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/9780197549469.003.0013
This chapter focuses on high frequency oscillations (HFOs), which are also known as transient oscillations in the broad 80 to 6000 Hz frequency range. HFOs typically represent a normal physiological activity and pathological phenomenon found in epileptic tissue. Technologies like intracranial EEG (iEEG) and scalp electrodes aid with recording HFOs and ripples that became an important biomarker of epilepsy and epileptogenesis. The chapter highlights the significance of identifying ripples and fast ripples of neuronal activity, which are neurophysiological events that can be defined by features such as association with sharp waves and particular laminar distribution. It then expounds on the differences between physiological HFOs and pathological HFOs.
For the purpose of this section, we define high-frequency oscillations (HFOs) as transient oscillations in the broad 80–600 Hz frequency range. While this range is usually divided into that for ripples (80–250 Hz) and that for fast ripples (250–500 or 600 Hz), it typically includes a variety of neuronal activity. Importantly, ripples and fast ripples are neurophysiological events that can also be defined by other features, like association with sharp waves and particular laminar distribution, which are relatively independent of frequency. Because action potential firing from an undetermined number of neurons may collapse and contribute energy at >250 Hz, their contribution should be carefully considered when examining HFOs. Hence, HFOs are most often considered as consisting of at least four oscillations that clearly stand out from the background. There is no formal limit to the maximum number of oscillations, but HFOs have a limited duration and are to be distinguished from high-frequency activity, which is the energy, usually measured by spectral analysis, in the corresponding frequency band: ripple-band activity in the 80–250 Hz band and fast ripple-band activity in the 250–500 Hz band. Such an activity is typically measured over a duration much longer than a single HFO, several hundred milliseconds to several seconds, and is assumed to be stationary during the measurement period. High-frequency activity is also measured as evoked activity during sensory, motor, or cognitive tasks. For the rest of this chapter, we will discuss HFOs, and not high-frequency activity.
HFOs represent normal physiological activity as well as a pathological phenomenon found in epileptic tissue. In this context, it is important to make a distinction between ripples, which can be physiological and pathological, and fast ripples, which are primarily pathological since they are very rare in healthy human brain tissue (Frauscher et al., 2018) and appear to be characteristic of epileptogenic tissue. We will discuss further below the distinction between physiological and pathological HFOs.
HFOs were first recorded with microelectrodes in experimental animal models of epilepsy and shortly thereafter with microelectrodes in epileptic patients (Bragin et al., 1999). Experimental work (Bragin et al., 2002) indicated that their generator could extend over 100 microns in the rat, which could correspond to about 200 microns in humans (Châtillon et al., 2011). In principle this is too small a generator to be visible with typical clinical intracerebral electrodes, with surface contacts of more than 1 cm2. Nevertheless, when clinical electrodes were used with amplifiers and computer systems capable of recording in the HFO frequency band, it was found that ripples and fast ripples were present at seizure onset (Akiyama et al., 2005; Jirsch et al., 2006) and in the interictal period (Jacobs et al., 2008). The fact that HFOs could be recorded with standard clinical electrodes opened a whole new field of investigation and of application, which we will review below.
Attempts have been made at comparing the ability of clinical electrodes and microelectrodes as well as electrodes of intermediate size to record HFOs. Comparing clinical subdural grid and depth contacts to superficial and penetrating microcontacts of 40 microns in diameter, it was found that the penetrating microcontacts detected substantially more HFOs, particularly fast ripples, than depth electrodes, whereas superficial microcontacts and standard subdural electrodes did not differ in their abilities to record HFOs (Blanco et al., 2011). Comparing macrocontacts of different sizes in humans and in rats (Châtillon et al., 2011, 2013) showed that smaller contacts had no significant advantage in recording HFOs. This looks quite straightforward when examining neocortex, because the sources of HFO activities are distributed in a relatively homogenous region. However, when subcortical structures are contributing, the anatomy and geometrical disposition of the sources may compromise how HFOs are detected by micro- and macroelectrodes.
Indeed, observations of HFOs in intracranial electroencephalogram (iEEG) recordings indicated that HFOs are most often present at a single contact, occasionally at two adjacent contacts, typically separated by 5 mm; when they occur simultaneously at several contacts, they are not synchronous (Worrell et al., 2008). For this reason, it would appear that HFO generators are too small to be seen on scalp EEG (it is often said that an iEEG generator surface of 6 to 10 cm2 is required for visibility on the scalp). It was therefore surprising when Kobayashi et al. (2010) demonstrated the presence of ripples in the scalp EEG of children with electrical status epilepticus during sleep (ESES). Several studies have subsequently observed ripples in scalp EEG and MEG (van Klink et al., 2017); even fast ripples have been recorded in the scalp EEG of young children with epilepsy (Bernardo et al., 2018). It is clear that the concept of a synchronized generator of 6 to 10 cm2 for visibility on the scalp (Tao et al., 2007) must be revised. According to the model and simulations of von Ellenrieder et al. (2016a), it is possible that a cortical region generating simultaneous but not necessarily synchronized events contribute to the generation of scalp EEG events; this could explain the presence of scalp HFOs.
