- Email: [email protected]
Acute mobile phone operation affects neural function in humans
Clinical Neurophysiology 113 (2002) 1623–1632 www.elsevier.com/locate/clinph
Acute mobile phone operation affects neural function in humans Rodney J. Croft a,b,*, Jody S. Chandler a, Adrian P. Burgess c, Robert J. Barry a, John D. Williams d, Adam R. Clarke a a
Brain and Behaviour Research Institute, University of Wollongong, Northfields Ave., Wollongong 2522, Australia b Centre for Neuropsychopharmacology, Swinburne University of Technology, Hawthorn 3122, Australia c Department of Cognitive Neuroscience and Behaviour, Imperial College Medical School, St. Dunstans Road, London W68RF, UK d Department of Psychology, Coventry University, Priory St., Coventry CV15FB, UK Accepted 24 June 2002
Abstract Objectives: Mobile phones (MP) are used extensively and yet little is known about the effects they may have on human physiology. There have been conflicting reports regarding the relation between MP use and the electroencephalogram (EEG). The present study suggests that this conflict may be due to methodological differences such as exposure durations, and tests whether exposure to an active MP affects EEG as a function of time. Methods: Twenty-four subjects participated in a single-blind fully counterbalanced cross-over design, where both resting EEG and phaselocked neural responses to auditory stimuli were measured while a MP was either operating or turned off. Results: MP exposure altered resting EEG, decreasing 1–4 Hz activity (right hemisphere sites), and increasing 8–12 Hz activity as a function of exposure duration (midline posterior sites). MP exposure also altered early phase-locked neural responses, attenuating the normal response decrement over time in the 4–8 Hz band, decreasing the response in the 12 30 Hz band globally and as a function of time, and increasing midline frontal and lateral posterior responses in the 30–45 Hz band. Conclusions: Active MPs affect neural function in humans and do so as a function of exposure duration. The temporal nature of this effect may contribute to the lack of consistent results reported in the literature. q 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Mobile phone; Electromagnetic field; Resting electroencephalogram; Event-related phase-locked power
1. Introduction Mobile phones (MP) operate on wireless technology, with communication typically occurring via a 900– 1800 MHz signal that is pulsed at 217 Hz. The signal carries essentially no power when the user is not talking or receiving, but when the user communicates the power of this pulsed electromagnetic field (EMF) reaches a maximum of 250 mW. There is concern that this pulsed EMF will reach neurons and directly affect membrane function (Adey and Bawin, 1979), and reflecting this concern is a body of research testing for MP-related changes in human physiology and pathology. Research has failed to find consistent relations between use of MP and human physiology/pathology, and, coupled with the lack of theoretical framework to explain the inconsistencies, there is little consensus on the issue. The electroencephalogram (EEG) has been employed to assess the effect of MP use on humans * Corresponding author. Tel.: 161-3-92148769; fax: 161-3-92145230. E-mail address: [email protected] (R.J. Croft).
because of its sensitivity to immediate changes in neural function. Although this too has failed to identify consistent effects, the present study assumed that the inconsistencies could be explained by methodological limitations, and was designed to determine whether exposure to active MPs could be demonstrated to affect neural function once these limitations were accounted for. The EEG has been employed in a variety of domains to assess the effect of active MPs on neural function. For example, it has been reported that the resting EEG is (Reiser et al., 1995; Lebedeva et al., 2000) and is not (Roschke and Mann, 1997; Hietanen et al., 2000) affected by active MPs, that the sleep EEG is (Mann and Roschke, 1996) and is not (Wagner et al., 2000) affected by active MPs, that phaselocked neural responses to stimuli are (Freude et al., 1998; Jech et al., 2001) and are not (Freude et al., 1998; Eulitz et al., 1998; Urban et al., 1998) affected by active MPs, and that non-phase-locked neural responses to stimuli are affected by active MPs (Eulitz et al., 1998; Krause et al., 2000). While statistical analysis problems may have affected some of these studies (see Section 4), the most
1388-2457/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S 1388-245 7(02)00215-8
CLINPH 2002032
1624
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
common explanation for these discrepant findings have been that only non-phase-locked neural responses to stimuli are affected by exposure to active MPs, particularly where higher cognitive processing is required (Krause et al., 2000; Eulitz et al., 1998). However, this explanation does not accommodate all the data. For example, it requires that we reject the resting EEG findings of Reiser et al. (1995) and Lebedeva et al. (2000), the sleep EEG findings of Mann and Roschke (1996) and trend-level findings of Wagner et al. (1998), and the phase-locked findings of Freude et al. (1998) and Jech et al. (2001), and further, that there is a mechanism that allows MPs to affect non-phase-locked response but not the resting EEG or phase-locked responses to stimuli. A more parsimonious explanation of the above inconsistencies is that the discrepant findings may be due to varying signal-to-noise ratios, the result of both the EEG method and the MP exposure-time used. For example, the most consistent finding is that non-phase-locked responses to stimuli are affected by active MPs (Eulitz et al., 1998; Krause et al., 2000). This dependent measure involves not only large neural responses, but also a powerful averaging technique that increases signal-to-noise ratios separately within each frequency band. Conversely, non-significant findings have tended to result from shorter exposure times (e.g. 3.5 min, Roschke and Mann, 1997; 3 min, Freude et al., 1998; 5 min, Urban et al., 1998) or less powerful analyses (e.g. spectral analysis, Roschke and Mann, 1997; Hietanen et al., 2000). Further, a difficulty for all analyses is that increasing exposure time also increases variability due to normal changes in the subject over the period of the experiment. For example, alpha activity (approximately 8–12 Hz) is sensitive to alterations in alertness, and so if the subject’s alertness changes over the course of the experiment, so too will alpha, error variance and the likelihood of failing to detect a MP-related alpha change. The present study tested the above hypothesis by assessing the effect of exposure to an active MP on the EEG as a function of exposure duration, and by statistically controlling for time differences between the MP and control conditions. To do so, it tested for EEG changes over time under single-blind conditions of either exposure to an active (but non-audible) MP, or the same period of time with no MP exposure. Measures obtained were resting EEG (the subject sits relaxed with their eyes open) and the early phase-locked neural response to auditory stimuli (the subject selectively responds to one of two tones differing only in frequency). These measures were chosen because, as described above, they are thought by some to be unaffected by MPs (e.g. Eulitz et al., 1998; Krause et al., 2000) and thus provide a strong test of the hypothesis that MP use affects the EEG in general. We required subjects to keep their eyes open in the resting EEG to keep them relatively awake, as drowsiness can affect the ‘alpha’ range, and such variance would act as noise and reduce the chances of detecting MP-related changes to alpha. Further, to help elucidate what any MP-
related changes to the EEG represent, we obtained measures of performance (reaction time) and psychological state (how ‘activated’ the subject is). 2. Materials and methods 2.1. Subjects Sixteen males and 8 females participated in the study and were paid A$10. Age ranged from 19 to 48 years (mean ¼ 27.5) and 20 were right-handed. Subjects reported no hearing deficits or neurological history, gave written informed consent and were free to withdraw from the study at any time without penalty. The study was approved by the local ethics committee. 2.2. Procedure On arrival at the laboratory, subjects completed an informed consent form followed by demographics and personality questionnaires. They were fitted with EEG recording apparatus and seated in a recording booth where an EMF stressor (EMF) and an EMF attenuator (EMFatt) were attached. Subjects then performed a 3 min auditory discrimination task after which the subject sat resting for 2 min with their eyes open (resting EEG). As is described in Fig. 1, the 5 min EEG protocol was repeated 4 times, and this 20 min series of tasks were performed 3 times, each with a different background condition (resulting in a total EEG recording time of 60 min). The 3 background conditions were ‘EMF’, consisting of exposure to an active MP, ‘EMFatt’, consisting of concurrent exposure to an active MP and a purported EMF attenuator, and ‘control’, consisting of neither MP nor EMF attenuator. Condition order was counterbalanced across subjects, and subjects were unaware of which condition they were engaged in (single-blind). Further to this, immediately before and after each 20 min series of tasks, the Thaya activation–deactivation adjective check list (AD–ACL; Thaya, 1967) was administered (Fig. 1). 2.3. Discrimination task Each discrimination task consisted of 90 70-dB binaural tones, of which half were 1100 Hz (high) and half were 1000 Hz (low). Tone duration was 60 ms, with a variable interstimulus interval (mean ¼ 2 s; range ¼ 1–3 s; rectangular distribution). Subjects were asked to respond as quickly and accurately as possible to the target tones with a button press. Target-frequency and response-hand were constant within subject and counterbalanced and randomly assigned between subjects. A trial block was given prior to the testing session to familiarise subjects with the task. Behavioural measures obtained were accuracy and reaction time, and to encourage peak performance throughout the session, after the practice block subjects were offered monetary
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
1625
Fig. 1. The experimental protocol is shown. There are 3 20-min conditions (control, EMF, EMFatt), each consisting of a discrimination task and a resting EEG recording that was repeated 4 times. In addition to this, subjects completed the Thaya AD–ACL before and after each condition to obtain a measure of subjects’ psychological ‘activation’ levels. Note: The EMFatt is not considered further in this paper.
