Interictal discharges in the human brain are travelling waves arising from an epileptogenic source

Abstract

While seizure activity may be electrographically widespread, increasing evidence has suggested that ictal discharges may in fact represent travelling waves propagated from a focal seizure source. Interictal epileptiform discharges (IEDs) are an electrographic manifestation of excessive hypersynchronization of cortical activity that occur between seizures and are considered a marker of potentially epileptogenic tissue. The precise relationship between brain regions demonstrating IEDs and those involved in seizure onset, however, remains poorly understood. Here, we hypothesize that IEDs likewise reflect the receipt of travelling waves propagated from the same regions which give rise to seizures.

Forty patients from our institution who underwent invasive monitoring for epilepsy, proceeded to surgery and had at least one year of follow-up were included in our study. Interictal epileptiform discharges were detected using custom software, validated by a clinical epileptologist.

We show that IEDs reach electrodes in sequences with a consistent temporal ordering, and this ordering matches the timing of receipt of ictal discharges, suggesting that both types of discharges spread as travelling waves. We use a novel approach for localization of ictal discharges, in which time differences of discharge receipt at nearby electrodes are used to compute source location; similar algorithms have been used in acoustics and geophysics. We find that interictal discharges co-localize with ictal discharges. Moreover, interictal discharges tend to localize to the resection territory in patients with good surgical outcome and outside of the resection territory in patients with poor outcome. The seizure source may originate at, and also travel to, spatially distinct IED foci.

Our data provide evidence that interictal discharges may represent travelling waves of pathological activity that are similar to their ictal counterparts, and that both ictal and interictal discharges emerge from common epileptogenic brain regions. Our findings have important clinical implications, as they suggest that seizure source localizations may be derived from interictal discharges, which are much more frequent than seizures.

The significance of interictal epileptiform discharges (IEDs) in patients with epilepsy is incompletely understood. Diamond et al. show that both ictal discharges and IEDs are travelling waves which emerge from common epileptogenic brain regions. These findings may underscore a new role for IEDs in seizure source localization.

See Schevon and Michalak (https://doi.org/10.1093/brain/awad108) for a scientific commentary on this article.

Introduction

In focal epilepsy, seizures tend to be stereotyped within a given patient, with relatively consistent patterns of hypersynchronous activity observed in EEG recordings.^1,2 The concept of the seizure focus arose from these observations, in concert with the ability of focal surgical resections to lead to seizure freedom. Visual identification of the seizure onset zone through intracranial EEG (iEEG) monitoring is currently the gold standard for approximating the source of epileptiform activity, particularly in non-lesional patients. Surgical resections guided by such clinical inspections of the iEEG recordings have offered patients with drug-resistant focal epilepsy improved chances for seizure freedom.

Despite the relative successes of this approach, however, recent studies based on microelectrode recordings have provided evidence that ictal discharges seen in macroelectrode recordings may not necessarily indicate brain regions that are actively seizing. The recorded discharges may instead reflect receipt of activity from a seizure focus either in the form of travelling waves along the cortical surface or through white matter propagation.^3-10 Rapid propagation of ictal discharges and slow propagation of the sources of this activity appear to be relatively stereotyped within a given patient across seizures, suggesting preferred propagation pathways that are likely driven by a combination of proximity to the source of the activity, functional connections and susceptibility to recruitment.^11-13 These observations together suggest that the source of epileptiform activity may be much more spatially circumscribed than the observed seizure onset zone.

Interictal epileptiform discharges (IEDs) are another electrographic manifestation of excessive hypersynchronization of cortical activity that occur between seizures and are considered a marker of potentially epileptogenic tissue.¹⁴ Even within individual patients, IEDs demonstrate significant spatial and temporal variability, likely related to changes in local and global brain states including state of arousal or temporal proximity to seizures.^14-18 Despite this variability, IEDs tend to involve a relatively consistent spatial core region, suggesting that the observed interictal discharges may also represent receipt of signal from a focal source.^14,19-22 Indeed, recent research has suggested that interictal discharges may also spread as travelling waves, much like ictal discharges.¹⁰ The precise relationship between brain regions demonstrating IEDs, those involved in seizure onset and subsequent propagation and the potential sources of this activity, however, remains poorly understood.^14,23

Here, we hypothesize that focal sources of epileptiform activity emit travelling waves that underlie both interictal and ictal activity observed in iEEG recordings. We reasoned that if both ictal and interictal discharges reflect the receipt of travelling waves emitted from the same focal source, we should expect similar timings of discharges in the interictal and ictal data. We show that the temporal order of receipt of interictal discharges across sequences of electrodes is relatively consistent over time and mirrors the order observed during ictal discharges. We then use the timing of interictal discharge receipt to localize the source of epileptiform activity. In previous work, we used phase differences of ictal discharges at adjacent electrodes to localize the source of seizure activity, and we adopt a similar approach here.⁹ Similar algorithms have been used extensively for signal detection in acoustics and radar^24-28 and for earthquake epicentre detection in geophysics.^29-33 We show that the source of IEDs, when determined in this way, co-localize with the source of seizure activity and with the resection territory in patients with good outcome.

Materials and methods

Participants

Participants underwent a surgical procedure for placement of intracranial electrodes followed by iEEG monitoring in anticipation of a surgical resection for drug-resistant epilepsy. We performed all surgical procedures and iEEG monitoring at the Clinical Center at the National Institutes of Health (NIH; Bethesda, MD). In each case, the clinical team determined the placement of the contacts to localize epileptogenic regions. Outcome was evaluated using the Engel class.³⁴ Here, Engel class 1 reflects freedom from disabling seizures. Engel class 2 reflects rare disabling seizures. Engel class 3 reflects a worthwhile improvement relative to prior to surgery. Finally, Engel class 4 reflects no worthwhile improvement compared to prior to surgery. All patients had at least 1 year of follow-up. Engel classes were assessed at two years or time of last follow-up, whichever was earlier. Mean time from surgery at outcomes assessment was 20.73 ± 5.87 months. Engel class outcomes and follow-up times are provided in Table 1. The research protocol was approved by the Institutional Review Board, and informed consent was obtained from the participants. Data were analysed using custom MATLAB scripts (Mathworks, Natick, MA). All data are reported as mean ± standard deviation unless otherwise reported.