It therefore appears that HFOs can be recorded with microelectrodes and with all the types of macroelectrode used to record EEGs. To record HFOs correctly, the antialiasing filter of the recording system should be set at about 600 Hz and the sampling rate should be at least 2000 Hz. Because HFOs are low-amplitude events, it is important to use low-noise amplifiers (Fedele et al., 2017a). These and many other methodological conundrums of HFOs should be carefully considered (Barth, 2003).
Physiological activity in the ripple frequency range can be evoked by cognitive or sensory stimuli in rodents (Barth, 2003), as well as in apparently healthy human brain and in epileptic tissue (Cervenka et al., 2013; Liu & Parvizi, 2019). We know that spontaneous ripples occur in healthy brain tissue in animals (Buzsáki, 2015) and in humans (Frauscher et al., 2018). In rodents, physiological ripples are more typically present in the hippocampal and para-hippocampal regions, as well as in hippocampal targets such as the retrosplenial cortex (Nitzan et al., 2020). In humans, hippocampal ripples are intimately related to the recollection and consolidation of episodic memories (Norman et al., 2019). It is therefore reasonable to assume that spontaneous physiological ripples also occur in epileptic hippocampal tissue. However, in the human brain, ripples and associated replay can be also recorded in other regions (Vaz et al., 2020), and so it may be difficult to separate pathological from physiological ripples on the basis of their signal characteristics. Although they tend to have slightly different frequency and amplitude characteristics (Matsumoto et al., 2013), as well as differences in the way in which they are activated (von Ellenrieder et al., 2016b, 2017; Song et al., 2017; Weiss et al., 2020), physiological and pathological ripples are not sufficiently different to be separated at the level of individual events. In the hippocampus of healthy and epileptic rodents, coexisting ripples and fast ripples can be relatively separated by using different spectral and nonspectral features (Valero et al., 2017).
In the epileptic brain, it has been well documented that many ripples occur at the same time as EEG spikes and sharp waves (interictal epileptic discharges [IEDs]). These are presumably pathological, and one approach to selecting only pathological ripples is to select only those occurring with IEDs, in iEEG (Wang et al., 2017), or in scalp EEG (Klotz et al., 2021). When trying to assess tissue epileptogenicity with ripples, another approach to deal with the presence of physiological ripples is to normalize ripple rates with the region-specific distribution of physiological ripples provided by the atlas of Frauscher et al. (2018). In the human epileptic brain, physiological ripples occur at very different rates in different brain regions: the atlas points to the occipital lobe, primary motor and sensory cortex, primary auditory cortex, and mesial temporal regions as areas with particularly high rates of physiological ripples. The atlas provides a distribution of rates for each region. If rates are significantly higher for a given region than the expected physiological rate, then that rate can presumably be considered pathological (Zweipenning et al., 2022). A similar approach was used but with a different atlas and incorporating measures of phase-amplitude coupling (Kuroda et al., 2021). The high inter-regional variability emphasizes the need for a region-specific correction. It is also important to note that physiological and pathological ripples fluctuate greatly with the stages of sleep and wakefulness, with the highest rates seen in non–rapid eye movement sleep (Bagshaw et al., 2009; Staba et al., 2004).
Fast ripples were originally defined as short-lived oscillations faster than 250 Hz associated with deflections of varying polarity in different layers of the hippocampus (Buzsáki, G., 2015). More generically, they also appear as riding upon low-amplitude deflections in intracranial EEG measurements elsewhere in cortical regions. Fast ripples appear much more specific to epileptic tissue because their occurrence in healthy brain tissue is very rare, as documented in experimental animals (Bragin et al., 1999) and in humans (Frauscher et al., 2018). They are therefore a presumably more specific marker of epileptic tissue than ripples. One difficulty in their practical use as such a marker is their rare occurrence (see below).
Although there is no simple explanation for this, it appears that physiological ripples are rare on scalp EEG, as ripples occur most often at the time of IEDs (Melani et al., 2013) and are therefore considered pathological. Physiological cortical ripples, however, have been recorded in the scalp EEG of healthy children and in the EEG of epileptic children showing no IEDs (Mooij et al., 2017). These occur primarily in the central region and are of very low amplitude (median around 1 microVolt). Possibly, these ripple events near the brain midline can be the human equivalent to retrosplenial ripples in rodents (Nitzan et al., 2020). Because of the restricted spatial distribution of scalp ripples, it is more effective to record them with a high-density electrode array (Kuhnke et al., 2018). Certainly, ethical issues impede a thoughtful mapping of the healthy human brain to allow for more consistent definition of the landscape of physiological ripples.