reward (A$20) to perform at least as accurately and fast during the remainder of the session as they did in the practice block. 2.4. Materials To enhance ecological validity, a standard Nokia 5110 mobile phone was used as a receiver to generate EMF (900 MHz EMF, 217 Hz pulse rate; 0.577 ms pulse width; estimated average power 3-4 mW; actual emissions during experiment not measured). This was positioned 5 cm radial to the subject’s scalp midway between Oz and Pz, using a non-metallic external bracket. The phone was removed from the recording room and reattached at the start of each condition (to allow the phone to be switched on/off without the subject being aware). To reduce variability due to changes in signal strength, a standard script was employed. The standard script consisted of spoken extracts from a radio play that were played to the sending phone via a CD player at approximately 80 dB SPL (peak level). These were not audible to the subject, but were played during each condition. A ‘Q Link Ally’ (Clarus Products International) was used as EMFatt. This was positioned over the subject’s chest (left sternum) using a comfortable chest-strap, remained in the same place for the duration of the experiment and was operated from a separate room. The EMF and EMFatt were in the same position during each condition. The EMFatt will not be discussed further as it is not relevant to the hypotheses being tested. 2.5. Data acquisition and analysis 2.5.1. Behavioural and psychological measures Accuracy and reaction time (RT) indices were derived from the discrimination task as the percentage of total targets responded to and the mean time taken to respond
to these targets, respectively. The first 4 stimuli from each 3 min task were excluded from the above measures to reduce variability due to confounding effects related to task recommencement. The AD–ACL (Thaya, 1967) consists of 20 words that describe mood or feelings. Subjects rate the degree to which these adjectives describe their mood at that particular point in time on a 4-point Likert-scale. Items relate to the adjectives ‘calm’, ‘excited’, ‘tired’, ‘tense’, and average to form an ‘activation’ scale (low scores represent high activation levels). A difference score was computed from this scale (i.e. ‘condition minus pre-experiment score’) and used as a dependent measure (this will be referred to as ‘activation’). 2.5.2. Electrophysiological measures EEG data were collected from 19 scalp sites (Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, O2) using an electrode cap with tin electrodes, referenced to linked ears. Electro-oculographic data were recorded using tin electrodes above and below the left eye, and on the outer canthus of each eye. Data were continuously sampled at 512 Hz with a 0.05–120 Hz bandpass. Impedances were kept below 5 kOhm. 2.5.3. Resting EEG For each 2 min resting EEG period, the 30–90 s period was EOG-corrected (Semlitsch et al., 1986), divided into 2 s bins and averaged in the frequency domain (fast Fourier transform; cosine window; 10%). Dependent variables were resultant EEG amplitude values grouped into 5 frequency bands: 1–4 Hz (delta); 4–8 Hz (theta); 8–12 Hz (alpha); 12–30 Hz (beta); 30–45 Hz (gamma). These will collectively be referred to as ‘FFT’. FFT data were grouped into the following scalp regions: front left (FL) ¼ mean (FP1, F3, F7); front midline (FM) ¼ Fz; front right (FR) ¼
1626
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
mean (FP2, F4, F8); centre left (CL) ¼ mean (C3, T3); centre midline (CM) ¼ Cz; centre right (CR) ¼ mean (C4, T4); posterior left (PL) ¼ mean (T5, P3, O1); posterior midline (PM) ¼ Pz; posterior right (PR) ¼ mean (T6, P4, O2).
were employed due to difficulties in normalising data, and as two tests were performed per significant contrast, a was reduced to 0.025.
2.5.4. Discrimination task For each 3 min discrimination task, data were EOGcorrected (Semlitsch et al., 1986), epoched 21500 to 1000 ms post-stimulus, and the power of the phase-locked component of these epochs computed for the target and nontarget stimuli separately, for each of the frequency bands defined above for FFT data (12 dB/octave roll-off; zero phase shift; envelope employed). The phase-locked component is derived using complex demodulation, whereby (in essence) epochs are filtered, squared and averaged, with the remaining signal measured relative to a pre-stimulus baseline and expressed as a percentage of this baseline (Kalcher and Pfurtsheller, 1995). This results in a measure of the change in power for each of the frequency bands of interest. Dependent measures for this task were peak percentage power change (relative to the 21000 to 2500 ms baseline) in the early sensory phase-locked response (0–200 ms poststimulus) for each of the frequency bands (early sensory change; ESC). ESC data were grouped into scalp regions as per the FFT data.
