Detection of Language Lateralization using Spectral Analysis of EEG

Abstract

Introduction:

Language lateralization relies on expensive equipment and can be difficult to tolerate. We assessed if lateralized brain responses to a language task can be detected with spectral analysis of EEG.

Methods:

Twenty right-handed, neurotypical adults (28+/−10 years; 5 males) performed a verb-generation task and two control tasks (word listening; repetition). We measured changes in EEG activity elicited by tasks (the event related spectral perturbation [ERSP]) in the theta, alpha, beta, and gamma frequency bands in two language (superior temporal and inferior frontal [ST, IF]) and one control (Occipital [Occ]) region bilaterally. We tested whether language tasks elicited: (1) changes in spectral power from baseline (significant ERSP) at any region; of (2) asymmetric ERSPs between matched left and right regions.

Results:

Left IF beta power (−0.37 +/− 0.53, t=−3.12, p=0.006) and gamma power in all regions decreased during verb generation. Asymmetric ERSPs occurred between the: (1) IF regions in the beta band (right ERSP > left IF by 0.23 +/− 0.37, t(19)=−2.80, p=0.0114); and (2) ST regions in the alpha band (right ERSP > left by 0.48 +/− 0.63, t(19)=−3.36, p=0.003). No changes from baseline or hemispheric asymmetries were noted in language regions during control tasks. On the individual level, 16 (80%) participants showed decreased left IF beta power from baseline, and 16 showed ST alpha asymmetry (right>left). Eighteen participants (90%) showed one of these two findings.

Conclusions:

Spectral EEG analysis detects lateralized responses during language tasks in frontal and temporal regions. Spectral EEG analysis could be developed into a readily-available language lateralization modality.

Lateralizing language is important both for the clinical evaluation of neurosurgical candidates and for studying language processing. Functional magnetic resonance imaging (fMRI) maps language based on elevations in blood oxygenation level dependent (BOLD) signal. Magnetoencephalography (MEG) is a newer modality that identifies language cortex by measuring changes in magnetic field elicited during language tasks.^1–4 MEG identifies decreased beta frequency band power in the left inferior frontal and superior temporal gyri as well as increased power in matched right hemisphere regions.^1,3–6 BOLD signal elevations have been correlated with decreased power on MEG,⁷ suggesting the reduction in power reflects increased metabolic demand in the underlying neural population.⁸ Right hemispheric power increases are hypothesized to reflect neural inhibition.^9–10

Both fMRI and MEG require highly specialized and costly equipment¹¹ and can be difficult to tolerate, especially for children and patients with neurological disorders.¹² Electroencephalography (EEG) has greater temporal specificity than fMRI and is more available than MEG, but its capacity to determine language lateralization is less clear. An early study¹³ found lateralized EEG responses using verb generation (VG), a classic word retrieval task that elicits lateralized fMRI¹⁴ and MEG^1,3 responses. The authors identified lateralized event-related potentials (ERPs) (reproducible perturbations of EEG amplitude) that occurred early in the temporal and later in the frontal region during VG but absent during passive listening or repetition. ERP analyses can miss brain signals whose timing varies across trials. In contrast, time-frequency analyses, looking for spectral power changes from baseline, can detect responses that have variability in onset as could be expected in language tasks. Surprisingly limited literature explores the feasibility of using EEG time-frequency analysis to lateralize language.

The main purpose of this pilot study is to test if spectral features measured by EEG lateralize language. To do this, we recorded EEG on healthy, right-handed, English-speaking adults as they participated in covert VG and two control tasks (word listening and repetition) and calculated the Event-Related Spectral Perturbation (ERSP), a measure that highlights task-related changes in power from baseline. Positive ERSP values, called event-related synchronizations (ERS), indicate increased power compared to baseline; negative ERSP values, called event-related desynchronizations (ERD), indicate a decrease from baseline power.¹⁵ We first assessed if language tasks evoked a significant change in power from baseline in frontal and temporal regions. We next tested whether there was a significant asymmetry in hemispheric responses to the language task by comparing left vs. right hemisphere ERSPs to each other. Based on MEG literature^1,3, we hypothesized that VG would elicit an ERD in left frontal and left temporal regions and an ERS in the right temporal region, and therefore significant asymmetry in temporal responses.