The cellular and synaptic mechanisms underlying HFOs have been extensively investigated. Since the discovery of fast ripples in temporal lobe epilepsy (TLE) (Bragin et al., 1999), the hippocampus has attracted most of the attention, given the preponderance of physiological ripples in this structure (Buzsáki, 2015). The fact that fast ripples were mostly described in the epileptic dentate gyrus (DG), a hippocampal region presumably lacking ripples in healthy rodents (but see Meier et al., 2020; Sasaki et al., 2018; Swaminathan et al., 2018), had a huge influence in the first attempts to identify their mechanisms.
In their seminal paper on pathological HFOs, Bragin and Engel identified both local field potential (LFP) population discharges accompanied by intense neuronal firing in the hippocampus and entorhinal cortex (EC) of epileptic rats and human (Bragin et al., 1999). They correctly inferred a major contribution of action potentials and pointed to synchronous bursting as the major underlying pathophysiological phenomenon. Few years later, they proposed the hypothesis of pathologically interconnected clusters (Bragin et al., 2000) and showed evidence of their local generation in the DG both in freely moving rats and in vitro slices (Bragin et al., 2002). These studies provided grounds for hypersynchronous firing as one of the most influential models of pathological HFOs >100 Hz (but see Menendez de la Prida and Trevelyan, 2011).
The potential role of intrinsic excitability underlying pathological brain activities entered into force supported by the discovery of acquired channelopathies in TLE (Bernard et al., 2004). Transcriptional and posttranslational mechanisms were associated with the loss of function of specific potassium channels, which resulted in amplification of neuronal firing at the single-cell level. A potential connection with fast ripples was established by combining simultaneous cell-attached recordings from individual CA3 pyramidal cells with LFP recordings (Dzhala and Staley, 2004). These experiments revealed a relatively synchronous onset of intrinsic bursts from subsets of cells, with the amplitude and frequency of LFP oscillations being modulated by pharmacologically changing action potential accommodation. Remarkably, synaptic mechanisms were also identified to regulate the fast ripple onset, which was attributed to recurrent excitatory connections (Miles et al., 1988).
By the time of these discoveries, there were debates in the field regarding the mechanisms of physiological ripples (100–200 Hz) (Buzsáki et al., 1992). A long disputed mechanism was gap junction and electrical interactions (Draguhn et al., 1998). However, this was found to contradict the necessity of synaptic transmission to sustain ripples (Buzsáki et al., 1983; Csicsvari et al., 1999) and fast ripples (Dzhala and Staley, 2004). Moreover, mice lacking the gap junction protein connexin 36 showed relatively normal sharp-wave ripples and impairment of gamma oscillations instead (Buhl et al., 2003). Thus, while the gap junction hypothesis kept reverberating for some years, for many researchers the basic underlying processes should be rather linked to synaptic and intrinsic factors associated with nonlinear microcircuit operation (Menendez De La Prida et al., 2006). Under this framework, the action potential inter-spike interval defined the maximal frequency of LFP oscillations (Dzhala and Staley, 2004), supporting the idea of oscillations in the 250–500 Hz range reflecting in-phase synchronous firing. However, how could pathological LFP oscillations emerge at 250–500 Hz, given lower firing rate of single neurons?
A more elaborated model proposed that subsets of neurons firing out-of-phase would make up higher frequency oscillations (Foffani et al., 2007). Intuitively it was not obvious how to explain that out-of-phase firing does not yield to destructive interference, but computational models proved such a regime was possible for groups of cells lagging at di- and polysynaptic delays across recruitment period (Ibarz et al., 2010; de la Prida et al., 2006). Since the mechanism relied on a recurrently connected microcircuit of low-frequency discharging cells, it was easily extrapolated to any epileptogenic territory and to broader HFO bands (Jefferys et al., 2012; Menendez de la Prida and Trevelyan, 2011). This extended framework was able to better accommodate experimental evidence, including interactions between local versus global dynamics in the transition to seizures (Jiruska et al., 2010).
There was, however, a point of dispute with the original paper, which proposed that fast ripples may be harmonics of physiological ripples (Foffani et al., 2007). The limiting concept was that ripples and fast ripples should not be related, given their totally opposed mechanisms (Engel et al., 2009).
