Short-term radiofrequency exposure from new generation mobile phones reduces EEG alpha power with no effects on cognitive performance

Although mobile phone (MP) use has been steadily increasing in the last decades and similar positive trends are expected for the near future, systematic investigations on neurophysiological and cognitive effects caused by recently developed technological standards for MPs are scarcely available. Here, we investigated the effects of radiofrequency (RF) fields emitted by new-generation mobile technologies, specifically, Universal Mobile Telecommunications System (UMTS) and Long-Term Evolution (LTE), on intrinsic scalp EEG activity in the alpha band (8–12 Hz) and cognitive performance in the Stroop test. The study involved 60 healthy, young-adult university students (34 for UMTS and 26 for LTE) with double-blind administration of Real and Sham exposure in separate sessions. EEG was recorded before, during and after RF exposure, and Stroop performance was assessed before and after EEG recording. Both RF exposure types caused a notable decrease in the alpha power over the whole scalp that persisted even after the cessation of the exposure, whereas no effects were found on any aspects of performance in the Stroop test. The results imply that the brain networks underlying global alpha oscillations might require minor reconfiguration to adapt to the local biophysical changes caused by focal RF exposure mimicking MP use.


Materials and Methods
Participants. Two identical investigations were performed, only differing in the type of RF exposure used (UMTS and LTE), which are hereinafter referred to as UMTS experiment and LTE experiment. Thirty-four healthy participants (20 females, aged 20 ± 3 years) participated in the UMTS experiment. In the LTE experiment, 26 healthy participants were involved (13 females, aged 21 ± 3 years).
Participants were right-handed and medication-free according to self-report. They were full-time students of the University of Pécs, Hungary. All participants had normal or corrected-to-normal vision. Participants reported that they use their right hand for phone calls. They were instructed to abstain from smoking, alcohol and caffeine consumption at least six hours prior to the investigation and to moderate phone use before investigation (less than 30 minutes on investigation day). They were compensated by practical course credit for the time spent in the experiment. All participants signed a written informed consent after the experiment was fully explained. The study was conducted in accordance with the declaration of Helsinki and the protocol was approved by the Regional  Experimental design. Two separate experimental sessions were conducted with at least a one-week interval between the sessions. The timelines of the two sessions are shown in Fig. 1. Both sessions were carried out at an identical time of the day (in the morning or in the afternoon) between 8 am and 6 pm, balanced across participants to avoid possible interference with circadian regulation effects 12 . Participants were seated in a comfortable armchair in front of the screen in a dimly lit room. Both the subject and the experimenter were present in the recording room, but no visual contact was possible during the Stroop tests and EEG recordings. To sustain the alertness of subjects, a muted nature film was played during the EEG recording; the film was started exactly at the beginning and stopped at the end of the resting EEG period (Fig. 1). This controlled visual stimulus contributed to diverting the possible imagery of the investigation or of the RF itself. The order of the two documentary clips, the order of the RF exposure types (Real, Sham) across the two sessions, and the order of the Stroop tasks (WN, CN) of one individual Stroop test within one session and between sessions were randomly assigned and counterbalanced across participants. Exposure to RF energy was administered in a double-blind fashion.

Stroop test.
A computerised version of the Stroop colour word test was designed to assess interference and facilitation effects between two different types of information: colour and word. The test was programmed in the freeware experiment programming language PEBL 43 , a free psychology software package that allows users to create or develop experiments freely without licence or charge (http://pebl.sourceforge.net). Test stimuli (Fig. 2) were coloured words sequentially projected on the screen and the participants had to respond with appropriate key presses according to the task rules on a keyboard from which task-irrelevant buttons were removed. RTs were recorded in ms accuracy on a PC microcomputer for later processing. Stimuli were projected on a screen at a distance of approx. 1.4 m in front of the participant, subtending vertical visual angles of approx. 0.3°. The monitor refresh rate was 120 Hz and the response time was 2 ms. Every Stroop test consisted of three blocks: one training block and two task blocks: CN (in which one had to react to the colour of the word-response by shade) and WN (in which one had to establish the meaning of the word-response by meaning). Each block consisted of only one kind of task. The type of the initial task (WN, CN) was randomly selected for each participant and then alternated within and between sessions (see Fig. 1, mixed elements are separated by horizontal lines). The number of trials per block was 20 in the training block and 60 in the task block, with an equal number of trials from each condition. Stimuli were coloured words "red", "green", "blue", "yellow" (in Hungarian). Three stimulus conditions were used: congruent (=, colour name and colour match), incongruent (≠, colour name does not match colour) and neutral (×, either the colour is black or colour name is 'xxxxx') as control stimuli for comparison purposes.
