Fifteen individuals with mTBI occurring in the past 3 years (mTBI group, 12 men, 3 women, mean age 27.1 ± 6.2 years) and 22 individuals with no history of head injury (control group, 21 men, 1 women, mean age 27.9 ± 6.1 years) were included in this study for comparisons. Both mTBI and controls were from the same cohort of military service members within 2 months of their return from a deployment in either Iraq or Afghanistan. The diagnosis of mTBI was determined by the Military Acute Concussion Evaluation (29). Most of the mTBI participants reported multiple head injury events. The types of head injury include combat-related blast exposure (n = 8), motor vehicle accident (n = 6), sports injury (n = 5), and accidental fall (n = 3).

All participants were administrated a series of validated questionnaires including the PTSD Checklist-Military Version (PCL-M), a self-administered screen for PTSD (30), the Patient Health Questionnaire-9 (PHQ-9), a widely used instrument for screening depression (31), and the Pittsburgh Sleep Quality Index (PSQI), a self-report instrument for assessing sleep over the prior month (32). Inclusion in the study required a score of less than 10 on the PHQ-9 and less than 50 on the PCL-M. In addition, an experienced, licensed psychologist conducted the Clinician-Administered PTSD Scale for DSM-IV (CAPS) (33) with all participants to confirm the absence of a PTSD diagnosis. Additional exclusion criteria for this study included: a current Glasgow Coma Scale score less than 13; a history of head injury resulting in loss of consciousness for 60 min or more; active suicidal or homicidal ideation; pregnancy; a diagnosis of postconcussive syndrome (PCS) according to International Classification of Diseases, 10th Clinical Modification criteria.

The data analyzed here are from a larger study exploring predictors for postdeployment delayed-onset PTSD, major depressive disorder and PCS. Therefore, the military service members who met the diagnostic criteria for PTSD, depression or PCS at the time of recruitment were excluded from the study. The experimental protocols were approved by institutional review boards at Uniformed Services University, Walter Reed National Military Medical Center, and the National Institutes of Health. All participants provided and signed written informed consent prior to participation.

Scalp EEG was recorded from all participants during eyes-closed resting state for about 2.5 min in an acoustically and electrically shielded room. The data were acquired using the EPA6 amplifier (Sensorium Inc.) and Grass electrodes (Natus Neurology Inc.) at Fz, Cz, Pz, Oz, C3, and C4 positioned according to the standard 10–20 electrode system, with linked earlobes as reference and a forehead ground. A limited number of electrodes was used to test the utility of a reduced montage that could be implemented in a far forward military medical environment. Electrode impedances were maintained under 5 kΩ. EOG was recorded from two electrodes placed below and above the right eye. The sampling rate was 2,048 Hz.

The EEG data were preprocessed offline using custom scripts written in MATLAB (www.mathworks.com). All data were first visually inspected to ensure that participants were awake throughout the recording period. Data from Oz channel were found to be contaminated by artifacts in three participants, and therefore excluded from further analysis. All the following analyses were conducted for Fz, Cz, Pz, C3, and C4 channels. EOG artifacts were corrected by using a regression approach (34). The data after EOG correction were high-pass filtered at 0.5 Hz, low-pass filtered at 50 Hz, and down-sampled to 256 Hz. To remove data portions contaminated by artifacts and to construct multiple trials for computing WPLI, we segmented the 2.5-min continuous EEG data into 1-s short epochs, resulting in 150 pseudotrials. Trials with activity exceeding 75 μV were excluded from analysis. The mean trial rejection rate was 2.83%.

As described by Vinck et al. (28), WPLI is a measure of phase synchronization based solely on the imaginary component of the cross-spectrum. Let X₁(f) and X₂(f) be the Fourier transform of the signals x₁(t) and x₂(t) from two separate sensors, respectively, at frequency f. Then the cross-spectrum between x₁(t) and x₂(t) is computed as (1)S(f)=X1(f)X2*(f), where X2* is the complex conjugate of X₂. If we write the Fourier transformed signals as X₁(f) = r₁exp(iθ₁) and X₂(f) = r₂exp(iθ₂) then the cross-spectrum can be rewritten as (2)S(f)=r1r2exp(iΔθ)=r1r2cos(Δθ)+i(r1r2sin(Δθ)), where Δθ = θ₁ − θ₂ is the phase difference between the two signals at frequency f. In what follows, we drop inclusion of f for notational convenience, but note that all values are frequency dependent. Let Im(S) denote the imaginary part of the cross-spectrum: (3)Im (S)=r1r2sin(Δθ).

Then the WPLI is defined as (4)WPLI=|〈Im(S)〉|〈|Im(S)|〉=|〈|Im(S)|sign(Im(S))〉|〈|Im(S)|〉, where 〈⋅〉 denotes an expectation. In practice the expectation is estimated by averaging over a large number of epochs. In Eq. 3, we see that the imaginary part of the cross-spectrum is only sensitive to synchronizations of two signals which have phase delays. From the quasistatic approximation, volume conduction does not induce a phase delay; therefore WPLI is insensitive to spurious connectivity created by volume conduction.

