Band power modulation through intracranial EEG stimulation and its cross-session consistency
Christoforos A Papasavvas, Gabrielle M Schroeder, Beate Diehl, Gerold Baier, Peter N Taylor, Yujiang Wang
BBand power modulation through intracranialEEG stimulation and its cross-sessionconsistency
Christoforos A Papasavvas , Gabrielle M Schroeder , Beate Diehl ,Gerold Baier , Peter N Taylor , , , Yujiang Wang ∗ , , June 1, 2020 a r X i v : . [ q - b i o . N C ] M a y bstract Background: Direct electrical stimulation of the brain through intracra-nial electrodes is currently used to probe the epileptic brain as part ofpre-surgical evaluation, and it is also being considered for therapeutictreatments through neuromodulation. It is still unknown, however, howconsistent intracranial direct electrical stimulation responses are acrosssessions, to allow effective neuromodulation design.Objective: To investigate the cross-session consistency of the electro-physiological effect of electrical stimulation delivered through intracranialEEG.Methods: We analysed data from 79 epilepsy patients implanted withintracranial EEG who underwent brain stimulation as part of a memoryexperiment. We quantified the effect of stimulation in terms of bandpower modulation and compared this effect from session to session. As areference, we applied the same measures during baseline periods.Results: In most sessions, the effect of stimulation on band power couldnot be distinguished from baseline fluctuations of band power. Stimulationeffect was also not consistent across sessions; only a third of the sessionpairs had a higher consistency than the baseline standards. Cross-sessionconsistency is mainly associated with the strength of positive stimulationeffects, and it also tends to be higher when the baseline conditions aremore similar between sessions.Conclusion: These findings can inform our practices for designingneuromodulation with greater efficacy when using direct electrical brainstimulation as a therapeutic treatment.
About 35% of patients with epilepsy are drug-resistant and require additionaltreatment [1, 2]. In this context, direct electrical stimulation through intracranialelectroencephalography (iEEG) has become an invaluable tool for clinicians.Direct electrical stimulation is currently used in three ways. First, functionalmapping of the cortex so that eloquent cortical areas are preserved in resectiveepilepsy surgery [3, 4]. Second, measuring the “epileptogenicity” of the stimulatedand surrounding areas [5]. Third, exploring the neuromodulatory potential ofdirect electrical stimulation which can be the basis for therapeutic interventions[6, 7]. In this work we will focus on the neuromodulatory potential of intracranialelectric stimulation. Arguably, to achieve any therapeutic goals, the effect ofstimulation should be consistent across multiple sessions [8]. To our knowledge,the consistency of iEEG stimulation effect has not been studied systematically.Neuromodulation has been explored as an alternative treatment for patientswith non-conclusive pre-surgical evaluation of the epileptogenic zone [9]. Insuch cases, without any candidate resection area, the goal is to modulate theepileptic network in a way that enhances physiological neural activity, andprevents pathological, or seizure activity. It is currently unknown how a targetedmodulatory effect can be achieved a priori , but several studies have begun tomap out how stimulation affects the brain both electrophysiologically as well2s behaviourally. For instance, Keller et al. showed that repeated stimulationmodulated the excitability of neighbouring areas around the stimulation site[6]. Memory enhancement has been reported after using a closed-loop electricalstimulation of the lateral temporal cortex [10]. Furthermore, stimulation appliedto the posterior cingulate cortex induced an increase of low gamma power inhippocampus which correlated with the magnitude of memory impairment [11].Muller and colleagues have reported a correlation between the modulation ofhigh gamma frequencies and somatosensory perception, both induced by directcurrent stimulation [12]. Khambhati and colleagues demonstrated functionalreconfiguration of brain networks after stimulation as indicated by alterationsin band-specific functional connectivity [7], while Huang and colleagues furtherdemonstrated the close relationship of functional connectivity and stimulation-induced band power modulation [13]. Similarly, another study showed thattemporal cortex stimulation increased theta band power in remote areas predictedby functional connectivity, especially when the stimulation was delivered closeto white matter [14]. These studies show the potential of using direct electricalstimulation in therapeutic neuromodulation, and intracranial stimulation throughiEEG can be a useful tool to rapidly explore possible stimulation locations andparameters for the design of effective neuromodulation.Consistent stimulation effects -electrophysiologically or behaviourally- acrosssessions are crucial for developing therapeutic neuromodulation treatments. Forexample, understanding the underlying electrophysiological effect of transcranialstimulation and its consistency is an important step towards taking advantageof its already demonstrated benefits on motor rehabilitation [15, 16]. Relevantinvestigations on cross-session consistency have been reported in non-invasivestimulation modalities (for a review see [8]). For instance, while the electro-physiological effect of transcranial magnetic stimulation has been reported tobe highly consistent across sessions [17], while transcranial direct current stim-ulation (tDCS) effect was found to be inconsistent [18, 19] (but see also [20]).The sources of such variability have been discussed extensively in the contextof inter-individual studies but some of them apply on an intra-individual basisas well (e.g., baseline physiological state, cognitive task at hand; for a reviewsee [21]). However, to our knowledge, the cross-session consistency of the elec-trophysiological effects of iEEG stimulation has not yet been systematicallyinvestigated.Here we investigate the consistency of the iEEG stimulation effect in termsof band power modulations between stimulation sessions from the same subject.We measure how stimulation modulates band power in five different frequencybands and investigate whether these modulations vary from one stimulationsession to the next for the same subject and stimulation location. We introducea measure of consistency that accounts for the distributed stimulation effectsrecorded across multiple iEEG channels. We finally investigate which featuresof the stimulation protocol, the measured stimulation effect, and the baselineconditions most influence between-session consistency.3
Methods
We used data that are publicly available as part of the Restoring ActiveMemory (RAM) project (managed by the University of Pennsylvania; http://memory.psych.upenn.edu/RAM ). As stated in the project’s website "Informedconsent has been obtained from each subject to share their data, and personallyidentifiable information has been removed to protect subject confidentiality".The original research protocol for data acquisition was approved by the relevantbodies at the participating institutions. Furthermore, the University EthicsCommittee at Newcastle University approved the current project involving thedata analysis reported here (Ref: 12721/2018). We extracted data from allpatients (n=87) that underwent at least one stimulation session while performingmemory tasks. We excluded 8 patients that either had substantial stimulationartefacts in almost all channels or their data were limited (single session with<18 stimulation trials). Thus, we analysed data from 79 subjects from which 36had at least 2 stimulation sessions with the same stimulation location (totalling101 pairs of sessions with same stimulation location).