Methods

Inclusion & Exclusion Criteria:

Right-handed adults with English as their first language were invited to participate. Exclusion criteria included history of prematurity (< 37-weeks gestation), epilepsy, brain injury, neurosurgery or medication use. We recorded age, sex, score on the Edinburgh handedness inventory¹⁶, and knowledge of other languages. This study was approved by the Stanford University School of Medicine Institutional Review Board and participants provided informed consent.

Data Acquisition

EEG Recording:

Patients sat in a comfortable chair facing a computer monitor with two speakers equally spaced on either side of the head. EEG was recorded via 257-channel saline cap (Electrical Geodesics, Inc.) with a sampling rate of 500Hz. Impedances were kept below 50 kΩ throughout the session. Stimuli were presented using E-Prime 3.0 software.¹⁷

Experimental Set-Up (Figure 1):

Figure 1: Experimental Set-Up.

A. I) The participant looked at a fixation cross while 500ms of resting baseline EEG was recorded. II) A noun was played via the speakers. III) During the VG task, the participant was instructed to think of an associated verb. During the repetition task, the participant was asked to remember the noun. There were no instructions for the listening task. IV) After 3 seconds, a tone was played. During the VG task, the participant stated the generated verb. During the repetition task, the participant repeated the noun. During the passive listening task, the participant sat quietly. B. Regions of Interest, Inferior Frontal (IF) (rectangles; left IF channels 47, 48, 49; right IF channels 2, 222, 213), Superior Temporal (ST) (triangles; left ST channels 62, 69, 74, 84; right ST channels 211, 202, 192, 179), and Occipital (Occ) (circles; left Occ channels 97, 98, 108, 109, 116; right Occ channels 140, 150, 151, 152, 161).

Participants completed three tasks (covert VG, passive listening, and repetition) in separate blocks; there were 2 blocks of VG and one each of listening and repetition. Block order was randomized per-participant. Each block consisted of 60 9-second trials, for a total experimental time of approximately 60 minutes.

Each trial began with a fixation cross on the screen while baseline EEG was recorded. A noun was then played via speakers. Separate word lists of concrete nouns balanced for word frequency and length were used for each block, adapted from a prior experiment.¹³ Noun presentation lasted 500–1000ms. During VG, the participant thought of a verb associated with that noun. After 3 seconds, a tone was played, indicating the trial’s end. After this tone, the participant stated the verb to confirm engagement. During repetition, the participant repeated the noun after the tone. During passive listening, the participant sat quietly. Before each block, the participant was prepared with written and auditory instructions and practice stimuli.

Data Analysis

Preprocessing:

Data were preprocessed using the semi-automated ARTIST pipeline¹⁸; see supplemental section for details. Each trial yielded a 4.9 second epoch of EEG data, consisting of 1.9 seconds before and 3 seconds after noun presentation.

Calculation of Event-Related Spectral Perturbation (ERSP):

We calculated time-frequency representations of the EEG using the FieldTrip Toolbox in MATLAB¹⁹ by applying sliding window spectral analysis with Morlet wavelets to each 4.9 second epoch of cleaned EEG. Frequencies between 5–55Hz were extracted in 1Hz bins using wavelets with a width of 3 cycles allowing for a frequency resolution (F) of 2F/(number of cycles), or 2F/3 (Hz). Wavelets were shifted by 10 ms per time window, yielding 490 time points per trial. We averaged the power spectra of all trials within each task. (Of note, in our primary analysis, we lumped trials from both VG blocks together, while in a supplemental analysis testing reproducibility, we considered trials from the first and second VG blocks separately.) We log-transformed this averaged power spectra and subtracted out the log of baseline power (measured 250 to 50ms prior to trial onset) from the task power (measured in the 3 seconds after noun presentation) to generate the ERSP (ERSP = log([Task Power]/[Baseline Power]). Once the ERSP was calculated for each electrode, frequency bin, and time point, we reduced the data to focus on specific components of interest as follows:

Frequencies of Interest:

We averaged across frequency bins to derive the ERSP for four frequency bands: theta (4 to <8 Hz), alpha (8 to <13 Hz), beta (13 to <30 Hz), gamma (30 to <50 Hz). We chose these canonical frequency bands based on prior language processing studies.^20–21

Regions of Interest:

Based on prior work^1,3,13, we hypothesized asymmetric hemispheric responses would be detectable during VG specifically in the superior temporal (ST) and inferior frontal (IF) regions. We included the Occipital region (Occ) as a control, because auditory VG has not been shown to elicit asymmetry between occipital regions.^3,13 We conducted analysis in channel space by averaging electrode clusters (Figure 1) previously noted to sample anatomic regions of interest.²²

Time Points of Interest:

Our primary analysis focused on spectral changes during the task period between 750–3000ms. We excluded the first 750ms during noun presentation, as the earliest EEG responses reflect primary auditory cortex responses and we were interested in asymmetries that develop with spread of information to specific language regions.²³ Furthermore, MEG shows spectral changes starting 500 to 650ms following stimulus presentation^1,3. As spectral responses may be detectable in temporal before frontal regions^1,13, we considered early (750–1500ms) and late epochs (1500–3000ms) separately in secondary analyses.

Statistical Analysis:

We first assessed if the ERSP during VG significantly differed from zero in any ROI/frequency band using one-sample t-tests. We used Bonferroni correction to account for the four frequency bands tested, setting the 2-sided significance threshold at p<0.0125. We did not correct for number of regions tested as we had a priori hypotheses for each region based on prior MEG and EEG studies. We hypothesized there would be ERD in the left IF and left ST, ERS in the right ST, and no change in the right IF or bilateral Occ.^3,4

We next assessed for asymmetry in spectral responses between hemispheres by comparing ERSPs in matched left and right ROIs using paired t-tests. We chose this analysis as prior studies have found hemispheric responses during language tasks diverge^1–3 and a paired comparison could identify divergence even when ERSP was not significantly different from zero. We again set significance at p<0.0125 to correct for number of frequency bands tested. We present an Asymmetry Score (AS) [right - left ERSP]; a positive AS indicates right hemisphere ERSP is more positive than left. We expected significant asymmetry in the IF and ST but not Occ regions.³

We conducted two secondary analyses of the VG data using similar statistical methods. First, we assessed if spectral responses were replicable by testing the two VG blocks separately. Second, we divided the task period into an early (750–1500ms) and late (1500–3000ms) block to test temporal specificity of responses.

We next tested if control tasks elicited either significant power changes from baseline or asymmetry between hemispheres, employing one-sample and paired t-tests respectively with a 2-sided significance threshold of 0.0125 as previously described.

Finally, we examined whether spectral changes from baseline or asymmetry between hemispheres were detectable at an individual level. We furthermore tested if the magnitude of the ERSP or Asymmetry Score correlated with age or handedness using Spearman’s correlation coefficient or differed based on sex (male/female) or bilingualism (yes/no) using the Mann-Whitney U test.

Results

Participants:

Twenty healthy adults (five males; 28.3 +/−7.7 years; range, 19–41 years) participated. Fourteen participants spoke only English, and six spoke additional languages, though English was their first and primary language. All participants were right-handed, supported by Edinburgh Handedness Inventory scores (77.27 +/− 22.56).

Change in Spectral Power from Baseline During VG (Table 1, Figure 2):

Table 1:

Hemispheric Spectral Responses Elicited by Verb Generation, Repetition, and Passive Listening