Table 1.

Patient demographic information

Patient ID	Sex	Age	Typical seizure types	MRI	Procedure	Pathology	Engel class	Time to outcome assessment (months)
1	F	33	FIAS	NL	LT topectomy	MDG	4	24
2	F	30	FIAS	NL	L SAH	HS, MDG	2b	25
3	F	59	FIAS	L MTS	L ATL	HS	1a	27
4	F	29	FAS, FIAS	R MTS	R ATL	HS	2b	25
5	F	49	FAS, FIAS, FTBTCS	L MTS	L SAH	HS	3a	24
6	F	34	FIAS	L MTS	L ATL + LT topectomy	HS	1a	12
7	M	45	FIAS, FTBTCS	NL	RF topectomy	MDG	1a	28
8	M	33	FIAS, FTBTCS	NL	L ATL	MDG	3a	22
9	M	39	FIAS, FTBTCS	R > L MTS	R ATL	HS	1a	23
10	F	35	FIAS, FTBTCS	B MTS	L ATL	HS, MDG	1a	17
11	F	43	FAS, FIAS	s/p R ATL	RT topectomy	MDG	4	29
12	M	33	FAS, FIAS	B PVNH	L ATL	MDG	1a	28
13	F	43	FIAS, FTBTCS	NL	L ATL	HS, MDG	1a	12
14	F	40	FIAS	R F FCD	RF lesionectomy	FCD	3a	24
15	F	28	FIAS	L MTS	L ATL	HS, MDG	1c	24
16	F	30	FIAS	RP encephalo-malacia, R MTS	R ATL	HS	1b	24
17	M	34	FTBTCS	NL	RP topectomy	MDG	2b	24
18	M	20	FIAS, FTBTCS	NL	LT topectomy	MDG	3a	24
19	M	27	FIAS, FTBTCS	L LT LGG	LT lesionectomy	LGG	1a	12
20	M	29	FAS, FTBTCS	NL	RP topectomy	MDG	4	14
21	M	32	FIAS, FTBTCS	B PVNH	L ATL	MDG	2b	24
22	F	33	FIAS	RF FCD	RF lesionectomy	FCD	1a	25
23	M	21	FIAS, FTBTCS	NL	R ATL	MDG	1a	28
24	M	57	FIAS, FTBTCS	RT encephalocele	R ATL	MDG	1a	24
25	F	53	FIAS	NL	R ATL	HS, FCD	2b	24
26	M	45	FIAS	R > L perisylvian polymicrogyria	R ATL	HS, MDG	1b	24
27	F	21	FIAS, FTBTCS	RP encephalo-malacia	RP lesionectomy	Post-traumatic	3a	16
28	M	34	FIAS, FTBTCS	L MTS	L ATL	HS, MDG	2b	11
29	M	22	FIAS	NL	RF topectomy	MDG	3a	24
30	M	46	FIAS	B OF encephalo-malacia	L SAH	Gliosis	3a	26
31	M	24	FIAS, FTBTCS	NL	L ATL	MDG	2b	12
32	M	36	FIAS	NL	LT topectomy	HS, MDG	2d	25
33	M	34	FIAS	NL	R ATL	HS, FCD	1a	15
34	F	57	FAS	s/p R ATL	RT topectomy	Gliosis	1a	12
35	F	33	FIAS	R MTS	RT topectomy	MDG	3a	24
36	M	21	FAS	NL	RF topectomy	FCD	1b	12
37	M	26	FIAS, FTBTCS	NL	LT topectomy	MDG	2b	12
38	M	39	FAS, FIAS, FTBTCS	2 RP FCDs	RP lesionectomy	FCD	1a	18
39	F	34	FIAS, FTBTCS	NL	LT topectomy	Gliosis	1b	19
40	M	40	FAS, FIAS, FTBTCS	NL	RP topectomy	MDG, FCD	4	12

ATL = anterior temporal lobectomy; B = bilateral; F = frontal; F2BTCS = focal to bilateral tonic–clonic seizures; FCD = focal cortical dysplasia; FIAS = focal impaired awareness seizure; HS = hippocampal sclerosis; L = left; LGG = low-grade glioma; MDG = microdysgenesis; MTS = mesial temporal sclerosis; NL = no lesion; P = parietal; PVNH = periventricular nodular heterotopia; R = right; SAH = selective amygdalo-hippocampectomy; T = temporal.

IED detector and IED sequences

For detection of interictal epileptiform discharges, we used a custom-built IED detector, used in previous research and validated by a clinical epileptologist.¹⁸ The IED detector works by detecting large fluctuations in the iEEG trace voltage.^35,36 For each electrode, we z-scored the iEEG traces over the duration of the 2-h epoch. We then identified negative deflections in each iEEG trace with a prominence of at least 3σ and with width less than or equal to 50 ms. When such a deflection was discovered, we searched for an additional positive deflection with a peak prominence of at least 3σ that occurs within 100 ms of the negative peak, either before or after. We additionally required that the difference between positive and negative peak height was at least 9σ. The time of the negative peak was marked for further analysis.

We defined an IED sequence as an event in which IEDs were detected in three distinct electrodes within the same 100 ms window.^37-39 Additional details are provided in the Supplementary material.