Stimulation was delivered using charge-balanced biphasic rectangular pulses (300 µ s pulse width) at 10, 25, 50, 100, or 200 Hz frequency 0.25–3.5 mA amplitude.The duration of the stimulation was 500 ms or 4.6 s, depending on the experiment. To measure stimulation effect, 1-second segments were extracted from the iEEGsignals around every stimulation trial; that is, we extracted one segment before(pre) and one after (post) the stimulation event, with a 50ms buffer betweeneach segment and the event. To assess baseline fluctuations, ‘pre’ and ‘post’segments were also extracted from the baseline activity during baseline epochs,with a pre-post interval equal to the one around the stimulation trials of thesame session. A baseline epoch was considered to be any inter-stimulus intervalwhich was at least 20 sec long and 5 sec away from the stimulation itself. Figure1 shows a schematic of the session timeline and the process of segment extraction.Since the stimulation trials were temporally organised in groups of three in atypical session (i.e., less than 10s interval between trials in the same group),we extracted baseline pre/post segments from each baseline epoch in groups ofthree as well (see Fig. S1 in Supplementary Material), such that the number ofsegments taken around stimulation and the number of baseline segments wereapproximately equal in each session. 4 trial 1trial 2trial 3 Baseline postpre Stimulation postpre1sband power calculationlog-transformpaired differences post - preBaseline ‘effect’U Stimulation effectU time stim. trial 3stim. trial 2stim. trial 1 stimulation sessiontimelinefor eachchannel baselineepoch baselineepoch baselineepoch . . . . . . Figure 1:
Measuring band power changes in response to stimulationand band power baseline fluctuations.
Top panel: Timeline of a typicalstimulation session. The schematic also shows how pre- and post-stimulationsegments are extracted from each stimulation trial and analysed in terms oftheir band power. While only three trials are shown here, a typical stimulationsession had 60 trials (median value with 13.9 SD). Lower panels: Band powerin five different frequency bands was calculated and log-transformed for eachextracted segment. The effect of stimulation on band power, and equivalently,band power’s fluctuations during baseline, are expressed by the effect U, whichis derived from a non-parametric test applied to the paired differences betweenpre and post segments.The time series of each segment were centred around zero and de-trended.De-trending was achieved by applying linear regression and then removing theleast-squares fit from the signal. Any channels with repeated artefacts were5xcluded (see below). A common average re-referencing was applied to theremaining set of channels. The stimulation channels were excluded from thecommon average calculation, but the calculated common average was applied tothem. The band power of each segment was calculated in 5 different bands [delta(2-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12-25 Hz), and gamma (25-55Hz)] after estimating the power spectral density of the segment using Welch’smethod (with window length equal to half of the segment length and overlaplength equal to a quarter of the segment length). Finally, the band powers werelog-transformed. Figure 1 shows a schematic of the preprocessing.Channels with repeated stimulation artefacts (i.e., voltage deflection) wereexcluded. A repeated stimulation artefact was detected based on two criteria.Either one of these two criteria was sufficient to indicate a channel with repeatedartefacts. First, a strong effect of stimulation on the average (across time)voltage of the first half of the post segment compared to the average (acrosstime) voltage of the second half of the pre segment. The effect was quantifiedusing the t-statistic of a paired t-test. Second, the second half of the average(across trials) post signal had a slow return to the average (across time) voltagevalue of the pre segments. This was detected by linear regression.