Physiological ripples (100–200 Hz) are typically recorded in the healthy CA1 in vivo (Buzsáki, 2015), and the idea that they exclusively reflect rhythmic inhibitory potentials (IPSPs) was dominant for years (Ylinen et al., 1995). In contrast, while in-phase pyramidal cell firing was systematically appreciated from the earliest reports (Buzsáki et al., 1992; Csicsvari et al., 2000), this angle was typically neglected because intrinsic bursting was always considered pathological. However, healthy CA1 pyramidal cells fire complex spike bursts which may actually accompany ripples (Kamondi et al., 1998), although at a relatively low level of synchronicity across the network.
Clarification to this debate took a few more years. Recent experiments now demonstrate that the physiological ripple generator is fine-tuned. First, specific microcircuit interactions between pyramidal cells and subsets of GABAergic interneurons underlie ripples in vivo (English et al., 2014; Stark et al., 2014) (Fig. 13–1A). Second, the contribution of IPSPs and in-phase firing was found to be distributed across different types of CA1 pyramidal cells making up deep and superficial sublayers (Valero et al., 2015). Under the new scenario, lamination of the CA1 stratum pyramidale hinders a potential segregation of IPSPs and burst firing across different cells of the normal hippocampus. Consequently, both excitatory and inhibitory potentials phase-lag each other to shape ripple oscillations (Maier et al., 2011). Disruption of the excitatory/inhibitory coordination results in pathological firing and fast ripple events (Valero et al., 2017) (Fig. 13–1A). This mechanism, which relies on the specific hippocampal physiology, supports commonalities between ripples and fast ripples and suggests that there may be multiple mechanisms leading to high-frequency LFP oscillations (Alvarado-Rojas et al., 2015; Smith et al., 2020).
Summary of basic mechanisms of fast ripple generation. A. In the hippocampus, fast ripples emerge from recurrent interactions of subset of cells. Excitatory intrinsically bursting neurons and regular spiking cells (dark triangles) interact with fast-spiking (more...)
Differences between preparations (in vivo versus in vitro) and recording conditions could favor different HFO mechanisms (Aivar et al., 2014). For example, in vitro ripples are more typically recorded in CA3 and GABAergic interneurons, and specifically parvalbumin (PV) basket cells, are claimed to play a major role in their initiation (Schlingloff et al., 2014). In contrast, optogenetic activation of PV basket cells in vivo has poor effect in triggering ripples, which are more likely to be initiated by pyramidal cell firing (Stark et al., 2014). The only way to reconcile these apparently disparate results is to understand that ripples emerge from microcircuit interactions that activate the appropriate sequence of firing between pyramidal cells and interneurons (Bazelot et al., 2016). GABAergic interneuron activity is thus essential to pace pyramidal cell firing and may critically influence generation of epileptifom activities (Khoshkhoo et al., 2017).
The fact that both rhythmic IPSPs and bursting can underlie LFP fast oscillations makes it difficult to separate their contribution. Indeed, distinction between hippocampal ripples and fast ripples in terms of frequency was always elusive (Engel et al., 2009), and separation should better rely on multimodal spectral information (Valero et al., 2017). Importantly, physiological ripples define a regime where individual neurons fire selectively and in a coordinate manner, reflecting the content of memory traces (Lee and Wilson, 2002; Norman et al., 2019; Vaz et al., 2020) (Fig. 13–1A). These neuronal sequences presumably carry relevant information for trace consolidation across memory systems (Atherton et al., 2015). In contrast, at least in CA1, fast ripples represent a shuffled version of physiological ripples caused by a collapse of excitatory/inhibitory synaptic sequences (Valero et al., 2017) (Fig. 13–1A, rightmost).
Neocortical epileptogenic territories also exhibit the presence of HFOs (Urrestarazu et al., 2007). The origin of neocortical HFOs is similarly linked to the local microcircuit, including the firing from fast rhythmic bursting cells and regular spiking neurons coordinated via recurrent connections (Grenier et al., 2001; Jefferys et al., 2012) (Fig. 13–1B). This local origin may well reflect the functional changes affecting intrinsic excitability, connectivity, and plasticity in epileptogenic regions of the cortex. Interestingly, regional differences across cortical territories may provide support for the emergence of fast oscillations, such as in the somatosentory cortex of the rodents (Barth, 2003).