Source localization

We hypothesized that there exists one or multiple focal sources which give rise to pathologic activity, including interictal as well as ictal discharges. IEDs may thus represent a single discharge that is released from a focal source and that radially spreads outward over the surface of the cerebral cortex over macroscopic-scale distances.^7-10 These waves thus reach different electrodes at different times. With this conceptualization, we should then be able to use the time of receipt of IEDs at different electrodes to estimate the location of the source of the pathological waves. Similar approaches have been used extensively in geophysics, acoustics and radar, and are referred to as multilateration algorithms.^24-33

Consider a source, s, which releases a signal that spreads outward at speed c, and reaches two electrodes, e_i and e_j. The signal reaches e_i with the time delay τ_i and e_j with the time delay τ_j. We are therefore able to express the difference in distance between s and e_i, on the one hand, and s and e_j, on the other, as:

$$d_{s,e_i}-d_{s,e_j}=c\cdot ({\tau }_i-{\tau }_j)$$

The relative distance between the source and the two electrodes can therefore be expressed using the difference in time delays τ_i − τ_j required for the signal to propagate from the source to the two electrodes.²⁸ The estimate $d_{s,e_i}-d_{s,e_j}$ can then be used to constrain the location of the source to a hyperbola over the brain surface. We model the brain surface for each patient as a geodesic mesh based on preoperative MRI. With multiple participating electrode pairs and multiple estimates of relative distance, we are able to create multiple hyperbolae, each of which contains the putative source. We then solve for the source location as the intersection of these hyperbolae (see Supplementary material).

Our algorithm assumes that the signal spreads outward evenly, in a concentric circular fashion, over the grey matter. This assumption is challenged in a simulation analysis in Supplementary Fig. 4.

Data and code availability

Data not provided in the article because of space limitations may be shared (anonymized) at the request of any qualified investigator for purposes of replicating procedures and results.

Our custom scripts used for IED detection and source localization are publicly available for download at https://research.ninds.nih.gov/zaghloul-lab/downloads.

Results

We examined iEEG recordings in 40 participants (35.40 ± 10.20 years old; 18 females) with drug-resistant epilepsy who were monitored for seizures with subdural and/or depth electrodes (Table 1). All patients had focal epilepsy and had seizures during the recording period. All patients proceeded to surgery for surgical resection of the brain regions thought to give rise to seizures and had at least one year of clinical follow-up. Seizure data were inaccessible in four patients due to the files becoming lost in our database; these patients were excluded from analyses which required the presence of seizure data. Engel class outcomes were assessed at 2 years (see ‘Materials and methods’ section). Of the 40 total participants, 19 achieved an Engel class 1 outcome following surgical resection, 9 achieved an Engel class 2 outcome, 8 achieved an Engel class 3 outcome and 4 achieved an Engel class 4 outcome. Mean follow-up was 20.73 ± 5.87 months. Engel class outcomes and follow-up times are provided in Table 1. For one patient (Patient 27), outcome changed from Engel class 3a to 1c at 60 months. This patient’s 16-month outcome, 3a, was considered in the analysis. No other patient went from non-seizure-free to seizure-free, or vice versa, between the 2-year outcome and long-term outcome. For the purposes of analysis, we divided our participants into those with good 2-year post-surgical seizure outcome (Engel 1, freedom from disabling seizures) and those with poor post-surgical seizure outcome (Engel 2 or greater, disabling seizures persist).

Interictal discharge sequences are stereotyped

We were interested in examining and understanding the origin of sequences of IEDs detected by the recording electrodes. We hypothesized that IEDs emerge from focal brain regions that emit travelling waves, which spread outward in a radial fashion. If this were the case, then we would expect to observe sequences of IEDs in recording electrodes that exhibit a consistent and predictable ordering.

To test this hypothesis, we used an automated IED detector to identify all sequences of IEDs recorded in the implanted electrodes (Fig. 1A and B; see ‘Material and methods’ section). We retained all sequences (or subsequences; see Supplementary material) of exactly three IEDs that occurred within a single 100 ms window (see ‘Materials and methods’ section). In an example participant, we observed many different IED sequences (Fig. 1C). Each sequence is defined by a specific order of activity across three electrodes, and we can quantify the probability of observing each specific three-member sequence in every participant. We observed that IED sequences involving a set of electrodes often exhibit a predictable and consistent temporal ordering, and that it is unusual to find the same electrodes involved in a sequence with a differing order.

Figure 1.

Interictal discharge sequences are stereotyped. (A) Implanted subdural and depth electrodes are shown on a cortical surface reconstruction in a single representative participant. (B) An example discharge is shown, detected in three electrodes in the anterolateral temporal lobe (ALT). (C) The most commonly occurring IED sequences involving three electrodes are shown for this participant. Each sequence begins with an IED observed in the first electrode (inner ring), followed by the second and third electrode, shown in the second and third rings respectively. Arc length of each ring represents the probability of observing an IED in that electrode in the sequence. Only the 10 electrodes that are most frequently observed to lead an IED sequence are represented in the first ring. Similarly, the second and third rings only display the 10 most frequently involved electrodes in the second and third position of the IED sequences, given the IED observed in the first electrode. Electrodes are coloured based on their membership in each subdural grid or strip. (D) Likelihood of observing one of the 20 most commonly occurring IED sequences across participants (black, mean likelihood 4.8 × 10^–03 ± 4.9 × 10^–03). Likelihood of observing IED sequences in the same electrodes but in a different order is significantly less across participants (red, shuffled, 1.6 × 10^–03 ± 1.9 × 10^–03). Observing IED sequences in randomly selected sets of three electrodes across participants are even less likely (grey, random, 3.1 × 10^–04 ± 2.9 × 10^–04). Error bars reflect standard error of the mean.