3.1. Behavioural and psychological measures
2.6. Statistical analyses For each of the dependent variables FFT, ESC, activation and RT, orthogonal repeated-measures polynomial contrasts were employed to test for relations between these indices and condition (EMF vs. control), and where appropriate, time (first 10 vs. second 10 min of each 20 min condition) and scalp location (sagittal [frontal, central, posterior]; lateral [left, midline, right]). As each condition occurred first, second and third an equal number of times, for each dependent variable ‘order’ was treated as a noise variable and removed by converting scores within each order (and, where appropriate, frequency) to z-scores (after being transformed to normality where appropriate). To maintain appropriate alpha levels, for each of the above indices, the number of planned contrasts was restricted to the number of error degrees of freedom (Tabacknick and Fidell, 1989). Only effects due to condition or interactions between condition and topography, time or type will be reported, and where significant, partial eta 2 values will be given as an estimate of effect size. As significant FFT or ESC contrasts represent changes in the EMF condition, for the EMF condition correlations were performed between the sagittal, lateral and temporal dimensions of significant contrasts, with both RT and activation. For example, for a condition p lateral (left vs. right) effect, a variable would be computed (mean left minus mean right hemisphere values) and correlations performed between this and both RT and activation. Non-parametric correlations
3. Results
For the Control and EMF conditions, mean accuracy levels were 99.0% (SD ¼ 1.8) and 99.3% (SD ¼ 1.3), and their RTs 385.4 ms (SD ¼ 83.8) and 383.4 ms (SD ¼ 75.4), respectively. To approximate normality, reaction time data were transformed according to ‘t_RT ¼ natural log (RT)’. There were no condition or condition p time effects for RT (F½1; 23 , 0:59; P . 0:451). Activation did not differ between the EMF (raw mean ¼ 4.73, SD ¼ 1.52; adjusted mean ¼ 20.085, SD ¼ 0.541) and control (raw mean ¼ 5.33, SD ¼ 1.77; adjusted mean ¼ 0.091, SD ¼ 0.614) conditions (F½1; 23 ¼ 3:31; P ¼ 0:082). 3.2. Electrophysiological measures Resting EEG: An example of a subject’s data (Fig. 2 right panel) as well as the grand average (Fig. 3) resting EEG data is shown for the control and EMF conditions separately. To approximate normality, data were transformed according to ‘t_FFT ¼ natural log (FFT) ’. As can be seen in Fig. 4, while there was less delta power over right hemisphere sites in the control condition, this differential was accentuated in the EMF condition (F½1; 23 ¼ 4:97; P ¼ 0:041; condition p lateral [left vs. right]; eta 2 ¼ 0.17). This pattern correlated inversely with activation (r½24 ¼ 20:55; P ¼ 0:006), and was not related to RT (r½24 ¼ 0:01; P ¼ 0:955). As can be seen in Fig. 4, while there was no effect of EMF on alpha at frontal sites, there was an enhancement of alpha over posterior sites in the EMF condition (F½1; 23 ¼ 6:59; P ¼ 0:017; condition p sagittal [frontal vs. posterior]; eta 2 ¼ 0.22). There was a tendency towards an inverse correlation between this enhancement and RT (r½24 ¼ 20:41; P ¼ 0:046), but not between this enhancement and activation (r½24 ¼ 0:02; P ¼ 0:929). Further, whereas alpha did not change over time in the control condition, it increased over time in the EMF condition at midline sites (F½1; 23 ¼ 5:47; P ¼ 0:028; condition p lateral [lateral vs. midline] p time; eta 2 ¼ 0.19; see Fig. 4 and Fig. 5). This alpha increase did not correlate with activation (r½24 ¼ 0:38; P ¼ 0:068) or RT (r½24 ¼ 0:15; P ¼ 0:418). There were no EMF effects in the theta, beta or gamma bands. 3.3. Discrimination task An example of a subject’s phase-locked beta response (Fig. 2, left panel) as well as grand means across subjects for each frequency band (Fig. 6) is shown for the control and EMF conditions separately. To approximate normality, ESC
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
data were transformed according to ‘t_ESC ¼ natural log (100 2 ESC) ’, such that more positive t_ESC values represent larger neural responses. The MP had no affect in the
1627
delta or alpha band. Whereas the early sensory theta response decreased strongly over time primarily at midline sites in the control condition, as can be seen in Fig. 4 this
Fig. 2. Subject 8’s phase-locked response in the 12–30 Hz range (left) and resting EEG (right) are shown at midline sites for the control and EMF conditions separately.
1628
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
Fig. 3. Grand average resting EEG spectra are shown at 9 sites for control (light grey lines) and EMF (black lines) conditions separately.
decrement was attenuated in the EMF condition (F½1; 23 ¼ 4:82; P ¼ 0:038; condition p lateral [lateral vs. midline] p time; eta 2 ¼ 0.17). This attenuation did not correlate with RT or activation scores (r½24 , 0:28; P . 0:192). Exposure to the active MP caused a reduction in the early phase-locked beta response globally (F½1; 23 ¼ 5:21; P ¼ 0:032; condition; eta 2 ¼ 0.18; Fig. 4). An example of this reduction, at Fz, can be seen in Fig. 6. This reduction was the result of an enhancement of the natural beta reduction over time that was greatest at frontal and posterior sites (F½1; 23 ¼ 5:01; P ¼ 0:035; condition p sagittal [frontoposterior versus central] p time; eta 2 ¼ 0.18). Although the phase-locked beta response values (mean over all sites) did not correlate with RT or activation (r½24 , 0:25; P . 0:245), the decrease in the beta response over time correlated inversely with RT (r½24 ¼ 20:56; P ¼ 0:005), but not with activation (r½24 ¼ 0:27; P ¼ 0:210). Further, whereas there was no effect of the active MP exposure at lateral frontal or midline posterior sites, it increased the
gamma response at midline frontal and lateral posterior sites (F½1; 23 ¼ 6:12; P ¼ 0:021; condition p lateral [lateral vs. midline] p sagittal [frontal vs. posterior]; eta 2 ¼ 0.20; Fig. 4). This differential enhancement was positively related to activation (r½24 ¼ 0:55; P ¼ 0:006), but did not correlate with RT (r½24 ¼ 0:18; P ¼ 0:395).
4. Discussion It has been demonstrated that exposure to an active MP does affect the resting EEG in that MP exposure increased alpha and decreased right hemisphere delta. This is consistent with previous findings in that MP-related changes were found in resting EEG (Hietanen et al., 2000; Lebedeva et al., 2000; Reiser et al., 1995), however, there were important differences in methodology that make direct comparison problematic. For example, Reiser et al. (1995) and Hietanen et al. (2000) analysed each site independently, which may
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
not maximise signal-to-noise ratios and also makes interpretation difficult in terms of type I error. The study of Lebedeva et al. (2000) is also difficult to compare with the present study because it employed a measure of global-
1629
complexity, which does not differentiate between the frequency bands. However, the present results are more clearly consistent with Mann and Roschke (1996) and the trend-level findings of Wagner et al. (1998) in that the alpha
Fig. 4. Effects of the MP are shown for resting EEG (left column) and phase-locked neural activity (right column). Control ¼ filled diamonds; EMF ¼ open squares; ‘a, b, c and d’ represent the 1st, 2nd, 3rd and 4th 5-min period of a 20 min condition, respectively. (A) Delta (1–4 Hz) is reduced at right relative to left hemisphere sites in EMF. (B) EMF exposure increased posterior alpha (8–12 Hz). (C) The EMF increased alpha (8–12 Hz) as a function of time, primarily at midline sites. (D) Whereas there was a reduction in evoked theta (4–8 Hz) in the control condition as a function of exposure duration, primarily at midline sites, this was attenuated in the EMF condition. (E) The EMF enhanced the natural decrease in the evoked beta response (12–30 Hz) that occurred as a function of exposure duration, particularly at fronto-posterior sites. (F) Evoked gamma (30–45 Hz) can be seen to be enhanced by the EMF at frontal midline (front-mid) and posterior lateral sites (post-lat).
1630
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
band was affected by exposure to an active MP. The phaselocked neural response in the discrimination task of the present study was also affected by MP exposure, with global reductions in the beta response, an attenuation of the normal reduction of the theta response, and an increase in the gamma response at midline frontal and lateral posterior sites. The results are thus consistent with Freude et al. (1998) and Jech et al. (2001) in that phase-locked responses are affected by active-MP exposure (although due to methodological differences it again cannot be determined whether, beyond this crucial point, the results are comparable). However, the present results are not consistent with a number of studies that failed to find MP-related effects upon resting and sleep EEG (Mann and Roschke, 1996; Roschke and Mann, 1997; Hietanen et al., 2000; Wagner et al., 2000) and phase-locked neural responses (Freude et al., 1998; Eulitz et al., 1998; Urban et al., 1998). However, as discussed in Section 1, the primary difference between these and the present study is the degree to which noise has been removed. For example, the present study removed the non-condition effect of time, employed topographic information to enhance the signal statistically, analysed different frequencies of phase-locked activity separately, reduced between-subject variability by considering the linear change of the EEG over time, and of particular importance, employed substantive exposure times. Thus, the present study is likely to have reduced type II error and to represent a more valid picture of the effect of active-MP exposure on the EEG. The magnitude of the above effects on neural function were moderate in terms of partial eta 2 estimates, with the effect sizes ranging from 0.17 to 0.22 for resting EEG, and 0.17–0.20 for phase-locked neural response data. However,
it is difficult to know whether such effect sizes are important functionally because resting alpha has not been tightly linked to any particular psychological function. For example, while Kenemans and Lorist (1995) found that 3 mg/kg of caffeine (the equivalent to approximately 2.5 standard cups of coffee) decreased eyes open resting alpha at Pz (5.5% of amplitude) and corresponding to this improved both reaction time (5.7%) and accuracy levels (6.0%), simply closing the eyes will result in a larger change to resting alpha but little change to performance. Consistent with this dissociation, there was a larger MP-related alpha change in the present study (29% amplitude at Pz; untransformed values; see Fig. 3) than might be expected due to a strong cup of coffee and less than might be expected due to eye closure, and yet the MP had no affect on performance and thus less than might be expected due to a strong cup of coffee. Three of the MP-related neural changes were seen as a function of time, with resting alpha increasing linearly, the phase-locked beta response decreasing exponentially, and the normal decrease over time in the phase-locked theta response attenuating as a function of exposure duration. As can be seen in Fig. 4, the effects thus paralleled the course of the experimental treatment and provide validation for their relation with the MP exposure. That a substantial portion of the MP-related effect was related to time suggests that an aspect of the phone’s effect was not due to a direct effect on the EEG, but rather represented an alteration of neural function that was evidenced in the EEG (i.e. the power of the MP did not vary over time, whereas the EEG response did). The reason for this non-linear neural response cannot be determined from the present study. It may be, for example, that the exposure disrupted function, with the longer exposure resulting in less functional reserve, but
Fig. 5. The change in resting alpha (8–12 Hz) over the 20 min testing session (second 10 minus first 10 min recordings) is shown for control and EMF conditions separately. Whereas there was no increase in the control condition from the first to second half of the recording, a large increase can be seen in the EMF condition that was largest at midline sites.