Additionally, fast ripples can be brought about by afferent inputs at specific cortical areas densely connected with the hippocampus. For instance, the frontal, cingulate and retrosplenial cortices exhibit conspicuous neuronal activation accompanying hippocampal ripples in normal rodents presumably reflecting dedicated hippocampal-subicular-cortical pathways (Abadchi et al., 2020; Nitzan et al., 2020; Remondes and Wilson, 2015). Similarly, several brainstem nuclei show firing of distinct cell types correlated with sharp-wave ripples, including the medio-dorsal thalamic nuclei, the medial raphe, and amygdala (Cox et al., 2020; Girardeau et al., 2017; Wang et al., 2015; Yang et al., 2019). Given that interregional connections interact brain-wide, coordinated activity between the hippocampus and neocortex can exhibit correlations during a myriad of physiological events, such as up/down states and spindles (Jiang et al., 2019; de la Prida and Huberfeld, 2019; Siapas and Wilson, 1998). Served by these physiological pathways, interictal activity can further promote the coupling between regions and enhance cortical excitability in the epileptic brain, thus affecting the normal cognitive operation (Gelinas et al., 2016; Tóth et al., 2018) (Fig. 13–1B).
Early experimental work relied on visual inspection to identify HFOs, but this is a very tedious task. Early human studies (Staba et al., 2002) used a simple semi-automatic detector identifying bursts of three or four sequential waves in the appropriate duration range, corresponding to ripples or fast ripples, having an amplitude several times that of the root-mean square of the background amplitude. Events detected by such a simple detector were then validated by visual analysis of a high-pass-filtered EEG (filtered with cut-off at 80 Hz or 250 Hz). Many later studies used visual analysis of the high-pass filtered signal partly in an effort to get acquainted with the characteristics of the high frequency activity. Such a visual analysis is extremely time consuming if one considers that it is necessary to display approximately 1 second per screen and that about 10 channels can be displayed at a time. If an intracerebral EEG includes 150 channels and if 10 minutes of EEG are to be analyzed, this corresponds to 9,000 screens and many hours of work. Visual analysis is, of course, subjective, and different raters do not have a good level of agreement on what they identify as HFOs (Spring et al., 2017).
Given the issues with visual analysis, many automatic analysis methods were developed and a recent review describes their main characteristics (Remakanthakurup Sindhu et al., 2020). Although the approaches differ, the fundamental principle of looking for oscillating activity in the correct frequency range and above the background remains. Most methods first high-pass filter the EEG, but it must be noted that this filtering may result in the appearance of false oscillations when sharp transients are present (Bénar et al., 2010). Some methods specifically deal with this issue (von Ellenrieder et al., 2012, 2016b), but many do not. More recently, both supervised and unsupervised machine learning algorithms are beginning to be applied to the issue of generically detecting HFOs (Blanco et al., 2010; Hagen et al., 2021).
A major issue with automatic detection is the definition of the gold standard: What should be detected and what should not? We saw above that visual inspection is not a particularly good way to define the gold standard because of high inter-rater variability. Rather than defining HFOs with qualitative and subjective characteristics, another approach to automatic detection is to define HFOs as whatever is detected by a particular detector. When taking this approach, it is critical that the EEG analyzed by the detector be of high technical quality and free of artifacts. The main artifacts that cause false HFO detections are EMG, visible on superficial contacts of iEEG electrodes, and technical artifacts related to transmission or electrode problems, which usually appear as very fast glitches. These can be quite easily identified by visual inspection of the original trace and the necessary sections or channels excluded from analysis, taking a “generous” approach of excluding any suspicious activity. It is also possible to include artifact-sensing features to the detection algorithm in case some small artifacts were missed visually. The “clean” EEG segments can then be subjected to automatic analysis, with the view that no visual validation is required. This is the approach taken by Nevalainen et al. (2020). It allows the analysis of much longer sections than the few minutes typically used when visual analysis is involved, and it is not subjective.
When comparing automatic detectors, one can be faced with the problem that different detectors may be set at different sensitivity levels and one detector may detect many more events than another. The absolute rate of HFOs (number per minute) has not been found to be of specific significance and what matters are the relative rates across channels within a patient or across patients within a study. When comparing two detectors for one patient, one can simply compare channel ranking: Do the same channels have most HFOs and the same channels have least HFOs in two different detectors (Zelmann et al., 2012)?
In humans, the first report of HFOs dealing with the localization of epileptogenicity were performed with microelectrodes; they indicated that HFOs were more frequent in the hippocampus generating seizures than in atrophic but not epileptogenic hippocampus (Staba et al., 2004). The combination of this finding and the ability to record HFOs with macroelectrodes at seizure onset (Jirsch et al., 2006) and in the interictal period (Urrestarazu et al., 2007) set the stage for the many subsequent studies that have tried to establish the relationship between the localization of HFOs and the localization of different facets of the epileptic tissue, epileptogenic zone, seizure-onset zone, irritative zone, and epileptogenic lesion. The field was most recently reviewed by Frauscher et al. (2017). The review concluded that many studies, as well as a meta-analysis, demonstrated that HFOs were most frequent in the epileptogenic zone of patients evaluated with intracerebral electrodes. It also concluded that larger and prospective studies were needed to assess the actual clinical applicability. In the following discussion, we will emphasize selected recent findings from iEEG and leave out, for the sake of brevity, the less common studies of HFOs during seizures and also the use of scalp-recorded HFOs in focus localization.