In order to quantify the extent to which the temporal orderings of the IED sequences are predictable, we identified the 20 most commonly observed three-member sequences in each participant. The likelihood of each of these sequences was computed, by taking the count and dividing by the total number of three-member sequences in each patient. Mean likelihood for the top 20 sequences was 4.8 × 10^–03 ± 4.9 × 10^–03. This analysis was then repeated after permuting the order of the top 20 sequences. For permuted top sequences, mean likelihood was 1.6 × 10^–03 ± 1.9 × 10^–03. Finally, it was repeated after retrieving 100 random sequences of three electrodes. We only chose sets of electrodes that were located within 30 mm of one another, to address the possible confound in which, in electrodes that are far from each other, sequences might be rare simply by virtue of the electrode spacing. Mean likelihood for random sequences was 3.1 × 10^–04 ± 2.9 × 10^–04. There was a significant difference between at least two groups [one-way ANOVA, F(2,117) = 23.76, P < 0.0001]. Tukey’s Honest Significant Difference (HSD) test for multiple comparisons revealed significant differences between the true top 20 sequences and their permuted counterparts (P < 0.0001) and between the top 20 sequences and the random sequences (P < 0.0001), but not between the permuted and random sequences (P = 0.15). These results are illustrated in Fig. 1D. Here, in each patient, likelihoods of sequences were binned, with the same binning scheme used in all patients. Bin counts were then averaged across patients. These data demonstrate that the observed IED sequences occur more frequently than the permuted or random sequences. This suggests that the observed IED sequences may in fact arise from focal regions of pathological tissue, and therefore arrive at the recording electrodes with reliable orderings.

Temporal order of IED sequences is preserved in the ictal data

We wished to test the hypothesis that ictal and interictal discharges arise from similar regions.¹⁰ If this were true, then we would expect that the temporal ordering of interictal discharges should be similar to that of the discharges observed during seizures.

In the same example participant as discussed in Fig. 1A–C, we identified two commonly occurring IED sequences. In the first sequence, the electrodes involved reside in the right parietal lobe, while in the other, the electrodes are in the right anterior temporal lobe. We then examined the time series captured from these same electrodes during a typical seizure (Fig. 2A). Earlier during the seizure, ictal discharges are present in the parietal electrodes but not in the temporal electrodes. Later, ictal discharges are present in the temporal lobe but not in the parietal lobe. Close examination in one-second epochs reveals that the ictal discharges appear to arrive with a reliable ordering that matches the order observed in the respective IED sequences. The change in ictal discharge patterns suggests that the source of seizure activity may migrate between distinct foci.⁹ Moreover, at each focus, it appears that the source of seizure activity emits discharges that match the order of observed IED sequences.

Figure 2.

Temporal order of IED sequences is preserved in the ictal data. (A) Ictal time series for two sets of three electrodes in a single seizure in an example participant. Each set of three electrodes represents a common IED sequence. One set is in the parietal lobe (PG) and the other is in the temporal lobe (ALT). Left: cortical reconstruction of the electrode locations. Two magnified 1-s epochs (bottom) are shown, captured at 23–24 s and 53–54 s, respectively, following seizure onset (top). Discharges from each set of electrodes arrive in a consistent temporal order. (B) The latencies of all observed IED sequences involving the same parietal and temporal sets of three electrodes. Latencies are arranged by the difference in time between an IED appearing on the first and second electrode members of the commonly observed IED sequence, and between the second and third member (black rings). Points in the bottom left quadrant therefore represent a temporal order identical to the commonly observed IED sequence in those electrodes. Latencies in the ictal discharges observed in the same electrodes during seizures demonstrate a similar order as the common IED sequences (filled circles, colour-coded by duration after seizure onset). (C) Number of IED sequences, and their relative latencies, observed in all sets of three electrodes that were involved in the 20 most commonly observed IED sequences in this participant (left). Number and relative latencies of ictal discharges observed in the same sets of electrodes (right). (D) Mean number of IED sequences (left) and ictal discharges (right), and their relative latencies, observed in each of the 20 most commonly observed IED sequences per patient, aggregated across all participants.

We directly compared the order of discharges in the ictal and interictal data. To do so, we first chose one particular common IED sequence of three members. We then retrieved every IED sequence in this patient involving the same three electrodes, but in any ordering. We then extracted the latency between IED discharges observed in the first and second electrode members of the common IED sequence, and between the second and third members. Because any ordering of the sequence is permitted, latencies can be positive or negative. In the two example sets of electrodes from the same participant, most combinations of IED latencies arise in the bottom left quadrant, suggesting that both latencies are negative. In other words, most IED latencies exhibit a temporal order similar to the commonly observed IED sequence, although all combinations of latencies across the set of three electrodes are also seen (Fig. 2B). We then examined sequences of ictal discharges detected in the same sets of electrodes. Similarly to the IED sequences, we considered ictal discharges involving the same electrodes, but in any order, and extracted the latencies between constituent members. The ictal discharges appear to exhibit a temporal ordering that is similar to that of the interictal discharges.

We repeated this analysis for the 20 most common three-member IED sequences observed in this participant. For each sequence, we took all sequences involving those electrodes, but in any order, and retrieved the latencies between discharges in the first and second members of the common sequence, and between the second and third members. We aggregated these latencies across all 20 IED sequences. As expected, most observed latencies exhibit a temporal ordering similar to the commonly observed IED sequences (Fig. 2C, left). We then examined ictal discharges involving these same sets of electrodes. The ictal discharges also exhibit a similar combination of latencies across the involved electrodes (Fig. 2C, right). We found a similar correspondence across all participants (Fig. 2D), suggesting that the temporal order of discharges during seizures appear to match the temporal order observed during IED sequences.