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
alternatively it might represent a cumulative enhancement of function, perhaps due to an accumulation of heat in the neurons (Preece et al., 1999). Thus the delta change that was not time dependent may represent a relatively direct response to the phone, whereas the alpha change that increased over time may represent a less direct response to the phone. Whatever the explanation, this latter nonlinear response offers a resolution to inconsistencies in the literature. That is, studies employing shorter exposure times have tended not to detect MP-related changes while those employing longer exposure times have, and this may merely be the result of the longer exposure times allowing sufficient time for observable neural changes.
1631
Of particular importance to the interpretation of the present findings is that relations were found between the MP-related EEG changes and psychological/behavioural measures. Specifically, the right hemisphere resting delta decrease correlated inversely and strongly with activation, and the enhancement of the phase-locked gamma response correlated directly with activation. This suggests that the neural response to active-MP exposure may be multidimensional (possibly with compensatory mechanisms operating), and may help explain why behavioural effects from activeMP exposure have been difficult to identify (Preece et al., 1999). For example, the phase-locked gamma change and related increase in activation may have been a direct result
Fig. 6. Grand-mean phase-locked responses are shown at Fz for control and EMF conditions separately, for each of the frequency bands analysed. The y-axis represents the change in the power of the response as a percentage of the baseline period.
1632
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
of the MP, with the resting delta change and related decrease in activation representing a compensatory mechanism. Similarly, although there was no overall affect of the MP on RT (P ¼ 0:451), the MP-related phase-locked beta increase was predictive of faster RTs, which again may suggest a multi-faceted neural response to the phone. It is important to note that the causal mechanism of the MP-related EEG changes has not been established in the present study. For example, the present results may have been due to the phone’s EMF acting directly on EEG generators, it may have been the result of a more general response to what may have been ‘interpreted’ as a stressor, or it may even represent subliminal processing of the signal through standard auditory pathways. However, that some of the changes occurred as a function of exposure time suggests that (at least) these changes were not due to a direct effect of the EMF on EEG generators. This study has demonstrated that exposure to an active MP affects neural function in humans, altering both resting EEG patterns and the evoked neural response to auditory stimuli, with a number of these changes occurring as a function of exposure duration. This demonstrates that exposure duration is an important determinant of MP-related effects on the EEG, and suggests that inconsistencies in the literature may be related to differences in exposure durations employed. Additional validation for these findings was that a number of these MP-related changes were correlated with behavioural and psychological measures. Further, this study highlights the need to account for the effect of error variance on the EEG when considering the impact of MP usage on human neural function. Having established that exposure to an active MP does affect neural function, it is important to investigate whether these effects are harmful. Acknowledgements This research was funded by Clarus Products International, LLC, San Rafael, CA, USA. References Adey WR, Bawin SM. Neurosci Res Prog Bull 1979;15:1–129. Eulitz C, Ullsperger P, Freude G, Elbert T. Mobile phones modulate
response patterns of human brain activity. NeuroReport 1998;9:3229– 3232. Freude G, Ullsperger P, Eggert S, Ruppe I. Effects of microwaves emitted by cellular phones on human slow brain potentials. Bioelectromagnetics 1998;19:384–387. Hietanen M, Kovala T, Hamalainen AM. Human brain activity during exposure to radiofrequency fields emitted by cellular phones. Scand J Work Environ Health 2000;26(2):87–92. Jech R, Sonka K, Ruzicka E, Nebuzelsky A, Bohm J, Juklickova M, Nevsimalova S. Electromagnetic field of mobile phones affects visual event related potential in patients with narcolepsy. Bioelectromagnetics 2001;22(7):519–528. Krause CM, Sillanmaki L, Koivisto M, Haggqvist A, Saarela C, Renonsuo A, Laine M, Hamalainen H. Effects of electromagnetic fields emitted by cellular phones on the EEG during a memory task. NeuroReport 2000;11:413–415. Kenemans JL, Lorist MM. Caffeine and selective visual processing. Pharmacol Biochem Behav 1995;52(3):461–471. Kalcher J, Pfurtscheller G. Discrimination between phase-locked and nonphase-locked event-related EEG activity. Electroenceph clin Neurophysiol 1995;94:381–384. Lebedeva NN, Sulimov AV, Sulimova OP, Kotrovskaya TI, Gailus T. Cellular phone electromagnetic field effects on bioelectric activity of human brain. Biomed Eng 2000;28(1 and 2):323–337. Mann K, Roschke J. Effects of pulsed high-frequency electromagnetic fields on human sleep. Neuropsychobiology 1996;33:41–47. Preece AW, Iwi G, Davies-Smith A, Wesnes K, Butler S, Lim E, Varey A. Effect of a 915 MHz simulated mobile phone signal on cognitive function in man. Int J Radiat Biol 1999;75:447–456. Reiser H, Dimpfel W, Schober F. The influence of electromagnetic fields on human brain activity. Eur J Med Res 1995;1:27–32. Roschke J, Mann K. No short term effects of digital mobile radio telephone on the awake human electroencephalogram. Bioelectromagnetics 1997;18:172–176. Semlitsch HV, Anderer P, Schuster P, Presslich O. A solution for reliable and valid reduction of ocular artifacts, applied to the P300. Psychophysiology 1986;23:695–703. Tabacknick BG, Fidell LS. Using multivariate statistics, 2nd ed. New York, NY: Harper Collins, 1989. Thaya RE. Activation, deactivation adjective checklist: measurement of activation through self-report. Psychol Rep 1967;20:663–678. Urban P, Lukas E, Roth Z. Does acute exposure to the electromagnetic field emitted by a mobile phone influence visual evoked potentials? A pilot study. Cent Eur J Public Health 1998;6:288–290. Wagner P, Roschke J, Mann K, Fell J, Hiller W, Frank C, Grozinger M. Human sleep EEG under the influence of pulsed radio frequency electromagnetic fields. Neuropsychobiology 2000;42:207–212. Wagner P, Roschke J, Mann K, Hiller W, Frank C. Human sleep EEG under the influence of pulsed radiofrequency electromagnetic fields: a polysomnographic study using standardised conditions. Bioelectromagnetics 1998;19:199–202.