Box plots were used to summarise various distributions in the Results. Centrallines indicate median values, while the boxes extend from the th to th percentile (interquartile range) of the distribution. Whiskers extend to the upperand lower adjacent values, that is, the most extreme values that are not outliers.Outliers are considered to be values that lie more than 1.5 × [interquartile range]away from the th or th percentile. The effect of stimulation on band power from pre to post was considered as thez-statistic (indicated by U throughout) produced by the Wilcoxon sign ranktest (paired non-parametric test; signrank function in MATLAB). A positiveU indicates an increase in band power from pre to post, whereas a negative Uindicates a decrease from pre to post. To also quantify the baseline fluctuationsof band power, the same measure was used on the ‘pre/post’ pairs of the baselineactivity (Fig. 1).The overall difference in stimulation effect between two sessions (across allchannel/band combinations) in Fig. 4B was quantified by using the absolutet-statistic of a paired t-test on the absolute effect U of the two sessions. We usedabsolute effects as we wanted to generally assess changes in effect magnitude.
The consistency of stimulation effect was measured for each pair of sessions withthe same stimulation location in the same subject. All possible combinations of6 sessions were considered, totalling 101 pairs. The consistency was computedby first pairing the effect values of corresponding channel/band combinationsbetween the two sessions. Note that only the intersection of valid channelsbetween the two sessions was considered (a channel can be excluded due toartefacts in one session but not the other). The consistency coefficient wasgiven by the Fisher-transformed zero-centred Pearson’s correlation. Consideringthe effect values of the two sessions as random variables S and S , then theconsistency coefficient is given by: r = E [ S S ] / (ˆ σ ˆ σ ) , where E denotesexpected value and ˆ σ refers to the average deviation from 0 ( ˆ σ = (cid:112) ( (cid:80) ni =1 s i ) /n ).We use the zero-centred Pearson’s correlation to only detect a zero-translatedagreement between the random variables, that is, in the form of S = kS , with0 intercept and k a non-zero constant. The consistency curve was used to express the consistency between two sessions bygradually considering fewer pairs of effect values at low effect sizes. Considering ascatter plot of all the effect value pairs, it was computed by gradually increasingthe radius of an exclusion circle emanating from (0,0). The consistency curve atradius = 0 gives the consistency when all points are included in the consistencycalculation, whereas the consistency curve at radius = x expresses the consistencyas computed after excluding every pair of effect values that lie inside a circlewith centre (0,0) and radius x. The circle was gradually enlarged with a stepof 0.2 and the enlargement stopped just before covering 98% of the scatteredvalues. We used this procedure to ensure that we can detect consistency even ifonly a few channels exhibited consistency, without the consistency being maskedby low effect channels.Each consistency curve is represented by its maximum consistency coefficient.The maximum consistency coefficient is the value on the curve that deviatesthe most from 0, being positive or negative. Thus, it expresses the strongestcorrelation or anti-correlation found between the effect values of the two sessions.
To explore which factors determine consistency across all 101 session pairs,we modelled the maximal value of consistency as a linear combination of thefollowing variables: • session time difference: absolute time difference between the sessions’starting timestamps. • difference in baseline (band power) means: mean absolute paired differencebetween the sessions’ mean values of band power during baseline (both‘pre’ and ‘post’). • difference in baseline (band power) standard deviations: mean absolutepaired difference between the sessions’ standard deviations of band powerduring baseline (both ‘pre’ and ‘post’).7 average max effect: average (between sessions) maximum effect (across allchannel/band combinations). • average min effect: average (between sessions) minimum effect (across allchannel/band combinations). • average stimulation amplitude: average stimulation amplitude betweensessions. • stimulation amplitude difference: difference in stimulation amplitude be-tween sessions. • stimulation frequency: frequency of stimulation pulse train (always commonbetween examined session pairs). • depth of the stimulation location: distance of stimulation location (midpointbetween anode and cathode) from brain surface. • task difference: difference in memory tasks (categorical variable) carriedout by the subject during recording; that is, 0 for same and 1 for differenttasks between sessions.The stimulation depth was computed as the Euclidean distance of the anode-cathode midpoint from the subject’s brain surface. If that midpoint was foundto be outside the provided surface, its depth was set to negative (minus theEuclidean distance). In order to quantify the explanatory power of all the different independentvariables on the consistency we used ANOVA test on the model built by theMultiple Linear Regression Analysis. We built the model and assessed theANOVA effects 200 times through bootstrapping. We used this bootstrappingapproach to check for the robustness of the model. The ANOVA effect, the R ,and the Adjusted R are reported. Figure 2 shows the measured stimulation effects across channels and frequencybands for one example session in each of two example subjects 1022 and 1069.These example sessions represent sessions with weak (Fig. 2, left) and strong(Fig. 2, right) stimulation effects. As a reference, the upper panels show the“effect” during baseline, that is, the background fluctuations of band power. Thelower panels show the stimulation effect in terms of band power changes, basedon multiple pre- and post-stimulation pairs (see example inset panels on the8ight and Fig. 1). Notice that, even in the example subject 1069, where somestrong stimulation effects are seen, these are restricted to a handful of channelsand specific frequency bands. This observation is typical for all the sessions thatexhibited a strong effect. Similarly, the example session on the left is a typicalexample of all the sessions that have a stimulation effect that is indistinguishablefrom the baseline fluctuations. channels channels -505 e ff e c t ( U ) ba s e li ne s t i m u l a t i on subject 1022 subject 1069
20 40 60 8020 40 60 80 100 stim. channelsstim. effect on band -0.5 0 0.5-2 0 2 post - prepost - pre
U = 5.5U = -2.9
Figure 2:
Examples of sessions with low and high stimulation effect.