The prospective multicenter study of Jacobs et al. (2018), including patients investigated with depth electrodes, with subdural grids and with acute electrocorticography, confirmed that HFOs were more frequent in the resected tissue of seizure-free patients than outside resected tissue but failed to demonstrate the possibility that these results could be used practically in individual patients. Indeed, several patients were seizure-free despite the fact that the region with most HFOs was not removed, and prognostication of seizure outcome from the removal of HFO-generating regions was only correct in 67% of patients. Multiple factors were investigated to try to explain these disappointing results, but none were found.
The study of Fedele et al. (2017b) showed great promise for the combined presence of ripples and fast ripples in multiple 5-minute sections to predict surgical outcome at the individual level in a group of 17 patients. In an acute electrocorticography study with a group of 54 patients (van ‘t Klooster et al., 2017), it was found that fast ripples present before the resection and persisting after the resection were a very solid predictor of poor surgical results. In another relatively small acute electrocorticography study with 16 patients (Weiss et al., 2018), it was shown that a single fast ripple in the non-resected tissue had a positive predictive value of the absence of seizure freedom of 86%.
We described earlier approaches that take into consideration the rate of occurrence of HFOs in the different parts of the healthy brain to define as abnormal HFO rates only if they are beyond the normal rates. These approaches, used in two large studies with 135 and 151 patients, increased significantly several measures reflecting the accuracy of prediction of surgical outcome (Kuroda et al., 2021; Zweiphenning et al., 2022). Both studies noted that the correction was effective when using ripples but did not improve classification with fast ripples. This is not surprising since fast ripples are very rare in healthy brain tissue (Frauscher et al., 2018).
We also saw above that some studies used fast ripples, but they had to rely on very low rates, sometimes on one fast ripple to declare a region abnormal; this is because fast ripples are generally infrequent, and this may cause unreliable measurements. Fast ripples have generally been found to be specific to epileptic tissue, but lacking in sensitivity, as they are often absent (Roehri et al., 2018). The studies relying on the more frequent ripples were faced with a necessary correction because of the presence of physiological ripples. Another approach can be taken: use fast ripples but analyze long recordings to increase measurement reliability (of course, this cannot be done with acute corticography). Studies having found fast ripples insensitive in the detection of the epileptogenic zone usually analyzed a few minutes of EEG (e.g., Roehri et al., 2018), as was recommended in the early days of HFO investigations (Zelmann et al., 2009). The study of Nevalainen et al. (2020), with 43 patients, performed an automatic fast ripple detection of a whole night of iEEG recording, after visual exclusion of any artifactual section. This considerably increased prediction accuracy compared to the analysis of shorter recordings. The study also found that if no channel showed a relatively high fast ripple rate, patients were very likely to have a poor outcome; this was interpreted to mean that the epileptogenic zone may have been missed by the implantation scheme in these patients. Fast ripples, but not ripples were also found to predict the epileptogenic lesion in patients with dual pathology (Schönberger et al., 2020).
In addition to the rate of occurrence that has been discussed so far, HFOs can also be analyzed from the point of view of propagating events creating HFO networks, and by analyzing the influence of other EEG rhythms on their occurrence. The propagation pattern of HFOs in iEEG has been studied in children (Tamilia et al., 2018) and in adults (González Otárula et al., 2019). Both studies found that the “source” oscillations were a better indicator of the epileptogenic zone than regions to which oscillations propagate. González Otárula also found that the channel with this source oscillation was as good as the channel with the highest HFO rate at finding the epileptogenic zone. The interaction between HFOs and lower EEG frequencies has been studied by determining if specific EEG waves (e.g., delta waves, spindles) influence HFO occurrence or by measuring phase-amplitude coupling, where activity in high frequencies, rather than individual HFO events, are related to activities at lower frequencies. Frauscher et al. (2015) found that the delta waves of slow-wave sleep facilitated HFO occurrence but with a different phase when separating epileptic ripples from physiologic ripples (von Ellenrieder et al., 2016). It was also found that phase-amplitude coupling between ripples and delta or theta rhythms is more marked in the seizure-onset region or the resection region than in other regions (Amiri et al., 2016; Weiss et al., 2016; Motoi et al., 2018).
It is quite clear at this point that HFOs, particularly fast ripples by themselves or in combination with ripples, or ripples corrected for physiological events, can play an important role in the multimodal presurgical search for the epileptogenic zone.