Stroop test data analysis. To test differences in performance in the two separate sessions (Real/Sham exposure), RTs were recorded and median RT values of different task conditions were computed. Considering the median of RTs instead of mean stems from a concern that RT distributions are positively skewed-similar to the ex-Gaussian distribution 44,45 -and the median value is less affected by outliers and less sensitive to deviation from normality 46 . Based on the distribution of all RTs from all participants in the Sham condition, responses with RTs more than two standard deviations above the mean were considered slow. Participants with more than 10% erroneous responses or more than 10% slow responses to non-training targets were excluded from data analysis, assuming that they were not sufficiently motivated or focused on the task 47 . Only the RTs of correct responses of remaining participants were analysed. In the UMTS experiment, 7 of the 34 participants were excluded from data analysis as exclusion criteria were met: 5 participants were excluded due to errors and 2 due to extreme slowness.  In the LTE experiment, 3 of the 26 participants were excluded: 2 participants due to errors and 1 due to extreme slowness. Reaction time data were analysed with mixed ANOVA to assess possible RF exposure effects on cognitive performance. As IF and FAC are surely detectable only in the CN task 37 , RTs of the WN task were not the focus of statistical analysis. Interference was probed by (2 × 2 × 2) × 2 ANOVA (within-subject factors: exposure (two levels: Real, Sham exposure) × time (two levels: Pre, Post), IF (two levels: CN incongruent, CN neutral)), between-subject factor: RF type (UMTS, LTE). Likewise, FAC was tested with ANOVA (within-subject factors: exposure (two levels: Real, Sham exposure) × time (two levels: Pre, Post), FAC (two levels: CN congruent, CN neutral)), between-subject factor: RF type (UMTS, LTE). In those significant interactions where condition IF or FAC were concerned, post-hoc tests were further investigated. Alpha level was set to 0.05 and Tukey correction was used to control for multiple testing. Assumption checks of normality and of sphericity (Mauchly's Test of Sphericity) were conducted.
Responses are fastest in the congruent condition because of facilitation and slowest in the incongruent condition due to interference, so the difference in RT between these conditions (congruent minus incongruent) yields a cumulative index of the two Stroop-effects (CUM), which is potentially more sensitive to concordant changes in the IF and FAC effects. CUM was studied with a design similar to that of the previous ones: (2 × 2 × 2) × 2 ANOVA (within-subject factors: exposure (two levels: Real, Sham exposure) × time (two levels: Pre, Post), CUM (two levels: CN incongruent, CN congruent)), between-subject factor: RF type (UMTS, LTE).

Subjective ratings.