Regions of Interest	Frequency Band
Verb Generation	Theta	p	Alpha	p	Beta	p	Gamma	p
Inferior Frontal
Left ERSP, Mean [SD]	0.11 [0.70]	0.51	−0.20 [0.74]	0.24	−0.37 [0.53]	0.006	−0.53 [0.56]	<0.001
Right ERSP, Mean [SD]	0.22 [0.73]	0.2	−0.02 [0.76]	0.91	−0.14 [0.59]	0.31	−0.52 [0.55]	<0.001
AS, Mean [SD]	0.11 [0.61]	0.43	0.18 [0.65]	0.23	0.23 [0.37]	0.01	0.01 [0.22]	0.79
Superior Temporal
Left ERSP, Mean [SD]	0.12 [0.56]	0.35	−0.18 [0.59]	0.19	−0.22 [0.48]	0.81	−0.37 [0.45]	0.002
Right ERSP, Mean [SD]	0.40 [0.74]	0.03	0.30 [0.81]	0.12	−0.03 [0.60]	0.06	−0.30 [0.46]	0.009
AS, Mean [SD]	0.28 [0.50]	0.02	0.48 [0.63]	0.003	0.19 [0.37]	0.04	0.07 [0.26]	0.26
Occipital
Left ERSP, Mean [SD]	−0.001 [0.64]	0.99	−0.20 [1.12]	0.43	−0.21 [0.67]	0.18	−0.42 [0.47]	<0.001
Right ERSP, Mean [SD]	0.04 [0.68]	0.78	−0.23 [1.03]	0.33	−0.19 [0.68]	0.22	−0.40 [0.45]	<0.001
AS, Mean [SD]	0.04 [0.48]	0.69	−0.03 [0.64]	0.86	0.02 [0.47]	0.91	0.02 [0.25]	0.65
Repetition	Theta	p	Alpha	p	Beta	p	Gamma	p
Inferior Frontal
Left ERSP, Mean [SD]	−0.23 [0.60]	0.10	−0.10 [0.65]	0.49	−0.16 [0.54]	0.20	−0.29 [0.79]	0.12
Right ERSP, Mean [SD]	−0.13 [0.55]	0.30	0.02 [0.63]	0.91	−0.11 [0.56]	0.39	−0.27 [0.60]	0.05
AS, Mean [SD]	0.10 [0.55]	0.43	0.12 [0.63]	0.41	0.05 [0.39]	0.55	0.02 [0.36]	0.81
Superior Temporal
Left ERSP, Mean [SD]	−0.19 [0.65]	0.20	−0.17 [0.78]	0.34	−0.19 [0.61]	0.17	−0.24 [0.57]	0.08
Right ERSP, Mean [SD]	−0.19 [0.54]	0.14	−0.11 [0.60]	0.43	−0.16 [0.58]	0.25	−0.23 [0.67]	0.14
AS, Mean [SD]	0.00 [0.55]	0.95	0.06 [0.54]	0.63	0.03 [0.38]	0.67	0.01 [0.29]	0.90
Occipital
Left ERSP, Mean [SD]	−0.21 [0.62]	0.14	−0.22 [0.72]	0.20	−0.29 [0.59]	0.04	−0.37 [0.63]	0.02
Right ERSP, Mean [SD]	−0.22 [0.78]	0.22	−0.22 [0.84]	0.26	−0.25 [0.72]	0.13	−0.38 [0.74]	0.03
AS, Mean [SD]	−0.01 [0.64]	0.94	0.00 [0.89]	0.99	0.04 [0.57]	0.78	−0.01 [0.32]	0.87
Passive Listening	Theta	p	Alpha	p	Beta	p	Gamma	p
Inferior Frontal
Left ERSP, Mean [SD]	−0.11 [0.82]	0.55	0.18 [0.77]	0.31	−0.06 [0.53]	0.47	−0.05 [0.54]	0.66
Right ERSP, Mean [SD]	−0.10 [0.58]	0.43	0.05 [0.65]	0.71	0.09 [0.38]	0.48	−0.18 [0.39]	0.05
AS, Mean [SD]	0.01 [0.77]	0.97	−0.13 [0.48]	0.26	0.15 [0.37]	0.09	−0.13 [0.40]	0.19
Superior Temporal
Left ERSP, Mean [SD]	−0.08 [0.67]	0.61	0.10 [0.70]	0.54	0.08 [0.39]	0.96	−0.11 [0.52]	0.36
Right ERSP, Mean [SD]	−0.11 [0.66]	0.45	0.23 [0.82]	0.22	0.00 [0.43]	0.41	−0.18 [0.48]	0.11
AS, Mean [SD]	−0.03 [0.65]	0.81	0.13 [0.69]	0.39	−0.03 [0.34]	0.32	−0.07 [0.32]	0.35
Occipital
Left ERSP, Mean [SD]	0.15 [0.64]	0.32	0.14 [0.74]	0.39	−0.04 [0.53]	0.72	−0.20 [0.35]	0.02
Right ERSP, Mean [SD]	0.08 [0.77]	0.63	0.33 [0.98]	0.15	0.21 [0.60]	0.13	0.01 [0.41]	0.94
AS, Mean [SD]	−0.07 [0.61]	0.65	0.19 [0.70]	0.26	0.25 [0.40]	0.01	0.21 [0.34]	0.02

Statistical significance set at p<0.0125 to account for multiple frequency bands.

