17 patients with mesial temporal lobe epilepsy EI values were computed in mesial and lateral structures of the temporal lobe. Structures which were involved early in the ictal process and produced rapid discharges at seizure onset demonstrated statistically high EI values . The highest EI values were computed from signals recorded in mesial structures in the majority of patients. Similar findings were observed in a mixed group of patients with temporal and neocortical epilepsy suggesting that widespread EI increases were linked to poorer seizure outcomes after surgery .

A particular limitation of the EI derives directly from the fact that it highly depends on changes in high-frequency activity. Ei will not be increased when ictal patterns do not involve fast activity, but rather present with either rhythmic spikes or slow waves as seen with a quarter to one-third of intracranial ictal patterns . Thus, some EI might fail to identify some seizure-onset areas.

The connectivity EI (cEI) was proposed as a continuous development from EI to overcome the inability of EI to evaluate slower electrographic onset patterns .

CEI measures changes in functional connectivity between SEEG signal trends shorty prior to and around the seizure onset . It combines a directed connectivity measure based on the nonlinear correlation coefficient ( h ²) and graph theory measures (out-degrees) with the EI. Like with EI there is a strong correlation between visually identified SOZ areas and high cEI measures. Data from 51 patients showed that cEI has a better yield in marking SOZ in seizures with slow onset patterns. The removal of EI areas has also been linked to better seizure outcomes.

In contrast to the individual biomarkers discussed in the beginning of this chapter EI and cEI do not only aim for an improved identification of epileptic tissue but also focus on automatizing the identification of epileptic areas. This reflects a constant movement from visual analysis which is largely reviewer dependent to more objective methods like automated assessment. A recent publication might be a peek into future applications and introduce the concept of virtual epileptic patient brain modeling. This method simulates seizure data using a virtual brain and combines information from stereoelectroencephalography, anatomical MRI data, connectivity measures and a computational neuronal model. In the publication the ability to identify epileptic areas was similar to that of cEI and resection of identified areas correlated with postsurgical seizure outcome .