At the end of the second session in the UMTS experiment, participants were asked the following question (in Hungarian): "What do you think, in which of the two sessions did you get exposed to the actual, Real irradiation? At the first (A) session?/At the second (B) session? /No idea?" Fourteen of them correctly identified the session of Real exposure, 14 participants gave a wrong answer and 6 could not tell the answer. When A and B responses in which the subjects expressed their uncertainty (saying e.g. "… but it is only a guess") were counted as "No idea", the proportion of right and wrong answers was still similar: 11 right, 10 wrong, 13 no idea. Thus, we can conclude that participants in the UMTS experiment really could not establish when the Real UMTS exposure was applied. In the previously conducted LTE experiment, no subjective ratings query was done. RF exposure. UMTS experiment. The UMTS MP exposure system was developed in our laboratories and was successfully used in previous studies 14, [48][49][50][51] . The UMTS RF source was a standard Nokia 6650 (Nokia, Espoo, Finland) MP operated by Phoenix Nokia Service Software (v. 2005/44_4_120; Nokia, Espoo, Finland). The MP was connected to an RF amplifier (Bonn Hungary Electronics Ltd., Hungary) via the external antenna output of the MP. A patch antenna (Reinheimer Elektronik, Wettenberg, Germany; model no: M30EXO-0250-XX) was connected to the output of the RF amplifier. The patch antenna was mounted in a position mimicking the normal use of an MP: the centre of the patch antenna was near the exit of the ear canal, above the tragus, at a distance of 7 mm. A two-position switch located on the front panel of the RF amplifier enabled double-blind experimental conditions: one position was associated with Real exposure, and the other with Sham exposure. Neither the investigator nor the participants were aware of the actual exposure condition. The system was set to WCDMA mode operating at the 1947 MHz carrier frequency (which corresponds to the operating frequency of UMTS MPs in Europe) with a wideband 5 MHz modulation of the RF carrier signal 6 . SAR measurements were performed using the same exposure setup as in this study in Trunk et al. 49 . The averaged SAR was below 2 W/kg in any position within the phantom, meeting the limit of public exposure to RF radiation in the European Union  LTE experiment. The LTE RF exposure system consisted of a programmable signal generator, a power amplifier, the antenna holding fixture and the same patch antenna as in the UMTS study. The LTE "signal cocktail" was generated by an Anritsu MG3700A (Anritsu Co., Japan) programmable signal generator. The signal generator was remotely controlled by a computer in a way that was not visible to the investigator or the participants, either enabling or disabling signal output (corresponding to Real or Sham exposure, respectively), thus enforcing double-blind experimental conditions. For this experiment, we chose a carrier frequency of 1750 MHz (corresponding to one of the operating frequencies of LTE systems in Europe), and the LTE signal used 20 MHz of bandwidth (the allowed maximum). The signal generator was connected to the RF power amplifier BPAM14 (Bonn Hungary Electronics Ltd., Hungary). In our experiment, we set the maximum peak SAR to 1.8 W/kg and the ear-antenna distance to 7 mm. Figure 3 shows a block diagram of both exposure setups. For a detailed description of exposure, see Supplementary Material 1 (Exposure Systems and RF Dosimetry).
EEG data acquisition. Participants were fitted with an elastic EEG cap (Easycap, Munich, Germany) mounted with 32 Ag/AgCl electrodes positioned according to the international 10/20 System, referenced to the nose and grounded on the forehead. Data were recorded with a BrainAmp amplifier (Brain Products GmbH, München, Germany) at sampling rate of 1 kHz, band-pass filtered online with 0.1 Hz low and 450 Hz high cut-off and 45-50 Hz notch. Electrode impedances were below 5 kOhm at the beginning of each session. Electrooculogram (EOG) was recorded from the outer canthus of the right eye for off-line artefact rejection. During the recording, to minimise muscle-derived artefacts, participants were asked to avoid head and eye movements as much as possible. In each session, participants watched a documentary clip with the sound turned off during EEG recording. They were informed that no film-related tasks would be given during the experiment.
EEG data processing. Pre-processing. EEG data were analysed off-line on a personal computer using  Statistical analysis. First, alpha power averaged across all EEG channels (the EOG channel was excluded) was analysed with three-way (2 × 3 × 2) mixed ANOVA: within-subject factors: exposure (two levels: Real, Sham exposure) × time (three levels: Pre, Mid, Post), between-subject factor: RF type (UMTS, LTE). To further characterise exposure effects across time, a follow-up analysis was conducted where pre-exposure alpha power was subtracted from mid-and post-exposure values (hence, the time factor has only two levels here, Mid and Post). Further ANOVAs were performed to examine the effects of the two different RF types separately. After this, we examined the topographic distribution of the effects using mass univariate analyses that were analogous with the ANOVA analyses we conducted. First, we tested for the exposure effect pooling data across RF types, followed by separate analyses for UMTS and LTE exposure. Finally, we explicitly tested whether exposure effects differed between UMTS and LTE exposure types. Each analysis was run on all three time windows and all the channels, yielding 90 tests per analysis (3 time windows × 30 electrodes); the within-analysis familywise error rate (α = 0.05, two-sided) was controlled by means of the cluster-based permutation method as implemented in the FieldTrip toolbox 53,54 . The tests were based on parametric statistics, which were paired-sample t-tests for exposure effects and independent samples t-test on individual Real minus Sham difference scores calculated for each subject for the exposure × RF type interaction. Spatiotemporal clusters were formed using the conventional two-sided α Cluster = 0.05 threshold of the local t-tests. During the clustering, channels were considered neighbours if they were closer than 5 cm from each other, except for Fp1, Fp2, O1 and O2, for which this threshold was 7 cm, because they were more distant from the rest of the electrodes. No minimal spatial cluster extent (in terms of minimum number of neighbouring channels) criterion was set. The number of permutations was 9999 and the maximum sum cluster statistic was used. The resulting p-values, reported as p Cluster , were multiplied by 2 to account for the fact that positive and negative effects were tested against separate null distributions (cfg.correcttail = 'prob' option in FieldTrip).