Weighted phase lag index has been proposed as an improved measure of phase synchronization compared to PLI (26), which is defined as (5)PLI=|〈sign(Im(S))〉|.

From the Eq. 3, we can see that the sign of the imaginary part of the cross-spectrum indicates whether signal x₁(t) tends to phase lead or lag signal x₂(t). Therefore PLI estimates to what extent the phase leads and lags between two signals are nonequiprobable. Compared to PLI, WPLI weights the contribution of the observed phase leads and lags by the magnitude of the imaginary part of cross-spectrum (see the Eq. 4). Cross-spectra around the real axis contribute to a lesser extent than cross-spectra around the imaginary axis (28). This makes WPLI less sensitive to noise as cross-spectra around the real axis are at risk of changing their true sign with small noise perturbations.

The estimate of WPLI obtained from Eq. 4 has a positive bias. To solve this problem, Vinck et al. (28) propose a debiased estimator of the squared WPLI, which is defined as (6)debiased WPLI-square=∑j=1N∑k≠jIm(Sj)Im(Sk)∑j=1N∑k≠j|Im(Sj)Im(Sk)|, where N is the number of epochs and S_j and S_k are the cross-spectrum between a pair of channels for the j-th and k-th epochs, respectively. In this study N was typically about 150. The smallest value of N used was 133. Equation 6 is debiased (strictly asymptotically unbiased) because it converges to the square of Eq. 4 in the limit of an infinite number of epochs. The theoretical development of this debiased estimator and its statistical performance can be found in the original method by Vinck et al. (28).

In the current study, we used the debiased WPLI-square estimator as the measure of EEG phase synchronization. We computed the debiased WPLI-square (which will be referred to as WPLI throughout the rest of the article) for all pairs of EEG electrodes based on Eq. 6. Cross-spectra were computed using the Fourier transforms of the preprocessed 1-s data epochs. The frequency resolution thus equaled 1 Hz.

Twelve out of 15 mTBI cases and 17 out of 22 controls had their DTI data recorded within a two day period from the EEG recording. DTI data were acquired on a Siemens Biograph mMR 3 T scanner with parameters TR (repetition time) = 17,000 ms, TE (echo time) = 98 ms, flip angle = 90°, voxel size = 2 mm × 2 mm × 2 mm, matrix size = 128 × 128, and slices = 75. The acquisition included 10 images at b = 0 s/mm², 10 images with non-collinear directional gradients at b = 300 s/mm², and 60 images with non-collinear directional gradients at b = 1,100 s/mm². Images were processed using the CATNAP software for tensor estimation (35). Two b-values were included for more robust tensor estimation (36) as well as for potentially exploring more sophisticated diffusion models in the future. All diffusion weighted images were included in the single tensor fit. Images were preprocessed for motion correction and eddy current correction, with adjustments to the gradient table performed based on patient position. Distortions due to echo planar imaging susceptibility artifacts were corrected by performing a deformable registration to an anatomic T2-weighted acquisition (TR = 3,200 ms, TE = 409 ms, flip angle = 120°, voxel size = 1 mm × 1 mm × 1 mm, matrix size = 256 × 256, slices = 176). Following preprocessing, all images were used to perform tensor estimation. Next, automated segmentation of white matter tracts was performed using the Diffusion-Oriented Tract Segmentation (DOTS) algorithm (37). DOTS combines statistical atlases of tract location and tensor orientation with a Markov random field model to map diffusion properties to tract labels. This enables average values of fractional anisotropy (FA) to be computed for each labeled tract. The statistical region of interest (ROI) atlas was based on the atlas of Ref. (38). We included the following fiber tracts which are susceptible to mTBI (39–43): corpus callosum, inferior and superior longitudinal and fronto-occipital fasciculi, cingulum, fornix, uncinate fasciculus, optic radiation, corticospinal and corticopontine tracts, inferior, middle, and superior cerebellar peduncles (separated for left and right when applicable).