The heat maps show the stimulation effect in two example sessions: one fromsubject 1022 with low effect and one from subject 1069 with high effect. Theeffect was measured for all combinations of channels and frequency bands. Noticethat the effect can be positive or negative, indicating increase or decrease of bandpower from pre to post stimulation (see example distributions of the differences(post-pre) in the rightmost panels). The fluctuations of band power duringbaseline are also shown for comparison. The channels are sorted based on theirEuclidean distance from the stimulation site. The lower panels show the spatialdistribution of the stimulation effect on theta band across the cortex.9he lower panels in Fig. 2 show the spatial layout of the iEEG stimulationand recording channels in the brain, with electrodes colour-coded by theircorresponding stimulation effect sizes. Note that a strong stimulation effect, inthis case on theta band, is not limited to contacts close to the stimulation sitebut also affected remote contacts (lower right panel).In order to assess if the effect of stimulation exceeded baseline fluctuationsin general across all 165 sessions and 79 patients, we compared the extremaof the stimulation effect to the extrema of the baseline fluctuations for eachfrequency band. Figure 3A shows the distributions of minima and maxima effecton theta band for baseline and stimulation. These extrema were taken acrosschannels to capture the strongest effect during a session. Generally, it is evidentthat, even the channel with the strongest stimulation effect does not have asubstantially larger effect size compared to the baseline fluctuations. In thetaband, only 10.2% of the sessions exhibit a minimum (negative) stimulation effectthat exceeds the adjacent value of the baseline minima. Similarly, only 18.1% ofthe sessions exhibit a maximum (positive) stimulation effect that exceeds theadjacent value of the baseline maxima (see Fig. 3A).The limited stimulation effect across all sessions was also evident whenwe computed the paired differences in effect between stimulation and baselineconditions. The histograms for the effect minima and maxima in Figure 3Bindicate that, in most sessions, even the most extreme effect sizes do not exceedthe band power fluctuations during baseline. However, these distributions arenot zero-centred (paired t-test for minima: p = 2 . · − , effect size for minima:-0.430; paired t-test for maxima : p = 4 . · − , effect size for maxima: 0.437),indicating that across patients and sessions there is a small but significantdifference between baseline and stimulation conditions in our dataset. Similarresults were found for all frequency bands (see Fig. S2 in SupplementaryMaterial). 10 AB paired difference in effect (U) paired difference in effect (U)session minima session maximasession minima session maxima Figure 3:
Low stimulation effect on theta band found in most sessions.A
Across all sessions, the distributions of their extrema effect values on thetaband are compared between baseline and stimulation. Each point corresponds tothe minimum (left) or maximum (right) effect value U of all recording channelsin a given session. A minority of sessions have stimulation extrema (10.2% formin and 18.1% for max) that are more extreme than the adjacent values seenin baseline distributions (adjacent values being the most extreme values thatare not outliers). B The histograms present the paired (per session) differencesin extreme values of effect on theta band (session stimulation effect – sessionbaseline effect).
Next, we investigated whether the low stimulation effect size in most sessions canbe attributed to the stimulation amplitude of the session. Figure 4A shows thatthere is no correlation between the effect size achieved in the session and thesession stimulation amplitude. The distributions of effect sizes for each session,across all channels and frequency bands, are represented by their minima andmaxima. Neither of these two measures tend to increase or decrease with thestimulation amplitude (range: 0.25 - 3.5 mA; see also Fig. S3 in SupplementaryMaterial for band specific results).Furthermore, we considered all the pairs of stimulation sessions with the samestimulation location in the same subject (101 pairs). We tested whether theirdifference in effect size is correlated with the difference of stimulation amplitudebetween the sessions. Figure 4B shows that the absolute difference in effect sizeis not correlated with the absolute difference in stimulation amplitude. Thus,11ven for the same subject and the same stimulation location, an increase instimulation amplitude does not necessarily produce a stronger effect. m a x e ff e c t ( U ) m i n e ff e c t ( U ) stim. amplitude (mA) s t i m . e ff e c t ( t - s t a t ) stim. amplitude (mA) stim. amplitude (mA) pair of stim. sessions AB session minima session maxima Figure 4:
No correlation found between stimulation amplitude andeffect. A
The effect minima and maxima (across all channels and frequencybands) from each session is scattered against the stimulation amplitude of thesession. B For all the pairs of sessions that come from the same subject andhave the same stimulation location, the difference in their stimulation effects isscattered versus the difference in their stimulation amplitudes.
To investigate the consistency of stimulation effect across sessions, we focusedon pairs of sessions in the same subject and stimulation location. Figure 5 showstwo examples of these session pairs. The example on the left (subject 1022) doesnot show positive correlation between the two sessions in terms of stimulationeffect in different channels and frequency bands. The example on the right(subject 1069) shows that the patient’s two sessions are positively correlated.Notice that this correlation is mainly driven by channels that exhibit a strongpositive stimulation effect in the first place.12 e ss i on s e ss i on s e ss i on s e ss i on -505 effect(U) -5 0 5-505 stimulation effect (U)session 1 s t i m u l a t i on e ff e c t ( U ) s e ss i on subject 1022 subject 1069 -5 0 5-505 stimulation effect (U)session 1 s t i m u l a t i on e ff e c t ( U ) s e ss i on channels20 40 60 80 100 channels20 40 60 80 Figure 5:
Examples of session pairs with low and high effect consistency.