Although factors were carefully balanced, additional ANOVAs were conducted as control analyses involving factors that may have affected our results, the order of exposure sessions and the order of movies played in the two sessions. The two ANOVAs were designed as follows: within-subject factors: exposure (two levels: Real, Sham exposure) × time (three levels: Pre, Mid, Post), and one of them with between-subject factor session order (Real exposure on first week, Sham exposure on first week) and the other with the between-subject factor film order (film A on first week, film B on first week).

Results
The effect of exposure on Stroop-effects. During   vs. LTE) as a between-subject factor found no significant main effect of (RF type: F 1,53 = 0.724, p = 0.399, η 2 p = 0.013) or interaction with RF type (time × RF type: F 1,53 = 2.927; p = 0.093, η 2 p = 0.052; exposure × RF type: F 1,53 = 0.061; p = 0.806, η 2 p = 0.001; time × exposure × RF type: F 1,53 = 1.562; p = 0.217, η 2 p = 0.029). The mass univariate analysis also yielded concordant results (Fig. 5). In the data pooled across RF types, a significant exposure effect was found (p Cluster = 0.005). The topographic distribution of the difference covered the whole scalp in the mid-and post-exposure period (with only very subtle apparent differences between them), but with no difference in the pre-exposure period (top row of Fig. 5A). UMTS and LTE data analysed separately resulted in marginally significant differences with this method as well (UMTS: p Cluster = 0.067, LTE: p Cluster = 0.055; see Fig. 5A, second row, left and right topoplots for each time period), whereas testing for an RF type × exposure interaction yielded no suprathreshold clusters (see Fig. 5A, second row, smaller topoplots in the middle for each time period). Note that there are several apparent subtle differences in topographic patterns depending on exposure type and between the mid-exposure and post-exposure time periods, but as detailed above, these differences did not result in any significant interaction.
Our analyses focused on the alpha band that was a priori selected based on previous results. We do not present statistics from any other frequency range, but it is nevertheless important to verify that the spectral pattern of the results matches our expectations 55 . In Fig. 6, it is evident that the effect is indeed confined to modulations of the peak in the alpha frequency band.
With regard to our results of control analyses, neither the order of the Real and Sham exposure sessions (session order:

Discussion
We examined the effect of 3G and 4G RF EMF exposure on cognitive functions manifested in Stroop test performance and resting state spontaneous EEG in a double-blind, randomised, counterbalanced, crossover study.
UMTS and LTE exposure caused no measurable effects on the observed parameters of attentional processing in the Stroop test-neither on IF nor on FAC. Both IF and FAC were reliably present during the Real RF exposure in both experiments. Moreover, we did not find any change in the effect of RF even when we considered a more robust variable-the difference between RTs in congruent and incongruent situations. In contrary to the present results obtained in the acute exposure conditions, some epidemiological studies showed shorter 19 or longer 16,17 IF RTs in groups with higher frequency of MP use. It is possible that the change in performance found in the epidemiological studies was not directly caused by RF exposure itself but by other (confounding) factors. Evaluating long-term epidemiological studies (involving data about cognitive abilities, personality and user habits) together with well-controlled laboratory studies could help us to differentiate the general effects of MP using habits from the effects of the RF energy exposure itself. Analysing the effects of RF exposure on resting EEG with data pooled across the UMTS and LTE RF types revealed that, relative to the same time periods in Sham exposure sessions, alpha band power decreased both during and after irradiation. There was no difference between Sham and Real sessions in the pre-exposure baseline, and the exposure effects (decrease of alpha power) in the mid-exposure and post-exposure periods were also significant when assessed relative to the baseline period, but did not differ from each other. The effect was not at all focal to the site of irradiation, but rather appeared to be global, suggesting a brain-wide modulation of alpha oscillations by the RF exposure. UMTS and LTE exposure had similar effects, which imply that either the biophysical mechanism that mediates the neural effect should be to some extent shared between the two RF types or that the differences are too subtle to detect with the current data and methodology.