Are pathological HFOs linked to anatomical abnormalities of the epileptic brain? In TLE, for instance, increased fast ripple to ripple ratios can be associated with unilateral and bilateral hippocampal atrophy in some clinical series (Ogren et al., 2009; Řehulka et al., 2019; Staba et al., 2007). At a finer scale, the rate of fast ripples exhibits some association with the presence of hippocampal sclerosis (HS), but correlations with atrophy do not necessarily apply (Agudelo Valencia et al., 2021). Importantly, simultaneous micro and macro electrode recordings in the human hippocampus suggest that fast ripples can be actually regionalized (Worrell et al., 2008), which possibly explains disparate results.
Possibly, the presence of HFOs can help to more precisely delineate structural alterations and assist in the pathophysiological diagnosis (van ’t Klooster et al., 2017). It is also tempting to speculate that these anatomical abnormalities may indeed facilitate the emergence of abnormal forms of oscillations by additional mechanisms to those previously mentioned. Recent data suggest there are substrates for correlation between features of fast ripples and the underlying anatomical alterations (Bernardo et al., 2018; Kerber et al., 2014), even though previous reports suggest they mark epileptogenicity rather than lesion type (Jacobs et al., 2009).
In TLE, microcircuit differences resulting from the different types of HS may strengthen recurrent excitatory pathways and promote abnormal oscillations. When no HS is present, connectivity between the hippocampus and the EC is tightly regulated by local circuit interneurons and inter-regional connections (Fig. 13–2A). This includes direct EC inputs to the DG, CA2, and CA1 regions coordinated by intra-hippocampal activity along the trisynaptic circuit. In physiological conditions, synaptic inhibition acts to limit hyperexcitable responses and properly gates activity flowing through the different regions (Basu et al., 2016; Fernandez-Lamo et al., 2019; Nasrallah et al., 2019; Srinivas et al., 2017; Valero and de la Prida, 2018).
Potential relationship between anatomopathological entities and fast ripples or high-frequency oscillations (HFOs). A. In the hippocampus with no HS, connectivity across areas is preserved. B. In type 1 HS, there is pronounced cell loss in CA4 and CA1, (more...)
In type 1 HS, breakdown of the DG gate causes abnormally regulated responses with both hyperexcitation and hyperinhibition of downstream targets (Sloviter et al., 2006; Sutula et al., 1988; Wozny et al., 2005) (Fig. 13–2B). Partial loss of CA3 leaves the more resilient CA2 area disinhibited, which together with associated mossy fiber sprouting into this region promotes substantial dysregulated responses (Ang et al., 2006; Freiman et al., 2021). Actually, in experimental TLE models, in vitro fast ripples correlate with the degree of mossy fiber sprouting at the DG (Foffani et al., 2007). Given massive neuronal loss of CA1 neurons, physiological ripples are hardly visible and the subiculum would take the lead in generating sharp-wave ripples (Alvarado-Rojas et al., 2015; Cohen et al., 2002; Huberfeld et al., 2011).
In contrast, in type 2 HS the hilus, CA4, CA3, and CA2 region may still provide some level of coordination in responses to EC inputs (Fig. 13–2C). In experimental models of epilepsy, loss of CA1 pyramidal cells is cell-type specific and progressive (Cid et al., 2021), and both patchy and laminar inhomogeneities are described in the region (Prada Jardim et al., 2018). Local inhomogeneities and variability along the longitudinal axis (Thom et al., 2012) will contribute to intermixed ripples and fast ripple events (Alvarado-Rojas et al., 2015; Valero et al., 2017).
Finally, in patients with dual pathologies there are two potentially epileptogenic foci interacting with each other (Fig. 13–2D). The precise functional organization of different lesions and their specific brain-wide connectivity, together with specific mechanisms at each place, will all shape the form and mechanisms of pathological HFOs (Cepeda et al., 2020; Cid et al., 2014; Kannan et al., 2016; Kerber et al., 2014; Tóth et al., 2018).
In an early study of the relationship between epilepsy treatment and HFOs (Zijlmans et al., 2009), it was demonstrated that HFO rates tended to fluctuate in accordance with medication changes, increasing when medication is reduced, thus following the effect of medication on seizures; this is contrary to the effect of medication on interictal spikes. Most studies on the relationship between seizure occurrence or treatment and HFOs, however, have been carried out on scalp EEG. The review of Frauscher et al. (2017) also concluded that scalp-recorded HFOs showed promise as a marker of drug treatment and of epilepsy severity, particularly in children. A relationship between scalp EEG ripple rates and treatment effectiveness was demonstrated in children with West syndrome (Kobayashi et al., 2015). van Klink et al. published an interesting study linking the presence of ripples in Rolandic epilepsy with the likelihood of seizures occurring (van Klink et al., 2016). In children with the tuberous sclerosis complex, fast ripples were only found in those with epilepsy (Bernardo et al., 2018). This study and that on Rolandic epilepsy indicate the potential of scalp HFOs in children to be a biomarker of epilepsy and of epileptogenesis. This potential was clearly put in evidence when scalp ripples were shown to be a strong predictor of an eventual epilepsy diagnosis in patients having a first unprovoked seizure (Klotz et al., 2021). Such a situation reflects that HFOs can be a predictor of epilepsy and therefore are a biomarker of epileptogenesis. This study also demonstrated that interictal spikes did not have this ability. HFOs were recorded in the EEG of neonates with seizures, but they were not frequent enough to be of clinical applicability (Noorlag et al., 2021).