Acute mobile phone operation affects neural function in humans Rodney J. Croft a,b,*, Jody S. Chandler a, Adrian P. Burgess c, Robert J. Barry a, John D. Williams d, Adam R. Clarke a a
Brain and Behaviour Research Institute, University of Wollongong, Northfields Ave., Wollongong 2522, Australia b Centre for Neuropsychopharmacology, Swinburne University of Technology, Hawthorn 3122, Australia c Department of Cognitive Neuroscience and Behaviour, Imperial College Medical School, St. Dunstans Road, London W68RF, UK d Department of Psychology, Coventry University, Priory St., Coventry CV15FB, UK Accepted 24 June 2002
Abstract Objectives: Mobile phones (MP) are used extensively and yet little is known about the effects they may have on human physiology. There have been conflicting reports regarding the relation between MP use and the electroencephalogram (EEG). The present study suggests that this conflict may be due to methodological differences such as exposure durations, and tests whether exposure to an active MP affects EEG as a function of time. Methods: Twenty-four subjects participated in a single-blind fully counterbalanced cross-over design, where both resting EEG and phaselocked neural responses to auditory stimuli were measured while a MP was either operating or turned off. Results: MP exposure altered resting EEG, decreasing 1–4 Hz activity (right hemisphere sites), and increasing 8–12 Hz activity as a function of exposure duration (midline posterior sites). MP exposure also altered early phase-locked neural responses, attenuating the normal response decrement over time in the 4–8 Hz band, decreasing the response in the 12 30 Hz band globally and as a function of time, and increasing midline frontal and lateral posterior responses in the 30–45 Hz band. Conclusions: Active MPs affect neural function in humans and do so as a function of exposure duration. The temporal nature of this effect may contribute to the lack of consistent results reported in the literature. q 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Mobile phone; Electromagnetic field; Resting electroencephalogram; Event-related phase-locked power
1. Introduction Mobile phones (MP) operate on wireless technology, with communication typically occurring via a 900– 1800 MHz signal that is pulsed at 217 Hz. The signal carries essentially no power when the user is not talking or receiving, but when the user communicates the power of this pulsed electromagnetic field (EMF) reaches a maximum of 250 mW. There is concern that this pulsed EMF will reach neurons and directly affect membrane function (Adey and Bawin, 1979), and reflecting this concern is a body of research testing for MP-related changes in human physiology and pathology. Research has failed to find consistent relations between use of MP and human physiology/pathology, and, coupled with the lack of theoretical framework to explain the inconsistencies, there is little consensus on the issue. The electroencephalogram (EEG) has been employed to assess the effect of MP use on humans * Corresponding author. Tel.: 161-3-92148769; fax: 161-3-92145230. E-mail address: [email protected] (R.J. Croft).
because of its sensitivity to immediate changes in neural function. Although this too has failed to identify consistent effects, the present study assumed that the inconsistencies could be explained by methodological limitations, and was designed to determine whether exposure to active MPs could be demonstrated to affect neural function once these limitations were accounted for. The EEG has been employed in a variety of domains to assess the effect of active MPs on neural function. For example, it has been reported that the resting EEG is (Reiser et al., 1995; Lebedeva et al., 2000) and is not (Roschke and Mann, 1997; Hietanen et al., 2000) affected by active MPs, that the sleep EEG is (Mann and Roschke, 1996) and is not (Wagner et al., 2000) affected by active MPs, that phaselocked neural responses to stimuli are (Freude et al., 1998; Jech et al., 2001) and are not (Freude et al., 1998; Eulitz et al., 1998; Urban et al., 1998) affected by active MPs, and that non-phase-locked neural responses to stimuli are affected by active MPs (Eulitz et al., 1998; Krause et al., 2000). While statistical analysis problems may have affected some of these studies (see Section 4), the most
1388-2457/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S 1388-245 7(02)00215-8
CLINPH 2002032
1624
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
common explanation for these discrepant findings have been that only non-phase-locked neural responses to stimuli are affected by exposure to active MPs, particularly where higher cognitive processing is required (Krause et al., 2000; Eulitz et al., 1998). However, this explanation does not accommodate all the data. For example, it requires that we reject the resting EEG findings of Reiser et al. (1995) and Lebedeva et al. (2000), the sleep EEG findings of Mann and Roschke (1996) and trend-level findings of Wagner et al. (1998), and the phase-locked findings of Freude et al. (1998) and Jech et al. (2001), and further, that there is a mechanism that allows MPs to affect non-phase-locked response but not the resting EEG or phase-locked responses to stimuli. A more parsimonious explanation of the above inconsistencies is that the discrepant findings may be due to varying signal-to-noise ratios, the result of both the EEG method and the MP exposure-time used. For example, the most consistent finding is that non-phase-locked responses to stimuli are affected by active MPs (Eulitz et al., 1998; Krause et al., 2000). This dependent measure involves not only large neural responses, but also a powerful averaging technique that increases signal-to-noise ratios separately within each frequency band. Conversely, non-significant findings have tended to result from shorter exposure times (e.g. 3.5 min, Roschke and Mann, 1997; 3 min, Freude et al., 1998; 5 min, Urban et al., 1998) or less powerful analyses (e.g. spectral analysis, Roschke and Mann, 1997; Hietanen et al., 2000). Further, a difficulty for all analyses is that increasing exposure time also increases variability due to normal changes in the subject over the period of the experiment. For example, alpha activity (approximately 8–12 Hz) is sensitive to alterations in alertness, and so if the subject’s alertness changes over the course of the experiment, so too will alpha, error variance and the likelihood of failing to detect a MP-related alpha change. The present study tested the above hypothesis by assessing the effect of exposure to an active MP on the EEG as a function of exposure duration, and by statistically controlling for time differences between the MP and control conditions. To do so, it tested for EEG changes over time under single-blind conditions of either exposure to an active (but non-audible) MP, or the same period of time with no MP exposure. Measures obtained were resting EEG (the subject sits relaxed with their eyes open) and the early phase-locked neural response to auditory stimuli (the subject selectively responds to one of two tones differing only in frequency). These measures were chosen because, as described above, they are thought by some to be unaffected by MPs (e.g. Eulitz et al., 1998; Krause et al., 2000) and thus provide a strong test of the hypothesis that MP use affects the EEG in general. We required subjects to keep their eyes open in the resting EEG to keep them relatively awake, as drowsiness can affect the ‘alpha’ range, and such variance would act as noise and reduce the chances of detecting MP-related changes to alpha. Further, to help elucidate what any MP-
related changes to the EEG represent, we obtained measures of performance (reaction time) and psychological state (how ‘activated’ the subject is). 2. Materials and methods 2.1. Subjects Sixteen males and 8 females participated in the study and were paid A$10. Age ranged from 19 to 48 years (mean ¼ 27.5) and 20 were right-handed. Subjects reported no hearing deficits or neurological history, gave written informed consent and were free to withdraw from the study at any time without penalty. The study was approved by the local ethics committee. 2.2. Procedure On arrival at the laboratory, subjects completed an informed consent form followed by demographics and personality questionnaires. They were fitted with EEG recording apparatus and seated in a recording booth where an EMF stressor (EMF) and an EMF attenuator (EMFatt) were attached. Subjects then performed a 3 min auditory discrimination task after which the subject sat resting for 2 min with their eyes open (resting EEG). As is described in Fig. 1, the 5 min EEG protocol was repeated 4 times, and this 20 min series of tasks were performed 3 times, each with a different background condition (resulting in a total EEG recording time of 60 min). The 3 background conditions were ‘EMF’, consisting of exposure to an active MP, ‘EMFatt’, consisting of concurrent exposure to an active MP and a purported EMF attenuator, and ‘control’, consisting of neither MP nor EMF attenuator. Condition order was counterbalanced across subjects, and subjects were unaware of which condition they were engaged in (single-blind). Further to this, immediately before and after each 20 min series of tasks, the Thaya activation–deactivation adjective check list (AD–ACL; Thaya, 1967) was administered (Fig. 1). 2.3. Discrimination task Each discrimination task consisted of 90 70-dB binaural tones, of which half were 1100 Hz (high) and half were 1000 Hz (low). Tone duration was 60 ms, with a variable interstimulus interval (mean ¼ 2 s; range ¼ 1–3 s; rectangular distribution). Subjects were asked to respond as quickly and accurately as possible to the target tones with a button press. Target-frequency and response-hand were constant within subject and counterbalanced and randomly assigned between subjects. A trial block was given prior to the testing session to familiarise subjects with the task. Behavioural measures obtained were accuracy and reaction time, and to encourage peak performance throughout the session, after the practice block subjects were offered monetary
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
1625
Fig. 1. The experimental protocol is shown. There are 3 20-min conditions (control, EMF, EMFatt), each consisting of a discrimination task and a resting EEG recording that was repeated 4 times. In addition to this, subjects completed the Thaya AD–ACL before and after each condition to obtain a measure of subjects’ psychological ‘activation’ levels. Note: The EMFatt is not considered further in this paper.