AS = Asymmetry Score (Right ERSP – Left ERSP). ERSP = Event-Related Spectral Perturbation (log([TaskPower]/[Baseline Power]). SD = Standard Deviation.

Figure 2: Spectral Responses during Verb Generation Task.

A. Average Left and Right ERSP over time, by frequency band (columns) and regions of interest (rows). The y-axis (black line) represents initiation of auditory noun presentation. Baseline is 250 to 50ms prior to auditory noun presentation. The left solid vertical line represents the beginning of the analyzed task period, which extends from 750–3000ms. B. Average ERSP Across Regions of Interest and Frequency Bands. Plus (+) indicate ROIs that demonstrate a significant change in power from baseline (p<0.0125). Shaded bars indicate regions and frequency bands where the left vs. right hemispheric ERSPs significantly differ from each other (p<0.0125).

Inferior Frontal:

There was significant beta ERD in the left IF (−0.37 +/− 0.53, t=−3.12, p=0.006). Additionally, there was significant gamma ERD in left (−0.53 +/− 0.56, t=−4.28, p=0.0004) and right (−0.52 +/− 0.55, t=−4.24, p=0.0004) IF.

Superior Temporal:

There was significant gamma ERD in left (−0.37 +/−0.45, t=−3.63, p=0.002) and right (−0.30 +/− 0.46, t=−2.91, p=0.009) ST.

Occipital:

There was significant gamma ERD in left (−0.42 +/−0.47, t=−4.01, p=0.0004) and right (−0.40 +/−0.45, t=−3.93, p=0.001) Occ.

Hemispheric Asymmetry in Spectral Response During VG (Table 1, Figure 2):

Inferior Frontal:

There was a larger decrease in beta power in the left (−0.37 +/−0.53) than right IF (−0.14 +/−0.59), leading to significant beta asymmetry (Asymmetry Score [AS] 0.23 +/− 0.37, t(19)=−2.80, p=0.0114).

Superior Temporal:

Alpha power decreased in the left (−0.18 +/−0.59) and increased in the right ST (0.30 +/−0.81), leading to significant alpha asymmetry (AS = 0.48 +/− 0.63, t(19)=−3.36, p=0.003).

Occipital:

There were no significant asymmetries in the Occ regions in any frequency band.

Reproducibility of Spectral Findings (Supplemental Table 1, Supplemental Figure 1):

Left IF beta ERD and alpha ST asymmetry were seen when analyzing each VG block separately, as was diffuse gamma ERD in most ROIs.

Time Course of Effects (Supplemental Table 2, Supplemental Figure 2):

Significant left IF beta ERD was present only in the latter part (1500–3000ms) of the task period (−0.40 +/− 0.54, t(19)=−3.31, p=0.004), though a significant asymmetry between the two IF (AS 0.27 +/−0.41, t(19)=2.96, p=0.008) was noted in the earlier time window (750–1500ms). The ST alpha asymmetry was significant in both early and later segments, but the magnitude was larger in the earlier time window (AS 0.57 +/− 0.80, t(19)=3.19, p=0.005).

Spectral Response During Control (Repetition and Listening) Tasks (Table 1, Figure 3)

Figure 3: Spectral Responses during Control Tasks.

Average left and right ERSP over time, by frequency band (columns) and regions of interest (rows) for Repetition (A) and Passive Listening (B). The y-axis (black line) represents initiation of auditory noun presentation. The solid vertical line represents the beginning of the analyzed whole task period (750–3000ms). Average ERSP for the task period elicited by Repetition (C) and Passive Listening (D). Shaded bars indicate regions and frequency bands where left vs. right hemispheric ERSPs significantly differ from each other (p<0.0125).