The stimulation effect in two pairs of sessions, from two different subjects, isshown in the top panels as examples. The pair of sessions on the left (subject1022) has low consistency whereas the pair on the right (subject 1069) has highconsistency. This disparity is more clearly shown in the lower panels, where thecorresponding effect pairs for each channel and frequency band are scattered.The low consistency on the left is shown by a circular cloud of points, whereasthe high consistency on the right is shown by an elongated cloud of points.In order to assess the level of consistency in stimulation effect across all101 session pairs, we computed the consistency curve for each pair. Figure 6Ashows the consistency curves of the two session pair examples in Fig. 5 alongsidesome illustrations on how the curve is computed: a circle of exclusion emanatingfrom (0,0) is gradually enlarged and the consistency coefficient is calculated forvarying values of the circle’s radius (see Methods). The consistency curve (as afunction of the radius) captures the consistency coefficient of the session pairwhen all effect values are considered (at radius 0) but also while increasinglyexcluding channel and frequency band combinations with weaker stimulationeffect (at higher radii). The exclusion of channel/band combinations with weak13ffect serves to minimise the influence of the inherently inconsistent band powerfluctuations on the consistency calculation. In addition, considering the selectiveconnectivity of brain areas, it is expected that only a subset of channels willrespond to a localised stimulation. The two curves shown in Figure 6A capturethe difference between high and low consistency as shown in the scatter plots ofFig. 5, not only when all values are considered, but also when only the strongeffect values are considered. This approach of gradually excluding the weakerstimulation effects (around the level of baseline fluctuations) essentially allows usto capture consistency in the few channels that display a discernible stimulationeffect in the first place.Figure 6B shows the consistency curves for all 101 session pairs and theconfidence interval of consistency coefficients of the baseline periods (blue back-ground). The overall consistency in this dataset is not high: 32.7% of the sessionpairs have higher consistency than the . th percentile of the baseline ‘effect’ atradius = 0; 12.9% of the session pairs have higher consistency than the . th percentile of the baseline ‘effect’ at radius = 3; and only 34.6% of the sessionpairs have a maximum consistency that is higher than the maximum value ofthe baseline ‘effect’ confidence interval. Four examples of maximum consistencycoefficients on four of these curves are indicated with brown markers in Fig.6B. We will consider these maximum consistency coefficients as a representativevalue of the session pair consistency in the following (i.e., highest consistencyachieved after exclusion of some not stimulation-related channels).In Fig. 6C, we demonstrate a strong and significant correlation between theaverage maximum effect and the maximum consistency coefficients (Pearson’s r = 0 . , p = 7 . × − ). Theta and alpha bands contribute more to thiscorrelation (see Fig. S4 in Supplementary Material). As a comparison, weapplied the same procedure to simulated data (normal distribution with mean 0and standard deviation matching the the sessions’ baseline), and the correlationis not present (Pearson’s r = − . , p = 0 . ). Essentially, the strongerstimulation effects also tend to be more consistent across sessions.Finally, we built a multiple linear regression model to explain the maximumconsistency coefficients as a linear combination of multiple independent variablesincluding the average maximum effect ( R = 0 . , Adjusted R = 0 . ).The high explanatory power of the average maximum effect is also evidentafter running ANOVA on the multiple linear regression model, with the resultsshown in Fig. 6D (distributions produced after 200 bootstrap samples). Otherthan the strong effect of the average maximum effect on consistency, Fig. 6Dshows a fair effect of both the task difference and the difference of baselinemean on consistency ( p = 0 . and p = 0 . , respectively), which are bothanti-correlated with the maximum consistency coefficient.14 c on s i s t en cy c oe ff. -5 0 5-505 stim. effect (U)session 1 s t i m . e ff e c t ( U ) s e ss i on -5 0 5-505 stim. effect (U)session 1 s t i m . e ff e c t ( U ) s e ss i on -5 0 5-505 stim. effect (U)session 1 s t i m . e ff e c t ( U ) s e ss i on radius of the circle of exclusion (U) AB c i r c l e o f e x c l u s i o n c on s i s t en cy c oe ff. examples of max consistency coefficients m a x c on s i s t en cy c oe ff. C A N O VA e ff e c t D -2 010203040 s t i m . a m p l . d i ff. a v e r . s t i m . a m p l . s e ss i on t i m e d i ff. s t i m . dep t h ba s e l . m ean s d i ff. ba s e l . S D s d i ff.t a sk d i ff. a v e r . m i n e ff e c t a v e r . m a x e ff e c t s t i m . f r equen cy correlatedunspecifiedanti-correlated Figure 6:
Cross-session consistency is found in a minority of subjectswhile it relies heavily on strong (positive) effect. A
Consistency curveswere computed by gradually enlarging the circle of exclusion and calculating theconsistency coefficient on the remaining scatter points. Three example radii forthe circle of exclusion are shown. The plotted consistency curves represent thetwo example session pairs in Fig. 5. B All 101 consistency curves, one for eachsession pair, are shown with five examples of maximal consistency coefficientsachieved (brown diamonds). The shaded blue region indicates the 95% two-sided confidence interval of the consistency coefficients of baseline activity. C Maximum consistency coefficient scattered versus average maximum effect revealsa strong correlation between them (Pearson’s r = 0 . , p = 7 . × − ). D Distributions of ANOVA effect values (produced through bootstrapping - seeMethods) for the independent variables used in the multiple linear regressionmodel which was used to explain the maximum consistency coefficients. Theaverage maximum effect between paired sessions has the strongest explanatorypower over consistency. Both task difference and difference in baseline meanhave a fair explanatory power over consistency. Outliers are omitted for clarity.15