reward (A$20) to perform at least as accurately and fast during the remainder of the session as they did in the practice block. 2.4. Materials To enhance ecological validity, a standard Nokia 5110 mobile phone was used as a receiver to generate EMF (900 MHz EMF, 217 Hz pulse rate; 0.577 ms pulse width; estimated average power 3-4 mW; actual emissions during experiment not measured). This was positioned 5 cm radial to the subject’s scalp midway between Oz and Pz, using a non-metallic external bracket. The phone was removed from the recording room and reattached at the start of each condition (to allow the phone to be switched on/off without the subject being aware). To reduce variability due to changes in signal strength, a standard script was employed. The standard script consisted of spoken extracts from a radio play that were played to the sending phone via a CD player at approximately 80 dB SPL (peak level). These were not audible to the subject, but were played during each condition. A ‘Q Link Ally’ (Clarus Products International) was used as EMFatt. This was positioned over the subject’s chest (left sternum) using a comfortable chest-strap, remained in the same place for the duration of the experiment and was operated from a separate room. The EMF and EMFatt were in the same position during each condition. The EMFatt will not be discussed further as it is not relevant to the hypotheses being tested. 2.5. Data acquisition and analysis 2.5.1. Behavioural and psychological measures Accuracy and reaction time (RT) indices were derived from the discrimination task as the percentage of total targets responded to and the mean time taken to respond
to these targets, respectively. The first 4 stimuli from each 3 min task were excluded from the above measures to reduce variability due to confounding effects related to task recommencement. The AD–ACL (Thaya, 1967) consists of 20 words that describe mood or feelings. Subjects rate the degree to which these adjectives describe their mood at that particular point in time on a 4-point Likert-scale. Items relate to the adjectives ‘calm’, ‘excited’, ‘tired’, ‘tense’, and average to form an ‘activation’ scale (low scores represent high activation levels). A difference score was computed from this scale (i.e. ‘condition minus pre-experiment score’) and used as a dependent measure (this will be referred to as ‘activation’). 2.5.2. Electrophysiological measures EEG data were collected from 19 scalp sites (Fp1, Fp2, F7, F3, Fz, F4, F8, T3, C3, Cz, C4, T4, T5, P3, Pz, P4, T6, O1, O2) using an electrode cap with tin electrodes, referenced to linked ears. Electro-oculographic data were recorded using tin electrodes above and below the left eye, and on the outer canthus of each eye. Data were continuously sampled at 512 Hz with a 0.05–120 Hz bandpass. Impedances were kept below 5 kOhm. 2.5.3. Resting EEG For each 2 min resting EEG period, the 30–90 s period was EOG-corrected (Semlitsch et al., 1986), divided into 2 s bins and averaged in the frequency domain (fast Fourier transform; cosine window; 10%). Dependent variables were resultant EEG amplitude values grouped into 5 frequency bands: 1–4 Hz (delta); 4–8 Hz (theta); 8–12 Hz (alpha); 12–30 Hz (beta); 30–45 Hz (gamma). These will collectively be referred to as ‘FFT’. FFT data were grouped into the following scalp regions: front left (FL) ¼ mean (FP1, F3, F7); front midline (FM) ¼ Fz; front right (FR) ¼
1626
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
mean (FP2, F4, F8); centre left (CL) ¼ mean (C3, T3); centre midline (CM) ¼ Cz; centre right (CR) ¼ mean (C4, T4); posterior left (PL) ¼ mean (T5, P3, O1); posterior midline (PM) ¼ Pz; posterior right (PR) ¼ mean (T6, P4, O2).
were employed due to difficulties in normalising data, and as two tests were performed per significant contrast, a was reduced to 0.025.
2.5.4. Discrimination task For each 3 min discrimination task, data were EOGcorrected (Semlitsch et al., 1986), epoched 21500 to 1000 ms post-stimulus, and the power of the phase-locked component of these epochs computed for the target and nontarget stimuli separately, for each of the frequency bands defined above for FFT data (12 dB/octave roll-off; zero phase shift; envelope employed). The phase-locked component is derived using complex demodulation, whereby (in essence) epochs are filtered, squared and averaged, with the remaining signal measured relative to a pre-stimulus baseline and expressed as a percentage of this baseline (Kalcher and Pfurtsheller, 1995). This results in a measure of the change in power for each of the frequency bands of interest. Dependent measures for this task were peak percentage power change (relative to the 21000 to 2500 ms baseline) in the early sensory phase-locked response (0–200 ms poststimulus) for each of the frequency bands (early sensory change; ESC). ESC data were grouped into scalp regions as per the FFT data.