Change in Spectral Power from Baseline:

There were no significant changes in spectral power from baseline in any frequency band/ROI during repetition or passive listening.

Hemispheric Asymmetry in Spectral Response:

During repetition, there was no asymmetry in spectral responses in any region. Passive listening elicited asymmetric beta responses only between Occ regions (Left Occ −0.04 +/− 0.53; Right Occ 0.22 +/− 0.60; AS 0.25+/− 0.40, t(19)=2.83, p=0.01).

Individual Responses to VG

We explored whether frontal beta or temporal alpha responses were consistent across individuals (Figure 4, Supplemental Table 3).

Figure 4: ERSP Values of Individual Participants.

The ERSP of individual participants in the (A) left and right inferior frontal regions; and (B) the left and right superior temporal regions. Left and right hemispheric values from an individual are connected with a line. The y-axis reflects the Event Related Spectral Perturbation (ERSP), with positive values indicating increased power from baseline (event related synchronization [ERS]) and negative values indicating decreased power from baseline (event related desynchronization [ERD]). The dashed lines reflect participants whose right ERSP was less than the left ERSP (negative Asymmetry Score), and solid lines reflect those whose right ERSP was greater than left ERSP (positive Asymmetry Score).

Inferior Frontal:

Sixteen of 20 (80%) participants showed left and 13/20 (65%) showed right IF beta ERD. Fourteen (70%) showed a positive beta IF AS, indicating that beta power was relatively higher in the right vs. left IF. Thirteen (65%) participants showed both a positive AS and decrease in left IF beta power.

Superior Temporal:

Eleven of twenty (55%) participants showed left ST alpha ERD and 14/20 (70%) showed right ST alpha ERS. Sixteen (80%) showed a positive AS, indicating relatively higher alpha power in the right vs. left ST.

Eighteen (90%) of subjects demonstrated either left IF beta ERD or an alpha ST positive AS.

Associated Clinical Characteristics (Table 2):

Table 2:

Clinical Characteristics are not associated with Hemispheric ERSP or Asymmetry

	Age		Handedness		Sex (F vs. M)		Bilingual (vs. Monolingual)
	r	p	r	p	U	p	U	p
Inferior Frontal: Beta
Left ERSP	0.40	0.08	−0.10	0.68	0.43	0.51	1.33	0.25
Right ERSP	0.29	0.21	0.10	0.68	0.15	0.69	0.33	0.56
Asymmetry Score	−0.07	0.76	0.31	0.18	0.02	0.90	0.11	0.74
Superior Temporal: Alpha
Left ERSP	0.20	0.40	0.14	0.56	1.83	0.18	0.17	0.68
Right ERSP	0.38	0.10	−0.09	0.69	0	0.97	0.03	0.87
Asymmetry Score	0.18	0.45	−0.19	0.42	1.01	0.32	0.17	0.68

We assessed whether the Event Related Spectral Perturbation (ERSP) or the Asymmetry Score (Right ERSP-Left ERSP) correlated with age (years) or handedness (Edinburgh Handedness Inventory Score; positive values are associated with right-handedness; negative values with left-handedness) using Spearman’s Correlation Coefficient. We assessed if median ERSP or Asymmetry Score differed between females vs. males and bilinguals vs. monolinguals using the Mann-Whitney U test.

Clinical characteristics of age, sex, handedness or bilingualism were not significantly associated with spectral responses in the IF beta band or the ST alpha band of hemispheric asymmetry.

Discussion

This study suggests spectral analysis of EEG can identify language lateralization. We assessed whether VG elicited either changes in spectral power from baseline in frontal or temporal regions or significant asymmetry in spectral responses between homologous regions in each hemisphere. When compared to baseline, the left frontal region showed decreased beta power (ERD) and all regions showed gamma ERD. Furthermore, VG elicited significant asymmetry between frontal beta responses and temporal alpha responses. Frontal and temporal changes in power or asymmetries were not evident during control tasks.