Discussion
We showed that the cross-session consistency of stimulation effect (in terms ofband power modulations) is relatively low in a group of 36 subjects who hadmultiple stimulation sessions through iEEG. A third of session pairs indicate aconsistency that is above the baseline consistency (Fig. 6B). High consistencyof stimulation effect was found to rely heavily on a strong positive effect ofstimulation, that is, high increase of band power (Fig. 6D). Thus, given thesefindings, the low consistency levels would be expected in this dataset since thestimulation effect was limited (Fig. 3). Other datasets with more pronouncedstimulation effect in terms of band power changes may exhibit a higher level ofconsistency between sessions.Variability in the baseline brain state may have impacted the consistency ofthe stimulation responses in our data set. Even the stimulation response withina session has been repeatedly found to depend on the underlying brain state[22, 23, 24]. This finding is corroborated here since consistency was found tobe anti-correlated with both the difference in baseline mean band power anddifference in memory task which can be understood as a difference in brain state(Fig. 6D). In other words, the more similar the brain states (as measured bytask, or band power configurations) were in this dataset, the more consistentthe stimulation effects tended to be. Therefore, a practical advice is to use thesame task across stimulation sessions if consistency across sessions is desired.Our multiple linear regression model included the stimulation depth as oneof the independent variables, and it did not exhibit a strong predictive powerover consistency. This is not surprising since we did not find any strong relationbetween stimulation depth and the effect U in the first place (see Fig. S5in Supplementary Material). However, it is worth noting that there was nodistinction between stimulation through surface and depth electrodes in ouranalysis. The difference between these two types of electrodes cannot be fullycaptured by the stimulation depth variable. Other confounding characteristics,like the physical dimensions of the contacts and the average distance from otherrecording electrodes, were not accounted for. Future work can investigate furtherwhether consistency depends on such factors.Considering the data across all subjects, the most represented stimulationsite is the right medial temporal lobe, but several other areas were stimulated.In addition, the spatial extent of the recording electrodes across the datasetcovers the whole cortex. A visual inspection of the stimulation sites and thehighly responsive sites did not reveal any specific area that was associated withhigh effect or consistency (see Fig. S6 in Supplementary Material).Surprisingly, the effect on band power was not correlated with the amplitudeof stimulation in this dataset. This finding agrees with the reported insensitivityof motor-cortical excitability to tDCS intensity increases [25]. However, anotheriEEG study has found stimulation intensity to correlate with high frequencyactivity (30-100Hz), a frequency range which extends beyond those we investi-gated [26]. Furthermore, multiple studies have reported correlations betweenstimulation intensity and motor improvements when deep brain stimulation of16ubthalamic nuclei is used for the treatment of Parkinson’s disease (e.g., [27]).This discrepancy might indicate a non-trivial or non-linear relationship betweenthe electrophysiological and behavioural effects of an increasing stimulationintensity. The potentially ‘all-or-nothing’ response may further depend on thestimulated area.In our study, the stimulation effect was measured based on the immediateresponses within a session only. Arguably, the effect of stimulation can manifestat longer timescales or in other features and those effects may be more consistentacross sessions [28, 29]. This also relates to our definition of baseline in thisstudy. Segments of baseline are taken from interstimulus intervals that may carrysome post-stimulus modulations of band power. Any consistency in long-termchanges due to stimulation should be investigated in future studies.Cross-session consistency of stimulation effect is critical for developing ther-apeutic neuromodulation treatments, both in terms of electrophysiological, aswell as behavioural stimulation effect. This is supported by recent studies whichestablished relationships between stimulation-induced modulation of specificfrequency bands and behavioural outcomes [12, 11]. Despite the fact that someanatomical factors (e.g., thicknesses of the skull and the cerebrospinal fluid layer)do not influence intracranial stimulation, as opposed to tDCS [30], we foundthat stimulation through iEEG still has low consistency in terms of band powermodulations across sessions in our dataset, similar to tDCS [18, 19, 20]. Ourresults suggest that ensuring a strong positive modulation of band power throughstimulation, by choosing the appropriate stimulation location and parameters,is prerequisite for a high consistency across sessions. In addition, our resultssuggest that the dynamical brain state needs to be taken into account and astate-depended framework of stimulation may be required. The present andprevious studies all show that more sophisticated protocol designs are needed tomaximise the benefit of neurostimulation interventions.
Acknowledgments
References [1] Dieter Schmidt. Drug treatment of epilepsy: options and limitations.