We wished to determine whether the propensity of observed discharges to match the ordering of common discharges was greater than expected by chance. To study this, we only considered latencies less than or equal to 20 s, to reduce the influence of noise. For latencies less than or equal to 20, the weighted centroid of the heatmap of IED latencies (Fig. 2D, left) was −1.92, −2.42 s, for the first and second latencies, respectively. We then divided latencies into quadrants, and asked if the bottom left quadrant had the greatest mean count. Mean count in the bottom left, bottom right, top left and top right quadrants were 10.08 ± 9.69, 5.06 ± 4.70, 5.45 ± 4.45 and 2.39 ± 3.58, respectively. There was a significant difference between at least two groups [one-way ANOVA, F(3,480) = 33.17, P < 0.0001]. Tukey’s HSD test for multiple comparisons revealed significant differences between the bottom left quadrant and all other quadrants (P < 0.0001). Next, we considered the ictal discharges. For latencies less than or equal to 20, the weighted centroid of the heatmap of ictal latencies (Fig. 2D, right) was −0.87, −0.81 s. We again divided latencies into quadrants. Mean count in the bottom left, bottom right, top left and top right quadrants were 4.37 ± 4.54, 3.14 ± 2.10, 3.18 ± 2.01 and 2.11 ± 1.93, respectively. There was a significant difference between at least two groups [one-way ANOVA, F(3,480) = 12.61, P < 0.0001]. Tukey’s HSD test for multiple comparisons revealed significant differences between the bottom left quadrant and the bottom right quadrant (P = 0.005), the top left quadrant (P < 0.007) and the top right quadrant (P < 0.001).

Interictal and ictal discharges localize to similar regions

The reliability of the IED sequences, and the preserved temporal order seen in the ictal data, suggest that the involved electrodes may receive pathological waves of activity from a particular focal source. In this conceptualization, a source of epileptiform activity emits a signal that travels radially outward over the cerebral cortex and reaches different electrodes at different times, depending on their distance from the source (Fig. 3A and B).⁹ We can therefore use the time differences observed across multiple pairs of electrodes to estimate the location of this potential source (Fig. 3C; see ‘Materials and methods' section). We were interested in comparing the results from source localization using the interictal discharges versus using the ictal data.

Figure 3.

Latencies between interictal and ictal discharges can be used to estimate the location of a hypothesized source of activity. (A) We hypothesize that one or multiple focal sources (star) emit pathological waves of activity that spread radially outward over the brain surface. Discharges are detected on electrodes with different latencies; these latencies are used to estimate Δd, the relative difference in distance between the electrodes and the hypothesized source. For each pair of electrodes, this estimate constrains the location of the hypothesized source to a single hyperbola. With multiple electrode pairs, we estimate the location of the source location from the intersection of the hyperbolae. (B) A representative IED sequence in a set of electrodes in the anterior temporal lobe. The latencies between each IED, Δt, can be used to estimate Δd. (C) The estimates of Δd for each electrode pair constrain the source to a hyperbola in geodesic space. We compute the source of these travelling waves (star) as the intersection of these hyperbolae. (D) Left: All IEDs are detected in a set of three electrodes in the parietal lobe (top) and another set of three electrodes in the anterior temporal lobe (bottom). We estimate the location of the hypothesized source for every IED sequence involving these electrodes and plot the number of times the estimated source location falls within each region of interest as a heatmap. Right: All ictal discharges during seizures are detected in the same sets of electrodes. Using the phase differences of ictal discharges across electrodes, we estimate the location of the hypothesized source of ictal discharges. Colour intensity indicates the number of times an estimated source of these ictal discharges localizes to each region of interest in these brain regions and colour hue indicates the mean time from seizure onset at which the source localization occurred. For each set of three electrodes, discharges tend to localize to similar regions in interictal versus ictal states.

We examined the same two commonly occurring IED sequences in the same example participant. We identified all IED sequences involving those three electrodes, but in any order. Based on those sequences, we estimated the location of the source of IED activity for each sequence. We mapped localized points to regions of interest that were evenly spaced across the cortical surface (see Supplementary material). Results are plotted on a heatmap on a cortical surface reconstruction, where warmer colours designate greater source localizations per region of interest (Fig. 3D, left). We then examined the discharges captured during a single representative seizure in these two sets of electrodes. This was the same seizure as was considered in Fig. 2. Results are again plotted on a cortical surface reconstruction (Fig. 3D, right). Here, hue designates time from seizure onset (blue is early; yellow is late). Meanwhile, intense colours designate a greater number of source localizations per region of interest. In both sets of electrodes, localization based on the ictal discharges is similar to localization based on the IED sequences, as would be expected given the similar temporal ordering of discharges observed in the ictal and interictal data. For the parietal leads, IEDs localize to a gyrus in the inferior parietal lobule, with the strongest region of interest posterolateral to the leads. The seizure also localizes to the inferior parietal lobule, here with the strongest region of interest posteromedial to the leads. In this analysis, the distance between strongest seizure region of interest and strongest IED region of interest is 20.5 mm. However, of note, there does appear to be a secondary IED focus, which is posteromedial to the parietal leads and which coincides spatially with the strongest seizure focus. For the temporal leads, both IEDs and seizure discharges localize to the temporal pole, with strongest activity in the inferior temporal gyrus, parahippocampal gyrus and middle temporal sulcus. In this analysis, the strongest IED region of interest is the same as the strongest seizure region of interest.

We next considered all electrodes in this patient, and examined all captured IED sequences, as well as all ictal discharges captured during the same representative seizure. In this analysis, we retained all full IED sequences of length at least three, rather than length exactly three, recorded in all electrodes (see Supplementary material). Using these sequences and our source localization procedure, we identified the brain regions most commonly identified as the estimated source of the observed IED sequences (Fig. 4A, left; see ‘Materials and methods' section). We then estimated the source of ictal data at every time point in the same representative seizure, based on the relative phase of ictal discharges observed across all recording electrodes. Localization based on the ictal data during this seizure is visually similar to that based on the IED sequences in this participant (Fig. 4A, right). The seizure source appears to originate from and travel to distinct IED foci (see Supplementary Fig. 2A). This patient went on to have a right parietal lesionectomy (grey region). Outcome was Engel class 3a at 16 months. Strongest points of IED and seizure localization were both outside of the resection territory, suggesting a successful prediction of our algorithm (see Supplementary Fig. 1, Patient 27).