Task-Evoked Changes in Spectral Power from Baseline

The left frontal region showed significant, task-evoked changes in beta power during VG and not control tasks, suggesting EEG spectral analysis identifies changes in brain activity specific to language processing. MEG studies also identify left frontal beta ERD during VG^1,3, thought to reflect increased demands on verbal working memory.^1,3,4,6,24 More generally, MEG and EEG spectral analysis studies have associated beta changes with higher order linguistic processing, including word discrimination and semantic unification, across a variety of language tasks and brain regions.^24–26 Alpha/beta ERD likely signify increased regional metabolic demand, as they correlate with increased fMRI BOLD signal.^{7,8,10,27–30} We speculate that left IF beta ERD we observe reflects demands of verbal working memory; fitting this hypothesis, we note that beta power decreases (though not significantly so) during repetition but not passive listening. An alternative explanation to our findings is that the frontal beta ERD reflects activation of regions engaged in motor preparation as has been previously described extensively.³¹ If the beta ERD represented solely motor processes, however, we would expect it to be of equal magnitude in the verb generation and repetition tasks. Therefore, we posit that either verbal working memory load or potentially attentional processes contribute to this spectral change; these hypotheses could be tested in future studies.

MEG studies have identified additional spectral changes we did not detect with EEG, including right frontal beta ERS that the authors link to active inhibition of the nondominant hemisphere.^1,3 However, they found right frontal ERS was less consistent across participants than left frontal ERD¹. Prior work has also described decreased alpha power (MEG) and negative evoked potentials (EEG) in temporal areas.^2,13 We did not observe significant temporal changes from baseline, though we found trends toward left alpha ERD and right alpha ERS. Our study may be underpowered. Alternatively, time windows and ROIs chosen may not have been ideal. While our analysis of early/later subsegments did not show stronger results, future studies could pursue a more thorough search of the parameter space with more participants.

We also found significant reduction in gamma power across language and occipital regions during VG but not control tasks. Some authors have reported that passive listening elicits changes in high gamma activity (65–140Hz) in left temporal and frontal areas on intracranial recordings.^33–35 Unfortunately it is particularly challenging to detect high gamma with scalp EEG due to its low amplitude and contamination with high-frequency muscle artifact³⁵, and thus we may miss high gamma activity with our current methodology; our analysis focused on lower gamma (30–50Hz). A past study showed gamma (25–50Hz) ERD in bilateral angular gyrus and occipital cortex, as well as left-lateralized regions during a syntactic processing task (mental verb conjugation into past tense).³⁷ The gamma ERD we observe during VG may reflect syntactic skills (word order) engaged by combining nouns with verbs, but not to the same extent by passive listening or repetition. Given that gamma ERD affects all regions and occurs, though reduced, even in control tasks, gamma changes may signify cognitive processes common to all tasks.

Task-Evoked Asymmetries between Left and Right Hemisphere Activity

In our second analysis, we examined whether spectral changes elicited by VG diverged between matched right and left regions. We performed this analysis because we hypothesized divergence in hemispheric activity might capture inter-hemispheric interactions that could be missed by comparison to baseline activity alone. Our analysis identified significant differences in left and right temporal alpha responses and frontal beta responses. We speculate that temporal alpha asymmetry reflects active inhibition of right temporal regions paired with release of inhibition in left temporal regions. Prior language processing studies have found right hemispheric power in several frequency bands increases to a greater extent in adolescents³ and adults⁴¹ than children, suggesting right hemispheric inhibition is important for language network consolidation and efficiency. Using EEG spectral analysis methods, right-hemispheric ERS has also been correlated with working memory storage demands³⁹ and task performance¹⁰, while left temporal alpha ERD has been correlated with sentence recall accuracy.³⁹ In line with prior studies, we believe the relative increase in right alpha power to reflect functional inhibition in task-irrelevant cortical areas, while the relative decrease in left alpha power may reflect word-level memory and syntactic processing⁴⁰. It is therefore possible that VG is demanding enough to detect asymmetric ERSP between hemispheres, but not demanding enough to elicit significant ERS³⁹ or ERD⁴⁰ measurable by EEG.