It is consistent that the EMF can affect the alpha band of resting EEG 56 , although the reason why alpha power reacts differently (increases/decreases) to the RF exposure between studies remains unclear. In terms of waking EEG, the effect of RF (considering GSM systems) has been predominantly shown to increase alpha band power [57][58][59][60][61] , although some reductions of alpha band power have also been reported 56,[62][63][64] . In a single-blind study by D'Costa et al. 62 , during a single 25-min Real GSM exposure, statistically significant decreases were found in EEG power at 8 Hz and 9 Hz in the occipital region, and at 7 Hz and 9 Hz in both the occipital end central recording sites compared to Sham exposure. Perentos et al. 64 observed a suppression of the global alpha band (8-12.75 Hz) activity under a 20-min low-level Real GSM-like extremely low frequency (ranging from direct current to at least 40 kHz) exposure compared to Sham in the exposed hemisphere only. In their follow-up study 63 , it was found that this decrease in alpha band power was present also both under pulse modulated and continuous GSM RF exposure. Another article 56 reported a statistically significant decrease of the alpha band spectral power in the post-exposure session after GSM RF exposure. The most relevant factors that may explain the bidirectional alpha power changes across the different studies are the use of diverse methods, intensities or frequencies and different experimental protocols as discussed by Loughran et al. 65 . Individual variability is also one of the factors that can cause discrepancies between the different results concerning the alpha power 65,66 .
In a recent investigation, of LTE exposure on resting EEG 67 , modulation in the alpha power was also detected. The double-blind, counterbalanced experiment involving 25 male adult subjects used a standard formulation for LTE signals and found reduced alpha band power during and after LTE exposure as compared to the pre-exposure period. In another recent LTE study 68 , the functional connectivity reflected in the neural synchronisation of EEG signals was assessed, and LTE exposure was found to modulate the synchronisation patterns of brain activation also in areas remote to the exposure-even in the contralateral hemisphere. This phenomenon was similar to a previous finding of the same group based on fMRI data 69 where a 30-min LTE RF exposure modulated spontaneous low frequency fluctuations in temporal and frontomedial brain regions, a finding which is also in line with other studies 57,70 .
The strength of alpha oscillations measurable on the scalp is associated with local modulations of excitability related to sensory or congnitive processes within an area: lower alpha power is often related to more intensive processing and higher cortical excitability [71][72][73] . From this perspective, our results are compatible with the findings of recent transcranial magnetic stimulation (TMS) experiments showing incresased motor cortical excitability following acute exposure to GSM-EMF 74 . Alpha oscillations are also thought to form important communication channels providing the basic scaffold of brain-wide visual attentional and memory networks [75][76][77] . The fact that the exposure was localised to the right temporal cortex in our experiment and the alpha power decrease appeared as a global modulation over the whole scalp implies that local biophysical effects of the exposure in the temporal cortex may have led to global network-level changes in the brain. For example, it could be speculated that a change in the local signal-to-noise ratio in a few nodes of a network might require the whole network to adapt by re-tuning the sensitivity of oscillatory communication channels represented by alpha oscillations. The notion that, in the present experiment, the alpha modulation persisted even after the cessation of the irradiation could be due to the local changes also persisting, but also possibly from the temporal dynamics of the underlying network-level adaptation mechanism. Using the Stroop test, we could not show that such network-level alteration would be followed by or manifested in any observable changes in cognitive function. It is possible, again, that a different task might be more sensitive to network-level changes that might be reflected by the alpha modulation, but the presence of an EEG effect and the absence of a cognitive effect together suggest that the effects on neural processing might be subtle and stay well within the adaptation range of the functional brain networks involved in successful performance in the Stroop test.