3.1. Behavioural and psychological measures
2.6. Statistical analyses For each of the dependent variables FFT, ESC, activation and RT, orthogonal repeated-measures polynomial contrasts were employed to test for relations between these indices and condition (EMF vs. control), and where appropriate, time (first 10 vs. second 10 min of each 20 min condition) and scalp location (sagittal [frontal, central, posterior]; lateral [left, midline, right]). As each condition occurred first, second and third an equal number of times, for each dependent variable ‘order’ was treated as a noise variable and removed by converting scores within each order (and, where appropriate, frequency) to z-scores (after being transformed to normality where appropriate). To maintain appropriate alpha levels, for each of the above indices, the number of planned contrasts was restricted to the number of error degrees of freedom (Tabacknick and Fidell, 1989). Only effects due to condition or interactions between condition and topography, time or type will be reported, and where significant, partial eta 2 values will be given as an estimate of effect size. As significant FFT or ESC contrasts represent changes in the EMF condition, for the EMF condition correlations were performed between the sagittal, lateral and temporal dimensions of significant contrasts, with both RT and activation. For example, for a condition p lateral (left vs. right) effect, a variable would be computed (mean left minus mean right hemisphere values) and correlations performed between this and both RT and activation. Non-parametric correlations
3. Results
For the Control and EMF conditions, mean accuracy levels were 99.0% (SD ¼ 1.8) and 99.3% (SD ¼ 1.3), and their RTs 385.4 ms (SD ¼ 83.8) and 383.4 ms (SD ¼ 75.4), respectively. To approximate normality, reaction time data were transformed according to ‘t_RT ¼ natural log (RT)’. There were no condition or condition p time effects for RT (F½1; 23 , 0:59; P . 0:451). Activation did not differ between the EMF (raw mean ¼ 4.73, SD ¼ 1.52; adjusted mean ¼ 20.085, SD ¼ 0.541) and control (raw mean ¼ 5.33, SD ¼ 1.77; adjusted mean ¼ 0.091, SD ¼ 0.614) conditions (F½1; 23 ¼ 3:31; P ¼ 0:082). 3.2. Electrophysiological measures Resting EEG: An example of a subject’s data (Fig. 2 right panel) as well as the grand average (Fig. 3) resting EEG data is shown for the control and EMF conditions separately. To approximate normality, data were transformed according to ‘t_FFT ¼ natural log (FFT) ’. As can be seen in Fig. 4, while there was less delta power over right hemisphere sites in the control condition, this differential was accentuated in the EMF condition (F½1; 23 ¼ 4:97; P ¼ 0:041; condition p lateral [left vs. right]; eta 2 ¼ 0.17). This pattern correlated inversely with activation (r½24 ¼ 20:55; P ¼ 0:006), and was not related to RT (r½24 ¼ 0:01; P ¼ 0:955). As can be seen in Fig. 4, while there was no effect of EMF on alpha at frontal sites, there was an enhancement of alpha over posterior sites in the EMF condition (F½1; 23 ¼ 6:59; P ¼ 0:017; condition p sagittal [frontal vs. posterior]; eta 2 ¼ 0.22). There was a tendency towards an inverse correlation between this enhancement and RT (r½24 ¼ 20:41; P ¼ 0:046), but not between this enhancement and activation (r½24 ¼ 0:02; P ¼ 0:929). Further, whereas alpha did not change over time in the control condition, it increased over time in the EMF condition at midline sites (F½1; 23 ¼ 5:47; P ¼ 0:028; condition p lateral [lateral vs. midline] p time; eta 2 ¼ 0.19; see Fig. 4 and Fig. 5). This alpha increase did not correlate with activation (r½24 ¼ 0:38; P ¼ 0:068) or RT (r½24 ¼ 0:15; P ¼ 0:418). There were no EMF effects in the theta, beta or gamma bands. 3.3. Discrimination task An example of a subject’s phase-locked beta response (Fig. 2, left panel) as well as grand means across subjects for each frequency band (Fig. 6) is shown for the control and EMF conditions separately. To approximate normality, ESC
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
data were transformed according to ‘t_ESC ¼ natural log (100 2 ESC) ’, such that more positive t_ESC values represent larger neural responses. The MP had no affect in the
1627
delta or alpha band. Whereas the early sensory theta response decreased strongly over time primarily at midline sites in the control condition, as can be seen in Fig. 4 this
Fig. 2. Subject 8’s phase-locked response in the 12–30 Hz range (left) and resting EEG (right) are shown at midline sites for the control and EMF conditions separately.
1628
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
Fig. 3. Grand average resting EEG spectra are shown at 9 sites for control (light grey lines) and EMF (black lines) conditions separately.
decrement was attenuated in the EMF condition (F½1; 23 ¼ 4:82; P ¼ 0:038; condition p lateral [lateral vs. midline] p time; eta 2 ¼ 0.17). This attenuation did not correlate with RT or activation scores (r½24 , 0:28; P . 0:192). Exposure to the active MP caused a reduction in the early phase-locked beta response globally (F½1; 23 ¼ 5:21; P ¼ 0:032; condition; eta 2 ¼ 0.18; Fig. 4). An example of this reduction, at Fz, can be seen in Fig. 6. This reduction was the result of an enhancement of the natural beta reduction over time that was greatest at frontal and posterior sites (F½1; 23 ¼ 5:01; P ¼ 0:035; condition p sagittal [frontoposterior versus central] p time; eta 2 ¼ 0.18). Although the phase-locked beta response values (mean over all sites) did not correlate with RT or activation (r½24 , 0:25; P . 0:245), the decrease in the beta response over time correlated inversely with RT (r½24 ¼ 20:56; P ¼ 0:005), but not with activation (r½24 ¼ 0:27; P ¼ 0:210). Further, whereas there was no effect of the active MP exposure at lateral frontal or midline posterior sites, it increased the
gamma response at midline frontal and lateral posterior sites (F½1; 23 ¼ 6:12; P ¼ 0:021; condition p lateral [lateral vs. midline] p sagittal [frontal vs. posterior]; eta 2 ¼ 0.20; Fig. 4). This differential enhancement was positively related to activation (r½24 ¼ 0:55; P ¼ 0:006), but did not correlate with RT (r½24 ¼ 0:18; P ¼ 0:395).
4. Discussion It has been demonstrated that exposure to an active MP does affect the resting EEG in that MP exposure increased alpha and decreased right hemisphere delta. This is consistent with previous findings in that MP-related changes were found in resting EEG (Hietanen et al., 2000; Lebedeva et al., 2000; Reiser et al., 1995), however, there were important differences in methodology that make direct comparison problematic. For example, Reiser et al. (1995) and Hietanen et al. (2000) analysed each site independently, which may
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
not maximise signal-to-noise ratios and also makes interpretation difficult in terms of type I error. The study of Lebedeva et al. (2000) is also difficult to compare with the present study because it employed a measure of global-
1629
complexity, which does not differentiate between the frequency bands. However, the present results are more clearly consistent with Mann and Roschke (1996) and the trend-level findings of Wagner et al. (1998) in that the alpha
Fig. 4. Effects of the MP are shown for resting EEG (left column) and phase-locked neural activity (right column). Control ¼ filled diamonds; EMF ¼ open squares; ‘a, b, c and d’ represent the 1st, 2nd, 3rd and 4th 5-min period of a 20 min condition, respectively. (A) Delta (1–4 Hz) is reduced at right relative to left hemisphere sites in EMF. (B) EMF exposure increased posterior alpha (8–12 Hz). (C) The EMF increased alpha (8–12 Hz) as a function of time, primarily at midline sites. (D) Whereas there was a reduction in evoked theta (4–8 Hz) in the control condition as a function of exposure duration, primarily at midline sites, this was attenuated in the EMF condition. (E) The EMF enhanced the natural decrease in the evoked beta response (12–30 Hz) that occurred as a function of exposure duration, particularly at fronto-posterior sites. (F) Evoked gamma (30–45 Hz) can be seen to be enhanced by the EMF at frontal midline (front-mid) and posterior lateral sites (post-lat).
1630
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
band was affected by exposure to an active MP. The phaselocked neural response in the discrimination task of the present study was also affected by MP exposure, with global reductions in the beta response, an attenuation of the normal reduction of the theta response, and an increase in the gamma response at midline frontal and lateral posterior sites. The results are thus consistent with Freude et al. (1998) and Jech et al. (2001) in that phase-locked responses are affected by active-MP exposure (although due to methodological differences it again cannot be determined whether, beyond this crucial point, the results are comparable). However, the present results are not consistent with a number of studies that failed to find MP-related effects upon resting and sleep EEG (Mann and Roschke, 1996; Roschke and Mann, 1997; Hietanen et al., 2000; Wagner et al., 2000) and phase-locked neural responses (Freude et al., 1998; Eulitz et al., 1998; Urban et al., 1998). However, as discussed in Section 1, the primary difference between these and the present study is the degree to which noise has been removed. For example, the present study removed the non-condition effect of time, employed topographic information to enhance the signal statistically, analysed different frequencies of phase-locked activity separately, reduced between-subject variability by considering the linear change of the EEG over time, and of particular importance, employed substantive exposure times. Thus, the present study is likely to have reduced type II error and to represent a more valid picture of the effect of active-MP exposure on the EEG. The magnitude of the above effects on neural function were moderate in terms of partial eta 2 estimates, with the effect sizes ranging from 0.17 to 0.22 for resting EEG, and 0.17–0.20 for phase-locked neural response data. However,
it is difficult to know whether such effect sizes are important functionally because resting alpha has not been tightly linked to any particular psychological function. For example, while Kenemans and Lorist (1995) found that 3 mg/kg of caffeine (the equivalent to approximately 2.5 standard cups of coffee) decreased eyes open resting alpha at Pz (5.5% of amplitude) and corresponding to this improved both reaction time (5.7%) and accuracy levels (6.0%), simply closing the eyes will result in a larger change to resting alpha but little change to performance. Consistent with this dissociation, there was a larger MP-related alpha change in the present study (29% amplitude at Pz; untransformed values; see Fig. 3) than might be expected due to a strong cup of coffee and less than might be expected due to eye closure, and yet the MP had no affect on performance and thus less than might be expected due to a strong cup of coffee. Three of the MP-related neural changes were seen as a function of time, with resting alpha increasing linearly, the phase-locked beta response decreasing exponentially, and the normal decrease over time in the phase-locked theta response attenuating as a function of exposure duration. As can be seen in Fig. 4, the effects thus paralleled the course of the experimental treatment and provide validation for their relation with the MP exposure. That a substantial portion of the MP-related effect was related to time suggests that an aspect of the phone’s effect was not due to a direct effect on the EEG, but rather represented an alteration of neural function that was evidenced in the EEG (i.e. the power of the MP did not vary over time, whereas the EEG response did). The reason for this non-linear neural response cannot be determined from the present study. It may be, for example, that the exposure disrupted function, with the longer exposure resulting in less functional reserve, but
Fig. 5. The change in resting alpha (8–12 Hz) over the 20 min testing session (second 10 minus first 10 min recordings) is shown for control and EMF conditions separately. Whereas there was no increase in the control condition from the first to second half of the recording, a large increase can be seen in the EMF condition that was largest at midline sites.