Clinical Implications

Since current language lateralizing modalities rely on expensive, highly-specialized equipment and can be difficult to tolerate, there is a clinical need to develop readily available, minimally-invasive alternatives. Our study suggests that spectral analysis of EEG could be developed into a clinically relevant lateralization technique. Similar to MEG, the majority (80%) of participants had left frontal beta ERD as well as asymmetry between frontal beta ERSPs (right>left). While change in temporal power from baseline varied across individuals, 80% of participants showed asymmetry in the temporal alpha ERSP (right>left) between hemispheres. Fourteen (70%) had both the frontal ERD and temporal asymmetry and 18 (90%) had one of these two findings. In this exploratory analysis, we utilize individuals’ mean ERSP and mean asymmetry, and future studies should investigate specific cut-offs for these values that best capture lateralization.

Limitations

Several limitations merit discussion. In this pilot, we did not obtain imaging, such as fMRI, to confirm left hemispheric language lateralization; we tailored our inclusion and exclusion criteria to mitigate this issue as there is good concordance between right-handedness left-lateralized language.⁴² A second consideration is that we may not have identified the optimal time-window or frequency band to lateralize language. We averaged over fairly long time windows even though EEG can be more temporally specific. We did this as the noun-stimuli had some variability in duration (500–1000ms), but in the future, more restricted word lists could be constructed to study more specific temporal changes. Similarly, our analysis may miss findings in more specific frequency bands because the time-frequency transform we used prioritizes temporal over frequency specificity. Despite this, we do see band-specific frequency differences. More precise characterization of temporal or frequency-specific changes could be pursued in future studies. We additionally corrected for the number of frequency bands but not the number of regions, because we had a priori hypotheses based on the MEG literature about how each region would respond. With application of a more stringent correction, the superior temporal alpha asymmetry and bifrontal and bioccipital gamma ERDs remain significant but the frontal beta ERD findings no longer clear this bar. However, as frontal beta ERD has been reported in the MEG literature, it may still be reasonable to explore the reliability of these frontal changes in future EEG studies. Further, given our sample size is relatively small and all our subjects were right-handed and healthy, it is important to replicate our findings with a larger sample size, and to assess if our technique can lateralize language in patients with neurological disorders, and those who are left-handed or ambidextrous.

Supplementary Material

SupplementalFigure 1

Supplemental Figure 1: Average Left and Right ERSP over time during first block of VG (1a) and second block of VG (1b), by frequency band (columns) and regions of interest (rows). The y-axis (black line) represents initiation of auditory noun presentation. Baseline is 250 to 50ms prior to auditory noun presentation. The left solid vertical line represents the beginning of the analyzed whole task period, which extends from 750–3000ms. The right solid vertical line indicates the point at which the whole task period is divided into early (750–1500ms) and late (1500–3000ms) segments for secondary analyses. 1c. Consistency of Responses between First and Second VG Blocks. Shaded bars indicate regions and frequency bands where the left vs. right hemispheric responses significantly differ from one another (p<0.0125), and Plus marks significant ERD or ERS (p<0.0125).

SupplementalFigure 2

Supplemental Figure 2. Average ERSP Across Regions of Interest and Frequency Bands in Whole task period (left column), Early task period (750–1500ms, middle column) and Late task period (1500–3000ms, right column). Plus (+) sign indicates ROIs that demonstrate a significant change in power from baseline (p<0.0125). Shaded bars indicate regions and frequency bands where the left vs. right hemispheric responses significantly differ from one another (p<0.0125).

Supplemental Figure 3

Supplemental Figure 3. Time-frequency plots of Group Average ERSP Across Regions of Interest for Verb Generation (A); Repetition (B); and Passive Listening (C). Time since noun presentation is represented on the x-axis and frequency is represented on the y-axis. For each of the three tasks, the left column represents the left hemisphere ERSP, the middle column the right hemisphere ERSP, and the right column the asymmetry score (right – left ERSP). For the ERSP plots, red indicates an increase in log power (ERS) and blue indicates a decrease in log power (ERD) from baseline. For the asymmetry score plots, red indicates that the log power change from baseline was more positive in the right vs. left hemisphere; blue indicates that the log power change from baseline was more positive in the left than right hemisphere.