Epilepsy & behavior , 15(1):56–65, 2009.[2] Jane De Tisi, Gail S Bell, Janet L Peacock, Andrew W McEvoy, WilliamF J Harkness, Josemir W Sander, and John S Duncan. The long-termoutcome of adult epilepsy surgery, patterns of seizure remission, and relapse:a cohort study.
The Lancet , 378(9800):1388–1395, 2011.173] Ronald P Lesser, Barry Gordon, and Sumio Uematsu. Electrical stimulationand language.
Journal of clinical neurophysiology: official publication of theAmerican Electroencephalographic Society , 11(2):191–204, 1994.[4] Stephanie Gollwitzer, Rüdiger Hopfengärtner, Karl Rössler, Tamara Müller,David Gerhard Olmes, Johannes Lang, Julia Köhn, Müjgan Dogan Onu-goren, Jana Heyne, Stefan Schwab, and Hajo Martinus Hamer. Afterdis-charges elicited by cortical electric stimulation in humans: When do theyoccur and what do they mean?
Epilepsy and Behavior , 87:173–179, 2018.[5] Antonio Valentín, Gonzalo Alarcón, Mrinalini Honavar, Jorge J. GarcíaSeoane, Richard P. Selway, Charles E. Polkey, and Colin D. Binnie. Singlepulse electrical stimulation for identification of structural abnormalities andprediction of seizure outcome after epilepsy surgery: A prospective study.
Lancet Neurology , 4(11):718–726, 2005.[6] Corey J. Keller, Yuhao Huang, Jose L. Herrero, Maria E. Fini, Victor Du,Fred A. Lado, Christopher J. Honey, and Ashesh D. Mehta. Induction andQuantification of Excitability Changes in Human Cortical Networks.
TheJournal of Neuroscience , 38(23):5384–5398, 2018.[7] Ankit N. Khambhati, Ari E. Kahn, Julia Costantini, Youssef Ezzyat,Ethan A. Solomon, Robert E. Gross, Barbara C. Jobst, Sameer A. Sheth,Kareem A. Zaghloul, Gregory Worrell, Sarah Seger, Bradley C. Lega, Shen-nan Weiss, Michael R. Sperling, Richard Gorniak, Sandhitsu R. Das, Joel M.Stein, Daniel S. Rizzuto, Michael J. Kahana, Timothy H. Lucas, Kathryn A.Davis, Joseph I. Tracy, and Danielle S Bassett. Functional control of electro-physiological network architecture using direct neurostimulation in humans.
Network Neuroscience , 3(3):848–877, jan 2019.[8] Andrea Guerra, Virginia López-Alonso, Binith Cheeran, and Antonio Suppa.Variability in non-invasive brain stimulation studies: Reasons and results.
Neuroscience Letters , 719(December 2017):133330, 2020.[9] Paul Boon, Robrecht Raedt, Veerle De Herdt, Tine Wyckhuys, and KristlVonck. Electrical stimulation for the treatment of epilepsy.
Neurotherapeu-tics , 6(2):218–227, 2009.[10] Youssef Ezzyat, Paul A. Wanda, Deborah F. Levy, Allison Kadel, AdaAka, Isaac Pedisich, Michael R Sperling, Ashwini D. Sharan, Bradley C.Lega, Alexis Burks, Robert E. Gross, Cory S. Inman, Barbara C. Jobst,Mark A. Gorenstein, Kathryn A. Davis, Gregory A. Worrell, Michal T.Kucewicz, Joel M. Stein, Richard Gorniak, Sandhitsu R. Das, Daniel SRizzuto, and Michael J. Kahana. Closed-loop stimulation of temporal cortexrescues functional networks and improves memory.
Nature Communications ,9(1):365, dec 2018.[11] Vaidehi S. Natu, Jui Jui Lin, Alexis Burks, Akshay Arora, Michael D. Rugg,and Bradley C. Lega. Stimulation of the Posterior Cingulate Cortex Impairs18pisodic Memory Encoding.
The Journal of neuroscience : the officialjournal of the Society for Neuroscience , 39(36):7173–7182, 2019.[12] Leah Muller, John D. Rolston, Neal P. Fox, Robert Knowlton, Vikram R.Rao, and Edward F. Chang. Direct electrical stimulation of human cor-tex evokes high gamma activity that predicts conscious somatosensoryperception.
Journal of Neural Engineering , 15(2), 2018.[13] Yuhao Huang, Boglárka Hajnal, László Entz, Dániel Fabó, Jose L. Herrero,Ashesh D. Mehta, and Corey J. Keller. Intracortical Dynamics UnderlyingRepetitive Stimulation Predicts Changes in Network Connectivity.
TheJournal of neuroscience : the official journal of the Society for Neuroscience ,39(31):6122–6135, 2019.[14] Ethan A Solomon, J E Kragel, R Gross, B Lega, Michael R Sperling,Gregory A. Worrell, Sameer A Sheth, Kareem A Zaghloul, B C Jobst,J M Stein, S Das, R Gorniak, C S Inman, Sarah Seger, Daniel S Rizzuto,and M J Kahana. Medial temporal lobe functional connectivity predictsstimulation-induced theta power.