Differences between groups in demographics and psychological measures were examined by two-sample t-tests whether the data are numerical or Fisher’s exact tests whether the data are categorical. The comparisons between the mTBI and control groups for the WPLI measures were tested using the Wilcoxon rank-sum tests. In order to identify oscillatory frequencies in neural synchronization showing abnormality following mTBI, we first performed the tests on the theta (4–7 Hz), alpha (8–13 Hz), beta (15–23 Hz), and low-gamma (25–40 Hz) frequency bands for WPLI spectra averaged across all electrode pairs. Then for the identified frequency band, we further performed the tests for each electrode pair to reveal the spatial pattern of the abnormality. To test whether or not the estimated WPLI values are significantly larger than zero, an empirical distribution technique using surrogate data was applied (44, 45). Specifically, we randomly and independently shuffled the time series data from each electrode to create a surrogate data set. Then the WPLI measures were computed from the surrogate data set. After repeating this process 1,000 times, we created empirical distributions for the WPLI measures. Since the shuffling process destroys all the temporal structure in the data, the empirical distributions provide variability for the null hypothesis case. We then used these distributions to assess the significance of the WPLI measures estimated from the actual data. For the FA measures from the DTI data, the comparisons between mTBI and control groups were tested using two-sample t-tests for each ROI and the p-values were corrected for multiple comparisons using the Holm-Bonferroni method. To examine the association between disruption of neural synchronization and the abnormality of white matter integrity, we computed Pearson’s correlation coefficient across participants between the WPLI measures and the FA measures that are found statistically significant between groups. For the WPLI measure, in order to accommodate the individual differences for the null hypothesis case, the measure of each participant was first normalized by its empirical null distribution (subtracted the mean and divided by the standard deviation) and then correlated with FA measure across participants. Statistical significance of correlation coefficients was tested by a t-statistic created from Fisher transform. The p-values less than 0.05 were considered statistically significant.

Table 1 summarizes the participant characteristics and psychological measures. Age, gender, and handedness were found to be statistically indistinguishable between the mTBI and the control groups (p > 0.05). Compared to controls, mTBI cases showed significantly higher PTSD scores including CAPS (total score, Criterion B subscore for Reexperiencing, Criterion D subscore for Hyperarousal) and PCL-M scores (p < 0.05). The PHQ-9 score for depression was also higher for mTBI compared to controls, but the difference did not reach significance (p = 0.076). The PSQI score for sleep quality over the past month was not statistically different between the two groups (p > 0.05).

Using the WPLI spectra averaged across all electrode pairs, comparisons between the control and mTBI groups revealed that the mean WPLI value in low-gamma frequency band (25–40 Hz) was significantly smaller (p = 0.0082) in those with a history of mTBI (Figures 1A,B). Moreover, application of testing using surrogate data found that the mean WPLI value in the low-gamma frequency band was significantly larger than zero for the control group (p < 0.001), but not significantly different from zero for the mTBI group (p > 0.05). The mean WPLI values in theta (4–7 Hz), alpha (8–13 Hz), and beta (15–23 Hz) frequency bands were not significantly different between two groups (p > 0.05).

The phase synchronization (measured by WPLI) at low-gamma frequency for different electrode pairs are displayed in Figures 1C,D for the control and the mTBI groups, respectively. The synchronization strength is indicated by line thickness and the significance of synchronization is indicated by red color. For the control population all pairs of electrodes except the C3-C4 pair exhibited significant (that is, greater synchronization than observed with surrogate data) low-gamma synchronization (Figure 1C). In contrast, none of the electrode pairs showed significant low-gamma synchronization for the mTBI cases (Figure 1D), suggesting a global reduction of synchronization instead of a reduction for only one or two electrode pairs. By comparing the two groups, we revealed that the reduction of low-gamma synchronization in the mTBI group was significant for the Cz-Pz (p = 0.0042, uncorrected), Fz-C3 (p = 0.023, uncorrected), C3-Pz (p = 0.017, uncorrected), and C4-Pz (p = 0.017, uncorrected) electrode pairs (see Figure 1E, significant electrode pairs were indicated by lines in green color).

Region of interest-based comparison revealed reduced FA in the white matter of the right inferior cerebellar peduncle (p = 0.0015, uncorrected, after correction p = 0.042) in those with a history of mTBI (Figure 2). The left inferior cerebellar peduncle ROI also showed a reduction of FA but did not reach significance after correction (p = 0.026, uncorrected, after correction p > 0.05). No significant differences were found for other ROIs (p > 0.05, uncorrected). The comparison results for all ROIs are shown in Table 2.

We have identified reduced low-gamma phase synchronization from the EEG data and reduced white matter integrity of the right inferior cerebellar peduncle from the DTI data for those with a history of mTBI. The correlation analysis further revealed that, among mTBI cases, these two measures were significantly correlated (r = 0.68, p = 0.016) (Figure 3). Weaker low-gamma phase synchronization (averaged across all electrode pairs) corresponded to smaller FA value of the right inferior cerebellar peduncle. Conversely, for controls there was no significant correlation between the two measures (p > 0.05).

Exploration of the relationship between low-gamma phase synchronization and psychological measures (CAPS, PCL-M, PHQ-9, and PSQI) yielded no significant correlations for either group (p > 0.05). Higher low-gamma phase synchronization tends to associate with lower CAPS Criterion D subscore for Hyperarousal in mTBI group, but the correlation (r = −0.43, p = 0.11) was not statistically significant. Exploration of the relationship between the FA value of the right inferior cerebellar peduncle and the psychological measures also yielded no significant correlations for either group (p > 0.05).