Figure 4.

Interictal and ictal discharges localize to similar brain regions. (A) Left: Estimated source location for all IED sequences captured in all electrodes in an example participant. Each IED sequence generates an estimated source location. Colour and size of points indicate the number of times a source was estimated to localize within 2 mm of that location across all observed IED sequences. Estimated locations are collapsed to two dimensions. Resection cavity indicated in grey. Right: Estimated source location for all ictal discharges captured in all electrodes during a seizure in the same participant. The size of each point indicates the number of times a source was estimated to localize within 2 mm of that location. Colour indicates time from seizure onset. (B) Correlation between the number of IED sequences that localized to each region of interest and the number of ictal discharges that localized to the same region of interest (Spearman’s ρ = 0.64). Each point represents one region of interest and is coloured based on the mean time from seizure onset when ictal discharges are localized to that region of interest. Axes are logarithmically scaled and normalized to the number of times discharges were observed in the region of interest containing the most localizations. (C) We compared the true correlation coefficient with the correlations observed by chance, generated by shuffling the lead labels 100 times. (D) We performed a similar shuffling procedure in each participant, comparing the true correlation coefficient (black) with the median and interquartile range of correlation coefficients generated by chance after shuffling lead labels 100 times in each participant (red). (E) Left: In every participant, we computed the Euclidian distance between the region of interest most frequently containing the estimated source location, based on the IED sequences and the region of interest most frequently containing the source location based on ictal discharges in the true data (black, histogram with 10 mm bins). We similarly computed distances between estimated IED and ictal source locations after shuffling lead labels 100 times in each participant (red). Right: The most frequently estimated location of the source of IED sequences falls within 10 mm of the most frequently estimated location of the source of ictal discharges in nine participants, significantly greater than the number observed by chance after shuffling the lead labels.

We were interested in more rigorously capturing the co-localization of interictal and ictal discharges. Therefore, we performed a similar localization procedure for all seizures recorded in this participant (eight seizures) and compared the results to the localization obtained from the IED sequences. Both ictal and interictal localized points were mapped to brain surface regions of interest. We found a strong correlation between the number of IED sequences that localized to each region of interest and the number of ictal discharge localizations to the same region of interest (Spearman’s ρ = 0.64; Fig. 4B). To confirm that this correspondence in localization between ictal and interictal activity is greater than chance while adjusting for confounding factors such as electrode placement, we used a shuffling procedure. During each of 100 iterations, we shuffled the electrode labels and recomputed IED sequences and IED localization. In other words, the mapping between raw time series and spatial location was scrambled over all electrodes in each patient’s implant. Shuffling was done prior to assembly of IEDs into sequences, so that electrodes did not need to record IEDs in order to be included in the shuffled sequences. An additional shuffling analysis, in which lead labels were shuffled after assembly into IED sequences, is considered in the legend of Supplementary Fig. 3. Ictal localization was left unchanged. The true value of ρ is in the 94th percentile of shuffled values (Fig. 4C).

We repeated this analysis in all participants and found that, in the vast majority of participants, the true correlation between IED and ictal localization is significantly greater than the correlation that would emerge by chance (Fig. 4D). In every participant, we additionally selected the region of interest that was most frequently identified as the source of interictal discharges, on the one hand, and of ictal discharges, on the other (see Methods). We then took the Euclidean distance between these two regions in each patient. Across participants, distance between ictal and ictal discharge sources is significantly less than that which would arise by chance, after shuffling the electrode labels [28.68 ± 20.75 mm versus 42.90 ± 21.97 mm; t(57) = 2.51, P = 0.01; Fig. 4E, left]. The most frequently estimated location of the source of IED sequences falls within 10 mm of the putative ictal source in nine participants, significantly greater than the number observed by chance after shuffling the electrode labels (3.70 ± 1.66, true value is in 100th percentile; Fig. 4E, right).

Note that the findings described in Fig. 4D would be expected given the results shown in Fig. 2D. However, the analysis shown in Fig. 4D provides additional information for several reasons. First, all electrodes and sequences are considered in Fig. 4D, rather than just the top 20 most common three-member sequences. Second, while in Fig. 2D we showed that ictal discharge latencies reside in the bottom left quadrant, the findings in Fig. 4D rest not just upon the fact that the signs of the latencies in the ictal data match those in the interictal data, but, additionally, that the values themselves are similar. Finally, in Fig. 4D, the temporal data provided in Fig. 2D are subject to spatial constraints, namely, the brain anatomy of each individual patient.

The hypothesized source of interictal and ictal discharges lies within the resection territory in participants with good outcome

Our data suggest a putative focal source of IEDs may account for the discharge sequences observed across the recording electrodes. It is challenging, however, to definitively establish which brain regions are indeed responsible for ictal and interictal epileptiform activity. A common approach to address this challenge is to determine whether the resection of a brain region leads to post-surgical seizure control. This would imply that epileptogenic tissues resided within the resection.

Therefore, in every participant, we identified the brain region of interest most frequently selected as the IED source. We similarly identified the region of interest most frequently selected as the ictal source. In participants with good surgical outcomes, the majority of IED and ictal localizations fell within the resection cavity (Fig. 5A). On the other hand, in participants with poor surgical outcomes, the majority fell outside the resection cavity. These data demonstrate that localization based on either the IED sequences or the ictal discharges appears to identify brain regions that are involved in epileptogenic activity. This therefore suggests that a focal source of pathologic travelling waves could underlie the observed time differences across electrodes in the IED sequences and in the ictal discharges.