Since the first studies on HFOs performed by Bragin et al. (1999) in the kainic acid model of focal epilepsy, numerous researchers have studied HFOs in different animal models. In the kainic acid model, it was demonstrated that HFOs occur shortly after kainic acid injection in the animals who later develop chronic seizures, whereas they were rare in the animals that did not develop seizures (Bragin et al., 2004), demonstrating here again the HFOs’ ability to be a biomarker of epileptogenesis. In a subsequent study with the same model, it was shown that a more widespread distribution of HFOs was also predictive of eventual occurrence of spontaneous seizures (Li et al., 2018). In the hippocampus, physiological ripples are associated with basic processes of activity replay in service for memory recollection and consolidation (Norman et al., 2019). In epileptic rodents, where ripples and fast ripples coexist, firing reactivation becomes randomized during these events and multimodal spectral measurements correlate with deficits in episodic-like memory tasks (Valero et al., 2017). This suggests that even though some HFOs may look spectrally similar to normal ripples in terms of frequencies, the mesoscopic structure of associated neuronal firing may be fundamentally impaired.
HFOs also occur just prior to seizures and during seizures. They are indeed occurring during LVF ictal discharges triggered with 4-aminopyridine (4AP) in vitro (Shiri et al., 2015) and in vivo (Salami et al., 2015), as well as before and during chronic seizures in the kainic acid and pilocarpine model of mesial temporal lobe epilepsy (Behr et al., 2017; Bragin et al., 2005; Lévesque et al., 2012), which indicates that these seizures could be linked to excessive GABAergic neurotransmission (Avoli et al., 2016). This hypothesis is further supported by recent optogenetic studies in which the activation of parvalbumin-positive GABAergic interneurons triggers LVF ictal discharges in vitro (Shiri et al., 2015, 2016). HYP seizures may instead be related to the involvement of glutamatergic principal cells since they are mostly associated with fast ripples (Behr et al., 2017; Lévesque et al., 2012; Salami et al, 2015). Optogenetic activation of CAMKII-positive glutamatergic principal cells indeed triggers HYP ictal discharges associated to fast ripples in vitro (Shiri et al., 2016). This points to two possible mechanisms of generation of these two seizure types and the importance to highlight the critically different role of cell-type-specific subcircuits in ictogenesis (Khoshkoo et al., 2017). Their mechanistic implications are discussed at length in Lévesque and Avoli (2019).
HFOs were also studied in humans with the two types of temporal lobe seizures, LVF and HYP, in an effort to see if parallel results were found in animals and in humans (Schönberger et al., 2019). The findings indicate a connection between experimental and human epilepsy regarding fast ripples: their association with the HYP onset suggests that out-of-phase firing of different pyramidal cell clusters underlying fast ripples contributes specifically to generation of these seizures, rather than to seizures with low-voltage fast onsets (Ibarz et al., 2010; Jiruska et al., 2010).
In the last 20 years, HFOs have appeared as a distinct phenomenon clearly linked to many facets of epilepsy. They can be recorded with microelectrodes, iEEG, and scalp electrodes, allowing them to be studied widely, but they show a multiscale invariance that remains incompletely understood. Fast ripples appear to be the pattern most specific to epileptic tissue, making them valuable markers of the epileptogenic zone; they are infrequent and may require long recordings for a reliable assessment. They have only been recorded on scalp EEG in infants, limiting their utilization in children and adults to intracerebral studies. Ripples are much more abundant and commonly recorded in scalp EEG but are not as specific as fast ripples to epileptic tissue because physiological ripples are common, and we do not have an unambiguous way of separating physiological from pathological ripples, although different methods have been proposed to deal with this issue. It is remarkable though that physiological ripples appear to be rare in scalp EEG, where ripples may become an important biomarker of epilepsy and epileptogenesis. Our current understanding of the mechanisms of generation of ripples and fast ripples may help us understand the imbalances that may lead to the appearance of different seizure types.
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...