Nature communications , 9(1):4437, 2018.[15] Bettina Pollok, Ann-Christin Boysen, and Vanessa Krause. The effectof transcranial alternating current stimulation (tACS) at alpha and betafrequency on motor learning.
Behavioural Brain Research , 293:234–240,2015.[16] Ethan R Buch, Emiliano Santarnecchi, Andrea Antal, Jan Born, Pablo ACelnik, Joseph Classen, Christian Gerloff, Mark Hallett, Friedhelm C Hum-mel, Michael A Nitsche, Alvaro Pascual-Leone, Walter J Paulus, Janine Reis,Edwin M Robertson, John C Rothwell, Marco Sandrini, Heidi M Schambra,Eric M Wassermann, Ulf Ziemann, and Leonardo G Cohen. Effects of tDCSon motor learning and memory formation: A consensus and critical positionpaper.
Clinical Neurophysiology , 128(4):589–603, 2017.[17] A. M. Hermsen, A. Haag, C. Duddek, K. Balkenhol, H. Bugiel, S. Bauer,V. Mylius, K. Menzler, and F. Rosenow. Test-retest reliability of singleand paired pulse transcranial magnetic stimulation parameters in healthysubjects.
Journal of the Neurological Sciences , 362:209–216, 2016.[18] Katherine Dyke, Soyoung Kim, Georgina M. Jackson, and Stephen R.Jackson. Intra-Subject Consistency and Reliability of Response Following2mA Transcranial Direct Current Stimulation.
Brain Stimulation , 9(6):819–825, 2016.[19] Jared Cooney Horvath, Simon J Vogrin, Olivia Carter, Mark J Cook, andJason D Forte. Effects of a common transcranial direct current stimulation(tDCS) protocol on motor evoked potentials found to be highly variablewithin individuals over 9 testing sessions.
Experimental brain research ,234(9):2629–2642, 2016. 1920] Claudia Ammann, Martin A. Lindquist, and Pablo A. Celnik. Response vari-ability of different anodal transcranial direct current stimulation intensitiesacross multiple sessions.
Brain Stimulation , 10(4):757–763, 2017.[21] Lucia M. Li, Kazumasa Uehara, and Takashi Hanakawa. The contributionof interindividual factors to variability of response in transcranial directcurrent stimulation studies.
Frontiers in Cellular Neuroscience , 9(MAY),2015.[22] Lia Papadopoulos, Christopher W. Lynn, Demian Battaglia, and Danielle SBassett. Relations between large scale brain connectivity and effects ofregional stimulation depend on collective dynamical state. bioRxiv , 2020.[23] Laura A. Ewell, Liang Liang, Caren Armstrong, Ivan Soltész, Stefan Leutgeb,and Jill K. Leutgeb. Brain state is a major factor in preseizure hippocampalnetwork activity and influences success of seizure intervention.
Journal ofNeuroscience , 35(47):15635–15648, 2015.[24] Ronald P Lesser, Hyang Woon Lee, W. R.S. Webber, Barry Prince, Nathan E.Crone, and Diana L. Miglioretti. Short-term variations in response distribu-tion to cortical stimulation.
Brain , 131(6):1528–1539, 2008.[25] Asif Jamil, Giorgi Batsikadze, Hsiao I. Kuo, Ludovica Labruna, AlkomietHasan, Walter Paulus, and Michael A. Nitsche. Systematic evaluation ofthe impact of stimulation intensity on neuroplastic after-effects induced bytranscranial direct current stimulation.
Journal of Physiology , 595(4):1273–1288, 2017.[26] Uma R. Mohan, Andrew J. Watrous, Jonathan F. Miller, Bradley C. Lega,Michael R. Sperling, Gregory A. Worrell, Robert E. Gross, Kareem A.Zaghloul, Barbara C. Jobst, Kathryn A. Davis, Sameer A. Sheth, Joel M.Stein, Sandhitsu R. Das, Richard Gorniak, Paul A. Wanda, Daniel S.Rizzuto, Michael J. Kahana, and Joshua Jacobs. The effects of direct brainstimulation in humans depend on frequency, amplitude, and white-matterproximity. bioRxiv , page 746834, 2019.[27] Gabriella Deli, Istvan Balas, Ferenc Nagy, Eva Balazs, Jozsef Janszky,Samuel Komoly, and Norbert Kovacs. Comparison of the efficacy of unipolarand bipolar electrode configuration during subthalamic deep brain stimula-tion.
Parkinsonism & Related Disorders , 17(1):50–54, jan 2011.[28] Ying Zu Huang, Mark J. Edwards, Elisabeth Rounis, Kailash P. Bhatia,and John C. Rothwell. Theta burst stimulation of the human motor cortex.
Neuron , 45(2):201–206, 2005.[29] A Suppa, Y-Z Huang, K Funke, M C Ridding, B Cheeran, V Di Lazzaro,U Ziemann, and J C Rothwell. Ten years of theta burst stimulation inhumans: established knowledge, unknowns and prospects.
Brain stimulation ,9(3):323–335, 2016. 2030] Alexander Opitz, Walter Paulus, Susanne Will, Andre Antunes, and AxelThielscher. Determinants of the electric field during transcranial directcurrent stimulation.