Figure 5.

The hypothesized source of interictal and ictal discharges lies within the resection territory in participants with good outcomes. (A) In every participant with good or poor surgical outcomes, we determined whether the region most frequently identified as the source of IED sequences or ictal discharges lay within or outside the resection cavity. (B) We compared the number of times in which source localization based on the IED sequences fell within the resection cavity, in participants with good surgical outcomes (13 participants out of 19 total, true data indicated by blue line), to the number of times this would occur by chance after shuffling the lead labels 100 times in each participant (8.88 ± 1.61, blue histogram). We also compared the number of times in which source localization based on the IED sequences identified a region that fell outside of the resection cavity in participants with poor surgical outcomes (17 participants out of 21 total, true data indicated with yellow line), to the number of times this would occur by chance after shuffling the lead labels 100 times in each participant (15.39 ± 1.72, yellow histogram). In this analysis, sensitivity corresponds to the number of participants with good surgical outcomes where localization falls within the resection cavity. Specificity corresponds to the number of participants with poor outcomes where localization falls outside the resection cavity.

We compared the results of localization with respect to the resection territory, using the IED sequences, to that obtained from chance. To do so, we shuffled the electrode labels 100 times in each participant and identified the region of interest most frequently identified as the source of the IED sequences in each shuffle (Fig. 5B). In participants with good surgical outcomes, the number of patients, on average, with greatest region of interest within the resection territory was 8.88 ± 1.61, compared to 13 of 19 patients in the true data. The true value is in the 100th percentile of shuffled values. On the other hand, in patients with poor outcome, the number of patients, on average, with greatest region of interest outside of the resection territory was 15.39 ± 1.72, compared to 17 of 21 total patients in the true data. The true value is in the 74th percentile of the shuffled data. Therefore, while in our results specificity is greater than sensitivity, on the other hand, the true positive rate is significantly greater than that expected by chance, while this is not true of the true negative rate.

We were interested in comparing the results of our IED source localization method to methods which are simpler to implement. Therefore, we considered two controls. First, we considered the electrode that most commonly led IED sequences. Second, we considered the electrode with the most frequent IED activity. Our technique outperformed both simpler methods (Supplementary Fig. 3A). Of 19 patients with good post-surgical outcome, 8 had earliest electrode in the resection territory and 7 had most frequent electrode in the resection territory, compared to 13 with source in the resection in the true data. Among 21 patients with poor post-surgical outcome, 18 had earliest electrode outside of the resection territory and 17 had most frequent electrode outside the resection territory, compared to 17 with source outside of the resection in the true data. Overall accuracy was 65% in the earliest approach and 60% in the most frequent approach, compared to 75% under our approach. Clinical accuracy was 47.5%, in that 19/40 patients had Engel class 1 outcome. The putative IED location resulting from the use of simpler techniques tended to be relatively far from the source of IEDs localized using our computational approach. Across patients, the earliest electrode was 42.49 ± 42.18 mm from the localized source. The most frequent electrode was 50.81 ± 49.42 mm from the localized source (Supplementary Fig. 3B).

We were curious as to whether the early seizure source holds particular consequence for seizure freedom, as opposed to the greatest seizure source during all seizure activity. Therefore, we performed an additional analysis where only early seizure activity is considered (see Supplementary material and Supplementary Table 1). Prediction is more accurate in some patients under the early seizure analysis but less accurate in others, and overall, accuracy is the same under both techniques. The strongest IED focus may serve as a marker for the site of seizure activity with greatest consequence for seizure control, even in cases where there is a discrepancy between the early source location and the eventual source location.

Discussion

Our results suggest that interictal epileptiform discharges may be generated by focal sources of pathological activity, which emit travelling waves that reach surrounding brain regions with consistent patterns. Interictal discharges arise as sequences, measured in sets of electrodes with consistent temporal orderings. Similar temporal orderings occur during ictal discharges, consistent with the hypothesis that both ictal and interictal discharges emerge from a common source. Using the times of receipt of the interictal discharges, we localized the source of interictal activity, and the location of this activity matches the results of ictal discharge localization. The majority of localizations lie within the resection territory in participants with good seizure outcomes and outside the resection in those with poor outcomes.

Visual inspection of the iEEG is currently the gold standard used to guide surgical resections in patients with drug-resistant focal epilepsy. Intracranial electrodes implanted for seizure monitoring, however, can only capture activity from a small fraction of the cortical volume,⁴⁰ and so much of the pathological activity recorded through these electrodes likely reflects propagated activity from nearby regions. Although the traditional interpretation of clinically observed ictal discharges in the electrode recordings has been that the underlying brain regions are actively seizing, studies using microelectrode recordings have challenged this interpretation, demonstrating that ictal discharges may in fact reflect receipt of travelling waves, propagated from a more spatially circumscribed seizure focus or ictal wavefront.^3-10 Our data provide evidence that interictal discharges may also reflect travelling waves of pathological activity, similar to their ictal counterparts.^3-10

Given the similar temporal ordering observed in both the interictal and ictal discharges, our results are consistent with recent evidence that the two types of activity may emerge from a common source.¹⁰ The suggestion that interictal and ictal discharges are directly related is consistent with animal work showing propagation over functional pathways of both interictal and ictal discharges from a focal source,¹² and with simultaneous regional offsets of seizure activity presumably emanating from a common source.¹³ In the clinical setting, brain regions with IED activity that leads other regions within the irritative zone appear to be correlated with the eventual site of seizure origin.^18,39,41-44 Patients with well-localized IEDs prior to surgery are more likely to have a favourable post-surgical outcome.^45,46 Finally, disappearance of IEDs postoperatively can be used to predict good seizure outcomes.^45-47 Notably, we did not find evidence of bidirectional interictal discharge propagation patterns that could correspond to a predominant as well as antipodal source, as has been previously reported.¹⁰ This difference may be explained at least in part by differing methodologies, as bidirectional interictal discharges were reported only in IEDs measured from a microelectrode array that had been recruited by the ictal wavefront. Without microelectrode array data, we cannot formally establish whether electrodes are recruited into the ictal wavefront, and we may also lack the spatial resolution to detect bidirectional discharges.