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
alternatively it might represent a cumulative enhancement of function, perhaps due to an accumulation of heat in the neurons (Preece et al., 1999). Thus the delta change that was not time dependent may represent a relatively direct response to the phone, whereas the alpha change that increased over time may represent a less direct response to the phone. Whatever the explanation, this latter nonlinear response offers a resolution to inconsistencies in the literature. That is, studies employing shorter exposure times have tended not to detect MP-related changes while those employing longer exposure times have, and this may merely be the result of the longer exposure times allowing sufficient time for observable neural changes.
1631
Of particular importance to the interpretation of the present findings is that relations were found between the MP-related EEG changes and psychological/behavioural measures. Specifically, the right hemisphere resting delta decrease correlated inversely and strongly with activation, and the enhancement of the phase-locked gamma response correlated directly with activation. This suggests that the neural response to active-MP exposure may be multidimensional (possibly with compensatory mechanisms operating), and may help explain why behavioural effects from activeMP exposure have been difficult to identify (Preece et al., 1999). For example, the phase-locked gamma change and related increase in activation may have been a direct result
Fig. 6. Grand-mean phase-locked responses are shown at Fz for control and EMF conditions separately, for each of the frequency bands analysed. The y-axis represents the change in the power of the response as a percentage of the baseline period.
1632
R.J. Croft et al. / Clinical Neurophysiology 113 (2002) 1623–1632
of the MP, with the resting delta change and related decrease in activation representing a compensatory mechanism. Similarly, although there was no overall affect of the MP on RT (P ¼ 0:451), the MP-related phase-locked beta increase was predictive of faster RTs, which again may suggest a multi-faceted neural response to the phone. It is important to note that the causal mechanism of the MP-related EEG changes has not been established in the present study. For example, the present results may have been due to the phone’s EMF acting directly on EEG generators, it may have been the result of a more general response to what may have been ‘interpreted’ as a stressor, or it may even represent subliminal processing of the signal through standard auditory pathways. However, that some of the changes occurred as a function of exposure time suggests that (at least) these changes were not due to a direct effect of the EMF on EEG generators. This study has demonstrated that exposure to an active MP affects neural function in humans, altering both resting EEG patterns and the evoked neural response to auditory stimuli, with a number of these changes occurring as a function of exposure duration. This demonstrates that exposure duration is an important determinant of MP-related effects on the EEG, and suggests that inconsistencies in the literature may be related to differences in exposure durations employed. Additional validation for these findings was that a number of these MP-related changes were correlated with behavioural and psychological measures. Further, this study highlights the need to account for the effect of error variance on the EEG when considering the impact of MP usage on human neural function. Having established that exposure to an active MP does affect neural function, it is important to investigate whether these effects are harmful. Acknowledgements This research was funded by Clarus Products International, LLC, San Rafael, CA, USA. References Adey WR, Bawin SM. Neurosci Res Prog Bull 1979;15:1–129. Eulitz C, Ullsperger P, Freude G, Elbert T. Mobile phones modulate
response patterns of human brain activity. NeuroReport 1998;9:3229– 3232. Freude G, Ullsperger P, Eggert S, Ruppe I. Effects of microwaves emitted by cellular phones on human slow brain potentials. Bioelectromagnetics 1998;19:384–387. Hietanen M, Kovala T, Hamalainen AM. Human brain activity during exposure to radiofrequency fields emitted by cellular phones. Scand J Work Environ Health 2000;26(2):87–92. Jech R, Sonka K, Ruzicka E, Nebuzelsky A, Bohm J, Juklickova M, Nevsimalova S. Electromagnetic field of mobile phones affects visual event related potential in patients with narcolepsy. Bioelectromagnetics 2001;22(7):519–528. Krause CM, Sillanmaki L, Koivisto M, Haggqvist A, Saarela C, Renonsuo A, Laine M, Hamalainen H. Effects of electromagnetic fields emitted by cellular phones on the EEG during a memory task. NeuroReport 2000;11:413–415. Kenemans JL, Lorist MM. Caffeine and selective visual processing. Pharmacol Biochem Behav 1995;52(3):461–471. Kalcher J, Pfurtscheller G. Discrimination between phase-locked and nonphase-locked event-related EEG activity. Electroenceph clin Neurophysiol 1995;94:381–384. Lebedeva NN, Sulimov AV, Sulimova OP, Kotrovskaya TI, Gailus T. Cellular phone electromagnetic field effects on bioelectric activity of human brain. Biomed Eng 2000;28(1 and 2):323–337. Mann K, Roschke J. Effects of pulsed high-frequency electromagnetic fields on human sleep. Neuropsychobiology 1996;33:41–47. Preece AW, Iwi G, Davies-Smith A, Wesnes K, Butler S, Lim E, Varey A. Effect of a 915 MHz simulated mobile phone signal on cognitive function in man. Int J Radiat Biol 1999;75:447–456. Reiser H, Dimpfel W, Schober F. The influence of electromagnetic fields on human brain activity. Eur J Med Res 1995;1:27–32. Roschke J, Mann K. No short term effects of digital mobile radio telephone on the awake human electroencephalogram. Bioelectromagnetics 1997;18:172–176. Semlitsch HV, Anderer P, Schuster P, Presslich O. A solution for reliable and valid reduction of ocular artifacts, applied to the P300. Psychophysiology 1986;23:695–703. Tabacknick BG, Fidell LS. Using multivariate statistics, 2nd ed. New York, NY: Harper Collins, 1989. Thaya RE. Activation, deactivation adjective checklist: measurement of activation through self-report. Psychol Rep 1967;20:663–678. Urban P, Lukas E, Roth Z. Does acute exposure to the electromagnetic field emitted by a mobile phone influence visual evoked potentials? A pilot study. Cent Eur J Public Health 1998;6:288–290. Wagner P, Roschke J, Mann K, Fell J, Hiller W, Frank C, Grozinger M. Human sleep EEG under the influence of pulsed radio frequency electromagnetic fields. Neuropsychobiology 2000;42:207–212. Wagner P, Roschke J, Mann K, Hiller W, Frank C. Human sleep EEG under the influence of pulsed radiofrequency electromagnetic fields: a polysomnographic study using standardised conditions. Bioelectromagnetics 1998;19:199–202.