Although our data suggest a strong correspondence between interictal and ictal discharges, the relation between brain regions involved in interictal epileptiform activity, referred to as the irritative zone, and those involved in seizure onset remains complex. There are several lines of evidence that suggest that the irritative zone is not monolithic. There are notable differences in the spatial extent and morphological appearance between interictal and ictal discharges, which could be due to variable propagation of this activity.⁴⁸ Similarly, variable spatial extent is often seen in interictal discharges over time even within the same patient, likely due to changes in local and global brain states such as state of arousal.^{14-18,21,49-51} Of note, this variability in spread may exist even if the underlying source of IEDs remains the same. This observation in fact underscores a strength of the computational approach we use for discharge source localization. That is, given multiple IEDs from a single source, each of which spreads over the brain surface to a different extent, our algorithm is nonetheless capable of recovering the single source common to each of the discharges.

Apart from the variability induced by any one particular source, we found that the irritative zone often consists of multiple subpopulations of activity, sometimes nearby to each other and sometimes separated by lobes or hemispheres. In many patients, spontaneous seizures appear to arise only from one of these source regions (for example, see Supplementary Fig. 1, Patients 17 and 27). The other independent IED populations perhaps represent less epileptogenic or only potentially epileptogenic brain regions that do not have the capacity to generate spontaneous seizures, but may be relatively more susceptible to recruitment as seizures evolve and spread. Our results are in agreement with recent work that found that there may be multiple co-incident seizure sources, some of which may move and others of which are stationary.⁵² Spread of the seizure source to distinct IED foci may occur by direct, contiguous spread over grey matter or by distant travel via white matter pathways. We do in fact find that when brain regions involved in IEDs are recruited into seizure activity, whether at seizure onset or later in seizure evolution, the temporal ordering of the ictal discharges within that given brain region remains consistent with the interictal data. This suggests that irritable brain regions that are regional sources of IED activity may indeed have a lower seizure threshold and may be preferentially recruited as seizures evolve. The development of in vivo, widefield cell-type–specific calcium imaging has offered an opportunity to capture stereotypic patterns of ictal and interictal discharge origin and spread in model organisms, and represents a promising future avenue.^53-55

Using a novel computational approach based on the temporal orderings of discharges, we found that ictal and interictal discharges tend to localize to similar brain regions. Additionally, discharges tend to localize to the resection territory in patients with good seizure outcomes and outside of it in patients with poor outcomes. Source localization based on the interictal discharges is in fact slightly stronger at predicting outcomes in our data. One possible reason for this is that sources of seizure discharges may move over time, leading to more variable source localizations. On the other hand, discovered IED sources are presumably stationary. We explored whether resection of the early seizure source is of particular importance to seizure freedom, but found that focusing only on early seizure activity did not improve prediction accuracy (Supplementary Table 1). It seems that both the site of seizure onset and of seizure travel may carry importance for seizure control. Resection of the site of seizure travel, for instance, may disconnect the seizure focus, rather than removing it, but may nonetheless provide seizure freedom. Finally, resection of the strongest focus of IED activity appears to be critical for seizure control, even in the setting of a moving seizure source, in which there is ambiguity in the seizure source location.

Our computational approach depends on the assumption that discharges arising from a pathological source spread outwardly over the grey matter at a fixed rate. Note that this limitation concerns the spread of discharges themselves, rather than the spread of the source of the discharges. We acknowledge that, rather than spreading evenly, discharges may instead spread over the grey matter but at variable rates or, alternatively, may spread over white matter tracts. Any non-local spread of discharges, whether over grey matter or white matter, may distort the results of our algorithm. Challenges to the assumptions that underlie our approach are explored in an analysis of simulated data (Supplementary Fig. 4). Incorporation of white matter spread of discharges using diffusion tensor imaging is a subject of ongoing work. Nonetheless, the success of our algorithm supports the notion that at least a significant portion of discharge spread is in fact radial. Candidate mechanisms for radial spread include local, isotropic synaptic spread^56-58 and ephaptic spread.^57-60 While our methodology supports the presence of radial spread of discharges, it is unable to elucidate the mechanism of this spread.

Despite the limitations of our approach, our results suggest that both interictal and ictal discharges observed in iEEG recordings represent propagated activity arising from more spatially constrained seizure foci. Seizures may originate from, and also travel to, IED foci. By examining the differences in time of receipt, we are able to localize the hypothesized source of epileptiform activity in both the interictal and ictal data. This approach therefore can have important clinical implications, as it would suggest that localizations can be derived from interictal discharges, which are much more frequent than seizures, and can be used to identify focal sources of activity underlying even the relatively broad epileptiform activity observed in iEEG recordings. More generally, our results provide an approach for better identifying the source and spread of both interictal and ictal activity, an important requirement for better understanding how seizure foci interact with epilepsy networks⁶¹ and for understanding the spatiotemporal evolution of seizures to guide both electrode implantation and targeted surgical interventions.⁶²

Funding

This work was supported by the National Institute for Neurological Disorders and Stroke [ZIA NS003144-09].

Competing interests

J.M.D. reports no disclosures relevant to the manuscript. C.P.W. reports no disclosures relevant to the manuscript. J.I.C. reports no disclosures relevant to the manuscript. S.R. reports no disclosures relevant to the manuscript. S.K.I. reports no disclosures relevant to the manuscript. K.A.Z. reports no disclosures relevant to the manuscript.

Supplementary material

Supplementary material is available at Brain online.